xref: /illumos-gate/usr/src/uts/common/inet/udp_impl.h (revision 3e8b6b84)
17c478bd9Sstevel@tonic-gate /*
27c478bd9Sstevel@tonic-gate  * CDDL HEADER START
37c478bd9Sstevel@tonic-gate  *
47c478bd9Sstevel@tonic-gate  * The contents of this file are subject to the terms of the
5e4f35dbaSgt  * Common Development and Distribution License (the "License").
6e4f35dbaSgt  * You may not use this file except in compliance with the License.
77c478bd9Sstevel@tonic-gate  *
87c478bd9Sstevel@tonic-gate  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
97c478bd9Sstevel@tonic-gate  * or http://www.opensolaris.org/os/licensing.
107c478bd9Sstevel@tonic-gate  * See the License for the specific language governing permissions
117c478bd9Sstevel@tonic-gate  * and limitations under the License.
127c478bd9Sstevel@tonic-gate  *
137c478bd9Sstevel@tonic-gate  * When distributing Covered Code, include this CDDL HEADER in each
147c478bd9Sstevel@tonic-gate  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
157c478bd9Sstevel@tonic-gate  * If applicable, add the following below this CDDL HEADER, with the
167c478bd9Sstevel@tonic-gate  * fields enclosed by brackets "[]" replaced with your own identifying
177c478bd9Sstevel@tonic-gate  * information: Portions Copyright [yyyy] [name of copyright owner]
187c478bd9Sstevel@tonic-gate  *
197c478bd9Sstevel@tonic-gate  * CDDL HEADER END
207c478bd9Sstevel@tonic-gate  */
217c478bd9Sstevel@tonic-gate /*
223e95bd4aSAnders Persson  * Copyright (c) 2000, 2010, Oracle and/or its affiliates. All rights reserved.
23*3e8b6b84SRobert Mustacchi  * Copyright 2015, Joyent, Inc.
247c478bd9Sstevel@tonic-gate  */
257c478bd9Sstevel@tonic-gate 
267c478bd9Sstevel@tonic-gate #ifndef	_UDP_IMPL_H
277c478bd9Sstevel@tonic-gate #define	_UDP_IMPL_H
287c478bd9Sstevel@tonic-gate 
29ff550d0eSmasputra /*
30ff550d0eSmasputra  * UDP implementation private declarations.  These interfaces are
31ff550d0eSmasputra  * used to build the IP module and are not meant to be accessed
32ff550d0eSmasputra  * by any modules except IP itself.  They are undocumented and are
33ff550d0eSmasputra  * subject to change without notice.
34ff550d0eSmasputra  */
35ff550d0eSmasputra 
367c478bd9Sstevel@tonic-gate #ifdef	__cplusplus
377c478bd9Sstevel@tonic-gate extern "C" {
387c478bd9Sstevel@tonic-gate #endif
397c478bd9Sstevel@tonic-gate 
407c478bd9Sstevel@tonic-gate #ifdef _KERNEL
417c478bd9Sstevel@tonic-gate 
427c478bd9Sstevel@tonic-gate #include <sys/int_types.h>
43f4b3ec61Sdh #include <sys/netstack.h>
447c478bd9Sstevel@tonic-gate 
457c478bd9Sstevel@tonic-gate #include <netinet/in.h>
467c478bd9Sstevel@tonic-gate #include <netinet/ip6.h>
477c478bd9Sstevel@tonic-gate 
487c478bd9Sstevel@tonic-gate #include <inet/common.h>
497c478bd9Sstevel@tonic-gate #include <inet/ip.h>
5045916cd2Sjpk #include <inet/optcom.h>
516e91bba0SGirish Moodalbail #include <inet/tunables.h>
527c478bd9Sstevel@tonic-gate 
53ff550d0eSmasputra #define	UDP_MOD_ID		5607
54ff550d0eSmasputra 
55f4b3ec61Sdh /*
56f4b3ec61Sdh  * Bind hash list size and hash function.  It has to be a power of 2 for
57f4b3ec61Sdh  * hashing.
58f4b3ec61Sdh  */
59f4b3ec61Sdh #define	UDP_BIND_FANOUT_SIZE	512
60f4b3ec61Sdh #define	UDP_BIND_HASH(lport, size) \
61f4b3ec61Sdh 	((ntohs((uint16_t)lport)) & (size - 1))
62f4b3ec61Sdh 
63f4b3ec61Sdh /* UDP bind fanout hash structure. */
64f4b3ec61Sdh typedef struct udp_fanout_s {
65f4b3ec61Sdh 	struct udp_s *uf_udp;
66f4b3ec61Sdh 	kmutex_t uf_lock;
67f4b3ec61Sdh #if defined(_LP64) || defined(_I32LPx)
68f4b3ec61Sdh 	char	uf_pad[48];
69f4b3ec61Sdh #else
70f4b3ec61Sdh 	char	uf_pad[56];
71f4b3ec61Sdh #endif
72f4b3ec61Sdh } udp_fanout_t;
73f4b3ec61Sdh 
74f4b3ec61Sdh /* Kstats */
75f4b3ec61Sdh typedef struct udp_stat {			/* Class "net" kstats */
76f4b3ec61Sdh 	kstat_named_t	udp_sock_fallback;
77f4b3ec61Sdh 	kstat_named_t	udp_out_opt;
78f4b3ec61Sdh 	kstat_named_t	udp_out_err_notconn;
79f4b3ec61Sdh 	kstat_named_t	udp_out_err_output;
80f4b3ec61Sdh 	kstat_named_t	udp_out_err_tudr;
81f4b3ec61Sdh #ifdef DEBUG
82f4b3ec61Sdh 	kstat_named_t	udp_data_conn;
83f4b3ec61Sdh 	kstat_named_t	udp_data_notconn;
84bd670b35SErik Nordmark 	kstat_named_t	udp_out_lastdst;
85bd670b35SErik Nordmark 	kstat_named_t	udp_out_diffdst;
86bd670b35SErik Nordmark 	kstat_named_t	udp_out_ipv6;
87bd670b35SErik Nordmark 	kstat_named_t	udp_out_mapped;
88bd670b35SErik Nordmark 	kstat_named_t	udp_out_ipv4;
89f4b3ec61Sdh #endif
90f4b3ec61Sdh } udp_stat_t;
91f4b3ec61Sdh 
925dd46ab5SKacheong Poon /*
935dd46ab5SKacheong Poon  * This struct contains only the counter part of udp_stat_t.  It is used
945dd46ab5SKacheong Poon  * in udp_stats_cpu_t instead of udp_stat_t to save memory space.
955dd46ab5SKacheong Poon  */
965dd46ab5SKacheong Poon typedef struct {
975dd46ab5SKacheong Poon 	uint64_t	udp_sock_fallback;
985dd46ab5SKacheong Poon 	uint64_t	udp_out_opt;
995dd46ab5SKacheong Poon 	uint64_t	udp_out_err_notconn;
1005dd46ab5SKacheong Poon 	uint64_t	udp_out_err_output;
1015dd46ab5SKacheong Poon 	uint64_t	udp_out_err_tudr;
1025dd46ab5SKacheong Poon #ifdef DEBUG
1035dd46ab5SKacheong Poon 	uint64_t	udp_data_conn;
1045dd46ab5SKacheong Poon 	uint64_t	udp_data_notconn;
1055dd46ab5SKacheong Poon 	uint64_t	udp_out_lastdst;
1065dd46ab5SKacheong Poon 	uint64_t	udp_out_diffdst;
1075dd46ab5SKacheong Poon 	uint64_t	udp_out_ipv6;
1085dd46ab5SKacheong Poon 	uint64_t	udp_out_mapped;
1095dd46ab5SKacheong Poon 	uint64_t	udp_out_ipv4;
1105dd46ab5SKacheong Poon #endif
1115dd46ab5SKacheong Poon } udp_stat_counter_t;
1125dd46ab5SKacheong Poon 
1135dd46ab5SKacheong Poon /* Per CPU stats: UDP MIB2 and UDP kstat. */
1145dd46ab5SKacheong Poon typedef struct {
1155dd46ab5SKacheong Poon 	mib2_udp_t		udp_sc_mib;
1165dd46ab5SKacheong Poon 	udp_stat_counter_t	udp_sc_stats;
1175dd46ab5SKacheong Poon } udp_stats_cpu_t;
1185dd46ab5SKacheong Poon 
119f4b3ec61Sdh #define	UDP_NUM_EPRIV_PORTS	64
120f4b3ec61Sdh 
1216e91bba0SGirish Moodalbail /* Default buffer size and flow control wake up threshold. */
1226e91bba0SGirish Moodalbail #define	UDP_RECV_HIWATER	(56 * 1024)
1236e91bba0SGirish Moodalbail #define	UDP_RECV_LOWATER	128
1246e91bba0SGirish Moodalbail #define	UDP_XMIT_HIWATER	(56 * 1024)
1256e91bba0SGirish Moodalbail #define	UDP_XMIT_LOWATER	1024
1266e91bba0SGirish Moodalbail 
127f4b3ec61Sdh /*
128f4b3ec61Sdh  * UDP stack instances
129f4b3ec61Sdh  */
130f4b3ec61Sdh struct udp_stack {
131f4b3ec61Sdh 	netstack_t	*us_netstack;	/* Common netstack */
132f4b3ec61Sdh 
133f4b3ec61Sdh 	uint_t		us_bind_fanout_size;
134f4b3ec61Sdh 	udp_fanout_t	*us_bind_fanout;
135f4b3ec61Sdh 
136f4b3ec61Sdh 	int		us_num_epriv_ports;
137f4b3ec61Sdh 	in_port_t	us_epriv_ports[UDP_NUM_EPRIV_PORTS];
1386e91bba0SGirish Moodalbail 	kmutex_t	us_epriv_port_lock;
139f4b3ec61Sdh 
140f4b3ec61Sdh 	/* Hint not protected by any lock */
141f4b3ec61Sdh 	in_port_t	us_next_port_to_try;
142f4b3ec61Sdh 
1436e91bba0SGirish Moodalbail 	/* UDP tunables table */
1446e91bba0SGirish Moodalbail 	struct mod_prop_info_s	*us_propinfo_tbl;
145f4b3ec61Sdh 
146f4b3ec61Sdh 	kstat_t		*us_mibkp;	/* kstats exporting mib data */
147f4b3ec61Sdh 	kstat_t		*us_kstat;
148f4b3ec61Sdh 
149f4b3ec61Sdh /*
150f4b3ec61Sdh  * The smallest anonymous port in the priviledged port range which UDP
151f4b3ec61Sdh  * looks for free port.  Use in the option UDP_ANONPRIVBIND.
152f4b3ec61Sdh  */
153f4b3ec61Sdh 	in_port_t	us_min_anonpriv_port;
154f4b3ec61Sdh 
1550f1702c5SYu Xiangning 	ldi_ident_t	us_ldi_ident;
1565dd46ab5SKacheong Poon 
1575dd46ab5SKacheong Poon 	udp_stats_cpu_t	**us_sc;
1585dd46ab5SKacheong Poon 	int		us_sc_cnt;
159f4b3ec61Sdh };
1600f1702c5SYu Xiangning 
161f4b3ec61Sdh typedef struct udp_stack udp_stack_t;
162f4b3ec61Sdh 
1637c478bd9Sstevel@tonic-gate /* Internal udp control structure, one per open stream */
1647c478bd9Sstevel@tonic-gate typedef	struct udp_s {
165fc80c0dfSnordmark 	/*
166bd670b35SErik Nordmark 	 * The addresses and ports in the conn_t and udp_state are protected by
167bd670b35SErik Nordmark 	 * conn_lock and the fanout lock i.e. uf_lock. Need both locks to change
168bd670b35SErik Nordmark 	 * the fields, either lock is sufficient for reading the field.
169bd670b35SErik Nordmark 	 * conn_lock also protects the content of udp_t.
170fc80c0dfSnordmark 	 */
171ff550d0eSmasputra 	uint32_t	udp_state;	/* TPI state */
172fc80c0dfSnordmark 
173bd670b35SErik Nordmark 	ip_pkt_t	udp_recv_ipp;	/* Used for IPv4 options received */
174fc80c0dfSnordmark 
175fc80c0dfSnordmark 	/* Written to only once at the time of opening the endpoint */
176ff550d0eSmasputra 	conn_t		*udp_connp;
1777c478bd9Sstevel@tonic-gate 
178bd670b35SErik Nordmark 	uint32_t
179bd670b35SErik Nordmark 		udp_issocket : 1,	/* socket mode; sockfs is on top */
180bd670b35SErik Nordmark 		udp_nat_t_endpoint : 1,	/* UDP_NAT_T_ENDPOINT option */
181bd670b35SErik Nordmark 		udp_rcvhdr : 1,		/* UDP_RCVHDR option */
182bd670b35SErik Nordmark 
183*3e8b6b84SRobert Mustacchi 		udp_vxlanhash: 1,	/* UDP_SRCPORT_HASH option */
184*3e8b6b84SRobert Mustacchi 					/* Because there's only VXLAN, cheat */
185*3e8b6b84SRobert Mustacchi 					/* and only use a single bit */
186*3e8b6b84SRobert Mustacchi 
187*3e8b6b84SRobert Mustacchi 		udp_pad_to_bit_31 : 28;
188fc80c0dfSnordmark 
189fc80c0dfSnordmark 	/* Following 2 fields protected by the uf_lock */
1907c478bd9Sstevel@tonic-gate 	struct udp_s	*udp_bind_hash; /* Bind hash chain */
1917c478bd9Sstevel@tonic-gate 	struct udp_s	**udp_ptpbhn; /* Pointer to previous bind hash next. */
192ff550d0eSmasputra 
1930f1702c5SYu Xiangning 	kmutex_t	udp_recv_lock;		/* recv lock */
1940f1702c5SYu Xiangning 	size_t		udp_rcv_disply_hiwat;	/* user's view of rcvbuf */
195ff550d0eSmasputra 	size_t		udp_rcv_hiwat;		/* receive high watermark */
196bd670b35SErik Nordmark 
197bd670b35SErik Nordmark 	/* Set at open time and never changed */
198f4b3ec61Sdh 	udp_stack_t	*udp_us;		/* Stack instance for zone */
199bd670b35SErik Nordmark 
2000f1702c5SYu Xiangning 	int		udp_delayed_error;
2010f1702c5SYu Xiangning 	mblk_t		*udp_fallback_queue_head;
2020f1702c5SYu Xiangning 	mblk_t		*udp_fallback_queue_tail;
2030f1702c5SYu Xiangning 	struct sockaddr_storage	udp_delayed_addr;
2047c478bd9Sstevel@tonic-gate } udp_t;
2057c478bd9Sstevel@tonic-gate 
2067c478bd9Sstevel@tonic-gate /* UDP Protocol header aligned */
2077c478bd9Sstevel@tonic-gate typedef	struct udpahdr_s {
2087c478bd9Sstevel@tonic-gate 	in_port_t	uha_src_port;		/* Source port */
2097c478bd9Sstevel@tonic-gate 	in_port_t	uha_dst_port;		/* Destination port */
2107c478bd9Sstevel@tonic-gate 	uint16_t	uha_length;		/* UDP length */
2117c478bd9Sstevel@tonic-gate 	uint16_t	uha_checksum;		/* UDP checksum */
2127c478bd9Sstevel@tonic-gate } udpha_t;
2137c478bd9Sstevel@tonic-gate 
2146e91bba0SGirish Moodalbail #define	us_wroff_extra			us_propinfo_tbl[0].prop_cur_uval
2156e91bba0SGirish Moodalbail #define	us_ipv4_ttl			us_propinfo_tbl[1].prop_cur_uval
2166e91bba0SGirish Moodalbail #define	us_ipv6_hoplimit		us_propinfo_tbl[2].prop_cur_uval
2176e91bba0SGirish Moodalbail #define	us_smallest_nonpriv_port	us_propinfo_tbl[3].prop_cur_uval
2186e91bba0SGirish Moodalbail #define	us_do_checksum			us_propinfo_tbl[4].prop_cur_bval
2196e91bba0SGirish Moodalbail #define	us_smallest_anon_port		us_propinfo_tbl[5].prop_cur_uval
2206e91bba0SGirish Moodalbail #define	us_largest_anon_port		us_propinfo_tbl[6].prop_cur_uval
2216e91bba0SGirish Moodalbail #define	us_xmit_hiwat			us_propinfo_tbl[7].prop_cur_uval
2226e91bba0SGirish Moodalbail #define	us_xmit_lowat			us_propinfo_tbl[8].prop_cur_uval
2236e91bba0SGirish Moodalbail #define	us_recv_hiwat			us_propinfo_tbl[9].prop_cur_uval
2246e91bba0SGirish Moodalbail #define	us_max_buf			us_propinfo_tbl[10].prop_cur_uval
2256e91bba0SGirish Moodalbail #define	us_pmtu_discovery		us_propinfo_tbl[11].prop_cur_bval
2266e91bba0SGirish Moodalbail #define	us_sendto_ignerr		us_propinfo_tbl[12].prop_cur_bval
227ff550d0eSmasputra 
2285dd46ab5SKacheong Poon #define	UDPS_BUMP_MIB(us, x)	\
2295dd46ab5SKacheong Poon 	BUMP_MIB(&(us)->us_sc[CPU->cpu_seqid]->udp_sc_mib, x)
2305dd46ab5SKacheong Poon 
2315dd46ab5SKacheong Poon #define	UDP_STAT(us, x)		((us)->us_sc[CPU->cpu_seqid]->udp_sc_stats.x++)
232f4b3ec61Sdh #define	UDP_STAT_UPDATE(us, x, n)	\
2335dd46ab5SKacheong Poon 	((us)->us->sc[CPU->cpu_seqid]->udp_sc_stats.x.value.ui64 += (n))
234ff550d0eSmasputra #ifdef DEBUG
235f4b3ec61Sdh #define	UDP_DBGSTAT(us, x)	UDP_STAT(us, x)
236ff550d0eSmasputra #else
237f4b3ec61Sdh #define	UDP_DBGSTAT(us, x)
238ff550d0eSmasputra #endif /* DEBUG */
239ff550d0eSmasputra 
240ff550d0eSmasputra extern int	udp_opt_default(queue_t *, t_scalar_t, t_scalar_t, uchar_t *);
2410f1702c5SYu Xiangning extern int	udp_tpi_opt_get(queue_t *, t_scalar_t, t_scalar_t, uchar_t *);
2420f1702c5SYu Xiangning extern int	udp_tpi_opt_set(queue_t *, uint_t, int, int, uint_t, uchar_t *,
243bd670b35SErik Nordmark 		    uint_t *, uchar_t *, void *, cred_t *);
2446f773e29SBaban Kenkre extern mblk_t	*udp_snmp_get(queue_t *, mblk_t *, boolean_t);
245ff550d0eSmasputra extern int	udp_snmp_set(queue_t *, t_scalar_t, t_scalar_t, uchar_t *, int);
2460f1702c5SYu Xiangning extern void	udp_ddi_g_init(void);
2470f1702c5SYu Xiangning extern void	udp_ddi_g_destroy(void);
2480f1702c5SYu Xiangning extern void	udp_output(conn_t *connp, mblk_t *mp, struct sockaddr *addr,
2490f1702c5SYu Xiangning 		    socklen_t addrlen);
2508a06b3d6SToomas Soome extern int	udp_wput(queue_t *, mblk_t *);
25145916cd2Sjpk 
2525dd46ab5SKacheong Poon extern void	*udp_kstat_init(netstackid_t stackid);
2535dd46ab5SKacheong Poon extern void	udp_kstat_fini(netstackid_t stackid, kstat_t *ksp);
2545dd46ab5SKacheong Poon extern void	*udp_kstat2_init(netstackid_t stackid);
2555dd46ab5SKacheong Poon extern void	udp_kstat2_fini(netstackid_t, kstat_t *);
2565dd46ab5SKacheong Poon 
2575dd46ab5SKacheong Poon extern void	udp_stack_cpu_add(udp_stack_t *, processorid_t);
2585dd46ab5SKacheong Poon 
25945916cd2Sjpk /*
26045916cd2Sjpk  * Object to represent database of options to search passed to
26145916cd2Sjpk  * {sock,tpi}optcom_req() interface routine to take care of option
26245916cd2Sjpk  * management and associated methods.
26345916cd2Sjpk  */
26445916cd2Sjpk extern optdb_obj_t	udp_opt_obj;
26545916cd2Sjpk extern uint_t		udp_max_optsize;
26645916cd2Sjpk 
2670f1702c5SYu Xiangning extern sock_lower_handle_t udp_create(int, int, int, sock_downcalls_t **,
2680f1702c5SYu Xiangning     uint_t *, int *, int, cred_t *);
26941174437SAnders Persson extern int udp_fallback(sock_lower_handle_t, queue_t *, boolean_t,
2703e95bd4aSAnders Persson     so_proto_quiesced_cb_t, sock_quiesce_arg_t *);
2710f1702c5SYu Xiangning 
2720f1702c5SYu Xiangning extern sock_downcalls_t sock_udp_downcalls;
2730f1702c5SYu Xiangning 
2747c478bd9Sstevel@tonic-gate #endif	/*  _KERNEL */
2757c478bd9Sstevel@tonic-gate 
2767c478bd9Sstevel@tonic-gate #ifdef	__cplusplus
2777c478bd9Sstevel@tonic-gate }
2787c478bd9Sstevel@tonic-gate #endif
2797c478bd9Sstevel@tonic-gate 
2807c478bd9Sstevel@tonic-gate #endif	/* _UDP_IMPL_H */
281