xref: /illumos-gate/usr/src/uts/common/inet/ip/ip.c (revision 0accf555)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 1990 Mentat Inc.
25  * Copyright (c) 2017 OmniTI Computer Consulting, Inc. All rights reserved.
26  * Copyright (c) 2016 by Delphix. All rights reserved.
27  * Copyright 2020 OmniOS Community Edition (OmniOSce) Association.
28  * Copyright 2021 Joyent, Inc.
29  * Copyright 2022 Oxide Computer Company
30  */
31 
32 #include <sys/types.h>
33 #include <sys/stream.h>
34 #include <sys/dlpi.h>
35 #include <sys/stropts.h>
36 #include <sys/sysmacros.h>
37 #include <sys/strsubr.h>
38 #include <sys/strlog.h>
39 #include <sys/strsun.h>
40 #include <sys/zone.h>
41 #define	_SUN_TPI_VERSION 2
42 #include <sys/tihdr.h>
43 #include <sys/xti_inet.h>
44 #include <sys/ddi.h>
45 #include <sys/suntpi.h>
46 #include <sys/cmn_err.h>
47 #include <sys/debug.h>
48 #include <sys/kobj.h>
49 #include <sys/modctl.h>
50 #include <sys/atomic.h>
51 #include <sys/policy.h>
52 #include <sys/priv.h>
53 #include <sys/taskq.h>
54 
55 #include <sys/systm.h>
56 #include <sys/param.h>
57 #include <sys/kmem.h>
58 #include <sys/sdt.h>
59 #include <sys/socket.h>
60 #include <sys/vtrace.h>
61 #include <sys/isa_defs.h>
62 #include <sys/mac.h>
63 #include <net/if.h>
64 #include <net/if_arp.h>
65 #include <net/route.h>
66 #include <sys/sockio.h>
67 #include <netinet/in.h>
68 #include <net/if_dl.h>
69 
70 #include <inet/common.h>
71 #include <inet/mi.h>
72 #include <inet/mib2.h>
73 #include <inet/nd.h>
74 #include <inet/arp.h>
75 #include <inet/snmpcom.h>
76 #include <inet/optcom.h>
77 #include <inet/kstatcom.h>
78 
79 #include <netinet/igmp_var.h>
80 #include <netinet/ip6.h>
81 #include <netinet/icmp6.h>
82 #include <netinet/sctp.h>
83 
84 #include <inet/ip.h>
85 #include <inet/ip_impl.h>
86 #include <inet/ip6.h>
87 #include <inet/ip6_asp.h>
88 #include <inet/tcp.h>
89 #include <inet/tcp_impl.h>
90 #include <inet/ip_multi.h>
91 #include <inet/ip_if.h>
92 #include <inet/ip_ire.h>
93 #include <inet/ip_ftable.h>
94 #include <inet/ip_rts.h>
95 #include <inet/ip_ndp.h>
96 #include <inet/ip_listutils.h>
97 #include <netinet/igmp.h>
98 #include <netinet/ip_mroute.h>
99 #include <inet/ipp_common.h>
100 #include <inet/cc.h>
101 
102 #include <net/pfkeyv2.h>
103 #include <inet/sadb.h>
104 #include <inet/ipsec_impl.h>
105 #include <inet/iptun/iptun_impl.h>
106 #include <inet/ipdrop.h>
107 #include <inet/ip_netinfo.h>
108 #include <inet/ilb_ip.h>
109 
110 #include <sys/ethernet.h>
111 #include <net/if_types.h>
112 #include <sys/cpuvar.h>
113 
114 #include <ipp/ipp.h>
115 #include <ipp/ipp_impl.h>
116 #include <ipp/ipgpc/ipgpc.h>
117 
118 #include <sys/pattr.h>
119 #include <inet/ipclassifier.h>
120 #include <inet/sctp_ip.h>
121 #include <inet/sctp/sctp_impl.h>
122 #include <inet/udp_impl.h>
123 #include <inet/rawip_impl.h>
124 #include <inet/rts_impl.h>
125 
126 #include <sys/tsol/label.h>
127 #include <sys/tsol/tnet.h>
128 
129 #include <sys/squeue_impl.h>
130 #include <inet/ip_arp.h>
131 
132 #include <sys/clock_impl.h>	/* For LBOLT_FASTPATH{,64} */
133 
134 /*
135  * Values for squeue switch:
136  * IP_SQUEUE_ENTER_NODRAIN: SQ_NODRAIN
137  * IP_SQUEUE_ENTER: SQ_PROCESS
138  * IP_SQUEUE_FILL: SQ_FILL
139  */
140 int ip_squeue_enter = IP_SQUEUE_ENTER;	/* Setable in /etc/system */
141 
142 int ip_squeue_flag;
143 
144 /*
145  * Setable in /etc/system
146  */
147 int ip_poll_normal_ms = 100;
148 int ip_poll_normal_ticks = 0;
149 int ip_modclose_ackwait_ms = 3000;
150 
151 /*
152  * It would be nice to have these present only in DEBUG systems, but the
153  * current design of the global symbol checking logic requires them to be
154  * unconditionally present.
155  */
156 uint_t ip_thread_data;			/* TSD key for debug support */
157 krwlock_t ip_thread_rwlock;
158 list_t	ip_thread_list;
159 
160 /*
161  * Structure to represent a linked list of msgblks. Used by ip_snmp_ functions.
162  */
163 
164 struct listptr_s {
165 	mblk_t	*lp_head;	/* pointer to the head of the list */
166 	mblk_t	*lp_tail;	/* pointer to the tail of the list */
167 };
168 
169 typedef struct listptr_s listptr_t;
170 
171 /*
172  * This is used by ip_snmp_get_mib2_ip_route_media and
173  * ip_snmp_get_mib2_ip6_route_media to carry the lists of return data.
174  */
175 typedef struct iproutedata_s {
176 	uint_t		ird_idx;
177 	uint_t		ird_flags;	/* see below */
178 	listptr_t	ird_route;	/* ipRouteEntryTable */
179 	listptr_t	ird_netmedia;	/* ipNetToMediaEntryTable */
180 	listptr_t	ird_attrs;	/* ipRouteAttributeTable */
181 } iproutedata_t;
182 
183 /* Include ire_testhidden and IRE_IF_CLONE routes */
184 #define	IRD_REPORT_ALL	0x01
185 
186 /*
187  * Cluster specific hooks. These should be NULL when booted as a non-cluster
188  */
189 
190 /*
191  * Hook functions to enable cluster networking
192  * On non-clustered systems these vectors must always be NULL.
193  *
194  * Hook function to Check ip specified ip address is a shared ip address
195  * in the cluster
196  *
197  */
198 int (*cl_inet_isclusterwide)(netstackid_t stack_id, uint8_t protocol,
199     sa_family_t addr_family, uint8_t *laddrp, void *args) = NULL;
200 
201 /*
202  * Hook function to generate cluster wide ip fragment identifier
203  */
204 uint32_t (*cl_inet_ipident)(netstackid_t stack_id, uint8_t protocol,
205     sa_family_t addr_family, uint8_t *laddrp, uint8_t *faddrp,
206     void *args) = NULL;
207 
208 /*
209  * Hook function to generate cluster wide SPI.
210  */
211 void (*cl_inet_getspi)(netstackid_t, uint8_t, uint8_t *, size_t,
212     void *) = NULL;
213 
214 /*
215  * Hook function to verify if the SPI is already utlized.
216  */
217 
218 int (*cl_inet_checkspi)(netstackid_t, uint8_t, uint32_t, void *) = NULL;
219 
220 /*
221  * Hook function to delete the SPI from the cluster wide repository.
222  */
223 
224 void (*cl_inet_deletespi)(netstackid_t, uint8_t, uint32_t, void *) = NULL;
225 
226 /*
227  * Hook function to inform the cluster when packet received on an IDLE SA
228  */
229 
230 void (*cl_inet_idlesa)(netstackid_t, uint8_t, uint32_t, sa_family_t,
231     in6_addr_t, in6_addr_t, void *) = NULL;
232 
233 /*
234  * Synchronization notes:
235  *
236  * IP is a fully D_MP STREAMS module/driver. Thus it does not depend on any
237  * MT level protection given by STREAMS. IP uses a combination of its own
238  * internal serialization mechanism and standard Solaris locking techniques.
239  * The internal serialization is per phyint.  This is used to serialize
240  * plumbing operations, IPMP operations, most set ioctls, etc.
241  *
242  * Plumbing is a long sequence of operations involving message
243  * exchanges between IP, ARP and device drivers. Many set ioctls are typically
244  * involved in plumbing operations. A natural model is to serialize these
245  * ioctls one per ill. For example plumbing of hme0 and qfe0 can go on in
246  * parallel without any interference. But various set ioctls on hme0 are best
247  * serialized, along with IPMP operations and processing of DLPI control
248  * messages received from drivers on a per phyint basis. This serialization is
249  * provided by the ipsq_t and primitives operating on this. Details can
250  * be found in ip_if.c above the core primitives operating on ipsq_t.
251  *
252  * Lookups of an ipif or ill by a thread return a refheld ipif / ill.
253  * Simiarly lookup of an ire by a thread also returns a refheld ire.
254  * In addition ipif's and ill's referenced by the ire are also indirectly
255  * refheld. Thus no ipif or ill can vanish as long as an ipif is refheld
256  * directly or indirectly. For example an SIOCSLIFADDR ioctl that changes the
257  * address of an ipif has to go through the ipsq_t. This ensures that only
258  * one such exclusive operation proceeds at any time on the ipif. It then
259  * waits for all refcnts
260  * associated with this ipif to come down to zero. The address is changed
261  * only after the ipif has been quiesced. Then the ipif is brought up again.
262  * More details are described above the comment in ip_sioctl_flags.
263  *
264  * Packet processing is based mostly on IREs and are fully multi-threaded
265  * using standard Solaris MT techniques.
266  *
267  * There are explicit locks in IP to handle:
268  * - The ip_g_head list maintained by mi_open_link() and friends.
269  *
270  * - The reassembly data structures (one lock per hash bucket)
271  *
272  * - conn_lock is meant to protect conn_t fields. The fields actually
273  *   protected by conn_lock are documented in the conn_t definition.
274  *
275  * - ire_lock to protect some of the fields of the ire, IRE tables
276  *   (one lock per hash bucket). Refer to ip_ire.c for details.
277  *
278  * - ndp_g_lock and ncec_lock for protecting NCEs.
279  *
280  * - ill_lock protects fields of the ill and ipif. Details in ip.h
281  *
282  * - ill_g_lock: This is a global reader/writer lock. Protects the following
283  *	* The AVL tree based global multi list of all ills.
284  *	* The linked list of all ipifs of an ill
285  *	* The <ipsq-xop> mapping
286  *	* <ill-phyint> association
287  *   Insertion/deletion of an ill in the system, insertion/deletion of an ipif
288  *   into an ill, changing the <ipsq-xop> mapping of an ill, changing the
289  *   <ill-phyint> assoc of an ill will all have to hold the ill_g_lock as
290  *   writer for the actual duration of the insertion/deletion/change.
291  *
292  * - ill_lock:  This is a per ill mutex.
293  *   It protects some members of the ill_t struct; see ip.h for details.
294  *   It also protects the <ill-phyint> assoc.
295  *   It also protects the list of ipifs hanging off the ill.
296  *
297  * - ipsq_lock: This is a per ipsq_t mutex lock.
298  *   This protects some members of the ipsq_t struct; see ip.h for details.
299  *   It also protects the <ipsq-ipxop> mapping
300  *
301  * - ipx_lock: This is a per ipxop_t mutex lock.
302  *   This protects some members of the ipxop_t struct; see ip.h for details.
303  *
304  * - phyint_lock: This is a per phyint mutex lock. Protects just the
305  *   phyint_flags
306  *
307  * - ip_addr_avail_lock: This is used to ensure the uniqueness of IP addresses.
308  *   This lock is held in ipif_up_done and the ipif is marked IPIF_UP and the
309  *   uniqueness check also done atomically.
310  *
311  * - ill_g_usesrc_lock: This readers/writer lock protects the usesrc
312  *   group list linked by ill_usesrc_grp_next. It also protects the
313  *   ill_usesrc_ifindex field. It is taken as a writer when a member of the
314  *   group is being added or deleted.  This lock is taken as a reader when
315  *   walking the list/group(eg: to get the number of members in a usesrc group).
316  *   Note, it is only necessary to take this lock if the ill_usesrc_grp_next
317  *   field is changing state i.e from NULL to non-NULL or vice-versa. For
318  *   example, it is not necessary to take this lock in the initial portion
319  *   of ip_sioctl_slifusesrc or at all in ip_sioctl_flags since these
320  *   operations are executed exclusively and that ensures that the "usesrc
321  *   group state" cannot change. The "usesrc group state" change can happen
322  *   only in the latter part of ip_sioctl_slifusesrc and in ill_delete.
323  *
324  * Changing <ill-phyint>, <ipsq-xop> assocications:
325  *
326  * To change the <ill-phyint> association, the ill_g_lock must be held
327  * as writer, and the ill_locks of both the v4 and v6 instance of the ill
328  * must be held.
329  *
330  * To change the <ipsq-xop> association, the ill_g_lock must be held as
331  * writer, the ipsq_lock must be held, and one must be writer on the ipsq.
332  * This is only done when ills are added or removed from IPMP groups.
333  *
334  * To add or delete an ipif from the list of ipifs hanging off the ill,
335  * ill_g_lock (writer) and ill_lock must be held and the thread must be
336  * a writer on the associated ipsq.
337  *
338  * To add or delete an ill to the system, the ill_g_lock must be held as
339  * writer and the thread must be a writer on the associated ipsq.
340  *
341  * To add or delete an ilm to an ill, the ill_lock must be held and the thread
342  * must be a writer on the associated ipsq.
343  *
344  * Lock hierarchy
345  *
346  * Some lock hierarchy scenarios are listed below.
347  *
348  * ill_g_lock -> conn_lock -> ill_lock -> ipsq_lock -> ipx_lock
349  * ill_g_lock -> ill_lock(s) -> phyint_lock
350  * ill_g_lock -> ndp_g_lock -> ill_lock -> ncec_lock
351  * ill_g_lock -> ip_addr_avail_lock
352  * conn_lock -> irb_lock -> ill_lock -> ire_lock
353  * ill_g_lock -> ip_g_nd_lock
354  * ill_g_lock -> ips_ipmp_lock -> ill_lock -> nce_lock
355  * ill_g_lock -> ndp_g_lock -> ill_lock -> ncec_lock -> nce_lock
356  * arl_lock -> ill_lock
357  * ips_ire_dep_lock -> irb_lock
358  *
359  * When more than 1 ill lock is needed to be held, all ill lock addresses
360  * are sorted on address and locked starting from highest addressed lock
361  * downward.
362  *
363  * Multicast scenarios
364  * ips_ill_g_lock -> ill_mcast_lock
365  * conn_ilg_lock -> ips_ill_g_lock -> ill_lock
366  * ill_mcast_serializer -> ill_mcast_lock -> ips_ipmp_lock -> ill_lock
367  * ill_mcast_serializer -> ill_mcast_lock -> connf_lock -> conn_lock
368  * ill_mcast_serializer -> ill_mcast_lock -> conn_ilg_lock
369  * ill_mcast_serializer -> ill_mcast_lock -> ips_igmp_timer_lock
370  *
371  * IPsec scenarios
372  *
373  * ipsa_lock -> ill_g_lock -> ill_lock
374  * ill_g_usesrc_lock -> ill_g_lock -> ill_lock
375  *
376  * Trusted Solaris scenarios
377  *
378  * igsa_lock -> gcgrp_rwlock -> gcgrp_lock
379  * igsa_lock -> gcdb_lock
380  * gcgrp_rwlock -> ire_lock
381  * gcgrp_rwlock -> gcdb_lock
382  *
383  * squeue(sq_lock), flow related (ft_lock, fe_lock) locking
384  *
385  * cpu_lock --> ill_lock --> sqset_lock --> sq_lock
386  * sq_lock -> conn_lock -> QLOCK(q)
387  * ill_lock -> ft_lock -> fe_lock
388  *
389  * Routing/forwarding table locking notes:
390  *
391  * Lock acquisition order: Radix tree lock, irb_lock.
392  * Requirements:
393  * i.  Walker must not hold any locks during the walker callback.
394  * ii  Walker must not see a truncated tree during the walk because of any node
395  *     deletion.
396  * iii Existing code assumes ire_bucket is valid if it is non-null and is used
397  *     in many places in the code to walk the irb list. Thus even if all the
398  *     ires in a bucket have been deleted, we still can't free the radix node
399  *     until the ires have actually been inactive'd (freed).
400  *
401  * Tree traversal - Need to hold the global tree lock in read mode.
402  * Before dropping the global tree lock, need to either increment the ire_refcnt
403  * to ensure that the radix node can't be deleted.
404  *
405  * Tree add - Need to hold the global tree lock in write mode to add a
406  * radix node. To prevent the node from being deleted, increment the
407  * irb_refcnt, after the node is added to the tree. The ire itself is
408  * added later while holding the irb_lock, but not the tree lock.
409  *
410  * Tree delete - Need to hold the global tree lock and irb_lock in write mode.
411  * All associated ires must be inactive (i.e. freed), and irb_refcnt
412  * must be zero.
413  *
414  * Walker - Increment irb_refcnt before calling the walker callback. Hold the
415  * global tree lock (read mode) for traversal.
416  *
417  * IRE dependencies - In some cases we hold ips_ire_dep_lock across ire_refrele
418  * hence we will acquire irb_lock while holding ips_ire_dep_lock.
419  *
420  * IPsec notes :
421  *
422  * IP interacts with the IPsec code (AH/ESP) by storing IPsec attributes
423  * in the ip_xmit_attr_t ip_recv_attr_t. For outbound datagrams, the
424  * ip_xmit_attr_t has the
425  * information used by the IPsec code for applying the right level of
426  * protection. The information initialized by IP in the ip_xmit_attr_t
427  * is determined by the per-socket policy or global policy in the system.
428  * For inbound datagrams, the ip_recv_attr_t
429  * starts out with nothing in it. It gets filled
430  * with the right information if it goes through the AH/ESP code, which
431  * happens if the incoming packet is secure. The information initialized
432  * by AH/ESP, is later used by IP (during fanouts to ULP) to see whether
433  * the policy requirements needed by per-socket policy or global policy
434  * is met or not.
435  *
436  * For fully connected sockets i.e dst, src [addr, port] is known,
437  * conn_policy_cached is set indicating that policy has been cached.
438  * conn_in_enforce_policy may or may not be set depending on whether
439  * there is a global policy match or per-socket policy match.
440  * Policy inheriting happpens in ip_policy_set once the destination is known.
441  * Once the right policy is set on the conn_t, policy cannot change for
442  * this socket. This makes life simpler for TCP (UDP ?) where
443  * re-transmissions go out with the same policy. For symmetry, policy
444  * is cached for fully connected UDP sockets also. Thus if policy is cached,
445  * it also implies that policy is latched i.e policy cannot change
446  * on these sockets. As we have the right policy on the conn, we don't
447  * have to lookup global policy for every outbound and inbound datagram
448  * and thus serving as an optimization. Note that a global policy change
449  * does not affect fully connected sockets if they have policy. If fully
450  * connected sockets did not have any policy associated with it, global
451  * policy change may affect them.
452  *
453  * IP Flow control notes:
454  * ---------------------
455  * Non-TCP streams are flow controlled by IP. The way this is accomplished
456  * differs when ILL_CAPAB_DLD_DIRECT is enabled for that IP instance. When
457  * ILL_DIRECT_CAPABLE(ill) is TRUE, IP can do direct function calls into
458  * GLDv3. Otherwise packets are sent down to lower layers using STREAMS
459  * functions.
460  *
461  * Per Tx ring udp flow control:
462  * This is applicable only when ILL_CAPAB_DLD_DIRECT capability is set in
463  * the ill (i.e. ILL_DIRECT_CAPABLE(ill) is true).
464  *
465  * The underlying link can expose multiple Tx rings to the GLDv3 mac layer.
466  * To achieve best performance, outgoing traffic need to be fanned out among
467  * these Tx ring. mac_tx() is called (via str_mdata_fastpath_put()) to send
468  * traffic out of the NIC and it takes a fanout hint. UDP connections pass
469  * the address of connp as fanout hint to mac_tx(). Under flow controlled
470  * condition, mac_tx() returns a non-NULL cookie (ip_mac_tx_cookie_t). This
471  * cookie points to a specific Tx ring that is blocked. The cookie is used to
472  * hash into an idl_tx_list[] entry in idl_tx_list[] array. Each idl_tx_list_t
473  * point to drain_lists (idl_t's). These drain list will store the blocked UDP
474  * connp's. The drain list is not a single list but a configurable number of
475  * lists.
476  *
477  * The diagram below shows idl_tx_list_t's and their drain_lists. ip_stack_t
478  * has an array of idl_tx_list_t. The size of the array is TX_FANOUT_SIZE
479  * which is equal to 128. This array in turn contains a pointer to idl_t[],
480  * the ip drain list. The idl_t[] array size is MIN(max_ncpus, 8). The drain
481  * list will point to the list of connp's that are flow controlled.
482  *
483  *                      ---------------   -------   -------   -------
484  *                   |->|drain_list[0]|-->|connp|-->|connp|-->|connp|-->
485  *                   |  ---------------   -------   -------   -------
486  *                   |  ---------------   -------   -------   -------
487  *                   |->|drain_list[1]|-->|connp|-->|connp|-->|connp|-->
488  * ----------------  |  ---------------   -------   -------   -------
489  * |idl_tx_list[0]|->|  ---------------   -------   -------   -------
490  * ----------------  |->|drain_list[2]|-->|connp|-->|connp|-->|connp|-->
491  *                   |  ---------------   -------   -------   -------
492  *                   .        .              .         .         .
493  *                   |  ---------------   -------   -------   -------
494  *                   |->|drain_list[n]|-->|connp|-->|connp|-->|connp|-->
495  *                      ---------------   -------   -------   -------
496  *                      ---------------   -------   -------   -------
497  *                   |->|drain_list[0]|-->|connp|-->|connp|-->|connp|-->
498  *                   |  ---------------   -------   -------   -------
499  *                   |  ---------------   -------   -------   -------
500  * ----------------  |->|drain_list[1]|-->|connp|-->|connp|-->|connp|-->
501  * |idl_tx_list[1]|->|  ---------------   -------   -------   -------
502  * ----------------  |        .              .         .         .
503  *                   |  ---------------   -------   -------   -------
504  *                   |->|drain_list[n]|-->|connp|-->|connp|-->|connp|-->
505  *                      ---------------   -------   -------   -------
506  *     .....
507  * ----------------
508  * |idl_tx_list[n]|-> ...
509  * ----------------
510  *
511  * When mac_tx() returns a cookie, the cookie is hashed into an index into
512  * ips_idl_tx_list[], and conn_drain_insert() is called with the idl_tx_list
513  * to insert the conn onto.  conn_drain_insert() asserts flow control for the
514  * sockets via su_txq_full() (non-STREAMS) or QFULL on conn_wq (STREAMS).
515  * Further, conn_blocked is set to indicate that the conn is blocked.
516  *
517  * GLDv3 calls ill_flow_enable() when flow control is relieved.  The cookie
518  * passed in the call to ill_flow_enable() identifies the blocked Tx ring and
519  * is again hashed to locate the appropriate idl_tx_list, which is then
520  * drained via conn_walk_drain().  conn_walk_drain() goes through each conn in
521  * the drain list and calls conn_drain_remove() to clear flow control (via
522  * calling su_txq_full() or clearing QFULL), and remove the conn from the
523  * drain list.
524  *
525  * Note that the drain list is not a single list but a (configurable) array of
526  * lists (8 elements by default).  Synchronization between drain insertion and
527  * flow control wakeup is handled by using idl_txl->txl_lock, and only
528  * conn_drain_insert() and conn_drain_remove() manipulate the drain list.
529  *
530  * Flow control via STREAMS is used when ILL_DIRECT_CAPABLE() returns FALSE.
531  * On the send side, if the packet cannot be sent down to the driver by IP
532  * (canput() fails), ip_xmit() drops the packet and returns EWOULDBLOCK to the
533  * caller, who may then invoke ixa_check_drain_insert() to insert the conn on
534  * the 0'th drain list.  When ip_wsrv() runs on the ill_wq because flow
535  * control has been relieved, the blocked conns in the 0'th drain list are
536  * drained as in the non-STREAMS case.
537  *
538  * In both the STREAMS and non-STREAMS cases, the sockfs upcall to set QFULL
539  * is done when the conn is inserted into the drain list (conn_drain_insert())
540  * and cleared when the conn is removed from the it (conn_drain_remove()).
541  *
542  * IPQOS notes:
543  *
544  * IPQoS Policies are applied to packets using IPPF (IP Policy framework)
545  * and IPQoS modules. IPPF includes hooks in IP at different control points
546  * (callout positions) which direct packets to IPQoS modules for policy
547  * processing. Policies, if present, are global.
548  *
549  * The callout positions are located in the following paths:
550  *		o local_in (packets destined for this host)
551  *		o local_out (packets orginating from this host )
552  *		o fwd_in  (packets forwarded by this m/c - inbound)
553  *		o fwd_out (packets forwarded by this m/c - outbound)
554  * Hooks at these callout points can be enabled/disabled using the ndd variable
555  * ip_policy_mask (a bit mask with the 4 LSB indicating the callout positions).
556  * By default all the callout positions are enabled.
557  *
558  * Outbound (local_out)
559  * Hooks are placed in ire_send_wire_v4 and ire_send_wire_v6.
560  *
561  * Inbound (local_in)
562  * Hooks are placed in ip_fanout_v4 and ip_fanout_v6.
563  *
564  * Forwarding (in and out)
565  * Hooks are placed in ire_recv_forward_v4/v6.
566  *
567  * IP Policy Framework processing (IPPF processing)
568  * Policy processing for a packet is initiated by ip_process, which ascertains
569  * that the classifier (ipgpc) is loaded and configured, failing which the
570  * packet resumes normal processing in IP. If the clasifier is present, the
571  * packet is acted upon by one or more IPQoS modules (action instances), per
572  * filters configured in ipgpc and resumes normal IP processing thereafter.
573  * An action instance can drop a packet in course of its processing.
574  *
575  * Zones notes:
576  *
577  * The partitioning rules for networking are as follows:
578  * 1) Packets coming from a zone must have a source address belonging to that
579  * zone.
580  * 2) Packets coming from a zone can only be sent on a physical interface on
581  * which the zone has an IP address.
582  * 3) Between two zones on the same machine, packet delivery is only allowed if
583  * there's a matching route for the destination and zone in the forwarding
584  * table.
585  * 4) The TCP and UDP port spaces are per-zone; that is, two processes in
586  * different zones can bind to the same port with the wildcard address
587  * (INADDR_ANY).
588  *
589  * The granularity of interface partitioning is at the logical interface level.
590  * Therefore, every zone has its own IP addresses, and incoming packets can be
591  * attributed to a zone unambiguously. A logical interface is placed into a zone
592  * using the SIOCSLIFZONE ioctl; this sets the ipif_zoneid field in the ipif_t
593  * structure. Rule (1) is implemented by modifying the source address selection
594  * algorithm so that the list of eligible addresses is filtered based on the
595  * sending process zone.
596  *
597  * The Internet Routing Entries (IREs) are either exclusive to a zone or shared
598  * across all zones, depending on their type. Here is the break-up:
599  *
600  * IRE type				Shared/exclusive
601  * --------				----------------
602  * IRE_BROADCAST			Exclusive
603  * IRE_DEFAULT (default routes)		Shared (*)
604  * IRE_LOCAL				Exclusive (x)
605  * IRE_LOOPBACK				Exclusive
606  * IRE_PREFIX (net routes)		Shared (*)
607  * IRE_IF_NORESOLVER (interface routes)	Exclusive
608  * IRE_IF_RESOLVER (interface routes)	Exclusive
609  * IRE_IF_CLONE (interface routes)	Exclusive
610  * IRE_HOST (host routes)		Shared (*)
611  *
612  * (*) A zone can only use a default or off-subnet route if the gateway is
613  * directly reachable from the zone, that is, if the gateway's address matches
614  * one of the zone's logical interfaces.
615  *
616  * (x) IRE_LOCAL are handled a bit differently.
617  * When ip_restrict_interzone_loopback is set (the default),
618  * ire_route_recursive restricts loopback using an IRE_LOCAL
619  * between zone to the case when L2 would have conceptually looped the packet
620  * back, i.e. the loopback which is required since neither Ethernet drivers
621  * nor Ethernet hardware loops them back. This is the case when the normal
622  * routes (ignoring IREs with different zoneids) would send out the packet on
623  * the same ill as the ill with which is IRE_LOCAL is associated.
624  *
625  * Multiple zones can share a common broadcast address; typically all zones
626  * share the 255.255.255.255 address. Incoming as well as locally originated
627  * broadcast packets must be dispatched to all the zones on the broadcast
628  * network. For directed broadcasts (e.g. 10.16.72.255) this is not trivial
629  * since some zones may not be on the 10.16.72/24 network. To handle this, each
630  * zone has its own set of IRE_BROADCAST entries; then, broadcast packets are
631  * sent to every zone that has an IRE_BROADCAST entry for the destination
632  * address on the input ill, see ip_input_broadcast().
633  *
634  * Applications in different zones can join the same multicast group address.
635  * The same logic applies for multicast as for broadcast. ip_input_multicast
636  * dispatches packets to all zones that have members on the physical interface.
637  */
638 
639 /*
640  * Squeue Fanout flags:
641  *	0: No fanout.
642  *	1: Fanout across all squeues
643  */
644 boolean_t	ip_squeue_fanout = 0;
645 
646 /*
647  * Maximum dups allowed per packet.
648  */
649 uint_t ip_max_frag_dups = 10;
650 
651 static int	ip_open(queue_t *q, dev_t *devp, int flag, int sflag,
652 		    cred_t *credp, boolean_t isv6);
653 static mblk_t	*ip_xmit_attach_llhdr(mblk_t *, nce_t *);
654 
655 static boolean_t icmp_inbound_verify_v4(mblk_t *, icmph_t *, ip_recv_attr_t *);
656 static void	icmp_inbound_too_big_v4(icmph_t *, ip_recv_attr_t *);
657 static void	icmp_inbound_error_fanout_v4(mblk_t *, icmph_t *,
658     ip_recv_attr_t *);
659 static void	icmp_options_update(ipha_t *);
660 static void	icmp_param_problem(mblk_t *, uint8_t,  ip_recv_attr_t *);
661 static void	icmp_pkt(mblk_t *, void *, size_t, ip_recv_attr_t *);
662 static mblk_t	*icmp_pkt_err_ok(mblk_t *, ip_recv_attr_t *);
663 static void	icmp_redirect_v4(mblk_t *mp, ipha_t *, icmph_t *,
664     ip_recv_attr_t *);
665 static void	icmp_send_redirect(mblk_t *, ipaddr_t, ip_recv_attr_t *);
666 static void	icmp_send_reply_v4(mblk_t *, ipha_t *, icmph_t *,
667     ip_recv_attr_t *);
668 
669 mblk_t		*ip_dlpi_alloc(size_t, t_uscalar_t);
670 char		*ip_dot_addr(ipaddr_t, char *);
671 mblk_t		*ip_carve_mp(mblk_t **, ssize_t);
672 static char	*ip_dot_saddr(uchar_t *, char *);
673 static int	ip_lrput(queue_t *, mblk_t *);
674 ipaddr_t	ip_net_mask(ipaddr_t);
675 char		*ip_nv_lookup(nv_t *, int);
676 int		ip_rput(queue_t *, mblk_t *);
677 static void	ip_rput_dlpi_writer(ipsq_t *dummy_sq, queue_t *q, mblk_t *mp,
678 		    void *dummy_arg);
679 int		ip_snmp_get(queue_t *, mblk_t *, int, boolean_t);
680 static mblk_t	*ip_snmp_get_mib2_ip(queue_t *, mblk_t *,
681 		    mib2_ipIfStatsEntry_t *, ip_stack_t *, boolean_t);
682 static mblk_t	*ip_snmp_get_mib2_ip_traffic_stats(queue_t *, mblk_t *,
683 		    ip_stack_t *, boolean_t);
684 static mblk_t	*ip_snmp_get_mib2_ip6(queue_t *, mblk_t *, ip_stack_t *,
685 		    boolean_t);
686 static mblk_t	*ip_snmp_get_mib2_icmp(queue_t *, mblk_t *, ip_stack_t *ipst);
687 static mblk_t	*ip_snmp_get_mib2_icmp6(queue_t *, mblk_t *, ip_stack_t *ipst);
688 static mblk_t	*ip_snmp_get_mib2_igmp(queue_t *, mblk_t *, ip_stack_t *ipst);
689 static mblk_t	*ip_snmp_get_mib2_multi(queue_t *, mblk_t *, ip_stack_t *ipst);
690 static mblk_t	*ip_snmp_get_mib2_ip_addr(queue_t *, mblk_t *,
691 		    ip_stack_t *ipst, boolean_t);
692 static mblk_t	*ip_snmp_get_mib2_ip6_addr(queue_t *, mblk_t *,
693 		    ip_stack_t *ipst, boolean_t);
694 static mblk_t	*ip_snmp_get_mib2_ip_group_src(queue_t *, mblk_t *,
695 		    ip_stack_t *ipst);
696 static mblk_t	*ip_snmp_get_mib2_ip6_group_src(queue_t *, mblk_t *,
697 		    ip_stack_t *ipst);
698 static mblk_t	*ip_snmp_get_mib2_ip_group_mem(queue_t *, mblk_t *,
699 		    ip_stack_t *ipst);
700 static mblk_t	*ip_snmp_get_mib2_ip6_group_mem(queue_t *, mblk_t *,
701 		    ip_stack_t *ipst);
702 static mblk_t	*ip_snmp_get_mib2_virt_multi(queue_t *, mblk_t *,
703 		    ip_stack_t *ipst);
704 static mblk_t	*ip_snmp_get_mib2_multi_rtable(queue_t *, mblk_t *,
705 		    ip_stack_t *ipst);
706 static mblk_t	*ip_snmp_get_mib2_ip_route_media(queue_t *, mblk_t *, int,
707 		    ip_stack_t *ipst);
708 static mblk_t	*ip_snmp_get_mib2_ip6_route_media(queue_t *, mblk_t *, int,
709 		    ip_stack_t *ipst);
710 static void	ip_snmp_get2_v4(ire_t *, iproutedata_t *);
711 static void	ip_snmp_get2_v6_route(ire_t *, iproutedata_t *);
712 static void	ip_snmp_get2_v4_media(ncec_t *, void *);
713 static void	ip_snmp_get2_v6_media(ncec_t *, void *);
714 int		ip_snmp_set(queue_t *, int, int, uchar_t *, int);
715 
716 static mblk_t	*ip_fragment_copyhdr(uchar_t *, int, int, ip_stack_t *,
717 		    mblk_t *);
718 
719 static void	conn_drain_init(ip_stack_t *);
720 static void	conn_drain_fini(ip_stack_t *);
721 static void	conn_drain(conn_t *connp, boolean_t closing);
722 
723 static void	conn_walk_drain(ip_stack_t *, idl_tx_list_t *);
724 static void	conn_walk_sctp(pfv_t, void *, zoneid_t, netstack_t *);
725 
726 static void	*ip_stack_init(netstackid_t stackid, netstack_t *ns);
727 static void	ip_stack_shutdown(netstackid_t stackid, void *arg);
728 static void	ip_stack_fini(netstackid_t stackid, void *arg);
729 
730 static int	ip_multirt_apply_membership(int (*fn)(conn_t *, boolean_t,
731     const in6_addr_t *, ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *),
732     ire_t *, conn_t *, boolean_t, const in6_addr_t *,  mcast_record_t,
733     const in6_addr_t *);
734 
735 static int	ip_squeue_switch(int);
736 
737 static void	*ip_kstat_init(netstackid_t, ip_stack_t *);
738 static void	ip_kstat_fini(netstackid_t, kstat_t *);
739 static int	ip_kstat_update(kstat_t *kp, int rw);
740 static void	*icmp_kstat_init(netstackid_t);
741 static void	icmp_kstat_fini(netstackid_t, kstat_t *);
742 static int	icmp_kstat_update(kstat_t *kp, int rw);
743 static void	*ip_kstat2_init(netstackid_t, ip_stat_t *);
744 static void	ip_kstat2_fini(netstackid_t, kstat_t *);
745 
746 static void	ipobs_init(ip_stack_t *);
747 static void	ipobs_fini(ip_stack_t *);
748 
749 static int	ip_tp_cpu_update(cpu_setup_t, int, void *);
750 
751 ipaddr_t	ip_g_all_ones = IP_HOST_MASK;
752 
753 static long ip_rput_pullups;
754 int	dohwcksum = 1;	/* use h/w cksum if supported by the hardware */
755 
756 vmem_t *ip_minor_arena_sa; /* for minor nos. from INET_MIN_DEV+2 thru 2^^18-1 */
757 vmem_t *ip_minor_arena_la; /* for minor nos. from 2^^18 thru 2^^32-1 */
758 
759 int	ip_debug;
760 
761 /*
762  * Multirouting/CGTP stuff
763  */
764 int	ip_cgtp_filter_rev = CGTP_FILTER_REV;	/* CGTP hooks version */
765 
766 /*
767  * IP tunables related declarations. Definitions are in ip_tunables.c
768  */
769 extern mod_prop_info_t ip_propinfo_tbl[];
770 extern int ip_propinfo_count;
771 
772 /*
773  * Table of IP ioctls encoding the various properties of the ioctl and
774  * indexed based on the last byte of the ioctl command. Occasionally there
775  * is a clash, and there is more than 1 ioctl with the same last byte.
776  * In such a case 1 ioctl is encoded in the ndx table and the remaining
777  * ioctls are encoded in the misc table. An entry in the ndx table is
778  * retrieved by indexing on the last byte of the ioctl command and comparing
779  * the ioctl command with the value in the ndx table. In the event of a
780  * mismatch the misc table is then searched sequentially for the desired
781  * ioctl command.
782  *
783  * Entry: <command> <copyin_size> <flags> <cmd_type> <function> <restart_func>
784  */
785 ip_ioctl_cmd_t ip_ndx_ioctl_table[] = {
786 	/* 000 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
787 	/* 001 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
788 	/* 002 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
789 	/* 003 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
790 	/* 004 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
791 	/* 005 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
792 	/* 006 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
793 	/* 007 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
794 	/* 008 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
795 	/* 009 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
796 
797 	/* 010 */ { SIOCADDRT,	sizeof (struct rtentry), IPI_PRIV,
798 			MISC_CMD, ip_siocaddrt, NULL },
799 	/* 011 */ { SIOCDELRT,	sizeof (struct rtentry), IPI_PRIV,
800 			MISC_CMD, ip_siocdelrt, NULL },
801 
802 	/* 012 */ { SIOCSIFADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
803 			IF_CMD, ip_sioctl_addr, ip_sioctl_addr_restart },
804 	/* 013 */ { SIOCGIFADDR, sizeof (struct ifreq), IPI_GET_CMD,
805 			IF_CMD, ip_sioctl_get_addr, NULL },
806 
807 	/* 014 */ { SIOCSIFDSTADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
808 			IF_CMD, ip_sioctl_dstaddr, ip_sioctl_dstaddr_restart },
809 	/* 015 */ { SIOCGIFDSTADDR, sizeof (struct ifreq),
810 			IPI_GET_CMD, IF_CMD, ip_sioctl_get_dstaddr, NULL },
811 
812 	/* 016 */ { SIOCSIFFLAGS, sizeof (struct ifreq),
813 			IPI_PRIV | IPI_WR,
814 			IF_CMD, ip_sioctl_flags, ip_sioctl_flags_restart },
815 	/* 017 */ { SIOCGIFFLAGS, sizeof (struct ifreq),
816 			IPI_MODOK | IPI_GET_CMD,
817 			IF_CMD, ip_sioctl_get_flags, NULL },
818 
819 	/* 018 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
820 	/* 019 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
821 
822 	/* copyin size cannot be coded for SIOCGIFCONF */
823 	/* 020 */ { O_SIOCGIFCONF, 0, IPI_GET_CMD,
824 			MISC_CMD, ip_sioctl_get_ifconf, NULL },
825 
826 	/* 021 */ { SIOCSIFMTU,	sizeof (struct ifreq), IPI_PRIV | IPI_WR,
827 			IF_CMD, ip_sioctl_mtu, NULL },
828 	/* 022 */ { SIOCGIFMTU,	sizeof (struct ifreq), IPI_GET_CMD,
829 			IF_CMD, ip_sioctl_get_mtu, NULL },
830 	/* 023 */ { SIOCGIFBRDADDR, sizeof (struct ifreq),
831 			IPI_GET_CMD, IF_CMD, ip_sioctl_get_brdaddr, NULL },
832 	/* 024 */ { SIOCSIFBRDADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
833 			IF_CMD, ip_sioctl_brdaddr, NULL },
834 	/* 025 */ { SIOCGIFNETMASK, sizeof (struct ifreq),
835 			IPI_GET_CMD, IF_CMD, ip_sioctl_get_netmask, NULL },
836 	/* 026 */ { SIOCSIFNETMASK, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
837 			IF_CMD, ip_sioctl_netmask, ip_sioctl_netmask_restart },
838 	/* 027 */ { SIOCGIFMETRIC, sizeof (struct ifreq),
839 			IPI_GET_CMD, IF_CMD, ip_sioctl_get_metric, NULL },
840 	/* 028 */ { SIOCSIFMETRIC, sizeof (struct ifreq), IPI_PRIV,
841 			IF_CMD, ip_sioctl_metric, NULL },
842 	/* 029 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
843 
844 	/* See 166-168 below for extended SIOC*XARP ioctls */
845 	/* 030 */ { SIOCSARP, sizeof (struct arpreq), IPI_PRIV | IPI_WR,
846 			ARP_CMD, ip_sioctl_arp, NULL },
847 	/* 031 */ { SIOCGARP, sizeof (struct arpreq), IPI_GET_CMD,
848 			ARP_CMD, ip_sioctl_arp, NULL },
849 	/* 032 */ { SIOCDARP, sizeof (struct arpreq), IPI_PRIV | IPI_WR,
850 			ARP_CMD, ip_sioctl_arp, NULL },
851 
852 	/* 033 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
853 	/* 034 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
854 	/* 035 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
855 	/* 036 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
856 	/* 037 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
857 	/* 038 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
858 	/* 039 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
859 	/* 040 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
860 	/* 041 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
861 	/* 042 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
862 	/* 043 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
863 	/* 044 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
864 	/* 045 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
865 	/* 046 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
866 	/* 047 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
867 	/* 048 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
868 	/* 049 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
869 	/* 050 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
870 	/* 051 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
871 	/* 052 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
872 	/* 053 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
873 
874 	/* 054 */ { IF_UNITSEL,	sizeof (int), IPI_PRIV | IPI_WR | IPI_MODOK,
875 			MISC_CMD, if_unitsel, if_unitsel_restart },
876 
877 	/* 055 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
878 	/* 056 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
879 	/* 057 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
880 	/* 058 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
881 	/* 059 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
882 	/* 060 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
883 	/* 061 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
884 	/* 062 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
885 	/* 063 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
886 	/* 064 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
887 	/* 065 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
888 	/* 066 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
889 	/* 067 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
890 	/* 068 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
891 	/* 069 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
892 	/* 070 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
893 	/* 071 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
894 	/* 072 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
895 
896 	/* 073 */ { SIOCSIFNAME, sizeof (struct ifreq),
897 			IPI_PRIV | IPI_WR | IPI_MODOK,
898 			IF_CMD, ip_sioctl_sifname, NULL },
899 
900 	/* 074 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
901 	/* 075 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
902 	/* 076 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
903 	/* 077 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
904 	/* 078 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
905 	/* 079 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
906 	/* 080 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
907 	/* 081 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
908 	/* 082 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
909 	/* 083 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
910 	/* 084 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
911 	/* 085 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
912 	/* 086 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
913 
914 	/* 087 */ { SIOCGIFNUM, sizeof (int), IPI_GET_CMD,
915 			MISC_CMD, ip_sioctl_get_ifnum, NULL },
916 	/* 088 */ { SIOCGIFMUXID, sizeof (struct ifreq), IPI_GET_CMD,
917 			IF_CMD, ip_sioctl_get_muxid, NULL },
918 	/* 089 */ { SIOCSIFMUXID, sizeof (struct ifreq),
919 			IPI_PRIV | IPI_WR, IF_CMD, ip_sioctl_muxid, NULL },
920 
921 	/* Both if and lif variants share same func */
922 	/* 090 */ { SIOCGIFINDEX, sizeof (struct ifreq), IPI_GET_CMD,
923 			IF_CMD, ip_sioctl_get_lifindex, NULL },
924 	/* Both if and lif variants share same func */
925 	/* 091 */ { SIOCSIFINDEX, sizeof (struct ifreq),
926 			IPI_PRIV | IPI_WR, IF_CMD, ip_sioctl_slifindex, NULL },
927 
928 	/* copyin size cannot be coded for SIOCGIFCONF */
929 	/* 092 */ { SIOCGIFCONF, 0, IPI_GET_CMD,
930 			MISC_CMD, ip_sioctl_get_ifconf, NULL },
931 	/* 093 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
932 	/* 094 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
933 	/* 095 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
934 	/* 096 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
935 	/* 097 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
936 	/* 098 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
937 	/* 099 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
938 	/* 100 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
939 	/* 101 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
940 	/* 102 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
941 	/* 103 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
942 	/* 104 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
943 	/* 105 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
944 	/* 106 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
945 	/* 107 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
946 	/* 108 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
947 	/* 109 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
948 
949 	/* 110 */ { SIOCLIFREMOVEIF, sizeof (struct lifreq),
950 			IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_removeif,
951 			ip_sioctl_removeif_restart },
952 	/* 111 */ { SIOCLIFADDIF, sizeof (struct lifreq),
953 			IPI_GET_CMD | IPI_PRIV | IPI_WR,
954 			LIF_CMD, ip_sioctl_addif, NULL },
955 #define	SIOCLIFADDR_NDX 112
956 	/* 112 */ { SIOCSLIFADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
957 			LIF_CMD, ip_sioctl_addr, ip_sioctl_addr_restart },
958 	/* 113 */ { SIOCGLIFADDR, sizeof (struct lifreq),
959 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_addr, NULL },
960 	/* 114 */ { SIOCSLIFDSTADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
961 			LIF_CMD, ip_sioctl_dstaddr, ip_sioctl_dstaddr_restart },
962 	/* 115 */ { SIOCGLIFDSTADDR, sizeof (struct lifreq),
963 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_dstaddr, NULL },
964 	/* 116 */ { SIOCSLIFFLAGS, sizeof (struct lifreq),
965 			IPI_PRIV | IPI_WR,
966 			LIF_CMD, ip_sioctl_flags, ip_sioctl_flags_restart },
967 	/* 117 */ { SIOCGLIFFLAGS, sizeof (struct lifreq),
968 			IPI_GET_CMD | IPI_MODOK,
969 			LIF_CMD, ip_sioctl_get_flags, NULL },
970 
971 	/* 118 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
972 	/* 119 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
973 
974 	/* 120 */ { O_SIOCGLIFCONF, 0, IPI_GET_CMD, MISC_CMD,
975 			ip_sioctl_get_lifconf, NULL },
976 	/* 121 */ { SIOCSLIFMTU, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
977 			LIF_CMD, ip_sioctl_mtu, NULL },
978 	/* 122 */ { SIOCGLIFMTU, sizeof (struct lifreq), IPI_GET_CMD,
979 			LIF_CMD, ip_sioctl_get_mtu, NULL },
980 	/* 123 */ { SIOCGLIFBRDADDR, sizeof (struct lifreq),
981 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_brdaddr, NULL },
982 	/* 124 */ { SIOCSLIFBRDADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
983 			LIF_CMD, ip_sioctl_brdaddr, NULL },
984 	/* 125 */ { SIOCGLIFNETMASK, sizeof (struct lifreq),
985 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_netmask, NULL },
986 	/* 126 */ { SIOCSLIFNETMASK, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
987 			LIF_CMD, ip_sioctl_netmask, ip_sioctl_netmask_restart },
988 	/* 127 */ { SIOCGLIFMETRIC, sizeof (struct lifreq),
989 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_metric, NULL },
990 	/* 128 */ { SIOCSLIFMETRIC, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
991 			LIF_CMD, ip_sioctl_metric, NULL },
992 	/* 129 */ { SIOCSLIFNAME, sizeof (struct lifreq),
993 			IPI_PRIV | IPI_WR | IPI_MODOK,
994 			LIF_CMD, ip_sioctl_slifname,
995 			ip_sioctl_slifname_restart },
996 
997 	/* 130 */ { SIOCGLIFNUM, sizeof (struct lifnum), IPI_GET_CMD,
998 			MISC_CMD, ip_sioctl_get_lifnum, NULL },
999 	/* 131 */ { SIOCGLIFMUXID, sizeof (struct lifreq),
1000 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_muxid, NULL },
1001 	/* 132 */ { SIOCSLIFMUXID, sizeof (struct lifreq),
1002 			IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_muxid, NULL },
1003 	/* 133 */ { SIOCGLIFINDEX, sizeof (struct lifreq),
1004 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lifindex, 0 },
1005 	/* 134 */ { SIOCSLIFINDEX, sizeof (struct lifreq),
1006 			IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_slifindex, 0 },
1007 	/* 135 */ { SIOCSLIFTOKEN, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1008 			LIF_CMD, ip_sioctl_token, NULL },
1009 	/* 136 */ { SIOCGLIFTOKEN, sizeof (struct lifreq),
1010 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_token, NULL },
1011 	/* 137 */ { SIOCSLIFSUBNET, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1012 			LIF_CMD, ip_sioctl_subnet, ip_sioctl_subnet_restart },
1013 	/* 138 */ { SIOCGLIFSUBNET, sizeof (struct lifreq),
1014 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_subnet, NULL },
1015 	/* 139 */ { SIOCSLIFLNKINFO, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1016 			LIF_CMD, ip_sioctl_lnkinfo, NULL },
1017 
1018 	/* 140 */ { SIOCGLIFLNKINFO, sizeof (struct lifreq),
1019 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lnkinfo, NULL },
1020 	/* 141 */ { SIOCLIFDELND, sizeof (struct lifreq), IPI_PRIV,
1021 			LIF_CMD, ip_siocdelndp_v6, NULL },
1022 	/* 142 */ { SIOCLIFGETND, sizeof (struct lifreq), IPI_GET_CMD,
1023 			LIF_CMD, ip_siocqueryndp_v6, NULL },
1024 	/* 143 */ { SIOCLIFSETND, sizeof (struct lifreq), IPI_PRIV,
1025 			LIF_CMD, ip_siocsetndp_v6, NULL },
1026 	/* 144 */ { SIOCTMYADDR, sizeof (struct sioc_addrreq), IPI_GET_CMD,
1027 			MISC_CMD, ip_sioctl_tmyaddr, NULL },
1028 	/* 145 */ { SIOCTONLINK, sizeof (struct sioc_addrreq), IPI_GET_CMD,
1029 			MISC_CMD, ip_sioctl_tonlink, NULL },
1030 	/* 146 */ { SIOCTMYSITE, sizeof (struct sioc_addrreq), 0,
1031 			MISC_CMD, ip_sioctl_tmysite, NULL },
1032 	/* 147 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1033 	/* 148 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1034 
1035 	/* Old *IPSECONFIG ioctls are now deprecated, now see spdsock.c */
1036 	/* 149 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1037 	/* 150 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1038 	/* 151 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1039 	/* 152 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1040 
1041 	/* 153 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1042 
1043 	/* 154 */ { SIOCGLIFBINDING, sizeof (struct lifreq), IPI_GET_CMD,
1044 			LIF_CMD, ip_sioctl_get_binding, NULL },
1045 	/* 155 */ { SIOCSLIFGROUPNAME, sizeof (struct lifreq),
1046 			IPI_PRIV | IPI_WR,
1047 			LIF_CMD, ip_sioctl_groupname, ip_sioctl_groupname },
1048 	/* 156 */ { SIOCGLIFGROUPNAME, sizeof (struct lifreq),
1049 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_groupname, NULL },
1050 	/* 157 */ { SIOCGLIFGROUPINFO, sizeof (lifgroupinfo_t),
1051 			IPI_GET_CMD, MISC_CMD, ip_sioctl_groupinfo, NULL },
1052 
1053 	/* Leave 158-160 unused; used to be SIOC*IFARP ioctls */
1054 	/* 158 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1055 	/* 159 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1056 	/* 160 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1057 
1058 	/* 161 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1059 
1060 	/* These are handled in ip_sioctl_copyin_setup itself */
1061 	/* 162 */ { SIOCGIP6ADDRPOLICY, 0, IPI_NULL_BCONT,
1062 			MISC_CMD, NULL, NULL },
1063 	/* 163 */ { SIOCSIP6ADDRPOLICY, 0, IPI_PRIV | IPI_NULL_BCONT,
1064 			MISC_CMD, NULL, NULL },
1065 	/* 164 */ { SIOCGDSTINFO, 0, IPI_GET_CMD, MISC_CMD, NULL, NULL },
1066 
1067 	/* 165 */ { SIOCGLIFCONF, 0, IPI_GET_CMD, MISC_CMD,
1068 			ip_sioctl_get_lifconf, NULL },
1069 
1070 	/* 166 */ { SIOCSXARP, sizeof (struct xarpreq), IPI_PRIV | IPI_WR,
1071 			XARP_CMD, ip_sioctl_arp, NULL },
1072 	/* 167 */ { SIOCGXARP, sizeof (struct xarpreq), IPI_GET_CMD,
1073 			XARP_CMD, ip_sioctl_arp, NULL },
1074 	/* 168 */ { SIOCDXARP, sizeof (struct xarpreq), IPI_PRIV | IPI_WR,
1075 			XARP_CMD, ip_sioctl_arp, NULL },
1076 
1077 	/* SIOCPOPSOCKFS is not handled by IP */
1078 	/* 169 */ { IPI_DONTCARE /* SIOCPOPSOCKFS */, 0, 0, 0, NULL, NULL },
1079 
1080 	/* 170 */ { SIOCGLIFZONE, sizeof (struct lifreq),
1081 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lifzone, NULL },
1082 	/* 171 */ { SIOCSLIFZONE, sizeof (struct lifreq),
1083 			IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_slifzone,
1084 			ip_sioctl_slifzone_restart },
1085 	/* 172-174 are SCTP ioctls and not handled by IP */
1086 	/* 172 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1087 	/* 173 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1088 	/* 174 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1089 	/* 175 */ { SIOCGLIFUSESRC, sizeof (struct lifreq),
1090 			IPI_GET_CMD, LIF_CMD,
1091 			ip_sioctl_get_lifusesrc, 0 },
1092 	/* 176 */ { SIOCSLIFUSESRC, sizeof (struct lifreq),
1093 			IPI_PRIV | IPI_WR,
1094 			LIF_CMD, ip_sioctl_slifusesrc,
1095 			NULL },
1096 	/* 177 */ { SIOCGLIFSRCOF, 0, IPI_GET_CMD, MISC_CMD,
1097 			ip_sioctl_get_lifsrcof, NULL },
1098 	/* 178 */ { SIOCGMSFILTER, sizeof (struct group_filter), IPI_GET_CMD,
1099 			MSFILT_CMD, ip_sioctl_msfilter, NULL },
1100 	/* 179 */ { SIOCSMSFILTER, sizeof (struct group_filter), 0,
1101 			MSFILT_CMD, ip_sioctl_msfilter, NULL },
1102 	/* 180 */ { SIOCGIPMSFILTER, sizeof (struct ip_msfilter), IPI_GET_CMD,
1103 			MSFILT_CMD, ip_sioctl_msfilter, NULL },
1104 	/* 181 */ { SIOCSIPMSFILTER, sizeof (struct ip_msfilter), 0,
1105 			MSFILT_CMD, ip_sioctl_msfilter, NULL },
1106 	/* 182 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1107 	/* SIOCSENABLESDP is handled by SDP */
1108 	/* 183 */ { IPI_DONTCARE /* SIOCSENABLESDP */, 0, 0, 0, NULL, NULL },
1109 	/* 184 */ { IPI_DONTCARE /* SIOCSQPTR */, 0, 0, 0, NULL, NULL },
1110 	/* 185 */ { SIOCGIFHWADDR, sizeof (struct ifreq), IPI_GET_CMD,
1111 			IF_CMD, ip_sioctl_get_ifhwaddr, NULL },
1112 	/* 186 */ { IPI_DONTCARE /* SIOCGSTAMP */, 0, 0, 0, NULL, NULL },
1113 	/* 187 */ { SIOCILB, 0, IPI_PRIV | IPI_GET_CMD, MISC_CMD,
1114 			ip_sioctl_ilb_cmd, NULL },
1115 	/* 188 */ { SIOCGETPROP, 0, IPI_GET_CMD, 0, NULL, NULL },
1116 	/* 189 */ { SIOCSETPROP, 0, IPI_PRIV | IPI_WR, 0, NULL, NULL},
1117 	/* 190 */ { SIOCGLIFDADSTATE, sizeof (struct lifreq),
1118 			IPI_GET_CMD, LIF_CMD, ip_sioctl_get_dadstate, NULL },
1119 	/* 191 */ { SIOCSLIFPREFIX, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1120 			LIF_CMD, ip_sioctl_prefix, ip_sioctl_prefix_restart },
1121 	/* 192 */ { SIOCGLIFHWADDR, sizeof (struct lifreq), IPI_GET_CMD,
1122 			LIF_CMD, ip_sioctl_get_lifhwaddr, NULL }
1123 };
1124 
1125 int ip_ndx_ioctl_count = sizeof (ip_ndx_ioctl_table) / sizeof (ip_ioctl_cmd_t);
1126 
1127 ip_ioctl_cmd_t ip_misc_ioctl_table[] = {
1128 	{ I_LINK,	0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1129 	{ I_UNLINK,	0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1130 	{ I_PLINK,	0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1131 	{ I_PUNLINK,	0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1132 	{ ND_GET,	0, 0, 0, NULL, NULL },
1133 	{ ND_SET,	0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1134 	{ IP_IOCTL,	0, 0, 0, NULL, NULL },
1135 	{ SIOCGETVIFCNT, sizeof (struct sioc_vif_req), IPI_GET_CMD,
1136 		MISC_CMD, mrt_ioctl},
1137 	{ SIOCGETSGCNT,	sizeof (struct sioc_sg_req), IPI_GET_CMD,
1138 		MISC_CMD, mrt_ioctl},
1139 	{ SIOCGETLSGCNT, sizeof (struct sioc_lsg_req), IPI_GET_CMD,
1140 		MISC_CMD, mrt_ioctl}
1141 };
1142 
1143 int ip_misc_ioctl_count =
1144     sizeof (ip_misc_ioctl_table) / sizeof (ip_ioctl_cmd_t);
1145 
1146 int	conn_drain_nthreads;		/* Number of drainers reqd. */
1147 					/* Settable in /etc/system */
1148 /* Defined in ip_ire.c */
1149 extern uint32_t ip_ire_max_bucket_cnt, ip6_ire_max_bucket_cnt;
1150 extern uint32_t ip_ire_min_bucket_cnt, ip6_ire_min_bucket_cnt;
1151 extern uint32_t ip_ire_mem_ratio, ip_ire_cpu_ratio;
1152 
1153 static nv_t	ire_nv_arr[] = {
1154 	{ IRE_BROADCAST, "BROADCAST" },
1155 	{ IRE_LOCAL, "LOCAL" },
1156 	{ IRE_LOOPBACK, "LOOPBACK" },
1157 	{ IRE_DEFAULT, "DEFAULT" },
1158 	{ IRE_PREFIX, "PREFIX" },
1159 	{ IRE_IF_NORESOLVER, "IF_NORESOL" },
1160 	{ IRE_IF_RESOLVER, "IF_RESOLV" },
1161 	{ IRE_IF_CLONE, "IF_CLONE" },
1162 	{ IRE_HOST, "HOST" },
1163 	{ IRE_MULTICAST, "MULTICAST" },
1164 	{ IRE_NOROUTE, "NOROUTE" },
1165 	{ 0 }
1166 };
1167 
1168 nv_t	*ire_nv_tbl = ire_nv_arr;
1169 
1170 /* Simple ICMP IP Header Template */
1171 static ipha_t icmp_ipha = {
1172 	IP_SIMPLE_HDR_VERSION, 0, 0, 0, 0, 0, IPPROTO_ICMP
1173 };
1174 
1175 struct module_info ip_mod_info = {
1176 	IP_MOD_ID, IP_MOD_NAME, IP_MOD_MINPSZ, IP_MOD_MAXPSZ, IP_MOD_HIWAT,
1177 	IP_MOD_LOWAT
1178 };
1179 
1180 /*
1181  * Duplicate static symbols within a module confuses mdb; so we avoid the
1182  * problem by making the symbols here distinct from those in udp.c.
1183  */
1184 
1185 /*
1186  * Entry points for IP as a device and as a module.
1187  * We have separate open functions for the /dev/ip and /dev/ip6 devices.
1188  */
1189 static struct qinit iprinitv4 = {
1190 	ip_rput, NULL, ip_openv4, ip_close, NULL, &ip_mod_info
1191 };
1192 
1193 struct qinit iprinitv6 = {
1194 	ip_rput_v6, NULL, ip_openv6, ip_close, NULL, &ip_mod_info
1195 };
1196 
1197 static struct qinit ipwinit = {
1198 	ip_wput_nondata, ip_wsrv, NULL, NULL, NULL, &ip_mod_info
1199 };
1200 
1201 static struct qinit iplrinit = {
1202 	ip_lrput, NULL, ip_openv4, ip_close, NULL, &ip_mod_info
1203 };
1204 
1205 static struct qinit iplwinit = {
1206 	ip_lwput, NULL, NULL, NULL, NULL, &ip_mod_info
1207 };
1208 
1209 /* For AF_INET aka /dev/ip */
1210 struct streamtab ipinfov4 = {
1211 	&iprinitv4, &ipwinit, &iplrinit, &iplwinit
1212 };
1213 
1214 /* For AF_INET6 aka /dev/ip6 */
1215 struct streamtab ipinfov6 = {
1216 	&iprinitv6, &ipwinit, &iplrinit, &iplwinit
1217 };
1218 
1219 #ifdef	DEBUG
1220 boolean_t skip_sctp_cksum = B_FALSE;
1221 #endif
1222 
1223 /*
1224  * Generate an ICMP fragmentation needed message.
1225  * When called from ip_output side a minimal ip_recv_attr_t needs to be
1226  * constructed by the caller.
1227  */
1228 void
icmp_frag_needed(mblk_t * mp,int mtu,ip_recv_attr_t * ira)1229 icmp_frag_needed(mblk_t *mp, int mtu, ip_recv_attr_t *ira)
1230 {
1231 	icmph_t	icmph;
1232 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
1233 
1234 	mp = icmp_pkt_err_ok(mp, ira);
1235 	if (mp == NULL)
1236 		return;
1237 
1238 	bzero(&icmph, sizeof (icmph_t));
1239 	icmph.icmph_type = ICMP_DEST_UNREACHABLE;
1240 	icmph.icmph_code = ICMP_FRAGMENTATION_NEEDED;
1241 	icmph.icmph_du_mtu = htons((uint16_t)mtu);
1242 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutFragNeeded);
1243 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDestUnreachs);
1244 
1245 	icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
1246 }
1247 
1248 /*
1249  * icmp_inbound_v4 deals with ICMP messages that are handled by IP.
1250  * If the ICMP message is consumed by IP, i.e., it should not be delivered
1251  * to any IPPROTO_ICMP raw sockets, then it returns NULL.
1252  * Likewise, if the ICMP error is misformed (too short, etc), then it
1253  * returns NULL. The caller uses this to determine whether or not to send
1254  * to raw sockets.
1255  *
1256  * All error messages are passed to the matching transport stream.
1257  *
1258  * The following cases are handled by icmp_inbound:
1259  * 1) It needs to send a reply back and possibly delivering it
1260  *    to the "interested" upper clients.
1261  * 2) Return the mblk so that the caller can pass it to the RAW socket clients.
1262  * 3) It needs to change some values in IP only.
1263  * 4) It needs to change some values in IP and upper layers e.g TCP
1264  *    by delivering an error to the upper layers.
1265  *
1266  * We handle the above three cases in the context of IPsec in the
1267  * following way :
1268  *
1269  * 1) Send the reply back in the same way as the request came in.
1270  *    If it came in encrypted, it goes out encrypted. If it came in
1271  *    clear, it goes out in clear. Thus, this will prevent chosen
1272  *    plain text attack.
1273  * 2) The client may or may not expect things to come in secure.
1274  *    If it comes in secure, the policy constraints are checked
1275  *    before delivering it to the upper layers. If it comes in
1276  *    clear, ipsec_inbound_accept_clear will decide whether to
1277  *    accept this in clear or not. In both the cases, if the returned
1278  *    message (IP header + 8 bytes) that caused the icmp message has
1279  *    AH/ESP headers, it is sent up to AH/ESP for validation before
1280  *    sending up. If there are only 8 bytes of returned message, then
1281  *    upper client will not be notified.
1282  * 3) Check with global policy to see whether it matches the constaints.
1283  *    But this will be done only if icmp_accept_messages_in_clear is
1284  *    zero.
1285  * 4) If we need to change both in IP and ULP, then the decision taken
1286  *    while affecting the values in IP and while delivering up to TCP
1287  *    should be the same.
1288  *
1289  *	There are two cases.
1290  *
1291  *	a) If we reject data at the IP layer (ipsec_check_global_policy()
1292  *	   failed), we will not deliver it to the ULP, even though they
1293  *	   are *willing* to accept in *clear*. This is fine as our global
1294  *	   disposition to icmp messages asks us reject the datagram.
1295  *
1296  *	b) If we accept data at the IP layer (ipsec_check_global_policy()
1297  *	   succeeded or icmp_accept_messages_in_clear is 1), and not able
1298  *	   to deliver it to ULP (policy failed), it can lead to
1299  *	   consistency problems. The cases known at this time are
1300  *	   ICMP_DESTINATION_UNREACHABLE  messages with following code
1301  *	   values :
1302  *
1303  *	   - ICMP_FRAGMENTATION_NEEDED : IP adapts to the new value
1304  *	     and Upper layer rejects. Then the communication will
1305  *	     come to a stop. This is solved by making similar decisions
1306  *	     at both levels. Currently, when we are unable to deliver
1307  *	     to the Upper Layer (due to policy failures) while IP has
1308  *	     adjusted dce_pmtu, the next outbound datagram would
1309  *	     generate a local ICMP_FRAGMENTATION_NEEDED message - which
1310  *	     will be with the right level of protection. Thus the right
1311  *	     value will be communicated even if we are not able to
1312  *	     communicate when we get from the wire initially. But this
1313  *	     assumes there would be at least one outbound datagram after
1314  *	     IP has adjusted its dce_pmtu value. To make things
1315  *	     simpler, we accept in clear after the validation of
1316  *	     AH/ESP headers.
1317  *
1318  *	   - Other ICMP ERRORS : We may not be able to deliver it to the
1319  *	     upper layer depending on the level of protection the upper
1320  *	     layer expects and the disposition in ipsec_inbound_accept_clear().
1321  *	     ipsec_inbound_accept_clear() decides whether a given ICMP error
1322  *	     should be accepted in clear when the Upper layer expects secure.
1323  *	     Thus the communication may get aborted by some bad ICMP
1324  *	     packets.
1325  */
1326 mblk_t *
icmp_inbound_v4(mblk_t * mp,ip_recv_attr_t * ira)1327 icmp_inbound_v4(mblk_t *mp, ip_recv_attr_t *ira)
1328 {
1329 	icmph_t		*icmph;
1330 	ipha_t		*ipha;		/* Outer header */
1331 	int		ip_hdr_length;	/* Outer header length */
1332 	boolean_t	interested;
1333 	ipif_t		*ipif;
1334 	uint32_t	ts;
1335 	uint32_t	*tsp;
1336 	timestruc_t	now;
1337 	ill_t		*ill = ira->ira_ill;
1338 	ip_stack_t	*ipst = ill->ill_ipst;
1339 	zoneid_t	zoneid = ira->ira_zoneid;
1340 	int		len_needed;
1341 	mblk_t		*mp_ret = NULL;
1342 
1343 	ipha = (ipha_t *)mp->b_rptr;
1344 
1345 	BUMP_MIB(&ipst->ips_icmp_mib, icmpInMsgs);
1346 
1347 	ip_hdr_length = ira->ira_ip_hdr_length;
1348 	if ((mp->b_wptr - mp->b_rptr) < (ip_hdr_length + ICMPH_SIZE)) {
1349 		if (ira->ira_pktlen < (ip_hdr_length + ICMPH_SIZE)) {
1350 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
1351 			ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
1352 			freemsg(mp);
1353 			return (NULL);
1354 		}
1355 		/* Last chance to get real. */
1356 		ipha = ip_pullup(mp, ip_hdr_length + ICMPH_SIZE, ira);
1357 		if (ipha == NULL) {
1358 			BUMP_MIB(&ipst->ips_icmp_mib, icmpInErrors);
1359 			freemsg(mp);
1360 			return (NULL);
1361 		}
1362 	}
1363 
1364 	/* The IP header will always be a multiple of four bytes */
1365 	icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1366 	ip2dbg(("icmp_inbound_v4: type %d code %d\n", icmph->icmph_type,
1367 	    icmph->icmph_code));
1368 
1369 	/*
1370 	 * We will set "interested" to "true" if we should pass a copy to
1371 	 * the transport or if we handle the packet locally.
1372 	 */
1373 	interested = B_FALSE;
1374 	switch (icmph->icmph_type) {
1375 	case ICMP_ECHO_REPLY:
1376 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInEchoReps);
1377 		break;
1378 	case ICMP_DEST_UNREACHABLE:
1379 		if (icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED)
1380 			BUMP_MIB(&ipst->ips_icmp_mib, icmpInFragNeeded);
1381 		interested = B_TRUE;	/* Pass up to transport */
1382 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInDestUnreachs);
1383 		break;
1384 	case ICMP_SOURCE_QUENCH:
1385 		interested = B_TRUE;	/* Pass up to transport */
1386 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInSrcQuenchs);
1387 		break;
1388 	case ICMP_REDIRECT:
1389 		if (!ipst->ips_ip_ignore_redirect)
1390 			interested = B_TRUE;
1391 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInRedirects);
1392 		break;
1393 	case ICMP_ECHO_REQUEST:
1394 		/*
1395 		 * Whether to respond to echo requests that come in as IP
1396 		 * broadcasts or as IP multicast is subject to debate
1397 		 * (what isn't?).  We aim to please, you pick it.
1398 		 * Default is do it.
1399 		 */
1400 		if (ira->ira_flags & IRAF_MULTICAST) {
1401 			/* multicast: respond based on tunable */
1402 			interested = ipst->ips_ip_g_resp_to_echo_mcast;
1403 		} else if (ira->ira_flags & IRAF_BROADCAST) {
1404 			/* broadcast: respond based on tunable */
1405 			interested = ipst->ips_ip_g_resp_to_echo_bcast;
1406 		} else {
1407 			/* unicast: always respond */
1408 			interested = B_TRUE;
1409 		}
1410 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInEchos);
1411 		if (!interested) {
1412 			/* We never pass these to RAW sockets */
1413 			freemsg(mp);
1414 			return (NULL);
1415 		}
1416 
1417 		/* Check db_ref to make sure we can modify the packet. */
1418 		if (mp->b_datap->db_ref > 1) {
1419 			mblk_t	*mp1;
1420 
1421 			mp1 = copymsg(mp);
1422 			freemsg(mp);
1423 			if (!mp1) {
1424 				BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1425 				return (NULL);
1426 			}
1427 			mp = mp1;
1428 			ipha = (ipha_t *)mp->b_rptr;
1429 			icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1430 		}
1431 		icmph->icmph_type = ICMP_ECHO_REPLY;
1432 		BUMP_MIB(&ipst->ips_icmp_mib, icmpOutEchoReps);
1433 		icmp_send_reply_v4(mp, ipha, icmph, ira);
1434 		return (NULL);
1435 
1436 	case ICMP_ROUTER_ADVERTISEMENT:
1437 	case ICMP_ROUTER_SOLICITATION:
1438 		break;
1439 	case ICMP_TIME_EXCEEDED:
1440 		interested = B_TRUE;	/* Pass up to transport */
1441 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimeExcds);
1442 		break;
1443 	case ICMP_PARAM_PROBLEM:
1444 		interested = B_TRUE;	/* Pass up to transport */
1445 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInParmProbs);
1446 		break;
1447 	case ICMP_TIME_STAMP_REQUEST:
1448 		/* Response to Time Stamp Requests is local policy. */
1449 		if (ipst->ips_ip_g_resp_to_timestamp) {
1450 			if (ira->ira_flags & IRAF_MULTIBROADCAST)
1451 				interested =
1452 				    ipst->ips_ip_g_resp_to_timestamp_bcast;
1453 			else
1454 				interested = B_TRUE;
1455 		}
1456 		if (!interested) {
1457 			/* We never pass these to RAW sockets */
1458 			freemsg(mp);
1459 			return (NULL);
1460 		}
1461 
1462 		/* Make sure we have enough of the packet */
1463 		len_needed = ip_hdr_length + ICMPH_SIZE +
1464 		    3 * sizeof (uint32_t);
1465 
1466 		if (mp->b_wptr - mp->b_rptr < len_needed) {
1467 			ipha = ip_pullup(mp, len_needed, ira);
1468 			if (ipha == NULL) {
1469 				BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1470 				ip_drop_input("ipIfStatsInDiscards - ip_pullup",
1471 				    mp, ill);
1472 				freemsg(mp);
1473 				return (NULL);
1474 			}
1475 			/* Refresh following the pullup. */
1476 			icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1477 		}
1478 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimestamps);
1479 		/* Check db_ref to make sure we can modify the packet. */
1480 		if (mp->b_datap->db_ref > 1) {
1481 			mblk_t	*mp1;
1482 
1483 			mp1 = copymsg(mp);
1484 			freemsg(mp);
1485 			if (!mp1) {
1486 				BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1487 				return (NULL);
1488 			}
1489 			mp = mp1;
1490 			ipha = (ipha_t *)mp->b_rptr;
1491 			icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1492 		}
1493 		icmph->icmph_type = ICMP_TIME_STAMP_REPLY;
1494 		tsp = (uint32_t *)&icmph[1];
1495 		tsp++;		/* Skip past 'originate time' */
1496 		/* Compute # of milliseconds since midnight */
1497 		gethrestime(&now);
1498 		ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
1499 		    NSEC2MSEC(now.tv_nsec);
1500 		*tsp++ = htonl(ts);	/* Lay in 'receive time' */
1501 		*tsp++ = htonl(ts);	/* Lay in 'send time' */
1502 		BUMP_MIB(&ipst->ips_icmp_mib, icmpOutTimestampReps);
1503 		icmp_send_reply_v4(mp, ipha, icmph, ira);
1504 		return (NULL);
1505 
1506 	case ICMP_TIME_STAMP_REPLY:
1507 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimestampReps);
1508 		break;
1509 	case ICMP_INFO_REQUEST:
1510 		/* Per RFC 1122 3.2.2.7, ignore this. */
1511 	case ICMP_INFO_REPLY:
1512 		break;
1513 	case ICMP_ADDRESS_MASK_REQUEST:
1514 		if (ira->ira_flags & IRAF_MULTIBROADCAST) {
1515 			interested =
1516 			    ipst->ips_ip_respond_to_address_mask_broadcast;
1517 		} else {
1518 			interested = B_TRUE;
1519 		}
1520 		if (!interested) {
1521 			/* We never pass these to RAW sockets */
1522 			freemsg(mp);
1523 			return (NULL);
1524 		}
1525 		len_needed = ip_hdr_length + ICMPH_SIZE + IP_ADDR_LEN;
1526 		if (mp->b_wptr - mp->b_rptr < len_needed) {
1527 			ipha = ip_pullup(mp, len_needed, ira);
1528 			if (ipha == NULL) {
1529 				BUMP_MIB(ill->ill_ip_mib,
1530 				    ipIfStatsInTruncatedPkts);
1531 				ip_drop_input("ipIfStatsInTruncatedPkts", mp,
1532 				    ill);
1533 				freemsg(mp);
1534 				return (NULL);
1535 			}
1536 			/* Refresh following the pullup. */
1537 			icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1538 		}
1539 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInAddrMasks);
1540 		/* Check db_ref to make sure we can modify the packet. */
1541 		if (mp->b_datap->db_ref > 1) {
1542 			mblk_t	*mp1;
1543 
1544 			mp1 = copymsg(mp);
1545 			freemsg(mp);
1546 			if (!mp1) {
1547 				BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1548 				return (NULL);
1549 			}
1550 			mp = mp1;
1551 			ipha = (ipha_t *)mp->b_rptr;
1552 			icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1553 		}
1554 		/*
1555 		 * Need the ipif with the mask be the same as the source
1556 		 * address of the mask reply. For unicast we have a specific
1557 		 * ipif. For multicast/broadcast we only handle onlink
1558 		 * senders, and use the source address to pick an ipif.
1559 		 */
1560 		ipif = ipif_lookup_addr(ipha->ipha_dst, ill, zoneid, ipst);
1561 		if (ipif == NULL) {
1562 			/* Broadcast or multicast */
1563 			ipif = ipif_lookup_remote(ill, ipha->ipha_src, zoneid);
1564 			if (ipif == NULL) {
1565 				freemsg(mp);
1566 				return (NULL);
1567 			}
1568 		}
1569 		icmph->icmph_type = ICMP_ADDRESS_MASK_REPLY;
1570 		bcopy(&ipif->ipif_net_mask, &icmph[1], IP_ADDR_LEN);
1571 		ipif_refrele(ipif);
1572 		BUMP_MIB(&ipst->ips_icmp_mib, icmpOutAddrMaskReps);
1573 		icmp_send_reply_v4(mp, ipha, icmph, ira);
1574 		return (NULL);
1575 
1576 	case ICMP_ADDRESS_MASK_REPLY:
1577 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInAddrMaskReps);
1578 		break;
1579 	default:
1580 		interested = B_TRUE;	/* Pass up to transport */
1581 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInUnknowns);
1582 		break;
1583 	}
1584 	/*
1585 	 * See if there is an ICMP client to avoid an extra copymsg/freemsg
1586 	 * if there isn't one.
1587 	 */
1588 	if (ipst->ips_ipcl_proto_fanout_v4[IPPROTO_ICMP].connf_head != NULL) {
1589 		/* If there is an ICMP client and we want one too, copy it. */
1590 
1591 		if (!interested) {
1592 			/* Caller will deliver to RAW sockets */
1593 			return (mp);
1594 		}
1595 		mp_ret = copymsg(mp);
1596 		if (mp_ret == NULL) {
1597 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1598 			ip_drop_input("ipIfStatsInDiscards - copymsg", mp, ill);
1599 		}
1600 	} else if (!interested) {
1601 		/* Neither we nor raw sockets are interested. Drop packet now */
1602 		freemsg(mp);
1603 		return (NULL);
1604 	}
1605 
1606 	/*
1607 	 * ICMP error or redirect packet. Make sure we have enough of
1608 	 * the header and that db_ref == 1 since we might end up modifying
1609 	 * the packet.
1610 	 */
1611 	if (mp->b_cont != NULL) {
1612 		if (ip_pullup(mp, -1, ira) == NULL) {
1613 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1614 			ip_drop_input("ipIfStatsInDiscards - ip_pullup",
1615 			    mp, ill);
1616 			freemsg(mp);
1617 			return (mp_ret);
1618 		}
1619 	}
1620 
1621 	if (mp->b_datap->db_ref > 1) {
1622 		mblk_t	*mp1;
1623 
1624 		mp1 = copymsg(mp);
1625 		if (mp1 == NULL) {
1626 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1627 			ip_drop_input("ipIfStatsInDiscards - copymsg", mp, ill);
1628 			freemsg(mp);
1629 			return (mp_ret);
1630 		}
1631 		freemsg(mp);
1632 		mp = mp1;
1633 	}
1634 
1635 	/*
1636 	 * In case mp has changed, verify the message before any further
1637 	 * processes.
1638 	 */
1639 	ipha = (ipha_t *)mp->b_rptr;
1640 	icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1641 	if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
1642 		freemsg(mp);
1643 		return (mp_ret);
1644 	}
1645 
1646 	switch (icmph->icmph_type) {
1647 	case ICMP_REDIRECT:
1648 		icmp_redirect_v4(mp, ipha, icmph, ira);
1649 		break;
1650 	case ICMP_DEST_UNREACHABLE:
1651 		if (icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED) {
1652 			/* Update DCE and adjust MTU is icmp header if needed */
1653 			icmp_inbound_too_big_v4(icmph, ira);
1654 		}
1655 		/* FALLTHROUGH */
1656 	default:
1657 		icmp_inbound_error_fanout_v4(mp, icmph, ira);
1658 		break;
1659 	}
1660 	return (mp_ret);
1661 }
1662 
1663 /*
1664  * Send an ICMP echo, timestamp or address mask reply.
1665  * The caller has already updated the payload part of the packet.
1666  * We handle the ICMP checksum, IP source address selection and feed
1667  * the packet into ip_output_simple.
1668  */
1669 static void
icmp_send_reply_v4(mblk_t * mp,ipha_t * ipha,icmph_t * icmph,ip_recv_attr_t * ira)1670 icmp_send_reply_v4(mblk_t *mp, ipha_t *ipha, icmph_t *icmph,
1671     ip_recv_attr_t *ira)
1672 {
1673 	uint_t		ip_hdr_length = ira->ira_ip_hdr_length;
1674 	ill_t		*ill = ira->ira_ill;
1675 	ip_stack_t	*ipst = ill->ill_ipst;
1676 	ip_xmit_attr_t	ixas;
1677 
1678 	/* Send out an ICMP packet */
1679 	icmph->icmph_checksum = 0;
1680 	icmph->icmph_checksum = IP_CSUM(mp, ip_hdr_length, 0);
1681 	/* Reset time to live. */
1682 	ipha->ipha_ttl = ipst->ips_ip_def_ttl;
1683 	{
1684 		/* Swap source and destination addresses */
1685 		ipaddr_t tmp;
1686 
1687 		tmp = ipha->ipha_src;
1688 		ipha->ipha_src = ipha->ipha_dst;
1689 		ipha->ipha_dst = tmp;
1690 	}
1691 	ipha->ipha_ident = 0;
1692 	if (!IS_SIMPLE_IPH(ipha))
1693 		icmp_options_update(ipha);
1694 
1695 	bzero(&ixas, sizeof (ixas));
1696 	ixas.ixa_flags = IXAF_BASIC_SIMPLE_V4;
1697 	ixas.ixa_zoneid = ira->ira_zoneid;
1698 	ixas.ixa_cred = kcred;
1699 	ixas.ixa_cpid = NOPID;
1700 	ixas.ixa_tsl = ira->ira_tsl;	/* Behave as a multi-level responder */
1701 	ixas.ixa_ifindex = 0;
1702 	ixas.ixa_ipst = ipst;
1703 	ixas.ixa_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1704 
1705 	if (!(ira->ira_flags & IRAF_IPSEC_SECURE)) {
1706 		/*
1707 		 * This packet should go out the same way as it
1708 		 * came in i.e in clear, independent of the IPsec policy
1709 		 * for transmitting packets.
1710 		 */
1711 		ixas.ixa_flags |= IXAF_NO_IPSEC;
1712 	} else {
1713 		if (!ipsec_in_to_out(ira, &ixas, mp, ipha, NULL)) {
1714 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1715 			/* Note: mp already consumed and ip_drop_packet done */
1716 			return;
1717 		}
1718 	}
1719 	if (ira->ira_flags & IRAF_MULTIBROADCAST) {
1720 		/*
1721 		 * Not one or our addresses (IRE_LOCALs), thus we let
1722 		 * ip_output_simple pick the source.
1723 		 */
1724 		ipha->ipha_src = INADDR_ANY;
1725 		ixas.ixa_flags |= IXAF_SET_SOURCE;
1726 	}
1727 	/* Should we send with DF and use dce_pmtu? */
1728 	if (ipst->ips_ipv4_icmp_return_pmtu) {
1729 		ixas.ixa_flags |= IXAF_PMTU_DISCOVERY;
1730 		ipha->ipha_fragment_offset_and_flags |= IPH_DF_HTONS;
1731 	}
1732 
1733 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutMsgs);
1734 
1735 	(void) ip_output_simple(mp, &ixas);
1736 	ixa_cleanup(&ixas);
1737 }
1738 
1739 /*
1740  * Verify the ICMP messages for either for ICMP error or redirect packet.
1741  * The caller should have fully pulled up the message. If it's a redirect
1742  * packet, only basic checks on IP header will be done; otherwise, verify
1743  * the packet by looking at the included ULP header.
1744  *
1745  * Called before icmp_inbound_error_fanout_v4 is called.
1746  */
1747 static boolean_t
icmp_inbound_verify_v4(mblk_t * mp,icmph_t * icmph,ip_recv_attr_t * ira)1748 icmp_inbound_verify_v4(mblk_t *mp, icmph_t *icmph, ip_recv_attr_t *ira)
1749 {
1750 	ill_t		*ill = ira->ira_ill;
1751 	int		hdr_length;
1752 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
1753 	conn_t		*connp;
1754 	ipha_t		*ipha;	/* Inner IP header */
1755 
1756 	ipha = (ipha_t *)&icmph[1];
1757 	if ((uchar_t *)ipha + IP_SIMPLE_HDR_LENGTH > mp->b_wptr)
1758 		goto truncated;
1759 
1760 	hdr_length = IPH_HDR_LENGTH(ipha);
1761 
1762 	if ((IPH_HDR_VERSION(ipha) != IPV4_VERSION))
1763 		goto discard_pkt;
1764 
1765 	if (hdr_length < sizeof (ipha_t))
1766 		goto truncated;
1767 
1768 	if ((uchar_t *)ipha + hdr_length > mp->b_wptr)
1769 		goto truncated;
1770 
1771 	/*
1772 	 * Stop here for ICMP_REDIRECT.
1773 	 */
1774 	if (icmph->icmph_type == ICMP_REDIRECT)
1775 		return (B_TRUE);
1776 
1777 	/*
1778 	 * ICMP errors only.
1779 	 */
1780 	switch (ipha->ipha_protocol) {
1781 	case IPPROTO_UDP:
1782 		/*
1783 		 * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1784 		 * transport header.
1785 		 */
1786 		if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1787 		    mp->b_wptr)
1788 			goto truncated;
1789 		break;
1790 	case IPPROTO_TCP: {
1791 		tcpha_t		*tcpha;
1792 
1793 		/*
1794 		 * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1795 		 * transport header.
1796 		 */
1797 		if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1798 		    mp->b_wptr)
1799 			goto truncated;
1800 
1801 		tcpha = (tcpha_t *)((uchar_t *)ipha + hdr_length);
1802 		connp = ipcl_tcp_lookup_reversed_ipv4(ipha, tcpha, TCPS_LISTEN,
1803 		    ipst);
1804 		if (connp == NULL)
1805 			goto discard_pkt;
1806 
1807 		if ((connp->conn_verifyicmp != NULL) &&
1808 		    !connp->conn_verifyicmp(connp, tcpha, icmph, NULL, ira)) {
1809 			CONN_DEC_REF(connp);
1810 			goto discard_pkt;
1811 		}
1812 		CONN_DEC_REF(connp);
1813 		break;
1814 	}
1815 	case IPPROTO_SCTP:
1816 		/*
1817 		 * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1818 		 * transport header.
1819 		 */
1820 		if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1821 		    mp->b_wptr)
1822 			goto truncated;
1823 		break;
1824 	case IPPROTO_ESP:
1825 	case IPPROTO_AH:
1826 		break;
1827 	case IPPROTO_ENCAP:
1828 		if ((uchar_t *)ipha + hdr_length + sizeof (ipha_t) >
1829 		    mp->b_wptr)
1830 			goto truncated;
1831 		break;
1832 	default:
1833 		break;
1834 	}
1835 
1836 	return (B_TRUE);
1837 
1838 discard_pkt:
1839 	/* Bogus ICMP error. */
1840 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1841 	return (B_FALSE);
1842 
1843 truncated:
1844 	/* We pulled up everthing already. Must be truncated */
1845 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
1846 	ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
1847 	return (B_FALSE);
1848 }
1849 
1850 /* Table from RFC 1191 */
1851 static int icmp_frag_size_table[] =
1852 { 32000, 17914, 8166, 4352, 2002, 1496, 1006, 508, 296, 68 };
1853 
1854 /*
1855  * Process received ICMP Packet too big.
1856  * Just handles the DCE create/update, including using the above table of
1857  * PMTU guesses. The caller is responsible for validating the packet before
1858  * passing it in and also to fanout the ICMP error to any matching transport
1859  * conns. Assumes the message has been fully pulled up and verified.
1860  *
1861  * Before getting here, the caller has called icmp_inbound_verify_v4()
1862  * that should have verified with ULP to prevent undoing the changes we're
1863  * going to make to DCE. For example, TCP might have verified that the packet
1864  * which generated error is in the send window.
1865  *
1866  * In some cases modified this MTU in the ICMP header packet; the caller
1867  * should pass to the matching ULP after this returns.
1868  */
1869 static void
icmp_inbound_too_big_v4(icmph_t * icmph,ip_recv_attr_t * ira)1870 icmp_inbound_too_big_v4(icmph_t *icmph, ip_recv_attr_t *ira)
1871 {
1872 	dce_t		*dce;
1873 	int		old_mtu;
1874 	int		mtu, orig_mtu;
1875 	ipaddr_t	dst;
1876 	boolean_t	disable_pmtud;
1877 	ill_t		*ill = ira->ira_ill;
1878 	ip_stack_t	*ipst = ill->ill_ipst;
1879 	uint_t		hdr_length;
1880 	ipha_t		*ipha;
1881 
1882 	/* Caller already pulled up everything. */
1883 	ipha = (ipha_t *)&icmph[1];
1884 	ASSERT(icmph->icmph_type == ICMP_DEST_UNREACHABLE &&
1885 	    icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED);
1886 	ASSERT(ill != NULL);
1887 
1888 	hdr_length = IPH_HDR_LENGTH(ipha);
1889 
1890 	/*
1891 	 * We handle path MTU for source routed packets since the DCE
1892 	 * is looked up using the final destination.
1893 	 */
1894 	dst = ip_get_dst(ipha);
1895 
1896 	dce = dce_lookup_and_add_v4(dst, ipst);
1897 	if (dce == NULL) {
1898 		/* Couldn't add a unique one - ENOMEM */
1899 		ip1dbg(("icmp_inbound_too_big_v4: no dce for 0x%x\n",
1900 		    ntohl(dst)));
1901 		return;
1902 	}
1903 
1904 	/* Check for MTU discovery advice as described in RFC 1191 */
1905 	mtu = ntohs(icmph->icmph_du_mtu);
1906 	orig_mtu = mtu;
1907 	disable_pmtud = B_FALSE;
1908 
1909 	mutex_enter(&dce->dce_lock);
1910 	if (dce->dce_flags & DCEF_PMTU)
1911 		old_mtu = dce->dce_pmtu;
1912 	else
1913 		old_mtu = ill->ill_mtu;
1914 
1915 	if (icmph->icmph_du_zero != 0 || mtu < ipst->ips_ip_pmtu_min) {
1916 		uint32_t length;
1917 		int	i;
1918 
1919 		/*
1920 		 * Use the table from RFC 1191 to figure out
1921 		 * the next "plateau" based on the length in
1922 		 * the original IP packet.
1923 		 */
1924 		length = ntohs(ipha->ipha_length);
1925 		DTRACE_PROBE2(ip4__pmtu__guess, dce_t *, dce,
1926 		    uint32_t, length);
1927 		if (old_mtu <= length &&
1928 		    old_mtu >= length - hdr_length) {
1929 			/*
1930 			 * Handle broken BSD 4.2 systems that
1931 			 * return the wrong ipha_length in ICMP
1932 			 * errors.
1933 			 */
1934 			ip1dbg(("Wrong mtu: sent %d, dce %d\n",
1935 			    length, old_mtu));
1936 			length -= hdr_length;
1937 		}
1938 		for (i = 0; i < A_CNT(icmp_frag_size_table); i++) {
1939 			if (length > icmp_frag_size_table[i])
1940 				break;
1941 		}
1942 		if (i == A_CNT(icmp_frag_size_table)) {
1943 			/* Smaller than IP_MIN_MTU! */
1944 			ip1dbg(("Too big for packet size %d\n",
1945 			    length));
1946 			disable_pmtud = B_TRUE;
1947 			mtu = ipst->ips_ip_pmtu_min;
1948 		} else {
1949 			mtu = icmp_frag_size_table[i];
1950 			ip1dbg(("Calculated mtu %d, packet size %d, "
1951 			    "before %d\n", mtu, length, old_mtu));
1952 			if (mtu < ipst->ips_ip_pmtu_min) {
1953 				mtu = ipst->ips_ip_pmtu_min;
1954 				disable_pmtud = B_TRUE;
1955 			}
1956 		}
1957 	}
1958 	if (disable_pmtud)
1959 		dce->dce_flags |= DCEF_TOO_SMALL_PMTU;
1960 	else
1961 		dce->dce_flags &= ~DCEF_TOO_SMALL_PMTU;
1962 
1963 	dce->dce_pmtu = MIN(old_mtu, mtu);
1964 	/* Prepare to send the new max frag size for the ULP. */
1965 	icmph->icmph_du_zero = 0;
1966 	icmph->icmph_du_mtu =  htons((uint16_t)dce->dce_pmtu);
1967 	DTRACE_PROBE4(ip4__pmtu__change, icmph_t *, icmph, dce_t *,
1968 	    dce, int, orig_mtu, int, mtu);
1969 
1970 	/* We now have a PMTU for sure */
1971 	dce->dce_flags |= DCEF_PMTU;
1972 	dce->dce_last_change_time = TICK_TO_SEC(ddi_get_lbolt64());
1973 	mutex_exit(&dce->dce_lock);
1974 	/*
1975 	 * After dropping the lock the new value is visible to everyone.
1976 	 * Then we bump the generation number so any cached values reinspect
1977 	 * the dce_t.
1978 	 */
1979 	dce_increment_generation(dce);
1980 	dce_refrele(dce);
1981 }
1982 
1983 /*
1984  * If the packet in error is Self-Encapsulated, icmp_inbound_error_fanout_v4
1985  * calls this function.
1986  */
1987 static mblk_t *
icmp_inbound_self_encap_error_v4(mblk_t * mp,ipha_t * ipha,ipha_t * in_ipha)1988 icmp_inbound_self_encap_error_v4(mblk_t *mp, ipha_t *ipha, ipha_t *in_ipha)
1989 {
1990 	int length;
1991 
1992 	ASSERT(mp->b_datap->db_type == M_DATA);
1993 
1994 	/* icmp_inbound_v4 has already pulled up the whole error packet */
1995 	ASSERT(mp->b_cont == NULL);
1996 
1997 	/*
1998 	 * The length that we want to overlay is the inner header
1999 	 * and what follows it.
2000 	 */
2001 	length = msgdsize(mp) - ((uchar_t *)in_ipha - mp->b_rptr);
2002 
2003 	/*
2004 	 * Overlay the inner header and whatever follows it over the
2005 	 * outer header.
2006 	 */
2007 	bcopy((uchar_t *)in_ipha, (uchar_t *)ipha, length);
2008 
2009 	/* Adjust for what we removed */
2010 	mp->b_wptr -= (uchar_t *)in_ipha - (uchar_t *)ipha;
2011 	return (mp);
2012 }
2013 
2014 /*
2015  * Try to pass the ICMP message upstream in case the ULP cares.
2016  *
2017  * If the packet that caused the ICMP error is secure, we send
2018  * it to AH/ESP to make sure that the attached packet has a
2019  * valid association. ipha in the code below points to the
2020  * IP header of the packet that caused the error.
2021  *
2022  * For IPsec cases, we let the next-layer-up (which has access to
2023  * cached policy on the conn_t, or can query the SPD directly)
2024  * subtract out any IPsec overhead if they must.  We therefore make no
2025  * adjustments here for IPsec overhead.
2026  *
2027  * IFN could have been generated locally or by some router.
2028  *
2029  * LOCAL : ire_send_wire (before calling ipsec_out_process) can call
2030  * icmp_frag_needed/icmp_pkt2big_v6 to generated a local IFN.
2031  *	    This happens because IP adjusted its value of MTU on an
2032  *	    earlier IFN message and could not tell the upper layer,
2033  *	    the new adjusted value of MTU e.g. Packet was encrypted
2034  *	    or there was not enough information to fanout to upper
2035  *	    layers. Thus on the next outbound datagram, ire_send_wire
2036  *	    generates the IFN, where IPsec processing has *not* been
2037  *	    done.
2038  *
2039  *	    Note that we retain ixa_fragsize across IPsec thus once
2040  *	    we have picking ixa_fragsize and entered ipsec_out_process we do
2041  *	    no change the fragsize even if the path MTU changes before
2042  *	    we reach ip_output_post_ipsec.
2043  *
2044  *	    In the local case, IRAF_LOOPBACK will be set indicating
2045  *	    that IFN was generated locally.
2046  *
2047  * ROUTER : IFN could be secure or non-secure.
2048  *
2049  *	    * SECURE : We use the IPSEC_IN to fanout to AH/ESP if the
2050  *	      packet in error has AH/ESP headers to validate the AH/ESP
2051  *	      headers. AH/ESP will verify whether there is a valid SA or
2052  *	      not and send it back. We will fanout again if we have more
2053  *	      data in the packet.
2054  *
2055  *	      If the packet in error does not have AH/ESP, we handle it
2056  *	      like any other case.
2057  *
2058  *	    * NON_SECURE : If the packet in error has AH/ESP headers, we send it
2059  *	      up to AH/ESP for validation. AH/ESP will verify whether there is a
2060  *	      valid SA or not and send it back. We will fanout again if
2061  *	      we have more data in the packet.
2062  *
2063  *	      If the packet in error does not have AH/ESP, we handle it
2064  *	      like any other case.
2065  *
2066  * The caller must have called icmp_inbound_verify_v4.
2067  */
2068 static void
icmp_inbound_error_fanout_v4(mblk_t * mp,icmph_t * icmph,ip_recv_attr_t * ira)2069 icmp_inbound_error_fanout_v4(mblk_t *mp, icmph_t *icmph, ip_recv_attr_t *ira)
2070 {
2071 	uint16_t	*up;	/* Pointer to ports in ULP header */
2072 	uint32_t	ports;	/* reversed ports for fanout */
2073 	ipha_t		ripha;	/* With reversed addresses */
2074 	ipha_t		*ipha;  /* Inner IP header */
2075 	uint_t		hdr_length; /* Inner IP header length */
2076 	tcpha_t		*tcpha;
2077 	conn_t		*connp;
2078 	ill_t		*ill = ira->ira_ill;
2079 	ip_stack_t	*ipst = ill->ill_ipst;
2080 	ipsec_stack_t	*ipss = ipst->ips_netstack->netstack_ipsec;
2081 	ill_t		*rill = ira->ira_rill;
2082 
2083 	/* Caller already pulled up everything. */
2084 	ipha = (ipha_t *)&icmph[1];
2085 	ASSERT((uchar_t *)&ipha[1] <= mp->b_wptr);
2086 	ASSERT(mp->b_cont == NULL);
2087 
2088 	hdr_length = IPH_HDR_LENGTH(ipha);
2089 	ira->ira_protocol = ipha->ipha_protocol;
2090 
2091 	/*
2092 	 * We need a separate IP header with the source and destination
2093 	 * addresses reversed to do fanout/classification because the ipha in
2094 	 * the ICMP error is in the form we sent it out.
2095 	 */
2096 	ripha.ipha_src = ipha->ipha_dst;
2097 	ripha.ipha_dst = ipha->ipha_src;
2098 	ripha.ipha_protocol = ipha->ipha_protocol;
2099 	ripha.ipha_version_and_hdr_length = ipha->ipha_version_and_hdr_length;
2100 
2101 	ip2dbg(("icmp_inbound_error_v4: proto %d %x to %x: %d/%d\n",
2102 	    ripha.ipha_protocol, ntohl(ipha->ipha_src),
2103 	    ntohl(ipha->ipha_dst),
2104 	    icmph->icmph_type, icmph->icmph_code));
2105 
2106 	switch (ipha->ipha_protocol) {
2107 	case IPPROTO_UDP:
2108 		up = (uint16_t *)((uchar_t *)ipha + hdr_length);
2109 
2110 		/* Attempt to find a client stream based on port. */
2111 		ip2dbg(("icmp_inbound_error_v4: UDP ports %d to %d\n",
2112 		    ntohs(up[0]), ntohs(up[1])));
2113 
2114 		/* Note that we send error to all matches. */
2115 		ira->ira_flags |= IRAF_ICMP_ERROR;
2116 		ip_fanout_udp_multi_v4(mp, &ripha, up[0], up[1], ira);
2117 		ira->ira_flags &= ~IRAF_ICMP_ERROR;
2118 		return;
2119 
2120 	case IPPROTO_TCP:
2121 		/*
2122 		 * Find a TCP client stream for this packet.
2123 		 * Note that we do a reverse lookup since the header is
2124 		 * in the form we sent it out.
2125 		 */
2126 		tcpha = (tcpha_t *)((uchar_t *)ipha + hdr_length);
2127 		connp = ipcl_tcp_lookup_reversed_ipv4(ipha, tcpha, TCPS_LISTEN,
2128 		    ipst);
2129 		if (connp == NULL)
2130 			goto discard_pkt;
2131 
2132 		if (CONN_INBOUND_POLICY_PRESENT(connp, ipss) ||
2133 		    (ira->ira_flags & IRAF_IPSEC_SECURE)) {
2134 			mp = ipsec_check_inbound_policy(mp, connp,
2135 			    ipha, NULL, ira);
2136 			if (mp == NULL) {
2137 				BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
2138 				/* Note that mp is NULL */
2139 				ip_drop_input("ipIfStatsInDiscards", mp, ill);
2140 				CONN_DEC_REF(connp);
2141 				return;
2142 			}
2143 		}
2144 
2145 		ira->ira_flags |= IRAF_ICMP_ERROR;
2146 		ira->ira_ill = ira->ira_rill = NULL;
2147 		if (IPCL_IS_TCP(connp)) {
2148 			SQUEUE_ENTER_ONE(connp->conn_sqp, mp,
2149 			    connp->conn_recvicmp, connp, ira, SQ_FILL,
2150 			    SQTAG_TCP_INPUT_ICMP_ERR);
2151 		} else {
2152 			/* Not TCP; must be SOCK_RAW, IPPROTO_TCP */
2153 			(connp->conn_recv)(connp, mp, NULL, ira);
2154 			CONN_DEC_REF(connp);
2155 		}
2156 		ira->ira_ill = ill;
2157 		ira->ira_rill = rill;
2158 		ira->ira_flags &= ~IRAF_ICMP_ERROR;
2159 		return;
2160 
2161 	case IPPROTO_SCTP:
2162 		up = (uint16_t *)((uchar_t *)ipha + hdr_length);
2163 		/* Find a SCTP client stream for this packet. */
2164 		((uint16_t *)&ports)[0] = up[1];
2165 		((uint16_t *)&ports)[1] = up[0];
2166 
2167 		ira->ira_flags |= IRAF_ICMP_ERROR;
2168 		ip_fanout_sctp(mp, &ripha, NULL, ports, ira);
2169 		ira->ira_flags &= ~IRAF_ICMP_ERROR;
2170 		return;
2171 
2172 	case IPPROTO_ESP:
2173 	case IPPROTO_AH:
2174 		if (!ipsec_loaded(ipss)) {
2175 			ip_proto_not_sup(mp, ira);
2176 			return;
2177 		}
2178 
2179 		if (ipha->ipha_protocol == IPPROTO_ESP)
2180 			mp = ipsecesp_icmp_error(mp, ira);
2181 		else
2182 			mp = ipsecah_icmp_error(mp, ira);
2183 		if (mp == NULL)
2184 			return;
2185 
2186 		/* Just in case ipsec didn't preserve the NULL b_cont */
2187 		if (mp->b_cont != NULL) {
2188 			if (!pullupmsg(mp, -1))
2189 				goto discard_pkt;
2190 		}
2191 
2192 		/*
2193 		 * Note that ira_pktlen and ira_ip_hdr_length are no longer
2194 		 * correct, but we don't use them any more here.
2195 		 *
2196 		 * If succesful, the mp has been modified to not include
2197 		 * the ESP/AH header so we can fanout to the ULP's icmp
2198 		 * error handler.
2199 		 */
2200 		if (mp->b_wptr - mp->b_rptr < IP_SIMPLE_HDR_LENGTH)
2201 			goto truncated;
2202 
2203 		/* Verify the modified message before any further processes. */
2204 		ipha = (ipha_t *)mp->b_rptr;
2205 		hdr_length = IPH_HDR_LENGTH(ipha);
2206 		icmph = (icmph_t *)&mp->b_rptr[hdr_length];
2207 		if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
2208 			freemsg(mp);
2209 			return;
2210 		}
2211 
2212 		icmp_inbound_error_fanout_v4(mp, icmph, ira);
2213 		return;
2214 
2215 	case IPPROTO_ENCAP: {
2216 		/* Look for self-encapsulated packets that caused an error */
2217 		ipha_t *in_ipha;
2218 
2219 		/*
2220 		 * Caller has verified that length has to be
2221 		 * at least the size of IP header.
2222 		 */
2223 		ASSERT(hdr_length >= sizeof (ipha_t));
2224 		/*
2225 		 * Check the sanity of the inner IP header like
2226 		 * we did for the outer header.
2227 		 */
2228 		in_ipha = (ipha_t *)((uchar_t *)ipha + hdr_length);
2229 		if ((IPH_HDR_VERSION(in_ipha) != IPV4_VERSION)) {
2230 			goto discard_pkt;
2231 		}
2232 		if (IPH_HDR_LENGTH(in_ipha) < sizeof (ipha_t)) {
2233 			goto discard_pkt;
2234 		}
2235 		/* Check for Self-encapsulated tunnels */
2236 		if (in_ipha->ipha_src == ipha->ipha_src &&
2237 		    in_ipha->ipha_dst == ipha->ipha_dst) {
2238 
2239 			mp = icmp_inbound_self_encap_error_v4(mp, ipha,
2240 			    in_ipha);
2241 			if (mp == NULL)
2242 				goto discard_pkt;
2243 
2244 			/*
2245 			 * Just in case self_encap didn't preserve the NULL
2246 			 * b_cont
2247 			 */
2248 			if (mp->b_cont != NULL) {
2249 				if (!pullupmsg(mp, -1))
2250 					goto discard_pkt;
2251 			}
2252 			/*
2253 			 * Note that ira_pktlen and ira_ip_hdr_length are no
2254 			 * longer correct, but we don't use them any more here.
2255 			 */
2256 			if (mp->b_wptr - mp->b_rptr < IP_SIMPLE_HDR_LENGTH)
2257 				goto truncated;
2258 
2259 			/*
2260 			 * Verify the modified message before any further
2261 			 * processes.
2262 			 */
2263 			ipha = (ipha_t *)mp->b_rptr;
2264 			hdr_length = IPH_HDR_LENGTH(ipha);
2265 			icmph = (icmph_t *)&mp->b_rptr[hdr_length];
2266 			if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
2267 				freemsg(mp);
2268 				return;
2269 			}
2270 
2271 			/*
2272 			 * The packet in error is self-encapsualted.
2273 			 * And we are finding it further encapsulated
2274 			 * which we could not have possibly generated.
2275 			 */
2276 			if (ipha->ipha_protocol == IPPROTO_ENCAP) {
2277 				goto discard_pkt;
2278 			}
2279 			icmp_inbound_error_fanout_v4(mp, icmph, ira);
2280 			return;
2281 		}
2282 		/* No self-encapsulated */
2283 	}
2284 	/* FALLTHROUGH */
2285 	case IPPROTO_IPV6:
2286 		if ((connp = ipcl_iptun_classify_v4(&ripha.ipha_src,
2287 		    &ripha.ipha_dst, ipst)) != NULL) {
2288 			ira->ira_flags |= IRAF_ICMP_ERROR;
2289 			connp->conn_recvicmp(connp, mp, NULL, ira);
2290 			CONN_DEC_REF(connp);
2291 			ira->ira_flags &= ~IRAF_ICMP_ERROR;
2292 			return;
2293 		}
2294 		/*
2295 		 * No IP tunnel is interested, fallthrough and see
2296 		 * if a raw socket will want it.
2297 		 */
2298 		/* FALLTHROUGH */
2299 	default:
2300 		ira->ira_flags |= IRAF_ICMP_ERROR;
2301 		ip_fanout_proto_v4(mp, &ripha, ira);
2302 		ira->ira_flags &= ~IRAF_ICMP_ERROR;
2303 		return;
2304 	}
2305 	/* NOTREACHED */
2306 discard_pkt:
2307 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
2308 	ip1dbg(("icmp_inbound_error_fanout_v4: drop pkt\n"));
2309 	ip_drop_input("ipIfStatsInDiscards", mp, ill);
2310 	freemsg(mp);
2311 	return;
2312 
2313 truncated:
2314 	/* We pulled up everthing already. Must be truncated */
2315 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
2316 	ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
2317 	freemsg(mp);
2318 }
2319 
2320 /*
2321  * Common IP options parser.
2322  *
2323  * Setup routine: fill in *optp with options-parsing state, then
2324  * tail-call ipoptp_next to return the first option.
2325  */
2326 uint8_t
ipoptp_first(ipoptp_t * optp,ipha_t * ipha)2327 ipoptp_first(ipoptp_t *optp, ipha_t *ipha)
2328 {
2329 	uint32_t totallen; /* total length of all options */
2330 
2331 	totallen = ipha->ipha_version_and_hdr_length -
2332 	    (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
2333 	totallen <<= 2;
2334 	optp->ipoptp_next = (uint8_t *)(&ipha[1]);
2335 	optp->ipoptp_end = optp->ipoptp_next + totallen;
2336 	optp->ipoptp_flags = 0;
2337 	return (ipoptp_next(optp));
2338 }
2339 
2340 /* Like above but without an ipha_t */
2341 uint8_t
ipoptp_first2(ipoptp_t * optp,uint32_t totallen,uint8_t * opt)2342 ipoptp_first2(ipoptp_t *optp, uint32_t totallen, uint8_t *opt)
2343 {
2344 	optp->ipoptp_next = opt;
2345 	optp->ipoptp_end = optp->ipoptp_next + totallen;
2346 	optp->ipoptp_flags = 0;
2347 	return (ipoptp_next(optp));
2348 }
2349 
2350 /*
2351  * Common IP options parser: extract next option.
2352  */
2353 uint8_t
ipoptp_next(ipoptp_t * optp)2354 ipoptp_next(ipoptp_t *optp)
2355 {
2356 	uint8_t *end = optp->ipoptp_end;
2357 	uint8_t *cur = optp->ipoptp_next;
2358 	uint8_t opt, len, pointer;
2359 
2360 	/*
2361 	 * If cur > end already, then the ipoptp_end or ipoptp_next pointer
2362 	 * has been corrupted.
2363 	 */
2364 	ASSERT(cur <= end);
2365 
2366 	if (cur == end)
2367 		return (IPOPT_EOL);
2368 
2369 	opt = cur[IPOPT_OPTVAL];
2370 
2371 	/*
2372 	 * Skip any NOP options.
2373 	 */
2374 	while (opt == IPOPT_NOP) {
2375 		cur++;
2376 		if (cur == end)
2377 			return (IPOPT_EOL);
2378 		opt = cur[IPOPT_OPTVAL];
2379 	}
2380 
2381 	if (opt == IPOPT_EOL)
2382 		return (IPOPT_EOL);
2383 
2384 	/*
2385 	 * Option requiring a length.
2386 	 */
2387 	if ((cur + 1) >= end) {
2388 		optp->ipoptp_flags |= IPOPTP_ERROR;
2389 		return (IPOPT_EOL);
2390 	}
2391 	len = cur[IPOPT_OLEN];
2392 	if (len < 2) {
2393 		optp->ipoptp_flags |= IPOPTP_ERROR;
2394 		return (IPOPT_EOL);
2395 	}
2396 	optp->ipoptp_cur = cur;
2397 	optp->ipoptp_len = len;
2398 	optp->ipoptp_next = cur + len;
2399 	if (cur + len > end) {
2400 		optp->ipoptp_flags |= IPOPTP_ERROR;
2401 		return (IPOPT_EOL);
2402 	}
2403 
2404 	/*
2405 	 * For the options which require a pointer field, make sure
2406 	 * its there, and make sure it points to either something
2407 	 * inside this option, or the end of the option.
2408 	 */
2409 	pointer = IPOPT_EOL;
2410 	switch (opt) {
2411 	case IPOPT_RR:
2412 	case IPOPT_TS:
2413 	case IPOPT_LSRR:
2414 	case IPOPT_SSRR:
2415 		if (len <= IPOPT_OFFSET) {
2416 			optp->ipoptp_flags |= IPOPTP_ERROR;
2417 			return (opt);
2418 		}
2419 		pointer = cur[IPOPT_OFFSET];
2420 		if (pointer - 1 > len) {
2421 			optp->ipoptp_flags |= IPOPTP_ERROR;
2422 			return (opt);
2423 		}
2424 		break;
2425 	}
2426 
2427 	/*
2428 	 * Sanity check the pointer field based on the type of the
2429 	 * option.
2430 	 */
2431 	switch (opt) {
2432 	case IPOPT_RR:
2433 	case IPOPT_SSRR:
2434 	case IPOPT_LSRR:
2435 		if (pointer < IPOPT_MINOFF_SR)
2436 			optp->ipoptp_flags |= IPOPTP_ERROR;
2437 		break;
2438 	case IPOPT_TS:
2439 		if (pointer < IPOPT_MINOFF_IT)
2440 			optp->ipoptp_flags |= IPOPTP_ERROR;
2441 		/*
2442 		 * Note that the Internet Timestamp option also
2443 		 * contains two four bit fields (the Overflow field,
2444 		 * and the Flag field), which follow the pointer
2445 		 * field.  We don't need to check that these fields
2446 		 * fall within the length of the option because this
2447 		 * was implicitely done above.  We've checked that the
2448 		 * pointer value is at least IPOPT_MINOFF_IT, and that
2449 		 * it falls within the option.  Since IPOPT_MINOFF_IT >
2450 		 * IPOPT_POS_OV_FLG, we don't need the explicit check.
2451 		 */
2452 		ASSERT(len > IPOPT_POS_OV_FLG);
2453 		break;
2454 	}
2455 
2456 	return (opt);
2457 }
2458 
2459 /*
2460  * Use the outgoing IP header to create an IP_OPTIONS option the way
2461  * it was passed down from the application.
2462  *
2463  * This is compatible with BSD in that it returns
2464  * the reverse source route with the final destination
2465  * as the last entry. The first 4 bytes of the option
2466  * will contain the final destination.
2467  */
2468 int
ip_opt_get_user(conn_t * connp,uchar_t * buf)2469 ip_opt_get_user(conn_t *connp, uchar_t *buf)
2470 {
2471 	ipoptp_t	opts;
2472 	uchar_t		*opt;
2473 	uint8_t		optval;
2474 	uint8_t		optlen;
2475 	uint32_t	len = 0;
2476 	uchar_t		*buf1 = buf;
2477 	uint32_t	totallen;
2478 	ipaddr_t	dst;
2479 	ip_pkt_t	*ipp = &connp->conn_xmit_ipp;
2480 
2481 	if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
2482 		return (0);
2483 
2484 	totallen = ipp->ipp_ipv4_options_len;
2485 	if (totallen & 0x3)
2486 		return (0);
2487 
2488 	buf += IP_ADDR_LEN;	/* Leave room for final destination */
2489 	len += IP_ADDR_LEN;
2490 	bzero(buf1, IP_ADDR_LEN);
2491 
2492 	dst = connp->conn_faddr_v4;
2493 
2494 	for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
2495 	    optval != IPOPT_EOL;
2496 	    optval = ipoptp_next(&opts)) {
2497 		int	off;
2498 
2499 		opt = opts.ipoptp_cur;
2500 		if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
2501 			break;
2502 		}
2503 		optlen = opts.ipoptp_len;
2504 
2505 		switch (optval) {
2506 		case IPOPT_SSRR:
2507 		case IPOPT_LSRR:
2508 
2509 			/*
2510 			 * Insert destination as the first entry in the source
2511 			 * route and move down the entries on step.
2512 			 * The last entry gets placed at buf1.
2513 			 */
2514 			buf[IPOPT_OPTVAL] = optval;
2515 			buf[IPOPT_OLEN] = optlen;
2516 			buf[IPOPT_OFFSET] = optlen;
2517 
2518 			off = optlen - IP_ADDR_LEN;
2519 			if (off < 0) {
2520 				/* No entries in source route */
2521 				break;
2522 			}
2523 			/* Last entry in source route if not already set */
2524 			if (dst == INADDR_ANY)
2525 				bcopy(opt + off, buf1, IP_ADDR_LEN);
2526 			off -= IP_ADDR_LEN;
2527 
2528 			while (off > 0) {
2529 				bcopy(opt + off,
2530 				    buf + off + IP_ADDR_LEN,
2531 				    IP_ADDR_LEN);
2532 				off -= IP_ADDR_LEN;
2533 			}
2534 			/* ipha_dst into first slot */
2535 			bcopy(&dst, buf + off + IP_ADDR_LEN,
2536 			    IP_ADDR_LEN);
2537 			buf += optlen;
2538 			len += optlen;
2539 			break;
2540 
2541 		default:
2542 			bcopy(opt, buf, optlen);
2543 			buf += optlen;
2544 			len += optlen;
2545 			break;
2546 		}
2547 	}
2548 done:
2549 	/* Pad the resulting options */
2550 	while (len & 0x3) {
2551 		*buf++ = IPOPT_EOL;
2552 		len++;
2553 	}
2554 	return (len);
2555 }
2556 
2557 /*
2558  * Update any record route or timestamp options to include this host.
2559  * Reverse any source route option.
2560  * This routine assumes that the options are well formed i.e. that they
2561  * have already been checked.
2562  */
2563 static void
icmp_options_update(ipha_t * ipha)2564 icmp_options_update(ipha_t *ipha)
2565 {
2566 	ipoptp_t	opts;
2567 	uchar_t		*opt;
2568 	uint8_t		optval;
2569 	ipaddr_t	src;		/* Our local address */
2570 	ipaddr_t	dst;
2571 
2572 	ip2dbg(("icmp_options_update\n"));
2573 	src = ipha->ipha_src;
2574 	dst = ipha->ipha_dst;
2575 
2576 	for (optval = ipoptp_first(&opts, ipha);
2577 	    optval != IPOPT_EOL;
2578 	    optval = ipoptp_next(&opts)) {
2579 		ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
2580 		opt = opts.ipoptp_cur;
2581 		ip2dbg(("icmp_options_update: opt %d, len %d\n",
2582 		    optval, opts.ipoptp_len));
2583 		switch (optval) {
2584 			int off1, off2;
2585 		case IPOPT_SSRR:
2586 		case IPOPT_LSRR:
2587 			/*
2588 			 * Reverse the source route.  The first entry
2589 			 * should be the next to last one in the current
2590 			 * source route (the last entry is our address).
2591 			 * The last entry should be the final destination.
2592 			 */
2593 			off1 = IPOPT_MINOFF_SR - 1;
2594 			off2 = opt[IPOPT_OFFSET] - IP_ADDR_LEN - 1;
2595 			if (off2 < 0) {
2596 				/* No entries in source route */
2597 				ip1dbg((
2598 				    "icmp_options_update: bad src route\n"));
2599 				break;
2600 			}
2601 			bcopy((char *)opt + off2, &dst, IP_ADDR_LEN);
2602 			bcopy(&ipha->ipha_dst, (char *)opt + off2, IP_ADDR_LEN);
2603 			bcopy(&dst, &ipha->ipha_dst, IP_ADDR_LEN);
2604 			off2 -= IP_ADDR_LEN;
2605 
2606 			while (off1 < off2) {
2607 				bcopy((char *)opt + off1, &src, IP_ADDR_LEN);
2608 				bcopy((char *)opt + off2, (char *)opt + off1,
2609 				    IP_ADDR_LEN);
2610 				bcopy(&src, (char *)opt + off2, IP_ADDR_LEN);
2611 				off1 += IP_ADDR_LEN;
2612 				off2 -= IP_ADDR_LEN;
2613 			}
2614 			opt[IPOPT_OFFSET] = IPOPT_MINOFF_SR;
2615 			break;
2616 		}
2617 	}
2618 }
2619 
2620 /*
2621  * Process received ICMP Redirect messages.
2622  * Assumes the caller has verified that the headers are in the pulled up mblk.
2623  * Consumes mp.
2624  */
2625 static void
icmp_redirect_v4(mblk_t * mp,ipha_t * ipha,icmph_t * icmph,ip_recv_attr_t * ira)2626 icmp_redirect_v4(mblk_t *mp, ipha_t *ipha, icmph_t *icmph, ip_recv_attr_t *ira)
2627 {
2628 	ire_t		*ire, *nire;
2629 	ire_t		*prev_ire;
2630 	ipaddr_t	src, dst, gateway;
2631 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
2632 	ipha_t		*inner_ipha;	/* Inner IP header */
2633 
2634 	/* Caller already pulled up everything. */
2635 	inner_ipha = (ipha_t *)&icmph[1];
2636 	src = ipha->ipha_src;
2637 	dst = inner_ipha->ipha_dst;
2638 	gateway = icmph->icmph_rd_gateway;
2639 	/* Make sure the new gateway is reachable somehow. */
2640 	ire = ire_ftable_lookup_v4(gateway, 0, 0, IRE_ONLINK, NULL,
2641 	    ALL_ZONES, NULL, MATCH_IRE_TYPE, 0, ipst, NULL);
2642 	/*
2643 	 * Make sure we had a route for the dest in question and that
2644 	 * that route was pointing to the old gateway (the source of the
2645 	 * redirect packet.)
2646 	 * We do longest match and then compare ire_gateway_addr below.
2647 	 */
2648 	prev_ire = ire_ftable_lookup_v4(dst, 0, 0, 0, NULL, ALL_ZONES,
2649 	    NULL, MATCH_IRE_DSTONLY, 0, ipst, NULL);
2650 	/*
2651 	 * Check that
2652 	 *	the redirect was not from ourselves
2653 	 *	the new gateway and the old gateway are directly reachable
2654 	 */
2655 	if (prev_ire == NULL || ire == NULL ||
2656 	    (prev_ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK)) ||
2657 	    (prev_ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) ||
2658 	    !(ire->ire_type & IRE_IF_ALL) ||
2659 	    prev_ire->ire_gateway_addr != src) {
2660 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInBadRedirects);
2661 		ip_drop_input("icmpInBadRedirects - ire", mp, ira->ira_ill);
2662 		freemsg(mp);
2663 		if (ire != NULL)
2664 			ire_refrele(ire);
2665 		if (prev_ire != NULL)
2666 			ire_refrele(prev_ire);
2667 		return;
2668 	}
2669 
2670 	ire_refrele(prev_ire);
2671 	ire_refrele(ire);
2672 
2673 	/*
2674 	 * TODO: more precise handling for cases 0, 2, 3, the latter two
2675 	 * require TOS routing
2676 	 */
2677 	switch (icmph->icmph_code) {
2678 	case 0:
2679 	case 1:
2680 		/* TODO: TOS specificity for cases 2 and 3 */
2681 	case 2:
2682 	case 3:
2683 		break;
2684 	default:
2685 		BUMP_MIB(&ipst->ips_icmp_mib, icmpInBadRedirects);
2686 		ip_drop_input("icmpInBadRedirects - code", mp, ira->ira_ill);
2687 		freemsg(mp);
2688 		return;
2689 	}
2690 	/*
2691 	 * Create a Route Association.  This will allow us to remember that
2692 	 * someone we believe told us to use the particular gateway.
2693 	 */
2694 	ire = ire_create(
2695 	    (uchar_t *)&dst,			/* dest addr */
2696 	    (uchar_t *)&ip_g_all_ones,		/* mask */
2697 	    (uchar_t *)&gateway,		/* gateway addr */
2698 	    IRE_HOST,
2699 	    NULL,				/* ill */
2700 	    ALL_ZONES,
2701 	    (RTF_DYNAMIC | RTF_GATEWAY | RTF_HOST),
2702 	    NULL,				/* tsol_gc_t */
2703 	    ipst);
2704 
2705 	if (ire == NULL) {
2706 		freemsg(mp);
2707 		return;
2708 	}
2709 	nire = ire_add(ire);
2710 	/* Check if it was a duplicate entry */
2711 	if (nire != NULL && nire != ire) {
2712 		ASSERT(nire->ire_identical_ref > 1);
2713 		ire_delete(nire);
2714 		ire_refrele(nire);
2715 		nire = NULL;
2716 	}
2717 	ire = nire;
2718 	if (ire != NULL) {
2719 		ire_refrele(ire);		/* Held in ire_add */
2720 
2721 		/* tell routing sockets that we received a redirect */
2722 		ip_rts_change(RTM_REDIRECT, dst, gateway, IP_HOST_MASK, 0, src,
2723 		    (RTF_DYNAMIC | RTF_GATEWAY | RTF_HOST), 0,
2724 		    (RTA_DST | RTA_GATEWAY | RTA_NETMASK | RTA_AUTHOR), ipst);
2725 	}
2726 
2727 	/*
2728 	 * Delete any existing IRE_HOST type redirect ires for this destination.
2729 	 * This together with the added IRE has the effect of
2730 	 * modifying an existing redirect.
2731 	 */
2732 	prev_ire = ire_ftable_lookup_v4(dst, 0, src, IRE_HOST, NULL,
2733 	    ALL_ZONES, NULL, (MATCH_IRE_GW | MATCH_IRE_TYPE), 0, ipst, NULL);
2734 	if (prev_ire != NULL) {
2735 		if (prev_ire ->ire_flags & RTF_DYNAMIC)
2736 			ire_delete(prev_ire);
2737 		ire_refrele(prev_ire);
2738 	}
2739 
2740 	freemsg(mp);
2741 }
2742 
2743 /*
2744  * Generate an ICMP parameter problem message.
2745  * When called from ip_output side a minimal ip_recv_attr_t needs to be
2746  * constructed by the caller.
2747  */
2748 static void
icmp_param_problem(mblk_t * mp,uint8_t ptr,ip_recv_attr_t * ira)2749 icmp_param_problem(mblk_t *mp, uint8_t ptr, ip_recv_attr_t *ira)
2750 {
2751 	icmph_t	icmph;
2752 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
2753 
2754 	mp = icmp_pkt_err_ok(mp, ira);
2755 	if (mp == NULL)
2756 		return;
2757 
2758 	bzero(&icmph, sizeof (icmph_t));
2759 	icmph.icmph_type = ICMP_PARAM_PROBLEM;
2760 	icmph.icmph_pp_ptr = ptr;
2761 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutParmProbs);
2762 	icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
2763 }
2764 
2765 /*
2766  * Build and ship an IPv4 ICMP message using the packet data in mp, and
2767  * the ICMP header pointed to by "stuff".  (May be called as writer.)
2768  * Note: assumes that icmp_pkt_err_ok has been called to verify that
2769  * an icmp error packet can be sent.
2770  * Assigns an appropriate source address to the packet. If ipha_dst is
2771  * one of our addresses use it for source. Otherwise let ip_output_simple
2772  * pick the source address.
2773  */
2774 static void
icmp_pkt(mblk_t * mp,void * stuff,size_t len,ip_recv_attr_t * ira)2775 icmp_pkt(mblk_t *mp, void *stuff, size_t len, ip_recv_attr_t *ira)
2776 {
2777 	ipaddr_t dst;
2778 	icmph_t	*icmph;
2779 	ipha_t	*ipha;
2780 	uint_t	len_needed;
2781 	size_t	msg_len;
2782 	mblk_t	*mp1;
2783 	ipaddr_t src;
2784 	ire_t	*ire;
2785 	ip_xmit_attr_t ixas;
2786 	ip_stack_t *ipst = ira->ira_ill->ill_ipst;
2787 
2788 	ipha = (ipha_t *)mp->b_rptr;
2789 
2790 	bzero(&ixas, sizeof (ixas));
2791 	ixas.ixa_flags = IXAF_BASIC_SIMPLE_V4;
2792 	ixas.ixa_zoneid = ira->ira_zoneid;
2793 	ixas.ixa_ifindex = 0;
2794 	ixas.ixa_ipst = ipst;
2795 	ixas.ixa_cred = kcred;
2796 	ixas.ixa_cpid = NOPID;
2797 	ixas.ixa_tsl = ira->ira_tsl;	/* Behave as a multi-level responder */
2798 	ixas.ixa_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
2799 
2800 	if (ira->ira_flags & IRAF_IPSEC_SECURE) {
2801 		/*
2802 		 * Apply IPsec based on how IPsec was applied to
2803 		 * the packet that had the error.
2804 		 *
2805 		 * If it was an outbound packet that caused the ICMP
2806 		 * error, then the caller will have setup the IRA
2807 		 * appropriately.
2808 		 */
2809 		if (!ipsec_in_to_out(ira, &ixas, mp, ipha, NULL)) {
2810 			BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
2811 			/* Note: mp already consumed and ip_drop_packet done */
2812 			return;
2813 		}
2814 	} else {
2815 		/*
2816 		 * This is in clear. The icmp message we are building
2817 		 * here should go out in clear, independent of our policy.
2818 		 */
2819 		ixas.ixa_flags |= IXAF_NO_IPSEC;
2820 	}
2821 
2822 	/* Remember our eventual destination */
2823 	dst = ipha->ipha_src;
2824 
2825 	/*
2826 	 * If the packet was for one of our unicast addresses, make
2827 	 * sure we respond with that as the source. Otherwise
2828 	 * have ip_output_simple pick the source address.
2829 	 */
2830 	ire = ire_ftable_lookup_v4(ipha->ipha_dst, 0, 0,
2831 	    (IRE_LOCAL|IRE_LOOPBACK), NULL, ira->ira_zoneid, NULL,
2832 	    MATCH_IRE_TYPE|MATCH_IRE_ZONEONLY, 0, ipst, NULL);
2833 	if (ire != NULL) {
2834 		ire_refrele(ire);
2835 		src = ipha->ipha_dst;
2836 	} else {
2837 		src = INADDR_ANY;
2838 		ixas.ixa_flags |= IXAF_SET_SOURCE;
2839 	}
2840 
2841 	/*
2842 	 * Check if we can send back more then 8 bytes in addition to
2843 	 * the IP header.  We try to send 64 bytes of data and the internal
2844 	 * header in the special cases of ipv4 encapsulated ipv4 or ipv6.
2845 	 */
2846 	len_needed = IPH_HDR_LENGTH(ipha);
2847 	if (ipha->ipha_protocol == IPPROTO_ENCAP ||
2848 	    ipha->ipha_protocol == IPPROTO_IPV6) {
2849 		/*
2850 		 * NOTE: It is posssible that the inner packet is poorly
2851 		 * formed (e.g. IP version is corrupt, or v6 extension headers
2852 		 * got cut off).  The receiver of the ICMP message should see
2853 		 * what we saw.  In the absence of a sane inner-packet (which
2854 		 * protocol types IPPPROTO_ENCAP and IPPROTO_IPV6 indicate
2855 		 * would be an IP header), we should send the size of what is
2856 		 * normally expected to be there (either sizeof (ipha_t) or
2857 		 * sizeof (ip6_t).  It may be useful for diagnostic purposes.
2858 		 *
2859 		 * ALSO NOTE: "inner_ip6h" is the inner packet header, v4 or v6.
2860 		 */
2861 		ip6_t *inner_ip6h = (ip6_t *)((uchar_t *)ipha + len_needed);
2862 
2863 		if (!pullupmsg(mp, -1)) {
2864 			BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
2865 			ip_drop_output("ipIfStatsOutDiscards", mp, NULL);
2866 			freemsg(mp);
2867 			return;
2868 		}
2869 		ipha = (ipha_t *)mp->b_rptr;
2870 
2871 		if (ipha->ipha_protocol == IPPROTO_ENCAP) {
2872 			/*
2873 			 * Check the inner IP version here to guard against
2874 			 * bogons.
2875 			 */
2876 			if (IPH_HDR_VERSION(inner_ip6h) == IPV4_VERSION) {
2877 				len_needed +=
2878 				    IPH_HDR_LENGTH(((uchar_t *)inner_ip6h));
2879 			} else {
2880 				len_needed = sizeof (ipha_t);
2881 			}
2882 		} else {
2883 			ASSERT(ipha->ipha_protocol == IPPROTO_IPV6);
2884 			/* function called next-line checks inner IP version */
2885 			len_needed += ip_hdr_length_v6(mp, inner_ip6h);
2886 		}
2887 	}
2888 	len_needed += ipst->ips_ip_icmp_return;
2889 	msg_len = msgdsize(mp);
2890 	if (msg_len > len_needed) {
2891 		(void) adjmsg(mp, len_needed - msg_len);
2892 		msg_len = len_needed;
2893 	}
2894 	mp1 = allocb(sizeof (icmp_ipha) + len, BPRI_MED);
2895 	if (mp1 == NULL) {
2896 		BUMP_MIB(&ipst->ips_icmp_mib, icmpOutErrors);
2897 		freemsg(mp);
2898 		return;
2899 	}
2900 	mp1->b_cont = mp;
2901 	mp = mp1;
2902 
2903 	/*
2904 	 * Set IXAF_TRUSTED_ICMP so we can let the ICMP messages this
2905 	 * node generates be accepted in peace by all on-host destinations.
2906 	 * If we do NOT assume that all on-host destinations trust
2907 	 * self-generated ICMP messages, then rework here, ip6.c, and spd.c.
2908 	 * (Look for IXAF_TRUSTED_ICMP).
2909 	 */
2910 	ixas.ixa_flags |= IXAF_TRUSTED_ICMP;
2911 
2912 	ipha = (ipha_t *)mp->b_rptr;
2913 	mp1->b_wptr = (uchar_t *)ipha + (sizeof (icmp_ipha) + len);
2914 	*ipha = icmp_ipha;
2915 	ipha->ipha_src = src;
2916 	ipha->ipha_dst = dst;
2917 	ipha->ipha_ttl = ipst->ips_ip_def_ttl;
2918 	msg_len += sizeof (icmp_ipha) + len;
2919 	if (msg_len > IP_MAXPACKET) {
2920 		(void) adjmsg(mp, IP_MAXPACKET - msg_len);
2921 		msg_len = IP_MAXPACKET;
2922 	}
2923 	ipha->ipha_length = htons((uint16_t)msg_len);
2924 	icmph = (icmph_t *)&ipha[1];
2925 	bcopy(stuff, icmph, len);
2926 	icmph->icmph_checksum = 0;
2927 	icmph->icmph_checksum = IP_CSUM(mp, (int32_t)sizeof (ipha_t), 0);
2928 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutMsgs);
2929 
2930 	(void) ip_output_simple(mp, &ixas);
2931 	ixa_cleanup(&ixas);
2932 }
2933 
2934 /*
2935  * Determine if an ICMP error packet can be sent given the rate limit.
2936  * The limit consists of an average frequency (icmp_pkt_err_interval measured
2937  * in milliseconds) and a burst size. Burst size number of packets can
2938  * be sent arbitrarely closely spaced.
2939  * The state is tracked using two variables to implement an approximate
2940  * token bucket filter:
2941  *	icmp_pkt_err_last - lbolt value when the last burst started
2942  *	icmp_pkt_err_sent - number of packets sent in current burst
2943  */
2944 boolean_t
icmp_err_rate_limit(ip_stack_t * ipst)2945 icmp_err_rate_limit(ip_stack_t *ipst)
2946 {
2947 	clock_t now = TICK_TO_MSEC(ddi_get_lbolt());
2948 	uint_t refilled; /* Number of packets refilled in tbf since last */
2949 	/* Guard against changes by loading into local variable */
2950 	uint_t err_interval = ipst->ips_ip_icmp_err_interval;
2951 
2952 	if (err_interval == 0)
2953 		return (B_FALSE);
2954 
2955 	if (ipst->ips_icmp_pkt_err_last > now) {
2956 		/* 100HZ lbolt in ms for 32bit arch wraps every 49.7 days */
2957 		ipst->ips_icmp_pkt_err_last = 0;
2958 		ipst->ips_icmp_pkt_err_sent = 0;
2959 	}
2960 	/*
2961 	 * If we are in a burst update the token bucket filter.
2962 	 * Update the "last" time to be close to "now" but make sure
2963 	 * we don't loose precision.
2964 	 */
2965 	if (ipst->ips_icmp_pkt_err_sent != 0) {
2966 		refilled = (now - ipst->ips_icmp_pkt_err_last)/err_interval;
2967 		if (refilled > ipst->ips_icmp_pkt_err_sent) {
2968 			ipst->ips_icmp_pkt_err_sent = 0;
2969 		} else {
2970 			ipst->ips_icmp_pkt_err_sent -= refilled;
2971 			ipst->ips_icmp_pkt_err_last += refilled * err_interval;
2972 		}
2973 	}
2974 	if (ipst->ips_icmp_pkt_err_sent == 0) {
2975 		/* Start of new burst */
2976 		ipst->ips_icmp_pkt_err_last = now;
2977 	}
2978 	if (ipst->ips_icmp_pkt_err_sent < ipst->ips_ip_icmp_err_burst) {
2979 		ipst->ips_icmp_pkt_err_sent++;
2980 		ip1dbg(("icmp_err_rate_limit: %d sent in burst\n",
2981 		    ipst->ips_icmp_pkt_err_sent));
2982 		return (B_FALSE);
2983 	}
2984 	ip1dbg(("icmp_err_rate_limit: dropped\n"));
2985 	return (B_TRUE);
2986 }
2987 
2988 /*
2989  * Check if it is ok to send an IPv4 ICMP error packet in
2990  * response to the IPv4 packet in mp.
2991  * Free the message and return null if no
2992  * ICMP error packet should be sent.
2993  */
2994 static mblk_t *
icmp_pkt_err_ok(mblk_t * mp,ip_recv_attr_t * ira)2995 icmp_pkt_err_ok(mblk_t *mp, ip_recv_attr_t *ira)
2996 {
2997 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
2998 	icmph_t	*icmph;
2999 	ipha_t	*ipha;
3000 	uint_t	len_needed;
3001 
3002 	if (!mp)
3003 		return (NULL);
3004 	ipha = (ipha_t *)mp->b_rptr;
3005 	if (ip_csum_hdr(ipha)) {
3006 		BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsInCksumErrs);
3007 		ip_drop_input("ipIfStatsInCksumErrs", mp, NULL);
3008 		freemsg(mp);
3009 		return (NULL);
3010 	}
3011 	if (ip_type_v4(ipha->ipha_dst, ipst) == IRE_BROADCAST ||
3012 	    ip_type_v4(ipha->ipha_src, ipst) == IRE_BROADCAST ||
3013 	    CLASSD(ipha->ipha_dst) ||
3014 	    CLASSD(ipha->ipha_src) ||
3015 	    (ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_OFFSET)) {
3016 		/* Note: only errors to the fragment with offset 0 */
3017 		BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
3018 		freemsg(mp);
3019 		return (NULL);
3020 	}
3021 	if (ipha->ipha_protocol == IPPROTO_ICMP) {
3022 		/*
3023 		 * Check the ICMP type.  RFC 1122 sez:  don't send ICMP
3024 		 * errors in response to any ICMP errors.
3025 		 */
3026 		len_needed = IPH_HDR_LENGTH(ipha) + ICMPH_SIZE;
3027 		if (mp->b_wptr - mp->b_rptr < len_needed) {
3028 			if (!pullupmsg(mp, len_needed)) {
3029 				BUMP_MIB(&ipst->ips_icmp_mib, icmpInErrors);
3030 				freemsg(mp);
3031 				return (NULL);
3032 			}
3033 			ipha = (ipha_t *)mp->b_rptr;
3034 		}
3035 		icmph = (icmph_t *)
3036 		    (&((char *)ipha)[IPH_HDR_LENGTH(ipha)]);
3037 		switch (icmph->icmph_type) {
3038 		case ICMP_DEST_UNREACHABLE:
3039 		case ICMP_SOURCE_QUENCH:
3040 		case ICMP_TIME_EXCEEDED:
3041 		case ICMP_PARAM_PROBLEM:
3042 		case ICMP_REDIRECT:
3043 			BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
3044 			freemsg(mp);
3045 			return (NULL);
3046 		default:
3047 			break;
3048 		}
3049 	}
3050 	/*
3051 	 * If this is a labeled system, then check to see if we're allowed to
3052 	 * send a response to this particular sender.  If not, then just drop.
3053 	 */
3054 	if (is_system_labeled() && !tsol_can_reply_error(mp, ira)) {
3055 		ip2dbg(("icmp_pkt_err_ok: can't respond to packet\n"));
3056 		BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
3057 		freemsg(mp);
3058 		return (NULL);
3059 	}
3060 	if (icmp_err_rate_limit(ipst)) {
3061 		/*
3062 		 * Only send ICMP error packets every so often.
3063 		 * This should be done on a per port/source basis,
3064 		 * but for now this will suffice.
3065 		 */
3066 		freemsg(mp);
3067 		return (NULL);
3068 	}
3069 	return (mp);
3070 }
3071 
3072 /*
3073  * Called when a packet was sent out the same link that it arrived on.
3074  * Check if it is ok to send a redirect and then send it.
3075  */
3076 void
ip_send_potential_redirect_v4(mblk_t * mp,ipha_t * ipha,ire_t * ire,ip_recv_attr_t * ira)3077 ip_send_potential_redirect_v4(mblk_t *mp, ipha_t *ipha, ire_t *ire,
3078     ip_recv_attr_t *ira)
3079 {
3080 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
3081 	ipaddr_t	src, nhop;
3082 	mblk_t		*mp1;
3083 	ire_t		*nhop_ire;
3084 
3085 	/*
3086 	 * Check the source address to see if it originated
3087 	 * on the same logical subnet it is going back out on.
3088 	 * If so, we should be able to send it a redirect.
3089 	 * Avoid sending a redirect if the destination
3090 	 * is directly connected (i.e., we matched an IRE_ONLINK),
3091 	 * or if the packet was source routed out this interface.
3092 	 *
3093 	 * We avoid sending a redirect if the
3094 	 * destination is directly connected
3095 	 * because it is possible that multiple
3096 	 * IP subnets may have been configured on
3097 	 * the link, and the source may not
3098 	 * be on the same subnet as ip destination,
3099 	 * even though they are on the same
3100 	 * physical link.
3101 	 */
3102 	if ((ire->ire_type & IRE_ONLINK) ||
3103 	    ip_source_routed(ipha, ipst))
3104 		return;
3105 
3106 	nhop_ire = ire_nexthop(ire);
3107 	if (nhop_ire == NULL)
3108 		return;
3109 
3110 	nhop = nhop_ire->ire_addr;
3111 
3112 	if (nhop_ire->ire_type & IRE_IF_CLONE) {
3113 		ire_t	*ire2;
3114 
3115 		/* Follow ire_dep_parent to find non-clone IRE_INTERFACE */
3116 		mutex_enter(&nhop_ire->ire_lock);
3117 		ire2 = nhop_ire->ire_dep_parent;
3118 		if (ire2 != NULL)
3119 			ire_refhold(ire2);
3120 		mutex_exit(&nhop_ire->ire_lock);
3121 		ire_refrele(nhop_ire);
3122 		nhop_ire = ire2;
3123 	}
3124 	if (nhop_ire == NULL)
3125 		return;
3126 
3127 	ASSERT(!(nhop_ire->ire_type & IRE_IF_CLONE));
3128 
3129 	src = ipha->ipha_src;
3130 
3131 	/*
3132 	 * We look at the interface ire for the nexthop,
3133 	 * to see if ipha_src is in the same subnet
3134 	 * as the nexthop.
3135 	 */
3136 	if ((src & nhop_ire->ire_mask) == (nhop & nhop_ire->ire_mask)) {
3137 		/*
3138 		 * The source is directly connected.
3139 		 */
3140 		mp1 = copymsg(mp);
3141 		if (mp1 != NULL) {
3142 			icmp_send_redirect(mp1, nhop, ira);
3143 		}
3144 	}
3145 	ire_refrele(nhop_ire);
3146 }
3147 
3148 /*
3149  * Generate an ICMP redirect message.
3150  */
3151 static void
icmp_send_redirect(mblk_t * mp,ipaddr_t gateway,ip_recv_attr_t * ira)3152 icmp_send_redirect(mblk_t *mp, ipaddr_t gateway, ip_recv_attr_t *ira)
3153 {
3154 	icmph_t	icmph;
3155 	ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3156 
3157 	mp = icmp_pkt_err_ok(mp, ira);
3158 	if (mp == NULL)
3159 		return;
3160 
3161 	bzero(&icmph, sizeof (icmph_t));
3162 	icmph.icmph_type = ICMP_REDIRECT;
3163 	icmph.icmph_code = 1;
3164 	icmph.icmph_rd_gateway = gateway;
3165 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutRedirects);
3166 	icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3167 }
3168 
3169 /*
3170  * Generate an ICMP time exceeded message.
3171  */
3172 void
icmp_time_exceeded(mblk_t * mp,uint8_t code,ip_recv_attr_t * ira)3173 icmp_time_exceeded(mblk_t *mp, uint8_t code, ip_recv_attr_t *ira)
3174 {
3175 	icmph_t	icmph;
3176 	ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3177 
3178 	mp = icmp_pkt_err_ok(mp, ira);
3179 	if (mp == NULL)
3180 		return;
3181 
3182 	bzero(&icmph, sizeof (icmph_t));
3183 	icmph.icmph_type = ICMP_TIME_EXCEEDED;
3184 	icmph.icmph_code = code;
3185 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutTimeExcds);
3186 	icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3187 }
3188 
3189 /*
3190  * Generate an ICMP unreachable message.
3191  * When called from ip_output side a minimal ip_recv_attr_t needs to be
3192  * constructed by the caller.
3193  */
3194 void
icmp_unreachable(mblk_t * mp,uint8_t code,ip_recv_attr_t * ira)3195 icmp_unreachable(mblk_t *mp, uint8_t code, ip_recv_attr_t *ira)
3196 {
3197 	icmph_t	icmph;
3198 	ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3199 
3200 	mp = icmp_pkt_err_ok(mp, ira);
3201 	if (mp == NULL)
3202 		return;
3203 
3204 	bzero(&icmph, sizeof (icmph_t));
3205 	icmph.icmph_type = ICMP_DEST_UNREACHABLE;
3206 	icmph.icmph_code = code;
3207 	BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDestUnreachs);
3208 	icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3209 }
3210 
3211 /*
3212  * Latch in the IPsec state for a stream based the policy in the listener
3213  * and the actions in the ip_recv_attr_t.
3214  * Called directly from TCP and SCTP.
3215  */
3216 boolean_t
ip_ipsec_policy_inherit(conn_t * connp,conn_t * lconnp,ip_recv_attr_t * ira)3217 ip_ipsec_policy_inherit(conn_t *connp, conn_t *lconnp, ip_recv_attr_t *ira)
3218 {
3219 	ASSERT(lconnp->conn_policy != NULL);
3220 	ASSERT(connp->conn_policy == NULL);
3221 
3222 	IPPH_REFHOLD(lconnp->conn_policy);
3223 	connp->conn_policy = lconnp->conn_policy;
3224 
3225 	if (ira->ira_ipsec_action != NULL) {
3226 		if (connp->conn_latch == NULL) {
3227 			connp->conn_latch = iplatch_create();
3228 			if (connp->conn_latch == NULL)
3229 				return (B_FALSE);
3230 		}
3231 		ipsec_latch_inbound(connp, ira);
3232 	}
3233 	return (B_TRUE);
3234 }
3235 
3236 /*
3237  * Verify whether or not the IP address is a valid local address.
3238  * Could be a unicast, including one for a down interface.
3239  * If allow_mcbc then a multicast or broadcast address is also
3240  * acceptable.
3241  *
3242  * In the case of a broadcast/multicast address, however, the
3243  * upper protocol is expected to reset the src address
3244  * to zero when we return IPVL_MCAST/IPVL_BCAST so that
3245  * no packets are emitted with broadcast/multicast address as
3246  * source address (that violates hosts requirements RFC 1122)
3247  * The addresses valid for bind are:
3248  *	(1) - INADDR_ANY (0)
3249  *	(2) - IP address of an UP interface
3250  *	(3) - IP address of a DOWN interface
3251  *	(4) - valid local IP broadcast addresses. In this case
3252  *	the conn will only receive packets destined to
3253  *	the specified broadcast address.
3254  *	(5) - a multicast address. In this case
3255  *	the conn will only receive packets destined to
3256  *	the specified multicast address. Note: the
3257  *	application still has to issue an
3258  *	IP_ADD_MEMBERSHIP socket option.
3259  *
3260  * In all the above cases, the bound address must be valid in the current zone.
3261  * When the address is loopback, multicast or broadcast, there might be many
3262  * matching IREs so bind has to look up based on the zone.
3263  */
3264 ip_laddr_t
ip_laddr_verify_v4(ipaddr_t src_addr,zoneid_t zoneid,ip_stack_t * ipst,boolean_t allow_mcbc)3265 ip_laddr_verify_v4(ipaddr_t src_addr, zoneid_t zoneid,
3266     ip_stack_t *ipst, boolean_t allow_mcbc)
3267 {
3268 	ire_t *src_ire;
3269 
3270 	ASSERT(src_addr != INADDR_ANY);
3271 
3272 	src_ire = ire_ftable_lookup_v4(src_addr, 0, 0, 0,
3273 	    NULL, zoneid, NULL, MATCH_IRE_ZONEONLY, 0, ipst, NULL);
3274 
3275 	/*
3276 	 * If an address other than in6addr_any is requested,
3277 	 * we verify that it is a valid address for bind
3278 	 * Note: Following code is in if-else-if form for
3279 	 * readability compared to a condition check.
3280 	 */
3281 	if (src_ire != NULL && (src_ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK))) {
3282 		/*
3283 		 * (2) Bind to address of local UP interface
3284 		 */
3285 		ire_refrele(src_ire);
3286 		return (IPVL_UNICAST_UP);
3287 	} else if (src_ire != NULL && src_ire->ire_type & IRE_BROADCAST) {
3288 		/*
3289 		 * (4) Bind to broadcast address
3290 		 */
3291 		ire_refrele(src_ire);
3292 		if (allow_mcbc)
3293 			return (IPVL_BCAST);
3294 		else
3295 			return (IPVL_BAD);
3296 	} else if (CLASSD(src_addr)) {
3297 		/* (5) bind to multicast address. */
3298 		if (src_ire != NULL)
3299 			ire_refrele(src_ire);
3300 
3301 		if (allow_mcbc)
3302 			return (IPVL_MCAST);
3303 		else
3304 			return (IPVL_BAD);
3305 	} else {
3306 		ipif_t *ipif;
3307 
3308 		/*
3309 		 * (3) Bind to address of local DOWN interface?
3310 		 * (ipif_lookup_addr() looks up all interfaces
3311 		 * but we do not get here for UP interfaces
3312 		 * - case (2) above)
3313 		 */
3314 		if (src_ire != NULL)
3315 			ire_refrele(src_ire);
3316 
3317 		ipif = ipif_lookup_addr(src_addr, NULL, zoneid, ipst);
3318 		if (ipif == NULL)
3319 			return (IPVL_BAD);
3320 
3321 		/* Not a useful source? */
3322 		if (ipif->ipif_flags & (IPIF_NOLOCAL | IPIF_ANYCAST)) {
3323 			ipif_refrele(ipif);
3324 			return (IPVL_BAD);
3325 		}
3326 		ipif_refrele(ipif);
3327 		return (IPVL_UNICAST_DOWN);
3328 	}
3329 }
3330 
3331 /*
3332  * Insert in the bind fanout for IPv4 and IPv6.
3333  * The caller should already have used ip_laddr_verify_v*() before calling
3334  * this.
3335  */
3336 int
ip_laddr_fanout_insert(conn_t * connp)3337 ip_laddr_fanout_insert(conn_t *connp)
3338 {
3339 	int		error;
3340 
3341 	/*
3342 	 * Allow setting new policies. For example, disconnects result
3343 	 * in us being called. As we would have set conn_policy_cached
3344 	 * to B_TRUE before, we should set it to B_FALSE, so that policy
3345 	 * can change after the disconnect.
3346 	 */
3347 	connp->conn_policy_cached = B_FALSE;
3348 
3349 	error = ipcl_bind_insert(connp);
3350 	if (error != 0) {
3351 		if (connp->conn_anon_port) {
3352 			(void) tsol_mlp_anon(crgetzone(connp->conn_cred),
3353 			    connp->conn_mlp_type, connp->conn_proto,
3354 			    ntohs(connp->conn_lport), B_FALSE);
3355 		}
3356 		connp->conn_mlp_type = mlptSingle;
3357 	}
3358 	return (error);
3359 }
3360 
3361 /*
3362  * Verify that both the source and destination addresses are valid. If
3363  * IPDF_VERIFY_DST is not set, then the destination address may be unreachable,
3364  * i.e. have no route to it.  Protocols like TCP want to verify destination
3365  * reachability, while tunnels do not.
3366  *
3367  * Determine the route, the interface, and (optionally) the source address
3368  * to use to reach a given destination.
3369  * Note that we allow connect to broadcast and multicast addresses when
3370  * IPDF_ALLOW_MCBC is set.
3371  * first_hop and dst_addr are normally the same, but if source routing
3372  * they will differ; in that case the first_hop is what we'll use for the
3373  * routing lookup but the dce and label checks will be done on dst_addr,
3374  *
3375  * If uinfo is set, then we fill in the best available information
3376  * we have for the destination. This is based on (in priority order) any
3377  * metrics and path MTU stored in a dce_t, route metrics, and finally the
3378  * ill_mtu/ill_mc_mtu.
3379  *
3380  * Tsol note: If we have a source route then dst_addr != firsthop. But we
3381  * always do the label check on dst_addr.
3382  */
3383 int
ip_set_destination_v4(ipaddr_t * src_addrp,ipaddr_t dst_addr,ipaddr_t firsthop,ip_xmit_attr_t * ixa,iulp_t * uinfo,uint32_t flags,uint_t mac_mode)3384 ip_set_destination_v4(ipaddr_t *src_addrp, ipaddr_t dst_addr, ipaddr_t firsthop,
3385     ip_xmit_attr_t *ixa, iulp_t *uinfo, uint32_t flags, uint_t mac_mode)
3386 {
3387 	ire_t		*ire = NULL;
3388 	int		error = 0;
3389 	ipaddr_t	setsrc;				/* RTF_SETSRC */
3390 	zoneid_t	zoneid = ixa->ixa_zoneid;	/* Honors SO_ALLZONES */
3391 	ip_stack_t	*ipst = ixa->ixa_ipst;
3392 	dce_t		*dce;
3393 	uint_t		pmtu;
3394 	uint_t		generation;
3395 	nce_t		*nce;
3396 	ill_t		*ill = NULL;
3397 	boolean_t	multirt = B_FALSE;
3398 
3399 	ASSERT(ixa->ixa_flags & IXAF_IS_IPV4);
3400 
3401 	/*
3402 	 * We never send to zero; the ULPs map it to the loopback address.
3403 	 * We can't allow it since we use zero to mean unitialized in some
3404 	 * places.
3405 	 */
3406 	ASSERT(dst_addr != INADDR_ANY);
3407 
3408 	if (is_system_labeled()) {
3409 		ts_label_t *tsl = NULL;
3410 
3411 		error = tsol_check_dest(ixa->ixa_tsl, &dst_addr, IPV4_VERSION,
3412 		    mac_mode, (flags & IPDF_ZONE_IS_GLOBAL) != 0, &tsl);
3413 		if (error != 0)
3414 			return (error);
3415 		if (tsl != NULL) {
3416 			/* Update the label */
3417 			ip_xmit_attr_replace_tsl(ixa, tsl);
3418 		}
3419 	}
3420 
3421 	setsrc = INADDR_ANY;
3422 	/*
3423 	 * Select a route; For IPMP interfaces, we would only select
3424 	 * a "hidden" route (i.e., going through a specific under_ill)
3425 	 * if ixa_ifindex has been specified.
3426 	 */
3427 	ire = ip_select_route_v4(firsthop, *src_addrp, ixa,
3428 	    &generation, &setsrc, &error, &multirt);
3429 	ASSERT(ire != NULL);	/* IRE_NOROUTE if none found */
3430 	if (error != 0)
3431 		goto bad_addr;
3432 
3433 	/*
3434 	 * ire can't be a broadcast or multicast unless IPDF_ALLOW_MCBC is set.
3435 	 * If IPDF_VERIFY_DST is set, the destination must be reachable;
3436 	 * Otherwise the destination needn't be reachable.
3437 	 *
3438 	 * If we match on a reject or black hole, then we've got a
3439 	 * local failure.  May as well fail out the connect() attempt,
3440 	 * since it's never going to succeed.
3441 	 */
3442 	if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
3443 		/*
3444 		 * If we're verifying destination reachability, we always want
3445 		 * to complain here.
3446 		 *
3447 		 * If we're not verifying destination reachability but the
3448 		 * destination has a route, we still want to fail on the
3449 		 * temporary address and broadcast address tests.
3450 		 *
3451 		 * In both cases do we let the code continue so some reasonable
3452 		 * information is returned to the caller. That enables the
3453 		 * caller to use (and even cache) the IRE. conn_ip_ouput will
3454 		 * use the generation mismatch path to check for the unreachable
3455 		 * case thereby avoiding any specific check in the main path.
3456 		 */
3457 		ASSERT(generation == IRE_GENERATION_VERIFY);
3458 		if (flags & IPDF_VERIFY_DST) {
3459 			/*
3460 			 * Set errno but continue to set up ixa_ire to be
3461 			 * the RTF_REJECT|RTF_BLACKHOLE IRE.
3462 			 * That allows callers to use ip_output to get an
3463 			 * ICMP error back.
3464 			 */
3465 			if (!(ire->ire_type & IRE_HOST))
3466 				error = ENETUNREACH;
3467 			else
3468 				error = EHOSTUNREACH;
3469 		}
3470 	}
3471 
3472 	if ((ire->ire_type & (IRE_BROADCAST|IRE_MULTICAST)) &&
3473 	    !(flags & IPDF_ALLOW_MCBC)) {
3474 		ire_refrele(ire);
3475 		ire = ire_reject(ipst, B_FALSE);
3476 		generation = IRE_GENERATION_VERIFY;
3477 		error = ENETUNREACH;
3478 	}
3479 
3480 	/* Cache things */
3481 	if (ixa->ixa_ire != NULL)
3482 		ire_refrele_notr(ixa->ixa_ire);
3483 #ifdef DEBUG
3484 	ire_refhold_notr(ire);
3485 	ire_refrele(ire);
3486 #endif
3487 	ixa->ixa_ire = ire;
3488 	ixa->ixa_ire_generation = generation;
3489 
3490 	/*
3491 	 * Ensure that ixa_dce is always set any time that ixa_ire is set,
3492 	 * since some callers will send a packet to conn_ip_output() even if
3493 	 * there's an error.
3494 	 */
3495 	if (flags & IPDF_UNIQUE_DCE) {
3496 		/* Fallback to the default dce if allocation fails */
3497 		dce = dce_lookup_and_add_v4(dst_addr, ipst);
3498 		if (dce != NULL)
3499 			generation = dce->dce_generation;
3500 		else
3501 			dce = dce_lookup_v4(dst_addr, ipst, &generation);
3502 	} else {
3503 		dce = dce_lookup_v4(dst_addr, ipst, &generation);
3504 	}
3505 	ASSERT(dce != NULL);
3506 	if (ixa->ixa_dce != NULL)
3507 		dce_refrele_notr(ixa->ixa_dce);
3508 #ifdef DEBUG
3509 	dce_refhold_notr(dce);
3510 	dce_refrele(dce);
3511 #endif
3512 	ixa->ixa_dce = dce;
3513 	ixa->ixa_dce_generation = generation;
3514 
3515 	/*
3516 	 * For multicast with multirt we have a flag passed back from
3517 	 * ire_lookup_multi_ill_v4 since we don't have an IRE for each
3518 	 * possible multicast address.
3519 	 * We also need a flag for multicast since we can't check
3520 	 * whether RTF_MULTIRT is set in ixa_ire for multicast.
3521 	 */
3522 	if (multirt) {
3523 		ixa->ixa_postfragfn = ip_postfrag_multirt_v4;
3524 		ixa->ixa_flags |= IXAF_MULTIRT_MULTICAST;
3525 	} else {
3526 		ixa->ixa_postfragfn = ire->ire_postfragfn;
3527 		ixa->ixa_flags &= ~IXAF_MULTIRT_MULTICAST;
3528 	}
3529 	if (!(ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))) {
3530 		/* Get an nce to cache. */
3531 		nce = ire_to_nce(ire, firsthop, NULL);
3532 		if (nce == NULL) {
3533 			/* Allocation failure? */
3534 			ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3535 		} else {
3536 			if (ixa->ixa_nce != NULL)
3537 				nce_refrele(ixa->ixa_nce);
3538 			ixa->ixa_nce = nce;
3539 		}
3540 	}
3541 
3542 	/*
3543 	 * If the source address is a loopback address, the
3544 	 * destination had best be local or multicast.
3545 	 * If we are sending to an IRE_LOCAL using a loopback source then
3546 	 * it had better be the same zoneid.
3547 	 */
3548 	if (*src_addrp == htonl(INADDR_LOOPBACK)) {
3549 		if ((ire->ire_type & IRE_LOCAL) && ire->ire_zoneid != zoneid) {
3550 			ire = NULL;	/* Stored in ixa_ire */
3551 			error = EADDRNOTAVAIL;
3552 			goto bad_addr;
3553 		}
3554 		if (!(ire->ire_type & (IRE_LOOPBACK|IRE_LOCAL|IRE_MULTICAST))) {
3555 			ire = NULL;	/* Stored in ixa_ire */
3556 			error = EADDRNOTAVAIL;
3557 			goto bad_addr;
3558 		}
3559 	}
3560 	if (ire->ire_type & IRE_BROADCAST) {
3561 		/*
3562 		 * If the ULP didn't have a specified source, then we
3563 		 * make sure we reselect the source when sending
3564 		 * broadcasts out different interfaces.
3565 		 */
3566 		if (flags & IPDF_SELECT_SRC)
3567 			ixa->ixa_flags |= IXAF_SET_SOURCE;
3568 		else
3569 			ixa->ixa_flags &= ~IXAF_SET_SOURCE;
3570 	}
3571 
3572 	/*
3573 	 * Does the caller want us to pick a source address?
3574 	 */
3575 	if (flags & IPDF_SELECT_SRC) {
3576 		ipaddr_t	src_addr;
3577 
3578 		/*
3579 		 * We use use ire_nexthop_ill to avoid the under ipmp
3580 		 * interface for source address selection. Note that for ipmp
3581 		 * probe packets, ixa_ifindex would have been specified, and
3582 		 * the ip_select_route() invocation would have picked an ire
3583 		 * will ire_ill pointing at an under interface.
3584 		 */
3585 		ill = ire_nexthop_ill(ire);
3586 
3587 		/* If unreachable we have no ill but need some source */
3588 		if (ill == NULL) {
3589 			src_addr = htonl(INADDR_LOOPBACK);
3590 			/* Make sure we look for a better source address */
3591 			generation = SRC_GENERATION_VERIFY;
3592 		} else {
3593 			error = ip_select_source_v4(ill, setsrc, dst_addr,
3594 			    ixa->ixa_multicast_ifaddr, zoneid,
3595 			    ipst, &src_addr, &generation, NULL);
3596 			if (error != 0) {
3597 				ire = NULL;	/* Stored in ixa_ire */
3598 				goto bad_addr;
3599 			}
3600 		}
3601 
3602 		/*
3603 		 * We allow the source address to to down.
3604 		 * However, we check that we don't use the loopback address
3605 		 * as a source when sending out on the wire.
3606 		 */
3607 		if ((src_addr == htonl(INADDR_LOOPBACK)) &&
3608 		    !(ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK|IRE_MULTICAST)) &&
3609 		    !(ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))) {
3610 			ire = NULL;	/* Stored in ixa_ire */
3611 			error = EADDRNOTAVAIL;
3612 			goto bad_addr;
3613 		}
3614 
3615 		*src_addrp = src_addr;
3616 		ixa->ixa_src_generation = generation;
3617 	}
3618 
3619 	/*
3620 	 * Make sure we don't leave an unreachable ixa_nce in place
3621 	 * since ip_select_route is used when we unplumb i.e., remove
3622 	 * references on ixa_ire, ixa_nce, and ixa_dce.
3623 	 */
3624 	nce = ixa->ixa_nce;
3625 	if (nce != NULL && nce->nce_is_condemned) {
3626 		nce_refrele(nce);
3627 		ixa->ixa_nce = NULL;
3628 		ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3629 	}
3630 
3631 	/*
3632 	 * The caller has set IXAF_PMTU_DISCOVERY if path MTU is desired.
3633 	 * However, we can't do it for IPv4 multicast or broadcast.
3634 	 */
3635 	if (ire->ire_type & (IRE_BROADCAST|IRE_MULTICAST))
3636 		ixa->ixa_flags &= ~IXAF_PMTU_DISCOVERY;
3637 
3638 	/*
3639 	 * Set initial value for fragmentation limit. Either conn_ip_output
3640 	 * or ULP might updates it when there are routing changes.
3641 	 * Handles a NULL ixa_ire->ire_ill or a NULL ixa_nce for RTF_REJECT.
3642 	 */
3643 	pmtu = ip_get_pmtu(ixa);
3644 	ixa->ixa_fragsize = pmtu;
3645 	/* Make sure ixa_fragsize and ixa_pmtu remain identical */
3646 	if (ixa->ixa_flags & IXAF_VERIFY_PMTU)
3647 		ixa->ixa_pmtu = pmtu;
3648 
3649 	/*
3650 	 * Extract information useful for some transports.
3651 	 * First we look for DCE metrics. Then we take what we have in
3652 	 * the metrics in the route, where the offlink is used if we have
3653 	 * one.
3654 	 */
3655 	if (uinfo != NULL) {
3656 		bzero(uinfo, sizeof (*uinfo));
3657 
3658 		if (dce->dce_flags & DCEF_UINFO)
3659 			*uinfo = dce->dce_uinfo;
3660 
3661 		rts_merge_metrics(uinfo, &ire->ire_metrics);
3662 
3663 		/* Allow ire_metrics to decrease the path MTU from above */
3664 		if (uinfo->iulp_mtu == 0 || uinfo->iulp_mtu > pmtu)
3665 			uinfo->iulp_mtu = pmtu;
3666 
3667 		uinfo->iulp_localnet = (ire->ire_type & IRE_ONLINK) != 0;
3668 		uinfo->iulp_loopback = (ire->ire_type & IRE_LOOPBACK) != 0;
3669 		uinfo->iulp_local = (ire->ire_type & IRE_LOCAL) != 0;
3670 	}
3671 
3672 	if (ill != NULL)
3673 		ill_refrele(ill);
3674 
3675 	return (error);
3676 
3677 bad_addr:
3678 	if (ire != NULL)
3679 		ire_refrele(ire);
3680 
3681 	if (ill != NULL)
3682 		ill_refrele(ill);
3683 
3684 	/*
3685 	 * Make sure we don't leave an unreachable ixa_nce in place
3686 	 * since ip_select_route is used when we unplumb i.e., remove
3687 	 * references on ixa_ire, ixa_nce, and ixa_dce.
3688 	 */
3689 	nce = ixa->ixa_nce;
3690 	if (nce != NULL && nce->nce_is_condemned) {
3691 		nce_refrele(nce);
3692 		ixa->ixa_nce = NULL;
3693 		ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3694 	}
3695 
3696 	return (error);
3697 }
3698 
3699 
3700 /*
3701  * Get the base MTU for the case when path MTU discovery is not used.
3702  * Takes the MTU of the IRE into account.
3703  */
3704 uint_t
ip_get_base_mtu(ill_t * ill,ire_t * ire)3705 ip_get_base_mtu(ill_t *ill, ire_t *ire)
3706 {
3707 	uint_t mtu;
3708 	uint_t iremtu = ire->ire_metrics.iulp_mtu;
3709 
3710 	if (ire->ire_type & (IRE_MULTICAST|IRE_BROADCAST))
3711 		mtu = ill->ill_mc_mtu;
3712 	else
3713 		mtu = ill->ill_mtu;
3714 
3715 	if (iremtu != 0 && iremtu < mtu)
3716 		mtu = iremtu;
3717 
3718 	return (mtu);
3719 }
3720 
3721 /*
3722  * Get the PMTU for the attributes. Handles both IPv4 and IPv6.
3723  * Assumes that ixa_ire, dce, and nce have already been set up.
3724  *
3725  * The caller has set IXAF_PMTU_DISCOVERY if path MTU discovery is desired.
3726  * We avoid path MTU discovery if it is disabled with ndd.
3727  * Furtermore, if the path MTU is too small, then we don't set DF for IPv4.
3728  *
3729  * NOTE: We also used to turn it off for source routed packets. That
3730  * is no longer required since the dce is per final destination.
3731  */
3732 uint_t
ip_get_pmtu(ip_xmit_attr_t * ixa)3733 ip_get_pmtu(ip_xmit_attr_t *ixa)
3734 {
3735 	ip_stack_t	*ipst = ixa->ixa_ipst;
3736 	dce_t		*dce;
3737 	nce_t		*nce;
3738 	ire_t		*ire;
3739 	uint_t		pmtu;
3740 
3741 	ire = ixa->ixa_ire;
3742 	dce = ixa->ixa_dce;
3743 	nce = ixa->ixa_nce;
3744 
3745 	/*
3746 	 * If path MTU discovery has been turned off by ndd, then we ignore
3747 	 * any dce_pmtu and for IPv4 we will not set DF.
3748 	 */
3749 	if (!ipst->ips_ip_path_mtu_discovery)
3750 		ixa->ixa_flags &= ~IXAF_PMTU_DISCOVERY;
3751 
3752 	pmtu = IP_MAXPACKET;
3753 	/*
3754 	 * We need to determine if it is acceptable to set DF for IPv4 or not
3755 	 * and for IPv6 if we need to use the minimum MTU. If a connection has
3756 	 * opted into path MTU discovery, then we can use 'DF' in IPv4 and do
3757 	 * not have to constrain ourselves to the IPv6 minimum MTU. There is a
3758 	 * second consideration here: IXAF_DONTFRAG. This is set as a result of
3759 	 * someone setting the IP_DONTFRAG or IPV6_DONTFRAG socket option. In
3760 	 * such a case, it is acceptable to set DF for IPv4 and to use a larger
3761 	 * MTU. Note, the actual MTU is constrained by the ill_t later on in
3762 	 * this function.
3763 	 */
3764 	if (ixa->ixa_flags & (IXAF_PMTU_DISCOVERY | IXAF_DONTFRAG)) {
3765 		ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3766 	} else {
3767 		ixa->ixa_flags &= ~IXAF_PMTU_IPV4_DF;
3768 		if (!(ixa->ixa_flags & IXAF_IS_IPV4))
3769 			pmtu = IPV6_MIN_MTU;
3770 	}
3771 
3772 	/* Check if the PMTU is to old before we use it */
3773 	if ((dce->dce_flags & DCEF_PMTU) &&
3774 	    TICK_TO_SEC(ddi_get_lbolt64()) - dce->dce_last_change_time >
3775 	    ipst->ips_ip_pathmtu_interval) {
3776 		/*
3777 		 * Older than 20 minutes. Drop the path MTU information.
3778 		 */
3779 		mutex_enter(&dce->dce_lock);
3780 		dce->dce_flags &= ~(DCEF_PMTU|DCEF_TOO_SMALL_PMTU);
3781 		dce->dce_last_change_time = TICK_TO_SEC(ddi_get_lbolt64());
3782 		mutex_exit(&dce->dce_lock);
3783 		dce_increment_generation(dce);
3784 	}
3785 
3786 	/* The metrics on the route can lower the path MTU */
3787 	if (ire->ire_metrics.iulp_mtu != 0 &&
3788 	    ire->ire_metrics.iulp_mtu < pmtu)
3789 		pmtu = ire->ire_metrics.iulp_mtu;
3790 
3791 	/*
3792 	 * If the path MTU is smaller than some minimum, we still use dce_pmtu
3793 	 * above (would be 576 for IPv4 and 1280 for IPv6), but we clear
3794 	 * IXAF_PMTU_IPV4_DF so that we avoid setting DF for IPv4.
3795 	 */
3796 	if (ixa->ixa_flags & IXAF_PMTU_DISCOVERY) {
3797 		if (dce->dce_flags & DCEF_PMTU) {
3798 			if (dce->dce_pmtu < pmtu)
3799 				pmtu = dce->dce_pmtu;
3800 
3801 			if (dce->dce_flags & DCEF_TOO_SMALL_PMTU) {
3802 				ixa->ixa_flags |= IXAF_PMTU_TOO_SMALL;
3803 				ixa->ixa_flags &= ~IXAF_PMTU_IPV4_DF;
3804 			} else {
3805 				ixa->ixa_flags &= ~IXAF_PMTU_TOO_SMALL;
3806 				ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3807 			}
3808 		} else {
3809 			ixa->ixa_flags &= ~IXAF_PMTU_TOO_SMALL;
3810 			ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3811 		}
3812 	}
3813 
3814 	/*
3815 	 * If we have an IRE_LOCAL we use the loopback mtu instead of
3816 	 * the ill for going out the wire i.e., IRE_LOCAL gets the same
3817 	 * mtu as IRE_LOOPBACK.
3818 	 */
3819 	if (ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK)) {
3820 		uint_t loopback_mtu;
3821 
3822 		loopback_mtu = (ire->ire_ipversion == IPV6_VERSION) ?
3823 		    ip_loopback_mtu_v6plus : ip_loopback_mtuplus;
3824 
3825 		if (loopback_mtu < pmtu)
3826 			pmtu = loopback_mtu;
3827 	} else if (nce != NULL) {
3828 		/*
3829 		 * Make sure we don't exceed the interface MTU.
3830 		 * In the case of RTF_REJECT or RTF_BLACKHOLE we might not have
3831 		 * an ill. We'd use the above IP_MAXPACKET in that case just
3832 		 * to tell the transport something larger than zero.
3833 		 */
3834 		if (ire->ire_type & (IRE_MULTICAST|IRE_BROADCAST)) {
3835 			if (nce->nce_common->ncec_ill->ill_mc_mtu < pmtu)
3836 				pmtu = nce->nce_common->ncec_ill->ill_mc_mtu;
3837 			if (nce->nce_common->ncec_ill != nce->nce_ill &&
3838 			    nce->nce_ill->ill_mc_mtu < pmtu) {
3839 				/*
3840 				 * for interfaces in an IPMP group, the mtu of
3841 				 * the nce_ill (under_ill) could be different
3842 				 * from the mtu of the ncec_ill, so we take the
3843 				 * min of the two.
3844 				 */
3845 				pmtu = nce->nce_ill->ill_mc_mtu;
3846 			}
3847 		} else {
3848 			if (nce->nce_common->ncec_ill->ill_mtu < pmtu)
3849 				pmtu = nce->nce_common->ncec_ill->ill_mtu;
3850 			if (nce->nce_common->ncec_ill != nce->nce_ill &&
3851 			    nce->nce_ill->ill_mtu < pmtu) {
3852 				/*
3853 				 * for interfaces in an IPMP group, the mtu of
3854 				 * the nce_ill (under_ill) could be different
3855 				 * from the mtu of the ncec_ill, so we take the
3856 				 * min of the two.
3857 				 */
3858 				pmtu = nce->nce_ill->ill_mtu;
3859 			}
3860 		}
3861 	}
3862 
3863 	/*
3864 	 * Handle the IPV6_USE_MIN_MTU socket option or ancillary data.
3865 	 * Only applies to IPv6.
3866 	 */
3867 	if (!(ixa->ixa_flags & IXAF_IS_IPV4)) {
3868 		if (ixa->ixa_flags & IXAF_USE_MIN_MTU) {
3869 			switch (ixa->ixa_use_min_mtu) {
3870 			case IPV6_USE_MIN_MTU_MULTICAST:
3871 				if (ire->ire_type & IRE_MULTICAST)
3872 					pmtu = IPV6_MIN_MTU;
3873 				break;
3874 			case IPV6_USE_MIN_MTU_ALWAYS:
3875 				pmtu = IPV6_MIN_MTU;
3876 				break;
3877 			case IPV6_USE_MIN_MTU_NEVER:
3878 				break;
3879 			}
3880 		} else {
3881 			/* Default is IPV6_USE_MIN_MTU_MULTICAST */
3882 			if (ire->ire_type & IRE_MULTICAST)
3883 				pmtu = IPV6_MIN_MTU;
3884 		}
3885 	}
3886 
3887 	/*
3888 	 * For multirouted IPv6 packets, the IP layer will insert a 8-byte
3889 	 * fragment header in every packet. We compensate for those cases by
3890 	 * returning a smaller path MTU to the ULP.
3891 	 *
3892 	 * In the case of CGTP then ip_output will add a fragment header.
3893 	 * Make sure there is room for it by telling a smaller number
3894 	 * to the transport.
3895 	 *
3896 	 * When IXAF_IPV6_ADDR_FRAGHDR we subtract the frag hdr here
3897 	 * so the ULPs consistently see a iulp_pmtu and ip_get_pmtu()
3898 	 * which is the size of the packets it can send.
3899 	 */
3900 	if (!(ixa->ixa_flags & IXAF_IS_IPV4)) {
3901 		if ((ire->ire_flags & RTF_MULTIRT) ||
3902 		    (ixa->ixa_flags & IXAF_MULTIRT_MULTICAST)) {
3903 			pmtu -= sizeof (ip6_frag_t);
3904 			ixa->ixa_flags |= IXAF_IPV6_ADD_FRAGHDR;
3905 		}
3906 	}
3907 
3908 	return (pmtu);
3909 }
3910 
3911 /*
3912  * Carve "len" bytes out of an mblk chain, consuming any we empty, and duping
3913  * the final piece where we don't.  Return a pointer to the first mblk in the
3914  * result, and update the pointer to the next mblk to chew on.  If anything
3915  * goes wrong (i.e., dupb fails), we waste everything in sight and return a
3916  * NULL pointer.
3917  */
3918 mblk_t *
ip_carve_mp(mblk_t ** mpp,ssize_t len)3919 ip_carve_mp(mblk_t **mpp, ssize_t len)
3920 {
3921 	mblk_t	*mp0;
3922 	mblk_t	*mp1;
3923 	mblk_t	*mp2;
3924 
3925 	if (!len || !mpp || !(mp0 = *mpp))
3926 		return (NULL);
3927 	/* If we aren't going to consume the first mblk, we need a dup. */
3928 	if (mp0->b_wptr - mp0->b_rptr > len) {
3929 		mp1 = dupb(mp0);
3930 		if (mp1) {
3931 			/* Partition the data between the two mblks. */
3932 			mp1->b_wptr = mp1->b_rptr + len;
3933 			mp0->b_rptr = mp1->b_wptr;
3934 			/*
3935 			 * after adjustments if mblk not consumed is now
3936 			 * unaligned, try to align it. If this fails free
3937 			 * all messages and let upper layer recover.
3938 			 */
3939 			if (!OK_32PTR(mp0->b_rptr)) {
3940 				if (!pullupmsg(mp0, -1)) {
3941 					freemsg(mp0);
3942 					freemsg(mp1);
3943 					*mpp = NULL;
3944 					return (NULL);
3945 				}
3946 			}
3947 		}
3948 		return (mp1);
3949 	}
3950 	/* Eat through as many mblks as we need to get len bytes. */
3951 	len -= mp0->b_wptr - mp0->b_rptr;
3952 	for (mp2 = mp1 = mp0; (mp2 = mp2->b_cont) != 0 && len; mp1 = mp2) {
3953 		if (mp2->b_wptr - mp2->b_rptr > len) {
3954 			/*
3955 			 * We won't consume the entire last mblk.  Like
3956 			 * above, dup and partition it.
3957 			 */
3958 			mp1->b_cont = dupb(mp2);
3959 			mp1 = mp1->b_cont;
3960 			if (!mp1) {
3961 				/*
3962 				 * Trouble.  Rather than go to a lot of
3963 				 * trouble to clean up, we free the messages.
3964 				 * This won't be any worse than losing it on
3965 				 * the wire.
3966 				 */
3967 				freemsg(mp0);
3968 				freemsg(mp2);
3969 				*mpp = NULL;
3970 				return (NULL);
3971 			}
3972 			mp1->b_wptr = mp1->b_rptr + len;
3973 			mp2->b_rptr = mp1->b_wptr;
3974 			/*
3975 			 * after adjustments if mblk not consumed is now
3976 			 * unaligned, try to align it. If this fails free
3977 			 * all messages and let upper layer recover.
3978 			 */
3979 			if (!OK_32PTR(mp2->b_rptr)) {
3980 				if (!pullupmsg(mp2, -1)) {
3981 					freemsg(mp0);
3982 					freemsg(mp2);
3983 					*mpp = NULL;
3984 					return (NULL);
3985 				}
3986 			}
3987 			*mpp = mp2;
3988 			return (mp0);
3989 		}
3990 		/* Decrement len by the amount we just got. */
3991 		len -= mp2->b_wptr - mp2->b_rptr;
3992 	}
3993 	/*
3994 	 * len should be reduced to zero now.  If not our caller has
3995 	 * screwed up.
3996 	 */
3997 	if (len) {
3998 		/* Shouldn't happen! */
3999 		freemsg(mp0);
4000 		*mpp = NULL;
4001 		return (NULL);
4002 	}
4003 	/*
4004 	 * We consumed up to exactly the end of an mblk.  Detach the part
4005 	 * we are returning from the rest of the chain.
4006 	 */
4007 	mp1->b_cont = NULL;
4008 	*mpp = mp2;
4009 	return (mp0);
4010 }
4011 
4012 /* The ill stream is being unplumbed. Called from ip_close */
4013 int
ip_modclose(ill_t * ill)4014 ip_modclose(ill_t *ill)
4015 {
4016 	boolean_t success;
4017 	ipsq_t	*ipsq;
4018 	ipif_t	*ipif;
4019 	queue_t	*q = ill->ill_rq;
4020 	ip_stack_t	*ipst = ill->ill_ipst;
4021 	int	i;
4022 	arl_ill_common_t *ai = ill->ill_common;
4023 
4024 	/*
4025 	 * The punlink prior to this may have initiated a capability
4026 	 * negotiation. But ipsq_enter will block until that finishes or
4027 	 * times out.
4028 	 */
4029 	success = ipsq_enter(ill, B_FALSE, NEW_OP);
4030 
4031 	/*
4032 	 * Open/close/push/pop is guaranteed to be single threaded
4033 	 * per stream by STREAMS. FS guarantees that all references
4034 	 * from top are gone before close is called. So there can't
4035 	 * be another close thread that has set CONDEMNED on this ill.
4036 	 * and cause ipsq_enter to return failure.
4037 	 */
4038 	ASSERT(success);
4039 	ipsq = ill->ill_phyint->phyint_ipsq;
4040 
4041 	/*
4042 	 * Mark it condemned. No new reference will be made to this ill.
4043 	 * Lookup functions will return an error. Threads that try to
4044 	 * increment the refcnt must check for ILL_CAN_LOOKUP. This ensures
4045 	 * that the refcnt will drop down to zero.
4046 	 */
4047 	mutex_enter(&ill->ill_lock);
4048 	ill->ill_state_flags |= ILL_CONDEMNED;
4049 	for (ipif = ill->ill_ipif; ipif != NULL;
4050 	    ipif = ipif->ipif_next) {
4051 		ipif->ipif_state_flags |= IPIF_CONDEMNED;
4052 	}
4053 	/*
4054 	 * Wake up anybody waiting to enter the ipsq. ipsq_enter
4055 	 * returns  error if ILL_CONDEMNED is set
4056 	 */
4057 	cv_broadcast(&ill->ill_cv);
4058 	mutex_exit(&ill->ill_lock);
4059 
4060 	/*
4061 	 * Send all the deferred DLPI messages downstream which came in
4062 	 * during the small window right before ipsq_enter(). We do this
4063 	 * without waiting for the ACKs because all the ACKs for M_PROTO
4064 	 * messages are ignored in ip_rput() when ILL_CONDEMNED is set.
4065 	 */
4066 	ill_dlpi_send_deferred(ill);
4067 
4068 	/*
4069 	 * Shut down fragmentation reassembly.
4070 	 * ill_frag_timer won't start a timer again.
4071 	 * Now cancel any existing timer
4072 	 */
4073 	(void) untimeout(ill->ill_frag_timer_id);
4074 	(void) ill_frag_timeout(ill, 0);
4075 
4076 	/*
4077 	 * Call ill_delete to bring down the ipifs, ilms and ill on
4078 	 * this ill. Then wait for the refcnts to drop to zero.
4079 	 * ill_is_freeable checks whether the ill is really quiescent.
4080 	 * Then make sure that threads that are waiting to enter the
4081 	 * ipsq have seen the error returned by ipsq_enter and have
4082 	 * gone away. Then we call ill_delete_tail which does the
4083 	 * DL_UNBIND_REQ with the driver and then qprocsoff.
4084 	 */
4085 	ill_delete(ill);
4086 	mutex_enter(&ill->ill_lock);
4087 	while (!ill_is_freeable(ill))
4088 		cv_wait(&ill->ill_cv, &ill->ill_lock);
4089 
4090 	while (ill->ill_waiters)
4091 		cv_wait(&ill->ill_cv, &ill->ill_lock);
4092 
4093 	mutex_exit(&ill->ill_lock);
4094 
4095 	/*
4096 	 * ill_delete_tail drops reference on ill_ipst, but we need to keep
4097 	 * it held until the end of the function since the cleanup
4098 	 * below needs to be able to use the ip_stack_t.
4099 	 */
4100 	netstack_hold(ipst->ips_netstack);
4101 
4102 	/* qprocsoff is done via ill_delete_tail */
4103 	ill_delete_tail(ill);
4104 	/*
4105 	 * synchronously wait for arp stream to unbind. After this, we
4106 	 * cannot get any data packets up from the driver.
4107 	 */
4108 	arp_unbind_complete(ill);
4109 	ASSERT(ill->ill_ipst == NULL);
4110 
4111 	/*
4112 	 * Walk through all conns and qenable those that have queued data.
4113 	 * Close synchronization needs this to
4114 	 * be done to ensure that all upper layers blocked
4115 	 * due to flow control to the closing device
4116 	 * get unblocked.
4117 	 */
4118 	ip1dbg(("ip_wsrv: walking\n"));
4119 	for (i = 0; i < TX_FANOUT_SIZE; i++) {
4120 		conn_walk_drain(ipst, &ipst->ips_idl_tx_list[i]);
4121 	}
4122 
4123 	/*
4124 	 * ai can be null if this is an IPv6 ill, or if the IPv4
4125 	 * stream is being torn down before ARP was plumbed (e.g.,
4126 	 * /sbin/ifconfig plumbing a stream twice, and encountering
4127 	 * an error
4128 	 */
4129 	if (ai != NULL) {
4130 		ASSERT(!ill->ill_isv6);
4131 		mutex_enter(&ai->ai_lock);
4132 		ai->ai_ill = NULL;
4133 		if (ai->ai_arl == NULL) {
4134 			mutex_destroy(&ai->ai_lock);
4135 			kmem_free(ai, sizeof (*ai));
4136 		} else {
4137 			cv_signal(&ai->ai_ill_unplumb_done);
4138 			mutex_exit(&ai->ai_lock);
4139 		}
4140 	}
4141 
4142 	mutex_enter(&ipst->ips_ip_mi_lock);
4143 	mi_close_unlink(&ipst->ips_ip_g_head, (IDP)ill);
4144 	mutex_exit(&ipst->ips_ip_mi_lock);
4145 
4146 	/*
4147 	 * credp could be null if the open didn't succeed and ip_modopen
4148 	 * itself calls ip_close.
4149 	 */
4150 	if (ill->ill_credp != NULL)
4151 		crfree(ill->ill_credp);
4152 
4153 	mutex_destroy(&ill->ill_saved_ire_lock);
4154 	mutex_destroy(&ill->ill_lock);
4155 	rw_destroy(&ill->ill_mcast_lock);
4156 	mutex_destroy(&ill->ill_mcast_serializer);
4157 	list_destroy(&ill->ill_nce);
4158 
4159 	/*
4160 	 * Now we are done with the module close pieces that
4161 	 * need the netstack_t.
4162 	 */
4163 	netstack_rele(ipst->ips_netstack);
4164 
4165 	mi_close_free((IDP)ill);
4166 	q->q_ptr = WR(q)->q_ptr = NULL;
4167 
4168 	ipsq_exit(ipsq);
4169 
4170 	return (0);
4171 }
4172 
4173 /*
4174  * This is called as part of close() for IP, UDP, ICMP, and RTS
4175  * in order to quiesce the conn.
4176  */
4177 void
ip_quiesce_conn(conn_t * connp)4178 ip_quiesce_conn(conn_t *connp)
4179 {
4180 	boolean_t	drain_cleanup_reqd = B_FALSE;
4181 	boolean_t	conn_ioctl_cleanup_reqd = B_FALSE;
4182 	boolean_t	ilg_cleanup_reqd = B_FALSE;
4183 	ip_stack_t	*ipst;
4184 
4185 	ASSERT(!IPCL_IS_TCP(connp));
4186 	ipst = connp->conn_netstack->netstack_ip;
4187 
4188 	/*
4189 	 * Mark the conn as closing, and this conn must not be
4190 	 * inserted in future into any list. Eg. conn_drain_insert(),
4191 	 * won't insert this conn into the conn_drain_list.
4192 	 *
4193 	 * conn_idl, and conn_ilg cannot get set henceforth.
4194 	 */
4195 	mutex_enter(&connp->conn_lock);
4196 	ASSERT(!(connp->conn_state_flags & CONN_QUIESCED));
4197 	connp->conn_state_flags |= CONN_CLOSING;
4198 	if (connp->conn_idl != NULL)
4199 		drain_cleanup_reqd = B_TRUE;
4200 	if (connp->conn_oper_pending_ill != NULL)
4201 		conn_ioctl_cleanup_reqd = B_TRUE;
4202 	if (connp->conn_dhcpinit_ill != NULL) {
4203 		ASSERT(connp->conn_dhcpinit_ill->ill_dhcpinit != 0);
4204 		atomic_dec_32(&connp->conn_dhcpinit_ill->ill_dhcpinit);
4205 		ill_set_inputfn(connp->conn_dhcpinit_ill);
4206 		connp->conn_dhcpinit_ill = NULL;
4207 	}
4208 	if (connp->conn_ilg != NULL)
4209 		ilg_cleanup_reqd = B_TRUE;
4210 	mutex_exit(&connp->conn_lock);
4211 
4212 	if (conn_ioctl_cleanup_reqd)
4213 		conn_ioctl_cleanup(connp);
4214 
4215 	if (is_system_labeled() && connp->conn_anon_port) {
4216 		(void) tsol_mlp_anon(crgetzone(connp->conn_cred),
4217 		    connp->conn_mlp_type, connp->conn_proto,
4218 		    ntohs(connp->conn_lport), B_FALSE);
4219 		connp->conn_anon_port = 0;
4220 	}
4221 	connp->conn_mlp_type = mlptSingle;
4222 
4223 	/*
4224 	 * Remove this conn from any fanout list it is on.
4225 	 * and then wait for any threads currently operating
4226 	 * on this endpoint to finish
4227 	 */
4228 	ipcl_hash_remove(connp);
4229 
4230 	/*
4231 	 * Remove this conn from the drain list, and do any other cleanup that
4232 	 * may be required.  (TCP conns are never flow controlled, and
4233 	 * conn_idl will be NULL.)
4234 	 */
4235 	if (drain_cleanup_reqd && connp->conn_idl != NULL) {
4236 		idl_t *idl = connp->conn_idl;
4237 
4238 		mutex_enter(&idl->idl_lock);
4239 		conn_drain(connp, B_TRUE);
4240 		mutex_exit(&idl->idl_lock);
4241 	}
4242 
4243 	if (connp == ipst->ips_ip_g_mrouter)
4244 		(void) ip_mrouter_done(ipst);
4245 
4246 	if (ilg_cleanup_reqd)
4247 		ilg_delete_all(connp);
4248 
4249 	/*
4250 	 * Now conn refcnt can increase only thru CONN_INC_REF_LOCKED.
4251 	 * callers from write side can't be there now because close
4252 	 * is in progress. The only other caller is ipcl_walk
4253 	 * which checks for the condemned flag.
4254 	 */
4255 	mutex_enter(&connp->conn_lock);
4256 	connp->conn_state_flags |= CONN_CONDEMNED;
4257 	while (connp->conn_ref != 1)
4258 		cv_wait(&connp->conn_cv, &connp->conn_lock);
4259 	connp->conn_state_flags |= CONN_QUIESCED;
4260 	mutex_exit(&connp->conn_lock);
4261 }
4262 
4263 /* ARGSUSED */
4264 int
ip_close(queue_t * q,int flags,cred_t * credp __unused)4265 ip_close(queue_t *q, int flags, cred_t *credp __unused)
4266 {
4267 	conn_t		*connp;
4268 
4269 	/*
4270 	 * Call the appropriate delete routine depending on whether this is
4271 	 * a module or device.
4272 	 */
4273 	if (WR(q)->q_next != NULL) {
4274 		/* This is a module close */
4275 		return (ip_modclose((ill_t *)q->q_ptr));
4276 	}
4277 
4278 	connp = q->q_ptr;
4279 	ip_quiesce_conn(connp);
4280 
4281 	qprocsoff(q);
4282 
4283 	/*
4284 	 * Now we are truly single threaded on this stream, and can
4285 	 * delete the things hanging off the connp, and finally the connp.
4286 	 * We removed this connp from the fanout list, it cannot be
4287 	 * accessed thru the fanouts, and we already waited for the
4288 	 * conn_ref to drop to 0. We are already in close, so
4289 	 * there cannot be any other thread from the top. qprocsoff
4290 	 * has completed, and service has completed or won't run in
4291 	 * future.
4292 	 */
4293 	ASSERT(connp->conn_ref == 1);
4294 
4295 	inet_minor_free(connp->conn_minor_arena, connp->conn_dev);
4296 
4297 	connp->conn_ref--;
4298 	ipcl_conn_destroy(connp);
4299 
4300 	q->q_ptr = WR(q)->q_ptr = NULL;
4301 	return (0);
4302 }
4303 
4304 /*
4305  * Wapper around putnext() so that ip_rts_request can merely use
4306  * conn_recv.
4307  */
4308 /*ARGSUSED2*/
4309 static void
ip_conn_input(void * arg1,mblk_t * mp,void * arg2,ip_recv_attr_t * ira)4310 ip_conn_input(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
4311 {
4312 	conn_t *connp = (conn_t *)arg1;
4313 
4314 	putnext(connp->conn_rq, mp);
4315 }
4316 
4317 /* Dummy in case ICMP error delivery is attempted to a /dev/ip instance */
4318 /* ARGSUSED */
4319 static void
ip_conn_input_icmp(void * arg1,mblk_t * mp,void * arg2,ip_recv_attr_t * ira)4320 ip_conn_input_icmp(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
4321 {
4322 	freemsg(mp);
4323 }
4324 
4325 /*
4326  * Called when the module is about to be unloaded
4327  */
4328 void
ip_ddi_destroy(void)4329 ip_ddi_destroy(void)
4330 {
4331 	/* This needs to be called before destroying any transports. */
4332 	mutex_enter(&cpu_lock);
4333 	unregister_cpu_setup_func(ip_tp_cpu_update, NULL);
4334 	mutex_exit(&cpu_lock);
4335 
4336 	tnet_fini();
4337 
4338 	icmp_ddi_g_destroy();
4339 	rts_ddi_g_destroy();
4340 	udp_ddi_g_destroy();
4341 	sctp_ddi_g_destroy();
4342 	tcp_ddi_g_destroy();
4343 	ilb_ddi_g_destroy();
4344 	dce_g_destroy();
4345 	ipsec_policy_g_destroy();
4346 	ipcl_g_destroy();
4347 	ip_net_g_destroy();
4348 	ip_ire_g_fini();
4349 	inet_minor_destroy(ip_minor_arena_sa);
4350 #if defined(_LP64)
4351 	inet_minor_destroy(ip_minor_arena_la);
4352 #endif
4353 
4354 #ifdef DEBUG
4355 	list_destroy(&ip_thread_list);
4356 	rw_destroy(&ip_thread_rwlock);
4357 	tsd_destroy(&ip_thread_data);
4358 #endif
4359 
4360 	netstack_unregister(NS_IP);
4361 }
4362 
4363 /*
4364  * First step in cleanup.
4365  */
4366 /* ARGSUSED */
4367 static void
ip_stack_shutdown(netstackid_t stackid,void * arg)4368 ip_stack_shutdown(netstackid_t stackid, void *arg)
4369 {
4370 	ip_stack_t *ipst = (ip_stack_t *)arg;
4371 	kt_did_t ktid;
4372 
4373 #ifdef NS_DEBUG
4374 	printf("ip_stack_shutdown(%p, stack %d)\n", (void *)ipst, stackid);
4375 #endif
4376 
4377 	/*
4378 	 * Perform cleanup for special interfaces (loopback and IPMP).
4379 	 */
4380 	ip_interface_cleanup(ipst);
4381 
4382 	/*
4383 	 * The *_hook_shutdown()s start the process of notifying any
4384 	 * consumers that things are going away.... nothing is destroyed.
4385 	 */
4386 	ipv4_hook_shutdown(ipst);
4387 	ipv6_hook_shutdown(ipst);
4388 	arp_hook_shutdown(ipst);
4389 
4390 	mutex_enter(&ipst->ips_capab_taskq_lock);
4391 	ktid = ipst->ips_capab_taskq_thread->t_did;
4392 	ipst->ips_capab_taskq_quit = B_TRUE;
4393 	cv_signal(&ipst->ips_capab_taskq_cv);
4394 	mutex_exit(&ipst->ips_capab_taskq_lock);
4395 
4396 	/*
4397 	 * In rare occurrences, particularly on virtual hardware where CPUs can
4398 	 * be de-scheduled, the thread that we just signaled will not run until
4399 	 * after we have gotten through parts of ip_stack_fini. If that happens
4400 	 * then we'll try to grab the ips_capab_taskq_lock as part of returning
4401 	 * from cv_wait which no longer exists.
4402 	 */
4403 	thread_join(ktid);
4404 }
4405 
4406 /*
4407  * Free the IP stack instance.
4408  */
4409 static void
ip_stack_fini(netstackid_t stackid,void * arg)4410 ip_stack_fini(netstackid_t stackid, void *arg)
4411 {
4412 	ip_stack_t *ipst = (ip_stack_t *)arg;
4413 	int ret;
4414 
4415 #ifdef NS_DEBUG
4416 	printf("ip_stack_fini(%p, stack %d)\n", (void *)ipst, stackid);
4417 #endif
4418 	/*
4419 	 * At this point, all of the notifications that the events and
4420 	 * protocols are going away have been run, meaning that we can
4421 	 * now set about starting to clean things up.
4422 	 */
4423 	ipobs_fini(ipst);
4424 	ipv4_hook_destroy(ipst);
4425 	ipv6_hook_destroy(ipst);
4426 	arp_hook_destroy(ipst);
4427 	ip_net_destroy(ipst);
4428 
4429 	ipmp_destroy(ipst);
4430 
4431 	ip_kstat_fini(stackid, ipst->ips_ip_mibkp);
4432 	ipst->ips_ip_mibkp = NULL;
4433 	icmp_kstat_fini(stackid, ipst->ips_icmp_mibkp);
4434 	ipst->ips_icmp_mibkp = NULL;
4435 	ip_kstat2_fini(stackid, ipst->ips_ip_kstat);
4436 	ipst->ips_ip_kstat = NULL;
4437 	bzero(&ipst->ips_ip_statistics, sizeof (ipst->ips_ip_statistics));
4438 	ip6_kstat_fini(stackid, ipst->ips_ip6_kstat);
4439 	ipst->ips_ip6_kstat = NULL;
4440 	bzero(&ipst->ips_ip6_statistics, sizeof (ipst->ips_ip6_statistics));
4441 
4442 	kmem_free(ipst->ips_propinfo_tbl,
4443 	    ip_propinfo_count * sizeof (mod_prop_info_t));
4444 	ipst->ips_propinfo_tbl = NULL;
4445 
4446 	dce_stack_destroy(ipst);
4447 	ip_mrouter_stack_destroy(ipst);
4448 
4449 	/*
4450 	 * Quiesce all of our timers. Note we set the quiesce flags before we
4451 	 * call untimeout. The slowtimers may actually kick off another instance
4452 	 * of the non-slow timers.
4453 	 */
4454 	mutex_enter(&ipst->ips_igmp_timer_lock);
4455 	ipst->ips_igmp_timer_quiesce = B_TRUE;
4456 	mutex_exit(&ipst->ips_igmp_timer_lock);
4457 
4458 	mutex_enter(&ipst->ips_mld_timer_lock);
4459 	ipst->ips_mld_timer_quiesce = B_TRUE;
4460 	mutex_exit(&ipst->ips_mld_timer_lock);
4461 
4462 	mutex_enter(&ipst->ips_igmp_slowtimeout_lock);
4463 	ipst->ips_igmp_slowtimeout_quiesce = B_TRUE;
4464 	mutex_exit(&ipst->ips_igmp_slowtimeout_lock);
4465 
4466 	mutex_enter(&ipst->ips_mld_slowtimeout_lock);
4467 	ipst->ips_mld_slowtimeout_quiesce = B_TRUE;
4468 	mutex_exit(&ipst->ips_mld_slowtimeout_lock);
4469 
4470 	ret = untimeout(ipst->ips_igmp_timeout_id);
4471 	if (ret == -1) {
4472 		ASSERT(ipst->ips_igmp_timeout_id == 0);
4473 	} else {
4474 		ASSERT(ipst->ips_igmp_timeout_id != 0);
4475 		ipst->ips_igmp_timeout_id = 0;
4476 	}
4477 	ret = untimeout(ipst->ips_igmp_slowtimeout_id);
4478 	if (ret == -1) {
4479 		ASSERT(ipst->ips_igmp_slowtimeout_id == 0);
4480 	} else {
4481 		ASSERT(ipst->ips_igmp_slowtimeout_id != 0);
4482 		ipst->ips_igmp_slowtimeout_id = 0;
4483 	}
4484 	ret = untimeout(ipst->ips_mld_timeout_id);
4485 	if (ret == -1) {
4486 		ASSERT(ipst->ips_mld_timeout_id == 0);
4487 	} else {
4488 		ASSERT(ipst->ips_mld_timeout_id != 0);
4489 		ipst->ips_mld_timeout_id = 0;
4490 	}
4491 	ret = untimeout(ipst->ips_mld_slowtimeout_id);
4492 	if (ret == -1) {
4493 		ASSERT(ipst->ips_mld_slowtimeout_id == 0);
4494 	} else {
4495 		ASSERT(ipst->ips_mld_slowtimeout_id != 0);
4496 		ipst->ips_mld_slowtimeout_id = 0;
4497 	}
4498 
4499 	ip_ire_fini(ipst);
4500 	ip6_asp_free(ipst);
4501 	conn_drain_fini(ipst);
4502 	ipcl_destroy(ipst);
4503 
4504 	mutex_destroy(&ipst->ips_ndp4->ndp_g_lock);
4505 	mutex_destroy(&ipst->ips_ndp6->ndp_g_lock);
4506 	kmem_free(ipst->ips_ndp4, sizeof (ndp_g_t));
4507 	ipst->ips_ndp4 = NULL;
4508 	kmem_free(ipst->ips_ndp6, sizeof (ndp_g_t));
4509 	ipst->ips_ndp6 = NULL;
4510 
4511 	if (ipst->ips_loopback_ksp != NULL) {
4512 		kstat_delete_netstack(ipst->ips_loopback_ksp, stackid);
4513 		ipst->ips_loopback_ksp = NULL;
4514 	}
4515 
4516 	mutex_destroy(&ipst->ips_capab_taskq_lock);
4517 	cv_destroy(&ipst->ips_capab_taskq_cv);
4518 
4519 	rw_destroy(&ipst->ips_srcid_lock);
4520 
4521 	mutex_destroy(&ipst->ips_ip_mi_lock);
4522 	rw_destroy(&ipst->ips_ill_g_usesrc_lock);
4523 
4524 	mutex_destroy(&ipst->ips_igmp_timer_lock);
4525 	mutex_destroy(&ipst->ips_mld_timer_lock);
4526 	mutex_destroy(&ipst->ips_igmp_slowtimeout_lock);
4527 	mutex_destroy(&ipst->ips_mld_slowtimeout_lock);
4528 	mutex_destroy(&ipst->ips_ip_addr_avail_lock);
4529 	rw_destroy(&ipst->ips_ill_g_lock);
4530 
4531 	kmem_free(ipst->ips_phyint_g_list, sizeof (phyint_list_t));
4532 	ipst->ips_phyint_g_list = NULL;
4533 	kmem_free(ipst->ips_ill_g_heads, sizeof (ill_g_head_t) * MAX_G_HEADS);
4534 	ipst->ips_ill_g_heads = NULL;
4535 
4536 	ldi_ident_release(ipst->ips_ldi_ident);
4537 	kmem_free(ipst, sizeof (*ipst));
4538 }
4539 
4540 /*
4541  * This function is called from the TSD destructor, and is used to debug
4542  * reference count issues in IP. See block comment in <inet/ip_if.h> for
4543  * details.
4544  */
4545 static void
ip_thread_exit(void * phash)4546 ip_thread_exit(void *phash)
4547 {
4548 	th_hash_t *thh = phash;
4549 
4550 	rw_enter(&ip_thread_rwlock, RW_WRITER);
4551 	list_remove(&ip_thread_list, thh);
4552 	rw_exit(&ip_thread_rwlock);
4553 	mod_hash_destroy_hash(thh->thh_hash);
4554 	kmem_free(thh, sizeof (*thh));
4555 }
4556 
4557 /*
4558  * Called when the IP kernel module is loaded into the kernel
4559  */
4560 void
ip_ddi_init(void)4561 ip_ddi_init(void)
4562 {
4563 	ip_squeue_flag = ip_squeue_switch(ip_squeue_enter);
4564 
4565 	/*
4566 	 * For IP and TCP the minor numbers should start from 2 since we have 4
4567 	 * initial devices: ip, ip6, tcp, tcp6.
4568 	 */
4569 	/*
4570 	 * If this is a 64-bit kernel, then create two separate arenas -
4571 	 * one for TLIs in the range of INET_MIN_DEV+2 through 2^^18-1, and the
4572 	 * other for socket apps in the range 2^^18 through 2^^32-1.
4573 	 */
4574 	ip_minor_arena_la = NULL;
4575 	ip_minor_arena_sa = NULL;
4576 #if defined(_LP64)
4577 	if ((ip_minor_arena_sa = inet_minor_create("ip_minor_arena_sa",
4578 	    INET_MIN_DEV + 2, MAXMIN32, KM_SLEEP)) == NULL) {
4579 		cmn_err(CE_PANIC,
4580 		    "ip_ddi_init: ip_minor_arena_sa creation failed\n");
4581 	}
4582 	if ((ip_minor_arena_la = inet_minor_create("ip_minor_arena_la",
4583 	    MAXMIN32 + 1, MAXMIN64, KM_SLEEP)) == NULL) {
4584 		cmn_err(CE_PANIC,
4585 		    "ip_ddi_init: ip_minor_arena_la creation failed\n");
4586 	}
4587 #else
4588 	if ((ip_minor_arena_sa = inet_minor_create("ip_minor_arena_sa",
4589 	    INET_MIN_DEV + 2, MAXMIN, KM_SLEEP)) == NULL) {
4590 		cmn_err(CE_PANIC,
4591 		    "ip_ddi_init: ip_minor_arena_sa creation failed\n");
4592 	}
4593 #endif
4594 	ip_poll_normal_ticks = MSEC_TO_TICK_ROUNDUP(ip_poll_normal_ms);
4595 
4596 	ipcl_g_init();
4597 	ip_ire_g_init();
4598 	ip_net_g_init();
4599 
4600 #ifdef DEBUG
4601 	tsd_create(&ip_thread_data, ip_thread_exit);
4602 	rw_init(&ip_thread_rwlock, NULL, RW_DEFAULT, NULL);
4603 	list_create(&ip_thread_list, sizeof (th_hash_t),
4604 	    offsetof(th_hash_t, thh_link));
4605 #endif
4606 	ipsec_policy_g_init();
4607 	tcp_ddi_g_init();
4608 	sctp_ddi_g_init();
4609 	dce_g_init();
4610 
4611 	/*
4612 	 * We want to be informed each time a stack is created or
4613 	 * destroyed in the kernel, so we can maintain the
4614 	 * set of udp_stack_t's.
4615 	 */
4616 	netstack_register(NS_IP, ip_stack_init, ip_stack_shutdown,
4617 	    ip_stack_fini);
4618 
4619 	tnet_init();
4620 
4621 	udp_ddi_g_init();
4622 	rts_ddi_g_init();
4623 	icmp_ddi_g_init();
4624 	ilb_ddi_g_init();
4625 
4626 	/* This needs to be called after all transports are initialized. */
4627 	mutex_enter(&cpu_lock);
4628 	register_cpu_setup_func(ip_tp_cpu_update, NULL);
4629 	mutex_exit(&cpu_lock);
4630 }
4631 
4632 /*
4633  * Initialize the IP stack instance.
4634  */
4635 static void *
ip_stack_init(netstackid_t stackid,netstack_t * ns)4636 ip_stack_init(netstackid_t stackid, netstack_t *ns)
4637 {
4638 	ip_stack_t	*ipst;
4639 	size_t		arrsz;
4640 	major_t		major;
4641 
4642 #ifdef NS_DEBUG
4643 	printf("ip_stack_init(stack %d)\n", stackid);
4644 #endif
4645 
4646 	ipst = (ip_stack_t *)kmem_zalloc(sizeof (*ipst), KM_SLEEP);
4647 	ipst->ips_netstack = ns;
4648 
4649 	ipst->ips_ill_g_heads = kmem_zalloc(sizeof (ill_g_head_t) * MAX_G_HEADS,
4650 	    KM_SLEEP);
4651 	ipst->ips_phyint_g_list = kmem_zalloc(sizeof (phyint_list_t),
4652 	    KM_SLEEP);
4653 	ipst->ips_ndp4 = kmem_zalloc(sizeof (ndp_g_t), KM_SLEEP);
4654 	ipst->ips_ndp6 = kmem_zalloc(sizeof (ndp_g_t), KM_SLEEP);
4655 	mutex_init(&ipst->ips_ndp4->ndp_g_lock, NULL, MUTEX_DEFAULT, NULL);
4656 	mutex_init(&ipst->ips_ndp6->ndp_g_lock, NULL, MUTEX_DEFAULT, NULL);
4657 
4658 	mutex_init(&ipst->ips_igmp_timer_lock, NULL, MUTEX_DEFAULT, NULL);
4659 	ipst->ips_igmp_deferred_next = INFINITY;
4660 	mutex_init(&ipst->ips_mld_timer_lock, NULL, MUTEX_DEFAULT, NULL);
4661 	ipst->ips_mld_deferred_next = INFINITY;
4662 	mutex_init(&ipst->ips_igmp_slowtimeout_lock, NULL, MUTEX_DEFAULT, NULL);
4663 	mutex_init(&ipst->ips_mld_slowtimeout_lock, NULL, MUTEX_DEFAULT, NULL);
4664 	mutex_init(&ipst->ips_ip_mi_lock, NULL, MUTEX_DEFAULT, NULL);
4665 	mutex_init(&ipst->ips_ip_addr_avail_lock, NULL, MUTEX_DEFAULT, NULL);
4666 	rw_init(&ipst->ips_ill_g_lock, NULL, RW_DEFAULT, NULL);
4667 	rw_init(&ipst->ips_ill_g_usesrc_lock, NULL, RW_DEFAULT, NULL);
4668 
4669 	ipcl_init(ipst);
4670 	ip_ire_init(ipst);
4671 	ip6_asp_init(ipst);
4672 	ipif_init(ipst);
4673 	conn_drain_init(ipst);
4674 	ip_mrouter_stack_init(ipst);
4675 	dce_stack_init(ipst);
4676 
4677 	ipst->ips_ip_multirt_log_interval = 1000;
4678 
4679 	ipst->ips_ill_index = 1;
4680 
4681 	ipst->ips_saved_ip_forwarding = -1;
4682 	ipst->ips_reg_vif_num = ALL_VIFS;	/* Index to Register vif */
4683 
4684 	arrsz = ip_propinfo_count * sizeof (mod_prop_info_t);
4685 	ipst->ips_propinfo_tbl = (mod_prop_info_t *)kmem_alloc(arrsz, KM_SLEEP);
4686 	bcopy(ip_propinfo_tbl, ipst->ips_propinfo_tbl, arrsz);
4687 
4688 	ipst->ips_ip_mibkp = ip_kstat_init(stackid, ipst);
4689 	ipst->ips_icmp_mibkp = icmp_kstat_init(stackid);
4690 	ipst->ips_ip_kstat = ip_kstat2_init(stackid, &ipst->ips_ip_statistics);
4691 	ipst->ips_ip6_kstat =
4692 	    ip6_kstat_init(stackid, &ipst->ips_ip6_statistics);
4693 
4694 	ipst->ips_ip_src_id = 1;
4695 	rw_init(&ipst->ips_srcid_lock, NULL, RW_DEFAULT, NULL);
4696 
4697 	ipst->ips_src_generation = SRC_GENERATION_INITIAL;
4698 
4699 	ip_net_init(ipst, ns);
4700 	ipv4_hook_init(ipst);
4701 	ipv6_hook_init(ipst);
4702 	arp_hook_init(ipst);
4703 	ipmp_init(ipst);
4704 	ipobs_init(ipst);
4705 
4706 	/*
4707 	 * Create the taskq dispatcher thread and initialize related stuff.
4708 	 */
4709 	mutex_init(&ipst->ips_capab_taskq_lock, NULL, MUTEX_DEFAULT, NULL);
4710 	cv_init(&ipst->ips_capab_taskq_cv, NULL, CV_DEFAULT, NULL);
4711 	ipst->ips_capab_taskq_thread = thread_create(NULL, 0,
4712 	    ill_taskq_dispatch, ipst, 0, &p0, TS_RUN, minclsyspri);
4713 
4714 	major = mod_name_to_major(INET_NAME);
4715 	(void) ldi_ident_from_major(major, &ipst->ips_ldi_ident);
4716 	return (ipst);
4717 }
4718 
4719 /*
4720  * Allocate and initialize a DLPI template of the specified length.  (May be
4721  * called as writer.)
4722  */
4723 mblk_t *
ip_dlpi_alloc(size_t len,t_uscalar_t prim)4724 ip_dlpi_alloc(size_t len, t_uscalar_t prim)
4725 {
4726 	mblk_t	*mp;
4727 
4728 	mp = allocb(len, BPRI_MED);
4729 	if (!mp)
4730 		return (NULL);
4731 
4732 	/*
4733 	 * DLPIv2 says that DL_INFO_REQ and DL_TOKEN_REQ (the latter
4734 	 * of which we don't seem to use) are sent with M_PCPROTO, and
4735 	 * that other DLPI are M_PROTO.
4736 	 */
4737 	if (prim == DL_INFO_REQ) {
4738 		mp->b_datap->db_type = M_PCPROTO;
4739 	} else {
4740 		mp->b_datap->db_type = M_PROTO;
4741 	}
4742 
4743 	mp->b_wptr = mp->b_rptr + len;
4744 	bzero(mp->b_rptr, len);
4745 	((dl_unitdata_req_t *)mp->b_rptr)->dl_primitive = prim;
4746 	return (mp);
4747 }
4748 
4749 /*
4750  * Allocate and initialize a DLPI notification.  (May be called as writer.)
4751  */
4752 mblk_t *
ip_dlnotify_alloc(uint_t notification,uint_t data)4753 ip_dlnotify_alloc(uint_t notification, uint_t data)
4754 {
4755 	dl_notify_ind_t	*notifyp;
4756 	mblk_t		*mp;
4757 
4758 	if ((mp = ip_dlpi_alloc(DL_NOTIFY_IND_SIZE, DL_NOTIFY_IND)) == NULL)
4759 		return (NULL);
4760 
4761 	notifyp = (dl_notify_ind_t *)mp->b_rptr;
4762 	notifyp->dl_notification = notification;
4763 	notifyp->dl_data = data;
4764 	return (mp);
4765 }
4766 
4767 mblk_t *
ip_dlnotify_alloc2(uint_t notification,uint_t data1,uint_t data2)4768 ip_dlnotify_alloc2(uint_t notification, uint_t data1, uint_t data2)
4769 {
4770 	dl_notify_ind_t	*notifyp;
4771 	mblk_t		*mp;
4772 
4773 	if ((mp = ip_dlpi_alloc(DL_NOTIFY_IND_SIZE, DL_NOTIFY_IND)) == NULL)
4774 		return (NULL);
4775 
4776 	notifyp = (dl_notify_ind_t *)mp->b_rptr;
4777 	notifyp->dl_notification = notification;
4778 	notifyp->dl_data1 = data1;
4779 	notifyp->dl_data2 = data2;
4780 	return (mp);
4781 }
4782 
4783 /*
4784  * Debug formatting routine.  Returns a character string representation of the
4785  * addr in buf, of the form xxx.xxx.xxx.xxx.  This routine takes the address
4786  * in the form of a ipaddr_t and calls ip_dot_saddr with a pointer.
4787  *
4788  * Once the ndd table-printing interfaces are removed, this can be changed to
4789  * standard dotted-decimal form.
4790  */
4791 char *
ip_dot_addr(ipaddr_t addr,char * buf)4792 ip_dot_addr(ipaddr_t addr, char *buf)
4793 {
4794 	uint8_t *ap = (uint8_t *)&addr;
4795 
4796 	(void) mi_sprintf(buf, "%03d.%03d.%03d.%03d",
4797 	    ap[0] & 0xFF, ap[1] & 0xFF, ap[2] & 0xFF, ap[3] & 0xFF);
4798 	return (buf);
4799 }
4800 
4801 /*
4802  * Write the given MAC address as a printable string in the usual colon-
4803  * separated format.
4804  */
4805 const char *
mac_colon_addr(const uint8_t * addr,size_t alen,char * buf,size_t buflen)4806 mac_colon_addr(const uint8_t *addr, size_t alen, char *buf, size_t buflen)
4807 {
4808 	char *bp;
4809 
4810 	if (alen == 0 || buflen < 4)
4811 		return ("?");
4812 	bp = buf;
4813 	for (;;) {
4814 		/*
4815 		 * If there are more MAC address bytes available, but we won't
4816 		 * have any room to print them, then add "..." to the string
4817 		 * instead.  See below for the 'magic number' explanation.
4818 		 */
4819 		if ((alen == 2 && buflen < 6) || (alen > 2 && buflen < 7)) {
4820 			(void) strcpy(bp, "...");
4821 			break;
4822 		}
4823 		(void) sprintf(bp, "%02x", *addr++);
4824 		bp += 2;
4825 		if (--alen == 0)
4826 			break;
4827 		*bp++ = ':';
4828 		buflen -= 3;
4829 		/*
4830 		 * At this point, based on the first 'if' statement above,
4831 		 * either alen == 1 and buflen >= 3, or alen > 1 and
4832 		 * buflen >= 4.  The first case leaves room for the final "xx"
4833 		 * number and trailing NUL byte.  The second leaves room for at
4834 		 * least "...".  Thus the apparently 'magic' numbers chosen for
4835 		 * that statement.
4836 		 */
4837 	}
4838 	return (buf);
4839 }
4840 
4841 /*
4842  * Called when it is conceptually a ULP that would sent the packet
4843  * e.g., port unreachable and protocol unreachable. Check that the packet
4844  * would have passed the IPsec global policy before sending the error.
4845  *
4846  * Send an ICMP error after patching up the packet appropriately.
4847  * Uses ip_drop_input and bumps the appropriate MIB.
4848  */
4849 void
ip_fanout_send_icmp_v4(mblk_t * mp,uint_t icmp_type,uint_t icmp_code,ip_recv_attr_t * ira)4850 ip_fanout_send_icmp_v4(mblk_t *mp, uint_t icmp_type, uint_t icmp_code,
4851     ip_recv_attr_t *ira)
4852 {
4853 	ipha_t		*ipha;
4854 	boolean_t	secure;
4855 	ill_t		*ill = ira->ira_ill;
4856 	ip_stack_t	*ipst = ill->ill_ipst;
4857 	netstack_t	*ns = ipst->ips_netstack;
4858 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
4859 
4860 	secure = ira->ira_flags & IRAF_IPSEC_SECURE;
4861 
4862 	/*
4863 	 * We are generating an icmp error for some inbound packet.
4864 	 * Called from all ip_fanout_(udp, tcp, proto) functions.
4865 	 * Before we generate an error, check with global policy
4866 	 * to see whether this is allowed to enter the system. As
4867 	 * there is no "conn", we are checking with global policy.
4868 	 */
4869 	ipha = (ipha_t *)mp->b_rptr;
4870 	if (secure || ipss->ipsec_inbound_v4_policy_present) {
4871 		mp = ipsec_check_global_policy(mp, NULL, ipha, NULL, ira, ns);
4872 		if (mp == NULL)
4873 			return;
4874 	}
4875 
4876 	/* We never send errors for protocols that we do implement */
4877 	if (ira->ira_protocol == IPPROTO_ICMP ||
4878 	    ira->ira_protocol == IPPROTO_IGMP) {
4879 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
4880 		ip_drop_input("ip_fanout_send_icmp_v4", mp, ill);
4881 		freemsg(mp);
4882 		return;
4883 	}
4884 	/*
4885 	 * Have to correct checksum since
4886 	 * the packet might have been
4887 	 * fragmented and the reassembly code in ip_rput
4888 	 * does not restore the IP checksum.
4889 	 */
4890 	ipha->ipha_hdr_checksum = 0;
4891 	ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
4892 
4893 	switch (icmp_type) {
4894 	case ICMP_DEST_UNREACHABLE:
4895 		switch (icmp_code) {
4896 		case ICMP_PROTOCOL_UNREACHABLE:
4897 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInUnknownProtos);
4898 			ip_drop_input("ipIfStatsInUnknownProtos", mp, ill);
4899 			break;
4900 		case ICMP_PORT_UNREACHABLE:
4901 			BUMP_MIB(ill->ill_ip_mib, udpIfStatsNoPorts);
4902 			ip_drop_input("ipIfStatsNoPorts", mp, ill);
4903 			break;
4904 		}
4905 
4906 		icmp_unreachable(mp, icmp_code, ira);
4907 		break;
4908 	default:
4909 #ifdef DEBUG
4910 		panic("ip_fanout_send_icmp_v4: wrong type");
4911 		/*NOTREACHED*/
4912 #else
4913 		freemsg(mp);
4914 		break;
4915 #endif
4916 	}
4917 }
4918 
4919 /*
4920  * Used to send an ICMP error message when a packet is received for
4921  * a protocol that is not supported. The mblk passed as argument
4922  * is consumed by this function.
4923  */
4924 void
ip_proto_not_sup(mblk_t * mp,ip_recv_attr_t * ira)4925 ip_proto_not_sup(mblk_t *mp, ip_recv_attr_t *ira)
4926 {
4927 	ipha_t		*ipha;
4928 
4929 	ipha = (ipha_t *)mp->b_rptr;
4930 	if (ira->ira_flags & IRAF_IS_IPV4) {
4931 		ASSERT(IPH_HDR_VERSION(ipha) == IP_VERSION);
4932 		ip_fanout_send_icmp_v4(mp, ICMP_DEST_UNREACHABLE,
4933 		    ICMP_PROTOCOL_UNREACHABLE, ira);
4934 	} else {
4935 		ASSERT(IPH_HDR_VERSION(ipha) == IPV6_VERSION);
4936 		ip_fanout_send_icmp_v6(mp, ICMP6_PARAM_PROB,
4937 		    ICMP6_PARAMPROB_NEXTHEADER, ira);
4938 	}
4939 }
4940 
4941 /*
4942  * Deliver a rawip packet to the given conn, possibly applying ipsec policy.
4943  * Handles IPv4 and IPv6.
4944  * We are responsible for disposing of mp, such as by freemsg() or putnext()
4945  * Caller is responsible for dropping references to the conn.
4946  */
4947 void
ip_fanout_proto_conn(conn_t * connp,mblk_t * mp,ipha_t * ipha,ip6_t * ip6h,ip_recv_attr_t * ira)4948 ip_fanout_proto_conn(conn_t *connp, mblk_t *mp, ipha_t *ipha, ip6_t *ip6h,
4949     ip_recv_attr_t *ira)
4950 {
4951 	ill_t		*ill = ira->ira_ill;
4952 	ip_stack_t	*ipst = ill->ill_ipst;
4953 	ipsec_stack_t	*ipss = ipst->ips_netstack->netstack_ipsec;
4954 	boolean_t	secure;
4955 	uint_t		protocol = ira->ira_protocol;
4956 	iaflags_t	iraflags = ira->ira_flags;
4957 	queue_t		*rq;
4958 
4959 	secure = iraflags & IRAF_IPSEC_SECURE;
4960 
4961 	rq = connp->conn_rq;
4962 	if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld : !canputnext(rq)) {
4963 		switch (protocol) {
4964 		case IPPROTO_ICMPV6:
4965 			BUMP_MIB(ill->ill_icmp6_mib, ipv6IfIcmpInOverflows);
4966 			break;
4967 		case IPPROTO_ICMP:
4968 			BUMP_MIB(&ipst->ips_icmp_mib, icmpInOverflows);
4969 			break;
4970 		default:
4971 			BUMP_MIB(ill->ill_ip_mib, rawipIfStatsInOverflows);
4972 			break;
4973 		}
4974 		freemsg(mp);
4975 		return;
4976 	}
4977 
4978 	ASSERT(!(IPCL_IS_IPTUN(connp)));
4979 
4980 	if (((iraflags & IRAF_IS_IPV4) ?
4981 	    CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
4982 	    CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
4983 	    secure) {
4984 		mp = ipsec_check_inbound_policy(mp, connp, ipha,
4985 		    ip6h, ira);
4986 		if (mp == NULL) {
4987 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
4988 			/* Note that mp is NULL */
4989 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
4990 			return;
4991 		}
4992 	}
4993 
4994 	if (iraflags & IRAF_ICMP_ERROR) {
4995 		(connp->conn_recvicmp)(connp, mp, NULL, ira);
4996 	} else {
4997 		ill_t *rill = ira->ira_rill;
4998 
4999 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
5000 		ira->ira_ill = ira->ira_rill = NULL;
5001 		/* Send it upstream */
5002 		(connp->conn_recv)(connp, mp, NULL, ira);
5003 		ira->ira_ill = ill;
5004 		ira->ira_rill = rill;
5005 	}
5006 }
5007 
5008 /*
5009  * Handle protocols with which IP is less intimate.  There
5010  * can be more than one stream bound to a particular
5011  * protocol.  When this is the case, normally each one gets a copy
5012  * of any incoming packets.
5013  *
5014  * IPsec NOTE :
5015  *
5016  * Don't allow a secure packet going up a non-secure connection.
5017  * We don't allow this because
5018  *
5019  * 1) Reply might go out in clear which will be dropped at
5020  *    the sending side.
5021  * 2) If the reply goes out in clear it will give the
5022  *    adversary enough information for getting the key in
5023  *    most of the cases.
5024  *
5025  * Moreover getting a secure packet when we expect clear
5026  * implies that SA's were added without checking for
5027  * policy on both ends. This should not happen once ISAKMP
5028  * is used to negotiate SAs as SAs will be added only after
5029  * verifying the policy.
5030  *
5031  * Zones notes:
5032  * Earlier in ip_input on a system with multiple shared-IP zones we
5033  * duplicate the multicast and broadcast packets and send them up
5034  * with each explicit zoneid that exists on that ill.
5035  * This means that here we can match the zoneid with SO_ALLZONES being special.
5036  */
5037 void
ip_fanout_proto_v4(mblk_t * mp,ipha_t * ipha,ip_recv_attr_t * ira)5038 ip_fanout_proto_v4(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
5039 {
5040 	mblk_t		*mp1;
5041 	ipaddr_t	laddr;
5042 	conn_t		*connp, *first_connp, *next_connp;
5043 	connf_t		*connfp;
5044 	ill_t		*ill = ira->ira_ill;
5045 	ip_stack_t	*ipst = ill->ill_ipst;
5046 
5047 	laddr = ipha->ipha_dst;
5048 
5049 	connfp = &ipst->ips_ipcl_proto_fanout_v4[ira->ira_protocol];
5050 	mutex_enter(&connfp->connf_lock);
5051 	connp = connfp->connf_head;
5052 	for (connp = connfp->connf_head; connp != NULL;
5053 	    connp = connp->conn_next) {
5054 		/* Note: IPCL_PROTO_MATCH includes conn_wantpacket */
5055 		if (IPCL_PROTO_MATCH(connp, ira, ipha) &&
5056 		    (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5057 		    tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp))) {
5058 			break;
5059 		}
5060 	}
5061 
5062 	if (connp == NULL) {
5063 		/*
5064 		 * No one bound to these addresses.  Is
5065 		 * there a client that wants all
5066 		 * unclaimed datagrams?
5067 		 */
5068 		mutex_exit(&connfp->connf_lock);
5069 		ip_fanout_send_icmp_v4(mp, ICMP_DEST_UNREACHABLE,
5070 		    ICMP_PROTOCOL_UNREACHABLE, ira);
5071 		return;
5072 	}
5073 
5074 	ASSERT(IPCL_IS_NONSTR(connp) || connp->conn_rq != NULL);
5075 
5076 	CONN_INC_REF(connp);
5077 	first_connp = connp;
5078 	connp = connp->conn_next;
5079 
5080 	for (;;) {
5081 		while (connp != NULL) {
5082 			/* Note: IPCL_PROTO_MATCH includes conn_wantpacket */
5083 			if (IPCL_PROTO_MATCH(connp, ira, ipha) &&
5084 			    (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5085 			    tsol_receive_local(mp, &laddr, IPV4_VERSION,
5086 			    ira, connp)))
5087 				break;
5088 			connp = connp->conn_next;
5089 		}
5090 
5091 		if (connp == NULL) {
5092 			/* No more interested clients */
5093 			connp = first_connp;
5094 			break;
5095 		}
5096 		if (((mp1 = dupmsg(mp)) == NULL) &&
5097 		    ((mp1 = copymsg(mp)) == NULL)) {
5098 			/* Memory allocation failed */
5099 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5100 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
5101 			connp = first_connp;
5102 			break;
5103 		}
5104 
5105 		CONN_INC_REF(connp);
5106 		mutex_exit(&connfp->connf_lock);
5107 
5108 		ip_fanout_proto_conn(connp, mp1, (ipha_t *)mp1->b_rptr, NULL,
5109 		    ira);
5110 
5111 		mutex_enter(&connfp->connf_lock);
5112 		/* Follow the next pointer before releasing the conn. */
5113 		next_connp = connp->conn_next;
5114 		CONN_DEC_REF(connp);
5115 		connp = next_connp;
5116 	}
5117 
5118 	/* Last one.  Send it upstream. */
5119 	mutex_exit(&connfp->connf_lock);
5120 
5121 	ip_fanout_proto_conn(connp, mp, ipha, NULL, ira);
5122 
5123 	CONN_DEC_REF(connp);
5124 }
5125 
5126 /*
5127  * If we have a IPsec NAT-Traversal packet, strip the zero-SPI or
5128  * pass it along to ESP if the SPI is non-zero.  Returns the mblk if the mblk
5129  * is not consumed.
5130  *
5131  * One of three things can happen, all of which affect the passed-in mblk:
5132  *
5133  * 1.) The packet is stock UDP and gets its zero-SPI stripped.  Return mblk..
5134  *
5135  * 2.) The packet is ESP-in-UDP, gets transformed into an equivalent
5136  *     ESP packet, and is passed along to ESP for consumption.  Return NULL.
5137  *
5138  * 3.) The packet is an ESP-in-UDP Keepalive.  Drop it and return NULL.
5139  */
5140 mblk_t *
zero_spi_check(mblk_t * mp,ip_recv_attr_t * ira)5141 zero_spi_check(mblk_t *mp, ip_recv_attr_t *ira)
5142 {
5143 	int shift, plen, iph_len;
5144 	ipha_t *ipha;
5145 	udpha_t *udpha;
5146 	uint32_t *spi;
5147 	uint32_t esp_ports;
5148 	uint8_t *orptr;
5149 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
5150 	ipsec_stack_t	*ipss = ipst->ips_netstack->netstack_ipsec;
5151 
5152 	ipha = (ipha_t *)mp->b_rptr;
5153 	iph_len = ira->ira_ip_hdr_length;
5154 	plen = ira->ira_pktlen;
5155 
5156 	if (plen - iph_len - sizeof (udpha_t) < sizeof (uint32_t)) {
5157 		/*
5158 		 * Most likely a keepalive for the benefit of an intervening
5159 		 * NAT.  These aren't for us, per se, so drop it.
5160 		 *
5161 		 * RFC 3947/8 doesn't say for sure what to do for 2-3
5162 		 * byte packets (keepalives are 1-byte), but we'll drop them
5163 		 * also.
5164 		 */
5165 		ip_drop_packet(mp, B_TRUE, ira->ira_ill,
5166 		    DROPPER(ipss, ipds_esp_nat_t_ka), &ipss->ipsec_dropper);
5167 		return (NULL);
5168 	}
5169 
5170 	if (MBLKL(mp) < iph_len + sizeof (udpha_t) + sizeof (*spi)) {
5171 		/* might as well pull it all up - it might be ESP. */
5172 		if (!pullupmsg(mp, -1)) {
5173 			ip_drop_packet(mp, B_TRUE, ira->ira_ill,
5174 			    DROPPER(ipss, ipds_esp_nomem),
5175 			    &ipss->ipsec_dropper);
5176 			return (NULL);
5177 		}
5178 
5179 		ipha = (ipha_t *)mp->b_rptr;
5180 	}
5181 	spi = (uint32_t *)(mp->b_rptr + iph_len + sizeof (udpha_t));
5182 	if (*spi == 0) {
5183 		/* UDP packet - remove 0-spi. */
5184 		shift = sizeof (uint32_t);
5185 	} else {
5186 		/* ESP-in-UDP packet - reduce to ESP. */
5187 		ipha->ipha_protocol = IPPROTO_ESP;
5188 		shift = sizeof (udpha_t);
5189 	}
5190 
5191 	/* Fix IP header */
5192 	ira->ira_pktlen = (plen - shift);
5193 	ipha->ipha_length = htons(ira->ira_pktlen);
5194 	ipha->ipha_hdr_checksum = 0;
5195 
5196 	orptr = mp->b_rptr;
5197 	mp->b_rptr += shift;
5198 
5199 	udpha = (udpha_t *)(orptr + iph_len);
5200 	if (*spi == 0) {
5201 		ASSERT((uint8_t *)ipha == orptr);
5202 		udpha->uha_length = htons(plen - shift - iph_len);
5203 		iph_len += sizeof (udpha_t);	/* For the call to ovbcopy(). */
5204 		esp_ports = 0;
5205 	} else {
5206 		esp_ports = *((uint32_t *)udpha);
5207 		ASSERT(esp_ports != 0);
5208 	}
5209 	ovbcopy(orptr, orptr + shift, iph_len);
5210 	if (esp_ports != 0) /* Punt up for ESP processing. */ {
5211 		ipha = (ipha_t *)(orptr + shift);
5212 
5213 		ira->ira_flags |= IRAF_ESP_UDP_PORTS;
5214 		ira->ira_esp_udp_ports = esp_ports;
5215 		ip_fanout_v4(mp, ipha, ira);
5216 		return (NULL);
5217 	}
5218 	return (mp);
5219 }
5220 
5221 /*
5222  * Deliver a udp packet to the given conn, possibly applying ipsec policy.
5223  * Handles IPv4 and IPv6.
5224  * We are responsible for disposing of mp, such as by freemsg() or putnext()
5225  * Caller is responsible for dropping references to the conn.
5226  */
5227 void
ip_fanout_udp_conn(conn_t * connp,mblk_t * mp,ipha_t * ipha,ip6_t * ip6h,ip_recv_attr_t * ira)5228 ip_fanout_udp_conn(conn_t *connp, mblk_t *mp, ipha_t *ipha, ip6_t *ip6h,
5229     ip_recv_attr_t *ira)
5230 {
5231 	ill_t		*ill = ira->ira_ill;
5232 	ip_stack_t	*ipst = ill->ill_ipst;
5233 	ipsec_stack_t	*ipss = ipst->ips_netstack->netstack_ipsec;
5234 	boolean_t	secure;
5235 	iaflags_t	iraflags = ira->ira_flags;
5236 
5237 	secure = iraflags & IRAF_IPSEC_SECURE;
5238 
5239 	if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld :
5240 	    !canputnext(connp->conn_rq)) {
5241 		BUMP_MIB(ill->ill_ip_mib, udpIfStatsInOverflows);
5242 		freemsg(mp);
5243 		return;
5244 	}
5245 
5246 	if (((iraflags & IRAF_IS_IPV4) ?
5247 	    CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
5248 	    CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
5249 	    secure) {
5250 		mp = ipsec_check_inbound_policy(mp, connp, ipha,
5251 		    ip6h, ira);
5252 		if (mp == NULL) {
5253 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5254 			/* Note that mp is NULL */
5255 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
5256 			return;
5257 		}
5258 	}
5259 
5260 	/*
5261 	 * Since this code is not used for UDP unicast we don't need a NAT_T
5262 	 * check. Only ip_fanout_v4 has that check.
5263 	 */
5264 	if (ira->ira_flags & IRAF_ICMP_ERROR) {
5265 		(connp->conn_recvicmp)(connp, mp, NULL, ira);
5266 	} else {
5267 		ill_t *rill = ira->ira_rill;
5268 
5269 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
5270 		ira->ira_ill = ira->ira_rill = NULL;
5271 		/* Send it upstream */
5272 		(connp->conn_recv)(connp, mp, NULL, ira);
5273 		ira->ira_ill = ill;
5274 		ira->ira_rill = rill;
5275 	}
5276 }
5277 
5278 /*
5279  * Fanout for UDP packets that are multicast or broadcast, and ICMP errors.
5280  * (Unicast fanout is handled in ip_input_v4.)
5281  *
5282  * If SO_REUSEADDR is set all multicast and broadcast packets
5283  * will be delivered to all conns bound to the same port.
5284  *
5285  * If there is at least one matching AF_INET receiver, then we will
5286  * ignore any AF_INET6 receivers.
5287  * In the special case where an AF_INET socket binds to 0.0.0.0/<port> and an
5288  * AF_INET6 socket binds to ::/<port>, only the AF_INET socket receives the IPv4
5289  * packets.
5290  *
5291  * Zones notes:
5292  * Earlier in ip_input on a system with multiple shared-IP zones we
5293  * duplicate the multicast and broadcast packets and send them up
5294  * with each explicit zoneid that exists on that ill.
5295  * This means that here we can match the zoneid with SO_ALLZONES being special.
5296  */
5297 void
ip_fanout_udp_multi_v4(mblk_t * mp,ipha_t * ipha,uint16_t lport,uint16_t fport,ip_recv_attr_t * ira)5298 ip_fanout_udp_multi_v4(mblk_t *mp, ipha_t *ipha, uint16_t lport, uint16_t fport,
5299     ip_recv_attr_t *ira)
5300 {
5301 	ipaddr_t	laddr;
5302 	in6_addr_t	v6faddr;
5303 	conn_t		*connp;
5304 	connf_t		*connfp;
5305 	ipaddr_t	faddr;
5306 	ill_t		*ill = ira->ira_ill;
5307 	ip_stack_t	*ipst = ill->ill_ipst;
5308 
5309 	ASSERT(ira->ira_flags & (IRAF_MULTIBROADCAST|IRAF_ICMP_ERROR));
5310 
5311 	laddr = ipha->ipha_dst;
5312 	faddr = ipha->ipha_src;
5313 
5314 	connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(lport, ipst)];
5315 	mutex_enter(&connfp->connf_lock);
5316 	connp = connfp->connf_head;
5317 
5318 	/*
5319 	 * If SO_REUSEADDR has been set on the first we send the
5320 	 * packet to all clients that have joined the group and
5321 	 * match the port.
5322 	 */
5323 	while (connp != NULL) {
5324 		if ((IPCL_UDP_MATCH(connp, lport, laddr, fport, faddr)) &&
5325 		    conn_wantpacket(connp, ira, ipha) &&
5326 		    (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5327 		    tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp)))
5328 			break;
5329 		connp = connp->conn_next;
5330 	}
5331 
5332 	if (connp == NULL)
5333 		goto notfound;
5334 
5335 	CONN_INC_REF(connp);
5336 
5337 	if (connp->conn_reuseaddr) {
5338 		conn_t		*first_connp = connp;
5339 		conn_t		*next_connp;
5340 		mblk_t		*mp1;
5341 
5342 		connp = connp->conn_next;
5343 		for (;;) {
5344 			while (connp != NULL) {
5345 				if (IPCL_UDP_MATCH(connp, lport, laddr,
5346 				    fport, faddr) &&
5347 				    conn_wantpacket(connp, ira, ipha) &&
5348 				    (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5349 				    tsol_receive_local(mp, &laddr, IPV4_VERSION,
5350 				    ira, connp)))
5351 					break;
5352 				connp = connp->conn_next;
5353 			}
5354 			if (connp == NULL) {
5355 				/* No more interested clients */
5356 				connp = first_connp;
5357 				break;
5358 			}
5359 			if (((mp1 = dupmsg(mp)) == NULL) &&
5360 			    ((mp1 = copymsg(mp)) == NULL)) {
5361 				/* Memory allocation failed */
5362 				BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5363 				ip_drop_input("ipIfStatsInDiscards", mp, ill);
5364 				connp = first_connp;
5365 				break;
5366 			}
5367 			CONN_INC_REF(connp);
5368 			mutex_exit(&connfp->connf_lock);
5369 
5370 			IP_STAT(ipst, ip_udp_fanmb);
5371 			ip_fanout_udp_conn(connp, mp1, (ipha_t *)mp1->b_rptr,
5372 			    NULL, ira);
5373 			mutex_enter(&connfp->connf_lock);
5374 			/* Follow the next pointer before releasing the conn */
5375 			next_connp = connp->conn_next;
5376 			CONN_DEC_REF(connp);
5377 			connp = next_connp;
5378 		}
5379 	}
5380 
5381 	/* Last one.  Send it upstream. */
5382 	mutex_exit(&connfp->connf_lock);
5383 	IP_STAT(ipst, ip_udp_fanmb);
5384 	ip_fanout_udp_conn(connp, mp, ipha, NULL, ira);
5385 	CONN_DEC_REF(connp);
5386 	return;
5387 
5388 notfound:
5389 	mutex_exit(&connfp->connf_lock);
5390 	/*
5391 	 * IPv6 endpoints bound to multicast IPv4-mapped addresses
5392 	 * have already been matched above, since they live in the IPv4
5393 	 * fanout tables. This implies we only need to
5394 	 * check for IPv6 in6addr_any endpoints here.
5395 	 * Thus we compare using ipv6_all_zeros instead of the destination
5396 	 * address, except for the multicast group membership lookup which
5397 	 * uses the IPv4 destination.
5398 	 */
5399 	IN6_IPADDR_TO_V4MAPPED(ipha->ipha_src, &v6faddr);
5400 	connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(lport, ipst)];
5401 	mutex_enter(&connfp->connf_lock);
5402 	connp = connfp->connf_head;
5403 	/*
5404 	 * IPv4 multicast packet being delivered to an AF_INET6
5405 	 * in6addr_any endpoint.
5406 	 * Need to check conn_wantpacket(). Note that we use conn_wantpacket()
5407 	 * and not conn_wantpacket_v6() since any multicast membership is
5408 	 * for an IPv4-mapped multicast address.
5409 	 */
5410 	while (connp != NULL) {
5411 		if (IPCL_UDP_MATCH_V6(connp, lport, ipv6_all_zeros,
5412 		    fport, v6faddr) &&
5413 		    conn_wantpacket(connp, ira, ipha) &&
5414 		    (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5415 		    tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp)))
5416 			break;
5417 		connp = connp->conn_next;
5418 	}
5419 
5420 	if (connp == NULL) {
5421 		/*
5422 		 * No one bound to this port.  Is
5423 		 * there a client that wants all
5424 		 * unclaimed datagrams?
5425 		 */
5426 		mutex_exit(&connfp->connf_lock);
5427 
5428 		if (ipst->ips_ipcl_proto_fanout_v4[IPPROTO_UDP].connf_head !=
5429 		    NULL) {
5430 			ASSERT(ira->ira_protocol == IPPROTO_UDP);
5431 			ip_fanout_proto_v4(mp, ipha, ira);
5432 		} else {
5433 			/*
5434 			 * We used to attempt to send an icmp error here, but
5435 			 * since this is known to be a multicast packet
5436 			 * and we don't send icmp errors in response to
5437 			 * multicast, just drop the packet and give up sooner.
5438 			 */
5439 			BUMP_MIB(ill->ill_ip_mib, udpIfStatsNoPorts);
5440 			freemsg(mp);
5441 		}
5442 		return;
5443 	}
5444 	CONN_INC_REF(connp);
5445 	ASSERT(IPCL_IS_NONSTR(connp) || connp->conn_rq != NULL);
5446 
5447 	/*
5448 	 * If SO_REUSEADDR has been set on the first we send the
5449 	 * packet to all clients that have joined the group and
5450 	 * match the port.
5451 	 */
5452 	if (connp->conn_reuseaddr) {
5453 		conn_t		*first_connp = connp;
5454 		conn_t		*next_connp;
5455 		mblk_t		*mp1;
5456 
5457 		connp = connp->conn_next;
5458 		for (;;) {
5459 			while (connp != NULL) {
5460 				if (IPCL_UDP_MATCH_V6(connp, lport,
5461 				    ipv6_all_zeros, fport, v6faddr) &&
5462 				    conn_wantpacket(connp, ira, ipha) &&
5463 				    (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5464 				    tsol_receive_local(mp, &laddr, IPV4_VERSION,
5465 				    ira, connp)))
5466 					break;
5467 				connp = connp->conn_next;
5468 			}
5469 			if (connp == NULL) {
5470 				/* No more interested clients */
5471 				connp = first_connp;
5472 				break;
5473 			}
5474 			if (((mp1 = dupmsg(mp)) == NULL) &&
5475 			    ((mp1 = copymsg(mp)) == NULL)) {
5476 				/* Memory allocation failed */
5477 				BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5478 				ip_drop_input("ipIfStatsInDiscards", mp, ill);
5479 				connp = first_connp;
5480 				break;
5481 			}
5482 			CONN_INC_REF(connp);
5483 			mutex_exit(&connfp->connf_lock);
5484 
5485 			IP_STAT(ipst, ip_udp_fanmb);
5486 			ip_fanout_udp_conn(connp, mp1, (ipha_t *)mp1->b_rptr,
5487 			    NULL, ira);
5488 			mutex_enter(&connfp->connf_lock);
5489 			/* Follow the next pointer before releasing the conn */
5490 			next_connp = connp->conn_next;
5491 			CONN_DEC_REF(connp);
5492 			connp = next_connp;
5493 		}
5494 	}
5495 
5496 	/* Last one.  Send it upstream. */
5497 	mutex_exit(&connfp->connf_lock);
5498 	IP_STAT(ipst, ip_udp_fanmb);
5499 	ip_fanout_udp_conn(connp, mp, ipha, NULL, ira);
5500 	CONN_DEC_REF(connp);
5501 }
5502 
5503 /*
5504  * Split an incoming packet's IPv4 options into the label and the other options.
5505  * If 'allocate' is set it does memory allocation for the ip_pkt_t, including
5506  * clearing out any leftover label or options.
5507  * Otherwise it just makes ipp point into the packet.
5508  *
5509  * Returns zero if ok; ENOMEM if the buffer couldn't be allocated.
5510  */
5511 int
ip_find_hdr_v4(ipha_t * ipha,ip_pkt_t * ipp,boolean_t allocate)5512 ip_find_hdr_v4(ipha_t *ipha, ip_pkt_t *ipp, boolean_t allocate)
5513 {
5514 	uchar_t		*opt;
5515 	uint32_t	totallen;
5516 	uint32_t	optval;
5517 	uint32_t	optlen;
5518 
5519 	ipp->ipp_fields |= IPPF_HOPLIMIT | IPPF_TCLASS | IPPF_ADDR;
5520 	ipp->ipp_hoplimit = ipha->ipha_ttl;
5521 	ipp->ipp_type_of_service = ipha->ipha_type_of_service;
5522 	IN6_IPADDR_TO_V4MAPPED(ipha->ipha_dst, &ipp->ipp_addr);
5523 
5524 	/*
5525 	 * Get length (in 4 byte octets) of IP header options.
5526 	 */
5527 	totallen = ipha->ipha_version_and_hdr_length -
5528 	    (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
5529 
5530 	if (totallen == 0) {
5531 		if (!allocate)
5532 			return (0);
5533 
5534 		/* Clear out anything from a previous packet */
5535 		if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
5536 			kmem_free(ipp->ipp_ipv4_options,
5537 			    ipp->ipp_ipv4_options_len);
5538 			ipp->ipp_ipv4_options = NULL;
5539 			ipp->ipp_ipv4_options_len = 0;
5540 			ipp->ipp_fields &= ~IPPF_IPV4_OPTIONS;
5541 		}
5542 		if (ipp->ipp_fields & IPPF_LABEL_V4) {
5543 			kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
5544 			ipp->ipp_label_v4 = NULL;
5545 			ipp->ipp_label_len_v4 = 0;
5546 			ipp->ipp_fields &= ~IPPF_LABEL_V4;
5547 		}
5548 		return (0);
5549 	}
5550 
5551 	totallen <<= 2;
5552 	opt = (uchar_t *)&ipha[1];
5553 	if (!is_system_labeled()) {
5554 
5555 	copyall:
5556 		if (!allocate) {
5557 			if (totallen != 0) {
5558 				ipp->ipp_ipv4_options = opt;
5559 				ipp->ipp_ipv4_options_len = totallen;
5560 				ipp->ipp_fields |= IPPF_IPV4_OPTIONS;
5561 			}
5562 			return (0);
5563 		}
5564 		/* Just copy all of options */
5565 		if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
5566 			if (totallen == ipp->ipp_ipv4_options_len) {
5567 				bcopy(opt, ipp->ipp_ipv4_options, totallen);
5568 				return (0);
5569 			}
5570 			kmem_free(ipp->ipp_ipv4_options,
5571 			    ipp->ipp_ipv4_options_len);
5572 			ipp->ipp_ipv4_options = NULL;
5573 			ipp->ipp_ipv4_options_len = 0;
5574 			ipp->ipp_fields &= ~IPPF_IPV4_OPTIONS;
5575 		}
5576 		if (totallen == 0)
5577 			return (0);
5578 
5579 		ipp->ipp_ipv4_options = kmem_alloc(totallen, KM_NOSLEEP);
5580 		if (ipp->ipp_ipv4_options == NULL)
5581 			return (ENOMEM);
5582 		ipp->ipp_ipv4_options_len = totallen;
5583 		ipp->ipp_fields |= IPPF_IPV4_OPTIONS;
5584 		bcopy(opt, ipp->ipp_ipv4_options, totallen);
5585 		return (0);
5586 	}
5587 
5588 	if (allocate && (ipp->ipp_fields & IPPF_LABEL_V4)) {
5589 		kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
5590 		ipp->ipp_label_v4 = NULL;
5591 		ipp->ipp_label_len_v4 = 0;
5592 		ipp->ipp_fields &= ~IPPF_LABEL_V4;
5593 	}
5594 
5595 	/*
5596 	 * Search for CIPSO option.
5597 	 * We assume CIPSO is first in options if it is present.
5598 	 * If it isn't, then ipp_opt_ipv4_options will not include the options
5599 	 * prior to the CIPSO option.
5600 	 */
5601 	while (totallen != 0) {
5602 		switch (optval = opt[IPOPT_OPTVAL]) {
5603 		case IPOPT_EOL:
5604 			return (0);
5605 		case IPOPT_NOP:
5606 			optlen = 1;
5607 			break;
5608 		default:
5609 			if (totallen <= IPOPT_OLEN)
5610 				return (EINVAL);
5611 			optlen = opt[IPOPT_OLEN];
5612 			if (optlen < 2)
5613 				return (EINVAL);
5614 		}
5615 		if (optlen > totallen)
5616 			return (EINVAL);
5617 
5618 		switch (optval) {
5619 		case IPOPT_COMSEC:
5620 			if (!allocate) {
5621 				ipp->ipp_label_v4 = opt;
5622 				ipp->ipp_label_len_v4 = optlen;
5623 				ipp->ipp_fields |= IPPF_LABEL_V4;
5624 			} else {
5625 				ipp->ipp_label_v4 = kmem_alloc(optlen,
5626 				    KM_NOSLEEP);
5627 				if (ipp->ipp_label_v4 == NULL)
5628 					return (ENOMEM);
5629 				ipp->ipp_label_len_v4 = optlen;
5630 				ipp->ipp_fields |= IPPF_LABEL_V4;
5631 				bcopy(opt, ipp->ipp_label_v4, optlen);
5632 			}
5633 			totallen -= optlen;
5634 			opt += optlen;
5635 
5636 			/* Skip padding bytes until we get to a multiple of 4 */
5637 			while ((totallen & 3) != 0 && opt[0] == IPOPT_NOP) {
5638 				totallen--;
5639 				opt++;
5640 			}
5641 			/* Remaining as ipp_ipv4_options */
5642 			goto copyall;
5643 		}
5644 		totallen -= optlen;
5645 		opt += optlen;
5646 	}
5647 	/* No CIPSO found; return everything as ipp_ipv4_options */
5648 	totallen = ipha->ipha_version_and_hdr_length -
5649 	    (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
5650 	totallen <<= 2;
5651 	opt = (uchar_t *)&ipha[1];
5652 	goto copyall;
5653 }
5654 
5655 /*
5656  * Efficient versions of lookup for an IRE when we only
5657  * match the address.
5658  * For RTF_REJECT or BLACKHOLE we return IRE_NOROUTE.
5659  * Does not handle multicast addresses.
5660  */
5661 uint_t
ip_type_v4(ipaddr_t addr,ip_stack_t * ipst)5662 ip_type_v4(ipaddr_t addr, ip_stack_t *ipst)
5663 {
5664 	ire_t *ire;
5665 	uint_t result;
5666 
5667 	ire = ire_ftable_lookup_simple_v4(addr, 0, ipst, NULL);
5668 	ASSERT(ire != NULL);
5669 	if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))
5670 		result = IRE_NOROUTE;
5671 	else
5672 		result = ire->ire_type;
5673 	ire_refrele(ire);
5674 	return (result);
5675 }
5676 
5677 /*
5678  * Efficient versions of lookup for an IRE when we only
5679  * match the address.
5680  * For RTF_REJECT or BLACKHOLE we return IRE_NOROUTE.
5681  * Does not handle multicast addresses.
5682  */
5683 uint_t
ip_type_v6(const in6_addr_t * addr,ip_stack_t * ipst)5684 ip_type_v6(const in6_addr_t *addr, ip_stack_t *ipst)
5685 {
5686 	ire_t *ire;
5687 	uint_t result;
5688 
5689 	ire = ire_ftable_lookup_simple_v6(addr, 0, ipst, NULL);
5690 	ASSERT(ire != NULL);
5691 	if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))
5692 		result = IRE_NOROUTE;
5693 	else
5694 		result = ire->ire_type;
5695 	ire_refrele(ire);
5696 	return (result);
5697 }
5698 
5699 /*
5700  * Nobody should be sending
5701  * packets up this stream
5702  */
5703 static int
ip_lrput(queue_t * q,mblk_t * mp)5704 ip_lrput(queue_t *q, mblk_t *mp)
5705 {
5706 	switch (mp->b_datap->db_type) {
5707 	case M_FLUSH:
5708 		/* Turn around */
5709 		if (*mp->b_rptr & FLUSHW) {
5710 			*mp->b_rptr &= ~FLUSHR;
5711 			qreply(q, mp);
5712 			return (0);
5713 		}
5714 		break;
5715 	}
5716 	freemsg(mp);
5717 	return (0);
5718 }
5719 
5720 /* Nobody should be sending packets down this stream */
5721 /* ARGSUSED */
5722 int
ip_lwput(queue_t * q,mblk_t * mp)5723 ip_lwput(queue_t *q, mblk_t *mp)
5724 {
5725 	freemsg(mp);
5726 	return (0);
5727 }
5728 
5729 /*
5730  * Move the first hop in any source route to ipha_dst and remove that part of
5731  * the source route.  Called by other protocols.  Errors in option formatting
5732  * are ignored - will be handled by ip_output_options. Return the final
5733  * destination (either ipha_dst or the last entry in a source route.)
5734  */
5735 ipaddr_t
ip_massage_options(ipha_t * ipha,netstack_t * ns)5736 ip_massage_options(ipha_t *ipha, netstack_t *ns)
5737 {
5738 	ipoptp_t	opts;
5739 	uchar_t		*opt;
5740 	uint8_t		optval;
5741 	uint8_t		optlen;
5742 	ipaddr_t	dst;
5743 	int		i;
5744 	ip_stack_t	*ipst = ns->netstack_ip;
5745 
5746 	ip2dbg(("ip_massage_options\n"));
5747 	dst = ipha->ipha_dst;
5748 	for (optval = ipoptp_first(&opts, ipha);
5749 	    optval != IPOPT_EOL;
5750 	    optval = ipoptp_next(&opts)) {
5751 		opt = opts.ipoptp_cur;
5752 		switch (optval) {
5753 			uint8_t off;
5754 		case IPOPT_SSRR:
5755 		case IPOPT_LSRR:
5756 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
5757 				ip1dbg(("ip_massage_options: bad src route\n"));
5758 				break;
5759 			}
5760 			optlen = opts.ipoptp_len;
5761 			off = opt[IPOPT_OFFSET];
5762 			off--;
5763 		redo_srr:
5764 			if (optlen < IP_ADDR_LEN ||
5765 			    off > optlen - IP_ADDR_LEN) {
5766 				/* End of source route */
5767 				ip1dbg(("ip_massage_options: end of SR\n"));
5768 				break;
5769 			}
5770 			bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
5771 			ip1dbg(("ip_massage_options: next hop 0x%x\n",
5772 			    ntohl(dst)));
5773 			/*
5774 			 * Check if our address is present more than
5775 			 * once as consecutive hops in source route.
5776 			 * XXX verify per-interface ip_forwarding
5777 			 * for source route?
5778 			 */
5779 			if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
5780 				off += IP_ADDR_LEN;
5781 				goto redo_srr;
5782 			}
5783 			if (dst == htonl(INADDR_LOOPBACK)) {
5784 				ip1dbg(("ip_massage_options: loopback addr in "
5785 				    "source route!\n"));
5786 				break;
5787 			}
5788 			/*
5789 			 * Update ipha_dst to be the first hop and remove the
5790 			 * first hop from the source route (by overwriting
5791 			 * part of the option with NOP options).
5792 			 */
5793 			ipha->ipha_dst = dst;
5794 			/* Put the last entry in dst */
5795 			off = ((optlen - IP_ADDR_LEN - 3) & ~(IP_ADDR_LEN-1)) +
5796 			    3;
5797 			bcopy(&opt[off], &dst, IP_ADDR_LEN);
5798 
5799 			ip1dbg(("ip_massage_options: last hop 0x%x\n",
5800 			    ntohl(dst)));
5801 			/* Move down and overwrite */
5802 			opt[IP_ADDR_LEN] = opt[0];
5803 			opt[IP_ADDR_LEN+1] = opt[IPOPT_OLEN] - IP_ADDR_LEN;
5804 			opt[IP_ADDR_LEN+2] = opt[IPOPT_OFFSET];
5805 			for (i = 0; i < IP_ADDR_LEN; i++)
5806 				opt[i] = IPOPT_NOP;
5807 			break;
5808 		}
5809 	}
5810 	return (dst);
5811 }
5812 
5813 /*
5814  * Return the network mask
5815  * associated with the specified address.
5816  */
5817 ipaddr_t
ip_net_mask(ipaddr_t addr)5818 ip_net_mask(ipaddr_t addr)
5819 {
5820 	uchar_t	*up = (uchar_t *)&addr;
5821 	ipaddr_t mask = 0;
5822 	uchar_t	*maskp = (uchar_t *)&mask;
5823 
5824 #if defined(__x86)
5825 #define	TOTALLY_BRAIN_DAMAGED_C_COMPILER
5826 #endif
5827 #ifdef  TOTALLY_BRAIN_DAMAGED_C_COMPILER
5828 	maskp[0] = maskp[1] = maskp[2] = maskp[3] = 0;
5829 #endif
5830 	if (CLASSD(addr)) {
5831 		maskp[0] = 0xF0;
5832 		return (mask);
5833 	}
5834 
5835 	/* We assume Class E default netmask to be 32 */
5836 	if (CLASSE(addr))
5837 		return (0xffffffffU);
5838 
5839 	if (addr == 0)
5840 		return (0);
5841 	maskp[0] = 0xFF;
5842 	if ((up[0] & 0x80) == 0)
5843 		return (mask);
5844 
5845 	maskp[1] = 0xFF;
5846 	if ((up[0] & 0xC0) == 0x80)
5847 		return (mask);
5848 
5849 	maskp[2] = 0xFF;
5850 	if ((up[0] & 0xE0) == 0xC0)
5851 		return (mask);
5852 
5853 	/* Otherwise return no mask */
5854 	return ((ipaddr_t)0);
5855 }
5856 
5857 /* Name/Value Table Lookup Routine */
5858 char *
ip_nv_lookup(nv_t * nv,int value)5859 ip_nv_lookup(nv_t *nv, int value)
5860 {
5861 	if (!nv)
5862 		return (NULL);
5863 	for (; nv->nv_name; nv++) {
5864 		if (nv->nv_value == value)
5865 			return (nv->nv_name);
5866 	}
5867 	return ("unknown");
5868 }
5869 
5870 static int
ip_wait_for_info_ack(ill_t * ill)5871 ip_wait_for_info_ack(ill_t *ill)
5872 {
5873 	int err;
5874 
5875 	mutex_enter(&ill->ill_lock);
5876 	while (ill->ill_state_flags & ILL_LL_SUBNET_PENDING) {
5877 		/*
5878 		 * Return value of 0 indicates a pending signal.
5879 		 */
5880 		err = cv_wait_sig(&ill->ill_cv, &ill->ill_lock);
5881 		if (err == 0) {
5882 			mutex_exit(&ill->ill_lock);
5883 			return (EINTR);
5884 		}
5885 	}
5886 	mutex_exit(&ill->ill_lock);
5887 	/*
5888 	 * ip_rput_other could have set an error  in ill_error on
5889 	 * receipt of M_ERROR.
5890 	 */
5891 	return (ill->ill_error);
5892 }
5893 
5894 /*
5895  * This is a module open, i.e. this is a control stream for access
5896  * to a DLPI device.  We allocate an ill_t as the instance data in
5897  * this case.
5898  */
5899 static int
ip_modopen(queue_t * q,dev_t * devp,int flag,int sflag,cred_t * credp)5900 ip_modopen(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5901 {
5902 	ill_t	*ill;
5903 	int	err;
5904 	zoneid_t zoneid;
5905 	netstack_t *ns;
5906 	ip_stack_t *ipst;
5907 
5908 	/*
5909 	 * Prevent unprivileged processes from pushing IP so that
5910 	 * they can't send raw IP.
5911 	 */
5912 	if (secpolicy_net_rawaccess(credp) != 0)
5913 		return (EPERM);
5914 
5915 	ns = netstack_find_by_cred(credp);
5916 	ASSERT(ns != NULL);
5917 	ipst = ns->netstack_ip;
5918 	ASSERT(ipst != NULL);
5919 
5920 	/*
5921 	 * For exclusive stacks we set the zoneid to zero
5922 	 * to make IP operate as if in the global zone.
5923 	 */
5924 	if (ipst->ips_netstack->netstack_stackid != GLOBAL_NETSTACKID)
5925 		zoneid = GLOBAL_ZONEID;
5926 	else
5927 		zoneid = crgetzoneid(credp);
5928 
5929 	ill = (ill_t *)mi_open_alloc_sleep(sizeof (ill_t));
5930 	q->q_ptr = WR(q)->q_ptr = ill;
5931 	ill->ill_ipst = ipst;
5932 	ill->ill_zoneid = zoneid;
5933 
5934 	/*
5935 	 * ill_init initializes the ill fields and then sends down
5936 	 * down a DL_INFO_REQ after calling qprocson.
5937 	 */
5938 	err = ill_init(q, ill);
5939 
5940 	if (err != 0) {
5941 		mi_free(ill);
5942 		netstack_rele(ipst->ips_netstack);
5943 		q->q_ptr = NULL;
5944 		WR(q)->q_ptr = NULL;
5945 		return (err);
5946 	}
5947 
5948 	/*
5949 	 * Wait for the DL_INFO_ACK if a DL_INFO_REQ was sent.
5950 	 *
5951 	 * ill_init initializes the ipsq marking this thread as
5952 	 * writer
5953 	 */
5954 	ipsq_exit(ill->ill_phyint->phyint_ipsq);
5955 	err = ip_wait_for_info_ack(ill);
5956 	if (err == 0)
5957 		ill->ill_credp = credp;
5958 	else
5959 		goto fail;
5960 
5961 	crhold(credp);
5962 
5963 	mutex_enter(&ipst->ips_ip_mi_lock);
5964 	err = mi_open_link(&ipst->ips_ip_g_head, (IDP)q->q_ptr, devp, flag,
5965 	    sflag, credp);
5966 	mutex_exit(&ipst->ips_ip_mi_lock);
5967 fail:
5968 	if (err) {
5969 		(void) ip_close(q, 0, credp);
5970 		return (err);
5971 	}
5972 	return (0);
5973 }
5974 
5975 /* For /dev/ip aka AF_INET open */
5976 int
ip_openv4(queue_t * q,dev_t * devp,int flag,int sflag,cred_t * credp)5977 ip_openv4(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5978 {
5979 	return (ip_open(q, devp, flag, sflag, credp, B_FALSE));
5980 }
5981 
5982 /* For /dev/ip6 aka AF_INET6 open */
5983 int
ip_openv6(queue_t * q,dev_t * devp,int flag,int sflag,cred_t * credp)5984 ip_openv6(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5985 {
5986 	return (ip_open(q, devp, flag, sflag, credp, B_TRUE));
5987 }
5988 
5989 /* IP open routine. */
5990 int
ip_open(queue_t * q,dev_t * devp,int flag,int sflag,cred_t * credp,boolean_t isv6)5991 ip_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp,
5992     boolean_t isv6)
5993 {
5994 	conn_t		*connp;
5995 	major_t		maj;
5996 	zoneid_t	zoneid;
5997 	netstack_t	*ns;
5998 	ip_stack_t	*ipst;
5999 
6000 	/* Allow reopen. */
6001 	if (q->q_ptr != NULL)
6002 		return (0);
6003 
6004 	if (sflag & MODOPEN) {
6005 		/* This is a module open */
6006 		return (ip_modopen(q, devp, flag, sflag, credp));
6007 	}
6008 
6009 	if ((flag & ~(FKLYR)) == IP_HELPER_STR) {
6010 		/*
6011 		 * Non streams based socket looking for a stream
6012 		 * to access IP
6013 		 */
6014 		return (ip_helper_stream_setup(q, devp, flag, sflag,
6015 		    credp, isv6));
6016 	}
6017 
6018 	ns = netstack_find_by_cred(credp);
6019 	ASSERT(ns != NULL);
6020 	ipst = ns->netstack_ip;
6021 	ASSERT(ipst != NULL);
6022 
6023 	/*
6024 	 * For exclusive stacks we set the zoneid to zero
6025 	 * to make IP operate as if in the global zone.
6026 	 */
6027 	if (ipst->ips_netstack->netstack_stackid != GLOBAL_NETSTACKID)
6028 		zoneid = GLOBAL_ZONEID;
6029 	else
6030 		zoneid = crgetzoneid(credp);
6031 
6032 	/*
6033 	 * We are opening as a device. This is an IP client stream, and we
6034 	 * allocate an conn_t as the instance data.
6035 	 */
6036 	connp = ipcl_conn_create(IPCL_IPCCONN, KM_SLEEP, ipst->ips_netstack);
6037 
6038 	/*
6039 	 * ipcl_conn_create did a netstack_hold. Undo the hold that was
6040 	 * done by netstack_find_by_cred()
6041 	 */
6042 	netstack_rele(ipst->ips_netstack);
6043 
6044 	connp->conn_ixa->ixa_flags |= IXAF_MULTICAST_LOOP | IXAF_SET_ULP_CKSUM;
6045 	/* conn_allzones can not be set this early, hence no IPCL_ZONEID */
6046 	connp->conn_ixa->ixa_zoneid = zoneid;
6047 	connp->conn_zoneid = zoneid;
6048 
6049 	connp->conn_rq = q;
6050 	q->q_ptr = WR(q)->q_ptr = connp;
6051 
6052 	/* Minor tells us which /dev entry was opened */
6053 	if (isv6) {
6054 		connp->conn_family = AF_INET6;
6055 		connp->conn_ipversion = IPV6_VERSION;
6056 		connp->conn_ixa->ixa_flags &= ~IXAF_IS_IPV4;
6057 		connp->conn_ixa->ixa_src_preferences = IPV6_PREFER_SRC_DEFAULT;
6058 	} else {
6059 		connp->conn_family = AF_INET;
6060 		connp->conn_ipversion = IPV4_VERSION;
6061 		connp->conn_ixa->ixa_flags |= IXAF_IS_IPV4;
6062 	}
6063 
6064 	if ((ip_minor_arena_la != NULL) && (flag & SO_SOCKSTR) &&
6065 	    ((connp->conn_dev = inet_minor_alloc(ip_minor_arena_la)) != 0)) {
6066 		connp->conn_minor_arena = ip_minor_arena_la;
6067 	} else {
6068 		/*
6069 		 * Either minor numbers in the large arena were exhausted
6070 		 * or a non socket application is doing the open.
6071 		 * Try to allocate from the small arena.
6072 		 */
6073 		if ((connp->conn_dev =
6074 		    inet_minor_alloc(ip_minor_arena_sa)) == 0) {
6075 			/* CONN_DEC_REF takes care of netstack_rele() */
6076 			q->q_ptr = WR(q)->q_ptr = NULL;
6077 			CONN_DEC_REF(connp);
6078 			return (EBUSY);
6079 		}
6080 		connp->conn_minor_arena = ip_minor_arena_sa;
6081 	}
6082 
6083 	maj = getemajor(*devp);
6084 	*devp = makedevice(maj, (minor_t)connp->conn_dev);
6085 
6086 	/*
6087 	 * connp->conn_cred is crfree()ed in ipcl_conn_destroy()
6088 	 */
6089 	connp->conn_cred = credp;
6090 	connp->conn_cpid = curproc->p_pid;
6091 	/* Cache things in ixa without an extra refhold */
6092 	ASSERT(!(connp->conn_ixa->ixa_free_flags & IXA_FREE_CRED));
6093 	connp->conn_ixa->ixa_cred = connp->conn_cred;
6094 	connp->conn_ixa->ixa_cpid = connp->conn_cpid;
6095 	if (is_system_labeled())
6096 		connp->conn_ixa->ixa_tsl = crgetlabel(connp->conn_cred);
6097 
6098 	/*
6099 	 * Handle IP_IOC_RTS_REQUEST and other ioctls which use conn_recv
6100 	 */
6101 	connp->conn_recv = ip_conn_input;
6102 	connp->conn_recvicmp = ip_conn_input_icmp;
6103 
6104 	crhold(connp->conn_cred);
6105 
6106 	/*
6107 	 * If the caller has the process-wide flag set, then default to MAC
6108 	 * exempt mode.  This allows read-down to unlabeled hosts.
6109 	 */
6110 	if (getpflags(NET_MAC_AWARE, credp) != 0)
6111 		connp->conn_mac_mode = CONN_MAC_AWARE;
6112 
6113 	connp->conn_zone_is_global = (crgetzoneid(credp) == GLOBAL_ZONEID);
6114 
6115 	connp->conn_rq = q;
6116 	connp->conn_wq = WR(q);
6117 
6118 	/* Non-zero default values */
6119 	connp->conn_ixa->ixa_flags |= IXAF_MULTICAST_LOOP;
6120 
6121 	/*
6122 	 * Make the conn globally visible to walkers
6123 	 */
6124 	ASSERT(connp->conn_ref == 1);
6125 	mutex_enter(&connp->conn_lock);
6126 	connp->conn_state_flags &= ~CONN_INCIPIENT;
6127 	mutex_exit(&connp->conn_lock);
6128 
6129 	qprocson(q);
6130 
6131 	return (0);
6132 }
6133 
6134 /*
6135  * Set IPsec policy from an ipsec_req_t. If the req is not "zero" and valid,
6136  * all of them are copied to the conn_t. If the req is "zero", the policy is
6137  * zeroed out. A "zero" policy has zero ipsr_{ah,req,self_encap}_req
6138  * fields.
6139  * We keep only the latest setting of the policy and thus policy setting
6140  * is not incremental/cumulative.
6141  *
6142  * Requests to set policies with multiple alternative actions will
6143  * go through a different API.
6144  */
6145 int
ipsec_set_req(cred_t * cr,conn_t * connp,ipsec_req_t * req)6146 ipsec_set_req(cred_t *cr, conn_t *connp, ipsec_req_t *req)
6147 {
6148 	uint_t ah_req = 0;
6149 	uint_t esp_req = 0;
6150 	uint_t se_req = 0;
6151 	ipsec_act_t *actp = NULL;
6152 	uint_t nact;
6153 	ipsec_policy_head_t *ph;
6154 	boolean_t is_pol_reset, is_pol_inserted = B_FALSE;
6155 	int error = 0;
6156 	netstack_t	*ns = connp->conn_netstack;
6157 	ip_stack_t	*ipst = ns->netstack_ip;
6158 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
6159 
6160 #define	REQ_MASK (IPSEC_PREF_REQUIRED|IPSEC_PREF_NEVER)
6161 
6162 	/*
6163 	 * The IP_SEC_OPT option does not allow variable length parameters,
6164 	 * hence a request cannot be NULL.
6165 	 */
6166 	if (req == NULL)
6167 		return (EINVAL);
6168 
6169 	ah_req = req->ipsr_ah_req;
6170 	esp_req = req->ipsr_esp_req;
6171 	se_req = req->ipsr_self_encap_req;
6172 
6173 	/* Don't allow setting self-encap without one or more of AH/ESP. */
6174 	if (se_req != 0 && esp_req == 0 && ah_req == 0)
6175 		return (EINVAL);
6176 
6177 	/*
6178 	 * Are we dealing with a request to reset the policy (i.e.
6179 	 * zero requests).
6180 	 */
6181 	is_pol_reset = ((ah_req & REQ_MASK) == 0 &&
6182 	    (esp_req & REQ_MASK) == 0 &&
6183 	    (se_req & REQ_MASK) == 0);
6184 
6185 	if (!is_pol_reset) {
6186 		/*
6187 		 * If we couldn't load IPsec, fail with "protocol
6188 		 * not supported".
6189 		 * IPsec may not have been loaded for a request with zero
6190 		 * policies, so we don't fail in this case.
6191 		 */
6192 		mutex_enter(&ipss->ipsec_loader_lock);
6193 		if (ipss->ipsec_loader_state != IPSEC_LOADER_SUCCEEDED) {
6194 			mutex_exit(&ipss->ipsec_loader_lock);
6195 			return (EPROTONOSUPPORT);
6196 		}
6197 		mutex_exit(&ipss->ipsec_loader_lock);
6198 
6199 		/*
6200 		 * Test for valid requests. Invalid algorithms
6201 		 * need to be tested by IPsec code because new
6202 		 * algorithms can be added dynamically.
6203 		 */
6204 		if ((ah_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0 ||
6205 		    (esp_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0 ||
6206 		    (se_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0) {
6207 			return (EINVAL);
6208 		}
6209 
6210 		/*
6211 		 * Only privileged users can issue these
6212 		 * requests.
6213 		 */
6214 		if (((ah_req & IPSEC_PREF_NEVER) ||
6215 		    (esp_req & IPSEC_PREF_NEVER) ||
6216 		    (se_req & IPSEC_PREF_NEVER)) &&
6217 		    secpolicy_ip_config(cr, B_FALSE) != 0) {
6218 			return (EPERM);
6219 		}
6220 
6221 		/*
6222 		 * The IPSEC_PREF_REQUIRED and IPSEC_PREF_NEVER
6223 		 * are mutually exclusive.
6224 		 */
6225 		if (((ah_req & REQ_MASK) == REQ_MASK) ||
6226 		    ((esp_req & REQ_MASK) == REQ_MASK) ||
6227 		    ((se_req & REQ_MASK) == REQ_MASK)) {
6228 			/* Both of them are set */
6229 			return (EINVAL);
6230 		}
6231 	}
6232 
6233 	ASSERT(MUTEX_HELD(&connp->conn_lock));
6234 
6235 	/*
6236 	 * If we have already cached policies in conn_connect(), don't
6237 	 * let them change now. We cache policies for connections
6238 	 * whose src,dst [addr, port] is known.
6239 	 */
6240 	if (connp->conn_policy_cached) {
6241 		return (EINVAL);
6242 	}
6243 
6244 	/*
6245 	 * We have a zero policies, reset the connection policy if already
6246 	 * set. This will cause the connection to inherit the
6247 	 * global policy, if any.
6248 	 */
6249 	if (is_pol_reset) {
6250 		if (connp->conn_policy != NULL) {
6251 			IPPH_REFRELE(connp->conn_policy, ipst->ips_netstack);
6252 			connp->conn_policy = NULL;
6253 		}
6254 		connp->conn_in_enforce_policy = B_FALSE;
6255 		connp->conn_out_enforce_policy = B_FALSE;
6256 		return (0);
6257 	}
6258 
6259 	ph = connp->conn_policy = ipsec_polhead_split(connp->conn_policy,
6260 	    ipst->ips_netstack);
6261 	if (ph == NULL)
6262 		goto enomem;
6263 
6264 	ipsec_actvec_from_req(req, &actp, &nact, ipst->ips_netstack);
6265 	if (actp == NULL)
6266 		goto enomem;
6267 
6268 	/*
6269 	 * Always insert IPv4 policy entries, since they can also apply to
6270 	 * ipv6 sockets being used in ipv4-compat mode.
6271 	 */
6272 	if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V4,
6273 	    IPSEC_TYPE_INBOUND, ns))
6274 		goto enomem;
6275 	is_pol_inserted = B_TRUE;
6276 	if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V4,
6277 	    IPSEC_TYPE_OUTBOUND, ns))
6278 		goto enomem;
6279 
6280 	/*
6281 	 * We're looking at a v6 socket, also insert the v6-specific
6282 	 * entries.
6283 	 */
6284 	if (connp->conn_family == AF_INET6) {
6285 		if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V6,
6286 		    IPSEC_TYPE_INBOUND, ns))
6287 			goto enomem;
6288 		if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V6,
6289 		    IPSEC_TYPE_OUTBOUND, ns))
6290 			goto enomem;
6291 	}
6292 
6293 	ipsec_actvec_free(actp, nact);
6294 
6295 	/*
6296 	 * If the requests need security, set enforce_policy.
6297 	 * If the requests are IPSEC_PREF_NEVER, one should
6298 	 * still set conn_out_enforce_policy so that ip_set_destination
6299 	 * marks the ip_xmit_attr_t appropriatly. This is needed so that
6300 	 * for connections that we don't cache policy in at connect time,
6301 	 * if global policy matches in ip_output_attach_policy, we
6302 	 * don't wrongly inherit global policy. Similarly, we need
6303 	 * to set conn_in_enforce_policy also so that we don't verify
6304 	 * policy wrongly.
6305 	 */
6306 	if ((ah_req & REQ_MASK) != 0 ||
6307 	    (esp_req & REQ_MASK) != 0 ||
6308 	    (se_req & REQ_MASK) != 0) {
6309 		connp->conn_in_enforce_policy = B_TRUE;
6310 		connp->conn_out_enforce_policy = B_TRUE;
6311 	}
6312 
6313 	return (error);
6314 #undef REQ_MASK
6315 
6316 	/*
6317 	 * Common memory-allocation-failure exit path.
6318 	 */
6319 enomem:
6320 	if (actp != NULL)
6321 		ipsec_actvec_free(actp, nact);
6322 	if (is_pol_inserted)
6323 		ipsec_polhead_flush(ph, ns);
6324 	return (ENOMEM);
6325 }
6326 
6327 /*
6328  * Set socket options for joining and leaving multicast groups.
6329  * Common to IPv4 and IPv6; inet6 indicates the type of socket.
6330  * The caller has already check that the option name is consistent with
6331  * the address family of the socket.
6332  */
6333 int
ip_opt_set_multicast_group(conn_t * connp,t_scalar_t name,uchar_t * invalp,boolean_t inet6,boolean_t checkonly)6334 ip_opt_set_multicast_group(conn_t *connp, t_scalar_t name,
6335     uchar_t *invalp, boolean_t inet6, boolean_t checkonly)
6336 {
6337 	int		*i1 = (int *)invalp;
6338 	int		error = 0;
6339 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
6340 	struct ip_mreq	*v4_mreqp;
6341 	struct ipv6_mreq *v6_mreqp;
6342 	struct group_req *greqp;
6343 	ire_t *ire;
6344 	boolean_t done = B_FALSE;
6345 	ipaddr_t ifaddr;
6346 	in6_addr_t v6group;
6347 	uint_t ifindex;
6348 	boolean_t mcast_opt = B_TRUE;
6349 	mcast_record_t fmode;
6350 	int (*optfn)(conn_t *, boolean_t, const in6_addr_t *,
6351 	    ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *);
6352 
6353 	switch (name) {
6354 	case IP_ADD_MEMBERSHIP:
6355 	case IPV6_JOIN_GROUP:
6356 		mcast_opt = B_FALSE;
6357 		/* FALLTHROUGH */
6358 	case MCAST_JOIN_GROUP:
6359 		fmode = MODE_IS_EXCLUDE;
6360 		optfn = ip_opt_add_group;
6361 		break;
6362 
6363 	case IP_DROP_MEMBERSHIP:
6364 	case IPV6_LEAVE_GROUP:
6365 		mcast_opt = B_FALSE;
6366 		/* FALLTHROUGH */
6367 	case MCAST_LEAVE_GROUP:
6368 		fmode = MODE_IS_INCLUDE;
6369 		optfn = ip_opt_delete_group;
6370 		break;
6371 	default:
6372 		/* Should not be reached. */
6373 		fmode = MODE_IS_INCLUDE;
6374 		optfn = NULL;
6375 		ASSERT(0);
6376 	}
6377 
6378 	if (mcast_opt) {
6379 		struct sockaddr_in *sin;
6380 		struct sockaddr_in6 *sin6;
6381 
6382 		greqp = (struct group_req *)i1;
6383 		if (greqp->gr_group.ss_family == AF_INET) {
6384 			sin = (struct sockaddr_in *)&(greqp->gr_group);
6385 			IN6_INADDR_TO_V4MAPPED(&sin->sin_addr, &v6group);
6386 		} else {
6387 			if (!inet6)
6388 				return (EINVAL);	/* Not on INET socket */
6389 
6390 			sin6 = (struct sockaddr_in6 *)&(greqp->gr_group);
6391 			v6group = sin6->sin6_addr;
6392 		}
6393 		ifaddr = INADDR_ANY;
6394 		ifindex = greqp->gr_interface;
6395 	} else if (inet6) {
6396 		v6_mreqp = (struct ipv6_mreq *)i1;
6397 		v6group = v6_mreqp->ipv6mr_multiaddr;
6398 		ifaddr = INADDR_ANY;
6399 		ifindex = v6_mreqp->ipv6mr_interface;
6400 	} else {
6401 		v4_mreqp = (struct ip_mreq *)i1;
6402 		IN6_INADDR_TO_V4MAPPED(&v4_mreqp->imr_multiaddr, &v6group);
6403 		ifaddr = (ipaddr_t)v4_mreqp->imr_interface.s_addr;
6404 		ifindex = 0;
6405 	}
6406 
6407 	/*
6408 	 * In the multirouting case, we need to replicate
6409 	 * the request on all interfaces that will take part
6410 	 * in replication.  We do so because multirouting is
6411 	 * reflective, thus we will probably receive multi-
6412 	 * casts on those interfaces.
6413 	 * The ip_multirt_apply_membership() succeeds if
6414 	 * the operation succeeds on at least one interface.
6415 	 */
6416 	if (IN6_IS_ADDR_V4MAPPED(&v6group)) {
6417 		ipaddr_t group;
6418 
6419 		IN6_V4MAPPED_TO_IPADDR(&v6group, group);
6420 
6421 		ire = ire_ftable_lookup_v4(group, IP_HOST_MASK, 0,
6422 		    IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6423 		    MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6424 	} else {
6425 		ire = ire_ftable_lookup_v6(&v6group, &ipv6_all_ones, 0,
6426 		    IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6427 		    MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6428 	}
6429 	if (ire != NULL) {
6430 		if (ire->ire_flags & RTF_MULTIRT) {
6431 			error = ip_multirt_apply_membership(optfn, ire, connp,
6432 			    checkonly, &v6group, fmode, &ipv6_all_zeros);
6433 			done = B_TRUE;
6434 		}
6435 		ire_refrele(ire);
6436 	}
6437 
6438 	if (!done) {
6439 		error = optfn(connp, checkonly, &v6group, ifaddr, ifindex,
6440 		    fmode, &ipv6_all_zeros);
6441 	}
6442 	return (error);
6443 }
6444 
6445 /*
6446  * Set socket options for joining and leaving multicast groups
6447  * for specific sources.
6448  * Common to IPv4 and IPv6; inet6 indicates the type of socket.
6449  * The caller has already check that the option name is consistent with
6450  * the address family of the socket.
6451  */
6452 int
ip_opt_set_multicast_sources(conn_t * connp,t_scalar_t name,uchar_t * invalp,boolean_t inet6,boolean_t checkonly)6453 ip_opt_set_multicast_sources(conn_t *connp, t_scalar_t name,
6454     uchar_t *invalp, boolean_t inet6, boolean_t checkonly)
6455 {
6456 	int		*i1 = (int *)invalp;
6457 	int		error = 0;
6458 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
6459 	struct ip_mreq_source *imreqp;
6460 	struct group_source_req *gsreqp;
6461 	in6_addr_t v6group, v6src;
6462 	uint32_t ifindex;
6463 	ipaddr_t ifaddr;
6464 	boolean_t mcast_opt = B_TRUE;
6465 	mcast_record_t fmode;
6466 	ire_t *ire;
6467 	boolean_t done = B_FALSE;
6468 	int (*optfn)(conn_t *, boolean_t, const in6_addr_t *,
6469 	    ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *);
6470 
6471 	switch (name) {
6472 	case IP_BLOCK_SOURCE:
6473 		mcast_opt = B_FALSE;
6474 		/* FALLTHROUGH */
6475 	case MCAST_BLOCK_SOURCE:
6476 		fmode = MODE_IS_EXCLUDE;
6477 		optfn = ip_opt_add_group;
6478 		break;
6479 
6480 	case IP_UNBLOCK_SOURCE:
6481 		mcast_opt = B_FALSE;
6482 		/* FALLTHROUGH */
6483 	case MCAST_UNBLOCK_SOURCE:
6484 		fmode = MODE_IS_EXCLUDE;
6485 		optfn = ip_opt_delete_group;
6486 		break;
6487 
6488 	case IP_ADD_SOURCE_MEMBERSHIP:
6489 		mcast_opt = B_FALSE;
6490 		/* FALLTHROUGH */
6491 	case MCAST_JOIN_SOURCE_GROUP:
6492 		fmode = MODE_IS_INCLUDE;
6493 		optfn = ip_opt_add_group;
6494 		break;
6495 
6496 	case IP_DROP_SOURCE_MEMBERSHIP:
6497 		mcast_opt = B_FALSE;
6498 		/* FALLTHROUGH */
6499 	case MCAST_LEAVE_SOURCE_GROUP:
6500 		fmode = MODE_IS_INCLUDE;
6501 		optfn = ip_opt_delete_group;
6502 		break;
6503 	default:
6504 		/* Should not be reached. */
6505 		optfn = NULL;
6506 		fmode = 0;
6507 		ASSERT(0);
6508 	}
6509 
6510 	if (mcast_opt) {
6511 		gsreqp = (struct group_source_req *)i1;
6512 		ifindex = gsreqp->gsr_interface;
6513 		if (gsreqp->gsr_group.ss_family == AF_INET) {
6514 			struct sockaddr_in *s;
6515 			s = (struct sockaddr_in *)&gsreqp->gsr_group;
6516 			IN6_INADDR_TO_V4MAPPED(&s->sin_addr, &v6group);
6517 			s = (struct sockaddr_in *)&gsreqp->gsr_source;
6518 			IN6_INADDR_TO_V4MAPPED(&s->sin_addr, &v6src);
6519 		} else {
6520 			struct sockaddr_in6 *s6;
6521 
6522 			if (!inet6)
6523 				return (EINVAL);	/* Not on INET socket */
6524 
6525 			s6 = (struct sockaddr_in6 *)&gsreqp->gsr_group;
6526 			v6group = s6->sin6_addr;
6527 			s6 = (struct sockaddr_in6 *)&gsreqp->gsr_source;
6528 			v6src = s6->sin6_addr;
6529 		}
6530 		ifaddr = INADDR_ANY;
6531 	} else {
6532 		imreqp = (struct ip_mreq_source *)i1;
6533 		IN6_INADDR_TO_V4MAPPED(&imreqp->imr_multiaddr, &v6group);
6534 		IN6_INADDR_TO_V4MAPPED(&imreqp->imr_sourceaddr, &v6src);
6535 		ifaddr = (ipaddr_t)imreqp->imr_interface.s_addr;
6536 		ifindex = 0;
6537 	}
6538 
6539 	/*
6540 	 * Handle src being mapped INADDR_ANY by changing it to unspecified.
6541 	 */
6542 	if (IN6_IS_ADDR_V4MAPPED_ANY(&v6src))
6543 		v6src = ipv6_all_zeros;
6544 
6545 	/*
6546 	 * In the multirouting case, we need to replicate
6547 	 * the request as noted in the mcast cases above.
6548 	 */
6549 	if (IN6_IS_ADDR_V4MAPPED(&v6group)) {
6550 		ipaddr_t group;
6551 
6552 		IN6_V4MAPPED_TO_IPADDR(&v6group, group);
6553 
6554 		ire = ire_ftable_lookup_v4(group, IP_HOST_MASK, 0,
6555 		    IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6556 		    MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6557 	} else {
6558 		ire = ire_ftable_lookup_v6(&v6group, &ipv6_all_ones, 0,
6559 		    IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6560 		    MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6561 	}
6562 	if (ire != NULL) {
6563 		if (ire->ire_flags & RTF_MULTIRT) {
6564 			error = ip_multirt_apply_membership(optfn, ire, connp,
6565 			    checkonly, &v6group, fmode, &v6src);
6566 			done = B_TRUE;
6567 		}
6568 		ire_refrele(ire);
6569 	}
6570 	if (!done) {
6571 		error = optfn(connp, checkonly, &v6group, ifaddr, ifindex,
6572 		    fmode, &v6src);
6573 	}
6574 	return (error);
6575 }
6576 
6577 /*
6578  * Given a destination address and a pointer to where to put the information
6579  * this routine fills in the mtuinfo.
6580  * The socket must be connected.
6581  * For sctp conn_faddr is the primary address.
6582  */
6583 int
ip_fill_mtuinfo(conn_t * connp,ip_xmit_attr_t * ixa,struct ip6_mtuinfo * mtuinfo)6584 ip_fill_mtuinfo(conn_t *connp, ip_xmit_attr_t *ixa, struct ip6_mtuinfo *mtuinfo)
6585 {
6586 	uint32_t	pmtu = IP_MAXPACKET;
6587 	uint_t		scopeid;
6588 
6589 	if (IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6))
6590 		return (-1);
6591 
6592 	/* In case we never sent or called ip_set_destination_v4/v6 */
6593 	if (ixa->ixa_ire != NULL)
6594 		pmtu = ip_get_pmtu(ixa);
6595 
6596 	if (ixa->ixa_flags & IXAF_SCOPEID_SET)
6597 		scopeid = ixa->ixa_scopeid;
6598 	else
6599 		scopeid = 0;
6600 
6601 	bzero(mtuinfo, sizeof (*mtuinfo));
6602 	mtuinfo->ip6m_addr.sin6_family = AF_INET6;
6603 	mtuinfo->ip6m_addr.sin6_port = connp->conn_fport;
6604 	mtuinfo->ip6m_addr.sin6_addr = connp->conn_faddr_v6;
6605 	mtuinfo->ip6m_addr.sin6_scope_id = scopeid;
6606 	mtuinfo->ip6m_mtu = pmtu;
6607 
6608 	return (sizeof (struct ip6_mtuinfo));
6609 }
6610 
6611 /*
6612  * When the src multihoming is changed from weak to [strong, preferred]
6613  * ip_ire_rebind_walker is called to walk the list of all ire_t entries
6614  * and identify routes that were created by user-applications in the
6615  * unbound state (i.e., without RTA_IFP), and for which an ire_ill is not
6616  * currently defined. These routes are then 'rebound', i.e., their ire_ill
6617  * is selected by finding an interface route for the gateway.
6618  */
6619 /* ARGSUSED */
6620 void
ip_ire_rebind_walker(ire_t * ire,void * notused)6621 ip_ire_rebind_walker(ire_t *ire, void *notused)
6622 {
6623 	if (!ire->ire_unbound || ire->ire_ill != NULL)
6624 		return;
6625 	ire_rebind(ire);
6626 	ire_delete(ire);
6627 }
6628 
6629 /*
6630  * When the src multihoming is changed from  [strong, preferred] to weak,
6631  * ip_ire_unbind_walker is called to walk the list of all ire_t entries, and
6632  * set any entries that were created by user-applications in the unbound state
6633  * (i.e., without RTA_IFP) back to having a NULL ire_ill.
6634  */
6635 /* ARGSUSED */
6636 void
ip_ire_unbind_walker(ire_t * ire,void * notused)6637 ip_ire_unbind_walker(ire_t *ire, void *notused)
6638 {
6639 	ire_t *new_ire;
6640 
6641 	if (!ire->ire_unbound || ire->ire_ill == NULL)
6642 		return;
6643 	if (ire->ire_ipversion == IPV6_VERSION) {
6644 		new_ire = ire_create_v6(&ire->ire_addr_v6, &ire->ire_mask_v6,
6645 		    &ire->ire_gateway_addr_v6, ire->ire_type, NULL,
6646 		    ire->ire_zoneid, ire->ire_flags, NULL, ire->ire_ipst);
6647 	} else {
6648 		new_ire = ire_create((uchar_t *)&ire->ire_addr,
6649 		    (uchar_t *)&ire->ire_mask,
6650 		    (uchar_t *)&ire->ire_gateway_addr, ire->ire_type, NULL,
6651 		    ire->ire_zoneid, ire->ire_flags, NULL, ire->ire_ipst);
6652 	}
6653 	if (new_ire == NULL)
6654 		return;
6655 	new_ire->ire_unbound = B_TRUE;
6656 	/*
6657 	 * The bound ire must first be deleted so that we don't return
6658 	 * the existing one on the attempt to add the unbound new_ire.
6659 	 */
6660 	ire_delete(ire);
6661 	new_ire = ire_add(new_ire);
6662 	if (new_ire != NULL)
6663 		ire_refrele(new_ire);
6664 }
6665 
6666 /*
6667  * When the settings of ip*_strict_src_multihoming tunables are changed,
6668  * all cached routes need to be recomputed. This recomputation needs to be
6669  * done when going from weaker to stronger modes so that the cached ire
6670  * for the connection does not violate the current ip*_strict_src_multihoming
6671  * setting. It also needs to be done when going from stronger to weaker modes,
6672  * so that we fall back to matching on the longest-matching-route (as opposed
6673  * to a shorter match that may have been selected in the strong mode
6674  * to satisfy src_multihoming settings).
6675  *
6676  * The cached ixa_ire entires for all conn_t entries are marked as
6677  * "verify" so that they will be recomputed for the next packet.
6678  */
6679 void
conn_ire_revalidate(conn_t * connp,void * arg)6680 conn_ire_revalidate(conn_t *connp, void *arg)
6681 {
6682 	boolean_t isv6 = (boolean_t)arg;
6683 
6684 	if ((isv6 && connp->conn_ipversion != IPV6_VERSION) ||
6685 	    (!isv6 && connp->conn_ipversion != IPV4_VERSION))
6686 		return;
6687 	connp->conn_ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
6688 }
6689 
6690 /*
6691  * Handles both IPv4 and IPv6 reassembly - doing the out-of-order cases,
6692  * When an ipf is passed here for the first time, if
6693  * we already have in-order fragments on the queue, we convert from the fast-
6694  * path reassembly scheme to the hard-case scheme.  From then on, additional
6695  * fragments are reassembled here.  We keep track of the start and end offsets
6696  * of each piece, and the number of holes in the chain.  When the hole count
6697  * goes to zero, we are done!
6698  *
6699  * The ipf_count will be updated to account for any mblk(s) added (pointed to
6700  * by mp) or subtracted (freeb()ed dups), upon return the caller must update
6701  * ipfb_count and ill_frag_count by the difference of ipf_count before and
6702  * after the call to ip_reassemble().
6703  */
6704 int
ip_reassemble(mblk_t * mp,ipf_t * ipf,uint_t start,boolean_t more,ill_t * ill,size_t msg_len)6705 ip_reassemble(mblk_t *mp, ipf_t *ipf, uint_t start, boolean_t more, ill_t *ill,
6706     size_t msg_len)
6707 {
6708 	uint_t	end;
6709 	mblk_t	*next_mp;
6710 	mblk_t	*mp1;
6711 	uint_t	offset;
6712 	boolean_t incr_dups = B_TRUE;
6713 	boolean_t offset_zero_seen = B_FALSE;
6714 	boolean_t pkt_boundary_checked = B_FALSE;
6715 
6716 	/* If start == 0 then ipf_nf_hdr_len has to be set. */
6717 	ASSERT(start != 0 || ipf->ipf_nf_hdr_len != 0);
6718 
6719 	/* Add in byte count */
6720 	ipf->ipf_count += msg_len;
6721 	if (ipf->ipf_end) {
6722 		/*
6723 		 * We were part way through in-order reassembly, but now there
6724 		 * is a hole.  We walk through messages already queued, and
6725 		 * mark them for hard case reassembly.  We know that up till
6726 		 * now they were in order starting from offset zero.
6727 		 */
6728 		offset = 0;
6729 		for (mp1 = ipf->ipf_mp->b_cont; mp1; mp1 = mp1->b_cont) {
6730 			IP_REASS_SET_START(mp1, offset);
6731 			if (offset == 0) {
6732 				ASSERT(ipf->ipf_nf_hdr_len != 0);
6733 				offset = -ipf->ipf_nf_hdr_len;
6734 			}
6735 			offset += mp1->b_wptr - mp1->b_rptr;
6736 			IP_REASS_SET_END(mp1, offset);
6737 		}
6738 		/* One hole at the end. */
6739 		ipf->ipf_hole_cnt = 1;
6740 		/* Brand it as a hard case, forever. */
6741 		ipf->ipf_end = 0;
6742 	}
6743 	/* Walk through all the new pieces. */
6744 	do {
6745 		end = start + (mp->b_wptr - mp->b_rptr);
6746 		/*
6747 		 * If start is 0, decrease 'end' only for the first mblk of
6748 		 * the fragment. Otherwise 'end' can get wrong value in the
6749 		 * second pass of the loop if first mblk is exactly the
6750 		 * size of ipf_nf_hdr_len.
6751 		 */
6752 		if (start == 0 && !offset_zero_seen) {
6753 			/* First segment */
6754 			ASSERT(ipf->ipf_nf_hdr_len != 0);
6755 			end -= ipf->ipf_nf_hdr_len;
6756 			offset_zero_seen = B_TRUE;
6757 		}
6758 		next_mp = mp->b_cont;
6759 		/*
6760 		 * We are checking to see if there is any interesing data
6761 		 * to process.  If there isn't and the mblk isn't the
6762 		 * one which carries the unfragmentable header then we
6763 		 * drop it.  It's possible to have just the unfragmentable
6764 		 * header come through without any data.  That needs to be
6765 		 * saved.
6766 		 *
6767 		 * If the assert at the top of this function holds then the
6768 		 * term "ipf->ipf_nf_hdr_len != 0" isn't needed.  This code
6769 		 * is infrequently traveled enough that the test is left in
6770 		 * to protect against future code changes which break that
6771 		 * invariant.
6772 		 */
6773 		if (start == end && start != 0 && ipf->ipf_nf_hdr_len != 0) {
6774 			/* Empty.  Blast it. */
6775 			IP_REASS_SET_START(mp, 0);
6776 			IP_REASS_SET_END(mp, 0);
6777 			/*
6778 			 * If the ipf points to the mblk we are about to free,
6779 			 * update ipf to point to the next mblk (or NULL
6780 			 * if none).
6781 			 */
6782 			if (ipf->ipf_mp->b_cont == mp)
6783 				ipf->ipf_mp->b_cont = next_mp;
6784 			freeb(mp);
6785 			continue;
6786 		}
6787 		mp->b_cont = NULL;
6788 		IP_REASS_SET_START(mp, start);
6789 		IP_REASS_SET_END(mp, end);
6790 		if (!ipf->ipf_tail_mp) {
6791 			ipf->ipf_tail_mp = mp;
6792 			ipf->ipf_mp->b_cont = mp;
6793 			if (start == 0 || !more) {
6794 				ipf->ipf_hole_cnt = 1;
6795 				/*
6796 				 * if the first fragment comes in more than one
6797 				 * mblk, this loop will be executed for each
6798 				 * mblk. Need to adjust hole count so exiting
6799 				 * this routine will leave hole count at 1.
6800 				 */
6801 				if (next_mp)
6802 					ipf->ipf_hole_cnt++;
6803 			} else
6804 				ipf->ipf_hole_cnt = 2;
6805 			continue;
6806 		} else if (ipf->ipf_last_frag_seen && !more &&
6807 		    !pkt_boundary_checked) {
6808 			/*
6809 			 * We check datagram boundary only if this fragment
6810 			 * claims to be the last fragment and we have seen a
6811 			 * last fragment in the past too. We do this only
6812 			 * once for a given fragment.
6813 			 *
6814 			 * start cannot be 0 here as fragments with start=0
6815 			 * and MF=0 gets handled as a complete packet. These
6816 			 * fragments should not reach here.
6817 			 */
6818 
6819 			if (start + msgdsize(mp) !=
6820 			    IP_REASS_END(ipf->ipf_tail_mp)) {
6821 				/*
6822 				 * We have two fragments both of which claim
6823 				 * to be the last fragment but gives conflicting
6824 				 * information about the whole datagram size.
6825 				 * Something fishy is going on. Drop the
6826 				 * fragment and free up the reassembly list.
6827 				 */
6828 				return (IP_REASS_FAILED);
6829 			}
6830 
6831 			/*
6832 			 * We shouldn't come to this code block again for this
6833 			 * particular fragment.
6834 			 */
6835 			pkt_boundary_checked = B_TRUE;
6836 		}
6837 
6838 		/* New stuff at or beyond tail? */
6839 		offset = IP_REASS_END(ipf->ipf_tail_mp);
6840 		if (start >= offset) {
6841 			if (ipf->ipf_last_frag_seen) {
6842 				/* current fragment is beyond last fragment */
6843 				return (IP_REASS_FAILED);
6844 			}
6845 			/* Link it on end. */
6846 			ipf->ipf_tail_mp->b_cont = mp;
6847 			ipf->ipf_tail_mp = mp;
6848 			if (more) {
6849 				if (start != offset)
6850 					ipf->ipf_hole_cnt++;
6851 			} else if (start == offset && next_mp == NULL)
6852 					ipf->ipf_hole_cnt--;
6853 			continue;
6854 		}
6855 		mp1 = ipf->ipf_mp->b_cont;
6856 		offset = IP_REASS_START(mp1);
6857 		/* New stuff at the front? */
6858 		if (start < offset) {
6859 			if (start == 0) {
6860 				if (end >= offset) {
6861 					/* Nailed the hole at the begining. */
6862 					ipf->ipf_hole_cnt--;
6863 				}
6864 			} else if (end < offset) {
6865 				/*
6866 				 * A hole, stuff, and a hole where there used
6867 				 * to be just a hole.
6868 				 */
6869 				ipf->ipf_hole_cnt++;
6870 			}
6871 			mp->b_cont = mp1;
6872 			/* Check for overlap. */
6873 			while (end > offset) {
6874 				if (end < IP_REASS_END(mp1)) {
6875 					mp->b_wptr -= end - offset;
6876 					IP_REASS_SET_END(mp, offset);
6877 					BUMP_MIB(ill->ill_ip_mib,
6878 					    ipIfStatsReasmPartDups);
6879 					break;
6880 				}
6881 				/* Did we cover another hole? */
6882 				if ((mp1->b_cont &&
6883 				    IP_REASS_END(mp1) !=
6884 				    IP_REASS_START(mp1->b_cont) &&
6885 				    end >= IP_REASS_START(mp1->b_cont)) ||
6886 				    (!ipf->ipf_last_frag_seen && !more)) {
6887 					ipf->ipf_hole_cnt--;
6888 				}
6889 				/* Clip out mp1. */
6890 				if ((mp->b_cont = mp1->b_cont) == NULL) {
6891 					/*
6892 					 * After clipping out mp1, this guy
6893 					 * is now hanging off the end.
6894 					 */
6895 					ipf->ipf_tail_mp = mp;
6896 				}
6897 				IP_REASS_SET_START(mp1, 0);
6898 				IP_REASS_SET_END(mp1, 0);
6899 				/* Subtract byte count */
6900 				ipf->ipf_count -= mp1->b_datap->db_lim -
6901 				    mp1->b_datap->db_base;
6902 				freeb(mp1);
6903 				BUMP_MIB(ill->ill_ip_mib,
6904 				    ipIfStatsReasmPartDups);
6905 				mp1 = mp->b_cont;
6906 				if (!mp1)
6907 					break;
6908 				offset = IP_REASS_START(mp1);
6909 			}
6910 			ipf->ipf_mp->b_cont = mp;
6911 			continue;
6912 		}
6913 		/*
6914 		 * The new piece starts somewhere between the start of the head
6915 		 * and before the end of the tail.
6916 		 */
6917 		for (; mp1; mp1 = mp1->b_cont) {
6918 			offset = IP_REASS_END(mp1);
6919 			if (start < offset) {
6920 				if (end <= offset) {
6921 					/* Nothing new. */
6922 					IP_REASS_SET_START(mp, 0);
6923 					IP_REASS_SET_END(mp, 0);
6924 					/* Subtract byte count */
6925 					ipf->ipf_count -= mp->b_datap->db_lim -
6926 					    mp->b_datap->db_base;
6927 					if (incr_dups) {
6928 						ipf->ipf_num_dups++;
6929 						incr_dups = B_FALSE;
6930 					}
6931 					freeb(mp);
6932 					BUMP_MIB(ill->ill_ip_mib,
6933 					    ipIfStatsReasmDuplicates);
6934 					break;
6935 				}
6936 				/*
6937 				 * Trim redundant stuff off beginning of new
6938 				 * piece.
6939 				 */
6940 				IP_REASS_SET_START(mp, offset);
6941 				mp->b_rptr += offset - start;
6942 				BUMP_MIB(ill->ill_ip_mib,
6943 				    ipIfStatsReasmPartDups);
6944 				start = offset;
6945 				if (!mp1->b_cont) {
6946 					/*
6947 					 * After trimming, this guy is now
6948 					 * hanging off the end.
6949 					 */
6950 					mp1->b_cont = mp;
6951 					ipf->ipf_tail_mp = mp;
6952 					if (!more) {
6953 						ipf->ipf_hole_cnt--;
6954 					}
6955 					break;
6956 				}
6957 			}
6958 			if (start >= IP_REASS_START(mp1->b_cont))
6959 				continue;
6960 			/* Fill a hole */
6961 			if (start > offset)
6962 				ipf->ipf_hole_cnt++;
6963 			mp->b_cont = mp1->b_cont;
6964 			mp1->b_cont = mp;
6965 			mp1 = mp->b_cont;
6966 			offset = IP_REASS_START(mp1);
6967 			if (end >= offset) {
6968 				ipf->ipf_hole_cnt--;
6969 				/* Check for overlap. */
6970 				while (end > offset) {
6971 					if (end < IP_REASS_END(mp1)) {
6972 						mp->b_wptr -= end - offset;
6973 						IP_REASS_SET_END(mp, offset);
6974 						/*
6975 						 * TODO we might bump
6976 						 * this up twice if there is
6977 						 * overlap at both ends.
6978 						 */
6979 						BUMP_MIB(ill->ill_ip_mib,
6980 						    ipIfStatsReasmPartDups);
6981 						break;
6982 					}
6983 					/* Did we cover another hole? */
6984 					if ((mp1->b_cont &&
6985 					    IP_REASS_END(mp1)
6986 					    != IP_REASS_START(mp1->b_cont) &&
6987 					    end >=
6988 					    IP_REASS_START(mp1->b_cont)) ||
6989 					    (!ipf->ipf_last_frag_seen &&
6990 					    !more)) {
6991 						ipf->ipf_hole_cnt--;
6992 					}
6993 					/* Clip out mp1. */
6994 					if ((mp->b_cont = mp1->b_cont) ==
6995 					    NULL) {
6996 						/*
6997 						 * After clipping out mp1,
6998 						 * this guy is now hanging
6999 						 * off the end.
7000 						 */
7001 						ipf->ipf_tail_mp = mp;
7002 					}
7003 					IP_REASS_SET_START(mp1, 0);
7004 					IP_REASS_SET_END(mp1, 0);
7005 					/* Subtract byte count */
7006 					ipf->ipf_count -=
7007 					    mp1->b_datap->db_lim -
7008 					    mp1->b_datap->db_base;
7009 					freeb(mp1);
7010 					BUMP_MIB(ill->ill_ip_mib,
7011 					    ipIfStatsReasmPartDups);
7012 					mp1 = mp->b_cont;
7013 					if (!mp1)
7014 						break;
7015 					offset = IP_REASS_START(mp1);
7016 				}
7017 			}
7018 			break;
7019 		}
7020 	} while (start = end, mp = next_mp);
7021 
7022 	/* Fragment just processed could be the last one. Remember this fact */
7023 	if (!more)
7024 		ipf->ipf_last_frag_seen = B_TRUE;
7025 
7026 	/* Still got holes? */
7027 	if (ipf->ipf_hole_cnt)
7028 		return (IP_REASS_PARTIAL);
7029 	/* Clean up overloaded fields to avoid upstream disasters. */
7030 	for (mp1 = ipf->ipf_mp->b_cont; mp1; mp1 = mp1->b_cont) {
7031 		IP_REASS_SET_START(mp1, 0);
7032 		IP_REASS_SET_END(mp1, 0);
7033 	}
7034 	return (IP_REASS_COMPLETE);
7035 }
7036 
7037 /*
7038  * Fragmentation reassembly.  Each ILL has a hash table for
7039  * queuing packets undergoing reassembly for all IPIFs
7040  * associated with the ILL.  The hash is based on the packet
7041  * IP ident field.  The ILL frag hash table was allocated
7042  * as a timer block at the time the ILL was created.  Whenever
7043  * there is anything on the reassembly queue, the timer will
7044  * be running.  Returns the reassembled packet if reassembly completes.
7045  */
7046 mblk_t *
ip_input_fragment(mblk_t * mp,ipha_t * ipha,ip_recv_attr_t * ira)7047 ip_input_fragment(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
7048 {
7049 	uint32_t	frag_offset_flags;
7050 	mblk_t		*t_mp;
7051 	ipaddr_t	dst;
7052 	uint8_t		proto = ipha->ipha_protocol;
7053 	uint32_t	sum_val;
7054 	uint16_t	sum_flags;
7055 	ipf_t		*ipf;
7056 	ipf_t		**ipfp;
7057 	ipfb_t		*ipfb;
7058 	uint16_t	ident;
7059 	uint32_t	offset;
7060 	ipaddr_t	src;
7061 	uint_t		hdr_length;
7062 	uint32_t	end;
7063 	mblk_t		*mp1;
7064 	mblk_t		*tail_mp;
7065 	size_t		count;
7066 	size_t		msg_len;
7067 	uint8_t		ecn_info = 0;
7068 	uint32_t	packet_size;
7069 	boolean_t	pruned = B_FALSE;
7070 	ill_t		*ill = ira->ira_ill;
7071 	ip_stack_t	*ipst = ill->ill_ipst;
7072 
7073 	/*
7074 	 * Drop the fragmented as early as possible, if
7075 	 * we don't have resource(s) to re-assemble.
7076 	 */
7077 	if (ipst->ips_ip_reass_queue_bytes == 0) {
7078 		freemsg(mp);
7079 		return (NULL);
7080 	}
7081 
7082 	/* Check for fragmentation offset; return if there's none */
7083 	if ((frag_offset_flags = ntohs(ipha->ipha_fragment_offset_and_flags) &
7084 	    (IPH_MF | IPH_OFFSET)) == 0)
7085 		return (mp);
7086 
7087 	/*
7088 	 * We utilize hardware computed checksum info only for UDP since
7089 	 * IP fragmentation is a normal occurrence for the protocol.  In
7090 	 * addition, checksum offload support for IP fragments carrying
7091 	 * UDP payload is commonly implemented across network adapters.
7092 	 */
7093 	ASSERT(ira->ira_rill != NULL);
7094 	if (proto == IPPROTO_UDP && dohwcksum &&
7095 	    ILL_HCKSUM_CAPABLE(ira->ira_rill) &&
7096 	    (DB_CKSUMFLAGS(mp) & (HCK_FULLCKSUM | HCK_PARTIALCKSUM))) {
7097 		mblk_t *mp1 = mp->b_cont;
7098 		int32_t len;
7099 
7100 		/* Record checksum information from the packet */
7101 		sum_val = (uint32_t)DB_CKSUM16(mp);
7102 		sum_flags = DB_CKSUMFLAGS(mp);
7103 
7104 		/* IP payload offset from beginning of mblk */
7105 		offset = ((uchar_t *)ipha + IPH_HDR_LENGTH(ipha)) - mp->b_rptr;
7106 
7107 		if ((sum_flags & HCK_PARTIALCKSUM) &&
7108 		    (mp1 == NULL || mp1->b_cont == NULL) &&
7109 		    offset >= DB_CKSUMSTART(mp) &&
7110 		    ((len = offset - DB_CKSUMSTART(mp)) & 1) == 0) {
7111 			uint32_t adj;
7112 			/*
7113 			 * Partial checksum has been calculated by hardware
7114 			 * and attached to the packet; in addition, any
7115 			 * prepended extraneous data is even byte aligned.
7116 			 * If any such data exists, we adjust the checksum;
7117 			 * this would also handle any postpended data.
7118 			 */
7119 			IP_ADJCKSUM_PARTIAL(mp->b_rptr + DB_CKSUMSTART(mp),
7120 			    mp, mp1, len, adj);
7121 
7122 			/* One's complement subtract extraneous checksum */
7123 			if (adj >= sum_val)
7124 				sum_val = ~(adj - sum_val) & 0xFFFF;
7125 			else
7126 				sum_val -= adj;
7127 		}
7128 	} else {
7129 		sum_val = 0;
7130 		sum_flags = 0;
7131 	}
7132 
7133 	/* Clear hardware checksumming flag */
7134 	DB_CKSUMFLAGS(mp) = 0;
7135 
7136 	ident = ipha->ipha_ident;
7137 	offset = (frag_offset_flags << 3) & 0xFFFF;
7138 	src = ipha->ipha_src;
7139 	dst = ipha->ipha_dst;
7140 	hdr_length = IPH_HDR_LENGTH(ipha);
7141 	end = ntohs(ipha->ipha_length) - hdr_length;
7142 
7143 	/* If end == 0 then we have a packet with no data, so just free it */
7144 	if (end == 0) {
7145 		freemsg(mp);
7146 		return (NULL);
7147 	}
7148 
7149 	/* Record the ECN field info. */
7150 	ecn_info = (ipha->ipha_type_of_service & 0x3);
7151 	if (offset != 0) {
7152 		/*
7153 		 * If this isn't the first piece, strip the header, and
7154 		 * add the offset to the end value.
7155 		 */
7156 		mp->b_rptr += hdr_length;
7157 		end += offset;
7158 	}
7159 
7160 	/* Handle vnic loopback of fragments */
7161 	if (mp->b_datap->db_ref > 2)
7162 		msg_len = 0;
7163 	else
7164 		msg_len = MBLKSIZE(mp);
7165 
7166 	tail_mp = mp;
7167 	while (tail_mp->b_cont != NULL) {
7168 		tail_mp = tail_mp->b_cont;
7169 		if (tail_mp->b_datap->db_ref <= 2)
7170 			msg_len += MBLKSIZE(tail_mp);
7171 	}
7172 
7173 	/* If the reassembly list for this ILL will get too big, prune it */
7174 	if ((msg_len + sizeof (*ipf) + ill->ill_frag_count) >=
7175 	    ipst->ips_ip_reass_queue_bytes) {
7176 		DTRACE_PROBE3(ip_reass_queue_bytes, uint_t, msg_len,
7177 		    uint_t, ill->ill_frag_count,
7178 		    uint_t, ipst->ips_ip_reass_queue_bytes);
7179 		ill_frag_prune(ill,
7180 		    (ipst->ips_ip_reass_queue_bytes < msg_len) ? 0 :
7181 		    (ipst->ips_ip_reass_queue_bytes - msg_len));
7182 		pruned = B_TRUE;
7183 	}
7184 
7185 	ipfb = &ill->ill_frag_hash_tbl[ILL_FRAG_HASH(src, ident)];
7186 	mutex_enter(&ipfb->ipfb_lock);
7187 
7188 	ipfp = &ipfb->ipfb_ipf;
7189 	/* Try to find an existing fragment queue for this packet. */
7190 	for (;;) {
7191 		ipf = ipfp[0];
7192 		if (ipf != NULL) {
7193 			/*
7194 			 * It has to match on ident and src/dst address.
7195 			 */
7196 			if (ipf->ipf_ident == ident &&
7197 			    ipf->ipf_src == src &&
7198 			    ipf->ipf_dst == dst &&
7199 			    ipf->ipf_protocol == proto) {
7200 				/*
7201 				 * If we have received too many
7202 				 * duplicate fragments for this packet
7203 				 * free it.
7204 				 */
7205 				if (ipf->ipf_num_dups > ip_max_frag_dups) {
7206 					ill_frag_free_pkts(ill, ipfb, ipf, 1);
7207 					freemsg(mp);
7208 					mutex_exit(&ipfb->ipfb_lock);
7209 					return (NULL);
7210 				}
7211 				/* Found it. */
7212 				break;
7213 			}
7214 			ipfp = &ipf->ipf_hash_next;
7215 			continue;
7216 		}
7217 
7218 		/*
7219 		 * If we pruned the list, do we want to store this new
7220 		 * fragment?. We apply an optimization here based on the
7221 		 * fact that most fragments will be received in order.
7222 		 * So if the offset of this incoming fragment is zero,
7223 		 * it is the first fragment of a new packet. We will
7224 		 * keep it.  Otherwise drop the fragment, as we have
7225 		 * probably pruned the packet already (since the
7226 		 * packet cannot be found).
7227 		 */
7228 		if (pruned && offset != 0) {
7229 			mutex_exit(&ipfb->ipfb_lock);
7230 			freemsg(mp);
7231 			return (NULL);
7232 		}
7233 
7234 		if (ipfb->ipfb_frag_pkts >= MAX_FRAG_PKTS(ipst))  {
7235 			/*
7236 			 * Too many fragmented packets in this hash
7237 			 * bucket. Free the oldest.
7238 			 */
7239 			ill_frag_free_pkts(ill, ipfb, ipfb->ipfb_ipf, 1);
7240 		}
7241 
7242 		/* New guy.  Allocate a frag message. */
7243 		mp1 = allocb(sizeof (*ipf), BPRI_MED);
7244 		if (mp1 == NULL) {
7245 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7246 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
7247 			freemsg(mp);
7248 reass_done:
7249 			mutex_exit(&ipfb->ipfb_lock);
7250 			return (NULL);
7251 		}
7252 
7253 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsReasmReqds);
7254 		mp1->b_cont = mp;
7255 
7256 		/* Initialize the fragment header. */
7257 		ipf = (ipf_t *)mp1->b_rptr;
7258 		ipf->ipf_mp = mp1;
7259 		ipf->ipf_ptphn = ipfp;
7260 		ipfp[0] = ipf;
7261 		ipf->ipf_hash_next = NULL;
7262 		ipf->ipf_ident = ident;
7263 		ipf->ipf_protocol = proto;
7264 		ipf->ipf_src = src;
7265 		ipf->ipf_dst = dst;
7266 		ipf->ipf_nf_hdr_len = 0;
7267 		/* Record reassembly start time. */
7268 		ipf->ipf_timestamp = gethrestime_sec();
7269 		/* Record ipf generation and account for frag header */
7270 		ipf->ipf_gen = ill->ill_ipf_gen++;
7271 		ipf->ipf_count = MBLKSIZE(mp1);
7272 		ipf->ipf_last_frag_seen = B_FALSE;
7273 		ipf->ipf_ecn = ecn_info;
7274 		ipf->ipf_num_dups = 0;
7275 		ipfb->ipfb_frag_pkts++;
7276 		ipf->ipf_checksum = 0;
7277 		ipf->ipf_checksum_flags = 0;
7278 
7279 		/* Store checksum value in fragment header */
7280 		if (sum_flags != 0) {
7281 			sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7282 			sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7283 			ipf->ipf_checksum = sum_val;
7284 			ipf->ipf_checksum_flags = sum_flags;
7285 		}
7286 
7287 		/*
7288 		 * We handle reassembly two ways.  In the easy case,
7289 		 * where all the fragments show up in order, we do
7290 		 * minimal bookkeeping, and just clip new pieces on
7291 		 * the end.  If we ever see a hole, then we go off
7292 		 * to ip_reassemble which has to mark the pieces and
7293 		 * keep track of the number of holes, etc.  Obviously,
7294 		 * the point of having both mechanisms is so we can
7295 		 * handle the easy case as efficiently as possible.
7296 		 */
7297 		if (offset == 0) {
7298 			/* Easy case, in-order reassembly so far. */
7299 			ipf->ipf_count += msg_len;
7300 			ipf->ipf_tail_mp = tail_mp;
7301 			/*
7302 			 * Keep track of next expected offset in
7303 			 * ipf_end.
7304 			 */
7305 			ipf->ipf_end = end;
7306 			ipf->ipf_nf_hdr_len = hdr_length;
7307 		} else {
7308 			/* Hard case, hole at the beginning. */
7309 			ipf->ipf_tail_mp = NULL;
7310 			/*
7311 			 * ipf_end == 0 means that we have given up
7312 			 * on easy reassembly.
7313 			 */
7314 			ipf->ipf_end = 0;
7315 
7316 			/* Forget checksum offload from now on */
7317 			ipf->ipf_checksum_flags = 0;
7318 
7319 			/*
7320 			 * ipf_hole_cnt is set by ip_reassemble.
7321 			 * ipf_count is updated by ip_reassemble.
7322 			 * No need to check for return value here
7323 			 * as we don't expect reassembly to complete
7324 			 * or fail for the first fragment itself.
7325 			 */
7326 			(void) ip_reassemble(mp, ipf,
7327 			    (frag_offset_flags & IPH_OFFSET) << 3,
7328 			    (frag_offset_flags & IPH_MF), ill, msg_len);
7329 		}
7330 		/* Update per ipfb and ill byte counts */
7331 		ipfb->ipfb_count += ipf->ipf_count;
7332 		ASSERT(ipfb->ipfb_count > 0);	/* Wraparound */
7333 		atomic_add_32(&ill->ill_frag_count, ipf->ipf_count);
7334 		/* If the frag timer wasn't already going, start it. */
7335 		mutex_enter(&ill->ill_lock);
7336 		ill_frag_timer_start(ill);
7337 		mutex_exit(&ill->ill_lock);
7338 		goto reass_done;
7339 	}
7340 
7341 	/*
7342 	 * If the packet's flag has changed (it could be coming up
7343 	 * from an interface different than the previous, therefore
7344 	 * possibly different checksum capability), then forget about
7345 	 * any stored checksum states.  Otherwise add the value to
7346 	 * the existing one stored in the fragment header.
7347 	 */
7348 	if (sum_flags != 0 && sum_flags == ipf->ipf_checksum_flags) {
7349 		sum_val += ipf->ipf_checksum;
7350 		sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7351 		sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7352 		ipf->ipf_checksum = sum_val;
7353 	} else if (ipf->ipf_checksum_flags != 0) {
7354 		/* Forget checksum offload from now on */
7355 		ipf->ipf_checksum_flags = 0;
7356 	}
7357 
7358 	/*
7359 	 * We have a new piece of a datagram which is already being
7360 	 * reassembled.  Update the ECN info if all IP fragments
7361 	 * are ECN capable.  If there is one which is not, clear
7362 	 * all the info.  If there is at least one which has CE
7363 	 * code point, IP needs to report that up to transport.
7364 	 */
7365 	if (ecn_info != IPH_ECN_NECT && ipf->ipf_ecn != IPH_ECN_NECT) {
7366 		if (ecn_info == IPH_ECN_CE)
7367 			ipf->ipf_ecn = IPH_ECN_CE;
7368 	} else {
7369 		ipf->ipf_ecn = IPH_ECN_NECT;
7370 	}
7371 	if (offset && ipf->ipf_end == offset) {
7372 		/* The new fragment fits at the end */
7373 		ipf->ipf_tail_mp->b_cont = mp;
7374 		/* Update the byte count */
7375 		ipf->ipf_count += msg_len;
7376 		/* Update per ipfb and ill byte counts */
7377 		ipfb->ipfb_count += msg_len;
7378 		ASSERT(ipfb->ipfb_count > 0);	/* Wraparound */
7379 		atomic_add_32(&ill->ill_frag_count, msg_len);
7380 		if (frag_offset_flags & IPH_MF) {
7381 			/* More to come. */
7382 			ipf->ipf_end = end;
7383 			ipf->ipf_tail_mp = tail_mp;
7384 			goto reass_done;
7385 		}
7386 	} else {
7387 		/* Go do the hard cases. */
7388 		int ret;
7389 
7390 		if (offset == 0)
7391 			ipf->ipf_nf_hdr_len = hdr_length;
7392 
7393 		/* Save current byte count */
7394 		count = ipf->ipf_count;
7395 		ret = ip_reassemble(mp, ipf,
7396 		    (frag_offset_flags & IPH_OFFSET) << 3,
7397 		    (frag_offset_flags & IPH_MF), ill, msg_len);
7398 		/* Count of bytes added and subtracted (freeb()ed) */
7399 		count = ipf->ipf_count - count;
7400 		if (count) {
7401 			/* Update per ipfb and ill byte counts */
7402 			ipfb->ipfb_count += count;
7403 			ASSERT(ipfb->ipfb_count > 0); /* Wraparound */
7404 			atomic_add_32(&ill->ill_frag_count, count);
7405 		}
7406 		if (ret == IP_REASS_PARTIAL) {
7407 			goto reass_done;
7408 		} else if (ret == IP_REASS_FAILED) {
7409 			/* Reassembly failed. Free up all resources */
7410 			ill_frag_free_pkts(ill, ipfb, ipf, 1);
7411 			for (t_mp = mp; t_mp != NULL; t_mp = t_mp->b_cont) {
7412 				IP_REASS_SET_START(t_mp, 0);
7413 				IP_REASS_SET_END(t_mp, 0);
7414 			}
7415 			freemsg(mp);
7416 			goto reass_done;
7417 		}
7418 		/* We will reach here iff 'ret' is IP_REASS_COMPLETE */
7419 	}
7420 	/*
7421 	 * We have completed reassembly.  Unhook the frag header from
7422 	 * the reassembly list.
7423 	 *
7424 	 * Before we free the frag header, record the ECN info
7425 	 * to report back to the transport.
7426 	 */
7427 	ecn_info = ipf->ipf_ecn;
7428 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsReasmOKs);
7429 	ipfp = ipf->ipf_ptphn;
7430 
7431 	/* We need to supply these to caller */
7432 	if ((sum_flags = ipf->ipf_checksum_flags) != 0)
7433 		sum_val = ipf->ipf_checksum;
7434 	else
7435 		sum_val = 0;
7436 
7437 	mp1 = ipf->ipf_mp;
7438 	count = ipf->ipf_count;
7439 	ipf = ipf->ipf_hash_next;
7440 	if (ipf != NULL)
7441 		ipf->ipf_ptphn = ipfp;
7442 	ipfp[0] = ipf;
7443 	atomic_add_32(&ill->ill_frag_count, -count);
7444 	ASSERT(ipfb->ipfb_count >= count);
7445 	ipfb->ipfb_count -= count;
7446 	ipfb->ipfb_frag_pkts--;
7447 	mutex_exit(&ipfb->ipfb_lock);
7448 	/* Ditch the frag header. */
7449 	mp = mp1->b_cont;
7450 
7451 	freeb(mp1);
7452 
7453 	/* Restore original IP length in header. */
7454 	packet_size = (uint32_t)msgdsize(mp);
7455 	if (packet_size > IP_MAXPACKET) {
7456 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7457 		ip_drop_input("Reassembled packet too large", mp, ill);
7458 		freemsg(mp);
7459 		return (NULL);
7460 	}
7461 
7462 	if (DB_REF(mp) > 1) {
7463 		mblk_t *mp2 = copymsg(mp);
7464 
7465 		if (mp2 == NULL) {
7466 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7467 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
7468 			freemsg(mp);
7469 			return (NULL);
7470 		}
7471 		freemsg(mp);
7472 		mp = mp2;
7473 	}
7474 	ipha = (ipha_t *)mp->b_rptr;
7475 
7476 	ipha->ipha_length = htons((uint16_t)packet_size);
7477 	/* We're now complete, zip the frag state */
7478 	ipha->ipha_fragment_offset_and_flags = 0;
7479 	/* Record the ECN info. */
7480 	ipha->ipha_type_of_service &= 0xFC;
7481 	ipha->ipha_type_of_service |= ecn_info;
7482 
7483 	/* Update the receive attributes */
7484 	ira->ira_pktlen = packet_size;
7485 	ira->ira_ip_hdr_length = IPH_HDR_LENGTH(ipha);
7486 
7487 	/* Reassembly is successful; set checksum information in packet */
7488 	DB_CKSUM16(mp) = (uint16_t)sum_val;
7489 	DB_CKSUMFLAGS(mp) = sum_flags;
7490 	DB_CKSUMSTART(mp) = ira->ira_ip_hdr_length;
7491 
7492 	return (mp);
7493 }
7494 
7495 /*
7496  * Pullup function that should be used for IP input in order to
7497  * ensure we do not loose the L2 source address; we need the l2 source
7498  * address for IP_RECVSLLA and for ndp_input.
7499  *
7500  * We return either NULL or b_rptr.
7501  */
7502 void *
ip_pullup(mblk_t * mp,ssize_t len,ip_recv_attr_t * ira)7503 ip_pullup(mblk_t *mp, ssize_t len, ip_recv_attr_t *ira)
7504 {
7505 	ill_t		*ill = ira->ira_ill;
7506 
7507 	if (ip_rput_pullups++ == 0) {
7508 		(void) mi_strlog(ill->ill_rq, 1, SL_ERROR|SL_TRACE,
7509 		    "ip_pullup: %s forced us to "
7510 		    " pullup pkt, hdr len %ld, hdr addr %p",
7511 		    ill->ill_name, len, (void *)mp->b_rptr);
7512 	}
7513 	if (!(ira->ira_flags & IRAF_L2SRC_SET))
7514 		ip_setl2src(mp, ira, ira->ira_rill);
7515 	ASSERT(ira->ira_flags & IRAF_L2SRC_SET);
7516 	if (!pullupmsg(mp, len))
7517 		return (NULL);
7518 	else
7519 		return (mp->b_rptr);
7520 }
7521 
7522 /*
7523  * Make sure ira_l2src has an address. If we don't have one fill with zeros.
7524  * When called from the ULP ira_rill will be NULL hence the caller has to
7525  * pass in the ill.
7526  */
7527 /* ARGSUSED */
7528 void
ip_setl2src(mblk_t * mp,ip_recv_attr_t * ira,ill_t * ill)7529 ip_setl2src(mblk_t *mp, ip_recv_attr_t *ira, ill_t *ill)
7530 {
7531 	const uchar_t *addr;
7532 	int alen;
7533 
7534 	if (ira->ira_flags & IRAF_L2SRC_SET)
7535 		return;
7536 
7537 	ASSERT(ill != NULL);
7538 	alen = ill->ill_phys_addr_length;
7539 	ASSERT(alen <= sizeof (ira->ira_l2src));
7540 	if (ira->ira_mhip != NULL &&
7541 	    (addr = ira->ira_mhip->mhi_saddr) != NULL) {
7542 		bcopy(addr, ira->ira_l2src, alen);
7543 	} else if ((ira->ira_flags & IRAF_L2SRC_LOOPBACK) &&
7544 	    (addr = ill->ill_phys_addr) != NULL) {
7545 		bcopy(addr, ira->ira_l2src, alen);
7546 	} else {
7547 		bzero(ira->ira_l2src, alen);
7548 	}
7549 	ira->ira_flags |= IRAF_L2SRC_SET;
7550 }
7551 
7552 /*
7553  * check ip header length and align it.
7554  */
7555 mblk_t *
ip_check_and_align_header(mblk_t * mp,uint_t min_size,ip_recv_attr_t * ira)7556 ip_check_and_align_header(mblk_t *mp, uint_t min_size, ip_recv_attr_t *ira)
7557 {
7558 	ill_t	*ill = ira->ira_ill;
7559 	ssize_t len;
7560 
7561 	len = MBLKL(mp);
7562 
7563 	if (!OK_32PTR(mp->b_rptr))
7564 		IP_STAT(ill->ill_ipst, ip_notaligned);
7565 	else
7566 		IP_STAT(ill->ill_ipst, ip_recv_pullup);
7567 
7568 	/* Guard against bogus device drivers */
7569 	if (len < 0) {
7570 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7571 		ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7572 		freemsg(mp);
7573 		return (NULL);
7574 	}
7575 
7576 	if (len == 0) {
7577 		/* GLD sometimes sends up mblk with b_rptr == b_wptr! */
7578 		mblk_t *mp1 = mp->b_cont;
7579 
7580 		if (!(ira->ira_flags & IRAF_L2SRC_SET))
7581 			ip_setl2src(mp, ira, ira->ira_rill);
7582 		ASSERT(ira->ira_flags & IRAF_L2SRC_SET);
7583 
7584 		freeb(mp);
7585 		mp = mp1;
7586 		if (mp == NULL)
7587 			return (NULL);
7588 
7589 		if (OK_32PTR(mp->b_rptr) && MBLKL(mp) >= min_size)
7590 			return (mp);
7591 	}
7592 	if (ip_pullup(mp, min_size, ira) == NULL) {
7593 		if (msgdsize(mp) < min_size) {
7594 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7595 			ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7596 		} else {
7597 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7598 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
7599 		}
7600 		freemsg(mp);
7601 		return (NULL);
7602 	}
7603 	return (mp);
7604 }
7605 
7606 /*
7607  * Common code for IPv4 and IPv6 to check and pullup multi-mblks
7608  */
7609 mblk_t *
ip_check_length(mblk_t * mp,uchar_t * rptr,ssize_t len,uint_t pkt_len,uint_t min_size,ip_recv_attr_t * ira)7610 ip_check_length(mblk_t *mp, uchar_t *rptr, ssize_t len,	uint_t pkt_len,
7611     uint_t min_size, ip_recv_attr_t *ira)
7612 {
7613 	ill_t	*ill = ira->ira_ill;
7614 
7615 	/*
7616 	 * Make sure we have data length consistent
7617 	 * with the IP header.
7618 	 */
7619 	if (mp->b_cont == NULL) {
7620 		/* pkt_len is based on ipha_len, not the mblk length */
7621 		if (pkt_len < min_size) {
7622 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7623 			ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7624 			freemsg(mp);
7625 			return (NULL);
7626 		}
7627 		if (len < 0) {
7628 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
7629 			ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
7630 			freemsg(mp);
7631 			return (NULL);
7632 		}
7633 		/* Drop any pad */
7634 		mp->b_wptr = rptr + pkt_len;
7635 	} else if ((len += msgdsize(mp->b_cont)) != 0) {
7636 		ASSERT(pkt_len >= min_size);
7637 		if (pkt_len < min_size) {
7638 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7639 			ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7640 			freemsg(mp);
7641 			return (NULL);
7642 		}
7643 		if (len < 0) {
7644 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
7645 			ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
7646 			freemsg(mp);
7647 			return (NULL);
7648 		}
7649 		/* Drop any pad */
7650 		(void) adjmsg(mp, -len);
7651 		/*
7652 		 * adjmsg may have freed an mblk from the chain, hence
7653 		 * invalidate any hw checksum here. This will force IP to
7654 		 * calculate the checksum in sw, but only for this packet.
7655 		 */
7656 		DB_CKSUMFLAGS(mp) = 0;
7657 		IP_STAT(ill->ill_ipst, ip_multimblk);
7658 	}
7659 	return (mp);
7660 }
7661 
7662 /*
7663  * Check that the IPv4 opt_len is consistent with the packet and pullup
7664  * the options.
7665  */
7666 mblk_t *
ip_check_optlen(mblk_t * mp,ipha_t * ipha,uint_t opt_len,uint_t pkt_len,ip_recv_attr_t * ira)7667 ip_check_optlen(mblk_t *mp, ipha_t *ipha, uint_t opt_len, uint_t pkt_len,
7668     ip_recv_attr_t *ira)
7669 {
7670 	ill_t	*ill = ira->ira_ill;
7671 	ssize_t len;
7672 
7673 	/* Assume no IPv6 packets arrive over the IPv4 queue */
7674 	if (IPH_HDR_VERSION(ipha) != IPV4_VERSION) {
7675 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7676 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInWrongIPVersion);
7677 		ip_drop_input("IPvN packet on IPv4 ill", mp, ill);
7678 		freemsg(mp);
7679 		return (NULL);
7680 	}
7681 
7682 	if (opt_len > (15 - IP_SIMPLE_HDR_LENGTH_IN_WORDS)) {
7683 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7684 		ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7685 		freemsg(mp);
7686 		return (NULL);
7687 	}
7688 	/*
7689 	 * Recompute complete header length and make sure we
7690 	 * have access to all of it.
7691 	 */
7692 	len = ((size_t)opt_len + IP_SIMPLE_HDR_LENGTH_IN_WORDS) << 2;
7693 	if (len > (mp->b_wptr - mp->b_rptr)) {
7694 		if (len > pkt_len) {
7695 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7696 			ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7697 			freemsg(mp);
7698 			return (NULL);
7699 		}
7700 		if (ip_pullup(mp, len, ira) == NULL) {
7701 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7702 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
7703 			freemsg(mp);
7704 			return (NULL);
7705 		}
7706 	}
7707 	return (mp);
7708 }
7709 
7710 /*
7711  * Returns a new ire, or the same ire, or NULL.
7712  * If a different IRE is returned, then it is held; the caller
7713  * needs to release it.
7714  * In no case is there any hold/release on the ire argument.
7715  */
7716 ire_t *
ip_check_multihome(void * addr,ire_t * ire,ill_t * ill)7717 ip_check_multihome(void *addr, ire_t *ire, ill_t *ill)
7718 {
7719 	ire_t		*new_ire;
7720 	ill_t		*ire_ill;
7721 	uint_t		ifindex;
7722 	ip_stack_t	*ipst = ill->ill_ipst;
7723 	boolean_t	strict_check = B_FALSE;
7724 
7725 	/*
7726 	 * IPMP common case: if IRE and ILL are in the same group, there's no
7727 	 * issue (e.g. packet received on an underlying interface matched an
7728 	 * IRE_LOCAL on its associated group interface).
7729 	 */
7730 	ASSERT(ire->ire_ill != NULL);
7731 	if (IS_IN_SAME_ILLGRP(ill, ire->ire_ill))
7732 		return (ire);
7733 
7734 	/*
7735 	 * Do another ire lookup here, using the ingress ill, to see if the
7736 	 * interface is in a usesrc group.
7737 	 * As long as the ills belong to the same group, we don't consider
7738 	 * them to be arriving on the wrong interface. Thus, if the switch
7739 	 * is doing inbound load spreading, we won't drop packets when the
7740 	 * ip*_strict_dst_multihoming switch is on.
7741 	 * We also need to check for IPIF_UNNUMBERED point2point interfaces
7742 	 * where the local address may not be unique. In this case we were
7743 	 * at the mercy of the initial ire lookup and the IRE_LOCAL it
7744 	 * actually returned. The new lookup, which is more specific, should
7745 	 * only find the IRE_LOCAL associated with the ingress ill if one
7746 	 * exists.
7747 	 */
7748 	if (ire->ire_ipversion == IPV4_VERSION) {
7749 		if (ipst->ips_ip_strict_dst_multihoming)
7750 			strict_check = B_TRUE;
7751 		new_ire = ire_ftable_lookup_v4(*((ipaddr_t *)addr), 0, 0,
7752 		    IRE_LOCAL, ill, ALL_ZONES, NULL,
7753 		    (MATCH_IRE_TYPE|MATCH_IRE_ILL), 0, ipst, NULL);
7754 	} else {
7755 		ASSERT(!IN6_IS_ADDR_MULTICAST((in6_addr_t *)addr));
7756 		if (ipst->ips_ipv6_strict_dst_multihoming)
7757 			strict_check = B_TRUE;
7758 		new_ire = ire_ftable_lookup_v6((in6_addr_t *)addr, NULL, NULL,
7759 		    IRE_LOCAL, ill, ALL_ZONES, NULL,
7760 		    (MATCH_IRE_TYPE|MATCH_IRE_ILL), 0, ipst, NULL);
7761 	}
7762 	/*
7763 	 * If the same ire that was returned in ip_input() is found then this
7764 	 * is an indication that usesrc groups are in use. The packet
7765 	 * arrived on a different ill in the group than the one associated with
7766 	 * the destination address.  If a different ire was found then the same
7767 	 * IP address must be hosted on multiple ills. This is possible with
7768 	 * unnumbered point2point interfaces. We switch to use this new ire in
7769 	 * order to have accurate interface statistics.
7770 	 */
7771 	if (new_ire != NULL) {
7772 		/* Note: held in one case but not the other? Caller handles */
7773 		if (new_ire != ire)
7774 			return (new_ire);
7775 		/* Unchanged */
7776 		ire_refrele(new_ire);
7777 		return (ire);
7778 	}
7779 
7780 	/*
7781 	 * Chase pointers once and store locally.
7782 	 */
7783 	ASSERT(ire->ire_ill != NULL);
7784 	ire_ill = ire->ire_ill;
7785 	ifindex = ill->ill_usesrc_ifindex;
7786 
7787 	/*
7788 	 * Check if it's a legal address on the 'usesrc' interface.
7789 	 * For IPMP data addresses the IRE_LOCAL is the upper, hence we
7790 	 * can just check phyint_ifindex.
7791 	 */
7792 	if (ifindex != 0 && ifindex == ire_ill->ill_phyint->phyint_ifindex) {
7793 		return (ire);
7794 	}
7795 
7796 	/*
7797 	 * If the ip*_strict_dst_multihoming switch is on then we can
7798 	 * only accept this packet if the interface is marked as routing.
7799 	 */
7800 	if (!(strict_check))
7801 		return (ire);
7802 
7803 	if ((ill->ill_flags & ire->ire_ill->ill_flags & ILLF_ROUTER) != 0) {
7804 		return (ire);
7805 	}
7806 	return (NULL);
7807 }
7808 
7809 /*
7810  * This function is used to construct a mac_header_info_s from a
7811  * DL_UNITDATA_IND message.
7812  * The address fields in the mhi structure points into the message,
7813  * thus the caller can't use those fields after freeing the message.
7814  *
7815  * We determine whether the packet received is a non-unicast packet
7816  * and in doing so, determine whether or not it is broadcast vs multicast.
7817  * For it to be a broadcast packet, we must have the appropriate mblk_t
7818  * hanging off the ill_t.  If this is either not present or doesn't match
7819  * the destination mac address in the DL_UNITDATA_IND, the packet is deemed
7820  * to be multicast.  Thus NICs that have no broadcast address (or no
7821  * capability for one, such as point to point links) cannot return as
7822  * the packet being broadcast.
7823  */
7824 void
ip_dlur_to_mhi(ill_t * ill,mblk_t * mb,struct mac_header_info_s * mhip)7825 ip_dlur_to_mhi(ill_t *ill, mblk_t *mb, struct mac_header_info_s *mhip)
7826 {
7827 	dl_unitdata_ind_t *ind = (dl_unitdata_ind_t *)mb->b_rptr;
7828 	mblk_t *bmp;
7829 	uint_t extra_offset;
7830 
7831 	bzero(mhip, sizeof (struct mac_header_info_s));
7832 
7833 	mhip->mhi_dsttype = MAC_ADDRTYPE_UNICAST;
7834 
7835 	if (ill->ill_sap_length < 0)
7836 		extra_offset = 0;
7837 	else
7838 		extra_offset = ill->ill_sap_length;
7839 
7840 	mhip->mhi_daddr = (uchar_t *)ind + ind->dl_dest_addr_offset +
7841 	    extra_offset;
7842 	mhip->mhi_saddr = (uchar_t *)ind + ind->dl_src_addr_offset +
7843 	    extra_offset;
7844 
7845 	if (!ind->dl_group_address)
7846 		return;
7847 
7848 	/* Multicast or broadcast */
7849 	mhip->mhi_dsttype = MAC_ADDRTYPE_MULTICAST;
7850 
7851 	if (ind->dl_dest_addr_offset > sizeof (*ind) &&
7852 	    ind->dl_dest_addr_offset + ind->dl_dest_addr_length < MBLKL(mb) &&
7853 	    (bmp = ill->ill_bcast_mp) != NULL) {
7854 		dl_unitdata_req_t *dlur;
7855 		uint8_t *bphys_addr;
7856 
7857 		dlur = (dl_unitdata_req_t *)bmp->b_rptr;
7858 		bphys_addr = (uchar_t *)dlur + dlur->dl_dest_addr_offset +
7859 		    extra_offset;
7860 
7861 		if (bcmp(mhip->mhi_daddr, bphys_addr,
7862 		    ind->dl_dest_addr_length) == 0)
7863 			mhip->mhi_dsttype = MAC_ADDRTYPE_BROADCAST;
7864 	}
7865 }
7866 
7867 /*
7868  * This function is used to construct a mac_header_info_s from a
7869  * M_DATA fastpath message from a DLPI driver.
7870  * The address fields in the mhi structure points into the message,
7871  * thus the caller can't use those fields after freeing the message.
7872  *
7873  * We determine whether the packet received is a non-unicast packet
7874  * and in doing so, determine whether or not it is broadcast vs multicast.
7875  * For it to be a broadcast packet, we must have the appropriate mblk_t
7876  * hanging off the ill_t.  If this is either not present or doesn't match
7877  * the destination mac address in the DL_UNITDATA_IND, the packet is deemed
7878  * to be multicast.  Thus NICs that have no broadcast address (or no
7879  * capability for one, such as point to point links) cannot return as
7880  * the packet being broadcast.
7881  */
7882 void
ip_mdata_to_mhi(ill_t * ill,mblk_t * mp,struct mac_header_info_s * mhip)7883 ip_mdata_to_mhi(ill_t *ill, mblk_t *mp, struct mac_header_info_s *mhip)
7884 {
7885 	mblk_t *bmp;
7886 	struct ether_header *pether;
7887 
7888 	bzero(mhip, sizeof (struct mac_header_info_s));
7889 
7890 	mhip->mhi_dsttype = MAC_ADDRTYPE_UNICAST;
7891 
7892 	pether = (struct ether_header *)((char *)mp->b_rptr
7893 	    - sizeof (struct ether_header));
7894 
7895 	/*
7896 	 * Make sure the interface is an ethernet type, since we don't
7897 	 * know the header format for anything but Ethernet. Also make
7898 	 * sure we are pointing correctly above db_base.
7899 	 */
7900 	if (ill->ill_type != IFT_ETHER)
7901 		return;
7902 
7903 retry:
7904 	if ((uchar_t *)pether < mp->b_datap->db_base)
7905 		return;
7906 
7907 	/* Is there a VLAN tag? */
7908 	if (ill->ill_isv6) {
7909 		if (pether->ether_type != htons(ETHERTYPE_IPV6)) {
7910 			pether = (struct ether_header *)((char *)pether - 4);
7911 			goto retry;
7912 		}
7913 	} else {
7914 		if (pether->ether_type != htons(ETHERTYPE_IP)) {
7915 			pether = (struct ether_header *)((char *)pether - 4);
7916 			goto retry;
7917 		}
7918 	}
7919 	mhip->mhi_daddr = (uchar_t *)&pether->ether_dhost;
7920 	mhip->mhi_saddr = (uchar_t *)&pether->ether_shost;
7921 
7922 	if (!(mhip->mhi_daddr[0] & 0x01))
7923 		return;
7924 
7925 	/* Multicast or broadcast */
7926 	mhip->mhi_dsttype = MAC_ADDRTYPE_MULTICAST;
7927 
7928 	if ((bmp = ill->ill_bcast_mp) != NULL) {
7929 		dl_unitdata_req_t *dlur;
7930 		uint8_t *bphys_addr;
7931 		uint_t	addrlen;
7932 
7933 		dlur = (dl_unitdata_req_t *)bmp->b_rptr;
7934 		addrlen = dlur->dl_dest_addr_length;
7935 		if (ill->ill_sap_length < 0) {
7936 			bphys_addr = (uchar_t *)dlur +
7937 			    dlur->dl_dest_addr_offset;
7938 			addrlen += ill->ill_sap_length;
7939 		} else {
7940 			bphys_addr = (uchar_t *)dlur +
7941 			    dlur->dl_dest_addr_offset +
7942 			    ill->ill_sap_length;
7943 			addrlen -= ill->ill_sap_length;
7944 		}
7945 		if (bcmp(mhip->mhi_daddr, bphys_addr, addrlen) == 0)
7946 			mhip->mhi_dsttype = MAC_ADDRTYPE_BROADCAST;
7947 	}
7948 }
7949 
7950 /*
7951  * Handle anything but M_DATA messages
7952  * We see the DL_UNITDATA_IND which are part
7953  * of the data path, and also the other messages from the driver.
7954  */
7955 void
ip_rput_notdata(ill_t * ill,mblk_t * mp)7956 ip_rput_notdata(ill_t *ill, mblk_t *mp)
7957 {
7958 	mblk_t		*first_mp;
7959 	struct iocblk   *iocp;
7960 	struct mac_header_info_s mhi;
7961 
7962 	switch (DB_TYPE(mp)) {
7963 	case M_PROTO:
7964 	case M_PCPROTO: {
7965 		if (((dl_unitdata_ind_t *)mp->b_rptr)->dl_primitive !=
7966 		    DL_UNITDATA_IND) {
7967 			/* Go handle anything other than data elsewhere. */
7968 			ip_rput_dlpi(ill, mp);
7969 			return;
7970 		}
7971 
7972 		first_mp = mp;
7973 		mp = first_mp->b_cont;
7974 		first_mp->b_cont = NULL;
7975 
7976 		if (mp == NULL) {
7977 			freeb(first_mp);
7978 			return;
7979 		}
7980 		ip_dlur_to_mhi(ill, first_mp, &mhi);
7981 		if (ill->ill_isv6)
7982 			ip_input_v6(ill, NULL, mp, &mhi);
7983 		else
7984 			ip_input(ill, NULL, mp, &mhi);
7985 
7986 		/* Ditch the DLPI header. */
7987 		freeb(first_mp);
7988 		return;
7989 	}
7990 	case M_IOCACK:
7991 		iocp = (struct iocblk *)mp->b_rptr;
7992 		switch (iocp->ioc_cmd) {
7993 		case DL_IOC_HDR_INFO:
7994 			ill_fastpath_ack(ill, mp);
7995 			return;
7996 		default:
7997 			putnext(ill->ill_rq, mp);
7998 			return;
7999 		}
8000 		/* FALLTHROUGH */
8001 	case M_ERROR:
8002 	case M_HANGUP:
8003 		mutex_enter(&ill->ill_lock);
8004 		if (ill->ill_state_flags & ILL_CONDEMNED) {
8005 			mutex_exit(&ill->ill_lock);
8006 			freemsg(mp);
8007 			return;
8008 		}
8009 		ill_refhold_locked(ill);
8010 		mutex_exit(&ill->ill_lock);
8011 		qwriter_ip(ill, ill->ill_rq, mp, ip_rput_other, CUR_OP,
8012 		    B_FALSE);
8013 		return;
8014 	case M_CTL:
8015 		putnext(ill->ill_rq, mp);
8016 		return;
8017 	case M_IOCNAK:
8018 		ip1dbg(("got iocnak "));
8019 		iocp = (struct iocblk *)mp->b_rptr;
8020 		switch (iocp->ioc_cmd) {
8021 		case DL_IOC_HDR_INFO:
8022 			ip_rput_other(NULL, ill->ill_rq, mp, NULL);
8023 			return;
8024 		default:
8025 			break;
8026 		}
8027 		/* FALLTHROUGH */
8028 	default:
8029 		putnext(ill->ill_rq, mp);
8030 		return;
8031 	}
8032 }
8033 
8034 /* Read side put procedure.  Packets coming from the wire arrive here. */
8035 int
ip_rput(queue_t * q,mblk_t * mp)8036 ip_rput(queue_t *q, mblk_t *mp)
8037 {
8038 	ill_t	*ill;
8039 	union DL_primitives *dl;
8040 
8041 	ill = (ill_t *)q->q_ptr;
8042 
8043 	if (ill->ill_state_flags & (ILL_CONDEMNED | ILL_LL_SUBNET_PENDING)) {
8044 		/*
8045 		 * If things are opening or closing, only accept high-priority
8046 		 * DLPI messages.  (On open ill->ill_ipif has not yet been
8047 		 * created; on close, things hanging off the ill may have been
8048 		 * freed already.)
8049 		 */
8050 		dl = (union DL_primitives *)mp->b_rptr;
8051 		if (DB_TYPE(mp) != M_PCPROTO ||
8052 		    dl->dl_primitive == DL_UNITDATA_IND) {
8053 			inet_freemsg(mp);
8054 			return (0);
8055 		}
8056 	}
8057 	if (DB_TYPE(mp) == M_DATA) {
8058 		struct mac_header_info_s mhi;
8059 
8060 		ip_mdata_to_mhi(ill, mp, &mhi);
8061 		ip_input(ill, NULL, mp, &mhi);
8062 	} else {
8063 		ip_rput_notdata(ill, mp);
8064 	}
8065 	return (0);
8066 }
8067 
8068 /*
8069  * Move the information to a copy.
8070  */
8071 mblk_t *
ip_fix_dbref(mblk_t * mp,ip_recv_attr_t * ira)8072 ip_fix_dbref(mblk_t *mp, ip_recv_attr_t *ira)
8073 {
8074 	mblk_t		*mp1;
8075 	ill_t		*ill = ira->ira_ill;
8076 	ip_stack_t	*ipst = ill->ill_ipst;
8077 
8078 	IP_STAT(ipst, ip_db_ref);
8079 
8080 	/* Make sure we have ira_l2src before we loose the original mblk */
8081 	if (!(ira->ira_flags & IRAF_L2SRC_SET))
8082 		ip_setl2src(mp, ira, ira->ira_rill);
8083 
8084 	mp1 = copymsg(mp);
8085 	if (mp1 == NULL) {
8086 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
8087 		ip_drop_input("ipIfStatsInDiscards", mp, ill);
8088 		freemsg(mp);
8089 		return (NULL);
8090 	}
8091 	/* preserve the hardware checksum flags and data, if present */
8092 	if (DB_CKSUMFLAGS(mp) != 0) {
8093 		DB_CKSUMFLAGS(mp1) = DB_CKSUMFLAGS(mp);
8094 		DB_CKSUMSTART(mp1) = DB_CKSUMSTART(mp);
8095 		DB_CKSUMSTUFF(mp1) = DB_CKSUMSTUFF(mp);
8096 		DB_CKSUMEND(mp1) = DB_CKSUMEND(mp);
8097 		DB_CKSUM16(mp1) = DB_CKSUM16(mp);
8098 	}
8099 	freemsg(mp);
8100 	return (mp1);
8101 }
8102 
8103 static void
ip_dlpi_error(ill_t * ill,t_uscalar_t prim,t_uscalar_t dl_err,t_uscalar_t err)8104 ip_dlpi_error(ill_t *ill, t_uscalar_t prim, t_uscalar_t dl_err,
8105     t_uscalar_t err)
8106 {
8107 	if (dl_err == DL_SYSERR) {
8108 		(void) mi_strlog(ill->ill_rq, 1, SL_CONSOLE|SL_ERROR|SL_TRACE,
8109 		    "%s: %s failed: DL_SYSERR (errno %u)\n",
8110 		    ill->ill_name, dl_primstr(prim), err);
8111 		return;
8112 	}
8113 
8114 	(void) mi_strlog(ill->ill_rq, 1, SL_CONSOLE|SL_ERROR|SL_TRACE,
8115 	    "%s: %s failed: %s\n", ill->ill_name, dl_primstr(prim),
8116 	    dl_errstr(dl_err));
8117 }
8118 
8119 /*
8120  * ip_rput_dlpi is called by ip_rput to handle all DLPI messages other
8121  * than DL_UNITDATA_IND messages. If we need to process this message
8122  * exclusively, we call qwriter_ip, in which case we also need to call
8123  * ill_refhold before that, since qwriter_ip does an ill_refrele.
8124  */
8125 void
ip_rput_dlpi(ill_t * ill,mblk_t * mp)8126 ip_rput_dlpi(ill_t *ill, mblk_t *mp)
8127 {
8128 	dl_ok_ack_t	*dloa = (dl_ok_ack_t *)mp->b_rptr;
8129 	dl_error_ack_t	*dlea = (dl_error_ack_t *)dloa;
8130 	queue_t		*q = ill->ill_rq;
8131 	t_uscalar_t	prim = dloa->dl_primitive;
8132 	t_uscalar_t	reqprim = DL_PRIM_INVAL;
8133 
8134 	DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi",
8135 	    char *, dl_primstr(prim), ill_t *, ill);
8136 	ip1dbg(("ip_rput_dlpi"));
8137 
8138 	/*
8139 	 * If we received an ACK but didn't send a request for it, then it
8140 	 * can't be part of any pending operation; discard up-front.
8141 	 */
8142 	switch (prim) {
8143 	case DL_ERROR_ACK:
8144 		reqprim = dlea->dl_error_primitive;
8145 		ip2dbg(("ip_rput_dlpi(%s): DL_ERROR_ACK for %s (0x%x): %s "
8146 		    "(0x%x), unix %u\n", ill->ill_name, dl_primstr(reqprim),
8147 		    reqprim, dl_errstr(dlea->dl_errno), dlea->dl_errno,
8148 		    dlea->dl_unix_errno));
8149 		break;
8150 	case DL_OK_ACK:
8151 		reqprim = dloa->dl_correct_primitive;
8152 		break;
8153 	case DL_INFO_ACK:
8154 		reqprim = DL_INFO_REQ;
8155 		break;
8156 	case DL_BIND_ACK:
8157 		reqprim = DL_BIND_REQ;
8158 		break;
8159 	case DL_PHYS_ADDR_ACK:
8160 		reqprim = DL_PHYS_ADDR_REQ;
8161 		break;
8162 	case DL_NOTIFY_ACK:
8163 		reqprim = DL_NOTIFY_REQ;
8164 		break;
8165 	case DL_CAPABILITY_ACK:
8166 		reqprim = DL_CAPABILITY_REQ;
8167 		break;
8168 	}
8169 
8170 	if (prim != DL_NOTIFY_IND) {
8171 		if (reqprim == DL_PRIM_INVAL ||
8172 		    !ill_dlpi_pending(ill, reqprim)) {
8173 			/* Not a DLPI message we support or expected */
8174 			freemsg(mp);
8175 			return;
8176 		}
8177 		ip1dbg(("ip_rput: received %s for %s\n", dl_primstr(prim),
8178 		    dl_primstr(reqprim)));
8179 	}
8180 
8181 	switch (reqprim) {
8182 	case DL_UNBIND_REQ:
8183 		/*
8184 		 * NOTE: we mark the unbind as complete even if we got a
8185 		 * DL_ERROR_ACK, since there's not much else we can do.
8186 		 */
8187 		mutex_enter(&ill->ill_lock);
8188 		ill->ill_state_flags &= ~ILL_DL_UNBIND_IN_PROGRESS;
8189 		cv_signal(&ill->ill_cv);
8190 		mutex_exit(&ill->ill_lock);
8191 		break;
8192 
8193 	case DL_ENABMULTI_REQ:
8194 		if (prim == DL_OK_ACK) {
8195 			if (ill->ill_dlpi_multicast_state == IDS_INPROGRESS)
8196 				ill->ill_dlpi_multicast_state = IDS_OK;
8197 		}
8198 		break;
8199 	}
8200 
8201 	/*
8202 	 * The message is one we're waiting for (or DL_NOTIFY_IND), but we
8203 	 * need to become writer to continue to process it.  Because an
8204 	 * exclusive operation doesn't complete until replies to all queued
8205 	 * DLPI messages have been received, we know we're in the middle of an
8206 	 * exclusive operation and pass CUR_OP (except for DL_NOTIFY_IND).
8207 	 *
8208 	 * As required by qwriter_ip(), we refhold the ill; it will refrele.
8209 	 * Since this is on the ill stream we unconditionally bump up the
8210 	 * refcount without doing ILL_CAN_LOOKUP().
8211 	 */
8212 	ill_refhold(ill);
8213 	if (prim == DL_NOTIFY_IND)
8214 		qwriter_ip(ill, q, mp, ip_rput_dlpi_writer, NEW_OP, B_FALSE);
8215 	else
8216 		qwriter_ip(ill, q, mp, ip_rput_dlpi_writer, CUR_OP, B_FALSE);
8217 }
8218 
8219 /*
8220  * Handling of DLPI messages that require exclusive access to the ipsq.
8221  *
8222  * Need to do ipsq_pending_mp_get on ioctl completion, which could
8223  * happen here. (along with mi_copy_done)
8224  */
8225 /* ARGSUSED */
8226 static void
ip_rput_dlpi_writer(ipsq_t * ipsq,queue_t * q,mblk_t * mp,void * dummy_arg)8227 ip_rput_dlpi_writer(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
8228 {
8229 	dl_ok_ack_t	*dloa = (dl_ok_ack_t *)mp->b_rptr;
8230 	dl_error_ack_t	*dlea = (dl_error_ack_t *)dloa;
8231 	int		err = 0;
8232 	ill_t		*ill = (ill_t *)q->q_ptr;
8233 	ipif_t		*ipif = NULL;
8234 	mblk_t		*mp1 = NULL;
8235 	conn_t		*connp = NULL;
8236 	t_uscalar_t	paddrreq;
8237 	mblk_t		*mp_hw;
8238 	boolean_t	success;
8239 	boolean_t	ioctl_aborted = B_FALSE;
8240 	boolean_t	log = B_TRUE;
8241 
8242 	DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer",
8243 	    char *, dl_primstr(dloa->dl_primitive), ill_t *, ill);
8244 
8245 	ip1dbg(("ip_rput_dlpi_writer .."));
8246 	ASSERT(ipsq->ipsq_xop == ill->ill_phyint->phyint_ipsq->ipsq_xop);
8247 	ASSERT(IAM_WRITER_ILL(ill));
8248 
8249 	ipif = ipsq->ipsq_xop->ipx_pending_ipif;
8250 	/*
8251 	 * The current ioctl could have been aborted by the user and a new
8252 	 * ioctl to bring up another ill could have started. We could still
8253 	 * get a response from the driver later.
8254 	 */
8255 	if (ipif != NULL && ipif->ipif_ill != ill)
8256 		ioctl_aborted = B_TRUE;
8257 
8258 	switch (dloa->dl_primitive) {
8259 	case DL_ERROR_ACK:
8260 		ip1dbg(("ip_rput_dlpi_writer: got DL_ERROR_ACK for %s\n",
8261 		    dl_primstr(dlea->dl_error_primitive)));
8262 
8263 		DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer error",
8264 		    char *, dl_primstr(dlea->dl_error_primitive),
8265 		    ill_t *, ill);
8266 
8267 		switch (dlea->dl_error_primitive) {
8268 		case DL_DISABMULTI_REQ:
8269 			ill_dlpi_done(ill, dlea->dl_error_primitive);
8270 			break;
8271 		case DL_PROMISCON_REQ:
8272 		case DL_PROMISCOFF_REQ:
8273 		case DL_UNBIND_REQ:
8274 		case DL_ATTACH_REQ:
8275 		case DL_INFO_REQ:
8276 			ill_dlpi_done(ill, dlea->dl_error_primitive);
8277 			break;
8278 		case DL_NOTIFY_REQ:
8279 			ill_dlpi_done(ill, DL_NOTIFY_REQ);
8280 			log = B_FALSE;
8281 			break;
8282 		case DL_PHYS_ADDR_REQ:
8283 			/*
8284 			 * For IPv6 only, there are two additional
8285 			 * phys_addr_req's sent to the driver to get the
8286 			 * IPv6 token and lla. This allows IP to acquire
8287 			 * the hardware address format for a given interface
8288 			 * without having built in knowledge of the hardware
8289 			 * address. ill_phys_addr_pend keeps track of the last
8290 			 * DL_PAR sent so we know which response we are
8291 			 * dealing with. ill_dlpi_done will update
8292 			 * ill_phys_addr_pend when it sends the next req.
8293 			 * We don't complete the IOCTL until all three DL_PARs
8294 			 * have been attempted, so set *_len to 0 and break.
8295 			 */
8296 			paddrreq = ill->ill_phys_addr_pend;
8297 			ill_dlpi_done(ill, DL_PHYS_ADDR_REQ);
8298 			if (paddrreq == DL_IPV6_TOKEN) {
8299 				ill->ill_token_length = 0;
8300 				log = B_FALSE;
8301 				break;
8302 			} else if (paddrreq == DL_IPV6_LINK_LAYER_ADDR) {
8303 				ill->ill_nd_lla_len = 0;
8304 				log = B_FALSE;
8305 				break;
8306 			}
8307 			/*
8308 			 * Something went wrong with the DL_PHYS_ADDR_REQ.
8309 			 * We presumably have an IOCTL hanging out waiting
8310 			 * for completion. Find it and complete the IOCTL
8311 			 * with the error noted.
8312 			 * However, ill_dl_phys was called on an ill queue
8313 			 * (from SIOCSLIFNAME), thus conn_pending_ill is not
8314 			 * set. But the ioctl is known to be pending on ill_wq.
8315 			 */
8316 			if (!ill->ill_ifname_pending)
8317 				break;
8318 			ill->ill_ifname_pending = 0;
8319 			if (!ioctl_aborted)
8320 				mp1 = ipsq_pending_mp_get(ipsq, &connp);
8321 			if (mp1 != NULL) {
8322 				/*
8323 				 * This operation (SIOCSLIFNAME) must have
8324 				 * happened on the ill. Assert there is no conn
8325 				 */
8326 				ASSERT(connp == NULL);
8327 				q = ill->ill_wq;
8328 			}
8329 			break;
8330 		case DL_BIND_REQ:
8331 			ill_dlpi_done(ill, DL_BIND_REQ);
8332 			if (ill->ill_ifname_pending)
8333 				break;
8334 			mutex_enter(&ill->ill_lock);
8335 			ill->ill_state_flags &= ~ILL_DOWN_IN_PROGRESS;
8336 			mutex_exit(&ill->ill_lock);
8337 			/*
8338 			 * Something went wrong with the bind.  We presumably
8339 			 * have an IOCTL hanging out waiting for completion.
8340 			 * Find it, take down the interface that was coming
8341 			 * up, and complete the IOCTL with the error noted.
8342 			 */
8343 			if (!ioctl_aborted)
8344 				mp1 = ipsq_pending_mp_get(ipsq, &connp);
8345 			if (mp1 != NULL) {
8346 				/*
8347 				 * This might be a result of a DL_NOTE_REPLUMB
8348 				 * notification. In that case, connp is NULL.
8349 				 */
8350 				if (connp != NULL)
8351 					q = CONNP_TO_WQ(connp);
8352 
8353 				(void) ipif_down(ipif, NULL, NULL);
8354 				/* error is set below the switch */
8355 			}
8356 			break;
8357 		case DL_ENABMULTI_REQ:
8358 			ill_dlpi_done(ill, DL_ENABMULTI_REQ);
8359 
8360 			if (ill->ill_dlpi_multicast_state == IDS_INPROGRESS)
8361 				ill->ill_dlpi_multicast_state = IDS_FAILED;
8362 			if (ill->ill_dlpi_multicast_state == IDS_FAILED) {
8363 
8364 				printf("ip: joining multicasts failed (%d)"
8365 				    " on %s - will use link layer "
8366 				    "broadcasts for multicast\n",
8367 				    dlea->dl_errno, ill->ill_name);
8368 
8369 				/*
8370 				 * Set up for multi_bcast; We are the
8371 				 * writer, so ok to access ill->ill_ipif
8372 				 * without any lock.
8373 				 */
8374 				mutex_enter(&ill->ill_phyint->phyint_lock);
8375 				ill->ill_phyint->phyint_flags |=
8376 				    PHYI_MULTI_BCAST;
8377 				mutex_exit(&ill->ill_phyint->phyint_lock);
8378 
8379 			}
8380 			freemsg(mp);	/* Don't want to pass this up */
8381 			return;
8382 		case DL_CAPABILITY_REQ:
8383 			ip1dbg(("ip_rput_dlpi_writer: got DL_ERROR_ACK for "
8384 			    "DL_CAPABILITY REQ\n"));
8385 			if (ill->ill_dlpi_capab_state == IDCS_PROBE_SENT)
8386 				ill->ill_dlpi_capab_state = IDCS_FAILED;
8387 			ill_capability_done(ill);
8388 			freemsg(mp);
8389 			return;
8390 		}
8391 		/*
8392 		 * Note the error for IOCTL completion (mp1 is set when
8393 		 * ready to complete ioctl). If ill_ifname_pending_err is
8394 		 * set, an error occured during plumbing (ill_ifname_pending),
8395 		 * so we want to report that error.
8396 		 *
8397 		 * NOTE: there are two addtional DL_PHYS_ADDR_REQ's
8398 		 * (DL_IPV6_TOKEN and DL_IPV6_LINK_LAYER_ADDR) that are
8399 		 * expected to get errack'd if the driver doesn't support
8400 		 * these flags (e.g. ethernet). log will be set to B_FALSE
8401 		 * if these error conditions are encountered.
8402 		 */
8403 		if (mp1 != NULL) {
8404 			if (ill->ill_ifname_pending_err != 0)  {
8405 				err = ill->ill_ifname_pending_err;
8406 				ill->ill_ifname_pending_err = 0;
8407 			} else {
8408 				err = dlea->dl_unix_errno ?
8409 				    dlea->dl_unix_errno : ENXIO;
8410 			}
8411 		/*
8412 		 * If we're plumbing an interface and an error hasn't already
8413 		 * been saved, set ill_ifname_pending_err to the error passed
8414 		 * up. Ignore the error if log is B_FALSE (see comment above).
8415 		 */
8416 		} else if (log && ill->ill_ifname_pending &&
8417 		    ill->ill_ifname_pending_err == 0) {
8418 			ill->ill_ifname_pending_err = dlea->dl_unix_errno ?
8419 			    dlea->dl_unix_errno : ENXIO;
8420 		}
8421 
8422 		if (log)
8423 			ip_dlpi_error(ill, dlea->dl_error_primitive,
8424 			    dlea->dl_errno, dlea->dl_unix_errno);
8425 		break;
8426 	case DL_CAPABILITY_ACK:
8427 		ill_capability_ack(ill, mp);
8428 		/*
8429 		 * The message has been handed off to ill_capability_ack
8430 		 * and must not be freed below
8431 		 */
8432 		mp = NULL;
8433 		break;
8434 
8435 	case DL_INFO_ACK:
8436 		/* Call a routine to handle this one. */
8437 		ill_dlpi_done(ill, DL_INFO_REQ);
8438 		ip_ll_subnet_defaults(ill, mp);
8439 		ASSERT(!MUTEX_HELD(&ill->ill_phyint->phyint_ipsq->ipsq_lock));
8440 		return;
8441 	case DL_BIND_ACK:
8442 		/*
8443 		 * We should have an IOCTL waiting on this unless
8444 		 * sent by ill_dl_phys, in which case just return
8445 		 */
8446 		ill_dlpi_done(ill, DL_BIND_REQ);
8447 
8448 		if (ill->ill_ifname_pending) {
8449 			DTRACE_PROBE2(ip__rput__dlpi__ifname__pending,
8450 			    ill_t *, ill, mblk_t *, mp);
8451 			break;
8452 		}
8453 		mutex_enter(&ill->ill_lock);
8454 		ill->ill_dl_up = 1;
8455 		ill->ill_state_flags &= ~ILL_DOWN_IN_PROGRESS;
8456 		mutex_exit(&ill->ill_lock);
8457 
8458 		if (!ioctl_aborted)
8459 			mp1 = ipsq_pending_mp_get(ipsq, &connp);
8460 		if (mp1 == NULL) {
8461 			DTRACE_PROBE1(ip__rput__dlpi__no__mblk, ill_t *, ill);
8462 			break;
8463 		}
8464 		/*
8465 		 * mp1 was added by ill_dl_up(). if that is a result of
8466 		 * a DL_NOTE_REPLUMB notification, connp could be NULL.
8467 		 */
8468 		if (connp != NULL)
8469 			q = CONNP_TO_WQ(connp);
8470 		/*
8471 		 * We are exclusive. So nothing can change even after
8472 		 * we get the pending mp.
8473 		 */
8474 		ip1dbg(("ip_rput_dlpi: bind_ack %s\n", ill->ill_name));
8475 		DTRACE_PROBE1(ip__rput__dlpi__bind__ack, ill_t *, ill);
8476 		ill_nic_event_dispatch(ill, 0, NE_UP, NULL, 0);
8477 
8478 		/*
8479 		 * Now bring up the resolver; when that is complete, we'll
8480 		 * create IREs.  Note that we intentionally mirror what
8481 		 * ipif_up() would have done, because we got here by way of
8482 		 * ill_dl_up(), which stopped ipif_up()'s processing.
8483 		 */
8484 		if (ill->ill_isv6) {
8485 			/*
8486 			 * v6 interfaces.
8487 			 * Unlike ARP which has to do another bind
8488 			 * and attach, once we get here we are
8489 			 * done with NDP
8490 			 */
8491 			(void) ipif_resolver_up(ipif, Res_act_initial);
8492 			if ((err = ipif_ndp_up(ipif, B_TRUE)) == 0)
8493 				err = ipif_up_done_v6(ipif);
8494 		} else if (ill->ill_net_type == IRE_IF_RESOLVER) {
8495 			/*
8496 			 * ARP and other v4 external resolvers.
8497 			 * Leave the pending mblk intact so that
8498 			 * the ioctl completes in ip_rput().
8499 			 */
8500 			if (connp != NULL)
8501 				mutex_enter(&connp->conn_lock);
8502 			mutex_enter(&ill->ill_lock);
8503 			success = ipsq_pending_mp_add(connp, ipif, q, mp1, 0);
8504 			mutex_exit(&ill->ill_lock);
8505 			if (connp != NULL)
8506 				mutex_exit(&connp->conn_lock);
8507 			if (success) {
8508 				err = ipif_resolver_up(ipif, Res_act_initial);
8509 				if (err == EINPROGRESS) {
8510 					freemsg(mp);
8511 					return;
8512 				}
8513 				mp1 = ipsq_pending_mp_get(ipsq, &connp);
8514 			} else {
8515 				/* The conn has started closing */
8516 				err = EINTR;
8517 			}
8518 		} else {
8519 			/*
8520 			 * This one is complete. Reply to pending ioctl.
8521 			 */
8522 			(void) ipif_resolver_up(ipif, Res_act_initial);
8523 			err = ipif_up_done(ipif);
8524 		}
8525 
8526 		if ((err == 0) && (ill->ill_up_ipifs)) {
8527 			err = ill_up_ipifs(ill, q, mp1);
8528 			if (err == EINPROGRESS) {
8529 				freemsg(mp);
8530 				return;
8531 			}
8532 		}
8533 
8534 		/*
8535 		 * If we have a moved ipif to bring up, and everything has
8536 		 * succeeded to this point, bring it up on the IPMP ill.
8537 		 * Otherwise, leave it down -- the admin can try to bring it
8538 		 * up by hand if need be.
8539 		 */
8540 		if (ill->ill_move_ipif != NULL) {
8541 			if (err != 0) {
8542 				ill->ill_move_ipif = NULL;
8543 			} else {
8544 				ipif = ill->ill_move_ipif;
8545 				ill->ill_move_ipif = NULL;
8546 				err = ipif_up(ipif, q, mp1);
8547 				if (err == EINPROGRESS) {
8548 					freemsg(mp);
8549 					return;
8550 				}
8551 			}
8552 		}
8553 		break;
8554 
8555 	case DL_NOTIFY_IND: {
8556 		dl_notify_ind_t *notify = (dl_notify_ind_t *)mp->b_rptr;
8557 		uint_t orig_mtu, orig_mc_mtu;
8558 
8559 		switch (notify->dl_notification) {
8560 		case DL_NOTE_PHYS_ADDR:
8561 			err = ill_set_phys_addr(ill, mp);
8562 			break;
8563 
8564 		case DL_NOTE_REPLUMB:
8565 			/*
8566 			 * Directly return after calling ill_replumb().
8567 			 * Note that we should not free mp as it is reused
8568 			 * in the ill_replumb() function.
8569 			 */
8570 			err = ill_replumb(ill, mp);
8571 			return;
8572 
8573 		case DL_NOTE_FASTPATH_FLUSH:
8574 			nce_flush(ill, B_FALSE);
8575 			break;
8576 
8577 		case DL_NOTE_SDU_SIZE:
8578 		case DL_NOTE_SDU_SIZE2:
8579 			/*
8580 			 * The dce and fragmentation code can cope with
8581 			 * this changing while packets are being sent.
8582 			 * When packets are sent ip_output will discover
8583 			 * a change.
8584 			 *
8585 			 * Change the MTU size of the interface.
8586 			 */
8587 			mutex_enter(&ill->ill_lock);
8588 			orig_mtu = ill->ill_mtu;
8589 			orig_mc_mtu = ill->ill_mc_mtu;
8590 			switch (notify->dl_notification) {
8591 			case DL_NOTE_SDU_SIZE:
8592 				ill->ill_current_frag =
8593 				    (uint_t)notify->dl_data;
8594 				ill->ill_mc_mtu = (uint_t)notify->dl_data;
8595 				break;
8596 			case DL_NOTE_SDU_SIZE2:
8597 				ill->ill_current_frag =
8598 				    (uint_t)notify->dl_data1;
8599 				ill->ill_mc_mtu = (uint_t)notify->dl_data2;
8600 				break;
8601 			}
8602 			if (ill->ill_current_frag > ill->ill_max_frag)
8603 				ill->ill_max_frag = ill->ill_current_frag;
8604 
8605 			if (!(ill->ill_flags & ILLF_FIXEDMTU)) {
8606 				ill->ill_mtu = ill->ill_current_frag;
8607 
8608 				/*
8609 				 * If ill_user_mtu was set (via
8610 				 * SIOCSLIFLNKINFO), clamp ill_mtu at it.
8611 				 */
8612 				if (ill->ill_user_mtu != 0 &&
8613 				    ill->ill_user_mtu < ill->ill_mtu)
8614 					ill->ill_mtu = ill->ill_user_mtu;
8615 
8616 				if (ill->ill_user_mtu != 0 &&
8617 				    ill->ill_user_mtu < ill->ill_mc_mtu)
8618 					ill->ill_mc_mtu = ill->ill_user_mtu;
8619 
8620 				if (ill->ill_isv6) {
8621 					if (ill->ill_mtu < IPV6_MIN_MTU)
8622 						ill->ill_mtu = IPV6_MIN_MTU;
8623 					if (ill->ill_mc_mtu < IPV6_MIN_MTU)
8624 						ill->ill_mc_mtu = IPV6_MIN_MTU;
8625 				} else {
8626 					if (ill->ill_mtu < IP_MIN_MTU)
8627 						ill->ill_mtu = IP_MIN_MTU;
8628 					if (ill->ill_mc_mtu < IP_MIN_MTU)
8629 						ill->ill_mc_mtu = IP_MIN_MTU;
8630 				}
8631 			} else if (ill->ill_mc_mtu > ill->ill_mtu) {
8632 				ill->ill_mc_mtu = ill->ill_mtu;
8633 			}
8634 
8635 			mutex_exit(&ill->ill_lock);
8636 			/*
8637 			 * Make sure all dce_generation checks find out
8638 			 * that ill_mtu/ill_mc_mtu has changed.
8639 			 */
8640 			if (orig_mtu != ill->ill_mtu ||
8641 			    orig_mc_mtu != ill->ill_mc_mtu) {
8642 				dce_increment_all_generations(ill->ill_isv6,
8643 				    ill->ill_ipst);
8644 			}
8645 
8646 			/*
8647 			 * Refresh IPMP meta-interface MTU if necessary.
8648 			 */
8649 			if (IS_UNDER_IPMP(ill))
8650 				ipmp_illgrp_refresh_mtu(ill->ill_grp);
8651 			break;
8652 
8653 		case DL_NOTE_LINK_UP:
8654 		case DL_NOTE_LINK_DOWN: {
8655 			/*
8656 			 * We are writer. ill / phyint / ipsq assocs stable.
8657 			 * The RUNNING flag reflects the state of the link.
8658 			 */
8659 			phyint_t *phyint = ill->ill_phyint;
8660 			uint64_t new_phyint_flags;
8661 			boolean_t changed = B_FALSE;
8662 			boolean_t went_up;
8663 
8664 			went_up = notify->dl_notification == DL_NOTE_LINK_UP;
8665 			mutex_enter(&phyint->phyint_lock);
8666 
8667 			new_phyint_flags = went_up ?
8668 			    phyint->phyint_flags | PHYI_RUNNING :
8669 			    phyint->phyint_flags & ~PHYI_RUNNING;
8670 
8671 			if (IS_IPMP(ill)) {
8672 				new_phyint_flags = went_up ?
8673 				    new_phyint_flags & ~PHYI_FAILED :
8674 				    new_phyint_flags | PHYI_FAILED;
8675 			}
8676 
8677 			if (new_phyint_flags != phyint->phyint_flags) {
8678 				phyint->phyint_flags = new_phyint_flags;
8679 				changed = B_TRUE;
8680 			}
8681 			mutex_exit(&phyint->phyint_lock);
8682 			/*
8683 			 * ill_restart_dad handles the DAD restart and routing
8684 			 * socket notification logic.
8685 			 */
8686 			if (changed) {
8687 				ill_restart_dad(phyint->phyint_illv4, went_up);
8688 				ill_restart_dad(phyint->phyint_illv6, went_up);
8689 			}
8690 			break;
8691 		}
8692 		case DL_NOTE_PROMISC_ON_PHYS: {
8693 			phyint_t *phyint = ill->ill_phyint;
8694 
8695 			mutex_enter(&phyint->phyint_lock);
8696 			phyint->phyint_flags |= PHYI_PROMISC;
8697 			mutex_exit(&phyint->phyint_lock);
8698 			break;
8699 		}
8700 		case DL_NOTE_PROMISC_OFF_PHYS: {
8701 			phyint_t *phyint = ill->ill_phyint;
8702 
8703 			mutex_enter(&phyint->phyint_lock);
8704 			phyint->phyint_flags &= ~PHYI_PROMISC;
8705 			mutex_exit(&phyint->phyint_lock);
8706 			break;
8707 		}
8708 		case DL_NOTE_CAPAB_RENEG:
8709 			/*
8710 			 * Something changed on the driver side.
8711 			 * It wants us to renegotiate the capabilities
8712 			 * on this ill. One possible cause is the aggregation
8713 			 * interface under us where a port got added or
8714 			 * went away.
8715 			 *
8716 			 * If the capability negotiation is already done
8717 			 * or is in progress, reset the capabilities and
8718 			 * mark the ill's ill_capab_reneg to be B_TRUE,
8719 			 * so that when the ack comes back, we can start
8720 			 * the renegotiation process.
8721 			 *
8722 			 * Note that if ill_capab_reneg is already B_TRUE
8723 			 * (ill_dlpi_capab_state is IDS_UNKNOWN in this case),
8724 			 * the capability resetting request has been sent
8725 			 * and the renegotiation has not been started yet;
8726 			 * nothing needs to be done in this case.
8727 			 */
8728 			ipsq_current_start(ipsq, ill->ill_ipif, 0);
8729 			ill_capability_reset(ill, B_TRUE);
8730 			ipsq_current_finish(ipsq);
8731 			break;
8732 
8733 		case DL_NOTE_ALLOWED_IPS:
8734 			ill_set_allowed_ips(ill, mp);
8735 			break;
8736 		default:
8737 			ip0dbg(("ip_rput_dlpi_writer: unknown notification "
8738 			    "type 0x%x for DL_NOTIFY_IND\n",
8739 			    notify->dl_notification));
8740 			break;
8741 		}
8742 
8743 		/*
8744 		 * As this is an asynchronous operation, we
8745 		 * should not call ill_dlpi_done
8746 		 */
8747 		break;
8748 	}
8749 	case DL_NOTIFY_ACK: {
8750 		dl_notify_ack_t *noteack = (dl_notify_ack_t *)mp->b_rptr;
8751 
8752 		if (noteack->dl_notifications & DL_NOTE_LINK_UP)
8753 			ill->ill_note_link = 1;
8754 		ill_dlpi_done(ill, DL_NOTIFY_REQ);
8755 		break;
8756 	}
8757 	case DL_PHYS_ADDR_ACK: {
8758 		/*
8759 		 * As part of plumbing the interface via SIOCSLIFNAME,
8760 		 * ill_dl_phys() will queue a series of DL_PHYS_ADDR_REQs,
8761 		 * whose answers we receive here.  As each answer is received,
8762 		 * we call ill_dlpi_done() to dispatch the next request as
8763 		 * we're processing the current one.  Once all answers have
8764 		 * been received, we use ipsq_pending_mp_get() to dequeue the
8765 		 * outstanding IOCTL and reply to it.  (Because ill_dl_phys()
8766 		 * is invoked from an ill queue, conn_oper_pending_ill is not
8767 		 * available, but we know the ioctl is pending on ill_wq.)
8768 		 */
8769 		uint_t	paddrlen, paddroff;
8770 		uint8_t	*addr;
8771 
8772 		paddrreq = ill->ill_phys_addr_pend;
8773 		paddrlen = ((dl_phys_addr_ack_t *)mp->b_rptr)->dl_addr_length;
8774 		paddroff = ((dl_phys_addr_ack_t *)mp->b_rptr)->dl_addr_offset;
8775 		addr = mp->b_rptr + paddroff;
8776 
8777 		ill_dlpi_done(ill, DL_PHYS_ADDR_REQ);
8778 		if (paddrreq == DL_IPV6_TOKEN) {
8779 			/*
8780 			 * bcopy to low-order bits of ill_token
8781 			 *
8782 			 * XXX Temporary hack - currently, all known tokens
8783 			 * are 64 bits, so I'll cheat for the moment.
8784 			 */
8785 			bcopy(addr, &ill->ill_token.s6_addr32[2], paddrlen);
8786 			ill->ill_token_length = paddrlen;
8787 			break;
8788 		} else if (paddrreq == DL_IPV6_LINK_LAYER_ADDR) {
8789 			ASSERT(ill->ill_nd_lla_mp == NULL);
8790 			ill_set_ndmp(ill, mp, paddroff, paddrlen);
8791 			mp = NULL;
8792 			break;
8793 		} else if (paddrreq == DL_CURR_DEST_ADDR) {
8794 			ASSERT(ill->ill_dest_addr_mp == NULL);
8795 			ill->ill_dest_addr_mp = mp;
8796 			ill->ill_dest_addr = addr;
8797 			mp = NULL;
8798 			if (ill->ill_isv6) {
8799 				ill_setdesttoken(ill);
8800 				ipif_setdestlinklocal(ill->ill_ipif);
8801 			}
8802 			break;
8803 		}
8804 
8805 		ASSERT(paddrreq == DL_CURR_PHYS_ADDR);
8806 		ASSERT(ill->ill_phys_addr_mp == NULL);
8807 		if (!ill->ill_ifname_pending)
8808 			break;
8809 		ill->ill_ifname_pending = 0;
8810 		if (!ioctl_aborted)
8811 			mp1 = ipsq_pending_mp_get(ipsq, &connp);
8812 		if (mp1 != NULL) {
8813 			ASSERT(connp == NULL);
8814 			q = ill->ill_wq;
8815 		}
8816 		/*
8817 		 * If any error acks received during the plumbing sequence,
8818 		 * ill_ifname_pending_err will be set. Break out and send up
8819 		 * the error to the pending ioctl.
8820 		 */
8821 		if (ill->ill_ifname_pending_err != 0) {
8822 			err = ill->ill_ifname_pending_err;
8823 			ill->ill_ifname_pending_err = 0;
8824 			break;
8825 		}
8826 
8827 		ill->ill_phys_addr_mp = mp;
8828 		ill->ill_phys_addr = (paddrlen == 0 ? NULL : addr);
8829 		mp = NULL;
8830 
8831 		/*
8832 		 * If paddrlen or ill_phys_addr_length is zero, the DLPI
8833 		 * provider doesn't support physical addresses.  We check both
8834 		 * paddrlen and ill_phys_addr_length because sppp (PPP) does
8835 		 * not have physical addresses, but historically adversises a
8836 		 * physical address length of 0 in its DL_INFO_ACK, but 6 in
8837 		 * its DL_PHYS_ADDR_ACK.
8838 		 */
8839 		if (paddrlen == 0 || ill->ill_phys_addr_length == 0) {
8840 			ill->ill_phys_addr = NULL;
8841 		} else if (paddrlen != ill->ill_phys_addr_length) {
8842 			ip0dbg(("DL_PHYS_ADDR_ACK: got addrlen %d, expected %d",
8843 			    paddrlen, ill->ill_phys_addr_length));
8844 			err = EINVAL;
8845 			break;
8846 		}
8847 
8848 		if (ill->ill_nd_lla_mp == NULL) {
8849 			if ((mp_hw = copyb(ill->ill_phys_addr_mp)) == NULL) {
8850 				err = ENOMEM;
8851 				break;
8852 			}
8853 			ill_set_ndmp(ill, mp_hw, paddroff, paddrlen);
8854 		}
8855 
8856 		if (ill->ill_isv6) {
8857 			ill_setdefaulttoken(ill);
8858 			ipif_setlinklocal(ill->ill_ipif);
8859 		}
8860 		break;
8861 	}
8862 	case DL_OK_ACK:
8863 		ip2dbg(("DL_OK_ACK %s (0x%x)\n",
8864 		    dl_primstr((int)dloa->dl_correct_primitive),
8865 		    dloa->dl_correct_primitive));
8866 		DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer ok",
8867 		    char *, dl_primstr(dloa->dl_correct_primitive),
8868 		    ill_t *, ill);
8869 
8870 		switch (dloa->dl_correct_primitive) {
8871 		case DL_ENABMULTI_REQ:
8872 		case DL_DISABMULTI_REQ:
8873 			ill_dlpi_done(ill, dloa->dl_correct_primitive);
8874 			break;
8875 		case DL_PROMISCON_REQ:
8876 		case DL_PROMISCOFF_REQ:
8877 		case DL_UNBIND_REQ:
8878 		case DL_ATTACH_REQ:
8879 			ill_dlpi_done(ill, dloa->dl_correct_primitive);
8880 			break;
8881 		}
8882 		break;
8883 	default:
8884 		break;
8885 	}
8886 
8887 	freemsg(mp);
8888 	if (mp1 == NULL)
8889 		return;
8890 
8891 	/*
8892 	 * The operation must complete without EINPROGRESS since
8893 	 * ipsq_pending_mp_get() has removed the mblk (mp1).  Otherwise,
8894 	 * the operation will be stuck forever inside the IPSQ.
8895 	 */
8896 	ASSERT(err != EINPROGRESS);
8897 
8898 	DTRACE_PROBE4(ipif__ioctl, char *, "ip_rput_dlpi_writer finish",
8899 	    int, ipsq->ipsq_xop->ipx_current_ioctl, ill_t *, ill,
8900 	    ipif_t *, NULL);
8901 
8902 	switch (ipsq->ipsq_xop->ipx_current_ioctl) {
8903 	case 0:
8904 		ipsq_current_finish(ipsq);
8905 		break;
8906 
8907 	case SIOCSLIFNAME:
8908 	case IF_UNITSEL: {
8909 		ill_t *ill_other = ILL_OTHER(ill);
8910 
8911 		/*
8912 		 * If SIOCSLIFNAME or IF_UNITSEL is about to succeed, and the
8913 		 * ill has a peer which is in an IPMP group, then place ill
8914 		 * into the same group.  One catch: although ifconfig plumbs
8915 		 * the appropriate IPMP meta-interface prior to plumbing this
8916 		 * ill, it is possible for multiple ifconfig applications to
8917 		 * race (or for another application to adjust plumbing), in
8918 		 * which case the IPMP meta-interface we need will be missing.
8919 		 * If so, kick the phyint out of the group.
8920 		 */
8921 		if (err == 0 && ill_other != NULL && IS_UNDER_IPMP(ill_other)) {
8922 			ipmp_grp_t	*grp = ill->ill_phyint->phyint_grp;
8923 			ipmp_illgrp_t	*illg;
8924 
8925 			illg = ill->ill_isv6 ? grp->gr_v6 : grp->gr_v4;
8926 			if (illg == NULL)
8927 				ipmp_phyint_leave_grp(ill->ill_phyint);
8928 			else
8929 				ipmp_ill_join_illgrp(ill, illg);
8930 		}
8931 
8932 		if (ipsq->ipsq_xop->ipx_current_ioctl == IF_UNITSEL)
8933 			ip_ioctl_finish(q, mp1, err, NO_COPYOUT, ipsq);
8934 		else
8935 			ip_ioctl_finish(q, mp1, err, COPYOUT, ipsq);
8936 		break;
8937 	}
8938 	case SIOCLIFADDIF:
8939 		ip_ioctl_finish(q, mp1, err, COPYOUT, ipsq);
8940 		break;
8941 
8942 	default:
8943 		ip_ioctl_finish(q, mp1, err, NO_COPYOUT, ipsq);
8944 		break;
8945 	}
8946 }
8947 
8948 /*
8949  * ip_rput_other is called by ip_rput to handle messages modifying the global
8950  * state in IP.  If 'ipsq' is non-NULL, caller is writer on it.
8951  */
8952 /* ARGSUSED */
8953 void
ip_rput_other(ipsq_t * ipsq,queue_t * q,mblk_t * mp,void * dummy_arg)8954 ip_rput_other(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
8955 {
8956 	ill_t		*ill = q->q_ptr;
8957 	struct iocblk	*iocp;
8958 
8959 	ip1dbg(("ip_rput_other "));
8960 	if (ipsq != NULL) {
8961 		ASSERT(IAM_WRITER_IPSQ(ipsq));
8962 		ASSERT(ipsq->ipsq_xop ==
8963 		    ill->ill_phyint->phyint_ipsq->ipsq_xop);
8964 	}
8965 
8966 	switch (mp->b_datap->db_type) {
8967 	case M_ERROR:
8968 	case M_HANGUP:
8969 		/*
8970 		 * The device has a problem.  We force the ILL down.  It can
8971 		 * be brought up again manually using SIOCSIFFLAGS (via
8972 		 * ifconfig or equivalent).
8973 		 */
8974 		ASSERT(ipsq != NULL);
8975 		if (mp->b_rptr < mp->b_wptr)
8976 			ill->ill_error = (int)(*mp->b_rptr & 0xFF);
8977 		if (ill->ill_error == 0)
8978 			ill->ill_error = ENXIO;
8979 		if (!ill_down_start(q, mp))
8980 			return;
8981 		ipif_all_down_tail(ipsq, q, mp, NULL);
8982 		break;
8983 	case M_IOCNAK: {
8984 		iocp = (struct iocblk *)mp->b_rptr;
8985 
8986 		ASSERT(iocp->ioc_cmd == DL_IOC_HDR_INFO);
8987 		/*
8988 		 * If this was the first attempt, turn off the fastpath
8989 		 * probing.
8990 		 */
8991 		mutex_enter(&ill->ill_lock);
8992 		if (ill->ill_dlpi_fastpath_state == IDS_INPROGRESS) {
8993 			ill->ill_dlpi_fastpath_state = IDS_FAILED;
8994 			mutex_exit(&ill->ill_lock);
8995 			/*
8996 			 * don't flush the nce_t entries: we use them
8997 			 * as an index to the ncec itself.
8998 			 */
8999 			ip1dbg(("ip_rput: DLPI fastpath off on interface %s\n",
9000 			    ill->ill_name));
9001 		} else {
9002 			mutex_exit(&ill->ill_lock);
9003 		}
9004 		freemsg(mp);
9005 		break;
9006 	}
9007 	default:
9008 		ASSERT(0);
9009 		break;
9010 	}
9011 }
9012 
9013 /*
9014  * Update any source route, record route or timestamp options
9015  * When it fails it has consumed the message and BUMPed the MIB.
9016  */
9017 boolean_t
ip_forward_options(mblk_t * mp,ipha_t * ipha,ill_t * dst_ill,ip_recv_attr_t * ira)9018 ip_forward_options(mblk_t *mp, ipha_t *ipha, ill_t *dst_ill,
9019     ip_recv_attr_t *ira)
9020 {
9021 	ipoptp_t	opts;
9022 	uchar_t		*opt;
9023 	uint8_t		optval;
9024 	uint8_t		optlen;
9025 	ipaddr_t	dst;
9026 	ipaddr_t	ifaddr;
9027 	uint32_t	ts;
9028 	timestruc_t	now;
9029 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
9030 
9031 	ip2dbg(("ip_forward_options\n"));
9032 	dst = ipha->ipha_dst;
9033 	opt = NULL;
9034 
9035 	for (optval = ipoptp_first(&opts, ipha);
9036 	    optval != IPOPT_EOL;
9037 	    optval = ipoptp_next(&opts)) {
9038 		ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
9039 		opt = opts.ipoptp_cur;
9040 		optlen = opts.ipoptp_len;
9041 		ip2dbg(("ip_forward_options: opt %d, len %d\n",
9042 		    optval, opts.ipoptp_len));
9043 		switch (optval) {
9044 			uint32_t off;
9045 		case IPOPT_SSRR:
9046 		case IPOPT_LSRR:
9047 			/* Check if adminstratively disabled */
9048 			if (!ipst->ips_ip_forward_src_routed) {
9049 				BUMP_MIB(dst_ill->ill_ip_mib,
9050 				    ipIfStatsForwProhibits);
9051 				ip_drop_input("ICMP_SOURCE_ROUTE_FAILED",
9052 				    mp, dst_ill);
9053 				icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED,
9054 				    ira);
9055 				return (B_FALSE);
9056 			}
9057 			if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9058 				/*
9059 				 * Must be partial since ip_input_options
9060 				 * checked for strict.
9061 				 */
9062 				break;
9063 			}
9064 			off = opt[IPOPT_OFFSET];
9065 			off--;
9066 		redo_srr:
9067 			if (optlen < IP_ADDR_LEN ||
9068 			    off > optlen - IP_ADDR_LEN) {
9069 				/* End of source route */
9070 				ip1dbg((
9071 				    "ip_forward_options: end of SR\n"));
9072 				break;
9073 			}
9074 			/* Pick a reasonable address on the outbound if */
9075 			ASSERT(dst_ill != NULL);
9076 			if (ip_select_source_v4(dst_ill, INADDR_ANY, dst,
9077 			    INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
9078 			    NULL) != 0) {
9079 				/* No source! Shouldn't happen */
9080 				ifaddr = INADDR_ANY;
9081 			}
9082 			bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9083 			bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9084 			ip1dbg(("ip_forward_options: next hop 0x%x\n",
9085 			    ntohl(dst)));
9086 
9087 			/*
9088 			 * Check if our address is present more than
9089 			 * once as consecutive hops in source route.
9090 			 */
9091 			if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
9092 				off += IP_ADDR_LEN;
9093 				opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9094 				goto redo_srr;
9095 			}
9096 			ipha->ipha_dst = dst;
9097 			opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9098 			break;
9099 		case IPOPT_RR:
9100 			off = opt[IPOPT_OFFSET];
9101 			off--;
9102 			if (optlen < IP_ADDR_LEN ||
9103 			    off > optlen - IP_ADDR_LEN) {
9104 				/* No more room - ignore */
9105 				ip1dbg((
9106 				    "ip_forward_options: end of RR\n"));
9107 				break;
9108 			}
9109 			/* Pick a reasonable address on the outbound if */
9110 			ASSERT(dst_ill != NULL);
9111 			if (ip_select_source_v4(dst_ill, INADDR_ANY, dst,
9112 			    INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
9113 			    NULL) != 0) {
9114 				/* No source! Shouldn't happen */
9115 				ifaddr = INADDR_ANY;
9116 			}
9117 			bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9118 			opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9119 			break;
9120 		case IPOPT_TS:
9121 			off = 0;
9122 			/* Insert timestamp if there is room */
9123 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9124 			case IPOPT_TS_TSONLY:
9125 				off = IPOPT_TS_TIMELEN;
9126 				break;
9127 			case IPOPT_TS_PRESPEC:
9128 			case IPOPT_TS_PRESPEC_RFC791:
9129 				/* Verify that the address matched */
9130 				off = opt[IPOPT_OFFSET] - 1;
9131 				bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9132 				if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9133 					/* Not for us */
9134 					break;
9135 				}
9136 				/* FALLTHROUGH */
9137 			case IPOPT_TS_TSANDADDR:
9138 				off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9139 				break;
9140 			default:
9141 				/*
9142 				 * ip_*put_options should have already
9143 				 * dropped this packet.
9144 				 */
9145 				cmn_err(CE_PANIC, "ip_forward_options: "
9146 				    "unknown IT - bug in ip_input_options?\n");
9147 			}
9148 			if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
9149 				/* Increase overflow counter */
9150 				off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
9151 				opt[IPOPT_POS_OV_FLG] =
9152 				    (uint8_t)((opt[IPOPT_POS_OV_FLG] & 0x0F) |
9153 				    (off << 4));
9154 				break;
9155 			}
9156 			off = opt[IPOPT_OFFSET] - 1;
9157 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9158 			case IPOPT_TS_PRESPEC:
9159 			case IPOPT_TS_PRESPEC_RFC791:
9160 			case IPOPT_TS_TSANDADDR:
9161 				/* Pick a reasonable addr on the outbound if */
9162 				ASSERT(dst_ill != NULL);
9163 				if (ip_select_source_v4(dst_ill, INADDR_ANY,
9164 				    dst, INADDR_ANY, ALL_ZONES, ipst, &ifaddr,
9165 				    NULL, NULL) != 0) {
9166 					/* No source! Shouldn't happen */
9167 					ifaddr = INADDR_ANY;
9168 				}
9169 				bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9170 				opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9171 				/* FALLTHROUGH */
9172 			case IPOPT_TS_TSONLY:
9173 				off = opt[IPOPT_OFFSET] - 1;
9174 				/* Compute # of milliseconds since midnight */
9175 				gethrestime(&now);
9176 				ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
9177 				    NSEC2MSEC(now.tv_nsec);
9178 				bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
9179 				opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
9180 				break;
9181 			}
9182 			break;
9183 		}
9184 	}
9185 	return (B_TRUE);
9186 }
9187 
9188 /*
9189  * Call ill_frag_timeout to do garbage collection. ill_frag_timeout
9190  * returns 'true' if there are still fragments left on the queue, in
9191  * which case we restart the timer.
9192  */
9193 void
ill_frag_timer(void * arg)9194 ill_frag_timer(void *arg)
9195 {
9196 	ill_t	*ill = (ill_t *)arg;
9197 	boolean_t frag_pending;
9198 	ip_stack_t *ipst = ill->ill_ipst;
9199 	time_t	timeout;
9200 
9201 	mutex_enter(&ill->ill_lock);
9202 	ASSERT(!ill->ill_fragtimer_executing);
9203 	if (ill->ill_state_flags & ILL_CONDEMNED) {
9204 		ill->ill_frag_timer_id = 0;
9205 		mutex_exit(&ill->ill_lock);
9206 		return;
9207 	}
9208 	ill->ill_fragtimer_executing = 1;
9209 	mutex_exit(&ill->ill_lock);
9210 
9211 	timeout = (ill->ill_isv6 ? ipst->ips_ipv6_reassembly_timeout :
9212 	    ipst->ips_ip_reassembly_timeout);
9213 
9214 	frag_pending = ill_frag_timeout(ill, timeout);
9215 
9216 	/*
9217 	 * Restart the timer, if we have fragments pending or if someone
9218 	 * wanted us to be scheduled again.
9219 	 */
9220 	mutex_enter(&ill->ill_lock);
9221 	ill->ill_fragtimer_executing = 0;
9222 	ill->ill_frag_timer_id = 0;
9223 	if (frag_pending || ill->ill_fragtimer_needrestart)
9224 		ill_frag_timer_start(ill);
9225 	mutex_exit(&ill->ill_lock);
9226 }
9227 
9228 void
ill_frag_timer_start(ill_t * ill)9229 ill_frag_timer_start(ill_t *ill)
9230 {
9231 	ip_stack_t *ipst = ill->ill_ipst;
9232 	clock_t	timeo_ms;
9233 
9234 	ASSERT(MUTEX_HELD(&ill->ill_lock));
9235 
9236 	/* If the ill is closing or opening don't proceed */
9237 	if (ill->ill_state_flags & ILL_CONDEMNED)
9238 		return;
9239 
9240 	if (ill->ill_fragtimer_executing) {
9241 		/*
9242 		 * ill_frag_timer is currently executing. Just record the
9243 		 * the fact that we want the timer to be restarted.
9244 		 * ill_frag_timer will post a timeout before it returns,
9245 		 * ensuring it will be called again.
9246 		 */
9247 		ill->ill_fragtimer_needrestart = 1;
9248 		return;
9249 	}
9250 
9251 	if (ill->ill_frag_timer_id == 0) {
9252 		timeo_ms = (ill->ill_isv6 ? ipst->ips_ipv6_reassembly_timeout :
9253 		    ipst->ips_ip_reassembly_timeout) * SECONDS;
9254 
9255 		/*
9256 		 * The timer is neither running nor is the timeout handler
9257 		 * executing. Post a timeout so that ill_frag_timer will be
9258 		 * called
9259 		 */
9260 		ill->ill_frag_timer_id = timeout(ill_frag_timer, ill,
9261 		    MSEC_TO_TICK(timeo_ms >> 1));
9262 		ill->ill_fragtimer_needrestart = 0;
9263 	}
9264 }
9265 
9266 /*
9267  * Update any source route, record route or timestamp options.
9268  * Check that we are at end of strict source route.
9269  * The options have already been checked for sanity in ip_input_options().
9270  */
9271 boolean_t
ip_input_local_options(mblk_t * mp,ipha_t * ipha,ip_recv_attr_t * ira)9272 ip_input_local_options(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
9273 {
9274 	ipoptp_t	opts;
9275 	uchar_t		*opt;
9276 	uint8_t		optval;
9277 	uint8_t		optlen;
9278 	ipaddr_t	dst;
9279 	ipaddr_t	ifaddr;
9280 	uint32_t	ts;
9281 	timestruc_t	now;
9282 	ill_t		*ill = ira->ira_ill;
9283 	ip_stack_t	*ipst = ill->ill_ipst;
9284 
9285 	ip2dbg(("ip_input_local_options\n"));
9286 	opt = NULL;
9287 
9288 	for (optval = ipoptp_first(&opts, ipha);
9289 	    optval != IPOPT_EOL;
9290 	    optval = ipoptp_next(&opts)) {
9291 		ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
9292 		opt = opts.ipoptp_cur;
9293 		optlen = opts.ipoptp_len;
9294 		ip2dbg(("ip_input_local_options: opt %d, len %d\n",
9295 		    optval, optlen));
9296 		switch (optval) {
9297 			uint32_t off;
9298 		case IPOPT_SSRR:
9299 		case IPOPT_LSRR:
9300 			off = opt[IPOPT_OFFSET];
9301 			off--;
9302 			if (optlen < IP_ADDR_LEN ||
9303 			    off > optlen - IP_ADDR_LEN) {
9304 				/* End of source route */
9305 				ip1dbg(("ip_input_local_options: end of SR\n"));
9306 				break;
9307 			}
9308 			/*
9309 			 * This will only happen if two consecutive entries
9310 			 * in the source route contains our address or if
9311 			 * it is a packet with a loose source route which
9312 			 * reaches us before consuming the whole source route
9313 			 */
9314 			ip1dbg(("ip_input_local_options: not end of SR\n"));
9315 			if (optval == IPOPT_SSRR) {
9316 				goto bad_src_route;
9317 			}
9318 			/*
9319 			 * Hack: instead of dropping the packet truncate the
9320 			 * source route to what has been used by filling the
9321 			 * rest with IPOPT_NOP.
9322 			 */
9323 			opt[IPOPT_OLEN] = (uint8_t)off;
9324 			while (off < optlen) {
9325 				opt[off++] = IPOPT_NOP;
9326 			}
9327 			break;
9328 		case IPOPT_RR:
9329 			off = opt[IPOPT_OFFSET];
9330 			off--;
9331 			if (optlen < IP_ADDR_LEN ||
9332 			    off > optlen - IP_ADDR_LEN) {
9333 				/* No more room - ignore */
9334 				ip1dbg((
9335 				    "ip_input_local_options: end of RR\n"));
9336 				break;
9337 			}
9338 			/* Pick a reasonable address on the outbound if */
9339 			if (ip_select_source_v4(ill, INADDR_ANY, ipha->ipha_dst,
9340 			    INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
9341 			    NULL) != 0) {
9342 				/* No source! Shouldn't happen */
9343 				ifaddr = INADDR_ANY;
9344 			}
9345 			bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9346 			opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9347 			break;
9348 		case IPOPT_TS:
9349 			off = 0;
9350 			/* Insert timestamp if there is romm */
9351 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9352 			case IPOPT_TS_TSONLY:
9353 				off = IPOPT_TS_TIMELEN;
9354 				break;
9355 			case IPOPT_TS_PRESPEC:
9356 			case IPOPT_TS_PRESPEC_RFC791:
9357 				/* Verify that the address matched */
9358 				off = opt[IPOPT_OFFSET] - 1;
9359 				bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9360 				if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9361 					/* Not for us */
9362 					break;
9363 				}
9364 				/* FALLTHROUGH */
9365 			case IPOPT_TS_TSANDADDR:
9366 				off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9367 				break;
9368 			default:
9369 				/*
9370 				 * ip_*put_options should have already
9371 				 * dropped this packet.
9372 				 */
9373 				cmn_err(CE_PANIC, "ip_input_local_options: "
9374 				    "unknown IT - bug in ip_input_options?\n");
9375 			}
9376 			if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
9377 				/* Increase overflow counter */
9378 				off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
9379 				opt[IPOPT_POS_OV_FLG] =
9380 				    (uint8_t)((opt[IPOPT_POS_OV_FLG] & 0x0F) |
9381 				    (off << 4));
9382 				break;
9383 			}
9384 			off = opt[IPOPT_OFFSET] - 1;
9385 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9386 			case IPOPT_TS_PRESPEC:
9387 			case IPOPT_TS_PRESPEC_RFC791:
9388 			case IPOPT_TS_TSANDADDR:
9389 				/* Pick a reasonable addr on the outbound if */
9390 				if (ip_select_source_v4(ill, INADDR_ANY,
9391 				    ipha->ipha_dst, INADDR_ANY, ALL_ZONES, ipst,
9392 				    &ifaddr, NULL, NULL) != 0) {
9393 					/* No source! Shouldn't happen */
9394 					ifaddr = INADDR_ANY;
9395 				}
9396 				bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9397 				opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9398 				/* FALLTHROUGH */
9399 			case IPOPT_TS_TSONLY:
9400 				off = opt[IPOPT_OFFSET] - 1;
9401 				/* Compute # of milliseconds since midnight */
9402 				gethrestime(&now);
9403 				ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
9404 				    NSEC2MSEC(now.tv_nsec);
9405 				bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
9406 				opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
9407 				break;
9408 			}
9409 			break;
9410 		}
9411 	}
9412 	return (B_TRUE);
9413 
9414 bad_src_route:
9415 	/* make sure we clear any indication of a hardware checksum */
9416 	DB_CKSUMFLAGS(mp) = 0;
9417 	ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ill);
9418 	icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, ira);
9419 	return (B_FALSE);
9420 
9421 }
9422 
9423 /*
9424  * Process IP options in an inbound packet.  Always returns the nexthop.
9425  * Normally this is the passed in nexthop, but if there is an option
9426  * that effects the nexthop (such as a source route) that will be returned.
9427  * Sets *errorp if there is an error, in which case an ICMP error has been sent
9428  * and mp freed.
9429  */
9430 ipaddr_t
ip_input_options(ipha_t * ipha,ipaddr_t dst,mblk_t * mp,ip_recv_attr_t * ira,int * errorp)9431 ip_input_options(ipha_t *ipha, ipaddr_t dst, mblk_t *mp,
9432     ip_recv_attr_t *ira, int *errorp)
9433 {
9434 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
9435 	ipoptp_t	opts;
9436 	uchar_t		*opt;
9437 	uint8_t		optval;
9438 	uint8_t		optlen;
9439 	intptr_t	code = 0;
9440 	ire_t		*ire;
9441 
9442 	ip2dbg(("ip_input_options\n"));
9443 	opt = NULL;
9444 	*errorp = 0;
9445 	for (optval = ipoptp_first(&opts, ipha);
9446 	    optval != IPOPT_EOL;
9447 	    optval = ipoptp_next(&opts)) {
9448 		opt = opts.ipoptp_cur;
9449 		optlen = opts.ipoptp_len;
9450 		ip2dbg(("ip_input_options: opt %d, len %d\n",
9451 		    optval, optlen));
9452 		/*
9453 		 * Note: we need to verify the checksum before we
9454 		 * modify anything thus this routine only extracts the next
9455 		 * hop dst from any source route.
9456 		 */
9457 		switch (optval) {
9458 			uint32_t off;
9459 		case IPOPT_SSRR:
9460 		case IPOPT_LSRR:
9461 			if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9462 				if (optval == IPOPT_SSRR) {
9463 					ip1dbg(("ip_input_options: not next"
9464 					    " strict source route 0x%x\n",
9465 					    ntohl(dst)));
9466 					code = (char *)&ipha->ipha_dst -
9467 					    (char *)ipha;
9468 					goto param_prob; /* RouterReq's */
9469 				}
9470 				ip2dbg(("ip_input_options: "
9471 				    "not next source route 0x%x\n",
9472 				    ntohl(dst)));
9473 				break;
9474 			}
9475 
9476 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9477 				ip1dbg((
9478 				    "ip_input_options: bad option offset\n"));
9479 				code = (char *)&opt[IPOPT_OLEN] -
9480 				    (char *)ipha;
9481 				goto param_prob;
9482 			}
9483 			off = opt[IPOPT_OFFSET];
9484 			off--;
9485 		redo_srr:
9486 			if (optlen < IP_ADDR_LEN ||
9487 			    off > optlen - IP_ADDR_LEN) {
9488 				/* End of source route */
9489 				ip1dbg(("ip_input_options: end of SR\n"));
9490 				break;
9491 			}
9492 			bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9493 			ip1dbg(("ip_input_options: next hop 0x%x\n",
9494 			    ntohl(dst)));
9495 
9496 			/*
9497 			 * Check if our address is present more than
9498 			 * once as consecutive hops in source route.
9499 			 * XXX verify per-interface ip_forwarding
9500 			 * for source route?
9501 			 */
9502 			if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
9503 				off += IP_ADDR_LEN;
9504 				goto redo_srr;
9505 			}
9506 
9507 			if (dst == htonl(INADDR_LOOPBACK)) {
9508 				ip1dbg(("ip_input_options: loopback addr in "
9509 				    "source route!\n"));
9510 				goto bad_src_route;
9511 			}
9512 			/*
9513 			 * For strict: verify that dst is directly
9514 			 * reachable.
9515 			 */
9516 			if (optval == IPOPT_SSRR) {
9517 				ire = ire_ftable_lookup_v4(dst, 0, 0,
9518 				    IRE_INTERFACE, NULL, ALL_ZONES,
9519 				    ira->ira_tsl,
9520 				    MATCH_IRE_TYPE | MATCH_IRE_SECATTR, 0, ipst,
9521 				    NULL);
9522 				if (ire == NULL) {
9523 					ip1dbg(("ip_input_options: SSRR not "
9524 					    "directly reachable: 0x%x\n",
9525 					    ntohl(dst)));
9526 					goto bad_src_route;
9527 				}
9528 				ire_refrele(ire);
9529 			}
9530 			/*
9531 			 * Defer update of the offset and the record route
9532 			 * until the packet is forwarded.
9533 			 */
9534 			break;
9535 		case IPOPT_RR:
9536 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9537 				ip1dbg((
9538 				    "ip_input_options: bad option offset\n"));
9539 				code = (char *)&opt[IPOPT_OLEN] -
9540 				    (char *)ipha;
9541 				goto param_prob;
9542 			}
9543 			break;
9544 		case IPOPT_TS:
9545 			/*
9546 			 * Verify that length >= 5 and that there is either
9547 			 * room for another timestamp or that the overflow
9548 			 * counter is not maxed out.
9549 			 */
9550 			code = (char *)&opt[IPOPT_OLEN] - (char *)ipha;
9551 			if (optlen < IPOPT_MINLEN_IT) {
9552 				goto param_prob;
9553 			}
9554 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9555 				ip1dbg((
9556 				    "ip_input_options: bad option offset\n"));
9557 				code = (char *)&opt[IPOPT_OFFSET] -
9558 				    (char *)ipha;
9559 				goto param_prob;
9560 			}
9561 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9562 			case IPOPT_TS_TSONLY:
9563 				off = IPOPT_TS_TIMELEN;
9564 				break;
9565 			case IPOPT_TS_TSANDADDR:
9566 			case IPOPT_TS_PRESPEC:
9567 			case IPOPT_TS_PRESPEC_RFC791:
9568 				off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9569 				break;
9570 			default:
9571 				code = (char *)&opt[IPOPT_POS_OV_FLG] -
9572 				    (char *)ipha;
9573 				goto param_prob;
9574 			}
9575 			if (opt[IPOPT_OFFSET] - 1 + off > optlen &&
9576 			    (opt[IPOPT_POS_OV_FLG] & 0xF0) == 0xF0) {
9577 				/*
9578 				 * No room and the overflow counter is 15
9579 				 * already.
9580 				 */
9581 				goto param_prob;
9582 			}
9583 			break;
9584 		}
9585 	}
9586 
9587 	if ((opts.ipoptp_flags & IPOPTP_ERROR) == 0) {
9588 		return (dst);
9589 	}
9590 
9591 	ip1dbg(("ip_input_options: error processing IP options."));
9592 	code = (char *)&opt[IPOPT_OFFSET] - (char *)ipha;
9593 
9594 param_prob:
9595 	/* make sure we clear any indication of a hardware checksum */
9596 	DB_CKSUMFLAGS(mp) = 0;
9597 	ip_drop_input("ICMP_PARAM_PROBLEM", mp, ira->ira_ill);
9598 	icmp_param_problem(mp, (uint8_t)code, ira);
9599 	*errorp = -1;
9600 	return (dst);
9601 
9602 bad_src_route:
9603 	/* make sure we clear any indication of a hardware checksum */
9604 	DB_CKSUMFLAGS(mp) = 0;
9605 	ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ira->ira_ill);
9606 	icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, ira);
9607 	*errorp = -1;
9608 	return (dst);
9609 }
9610 
9611 /*
9612  * IP & ICMP info in >=14 msg's ...
9613  *  - ip fixed part (mib2_ip_t)
9614  *  - icmp fixed part (mib2_icmp_t)
9615  *  - ipAddrEntryTable (ip 20)		all IPv4 ipifs
9616  *  - ipRouteEntryTable (ip 21)		all IPv4 IREs
9617  *  - ipNetToMediaEntryTable (ip 22)	all IPv4 Neighbor Cache entries
9618  *  - ipRouteAttributeTable (ip 102)	labeled routes
9619  *  - ip multicast membership (ip_member_t)
9620  *  - ip multicast source filtering (ip_grpsrc_t)
9621  *  - igmp fixed part (struct igmpstat)
9622  *  - multicast routing stats (struct mrtstat)
9623  *  - multicast routing vifs (array of struct vifctl)
9624  *  - multicast routing routes (array of struct mfcctl)
9625  *  - ip6 fixed part (mib2_ipv6IfStatsEntry_t)
9626  *					One per ill plus one generic
9627  *  - icmp6 fixed part (mib2_ipv6IfIcmpEntry_t)
9628  *					One per ill plus one generic
9629  *  - ipv6RouteEntry			all IPv6 IREs
9630  *  - ipv6RouteAttributeTable (ip6 102)	labeled routes
9631  *  - ipv6NetToMediaEntry		all IPv6 Neighbor Cache entries
9632  *  - ipv6AddrEntry			all IPv6 ipifs
9633  *  - ipv6 multicast membership (ipv6_member_t)
9634  *  - ipv6 multicast source filtering (ipv6_grpsrc_t)
9635  *
9636  * NOTE: original mpctl is copied for msg's 2..N, since its ctl part is
9637  * already filled in by the caller.
9638  * If legacy_req is true then MIB structures needs to be truncated to their
9639  * legacy sizes before being returned.
9640  * Return value of 0 indicates that no messages were sent and caller
9641  * should free mpctl.
9642  */
9643 int
ip_snmp_get(queue_t * q,mblk_t * mpctl,int level,boolean_t legacy_req)9644 ip_snmp_get(queue_t *q, mblk_t *mpctl, int level, boolean_t legacy_req)
9645 {
9646 	ip_stack_t *ipst;
9647 	sctp_stack_t *sctps;
9648 
9649 	if (q->q_next != NULL) {
9650 		ipst = ILLQ_TO_IPST(q);
9651 	} else {
9652 		ipst = CONNQ_TO_IPST(q);
9653 	}
9654 	ASSERT(ipst != NULL);
9655 	sctps = ipst->ips_netstack->netstack_sctp;
9656 
9657 	if (mpctl == NULL || mpctl->b_cont == NULL) {
9658 		return (0);
9659 	}
9660 
9661 	/*
9662 	 * For the purposes of the (broken) packet shell use
9663 	 * of the level we make sure MIB2_TCP/MIB2_UDP can be used
9664 	 * to make TCP and UDP appear first in the list of mib items.
9665 	 * TBD: We could expand this and use it in netstat so that
9666 	 * the kernel doesn't have to produce large tables (connections,
9667 	 * routes, etc) when netstat only wants the statistics or a particular
9668 	 * table.
9669 	 */
9670 	if (!(level == MIB2_TCP || level == MIB2_UDP)) {
9671 		if ((mpctl = icmp_snmp_get(q, mpctl)) == NULL) {
9672 			return (1);
9673 		}
9674 	}
9675 
9676 	if (level != MIB2_TCP) {
9677 		if ((mpctl = udp_snmp_get(q, mpctl, legacy_req)) == NULL) {
9678 			return (1);
9679 		}
9680 		if (level == MIB2_UDP) {
9681 			goto done;
9682 		}
9683 	}
9684 
9685 	if (level != MIB2_UDP) {
9686 		if ((mpctl = tcp_snmp_get(q, mpctl, legacy_req)) == NULL) {
9687 			return (1);
9688 		}
9689 		if (level == MIB2_TCP) {
9690 			goto done;
9691 		}
9692 	}
9693 
9694 	if ((mpctl = ip_snmp_get_mib2_ip_traffic_stats(q, mpctl,
9695 	    ipst, legacy_req)) == NULL) {
9696 		return (1);
9697 	}
9698 
9699 	if ((mpctl = ip_snmp_get_mib2_ip6(q, mpctl, ipst,
9700 	    legacy_req)) == NULL) {
9701 		return (1);
9702 	}
9703 
9704 	if ((mpctl = ip_snmp_get_mib2_icmp(q, mpctl, ipst)) == NULL) {
9705 		return (1);
9706 	}
9707 
9708 	if ((mpctl = ip_snmp_get_mib2_icmp6(q, mpctl, ipst)) == NULL) {
9709 		return (1);
9710 	}
9711 
9712 	if ((mpctl = ip_snmp_get_mib2_igmp(q, mpctl, ipst)) == NULL) {
9713 		return (1);
9714 	}
9715 
9716 	if ((mpctl = ip_snmp_get_mib2_multi(q, mpctl, ipst)) == NULL) {
9717 		return (1);
9718 	}
9719 
9720 	if ((mpctl = ip_snmp_get_mib2_ip_addr(q, mpctl, ipst,
9721 	    legacy_req)) == NULL) {
9722 		return (1);
9723 	}
9724 
9725 	if ((mpctl = ip_snmp_get_mib2_ip6_addr(q, mpctl, ipst,
9726 	    legacy_req)) == NULL) {
9727 		return (1);
9728 	}
9729 
9730 	if ((mpctl = ip_snmp_get_mib2_ip_group_mem(q, mpctl, ipst)) == NULL) {
9731 		return (1);
9732 	}
9733 
9734 	if ((mpctl = ip_snmp_get_mib2_ip6_group_mem(q, mpctl, ipst)) == NULL) {
9735 		return (1);
9736 	}
9737 
9738 	if ((mpctl = ip_snmp_get_mib2_ip_group_src(q, mpctl, ipst)) == NULL) {
9739 		return (1);
9740 	}
9741 
9742 	if ((mpctl = ip_snmp_get_mib2_ip6_group_src(q, mpctl, ipst)) == NULL) {
9743 		return (1);
9744 	}
9745 
9746 	if ((mpctl = ip_snmp_get_mib2_virt_multi(q, mpctl, ipst)) == NULL) {
9747 		return (1);
9748 	}
9749 
9750 	if ((mpctl = ip_snmp_get_mib2_multi_rtable(q, mpctl, ipst)) == NULL) {
9751 		return (1);
9752 	}
9753 
9754 	mpctl = ip_snmp_get_mib2_ip_route_media(q, mpctl, level, ipst);
9755 	if (mpctl == NULL)
9756 		return (1);
9757 
9758 	mpctl = ip_snmp_get_mib2_ip6_route_media(q, mpctl, level, ipst);
9759 	if (mpctl == NULL)
9760 		return (1);
9761 
9762 	if ((mpctl = sctp_snmp_get_mib2(q, mpctl, sctps)) == NULL) {
9763 		return (1);
9764 	}
9765 	if ((mpctl = ip_snmp_get_mib2_ip_dce(q, mpctl, ipst)) == NULL) {
9766 		return (1);
9767 	}
9768 done:
9769 	freemsg(mpctl);
9770 	return (1);
9771 }
9772 
9773 /* Get global (legacy) IPv4 statistics */
9774 static mblk_t *
ip_snmp_get_mib2_ip(queue_t * q,mblk_t * mpctl,mib2_ipIfStatsEntry_t * ipmib,ip_stack_t * ipst,boolean_t legacy_req)9775 ip_snmp_get_mib2_ip(queue_t *q, mblk_t *mpctl, mib2_ipIfStatsEntry_t *ipmib,
9776     ip_stack_t *ipst, boolean_t legacy_req)
9777 {
9778 	mib2_ip_t		old_ip_mib;
9779 	struct opthdr		*optp;
9780 	mblk_t			*mp2ctl;
9781 	mib2_ipAddrEntry_t	mae;
9782 
9783 	/*
9784 	 * make a copy of the original message
9785 	 */
9786 	mp2ctl = copymsg(mpctl);
9787 
9788 	/* fixed length IP structure... */
9789 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9790 	optp->level = MIB2_IP;
9791 	optp->name = 0;
9792 	SET_MIB(old_ip_mib.ipForwarding,
9793 	    (WE_ARE_FORWARDING(ipst) ? 1 : 2));
9794 	SET_MIB(old_ip_mib.ipDefaultTTL,
9795 	    (uint32_t)ipst->ips_ip_def_ttl);
9796 	SET_MIB(old_ip_mib.ipReasmTimeout,
9797 	    ipst->ips_ip_reassembly_timeout);
9798 	SET_MIB(old_ip_mib.ipAddrEntrySize,
9799 	    (legacy_req) ? LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t) :
9800 	    sizeof (mib2_ipAddrEntry_t));
9801 	SET_MIB(old_ip_mib.ipRouteEntrySize,
9802 	    sizeof (mib2_ipRouteEntry_t));
9803 	SET_MIB(old_ip_mib.ipNetToMediaEntrySize,
9804 	    sizeof (mib2_ipNetToMediaEntry_t));
9805 	SET_MIB(old_ip_mib.ipMemberEntrySize, sizeof (ip_member_t));
9806 	SET_MIB(old_ip_mib.ipGroupSourceEntrySize, sizeof (ip_grpsrc_t));
9807 	SET_MIB(old_ip_mib.ipRouteAttributeSize,
9808 	    sizeof (mib2_ipAttributeEntry_t));
9809 	SET_MIB(old_ip_mib.transportMLPSize, sizeof (mib2_transportMLPEntry_t));
9810 	SET_MIB(old_ip_mib.ipDestEntrySize, sizeof (dest_cache_entry_t));
9811 
9812 	/*
9813 	 * Grab the statistics from the new IP MIB
9814 	 */
9815 	SET_MIB(old_ip_mib.ipInReceives,
9816 	    (uint32_t)ipmib->ipIfStatsHCInReceives);
9817 	SET_MIB(old_ip_mib.ipInHdrErrors, ipmib->ipIfStatsInHdrErrors);
9818 	SET_MIB(old_ip_mib.ipInAddrErrors, ipmib->ipIfStatsInAddrErrors);
9819 	SET_MIB(old_ip_mib.ipForwDatagrams,
9820 	    (uint32_t)ipmib->ipIfStatsHCOutForwDatagrams);
9821 	SET_MIB(old_ip_mib.ipInUnknownProtos,
9822 	    ipmib->ipIfStatsInUnknownProtos);
9823 	SET_MIB(old_ip_mib.ipInDiscards, ipmib->ipIfStatsInDiscards);
9824 	SET_MIB(old_ip_mib.ipInDelivers,
9825 	    (uint32_t)ipmib->ipIfStatsHCInDelivers);
9826 	SET_MIB(old_ip_mib.ipOutRequests,
9827 	    (uint32_t)ipmib->ipIfStatsHCOutRequests);
9828 	SET_MIB(old_ip_mib.ipOutDiscards, ipmib->ipIfStatsOutDiscards);
9829 	SET_MIB(old_ip_mib.ipOutNoRoutes, ipmib->ipIfStatsOutNoRoutes);
9830 	SET_MIB(old_ip_mib.ipReasmReqds, ipmib->ipIfStatsReasmReqds);
9831 	SET_MIB(old_ip_mib.ipReasmOKs, ipmib->ipIfStatsReasmOKs);
9832 	SET_MIB(old_ip_mib.ipReasmFails, ipmib->ipIfStatsReasmFails);
9833 	SET_MIB(old_ip_mib.ipFragOKs, ipmib->ipIfStatsOutFragOKs);
9834 	SET_MIB(old_ip_mib.ipFragFails, ipmib->ipIfStatsOutFragFails);
9835 	SET_MIB(old_ip_mib.ipFragCreates, ipmib->ipIfStatsOutFragCreates);
9836 
9837 	/* ipRoutingDiscards is not being used */
9838 	SET_MIB(old_ip_mib.ipRoutingDiscards, 0);
9839 	SET_MIB(old_ip_mib.tcpInErrs, ipmib->tcpIfStatsInErrs);
9840 	SET_MIB(old_ip_mib.udpNoPorts, ipmib->udpIfStatsNoPorts);
9841 	SET_MIB(old_ip_mib.ipInCksumErrs, ipmib->ipIfStatsInCksumErrs);
9842 	SET_MIB(old_ip_mib.ipReasmDuplicates,
9843 	    ipmib->ipIfStatsReasmDuplicates);
9844 	SET_MIB(old_ip_mib.ipReasmPartDups, ipmib->ipIfStatsReasmPartDups);
9845 	SET_MIB(old_ip_mib.ipForwProhibits, ipmib->ipIfStatsForwProhibits);
9846 	SET_MIB(old_ip_mib.udpInCksumErrs, ipmib->udpIfStatsInCksumErrs);
9847 	SET_MIB(old_ip_mib.udpInOverflows, ipmib->udpIfStatsInOverflows);
9848 	SET_MIB(old_ip_mib.rawipInOverflows,
9849 	    ipmib->rawipIfStatsInOverflows);
9850 
9851 	SET_MIB(old_ip_mib.ipsecInSucceeded, ipmib->ipsecIfStatsInSucceeded);
9852 	SET_MIB(old_ip_mib.ipsecInFailed, ipmib->ipsecIfStatsInFailed);
9853 	SET_MIB(old_ip_mib.ipInIPv6, ipmib->ipIfStatsInWrongIPVersion);
9854 	SET_MIB(old_ip_mib.ipOutIPv6, ipmib->ipIfStatsOutWrongIPVersion);
9855 	SET_MIB(old_ip_mib.ipOutSwitchIPv6,
9856 	    ipmib->ipIfStatsOutSwitchIPVersion);
9857 
9858 	if (!snmp_append_data(mpctl->b_cont, (char *)&old_ip_mib,
9859 	    (int)sizeof (old_ip_mib))) {
9860 		ip1dbg(("ip_snmp_get_mib2_ip: failed to allocate %u bytes\n",
9861 		    (uint_t)sizeof (old_ip_mib)));
9862 	}
9863 
9864 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9865 	ip3dbg(("ip_snmp_get_mib2_ip: level %d, name %d, len %d\n",
9866 	    (int)optp->level, (int)optp->name, (int)optp->len));
9867 	qreply(q, mpctl);
9868 	return (mp2ctl);
9869 }
9870 
9871 /* Per interface IPv4 statistics */
9872 static mblk_t *
ip_snmp_get_mib2_ip_traffic_stats(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst,boolean_t legacy_req)9873 ip_snmp_get_mib2_ip_traffic_stats(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
9874     boolean_t legacy_req)
9875 {
9876 	struct opthdr		*optp;
9877 	mblk_t			*mp2ctl;
9878 	ill_t			*ill;
9879 	ill_walk_context_t	ctx;
9880 	mblk_t			*mp_tail = NULL;
9881 	mib2_ipIfStatsEntry_t	global_ip_mib;
9882 	mib2_ipAddrEntry_t	mae;
9883 
9884 	/*
9885 	 * Make a copy of the original message
9886 	 */
9887 	mp2ctl = copymsg(mpctl);
9888 
9889 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9890 	optp->level = MIB2_IP;
9891 	optp->name = MIB2_IP_TRAFFIC_STATS;
9892 	/* Include "unknown interface" ip_mib */
9893 	ipst->ips_ip_mib.ipIfStatsIPVersion = MIB2_INETADDRESSTYPE_ipv4;
9894 	ipst->ips_ip_mib.ipIfStatsIfIndex =
9895 	    MIB2_UNKNOWN_INTERFACE; /* Flag to netstat */
9896 	SET_MIB(ipst->ips_ip_mib.ipIfStatsForwarding,
9897 	    (ipst->ips_ip_forwarding ? 1 : 2));
9898 	SET_MIB(ipst->ips_ip_mib.ipIfStatsDefaultTTL,
9899 	    (uint32_t)ipst->ips_ip_def_ttl);
9900 	SET_MIB(ipst->ips_ip_mib.ipIfStatsEntrySize,
9901 	    sizeof (mib2_ipIfStatsEntry_t));
9902 	SET_MIB(ipst->ips_ip_mib.ipIfStatsAddrEntrySize,
9903 	    sizeof (mib2_ipAddrEntry_t));
9904 	SET_MIB(ipst->ips_ip_mib.ipIfStatsRouteEntrySize,
9905 	    sizeof (mib2_ipRouteEntry_t));
9906 	SET_MIB(ipst->ips_ip_mib.ipIfStatsNetToMediaEntrySize,
9907 	    sizeof (mib2_ipNetToMediaEntry_t));
9908 	SET_MIB(ipst->ips_ip_mib.ipIfStatsMemberEntrySize,
9909 	    sizeof (ip_member_t));
9910 	SET_MIB(ipst->ips_ip_mib.ipIfStatsGroupSourceEntrySize,
9911 	    sizeof (ip_grpsrc_t));
9912 
9913 	bcopy(&ipst->ips_ip_mib, &global_ip_mib, sizeof (global_ip_mib));
9914 
9915 	if (legacy_req) {
9916 		SET_MIB(global_ip_mib.ipIfStatsAddrEntrySize,
9917 		    LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t));
9918 	}
9919 
9920 	if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
9921 	    (char *)&global_ip_mib, (int)sizeof (global_ip_mib))) {
9922 		ip1dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9923 		    "failed to allocate %u bytes\n",
9924 		    (uint_t)sizeof (global_ip_mib)));
9925 	}
9926 
9927 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
9928 	ill = ILL_START_WALK_V4(&ctx, ipst);
9929 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
9930 		ill->ill_ip_mib->ipIfStatsIfIndex =
9931 		    ill->ill_phyint->phyint_ifindex;
9932 		SET_MIB(ill->ill_ip_mib->ipIfStatsForwarding,
9933 		    (ipst->ips_ip_forwarding ? 1 : 2));
9934 		SET_MIB(ill->ill_ip_mib->ipIfStatsDefaultTTL,
9935 		    (uint32_t)ipst->ips_ip_def_ttl);
9936 
9937 		ip_mib2_add_ip_stats(&global_ip_mib, ill->ill_ip_mib);
9938 		if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
9939 		    (char *)ill->ill_ip_mib,
9940 		    (int)sizeof (*ill->ill_ip_mib))) {
9941 			ip1dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9942 			    "failed to allocate %u bytes\n",
9943 			    (uint_t)sizeof (*ill->ill_ip_mib)));
9944 		}
9945 	}
9946 	rw_exit(&ipst->ips_ill_g_lock);
9947 
9948 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9949 	ip3dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9950 	    "level %d, name %d, len %d\n",
9951 	    (int)optp->level, (int)optp->name, (int)optp->len));
9952 	qreply(q, mpctl);
9953 
9954 	if (mp2ctl == NULL)
9955 		return (NULL);
9956 
9957 	return (ip_snmp_get_mib2_ip(q, mp2ctl, &global_ip_mib, ipst,
9958 	    legacy_req));
9959 }
9960 
9961 /* Global IPv4 ICMP statistics */
9962 static mblk_t *
ip_snmp_get_mib2_icmp(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)9963 ip_snmp_get_mib2_icmp(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
9964 {
9965 	struct opthdr		*optp;
9966 	mblk_t			*mp2ctl;
9967 
9968 	/*
9969 	 * Make a copy of the original message
9970 	 */
9971 	mp2ctl = copymsg(mpctl);
9972 
9973 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9974 	optp->level = MIB2_ICMP;
9975 	optp->name = 0;
9976 	if (!snmp_append_data(mpctl->b_cont, (char *)&ipst->ips_icmp_mib,
9977 	    (int)sizeof (ipst->ips_icmp_mib))) {
9978 		ip1dbg(("ip_snmp_get_mib2_icmp: failed to allocate %u bytes\n",
9979 		    (uint_t)sizeof (ipst->ips_icmp_mib)));
9980 	}
9981 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9982 	ip3dbg(("ip_snmp_get_mib2_icmp: level %d, name %d, len %d\n",
9983 	    (int)optp->level, (int)optp->name, (int)optp->len));
9984 	qreply(q, mpctl);
9985 	return (mp2ctl);
9986 }
9987 
9988 /* Global IPv4 IGMP statistics */
9989 static mblk_t *
ip_snmp_get_mib2_igmp(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)9990 ip_snmp_get_mib2_igmp(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
9991 {
9992 	struct opthdr		*optp;
9993 	mblk_t			*mp2ctl;
9994 
9995 	/*
9996 	 * make a copy of the original message
9997 	 */
9998 	mp2ctl = copymsg(mpctl);
9999 
10000 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10001 	optp->level = EXPER_IGMP;
10002 	optp->name = 0;
10003 	if (!snmp_append_data(mpctl->b_cont, (char *)&ipst->ips_igmpstat,
10004 	    (int)sizeof (ipst->ips_igmpstat))) {
10005 		ip1dbg(("ip_snmp_get_mib2_igmp: failed to allocate %u bytes\n",
10006 		    (uint_t)sizeof (ipst->ips_igmpstat)));
10007 	}
10008 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10009 	ip3dbg(("ip_snmp_get_mib2_igmp: level %d, name %d, len %d\n",
10010 	    (int)optp->level, (int)optp->name, (int)optp->len));
10011 	qreply(q, mpctl);
10012 	return (mp2ctl);
10013 }
10014 
10015 /* Global IPv4 Multicast Routing statistics */
10016 static mblk_t *
ip_snmp_get_mib2_multi(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10017 ip_snmp_get_mib2_multi(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10018 {
10019 	struct opthdr		*optp;
10020 	mblk_t			*mp2ctl;
10021 
10022 	/*
10023 	 * make a copy of the original message
10024 	 */
10025 	mp2ctl = copymsg(mpctl);
10026 
10027 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10028 	optp->level = EXPER_DVMRP;
10029 	optp->name = 0;
10030 	if (!ip_mroute_stats(mpctl->b_cont, ipst)) {
10031 		ip0dbg(("ip_mroute_stats: failed\n"));
10032 	}
10033 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10034 	ip3dbg(("ip_snmp_get_mib2_multi: level %d, name %d, len %d\n",
10035 	    (int)optp->level, (int)optp->name, (int)optp->len));
10036 	qreply(q, mpctl);
10037 	return (mp2ctl);
10038 }
10039 
10040 /* IPv4 address information */
10041 static mblk_t *
ip_snmp_get_mib2_ip_addr(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst,boolean_t legacy_req)10042 ip_snmp_get_mib2_ip_addr(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
10043     boolean_t legacy_req)
10044 {
10045 	struct opthdr		*optp;
10046 	mblk_t			*mp2ctl;
10047 	mblk_t			*mp_tail = NULL;
10048 	ill_t			*ill;
10049 	ipif_t			*ipif;
10050 	uint_t			bitval;
10051 	mib2_ipAddrEntry_t	mae;
10052 	size_t			mae_size;
10053 	zoneid_t		zoneid;
10054 	ill_walk_context_t	ctx;
10055 
10056 	/*
10057 	 * make a copy of the original message
10058 	 */
10059 	mp2ctl = copymsg(mpctl);
10060 
10061 	mae_size = (legacy_req) ? LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t) :
10062 	    sizeof (mib2_ipAddrEntry_t);
10063 
10064 	/* ipAddrEntryTable */
10065 
10066 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10067 	optp->level = MIB2_IP;
10068 	optp->name = MIB2_IP_ADDR;
10069 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10070 
10071 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10072 	ill = ILL_START_WALK_V4(&ctx, ipst);
10073 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10074 		for (ipif = ill->ill_ipif; ipif != NULL;
10075 		    ipif = ipif->ipif_next) {
10076 			if (ipif->ipif_zoneid != zoneid &&
10077 			    ipif->ipif_zoneid != ALL_ZONES)
10078 				continue;
10079 			/* Sum of count from dead IRE_LO* and our current */
10080 			mae.ipAdEntInfo.ae_ibcnt = ipif->ipif_ib_pkt_count;
10081 			if (ipif->ipif_ire_local != NULL) {
10082 				mae.ipAdEntInfo.ae_ibcnt +=
10083 				    ipif->ipif_ire_local->ire_ib_pkt_count;
10084 			}
10085 			mae.ipAdEntInfo.ae_obcnt = 0;
10086 			mae.ipAdEntInfo.ae_focnt = 0;
10087 
10088 			ipif_get_name(ipif, mae.ipAdEntIfIndex.o_bytes,
10089 			    OCTET_LENGTH);
10090 			mae.ipAdEntIfIndex.o_length =
10091 			    mi_strlen(mae.ipAdEntIfIndex.o_bytes);
10092 			mae.ipAdEntAddr = ipif->ipif_lcl_addr;
10093 			mae.ipAdEntNetMask = ipif->ipif_net_mask;
10094 			mae.ipAdEntInfo.ae_subnet = ipif->ipif_subnet;
10095 			mae.ipAdEntInfo.ae_subnet_len =
10096 			    ip_mask_to_plen(ipif->ipif_net_mask);
10097 			mae.ipAdEntInfo.ae_src_addr = ipif->ipif_lcl_addr;
10098 			for (bitval = 1;
10099 			    bitval &&
10100 			    !(bitval & ipif->ipif_brd_addr);
10101 			    bitval <<= 1)
10102 				noop;
10103 			mae.ipAdEntBcastAddr = bitval;
10104 			mae.ipAdEntReasmMaxSize = IP_MAXPACKET;
10105 			mae.ipAdEntInfo.ae_mtu = ipif->ipif_ill->ill_mtu;
10106 			mae.ipAdEntInfo.ae_metric  = ipif->ipif_ill->ill_metric;
10107 			mae.ipAdEntInfo.ae_broadcast_addr =
10108 			    ipif->ipif_brd_addr;
10109 			mae.ipAdEntInfo.ae_pp_dst_addr =
10110 			    ipif->ipif_pp_dst_addr;
10111 			mae.ipAdEntInfo.ae_flags = ipif->ipif_flags |
10112 			    ill->ill_flags | ill->ill_phyint->phyint_flags;
10113 			mae.ipAdEntRetransmitTime =
10114 			    ill->ill_reachable_retrans_time;
10115 
10116 			if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10117 			    (char *)&mae, (int)mae_size)) {
10118 				ip1dbg(("ip_snmp_get_mib2_ip_addr: failed to "
10119 				    "allocate %u bytes\n", (uint_t)mae_size));
10120 			}
10121 		}
10122 	}
10123 	rw_exit(&ipst->ips_ill_g_lock);
10124 
10125 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10126 	ip3dbg(("ip_snmp_get_mib2_ip_addr: level %d, name %d, len %d\n",
10127 	    (int)optp->level, (int)optp->name, (int)optp->len));
10128 	qreply(q, mpctl);
10129 	return (mp2ctl);
10130 }
10131 
10132 /* IPv6 address information */
10133 static mblk_t *
ip_snmp_get_mib2_ip6_addr(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst,boolean_t legacy_req)10134 ip_snmp_get_mib2_ip6_addr(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
10135     boolean_t legacy_req)
10136 {
10137 	struct opthdr		*optp;
10138 	mblk_t			*mp2ctl;
10139 	mblk_t			*mp_tail = NULL;
10140 	ill_t			*ill;
10141 	ipif_t			*ipif;
10142 	mib2_ipv6AddrEntry_t	mae6;
10143 	size_t			mae6_size;
10144 	zoneid_t		zoneid;
10145 	ill_walk_context_t	ctx;
10146 
10147 	/*
10148 	 * make a copy of the original message
10149 	 */
10150 	mp2ctl = copymsg(mpctl);
10151 
10152 	mae6_size = (legacy_req) ?
10153 	    LEGACY_MIB_SIZE(&mae6, mib2_ipv6AddrEntry_t) :
10154 	    sizeof (mib2_ipv6AddrEntry_t);
10155 
10156 	/* ipv6AddrEntryTable */
10157 
10158 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10159 	optp->level = MIB2_IP6;
10160 	optp->name = MIB2_IP6_ADDR;
10161 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10162 
10163 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10164 	ill = ILL_START_WALK_V6(&ctx, ipst);
10165 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10166 		for (ipif = ill->ill_ipif; ipif != NULL;
10167 		    ipif = ipif->ipif_next) {
10168 			if (ipif->ipif_zoneid != zoneid &&
10169 			    ipif->ipif_zoneid != ALL_ZONES)
10170 				continue;
10171 			/* Sum of count from dead IRE_LO* and our current */
10172 			mae6.ipv6AddrInfo.ae_ibcnt = ipif->ipif_ib_pkt_count;
10173 			if (ipif->ipif_ire_local != NULL) {
10174 				mae6.ipv6AddrInfo.ae_ibcnt +=
10175 				    ipif->ipif_ire_local->ire_ib_pkt_count;
10176 			}
10177 			mae6.ipv6AddrInfo.ae_obcnt = 0;
10178 			mae6.ipv6AddrInfo.ae_focnt = 0;
10179 
10180 			ipif_get_name(ipif, mae6.ipv6AddrIfIndex.o_bytes,
10181 			    OCTET_LENGTH);
10182 			mae6.ipv6AddrIfIndex.o_length =
10183 			    mi_strlen(mae6.ipv6AddrIfIndex.o_bytes);
10184 			mae6.ipv6AddrAddress = ipif->ipif_v6lcl_addr;
10185 			mae6.ipv6AddrPfxLength =
10186 			    ip_mask_to_plen_v6(&ipif->ipif_v6net_mask);
10187 			mae6.ipv6AddrInfo.ae_subnet = ipif->ipif_v6subnet;
10188 			mae6.ipv6AddrInfo.ae_subnet_len =
10189 			    mae6.ipv6AddrPfxLength;
10190 			mae6.ipv6AddrInfo.ae_src_addr = ipif->ipif_v6lcl_addr;
10191 
10192 			/* Type: stateless(1), stateful(2), unknown(3) */
10193 			if (ipif->ipif_flags & IPIF_ADDRCONF)
10194 				mae6.ipv6AddrType = 1;
10195 			else
10196 				mae6.ipv6AddrType = 2;
10197 			/* Anycast: true(1), false(2) */
10198 			if (ipif->ipif_flags & IPIF_ANYCAST)
10199 				mae6.ipv6AddrAnycastFlag = 1;
10200 			else
10201 				mae6.ipv6AddrAnycastFlag = 2;
10202 
10203 			/*
10204 			 * Address status: preferred(1), deprecated(2),
10205 			 * invalid(3), inaccessible(4), unknown(5)
10206 			 */
10207 			if (ipif->ipif_flags & IPIF_NOLOCAL)
10208 				mae6.ipv6AddrStatus = 3;
10209 			else if (ipif->ipif_flags & IPIF_DEPRECATED)
10210 				mae6.ipv6AddrStatus = 2;
10211 			else
10212 				mae6.ipv6AddrStatus = 1;
10213 			mae6.ipv6AddrInfo.ae_mtu = ipif->ipif_ill->ill_mtu;
10214 			mae6.ipv6AddrInfo.ae_metric  =
10215 			    ipif->ipif_ill->ill_metric;
10216 			mae6.ipv6AddrInfo.ae_pp_dst_addr =
10217 			    ipif->ipif_v6pp_dst_addr;
10218 			mae6.ipv6AddrInfo.ae_flags = ipif->ipif_flags |
10219 			    ill->ill_flags | ill->ill_phyint->phyint_flags;
10220 			mae6.ipv6AddrReasmMaxSize = IP_MAXPACKET;
10221 			mae6.ipv6AddrIdentifier = ill->ill_token;
10222 			mae6.ipv6AddrIdentifierLen = ill->ill_token_length;
10223 			mae6.ipv6AddrReachableTime = ill->ill_reachable_time;
10224 			mae6.ipv6AddrRetransmitTime =
10225 			    ill->ill_reachable_retrans_time;
10226 			if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10227 			    (char *)&mae6, (int)mae6_size)) {
10228 				ip1dbg(("ip_snmp_get_mib2_ip6_addr: failed to "
10229 				    "allocate %u bytes\n",
10230 				    (uint_t)mae6_size));
10231 			}
10232 		}
10233 	}
10234 	rw_exit(&ipst->ips_ill_g_lock);
10235 
10236 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10237 	ip3dbg(("ip_snmp_get_mib2_ip6_addr: level %d, name %d, len %d\n",
10238 	    (int)optp->level, (int)optp->name, (int)optp->len));
10239 	qreply(q, mpctl);
10240 	return (mp2ctl);
10241 }
10242 
10243 /* IPv4 multicast group membership. */
10244 static mblk_t *
ip_snmp_get_mib2_ip_group_mem(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10245 ip_snmp_get_mib2_ip_group_mem(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10246 {
10247 	struct opthdr		*optp;
10248 	mblk_t			*mp2ctl;
10249 	ill_t			*ill;
10250 	ipif_t			*ipif;
10251 	ilm_t			*ilm;
10252 	ip_member_t		ipm;
10253 	mblk_t			*mp_tail = NULL;
10254 	ill_walk_context_t	ctx;
10255 	zoneid_t		zoneid;
10256 
10257 	/*
10258 	 * make a copy of the original message
10259 	 */
10260 	mp2ctl = copymsg(mpctl);
10261 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10262 
10263 	/* ipGroupMember table */
10264 	optp = (struct opthdr *)&mpctl->b_rptr[
10265 	    sizeof (struct T_optmgmt_ack)];
10266 	optp->level = MIB2_IP;
10267 	optp->name = EXPER_IP_GROUP_MEMBERSHIP;
10268 
10269 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10270 	ill = ILL_START_WALK_V4(&ctx, ipst);
10271 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10272 		/* Make sure the ill isn't going away. */
10273 		if (!ill_check_and_refhold(ill))
10274 			continue;
10275 		rw_exit(&ipst->ips_ill_g_lock);
10276 		rw_enter(&ill->ill_mcast_lock, RW_READER);
10277 		for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10278 			if (ilm->ilm_zoneid != zoneid &&
10279 			    ilm->ilm_zoneid != ALL_ZONES)
10280 				continue;
10281 
10282 			/* Is there an ipif for ilm_ifaddr? */
10283 			for (ipif = ill->ill_ipif; ipif != NULL;
10284 			    ipif = ipif->ipif_next) {
10285 				if (!IPIF_IS_CONDEMNED(ipif) &&
10286 				    ipif->ipif_lcl_addr == ilm->ilm_ifaddr &&
10287 				    ilm->ilm_ifaddr != INADDR_ANY)
10288 					break;
10289 			}
10290 			if (ipif != NULL) {
10291 				ipif_get_name(ipif,
10292 				    ipm.ipGroupMemberIfIndex.o_bytes,
10293 				    OCTET_LENGTH);
10294 			} else {
10295 				ill_get_name(ill,
10296 				    ipm.ipGroupMemberIfIndex.o_bytes,
10297 				    OCTET_LENGTH);
10298 			}
10299 			ipm.ipGroupMemberIfIndex.o_length =
10300 			    mi_strlen(ipm.ipGroupMemberIfIndex.o_bytes);
10301 
10302 			ipm.ipGroupMemberAddress = ilm->ilm_addr;
10303 			ipm.ipGroupMemberRefCnt = ilm->ilm_refcnt;
10304 			ipm.ipGroupMemberFilterMode = ilm->ilm_fmode;
10305 			if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10306 			    (char *)&ipm, (int)sizeof (ipm))) {
10307 				ip1dbg(("ip_snmp_get_mib2_ip_group: "
10308 				    "failed to allocate %u bytes\n",
10309 				    (uint_t)sizeof (ipm)));
10310 			}
10311 		}
10312 		rw_exit(&ill->ill_mcast_lock);
10313 		ill_refrele(ill);
10314 		rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10315 	}
10316 	rw_exit(&ipst->ips_ill_g_lock);
10317 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10318 	ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10319 	    (int)optp->level, (int)optp->name, (int)optp->len));
10320 	qreply(q, mpctl);
10321 	return (mp2ctl);
10322 }
10323 
10324 /* IPv6 multicast group membership. */
10325 static mblk_t *
ip_snmp_get_mib2_ip6_group_mem(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10326 ip_snmp_get_mib2_ip6_group_mem(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10327 {
10328 	struct opthdr		*optp;
10329 	mblk_t			*mp2ctl;
10330 	ill_t			*ill;
10331 	ilm_t			*ilm;
10332 	ipv6_member_t		ipm6;
10333 	mblk_t			*mp_tail = NULL;
10334 	ill_walk_context_t	ctx;
10335 	zoneid_t		zoneid;
10336 
10337 	/*
10338 	 * make a copy of the original message
10339 	 */
10340 	mp2ctl = copymsg(mpctl);
10341 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10342 
10343 	/* ip6GroupMember table */
10344 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10345 	optp->level = MIB2_IP6;
10346 	optp->name = EXPER_IP6_GROUP_MEMBERSHIP;
10347 
10348 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10349 	ill = ILL_START_WALK_V6(&ctx, ipst);
10350 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10351 		/* Make sure the ill isn't going away. */
10352 		if (!ill_check_and_refhold(ill))
10353 			continue;
10354 		rw_exit(&ipst->ips_ill_g_lock);
10355 		/*
10356 		 * Normally we don't have any members on under IPMP interfaces.
10357 		 * We report them as a debugging aid.
10358 		 */
10359 		rw_enter(&ill->ill_mcast_lock, RW_READER);
10360 		ipm6.ipv6GroupMemberIfIndex = ill->ill_phyint->phyint_ifindex;
10361 		for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10362 			if (ilm->ilm_zoneid != zoneid &&
10363 			    ilm->ilm_zoneid != ALL_ZONES)
10364 				continue;	/* not this zone */
10365 			ipm6.ipv6GroupMemberAddress = ilm->ilm_v6addr;
10366 			ipm6.ipv6GroupMemberRefCnt = ilm->ilm_refcnt;
10367 			ipm6.ipv6GroupMemberFilterMode = ilm->ilm_fmode;
10368 			if (!snmp_append_data2(mpctl->b_cont,
10369 			    &mp_tail,
10370 			    (char *)&ipm6, (int)sizeof (ipm6))) {
10371 				ip1dbg(("ip_snmp_get_mib2_ip6_group: "
10372 				    "failed to allocate %u bytes\n",
10373 				    (uint_t)sizeof (ipm6)));
10374 			}
10375 		}
10376 		rw_exit(&ill->ill_mcast_lock);
10377 		ill_refrele(ill);
10378 		rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10379 	}
10380 	rw_exit(&ipst->ips_ill_g_lock);
10381 
10382 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10383 	ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10384 	    (int)optp->level, (int)optp->name, (int)optp->len));
10385 	qreply(q, mpctl);
10386 	return (mp2ctl);
10387 }
10388 
10389 /* IP multicast filtered sources */
10390 static mblk_t *
ip_snmp_get_mib2_ip_group_src(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10391 ip_snmp_get_mib2_ip_group_src(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10392 {
10393 	struct opthdr		*optp;
10394 	mblk_t			*mp2ctl;
10395 	ill_t			*ill;
10396 	ipif_t			*ipif;
10397 	ilm_t			*ilm;
10398 	ip_grpsrc_t		ips;
10399 	mblk_t			*mp_tail = NULL;
10400 	ill_walk_context_t	ctx;
10401 	zoneid_t		zoneid;
10402 	int			i;
10403 	slist_t			*sl;
10404 
10405 	/*
10406 	 * make a copy of the original message
10407 	 */
10408 	mp2ctl = copymsg(mpctl);
10409 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10410 
10411 	/* ipGroupSource table */
10412 	optp = (struct opthdr *)&mpctl->b_rptr[
10413 	    sizeof (struct T_optmgmt_ack)];
10414 	optp->level = MIB2_IP;
10415 	optp->name = EXPER_IP_GROUP_SOURCES;
10416 
10417 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10418 	ill = ILL_START_WALK_V4(&ctx, ipst);
10419 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10420 		/* Make sure the ill isn't going away. */
10421 		if (!ill_check_and_refhold(ill))
10422 			continue;
10423 		rw_exit(&ipst->ips_ill_g_lock);
10424 		rw_enter(&ill->ill_mcast_lock, RW_READER);
10425 		for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10426 			sl = ilm->ilm_filter;
10427 			if (ilm->ilm_zoneid != zoneid &&
10428 			    ilm->ilm_zoneid != ALL_ZONES)
10429 				continue;
10430 			if (SLIST_IS_EMPTY(sl))
10431 				continue;
10432 
10433 			/* Is there an ipif for ilm_ifaddr? */
10434 			for (ipif = ill->ill_ipif; ipif != NULL;
10435 			    ipif = ipif->ipif_next) {
10436 				if (!IPIF_IS_CONDEMNED(ipif) &&
10437 				    ipif->ipif_lcl_addr == ilm->ilm_ifaddr &&
10438 				    ilm->ilm_ifaddr != INADDR_ANY)
10439 					break;
10440 			}
10441 			if (ipif != NULL) {
10442 				ipif_get_name(ipif,
10443 				    ips.ipGroupSourceIfIndex.o_bytes,
10444 				    OCTET_LENGTH);
10445 			} else {
10446 				ill_get_name(ill,
10447 				    ips.ipGroupSourceIfIndex.o_bytes,
10448 				    OCTET_LENGTH);
10449 			}
10450 			ips.ipGroupSourceIfIndex.o_length =
10451 			    mi_strlen(ips.ipGroupSourceIfIndex.o_bytes);
10452 
10453 			ips.ipGroupSourceGroup = ilm->ilm_addr;
10454 			for (i = 0; i < sl->sl_numsrc; i++) {
10455 				if (!IN6_IS_ADDR_V4MAPPED(&sl->sl_addr[i]))
10456 					continue;
10457 				IN6_V4MAPPED_TO_IPADDR(&sl->sl_addr[i],
10458 				    ips.ipGroupSourceAddress);
10459 				if (snmp_append_data2(mpctl->b_cont, &mp_tail,
10460 				    (char *)&ips, (int)sizeof (ips)) == 0) {
10461 					ip1dbg(("ip_snmp_get_mib2_ip_group_src:"
10462 					    " failed to allocate %u bytes\n",
10463 					    (uint_t)sizeof (ips)));
10464 				}
10465 			}
10466 		}
10467 		rw_exit(&ill->ill_mcast_lock);
10468 		ill_refrele(ill);
10469 		rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10470 	}
10471 	rw_exit(&ipst->ips_ill_g_lock);
10472 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10473 	ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10474 	    (int)optp->level, (int)optp->name, (int)optp->len));
10475 	qreply(q, mpctl);
10476 	return (mp2ctl);
10477 }
10478 
10479 /* IPv6 multicast filtered sources. */
10480 static mblk_t *
ip_snmp_get_mib2_ip6_group_src(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10481 ip_snmp_get_mib2_ip6_group_src(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10482 {
10483 	struct opthdr		*optp;
10484 	mblk_t			*mp2ctl;
10485 	ill_t			*ill;
10486 	ilm_t			*ilm;
10487 	ipv6_grpsrc_t		ips6;
10488 	mblk_t			*mp_tail = NULL;
10489 	ill_walk_context_t	ctx;
10490 	zoneid_t		zoneid;
10491 	int			i;
10492 	slist_t			*sl;
10493 
10494 	/*
10495 	 * make a copy of the original message
10496 	 */
10497 	mp2ctl = copymsg(mpctl);
10498 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10499 
10500 	/* ip6GroupMember table */
10501 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10502 	optp->level = MIB2_IP6;
10503 	optp->name = EXPER_IP6_GROUP_SOURCES;
10504 
10505 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10506 	ill = ILL_START_WALK_V6(&ctx, ipst);
10507 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10508 		/* Make sure the ill isn't going away. */
10509 		if (!ill_check_and_refhold(ill))
10510 			continue;
10511 		rw_exit(&ipst->ips_ill_g_lock);
10512 		/*
10513 		 * Normally we don't have any members on under IPMP interfaces.
10514 		 * We report them as a debugging aid.
10515 		 */
10516 		rw_enter(&ill->ill_mcast_lock, RW_READER);
10517 		ips6.ipv6GroupSourceIfIndex = ill->ill_phyint->phyint_ifindex;
10518 		for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10519 			sl = ilm->ilm_filter;
10520 			if (ilm->ilm_zoneid != zoneid &&
10521 			    ilm->ilm_zoneid != ALL_ZONES)
10522 				continue;
10523 			if (SLIST_IS_EMPTY(sl))
10524 				continue;
10525 			ips6.ipv6GroupSourceGroup = ilm->ilm_v6addr;
10526 			for (i = 0; i < sl->sl_numsrc; i++) {
10527 				ips6.ipv6GroupSourceAddress = sl->sl_addr[i];
10528 				if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10529 				    (char *)&ips6, (int)sizeof (ips6))) {
10530 					ip1dbg(("ip_snmp_get_mib2_ip6_"
10531 					    "group_src: failed to allocate "
10532 					    "%u bytes\n",
10533 					    (uint_t)sizeof (ips6)));
10534 				}
10535 			}
10536 		}
10537 		rw_exit(&ill->ill_mcast_lock);
10538 		ill_refrele(ill);
10539 		rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10540 	}
10541 	rw_exit(&ipst->ips_ill_g_lock);
10542 
10543 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10544 	ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10545 	    (int)optp->level, (int)optp->name, (int)optp->len));
10546 	qreply(q, mpctl);
10547 	return (mp2ctl);
10548 }
10549 
10550 /* Multicast routing virtual interface table. */
10551 static mblk_t *
ip_snmp_get_mib2_virt_multi(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10552 ip_snmp_get_mib2_virt_multi(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10553 {
10554 	struct opthdr		*optp;
10555 	mblk_t			*mp2ctl;
10556 
10557 	/*
10558 	 * make a copy of the original message
10559 	 */
10560 	mp2ctl = copymsg(mpctl);
10561 
10562 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10563 	optp->level = EXPER_DVMRP;
10564 	optp->name = EXPER_DVMRP_VIF;
10565 	if (!ip_mroute_vif(mpctl->b_cont, ipst)) {
10566 		ip0dbg(("ip_mroute_vif: failed\n"));
10567 	}
10568 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10569 	ip3dbg(("ip_snmp_get_mib2_virt_multi: level %d, name %d, len %d\n",
10570 	    (int)optp->level, (int)optp->name, (int)optp->len));
10571 	qreply(q, mpctl);
10572 	return (mp2ctl);
10573 }
10574 
10575 /* Multicast routing table. */
10576 static mblk_t *
ip_snmp_get_mib2_multi_rtable(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10577 ip_snmp_get_mib2_multi_rtable(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10578 {
10579 	struct opthdr		*optp;
10580 	mblk_t			*mp2ctl;
10581 
10582 	/*
10583 	 * make a copy of the original message
10584 	 */
10585 	mp2ctl = copymsg(mpctl);
10586 
10587 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10588 	optp->level = EXPER_DVMRP;
10589 	optp->name = EXPER_DVMRP_MRT;
10590 	if (!ip_mroute_mrt(mpctl->b_cont, ipst)) {
10591 		ip0dbg(("ip_mroute_mrt: failed\n"));
10592 	}
10593 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10594 	ip3dbg(("ip_snmp_get_mib2_multi_rtable: level %d, name %d, len %d\n",
10595 	    (int)optp->level, (int)optp->name, (int)optp->len));
10596 	qreply(q, mpctl);
10597 	return (mp2ctl);
10598 }
10599 
10600 /*
10601  * Return ipRouteEntryTable, ipNetToMediaEntryTable, and ipRouteAttributeTable
10602  * in one IRE walk.
10603  */
10604 static mblk_t *
ip_snmp_get_mib2_ip_route_media(queue_t * q,mblk_t * mpctl,int level,ip_stack_t * ipst)10605 ip_snmp_get_mib2_ip_route_media(queue_t *q, mblk_t *mpctl, int level,
10606     ip_stack_t *ipst)
10607 {
10608 	struct opthdr	*optp;
10609 	mblk_t		*mp2ctl;	/* Returned */
10610 	mblk_t		*mp3ctl;	/* nettomedia */
10611 	mblk_t		*mp4ctl;	/* routeattrs */
10612 	iproutedata_t	ird;
10613 	zoneid_t	zoneid;
10614 
10615 	/*
10616 	 * make copies of the original message
10617 	 *	- mp2ctl is returned unchanged to the caller for its use
10618 	 *	- mpctl is sent upstream as ipRouteEntryTable
10619 	 *	- mp3ctl is sent upstream as ipNetToMediaEntryTable
10620 	 *	- mp4ctl is sent upstream as ipRouteAttributeTable
10621 	 */
10622 	mp2ctl = copymsg(mpctl);
10623 	mp3ctl = copymsg(mpctl);
10624 	mp4ctl = copymsg(mpctl);
10625 	if (mp3ctl == NULL || mp4ctl == NULL) {
10626 		freemsg(mp4ctl);
10627 		freemsg(mp3ctl);
10628 		freemsg(mp2ctl);
10629 		freemsg(mpctl);
10630 		return (NULL);
10631 	}
10632 
10633 	bzero(&ird, sizeof (ird));
10634 
10635 	ird.ird_route.lp_head = mpctl->b_cont;
10636 	ird.ird_netmedia.lp_head = mp3ctl->b_cont;
10637 	ird.ird_attrs.lp_head = mp4ctl->b_cont;
10638 	/*
10639 	 * If the level has been set the special EXPER_IP_AND_ALL_IRES value,
10640 	 * then also include ire_testhidden IREs and IRE_IF_CLONE.  This is
10641 	 * intended a temporary solution until a proper MIB API is provided
10642 	 * that provides complete filtering/caller-opt-in.
10643 	 */
10644 	if (level == EXPER_IP_AND_ALL_IRES)
10645 		ird.ird_flags |= IRD_REPORT_ALL;
10646 
10647 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10648 	ire_walk_v4(ip_snmp_get2_v4, &ird, zoneid, ipst);
10649 
10650 	/* ipRouteEntryTable in mpctl */
10651 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10652 	optp->level = MIB2_IP;
10653 	optp->name = MIB2_IP_ROUTE;
10654 	optp->len = msgdsize(ird.ird_route.lp_head);
10655 	ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10656 	    (int)optp->level, (int)optp->name, (int)optp->len));
10657 	qreply(q, mpctl);
10658 
10659 	/* ipNetToMediaEntryTable in mp3ctl */
10660 	ncec_walk(NULL, ip_snmp_get2_v4_media, &ird, ipst);
10661 
10662 	optp = (struct opthdr *)&mp3ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10663 	optp->level = MIB2_IP;
10664 	optp->name = MIB2_IP_MEDIA;
10665 	optp->len = msgdsize(ird.ird_netmedia.lp_head);
10666 	ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10667 	    (int)optp->level, (int)optp->name, (int)optp->len));
10668 	qreply(q, mp3ctl);
10669 
10670 	/* ipRouteAttributeTable in mp4ctl */
10671 	optp = (struct opthdr *)&mp4ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10672 	optp->level = MIB2_IP;
10673 	optp->name = EXPER_IP_RTATTR;
10674 	optp->len = msgdsize(ird.ird_attrs.lp_head);
10675 	ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10676 	    (int)optp->level, (int)optp->name, (int)optp->len));
10677 	if (optp->len == 0)
10678 		freemsg(mp4ctl);
10679 	else
10680 		qreply(q, mp4ctl);
10681 
10682 	return (mp2ctl);
10683 }
10684 
10685 /*
10686  * Return ipv6RouteEntryTable and ipv6RouteAttributeTable in one IRE walk, and
10687  * ipv6NetToMediaEntryTable in an NDP walk.
10688  */
10689 static mblk_t *
ip_snmp_get_mib2_ip6_route_media(queue_t * q,mblk_t * mpctl,int level,ip_stack_t * ipst)10690 ip_snmp_get_mib2_ip6_route_media(queue_t *q, mblk_t *mpctl, int level,
10691     ip_stack_t *ipst)
10692 {
10693 	struct opthdr	*optp;
10694 	mblk_t		*mp2ctl;	/* Returned */
10695 	mblk_t		*mp3ctl;	/* nettomedia */
10696 	mblk_t		*mp4ctl;	/* routeattrs */
10697 	iproutedata_t	ird;
10698 	zoneid_t	zoneid;
10699 
10700 	/*
10701 	 * make copies of the original message
10702 	 *	- mp2ctl is returned unchanged to the caller for its use
10703 	 *	- mpctl is sent upstream as ipv6RouteEntryTable
10704 	 *	- mp3ctl is sent upstream as ipv6NetToMediaEntryTable
10705 	 *	- mp4ctl is sent upstream as ipv6RouteAttributeTable
10706 	 */
10707 	mp2ctl = copymsg(mpctl);
10708 	mp3ctl = copymsg(mpctl);
10709 	mp4ctl = copymsg(mpctl);
10710 	if (mp3ctl == NULL || mp4ctl == NULL) {
10711 		freemsg(mp4ctl);
10712 		freemsg(mp3ctl);
10713 		freemsg(mp2ctl);
10714 		freemsg(mpctl);
10715 		return (NULL);
10716 	}
10717 
10718 	bzero(&ird, sizeof (ird));
10719 
10720 	ird.ird_route.lp_head = mpctl->b_cont;
10721 	ird.ird_netmedia.lp_head = mp3ctl->b_cont;
10722 	ird.ird_attrs.lp_head = mp4ctl->b_cont;
10723 	/*
10724 	 * If the level has been set the special EXPER_IP_AND_ALL_IRES value,
10725 	 * then also include ire_testhidden IREs and IRE_IF_CLONE.  This is
10726 	 * intended a temporary solution until a proper MIB API is provided
10727 	 * that provides complete filtering/caller-opt-in.
10728 	 */
10729 	if (level == EXPER_IP_AND_ALL_IRES)
10730 		ird.ird_flags |= IRD_REPORT_ALL;
10731 
10732 	zoneid = Q_TO_CONN(q)->conn_zoneid;
10733 	ire_walk_v6(ip_snmp_get2_v6_route, &ird, zoneid, ipst);
10734 
10735 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10736 	optp->level = MIB2_IP6;
10737 	optp->name = MIB2_IP6_ROUTE;
10738 	optp->len = msgdsize(ird.ird_route.lp_head);
10739 	ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10740 	    (int)optp->level, (int)optp->name, (int)optp->len));
10741 	qreply(q, mpctl);
10742 
10743 	/* ipv6NetToMediaEntryTable in mp3ctl */
10744 	ncec_walk(NULL, ip_snmp_get2_v6_media, &ird, ipst);
10745 
10746 	optp = (struct opthdr *)&mp3ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10747 	optp->level = MIB2_IP6;
10748 	optp->name = MIB2_IP6_MEDIA;
10749 	optp->len = msgdsize(ird.ird_netmedia.lp_head);
10750 	ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10751 	    (int)optp->level, (int)optp->name, (int)optp->len));
10752 	qreply(q, mp3ctl);
10753 
10754 	/* ipv6RouteAttributeTable in mp4ctl */
10755 	optp = (struct opthdr *)&mp4ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10756 	optp->level = MIB2_IP6;
10757 	optp->name = EXPER_IP_RTATTR;
10758 	optp->len = msgdsize(ird.ird_attrs.lp_head);
10759 	ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10760 	    (int)optp->level, (int)optp->name, (int)optp->len));
10761 	if (optp->len == 0)
10762 		freemsg(mp4ctl);
10763 	else
10764 		qreply(q, mp4ctl);
10765 
10766 	return (mp2ctl);
10767 }
10768 
10769 /*
10770  * IPv6 mib: One per ill
10771  */
10772 static mblk_t *
ip_snmp_get_mib2_ip6(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst,boolean_t legacy_req)10773 ip_snmp_get_mib2_ip6(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
10774     boolean_t legacy_req)
10775 {
10776 	struct opthdr		*optp;
10777 	mblk_t			*mp2ctl;
10778 	ill_t			*ill;
10779 	ill_walk_context_t	ctx;
10780 	mblk_t			*mp_tail = NULL;
10781 	mib2_ipv6AddrEntry_t	mae6;
10782 	mib2_ipIfStatsEntry_t	*ise;
10783 	size_t			ise_size, iae_size;
10784 
10785 	/*
10786 	 * Make a copy of the original message
10787 	 */
10788 	mp2ctl = copymsg(mpctl);
10789 
10790 	/* fixed length IPv6 structure ... */
10791 
10792 	if (legacy_req) {
10793 		ise_size = LEGACY_MIB_SIZE(&ipst->ips_ip6_mib,
10794 		    mib2_ipIfStatsEntry_t);
10795 		iae_size = LEGACY_MIB_SIZE(&mae6, mib2_ipv6AddrEntry_t);
10796 	} else {
10797 		ise_size = sizeof (mib2_ipIfStatsEntry_t);
10798 		iae_size = sizeof (mib2_ipv6AddrEntry_t);
10799 	}
10800 
10801 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10802 	optp->level = MIB2_IP6;
10803 	optp->name = 0;
10804 	/* Include "unknown interface" ip6_mib */
10805 	ipst->ips_ip6_mib.ipIfStatsIPVersion = MIB2_INETADDRESSTYPE_ipv6;
10806 	ipst->ips_ip6_mib.ipIfStatsIfIndex =
10807 	    MIB2_UNKNOWN_INTERFACE; /* Flag to netstat */
10808 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsForwarding,
10809 	    ipst->ips_ipv6_forwarding ? 1 : 2);
10810 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsDefaultHopLimit,
10811 	    ipst->ips_ipv6_def_hops);
10812 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsEntrySize,
10813 	    sizeof (mib2_ipIfStatsEntry_t));
10814 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsAddrEntrySize,
10815 	    sizeof (mib2_ipv6AddrEntry_t));
10816 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsRouteEntrySize,
10817 	    sizeof (mib2_ipv6RouteEntry_t));
10818 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsNetToMediaEntrySize,
10819 	    sizeof (mib2_ipv6NetToMediaEntry_t));
10820 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsMemberEntrySize,
10821 	    sizeof (ipv6_member_t));
10822 	SET_MIB(ipst->ips_ip6_mib.ipIfStatsGroupSourceEntrySize,
10823 	    sizeof (ipv6_grpsrc_t));
10824 
10825 	/*
10826 	 * Synchronize 64- and 32-bit counters
10827 	 */
10828 	SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInReceives,
10829 	    ipIfStatsHCInReceives);
10830 	SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInDelivers,
10831 	    ipIfStatsHCInDelivers);
10832 	SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutRequests,
10833 	    ipIfStatsHCOutRequests);
10834 	SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutForwDatagrams,
10835 	    ipIfStatsHCOutForwDatagrams);
10836 	SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutMcastPkts,
10837 	    ipIfStatsHCOutMcastPkts);
10838 	SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInMcastPkts,
10839 	    ipIfStatsHCInMcastPkts);
10840 
10841 	if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10842 	    (char *)&ipst->ips_ip6_mib, (int)ise_size)) {
10843 		ip1dbg(("ip_snmp_get_mib2_ip6: failed to allocate %u bytes\n",
10844 		    (uint_t)ise_size));
10845 	} else if (legacy_req) {
10846 		/* Adjust the EntrySize fields for legacy requests. */
10847 		ise =
10848 		    (mib2_ipIfStatsEntry_t *)(mp_tail->b_wptr - (int)ise_size);
10849 		SET_MIB(ise->ipIfStatsEntrySize, ise_size);
10850 		SET_MIB(ise->ipIfStatsAddrEntrySize, iae_size);
10851 	}
10852 
10853 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10854 	ill = ILL_START_WALK_V6(&ctx, ipst);
10855 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10856 		ill->ill_ip_mib->ipIfStatsIfIndex =
10857 		    ill->ill_phyint->phyint_ifindex;
10858 		SET_MIB(ill->ill_ip_mib->ipIfStatsForwarding,
10859 		    ipst->ips_ipv6_forwarding ? 1 : 2);
10860 		SET_MIB(ill->ill_ip_mib->ipIfStatsDefaultHopLimit,
10861 		    ill->ill_max_hops);
10862 
10863 		/*
10864 		 * Synchronize 64- and 32-bit counters
10865 		 */
10866 		SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInReceives,
10867 		    ipIfStatsHCInReceives);
10868 		SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInDelivers,
10869 		    ipIfStatsHCInDelivers);
10870 		SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutRequests,
10871 		    ipIfStatsHCOutRequests);
10872 		SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutForwDatagrams,
10873 		    ipIfStatsHCOutForwDatagrams);
10874 		SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutMcastPkts,
10875 		    ipIfStatsHCOutMcastPkts);
10876 		SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInMcastPkts,
10877 		    ipIfStatsHCInMcastPkts);
10878 
10879 		if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10880 		    (char *)ill->ill_ip_mib, (int)ise_size)) {
10881 			ip1dbg(("ip_snmp_get_mib2_ip6: failed to allocate "
10882 			"%u bytes\n", (uint_t)ise_size));
10883 		} else if (legacy_req) {
10884 			/* Adjust the EntrySize fields for legacy requests. */
10885 			ise = (mib2_ipIfStatsEntry_t *)(mp_tail->b_wptr -
10886 			    (int)ise_size);
10887 			SET_MIB(ise->ipIfStatsEntrySize, ise_size);
10888 			SET_MIB(ise->ipIfStatsAddrEntrySize, iae_size);
10889 		}
10890 	}
10891 	rw_exit(&ipst->ips_ill_g_lock);
10892 
10893 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10894 	ip3dbg(("ip_snmp_get_mib2_ip6: level %d, name %d, len %d\n",
10895 	    (int)optp->level, (int)optp->name, (int)optp->len));
10896 	qreply(q, mpctl);
10897 	return (mp2ctl);
10898 }
10899 
10900 /*
10901  * ICMPv6 mib: One per ill
10902  */
10903 static mblk_t *
ip_snmp_get_mib2_icmp6(queue_t * q,mblk_t * mpctl,ip_stack_t * ipst)10904 ip_snmp_get_mib2_icmp6(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10905 {
10906 	struct opthdr		*optp;
10907 	mblk_t			*mp2ctl;
10908 	ill_t			*ill;
10909 	ill_walk_context_t	ctx;
10910 	mblk_t			*mp_tail = NULL;
10911 	/*
10912 	 * Make a copy of the original message
10913 	 */
10914 	mp2ctl = copymsg(mpctl);
10915 
10916 	/* fixed length ICMPv6 structure ... */
10917 
10918 	optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10919 	optp->level = MIB2_ICMP6;
10920 	optp->name = 0;
10921 	/* Include "unknown interface" icmp6_mib */
10922 	ipst->ips_icmp6_mib.ipv6IfIcmpIfIndex =
10923 	    MIB2_UNKNOWN_INTERFACE; /* netstat flag */
10924 	ipst->ips_icmp6_mib.ipv6IfIcmpEntrySize =
10925 	    sizeof (mib2_ipv6IfIcmpEntry_t);
10926 	if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10927 	    (char *)&ipst->ips_icmp6_mib,
10928 	    (int)sizeof (ipst->ips_icmp6_mib))) {
10929 		ip1dbg(("ip_snmp_get_mib2_icmp6: failed to allocate %u bytes\n",
10930 		    (uint_t)sizeof (ipst->ips_icmp6_mib)));
10931 	}
10932 
10933 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10934 	ill = ILL_START_WALK_V6(&ctx, ipst);
10935 	for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10936 		ill->ill_icmp6_mib->ipv6IfIcmpIfIndex =
10937 		    ill->ill_phyint->phyint_ifindex;
10938 		if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10939 		    (char *)ill->ill_icmp6_mib,
10940 		    (int)sizeof (*ill->ill_icmp6_mib))) {
10941 			ip1dbg(("ip_snmp_get_mib2_icmp6: failed to allocate "
10942 			    "%u bytes\n",
10943 			    (uint_t)sizeof (*ill->ill_icmp6_mib)));
10944 		}
10945 	}
10946 	rw_exit(&ipst->ips_ill_g_lock);
10947 
10948 	optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10949 	ip3dbg(("ip_snmp_get_mib2_icmp6: level %d, name %d, len %d\n",
10950 	    (int)optp->level, (int)optp->name, (int)optp->len));
10951 	qreply(q, mpctl);
10952 	return (mp2ctl);
10953 }
10954 
10955 /*
10956  * ire_walk routine to create both ipRouteEntryTable and
10957  * ipRouteAttributeTable in one IRE walk
10958  */
10959 static void
ip_snmp_get2_v4(ire_t * ire,iproutedata_t * ird)10960 ip_snmp_get2_v4(ire_t *ire, iproutedata_t *ird)
10961 {
10962 	ill_t				*ill;
10963 	mib2_ipRouteEntry_t		*re;
10964 	mib2_ipAttributeEntry_t		iaes;
10965 	tsol_ire_gw_secattr_t		*attrp;
10966 	tsol_gc_t			*gc = NULL;
10967 	tsol_gcgrp_t			*gcgrp = NULL;
10968 	ip_stack_t			*ipst = ire->ire_ipst;
10969 
10970 	ASSERT(ire->ire_ipversion == IPV4_VERSION);
10971 
10972 	if (!(ird->ird_flags & IRD_REPORT_ALL)) {
10973 		if (ire->ire_testhidden)
10974 			return;
10975 		if (ire->ire_type & IRE_IF_CLONE)
10976 			return;
10977 	}
10978 
10979 	if ((re = kmem_zalloc(sizeof (*re), KM_NOSLEEP)) == NULL)
10980 		return;
10981 
10982 	if ((attrp = ire->ire_gw_secattr) != NULL) {
10983 		mutex_enter(&attrp->igsa_lock);
10984 		if ((gc = attrp->igsa_gc) != NULL) {
10985 			gcgrp = gc->gc_grp;
10986 			ASSERT(gcgrp != NULL);
10987 			rw_enter(&gcgrp->gcgrp_rwlock, RW_READER);
10988 		}
10989 		mutex_exit(&attrp->igsa_lock);
10990 	}
10991 	/*
10992 	 * Return all IRE types for route table... let caller pick and choose
10993 	 */
10994 	re->ipRouteDest = ire->ire_addr;
10995 	ill = ire->ire_ill;
10996 	re->ipRouteIfIndex.o_length = 0;
10997 	if (ill != NULL) {
10998 		ill_get_name(ill, re->ipRouteIfIndex.o_bytes, OCTET_LENGTH);
10999 		re->ipRouteIfIndex.o_length =
11000 		    mi_strlen(re->ipRouteIfIndex.o_bytes);
11001 	}
11002 	re->ipRouteMetric1 = -1;
11003 	re->ipRouteMetric2 = -1;
11004 	re->ipRouteMetric3 = -1;
11005 	re->ipRouteMetric4 = -1;
11006 
11007 	re->ipRouteNextHop = ire->ire_gateway_addr;
11008 	/* indirect(4), direct(3), or invalid(2) */
11009 	if (ire->ire_flags & (RTF_REJECT | RTF_BLACKHOLE))
11010 		re->ipRouteType = 2;
11011 	else if (ire->ire_type & IRE_ONLINK)
11012 		re->ipRouteType = 3;
11013 	else
11014 		re->ipRouteType = 4;
11015 
11016 	re->ipRouteProto = -1;
11017 	re->ipRouteAge = gethrestime_sec() - ire->ire_create_time;
11018 	re->ipRouteMask = ire->ire_mask;
11019 	re->ipRouteMetric5 = -1;
11020 	re->ipRouteInfo.re_max_frag = ire->ire_metrics.iulp_mtu;
11021 	if (ire->ire_ill != NULL && re->ipRouteInfo.re_max_frag == 0)
11022 		re->ipRouteInfo.re_max_frag = ire->ire_ill->ill_mtu;
11023 
11024 	re->ipRouteInfo.re_frag_flag	= 0;
11025 	re->ipRouteInfo.re_rtt		= 0;
11026 	re->ipRouteInfo.re_src_addr	= 0;
11027 	re->ipRouteInfo.re_ref		= ire->ire_refcnt;
11028 	re->ipRouteInfo.re_obpkt	= ire->ire_ob_pkt_count;
11029 	re->ipRouteInfo.re_ibpkt	= ire->ire_ib_pkt_count;
11030 	re->ipRouteInfo.re_flags	= ire->ire_flags;
11031 
11032 	/* Add the IRE_IF_CLONE's counters to their parent IRE_INTERFACE */
11033 	if (ire->ire_type & IRE_INTERFACE) {
11034 		ire_t *child;
11035 
11036 		rw_enter(&ipst->ips_ire_dep_lock, RW_READER);
11037 		child = ire->ire_dep_children;
11038 		while (child != NULL) {
11039 			re->ipRouteInfo.re_obpkt += child->ire_ob_pkt_count;
11040 			re->ipRouteInfo.re_ibpkt += child->ire_ib_pkt_count;
11041 			child = child->ire_dep_sib_next;
11042 		}
11043 		rw_exit(&ipst->ips_ire_dep_lock);
11044 	}
11045 
11046 	if (ire->ire_flags & RTF_DYNAMIC) {
11047 		re->ipRouteInfo.re_ire_type	= IRE_HOST_REDIRECT;
11048 	} else {
11049 		re->ipRouteInfo.re_ire_type	= ire->ire_type;
11050 	}
11051 
11052 	if (!snmp_append_data2(ird->ird_route.lp_head, &ird->ird_route.lp_tail,
11053 	    (char *)re, (int)sizeof (*re))) {
11054 		ip1dbg(("ip_snmp_get2_v4: failed to allocate %u bytes\n",
11055 		    (uint_t)sizeof (*re)));
11056 	}
11057 
11058 	if (gc != NULL) {
11059 		iaes.iae_routeidx = ird->ird_idx;
11060 		iaes.iae_doi = gc->gc_db->gcdb_doi;
11061 		iaes.iae_slrange = gc->gc_db->gcdb_slrange;
11062 
11063 		if (!snmp_append_data2(ird->ird_attrs.lp_head,
11064 		    &ird->ird_attrs.lp_tail, (char *)&iaes, sizeof (iaes))) {
11065 			ip1dbg(("ip_snmp_get2_v4: failed to allocate %u "
11066 			    "bytes\n", (uint_t)sizeof (iaes)));
11067 		}
11068 	}
11069 
11070 	/* bump route index for next pass */
11071 	ird->ird_idx++;
11072 
11073 	kmem_free(re, sizeof (*re));
11074 	if (gcgrp != NULL)
11075 		rw_exit(&gcgrp->gcgrp_rwlock);
11076 }
11077 
11078 /*
11079  * ire_walk routine to create ipv6RouteEntryTable and ipRouteEntryTable.
11080  */
11081 static void
ip_snmp_get2_v6_route(ire_t * ire,iproutedata_t * ird)11082 ip_snmp_get2_v6_route(ire_t *ire, iproutedata_t *ird)
11083 {
11084 	ill_t				*ill;
11085 	mib2_ipv6RouteEntry_t		*re;
11086 	mib2_ipAttributeEntry_t		iaes;
11087 	tsol_ire_gw_secattr_t		*attrp;
11088 	tsol_gc_t			*gc = NULL;
11089 	tsol_gcgrp_t			*gcgrp = NULL;
11090 	ip_stack_t			*ipst = ire->ire_ipst;
11091 
11092 	ASSERT(ire->ire_ipversion == IPV6_VERSION);
11093 
11094 	if (!(ird->ird_flags & IRD_REPORT_ALL)) {
11095 		if (ire->ire_testhidden)
11096 			return;
11097 		if (ire->ire_type & IRE_IF_CLONE)
11098 			return;
11099 	}
11100 
11101 	if ((re = kmem_zalloc(sizeof (*re), KM_NOSLEEP)) == NULL)
11102 		return;
11103 
11104 	if ((attrp = ire->ire_gw_secattr) != NULL) {
11105 		mutex_enter(&attrp->igsa_lock);
11106 		if ((gc = attrp->igsa_gc) != NULL) {
11107 			gcgrp = gc->gc_grp;
11108 			ASSERT(gcgrp != NULL);
11109 			rw_enter(&gcgrp->gcgrp_rwlock, RW_READER);
11110 		}
11111 		mutex_exit(&attrp->igsa_lock);
11112 	}
11113 	/*
11114 	 * Return all IRE types for route table... let caller pick and choose
11115 	 */
11116 	re->ipv6RouteDest = ire->ire_addr_v6;
11117 	re->ipv6RoutePfxLength = ip_mask_to_plen_v6(&ire->ire_mask_v6);
11118 	re->ipv6RouteIndex = 0;	/* Unique when multiple with same dest/plen */
11119 	re->ipv6RouteIfIndex.o_length = 0;
11120 	ill = ire->ire_ill;
11121 	if (ill != NULL) {
11122 		ill_get_name(ill, re->ipv6RouteIfIndex.o_bytes, OCTET_LENGTH);
11123 		re->ipv6RouteIfIndex.o_length =
11124 		    mi_strlen(re->ipv6RouteIfIndex.o_bytes);
11125 	}
11126 
11127 	ASSERT(!(ire->ire_type & IRE_BROADCAST));
11128 
11129 	mutex_enter(&ire->ire_lock);
11130 	re->ipv6RouteNextHop = ire->ire_gateway_addr_v6;
11131 	mutex_exit(&ire->ire_lock);
11132 
11133 	/* remote(4), local(3), or discard(2) */
11134 	if (ire->ire_flags & (RTF_REJECT | RTF_BLACKHOLE))
11135 		re->ipv6RouteType = 2;
11136 	else if (ire->ire_type & IRE_ONLINK)
11137 		re->ipv6RouteType = 3;
11138 	else
11139 		re->ipv6RouteType = 4;
11140 
11141 	re->ipv6RouteProtocol	= -1;
11142 	re->ipv6RoutePolicy	= 0;
11143 	re->ipv6RouteAge	= gethrestime_sec() - ire->ire_create_time;
11144 	re->ipv6RouteNextHopRDI	= 0;
11145 	re->ipv6RouteWeight	= 0;
11146 	re->ipv6RouteMetric	= 0;
11147 	re->ipv6RouteInfo.re_max_frag = ire->ire_metrics.iulp_mtu;
11148 	if (ire->ire_ill != NULL && re->ipv6RouteInfo.re_max_frag == 0)
11149 		re->ipv6RouteInfo.re_max_frag = ire->ire_ill->ill_mtu;
11150 
11151 	re->ipv6RouteInfo.re_frag_flag	= 0;
11152 	re->ipv6RouteInfo.re_rtt	= 0;
11153 	re->ipv6RouteInfo.re_src_addr	= ipv6_all_zeros;
11154 	re->ipv6RouteInfo.re_obpkt	= ire->ire_ob_pkt_count;
11155 	re->ipv6RouteInfo.re_ibpkt	= ire->ire_ib_pkt_count;
11156 	re->ipv6RouteInfo.re_ref	= ire->ire_refcnt;
11157 	re->ipv6RouteInfo.re_flags	= ire->ire_flags;
11158 
11159 	/* Add the IRE_IF_CLONE's counters to their parent IRE_INTERFACE */
11160 	if (ire->ire_type & IRE_INTERFACE) {
11161 		ire_t *child;
11162 
11163 		rw_enter(&ipst->ips_ire_dep_lock, RW_READER);
11164 		child = ire->ire_dep_children;
11165 		while (child != NULL) {
11166 			re->ipv6RouteInfo.re_obpkt += child->ire_ob_pkt_count;
11167 			re->ipv6RouteInfo.re_ibpkt += child->ire_ib_pkt_count;
11168 			child = child->ire_dep_sib_next;
11169 		}
11170 		rw_exit(&ipst->ips_ire_dep_lock);
11171 	}
11172 	if (ire->ire_flags & RTF_DYNAMIC) {
11173 		re->ipv6RouteInfo.re_ire_type	= IRE_HOST_REDIRECT;
11174 	} else {
11175 		re->ipv6RouteInfo.re_ire_type	= ire->ire_type;
11176 	}
11177 
11178 	if (!snmp_append_data2(ird->ird_route.lp_head, &ird->ird_route.lp_tail,
11179 	    (char *)re, (int)sizeof (*re))) {
11180 		ip1dbg(("ip_snmp_get2_v6: failed to allocate %u bytes\n",
11181 		    (uint_t)sizeof (*re)));
11182 	}
11183 
11184 	if (gc != NULL) {
11185 		iaes.iae_routeidx = ird->ird_idx;
11186 		iaes.iae_doi = gc->gc_db->gcdb_doi;
11187 		iaes.iae_slrange = gc->gc_db->gcdb_slrange;
11188 
11189 		if (!snmp_append_data2(ird->ird_attrs.lp_head,
11190 		    &ird->ird_attrs.lp_tail, (char *)&iaes, sizeof (iaes))) {
11191 			ip1dbg(("ip_snmp_get2_v6: failed to allocate %u "
11192 			    "bytes\n", (uint_t)sizeof (iaes)));
11193 		}
11194 	}
11195 
11196 	/* bump route index for next pass */
11197 	ird->ird_idx++;
11198 
11199 	kmem_free(re, sizeof (*re));
11200 	if (gcgrp != NULL)
11201 		rw_exit(&gcgrp->gcgrp_rwlock);
11202 }
11203 
11204 /*
11205  * ncec_walk routine to create ipv6NetToMediaEntryTable
11206  */
11207 static void
ip_snmp_get2_v6_media(ncec_t * ncec,void * ptr)11208 ip_snmp_get2_v6_media(ncec_t *ncec, void *ptr)
11209 {
11210 	iproutedata_t *ird		= ptr;
11211 	ill_t				*ill;
11212 	mib2_ipv6NetToMediaEntry_t	ntme;
11213 
11214 	ill = ncec->ncec_ill;
11215 	/* skip arpce entries, and loopback ncec entries */
11216 	if (ill->ill_isv6 == B_FALSE || ill->ill_net_type == IRE_LOOPBACK)
11217 		return;
11218 	/*
11219 	 * Neighbor cache entry attached to IRE with on-link
11220 	 * destination.
11221 	 * We report all IPMP groups on ncec_ill which is normally the upper.
11222 	 */
11223 	ntme.ipv6NetToMediaIfIndex = ill->ill_phyint->phyint_ifindex;
11224 	ntme.ipv6NetToMediaNetAddress = ncec->ncec_addr;
11225 	ntme.ipv6NetToMediaPhysAddress.o_length = ill->ill_phys_addr_length;
11226 	if (ncec->ncec_lladdr != NULL) {
11227 		bcopy(ncec->ncec_lladdr, ntme.ipv6NetToMediaPhysAddress.o_bytes,
11228 		    ntme.ipv6NetToMediaPhysAddress.o_length);
11229 	}
11230 	/*
11231 	 * Note: Returns ND_* states. Should be:
11232 	 * reachable(1), stale(2), delay(3), probe(4),
11233 	 * invalid(5), unknown(6)
11234 	 */
11235 	ntme.ipv6NetToMediaState = ncec->ncec_state;
11236 	ntme.ipv6NetToMediaLastUpdated = 0;
11237 
11238 	/* other(1), dynamic(2), static(3), local(4) */
11239 	if (NCE_MYADDR(ncec)) {
11240 		ntme.ipv6NetToMediaType = 4;
11241 	} else if (ncec->ncec_flags & NCE_F_PUBLISH) {
11242 		ntme.ipv6NetToMediaType = 1; /* proxy */
11243 	} else if (ncec->ncec_flags & NCE_F_STATIC) {
11244 		ntme.ipv6NetToMediaType = 3;
11245 	} else if (ncec->ncec_flags & (NCE_F_MCAST|NCE_F_BCAST)) {
11246 		ntme.ipv6NetToMediaType = 1;
11247 	} else {
11248 		ntme.ipv6NetToMediaType = 2;
11249 	}
11250 
11251 	if (!snmp_append_data2(ird->ird_netmedia.lp_head,
11252 	    &ird->ird_netmedia.lp_tail, (char *)&ntme, sizeof (ntme))) {
11253 		ip1dbg(("ip_snmp_get2_v6_media: failed to allocate %u bytes\n",
11254 		    (uint_t)sizeof (ntme)));
11255 	}
11256 }
11257 
11258 int
nce2ace(ncec_t * ncec)11259 nce2ace(ncec_t *ncec)
11260 {
11261 	int flags = 0;
11262 
11263 	if (NCE_ISREACHABLE(ncec))
11264 		flags |= ACE_F_RESOLVED;
11265 	if (ncec->ncec_flags & NCE_F_AUTHORITY)
11266 		flags |= ACE_F_AUTHORITY;
11267 	if (ncec->ncec_flags & NCE_F_PUBLISH)
11268 		flags |= ACE_F_PUBLISH;
11269 	if ((ncec->ncec_flags & NCE_F_NONUD) != 0)
11270 		flags |= ACE_F_PERMANENT;
11271 	if (NCE_MYADDR(ncec))
11272 		flags |= (ACE_F_MYADDR | ACE_F_AUTHORITY);
11273 	if (ncec->ncec_flags & NCE_F_UNVERIFIED)
11274 		flags |= ACE_F_UNVERIFIED;
11275 	if (ncec->ncec_flags & NCE_F_AUTHORITY)
11276 		flags |= ACE_F_AUTHORITY;
11277 	if (ncec->ncec_flags & NCE_F_DELAYED)
11278 		flags |= ACE_F_DELAYED;
11279 	return (flags);
11280 }
11281 
11282 /*
11283  * ncec_walk routine to create ipNetToMediaEntryTable
11284  */
11285 static void
ip_snmp_get2_v4_media(ncec_t * ncec,void * ptr)11286 ip_snmp_get2_v4_media(ncec_t *ncec, void *ptr)
11287 {
11288 	iproutedata_t *ird		= ptr;
11289 	ill_t				*ill;
11290 	mib2_ipNetToMediaEntry_t	ntme;
11291 	const char			*name = "unknown";
11292 	ipaddr_t			ncec_addr;
11293 
11294 	ill = ncec->ncec_ill;
11295 	if (ill->ill_isv6 || (ncec->ncec_flags & NCE_F_BCAST) ||
11296 	    ill->ill_net_type == IRE_LOOPBACK)
11297 		return;
11298 
11299 	/* We report all IPMP groups on ncec_ill which is normally the upper. */
11300 	name = ill->ill_name;
11301 	/* Based on RFC 4293: other(1), inval(2), dyn(3), stat(4) */
11302 	if (NCE_MYADDR(ncec)) {
11303 		ntme.ipNetToMediaType = 4;
11304 	} else if (ncec->ncec_flags & (NCE_F_MCAST|NCE_F_BCAST|NCE_F_PUBLISH)) {
11305 		ntme.ipNetToMediaType = 1;
11306 	} else {
11307 		ntme.ipNetToMediaType = 3;
11308 	}
11309 	ntme.ipNetToMediaIfIndex.o_length = MIN(OCTET_LENGTH, strlen(name));
11310 	bcopy(name, ntme.ipNetToMediaIfIndex.o_bytes,
11311 	    ntme.ipNetToMediaIfIndex.o_length);
11312 
11313 	IN6_V4MAPPED_TO_IPADDR(&ncec->ncec_addr, ncec_addr);
11314 	bcopy(&ncec_addr, &ntme.ipNetToMediaNetAddress, sizeof (ncec_addr));
11315 
11316 	ntme.ipNetToMediaInfo.ntm_mask.o_length = sizeof (ipaddr_t);
11317 	ncec_addr = INADDR_BROADCAST;
11318 	bcopy(&ncec_addr, ntme.ipNetToMediaInfo.ntm_mask.o_bytes,
11319 	    sizeof (ncec_addr));
11320 	/*
11321 	 * map all the flags to the ACE counterpart.
11322 	 */
11323 	ntme.ipNetToMediaInfo.ntm_flags = nce2ace(ncec);
11324 
11325 	ntme.ipNetToMediaPhysAddress.o_length =
11326 	    MIN(OCTET_LENGTH, ill->ill_phys_addr_length);
11327 
11328 	if (!NCE_ISREACHABLE(ncec))
11329 		ntme.ipNetToMediaPhysAddress.o_length = 0;
11330 	else {
11331 		if (ncec->ncec_lladdr != NULL) {
11332 			bcopy(ncec->ncec_lladdr,
11333 			    ntme.ipNetToMediaPhysAddress.o_bytes,
11334 			    ntme.ipNetToMediaPhysAddress.o_length);
11335 		}
11336 	}
11337 
11338 	if (!snmp_append_data2(ird->ird_netmedia.lp_head,
11339 	    &ird->ird_netmedia.lp_tail, (char *)&ntme, sizeof (ntme))) {
11340 		ip1dbg(("ip_snmp_get2_v4_media: failed to allocate %u bytes\n",
11341 		    (uint_t)sizeof (ntme)));
11342 	}
11343 }
11344 
11345 /*
11346  * return (0) if invalid set request, 1 otherwise, including non-tcp requests
11347  */
11348 /* ARGSUSED */
11349 int
ip_snmp_set(queue_t * q,int level,int name,uchar_t * ptr,int len)11350 ip_snmp_set(queue_t *q, int level, int name, uchar_t *ptr, int len)
11351 {
11352 	switch (level) {
11353 	case MIB2_IP:
11354 	case MIB2_ICMP:
11355 		switch (name) {
11356 		default:
11357 			break;
11358 		}
11359 		return (1);
11360 	default:
11361 		return (1);
11362 	}
11363 }
11364 
11365 /*
11366  * When there exists both a 64- and 32-bit counter of a particular type
11367  * (i.e., InReceives), only the 64-bit counters are added.
11368  */
11369 void
ip_mib2_add_ip_stats(mib2_ipIfStatsEntry_t * o1,mib2_ipIfStatsEntry_t * o2)11370 ip_mib2_add_ip_stats(mib2_ipIfStatsEntry_t *o1, mib2_ipIfStatsEntry_t *o2)
11371 {
11372 	UPDATE_MIB(o1, ipIfStatsInHdrErrors, o2->ipIfStatsInHdrErrors);
11373 	UPDATE_MIB(o1, ipIfStatsInTooBigErrors, o2->ipIfStatsInTooBigErrors);
11374 	UPDATE_MIB(o1, ipIfStatsInNoRoutes, o2->ipIfStatsInNoRoutes);
11375 	UPDATE_MIB(o1, ipIfStatsInAddrErrors, o2->ipIfStatsInAddrErrors);
11376 	UPDATE_MIB(o1, ipIfStatsInUnknownProtos, o2->ipIfStatsInUnknownProtos);
11377 	UPDATE_MIB(o1, ipIfStatsInTruncatedPkts, o2->ipIfStatsInTruncatedPkts);
11378 	UPDATE_MIB(o1, ipIfStatsInDiscards, o2->ipIfStatsInDiscards);
11379 	UPDATE_MIB(o1, ipIfStatsOutDiscards, o2->ipIfStatsOutDiscards);
11380 	UPDATE_MIB(o1, ipIfStatsOutFragOKs, o2->ipIfStatsOutFragOKs);
11381 	UPDATE_MIB(o1, ipIfStatsOutFragFails, o2->ipIfStatsOutFragFails);
11382 	UPDATE_MIB(o1, ipIfStatsOutFragCreates, o2->ipIfStatsOutFragCreates);
11383 	UPDATE_MIB(o1, ipIfStatsReasmReqds, o2->ipIfStatsReasmReqds);
11384 	UPDATE_MIB(o1, ipIfStatsReasmOKs, o2->ipIfStatsReasmOKs);
11385 	UPDATE_MIB(o1, ipIfStatsReasmFails, o2->ipIfStatsReasmFails);
11386 	UPDATE_MIB(o1, ipIfStatsOutNoRoutes, o2->ipIfStatsOutNoRoutes);
11387 	UPDATE_MIB(o1, ipIfStatsReasmDuplicates, o2->ipIfStatsReasmDuplicates);
11388 	UPDATE_MIB(o1, ipIfStatsReasmPartDups, o2->ipIfStatsReasmPartDups);
11389 	UPDATE_MIB(o1, ipIfStatsForwProhibits, o2->ipIfStatsForwProhibits);
11390 	UPDATE_MIB(o1, udpInCksumErrs, o2->udpInCksumErrs);
11391 	UPDATE_MIB(o1, udpInOverflows, o2->udpInOverflows);
11392 	UPDATE_MIB(o1, rawipInOverflows, o2->rawipInOverflows);
11393 	UPDATE_MIB(o1, ipIfStatsInWrongIPVersion,
11394 	    o2->ipIfStatsInWrongIPVersion);
11395 	UPDATE_MIB(o1, ipIfStatsOutWrongIPVersion,
11396 	    o2->ipIfStatsInWrongIPVersion);
11397 	UPDATE_MIB(o1, ipIfStatsOutSwitchIPVersion,
11398 	    o2->ipIfStatsOutSwitchIPVersion);
11399 	UPDATE_MIB(o1, ipIfStatsHCInReceives, o2->ipIfStatsHCInReceives);
11400 	UPDATE_MIB(o1, ipIfStatsHCInOctets, o2->ipIfStatsHCInOctets);
11401 	UPDATE_MIB(o1, ipIfStatsHCInForwDatagrams,
11402 	    o2->ipIfStatsHCInForwDatagrams);
11403 	UPDATE_MIB(o1, ipIfStatsHCInDelivers, o2->ipIfStatsHCInDelivers);
11404 	UPDATE_MIB(o1, ipIfStatsHCOutRequests, o2->ipIfStatsHCOutRequests);
11405 	UPDATE_MIB(o1, ipIfStatsHCOutForwDatagrams,
11406 	    o2->ipIfStatsHCOutForwDatagrams);
11407 	UPDATE_MIB(o1, ipIfStatsOutFragReqds, o2->ipIfStatsOutFragReqds);
11408 	UPDATE_MIB(o1, ipIfStatsHCOutTransmits, o2->ipIfStatsHCOutTransmits);
11409 	UPDATE_MIB(o1, ipIfStatsHCOutOctets, o2->ipIfStatsHCOutOctets);
11410 	UPDATE_MIB(o1, ipIfStatsHCInMcastPkts, o2->ipIfStatsHCInMcastPkts);
11411 	UPDATE_MIB(o1, ipIfStatsHCInMcastOctets, o2->ipIfStatsHCInMcastOctets);
11412 	UPDATE_MIB(o1, ipIfStatsHCOutMcastPkts, o2->ipIfStatsHCOutMcastPkts);
11413 	UPDATE_MIB(o1, ipIfStatsHCOutMcastOctets,
11414 	    o2->ipIfStatsHCOutMcastOctets);
11415 	UPDATE_MIB(o1, ipIfStatsHCInBcastPkts, o2->ipIfStatsHCInBcastPkts);
11416 	UPDATE_MIB(o1, ipIfStatsHCOutBcastPkts, o2->ipIfStatsHCOutBcastPkts);
11417 	UPDATE_MIB(o1, ipsecInSucceeded, o2->ipsecInSucceeded);
11418 	UPDATE_MIB(o1, ipsecInFailed, o2->ipsecInFailed);
11419 	UPDATE_MIB(o1, ipInCksumErrs, o2->ipInCksumErrs);
11420 	UPDATE_MIB(o1, tcpInErrs, o2->tcpInErrs);
11421 	UPDATE_MIB(o1, udpNoPorts, o2->udpNoPorts);
11422 }
11423 
11424 void
ip_mib2_add_icmp6_stats(mib2_ipv6IfIcmpEntry_t * o1,mib2_ipv6IfIcmpEntry_t * o2)11425 ip_mib2_add_icmp6_stats(mib2_ipv6IfIcmpEntry_t *o1, mib2_ipv6IfIcmpEntry_t *o2)
11426 {
11427 	UPDATE_MIB(o1, ipv6IfIcmpInMsgs, o2->ipv6IfIcmpInMsgs);
11428 	UPDATE_MIB(o1, ipv6IfIcmpInErrors, o2->ipv6IfIcmpInErrors);
11429 	UPDATE_MIB(o1, ipv6IfIcmpInDestUnreachs, o2->ipv6IfIcmpInDestUnreachs);
11430 	UPDATE_MIB(o1, ipv6IfIcmpInAdminProhibs, o2->ipv6IfIcmpInAdminProhibs);
11431 	UPDATE_MIB(o1, ipv6IfIcmpInTimeExcds, o2->ipv6IfIcmpInTimeExcds);
11432 	UPDATE_MIB(o1, ipv6IfIcmpInParmProblems, o2->ipv6IfIcmpInParmProblems);
11433 	UPDATE_MIB(o1, ipv6IfIcmpInPktTooBigs, o2->ipv6IfIcmpInPktTooBigs);
11434 	UPDATE_MIB(o1, ipv6IfIcmpInEchos, o2->ipv6IfIcmpInEchos);
11435 	UPDATE_MIB(o1, ipv6IfIcmpInEchoReplies, o2->ipv6IfIcmpInEchoReplies);
11436 	UPDATE_MIB(o1, ipv6IfIcmpInRouterSolicits,
11437 	    o2->ipv6IfIcmpInRouterSolicits);
11438 	UPDATE_MIB(o1, ipv6IfIcmpInRouterAdvertisements,
11439 	    o2->ipv6IfIcmpInRouterAdvertisements);
11440 	UPDATE_MIB(o1, ipv6IfIcmpInNeighborSolicits,
11441 	    o2->ipv6IfIcmpInNeighborSolicits);
11442 	UPDATE_MIB(o1, ipv6IfIcmpInNeighborAdvertisements,
11443 	    o2->ipv6IfIcmpInNeighborAdvertisements);
11444 	UPDATE_MIB(o1, ipv6IfIcmpInRedirects, o2->ipv6IfIcmpInRedirects);
11445 	UPDATE_MIB(o1, ipv6IfIcmpInGroupMembQueries,
11446 	    o2->ipv6IfIcmpInGroupMembQueries);
11447 	UPDATE_MIB(o1, ipv6IfIcmpInGroupMembResponses,
11448 	    o2->ipv6IfIcmpInGroupMembResponses);
11449 	UPDATE_MIB(o1, ipv6IfIcmpInGroupMembReductions,
11450 	    o2->ipv6IfIcmpInGroupMembReductions);
11451 	UPDATE_MIB(o1, ipv6IfIcmpOutMsgs, o2->ipv6IfIcmpOutMsgs);
11452 	UPDATE_MIB(o1, ipv6IfIcmpOutErrors, o2->ipv6IfIcmpOutErrors);
11453 	UPDATE_MIB(o1, ipv6IfIcmpOutDestUnreachs,
11454 	    o2->ipv6IfIcmpOutDestUnreachs);
11455 	UPDATE_MIB(o1, ipv6IfIcmpOutAdminProhibs,
11456 	    o2->ipv6IfIcmpOutAdminProhibs);
11457 	UPDATE_MIB(o1, ipv6IfIcmpOutTimeExcds, o2->ipv6IfIcmpOutTimeExcds);
11458 	UPDATE_MIB(o1, ipv6IfIcmpOutParmProblems,
11459 	    o2->ipv6IfIcmpOutParmProblems);
11460 	UPDATE_MIB(o1, ipv6IfIcmpOutPktTooBigs, o2->ipv6IfIcmpOutPktTooBigs);
11461 	UPDATE_MIB(o1, ipv6IfIcmpOutEchos, o2->ipv6IfIcmpOutEchos);
11462 	UPDATE_MIB(o1, ipv6IfIcmpOutEchoReplies, o2->ipv6IfIcmpOutEchoReplies);
11463 	UPDATE_MIB(o1, ipv6IfIcmpOutRouterSolicits,
11464 	    o2->ipv6IfIcmpOutRouterSolicits);
11465 	UPDATE_MIB(o1, ipv6IfIcmpOutRouterAdvertisements,
11466 	    o2->ipv6IfIcmpOutRouterAdvertisements);
11467 	UPDATE_MIB(o1, ipv6IfIcmpOutNeighborSolicits,
11468 	    o2->ipv6IfIcmpOutNeighborSolicits);
11469 	UPDATE_MIB(o1, ipv6IfIcmpOutNeighborAdvertisements,
11470 	    o2->ipv6IfIcmpOutNeighborAdvertisements);
11471 	UPDATE_MIB(o1, ipv6IfIcmpOutRedirects, o2->ipv6IfIcmpOutRedirects);
11472 	UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembQueries,
11473 	    o2->ipv6IfIcmpOutGroupMembQueries);
11474 	UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembResponses,
11475 	    o2->ipv6IfIcmpOutGroupMembResponses);
11476 	UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembReductions,
11477 	    o2->ipv6IfIcmpOutGroupMembReductions);
11478 	UPDATE_MIB(o1, ipv6IfIcmpInOverflows, o2->ipv6IfIcmpInOverflows);
11479 	UPDATE_MIB(o1, ipv6IfIcmpBadHoplimit, o2->ipv6IfIcmpBadHoplimit);
11480 	UPDATE_MIB(o1, ipv6IfIcmpInBadNeighborAdvertisements,
11481 	    o2->ipv6IfIcmpInBadNeighborAdvertisements);
11482 	UPDATE_MIB(o1, ipv6IfIcmpInBadNeighborSolicitations,
11483 	    o2->ipv6IfIcmpInBadNeighborSolicitations);
11484 	UPDATE_MIB(o1, ipv6IfIcmpInBadRedirects, o2->ipv6IfIcmpInBadRedirects);
11485 	UPDATE_MIB(o1, ipv6IfIcmpInGroupMembTotal,
11486 	    o2->ipv6IfIcmpInGroupMembTotal);
11487 	UPDATE_MIB(o1, ipv6IfIcmpInGroupMembBadQueries,
11488 	    o2->ipv6IfIcmpInGroupMembBadQueries);
11489 	UPDATE_MIB(o1, ipv6IfIcmpInGroupMembBadReports,
11490 	    o2->ipv6IfIcmpInGroupMembBadReports);
11491 	UPDATE_MIB(o1, ipv6IfIcmpInGroupMembOurReports,
11492 	    o2->ipv6IfIcmpInGroupMembOurReports);
11493 }
11494 
11495 /*
11496  * Called before the options are updated to check if this packet will
11497  * be source routed from here.
11498  * This routine assumes that the options are well formed i.e. that they
11499  * have already been checked.
11500  */
11501 boolean_t
ip_source_routed(ipha_t * ipha,ip_stack_t * ipst)11502 ip_source_routed(ipha_t *ipha, ip_stack_t *ipst)
11503 {
11504 	ipoptp_t	opts;
11505 	uchar_t		*opt;
11506 	uint8_t		optval;
11507 	uint8_t		optlen;
11508 	ipaddr_t	dst;
11509 
11510 	if (IS_SIMPLE_IPH(ipha)) {
11511 		ip2dbg(("not source routed\n"));
11512 		return (B_FALSE);
11513 	}
11514 	dst = ipha->ipha_dst;
11515 	for (optval = ipoptp_first(&opts, ipha);
11516 	    optval != IPOPT_EOL;
11517 	    optval = ipoptp_next(&opts)) {
11518 		ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
11519 		opt = opts.ipoptp_cur;
11520 		optlen = opts.ipoptp_len;
11521 		ip2dbg(("ip_source_routed: opt %d, len %d\n",
11522 		    optval, optlen));
11523 		switch (optval) {
11524 			uint32_t off;
11525 		case IPOPT_SSRR:
11526 		case IPOPT_LSRR:
11527 			/*
11528 			 * If dst is one of our addresses and there are some
11529 			 * entries left in the source route return (true).
11530 			 */
11531 			if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
11532 				ip2dbg(("ip_source_routed: not next"
11533 				    " source route 0x%x\n",
11534 				    ntohl(dst)));
11535 				return (B_FALSE);
11536 			}
11537 			off = opt[IPOPT_OFFSET];
11538 			off--;
11539 			if (optlen < IP_ADDR_LEN ||
11540 			    off > optlen - IP_ADDR_LEN) {
11541 				/* End of source route */
11542 				ip1dbg(("ip_source_routed: end of SR\n"));
11543 				return (B_FALSE);
11544 			}
11545 			return (B_TRUE);
11546 		}
11547 	}
11548 	ip2dbg(("not source routed\n"));
11549 	return (B_FALSE);
11550 }
11551 
11552 /*
11553  * ip_unbind is called by the transports to remove a conn from
11554  * the fanout table.
11555  */
11556 void
ip_unbind(conn_t * connp)11557 ip_unbind(conn_t *connp)
11558 {
11559 
11560 	ASSERT(!MUTEX_HELD(&connp->conn_lock));
11561 
11562 	if (is_system_labeled() && connp->conn_anon_port) {
11563 		(void) tsol_mlp_anon(crgetzone(connp->conn_cred),
11564 		    connp->conn_mlp_type, connp->conn_proto,
11565 		    ntohs(connp->conn_lport), B_FALSE);
11566 		connp->conn_anon_port = 0;
11567 	}
11568 	connp->conn_mlp_type = mlptSingle;
11569 
11570 	ipcl_hash_remove(connp);
11571 }
11572 
11573 /*
11574  * Used for deciding the MSS size for the upper layer. Thus
11575  * we need to check the outbound policy values in the conn.
11576  */
11577 int
conn_ipsec_length(conn_t * connp)11578 conn_ipsec_length(conn_t *connp)
11579 {
11580 	ipsec_latch_t *ipl;
11581 
11582 	ipl = connp->conn_latch;
11583 	if (ipl == NULL)
11584 		return (0);
11585 
11586 	if (connp->conn_ixa->ixa_ipsec_policy == NULL)
11587 		return (0);
11588 
11589 	return (connp->conn_ixa->ixa_ipsec_policy->ipsp_act->ipa_ovhd);
11590 }
11591 
11592 /*
11593  * Returns an estimate of the IPsec headers size. This is used if
11594  * we don't want to call into IPsec to get the exact size.
11595  */
11596 int
ipsec_out_extra_length(ip_xmit_attr_t * ixa)11597 ipsec_out_extra_length(ip_xmit_attr_t *ixa)
11598 {
11599 	ipsec_action_t *a;
11600 
11601 	if (!(ixa->ixa_flags & IXAF_IPSEC_SECURE))
11602 		return (0);
11603 
11604 	a = ixa->ixa_ipsec_action;
11605 	if (a == NULL) {
11606 		ASSERT(ixa->ixa_ipsec_policy != NULL);
11607 		a = ixa->ixa_ipsec_policy->ipsp_act;
11608 	}
11609 	ASSERT(a != NULL);
11610 
11611 	return (a->ipa_ovhd);
11612 }
11613 
11614 /*
11615  * If there are any source route options, return the true final
11616  * destination. Otherwise, return the destination.
11617  */
11618 ipaddr_t
ip_get_dst(ipha_t * ipha)11619 ip_get_dst(ipha_t *ipha)
11620 {
11621 	ipoptp_t	opts;
11622 	uchar_t		*opt;
11623 	uint8_t		optval;
11624 	uint8_t		optlen;
11625 	ipaddr_t	dst;
11626 	uint32_t off;
11627 
11628 	dst = ipha->ipha_dst;
11629 
11630 	if (IS_SIMPLE_IPH(ipha))
11631 		return (dst);
11632 
11633 	for (optval = ipoptp_first(&opts, ipha);
11634 	    optval != IPOPT_EOL;
11635 	    optval = ipoptp_next(&opts)) {
11636 		opt = opts.ipoptp_cur;
11637 		optlen = opts.ipoptp_len;
11638 		ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
11639 		switch (optval) {
11640 		case IPOPT_SSRR:
11641 		case IPOPT_LSRR:
11642 			off = opt[IPOPT_OFFSET];
11643 			/*
11644 			 * If one of the conditions is true, it means
11645 			 * end of options and dst already has the right
11646 			 * value.
11647 			 */
11648 			if (!(optlen < IP_ADDR_LEN || off > optlen - 3)) {
11649 				off = optlen - IP_ADDR_LEN;
11650 				bcopy(&opt[off], &dst, IP_ADDR_LEN);
11651 			}
11652 			return (dst);
11653 		default:
11654 			break;
11655 		}
11656 	}
11657 
11658 	return (dst);
11659 }
11660 
11661 /*
11662  * Outbound IP fragmentation routine.
11663  * Assumes the caller has checked whether or not fragmentation should
11664  * be allowed. Here we copy the DF bit from the header to all the generated
11665  * fragments.
11666  */
11667 int
ip_fragment_v4(mblk_t * mp_orig,nce_t * nce,iaflags_t ixaflags,uint_t pkt_len,uint32_t max_frag,uint32_t xmit_hint,zoneid_t szone,zoneid_t nolzid,pfirepostfrag_t postfragfn,uintptr_t * ixa_cookie)11668 ip_fragment_v4(mblk_t *mp_orig, nce_t *nce, iaflags_t ixaflags,
11669     uint_t pkt_len, uint32_t max_frag, uint32_t xmit_hint, zoneid_t szone,
11670     zoneid_t nolzid, pfirepostfrag_t postfragfn, uintptr_t *ixa_cookie)
11671 {
11672 	int		i1;
11673 	int		hdr_len;
11674 	mblk_t		*hdr_mp;
11675 	ipha_t		*ipha;
11676 	int		ip_data_end;
11677 	int		len;
11678 	mblk_t		*mp = mp_orig;
11679 	int		offset;
11680 	ill_t		*ill = nce->nce_ill;
11681 	ip_stack_t	*ipst = ill->ill_ipst;
11682 	mblk_t		*carve_mp;
11683 	uint32_t	frag_flag;
11684 	uint_t		priority = mp->b_band;
11685 	int		error = 0;
11686 
11687 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragReqds);
11688 
11689 	if (pkt_len != msgdsize(mp)) {
11690 		ip0dbg(("Packet length mismatch: %d, %ld\n",
11691 		    pkt_len, msgdsize(mp)));
11692 		freemsg(mp);
11693 		return (EINVAL);
11694 	}
11695 
11696 	if (max_frag == 0) {
11697 		ip1dbg(("ip_fragment_v4: max_frag is zero. Dropping packet\n"));
11698 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11699 		ip_drop_output("FragFails: zero max_frag", mp, ill);
11700 		freemsg(mp);
11701 		return (EINVAL);
11702 	}
11703 
11704 	ASSERT(MBLKL(mp) >= sizeof (ipha_t));
11705 	ipha = (ipha_t *)mp->b_rptr;
11706 	ASSERT(ntohs(ipha->ipha_length) == pkt_len);
11707 	frag_flag = ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_DF;
11708 
11709 	/*
11710 	 * Establish the starting offset.  May not be zero if we are fragging
11711 	 * a fragment that is being forwarded.
11712 	 */
11713 	offset = ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_OFFSET;
11714 
11715 	/* TODO why is this test needed? */
11716 	if (((max_frag - ntohs(ipha->ipha_length)) & ~7) < 8) {
11717 		/* TODO: notify ulp somehow */
11718 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11719 		ip_drop_output("FragFails: bad starting offset", mp, ill);
11720 		freemsg(mp);
11721 		return (EINVAL);
11722 	}
11723 
11724 	hdr_len = IPH_HDR_LENGTH(ipha);
11725 	ipha->ipha_hdr_checksum = 0;
11726 
11727 	/*
11728 	 * Establish the number of bytes maximum per frag, after putting
11729 	 * in the header.
11730 	 */
11731 	len = (max_frag - hdr_len) & ~7;
11732 
11733 	/* Get a copy of the header for the trailing frags */
11734 	hdr_mp = ip_fragment_copyhdr((uchar_t *)ipha, hdr_len, offset, ipst,
11735 	    mp);
11736 	if (hdr_mp == NULL) {
11737 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11738 		ip_drop_output("FragFails: no hdr_mp", mp, ill);
11739 		freemsg(mp);
11740 		return (ENOBUFS);
11741 	}
11742 
11743 	/* Store the starting offset, with the MoreFrags flag. */
11744 	i1 = offset | IPH_MF | frag_flag;
11745 	ipha->ipha_fragment_offset_and_flags = htons((uint16_t)i1);
11746 
11747 	/* Establish the ending byte offset, based on the starting offset. */
11748 	offset <<= 3;
11749 	ip_data_end = offset + ntohs(ipha->ipha_length) - hdr_len;
11750 
11751 	/* Store the length of the first fragment in the IP header. */
11752 	i1 = len + hdr_len;
11753 	ASSERT(i1 <= IP_MAXPACKET);
11754 	ipha->ipha_length = htons((uint16_t)i1);
11755 
11756 	/*
11757 	 * Compute the IP header checksum for the first frag.  We have to
11758 	 * watch out that we stop at the end of the header.
11759 	 */
11760 	ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
11761 
11762 	/*
11763 	 * Now carve off the first frag.  Note that this will include the
11764 	 * original IP header.
11765 	 */
11766 	if (!(mp = ip_carve_mp(&mp_orig, i1))) {
11767 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11768 		ip_drop_output("FragFails: could not carve mp", mp_orig, ill);
11769 		freeb(hdr_mp);
11770 		freemsg(mp_orig);
11771 		return (ENOBUFS);
11772 	}
11773 
11774 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragCreates);
11775 
11776 	error = postfragfn(mp, nce, ixaflags, i1, xmit_hint, szone, nolzid,
11777 	    ixa_cookie);
11778 	if (error != 0 && error != EWOULDBLOCK) {
11779 		/* No point in sending the other fragments */
11780 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11781 		ip_drop_output("FragFails: postfragfn failed", mp_orig, ill);
11782 		freeb(hdr_mp);
11783 		freemsg(mp_orig);
11784 		return (error);
11785 	}
11786 
11787 	/* No need to redo state machine in loop */
11788 	ixaflags &= ~IXAF_REACH_CONF;
11789 
11790 	/* Advance the offset to the second frag starting point. */
11791 	offset += len;
11792 	/*
11793 	 * Update hdr_len from the copied header - there might be less options
11794 	 * in the later fragments.
11795 	 */
11796 	hdr_len = IPH_HDR_LENGTH(hdr_mp->b_rptr);
11797 	/* Loop until done. */
11798 	for (;;) {
11799 		uint16_t	offset_and_flags;
11800 		uint16_t	ip_len;
11801 
11802 		if (ip_data_end - offset > len) {
11803 			/*
11804 			 * Carve off the appropriate amount from the original
11805 			 * datagram.
11806 			 */
11807 			if (!(carve_mp = ip_carve_mp(&mp_orig, len))) {
11808 				mp = NULL;
11809 				break;
11810 			}
11811 			/*
11812 			 * More frags after this one.  Get another copy
11813 			 * of the header.
11814 			 */
11815 			if (carve_mp->b_datap->db_ref == 1 &&
11816 			    hdr_mp->b_wptr - hdr_mp->b_rptr <
11817 			    carve_mp->b_rptr - carve_mp->b_datap->db_base) {
11818 				/* Inline IP header */
11819 				carve_mp->b_rptr -= hdr_mp->b_wptr -
11820 				    hdr_mp->b_rptr;
11821 				bcopy(hdr_mp->b_rptr, carve_mp->b_rptr,
11822 				    hdr_mp->b_wptr - hdr_mp->b_rptr);
11823 				mp = carve_mp;
11824 			} else {
11825 				if (!(mp = copyb(hdr_mp))) {
11826 					freemsg(carve_mp);
11827 					break;
11828 				}
11829 				/* Get priority marking, if any. */
11830 				mp->b_band = priority;
11831 				mp->b_cont = carve_mp;
11832 			}
11833 			ipha = (ipha_t *)mp->b_rptr;
11834 			offset_and_flags = IPH_MF;
11835 		} else {
11836 			/*
11837 			 * Last frag.  Consume the header. Set len to
11838 			 * the length of this last piece.
11839 			 */
11840 			len = ip_data_end - offset;
11841 
11842 			/*
11843 			 * Carve off the appropriate amount from the original
11844 			 * datagram.
11845 			 */
11846 			if (!(carve_mp = ip_carve_mp(&mp_orig, len))) {
11847 				mp = NULL;
11848 				break;
11849 			}
11850 			if (carve_mp->b_datap->db_ref == 1 &&
11851 			    hdr_mp->b_wptr - hdr_mp->b_rptr <
11852 			    carve_mp->b_rptr - carve_mp->b_datap->db_base) {
11853 				/* Inline IP header */
11854 				carve_mp->b_rptr -= hdr_mp->b_wptr -
11855 				    hdr_mp->b_rptr;
11856 				bcopy(hdr_mp->b_rptr, carve_mp->b_rptr,
11857 				    hdr_mp->b_wptr - hdr_mp->b_rptr);
11858 				mp = carve_mp;
11859 				freeb(hdr_mp);
11860 				hdr_mp = mp;
11861 			} else {
11862 				mp = hdr_mp;
11863 				/* Get priority marking, if any. */
11864 				mp->b_band = priority;
11865 				mp->b_cont = carve_mp;
11866 			}
11867 			ipha = (ipha_t *)mp->b_rptr;
11868 			/* A frag of a frag might have IPH_MF non-zero */
11869 			offset_and_flags =
11870 			    ntohs(ipha->ipha_fragment_offset_and_flags) &
11871 			    IPH_MF;
11872 		}
11873 		offset_and_flags |= (uint16_t)(offset >> 3);
11874 		offset_and_flags |= (uint16_t)frag_flag;
11875 		/* Store the offset and flags in the IP header. */
11876 		ipha->ipha_fragment_offset_and_flags = htons(offset_and_flags);
11877 
11878 		/* Store the length in the IP header. */
11879 		ip_len = (uint16_t)(len + hdr_len);
11880 		ipha->ipha_length = htons(ip_len);
11881 
11882 		/*
11883 		 * Set the IP header checksum.	Note that mp is just
11884 		 * the header, so this is easy to pass to ip_csum.
11885 		 */
11886 		ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
11887 
11888 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragCreates);
11889 
11890 		error = postfragfn(mp, nce, ixaflags, ip_len, xmit_hint, szone,
11891 		    nolzid, ixa_cookie);
11892 		/* All done if we just consumed the hdr_mp. */
11893 		if (mp == hdr_mp) {
11894 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragOKs);
11895 			return (error);
11896 		}
11897 		if (error != 0 && error != EWOULDBLOCK) {
11898 			DTRACE_PROBE2(ip__xmit__frag__fail, ill_t *, ill,
11899 			    mblk_t *, hdr_mp);
11900 			/* No point in sending the other fragments */
11901 			break;
11902 		}
11903 
11904 		/* Otherwise, advance and loop. */
11905 		offset += len;
11906 	}
11907 	/* Clean up following allocation failure. */
11908 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11909 	ip_drop_output("FragFails: loop ended", NULL, ill);
11910 	if (mp != hdr_mp)
11911 		freeb(hdr_mp);
11912 	if (mp != mp_orig)
11913 		freemsg(mp_orig);
11914 	return (error);
11915 }
11916 
11917 /*
11918  * Copy the header plus those options which have the copy bit set
11919  */
11920 static mblk_t *
ip_fragment_copyhdr(uchar_t * rptr,int hdr_len,int offset,ip_stack_t * ipst,mblk_t * src)11921 ip_fragment_copyhdr(uchar_t *rptr, int hdr_len, int offset, ip_stack_t *ipst,
11922     mblk_t *src)
11923 {
11924 	mblk_t	*mp;
11925 	uchar_t	*up;
11926 
11927 	/*
11928 	 * Quick check if we need to look for options without the copy bit
11929 	 * set
11930 	 */
11931 	mp = allocb_tmpl(ipst->ips_ip_wroff_extra + hdr_len, src);
11932 	if (!mp)
11933 		return (mp);
11934 	mp->b_rptr += ipst->ips_ip_wroff_extra;
11935 	if (hdr_len == IP_SIMPLE_HDR_LENGTH || offset != 0) {
11936 		bcopy(rptr, mp->b_rptr, hdr_len);
11937 		mp->b_wptr += hdr_len + ipst->ips_ip_wroff_extra;
11938 		return (mp);
11939 	}
11940 	up  = mp->b_rptr;
11941 	bcopy(rptr, up, IP_SIMPLE_HDR_LENGTH);
11942 	up += IP_SIMPLE_HDR_LENGTH;
11943 	rptr += IP_SIMPLE_HDR_LENGTH;
11944 	hdr_len -= IP_SIMPLE_HDR_LENGTH;
11945 	while (hdr_len > 0) {
11946 		uint32_t optval;
11947 		uint32_t optlen;
11948 
11949 		optval = *rptr;
11950 		if (optval == IPOPT_EOL)
11951 			break;
11952 		if (optval == IPOPT_NOP)
11953 			optlen = 1;
11954 		else
11955 			optlen = rptr[1];
11956 		if (optval & IPOPT_COPY) {
11957 			bcopy(rptr, up, optlen);
11958 			up += optlen;
11959 		}
11960 		rptr += optlen;
11961 		hdr_len -= optlen;
11962 	}
11963 	/*
11964 	 * Make sure that we drop an even number of words by filling
11965 	 * with EOL to the next word boundary.
11966 	 */
11967 	for (hdr_len = up - (mp->b_rptr + IP_SIMPLE_HDR_LENGTH);
11968 	    hdr_len & 0x3; hdr_len++)
11969 		*up++ = IPOPT_EOL;
11970 	mp->b_wptr = up;
11971 	/* Update header length */
11972 	mp->b_rptr[0] = (uint8_t)((IP_VERSION << 4) | ((up - mp->b_rptr) >> 2));
11973 	return (mp);
11974 }
11975 
11976 /*
11977  * Update any source route, record route, or timestamp options when
11978  * sending a packet back to ourselves.
11979  * Check that we are at end of strict source route.
11980  * The options have been sanity checked by ip_output_options().
11981  */
11982 void
ip_output_local_options(ipha_t * ipha,ip_stack_t * ipst)11983 ip_output_local_options(ipha_t *ipha, ip_stack_t *ipst)
11984 {
11985 	ipoptp_t	opts;
11986 	uchar_t		*opt;
11987 	uint8_t		optval;
11988 	uint8_t		optlen;
11989 	ipaddr_t	dst;
11990 	uint32_t	ts;
11991 	timestruc_t	now;
11992 	uint32_t	off = 0;
11993 
11994 	for (optval = ipoptp_first(&opts, ipha);
11995 	    optval != IPOPT_EOL;
11996 	    optval = ipoptp_next(&opts)) {
11997 		opt = opts.ipoptp_cur;
11998 		optlen = opts.ipoptp_len;
11999 		ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
12000 		switch (optval) {
12001 		case IPOPT_SSRR:
12002 		case IPOPT_LSRR:
12003 			off = opt[IPOPT_OFFSET];
12004 			off--;
12005 			if (optlen < IP_ADDR_LEN ||
12006 			    off > optlen - IP_ADDR_LEN) {
12007 				/* End of source route */
12008 				break;
12009 			}
12010 			/*
12011 			 * This will only happen if two consecutive entries
12012 			 * in the source route contains our address or if
12013 			 * it is a packet with a loose source route which
12014 			 * reaches us before consuming the whole source route
12015 			 */
12016 
12017 			if (optval == IPOPT_SSRR) {
12018 				return;
12019 			}
12020 			/*
12021 			 * Hack: instead of dropping the packet truncate the
12022 			 * source route to what has been used by filling the
12023 			 * rest with IPOPT_NOP.
12024 			 */
12025 			opt[IPOPT_OLEN] = (uint8_t)off;
12026 			while (off < optlen) {
12027 				opt[off++] = IPOPT_NOP;
12028 			}
12029 			break;
12030 		case IPOPT_RR:
12031 			off = opt[IPOPT_OFFSET];
12032 			off--;
12033 			if (optlen < IP_ADDR_LEN ||
12034 			    off > optlen - IP_ADDR_LEN) {
12035 				/* No more room - ignore */
12036 				ip1dbg((
12037 				    "ip_output_local_options: end of RR\n"));
12038 				break;
12039 			}
12040 			dst = htonl(INADDR_LOOPBACK);
12041 			bcopy(&dst, (char *)opt + off, IP_ADDR_LEN);
12042 			opt[IPOPT_OFFSET] += IP_ADDR_LEN;
12043 			break;
12044 		case IPOPT_TS:
12045 			/* Insert timestamp if there is romm */
12046 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
12047 			case IPOPT_TS_TSONLY:
12048 				off = IPOPT_TS_TIMELEN;
12049 				break;
12050 			case IPOPT_TS_PRESPEC:
12051 			case IPOPT_TS_PRESPEC_RFC791:
12052 				/* Verify that the address matched */
12053 				off = opt[IPOPT_OFFSET] - 1;
12054 				bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
12055 				if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
12056 					/* Not for us */
12057 					break;
12058 				}
12059 				/* FALLTHROUGH */
12060 			case IPOPT_TS_TSANDADDR:
12061 				off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
12062 				break;
12063 			default:
12064 				/*
12065 				 * ip_*put_options should have already
12066 				 * dropped this packet.
12067 				 */
12068 				cmn_err(CE_PANIC, "ip_output_local_options: "
12069 				    "unknown IT - bug in ip_output_options?\n");
12070 			}
12071 			if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
12072 				/* Increase overflow counter */
12073 				off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
12074 				opt[IPOPT_POS_OV_FLG] = (uint8_t)
12075 				    (opt[IPOPT_POS_OV_FLG] & 0x0F) |
12076 				    (off << 4);
12077 				break;
12078 			}
12079 			off = opt[IPOPT_OFFSET] - 1;
12080 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
12081 			case IPOPT_TS_PRESPEC:
12082 			case IPOPT_TS_PRESPEC_RFC791:
12083 			case IPOPT_TS_TSANDADDR:
12084 				dst = htonl(INADDR_LOOPBACK);
12085 				bcopy(&dst, (char *)opt + off, IP_ADDR_LEN);
12086 				opt[IPOPT_OFFSET] += IP_ADDR_LEN;
12087 				/* FALLTHROUGH */
12088 			case IPOPT_TS_TSONLY:
12089 				off = opt[IPOPT_OFFSET] - 1;
12090 				/* Compute # of milliseconds since midnight */
12091 				gethrestime(&now);
12092 				ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
12093 				    NSEC2MSEC(now.tv_nsec);
12094 				bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
12095 				opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
12096 				break;
12097 			}
12098 			break;
12099 		}
12100 	}
12101 }
12102 
12103 /*
12104  * Prepend an M_DATA fastpath header, and if none present prepend a
12105  * DL_UNITDATA_REQ. Frees the mblk on failure.
12106  *
12107  * nce_dlur_mp and nce_fp_mp can not disappear once they have been set.
12108  * If there is a change to them, the nce will be deleted (condemned) and
12109  * a new nce_t will be created when packets are sent. Thus we need no locks
12110  * to access those fields.
12111  *
12112  * We preserve b_band to support IPQoS. If a DL_UNITDATA_REQ is prepended
12113  * we place b_band in dl_priority.dl_max.
12114  */
12115 static mblk_t *
ip_xmit_attach_llhdr(mblk_t * mp,nce_t * nce)12116 ip_xmit_attach_llhdr(mblk_t *mp, nce_t *nce)
12117 {
12118 	uint_t	hlen;
12119 	mblk_t *mp1;
12120 	uint_t	priority;
12121 	uchar_t *rptr;
12122 
12123 	rptr = mp->b_rptr;
12124 
12125 	ASSERT(DB_TYPE(mp) == M_DATA);
12126 	priority = mp->b_band;
12127 
12128 	ASSERT(nce != NULL);
12129 	if ((mp1 = nce->nce_fp_mp) != NULL) {
12130 		hlen = MBLKL(mp1);
12131 		/*
12132 		 * Check if we have enough room to prepend fastpath
12133 		 * header
12134 		 */
12135 		if (hlen != 0 && (rptr - mp->b_datap->db_base) >= hlen) {
12136 			rptr -= hlen;
12137 			bcopy(mp1->b_rptr, rptr, hlen);
12138 			/*
12139 			 * Set the b_rptr to the start of the link layer
12140 			 * header
12141 			 */
12142 			mp->b_rptr = rptr;
12143 			return (mp);
12144 		}
12145 		mp1 = copyb(mp1);
12146 		if (mp1 == NULL) {
12147 			ill_t *ill = nce->nce_ill;
12148 
12149 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12150 			ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12151 			freemsg(mp);
12152 			return (NULL);
12153 		}
12154 		mp1->b_band = priority;
12155 		mp1->b_cont = mp;
12156 		DB_CKSUMSTART(mp1) = DB_CKSUMSTART(mp);
12157 		DB_CKSUMSTUFF(mp1) = DB_CKSUMSTUFF(mp);
12158 		DB_CKSUMEND(mp1) = DB_CKSUMEND(mp);
12159 		DB_CKSUMFLAGS(mp1) = DB_CKSUMFLAGS(mp);
12160 		DB_LSOMSS(mp1) = DB_LSOMSS(mp);
12161 		DTRACE_PROBE1(ip__xmit__copyb, (mblk_t *), mp1);
12162 		/*
12163 		 * XXX disable ICK_VALID and compute checksum
12164 		 * here; can happen if nce_fp_mp changes and
12165 		 * it can't be copied now due to insufficient
12166 		 * space. (unlikely, fp mp can change, but it
12167 		 * does not increase in length)
12168 		 */
12169 		return (mp1);
12170 	}
12171 	mp1 = copyb(nce->nce_dlur_mp);
12172 
12173 	if (mp1 == NULL) {
12174 		ill_t *ill = nce->nce_ill;
12175 
12176 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12177 		ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12178 		freemsg(mp);
12179 		return (NULL);
12180 	}
12181 	mp1->b_cont = mp;
12182 	if (priority != 0) {
12183 		mp1->b_band = priority;
12184 		((dl_unitdata_req_t *)(mp1->b_rptr))->dl_priority.dl_max =
12185 		    priority;
12186 	}
12187 	return (mp1);
12188 }
12189 
12190 /*
12191  * Finish the outbound IPsec processing. This function is called from
12192  * ipsec_out_process() if the IPsec packet was processed
12193  * synchronously, or from {ah,esp}_kcf_callback_outbound() if it was processed
12194  * asynchronously.
12195  *
12196  * This is common to IPv4 and IPv6.
12197  */
12198 int
ip_output_post_ipsec(mblk_t * mp,ip_xmit_attr_t * ixa)12199 ip_output_post_ipsec(mblk_t *mp, ip_xmit_attr_t *ixa)
12200 {
12201 	iaflags_t	ixaflags = ixa->ixa_flags;
12202 	uint_t		pktlen;
12203 
12204 
12205 	/* AH/ESP don't update ixa_pktlen when they modify the packet */
12206 	if (ixaflags & IXAF_IS_IPV4) {
12207 		ipha_t		*ipha = (ipha_t *)mp->b_rptr;
12208 
12209 		ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
12210 		pktlen = ntohs(ipha->ipha_length);
12211 	} else {
12212 		ip6_t		*ip6h = (ip6_t *)mp->b_rptr;
12213 
12214 		ASSERT(IPH_HDR_VERSION(mp->b_rptr) == IPV6_VERSION);
12215 		pktlen = ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN;
12216 	}
12217 
12218 	/*
12219 	 * We release any hard reference on the SAs here to make
12220 	 * sure the SAs can be garbage collected. ipsr_sa has a soft reference
12221 	 * on the SAs.
12222 	 * If in the future we want the hard latching of the SAs in the
12223 	 * ip_xmit_attr_t then we should remove this.
12224 	 */
12225 	if (ixa->ixa_ipsec_esp_sa != NULL) {
12226 		IPSA_REFRELE(ixa->ixa_ipsec_esp_sa);
12227 		ixa->ixa_ipsec_esp_sa = NULL;
12228 	}
12229 	if (ixa->ixa_ipsec_ah_sa != NULL) {
12230 		IPSA_REFRELE(ixa->ixa_ipsec_ah_sa);
12231 		ixa->ixa_ipsec_ah_sa = NULL;
12232 	}
12233 
12234 	/* Do we need to fragment? */
12235 	if ((ixa->ixa_flags & IXAF_IPV6_ADD_FRAGHDR) ||
12236 	    pktlen > ixa->ixa_fragsize) {
12237 		if (ixaflags & IXAF_IS_IPV4) {
12238 			ASSERT(!(ixa->ixa_flags & IXAF_IPV6_ADD_FRAGHDR));
12239 			/*
12240 			 * We check for the DF case in ipsec_out_process
12241 			 * hence this only handles the non-DF case.
12242 			 */
12243 			return (ip_fragment_v4(mp, ixa->ixa_nce, ixa->ixa_flags,
12244 			    pktlen, ixa->ixa_fragsize,
12245 			    ixa->ixa_xmit_hint, ixa->ixa_zoneid,
12246 			    ixa->ixa_no_loop_zoneid, ixa->ixa_postfragfn,
12247 			    &ixa->ixa_cookie));
12248 		} else {
12249 			mp = ip_fraghdr_add_v6(mp, ixa->ixa_ident, ixa);
12250 			if (mp == NULL) {
12251 				/* MIB and ip_drop_output already done */
12252 				return (ENOMEM);
12253 			}
12254 			pktlen += sizeof (ip6_frag_t);
12255 			if (pktlen > ixa->ixa_fragsize) {
12256 				return (ip_fragment_v6(mp, ixa->ixa_nce,
12257 				    ixa->ixa_flags, pktlen,
12258 				    ixa->ixa_fragsize, ixa->ixa_xmit_hint,
12259 				    ixa->ixa_zoneid, ixa->ixa_no_loop_zoneid,
12260 				    ixa->ixa_postfragfn, &ixa->ixa_cookie));
12261 			}
12262 		}
12263 	}
12264 	return ((ixa->ixa_postfragfn)(mp, ixa->ixa_nce, ixa->ixa_flags,
12265 	    pktlen, ixa->ixa_xmit_hint, ixa->ixa_zoneid,
12266 	    ixa->ixa_no_loop_zoneid, NULL));
12267 }
12268 
12269 /*
12270  * Finish the inbound IPsec processing. This function is called from
12271  * ipsec_out_process() if the IPsec packet was processed
12272  * synchronously, or from {ah,esp}_kcf_callback_outbound() if it was processed
12273  * asynchronously.
12274  *
12275  * This is common to IPv4 and IPv6.
12276  */
12277 void
ip_input_post_ipsec(mblk_t * mp,ip_recv_attr_t * ira)12278 ip_input_post_ipsec(mblk_t *mp, ip_recv_attr_t *ira)
12279 {
12280 	iaflags_t	iraflags = ira->ira_flags;
12281 
12282 	/* Length might have changed */
12283 	if (iraflags & IRAF_IS_IPV4) {
12284 		ipha_t		*ipha = (ipha_t *)mp->b_rptr;
12285 
12286 		ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
12287 		ira->ira_pktlen = ntohs(ipha->ipha_length);
12288 		ira->ira_ip_hdr_length = IPH_HDR_LENGTH(ipha);
12289 		ira->ira_protocol = ipha->ipha_protocol;
12290 
12291 		ip_fanout_v4(mp, ipha, ira);
12292 	} else {
12293 		ip6_t		*ip6h = (ip6_t *)mp->b_rptr;
12294 		uint8_t		*nexthdrp;
12295 
12296 		ASSERT(IPH_HDR_VERSION(mp->b_rptr) == IPV6_VERSION);
12297 		ira->ira_pktlen = ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN;
12298 		if (!ip_hdr_length_nexthdr_v6(mp, ip6h, &ira->ira_ip_hdr_length,
12299 		    &nexthdrp)) {
12300 			/* Malformed packet */
12301 			BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
12302 			ip_drop_input("ipIfStatsInDiscards", mp, ira->ira_ill);
12303 			freemsg(mp);
12304 			return;
12305 		}
12306 		ira->ira_protocol = *nexthdrp;
12307 		ip_fanout_v6(mp, ip6h, ira);
12308 	}
12309 }
12310 
12311 /*
12312  * Select which AH & ESP SA's to use (if any) for the outbound packet.
12313  *
12314  * If this function returns B_TRUE, the requested SA's have been filled
12315  * into the ixa_ipsec_*_sa pointers.
12316  *
12317  * If the function returns B_FALSE, the packet has been "consumed", most
12318  * likely by an ACQUIRE sent up via PF_KEY to a key management daemon.
12319  *
12320  * The SA references created by the protocol-specific "select"
12321  * function will be released in ip_output_post_ipsec.
12322  */
12323 static boolean_t
ipsec_out_select_sa(mblk_t * mp,ip_xmit_attr_t * ixa)12324 ipsec_out_select_sa(mblk_t *mp, ip_xmit_attr_t *ixa)
12325 {
12326 	boolean_t need_ah_acquire = B_FALSE, need_esp_acquire = B_FALSE;
12327 	ipsec_policy_t *pp;
12328 	ipsec_action_t *ap;
12329 
12330 	ASSERT(ixa->ixa_flags & IXAF_IPSEC_SECURE);
12331 	ASSERT((ixa->ixa_ipsec_policy != NULL) ||
12332 	    (ixa->ixa_ipsec_action != NULL));
12333 
12334 	ap = ixa->ixa_ipsec_action;
12335 	if (ap == NULL) {
12336 		pp = ixa->ixa_ipsec_policy;
12337 		ASSERT(pp != NULL);
12338 		ap = pp->ipsp_act;
12339 		ASSERT(ap != NULL);
12340 	}
12341 
12342 	/*
12343 	 * We have an action.  now, let's select SA's.
12344 	 * A side effect of setting ixa_ipsec_*_sa is that it will
12345 	 * be cached in the conn_t.
12346 	 */
12347 	if (ap->ipa_want_esp) {
12348 		if (ixa->ixa_ipsec_esp_sa == NULL) {
12349 			need_esp_acquire = !ipsec_outbound_sa(mp, ixa,
12350 			    IPPROTO_ESP);
12351 		}
12352 		ASSERT(need_esp_acquire || ixa->ixa_ipsec_esp_sa != NULL);
12353 	}
12354 
12355 	if (ap->ipa_want_ah) {
12356 		if (ixa->ixa_ipsec_ah_sa == NULL) {
12357 			need_ah_acquire = !ipsec_outbound_sa(mp, ixa,
12358 			    IPPROTO_AH);
12359 		}
12360 		ASSERT(need_ah_acquire || ixa->ixa_ipsec_ah_sa != NULL);
12361 		/*
12362 		 * The ESP and AH processing order needs to be preserved
12363 		 * when both protocols are required (ESP should be applied
12364 		 * before AH for an outbound packet). Force an ESP ACQUIRE
12365 		 * when both ESP and AH are required, and an AH ACQUIRE
12366 		 * is needed.
12367 		 */
12368 		if (ap->ipa_want_esp && need_ah_acquire)
12369 			need_esp_acquire = B_TRUE;
12370 	}
12371 
12372 	/*
12373 	 * Send an ACQUIRE (extended, regular, or both) if we need one.
12374 	 * Release SAs that got referenced, but will not be used until we
12375 	 * acquire _all_ of the SAs we need.
12376 	 */
12377 	if (need_ah_acquire || need_esp_acquire) {
12378 		if (ixa->ixa_ipsec_ah_sa != NULL) {
12379 			IPSA_REFRELE(ixa->ixa_ipsec_ah_sa);
12380 			ixa->ixa_ipsec_ah_sa = NULL;
12381 		}
12382 		if (ixa->ixa_ipsec_esp_sa != NULL) {
12383 			IPSA_REFRELE(ixa->ixa_ipsec_esp_sa);
12384 			ixa->ixa_ipsec_esp_sa = NULL;
12385 		}
12386 
12387 		sadb_acquire(mp, ixa, need_ah_acquire, need_esp_acquire);
12388 		return (B_FALSE);
12389 	}
12390 
12391 	return (B_TRUE);
12392 }
12393 
12394 /*
12395  * Handle IPsec output processing.
12396  * This function is only entered once for a given packet.
12397  * We try to do things synchronously, but if we need to have user-level
12398  * set up SAs, or ESP or AH uses asynchronous kEF, then the operation
12399  * will be completed
12400  *  - when the SAs are added in esp_add_sa_finish/ah_add_sa_finish
12401  *  - when asynchronous ESP is done it will do AH
12402  *
12403  * In all cases we come back in ip_output_post_ipsec() to fragment and
12404  * send out the packet.
12405  */
12406 int
ipsec_out_process(mblk_t * mp,ip_xmit_attr_t * ixa)12407 ipsec_out_process(mblk_t *mp, ip_xmit_attr_t *ixa)
12408 {
12409 	ill_t		*ill = ixa->ixa_nce->nce_ill;
12410 	ip_stack_t	*ipst = ixa->ixa_ipst;
12411 	ipsec_stack_t	*ipss;
12412 	ipsec_policy_t	*pp;
12413 	ipsec_action_t	*ap;
12414 
12415 	ASSERT(ixa->ixa_flags & IXAF_IPSEC_SECURE);
12416 
12417 	ASSERT((ixa->ixa_ipsec_policy != NULL) ||
12418 	    (ixa->ixa_ipsec_action != NULL));
12419 
12420 	ipss = ipst->ips_netstack->netstack_ipsec;
12421 	if (!ipsec_loaded(ipss)) {
12422 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12423 		ip_drop_packet(mp, B_TRUE, ill,
12424 		    DROPPER(ipss, ipds_ip_ipsec_not_loaded),
12425 		    &ipss->ipsec_dropper);
12426 		return (ENOTSUP);
12427 	}
12428 
12429 	ap = ixa->ixa_ipsec_action;
12430 	if (ap == NULL) {
12431 		pp = ixa->ixa_ipsec_policy;
12432 		ASSERT(pp != NULL);
12433 		ap = pp->ipsp_act;
12434 		ASSERT(ap != NULL);
12435 	}
12436 
12437 	/* Handle explicit drop action and bypass. */
12438 	switch (ap->ipa_act.ipa_type) {
12439 	case IPSEC_ACT_DISCARD:
12440 	case IPSEC_ACT_REJECT:
12441 		ip_drop_packet(mp, B_FALSE, ill,
12442 		    DROPPER(ipss, ipds_spd_explicit), &ipss->ipsec_spd_dropper);
12443 		return (EHOSTUNREACH);	/* IPsec policy failure */
12444 	case IPSEC_ACT_BYPASS:
12445 		return (ip_output_post_ipsec(mp, ixa));
12446 	}
12447 
12448 	/*
12449 	 * The order of processing is first insert a IP header if needed.
12450 	 * Then insert the ESP header and then the AH header.
12451 	 */
12452 	if ((ixa->ixa_flags & IXAF_IS_IPV4) && ap->ipa_want_se) {
12453 		/*
12454 		 * First get the outer IP header before sending
12455 		 * it to ESP.
12456 		 */
12457 		ipha_t *oipha, *iipha;
12458 		mblk_t *outer_mp, *inner_mp;
12459 
12460 		if ((outer_mp = allocb(sizeof (ipha_t), BPRI_HI)) == NULL) {
12461 			(void) mi_strlog(ill->ill_rq, 0,
12462 			    SL_ERROR|SL_TRACE|SL_CONSOLE,
12463 			    "ipsec_out_process: "
12464 			    "Self-Encapsulation failed: Out of memory\n");
12465 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12466 			ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12467 			freemsg(mp);
12468 			return (ENOBUFS);
12469 		}
12470 		inner_mp = mp;
12471 		ASSERT(inner_mp->b_datap->db_type == M_DATA);
12472 		oipha = (ipha_t *)outer_mp->b_rptr;
12473 		iipha = (ipha_t *)inner_mp->b_rptr;
12474 		*oipha = *iipha;
12475 		outer_mp->b_wptr += sizeof (ipha_t);
12476 		oipha->ipha_length = htons(ntohs(iipha->ipha_length) +
12477 		    sizeof (ipha_t));
12478 		oipha->ipha_protocol = IPPROTO_ENCAP;
12479 		oipha->ipha_version_and_hdr_length =
12480 		    IP_SIMPLE_HDR_VERSION;
12481 		oipha->ipha_hdr_checksum = 0;
12482 		oipha->ipha_hdr_checksum = ip_csum_hdr(oipha);
12483 		outer_mp->b_cont = inner_mp;
12484 		mp = outer_mp;
12485 
12486 		ixa->ixa_flags |= IXAF_IPSEC_TUNNEL;
12487 	}
12488 
12489 	/* If we need to wait for a SA then we can't return any errno */
12490 	if (((ap->ipa_want_ah && (ixa->ixa_ipsec_ah_sa == NULL)) ||
12491 	    (ap->ipa_want_esp && (ixa->ixa_ipsec_esp_sa == NULL))) &&
12492 	    !ipsec_out_select_sa(mp, ixa))
12493 		return (0);
12494 
12495 	/*
12496 	 * By now, we know what SA's to use.  Toss over to ESP & AH
12497 	 * to do the heavy lifting.
12498 	 */
12499 	if (ap->ipa_want_esp) {
12500 		ASSERT(ixa->ixa_ipsec_esp_sa != NULL);
12501 
12502 		mp = ixa->ixa_ipsec_esp_sa->ipsa_output_func(mp, ixa);
12503 		if (mp == NULL) {
12504 			/*
12505 			 * Either it failed or is pending. In the former case
12506 			 * ipIfStatsInDiscards was increased.
12507 			 */
12508 			return (0);
12509 		}
12510 	}
12511 
12512 	if (ap->ipa_want_ah) {
12513 		ASSERT(ixa->ixa_ipsec_ah_sa != NULL);
12514 
12515 		mp = ixa->ixa_ipsec_ah_sa->ipsa_output_func(mp, ixa);
12516 		if (mp == NULL) {
12517 			/*
12518 			 * Either it failed or is pending. In the former case
12519 			 * ipIfStatsInDiscards was increased.
12520 			 */
12521 			return (0);
12522 		}
12523 	}
12524 	/*
12525 	 * We are done with IPsec processing. Send it over
12526 	 * the wire.
12527 	 */
12528 	return (ip_output_post_ipsec(mp, ixa));
12529 }
12530 
12531 /*
12532  * ioctls that go through a down/up sequence may need to wait for the down
12533  * to complete. This involves waiting for the ire and ipif refcnts to go down
12534  * to zero. Subsequently the ioctl is restarted from ipif_ill_refrele_tail.
12535  */
12536 /* ARGSUSED */
12537 void
ip_reprocess_ioctl(ipsq_t * ipsq,queue_t * q,mblk_t * mp,void * dummy_arg)12538 ip_reprocess_ioctl(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
12539 {
12540 	struct iocblk *iocp;
12541 	mblk_t *mp1;
12542 	ip_ioctl_cmd_t *ipip;
12543 	int err;
12544 	sin_t	*sin;
12545 	struct lifreq *lifr;
12546 	struct ifreq *ifr;
12547 
12548 	iocp = (struct iocblk *)mp->b_rptr;
12549 	ASSERT(ipsq != NULL);
12550 	/* Existence of mp1 verified in ip_wput_nondata */
12551 	mp1 = mp->b_cont->b_cont;
12552 	ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12553 	if (ipip->ipi_cmd == SIOCSLIFNAME || ipip->ipi_cmd == IF_UNITSEL) {
12554 		/*
12555 		 * Special case where ipx_current_ipif is not set:
12556 		 * ill_phyint_reinit merged the v4 and v6 into a single ipsq.
12557 		 * We are here as were not able to complete the operation in
12558 		 * ipif_set_values because we could not become exclusive on
12559 		 * the new ipsq.
12560 		 */
12561 		ill_t *ill = q->q_ptr;
12562 		ipsq_current_start(ipsq, ill->ill_ipif, ipip->ipi_cmd);
12563 	}
12564 	ASSERT(ipsq->ipsq_xop->ipx_current_ipif != NULL);
12565 
12566 	if (ipip->ipi_cmd_type == IF_CMD) {
12567 		/* This a old style SIOC[GS]IF* command */
12568 		ifr = (struct ifreq *)mp1->b_rptr;
12569 		sin = (sin_t *)&ifr->ifr_addr;
12570 	} else if (ipip->ipi_cmd_type == LIF_CMD) {
12571 		/* This a new style SIOC[GS]LIF* command */
12572 		lifr = (struct lifreq *)mp1->b_rptr;
12573 		sin = (sin_t *)&lifr->lifr_addr;
12574 	} else {
12575 		sin = NULL;
12576 	}
12577 
12578 	err = (*ipip->ipi_func_restart)(ipsq->ipsq_xop->ipx_current_ipif, sin,
12579 	    q, mp, ipip, mp1->b_rptr);
12580 
12581 	DTRACE_PROBE4(ipif__ioctl, char *, "ip_reprocess_ioctl finish",
12582 	    int, ipip->ipi_cmd,
12583 	    ill_t *, ipsq->ipsq_xop->ipx_current_ipif->ipif_ill,
12584 	    ipif_t *, ipsq->ipsq_xop->ipx_current_ipif);
12585 
12586 	ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), ipsq);
12587 }
12588 
12589 /*
12590  * ioctl processing
12591  *
12592  * ioctl processing starts with ip_sioctl_copyin_setup(), which looks up
12593  * the ioctl command in the ioctl tables, determines the copyin data size
12594  * from the ipi_copyin_size field, and does an mi_copyin() of that size.
12595  *
12596  * ioctl processing then continues when the M_IOCDATA makes its way down to
12597  * ip_wput_nondata().  The ioctl is looked up again in the ioctl table, its
12598  * associated 'conn' is refheld till the end of the ioctl and the general
12599  * ioctl processing function ip_process_ioctl() is called to extract the
12600  * arguments and process the ioctl.  To simplify extraction, ioctl commands
12601  * are "typed" based on the arguments they take (e.g., LIF_CMD which takes a
12602  * `struct lifreq'), and a common extract function (e.g., ip_extract_lifreq())
12603  * is used to extract the ioctl's arguments.
12604  *
12605  * ip_process_ioctl determines if the ioctl needs to be serialized, and if
12606  * so goes thru the serialization primitive ipsq_try_enter. Then the
12607  * appropriate function to handle the ioctl is called based on the entry in
12608  * the ioctl table. ioctl completion is encapsulated in ip_ioctl_finish
12609  * which also refreleases the 'conn' that was refheld at the start of the
12610  * ioctl. Finally ipsq_exit is called if needed to exit the ipsq.
12611  *
12612  * Many exclusive ioctls go thru an internal down up sequence as part of
12613  * the operation. For example an attempt to change the IP address of an
12614  * ipif entails ipif_down, set address, ipif_up. Bringing down the interface
12615  * does all the cleanup such as deleting all ires that use this address.
12616  * Then we need to wait till all references to the interface go away.
12617  */
12618 void
ip_process_ioctl(ipsq_t * ipsq,queue_t * q,mblk_t * mp,void * arg)12619 ip_process_ioctl(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *arg)
12620 {
12621 	struct iocblk *iocp = (struct iocblk *)mp->b_rptr;
12622 	ip_ioctl_cmd_t *ipip = arg;
12623 	ip_extract_func_t *extract_funcp;
12624 	cmd_info_t ci;
12625 	int err;
12626 	boolean_t entered_ipsq = B_FALSE;
12627 
12628 	ip3dbg(("ip_process_ioctl: ioctl %X\n", iocp->ioc_cmd));
12629 
12630 	if (ipip == NULL)
12631 		ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12632 
12633 	/*
12634 	 * SIOCLIFADDIF needs to go thru a special path since the
12635 	 * ill may not exist yet. This happens in the case of lo0
12636 	 * which is created using this ioctl.
12637 	 */
12638 	if (ipip->ipi_cmd == SIOCLIFADDIF) {
12639 		err = ip_sioctl_addif(NULL, NULL, q, mp, NULL, NULL);
12640 		DTRACE_PROBE4(ipif__ioctl, char *, "ip_process_ioctl finish",
12641 		    int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12642 		ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12643 		return;
12644 	}
12645 
12646 	ci.ci_ipif = NULL;
12647 	extract_funcp = NULL;
12648 	switch (ipip->ipi_cmd_type) {
12649 	case MISC_CMD:
12650 	case MSFILT_CMD:
12651 		/*
12652 		 * All MISC_CMD ioctls come in here -- e.g. SIOCGLIFCONF.
12653 		 */
12654 		if (ipip->ipi_cmd == IF_UNITSEL) {
12655 			/* ioctl comes down the ill */
12656 			ci.ci_ipif = ((ill_t *)q->q_ptr)->ill_ipif;
12657 			ipif_refhold(ci.ci_ipif);
12658 		}
12659 		err = 0;
12660 		ci.ci_sin = NULL;
12661 		ci.ci_sin6 = NULL;
12662 		ci.ci_lifr = NULL;
12663 		extract_funcp = NULL;
12664 		break;
12665 
12666 	case IF_CMD:
12667 	case LIF_CMD:
12668 		extract_funcp = ip_extract_lifreq;
12669 		break;
12670 
12671 	case ARP_CMD:
12672 	case XARP_CMD:
12673 		extract_funcp = ip_extract_arpreq;
12674 		break;
12675 
12676 	default:
12677 		ASSERT(0);
12678 	}
12679 
12680 	if (extract_funcp != NULL) {
12681 		err = (*extract_funcp)(q, mp, ipip, &ci);
12682 		if (err != 0) {
12683 			DTRACE_PROBE4(ipif__ioctl,
12684 			    char *, "ip_process_ioctl finish err",
12685 			    int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12686 			ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12687 			return;
12688 		}
12689 
12690 		/*
12691 		 * All of the extraction functions return a refheld ipif.
12692 		 */
12693 		ASSERT(ci.ci_ipif != NULL);
12694 	}
12695 
12696 	if (!(ipip->ipi_flags & IPI_WR)) {
12697 		/*
12698 		 * A return value of EINPROGRESS means the ioctl is
12699 		 * either queued and waiting for some reason or has
12700 		 * already completed.
12701 		 */
12702 		err = (*ipip->ipi_func)(ci.ci_ipif, ci.ci_sin, q, mp, ipip,
12703 		    ci.ci_lifr);
12704 		if (ci.ci_ipif != NULL) {
12705 			DTRACE_PROBE4(ipif__ioctl,
12706 			    char *, "ip_process_ioctl finish RD",
12707 			    int, ipip->ipi_cmd, ill_t *, ci.ci_ipif->ipif_ill,
12708 			    ipif_t *, ci.ci_ipif);
12709 			ipif_refrele(ci.ci_ipif);
12710 		} else {
12711 			DTRACE_PROBE4(ipif__ioctl,
12712 			    char *, "ip_process_ioctl finish RD",
12713 			    int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12714 		}
12715 		ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12716 		return;
12717 	}
12718 
12719 	ASSERT(ci.ci_ipif != NULL);
12720 
12721 	/*
12722 	 * If ipsq is non-NULL, we are already being called exclusively
12723 	 */
12724 	ASSERT(ipsq == NULL || IAM_WRITER_IPSQ(ipsq));
12725 	if (ipsq == NULL) {
12726 		ipsq = ipsq_try_enter(ci.ci_ipif, NULL, q, mp, ip_process_ioctl,
12727 		    NEW_OP, B_TRUE);
12728 		if (ipsq == NULL) {
12729 			ipif_refrele(ci.ci_ipif);
12730 			return;
12731 		}
12732 		entered_ipsq = B_TRUE;
12733 	}
12734 	/*
12735 	 * Release the ipif so that ipif_down and friends that wait for
12736 	 * references to go away are not misled about the current ipif_refcnt
12737 	 * values. We are writer so we can access the ipif even after releasing
12738 	 * the ipif.
12739 	 */
12740 	ipif_refrele(ci.ci_ipif);
12741 
12742 	ipsq_current_start(ipsq, ci.ci_ipif, ipip->ipi_cmd);
12743 
12744 	/*
12745 	 * A return value of EINPROGRESS means the ioctl is
12746 	 * either queued and waiting for some reason or has
12747 	 * already completed.
12748 	 */
12749 	err = (*ipip->ipi_func)(ci.ci_ipif, ci.ci_sin, q, mp, ipip, ci.ci_lifr);
12750 
12751 	DTRACE_PROBE4(ipif__ioctl, char *, "ip_process_ioctl finish WR",
12752 	    int, ipip->ipi_cmd,
12753 	    ill_t *, ci.ci_ipif == NULL ? NULL : ci.ci_ipif->ipif_ill,
12754 	    ipif_t *, ci.ci_ipif);
12755 	ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), ipsq);
12756 
12757 	if (entered_ipsq)
12758 		ipsq_exit(ipsq);
12759 }
12760 
12761 /*
12762  * Complete the ioctl. Typically ioctls use the mi package and need to
12763  * do mi_copyout/mi_copy_done.
12764  */
12765 void
ip_ioctl_finish(queue_t * q,mblk_t * mp,int err,int mode,ipsq_t * ipsq)12766 ip_ioctl_finish(queue_t *q, mblk_t *mp, int err, int mode, ipsq_t *ipsq)
12767 {
12768 	conn_t	*connp = NULL;
12769 
12770 	if (err == EINPROGRESS)
12771 		return;
12772 
12773 	if (CONN_Q(q)) {
12774 		connp = Q_TO_CONN(q);
12775 		ASSERT(connp->conn_ref >= 2);
12776 	}
12777 
12778 	switch (mode) {
12779 	case COPYOUT:
12780 		if (err == 0)
12781 			mi_copyout(q, mp);
12782 		else
12783 			mi_copy_done(q, mp, err);
12784 		break;
12785 
12786 	case NO_COPYOUT:
12787 		mi_copy_done(q, mp, err);
12788 		break;
12789 
12790 	default:
12791 		ASSERT(mode == CONN_CLOSE);	/* aborted through CONN_CLOSE */
12792 		break;
12793 	}
12794 
12795 	/*
12796 	 * The conn refhold and ioctlref placed on the conn at the start of the
12797 	 * ioctl are released here.
12798 	 */
12799 	if (connp != NULL) {
12800 		CONN_DEC_IOCTLREF(connp);
12801 		CONN_OPER_PENDING_DONE(connp);
12802 	}
12803 
12804 	if (ipsq != NULL)
12805 		ipsq_current_finish(ipsq);
12806 }
12807 
12808 /* Handles all non data messages */
12809 int
ip_wput_nondata(queue_t * q,mblk_t * mp)12810 ip_wput_nondata(queue_t *q, mblk_t *mp)
12811 {
12812 	mblk_t		*mp1;
12813 	struct iocblk	*iocp;
12814 	ip_ioctl_cmd_t	*ipip;
12815 	conn_t		*connp;
12816 	cred_t		*cr;
12817 	char		*proto_str;
12818 
12819 	if (CONN_Q(q))
12820 		connp = Q_TO_CONN(q);
12821 	else
12822 		connp = NULL;
12823 
12824 	iocp = NULL;
12825 	switch (DB_TYPE(mp)) {
12826 	case M_IOCTL:
12827 		/*
12828 		 * IOCTL processing begins in ip_sioctl_copyin_setup which
12829 		 * will arrange to copy in associated control structures.
12830 		 */
12831 		ip_sioctl_copyin_setup(q, mp);
12832 		return (0);
12833 	case M_IOCDATA:
12834 		/*
12835 		 * Ensure that this is associated with one of our trans-
12836 		 * parent ioctls.  If it's not ours, discard it if we're
12837 		 * running as a driver, or pass it on if we're a module.
12838 		 */
12839 		iocp = (struct iocblk *)mp->b_rptr;
12840 		ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12841 		if (ipip == NULL) {
12842 			if (q->q_next == NULL) {
12843 				goto nak;
12844 			} else {
12845 				putnext(q, mp);
12846 			}
12847 			return (0);
12848 		}
12849 		if ((q->q_next != NULL) && !(ipip->ipi_flags & IPI_MODOK)) {
12850 			/*
12851 			 * The ioctl is one we recognise, but is not consumed
12852 			 * by IP as a module and we are a module, so we drop
12853 			 */
12854 			goto nak;
12855 		}
12856 
12857 		/* IOCTL continuation following copyin or copyout. */
12858 		if (mi_copy_state(q, mp, NULL) == -1) {
12859 			/*
12860 			 * The copy operation failed.  mi_copy_state already
12861 			 * cleaned up, so we're out of here.
12862 			 */
12863 			return (0);
12864 		}
12865 		/*
12866 		 * If we just completed a copy in, we become writer and
12867 		 * continue processing in ip_sioctl_copyin_done.  If it
12868 		 * was a copy out, we call mi_copyout again.  If there is
12869 		 * nothing more to copy out, it will complete the IOCTL.
12870 		 */
12871 		if (MI_COPY_DIRECTION(mp) == MI_COPY_IN) {
12872 			if (!(mp1 = mp->b_cont) || !(mp1 = mp1->b_cont)) {
12873 				mi_copy_done(q, mp, EPROTO);
12874 				return (0);
12875 			}
12876 			/*
12877 			 * Check for cases that need more copying.  A return
12878 			 * value of 0 means a second copyin has been started,
12879 			 * so we return; a return value of 1 means no more
12880 			 * copying is needed, so we continue.
12881 			 */
12882 			if (ipip->ipi_cmd_type == MSFILT_CMD &&
12883 			    MI_COPY_COUNT(mp) == 1) {
12884 				if (ip_copyin_msfilter(q, mp) == 0)
12885 					return (0);
12886 			}
12887 			/*
12888 			 * Refhold the conn, till the ioctl completes. This is
12889 			 * needed in case the ioctl ends up in the pending mp
12890 			 * list. Every mp in the ipx_pending_mp list must have
12891 			 * a refhold on the conn to resume processing. The
12892 			 * refhold is released when the ioctl completes
12893 			 * (whether normally or abnormally). An ioctlref is also
12894 			 * placed on the conn to prevent TCP from removing the
12895 			 * queue needed to send the ioctl reply back.
12896 			 * In all cases ip_ioctl_finish is called to finish
12897 			 * the ioctl and release the refholds.
12898 			 */
12899 			if (connp != NULL) {
12900 				/* This is not a reentry */
12901 				CONN_INC_REF(connp);
12902 				CONN_INC_IOCTLREF(connp);
12903 			} else {
12904 				if (!(ipip->ipi_flags & IPI_MODOK)) {
12905 					mi_copy_done(q, mp, EINVAL);
12906 					return (0);
12907 				}
12908 			}
12909 
12910 			ip_process_ioctl(NULL, q, mp, ipip);
12911 
12912 		} else {
12913 			mi_copyout(q, mp);
12914 		}
12915 		return (0);
12916 
12917 	case M_IOCNAK:
12918 		/*
12919 		 * The only way we could get here is if a resolver didn't like
12920 		 * an IOCTL we sent it.	 This shouldn't happen.
12921 		 */
12922 		(void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
12923 		    "ip_wput_nondata: unexpected M_IOCNAK, ioc_cmd 0x%x",
12924 		    ((struct iocblk *)mp->b_rptr)->ioc_cmd);
12925 		freemsg(mp);
12926 		return (0);
12927 	case M_IOCACK:
12928 		/* /dev/ip shouldn't see this */
12929 		goto nak;
12930 	case M_FLUSH:
12931 		if (*mp->b_rptr & FLUSHW)
12932 			flushq(q, FLUSHALL);
12933 		if (q->q_next) {
12934 			putnext(q, mp);
12935 			return (0);
12936 		}
12937 		if (*mp->b_rptr & FLUSHR) {
12938 			*mp->b_rptr &= ~FLUSHW;
12939 			qreply(q, mp);
12940 			return (0);
12941 		}
12942 		freemsg(mp);
12943 		return (0);
12944 	case M_CTL:
12945 		break;
12946 	case M_PROTO:
12947 	case M_PCPROTO:
12948 		/*
12949 		 * The only PROTO messages we expect are SNMP-related.
12950 		 */
12951 		switch (((union T_primitives *)mp->b_rptr)->type) {
12952 		case T_SVR4_OPTMGMT_REQ:
12953 			ip2dbg(("ip_wput_nondata: T_SVR4_OPTMGMT_REQ "
12954 			    "flags %x\n",
12955 			    ((struct T_optmgmt_req *)mp->b_rptr)->MGMT_flags));
12956 
12957 			if (connp == NULL) {
12958 				proto_str = "T_SVR4_OPTMGMT_REQ";
12959 				goto protonak;
12960 			}
12961 
12962 			/*
12963 			 * All Solaris components should pass a db_credp
12964 			 * for this TPI message, hence we ASSERT.
12965 			 * But in case there is some other M_PROTO that looks
12966 			 * like a TPI message sent by some other kernel
12967 			 * component, we check and return an error.
12968 			 */
12969 			cr = msg_getcred(mp, NULL);
12970 			ASSERT(cr != NULL);
12971 			if (cr == NULL) {
12972 				mp = mi_tpi_err_ack_alloc(mp, TSYSERR, EINVAL);
12973 				if (mp != NULL)
12974 					qreply(q, mp);
12975 				return (0);
12976 			}
12977 
12978 			if (!snmpcom_req(q, mp, ip_snmp_set, ip_snmp_get, cr)) {
12979 				proto_str = "Bad SNMPCOM request?";
12980 				goto protonak;
12981 			}
12982 			return (0);
12983 		default:
12984 			ip1dbg(("ip_wput_nondata: dropping M_PROTO prim %u\n",
12985 			    (int)*(uint_t *)mp->b_rptr));
12986 			freemsg(mp);
12987 			return (0);
12988 		}
12989 	default:
12990 		break;
12991 	}
12992 	if (q->q_next) {
12993 		putnext(q, mp);
12994 	} else
12995 		freemsg(mp);
12996 	return (0);
12997 
12998 nak:
12999 	iocp->ioc_error = EINVAL;
13000 	mp->b_datap->db_type = M_IOCNAK;
13001 	iocp->ioc_count = 0;
13002 	qreply(q, mp);
13003 	return (0);
13004 
13005 protonak:
13006 	cmn_err(CE_NOTE, "IP doesn't process %s as a module", proto_str);
13007 	if ((mp = mi_tpi_err_ack_alloc(mp, TPROTO, EINVAL)) != NULL)
13008 		qreply(q, mp);
13009 	return (0);
13010 }
13011 
13012 /*
13013  * Process IP options in an outbound packet.  Verify that the nexthop in a
13014  * strict source route is onlink.
13015  * Returns non-zero if something fails in which case an ICMP error has been
13016  * sent and mp freed.
13017  *
13018  * Assumes the ULP has called ip_massage_options to move nexthop into ipha_dst.
13019  */
13020 int
ip_output_options(mblk_t * mp,ipha_t * ipha,ip_xmit_attr_t * ixa,ill_t * ill)13021 ip_output_options(mblk_t *mp, ipha_t *ipha, ip_xmit_attr_t *ixa, ill_t *ill)
13022 {
13023 	ipoptp_t	opts;
13024 	uchar_t		*opt;
13025 	uint8_t		optval;
13026 	uint8_t		optlen;
13027 	ipaddr_t	dst;
13028 	intptr_t	code = 0;
13029 	ire_t		*ire;
13030 	ip_stack_t	*ipst = ixa->ixa_ipst;
13031 	ip_recv_attr_t	iras;
13032 
13033 	ip2dbg(("ip_output_options\n"));
13034 
13035 	opt = NULL;
13036 	dst = ipha->ipha_dst;
13037 	for (optval = ipoptp_first(&opts, ipha);
13038 	    optval != IPOPT_EOL;
13039 	    optval = ipoptp_next(&opts)) {
13040 		opt = opts.ipoptp_cur;
13041 		optlen = opts.ipoptp_len;
13042 		ip2dbg(("ip_output_options: opt %d, len %d\n",
13043 		    optval, optlen));
13044 		switch (optval) {
13045 			uint32_t off;
13046 		case IPOPT_SSRR:
13047 		case IPOPT_LSRR:
13048 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
13049 				ip1dbg((
13050 				    "ip_output_options: bad option offset\n"));
13051 				code = (char *)&opt[IPOPT_OLEN] -
13052 				    (char *)ipha;
13053 				goto param_prob;
13054 			}
13055 			off = opt[IPOPT_OFFSET];
13056 			ip1dbg(("ip_output_options: next hop 0x%x\n",
13057 			    ntohl(dst)));
13058 			/*
13059 			 * For strict: verify that dst is directly
13060 			 * reachable.
13061 			 */
13062 			if (optval == IPOPT_SSRR) {
13063 				ire = ire_ftable_lookup_v4(dst, 0, 0,
13064 				    IRE_INTERFACE, NULL, ALL_ZONES,
13065 				    ixa->ixa_tsl,
13066 				    MATCH_IRE_TYPE | MATCH_IRE_SECATTR, 0, ipst,
13067 				    NULL);
13068 				if (ire == NULL) {
13069 					ip1dbg(("ip_output_options: SSRR not"
13070 					    " directly reachable: 0x%x\n",
13071 					    ntohl(dst)));
13072 					goto bad_src_route;
13073 				}
13074 				ire_refrele(ire);
13075 			}
13076 			break;
13077 		case IPOPT_RR:
13078 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
13079 				ip1dbg((
13080 				    "ip_output_options: bad option offset\n"));
13081 				code = (char *)&opt[IPOPT_OLEN] -
13082 				    (char *)ipha;
13083 				goto param_prob;
13084 			}
13085 			break;
13086 		case IPOPT_TS:
13087 			/*
13088 			 * Verify that length >=5 and that there is either
13089 			 * room for another timestamp or that the overflow
13090 			 * counter is not maxed out.
13091 			 */
13092 			code = (char *)&opt[IPOPT_OLEN] - (char *)ipha;
13093 			if (optlen < IPOPT_MINLEN_IT) {
13094 				goto param_prob;
13095 			}
13096 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
13097 				ip1dbg((
13098 				    "ip_output_options: bad option offset\n"));
13099 				code = (char *)&opt[IPOPT_OFFSET] -
13100 				    (char *)ipha;
13101 				goto param_prob;
13102 			}
13103 			switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
13104 			case IPOPT_TS_TSONLY:
13105 				off = IPOPT_TS_TIMELEN;
13106 				break;
13107 			case IPOPT_TS_TSANDADDR:
13108 			case IPOPT_TS_PRESPEC:
13109 			case IPOPT_TS_PRESPEC_RFC791:
13110 				off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
13111 				break;
13112 			default:
13113 				code = (char *)&opt[IPOPT_POS_OV_FLG] -
13114 				    (char *)ipha;
13115 				goto param_prob;
13116 			}
13117 			if (opt[IPOPT_OFFSET] - 1 + off > optlen &&
13118 			    (opt[IPOPT_POS_OV_FLG] & 0xF0) == 0xF0) {
13119 				/*
13120 				 * No room and the overflow counter is 15
13121 				 * already.
13122 				 */
13123 				goto param_prob;
13124 			}
13125 			break;
13126 		}
13127 	}
13128 
13129 	if ((opts.ipoptp_flags & IPOPTP_ERROR) == 0)
13130 		return (0);
13131 
13132 	ip1dbg(("ip_output_options: error processing IP options."));
13133 	code = (char *)&opt[IPOPT_OFFSET] - (char *)ipha;
13134 
13135 param_prob:
13136 	bzero(&iras, sizeof (iras));
13137 	iras.ira_ill = iras.ira_rill = ill;
13138 	iras.ira_ruifindex = ill->ill_phyint->phyint_ifindex;
13139 	iras.ira_rifindex = iras.ira_ruifindex;
13140 	iras.ira_flags = IRAF_IS_IPV4;
13141 
13142 	ip_drop_output("ip_output_options", mp, ill);
13143 	icmp_param_problem(mp, (uint8_t)code, &iras);
13144 	ASSERT(!(iras.ira_flags & IRAF_IPSEC_SECURE));
13145 	return (-1);
13146 
13147 bad_src_route:
13148 	bzero(&iras, sizeof (iras));
13149 	iras.ira_ill = iras.ira_rill = ill;
13150 	iras.ira_ruifindex = ill->ill_phyint->phyint_ifindex;
13151 	iras.ira_rifindex = iras.ira_ruifindex;
13152 	iras.ira_flags = IRAF_IS_IPV4;
13153 
13154 	ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ill);
13155 	icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, &iras);
13156 	ASSERT(!(iras.ira_flags & IRAF_IPSEC_SECURE));
13157 	return (-1);
13158 }
13159 
13160 /*
13161  * The maximum value of conn_drain_list_cnt is CONN_MAXDRAINCNT.
13162  * conn_drain_list_cnt can be changed by setting conn_drain_nthreads
13163  * thru /etc/system.
13164  */
13165 #define	CONN_MAXDRAINCNT	64
13166 
13167 static void
conn_drain_init(ip_stack_t * ipst)13168 conn_drain_init(ip_stack_t *ipst)
13169 {
13170 	int i, j;
13171 	idl_tx_list_t *itl_tx;
13172 
13173 	ipst->ips_conn_drain_list_cnt = conn_drain_nthreads;
13174 
13175 	if ((ipst->ips_conn_drain_list_cnt == 0) ||
13176 	    (ipst->ips_conn_drain_list_cnt > CONN_MAXDRAINCNT)) {
13177 		/*
13178 		 * Default value of the number of drainers is the
13179 		 * number of cpus, subject to maximum of 8 drainers.
13180 		 */
13181 		if (boot_max_ncpus != -1)
13182 			ipst->ips_conn_drain_list_cnt = MIN(boot_max_ncpus, 8);
13183 		else
13184 			ipst->ips_conn_drain_list_cnt = MIN(max_ncpus, 8);
13185 	}
13186 
13187 	ipst->ips_idl_tx_list =
13188 	    kmem_zalloc(TX_FANOUT_SIZE * sizeof (idl_tx_list_t), KM_SLEEP);
13189 	for (i = 0; i < TX_FANOUT_SIZE; i++) {
13190 		itl_tx =  &ipst->ips_idl_tx_list[i];
13191 		itl_tx->txl_drain_list =
13192 		    kmem_zalloc(ipst->ips_conn_drain_list_cnt *
13193 		    sizeof (idl_t), KM_SLEEP);
13194 		mutex_init(&itl_tx->txl_lock, NULL, MUTEX_DEFAULT, NULL);
13195 		for (j = 0; j < ipst->ips_conn_drain_list_cnt; j++) {
13196 			mutex_init(&itl_tx->txl_drain_list[j].idl_lock, NULL,
13197 			    MUTEX_DEFAULT, NULL);
13198 			itl_tx->txl_drain_list[j].idl_itl = itl_tx;
13199 		}
13200 	}
13201 }
13202 
13203 static void
conn_drain_fini(ip_stack_t * ipst)13204 conn_drain_fini(ip_stack_t *ipst)
13205 {
13206 	int i;
13207 	idl_tx_list_t *itl_tx;
13208 
13209 	for (i = 0; i < TX_FANOUT_SIZE; i++) {
13210 		itl_tx =  &ipst->ips_idl_tx_list[i];
13211 		kmem_free(itl_tx->txl_drain_list,
13212 		    ipst->ips_conn_drain_list_cnt * sizeof (idl_t));
13213 	}
13214 	kmem_free(ipst->ips_idl_tx_list,
13215 	    TX_FANOUT_SIZE * sizeof (idl_tx_list_t));
13216 	ipst->ips_idl_tx_list = NULL;
13217 }
13218 
13219 /*
13220  * Flow control has blocked us from proceeding.  Insert the given conn in one
13221  * of the conn drain lists.  When flow control is unblocked, either ip_wsrv()
13222  * (STREAMS) or ill_flow_enable() (direct) will be called back, which in turn
13223  * will call conn_walk_drain().  See the flow control notes at the top of this
13224  * file for more details.
13225  */
13226 void
conn_drain_insert(conn_t * connp,idl_tx_list_t * tx_list)13227 conn_drain_insert(conn_t *connp, idl_tx_list_t *tx_list)
13228 {
13229 	idl_t	*idl = tx_list->txl_drain_list;
13230 	uint_t	index;
13231 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
13232 
13233 	mutex_enter(&connp->conn_lock);
13234 	if (connp->conn_state_flags & CONN_CLOSING) {
13235 		/*
13236 		 * The conn is closing as a result of which CONN_CLOSING
13237 		 * is set. Return.
13238 		 */
13239 		mutex_exit(&connp->conn_lock);
13240 		return;
13241 	} else if (connp->conn_idl == NULL) {
13242 		/*
13243 		 * Assign the next drain list round robin. We dont' use
13244 		 * a lock, and thus it may not be strictly round robin.
13245 		 * Atomicity of load/stores is enough to make sure that
13246 		 * conn_drain_list_index is always within bounds.
13247 		 */
13248 		index = tx_list->txl_drain_index;
13249 		ASSERT(index < ipst->ips_conn_drain_list_cnt);
13250 		connp->conn_idl = &tx_list->txl_drain_list[index];
13251 		index++;
13252 		if (index == ipst->ips_conn_drain_list_cnt)
13253 			index = 0;
13254 		tx_list->txl_drain_index = index;
13255 	} else {
13256 		ASSERT(connp->conn_idl->idl_itl == tx_list);
13257 	}
13258 	mutex_exit(&connp->conn_lock);
13259 
13260 	idl = connp->conn_idl;
13261 	mutex_enter(&idl->idl_lock);
13262 	if ((connp->conn_drain_prev != NULL) ||
13263 	    (connp->conn_state_flags & CONN_CLOSING)) {
13264 		/*
13265 		 * The conn is either already in the drain list or closing.
13266 		 * (We needed to check for CONN_CLOSING again since close can
13267 		 * sneak in between dropping conn_lock and acquiring idl_lock.)
13268 		 */
13269 		mutex_exit(&idl->idl_lock);
13270 		return;
13271 	}
13272 
13273 	/*
13274 	 * The conn is not in the drain list. Insert it at the
13275 	 * tail of the drain list. The drain list is circular
13276 	 * and doubly linked. idl_conn points to the 1st element
13277 	 * in the list.
13278 	 */
13279 	if (idl->idl_conn == NULL) {
13280 		idl->idl_conn = connp;
13281 		connp->conn_drain_next = connp;
13282 		connp->conn_drain_prev = connp;
13283 	} else {
13284 		conn_t *head = idl->idl_conn;
13285 
13286 		connp->conn_drain_next = head;
13287 		connp->conn_drain_prev = head->conn_drain_prev;
13288 		head->conn_drain_prev->conn_drain_next = connp;
13289 		head->conn_drain_prev = connp;
13290 	}
13291 	/*
13292 	 * For non streams based sockets assert flow control.
13293 	 */
13294 	conn_setqfull(connp, NULL);
13295 	mutex_exit(&idl->idl_lock);
13296 }
13297 
13298 static void
conn_drain_remove(conn_t * connp)13299 conn_drain_remove(conn_t *connp)
13300 {
13301 	idl_t *idl = connp->conn_idl;
13302 
13303 	if (idl != NULL) {
13304 		/*
13305 		 * Remove ourself from the drain list.
13306 		 */
13307 		if (connp->conn_drain_next == connp) {
13308 			/* Singleton in the list */
13309 			ASSERT(connp->conn_drain_prev == connp);
13310 			idl->idl_conn = NULL;
13311 		} else {
13312 			connp->conn_drain_prev->conn_drain_next =
13313 			    connp->conn_drain_next;
13314 			connp->conn_drain_next->conn_drain_prev =
13315 			    connp->conn_drain_prev;
13316 			if (idl->idl_conn == connp)
13317 				idl->idl_conn = connp->conn_drain_next;
13318 		}
13319 
13320 		/*
13321 		 * NOTE: because conn_idl is associated with a specific drain
13322 		 * list which in turn is tied to the index the TX ring
13323 		 * (txl_cookie) hashes to, and because the TX ring can change
13324 		 * over the lifetime of the conn_t, we must clear conn_idl so
13325 		 * a subsequent conn_drain_insert() will set conn_idl again
13326 		 * based on the latest txl_cookie.
13327 		 */
13328 		connp->conn_idl = NULL;
13329 	}
13330 	connp->conn_drain_next = NULL;
13331 	connp->conn_drain_prev = NULL;
13332 
13333 	conn_clrqfull(connp, NULL);
13334 	/*
13335 	 * For streams based sockets open up flow control.
13336 	 */
13337 	if (!IPCL_IS_NONSTR(connp))
13338 		enableok(connp->conn_wq);
13339 }
13340 
13341 /*
13342  * This conn is closing, and we are called from ip_close. OR
13343  * this conn is draining because flow-control on the ill has been relieved.
13344  *
13345  * We must also need to remove conn's on this idl from the list, and also
13346  * inform the sockfs upcalls about the change in flow-control.
13347  */
13348 static void
conn_drain(conn_t * connp,boolean_t closing)13349 conn_drain(conn_t *connp, boolean_t closing)
13350 {
13351 	idl_t *idl;
13352 	conn_t *next_connp;
13353 
13354 	/*
13355 	 * connp->conn_idl is stable at this point, and no lock is needed
13356 	 * to check it. If we are called from ip_close, close has already
13357 	 * set CONN_CLOSING, thus freezing the value of conn_idl, and
13358 	 * called us only because conn_idl is non-null. If we are called thru
13359 	 * service, conn_idl could be null, but it cannot change because
13360 	 * service is single-threaded per queue, and there cannot be another
13361 	 * instance of service trying to call conn_drain_insert on this conn
13362 	 * now.
13363 	 */
13364 	ASSERT(!closing || connp == NULL || connp->conn_idl != NULL);
13365 
13366 	/*
13367 	 * If the conn doesn't exist or is not on a drain list, bail.
13368 	 */
13369 	if (connp == NULL || connp->conn_idl == NULL ||
13370 	    connp->conn_drain_prev == NULL) {
13371 		return;
13372 	}
13373 
13374 	idl = connp->conn_idl;
13375 	ASSERT(MUTEX_HELD(&idl->idl_lock));
13376 
13377 	if (!closing) {
13378 		next_connp = connp->conn_drain_next;
13379 		while (next_connp != connp) {
13380 			conn_t *delconnp = next_connp;
13381 
13382 			next_connp = next_connp->conn_drain_next;
13383 			conn_drain_remove(delconnp);
13384 		}
13385 		ASSERT(connp->conn_drain_next == idl->idl_conn);
13386 	}
13387 	conn_drain_remove(connp);
13388 }
13389 
13390 /*
13391  * Write service routine. Shared perimeter entry point.
13392  * The device queue's messages has fallen below the low water mark and STREAMS
13393  * has backenabled the ill_wq. Send sockfs notification about flow-control on
13394  * each waiting conn.
13395  */
13396 int
ip_wsrv(queue_t * q)13397 ip_wsrv(queue_t *q)
13398 {
13399 	ill_t	*ill;
13400 
13401 	ill = (ill_t *)q->q_ptr;
13402 	if (ill->ill_state_flags == 0) {
13403 		ip_stack_t *ipst = ill->ill_ipst;
13404 
13405 		/*
13406 		 * The device flow control has opened up.
13407 		 * Walk through conn drain lists and qenable the
13408 		 * first conn in each list. This makes sense only
13409 		 * if the stream is fully plumbed and setup.
13410 		 * Hence the ill_state_flags check above.
13411 		 */
13412 		ip1dbg(("ip_wsrv: walking\n"));
13413 		conn_walk_drain(ipst, &ipst->ips_idl_tx_list[0]);
13414 		enableok(ill->ill_wq);
13415 	}
13416 	return (0);
13417 }
13418 
13419 /*
13420  * Callback to disable flow control in IP.
13421  *
13422  * This is a mac client callback added when the DLD_CAPAB_DIRECT capability
13423  * is enabled.
13424  *
13425  * When MAC_TX() is not able to send any more packets, dld sets its queue
13426  * to QFULL and enable the STREAMS flow control. Later, when the underlying
13427  * driver is able to continue to send packets, it calls mac_tx_(ring_)update()
13428  * function and wakes up corresponding mac worker threads, which in turn
13429  * calls this callback function, and disables flow control.
13430  */
13431 void
ill_flow_enable(void * arg,ip_mac_tx_cookie_t cookie)13432 ill_flow_enable(void *arg, ip_mac_tx_cookie_t cookie)
13433 {
13434 	ill_t *ill = (ill_t *)arg;
13435 	ip_stack_t *ipst = ill->ill_ipst;
13436 	idl_tx_list_t *idl_txl;
13437 
13438 	idl_txl = &ipst->ips_idl_tx_list[IDLHASHINDEX(cookie)];
13439 	mutex_enter(&idl_txl->txl_lock);
13440 	/* add code to to set a flag to indicate idl_txl is enabled */
13441 	conn_walk_drain(ipst, idl_txl);
13442 	mutex_exit(&idl_txl->txl_lock);
13443 }
13444 
13445 /*
13446  * Flow control has been relieved and STREAMS has backenabled us; drain
13447  * all the conn lists on `tx_list'.
13448  */
13449 static void
conn_walk_drain(ip_stack_t * ipst,idl_tx_list_t * tx_list)13450 conn_walk_drain(ip_stack_t *ipst, idl_tx_list_t *tx_list)
13451 {
13452 	int i;
13453 	idl_t *idl;
13454 
13455 	IP_STAT(ipst, ip_conn_walk_drain);
13456 
13457 	for (i = 0; i < ipst->ips_conn_drain_list_cnt; i++) {
13458 		idl = &tx_list->txl_drain_list[i];
13459 		mutex_enter(&idl->idl_lock);
13460 		conn_drain(idl->idl_conn, B_FALSE);
13461 		mutex_exit(&idl->idl_lock);
13462 	}
13463 }
13464 
13465 /*
13466  * Determine if the ill and multicast aspects of that packets
13467  * "matches" the conn.
13468  */
13469 boolean_t
conn_wantpacket(conn_t * connp,ip_recv_attr_t * ira,ipha_t * ipha)13470 conn_wantpacket(conn_t *connp, ip_recv_attr_t *ira, ipha_t *ipha)
13471 {
13472 	ill_t		*ill = ira->ira_rill;
13473 	zoneid_t	zoneid = ira->ira_zoneid;
13474 	uint_t		in_ifindex;
13475 	ipaddr_t	dst, src;
13476 
13477 	dst = ipha->ipha_dst;
13478 	src = ipha->ipha_src;
13479 
13480 	/*
13481 	 * conn_incoming_ifindex is set by IP_BOUND_IF which limits
13482 	 * unicast, broadcast and multicast reception to
13483 	 * conn_incoming_ifindex.
13484 	 * conn_wantpacket is called for unicast, broadcast and
13485 	 * multicast packets.
13486 	 */
13487 	in_ifindex = connp->conn_incoming_ifindex;
13488 
13489 	/* mpathd can bind to the under IPMP interface, which we allow */
13490 	if (in_ifindex != 0 && in_ifindex != ill->ill_phyint->phyint_ifindex) {
13491 		if (!IS_UNDER_IPMP(ill))
13492 			return (B_FALSE);
13493 
13494 		if (in_ifindex != ipmp_ill_get_ipmp_ifindex(ill))
13495 			return (B_FALSE);
13496 	}
13497 
13498 	if (!IPCL_ZONE_MATCH(connp, zoneid))
13499 		return (B_FALSE);
13500 
13501 	if (!(ira->ira_flags & IRAF_MULTICAST))
13502 		return (B_TRUE);
13503 
13504 	if (connp->conn_multi_router) {
13505 		/* multicast packet and multicast router socket: send up */
13506 		return (B_TRUE);
13507 	}
13508 
13509 	if (ipha->ipha_protocol == IPPROTO_PIM ||
13510 	    ipha->ipha_protocol == IPPROTO_RSVP)
13511 		return (B_TRUE);
13512 
13513 	return (conn_hasmembers_ill_withsrc_v4(connp, dst, src, ira->ira_ill));
13514 }
13515 
13516 void
conn_setqfull(conn_t * connp,boolean_t * flow_stopped)13517 conn_setqfull(conn_t *connp, boolean_t *flow_stopped)
13518 {
13519 	if (IPCL_IS_NONSTR(connp)) {
13520 		(*connp->conn_upcalls->su_txq_full)
13521 		    (connp->conn_upper_handle, B_TRUE);
13522 		if (flow_stopped != NULL)
13523 			*flow_stopped = B_TRUE;
13524 	} else {
13525 		queue_t *q = connp->conn_wq;
13526 
13527 		ASSERT(q != NULL);
13528 		if (!(q->q_flag & QFULL)) {
13529 			mutex_enter(QLOCK(q));
13530 			if (!(q->q_flag & QFULL)) {
13531 				/* still need to set QFULL */
13532 				q->q_flag |= QFULL;
13533 				/* set flow_stopped to true under QLOCK */
13534 				if (flow_stopped != NULL)
13535 					*flow_stopped = B_TRUE;
13536 				mutex_exit(QLOCK(q));
13537 			} else {
13538 				/* flow_stopped is left unchanged */
13539 				mutex_exit(QLOCK(q));
13540 			}
13541 		}
13542 	}
13543 }
13544 
13545 void
conn_clrqfull(conn_t * connp,boolean_t * flow_stopped)13546 conn_clrqfull(conn_t *connp, boolean_t *flow_stopped)
13547 {
13548 	if (IPCL_IS_NONSTR(connp)) {
13549 		(*connp->conn_upcalls->su_txq_full)
13550 		    (connp->conn_upper_handle, B_FALSE);
13551 		if (flow_stopped != NULL)
13552 			*flow_stopped = B_FALSE;
13553 	} else {
13554 		queue_t *q = connp->conn_wq;
13555 
13556 		ASSERT(q != NULL);
13557 		if (q->q_flag & QFULL) {
13558 			mutex_enter(QLOCK(q));
13559 			if (q->q_flag & QFULL) {
13560 				q->q_flag &= ~QFULL;
13561 				/* set flow_stopped to false under QLOCK */
13562 				if (flow_stopped != NULL)
13563 					*flow_stopped = B_FALSE;
13564 				mutex_exit(QLOCK(q));
13565 				if (q->q_flag & QWANTW)
13566 					qbackenable(q, 0);
13567 			} else {
13568 				/* flow_stopped is left unchanged */
13569 				mutex_exit(QLOCK(q));
13570 			}
13571 		}
13572 	}
13573 
13574 	mutex_enter(&connp->conn_lock);
13575 	connp->conn_blocked = B_FALSE;
13576 	mutex_exit(&connp->conn_lock);
13577 }
13578 
13579 /*
13580  * Return the length in bytes of the IPv4 headers (base header, label, and
13581  * other IP options) that will be needed based on the
13582  * ip_pkt_t structure passed by the caller.
13583  *
13584  * The returned length does not include the length of the upper level
13585  * protocol (ULP) header.
13586  * The caller needs to check that the length doesn't exceed the max for IPv4.
13587  */
13588 int
ip_total_hdrs_len_v4(const ip_pkt_t * ipp)13589 ip_total_hdrs_len_v4(const ip_pkt_t *ipp)
13590 {
13591 	int len;
13592 
13593 	len = IP_SIMPLE_HDR_LENGTH;
13594 	if (ipp->ipp_fields & IPPF_LABEL_V4) {
13595 		ASSERT(ipp->ipp_label_len_v4 != 0);
13596 		/* We need to round up here */
13597 		len += (ipp->ipp_label_len_v4 + 3) & ~3;
13598 	}
13599 
13600 	if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
13601 		ASSERT(ipp->ipp_ipv4_options_len != 0);
13602 		ASSERT((ipp->ipp_ipv4_options_len & 3) == 0);
13603 		len += ipp->ipp_ipv4_options_len;
13604 	}
13605 	return (len);
13606 }
13607 
13608 /*
13609  * All-purpose routine to build an IPv4 header with options based
13610  * on the abstract ip_pkt_t.
13611  *
13612  * The caller has to set the source and destination address as well as
13613  * ipha_length. The caller has to massage any source route and compensate
13614  * for the ULP pseudo-header checksum due to the source route.
13615  */
13616 void
ip_build_hdrs_v4(uchar_t * buf,uint_t buf_len,const ip_pkt_t * ipp,uint8_t protocol)13617 ip_build_hdrs_v4(uchar_t *buf, uint_t buf_len, const ip_pkt_t *ipp,
13618     uint8_t protocol)
13619 {
13620 	ipha_t	*ipha = (ipha_t *)buf;
13621 	uint8_t *cp;
13622 
13623 	/* Initialize IPv4 header */
13624 	ipha->ipha_type_of_service = ipp->ipp_type_of_service;
13625 	ipha->ipha_length = 0;	/* Caller will set later */
13626 	ipha->ipha_ident = 0;
13627 	ipha->ipha_fragment_offset_and_flags = 0;
13628 	ipha->ipha_ttl = ipp->ipp_unicast_hops;
13629 	ipha->ipha_protocol = protocol;
13630 	ipha->ipha_hdr_checksum = 0;
13631 
13632 	if ((ipp->ipp_fields & IPPF_ADDR) &&
13633 	    IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
13634 		ipha->ipha_src = ipp->ipp_addr_v4;
13635 
13636 	cp = (uint8_t *)&ipha[1];
13637 	if (ipp->ipp_fields & IPPF_LABEL_V4) {
13638 		ASSERT(ipp->ipp_label_len_v4 != 0);
13639 		bcopy(ipp->ipp_label_v4, cp, ipp->ipp_label_len_v4);
13640 		cp += ipp->ipp_label_len_v4;
13641 		/* We need to round up here */
13642 		while ((uintptr_t)cp & 0x3) {
13643 			*cp++ = IPOPT_NOP;
13644 		}
13645 	}
13646 
13647 	if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
13648 		ASSERT(ipp->ipp_ipv4_options_len != 0);
13649 		ASSERT((ipp->ipp_ipv4_options_len & 3) == 0);
13650 		bcopy(ipp->ipp_ipv4_options, cp, ipp->ipp_ipv4_options_len);
13651 		cp += ipp->ipp_ipv4_options_len;
13652 	}
13653 	ipha->ipha_version_and_hdr_length =
13654 	    (uint8_t)((IP_VERSION << 4) + buf_len / 4);
13655 
13656 	ASSERT((int)(cp - buf) == buf_len);
13657 }
13658 
13659 /* Allocate the private structure */
13660 static int
ip_priv_alloc(void ** bufp)13661 ip_priv_alloc(void **bufp)
13662 {
13663 	void	*buf;
13664 
13665 	if ((buf = kmem_alloc(sizeof (ip_priv_t), KM_NOSLEEP)) == NULL)
13666 		return (ENOMEM);
13667 
13668 	*bufp = buf;
13669 	return (0);
13670 }
13671 
13672 /* Function to delete the private structure */
13673 void
ip_priv_free(void * buf)13674 ip_priv_free(void *buf)
13675 {
13676 	ASSERT(buf != NULL);
13677 	kmem_free(buf, sizeof (ip_priv_t));
13678 }
13679 
13680 /*
13681  * The entry point for IPPF processing.
13682  * If the classifier (IPGPC_CLASSIFY) is not loaded and configured, the
13683  * routine just returns.
13684  *
13685  * When called, ip_process generates an ipp_packet_t structure
13686  * which holds the state information for this packet and invokes the
13687  * the classifier (via ipp_packet_process). The classification, depending on
13688  * configured filters, results in a list of actions for this packet. Invoking
13689  * an action may cause the packet to be dropped, in which case we return NULL.
13690  * proc indicates the callout position for
13691  * this packet and ill is the interface this packet arrived on or will leave
13692  * on (inbound and outbound resp.).
13693  *
13694  * We do the processing on the rill (mapped to the upper if ipmp), but MIB
13695  * on the ill corrsponding to the destination IP address.
13696  */
13697 mblk_t *
ip_process(ip_proc_t proc,mblk_t * mp,ill_t * rill,ill_t * ill)13698 ip_process(ip_proc_t proc, mblk_t *mp, ill_t *rill, ill_t *ill)
13699 {
13700 	ip_priv_t	*priv;
13701 	ipp_action_id_t	aid;
13702 	int		rc = 0;
13703 	ipp_packet_t	*pp;
13704 
13705 	/* If the classifier is not loaded, return  */
13706 	if ((aid = ipp_action_lookup(IPGPC_CLASSIFY)) == IPP_ACTION_INVAL) {
13707 		return (mp);
13708 	}
13709 
13710 	ASSERT(mp != NULL);
13711 
13712 	/* Allocate the packet structure */
13713 	rc = ipp_packet_alloc(&pp, "ip", aid);
13714 	if (rc != 0)
13715 		goto drop;
13716 
13717 	/* Allocate the private structure */
13718 	rc = ip_priv_alloc((void **)&priv);
13719 	if (rc != 0) {
13720 		ipp_packet_free(pp);
13721 		goto drop;
13722 	}
13723 	priv->proc = proc;
13724 	priv->ill_index = ill_get_upper_ifindex(rill);
13725 
13726 	ipp_packet_set_private(pp, priv, ip_priv_free);
13727 	ipp_packet_set_data(pp, mp);
13728 
13729 	/* Invoke the classifier */
13730 	rc = ipp_packet_process(&pp);
13731 	if (pp != NULL) {
13732 		mp = ipp_packet_get_data(pp);
13733 		ipp_packet_free(pp);
13734 		if (rc != 0)
13735 			goto drop;
13736 		return (mp);
13737 	} else {
13738 		/* No mp to trace in ip_drop_input/ip_drop_output  */
13739 		mp = NULL;
13740 	}
13741 drop:
13742 	if (proc == IPP_LOCAL_IN || proc == IPP_FWD_IN) {
13743 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
13744 		ip_drop_input("ip_process", mp, ill);
13745 	} else {
13746 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
13747 		ip_drop_output("ip_process", mp, ill);
13748 	}
13749 	freemsg(mp);
13750 	return (NULL);
13751 }
13752 
13753 /*
13754  * Propagate a multicast group membership operation (add/drop) on
13755  * all the interfaces crossed by the related multirt routes.
13756  * The call is considered successful if the operation succeeds
13757  * on at least one interface.
13758  *
13759  * This assumes that a set of IRE_HOST/RTF_MULTIRT has been created for the
13760  * multicast addresses with the ire argument being the first one.
13761  * We walk the bucket to find all the of those.
13762  *
13763  * Common to IPv4 and IPv6.
13764  */
13765 static int
ip_multirt_apply_membership(int (* fn)(conn_t *,boolean_t,const in6_addr_t *,ipaddr_t,uint_t,mcast_record_t,const in6_addr_t *),ire_t * ire,conn_t * connp,boolean_t checkonly,const in6_addr_t * v6group,mcast_record_t fmode,const in6_addr_t * v6src)13766 ip_multirt_apply_membership(int (*fn)(conn_t *, boolean_t,
13767     const in6_addr_t *, ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *),
13768     ire_t *ire, conn_t *connp, boolean_t checkonly, const in6_addr_t *v6group,
13769     mcast_record_t fmode, const in6_addr_t *v6src)
13770 {
13771 	ire_t		*ire_gw;
13772 	irb_t		*irb;
13773 	int		ifindex;
13774 	int		error = 0;
13775 	int		result;
13776 	ip_stack_t	*ipst = ire->ire_ipst;
13777 	ipaddr_t	group;
13778 	boolean_t	isv6;
13779 	int		match_flags;
13780 
13781 	if (IN6_IS_ADDR_V4MAPPED(v6group)) {
13782 		IN6_V4MAPPED_TO_IPADDR(v6group, group);
13783 		isv6 = B_FALSE;
13784 	} else {
13785 		isv6 = B_TRUE;
13786 	}
13787 
13788 	irb = ire->ire_bucket;
13789 	ASSERT(irb != NULL);
13790 
13791 	result = 0;
13792 	irb_refhold(irb);
13793 	for (; ire != NULL; ire = ire->ire_next) {
13794 		if ((ire->ire_flags & RTF_MULTIRT) == 0)
13795 			continue;
13796 
13797 		/* We handle -ifp routes by matching on the ill if set */
13798 		match_flags = MATCH_IRE_TYPE;
13799 		if (ire->ire_ill != NULL)
13800 			match_flags |= MATCH_IRE_ILL;
13801 
13802 		if (isv6) {
13803 			if (!IN6_ARE_ADDR_EQUAL(&ire->ire_addr_v6, v6group))
13804 				continue;
13805 
13806 			ire_gw = ire_ftable_lookup_v6(&ire->ire_gateway_addr_v6,
13807 			    0, 0, IRE_INTERFACE, ire->ire_ill, ALL_ZONES, NULL,
13808 			    match_flags, 0, ipst, NULL);
13809 		} else {
13810 			if (ire->ire_addr != group)
13811 				continue;
13812 
13813 			ire_gw = ire_ftable_lookup_v4(ire->ire_gateway_addr,
13814 			    0, 0, IRE_INTERFACE, ire->ire_ill, ALL_ZONES, NULL,
13815 			    match_flags, 0, ipst, NULL);
13816 		}
13817 		/* No interface route exists for the gateway; skip this ire. */
13818 		if (ire_gw == NULL)
13819 			continue;
13820 		if (ire_gw->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
13821 			ire_refrele(ire_gw);
13822 			continue;
13823 		}
13824 		ASSERT(ire_gw->ire_ill != NULL);	/* IRE_INTERFACE */
13825 		ifindex = ire_gw->ire_ill->ill_phyint->phyint_ifindex;
13826 
13827 		/*
13828 		 * The operation is considered a success if
13829 		 * it succeeds at least once on any one interface.
13830 		 */
13831 		error = fn(connp, checkonly, v6group, INADDR_ANY, ifindex,
13832 		    fmode, v6src);
13833 		if (error == 0)
13834 			result = CGTP_MCAST_SUCCESS;
13835 
13836 		ire_refrele(ire_gw);
13837 	}
13838 	irb_refrele(irb);
13839 	/*
13840 	 * Consider the call as successful if we succeeded on at least
13841 	 * one interface. Otherwise, return the last encountered error.
13842 	 */
13843 	return (result == CGTP_MCAST_SUCCESS ? 0 : error);
13844 }
13845 
13846 /*
13847  * Return the expected CGTP hooks version number.
13848  */
13849 int
ip_cgtp_filter_supported(void)13850 ip_cgtp_filter_supported(void)
13851 {
13852 	return (ip_cgtp_filter_rev);
13853 }
13854 
13855 /*
13856  * CGTP hooks can be registered by invoking this function.
13857  * Checks that the version number matches.
13858  */
13859 int
ip_cgtp_filter_register(netstackid_t stackid,cgtp_filter_ops_t * ops)13860 ip_cgtp_filter_register(netstackid_t stackid, cgtp_filter_ops_t *ops)
13861 {
13862 	netstack_t *ns;
13863 	ip_stack_t *ipst;
13864 
13865 	if (ops->cfo_filter_rev != CGTP_FILTER_REV)
13866 		return (ENOTSUP);
13867 
13868 	ns = netstack_find_by_stackid(stackid);
13869 	if (ns == NULL)
13870 		return (EINVAL);
13871 	ipst = ns->netstack_ip;
13872 	ASSERT(ipst != NULL);
13873 
13874 	if (ipst->ips_ip_cgtp_filter_ops != NULL) {
13875 		netstack_rele(ns);
13876 		return (EALREADY);
13877 	}
13878 
13879 	ipst->ips_ip_cgtp_filter_ops = ops;
13880 
13881 	ill_set_inputfn_all(ipst);
13882 
13883 	netstack_rele(ns);
13884 	return (0);
13885 }
13886 
13887 /*
13888  * CGTP hooks can be unregistered by invoking this function.
13889  * Returns ENXIO if there was no registration.
13890  * Returns EBUSY if the ndd variable has not been turned off.
13891  */
13892 int
ip_cgtp_filter_unregister(netstackid_t stackid)13893 ip_cgtp_filter_unregister(netstackid_t stackid)
13894 {
13895 	netstack_t *ns;
13896 	ip_stack_t *ipst;
13897 
13898 	ns = netstack_find_by_stackid(stackid);
13899 	if (ns == NULL)
13900 		return (EINVAL);
13901 	ipst = ns->netstack_ip;
13902 	ASSERT(ipst != NULL);
13903 
13904 	if (ipst->ips_ip_cgtp_filter) {
13905 		netstack_rele(ns);
13906 		return (EBUSY);
13907 	}
13908 
13909 	if (ipst->ips_ip_cgtp_filter_ops == NULL) {
13910 		netstack_rele(ns);
13911 		return (ENXIO);
13912 	}
13913 	ipst->ips_ip_cgtp_filter_ops = NULL;
13914 
13915 	ill_set_inputfn_all(ipst);
13916 
13917 	netstack_rele(ns);
13918 	return (0);
13919 }
13920 
13921 /*
13922  * Check whether there is a CGTP filter registration.
13923  * Returns non-zero if there is a registration, otherwise returns zero.
13924  * Note: returns zero if bad stackid.
13925  */
13926 int
ip_cgtp_filter_is_registered(netstackid_t stackid)13927 ip_cgtp_filter_is_registered(netstackid_t stackid)
13928 {
13929 	netstack_t *ns;
13930 	ip_stack_t *ipst;
13931 	int ret;
13932 
13933 	ns = netstack_find_by_stackid(stackid);
13934 	if (ns == NULL)
13935 		return (0);
13936 	ipst = ns->netstack_ip;
13937 	ASSERT(ipst != NULL);
13938 
13939 	if (ipst->ips_ip_cgtp_filter_ops != NULL)
13940 		ret = 1;
13941 	else
13942 		ret = 0;
13943 
13944 	netstack_rele(ns);
13945 	return (ret);
13946 }
13947 
13948 static int
ip_squeue_switch(int val)13949 ip_squeue_switch(int val)
13950 {
13951 	int rval;
13952 
13953 	switch (val) {
13954 	case IP_SQUEUE_ENTER_NODRAIN:
13955 		rval = SQ_NODRAIN;
13956 		break;
13957 	case IP_SQUEUE_ENTER:
13958 		rval = SQ_PROCESS;
13959 		break;
13960 	case IP_SQUEUE_FILL:
13961 	default:
13962 		rval = SQ_FILL;
13963 		break;
13964 	}
13965 	return (rval);
13966 }
13967 
13968 static void *
ip_kstat2_init(netstackid_t stackid,ip_stat_t * ip_statisticsp)13969 ip_kstat2_init(netstackid_t stackid, ip_stat_t *ip_statisticsp)
13970 {
13971 	kstat_t *ksp;
13972 
13973 	ip_stat_t template = {
13974 		{ "ip_udp_fannorm",		KSTAT_DATA_UINT64 },
13975 		{ "ip_udp_fanmb",		KSTAT_DATA_UINT64 },
13976 		{ "ip_recv_pullup",		KSTAT_DATA_UINT64 },
13977 		{ "ip_db_ref",			KSTAT_DATA_UINT64 },
13978 		{ "ip_notaligned",		KSTAT_DATA_UINT64 },
13979 		{ "ip_multimblk",		KSTAT_DATA_UINT64 },
13980 		{ "ip_opt",			KSTAT_DATA_UINT64 },
13981 		{ "ipsec_proto_ahesp",		KSTAT_DATA_UINT64 },
13982 		{ "ip_conn_flputbq",		KSTAT_DATA_UINT64 },
13983 		{ "ip_conn_walk_drain",		KSTAT_DATA_UINT64 },
13984 		{ "ip_out_sw_cksum",		KSTAT_DATA_UINT64 },
13985 		{ "ip_out_sw_cksum_bytes",	KSTAT_DATA_UINT64 },
13986 		{ "ip_in_sw_cksum",		KSTAT_DATA_UINT64 },
13987 		{ "ip_ire_reclaim_calls",	KSTAT_DATA_UINT64 },
13988 		{ "ip_ire_reclaim_deleted",	KSTAT_DATA_UINT64 },
13989 		{ "ip_nce_reclaim_calls",	KSTAT_DATA_UINT64 },
13990 		{ "ip_nce_reclaim_deleted",	KSTAT_DATA_UINT64 },
13991 		{ "ip_nce_mcast_reclaim_calls",	KSTAT_DATA_UINT64 },
13992 		{ "ip_nce_mcast_reclaim_deleted",	KSTAT_DATA_UINT64 },
13993 		{ "ip_nce_mcast_reclaim_tqfail",	KSTAT_DATA_UINT64 },
13994 		{ "ip_dce_reclaim_calls",	KSTAT_DATA_UINT64 },
13995 		{ "ip_dce_reclaim_deleted",	KSTAT_DATA_UINT64 },
13996 		{ "ip_tcp_in_full_hw_cksum_err",	KSTAT_DATA_UINT64 },
13997 		{ "ip_tcp_in_part_hw_cksum_err",	KSTAT_DATA_UINT64 },
13998 		{ "ip_tcp_in_sw_cksum_err",		KSTAT_DATA_UINT64 },
13999 		{ "ip_udp_in_full_hw_cksum_err",	KSTAT_DATA_UINT64 },
14000 		{ "ip_udp_in_part_hw_cksum_err",	KSTAT_DATA_UINT64 },
14001 		{ "ip_udp_in_sw_cksum_err",	KSTAT_DATA_UINT64 },
14002 		{ "conn_in_recvdstaddr",	KSTAT_DATA_UINT64 },
14003 		{ "conn_in_recvopts",		KSTAT_DATA_UINT64 },
14004 		{ "conn_in_recvif",		KSTAT_DATA_UINT64 },
14005 		{ "conn_in_recvslla",		KSTAT_DATA_UINT64 },
14006 		{ "conn_in_recvucred",		KSTAT_DATA_UINT64 },
14007 		{ "conn_in_recvttl",		KSTAT_DATA_UINT64 },
14008 		{ "conn_in_recvtos",		KSTAT_DATA_UINT64 },
14009 		{ "conn_in_recvhopopts",	KSTAT_DATA_UINT64 },
14010 		{ "conn_in_recvhoplimit",	KSTAT_DATA_UINT64 },
14011 		{ "conn_in_recvdstopts",	KSTAT_DATA_UINT64 },
14012 		{ "conn_in_recvrthdrdstopts",	KSTAT_DATA_UINT64 },
14013 		{ "conn_in_recvrthdr",		KSTAT_DATA_UINT64 },
14014 		{ "conn_in_recvpktinfo",	KSTAT_DATA_UINT64 },
14015 		{ "conn_in_recvtclass",		KSTAT_DATA_UINT64 },
14016 		{ "conn_in_timestamp",		KSTAT_DATA_UINT64 },
14017 	};
14018 
14019 	ksp = kstat_create_netstack("ip", 0, "ipstat", "net",
14020 	    KSTAT_TYPE_NAMED, sizeof (template) / sizeof (kstat_named_t),
14021 	    KSTAT_FLAG_VIRTUAL, stackid);
14022 
14023 	if (ksp == NULL)
14024 		return (NULL);
14025 
14026 	bcopy(&template, ip_statisticsp, sizeof (template));
14027 	ksp->ks_data = (void *)ip_statisticsp;
14028 	ksp->ks_private = (void *)(uintptr_t)stackid;
14029 
14030 	kstat_install(ksp);
14031 	return (ksp);
14032 }
14033 
14034 static void
ip_kstat2_fini(netstackid_t stackid,kstat_t * ksp)14035 ip_kstat2_fini(netstackid_t stackid, kstat_t *ksp)
14036 {
14037 	if (ksp != NULL) {
14038 		ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
14039 		kstat_delete_netstack(ksp, stackid);
14040 	}
14041 }
14042 
14043 static void *
ip_kstat_init(netstackid_t stackid,ip_stack_t * ipst)14044 ip_kstat_init(netstackid_t stackid, ip_stack_t *ipst)
14045 {
14046 	kstat_t	*ksp;
14047 
14048 	ip_named_kstat_t template = {
14049 		{ "forwarding",		KSTAT_DATA_UINT32, 0 },
14050 		{ "defaultTTL",		KSTAT_DATA_UINT32, 0 },
14051 		{ "inReceives",		KSTAT_DATA_UINT64, 0 },
14052 		{ "inHdrErrors",	KSTAT_DATA_UINT32, 0 },
14053 		{ "inAddrErrors",	KSTAT_DATA_UINT32, 0 },
14054 		{ "forwDatagrams",	KSTAT_DATA_UINT64, 0 },
14055 		{ "inUnknownProtos",	KSTAT_DATA_UINT32, 0 },
14056 		{ "inDiscards",		KSTAT_DATA_UINT32, 0 },
14057 		{ "inDelivers",		KSTAT_DATA_UINT64, 0 },
14058 		{ "outRequests",	KSTAT_DATA_UINT64, 0 },
14059 		{ "outDiscards",	KSTAT_DATA_UINT32, 0 },
14060 		{ "outNoRoutes",	KSTAT_DATA_UINT32, 0 },
14061 		{ "reasmTimeout",	KSTAT_DATA_UINT32, 0 },
14062 		{ "reasmReqds",		KSTAT_DATA_UINT32, 0 },
14063 		{ "reasmOKs",		KSTAT_DATA_UINT32, 0 },
14064 		{ "reasmFails",		KSTAT_DATA_UINT32, 0 },
14065 		{ "fragOKs",		KSTAT_DATA_UINT32, 0 },
14066 		{ "fragFails",		KSTAT_DATA_UINT32, 0 },
14067 		{ "fragCreates",	KSTAT_DATA_UINT32, 0 },
14068 		{ "addrEntrySize",	KSTAT_DATA_INT32, 0 },
14069 		{ "routeEntrySize",	KSTAT_DATA_INT32, 0 },
14070 		{ "netToMediaEntrySize",	KSTAT_DATA_INT32, 0 },
14071 		{ "routingDiscards",	KSTAT_DATA_UINT32, 0 },
14072 		{ "inErrs",		KSTAT_DATA_UINT32, 0 },
14073 		{ "noPorts",		KSTAT_DATA_UINT32, 0 },
14074 		{ "inCksumErrs",	KSTAT_DATA_UINT32, 0 },
14075 		{ "reasmDuplicates",	KSTAT_DATA_UINT32, 0 },
14076 		{ "reasmPartDups",	KSTAT_DATA_UINT32, 0 },
14077 		{ "forwProhibits",	KSTAT_DATA_UINT32, 0 },
14078 		{ "udpInCksumErrs",	KSTAT_DATA_UINT32, 0 },
14079 		{ "udpInOverflows",	KSTAT_DATA_UINT32, 0 },
14080 		{ "rawipInOverflows",	KSTAT_DATA_UINT32, 0 },
14081 		{ "ipsecInSucceeded",	KSTAT_DATA_UINT32, 0 },
14082 		{ "ipsecInFailed",	KSTAT_DATA_INT32, 0 },
14083 		{ "memberEntrySize",	KSTAT_DATA_INT32, 0 },
14084 		{ "inIPv6",		KSTAT_DATA_UINT32, 0 },
14085 		{ "outIPv6",		KSTAT_DATA_UINT32, 0 },
14086 		{ "outSwitchIPv6",	KSTAT_DATA_UINT32, 0 },
14087 	};
14088 
14089 	ksp = kstat_create_netstack("ip", 0, "ip", "mib2", KSTAT_TYPE_NAMED,
14090 	    NUM_OF_FIELDS(ip_named_kstat_t), 0, stackid);
14091 	if (ksp == NULL || ksp->ks_data == NULL)
14092 		return (NULL);
14093 
14094 	template.forwarding.value.ui32 = WE_ARE_FORWARDING(ipst) ? 1:2;
14095 	template.defaultTTL.value.ui32 = (uint32_t)ipst->ips_ip_def_ttl;
14096 	template.reasmTimeout.value.ui32 = ipst->ips_ip_reassembly_timeout;
14097 	template.addrEntrySize.value.i32 = sizeof (mib2_ipAddrEntry_t);
14098 	template.routeEntrySize.value.i32 = sizeof (mib2_ipRouteEntry_t);
14099 
14100 	template.netToMediaEntrySize.value.i32 =
14101 	    sizeof (mib2_ipNetToMediaEntry_t);
14102 
14103 	template.memberEntrySize.value.i32 = sizeof (ipv6_member_t);
14104 
14105 	bcopy(&template, ksp->ks_data, sizeof (template));
14106 	ksp->ks_update = ip_kstat_update;
14107 	ksp->ks_private = (void *)(uintptr_t)stackid;
14108 
14109 	kstat_install(ksp);
14110 	return (ksp);
14111 }
14112 
14113 static void
ip_kstat_fini(netstackid_t stackid,kstat_t * ksp)14114 ip_kstat_fini(netstackid_t stackid, kstat_t *ksp)
14115 {
14116 	if (ksp != NULL) {
14117 		ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
14118 		kstat_delete_netstack(ksp, stackid);
14119 	}
14120 }
14121 
14122 static int
ip_kstat_update(kstat_t * kp,int rw)14123 ip_kstat_update(kstat_t *kp, int rw)
14124 {
14125 	ip_named_kstat_t *ipkp;
14126 	mib2_ipIfStatsEntry_t ipmib;
14127 	ill_walk_context_t ctx;
14128 	ill_t *ill;
14129 	netstackid_t	stackid = (zoneid_t)(uintptr_t)kp->ks_private;
14130 	netstack_t	*ns;
14131 	ip_stack_t	*ipst;
14132 
14133 	if (kp->ks_data == NULL)
14134 		return (EIO);
14135 
14136 	if (rw == KSTAT_WRITE)
14137 		return (EACCES);
14138 
14139 	ns = netstack_find_by_stackid(stackid);
14140 	if (ns == NULL)
14141 		return (-1);
14142 	ipst = ns->netstack_ip;
14143 	if (ipst == NULL) {
14144 		netstack_rele(ns);
14145 		return (-1);
14146 	}
14147 	ipkp = (ip_named_kstat_t *)kp->ks_data;
14148 
14149 	bcopy(&ipst->ips_ip_mib, &ipmib, sizeof (ipmib));
14150 	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
14151 	ill = ILL_START_WALK_V4(&ctx, ipst);
14152 	for (; ill != NULL; ill = ill_next(&ctx, ill))
14153 		ip_mib2_add_ip_stats(&ipmib, ill->ill_ip_mib);
14154 	rw_exit(&ipst->ips_ill_g_lock);
14155 
14156 	ipkp->forwarding.value.ui32 =		ipmib.ipIfStatsForwarding;
14157 	ipkp->defaultTTL.value.ui32 =		ipmib.ipIfStatsDefaultTTL;
14158 	ipkp->inReceives.value.ui64 =		ipmib.ipIfStatsHCInReceives;
14159 	ipkp->inHdrErrors.value.ui32 =		ipmib.ipIfStatsInHdrErrors;
14160 	ipkp->inAddrErrors.value.ui32 =		ipmib.ipIfStatsInAddrErrors;
14161 	ipkp->forwDatagrams.value.ui64 = ipmib.ipIfStatsHCOutForwDatagrams;
14162 	ipkp->inUnknownProtos.value.ui32 =	ipmib.ipIfStatsInUnknownProtos;
14163 	ipkp->inDiscards.value.ui32 =		ipmib.ipIfStatsInDiscards;
14164 	ipkp->inDelivers.value.ui64 =		ipmib.ipIfStatsHCInDelivers;
14165 	ipkp->outRequests.value.ui64 =		ipmib.ipIfStatsHCOutRequests;
14166 	ipkp->outDiscards.value.ui32 =		ipmib.ipIfStatsOutDiscards;
14167 	ipkp->outNoRoutes.value.ui32 =		ipmib.ipIfStatsOutNoRoutes;
14168 	ipkp->reasmTimeout.value.ui32 =		ipst->ips_ip_reassembly_timeout;
14169 	ipkp->reasmReqds.value.ui32 =		ipmib.ipIfStatsReasmReqds;
14170 	ipkp->reasmOKs.value.ui32 =		ipmib.ipIfStatsReasmOKs;
14171 	ipkp->reasmFails.value.ui32 =		ipmib.ipIfStatsReasmFails;
14172 	ipkp->fragOKs.value.ui32 =		ipmib.ipIfStatsOutFragOKs;
14173 	ipkp->fragFails.value.ui32 =		ipmib.ipIfStatsOutFragFails;
14174 	ipkp->fragCreates.value.ui32 =		ipmib.ipIfStatsOutFragCreates;
14175 
14176 	ipkp->routingDiscards.value.ui32 =	0;
14177 	ipkp->inErrs.value.ui32 =		ipmib.tcpIfStatsInErrs;
14178 	ipkp->noPorts.value.ui32 =		ipmib.udpIfStatsNoPorts;
14179 	ipkp->inCksumErrs.value.ui32 =		ipmib.ipIfStatsInCksumErrs;
14180 	ipkp->reasmDuplicates.value.ui32 =	ipmib.ipIfStatsReasmDuplicates;
14181 	ipkp->reasmPartDups.value.ui32 =	ipmib.ipIfStatsReasmPartDups;
14182 	ipkp->forwProhibits.value.ui32 =	ipmib.ipIfStatsForwProhibits;
14183 	ipkp->udpInCksumErrs.value.ui32 =	ipmib.udpIfStatsInCksumErrs;
14184 	ipkp->udpInOverflows.value.ui32 =	ipmib.udpIfStatsInOverflows;
14185 	ipkp->rawipInOverflows.value.ui32 =	ipmib.rawipIfStatsInOverflows;
14186 	ipkp->ipsecInSucceeded.value.ui32 =	ipmib.ipsecIfStatsInSucceeded;
14187 	ipkp->ipsecInFailed.value.i32 =		ipmib.ipsecIfStatsInFailed;
14188 
14189 	ipkp->inIPv6.value.ui32 =	ipmib.ipIfStatsInWrongIPVersion;
14190 	ipkp->outIPv6.value.ui32 =	ipmib.ipIfStatsOutWrongIPVersion;
14191 	ipkp->outSwitchIPv6.value.ui32 = ipmib.ipIfStatsOutSwitchIPVersion;
14192 
14193 	netstack_rele(ns);
14194 
14195 	return (0);
14196 }
14197 
14198 static void *
icmp_kstat_init(netstackid_t stackid)14199 icmp_kstat_init(netstackid_t stackid)
14200 {
14201 	kstat_t	*ksp;
14202 
14203 	icmp_named_kstat_t template = {
14204 		{ "inMsgs",		KSTAT_DATA_UINT32 },
14205 		{ "inErrors",		KSTAT_DATA_UINT32 },
14206 		{ "inDestUnreachs",	KSTAT_DATA_UINT32 },
14207 		{ "inTimeExcds",	KSTAT_DATA_UINT32 },
14208 		{ "inParmProbs",	KSTAT_DATA_UINT32 },
14209 		{ "inSrcQuenchs",	KSTAT_DATA_UINT32 },
14210 		{ "inRedirects",	KSTAT_DATA_UINT32 },
14211 		{ "inEchos",		KSTAT_DATA_UINT32 },
14212 		{ "inEchoReps",		KSTAT_DATA_UINT32 },
14213 		{ "inTimestamps",	KSTAT_DATA_UINT32 },
14214 		{ "inTimestampReps",	KSTAT_DATA_UINT32 },
14215 		{ "inAddrMasks",	KSTAT_DATA_UINT32 },
14216 		{ "inAddrMaskReps",	KSTAT_DATA_UINT32 },
14217 		{ "outMsgs",		KSTAT_DATA_UINT32 },
14218 		{ "outErrors",		KSTAT_DATA_UINT32 },
14219 		{ "outDestUnreachs",	KSTAT_DATA_UINT32 },
14220 		{ "outTimeExcds",	KSTAT_DATA_UINT32 },
14221 		{ "outParmProbs",	KSTAT_DATA_UINT32 },
14222 		{ "outSrcQuenchs",	KSTAT_DATA_UINT32 },
14223 		{ "outRedirects",	KSTAT_DATA_UINT32 },
14224 		{ "outEchos",		KSTAT_DATA_UINT32 },
14225 		{ "outEchoReps",	KSTAT_DATA_UINT32 },
14226 		{ "outTimestamps",	KSTAT_DATA_UINT32 },
14227 		{ "outTimestampReps",	KSTAT_DATA_UINT32 },
14228 		{ "outAddrMasks",	KSTAT_DATA_UINT32 },
14229 		{ "outAddrMaskReps",	KSTAT_DATA_UINT32 },
14230 		{ "inChksumErrs",	KSTAT_DATA_UINT32 },
14231 		{ "inUnknowns",		KSTAT_DATA_UINT32 },
14232 		{ "inFragNeeded",	KSTAT_DATA_UINT32 },
14233 		{ "outFragNeeded",	KSTAT_DATA_UINT32 },
14234 		{ "outDrops",		KSTAT_DATA_UINT32 },
14235 		{ "inOverFlows",	KSTAT_DATA_UINT32 },
14236 		{ "inBadRedirects",	KSTAT_DATA_UINT32 },
14237 	};
14238 
14239 	ksp = kstat_create_netstack("ip", 0, "icmp", "mib2", KSTAT_TYPE_NAMED,
14240 	    NUM_OF_FIELDS(icmp_named_kstat_t), 0, stackid);
14241 	if (ksp == NULL || ksp->ks_data == NULL)
14242 		return (NULL);
14243 
14244 	bcopy(&template, ksp->ks_data, sizeof (template));
14245 
14246 	ksp->ks_update = icmp_kstat_update;
14247 	ksp->ks_private = (void *)(uintptr_t)stackid;
14248 
14249 	kstat_install(ksp);
14250 	return (ksp);
14251 }
14252 
14253 static void
icmp_kstat_fini(netstackid_t stackid,kstat_t * ksp)14254 icmp_kstat_fini(netstackid_t stackid, kstat_t *ksp)
14255 {
14256 	if (ksp != NULL) {
14257 		ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
14258 		kstat_delete_netstack(ksp, stackid);
14259 	}
14260 }
14261 
14262 static int
icmp_kstat_update(kstat_t * kp,int rw)14263 icmp_kstat_update(kstat_t *kp, int rw)
14264 {
14265 	icmp_named_kstat_t *icmpkp;
14266 	netstackid_t	stackid = (zoneid_t)(uintptr_t)kp->ks_private;
14267 	netstack_t	*ns;
14268 	ip_stack_t	*ipst;
14269 
14270 	if (kp->ks_data == NULL)
14271 		return (EIO);
14272 
14273 	if (rw == KSTAT_WRITE)
14274 		return (EACCES);
14275 
14276 	ns = netstack_find_by_stackid(stackid);
14277 	if (ns == NULL)
14278 		return (-1);
14279 	ipst = ns->netstack_ip;
14280 	if (ipst == NULL) {
14281 		netstack_rele(ns);
14282 		return (-1);
14283 	}
14284 	icmpkp = (icmp_named_kstat_t *)kp->ks_data;
14285 
14286 	icmpkp->inMsgs.value.ui32 =	    ipst->ips_icmp_mib.icmpInMsgs;
14287 	icmpkp->inErrors.value.ui32 =	    ipst->ips_icmp_mib.icmpInErrors;
14288 	icmpkp->inDestUnreachs.value.ui32 =
14289 	    ipst->ips_icmp_mib.icmpInDestUnreachs;
14290 	icmpkp->inTimeExcds.value.ui32 =    ipst->ips_icmp_mib.icmpInTimeExcds;
14291 	icmpkp->inParmProbs.value.ui32 =    ipst->ips_icmp_mib.icmpInParmProbs;
14292 	icmpkp->inSrcQuenchs.value.ui32 =   ipst->ips_icmp_mib.icmpInSrcQuenchs;
14293 	icmpkp->inRedirects.value.ui32 =    ipst->ips_icmp_mib.icmpInRedirects;
14294 	icmpkp->inEchos.value.ui32 =	    ipst->ips_icmp_mib.icmpInEchos;
14295 	icmpkp->inEchoReps.value.ui32 =	    ipst->ips_icmp_mib.icmpInEchoReps;
14296 	icmpkp->inTimestamps.value.ui32 =   ipst->ips_icmp_mib.icmpInTimestamps;
14297 	icmpkp->inTimestampReps.value.ui32 =
14298 	    ipst->ips_icmp_mib.icmpInTimestampReps;
14299 	icmpkp->inAddrMasks.value.ui32 =    ipst->ips_icmp_mib.icmpInAddrMasks;
14300 	icmpkp->inAddrMaskReps.value.ui32 =
14301 	    ipst->ips_icmp_mib.icmpInAddrMaskReps;
14302 	icmpkp->outMsgs.value.ui32 =	    ipst->ips_icmp_mib.icmpOutMsgs;
14303 	icmpkp->outErrors.value.ui32 =	    ipst->ips_icmp_mib.icmpOutErrors;
14304 	icmpkp->outDestUnreachs.value.ui32 =
14305 	    ipst->ips_icmp_mib.icmpOutDestUnreachs;
14306 	icmpkp->outTimeExcds.value.ui32 =   ipst->ips_icmp_mib.icmpOutTimeExcds;
14307 	icmpkp->outParmProbs.value.ui32 =   ipst->ips_icmp_mib.icmpOutParmProbs;
14308 	icmpkp->outSrcQuenchs.value.ui32 =
14309 	    ipst->ips_icmp_mib.icmpOutSrcQuenchs;
14310 	icmpkp->outRedirects.value.ui32 =   ipst->ips_icmp_mib.icmpOutRedirects;
14311 	icmpkp->outEchos.value.ui32 =	    ipst->ips_icmp_mib.icmpOutEchos;
14312 	icmpkp->outEchoReps.value.ui32 =    ipst->ips_icmp_mib.icmpOutEchoReps;
14313 	icmpkp->outTimestamps.value.ui32 =
14314 	    ipst->ips_icmp_mib.icmpOutTimestamps;
14315 	icmpkp->outTimestampReps.value.ui32 =
14316 	    ipst->ips_icmp_mib.icmpOutTimestampReps;
14317 	icmpkp->outAddrMasks.value.ui32 =
14318 	    ipst->ips_icmp_mib.icmpOutAddrMasks;
14319 	icmpkp->outAddrMaskReps.value.ui32 =
14320 	    ipst->ips_icmp_mib.icmpOutAddrMaskReps;
14321 	icmpkp->inCksumErrs.value.ui32 =    ipst->ips_icmp_mib.icmpInCksumErrs;
14322 	icmpkp->inUnknowns.value.ui32 =	    ipst->ips_icmp_mib.icmpInUnknowns;
14323 	icmpkp->inFragNeeded.value.ui32 =   ipst->ips_icmp_mib.icmpInFragNeeded;
14324 	icmpkp->outFragNeeded.value.ui32 =
14325 	    ipst->ips_icmp_mib.icmpOutFragNeeded;
14326 	icmpkp->outDrops.value.ui32 =	    ipst->ips_icmp_mib.icmpOutDrops;
14327 	icmpkp->inOverflows.value.ui32 =    ipst->ips_icmp_mib.icmpInOverflows;
14328 	icmpkp->inBadRedirects.value.ui32 =
14329 	    ipst->ips_icmp_mib.icmpInBadRedirects;
14330 
14331 	netstack_rele(ns);
14332 	return (0);
14333 }
14334 
14335 /*
14336  * This is the fanout function for raw socket opened for SCTP.  Note
14337  * that it is called after SCTP checks that there is no socket which
14338  * wants a packet.  Then before SCTP handles this out of the blue packet,
14339  * this function is called to see if there is any raw socket for SCTP.
14340  * If there is and it is bound to the correct address, the packet will
14341  * be sent to that socket.  Note that only one raw socket can be bound to
14342  * a port.  This is assured in ipcl_sctp_hash_insert();
14343  */
14344 void
ip_fanout_sctp_raw(mblk_t * mp,ipha_t * ipha,ip6_t * ip6h,uint32_t ports,ip_recv_attr_t * ira)14345 ip_fanout_sctp_raw(mblk_t *mp, ipha_t *ipha, ip6_t *ip6h, uint32_t ports,
14346     ip_recv_attr_t *ira)
14347 {
14348 	conn_t		*connp;
14349 	queue_t		*rq;
14350 	boolean_t	secure;
14351 	ill_t		*ill = ira->ira_ill;
14352 	ip_stack_t	*ipst = ill->ill_ipst;
14353 	ipsec_stack_t	*ipss = ipst->ips_netstack->netstack_ipsec;
14354 	sctp_stack_t	*sctps = ipst->ips_netstack->netstack_sctp;
14355 	iaflags_t	iraflags = ira->ira_flags;
14356 	ill_t		*rill = ira->ira_rill;
14357 
14358 	secure = iraflags & IRAF_IPSEC_SECURE;
14359 
14360 	connp = ipcl_classify_raw(mp, IPPROTO_SCTP, ports, ipha, ip6h,
14361 	    ira, ipst);
14362 	if (connp == NULL) {
14363 		/*
14364 		 * Although raw sctp is not summed, OOB chunks must be.
14365 		 * Drop the packet here if the sctp checksum failed.
14366 		 */
14367 		if (iraflags & IRAF_SCTP_CSUM_ERR) {
14368 			SCTPS_BUMP_MIB(sctps, sctpChecksumError);
14369 			freemsg(mp);
14370 			return;
14371 		}
14372 		ira->ira_ill = ira->ira_rill = NULL;
14373 		sctp_ootb_input(mp, ira, ipst);
14374 		ira->ira_ill = ill;
14375 		ira->ira_rill = rill;
14376 		return;
14377 	}
14378 	rq = connp->conn_rq;
14379 	if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld : !canputnext(rq)) {
14380 		CONN_DEC_REF(connp);
14381 		BUMP_MIB(ill->ill_ip_mib, rawipIfStatsInOverflows);
14382 		freemsg(mp);
14383 		return;
14384 	}
14385 	if (((iraflags & IRAF_IS_IPV4) ?
14386 	    CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
14387 	    CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
14388 	    secure) {
14389 		mp = ipsec_check_inbound_policy(mp, connp, ipha,
14390 		    ip6h, ira);
14391 		if (mp == NULL) {
14392 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
14393 			/* Note that mp is NULL */
14394 			ip_drop_input("ipIfStatsInDiscards", mp, ill);
14395 			CONN_DEC_REF(connp);
14396 			return;
14397 		}
14398 	}
14399 
14400 	if (iraflags & IRAF_ICMP_ERROR) {
14401 		(connp->conn_recvicmp)(connp, mp, NULL, ira);
14402 	} else {
14403 		ill_t *rill = ira->ira_rill;
14404 
14405 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
14406 		/* This is the SOCK_RAW, IPPROTO_SCTP case. */
14407 		ira->ira_ill = ira->ira_rill = NULL;
14408 		(connp->conn_recv)(connp, mp, NULL, ira);
14409 		ira->ira_ill = ill;
14410 		ira->ira_rill = rill;
14411 	}
14412 	CONN_DEC_REF(connp);
14413 }
14414 
14415 /*
14416  * Free a packet that has the link-layer dl_unitdata_req_t or fast-path
14417  * header before the ip payload.
14418  */
14419 static void
ip_xmit_flowctl_drop(ill_t * ill,mblk_t * mp,boolean_t is_fp_mp,int fp_mp_len)14420 ip_xmit_flowctl_drop(ill_t *ill, mblk_t *mp, boolean_t is_fp_mp, int fp_mp_len)
14421 {
14422 	int len = (mp->b_wptr - mp->b_rptr);
14423 	mblk_t *ip_mp;
14424 
14425 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14426 	if (is_fp_mp || len != fp_mp_len) {
14427 		if (len > fp_mp_len) {
14428 			/*
14429 			 * fastpath header and ip header in the first mblk
14430 			 */
14431 			mp->b_rptr += fp_mp_len;
14432 		} else {
14433 			/*
14434 			 * ip_xmit_attach_llhdr had to prepend an mblk to
14435 			 * attach the fastpath header before ip header.
14436 			 */
14437 			ip_mp = mp->b_cont;
14438 			freeb(mp);
14439 			mp = ip_mp;
14440 			mp->b_rptr += (fp_mp_len - len);
14441 		}
14442 	} else {
14443 		ip_mp = mp->b_cont;
14444 		freeb(mp);
14445 		mp = ip_mp;
14446 	}
14447 	ip_drop_output("ipIfStatsOutDiscards - flow ctl", mp, ill);
14448 	freemsg(mp);
14449 }
14450 
14451 /*
14452  * Normal post fragmentation function.
14453  *
14454  * Send a packet using the passed in nce. This handles both IPv4 and IPv6
14455  * using the same state machine.
14456  *
14457  * We return an error on failure. In particular we return EWOULDBLOCK
14458  * when the driver flow controls. In that case this ensures that ip_wsrv runs
14459  * (currently by canputnext failure resulting in backenabling from GLD.)
14460  * This allows the callers of conn_ip_output() to use EWOULDBLOCK as an
14461  * indication that they can flow control until ip_wsrv() tells then to restart.
14462  *
14463  * If the nce passed by caller is incomplete, this function
14464  * queues the packet and if necessary, sends ARP request and bails.
14465  * If the Neighbor Cache passed is fully resolved, we simply prepend
14466  * the link-layer header to the packet, do ipsec hw acceleration
14467  * work if necessary, and send the packet out on the wire.
14468  */
14469 /* ARGSUSED6 */
14470 int
ip_xmit(mblk_t * mp,nce_t * nce,iaflags_t ixaflags,uint_t pkt_len,uint32_t xmit_hint,zoneid_t szone,zoneid_t nolzid,uintptr_t * ixacookie)14471 ip_xmit(mblk_t *mp, nce_t *nce, iaflags_t ixaflags, uint_t pkt_len,
14472     uint32_t xmit_hint, zoneid_t szone, zoneid_t nolzid, uintptr_t *ixacookie)
14473 {
14474 	queue_t		*wq;
14475 	ill_t		*ill = nce->nce_ill;
14476 	ip_stack_t	*ipst = ill->ill_ipst;
14477 	uint64_t	delta;
14478 	boolean_t	isv6 = ill->ill_isv6;
14479 	boolean_t	fp_mp;
14480 	ncec_t		*ncec = nce->nce_common;
14481 	int64_t		now = LBOLT_FASTPATH64;
14482 	boolean_t	is_probe;
14483 
14484 	DTRACE_PROBE1(ip__xmit, nce_t *, nce);
14485 
14486 	ASSERT(mp != NULL);
14487 	ASSERT(mp->b_datap->db_type == M_DATA);
14488 	ASSERT(pkt_len == msgdsize(mp));
14489 
14490 	/*
14491 	 * If we have already been here and are coming back after ARP/ND.
14492 	 * the IXAF_NO_TRACE flag is set. We skip FW_HOOKS, DTRACE and ipobs
14493 	 * in that case since they have seen the packet when it came here
14494 	 * the first time.
14495 	 */
14496 	if (ixaflags & IXAF_NO_TRACE)
14497 		goto sendit;
14498 
14499 	if (ixaflags & IXAF_IS_IPV4) {
14500 		ipha_t *ipha = (ipha_t *)mp->b_rptr;
14501 
14502 		ASSERT(!isv6);
14503 		ASSERT(pkt_len == ntohs(((ipha_t *)mp->b_rptr)->ipha_length));
14504 		if (HOOKS4_INTERESTED_PHYSICAL_OUT(ipst) &&
14505 		    !(ixaflags & IXAF_NO_PFHOOK)) {
14506 			int	error;
14507 
14508 			FW_HOOKS(ipst->ips_ip4_physical_out_event,
14509 			    ipst->ips_ipv4firewall_physical_out,
14510 			    NULL, ill, ipha, mp, mp, 0, ipst, error);
14511 			DTRACE_PROBE1(ip4__physical__out__end,
14512 			    mblk_t *, mp);
14513 			if (mp == NULL)
14514 				return (error);
14515 
14516 			/* The length could have changed */
14517 			pkt_len = msgdsize(mp);
14518 		}
14519 		if (ipst->ips_ip4_observe.he_interested) {
14520 			/*
14521 			 * Note that for TX the zoneid is the sending
14522 			 * zone, whether or not MLP is in play.
14523 			 * Since the szone argument is the IP zoneid (i.e.,
14524 			 * zero for exclusive-IP zones) and ipobs wants
14525 			 * the system zoneid, we map it here.
14526 			 */
14527 			szone = IP_REAL_ZONEID(szone, ipst);
14528 
14529 			/*
14530 			 * On the outbound path the destination zone will be
14531 			 * unknown as we're sending this packet out on the
14532 			 * wire.
14533 			 */
14534 			ipobs_hook(mp, IPOBS_HOOK_OUTBOUND, szone, ALL_ZONES,
14535 			    ill, ipst);
14536 		}
14537 		DTRACE_IP7(send, mblk_t *, mp,  conn_t *, NULL,
14538 		    void_ip_t *, ipha,  __dtrace_ipsr_ill_t *, ill,
14539 		    ipha_t *, ipha, ip6_t *, NULL, int, 0);
14540 	} else {
14541 		ip6_t *ip6h = (ip6_t *)mp->b_rptr;
14542 
14543 		ASSERT(isv6);
14544 		ASSERT(pkt_len ==
14545 		    ntohs(((ip6_t *)mp->b_rptr)->ip6_plen) + IPV6_HDR_LEN);
14546 		if (HOOKS6_INTERESTED_PHYSICAL_OUT(ipst) &&
14547 		    !(ixaflags & IXAF_NO_PFHOOK)) {
14548 			int	error;
14549 
14550 			FW_HOOKS6(ipst->ips_ip6_physical_out_event,
14551 			    ipst->ips_ipv6firewall_physical_out,
14552 			    NULL, ill, ip6h, mp, mp, 0, ipst, error);
14553 			DTRACE_PROBE1(ip6__physical__out__end,
14554 			    mblk_t *, mp);
14555 			if (mp == NULL)
14556 				return (error);
14557 
14558 			/* The length could have changed */
14559 			pkt_len = msgdsize(mp);
14560 		}
14561 		if (ipst->ips_ip6_observe.he_interested) {
14562 			/* See above */
14563 			szone = IP_REAL_ZONEID(szone, ipst);
14564 
14565 			ipobs_hook(mp, IPOBS_HOOK_OUTBOUND, szone, ALL_ZONES,
14566 			    ill, ipst);
14567 		}
14568 		DTRACE_IP7(send, mblk_t *, mp,  conn_t *, NULL,
14569 		    void_ip_t *, ip6h,  __dtrace_ipsr_ill_t *, ill,
14570 		    ipha_t *, NULL, ip6_t *, ip6h, int, 0);
14571 	}
14572 
14573 sendit:
14574 	/*
14575 	 * We check the state without a lock because the state can never
14576 	 * move "backwards" to initial or incomplete.
14577 	 */
14578 	switch (ncec->ncec_state) {
14579 	case ND_REACHABLE:
14580 	case ND_STALE:
14581 	case ND_DELAY:
14582 	case ND_PROBE:
14583 		mp = ip_xmit_attach_llhdr(mp, nce);
14584 		if (mp == NULL) {
14585 			/*
14586 			 * ip_xmit_attach_llhdr has increased
14587 			 * ipIfStatsOutDiscards and called ip_drop_output()
14588 			 */
14589 			return (ENOBUFS);
14590 		}
14591 		/*
14592 		 * check if nce_fastpath completed and we tagged on a
14593 		 * copy of nce_fp_mp in ip_xmit_attach_llhdr().
14594 		 */
14595 		fp_mp = (mp->b_datap->db_type == M_DATA);
14596 
14597 		if (fp_mp &&
14598 		    (ill->ill_capabilities & ILL_CAPAB_DLD_DIRECT)) {
14599 			ill_dld_direct_t *idd;
14600 
14601 			idd = &ill->ill_dld_capab->idc_direct;
14602 			/*
14603 			 * Send the packet directly to DLD, where it
14604 			 * may be queued depending on the availability
14605 			 * of transmit resources at the media layer.
14606 			 * Return value should be taken into
14607 			 * account and flow control the TCP.
14608 			 */
14609 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCOutTransmits);
14610 			UPDATE_MIB(ill->ill_ip_mib, ipIfStatsHCOutOctets,
14611 			    pkt_len);
14612 
14613 			if (ixaflags & IXAF_NO_DEV_FLOW_CTL) {
14614 				(void) idd->idd_tx_df(idd->idd_tx_dh, mp,
14615 				    (uintptr_t)xmit_hint, IP_DROP_ON_NO_DESC);
14616 			} else {
14617 				uintptr_t cookie;
14618 
14619 				if ((cookie = idd->idd_tx_df(idd->idd_tx_dh,
14620 				    mp, (uintptr_t)xmit_hint, 0)) != 0) {
14621 					if (ixacookie != NULL)
14622 						*ixacookie = cookie;
14623 					return (EWOULDBLOCK);
14624 				}
14625 			}
14626 		} else {
14627 			wq = ill->ill_wq;
14628 
14629 			if (!(ixaflags & IXAF_NO_DEV_FLOW_CTL) &&
14630 			    !canputnext(wq)) {
14631 				if (ixacookie != NULL)
14632 					*ixacookie = 0;
14633 				ip_xmit_flowctl_drop(ill, mp, fp_mp,
14634 				    nce->nce_fp_mp != NULL ?
14635 				    MBLKL(nce->nce_fp_mp) : 0);
14636 				return (EWOULDBLOCK);
14637 			}
14638 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCOutTransmits);
14639 			UPDATE_MIB(ill->ill_ip_mib, ipIfStatsHCOutOctets,
14640 			    pkt_len);
14641 			putnext(wq, mp);
14642 		}
14643 
14644 		/*
14645 		 * The rest of this function implements Neighbor Unreachability
14646 		 * detection. Determine if the ncec is eligible for NUD.
14647 		 */
14648 		if (ncec->ncec_flags & NCE_F_NONUD)
14649 			return (0);
14650 
14651 		ASSERT(ncec->ncec_state != ND_INCOMPLETE);
14652 
14653 		/*
14654 		 * Check for upper layer advice
14655 		 */
14656 		if (ixaflags & IXAF_REACH_CONF) {
14657 			timeout_id_t tid;
14658 
14659 			/*
14660 			 * It should be o.k. to check the state without
14661 			 * a lock here, at most we lose an advice.
14662 			 */
14663 			ncec->ncec_last = TICK_TO_MSEC(now);
14664 			if (ncec->ncec_state != ND_REACHABLE) {
14665 				mutex_enter(&ncec->ncec_lock);
14666 				ncec->ncec_state = ND_REACHABLE;
14667 				tid = ncec->ncec_timeout_id;
14668 				ncec->ncec_timeout_id = 0;
14669 				mutex_exit(&ncec->ncec_lock);
14670 				(void) untimeout(tid);
14671 				if (ip_debug > 2) {
14672 					/* ip1dbg */
14673 					pr_addr_dbg("ip_xmit: state"
14674 					    " for %s changed to"
14675 					    " REACHABLE\n", AF_INET6,
14676 					    &ncec->ncec_addr);
14677 				}
14678 			}
14679 			return (0);
14680 		}
14681 
14682 		delta =  TICK_TO_MSEC(now) - ncec->ncec_last;
14683 		ip1dbg(("ip_xmit: delta = %" PRId64
14684 		    " ill_reachable_time = %d \n", delta,
14685 		    ill->ill_reachable_time));
14686 		if (delta > (uint64_t)ill->ill_reachable_time) {
14687 			mutex_enter(&ncec->ncec_lock);
14688 			switch (ncec->ncec_state) {
14689 			case ND_REACHABLE:
14690 				ASSERT((ncec->ncec_flags & NCE_F_NONUD) == 0);
14691 				/* FALLTHROUGH */
14692 			case ND_STALE:
14693 				/*
14694 				 * ND_REACHABLE is identical to
14695 				 * ND_STALE in this specific case. If
14696 				 * reachable time has expired for this
14697 				 * neighbor (delta is greater than
14698 				 * reachable time), conceptually, the
14699 				 * neighbor cache is no longer in
14700 				 * REACHABLE state, but already in
14701 				 * STALE state.  So the correct
14702 				 * transition here is to ND_DELAY.
14703 				 */
14704 				ncec->ncec_state = ND_DELAY;
14705 				mutex_exit(&ncec->ncec_lock);
14706 				nce_restart_timer(ncec,
14707 				    ipst->ips_delay_first_probe_time);
14708 				if (ip_debug > 3) {
14709 					/* ip2dbg */
14710 					pr_addr_dbg("ip_xmit: state"
14711 					    " for %s changed to"
14712 					    " DELAY\n", AF_INET6,
14713 					    &ncec->ncec_addr);
14714 				}
14715 				break;
14716 			case ND_DELAY:
14717 			case ND_PROBE:
14718 				mutex_exit(&ncec->ncec_lock);
14719 				/* Timers have already started */
14720 				break;
14721 			case ND_UNREACHABLE:
14722 				/*
14723 				 * nce_timer has detected that this ncec
14724 				 * is unreachable and initiated deleting
14725 				 * this ncec.
14726 				 * This is a harmless race where we found the
14727 				 * ncec before it was deleted and have
14728 				 * just sent out a packet using this
14729 				 * unreachable ncec.
14730 				 */
14731 				mutex_exit(&ncec->ncec_lock);
14732 				break;
14733 			default:
14734 				ASSERT(0);
14735 				mutex_exit(&ncec->ncec_lock);
14736 			}
14737 		}
14738 		return (0);
14739 
14740 	case ND_INCOMPLETE:
14741 		/*
14742 		 * the state could have changed since we didn't hold the lock.
14743 		 * Re-verify state under lock.
14744 		 */
14745 		is_probe = ipmp_packet_is_probe(mp, nce->nce_ill);
14746 		mutex_enter(&ncec->ncec_lock);
14747 		if (NCE_ISREACHABLE(ncec)) {
14748 			mutex_exit(&ncec->ncec_lock);
14749 			goto sendit;
14750 		}
14751 		/* queue the packet */
14752 		nce_queue_mp(ncec, mp, is_probe);
14753 		mutex_exit(&ncec->ncec_lock);
14754 		DTRACE_PROBE2(ip__xmit__incomplete,
14755 		    (ncec_t *), ncec, (mblk_t *), mp);
14756 		return (0);
14757 
14758 	case ND_INITIAL:
14759 		/*
14760 		 * State could have changed since we didn't hold the lock, so
14761 		 * re-verify state.
14762 		 */
14763 		is_probe = ipmp_packet_is_probe(mp, nce->nce_ill);
14764 		mutex_enter(&ncec->ncec_lock);
14765 		if (NCE_ISREACHABLE(ncec))  {
14766 			mutex_exit(&ncec->ncec_lock);
14767 			goto sendit;
14768 		}
14769 		nce_queue_mp(ncec, mp, is_probe);
14770 		if (ncec->ncec_state == ND_INITIAL) {
14771 			ncec->ncec_state = ND_INCOMPLETE;
14772 			mutex_exit(&ncec->ncec_lock);
14773 			/*
14774 			 * figure out the source we want to use
14775 			 * and resolve it.
14776 			 */
14777 			ip_ndp_resolve(ncec);
14778 		} else  {
14779 			mutex_exit(&ncec->ncec_lock);
14780 		}
14781 		return (0);
14782 
14783 	case ND_UNREACHABLE:
14784 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14785 		ip_drop_output("ipIfStatsOutDiscards - ND_UNREACHABLE",
14786 		    mp, ill);
14787 		freemsg(mp);
14788 		return (0);
14789 
14790 	default:
14791 		ASSERT(0);
14792 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14793 		ip_drop_output("ipIfStatsOutDiscards - ND_other",
14794 		    mp, ill);
14795 		freemsg(mp);
14796 		return (ENETUNREACH);
14797 	}
14798 }
14799 
14800 /*
14801  * Return B_TRUE if the buffers differ in length or content.
14802  * This is used for comparing extension header buffers.
14803  * Note that an extension header would be declared different
14804  * even if all that changed was the next header value in that header i.e.
14805  * what really changed is the next extension header.
14806  */
14807 boolean_t
ip_cmpbuf(const void * abuf,uint_t alen,boolean_t b_valid,const void * bbuf,uint_t blen)14808 ip_cmpbuf(const void *abuf, uint_t alen, boolean_t b_valid, const void *bbuf,
14809     uint_t blen)
14810 {
14811 	if (!b_valid)
14812 		blen = 0;
14813 
14814 	if (alen != blen)
14815 		return (B_TRUE);
14816 	if (alen == 0)
14817 		return (B_FALSE);	/* Both zero length */
14818 	return (bcmp(abuf, bbuf, alen));
14819 }
14820 
14821 /*
14822  * Preallocate memory for ip_savebuf(). Returns B_TRUE if ok.
14823  * Return B_FALSE if memory allocation fails - don't change any state!
14824  */
14825 boolean_t
ip_allocbuf(void ** dstp,uint_t * dstlenp,boolean_t src_valid,const void * src,uint_t srclen)14826 ip_allocbuf(void **dstp, uint_t *dstlenp, boolean_t src_valid,
14827     const void *src, uint_t srclen)
14828 {
14829 	void *dst;
14830 
14831 	if (!src_valid)
14832 		srclen = 0;
14833 
14834 	ASSERT(*dstlenp == 0);
14835 	if (src != NULL && srclen != 0) {
14836 		dst = mi_alloc(srclen, BPRI_MED);
14837 		if (dst == NULL)
14838 			return (B_FALSE);
14839 	} else {
14840 		dst = NULL;
14841 	}
14842 	if (*dstp != NULL)
14843 		mi_free(*dstp);
14844 	*dstp = dst;
14845 	*dstlenp = dst == NULL ? 0 : srclen;
14846 	return (B_TRUE);
14847 }
14848 
14849 /*
14850  * Replace what is in *dst, *dstlen with the source.
14851  * Assumes ip_allocbuf has already been called.
14852  */
14853 void
ip_savebuf(void ** dstp,uint_t * dstlenp,boolean_t src_valid,const void * src,uint_t srclen)14854 ip_savebuf(void **dstp, uint_t *dstlenp, boolean_t src_valid,
14855     const void *src, uint_t srclen)
14856 {
14857 	if (!src_valid)
14858 		srclen = 0;
14859 
14860 	ASSERT(*dstlenp == srclen);
14861 	if (src != NULL && srclen != 0)
14862 		bcopy(src, *dstp, srclen);
14863 }
14864 
14865 /*
14866  * Free the storage pointed to by the members of an ip_pkt_t.
14867  */
14868 void
ip_pkt_free(ip_pkt_t * ipp)14869 ip_pkt_free(ip_pkt_t *ipp)
14870 {
14871 	uint_t	fields = ipp->ipp_fields;
14872 
14873 	if (fields & IPPF_HOPOPTS) {
14874 		kmem_free(ipp->ipp_hopopts, ipp->ipp_hopoptslen);
14875 		ipp->ipp_hopopts = NULL;
14876 		ipp->ipp_hopoptslen = 0;
14877 	}
14878 	if (fields & IPPF_RTHDRDSTOPTS) {
14879 		kmem_free(ipp->ipp_rthdrdstopts, ipp->ipp_rthdrdstoptslen);
14880 		ipp->ipp_rthdrdstopts = NULL;
14881 		ipp->ipp_rthdrdstoptslen = 0;
14882 	}
14883 	if (fields & IPPF_DSTOPTS) {
14884 		kmem_free(ipp->ipp_dstopts, ipp->ipp_dstoptslen);
14885 		ipp->ipp_dstopts = NULL;
14886 		ipp->ipp_dstoptslen = 0;
14887 	}
14888 	if (fields & IPPF_RTHDR) {
14889 		kmem_free(ipp->ipp_rthdr, ipp->ipp_rthdrlen);
14890 		ipp->ipp_rthdr = NULL;
14891 		ipp->ipp_rthdrlen = 0;
14892 	}
14893 	if (fields & IPPF_IPV4_OPTIONS) {
14894 		kmem_free(ipp->ipp_ipv4_options, ipp->ipp_ipv4_options_len);
14895 		ipp->ipp_ipv4_options = NULL;
14896 		ipp->ipp_ipv4_options_len = 0;
14897 	}
14898 	if (fields & IPPF_LABEL_V4) {
14899 		kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
14900 		ipp->ipp_label_v4 = NULL;
14901 		ipp->ipp_label_len_v4 = 0;
14902 	}
14903 	if (fields & IPPF_LABEL_V6) {
14904 		kmem_free(ipp->ipp_label_v6, ipp->ipp_label_len_v6);
14905 		ipp->ipp_label_v6 = NULL;
14906 		ipp->ipp_label_len_v6 = 0;
14907 	}
14908 	ipp->ipp_fields &= ~(IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14909 	    IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6);
14910 }
14911 
14912 /*
14913  * Copy from src to dst and allocate as needed.
14914  * Returns zero or ENOMEM.
14915  *
14916  * The caller must initialize dst to zero.
14917  */
14918 int
ip_pkt_copy(ip_pkt_t * src,ip_pkt_t * dst,int kmflag)14919 ip_pkt_copy(ip_pkt_t *src, ip_pkt_t *dst, int kmflag)
14920 {
14921 	uint_t	fields = src->ipp_fields;
14922 
14923 	/* Start with fields that don't require memory allocation */
14924 	dst->ipp_fields = fields &
14925 	    ~(IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14926 	    IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6);
14927 
14928 	dst->ipp_addr = src->ipp_addr;
14929 	dst->ipp_unicast_hops = src->ipp_unicast_hops;
14930 	dst->ipp_hoplimit = src->ipp_hoplimit;
14931 	dst->ipp_tclass = src->ipp_tclass;
14932 	dst->ipp_type_of_service = src->ipp_type_of_service;
14933 
14934 	if (!(fields & (IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14935 	    IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6)))
14936 		return (0);
14937 
14938 	if (fields & IPPF_HOPOPTS) {
14939 		dst->ipp_hopopts = kmem_alloc(src->ipp_hopoptslen, kmflag);
14940 		if (dst->ipp_hopopts == NULL) {
14941 			ip_pkt_free(dst);
14942 			return (ENOMEM);
14943 		}
14944 		dst->ipp_fields |= IPPF_HOPOPTS;
14945 		bcopy(src->ipp_hopopts, dst->ipp_hopopts,
14946 		    src->ipp_hopoptslen);
14947 		dst->ipp_hopoptslen = src->ipp_hopoptslen;
14948 	}
14949 	if (fields & IPPF_RTHDRDSTOPTS) {
14950 		dst->ipp_rthdrdstopts = kmem_alloc(src->ipp_rthdrdstoptslen,
14951 		    kmflag);
14952 		if (dst->ipp_rthdrdstopts == NULL) {
14953 			ip_pkt_free(dst);
14954 			return (ENOMEM);
14955 		}
14956 		dst->ipp_fields |= IPPF_RTHDRDSTOPTS;
14957 		bcopy(src->ipp_rthdrdstopts, dst->ipp_rthdrdstopts,
14958 		    src->ipp_rthdrdstoptslen);
14959 		dst->ipp_rthdrdstoptslen = src->ipp_rthdrdstoptslen;
14960 	}
14961 	if (fields & IPPF_DSTOPTS) {
14962 		dst->ipp_dstopts = kmem_alloc(src->ipp_dstoptslen, kmflag);
14963 		if (dst->ipp_dstopts == NULL) {
14964 			ip_pkt_free(dst);
14965 			return (ENOMEM);
14966 		}
14967 		dst->ipp_fields |= IPPF_DSTOPTS;
14968 		bcopy(src->ipp_dstopts, dst->ipp_dstopts,
14969 		    src->ipp_dstoptslen);
14970 		dst->ipp_dstoptslen = src->ipp_dstoptslen;
14971 	}
14972 	if (fields & IPPF_RTHDR) {
14973 		dst->ipp_rthdr = kmem_alloc(src->ipp_rthdrlen, kmflag);
14974 		if (dst->ipp_rthdr == NULL) {
14975 			ip_pkt_free(dst);
14976 			return (ENOMEM);
14977 		}
14978 		dst->ipp_fields |= IPPF_RTHDR;
14979 		bcopy(src->ipp_rthdr, dst->ipp_rthdr,
14980 		    src->ipp_rthdrlen);
14981 		dst->ipp_rthdrlen = src->ipp_rthdrlen;
14982 	}
14983 	if (fields & IPPF_IPV4_OPTIONS) {
14984 		dst->ipp_ipv4_options = kmem_alloc(src->ipp_ipv4_options_len,
14985 		    kmflag);
14986 		if (dst->ipp_ipv4_options == NULL) {
14987 			ip_pkt_free(dst);
14988 			return (ENOMEM);
14989 		}
14990 		dst->ipp_fields |= IPPF_IPV4_OPTIONS;
14991 		bcopy(src->ipp_ipv4_options, dst->ipp_ipv4_options,
14992 		    src->ipp_ipv4_options_len);
14993 		dst->ipp_ipv4_options_len = src->ipp_ipv4_options_len;
14994 	}
14995 	if (fields & IPPF_LABEL_V4) {
14996 		dst->ipp_label_v4 = kmem_alloc(src->ipp_label_len_v4, kmflag);
14997 		if (dst->ipp_label_v4 == NULL) {
14998 			ip_pkt_free(dst);
14999 			return (ENOMEM);
15000 		}
15001 		dst->ipp_fields |= IPPF_LABEL_V4;
15002 		bcopy(src->ipp_label_v4, dst->ipp_label_v4,
15003 		    src->ipp_label_len_v4);
15004 		dst->ipp_label_len_v4 = src->ipp_label_len_v4;
15005 	}
15006 	if (fields & IPPF_LABEL_V6) {
15007 		dst->ipp_label_v6 = kmem_alloc(src->ipp_label_len_v6, kmflag);
15008 		if (dst->ipp_label_v6 == NULL) {
15009 			ip_pkt_free(dst);
15010 			return (ENOMEM);
15011 		}
15012 		dst->ipp_fields |= IPPF_LABEL_V6;
15013 		bcopy(src->ipp_label_v6, dst->ipp_label_v6,
15014 		    src->ipp_label_len_v6);
15015 		dst->ipp_label_len_v6 = src->ipp_label_len_v6;
15016 	}
15017 	if (fields & IPPF_FRAGHDR) {
15018 		dst->ipp_fraghdr = kmem_alloc(src->ipp_fraghdrlen, kmflag);
15019 		if (dst->ipp_fraghdr == NULL) {
15020 			ip_pkt_free(dst);
15021 			return (ENOMEM);
15022 		}
15023 		dst->ipp_fields |= IPPF_FRAGHDR;
15024 		bcopy(src->ipp_fraghdr, dst->ipp_fraghdr,
15025 		    src->ipp_fraghdrlen);
15026 		dst->ipp_fraghdrlen = src->ipp_fraghdrlen;
15027 	}
15028 	return (0);
15029 }
15030 
15031 /*
15032  * Returns INADDR_ANY if no source route
15033  */
15034 ipaddr_t
ip_pkt_source_route_v4(const ip_pkt_t * ipp)15035 ip_pkt_source_route_v4(const ip_pkt_t *ipp)
15036 {
15037 	ipaddr_t	nexthop = INADDR_ANY;
15038 	ipoptp_t	opts;
15039 	uchar_t		*opt;
15040 	uint8_t		optval;
15041 	uint8_t		optlen;
15042 	uint32_t	totallen;
15043 
15044 	if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
15045 		return (INADDR_ANY);
15046 
15047 	totallen = ipp->ipp_ipv4_options_len;
15048 	if (totallen & 0x3)
15049 		return (INADDR_ANY);
15050 
15051 	for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
15052 	    optval != IPOPT_EOL;
15053 	    optval = ipoptp_next(&opts)) {
15054 		opt = opts.ipoptp_cur;
15055 		switch (optval) {
15056 			uint8_t off;
15057 		case IPOPT_SSRR:
15058 		case IPOPT_LSRR:
15059 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
15060 				break;
15061 			}
15062 			optlen = opts.ipoptp_len;
15063 			off = opt[IPOPT_OFFSET];
15064 			off--;
15065 			if (optlen < IP_ADDR_LEN ||
15066 			    off > optlen - IP_ADDR_LEN) {
15067 				/* End of source route */
15068 				break;
15069 			}
15070 			bcopy((char *)opt + off, &nexthop, IP_ADDR_LEN);
15071 			if (nexthop == htonl(INADDR_LOOPBACK)) {
15072 				/* Ignore */
15073 				nexthop = INADDR_ANY;
15074 				break;
15075 			}
15076 			break;
15077 		}
15078 	}
15079 	return (nexthop);
15080 }
15081 
15082 /*
15083  * Reverse a source route.
15084  */
15085 void
ip_pkt_source_route_reverse_v4(ip_pkt_t * ipp)15086 ip_pkt_source_route_reverse_v4(ip_pkt_t *ipp)
15087 {
15088 	ipaddr_t	tmp;
15089 	ipoptp_t	opts;
15090 	uchar_t		*opt;
15091 	uint8_t		optval;
15092 	uint32_t	totallen;
15093 
15094 	if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
15095 		return;
15096 
15097 	totallen = ipp->ipp_ipv4_options_len;
15098 	if (totallen & 0x3)
15099 		return;
15100 
15101 	for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
15102 	    optval != IPOPT_EOL;
15103 	    optval = ipoptp_next(&opts)) {
15104 		uint8_t off1, off2;
15105 
15106 		opt = opts.ipoptp_cur;
15107 		switch (optval) {
15108 		case IPOPT_SSRR:
15109 		case IPOPT_LSRR:
15110 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
15111 				break;
15112 			}
15113 			off1 = IPOPT_MINOFF_SR - 1;
15114 			off2 = opt[IPOPT_OFFSET] - IP_ADDR_LEN - 1;
15115 			while (off2 > off1) {
15116 				bcopy(opt + off2, &tmp, IP_ADDR_LEN);
15117 				bcopy(opt + off1, opt + off2, IP_ADDR_LEN);
15118 				bcopy(&tmp, opt + off2, IP_ADDR_LEN);
15119 				off2 -= IP_ADDR_LEN;
15120 				off1 += IP_ADDR_LEN;
15121 			}
15122 			opt[IPOPT_OFFSET] = IPOPT_MINOFF_SR;
15123 			break;
15124 		}
15125 	}
15126 }
15127 
15128 /*
15129  * Returns NULL if no routing header
15130  */
15131 in6_addr_t *
ip_pkt_source_route_v6(const ip_pkt_t * ipp)15132 ip_pkt_source_route_v6(const ip_pkt_t *ipp)
15133 {
15134 	in6_addr_t	*nexthop = NULL;
15135 	ip6_rthdr0_t	*rthdr;
15136 
15137 	if (!(ipp->ipp_fields & IPPF_RTHDR))
15138 		return (NULL);
15139 
15140 	rthdr = (ip6_rthdr0_t *)ipp->ipp_rthdr;
15141 	if (rthdr->ip6r0_segleft == 0)
15142 		return (NULL);
15143 
15144 	nexthop = (in6_addr_t *)((char *)rthdr + sizeof (*rthdr));
15145 	return (nexthop);
15146 }
15147 
15148 zoneid_t
ip_get_zoneid_v4(ipaddr_t addr,mblk_t * mp,ip_recv_attr_t * ira,zoneid_t lookup_zoneid)15149 ip_get_zoneid_v4(ipaddr_t addr, mblk_t *mp, ip_recv_attr_t *ira,
15150     zoneid_t lookup_zoneid)
15151 {
15152 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
15153 	ire_t		*ire;
15154 	int		ire_flags = MATCH_IRE_TYPE;
15155 	zoneid_t	zoneid = ALL_ZONES;
15156 
15157 	if (is_system_labeled() && !tsol_can_accept_raw(mp, ira, B_FALSE))
15158 		return (ALL_ZONES);
15159 
15160 	if (lookup_zoneid != ALL_ZONES)
15161 		ire_flags |= MATCH_IRE_ZONEONLY;
15162 	ire = ire_ftable_lookup_v4(addr, 0, 0, IRE_LOCAL | IRE_LOOPBACK,
15163 	    NULL, lookup_zoneid, NULL, ire_flags, 0, ipst, NULL);
15164 	if (ire != NULL) {
15165 		zoneid = IP_REAL_ZONEID(ire->ire_zoneid, ipst);
15166 		ire_refrele(ire);
15167 	}
15168 	return (zoneid);
15169 }
15170 
15171 zoneid_t
ip_get_zoneid_v6(in6_addr_t * addr,mblk_t * mp,const ill_t * ill,ip_recv_attr_t * ira,zoneid_t lookup_zoneid)15172 ip_get_zoneid_v6(in6_addr_t *addr, mblk_t *mp, const ill_t *ill,
15173     ip_recv_attr_t *ira, zoneid_t lookup_zoneid)
15174 {
15175 	ip_stack_t	*ipst = ira->ira_ill->ill_ipst;
15176 	ire_t		*ire;
15177 	int		ire_flags = MATCH_IRE_TYPE;
15178 	zoneid_t	zoneid = ALL_ZONES;
15179 
15180 	if (is_system_labeled() && !tsol_can_accept_raw(mp, ira, B_FALSE))
15181 		return (ALL_ZONES);
15182 
15183 	if (IN6_IS_ADDR_LINKLOCAL(addr))
15184 		ire_flags |= MATCH_IRE_ILL;
15185 
15186 	if (lookup_zoneid != ALL_ZONES)
15187 		ire_flags |= MATCH_IRE_ZONEONLY;
15188 	ire = ire_ftable_lookup_v6(addr, NULL, NULL, IRE_LOCAL | IRE_LOOPBACK,
15189 	    ill, lookup_zoneid, NULL, ire_flags, 0, ipst, NULL);
15190 	if (ire != NULL) {
15191 		zoneid = IP_REAL_ZONEID(ire->ire_zoneid, ipst);
15192 		ire_refrele(ire);
15193 	}
15194 	return (zoneid);
15195 }
15196 
15197 /*
15198  * IP obserability hook support functions.
15199  */
15200 static void
ipobs_init(ip_stack_t * ipst)15201 ipobs_init(ip_stack_t *ipst)
15202 {
15203 	netid_t id;
15204 
15205 	id = net_getnetidbynetstackid(ipst->ips_netstack->netstack_stackid);
15206 
15207 	ipst->ips_ip4_observe_pr = net_protocol_lookup(id, NHF_INET);
15208 	VERIFY(ipst->ips_ip4_observe_pr != NULL);
15209 
15210 	ipst->ips_ip6_observe_pr = net_protocol_lookup(id, NHF_INET6);
15211 	VERIFY(ipst->ips_ip6_observe_pr != NULL);
15212 }
15213 
15214 static void
ipobs_fini(ip_stack_t * ipst)15215 ipobs_fini(ip_stack_t *ipst)
15216 {
15217 
15218 	VERIFY(net_protocol_release(ipst->ips_ip4_observe_pr) == 0);
15219 	VERIFY(net_protocol_release(ipst->ips_ip6_observe_pr) == 0);
15220 }
15221 
15222 /*
15223  * hook_pkt_observe_t is composed in network byte order so that the
15224  * entire mblk_t chain handed into hook_run can be used as-is.
15225  * The caveat is that use of the fields, such as the zone fields,
15226  * requires conversion into host byte order first.
15227  */
15228 void
ipobs_hook(mblk_t * mp,int htype,zoneid_t zsrc,zoneid_t zdst,const ill_t * ill,ip_stack_t * ipst)15229 ipobs_hook(mblk_t *mp, int htype, zoneid_t zsrc, zoneid_t zdst,
15230     const ill_t *ill, ip_stack_t *ipst)
15231 {
15232 	hook_pkt_observe_t *hdr;
15233 	uint64_t grifindex;
15234 	mblk_t *imp;
15235 
15236 	imp = allocb(sizeof (*hdr), BPRI_HI);
15237 	if (imp == NULL)
15238 		return;
15239 
15240 	hdr = (hook_pkt_observe_t *)imp->b_rptr;
15241 	/*
15242 	 * b_wptr is set to make the apparent size of the data in the mblk_t
15243 	 * to exclude the pointers at the end of hook_pkt_observer_t.
15244 	 */
15245 	imp->b_wptr = imp->b_rptr + sizeof (dl_ipnetinfo_t);
15246 	imp->b_cont = mp;
15247 
15248 	ASSERT(DB_TYPE(mp) == M_DATA);
15249 
15250 	if (IS_UNDER_IPMP(ill))
15251 		grifindex = ipmp_ill_get_ipmp_ifindex(ill);
15252 	else
15253 		grifindex = 0;
15254 
15255 	hdr->hpo_version = 1;
15256 	hdr->hpo_htype = htons(htype);
15257 	hdr->hpo_pktlen = htonl((ulong_t)msgdsize(mp));
15258 	hdr->hpo_ifindex = htonl(ill->ill_phyint->phyint_ifindex);
15259 	hdr->hpo_grifindex = htonl(grifindex);
15260 	hdr->hpo_zsrc = htonl(zsrc);
15261 	hdr->hpo_zdst = htonl(zdst);
15262 	hdr->hpo_pkt = imp;
15263 	hdr->hpo_ctx = ipst->ips_netstack;
15264 
15265 	if (ill->ill_isv6) {
15266 		hdr->hpo_family = AF_INET6;
15267 		(void) hook_run(ipst->ips_ipv6_net_data->netd_hooks,
15268 		    ipst->ips_ipv6observing, (hook_data_t)hdr);
15269 	} else {
15270 		hdr->hpo_family = AF_INET;
15271 		(void) hook_run(ipst->ips_ipv4_net_data->netd_hooks,
15272 		    ipst->ips_ipv4observing, (hook_data_t)hdr);
15273 	}
15274 
15275 	imp->b_cont = NULL;
15276 	freemsg(imp);
15277 }
15278 
15279 /*
15280  * Utility routine that checks if `v4srcp' is a valid address on underlying
15281  * interface `ill'.  If `ipifp' is non-NULL, it's set to a held ipif
15282  * associated with `v4srcp' on success.  NOTE: if this is not called from
15283  * inside the IPSQ (ill_g_lock is not held), `ill' may be removed from the
15284  * group during or after this lookup.
15285  */
15286 boolean_t
ipif_lookup_testaddr_v4(ill_t * ill,const in_addr_t * v4srcp,ipif_t ** ipifp)15287 ipif_lookup_testaddr_v4(ill_t *ill, const in_addr_t *v4srcp, ipif_t **ipifp)
15288 {
15289 	ipif_t *ipif;
15290 
15291 	ipif = ipif_lookup_addr_exact(*v4srcp, ill, ill->ill_ipst);
15292 	if (ipif != NULL) {
15293 		if (ipifp != NULL)
15294 			*ipifp = ipif;
15295 		else
15296 			ipif_refrele(ipif);
15297 		return (B_TRUE);
15298 	}
15299 
15300 	ip1dbg(("ipif_lookup_testaddr_v4: cannot find ipif for src %x\n",
15301 	    *v4srcp));
15302 	return (B_FALSE);
15303 }
15304 
15305 /*
15306  * Transport protocol call back function for CPU state change.
15307  */
15308 /* ARGSUSED */
15309 static int
ip_tp_cpu_update(cpu_setup_t what,int id,void * arg)15310 ip_tp_cpu_update(cpu_setup_t what, int id, void *arg)
15311 {
15312 	processorid_t cpu_seqid;
15313 	netstack_handle_t nh;
15314 	netstack_t *ns;
15315 
15316 	ASSERT(MUTEX_HELD(&cpu_lock));
15317 
15318 	switch (what) {
15319 	case CPU_CONFIG:
15320 	case CPU_ON:
15321 	case CPU_INIT:
15322 	case CPU_CPUPART_IN:
15323 		cpu_seqid = cpu[id]->cpu_seqid;
15324 		netstack_next_init(&nh);
15325 		while ((ns = netstack_next(&nh)) != NULL) {
15326 			tcp_stack_cpu_add(ns->netstack_tcp, cpu_seqid);
15327 			sctp_stack_cpu_add(ns->netstack_sctp, cpu_seqid);
15328 			udp_stack_cpu_add(ns->netstack_udp, cpu_seqid);
15329 			netstack_rele(ns);
15330 		}
15331 		netstack_next_fini(&nh);
15332 		break;
15333 	case CPU_UNCONFIG:
15334 	case CPU_OFF:
15335 	case CPU_CPUPART_OUT:
15336 		/*
15337 		 * Nothing to do.  We don't remove the per CPU stats from
15338 		 * the IP stack even when the CPU goes offline.
15339 		 */
15340 		break;
15341 	default:
15342 		break;
15343 	}
15344 	return (0);
15345 }
15346