xref: /illumos-gate/usr/src/uts/common/io/mac/mac_client.c (revision 84de666edc7f7d835057ae4807a387447c086bcf)
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) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright 2018 Joyent, Inc.
25  * Copyright 2017 RackTop Systems.
26  */
27 
28 /*
29  * - General Introduction:
30  *
31  * This file contains the implementation of the MAC client kernel
32  * API and related code. The MAC client API allows a kernel module
33  * to gain access to a MAC instance (physical NIC, link aggregation, etc).
34  * It allows a MAC client to associate itself with a MAC address,
35  * VLANs, callback functions for data traffic and for promiscuous mode.
36  * The MAC client API is also used to specify the properties associated
37  * with a MAC client, such as bandwidth limits, priority, CPUS, etc.
38  * These properties are further used to determine the hardware resources
39  * to allocate to the various MAC clients.
40  *
41  * - Primary MAC clients:
42  *
43  * The MAC client API refers to "primary MAC clients". A primary MAC
44  * client is a client which "owns" the primary MAC address of
45  * the underlying MAC instance. The primary MAC address is called out
46  * since it is associated with specific semantics: the primary MAC
47  * address is the MAC address which is assigned to the IP interface
48  * when it is plumbed, and the primary MAC address is assigned
49  * to VLAN data-links. The primary address of a MAC instance can
50  * also change dynamically from under the MAC client, for example
51  * as a result of a change of state of a link aggregation. In that
52  * case the MAC layer automatically updates all data-structures which
53  * refer to the current value of the primary MAC address. Typical
54  * primary MAC clients are dls, aggr, and xnb. A typical non-primary
55  * MAC client is the vnic driver.
56  *
57  * - Virtual Switching:
58  *
59  * The MAC layer implements a virtual switch between the MAC clients
60  * (primary and non-primary) defined on top of the same underlying
61  * NIC (physical, link aggregation, etc). The virtual switch is
62  * VLAN-aware, i.e. it allows multiple MAC clients to be member
63  * of one or more VLANs, and the virtual switch will distribute
64  * multicast tagged packets only to the member of the corresponding
65  * VLANs.
66  *
67  * - Upper vs Lower MAC:
68  *
69  * Creating a VNIC on top of a MAC instance effectively causes
70  * two MAC instances to be layered on top of each other, one for
71  * the VNIC(s), one for the underlying MAC instance (physical NIC,
72  * link aggregation, etc). In the code below we refer to the
73  * underlying NIC as the "lower MAC", and we refer to VNICs as
74  * the "upper MAC".
75  *
76  * - Pass-through for VNICs:
77  *
78  * When VNICs are created on top of an underlying MAC, this causes
79  * a layering of two MAC instances. Since the lower MAC already
80  * does the switching and demultiplexing to its MAC clients, the
81  * upper MAC would simply have to pass packets to the layer below
82  * or above it, which would introduce overhead. In order to avoid
83  * this overhead, the MAC layer implements a pass-through mechanism
84  * for VNICs. When a VNIC opens the lower MAC instance, it saves
85  * the MAC client handle it optains from the MAC layer. When a MAC
86  * client opens a VNIC (upper MAC), the MAC layer detects that
87  * the MAC being opened is a VNIC, and gets the MAC client handle
88  * that the VNIC driver obtained from the lower MAC. This exchange
89  * is done through a private capability between the MAC layer
90  * and the VNIC driver. The upper MAC then returns that handle
91  * directly to its MAC client. Any operation done by the upper
92  * MAC client is now done on the lower MAC client handle, which
93  * allows the VNIC driver to be completely bypassed for the
94  * performance sensitive data-path.
95  *
96  * - Secondary MACs for VNICs:
97  *
98  * VNICs support multiple upper mac clients to enable support for
99  * multiple MAC addresses on the VNIC. When the VNIC is created the
100  * initial mac client is the primary upper mac. Any additional mac
101  * clients are secondary macs. These are kept in sync with the primary
102  * (for things such as the rx function and resource control settings)
103  * using the same private capability interface between the MAC layer
104  * and the VNIC layer.
105  *
106  */
107 
108 #include <sys/types.h>
109 #include <sys/conf.h>
110 #include <sys/id_space.h>
111 #include <sys/esunddi.h>
112 #include <sys/stat.h>
113 #include <sys/mkdev.h>
114 #include <sys/stream.h>
115 #include <sys/strsun.h>
116 #include <sys/strsubr.h>
117 #include <sys/dlpi.h>
118 #include <sys/modhash.h>
119 #include <sys/mac_impl.h>
120 #include <sys/mac_client_impl.h>
121 #include <sys/mac_soft_ring.h>
122 #include <sys/mac_stat.h>
123 #include <sys/dls.h>
124 #include <sys/dld.h>
125 #include <sys/modctl.h>
126 #include <sys/fs/dv_node.h>
127 #include <sys/thread.h>
128 #include <sys/proc.h>
129 #include <sys/callb.h>
130 #include <sys/cpuvar.h>
131 #include <sys/atomic.h>
132 #include <sys/sdt.h>
133 #include <sys/mac_flow.h>
134 #include <sys/ddi_intr_impl.h>
135 #include <sys/disp.h>
136 #include <sys/sdt.h>
137 #include <sys/vnic.h>
138 #include <sys/vnic_impl.h>
139 #include <sys/vlan.h>
140 #include <inet/ip.h>
141 #include <inet/ip6.h>
142 #include <sys/exacct.h>
143 #include <sys/exacct_impl.h>
144 #include <inet/nd.h>
145 #include <sys/ethernet.h>
146 
147 kmem_cache_t	*mac_client_impl_cache;
148 kmem_cache_t	*mac_promisc_impl_cache;
149 
150 static boolean_t mac_client_single_rcvr(mac_client_impl_t *);
151 static flow_entry_t *mac_client_swap_mciflent(mac_client_impl_t *);
152 static flow_entry_t *mac_client_get_flow(mac_client_impl_t *,
153     mac_unicast_impl_t *);
154 static void mac_client_remove_flow_from_list(mac_client_impl_t *,
155     flow_entry_t *);
156 static void mac_client_add_to_flow_list(mac_client_impl_t *, flow_entry_t *);
157 static void mac_rename_flow_names(mac_client_impl_t *, const char *);
158 static void mac_virtual_link_update(mac_impl_t *);
159 static int mac_client_datapath_setup(mac_client_impl_t *, uint16_t,
160     uint8_t *, mac_resource_props_t *, boolean_t, mac_unicast_impl_t *);
161 static void mac_client_datapath_teardown(mac_client_handle_t,
162     mac_unicast_impl_t *, flow_entry_t *);
163 static int mac_resource_ctl_set(mac_client_handle_t, mac_resource_props_t *);
164 
165 /* ARGSUSED */
166 static int
167 i_mac_client_impl_ctor(void *buf, void *arg, int kmflag)
168 {
169 	int	i;
170 	mac_client_impl_t	*mcip = buf;
171 
172 	bzero(buf, MAC_CLIENT_IMPL_SIZE);
173 	mutex_init(&mcip->mci_tx_cb_lock, NULL, MUTEX_DRIVER, NULL);
174 	mcip->mci_tx_notify_cb_info.mcbi_lockp = &mcip->mci_tx_cb_lock;
175 
176 	ASSERT(mac_tx_percpu_cnt >= 0);
177 	for (i = 0; i <= mac_tx_percpu_cnt; i++) {
178 		mutex_init(&mcip->mci_tx_pcpu[i].pcpu_tx_lock, NULL,
179 		    MUTEX_DRIVER, NULL);
180 	}
181 	cv_init(&mcip->mci_tx_cv, NULL, CV_DRIVER, NULL);
182 
183 	return (0);
184 }
185 
186 /* ARGSUSED */
187 static void
188 i_mac_client_impl_dtor(void *buf, void *arg)
189 {
190 	int	i;
191 	mac_client_impl_t *mcip = buf;
192 
193 	ASSERT(mcip->mci_promisc_list == NULL);
194 	ASSERT(mcip->mci_unicast_list == NULL);
195 	ASSERT(mcip->mci_state_flags == 0);
196 	ASSERT(mcip->mci_tx_flag == 0);
197 
198 	mutex_destroy(&mcip->mci_tx_cb_lock);
199 
200 	ASSERT(mac_tx_percpu_cnt >= 0);
201 	for (i = 0; i <= mac_tx_percpu_cnt; i++) {
202 		ASSERT(mcip->mci_tx_pcpu[i].pcpu_tx_refcnt == 0);
203 		mutex_destroy(&mcip->mci_tx_pcpu[i].pcpu_tx_lock);
204 	}
205 	cv_destroy(&mcip->mci_tx_cv);
206 }
207 
208 /* ARGSUSED */
209 static int
210 i_mac_promisc_impl_ctor(void *buf, void *arg, int kmflag)
211 {
212 	mac_promisc_impl_t	*mpip = buf;
213 
214 	bzero(buf, sizeof (mac_promisc_impl_t));
215 	mpip->mpi_mci_link.mcb_objp = buf;
216 	mpip->mpi_mci_link.mcb_objsize = sizeof (mac_promisc_impl_t);
217 	mpip->mpi_mi_link.mcb_objp = buf;
218 	mpip->mpi_mi_link.mcb_objsize = sizeof (mac_promisc_impl_t);
219 	return (0);
220 }
221 
222 /* ARGSUSED */
223 static void
224 i_mac_promisc_impl_dtor(void *buf, void *arg)
225 {
226 	mac_promisc_impl_t	*mpip = buf;
227 
228 	ASSERT(mpip->mpi_mci_link.mcb_objp != NULL);
229 	ASSERT(mpip->mpi_mci_link.mcb_objsize == sizeof (mac_promisc_impl_t));
230 	ASSERT(mpip->mpi_mi_link.mcb_objp == mpip->mpi_mci_link.mcb_objp);
231 	ASSERT(mpip->mpi_mi_link.mcb_objsize == sizeof (mac_promisc_impl_t));
232 
233 	mpip->mpi_mci_link.mcb_objp = NULL;
234 	mpip->mpi_mci_link.mcb_objsize = 0;
235 	mpip->mpi_mi_link.mcb_objp = NULL;
236 	mpip->mpi_mi_link.mcb_objsize = 0;
237 
238 	ASSERT(mpip->mpi_mci_link.mcb_flags == 0);
239 	mpip->mpi_mci_link.mcb_objsize = 0;
240 }
241 
242 void
243 mac_client_init(void)
244 {
245 	ASSERT(mac_tx_percpu_cnt >= 0);
246 
247 	mac_client_impl_cache = kmem_cache_create("mac_client_impl_cache",
248 	    MAC_CLIENT_IMPL_SIZE, 0, i_mac_client_impl_ctor,
249 	    i_mac_client_impl_dtor, NULL, NULL, NULL, 0);
250 	ASSERT(mac_client_impl_cache != NULL);
251 
252 	mac_promisc_impl_cache = kmem_cache_create("mac_promisc_impl_cache",
253 	    sizeof (mac_promisc_impl_t), 0, i_mac_promisc_impl_ctor,
254 	    i_mac_promisc_impl_dtor, NULL, NULL, NULL, 0);
255 	ASSERT(mac_promisc_impl_cache != NULL);
256 }
257 
258 void
259 mac_client_fini(void)
260 {
261 	kmem_cache_destroy(mac_client_impl_cache);
262 	kmem_cache_destroy(mac_promisc_impl_cache);
263 }
264 
265 /*
266  * Return the lower MAC client handle from the VNIC driver for the
267  * specified VNIC MAC instance.
268  */
269 mac_client_impl_t *
270 mac_vnic_lower(mac_impl_t *mip)
271 {
272 	mac_capab_vnic_t cap;
273 	mac_client_impl_t *mcip;
274 
275 	VERIFY(i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_VNIC, &cap));
276 	mcip = cap.mcv_mac_client_handle(cap.mcv_arg);
277 
278 	return (mcip);
279 }
280 
281 /*
282  * Update the secondary macs
283  */
284 void
285 mac_vnic_secondary_update(mac_impl_t *mip)
286 {
287 	mac_capab_vnic_t cap;
288 
289 	VERIFY(i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_VNIC, &cap));
290 	cap.mcv_mac_secondary_update(cap.mcv_arg);
291 }
292 
293 /*
294  * Return the MAC client handle of the primary MAC client for the
295  * specified MAC instance, or NULL otherwise.
296  */
297 mac_client_impl_t *
298 mac_primary_client_handle(mac_impl_t *mip)
299 {
300 	mac_client_impl_t *mcip;
301 
302 	if (mip->mi_state_flags & MIS_IS_VNIC)
303 		return (mac_vnic_lower(mip));
304 
305 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
306 
307 	for (mcip = mip->mi_clients_list; mcip != NULL;
308 	    mcip = mcip->mci_client_next) {
309 		if (MCIP_DATAPATH_SETUP(mcip) && mac_is_primary_client(mcip))
310 			return (mcip);
311 	}
312 	return (NULL);
313 }
314 
315 /*
316  * Open a MAC specified by its MAC name.
317  */
318 int
319 mac_open(const char *macname, mac_handle_t *mhp)
320 {
321 	mac_impl_t	*mip;
322 	int		err;
323 
324 	/*
325 	 * Look up its entry in the global hash table.
326 	 */
327 	if ((err = mac_hold(macname, &mip)) != 0)
328 		return (err);
329 
330 	/*
331 	 * Hold the dip associated to the MAC to prevent it from being
332 	 * detached. For a softmac, its underlying dip is held by the
333 	 * mi_open() callback.
334 	 *
335 	 * This is done to be more tolerant with some defective drivers,
336 	 * which incorrectly handle mac_unregister() failure in their
337 	 * xxx_detach() routine. For example, some drivers ignore the
338 	 * failure of mac_unregister() and free all resources that
339 	 * that are needed for data transmition.
340 	 */
341 	e_ddi_hold_devi(mip->mi_dip);
342 
343 	if (!(mip->mi_callbacks->mc_callbacks & MC_OPEN)) {
344 		*mhp = (mac_handle_t)mip;
345 		return (0);
346 	}
347 
348 	/*
349 	 * The mac perimeter is used in both mac_open and mac_close by the
350 	 * framework to single thread the MC_OPEN/MC_CLOSE of drivers.
351 	 */
352 	i_mac_perim_enter(mip);
353 	mip->mi_oref++;
354 	if (mip->mi_oref != 1 || ((err = mip->mi_open(mip->mi_driver)) == 0)) {
355 		*mhp = (mac_handle_t)mip;
356 		i_mac_perim_exit(mip);
357 		return (0);
358 	}
359 	mip->mi_oref--;
360 	ddi_release_devi(mip->mi_dip);
361 	mac_rele(mip);
362 	i_mac_perim_exit(mip);
363 	return (err);
364 }
365 
366 /*
367  * Open a MAC specified by its linkid.
368  */
369 int
370 mac_open_by_linkid(datalink_id_t linkid, mac_handle_t *mhp)
371 {
372 	dls_dl_handle_t	dlh;
373 	int		err;
374 
375 	if ((err = dls_devnet_hold_tmp(linkid, &dlh)) != 0)
376 		return (err);
377 
378 	dls_devnet_prop_task_wait(dlh);
379 
380 	err = mac_open(dls_devnet_mac(dlh), mhp);
381 
382 	dls_devnet_rele_tmp(dlh);
383 	return (err);
384 }
385 
386 /*
387  * Open a MAC specified by its link name.
388  */
389 int
390 mac_open_by_linkname(const char *link, mac_handle_t *mhp)
391 {
392 	datalink_id_t	linkid;
393 	int		err;
394 
395 	if ((err = dls_mgmt_get_linkid(link, &linkid)) != 0)
396 		return (err);
397 	return (mac_open_by_linkid(linkid, mhp));
398 }
399 
400 /*
401  * Close the specified MAC.
402  */
403 void
404 mac_close(mac_handle_t mh)
405 {
406 	mac_impl_t	*mip = (mac_impl_t *)mh;
407 
408 	i_mac_perim_enter(mip);
409 	/*
410 	 * The mac perimeter is used in both mac_open and mac_close by the
411 	 * framework to single thread the MC_OPEN/MC_CLOSE of drivers.
412 	 */
413 	if (mip->mi_callbacks->mc_callbacks & MC_OPEN) {
414 		ASSERT(mip->mi_oref != 0);
415 		if (--mip->mi_oref == 0) {
416 			if ((mip->mi_callbacks->mc_callbacks & MC_CLOSE))
417 				mip->mi_close(mip->mi_driver);
418 		}
419 	}
420 	i_mac_perim_exit(mip);
421 	ddi_release_devi(mip->mi_dip);
422 	mac_rele(mip);
423 }
424 
425 /*
426  * Misc utility functions to retrieve various information about a MAC
427  * instance or a MAC client.
428  */
429 
430 const mac_info_t *
431 mac_info(mac_handle_t mh)
432 {
433 	return (&((mac_impl_t *)mh)->mi_info);
434 }
435 
436 dev_info_t *
437 mac_devinfo_get(mac_handle_t mh)
438 {
439 	return (((mac_impl_t *)mh)->mi_dip);
440 }
441 
442 void *
443 mac_driver(mac_handle_t mh)
444 {
445 	return (((mac_impl_t *)mh)->mi_driver);
446 }
447 
448 const char *
449 mac_name(mac_handle_t mh)
450 {
451 	return (((mac_impl_t *)mh)->mi_name);
452 }
453 
454 int
455 mac_type(mac_handle_t mh)
456 {
457 	return (((mac_impl_t *)mh)->mi_type->mt_type);
458 }
459 
460 int
461 mac_nativetype(mac_handle_t mh)
462 {
463 	return (((mac_impl_t *)mh)->mi_type->mt_nativetype);
464 }
465 
466 char *
467 mac_client_name(mac_client_handle_t mch)
468 {
469 	return (((mac_client_impl_t *)mch)->mci_name);
470 }
471 
472 minor_t
473 mac_minor(mac_handle_t mh)
474 {
475 	return (((mac_impl_t *)mh)->mi_minor);
476 }
477 
478 /*
479  * Return the VID associated with a MAC client. This function should
480  * be called for clients which are associated with only one VID.
481  */
482 uint16_t
483 mac_client_vid(mac_client_handle_t mch)
484 {
485 	uint16_t		vid = VLAN_ID_NONE;
486 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
487 	flow_desc_t		flow_desc;
488 
489 	if (mcip->mci_nflents == 0)
490 		return (vid);
491 
492 	ASSERT(MCIP_DATAPATH_SETUP(mcip) && mac_client_single_rcvr(mcip));
493 
494 	mac_flow_get_desc(mcip->mci_flent, &flow_desc);
495 	if ((flow_desc.fd_mask & FLOW_LINK_VID) != 0)
496 		vid = flow_desc.fd_vid;
497 
498 	return (vid);
499 }
500 
501 /*
502  * Return whether the specified MAC client corresponds to a VLAN VNIC.
503  */
504 boolean_t
505 mac_client_is_vlan_vnic(mac_client_handle_t mch)
506 {
507 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
508 
509 	return (((mcip->mci_state_flags & MCIS_IS_VNIC) != 0) &&
510 	    ((mcip->mci_flent->fe_type & FLOW_PRIMARY_MAC) != 0));
511 }
512 
513 /*
514  * Return the link speed associated with the specified MAC client.
515  *
516  * The link speed of a MAC client is equal to the smallest value of
517  * 1) the current link speed of the underlying NIC, or
518  * 2) the bandwidth limit set for the MAC client.
519  *
520  * Note that the bandwidth limit can be higher than the speed
521  * of the underlying NIC. This is allowed to avoid spurious
522  * administration action failures or artifically lowering the
523  * bandwidth limit of a link that may  have temporarily lowered
524  * its link speed due to hardware problem or administrator action.
525  */
526 static uint64_t
527 mac_client_ifspeed(mac_client_impl_t *mcip)
528 {
529 	mac_impl_t *mip = mcip->mci_mip;
530 	uint64_t nic_speed;
531 
532 	nic_speed = mac_stat_get((mac_handle_t)mip, MAC_STAT_IFSPEED);
533 
534 	if (nic_speed == 0) {
535 		return (0);
536 	} else {
537 		uint64_t policy_limit = (uint64_t)-1;
538 
539 		if (MCIP_RESOURCE_PROPS_MASK(mcip) & MRP_MAXBW)
540 			policy_limit = MCIP_RESOURCE_PROPS_MAXBW(mcip);
541 
542 		return (MIN(policy_limit, nic_speed));
543 	}
544 }
545 
546 /*
547  * Return the link state of the specified client. If here are more
548  * than one clients of the underying mac_impl_t, the link state
549  * will always be UP regardless of the link state of the underlying
550  * mac_impl_t. This is needed to allow the MAC clients to continue
551  * to communicate with each other even when the physical link of
552  * their mac_impl_t is down.
553  */
554 static uint64_t
555 mac_client_link_state(mac_client_impl_t *mcip)
556 {
557 	mac_impl_t *mip = mcip->mci_mip;
558 	uint16_t vid;
559 	mac_client_impl_t *mci_list;
560 	mac_unicast_impl_t *mui_list, *oth_mui_list;
561 
562 	/*
563 	 * Returns LINK_STATE_UP if there are other MAC clients defined on
564 	 * mac_impl_t which share same VLAN ID as that of mcip. Note that
565 	 * if 'mcip' has more than one VID's then we match ANY one of the
566 	 * VID's with other MAC client's VID's and return LINK_STATE_UP.
567 	 */
568 	rw_enter(&mcip->mci_rw_lock, RW_READER);
569 	for (mui_list = mcip->mci_unicast_list; mui_list != NULL;
570 	    mui_list = mui_list->mui_next) {
571 		vid = mui_list->mui_vid;
572 		for (mci_list = mip->mi_clients_list; mci_list != NULL;
573 		    mci_list = mci_list->mci_client_next) {
574 			if (mci_list == mcip)
575 				continue;
576 			for (oth_mui_list = mci_list->mci_unicast_list;
577 			    oth_mui_list != NULL; oth_mui_list = oth_mui_list->
578 			    mui_next) {
579 				if (vid == oth_mui_list->mui_vid) {
580 					rw_exit(&mcip->mci_rw_lock);
581 					return (LINK_STATE_UP);
582 				}
583 			}
584 		}
585 	}
586 	rw_exit(&mcip->mci_rw_lock);
587 
588 	return (mac_stat_get((mac_handle_t)mip, MAC_STAT_LINK_STATE));
589 }
590 
591 /*
592  * These statistics are consumed by dladm show-link -s <vnic>,
593  * dladm show-vnic -s and netstat. With the introduction of dlstat,
594  * dladm show-link -s and dladm show-vnic -s witll be EOL'ed while
595  * netstat will consume from kstats introduced for dlstat. This code
596  * will be removed at that time.
597  */
598 
599 /*
600  * Return the statistics of a MAC client. These statistics are different
601  * then the statistics of the underlying MAC which are returned by
602  * mac_stat_get().
603  *
604  * Note that for things based on the tx and rx stats, mac will end up clobbering
605  * those stats when the underlying set of rings in the srs changes. As such, we
606  * need to source not only the current set, but also the historical set when
607  * returning to the client, lest our counters appear to go backwards.
608  */
609 uint64_t
610 mac_client_stat_get(mac_client_handle_t mch, uint_t stat)
611 {
612 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
613 	mac_impl_t		*mip = mcip->mci_mip;
614 	flow_entry_t		*flent = mcip->mci_flent;
615 	mac_soft_ring_set_t	*mac_srs;
616 	mac_rx_stats_t		*mac_rx_stat, *old_rx_stat;
617 	mac_tx_stats_t		*mac_tx_stat, *old_tx_stat;
618 	int i;
619 	uint64_t val = 0;
620 
621 	mac_srs = (mac_soft_ring_set_t *)(flent->fe_tx_srs);
622 	mac_tx_stat = &mac_srs->srs_tx.st_stat;
623 	old_rx_stat = &mcip->mci_misc_stat.mms_defunctrxlanestats;
624 	old_tx_stat = &mcip->mci_misc_stat.mms_defuncttxlanestats;
625 
626 	switch (stat) {
627 	case MAC_STAT_LINK_STATE:
628 		val = mac_client_link_state(mcip);
629 		break;
630 	case MAC_STAT_LINK_UP:
631 		val = (mac_client_link_state(mcip) == LINK_STATE_UP);
632 		break;
633 	case MAC_STAT_PROMISC:
634 		val = mac_stat_get((mac_handle_t)mip, MAC_STAT_PROMISC);
635 		break;
636 	case MAC_STAT_LOWLINK_STATE:
637 		val = mac_stat_get((mac_handle_t)mip, MAC_STAT_LOWLINK_STATE);
638 		break;
639 	case MAC_STAT_IFSPEED:
640 		val = mac_client_ifspeed(mcip);
641 		break;
642 	case MAC_STAT_MULTIRCV:
643 		val = mcip->mci_misc_stat.mms_multircv;
644 		break;
645 	case MAC_STAT_BRDCSTRCV:
646 		val = mcip->mci_misc_stat.mms_brdcstrcv;
647 		break;
648 	case MAC_STAT_MULTIXMT:
649 		val = mcip->mci_misc_stat.mms_multixmt;
650 		break;
651 	case MAC_STAT_BRDCSTXMT:
652 		val = mcip->mci_misc_stat.mms_brdcstxmt;
653 		break;
654 	case MAC_STAT_OBYTES:
655 		val = mac_tx_stat->mts_obytes;
656 		val += old_tx_stat->mts_obytes;
657 		break;
658 	case MAC_STAT_OPACKETS:
659 		val = mac_tx_stat->mts_opackets;
660 		val += old_tx_stat->mts_opackets;
661 		break;
662 	case MAC_STAT_OERRORS:
663 		val = mac_tx_stat->mts_oerrors;
664 		val += old_tx_stat->mts_oerrors;
665 		break;
666 	case MAC_STAT_IPACKETS:
667 		for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
668 			mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
669 			mac_rx_stat = &mac_srs->srs_rx.sr_stat;
670 			val += mac_rx_stat->mrs_intrcnt +
671 			    mac_rx_stat->mrs_pollcnt + mac_rx_stat->mrs_lclcnt;
672 		}
673 		val += old_rx_stat->mrs_intrcnt + old_rx_stat->mrs_pollcnt +
674 		    old_rx_stat->mrs_lclcnt;
675 		break;
676 	case MAC_STAT_RBYTES:
677 		for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
678 			mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
679 			mac_rx_stat = &mac_srs->srs_rx.sr_stat;
680 			val += mac_rx_stat->mrs_intrbytes +
681 			    mac_rx_stat->mrs_pollbytes +
682 			    mac_rx_stat->mrs_lclbytes;
683 		}
684 		val += old_rx_stat->mrs_intrbytes + old_rx_stat->mrs_pollbytes +
685 		    old_rx_stat->mrs_lclbytes;
686 		break;
687 	case MAC_STAT_IERRORS:
688 		for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
689 			mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
690 			mac_rx_stat = &mac_srs->srs_rx.sr_stat;
691 			val += mac_rx_stat->mrs_ierrors;
692 		}
693 		val += old_rx_stat->mrs_ierrors;
694 		break;
695 	default:
696 		val = mac_driver_stat_default(mip, stat);
697 		break;
698 	}
699 
700 	return (val);
701 }
702 
703 /*
704  * Return the statistics of the specified MAC instance.
705  */
706 uint64_t
707 mac_stat_get(mac_handle_t mh, uint_t stat)
708 {
709 	mac_impl_t	*mip = (mac_impl_t *)mh;
710 	uint64_t	val;
711 	int		ret;
712 
713 	/*
714 	 * The range of stat determines where it is maintained.  Stat
715 	 * values from 0 up to (but not including) MAC_STAT_MIN are
716 	 * mainteined by the mac module itself.  Everything else is
717 	 * maintained by the driver.
718 	 *
719 	 * If the mac_impl_t being queried corresponds to a VNIC,
720 	 * the stats need to be queried from the lower MAC client
721 	 * corresponding to the VNIC. (The mac_link_update()
722 	 * invoked by the driver to the lower MAC causes the *lower
723 	 * MAC* to update its mi_linkstate, and send a notification
724 	 * to its MAC clients. Due to the VNIC passthrough,
725 	 * these notifications are sent to the upper MAC clients
726 	 * of the VNIC directly, and the upper mac_impl_t of the VNIC
727 	 * does not have a valid mi_linkstate.
728 	 */
729 	if (stat < MAC_STAT_MIN && !(mip->mi_state_flags & MIS_IS_VNIC)) {
730 		/* these stats are maintained by the mac module itself */
731 		switch (stat) {
732 		case MAC_STAT_LINK_STATE:
733 			return (mip->mi_linkstate);
734 		case MAC_STAT_LINK_UP:
735 			return (mip->mi_linkstate == LINK_STATE_UP);
736 		case MAC_STAT_PROMISC:
737 			return (mip->mi_devpromisc != 0);
738 		case MAC_STAT_LOWLINK_STATE:
739 			return (mip->mi_lowlinkstate);
740 		default:
741 			ASSERT(B_FALSE);
742 		}
743 	}
744 
745 	/*
746 	 * Call the driver to get the given statistic.
747 	 */
748 	ret = mip->mi_getstat(mip->mi_driver, stat, &val);
749 	if (ret != 0) {
750 		/*
751 		 * The driver doesn't support this statistic.  Get the
752 		 * statistic's default value.
753 		 */
754 		val = mac_driver_stat_default(mip, stat);
755 	}
756 	return (val);
757 }
758 
759 /*
760  * Query hardware rx ring corresponding to the pseudo ring.
761  */
762 uint64_t
763 mac_pseudo_rx_ring_stat_get(mac_ring_handle_t handle, uint_t stat)
764 {
765 	return (mac_rx_ring_stat_get(handle, stat));
766 }
767 
768 /*
769  * Query hardware tx ring corresponding to the pseudo ring.
770  */
771 uint64_t
772 mac_pseudo_tx_ring_stat_get(mac_ring_handle_t handle, uint_t stat)
773 {
774 	return (mac_tx_ring_stat_get(handle, stat));
775 }
776 
777 /*
778  * Utility function which returns the VID associated with a flow entry.
779  */
780 uint16_t
781 i_mac_flow_vid(flow_entry_t *flent)
782 {
783 	flow_desc_t	flow_desc;
784 
785 	mac_flow_get_desc(flent, &flow_desc);
786 
787 	if ((flow_desc.fd_mask & FLOW_LINK_VID) != 0)
788 		return (flow_desc.fd_vid);
789 	return (VLAN_ID_NONE);
790 }
791 
792 /*
793  * Verify the validity of the specified unicast MAC address. Returns B_TRUE
794  * if the address is valid, B_FALSE otherwise (multicast address, or incorrect
795  * length.
796  */
797 boolean_t
798 mac_unicst_verify(mac_handle_t mh, const uint8_t *addr, uint_t len)
799 {
800 	mac_impl_t	*mip = (mac_impl_t *)mh;
801 
802 	/*
803 	 * Verify the address. No lock is needed since mi_type and plugin
804 	 * details don't change after mac_register().
805 	 */
806 	if ((len != mip->mi_type->mt_addr_length) ||
807 	    (mip->mi_type->mt_ops.mtops_unicst_verify(addr,
808 	    mip->mi_pdata)) != 0) {
809 		return (B_FALSE);
810 	} else {
811 		return (B_TRUE);
812 	}
813 }
814 
815 void
816 mac_sdu_get(mac_handle_t mh, uint_t *min_sdu, uint_t *max_sdu)
817 {
818 	mac_impl_t	*mip = (mac_impl_t *)mh;
819 
820 	if (min_sdu != NULL)
821 		*min_sdu = mip->mi_sdu_min;
822 	if (max_sdu != NULL)
823 		*max_sdu = mip->mi_sdu_max;
824 }
825 
826 void
827 mac_sdu_get2(mac_handle_t mh, uint_t *min_sdu, uint_t *max_sdu,
828     uint_t *multicast_sdu)
829 {
830 	mac_impl_t	*mip = (mac_impl_t *)mh;
831 
832 	if (min_sdu != NULL)
833 		*min_sdu = mip->mi_sdu_min;
834 	if (max_sdu != NULL)
835 		*max_sdu = mip->mi_sdu_max;
836 	if (multicast_sdu != NULL)
837 		*multicast_sdu = mip->mi_sdu_multicast;
838 }
839 
840 /*
841  * Update the MAC unicast address of the specified client's flows. Currently
842  * only one unicast MAC unicast address is allowed per client.
843  */
844 static void
845 mac_unicast_update_client_flow(mac_client_impl_t *mcip)
846 {
847 	mac_impl_t *mip = mcip->mci_mip;
848 	flow_entry_t *flent = mcip->mci_flent;
849 	mac_address_t *map = mcip->mci_unicast;
850 	flow_desc_t flow_desc;
851 
852 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
853 	ASSERT(flent != NULL);
854 
855 	mac_flow_get_desc(flent, &flow_desc);
856 	ASSERT(flow_desc.fd_mask & FLOW_LINK_DST);
857 
858 	bcopy(map->ma_addr, flow_desc.fd_dst_mac, map->ma_len);
859 	mac_flow_set_desc(flent, &flow_desc);
860 
861 	/*
862 	 * The v6 local and SLAAC addrs (used by mac protection) need to be
863 	 * regenerated because our mac address has changed.
864 	 */
865 	mac_protect_update_mac_token(mcip);
866 
867 	/*
868 	 * When there are multiple VLANs sharing the same MAC address,
869 	 * each gets its own MAC client, except when running on sun4v
870 	 * vsw. In that case the mci_flent_list is used to place
871 	 * multiple VLAN flows on one MAC client. If we ever get rid
872 	 * of vsw then this code can go, but until then we need to
873 	 * update all flow entries.
874 	 */
875 	for (flent = mcip->mci_flent_list; flent != NULL;
876 	    flent = flent->fe_client_next) {
877 		mac_flow_get_desc(flent, &flow_desc);
878 		if (!(flent->fe_type & FLOW_PRIMARY_MAC ||
879 		    flent->fe_type & FLOW_VNIC_MAC))
880 			continue;
881 
882 		bcopy(map->ma_addr, flow_desc.fd_dst_mac, map->ma_len);
883 		mac_flow_set_desc(flent, &flow_desc);
884 	}
885 }
886 
887 /*
888  * Update all clients that share the same unicast address.
889  */
890 void
891 mac_unicast_update_clients(mac_impl_t *mip, mac_address_t *map)
892 {
893 	mac_client_impl_t *mcip;
894 
895 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
896 
897 	/*
898 	 * Find all clients that share the same unicast MAC address and update
899 	 * them appropriately.
900 	 */
901 	for (mcip = mip->mi_clients_list; mcip != NULL;
902 	    mcip = mcip->mci_client_next) {
903 		/*
904 		 * Ignore clients that don't share this MAC address.
905 		 */
906 		if (map != mcip->mci_unicast)
907 			continue;
908 
909 		/*
910 		 * Update those clients with same old unicast MAC address.
911 		 */
912 		mac_unicast_update_client_flow(mcip);
913 	}
914 }
915 
916 /*
917  * Update the unicast MAC address of the specified VNIC MAC client.
918  *
919  * Check whether the operation is valid. Any of following cases should fail:
920  *
921  * 1. It's a VLAN type of VNIC.
922  * 2. The new value is current "primary" MAC address.
923  * 3. The current MAC address is shared with other clients.
924  * 4. The new MAC address has been used. This case will be valid when
925  *    client migration is fully supported.
926  */
927 int
928 mac_vnic_unicast_set(mac_client_handle_t mch, const uint8_t *addr)
929 {
930 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
931 	mac_impl_t *mip = mcip->mci_mip;
932 	mac_address_t *map = mcip->mci_unicast;
933 	int err;
934 
935 	ASSERT(!(mip->mi_state_flags & MIS_IS_VNIC));
936 	ASSERT(mcip->mci_state_flags & MCIS_IS_VNIC);
937 	ASSERT(mcip->mci_flags != MAC_CLIENT_FLAGS_PRIMARY);
938 
939 	i_mac_perim_enter(mip);
940 
941 	/*
942 	 * If this is a VLAN type of VNIC, it's using "primary" MAC address
943 	 * of the underlying interface. Must fail here. Refer to case 1 above.
944 	 */
945 	if (bcmp(map->ma_addr, mip->mi_addr, map->ma_len) == 0) {
946 		i_mac_perim_exit(mip);
947 		return (ENOTSUP);
948 	}
949 
950 	/*
951 	 * If the new address is the "primary" one, must fail. Refer to
952 	 * case 2 above.
953 	 */
954 	if (bcmp(addr, mip->mi_addr, map->ma_len) == 0) {
955 		i_mac_perim_exit(mip);
956 		return (EACCES);
957 	}
958 
959 	/*
960 	 * If the address is shared by multiple clients, must fail. Refer
961 	 * to case 3 above.
962 	 */
963 	if (mac_check_macaddr_shared(map)) {
964 		i_mac_perim_exit(mip);
965 		return (EBUSY);
966 	}
967 
968 	/*
969 	 * If the new address has been used, must fail for now. Refer to
970 	 * case 4 above.
971 	 */
972 	if (mac_find_macaddr(mip, (uint8_t *)addr) != NULL) {
973 		i_mac_perim_exit(mip);
974 		return (ENOTSUP);
975 	}
976 
977 	/*
978 	 * Update the MAC address.
979 	 */
980 	err = mac_update_macaddr(map, (uint8_t *)addr);
981 
982 	if (err != 0) {
983 		i_mac_perim_exit(mip);
984 		return (err);
985 	}
986 
987 	/*
988 	 * Update all flows of this MAC client.
989 	 */
990 	mac_unicast_update_client_flow(mcip);
991 
992 	i_mac_perim_exit(mip);
993 	return (0);
994 }
995 
996 /*
997  * Program the new primary unicast address of the specified MAC.
998  *
999  * Function mac_update_macaddr() takes care different types of underlying
1000  * MAC. If the underlying MAC is VNIC, the VNIC driver must have registerd
1001  * mi_unicst() entry point, that indirectly calls mac_vnic_unicast_set()
1002  * which will take care of updating the MAC address of the corresponding
1003  * MAC client.
1004  *
1005  * This is the only interface that allow the client to update the "primary"
1006  * MAC address of the underlying MAC. The new value must have not been
1007  * used by other clients.
1008  */
1009 int
1010 mac_unicast_primary_set(mac_handle_t mh, const uint8_t *addr)
1011 {
1012 	mac_impl_t *mip = (mac_impl_t *)mh;
1013 	mac_address_t *map;
1014 	int err;
1015 
1016 	/* verify the address validity */
1017 	if (!mac_unicst_verify(mh, addr, mip->mi_type->mt_addr_length))
1018 		return (EINVAL);
1019 
1020 	i_mac_perim_enter(mip);
1021 
1022 	/*
1023 	 * If the new value is the same as the current primary address value,
1024 	 * there's nothing to do.
1025 	 */
1026 	if (bcmp(addr, mip->mi_addr, mip->mi_type->mt_addr_length) == 0) {
1027 		i_mac_perim_exit(mip);
1028 		return (0);
1029 	}
1030 
1031 	if (mac_find_macaddr(mip, (uint8_t *)addr) != NULL) {
1032 		i_mac_perim_exit(mip);
1033 		return (EBUSY);
1034 	}
1035 
1036 	map = mac_find_macaddr(mip, mip->mi_addr);
1037 	ASSERT(map != NULL);
1038 
1039 	/*
1040 	 * Update the MAC address.
1041 	 */
1042 	if (mip->mi_state_flags & MIS_IS_AGGR) {
1043 		mac_capab_aggr_t aggr_cap;
1044 
1045 		/*
1046 		 * If the MAC is an aggregation, other than the unicast
1047 		 * addresses programming, aggr must be informed about this
1048 		 * primary unicst address change to change its MAC address
1049 		 * policy to be user-specified.
1050 		 */
1051 		ASSERT(map->ma_type == MAC_ADDRESS_TYPE_UNICAST_CLASSIFIED);
1052 		VERIFY(i_mac_capab_get(mh, MAC_CAPAB_AGGR, &aggr_cap));
1053 		err = aggr_cap.mca_unicst(mip->mi_driver, addr);
1054 		if (err == 0)
1055 			bcopy(addr, map->ma_addr, map->ma_len);
1056 	} else {
1057 		err = mac_update_macaddr(map, (uint8_t *)addr);
1058 	}
1059 
1060 	if (err != 0) {
1061 		i_mac_perim_exit(mip);
1062 		return (err);
1063 	}
1064 
1065 	mac_unicast_update_clients(mip, map);
1066 
1067 	/*
1068 	 * Save the new primary MAC address in mac_impl_t.
1069 	 */
1070 	bcopy(addr, mip->mi_addr, mip->mi_type->mt_addr_length);
1071 
1072 	i_mac_perim_exit(mip);
1073 
1074 	if (err == 0)
1075 		i_mac_notify(mip, MAC_NOTE_UNICST);
1076 
1077 	return (err);
1078 }
1079 
1080 /*
1081  * Return the current primary MAC address of the specified MAC.
1082  */
1083 void
1084 mac_unicast_primary_get(mac_handle_t mh, uint8_t *addr)
1085 {
1086 	mac_impl_t *mip = (mac_impl_t *)mh;
1087 
1088 	rw_enter(&mip->mi_rw_lock, RW_READER);
1089 	bcopy(mip->mi_addr, addr, mip->mi_type->mt_addr_length);
1090 	rw_exit(&mip->mi_rw_lock);
1091 }
1092 
1093 /*
1094  * Return the secondary MAC address for the specified handle
1095  */
1096 void
1097 mac_unicast_secondary_get(mac_client_handle_t mh, uint8_t *addr)
1098 {
1099 	mac_client_impl_t *mcip = (mac_client_impl_t *)mh;
1100 
1101 	ASSERT(mcip->mci_unicast != NULL);
1102 	bcopy(mcip->mci_unicast->ma_addr, addr, mcip->mci_unicast->ma_len);
1103 }
1104 
1105 /*
1106  * Return information about the use of the primary MAC address of the
1107  * specified MAC instance:
1108  *
1109  * - if client_name is non-NULL, it must point to a string of at
1110  *   least MAXNAMELEN bytes, and will be set to the name of the MAC
1111  *   client which uses the primary MAC address.
1112  *
1113  * - if in_use is non-NULL, used to return whether the primary MAC
1114  *   address is currently in use.
1115  */
1116 void
1117 mac_unicast_primary_info(mac_handle_t mh, char *client_name, boolean_t *in_use)
1118 {
1119 	mac_impl_t *mip = (mac_impl_t *)mh;
1120 	mac_client_impl_t *cur_client;
1121 
1122 	if (in_use != NULL)
1123 		*in_use = B_FALSE;
1124 	if (client_name != NULL)
1125 		bzero(client_name, MAXNAMELEN);
1126 
1127 	/*
1128 	 * The mi_rw_lock is used to protect threads that don't hold the
1129 	 * mac perimeter to get a consistent view of the mi_clients_list.
1130 	 * Threads that modify the list must hold both the mac perimeter and
1131 	 * mi_rw_lock(RW_WRITER)
1132 	 */
1133 	rw_enter(&mip->mi_rw_lock, RW_READER);
1134 	for (cur_client = mip->mi_clients_list; cur_client != NULL;
1135 	    cur_client = cur_client->mci_client_next) {
1136 		if (mac_is_primary_client(cur_client) ||
1137 		    (mip->mi_state_flags & MIS_IS_VNIC)) {
1138 			rw_exit(&mip->mi_rw_lock);
1139 			if (in_use != NULL)
1140 				*in_use = B_TRUE;
1141 			if (client_name != NULL) {
1142 				bcopy(cur_client->mci_name, client_name,
1143 				    MAXNAMELEN);
1144 			}
1145 			return;
1146 		}
1147 	}
1148 	rw_exit(&mip->mi_rw_lock);
1149 }
1150 
1151 /*
1152  * Return the current destination MAC address of the specified MAC.
1153  */
1154 boolean_t
1155 mac_dst_get(mac_handle_t mh, uint8_t *addr)
1156 {
1157 	mac_impl_t *mip = (mac_impl_t *)mh;
1158 
1159 	rw_enter(&mip->mi_rw_lock, RW_READER);
1160 	if (mip->mi_dstaddr_set)
1161 		bcopy(mip->mi_dstaddr, addr, mip->mi_type->mt_addr_length);
1162 	rw_exit(&mip->mi_rw_lock);
1163 	return (mip->mi_dstaddr_set);
1164 }
1165 
1166 /*
1167  * Add the specified MAC client to the list of clients which opened
1168  * the specified MAC.
1169  */
1170 static void
1171 mac_client_add(mac_client_impl_t *mcip)
1172 {
1173 	mac_impl_t *mip = mcip->mci_mip;
1174 
1175 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1176 
1177 	/* add VNIC to the front of the list */
1178 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
1179 	mcip->mci_client_next = mip->mi_clients_list;
1180 	mip->mi_clients_list = mcip;
1181 	mip->mi_nclients++;
1182 	rw_exit(&mip->mi_rw_lock);
1183 }
1184 
1185 /*
1186  * Remove the specified MAC client from the list of clients which opened
1187  * the specified MAC.
1188  */
1189 static void
1190 mac_client_remove(mac_client_impl_t *mcip)
1191 {
1192 	mac_impl_t *mip = mcip->mci_mip;
1193 	mac_client_impl_t **prev, *cclient;
1194 
1195 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1196 
1197 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
1198 	prev = &mip->mi_clients_list;
1199 	cclient = *prev;
1200 	while (cclient != NULL && cclient != mcip) {
1201 		prev = &cclient->mci_client_next;
1202 		cclient = *prev;
1203 	}
1204 	ASSERT(cclient != NULL);
1205 	*prev = cclient->mci_client_next;
1206 	mip->mi_nclients--;
1207 	rw_exit(&mip->mi_rw_lock);
1208 }
1209 
1210 static mac_unicast_impl_t *
1211 mac_client_find_vid(mac_client_impl_t *mcip, uint16_t vid)
1212 {
1213 	mac_unicast_impl_t *muip = mcip->mci_unicast_list;
1214 
1215 	while ((muip != NULL) && (muip->mui_vid != vid))
1216 		muip = muip->mui_next;
1217 
1218 	return (muip);
1219 }
1220 
1221 /*
1222  * Return whether the specified (MAC address, VID) tuple is already used by
1223  * one of the MAC clients associated with the specified MAC.
1224  */
1225 static boolean_t
1226 mac_addr_in_use(mac_impl_t *mip, uint8_t *mac_addr, uint16_t vid)
1227 {
1228 	mac_client_impl_t *client;
1229 	mac_address_t *map;
1230 
1231 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1232 
1233 	for (client = mip->mi_clients_list; client != NULL;
1234 	    client = client->mci_client_next) {
1235 
1236 		/*
1237 		 * Ignore clients that don't have unicast address.
1238 		 */
1239 		if (client->mci_unicast_list == NULL)
1240 			continue;
1241 
1242 		map = client->mci_unicast;
1243 
1244 		if ((bcmp(mac_addr, map->ma_addr, map->ma_len) == 0) &&
1245 		    (mac_client_find_vid(client, vid) != NULL)) {
1246 			return (B_TRUE);
1247 		}
1248 	}
1249 
1250 	return (B_FALSE);
1251 }
1252 
1253 /*
1254  * Generate a random MAC address. The MAC address prefix is
1255  * stored in the array pointed to by mac_addr, and its length, in bytes,
1256  * is specified by prefix_len. The least significant bits
1257  * after prefix_len bytes are generated, and stored after the prefix
1258  * in the mac_addr array.
1259  */
1260 int
1261 mac_addr_random(mac_client_handle_t mch, uint_t prefix_len,
1262     uint8_t *mac_addr, mac_diag_t *diag)
1263 {
1264 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1265 	mac_impl_t *mip = mcip->mci_mip;
1266 	size_t addr_len = mip->mi_type->mt_addr_length;
1267 
1268 	if (prefix_len >= addr_len) {
1269 		*diag = MAC_DIAG_MACPREFIXLEN_INVALID;
1270 		return (EINVAL);
1271 	}
1272 
1273 	/* check the prefix value */
1274 	if (prefix_len > 0) {
1275 		bzero(mac_addr + prefix_len, addr_len - prefix_len);
1276 		if (!mac_unicst_verify((mac_handle_t)mip, mac_addr,
1277 		    addr_len)) {
1278 			*diag = MAC_DIAG_MACPREFIX_INVALID;
1279 			return (EINVAL);
1280 		}
1281 	}
1282 
1283 	/* generate the MAC address */
1284 	if (prefix_len < addr_len) {
1285 		(void) random_get_pseudo_bytes(mac_addr +
1286 		    prefix_len, addr_len - prefix_len);
1287 	}
1288 
1289 	*diag = 0;
1290 	return (0);
1291 }
1292 
1293 /*
1294  * Set the priority range for this MAC client. This will be used to
1295  * determine the absolute priority for the threads created for this
1296  * MAC client using the specified "low", "medium" and "high" level.
1297  * This will also be used for any subflows on this MAC client.
1298  */
1299 #define	MAC_CLIENT_SET_PRIORITY_RANGE(mcip, pri) {			\
1300 	(mcip)->mci_min_pri = FLOW_MIN_PRIORITY(MINCLSYSPRI,	\
1301 	    MAXCLSYSPRI, (pri));					\
1302 	(mcip)->mci_max_pri = FLOW_MAX_PRIORITY(MINCLSYSPRI,	\
1303 	    MAXCLSYSPRI, (mcip)->mci_min_pri);				\
1304 	}
1305 
1306 /*
1307  * MAC client open entry point. Return a new MAC client handle. Each
1308  * MAC client is associated with a name, specified through the 'name'
1309  * argument.
1310  */
1311 int
1312 mac_client_open(mac_handle_t mh, mac_client_handle_t *mchp, char *name,
1313     uint16_t flags)
1314 {
1315 	mac_impl_t		*mip = (mac_impl_t *)mh;
1316 	mac_client_impl_t	*mcip;
1317 	int			err = 0;
1318 	boolean_t		share_desired;
1319 	flow_entry_t		*flent = NULL;
1320 
1321 	share_desired = (flags & MAC_OPEN_FLAGS_SHARES_DESIRED) != 0;
1322 	*mchp = NULL;
1323 
1324 	i_mac_perim_enter(mip);
1325 
1326 	if (mip->mi_state_flags & MIS_IS_VNIC) {
1327 		/*
1328 		 * The underlying MAC is a VNIC. Return the MAC client
1329 		 * handle of the lower MAC which was obtained by
1330 		 * the VNIC driver when it did its mac_client_open().
1331 		 */
1332 
1333 		mcip = mac_vnic_lower(mip);
1334 
1335 		/*
1336 		 * Note that multiple mac clients share the same mcip in
1337 		 * this case.
1338 		 */
1339 		if (flags & MAC_OPEN_FLAGS_EXCLUSIVE)
1340 			mcip->mci_state_flags |= MCIS_EXCLUSIVE;
1341 
1342 		if (flags & MAC_OPEN_FLAGS_MULTI_PRIMARY)
1343 			mcip->mci_flags |= MAC_CLIENT_FLAGS_MULTI_PRIMARY;
1344 
1345 		mip->mi_clients_list = mcip;
1346 		i_mac_perim_exit(mip);
1347 		*mchp = (mac_client_handle_t)mcip;
1348 
1349 		DTRACE_PROBE2(mac__client__open__nonallocated, mac_impl_t *,
1350 		    mcip->mci_mip, mac_client_impl_t *, mcip);
1351 
1352 		return (err);
1353 	}
1354 
1355 	mcip = kmem_cache_alloc(mac_client_impl_cache, KM_SLEEP);
1356 
1357 	mcip->mci_mip = mip;
1358 	mcip->mci_upper_mip = NULL;
1359 	mcip->mci_rx_fn = mac_pkt_drop;
1360 	mcip->mci_rx_arg = NULL;
1361 	mcip->mci_rx_p_fn = NULL;
1362 	mcip->mci_rx_p_arg = NULL;
1363 	mcip->mci_p_unicast_list = NULL;
1364 	mcip->mci_direct_rx_fn = NULL;
1365 	mcip->mci_direct_rx_arg = NULL;
1366 	mcip->mci_vidcache = MCIP_VIDCACHE_INVALID;
1367 
1368 	mcip->mci_unicast_list = NULL;
1369 
1370 	if ((flags & MAC_OPEN_FLAGS_IS_VNIC) != 0)
1371 		mcip->mci_state_flags |= MCIS_IS_VNIC;
1372 
1373 	if ((flags & MAC_OPEN_FLAGS_EXCLUSIVE) != 0)
1374 		mcip->mci_state_flags |= MCIS_EXCLUSIVE;
1375 
1376 	if ((flags & MAC_OPEN_FLAGS_IS_AGGR_PORT) != 0)
1377 		mcip->mci_state_flags |= MCIS_IS_AGGR_PORT;
1378 
1379 	if (mip->mi_state_flags & MIS_IS_AGGR)
1380 		mcip->mci_state_flags |= MCIS_IS_AGGR_CLIENT;
1381 
1382 	if ((flags & MAC_OPEN_FLAGS_USE_DATALINK_NAME) != 0) {
1383 		datalink_id_t	linkid;
1384 
1385 		ASSERT(name == NULL);
1386 		if ((err = dls_devnet_macname2linkid(mip->mi_name,
1387 		    &linkid)) != 0) {
1388 			goto done;
1389 		}
1390 		if ((err = dls_mgmt_get_linkinfo(linkid, mcip->mci_name, NULL,
1391 		    NULL, NULL)) != 0) {
1392 			/*
1393 			 * Use mac name if dlmgmtd is not available.
1394 			 */
1395 			if (err == EBADF) {
1396 				(void) strlcpy(mcip->mci_name, mip->mi_name,
1397 				    sizeof (mcip->mci_name));
1398 				err = 0;
1399 			} else {
1400 				goto done;
1401 			}
1402 		}
1403 		mcip->mci_state_flags |= MCIS_USE_DATALINK_NAME;
1404 	} else {
1405 		ASSERT(name != NULL);
1406 		if (strlen(name) > MAXNAMELEN) {
1407 			err = EINVAL;
1408 			goto done;
1409 		}
1410 		(void) strlcpy(mcip->mci_name, name, sizeof (mcip->mci_name));
1411 	}
1412 
1413 	if (flags & MAC_OPEN_FLAGS_MULTI_PRIMARY)
1414 		mcip->mci_flags |= MAC_CLIENT_FLAGS_MULTI_PRIMARY;
1415 
1416 	if (flags & MAC_OPEN_FLAGS_NO_UNICAST_ADDR)
1417 		mcip->mci_state_flags |= MCIS_NO_UNICAST_ADDR;
1418 
1419 	mac_protect_init(mcip);
1420 
1421 	/* the subflow table will be created dynamically */
1422 	mcip->mci_subflow_tab = NULL;
1423 
1424 	mcip->mci_misc_stat.mms_multircv = 0;
1425 	mcip->mci_misc_stat.mms_brdcstrcv = 0;
1426 	mcip->mci_misc_stat.mms_multixmt = 0;
1427 	mcip->mci_misc_stat.mms_brdcstxmt = 0;
1428 
1429 	/* Create an initial flow */
1430 
1431 	err = mac_flow_create(NULL, NULL, mcip->mci_name, NULL,
1432 	    mcip->mci_state_flags & MCIS_IS_VNIC ? FLOW_VNIC_MAC :
1433 	    FLOW_PRIMARY_MAC, &flent);
1434 	if (err != 0)
1435 		goto done;
1436 	mcip->mci_flent = flent;
1437 	FLOW_MARK(flent, FE_MC_NO_DATAPATH);
1438 	flent->fe_mcip = mcip;
1439 	/*
1440 	 * Place initial creation reference on the flow. This reference
1441 	 * is released in the corresponding delete action viz.
1442 	 * mac_unicast_remove after waiting for all transient refs to
1443 	 * to go away. The wait happens in mac_flow_wait.
1444 	 */
1445 	FLOW_REFHOLD(flent);
1446 
1447 	/*
1448 	 * Do this ahead of the mac_bcast_add() below so that the mi_nclients
1449 	 * will have the right value for mac_rx_srs_setup().
1450 	 */
1451 	mac_client_add(mcip);
1452 
1453 	mcip->mci_share = 0;
1454 	if (share_desired)
1455 		i_mac_share_alloc(mcip);
1456 
1457 	/*
1458 	 * We will do mimimal datapath setup to allow a MAC client to
1459 	 * transmit or receive non-unicast packets without waiting
1460 	 * for mac_unicast_add.
1461 	 */
1462 	if (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR) {
1463 		if ((err = mac_client_datapath_setup(mcip, VLAN_ID_NONE,
1464 		    NULL, NULL, B_TRUE, NULL)) != 0) {
1465 			goto done;
1466 		}
1467 	}
1468 
1469 	DTRACE_PROBE2(mac__client__open__allocated, mac_impl_t *,
1470 	    mcip->mci_mip, mac_client_impl_t *, mcip);
1471 
1472 	*mchp = (mac_client_handle_t)mcip;
1473 	i_mac_perim_exit(mip);
1474 	return (0);
1475 
1476 done:
1477 	i_mac_perim_exit(mip);
1478 	mcip->mci_state_flags = 0;
1479 	mcip->mci_tx_flag = 0;
1480 	kmem_cache_free(mac_client_impl_cache, mcip);
1481 	return (err);
1482 }
1483 
1484 /*
1485  * Close the specified MAC client handle.
1486  */
1487 void
1488 mac_client_close(mac_client_handle_t mch, uint16_t flags)
1489 {
1490 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
1491 	mac_impl_t		*mip = mcip->mci_mip;
1492 	flow_entry_t		*flent;
1493 
1494 	i_mac_perim_enter(mip);
1495 
1496 	if (flags & MAC_CLOSE_FLAGS_EXCLUSIVE)
1497 		mcip->mci_state_flags &= ~MCIS_EXCLUSIVE;
1498 
1499 	if ((mcip->mci_state_flags & MCIS_IS_VNIC) &&
1500 	    !(flags & MAC_CLOSE_FLAGS_IS_VNIC)) {
1501 		/*
1502 		 * This is an upper VNIC client initiated operation.
1503 		 * The lower MAC client will be closed by the VNIC driver
1504 		 * when the VNIC is deleted.
1505 		 */
1506 
1507 		i_mac_perim_exit(mip);
1508 		return;
1509 	}
1510 
1511 	/* If we have only setup up minimal datapth setup, tear it down */
1512 	if (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR) {
1513 		mac_client_datapath_teardown((mac_client_handle_t)mcip, NULL,
1514 		    mcip->mci_flent);
1515 		mcip->mci_state_flags &= ~MCIS_NO_UNICAST_ADDR;
1516 	}
1517 
1518 	/*
1519 	 * Remove the flent associated with the MAC client
1520 	 */
1521 	flent = mcip->mci_flent;
1522 	mcip->mci_flent = NULL;
1523 	FLOW_FINAL_REFRELE(flent);
1524 
1525 	/*
1526 	 * MAC clients must remove the unicast addresses and promisc callbacks
1527 	 * they added before issuing a mac_client_close().
1528 	 */
1529 	ASSERT(mcip->mci_unicast_list == NULL);
1530 	ASSERT(mcip->mci_promisc_list == NULL);
1531 	ASSERT(mcip->mci_tx_notify_cb_list == NULL);
1532 
1533 	i_mac_share_free(mcip);
1534 	mac_protect_fini(mcip);
1535 	mac_client_remove(mcip);
1536 
1537 	i_mac_perim_exit(mip);
1538 	mcip->mci_subflow_tab = NULL;
1539 	mcip->mci_state_flags = 0;
1540 	mcip->mci_tx_flag = 0;
1541 	kmem_cache_free(mac_client_impl_cache, mch);
1542 }
1543 
1544 /*
1545  * Set the Rx bypass receive callback and return B_TRUE. Return
1546  * B_FALSE if it's not possible to enable bypass.
1547  */
1548 boolean_t
1549 mac_rx_bypass_set(mac_client_handle_t mch, mac_direct_rx_t rx_fn, void *arg1)
1550 {
1551 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
1552 	mac_impl_t		*mip = mcip->mci_mip;
1553 
1554 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1555 
1556 	/*
1557 	 * If the client has more than one VLAN then process packets
1558 	 * through DLS. This should happen only when sun4v vsw is on
1559 	 * the scene.
1560 	 */
1561 	if (mcip->mci_nvids > 1)
1562 		return (B_FALSE);
1563 
1564 	/*
1565 	 * These are not accessed directly in the data path, and hence
1566 	 * don't need any protection
1567 	 */
1568 	mcip->mci_direct_rx_fn = rx_fn;
1569 	mcip->mci_direct_rx_arg = arg1;
1570 	return (B_TRUE);
1571 }
1572 
1573 /*
1574  * Enable/Disable rx bypass. By default, bypass is assumed to be enabled.
1575  */
1576 void
1577 mac_rx_bypass_enable(mac_client_handle_t mch)
1578 {
1579 	((mac_client_impl_t *)mch)->mci_state_flags &= ~MCIS_RX_BYPASS_DISABLE;
1580 }
1581 
1582 void
1583 mac_rx_bypass_disable(mac_client_handle_t mch)
1584 {
1585 	((mac_client_impl_t *)mch)->mci_state_flags |= MCIS_RX_BYPASS_DISABLE;
1586 }
1587 
1588 /*
1589  * Set the receive callback for the specified MAC client. There can be
1590  * at most one such callback per MAC client.
1591  */
1592 void
1593 mac_rx_set(mac_client_handle_t mch, mac_rx_t rx_fn, void *arg)
1594 {
1595 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1596 	mac_impl_t	*mip = mcip->mci_mip;
1597 	mac_impl_t	*umip = mcip->mci_upper_mip;
1598 
1599 	/*
1600 	 * Instead of adding an extra set of locks and refcnts in
1601 	 * the datapath at the mac client boundary, we temporarily quiesce
1602 	 * the SRS and related entities. We then change the receive function
1603 	 * without interference from any receive data thread and then reenable
1604 	 * the data flow subsequently.
1605 	 */
1606 	i_mac_perim_enter(mip);
1607 	mac_rx_client_quiesce(mch);
1608 
1609 	mcip->mci_rx_fn = rx_fn;
1610 	mcip->mci_rx_arg = arg;
1611 	mac_rx_client_restart(mch);
1612 	i_mac_perim_exit(mip);
1613 
1614 	/*
1615 	 * If we're changing the Rx function on the primary MAC of a VNIC,
1616 	 * make sure any secondary addresses on the VNIC are updated as well.
1617 	 */
1618 	if (umip != NULL) {
1619 		ASSERT((umip->mi_state_flags & MIS_IS_VNIC) != 0);
1620 		mac_vnic_secondary_update(umip);
1621 	}
1622 }
1623 
1624 /*
1625  * Reset the receive callback for the specified MAC client.
1626  */
1627 void
1628 mac_rx_clear(mac_client_handle_t mch)
1629 {
1630 	mac_rx_set(mch, mac_pkt_drop, NULL);
1631 }
1632 
1633 void
1634 mac_secondary_dup(mac_client_handle_t smch, mac_client_handle_t dmch)
1635 {
1636 	mac_client_impl_t *smcip = (mac_client_impl_t *)smch;
1637 	mac_client_impl_t *dmcip = (mac_client_impl_t *)dmch;
1638 	flow_entry_t *flent = dmcip->mci_flent;
1639 
1640 	/* This should only be called to setup secondary macs */
1641 	ASSERT((flent->fe_type & FLOW_PRIMARY_MAC) == 0);
1642 
1643 	mac_rx_set(dmch, smcip->mci_rx_fn, smcip->mci_rx_arg);
1644 	dmcip->mci_promisc_list = smcip->mci_promisc_list;
1645 
1646 	/*
1647 	 * Duplicate the primary mac resources to the secondary.
1648 	 * Since we already validated the resource controls when setting
1649 	 * them on the primary, we can ignore errors here.
1650 	 */
1651 	(void) mac_resource_ctl_set(dmch, MCIP_RESOURCE_PROPS(smcip));
1652 }
1653 
1654 /*
1655  * Called when removing a secondary MAC. Currently only clears the promisc_list
1656  * since we share the primary mac's promisc_list.
1657  */
1658 void
1659 mac_secondary_cleanup(mac_client_handle_t mch)
1660 {
1661 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1662 	flow_entry_t *flent = mcip->mci_flent;
1663 
1664 	/* This should only be called for secondary macs */
1665 	ASSERT((flent->fe_type & FLOW_PRIMARY_MAC) == 0);
1666 	mcip->mci_promisc_list = NULL;
1667 }
1668 
1669 /*
1670  * Walk the MAC client subflow table and updates their priority values.
1671  */
1672 static int
1673 mac_update_subflow_priority_cb(flow_entry_t *flent, void *arg)
1674 {
1675 	mac_flow_update_priority(arg, flent);
1676 	return (0);
1677 }
1678 
1679 void
1680 mac_update_subflow_priority(mac_client_impl_t *mcip)
1681 {
1682 	(void) mac_flow_walk(mcip->mci_subflow_tab,
1683 	    mac_update_subflow_priority_cb, mcip);
1684 }
1685 
1686 /*
1687  * Modify the TX or RX ring properties. We could either just move around
1688  * rings, i.e add/remove rings given to a client. Or this might cause the
1689  * client to move from hardware based to software or the other way around.
1690  * If we want to reset this property, then we clear the mask, additionally
1691  * if the client was given a non-default group we remove all rings except
1692  * for 1 and give it back to the default group.
1693  */
1694 int
1695 mac_client_set_rings_prop(mac_client_impl_t *mcip, mac_resource_props_t *mrp,
1696     mac_resource_props_t *tmrp)
1697 {
1698 	mac_impl_t		*mip = mcip->mci_mip;
1699 	flow_entry_t		*flent = mcip->mci_flent;
1700 	uint8_t			*mac_addr;
1701 	int			err = 0;
1702 	mac_group_t		*defgrp;
1703 	mac_group_t		*group;
1704 	mac_group_t		*ngrp;
1705 	mac_resource_props_t	*cmrp = MCIP_RESOURCE_PROPS(mcip);
1706 	uint_t			ringcnt;
1707 	boolean_t		unspec;
1708 
1709 	if (mcip->mci_share != 0)
1710 		return (EINVAL);
1711 
1712 	if (mrp->mrp_mask & MRP_RX_RINGS) {
1713 		unspec = mrp->mrp_mask & MRP_RXRINGS_UNSPEC;
1714 		group = flent->fe_rx_ring_group;
1715 		defgrp = MAC_DEFAULT_RX_GROUP(mip);
1716 		mac_addr = flent->fe_flow_desc.fd_dst_mac;
1717 
1718 		/*
1719 		 * No resulting change. If we are resetting on a client on
1720 		 * which there was no rx rings property. For dynamic group
1721 		 * if we are setting the same number of rings already set.
1722 		 * For static group if we are requesting a group again.
1723 		 */
1724 		if (mrp->mrp_mask & MRP_RINGS_RESET) {
1725 			if (!(tmrp->mrp_mask & MRP_RX_RINGS))
1726 				return (0);
1727 		} else {
1728 			if (unspec) {
1729 				if (tmrp->mrp_mask & MRP_RXRINGS_UNSPEC)
1730 					return (0);
1731 			} else if (mip->mi_rx_group_type ==
1732 			    MAC_GROUP_TYPE_DYNAMIC) {
1733 				if ((tmrp->mrp_mask & MRP_RX_RINGS) &&
1734 				    !(tmrp->mrp_mask & MRP_RXRINGS_UNSPEC) &&
1735 				    mrp->mrp_nrxrings == tmrp->mrp_nrxrings) {
1736 					return (0);
1737 				}
1738 			}
1739 		}
1740 		/* Resetting the prop */
1741 		if (mrp->mrp_mask & MRP_RINGS_RESET) {
1742 			/*
1743 			 * We will just keep one ring and give others back if
1744 			 * we are not the primary. For the primary we give
1745 			 * all the rings in the default group except the
1746 			 * default ring. If it is a static group, then
1747 			 * we don't do anything, but clear the MRP_RX_RINGS
1748 			 * flag.
1749 			 */
1750 			if (group != defgrp) {
1751 				if (mip->mi_rx_group_type ==
1752 				    MAC_GROUP_TYPE_DYNAMIC) {
1753 					/*
1754 					 * This group has reserved rings
1755 					 * that need to be released now,
1756 					 * so does the group.
1757 					 */
1758 					MAC_RX_RING_RELEASED(mip,
1759 					    group->mrg_cur_count);
1760 					MAC_RX_GRP_RELEASED(mip);
1761 					if ((flent->fe_type &
1762 					    FLOW_PRIMARY_MAC) != 0) {
1763 						if (mip->mi_nactiveclients ==
1764 						    1) {
1765 							(void)
1766 							    mac_rx_switch_group(
1767 							    mcip, group,
1768 							    defgrp);
1769 							return (0);
1770 						} else {
1771 							cmrp->mrp_nrxrings =
1772 							    group->
1773 							    mrg_cur_count +
1774 							    defgrp->
1775 							    mrg_cur_count - 1;
1776 						}
1777 					} else {
1778 						cmrp->mrp_nrxrings = 1;
1779 					}
1780 					(void) mac_group_ring_modify(mcip,
1781 					    group, defgrp);
1782 				} else {
1783 					/*
1784 					 * If this is a static group, we
1785 					 * need to release the group. The
1786 					 * client will remain in the same
1787 					 * group till some other client
1788 					 * needs this group.
1789 					 */
1790 					MAC_RX_GRP_RELEASED(mip);
1791 				}
1792 			/* Let check if we can give this an excl group */
1793 			} else if (group == defgrp) {
1794 				/*
1795 				 * If multiple clients share an
1796 				 * address then they must stay on the
1797 				 * default group.
1798 				 */
1799 				if (mac_check_macaddr_shared(mcip->mci_unicast))
1800 					return (0);
1801 
1802 				ngrp = mac_reserve_rx_group(mcip, mac_addr,
1803 				    B_TRUE);
1804 				/* Couldn't give it a group, that's fine */
1805 				if (ngrp == NULL)
1806 					return (0);
1807 				/* Switch to H/W */
1808 				if (mac_rx_switch_group(mcip, defgrp, ngrp) !=
1809 				    0) {
1810 					mac_stop_group(ngrp);
1811 					return (0);
1812 				}
1813 			}
1814 			/*
1815 			 * If the client is in the default group, we will
1816 			 * just clear the MRP_RX_RINGS and leave it as
1817 			 * it rather than look for an exclusive group
1818 			 * for it.
1819 			 */
1820 			return (0);
1821 		}
1822 
1823 		if (group == defgrp && ((mrp->mrp_nrxrings > 0) || unspec)) {
1824 			/*
1825 			 * We are requesting Rx rings. Try to reserve
1826 			 * a non-default group.
1827 			 *
1828 			 * If multiple clients share an address then
1829 			 * they must stay on the default group.
1830 			 */
1831 			if (mac_check_macaddr_shared(mcip->mci_unicast))
1832 				return (EINVAL);
1833 
1834 			ngrp = mac_reserve_rx_group(mcip, mac_addr, B_TRUE);
1835 			if (ngrp == NULL)
1836 				return (ENOSPC);
1837 
1838 			/* Switch to H/W */
1839 			if (mac_rx_switch_group(mcip, defgrp, ngrp) != 0) {
1840 				mac_release_rx_group(mcip, ngrp);
1841 				return (ENOSPC);
1842 			}
1843 			MAC_RX_GRP_RESERVED(mip);
1844 			if (mip->mi_rx_group_type == MAC_GROUP_TYPE_DYNAMIC)
1845 				MAC_RX_RING_RESERVED(mip, ngrp->mrg_cur_count);
1846 		} else if (group != defgrp && !unspec &&
1847 		    mrp->mrp_nrxrings == 0) {
1848 			/* Switch to S/W */
1849 			ringcnt = group->mrg_cur_count;
1850 			if (mac_rx_switch_group(mcip, group, defgrp) != 0)
1851 				return (ENOSPC);
1852 			if (tmrp->mrp_mask & MRP_RX_RINGS) {
1853 				MAC_RX_GRP_RELEASED(mip);
1854 				if (mip->mi_rx_group_type ==
1855 				    MAC_GROUP_TYPE_DYNAMIC) {
1856 					MAC_RX_RING_RELEASED(mip, ringcnt);
1857 				}
1858 			}
1859 		} else if (group != defgrp && mip->mi_rx_group_type ==
1860 		    MAC_GROUP_TYPE_DYNAMIC) {
1861 			ringcnt = group->mrg_cur_count;
1862 			err = mac_group_ring_modify(mcip, group, defgrp);
1863 			if (err != 0)
1864 				return (err);
1865 			/*
1866 			 * Update the accounting. If this group
1867 			 * already had explicitly reserved rings,
1868 			 * we need to update the rings based on
1869 			 * the new ring count. If this group
1870 			 * had not explicitly reserved rings,
1871 			 * then we just reserve the rings asked for
1872 			 * and reserve the group.
1873 			 */
1874 			if (tmrp->mrp_mask & MRP_RX_RINGS) {
1875 				if (ringcnt > group->mrg_cur_count) {
1876 					MAC_RX_RING_RELEASED(mip,
1877 					    ringcnt - group->mrg_cur_count);
1878 				} else {
1879 					MAC_RX_RING_RESERVED(mip,
1880 					    group->mrg_cur_count - ringcnt);
1881 				}
1882 			} else {
1883 				MAC_RX_RING_RESERVED(mip, group->mrg_cur_count);
1884 				MAC_RX_GRP_RESERVED(mip);
1885 			}
1886 		}
1887 	}
1888 	if (mrp->mrp_mask & MRP_TX_RINGS) {
1889 		unspec = mrp->mrp_mask & MRP_TXRINGS_UNSPEC;
1890 		group = flent->fe_tx_ring_group;
1891 		defgrp = MAC_DEFAULT_TX_GROUP(mip);
1892 
1893 		/*
1894 		 * For static groups we only allow rings=0 or resetting the
1895 		 * rings property.
1896 		 */
1897 		if (mrp->mrp_ntxrings > 0 &&
1898 		    mip->mi_tx_group_type != MAC_GROUP_TYPE_DYNAMIC) {
1899 			return (ENOTSUP);
1900 		}
1901 		if (mrp->mrp_mask & MRP_RINGS_RESET) {
1902 			if (!(tmrp->mrp_mask & MRP_TX_RINGS))
1903 				return (0);
1904 		} else {
1905 			if (unspec) {
1906 				if (tmrp->mrp_mask & MRP_TXRINGS_UNSPEC)
1907 					return (0);
1908 			} else if (mip->mi_tx_group_type ==
1909 			    MAC_GROUP_TYPE_DYNAMIC) {
1910 				if ((tmrp->mrp_mask & MRP_TX_RINGS) &&
1911 				    !(tmrp->mrp_mask & MRP_TXRINGS_UNSPEC) &&
1912 				    mrp->mrp_ntxrings == tmrp->mrp_ntxrings) {
1913 					return (0);
1914 				}
1915 			}
1916 		}
1917 		/* Resetting the prop */
1918 		if (mrp->mrp_mask & MRP_RINGS_RESET) {
1919 			if (group != defgrp) {
1920 				if (mip->mi_tx_group_type ==
1921 				    MAC_GROUP_TYPE_DYNAMIC) {
1922 					ringcnt = group->mrg_cur_count;
1923 					if ((flent->fe_type &
1924 					    FLOW_PRIMARY_MAC) != 0) {
1925 						mac_tx_client_quiesce(
1926 						    (mac_client_handle_t)
1927 						    mcip);
1928 						mac_tx_switch_group(mcip,
1929 						    group, defgrp);
1930 						mac_tx_client_restart(
1931 						    (mac_client_handle_t)
1932 						    mcip);
1933 						MAC_TX_GRP_RELEASED(mip);
1934 						MAC_TX_RING_RELEASED(mip,
1935 						    ringcnt);
1936 						return (0);
1937 					}
1938 					cmrp->mrp_ntxrings = 1;
1939 					(void) mac_group_ring_modify(mcip,
1940 					    group, defgrp);
1941 					/*
1942 					 * This group has reserved rings
1943 					 * that need to be released now.
1944 					 */
1945 					MAC_TX_RING_RELEASED(mip, ringcnt);
1946 				}
1947 				/*
1948 				 * If this is a static group, we
1949 				 * need to release the group. The
1950 				 * client will remain in the same
1951 				 * group till some other client
1952 				 * needs this group.
1953 				 */
1954 				MAC_TX_GRP_RELEASED(mip);
1955 			} else if (group == defgrp &&
1956 			    (flent->fe_type & FLOW_PRIMARY_MAC) == 0) {
1957 				ngrp = mac_reserve_tx_group(mcip, B_TRUE);
1958 				if (ngrp == NULL)
1959 					return (0);
1960 				mac_tx_client_quiesce(
1961 				    (mac_client_handle_t)mcip);
1962 				mac_tx_switch_group(mcip, defgrp, ngrp);
1963 				mac_tx_client_restart(
1964 				    (mac_client_handle_t)mcip);
1965 			}
1966 			/*
1967 			 * If the client is in the default group, we will
1968 			 * just clear the MRP_TX_RINGS and leave it as
1969 			 * it rather than look for an exclusive group
1970 			 * for it.
1971 			 */
1972 			return (0);
1973 		}
1974 
1975 		/* Switch to H/W */
1976 		if (group == defgrp && ((mrp->mrp_ntxrings > 0) || unspec)) {
1977 			ngrp = mac_reserve_tx_group(mcip, B_TRUE);
1978 			if (ngrp == NULL)
1979 				return (ENOSPC);
1980 			mac_tx_client_quiesce((mac_client_handle_t)mcip);
1981 			mac_tx_switch_group(mcip, defgrp, ngrp);
1982 			mac_tx_client_restart((mac_client_handle_t)mcip);
1983 			MAC_TX_GRP_RESERVED(mip);
1984 			if (mip->mi_tx_group_type == MAC_GROUP_TYPE_DYNAMIC)
1985 				MAC_TX_RING_RESERVED(mip, ngrp->mrg_cur_count);
1986 		/* Switch to S/W */
1987 		} else if (group != defgrp && !unspec &&
1988 		    mrp->mrp_ntxrings == 0) {
1989 			/* Switch to S/W */
1990 			ringcnt = group->mrg_cur_count;
1991 			mac_tx_client_quiesce((mac_client_handle_t)mcip);
1992 			mac_tx_switch_group(mcip, group, defgrp);
1993 			mac_tx_client_restart((mac_client_handle_t)mcip);
1994 			if (tmrp->mrp_mask & MRP_TX_RINGS) {
1995 				MAC_TX_GRP_RELEASED(mip);
1996 				if (mip->mi_tx_group_type ==
1997 				    MAC_GROUP_TYPE_DYNAMIC) {
1998 					MAC_TX_RING_RELEASED(mip, ringcnt);
1999 				}
2000 			}
2001 		} else if (group != defgrp && mip->mi_tx_group_type ==
2002 		    MAC_GROUP_TYPE_DYNAMIC) {
2003 			ringcnt = group->mrg_cur_count;
2004 			err = mac_group_ring_modify(mcip, group, defgrp);
2005 			if (err != 0)
2006 				return (err);
2007 			/*
2008 			 * Update the accounting. If this group
2009 			 * already had explicitly reserved rings,
2010 			 * we need to update the rings based on
2011 			 * the new ring count. If this group
2012 			 * had not explicitly reserved rings,
2013 			 * then we just reserve the rings asked for
2014 			 * and reserve the group.
2015 			 */
2016 			if (tmrp->mrp_mask & MRP_TX_RINGS) {
2017 				if (ringcnt > group->mrg_cur_count) {
2018 					MAC_TX_RING_RELEASED(mip,
2019 					    ringcnt - group->mrg_cur_count);
2020 				} else {
2021 					MAC_TX_RING_RESERVED(mip,
2022 					    group->mrg_cur_count - ringcnt);
2023 				}
2024 			} else {
2025 				MAC_TX_RING_RESERVED(mip, group->mrg_cur_count);
2026 				MAC_TX_GRP_RESERVED(mip);
2027 			}
2028 		}
2029 	}
2030 	return (0);
2031 }
2032 
2033 /*
2034  * When the MAC client is being brought up (i.e. we do a unicast_add) we need
2035  * to initialize the cpu and resource control structure in the
2036  * mac_client_impl_t from the mac_impl_t (i.e if there are any cached
2037  * properties before the flow entry for the unicast address was created).
2038  */
2039 static int
2040 mac_resource_ctl_set(mac_client_handle_t mch, mac_resource_props_t *mrp)
2041 {
2042 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2043 	mac_impl_t		*mip = (mac_impl_t *)mcip->mci_mip;
2044 	mac_impl_t		*umip = mcip->mci_upper_mip;
2045 	int			err = 0;
2046 	flow_entry_t		*flent = mcip->mci_flent;
2047 	mac_resource_props_t	*omrp, *nmrp = MCIP_RESOURCE_PROPS(mcip);
2048 
2049 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
2050 
2051 	err = mac_validate_props(mcip->mci_state_flags & MCIS_IS_VNIC ?
2052 	    mcip->mci_upper_mip : mip, mrp);
2053 	if (err != 0)
2054 		return (err);
2055 
2056 	/*
2057 	 * Copy over the existing properties since mac_update_resources
2058 	 * will modify the client's mrp. Currently, the saved property
2059 	 * is used to determine the difference between existing and
2060 	 * modified rings property.
2061 	 */
2062 	omrp = kmem_zalloc(sizeof (*omrp), KM_SLEEP);
2063 	bcopy(nmrp, omrp, sizeof (*omrp));
2064 	mac_update_resources(mrp, MCIP_RESOURCE_PROPS(mcip), B_FALSE);
2065 	if (MCIP_DATAPATH_SETUP(mcip)) {
2066 		/*
2067 		 * We support rings only for primary client when there are
2068 		 * multiple clients sharing the same MAC address (e.g. VLAN).
2069 		 */
2070 		if (mrp->mrp_mask & MRP_RX_RINGS ||
2071 		    mrp->mrp_mask & MRP_TX_RINGS) {
2072 
2073 			if ((err = mac_client_set_rings_prop(mcip, mrp,
2074 			    omrp)) != 0) {
2075 				if (omrp->mrp_mask & MRP_RX_RINGS) {
2076 					nmrp->mrp_mask |= MRP_RX_RINGS;
2077 					nmrp->mrp_nrxrings = omrp->mrp_nrxrings;
2078 				} else {
2079 					nmrp->mrp_mask &= ~MRP_RX_RINGS;
2080 					nmrp->mrp_nrxrings = 0;
2081 				}
2082 				if (omrp->mrp_mask & MRP_TX_RINGS) {
2083 					nmrp->mrp_mask |= MRP_TX_RINGS;
2084 					nmrp->mrp_ntxrings = omrp->mrp_ntxrings;
2085 				} else {
2086 					nmrp->mrp_mask &= ~MRP_TX_RINGS;
2087 					nmrp->mrp_ntxrings = 0;
2088 				}
2089 				if (omrp->mrp_mask & MRP_RXRINGS_UNSPEC)
2090 					omrp->mrp_mask |= MRP_RXRINGS_UNSPEC;
2091 				else
2092 					omrp->mrp_mask &= ~MRP_RXRINGS_UNSPEC;
2093 
2094 				if (omrp->mrp_mask & MRP_TXRINGS_UNSPEC)
2095 					omrp->mrp_mask |= MRP_TXRINGS_UNSPEC;
2096 				else
2097 					omrp->mrp_mask &= ~MRP_TXRINGS_UNSPEC;
2098 				kmem_free(omrp, sizeof (*omrp));
2099 				return (err);
2100 			}
2101 
2102 			/*
2103 			 * If we modified the rings property of the primary
2104 			 * we need to update the property fields of its
2105 			 * VLANs as they inherit the primary's properites.
2106 			 */
2107 			if (mac_is_primary_client(mcip)) {
2108 				mac_set_prim_vlan_rings(mip,
2109 				    MCIP_RESOURCE_PROPS(mcip));
2110 			}
2111 		}
2112 		/*
2113 		 * We have to set this prior to calling mac_flow_modify.
2114 		 */
2115 		if (mrp->mrp_mask & MRP_PRIORITY) {
2116 			if (mrp->mrp_priority == MPL_RESET) {
2117 				MAC_CLIENT_SET_PRIORITY_RANGE(mcip,
2118 				    MPL_LINK_DEFAULT);
2119 			} else {
2120 				MAC_CLIENT_SET_PRIORITY_RANGE(mcip,
2121 				    mrp->mrp_priority);
2122 			}
2123 		}
2124 
2125 		mac_flow_modify(mip->mi_flow_tab, flent, mrp);
2126 		if (mrp->mrp_mask & MRP_PRIORITY)
2127 			mac_update_subflow_priority(mcip);
2128 
2129 		/* Apply these resource settings to any secondary macs */
2130 		if (umip != NULL) {
2131 			ASSERT((umip->mi_state_flags & MIS_IS_VNIC) != 0);
2132 			mac_vnic_secondary_update(umip);
2133 		}
2134 	}
2135 	kmem_free(omrp, sizeof (*omrp));
2136 	return (0);
2137 }
2138 
2139 static int
2140 mac_unicast_flow_create(mac_client_impl_t *mcip, uint8_t *mac_addr,
2141     uint16_t vid, boolean_t is_primary, boolean_t first_flow,
2142     flow_entry_t **flent, mac_resource_props_t *mrp)
2143 {
2144 	mac_impl_t	*mip = (mac_impl_t *)mcip->mci_mip;
2145 	flow_desc_t	flow_desc;
2146 	char		flowname[MAXFLOWNAMELEN];
2147 	int		err;
2148 	uint_t		flent_flags;
2149 
2150 	/*
2151 	 * First unicast address being added, create a new flow
2152 	 * for that MAC client.
2153 	 */
2154 	bzero(&flow_desc, sizeof (flow_desc));
2155 
2156 	ASSERT(mac_addr != NULL ||
2157 	    (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR));
2158 	if (mac_addr != NULL) {
2159 		flow_desc.fd_mac_len = mip->mi_type->mt_addr_length;
2160 		bcopy(mac_addr, flow_desc.fd_dst_mac, flow_desc.fd_mac_len);
2161 	}
2162 	flow_desc.fd_mask = FLOW_LINK_DST;
2163 	if (vid != 0) {
2164 		flow_desc.fd_vid = vid;
2165 		flow_desc.fd_mask |= FLOW_LINK_VID;
2166 	}
2167 
2168 	/*
2169 	 * XXX-nicolas. For now I'm keeping the FLOW_PRIMARY_MAC
2170 	 * and FLOW_VNIC. Even though they're a hack inherited
2171 	 * from the SRS code, we'll keep them for now. They're currently
2172 	 * consumed by mac_datapath_setup() to create the SRS.
2173 	 * That code should be eventually moved out of
2174 	 * mac_datapath_setup() and moved to a mac_srs_create()
2175 	 * function of some sort to keep things clean.
2176 	 *
2177 	 * Also, there's no reason why the SRS for the primary MAC
2178 	 * client should be different than any other MAC client. Until
2179 	 * this is cleaned-up, we support only one MAC unicast address
2180 	 * per client.
2181 	 *
2182 	 * We set FLOW_PRIMARY_MAC for the primary MAC address,
2183 	 * FLOW_VNIC for everything else.
2184 	 */
2185 	if (is_primary)
2186 		flent_flags = FLOW_PRIMARY_MAC;
2187 	else
2188 		flent_flags = FLOW_VNIC_MAC;
2189 
2190 	/*
2191 	 * For the first flow we use the MAC client's name - mci_name, for
2192 	 * subsequent ones we just create a name with the VID. This is
2193 	 * so that we can add these flows to the same flow table. This is
2194 	 * fine as the flow name (except for the one with the MAC client's
2195 	 * name) is not visible. When the first flow is removed, we just replace
2196 	 * its fdesc with another from the list, so we will still retain the
2197 	 * flent with the MAC client's flow name.
2198 	 */
2199 	if (first_flow) {
2200 		bcopy(mcip->mci_name, flowname, MAXFLOWNAMELEN);
2201 	} else {
2202 		(void) sprintf(flowname, "%s%u", mcip->mci_name, vid);
2203 		flent_flags = FLOW_NO_STATS;
2204 	}
2205 
2206 	if ((err = mac_flow_create(&flow_desc, mrp, flowname, NULL,
2207 	    flent_flags, flent)) != 0)
2208 		return (err);
2209 
2210 	mac_misc_stat_create(*flent);
2211 	FLOW_MARK(*flent, FE_INCIPIENT);
2212 	(*flent)->fe_mcip = mcip;
2213 
2214 	/*
2215 	 * Place initial creation reference on the flow. This reference
2216 	 * is released in the corresponding delete action viz.
2217 	 * mac_unicast_remove after waiting for all transient refs to
2218 	 * to go away. The wait happens in mac_flow_wait.
2219 	 * We have already held the reference in mac_client_open().
2220 	 */
2221 	if (!first_flow)
2222 		FLOW_REFHOLD(*flent);
2223 	return (0);
2224 }
2225 
2226 /* Refresh the multicast grouping for this VID. */
2227 int
2228 mac_client_update_mcast(void *arg, boolean_t add, const uint8_t *addrp)
2229 {
2230 	flow_entry_t		*flent = arg;
2231 	mac_client_impl_t	*mcip = flent->fe_mcip;
2232 	uint16_t		vid;
2233 	flow_desc_t		flow_desc;
2234 
2235 	mac_flow_get_desc(flent, &flow_desc);
2236 	vid = (flow_desc.fd_mask & FLOW_LINK_VID) != 0 ?
2237 	    flow_desc.fd_vid : VLAN_ID_NONE;
2238 
2239 	/*
2240 	 * We don't call mac_multicast_add()/mac_multicast_remove() as
2241 	 * we want to add/remove for this specific vid.
2242 	 */
2243 	if (add) {
2244 		return (mac_bcast_add(mcip, addrp, vid,
2245 		    MAC_ADDRTYPE_MULTICAST));
2246 	} else {
2247 		mac_bcast_delete(mcip, addrp, vid);
2248 		return (0);
2249 	}
2250 }
2251 
2252 static void
2253 mac_update_single_active_client(mac_impl_t *mip)
2254 {
2255 	mac_client_impl_t *client = NULL;
2256 
2257 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
2258 
2259 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
2260 	if (mip->mi_nactiveclients == 1) {
2261 		/*
2262 		 * Find the one active MAC client from the list of MAC
2263 		 * clients. The active MAC client has at least one
2264 		 * unicast address.
2265 		 */
2266 		for (client = mip->mi_clients_list; client != NULL;
2267 		    client = client->mci_client_next) {
2268 			if (client->mci_unicast_list != NULL)
2269 				break;
2270 		}
2271 		ASSERT(client != NULL);
2272 	}
2273 
2274 	/*
2275 	 * mi_single_active_client is protected by the MAC impl's read/writer
2276 	 * lock, which allows mac_rx() to check the value of that pointer
2277 	 * as a reader.
2278 	 */
2279 	mip->mi_single_active_client = client;
2280 	rw_exit(&mip->mi_rw_lock);
2281 }
2282 
2283 /*
2284  * Set up the data path. Called from i_mac_unicast_add after having
2285  * done all the validations including making sure this is an active
2286  * client (i.e that is ready to process packets.)
2287  */
2288 static int
2289 mac_client_datapath_setup(mac_client_impl_t *mcip, uint16_t vid,
2290     uint8_t *mac_addr, mac_resource_props_t *mrp, boolean_t isprimary,
2291     mac_unicast_impl_t *muip)
2292 {
2293 	mac_impl_t	*mip = mcip->mci_mip;
2294 	boolean_t	mac_started = B_FALSE;
2295 	boolean_t	bcast_added = B_FALSE;
2296 	boolean_t	nactiveclients_added = B_FALSE;
2297 	flow_entry_t	*flent;
2298 	int		err = 0;
2299 	boolean_t	no_unicast;
2300 
2301 	no_unicast = mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR;
2302 
2303 	if ((err = mac_start((mac_handle_t)mip)) != 0)
2304 		goto bail;
2305 
2306 	mac_started = B_TRUE;
2307 
2308 	/* add the MAC client to the broadcast address group by default */
2309 	if (mip->mi_type->mt_brdcst_addr != NULL) {
2310 		err = mac_bcast_add(mcip, mip->mi_type->mt_brdcst_addr, vid,
2311 		    MAC_ADDRTYPE_BROADCAST);
2312 		if (err != 0)
2313 			goto bail;
2314 		bcast_added = B_TRUE;
2315 	}
2316 
2317 	/*
2318 	 * If this is the first unicast address addition for this
2319 	 * client, reuse the pre-allocated larval flow entry associated with
2320 	 * the MAC client.
2321 	 */
2322 	flent = (mcip->mci_nflents == 0) ? mcip->mci_flent : NULL;
2323 
2324 	/* We are configuring the unicast flow now */
2325 	if (!MCIP_DATAPATH_SETUP(mcip)) {
2326 
2327 		if (mrp != NULL) {
2328 			MAC_CLIENT_SET_PRIORITY_RANGE(mcip,
2329 			    (mrp->mrp_mask & MRP_PRIORITY) ? mrp->mrp_priority :
2330 			    MPL_LINK_DEFAULT);
2331 		}
2332 		if ((err = mac_unicast_flow_create(mcip, mac_addr, vid,
2333 		    isprimary, B_TRUE, &flent, mrp)) != 0)
2334 			goto bail;
2335 
2336 		mip->mi_nactiveclients++;
2337 		nactiveclients_added = B_TRUE;
2338 
2339 		/*
2340 		 * This will allocate the RX ring group if possible for the
2341 		 * flow and program the software classifier as needed.
2342 		 */
2343 		if ((err = mac_datapath_setup(mcip, flent, SRST_LINK)) != 0)
2344 			goto bail;
2345 
2346 		if (no_unicast)
2347 			goto done_setup;
2348 		/*
2349 		 * The unicast MAC address must have been added successfully.
2350 		 */
2351 		ASSERT(mcip->mci_unicast != NULL);
2352 
2353 		/*
2354 		 * Push down the sub-flows that were defined on this link
2355 		 * hitherto. The flows are added to the active flow table
2356 		 * and SRS, softrings etc. are created as needed.
2357 		 */
2358 		mac_link_init_flows((mac_client_handle_t)mcip);
2359 	} else {
2360 		mac_address_t *map = mcip->mci_unicast;
2361 
2362 		ASSERT(!no_unicast);
2363 		/*
2364 		 * A unicast flow already exists for that MAC client
2365 		 * so this flow must be the same MAC address but with
2366 		 * a different VID. It has been checked by
2367 		 * mac_addr_in_use().
2368 		 *
2369 		 * We will use the SRS etc. from the initial
2370 		 * mci_flent. We don't need to create a kstat for
2371 		 * this, as except for the fdesc, everything will be
2372 		 * used from the first flent.
2373 		 *
2374 		 * The only time we should see multiple flents on the
2375 		 * same MAC client is on the sun4v vsw. If we removed
2376 		 * that code we should be able to remove the entire
2377 		 * notion of multiple flents on a MAC client (this
2378 		 * doesn't affect sub/user flows because they have
2379 		 * their own list unrelated to mci_flent_list).
2380 		 */
2381 		if (bcmp(mac_addr, map->ma_addr, map->ma_len) != 0) {
2382 			err = EINVAL;
2383 			goto bail;
2384 		}
2385 
2386 		if ((err = mac_unicast_flow_create(mcip, mac_addr, vid,
2387 		    isprimary, B_FALSE, &flent, NULL)) != 0) {
2388 			goto bail;
2389 		}
2390 		if ((err = mac_flow_add(mip->mi_flow_tab, flent)) != 0) {
2391 			FLOW_FINAL_REFRELE(flent);
2392 			goto bail;
2393 		}
2394 
2395 		/* update the multicast group for this vid */
2396 		mac_client_bcast_refresh(mcip, mac_client_update_mcast,
2397 		    (void *)flent, B_TRUE);
2398 
2399 	}
2400 
2401 	/* populate the shared MAC address */
2402 	muip->mui_map = mcip->mci_unicast;
2403 
2404 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
2405 	muip->mui_next = mcip->mci_unicast_list;
2406 	mcip->mci_unicast_list = muip;
2407 	rw_exit(&mcip->mci_rw_lock);
2408 
2409 done_setup:
2410 	/*
2411 	 * First add the flent to the flow list of this mcip. Then set
2412 	 * the mip's mi_single_active_client if needed. The Rx path assumes
2413 	 * that mip->mi_single_active_client will always have an associated
2414 	 * flent.
2415 	 */
2416 	mac_client_add_to_flow_list(mcip, flent);
2417 	if (nactiveclients_added)
2418 		mac_update_single_active_client(mip);
2419 	/*
2420 	 * Trigger a renegotiation of the capabilities when the number of
2421 	 * active clients changes from 1 to 2, since some of the capabilities
2422 	 * might have to be disabled. Also send a MAC_NOTE_LINK notification
2423 	 * to all the MAC clients whenever physical link is DOWN.
2424 	 */
2425 	if (mip->mi_nactiveclients == 2) {
2426 		mac_capab_update((mac_handle_t)mip);
2427 		mac_virtual_link_update(mip);
2428 	}
2429 	/*
2430 	 * Now that the setup is complete, clear the INCIPIENT flag.
2431 	 * The flag was set to avoid incoming packets seeing inconsistent
2432 	 * structures while the setup was in progress. Clear the mci_tx_flag
2433 	 * by calling mac_tx_client_block. It is possible that
2434 	 * mac_unicast_remove was called prior to this mac_unicast_add which
2435 	 * could have set the MCI_TX_QUIESCE flag.
2436 	 */
2437 	if (flent->fe_rx_ring_group != NULL)
2438 		mac_rx_group_unmark(flent->fe_rx_ring_group, MR_INCIPIENT);
2439 	FLOW_UNMARK(flent, FE_INCIPIENT);
2440 	FLOW_UNMARK(flent, FE_MC_NO_DATAPATH);
2441 	mac_tx_client_unblock(mcip);
2442 	return (0);
2443 bail:
2444 	if (bcast_added)
2445 		mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr, vid);
2446 
2447 	if (nactiveclients_added)
2448 		mip->mi_nactiveclients--;
2449 
2450 	if (mac_started)
2451 		mac_stop((mac_handle_t)mip);
2452 
2453 	return (err);
2454 }
2455 
2456 /*
2457  * Return the passive primary MAC client, if present. The passive client is
2458  * a stand-by client that has the same unicast address as another that is
2459  * currenly active. Once the active client goes away, the passive client
2460  * becomes active.
2461  */
2462 static mac_client_impl_t *
2463 mac_get_passive_primary_client(mac_impl_t *mip)
2464 {
2465 	mac_client_impl_t	*mcip;
2466 
2467 	for (mcip = mip->mi_clients_list; mcip != NULL;
2468 	    mcip = mcip->mci_client_next) {
2469 		if (mac_is_primary_client(mcip) &&
2470 		    (mcip->mci_flags & MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
2471 			return (mcip);
2472 		}
2473 	}
2474 	return (NULL);
2475 }
2476 
2477 /*
2478  * Add a new unicast address to the MAC client.
2479  *
2480  * The MAC address can be specified either by value, or the MAC client
2481  * can specify that it wants to use the primary MAC address of the
2482  * underlying MAC. See the introductory comments at the beginning
2483  * of this file for more more information on primary MAC addresses.
2484  *
2485  * Note also the tuple (MAC address, VID) must be unique
2486  * for the MAC clients defined on top of the same underlying MAC
2487  * instance, unless the MAC_UNICAST_NODUPCHECK is specified.
2488  *
2489  * In no case can a client use the PVID for the MAC, if the MAC has one set.
2490  */
2491 int
2492 i_mac_unicast_add(mac_client_handle_t mch, uint8_t *mac_addr, uint16_t flags,
2493     mac_unicast_handle_t *mah, uint16_t vid, mac_diag_t *diag)
2494 {
2495 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2496 	mac_impl_t		*mip = mcip->mci_mip;
2497 	int			err;
2498 	uint_t			mac_len = mip->mi_type->mt_addr_length;
2499 	boolean_t		check_dups = !(flags & MAC_UNICAST_NODUPCHECK);
2500 	boolean_t		fastpath_disabled = B_FALSE;
2501 	boolean_t		is_primary = (flags & MAC_UNICAST_PRIMARY);
2502 	boolean_t		is_unicast_hw = (flags & MAC_UNICAST_HW);
2503 	mac_resource_props_t	*mrp;
2504 	boolean_t		passive_client = B_FALSE;
2505 	mac_unicast_impl_t	*muip;
2506 	boolean_t		is_vnic_primary =
2507 	    (flags & MAC_UNICAST_VNIC_PRIMARY);
2508 
2509 	/*
2510 	 * When the VID is non-zero the underlying MAC cannot be a
2511 	 * VNIC. I.e., dladm create-vlan cannot take a VNIC as
2512 	 * argument, only the primary MAC client.
2513 	 */
2514 	ASSERT(!((mip->mi_state_flags & MIS_IS_VNIC) && (vid != VLAN_ID_NONE)));
2515 
2516 	/*
2517 	 * Can't unicast add if the client asked only for minimal datapath
2518 	 * setup.
2519 	 */
2520 	if (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR)
2521 		return (ENOTSUP);
2522 
2523 	/*
2524 	 * Check for an attempted use of the current Port VLAN ID, if enabled.
2525 	 * No client may use it.
2526 	 */
2527 	if (mip->mi_pvid != VLAN_ID_NONE && vid == mip->mi_pvid)
2528 		return (EBUSY);
2529 
2530 	/*
2531 	 * Check whether it's the primary client and flag it.
2532 	 */
2533 	if (!(mcip->mci_state_flags & MCIS_IS_VNIC) && is_primary &&
2534 	    vid == VLAN_ID_NONE)
2535 		mcip->mci_flags |= MAC_CLIENT_FLAGS_PRIMARY;
2536 
2537 	/*
2538 	 * is_vnic_primary is true when we come here as a VLAN VNIC
2539 	 * which uses the primary MAC client's address but with a non-zero
2540 	 * VID. In this case the MAC address is not specified by an upper
2541 	 * MAC client.
2542 	 */
2543 	if ((mcip->mci_state_flags & MCIS_IS_VNIC) && is_primary &&
2544 	    !is_vnic_primary) {
2545 		/*
2546 		 * The address is being set by the upper MAC client
2547 		 * of a VNIC. The MAC address was already set by the
2548 		 * VNIC driver during VNIC creation.
2549 		 *
2550 		 * Note: a VNIC has only one MAC address. We return
2551 		 * the MAC unicast address handle of the lower MAC client
2552 		 * corresponding to the VNIC. We allocate a new entry
2553 		 * which is flagged appropriately, so that mac_unicast_remove()
2554 		 * doesn't attempt to free the original entry that
2555 		 * was allocated by the VNIC driver.
2556 		 */
2557 		ASSERT(mcip->mci_unicast != NULL);
2558 
2559 		/* Check for VLAN flags, if present */
2560 		if ((flags & MAC_UNICAST_TAG_DISABLE) != 0)
2561 			mcip->mci_state_flags |= MCIS_TAG_DISABLE;
2562 
2563 		if ((flags & MAC_UNICAST_STRIP_DISABLE) != 0)
2564 			mcip->mci_state_flags |= MCIS_STRIP_DISABLE;
2565 
2566 		if ((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) != 0)
2567 			mcip->mci_state_flags |= MCIS_DISABLE_TX_VID_CHECK;
2568 
2569 		/*
2570 		 * Ensure that the primary unicast address of the VNIC
2571 		 * is added only once unless we have the
2572 		 * MAC_CLIENT_FLAGS_MULTI_PRIMARY set (and this is not
2573 		 * a passive MAC client).
2574 		 */
2575 		if ((mcip->mci_flags & MAC_CLIENT_FLAGS_VNIC_PRIMARY) != 0) {
2576 			if ((mcip->mci_flags &
2577 			    MAC_CLIENT_FLAGS_MULTI_PRIMARY) == 0 ||
2578 			    (mcip->mci_flags &
2579 			    MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
2580 				return (EBUSY);
2581 			}
2582 			mcip->mci_flags |= MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
2583 			passive_client = B_TRUE;
2584 		}
2585 
2586 		mcip->mci_flags |= MAC_CLIENT_FLAGS_VNIC_PRIMARY;
2587 
2588 		/*
2589 		 * Create a handle for vid 0.
2590 		 */
2591 		ASSERT(vid == VLAN_ID_NONE);
2592 		muip = kmem_zalloc(sizeof (mac_unicast_impl_t), KM_SLEEP);
2593 		muip->mui_vid = vid;
2594 		*mah = (mac_unicast_handle_t)muip;
2595 		/*
2596 		 * This will be used by the caller to defer setting the
2597 		 * rx functions.
2598 		 */
2599 		if (passive_client)
2600 			return (EAGAIN);
2601 		return (0);
2602 	}
2603 
2604 	/* primary MAC clients cannot be opened on top of anchor VNICs */
2605 	if ((is_vnic_primary || is_primary) &&
2606 	    i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_ANCHOR_VNIC, NULL)) {
2607 		return (ENXIO);
2608 	}
2609 
2610 	/*
2611 	 * If this is a VNIC/VLAN, disable softmac fast-path. This is
2612 	 * only relevant to legacy devices which use softmac to
2613 	 * interface with GLDv3.
2614 	 */
2615 	if (mcip->mci_state_flags & MCIS_IS_VNIC) {
2616 		err = mac_fastpath_disable((mac_handle_t)mip);
2617 		if (err != 0)
2618 			return (err);
2619 		fastpath_disabled = B_TRUE;
2620 	}
2621 
2622 	/*
2623 	 * Return EBUSY if:
2624 	 *  - there is an exclusively active mac client exists.
2625 	 *  - this is an exclusive active mac client but
2626 	 *	a. there is already active mac clients exist, or
2627 	 *	b. fastpath streams are already plumbed on this legacy device
2628 	 *  - the mac creator has disallowed active mac clients.
2629 	 */
2630 	if (mip->mi_state_flags & (MIS_EXCLUSIVE|MIS_NO_ACTIVE)) {
2631 		if (fastpath_disabled)
2632 			mac_fastpath_enable((mac_handle_t)mip);
2633 		return (EBUSY);
2634 	}
2635 
2636 	if (mcip->mci_state_flags & MCIS_EXCLUSIVE) {
2637 		ASSERT(!fastpath_disabled);
2638 		if (mip->mi_nactiveclients != 0)
2639 			return (EBUSY);
2640 
2641 		if ((mip->mi_state_flags & MIS_LEGACY) &&
2642 		    !(mip->mi_capab_legacy.ml_active_set(mip->mi_driver))) {
2643 			return (EBUSY);
2644 		}
2645 		mip->mi_state_flags |= MIS_EXCLUSIVE;
2646 	}
2647 
2648 	mrp = kmem_zalloc(sizeof (*mrp), KM_SLEEP);
2649 	if (is_primary && !(mcip->mci_state_flags & (MCIS_IS_VNIC |
2650 	    MCIS_IS_AGGR_PORT))) {
2651 		/*
2652 		 * Apply the property cached in the mac_impl_t to the primary
2653 		 * mac client. If the mac client is a VNIC or an aggregation
2654 		 * port, its property should be set in the mcip when the
2655 		 * VNIC/aggr was created.
2656 		 */
2657 		mac_get_resources((mac_handle_t)mip, mrp);
2658 		(void) mac_client_set_resources(mch, mrp);
2659 	} else if (mcip->mci_state_flags & MCIS_IS_VNIC) {
2660 		/*
2661 		 * This is a VLAN client sharing the address of the
2662 		 * primary MAC client; i.e., one created via dladm
2663 		 * create-vlan. We don't support specifying ring
2664 		 * properties for this type of client as it inherits
2665 		 * these from the primary MAC client.
2666 		 */
2667 		if (is_vnic_primary) {
2668 			mac_resource_props_t	*vmrp;
2669 
2670 			vmrp = MCIP_RESOURCE_PROPS(mcip);
2671 			if (vmrp->mrp_mask & MRP_RX_RINGS ||
2672 			    vmrp->mrp_mask & MRP_TX_RINGS) {
2673 				if (fastpath_disabled)
2674 					mac_fastpath_enable((mac_handle_t)mip);
2675 				kmem_free(mrp, sizeof (*mrp));
2676 				return (ENOTSUP);
2677 			}
2678 			/*
2679 			 * Additionally we also need to inherit any
2680 			 * rings property from the MAC.
2681 			 */
2682 			mac_get_resources((mac_handle_t)mip, mrp);
2683 			if (mrp->mrp_mask & MRP_RX_RINGS) {
2684 				vmrp->mrp_mask |= MRP_RX_RINGS;
2685 				vmrp->mrp_nrxrings = mrp->mrp_nrxrings;
2686 			}
2687 			if (mrp->mrp_mask & MRP_TX_RINGS) {
2688 				vmrp->mrp_mask |= MRP_TX_RINGS;
2689 				vmrp->mrp_ntxrings = mrp->mrp_ntxrings;
2690 			}
2691 		}
2692 		bcopy(MCIP_RESOURCE_PROPS(mcip), mrp, sizeof (*mrp));
2693 	}
2694 
2695 	muip = kmem_zalloc(sizeof (mac_unicast_impl_t), KM_SLEEP);
2696 	muip->mui_vid = vid;
2697 
2698 	if (is_primary || is_vnic_primary) {
2699 		mac_addr = mip->mi_addr;
2700 	} else {
2701 
2702 		/*
2703 		 * Verify the validity of the specified MAC addresses value.
2704 		 */
2705 		if (!mac_unicst_verify((mac_handle_t)mip, mac_addr, mac_len)) {
2706 			*diag = MAC_DIAG_MACADDR_INVALID;
2707 			err = EINVAL;
2708 			goto bail_out;
2709 		}
2710 
2711 		/*
2712 		 * Make sure that the specified MAC address is different
2713 		 * than the unicast MAC address of the underlying NIC.
2714 		 */
2715 		if (check_dups && bcmp(mip->mi_addr, mac_addr, mac_len) == 0) {
2716 			*diag = MAC_DIAG_MACADDR_NIC;
2717 			err = EINVAL;
2718 			goto bail_out;
2719 		}
2720 	}
2721 
2722 	/*
2723 	 * Set the flags here so that if this is a passive client, we
2724 	 * can return and set it when we call mac_client_datapath_setup
2725 	 * when this becomes the active client. If we defer to using these
2726 	 * flags to mac_client_datapath_setup, then for a passive client,
2727 	 * we'd have to store the flags somewhere (probably fe_flags)
2728 	 * and then use it.
2729 	 */
2730 	if (!MCIP_DATAPATH_SETUP(mcip)) {
2731 		if (is_unicast_hw) {
2732 			/*
2733 			 * The client requires a hardware MAC address slot
2734 			 * for that unicast address. Since we support only
2735 			 * one unicast MAC address per client, flag the
2736 			 * MAC client itself.
2737 			 */
2738 			mcip->mci_state_flags |= MCIS_UNICAST_HW;
2739 		}
2740 
2741 		/* Check for VLAN flags, if present */
2742 		if ((flags & MAC_UNICAST_TAG_DISABLE) != 0)
2743 			mcip->mci_state_flags |= MCIS_TAG_DISABLE;
2744 
2745 		if ((flags & MAC_UNICAST_STRIP_DISABLE) != 0)
2746 			mcip->mci_state_flags |= MCIS_STRIP_DISABLE;
2747 
2748 		if ((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) != 0)
2749 			mcip->mci_state_flags |= MCIS_DISABLE_TX_VID_CHECK;
2750 	} else {
2751 		/*
2752 		 * Assert that the specified flags are consistent with the
2753 		 * flags specified by previous calls to mac_unicast_add().
2754 		 */
2755 		ASSERT(((flags & MAC_UNICAST_TAG_DISABLE) != 0 &&
2756 		    (mcip->mci_state_flags & MCIS_TAG_DISABLE) != 0) ||
2757 		    ((flags & MAC_UNICAST_TAG_DISABLE) == 0 &&
2758 		    (mcip->mci_state_flags & MCIS_TAG_DISABLE) == 0));
2759 
2760 		ASSERT(((flags & MAC_UNICAST_STRIP_DISABLE) != 0 &&
2761 		    (mcip->mci_state_flags & MCIS_STRIP_DISABLE) != 0) ||
2762 		    ((flags & MAC_UNICAST_STRIP_DISABLE) == 0 &&
2763 		    (mcip->mci_state_flags & MCIS_STRIP_DISABLE) == 0));
2764 
2765 		ASSERT(((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) != 0 &&
2766 		    (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK) != 0) ||
2767 		    ((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) == 0 &&
2768 		    (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK) == 0));
2769 
2770 		/*
2771 		 * Make sure the client is consistent about its requests
2772 		 * for MAC addresses. I.e. all requests from the clients
2773 		 * must have the MAC_UNICAST_HW flag set or clear.
2774 		 */
2775 		if (((mcip->mci_state_flags & MCIS_UNICAST_HW) != 0 &&
2776 		    !is_unicast_hw) ||
2777 		    ((mcip->mci_state_flags & MCIS_UNICAST_HW) == 0 &&
2778 		    is_unicast_hw)) {
2779 			err = EINVAL;
2780 			goto bail_out;
2781 		}
2782 	}
2783 	/*
2784 	 * Make sure the MAC address is not already used by
2785 	 * another MAC client defined on top of the same
2786 	 * underlying NIC. Unless we have MAC_CLIENT_FLAGS_MULTI_PRIMARY
2787 	 * set when we allow a passive client to be present which will
2788 	 * be activated when the currently active client goes away - this
2789 	 * works only with primary addresses.
2790 	 */
2791 	if ((check_dups || is_primary || is_vnic_primary) &&
2792 	    mac_addr_in_use(mip, mac_addr, vid)) {
2793 		/*
2794 		 * Must have set the multiple primary address flag when
2795 		 * we did a mac_client_open AND this should be a primary
2796 		 * MAC client AND there should not already be a passive
2797 		 * primary. If all is true then we let this succeed
2798 		 * even if the address is a dup.
2799 		 */
2800 		if ((mcip->mci_flags & MAC_CLIENT_FLAGS_MULTI_PRIMARY) == 0 ||
2801 		    (mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY) == 0 ||
2802 		    mac_get_passive_primary_client(mip) != NULL) {
2803 			*diag = MAC_DIAG_MACADDR_INUSE;
2804 			err = EEXIST;
2805 			goto bail_out;
2806 		}
2807 		ASSERT((mcip->mci_flags &
2808 		    MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) == 0);
2809 		mcip->mci_flags |= MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
2810 		kmem_free(mrp, sizeof (*mrp));
2811 
2812 		/*
2813 		 * Stash the unicast address handle, we will use it when
2814 		 * we set up the passive client.
2815 		 */
2816 		mcip->mci_p_unicast_list = muip;
2817 		*mah = (mac_unicast_handle_t)muip;
2818 		return (0);
2819 	}
2820 
2821 	err = mac_client_datapath_setup(mcip, vid, mac_addr, mrp,
2822 	    is_primary || is_vnic_primary, muip);
2823 	if (err != 0)
2824 		goto bail_out;
2825 
2826 	kmem_free(mrp, sizeof (*mrp));
2827 	*mah = (mac_unicast_handle_t)muip;
2828 	return (0);
2829 
2830 bail_out:
2831 	if (fastpath_disabled)
2832 		mac_fastpath_enable((mac_handle_t)mip);
2833 	if (mcip->mci_state_flags & MCIS_EXCLUSIVE) {
2834 		mip->mi_state_flags &= ~MIS_EXCLUSIVE;
2835 		if (mip->mi_state_flags & MIS_LEGACY) {
2836 			mip->mi_capab_legacy.ml_active_clear(
2837 			    mip->mi_driver);
2838 		}
2839 	}
2840 	kmem_free(mrp, sizeof (*mrp));
2841 	kmem_free(muip, sizeof (mac_unicast_impl_t));
2842 	return (err);
2843 }
2844 
2845 /*
2846  * Wrapper function to mac_unicast_add when we want to have the same mac
2847  * client open for two instances, one that is currently active and another
2848  * that will become active when the current one is removed. In this case
2849  * mac_unicast_add will return EGAIN and we will save the rx function and
2850  * arg which will be used when we activate the passive client in
2851  * mac_unicast_remove.
2852  */
2853 int
2854 mac_unicast_add_set_rx(mac_client_handle_t mch, uint8_t *mac_addr,
2855     uint16_t flags, mac_unicast_handle_t *mah,  uint16_t vid, mac_diag_t *diag,
2856     mac_rx_t rx_fn, void *arg)
2857 {
2858 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2859 	uint_t			err;
2860 
2861 	err = mac_unicast_add(mch, mac_addr, flags, mah, vid, diag);
2862 	if (err != 0 && err != EAGAIN)
2863 		return (err);
2864 	if (err == EAGAIN) {
2865 		if (rx_fn != NULL) {
2866 			mcip->mci_rx_p_fn = rx_fn;
2867 			mcip->mci_rx_p_arg = arg;
2868 		}
2869 		return (0);
2870 	}
2871 	if (rx_fn != NULL)
2872 		mac_rx_set(mch, rx_fn, arg);
2873 	return (err);
2874 }
2875 
2876 int
2877 mac_unicast_add(mac_client_handle_t mch, uint8_t *mac_addr, uint16_t flags,
2878     mac_unicast_handle_t *mah, uint16_t vid, mac_diag_t *diag)
2879 {
2880 	mac_impl_t *mip = ((mac_client_impl_t *)mch)->mci_mip;
2881 	uint_t err;
2882 
2883 	i_mac_perim_enter(mip);
2884 	err = i_mac_unicast_add(mch, mac_addr, flags, mah, vid, diag);
2885 	i_mac_perim_exit(mip);
2886 
2887 	return (err);
2888 }
2889 
2890 static void
2891 mac_client_datapath_teardown(mac_client_handle_t mch, mac_unicast_impl_t *muip,
2892     flow_entry_t *flent)
2893 {
2894 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2895 	mac_impl_t		*mip = mcip->mci_mip;
2896 	boolean_t		no_unicast;
2897 
2898 	/*
2899 	 * If we have not added a unicast address for this MAC client, just
2900 	 * teardown the datapath.
2901 	 */
2902 	no_unicast = mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR;
2903 
2904 	if (!no_unicast) {
2905 		/*
2906 		 * We would have initialized subflows etc. only if we brought
2907 		 * up the primary client and set the unicast unicast address
2908 		 * etc. Deactivate the flows. The flow entry will be removed
2909 		 * from the active flow tables, and the associated SRS,
2910 		 * softrings etc will be deleted. But the flow entry itself
2911 		 * won't be destroyed, instead it will continue to be archived
2912 		 * off the  the global flow hash list, for a possible future
2913 		 * activation when say IP is plumbed again.
2914 		 */
2915 		mac_link_release_flows(mch);
2916 	}
2917 	mip->mi_nactiveclients--;
2918 	mac_update_single_active_client(mip);
2919 
2920 	/* Tear down the data path */
2921 	mac_datapath_teardown(mcip, mcip->mci_flent, SRST_LINK);
2922 
2923 	/*
2924 	 * Prevent any future access to the flow entry through the mci_flent
2925 	 * pointer by setting the mci_flent to NULL. Access to mci_flent in
2926 	 * mac_bcast_send is also under mi_rw_lock.
2927 	 */
2928 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
2929 	flent = mcip->mci_flent;
2930 	mac_client_remove_flow_from_list(mcip, flent);
2931 
2932 	if (mcip->mci_state_flags & MCIS_DESC_LOGGED)
2933 		mcip->mci_state_flags &= ~MCIS_DESC_LOGGED;
2934 
2935 	/*
2936 	 * This is the last unicast address being removed and there shouldn't
2937 	 * be any outbound data threads at this point coming down from mac
2938 	 * clients. We have waited for the data threads to finish before
2939 	 * starting dld_str_detach. Non-data threads must access TX SRS
2940 	 * under mi_rw_lock.
2941 	 */
2942 	rw_exit(&mip->mi_rw_lock);
2943 
2944 	/*
2945 	 * Don't use FLOW_MARK with FE_MC_NO_DATAPATH, as the flow might
2946 	 * contain other flags, such as FE_CONDEMNED, which we need to
2947 	 * cleared. We don't call mac_flow_cleanup() for this unicast
2948 	 * flow as we have a already cleaned up SRSs etc. (via the teadown
2949 	 * path). We just clear the stats and reset the initial callback
2950 	 * function, the rest will be set when we call mac_flow_create,
2951 	 * if at all.
2952 	 */
2953 	mutex_enter(&flent->fe_lock);
2954 	ASSERT(flent->fe_refcnt == 1 && flent->fe_mbg == NULL &&
2955 	    flent->fe_tx_srs == NULL && flent->fe_rx_srs_cnt == 0);
2956 	flent->fe_flags = FE_MC_NO_DATAPATH;
2957 	flow_stat_destroy(flent);
2958 	mac_misc_stat_delete(flent);
2959 
2960 	/* Initialize the receiver function to a safe routine */
2961 	flent->fe_cb_fn = (flow_fn_t)mac_pkt_drop;
2962 	flent->fe_cb_arg1 = NULL;
2963 	flent->fe_cb_arg2 = NULL;
2964 
2965 	flent->fe_index = -1;
2966 	mutex_exit(&flent->fe_lock);
2967 
2968 	if (mip->mi_type->mt_brdcst_addr != NULL) {
2969 		ASSERT(muip != NULL || no_unicast);
2970 		mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr,
2971 		    muip != NULL ? muip->mui_vid : VLAN_ID_NONE);
2972 	}
2973 
2974 	if (mip->mi_nactiveclients == 1) {
2975 		mac_capab_update((mac_handle_t)mip);
2976 		mac_virtual_link_update(mip);
2977 	}
2978 
2979 	if (mcip->mci_state_flags & MCIS_EXCLUSIVE) {
2980 		mip->mi_state_flags &= ~MIS_EXCLUSIVE;
2981 
2982 		if (mip->mi_state_flags & MIS_LEGACY)
2983 			mip->mi_capab_legacy.ml_active_clear(mip->mi_driver);
2984 	}
2985 
2986 	mcip->mci_state_flags &= ~MCIS_UNICAST_HW;
2987 
2988 	if (mcip->mci_state_flags & MCIS_TAG_DISABLE)
2989 		mcip->mci_state_flags &= ~MCIS_TAG_DISABLE;
2990 
2991 	if (mcip->mci_state_flags & MCIS_STRIP_DISABLE)
2992 		mcip->mci_state_flags &= ~MCIS_STRIP_DISABLE;
2993 
2994 	if (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK)
2995 		mcip->mci_state_flags &= ~MCIS_DISABLE_TX_VID_CHECK;
2996 
2997 	if (muip != NULL)
2998 		kmem_free(muip, sizeof (mac_unicast_impl_t));
2999 	mac_protect_cancel_timer(mcip);
3000 	mac_protect_flush_dynamic(mcip);
3001 
3002 	bzero(&mcip->mci_misc_stat, sizeof (mcip->mci_misc_stat));
3003 	/*
3004 	 * Disable fastpath if this is a VNIC or a VLAN.
3005 	 */
3006 	if (mcip->mci_state_flags & MCIS_IS_VNIC)
3007 		mac_fastpath_enable((mac_handle_t)mip);
3008 	mac_stop((mac_handle_t)mip);
3009 }
3010 
3011 /*
3012  * Remove a MAC address which was previously added by mac_unicast_add().
3013  */
3014 int
3015 mac_unicast_remove(mac_client_handle_t mch, mac_unicast_handle_t mah)
3016 {
3017 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3018 	mac_unicast_impl_t *muip = (mac_unicast_impl_t *)mah;
3019 	mac_unicast_impl_t *pre;
3020 	mac_impl_t *mip = mcip->mci_mip;
3021 	flow_entry_t		*flent;
3022 	uint16_t mui_vid;
3023 
3024 	i_mac_perim_enter(mip);
3025 	if (mcip->mci_flags & MAC_CLIENT_FLAGS_VNIC_PRIMARY) {
3026 		/*
3027 		 * Call made by the upper MAC client of a VNIC.
3028 		 * There's nothing much to do, the unicast address will
3029 		 * be removed by the VNIC driver when the VNIC is deleted,
3030 		 * but let's ensure that all our transmit is done before
3031 		 * the client does a mac_client_stop lest it trigger an
3032 		 * assert in the driver.
3033 		 */
3034 		ASSERT(muip->mui_vid == VLAN_ID_NONE);
3035 
3036 		mac_tx_client_flush(mcip);
3037 
3038 		if ((mcip->mci_flags & MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
3039 			mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
3040 			if (mcip->mci_rx_p_fn != NULL) {
3041 				mac_rx_set(mch, mcip->mci_rx_p_fn,
3042 				    mcip->mci_rx_p_arg);
3043 				mcip->mci_rx_p_fn = NULL;
3044 				mcip->mci_rx_p_arg = NULL;
3045 			}
3046 			kmem_free(muip, sizeof (mac_unicast_impl_t));
3047 			i_mac_perim_exit(mip);
3048 			return (0);
3049 		}
3050 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_VNIC_PRIMARY;
3051 
3052 		if (mcip->mci_state_flags & MCIS_TAG_DISABLE)
3053 			mcip->mci_state_flags &= ~MCIS_TAG_DISABLE;
3054 
3055 		if (mcip->mci_state_flags & MCIS_STRIP_DISABLE)
3056 			mcip->mci_state_flags &= ~MCIS_STRIP_DISABLE;
3057 
3058 		if (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK)
3059 			mcip->mci_state_flags &= ~MCIS_DISABLE_TX_VID_CHECK;
3060 
3061 		kmem_free(muip, sizeof (mac_unicast_impl_t));
3062 		i_mac_perim_exit(mip);
3063 		return (0);
3064 	}
3065 
3066 	ASSERT(muip != NULL);
3067 
3068 	/*
3069 	 * We are removing a passive client, we haven't setup the datapath
3070 	 * for this yet, so nothing much to do.
3071 	 */
3072 	if ((mcip->mci_flags & MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
3073 
3074 		ASSERT((mcip->mci_flent->fe_flags & FE_MC_NO_DATAPATH) != 0);
3075 		ASSERT(mcip->mci_p_unicast_list == muip);
3076 
3077 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
3078 
3079 		mcip->mci_p_unicast_list = NULL;
3080 		mcip->mci_rx_p_fn = NULL;
3081 		mcip->mci_rx_p_arg = NULL;
3082 
3083 		mcip->mci_state_flags &= ~MCIS_UNICAST_HW;
3084 
3085 		if (mcip->mci_state_flags & MCIS_TAG_DISABLE)
3086 			mcip->mci_state_flags &= ~MCIS_TAG_DISABLE;
3087 
3088 		if (mcip->mci_state_flags & MCIS_STRIP_DISABLE)
3089 			mcip->mci_state_flags &= ~MCIS_STRIP_DISABLE;
3090 
3091 		if (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK)
3092 			mcip->mci_state_flags &= ~MCIS_DISABLE_TX_VID_CHECK;
3093 
3094 		kmem_free(muip, sizeof (mac_unicast_impl_t));
3095 		i_mac_perim_exit(mip);
3096 		return (0);
3097 	}
3098 
3099 	/*
3100 	 * Remove the VID from the list of client's VIDs.
3101 	 */
3102 	pre = mcip->mci_unicast_list;
3103 	if (muip == pre) {
3104 		mcip->mci_unicast_list = muip->mui_next;
3105 	} else {
3106 		while ((pre->mui_next != NULL) && (pre->mui_next != muip))
3107 			pre = pre->mui_next;
3108 		ASSERT(pre->mui_next == muip);
3109 		rw_enter(&mcip->mci_rw_lock, RW_WRITER);
3110 		pre->mui_next = muip->mui_next;
3111 		rw_exit(&mcip->mci_rw_lock);
3112 	}
3113 
3114 	if (!mac_client_single_rcvr(mcip)) {
3115 		/*
3116 		 * This MAC client is shared by more than one unicast
3117 		 * addresses, so we will just remove the flent
3118 		 * corresponding to the address being removed. We don't invoke
3119 		 * mac_rx_classify_flow_rem() since the additional flow is
3120 		 * not associated with its own separate set of SRS and rings,
3121 		 * and these constructs are still needed for the remaining
3122 		 * flows.
3123 		 */
3124 		flent = mac_client_get_flow(mcip, muip);
3125 		VERIFY3P(flent, !=, NULL);
3126 
3127 		/*
3128 		 * The first one is disappearing, need to make sure
3129 		 * we replace it with another from the list of
3130 		 * shared clients.
3131 		 */
3132 		if (flent == mcip->mci_flent)
3133 			flent = mac_client_swap_mciflent(mcip);
3134 		mac_client_remove_flow_from_list(mcip, flent);
3135 		mac_flow_remove(mip->mi_flow_tab, flent, B_FALSE);
3136 		mac_flow_wait(flent, FLOW_DRIVER_UPCALL);
3137 
3138 		/*
3139 		 * The multicast groups that were added by the client so
3140 		 * far must be removed from the brodcast domain corresponding
3141 		 * to the VID being removed.
3142 		 */
3143 		mac_client_bcast_refresh(mcip, mac_client_update_mcast,
3144 		    (void *)flent, B_FALSE);
3145 
3146 		if (mip->mi_type->mt_brdcst_addr != NULL) {
3147 			mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr,
3148 			    muip->mui_vid);
3149 		}
3150 
3151 		FLOW_FINAL_REFRELE(flent);
3152 		ASSERT(!(mcip->mci_state_flags & MCIS_EXCLUSIVE));
3153 
3154 		/*
3155 		 * Enable fastpath if this is a VNIC or a VLAN.
3156 		 */
3157 		if (mcip->mci_state_flags & MCIS_IS_VNIC)
3158 			mac_fastpath_enable((mac_handle_t)mip);
3159 		mac_stop((mac_handle_t)mip);
3160 		i_mac_perim_exit(mip);
3161 		return (0);
3162 	}
3163 
3164 	mui_vid = muip->mui_vid;
3165 	mac_client_datapath_teardown(mch, muip, flent);
3166 
3167 	if ((mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY) &&
3168 	    mui_vid == VLAN_ID_NONE) {
3169 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PRIMARY;
3170 	} else {
3171 		i_mac_perim_exit(mip);
3172 		return (0);
3173 	}
3174 
3175 	/*
3176 	 * If we are removing the primary, check if we have a passive primary
3177 	 * client that we need to activate now.
3178 	 */
3179 	mcip = mac_get_passive_primary_client(mip);
3180 	if (mcip != NULL) {
3181 		mac_resource_props_t	*mrp;
3182 		mac_unicast_impl_t	*muip;
3183 
3184 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
3185 		mrp = kmem_zalloc(sizeof (*mrp), KM_SLEEP);
3186 
3187 		/*
3188 		 * Apply the property cached in the mac_impl_t to the
3189 		 * primary mac client.
3190 		 */
3191 		mac_get_resources((mac_handle_t)mip, mrp);
3192 		(void) mac_client_set_resources(mch, mrp);
3193 		ASSERT(mcip->mci_p_unicast_list != NULL);
3194 		muip = mcip->mci_p_unicast_list;
3195 		mcip->mci_p_unicast_list = NULL;
3196 		if (mac_client_datapath_setup(mcip, VLAN_ID_NONE,
3197 		    mip->mi_addr, mrp, B_TRUE, muip) == 0) {
3198 			if (mcip->mci_rx_p_fn != NULL) {
3199 				mac_rx_set(mch, mcip->mci_rx_p_fn,
3200 				    mcip->mci_rx_p_arg);
3201 				mcip->mci_rx_p_fn = NULL;
3202 				mcip->mci_rx_p_arg = NULL;
3203 			}
3204 		} else {
3205 			kmem_free(muip, sizeof (mac_unicast_impl_t));
3206 		}
3207 		kmem_free(mrp, sizeof (*mrp));
3208 	}
3209 	i_mac_perim_exit(mip);
3210 	return (0);
3211 }
3212 
3213 /*
3214  * Multicast add function invoked by MAC clients.
3215  */
3216 int
3217 mac_multicast_add(mac_client_handle_t mch, const uint8_t *addr)
3218 {
3219 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3220 	mac_impl_t		*mip = mcip->mci_mip;
3221 	flow_entry_t		*flent = mcip->mci_flent_list;
3222 	flow_entry_t		*prev_fe = NULL;
3223 	uint16_t		vid;
3224 	int			err = 0;
3225 
3226 	/* Verify the address is a valid multicast address */
3227 	if ((err = mip->mi_type->mt_ops.mtops_multicst_verify(addr,
3228 	    mip->mi_pdata)) != 0)
3229 		return (err);
3230 
3231 	i_mac_perim_enter(mip);
3232 	while (flent != NULL) {
3233 		vid = i_mac_flow_vid(flent);
3234 
3235 		err = mac_bcast_add((mac_client_impl_t *)mch, addr, vid,
3236 		    MAC_ADDRTYPE_MULTICAST);
3237 		if (err != 0)
3238 			break;
3239 		prev_fe = flent;
3240 		flent = flent->fe_client_next;
3241 	}
3242 
3243 	/*
3244 	 * If we failed adding, then undo all, rather than partial
3245 	 * success.
3246 	 */
3247 	if (flent != NULL && prev_fe != NULL) {
3248 		flent = mcip->mci_flent_list;
3249 		while (flent != prev_fe->fe_client_next) {
3250 			vid = i_mac_flow_vid(flent);
3251 			mac_bcast_delete((mac_client_impl_t *)mch, addr, vid);
3252 			flent = flent->fe_client_next;
3253 		}
3254 	}
3255 	i_mac_perim_exit(mip);
3256 	return (err);
3257 }
3258 
3259 /*
3260  * Multicast delete function invoked by MAC clients.
3261  */
3262 void
3263 mac_multicast_remove(mac_client_handle_t mch, const uint8_t *addr)
3264 {
3265 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3266 	mac_impl_t		*mip = mcip->mci_mip;
3267 	flow_entry_t		*flent;
3268 	uint16_t		vid;
3269 
3270 	i_mac_perim_enter(mip);
3271 	for (flent = mcip->mci_flent_list; flent != NULL;
3272 	    flent = flent->fe_client_next) {
3273 		vid = i_mac_flow_vid(flent);
3274 		mac_bcast_delete((mac_client_impl_t *)mch, addr, vid);
3275 	}
3276 	i_mac_perim_exit(mip);
3277 }
3278 
3279 /*
3280  * When a MAC client desires to capture packets on an interface,
3281  * it registers a promiscuous call back with mac_promisc_add().
3282  * There are three types of promiscuous callbacks:
3283  *
3284  * * MAC_CLIENT_PROMISC_ALL
3285  *   Captures all packets sent and received by the MAC client,
3286  *   the physical interface, as well as all other MAC clients
3287  *   defined on top of the same MAC.
3288  *
3289  * * MAC_CLIENT_PROMISC_FILTERED
3290  *   Captures all packets sent and received by the MAC client,
3291  *   plus all multicast traffic sent and received by the phyisical
3292  *   interface and the other MAC clients.
3293  *
3294  * * MAC_CLIENT_PROMISC_MULTI
3295  *   Captures all broadcast and multicast packets sent and
3296  *   received by the MAC clients as well as the physical interface.
3297  *
3298  * In all cases, the underlying MAC is put in promiscuous mode.
3299  */
3300 int
3301 mac_promisc_add(mac_client_handle_t mch, mac_client_promisc_type_t type,
3302     mac_rx_t fn, void *arg, mac_promisc_handle_t *mphp, uint16_t flags)
3303 {
3304 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3305 	mac_impl_t *mip = mcip->mci_mip;
3306 	mac_promisc_impl_t *mpip;
3307 	mac_cb_info_t	*mcbi;
3308 	int rc;
3309 
3310 	i_mac_perim_enter(mip);
3311 
3312 	if ((rc = mac_start((mac_handle_t)mip)) != 0) {
3313 		i_mac_perim_exit(mip);
3314 		return (rc);
3315 	}
3316 
3317 	if ((mcip->mci_state_flags & MCIS_IS_VNIC) &&
3318 	    type == MAC_CLIENT_PROMISC_ALL &&
3319 	    (mcip->mci_protect_flags & MPT_FLAG_PROMISC_FILTERED)) {
3320 		/*
3321 		 * The function is being invoked by the upper MAC client
3322 		 * of a VNIC. The VNIC should only see the traffic
3323 		 * it is entitled to.
3324 		 */
3325 		type = MAC_CLIENT_PROMISC_FILTERED;
3326 	}
3327 
3328 
3329 	/*
3330 	 * Turn on promiscuous mode for the underlying NIC.
3331 	 * This is needed even for filtered callbacks which
3332 	 * expect to receive all multicast traffic on the wire.
3333 	 *
3334 	 * Physical promiscuous mode should not be turned on if
3335 	 * MAC_PROMISC_FLAGS_NO_PHYS is set.
3336 	 */
3337 	if ((flags & MAC_PROMISC_FLAGS_NO_PHYS) == 0) {
3338 		if ((rc = i_mac_promisc_set(mip, B_TRUE)) != 0) {
3339 			mac_stop((mac_handle_t)mip);
3340 			i_mac_perim_exit(mip);
3341 			return (rc);
3342 		}
3343 	}
3344 
3345 	mpip = kmem_cache_alloc(mac_promisc_impl_cache, KM_SLEEP);
3346 
3347 	mpip->mpi_type = type;
3348 	mpip->mpi_fn = fn;
3349 	mpip->mpi_arg = arg;
3350 	mpip->mpi_mcip = mcip;
3351 	mpip->mpi_no_tx_loop = ((flags & MAC_PROMISC_FLAGS_NO_TX_LOOP) != 0);
3352 	mpip->mpi_no_phys = ((flags & MAC_PROMISC_FLAGS_NO_PHYS) != 0);
3353 	mpip->mpi_strip_vlan_tag =
3354 	    ((flags & MAC_PROMISC_FLAGS_VLAN_TAG_STRIP) != 0);
3355 	mpip->mpi_no_copy = ((flags & MAC_PROMISC_FLAGS_NO_COPY) != 0);
3356 
3357 	mcbi = &mip->mi_promisc_cb_info;
3358 	mutex_enter(mcbi->mcbi_lockp);
3359 
3360 	mac_callback_add(&mip->mi_promisc_cb_info, &mcip->mci_promisc_list,
3361 	    &mpip->mpi_mci_link);
3362 	mac_callback_add(&mip->mi_promisc_cb_info, &mip->mi_promisc_list,
3363 	    &mpip->mpi_mi_link);
3364 
3365 	mutex_exit(mcbi->mcbi_lockp);
3366 
3367 	*mphp = (mac_promisc_handle_t)mpip;
3368 
3369 	if (mcip->mci_state_flags & MCIS_IS_VNIC) {
3370 		mac_impl_t *umip = mcip->mci_upper_mip;
3371 
3372 		ASSERT(umip != NULL);
3373 		mac_vnic_secondary_update(umip);
3374 	}
3375 
3376 	i_mac_perim_exit(mip);
3377 
3378 	return (0);
3379 }
3380 
3381 /*
3382  * Remove a multicast address previously aded through mac_promisc_add().
3383  */
3384 void
3385 mac_promisc_remove(mac_promisc_handle_t mph)
3386 {
3387 	mac_promisc_impl_t *mpip = (mac_promisc_impl_t *)mph;
3388 	mac_client_impl_t *mcip = mpip->mpi_mcip;
3389 	mac_impl_t *mip = mcip->mci_mip;
3390 	mac_cb_info_t *mcbi;
3391 	int rv;
3392 
3393 	i_mac_perim_enter(mip);
3394 
3395 	/*
3396 	 * Even if the device can't be reset into normal mode, we still
3397 	 * need to clear the client promisc callbacks. The client may want
3398 	 * to close the mac end point and we can't have stale callbacks.
3399 	 */
3400 	if (!(mpip->mpi_no_phys)) {
3401 		if ((rv = i_mac_promisc_set(mip, B_FALSE)) != 0) {
3402 			cmn_err(CE_WARN, "%s: failed to switch OFF promiscuous"
3403 			    " mode because of error 0x%x", mip->mi_name, rv);
3404 		}
3405 	}
3406 	mcbi = &mip->mi_promisc_cb_info;
3407 	mutex_enter(mcbi->mcbi_lockp);
3408 	if (mac_callback_remove(mcbi, &mip->mi_promisc_list,
3409 	    &mpip->mpi_mi_link)) {
3410 		VERIFY(mac_callback_remove(&mip->mi_promisc_cb_info,
3411 		    &mcip->mci_promisc_list, &mpip->mpi_mci_link));
3412 		kmem_cache_free(mac_promisc_impl_cache, mpip);
3413 	} else {
3414 		mac_callback_remove_wait(&mip->mi_promisc_cb_info);
3415 	}
3416 
3417 	if (mcip->mci_state_flags & MCIS_IS_VNIC) {
3418 		mac_impl_t *umip = mcip->mci_upper_mip;
3419 
3420 		ASSERT(umip != NULL);
3421 		mac_vnic_secondary_update(umip);
3422 	}
3423 
3424 	mutex_exit(mcbi->mcbi_lockp);
3425 	mac_stop((mac_handle_t)mip);
3426 
3427 	i_mac_perim_exit(mip);
3428 }
3429 
3430 /*
3431  * Reference count the number of active Tx threads. MCI_TX_QUIESCE indicates
3432  * that a control operation wants to quiesce the Tx data flow in which case
3433  * we return an error. Holding any of the per cpu locks ensures that the
3434  * mci_tx_flag won't change.
3435  *
3436  * 'CPU' must be accessed just once and used to compute the index into the
3437  * percpu array, and that index must be used for the entire duration of the
3438  * packet send operation. Note that the thread may be preempted and run on
3439  * another cpu any time and so we can't use 'CPU' more than once for the
3440  * operation.
3441  */
3442 #define	MAC_TX_TRY_HOLD(mcip, mytx, error)				\
3443 {									\
3444 	(error) = 0;							\
3445 	(mytx) = &(mcip)->mci_tx_pcpu[CPU->cpu_seqid & mac_tx_percpu_cnt]; \
3446 	mutex_enter(&(mytx)->pcpu_tx_lock);				\
3447 	if (!((mcip)->mci_tx_flag & MCI_TX_QUIESCE)) {			\
3448 		(mytx)->pcpu_tx_refcnt++;				\
3449 	} else {							\
3450 		(error) = -1;						\
3451 	}								\
3452 	mutex_exit(&(mytx)->pcpu_tx_lock);				\
3453 }
3454 
3455 /*
3456  * Release the reference. If needed, signal any control operation waiting
3457  * for Tx quiescence. The wait and signal are always done using the
3458  * mci_tx_pcpu[0]'s lock
3459  */
3460 #define	MAC_TX_RELE(mcip, mytx) {					\
3461 	mutex_enter(&(mytx)->pcpu_tx_lock);				\
3462 	if (--(mytx)->pcpu_tx_refcnt == 0 &&				\
3463 	    (mcip)->mci_tx_flag & MCI_TX_QUIESCE) {			\
3464 		mutex_exit(&(mytx)->pcpu_tx_lock);			\
3465 		mutex_enter(&(mcip)->mci_tx_pcpu[0].pcpu_tx_lock);	\
3466 		cv_signal(&(mcip)->mci_tx_cv);				\
3467 		mutex_exit(&(mcip)->mci_tx_pcpu[0].pcpu_tx_lock);	\
3468 	} else {							\
3469 		mutex_exit(&(mytx)->pcpu_tx_lock);			\
3470 	}								\
3471 }
3472 
3473 /*
3474  * Send function invoked by MAC clients.
3475  */
3476 mac_tx_cookie_t
3477 mac_tx(mac_client_handle_t mch, mblk_t *mp_chain, uintptr_t hint,
3478     uint16_t flag, mblk_t **ret_mp)
3479 {
3480 	mac_tx_cookie_t		cookie = 0;
3481 	int			error;
3482 	mac_tx_percpu_t		*mytx;
3483 	mac_soft_ring_set_t	*srs;
3484 	flow_entry_t		*flent;
3485 	boolean_t		is_subflow = B_FALSE;
3486 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3487 	mac_impl_t		*mip = mcip->mci_mip;
3488 	mac_srs_tx_t		*srs_tx;
3489 
3490 	/*
3491 	 * Check whether the active Tx threads count is bumped already.
3492 	 */
3493 	if (!(flag & MAC_TX_NO_HOLD)) {
3494 		MAC_TX_TRY_HOLD(mcip, mytx, error);
3495 		if (error != 0) {
3496 			freemsgchain(mp_chain);
3497 			return (0);
3498 		}
3499 	}
3500 
3501 	/*
3502 	 * If mac protection is enabled, only the permissible packets will be
3503 	 * returned by mac_protect_check().
3504 	 */
3505 	if ((mcip->mci_flent->
3506 	    fe_resource_props.mrp_mask & MRP_PROTECT) != 0 &&
3507 	    (mp_chain = mac_protect_check(mch, mp_chain)) == NULL)
3508 		goto done;
3509 
3510 	if (mcip->mci_subflow_tab != NULL &&
3511 	    mcip->mci_subflow_tab->ft_flow_count > 0 &&
3512 	    mac_flow_lookup(mcip->mci_subflow_tab, mp_chain,
3513 	    FLOW_OUTBOUND, &flent) == 0) {
3514 		/*
3515 		 * The main assumption here is that if in the event
3516 		 * we get a chain, all the packets will be classified
3517 		 * to the same Flow/SRS. If this changes for any
3518 		 * reason, the following logic should change as well.
3519 		 * I suppose the fanout_hint also assumes this .
3520 		 */
3521 		ASSERT(flent != NULL);
3522 		is_subflow = B_TRUE;
3523 	} else {
3524 		flent = mcip->mci_flent;
3525 	}
3526 
3527 	srs = flent->fe_tx_srs;
3528 	/*
3529 	 * This is to avoid panics with PF_PACKET that can call mac_tx()
3530 	 * against an interface that is not capable of sending. A rewrite
3531 	 * of the mac datapath is required to remove this limitation.
3532 	 */
3533 	if (srs == NULL) {
3534 		freemsgchain(mp_chain);
3535 		goto done;
3536 	}
3537 
3538 	srs_tx = &srs->srs_tx;
3539 	if (srs_tx->st_mode == SRS_TX_DEFAULT &&
3540 	    (srs->srs_state & SRS_ENQUEUED) == 0 &&
3541 	    mip->mi_nactiveclients == 1 && mp_chain->b_next == NULL) {
3542 		uint64_t	obytes;
3543 
3544 		/*
3545 		 * Since dls always opens the underlying MAC, nclients equals
3546 		 * to 1 means that the only active client is dls itself acting
3547 		 * as a primary client of the MAC instance. Since dls will not
3548 		 * send tagged packets in that case, and dls is trusted to send
3549 		 * packets for its allowed VLAN(s), the VLAN tag insertion and
3550 		 * check is required only if nclients is greater than 1.
3551 		 */
3552 		if (mip->mi_nclients > 1) {
3553 			if (MAC_VID_CHECK_NEEDED(mcip)) {
3554 				int	err = 0;
3555 
3556 				MAC_VID_CHECK(mcip, mp_chain, err);
3557 				if (err != 0) {
3558 					freemsg(mp_chain);
3559 					mcip->mci_misc_stat.mms_txerrors++;
3560 					goto done;
3561 				}
3562 			}
3563 			if (MAC_TAG_NEEDED(mcip)) {
3564 				mp_chain = mac_add_vlan_tag(mp_chain, 0,
3565 				    mac_client_vid(mch));
3566 				if (mp_chain == NULL) {
3567 					mcip->mci_misc_stat.mms_txerrors++;
3568 					goto done;
3569 				}
3570 			}
3571 		}
3572 
3573 		obytes = (mp_chain->b_cont == NULL ? MBLKL(mp_chain) :
3574 		    msgdsize(mp_chain));
3575 
3576 		MAC_TX(mip, srs_tx->st_arg2, mp_chain, mcip);
3577 		if (mp_chain == NULL) {
3578 			cookie = 0;
3579 			SRS_TX_STAT_UPDATE(srs, opackets, 1);
3580 			SRS_TX_STAT_UPDATE(srs, obytes, obytes);
3581 		} else {
3582 			mutex_enter(&srs->srs_lock);
3583 			cookie = mac_tx_srs_no_desc(srs, mp_chain,
3584 			    flag, ret_mp);
3585 			mutex_exit(&srs->srs_lock);
3586 		}
3587 	} else {
3588 		cookie = srs_tx->st_func(srs, mp_chain, hint, flag, ret_mp);
3589 	}
3590 
3591 done:
3592 	if (is_subflow)
3593 		FLOW_REFRELE(flent);
3594 
3595 	if (!(flag & MAC_TX_NO_HOLD))
3596 		MAC_TX_RELE(mcip, mytx);
3597 
3598 	return (cookie);
3599 }
3600 
3601 /*
3602  * mac_tx_is_blocked
3603  *
3604  * Given a cookie, it returns if the ring identified by the cookie is
3605  * flow-controlled or not. If NULL is passed in place of a cookie,
3606  * then it finds out if any of the underlying rings belonging to the
3607  * SRS is flow controlled or not and returns that status.
3608  */
3609 /* ARGSUSED */
3610 boolean_t
3611 mac_tx_is_flow_blocked(mac_client_handle_t mch, mac_tx_cookie_t cookie)
3612 {
3613 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3614 	mac_soft_ring_set_t *mac_srs;
3615 	mac_soft_ring_t *sringp;
3616 	boolean_t blocked = B_FALSE;
3617 	mac_tx_percpu_t *mytx;
3618 	int err;
3619 	int i;
3620 
3621 	/*
3622 	 * Bump the reference count so that mac_srs won't be deleted.
3623 	 * If the client is currently quiesced and we failed to bump
3624 	 * the reference, return B_TRUE so that flow control stays
3625 	 * as enabled.
3626 	 *
3627 	 * Flow control will then be disabled once the client is no
3628 	 * longer quiesced.
3629 	 */
3630 	MAC_TX_TRY_HOLD(mcip, mytx, err);
3631 	if (err != 0)
3632 		return (B_TRUE);
3633 
3634 	if ((mac_srs = MCIP_TX_SRS(mcip)) == NULL) {
3635 		MAC_TX_RELE(mcip, mytx);
3636 		return (B_FALSE);
3637 	}
3638 
3639 	mutex_enter(&mac_srs->srs_lock);
3640 	/*
3641 	 * Only in the case of TX_FANOUT and TX_AGGR, the underlying
3642 	 * softring (s_ring_state) will have the HIWAT set. This is
3643 	 * the multiple Tx ring flow control case. For all other
3644 	 * case, SRS (srs_state) will store the condition.
3645 	 */
3646 	if (mac_srs->srs_tx.st_mode == SRS_TX_FANOUT ||
3647 	    mac_srs->srs_tx.st_mode == SRS_TX_AGGR) {
3648 		if (cookie != 0) {
3649 			sringp = (mac_soft_ring_t *)cookie;
3650 			mutex_enter(&sringp->s_ring_lock);
3651 			if (sringp->s_ring_state & S_RING_TX_HIWAT)
3652 				blocked = B_TRUE;
3653 			mutex_exit(&sringp->s_ring_lock);
3654 		} else {
3655 			for (i = 0; i < mac_srs->srs_tx_ring_count; i++) {
3656 				sringp = mac_srs->srs_tx_soft_rings[i];
3657 				mutex_enter(&sringp->s_ring_lock);
3658 				if (sringp->s_ring_state & S_RING_TX_HIWAT) {
3659 					blocked = B_TRUE;
3660 					mutex_exit(&sringp->s_ring_lock);
3661 					break;
3662 				}
3663 				mutex_exit(&sringp->s_ring_lock);
3664 			}
3665 		}
3666 	} else {
3667 		blocked = (mac_srs->srs_state & SRS_TX_HIWAT);
3668 	}
3669 	mutex_exit(&mac_srs->srs_lock);
3670 	MAC_TX_RELE(mcip, mytx);
3671 	return (blocked);
3672 }
3673 
3674 /*
3675  * Check if the MAC client is the primary MAC client.
3676  */
3677 boolean_t
3678 mac_is_primary_client(mac_client_impl_t *mcip)
3679 {
3680 	return (mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY);
3681 }
3682 
3683 void
3684 mac_ioctl(mac_handle_t mh, queue_t *wq, mblk_t *bp)
3685 {
3686 	mac_impl_t	*mip = (mac_impl_t *)mh;
3687 	int cmd = ((struct iocblk *)bp->b_rptr)->ioc_cmd;
3688 
3689 	if ((cmd == ND_GET && (mip->mi_callbacks->mc_callbacks & MC_GETPROP)) ||
3690 	    (cmd == ND_SET && (mip->mi_callbacks->mc_callbacks & MC_SETPROP))) {
3691 		/*
3692 		 * If ndd props were registered, call them.
3693 		 * Note that ndd ioctls are Obsolete
3694 		 */
3695 		mac_ndd_ioctl(mip, wq, bp);
3696 		return;
3697 	}
3698 
3699 	/*
3700 	 * Call the driver to handle the ioctl.  The driver may not support
3701 	 * any ioctls, in which case we reply with a NAK on its behalf.
3702 	 */
3703 	if (mip->mi_callbacks->mc_callbacks & MC_IOCTL)
3704 		mip->mi_ioctl(mip->mi_driver, wq, bp);
3705 	else
3706 		miocnak(wq, bp, 0, EINVAL);
3707 }
3708 
3709 /*
3710  * Return the link state of the specified MAC instance.
3711  */
3712 link_state_t
3713 mac_link_get(mac_handle_t mh)
3714 {
3715 	return (((mac_impl_t *)mh)->mi_linkstate);
3716 }
3717 
3718 /*
3719  * Add a mac client specified notification callback. Please see the comments
3720  * above mac_callback_add() for general information about mac callback
3721  * addition/deletion in the presence of mac callback list walkers
3722  */
3723 mac_notify_handle_t
3724 mac_notify_add(mac_handle_t mh, mac_notify_t notify_fn, void *arg)
3725 {
3726 	mac_impl_t		*mip = (mac_impl_t *)mh;
3727 	mac_notify_cb_t		*mncb;
3728 	mac_cb_info_t		*mcbi;
3729 
3730 	/*
3731 	 * Allocate a notify callback structure, fill in the details and
3732 	 * use the mac callback list manipulation functions to chain into
3733 	 * the list of callbacks.
3734 	 */
3735 	mncb = kmem_zalloc(sizeof (mac_notify_cb_t), KM_SLEEP);
3736 	mncb->mncb_fn = notify_fn;
3737 	mncb->mncb_arg = arg;
3738 	mncb->mncb_mip = mip;
3739 	mncb->mncb_link.mcb_objp = mncb;
3740 	mncb->mncb_link.mcb_objsize = sizeof (mac_notify_cb_t);
3741 	mncb->mncb_link.mcb_flags = MCB_NOTIFY_CB_T;
3742 
3743 	mcbi = &mip->mi_notify_cb_info;
3744 
3745 	i_mac_perim_enter(mip);
3746 	mutex_enter(mcbi->mcbi_lockp);
3747 
3748 	mac_callback_add(&mip->mi_notify_cb_info, &mip->mi_notify_cb_list,
3749 	    &mncb->mncb_link);
3750 
3751 	mutex_exit(mcbi->mcbi_lockp);
3752 	i_mac_perim_exit(mip);
3753 	return ((mac_notify_handle_t)mncb);
3754 }
3755 
3756 void
3757 mac_notify_remove_wait(mac_handle_t mh)
3758 {
3759 	mac_impl_t	*mip = (mac_impl_t *)mh;
3760 	mac_cb_info_t	*mcbi = &mip->mi_notify_cb_info;
3761 
3762 	mutex_enter(mcbi->mcbi_lockp);
3763 	mac_callback_remove_wait(&mip->mi_notify_cb_info);
3764 	mutex_exit(mcbi->mcbi_lockp);
3765 }
3766 
3767 /*
3768  * Remove a mac client specified notification callback
3769  */
3770 int
3771 mac_notify_remove(mac_notify_handle_t mnh, boolean_t wait)
3772 {
3773 	mac_notify_cb_t	*mncb = (mac_notify_cb_t *)mnh;
3774 	mac_impl_t	*mip = mncb->mncb_mip;
3775 	mac_cb_info_t	*mcbi;
3776 	int		err = 0;
3777 
3778 	mcbi = &mip->mi_notify_cb_info;
3779 
3780 	i_mac_perim_enter(mip);
3781 	mutex_enter(mcbi->mcbi_lockp);
3782 
3783 	ASSERT(mncb->mncb_link.mcb_objp == mncb);
3784 	/*
3785 	 * If there aren't any list walkers, the remove would succeed
3786 	 * inline, else we wait for the deferred remove to complete
3787 	 */
3788 	if (mac_callback_remove(&mip->mi_notify_cb_info,
3789 	    &mip->mi_notify_cb_list, &mncb->mncb_link)) {
3790 		kmem_free(mncb, sizeof (mac_notify_cb_t));
3791 	} else {
3792 		err = EBUSY;
3793 	}
3794 
3795 	mutex_exit(mcbi->mcbi_lockp);
3796 	i_mac_perim_exit(mip);
3797 
3798 	/*
3799 	 * If we failed to remove the notification callback and "wait" is set
3800 	 * to be B_TRUE, wait for the callback to finish after we exit the
3801 	 * mac perimeter.
3802 	 */
3803 	if (err != 0 && wait) {
3804 		mac_notify_remove_wait((mac_handle_t)mip);
3805 		return (0);
3806 	}
3807 
3808 	return (err);
3809 }
3810 
3811 /*
3812  * Associate resource management callbacks with the specified MAC
3813  * clients.
3814  */
3815 
3816 void
3817 mac_resource_set_common(mac_client_handle_t mch, mac_resource_add_t add,
3818     mac_resource_remove_t remove, mac_resource_quiesce_t quiesce,
3819     mac_resource_restart_t restart, mac_resource_bind_t bind,
3820     void *arg)
3821 {
3822 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3823 
3824 	mcip->mci_resource_add = add;
3825 	mcip->mci_resource_remove = remove;
3826 	mcip->mci_resource_quiesce = quiesce;
3827 	mcip->mci_resource_restart = restart;
3828 	mcip->mci_resource_bind = bind;
3829 	mcip->mci_resource_arg = arg;
3830 }
3831 
3832 void
3833 mac_resource_set(mac_client_handle_t mch, mac_resource_add_t add, void *arg)
3834 {
3835 	/* update the 'resource_add' callback */
3836 	mac_resource_set_common(mch, add, NULL, NULL, NULL, NULL, arg);
3837 }
3838 
3839 /*
3840  * Sets up the client resources and enable the polling interface over all the
3841  * SRS's and the soft rings of the client
3842  */
3843 void
3844 mac_client_poll_enable(mac_client_handle_t mch)
3845 {
3846 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3847 	mac_soft_ring_set_t	*mac_srs;
3848 	flow_entry_t		*flent;
3849 	int			i;
3850 
3851 	flent = mcip->mci_flent;
3852 	ASSERT(flent != NULL);
3853 
3854 	mcip->mci_state_flags |= MCIS_CLIENT_POLL_CAPABLE;
3855 	for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
3856 		mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
3857 		ASSERT(mac_srs->srs_mcip == mcip);
3858 		mac_srs_client_poll_enable(mcip, mac_srs);
3859 	}
3860 }
3861 
3862 /*
3863  * Tears down the client resources and disable the polling interface over all
3864  * the SRS's and the soft rings of the client
3865  */
3866 void
3867 mac_client_poll_disable(mac_client_handle_t mch)
3868 {
3869 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3870 	mac_soft_ring_set_t	*mac_srs;
3871 	flow_entry_t		*flent;
3872 	int			i;
3873 
3874 	flent = mcip->mci_flent;
3875 	ASSERT(flent != NULL);
3876 
3877 	mcip->mci_state_flags &= ~MCIS_CLIENT_POLL_CAPABLE;
3878 	for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
3879 		mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
3880 		ASSERT(mac_srs->srs_mcip == mcip);
3881 		mac_srs_client_poll_disable(mcip, mac_srs);
3882 	}
3883 }
3884 
3885 /*
3886  * Associate the CPUs specified by the given property with a MAC client.
3887  */
3888 int
3889 mac_cpu_set(mac_client_handle_t mch, mac_resource_props_t *mrp)
3890 {
3891 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3892 	mac_impl_t *mip = mcip->mci_mip;
3893 	int err = 0;
3894 
3895 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
3896 
3897 	if ((err = mac_validate_props(mcip->mci_state_flags & MCIS_IS_VNIC ?
3898 	    mcip->mci_upper_mip : mip, mrp)) != 0) {
3899 		return (err);
3900 	}
3901 	if (MCIP_DATAPATH_SETUP(mcip))
3902 		mac_flow_modify(mip->mi_flow_tab, mcip->mci_flent, mrp);
3903 
3904 	mac_update_resources(mrp, MCIP_RESOURCE_PROPS(mcip), B_FALSE);
3905 	return (0);
3906 }
3907 
3908 /*
3909  * Apply the specified properties to the specified MAC client.
3910  */
3911 int
3912 mac_client_set_resources(mac_client_handle_t mch, mac_resource_props_t *mrp)
3913 {
3914 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3915 	mac_impl_t *mip = mcip->mci_mip;
3916 	int err = 0;
3917 
3918 	i_mac_perim_enter(mip);
3919 
3920 	if ((mrp->mrp_mask & MRP_MAXBW) || (mrp->mrp_mask & MRP_PRIORITY)) {
3921 		err = mac_resource_ctl_set(mch, mrp);
3922 		if (err != 0)
3923 			goto done;
3924 	}
3925 
3926 	if (mrp->mrp_mask & (MRP_CPUS|MRP_POOL)) {
3927 		err = mac_cpu_set(mch, mrp);
3928 		if (err != 0)
3929 			goto done;
3930 	}
3931 
3932 	if (mrp->mrp_mask & MRP_PROTECT) {
3933 		err = mac_protect_set(mch, mrp);
3934 		if (err != 0)
3935 			goto done;
3936 	}
3937 
3938 	if ((mrp->mrp_mask & MRP_RX_RINGS) || (mrp->mrp_mask & MRP_TX_RINGS))
3939 		err = mac_resource_ctl_set(mch, mrp);
3940 
3941 done:
3942 	i_mac_perim_exit(mip);
3943 	return (err);
3944 }
3945 
3946 /*
3947  * Return the properties currently associated with the specified MAC client.
3948  */
3949 void
3950 mac_client_get_resources(mac_client_handle_t mch, mac_resource_props_t *mrp)
3951 {
3952 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3953 	mac_resource_props_t	*mcip_mrp = MCIP_RESOURCE_PROPS(mcip);
3954 
3955 	bcopy(mcip_mrp, mrp, sizeof (mac_resource_props_t));
3956 }
3957 
3958 /*
3959  * Return the effective properties currently associated with the specified
3960  * MAC client.
3961  */
3962 void
3963 mac_client_get_effective_resources(mac_client_handle_t mch,
3964     mac_resource_props_t *mrp)
3965 {
3966 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3967 	mac_resource_props_t	*mcip_mrp = MCIP_EFFECTIVE_PROPS(mcip);
3968 
3969 	bcopy(mcip_mrp, mrp, sizeof (mac_resource_props_t));
3970 }
3971 
3972 /*
3973  * Pass a copy of the specified packet to the promiscuous callbacks
3974  * of the specified MAC.
3975  *
3976  * If sender is NULL, the function is being invoked for a packet chain
3977  * received from the wire. If sender is non-NULL, it points to
3978  * the MAC client from which the packet is being sent.
3979  *
3980  * The packets are distributed to the promiscuous callbacks as follows:
3981  *
3982  * - all packets are sent to the MAC_CLIENT_PROMISC_ALL callbacks
3983  * - all broadcast and multicast packets are sent to the
3984  *   MAC_CLIENT_PROMISC_FILTER and MAC_CLIENT_PROMISC_MULTI.
3985  *
3986  * The unicast packets of MAC_CLIENT_PROMISC_FILTER callbacks are dispatched
3987  * after classification by mac_rx_deliver().
3988  */
3989 
3990 static void
3991 mac_promisc_dispatch_one(mac_promisc_impl_t *mpip, mblk_t *mp,
3992     boolean_t loopback)
3993 {
3994 	mblk_t *mp_copy, *mp_next;
3995 
3996 	if (!mpip->mpi_no_copy || mpip->mpi_strip_vlan_tag) {
3997 		mp_copy = copymsg(mp);
3998 		if (mp_copy == NULL)
3999 			return;
4000 
4001 		if (mpip->mpi_strip_vlan_tag) {
4002 			mp_copy = mac_strip_vlan_tag_chain(mp_copy);
4003 			if (mp_copy == NULL)
4004 				return;
4005 		}
4006 		mp_next = NULL;
4007 	} else {
4008 		mp_copy = mp;
4009 		mp_next = mp->b_next;
4010 	}
4011 	mp_copy->b_next = NULL;
4012 
4013 	mpip->mpi_fn(mpip->mpi_arg, NULL, mp_copy, loopback);
4014 	if (mp_copy == mp)
4015 		mp->b_next = mp_next;
4016 }
4017 
4018 /*
4019  * Return the VID of a packet. Zero if the packet is not tagged.
4020  */
4021 static uint16_t
4022 mac_ether_vid(mblk_t *mp)
4023 {
4024 	struct ether_header *eth = (struct ether_header *)mp->b_rptr;
4025 
4026 	if (ntohs(eth->ether_type) == ETHERTYPE_VLAN) {
4027 		struct ether_vlan_header *t_evhp =
4028 		    (struct ether_vlan_header *)mp->b_rptr;
4029 		return (VLAN_ID(ntohs(t_evhp->ether_tci)));
4030 	}
4031 
4032 	return (0);
4033 }
4034 
4035 /*
4036  * Return whether the specified packet contains a multicast or broadcast
4037  * destination MAC address.
4038  */
4039 static boolean_t
4040 mac_is_mcast(mac_impl_t *mip, mblk_t *mp)
4041 {
4042 	mac_header_info_t hdr_info;
4043 
4044 	if (mac_header_info((mac_handle_t)mip, mp, &hdr_info) != 0)
4045 		return (B_FALSE);
4046 	return ((hdr_info.mhi_dsttype == MAC_ADDRTYPE_BROADCAST) ||
4047 	    (hdr_info.mhi_dsttype == MAC_ADDRTYPE_MULTICAST));
4048 }
4049 
4050 /*
4051  * Send a copy of an mblk chain to the MAC clients of the specified MAC.
4052  * "sender" points to the sender MAC client for outbound packets, and
4053  * is set to NULL for inbound packets.
4054  */
4055 void
4056 mac_promisc_dispatch(mac_impl_t *mip, mblk_t *mp_chain,
4057     mac_client_impl_t *sender)
4058 {
4059 	mac_promisc_impl_t *mpip;
4060 	mac_cb_t *mcb;
4061 	mblk_t *mp;
4062 	boolean_t is_mcast, is_sender;
4063 
4064 	MAC_PROMISC_WALKER_INC(mip);
4065 	for (mp = mp_chain; mp != NULL; mp = mp->b_next) {
4066 		is_mcast = mac_is_mcast(mip, mp);
4067 		/* send packet to interested callbacks */
4068 		for (mcb = mip->mi_promisc_list; mcb != NULL;
4069 		    mcb = mcb->mcb_nextp) {
4070 			mpip = (mac_promisc_impl_t *)mcb->mcb_objp;
4071 			is_sender = (mpip->mpi_mcip == sender);
4072 
4073 			if (is_sender && mpip->mpi_no_tx_loop)
4074 				/*
4075 				 * The sender doesn't want to receive
4076 				 * copies of the packets it sends.
4077 				 */
4078 				continue;
4079 
4080 			/* this client doesn't need any packets (bridge) */
4081 			if (mpip->mpi_fn == NULL)
4082 				continue;
4083 
4084 			/*
4085 			 * For an ethernet MAC, don't displatch a multicast
4086 			 * packet to a non-PROMISC_ALL callbacks unless the VID
4087 			 * of the packet matches the VID of the client.
4088 			 */
4089 			if (is_mcast &&
4090 			    mpip->mpi_type != MAC_CLIENT_PROMISC_ALL &&
4091 			    !mac_client_check_flow_vid(mpip->mpi_mcip,
4092 			    mac_ether_vid(mp)))
4093 				continue;
4094 
4095 			if (is_sender ||
4096 			    mpip->mpi_type == MAC_CLIENT_PROMISC_ALL ||
4097 			    is_mcast)
4098 				mac_promisc_dispatch_one(mpip, mp, is_sender);
4099 		}
4100 	}
4101 	MAC_PROMISC_WALKER_DCR(mip);
4102 }
4103 
4104 void
4105 mac_promisc_client_dispatch(mac_client_impl_t *mcip, mblk_t *mp_chain)
4106 {
4107 	mac_impl_t		*mip = mcip->mci_mip;
4108 	mac_promisc_impl_t	*mpip;
4109 	boolean_t		is_mcast;
4110 	mblk_t			*mp;
4111 	mac_cb_t		*mcb;
4112 
4113 	/*
4114 	 * The unicast packets for the MAC client still
4115 	 * need to be delivered to the MAC_CLIENT_PROMISC_FILTERED
4116 	 * promiscuous callbacks. The broadcast and multicast
4117 	 * packets were delivered from mac_rx().
4118 	 */
4119 	MAC_PROMISC_WALKER_INC(mip);
4120 	for (mp = mp_chain; mp != NULL; mp = mp->b_next) {
4121 		is_mcast = mac_is_mcast(mip, mp);
4122 		for (mcb = mcip->mci_promisc_list; mcb != NULL;
4123 		    mcb = mcb->mcb_nextp) {
4124 			mpip = (mac_promisc_impl_t *)mcb->mcb_objp;
4125 			if (mpip->mpi_type == MAC_CLIENT_PROMISC_FILTERED &&
4126 			    !is_mcast) {
4127 				mac_promisc_dispatch_one(mpip, mp, B_FALSE);
4128 			}
4129 		}
4130 	}
4131 	MAC_PROMISC_WALKER_DCR(mip);
4132 }
4133 
4134 /*
4135  * Return the margin value currently assigned to the specified MAC instance.
4136  */
4137 void
4138 mac_margin_get(mac_handle_t mh, uint32_t *marginp)
4139 {
4140 	mac_impl_t *mip = (mac_impl_t *)mh;
4141 
4142 	rw_enter(&(mip->mi_rw_lock), RW_READER);
4143 	*marginp = mip->mi_margin;
4144 	rw_exit(&(mip->mi_rw_lock));
4145 }
4146 
4147 /*
4148  * mac_info_get() is used for retrieving the mac_info when a DL_INFO_REQ is
4149  * issued before a DL_ATTACH_REQ. we walk the i_mac_impl_hash table and find
4150  * the first mac_impl_t with a matching driver name; then we copy its mac_info_t
4151  * to the caller. we do all this with i_mac_impl_lock held so the mac_impl_t
4152  * cannot disappear while we are accessing it.
4153  */
4154 typedef struct i_mac_info_state_s {
4155 	const char	*mi_name;
4156 	mac_info_t	*mi_infop;
4157 } i_mac_info_state_t;
4158 
4159 /*ARGSUSED*/
4160 static uint_t
4161 i_mac_info_walker(mod_hash_key_t key, mod_hash_val_t *val, void *arg)
4162 {
4163 	i_mac_info_state_t *statep = arg;
4164 	mac_impl_t *mip = (mac_impl_t *)val;
4165 
4166 	if (mip->mi_state_flags & MIS_DISABLED)
4167 		return (MH_WALK_CONTINUE);
4168 
4169 	if (strcmp(statep->mi_name,
4170 	    ddi_driver_name(mip->mi_dip)) != 0)
4171 		return (MH_WALK_CONTINUE);
4172 
4173 	statep->mi_infop = &mip->mi_info;
4174 	return (MH_WALK_TERMINATE);
4175 }
4176 
4177 boolean_t
4178 mac_info_get(const char *name, mac_info_t *minfop)
4179 {
4180 	i_mac_info_state_t state;
4181 
4182 	rw_enter(&i_mac_impl_lock, RW_READER);
4183 	state.mi_name = name;
4184 	state.mi_infop = NULL;
4185 	mod_hash_walk(i_mac_impl_hash, i_mac_info_walker, &state);
4186 	if (state.mi_infop == NULL) {
4187 		rw_exit(&i_mac_impl_lock);
4188 		return (B_FALSE);
4189 	}
4190 	*minfop = *state.mi_infop;
4191 	rw_exit(&i_mac_impl_lock);
4192 	return (B_TRUE);
4193 }
4194 
4195 /*
4196  * To get the capabilities that MAC layer cares about, such as rings, factory
4197  * mac address, vnic or not, it should directly invoke this function.  If the
4198  * link is part of a bridge, then the only "capability" it has is the inability
4199  * to do zero copy.
4200  */
4201 boolean_t
4202 i_mac_capab_get(mac_handle_t mh, mac_capab_t cap, void *cap_data)
4203 {
4204 	mac_impl_t *mip = (mac_impl_t *)mh;
4205 
4206 	if (mip->mi_bridge_link != NULL)
4207 		return (cap == MAC_CAPAB_NO_ZCOPY);
4208 	else if (mip->mi_callbacks->mc_callbacks & MC_GETCAPAB)
4209 		return (mip->mi_getcapab(mip->mi_driver, cap, cap_data));
4210 	else
4211 		return (B_FALSE);
4212 }
4213 
4214 /*
4215  * Capability query function. If number of active mac clients is greater than
4216  * 1, only limited capabilities can be advertised to the caller no matter the
4217  * driver has certain capability or not. Else, we query the driver to get the
4218  * capability.
4219  */
4220 boolean_t
4221 mac_capab_get(mac_handle_t mh, mac_capab_t cap, void *cap_data)
4222 {
4223 	mac_impl_t *mip = (mac_impl_t *)mh;
4224 
4225 	/*
4226 	 * if mi_nactiveclients > 1, only MAC_CAPAB_LEGACY, MAC_CAPAB_HCKSUM,
4227 	 * MAC_CAPAB_NO_NATIVEVLAN and MAC_CAPAB_NO_ZCOPY can be advertised.
4228 	 */
4229 	if (mip->mi_nactiveclients > 1) {
4230 		switch (cap) {
4231 		case MAC_CAPAB_NO_ZCOPY:
4232 			return (B_TRUE);
4233 		case MAC_CAPAB_LEGACY:
4234 		case MAC_CAPAB_HCKSUM:
4235 		case MAC_CAPAB_NO_NATIVEVLAN:
4236 			break;
4237 		default:
4238 			return (B_FALSE);
4239 		}
4240 	}
4241 
4242 	/* else get capab from driver */
4243 	return (i_mac_capab_get(mh, cap, cap_data));
4244 }
4245 
4246 boolean_t
4247 mac_sap_verify(mac_handle_t mh, uint32_t sap, uint32_t *bind_sap)
4248 {
4249 	mac_impl_t *mip = (mac_impl_t *)mh;
4250 
4251 	return (mip->mi_type->mt_ops.mtops_sap_verify(sap, bind_sap,
4252 	    mip->mi_pdata));
4253 }
4254 
4255 mblk_t *
4256 mac_header(mac_handle_t mh, const uint8_t *daddr, uint32_t sap, mblk_t *payload,
4257     size_t extra_len)
4258 {
4259 	mac_impl_t	*mip = (mac_impl_t *)mh;
4260 	const uint8_t	*hdr_daddr;
4261 
4262 	/*
4263 	 * If the MAC is point-to-point with a fixed destination address, then
4264 	 * we must always use that destination in the MAC header.
4265 	 */
4266 	hdr_daddr = (mip->mi_dstaddr_set ? mip->mi_dstaddr : daddr);
4267 	return (mip->mi_type->mt_ops.mtops_header(mip->mi_addr, hdr_daddr, sap,
4268 	    mip->mi_pdata, payload, extra_len));
4269 }
4270 
4271 int
4272 mac_header_info(mac_handle_t mh, mblk_t *mp, mac_header_info_t *mhip)
4273 {
4274 	mac_impl_t *mip = (mac_impl_t *)mh;
4275 
4276 	return (mip->mi_type->mt_ops.mtops_header_info(mp, mip->mi_pdata,
4277 	    mhip));
4278 }
4279 
4280 int
4281 mac_vlan_header_info(mac_handle_t mh, mblk_t *mp, mac_header_info_t *mhip)
4282 {
4283 	mac_impl_t	*mip = (mac_impl_t *)mh;
4284 	boolean_t	is_ethernet = (mip->mi_info.mi_media == DL_ETHER);
4285 	int		err = 0;
4286 
4287 	/*
4288 	 * Packets should always be at least 16 bit aligned.
4289 	 */
4290 	ASSERT(IS_P2ALIGNED(mp->b_rptr, sizeof (uint16_t)));
4291 
4292 	if ((err = mac_header_info(mh, mp, mhip)) != 0)
4293 		return (err);
4294 
4295 	/*
4296 	 * If this is a VLAN-tagged Ethernet packet, then the SAP in the
4297 	 * mac_header_info_t as returned by mac_header_info() is
4298 	 * ETHERTYPE_VLAN. We need to grab the ethertype from the VLAN header.
4299 	 */
4300 	if (is_ethernet && (mhip->mhi_bindsap == ETHERTYPE_VLAN)) {
4301 		struct ether_vlan_header *evhp;
4302 		uint16_t sap;
4303 		mblk_t *tmp = NULL;
4304 		size_t size;
4305 
4306 		size = sizeof (struct ether_vlan_header);
4307 		if (MBLKL(mp) < size) {
4308 			/*
4309 			 * Pullup the message in order to get the MAC header
4310 			 * infomation. Note that this is a read-only function,
4311 			 * we keep the input packet intact.
4312 			 */
4313 			if ((tmp = msgpullup(mp, size)) == NULL)
4314 				return (EINVAL);
4315 
4316 			mp = tmp;
4317 		}
4318 		evhp = (struct ether_vlan_header *)mp->b_rptr;
4319 		sap = ntohs(evhp->ether_type);
4320 		(void) mac_sap_verify(mh, sap, &mhip->mhi_bindsap);
4321 		mhip->mhi_hdrsize = sizeof (struct ether_vlan_header);
4322 		mhip->mhi_tci = ntohs(evhp->ether_tci);
4323 		mhip->mhi_istagged = B_TRUE;
4324 		freemsg(tmp);
4325 
4326 		if (VLAN_CFI(mhip->mhi_tci) != ETHER_CFI)
4327 			return (EINVAL);
4328 	} else {
4329 		mhip->mhi_istagged = B_FALSE;
4330 		mhip->mhi_tci = 0;
4331 	}
4332 
4333 	return (0);
4334 }
4335 
4336 mblk_t *
4337 mac_header_cook(mac_handle_t mh, mblk_t *mp)
4338 {
4339 	mac_impl_t *mip = (mac_impl_t *)mh;
4340 
4341 	if (mip->mi_type->mt_ops.mtops_ops & MTOPS_HEADER_COOK) {
4342 		if (DB_REF(mp) > 1) {
4343 			mblk_t *newmp = copymsg(mp);
4344 			if (newmp == NULL)
4345 				return (NULL);
4346 			freemsg(mp);
4347 			mp = newmp;
4348 		}
4349 		return (mip->mi_type->mt_ops.mtops_header_cook(mp,
4350 		    mip->mi_pdata));
4351 	}
4352 	return (mp);
4353 }
4354 
4355 mblk_t *
4356 mac_header_uncook(mac_handle_t mh, mblk_t *mp)
4357 {
4358 	mac_impl_t *mip = (mac_impl_t *)mh;
4359 
4360 	if (mip->mi_type->mt_ops.mtops_ops & MTOPS_HEADER_UNCOOK) {
4361 		if (DB_REF(mp) > 1) {
4362 			mblk_t *newmp = copymsg(mp);
4363 			if (newmp == NULL)
4364 				return (NULL);
4365 			freemsg(mp);
4366 			mp = newmp;
4367 		}
4368 		return (mip->mi_type->mt_ops.mtops_header_uncook(mp,
4369 		    mip->mi_pdata));
4370 	}
4371 	return (mp);
4372 }
4373 
4374 uint_t
4375 mac_addr_len(mac_handle_t mh)
4376 {
4377 	mac_impl_t *mip = (mac_impl_t *)mh;
4378 
4379 	return (mip->mi_type->mt_addr_length);
4380 }
4381 
4382 /* True if a MAC is a VNIC */
4383 boolean_t
4384 mac_is_vnic(mac_handle_t mh)
4385 {
4386 	return (((mac_impl_t *)mh)->mi_state_flags & MIS_IS_VNIC);
4387 }
4388 
4389 mac_handle_t
4390 mac_get_lower_mac_handle(mac_handle_t mh)
4391 {
4392 	mac_impl_t *mip = (mac_impl_t *)mh;
4393 
4394 	ASSERT(mac_is_vnic(mh));
4395 	return (((vnic_t *)mip->mi_driver)->vn_lower_mh);
4396 }
4397 
4398 boolean_t
4399 mac_is_vnic_primary(mac_handle_t mh)
4400 {
4401 	mac_impl_t *mip = (mac_impl_t *)mh;
4402 
4403 	ASSERT(mac_is_vnic(mh));
4404 	return (((vnic_t *)mip->mi_driver)->vn_addr_type ==
4405 	    VNIC_MAC_ADDR_TYPE_PRIMARY);
4406 }
4407 
4408 void
4409 mac_update_resources(mac_resource_props_t *nmrp, mac_resource_props_t *cmrp,
4410     boolean_t is_user_flow)
4411 {
4412 	if (nmrp != NULL && cmrp != NULL) {
4413 		if (nmrp->mrp_mask & MRP_PRIORITY) {
4414 			if (nmrp->mrp_priority == MPL_RESET) {
4415 				cmrp->mrp_mask &= ~MRP_PRIORITY;
4416 				if (is_user_flow) {
4417 					cmrp->mrp_priority =
4418 					    MPL_SUBFLOW_DEFAULT;
4419 				} else {
4420 					cmrp->mrp_priority = MPL_LINK_DEFAULT;
4421 				}
4422 			} else {
4423 				cmrp->mrp_mask |= MRP_PRIORITY;
4424 				cmrp->mrp_priority = nmrp->mrp_priority;
4425 			}
4426 		}
4427 		if (nmrp->mrp_mask & MRP_MAXBW) {
4428 			if (nmrp->mrp_maxbw == MRP_MAXBW_RESETVAL) {
4429 				cmrp->mrp_mask &= ~MRP_MAXBW;
4430 				cmrp->mrp_maxbw = 0;
4431 			} else {
4432 				cmrp->mrp_mask |= MRP_MAXBW;
4433 				cmrp->mrp_maxbw = nmrp->mrp_maxbw;
4434 			}
4435 		}
4436 		if (nmrp->mrp_mask & MRP_CPUS)
4437 			MAC_COPY_CPUS(nmrp, cmrp);
4438 
4439 		if (nmrp->mrp_mask & MRP_POOL) {
4440 			if (strlen(nmrp->mrp_pool) == 0) {
4441 				cmrp->mrp_mask &= ~MRP_POOL;
4442 				bzero(cmrp->mrp_pool, sizeof (cmrp->mrp_pool));
4443 			} else {
4444 				cmrp->mrp_mask |= MRP_POOL;
4445 				(void) strncpy(cmrp->mrp_pool, nmrp->mrp_pool,
4446 				    sizeof (cmrp->mrp_pool));
4447 			}
4448 
4449 		}
4450 
4451 		if (nmrp->mrp_mask & MRP_PROTECT)
4452 			mac_protect_update(nmrp, cmrp);
4453 
4454 		/*
4455 		 * Update the rings specified.
4456 		 */
4457 		if (nmrp->mrp_mask & MRP_RX_RINGS) {
4458 			if (nmrp->mrp_mask & MRP_RINGS_RESET) {
4459 				cmrp->mrp_mask &= ~MRP_RX_RINGS;
4460 				if (cmrp->mrp_mask & MRP_RXRINGS_UNSPEC)
4461 					cmrp->mrp_mask &= ~MRP_RXRINGS_UNSPEC;
4462 				cmrp->mrp_nrxrings = 0;
4463 			} else {
4464 				cmrp->mrp_mask |= MRP_RX_RINGS;
4465 				cmrp->mrp_nrxrings = nmrp->mrp_nrxrings;
4466 			}
4467 		}
4468 		if (nmrp->mrp_mask & MRP_TX_RINGS) {
4469 			if (nmrp->mrp_mask & MRP_RINGS_RESET) {
4470 				cmrp->mrp_mask &= ~MRP_TX_RINGS;
4471 				if (cmrp->mrp_mask & MRP_TXRINGS_UNSPEC)
4472 					cmrp->mrp_mask &= ~MRP_TXRINGS_UNSPEC;
4473 				cmrp->mrp_ntxrings = 0;
4474 			} else {
4475 				cmrp->mrp_mask |= MRP_TX_RINGS;
4476 				cmrp->mrp_ntxrings = nmrp->mrp_ntxrings;
4477 			}
4478 		}
4479 		if (nmrp->mrp_mask & MRP_RXRINGS_UNSPEC)
4480 			cmrp->mrp_mask |= MRP_RXRINGS_UNSPEC;
4481 		else if (cmrp->mrp_mask & MRP_RXRINGS_UNSPEC)
4482 			cmrp->mrp_mask &= ~MRP_RXRINGS_UNSPEC;
4483 
4484 		if (nmrp->mrp_mask & MRP_TXRINGS_UNSPEC)
4485 			cmrp->mrp_mask |= MRP_TXRINGS_UNSPEC;
4486 		else if (cmrp->mrp_mask & MRP_TXRINGS_UNSPEC)
4487 			cmrp->mrp_mask &= ~MRP_TXRINGS_UNSPEC;
4488 	}
4489 }
4490 
4491 /*
4492  * i_mac_set_resources:
4493  *
4494  * This routine associates properties with the primary MAC client of
4495  * the specified MAC instance.
4496  * - Cache the properties in mac_impl_t
4497  * - Apply the properties to the primary MAC client if exists
4498  */
4499 int
4500 i_mac_set_resources(mac_handle_t mh, mac_resource_props_t *mrp)
4501 {
4502 	mac_impl_t		*mip = (mac_impl_t *)mh;
4503 	mac_client_impl_t	*mcip;
4504 	int			err = 0;
4505 	uint32_t		resmask, newresmask;
4506 	mac_resource_props_t	*tmrp, *umrp;
4507 
4508 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
4509 
4510 	err = mac_validate_props(mip, mrp);
4511 	if (err != 0)
4512 		return (err);
4513 
4514 	umrp = kmem_zalloc(sizeof (*umrp), KM_SLEEP);
4515 	bcopy(&mip->mi_resource_props, umrp, sizeof (*umrp));
4516 	resmask = umrp->mrp_mask;
4517 	mac_update_resources(mrp, umrp, B_FALSE);
4518 	newresmask = umrp->mrp_mask;
4519 
4520 	if (resmask == 0 && newresmask != 0) {
4521 		/*
4522 		 * Bandwidth, priority, cpu or pool link properties configured,
4523 		 * must disable fastpath.
4524 		 */
4525 		if ((err = mac_fastpath_disable((mac_handle_t)mip)) != 0) {
4526 			kmem_free(umrp, sizeof (*umrp));
4527 			return (err);
4528 		}
4529 	}
4530 
4531 	/*
4532 	 * Since bind_cpu may be modified by mac_client_set_resources()
4533 	 * we use a copy of bind_cpu and finally cache bind_cpu in mip.
4534 	 * This allows us to cache only user edits in mip.
4535 	 */
4536 	tmrp = kmem_zalloc(sizeof (*tmrp), KM_SLEEP);
4537 	bcopy(mrp, tmrp, sizeof (*tmrp));
4538 	mcip = mac_primary_client_handle(mip);
4539 	if (mcip != NULL && (mcip->mci_state_flags & MCIS_IS_AGGR_PORT) == 0) {
4540 		err = mac_client_set_resources((mac_client_handle_t)mcip, tmrp);
4541 	} else if ((mrp->mrp_mask & MRP_RX_RINGS ||
4542 	    mrp->mrp_mask & MRP_TX_RINGS)) {
4543 		mac_client_impl_t	*vmcip;
4544 
4545 		/*
4546 		 * If the primary is not up, we need to check if there
4547 		 * are any VLANs on this primary. If there are then
4548 		 * we need to set this property on the VLANs since
4549 		 * VLANs follow the primary they are based on. Just
4550 		 * look for the first VLAN and change its properties,
4551 		 * all the other VLANs should be in the same group.
4552 		 */
4553 		for (vmcip = mip->mi_clients_list; vmcip != NULL;
4554 		    vmcip = vmcip->mci_client_next) {
4555 			if ((vmcip->mci_flent->fe_type & FLOW_PRIMARY_MAC) &&
4556 			    mac_client_vid((mac_client_handle_t)vmcip) !=
4557 			    VLAN_ID_NONE) {
4558 				break;
4559 			}
4560 		}
4561 		if (vmcip != NULL) {
4562 			mac_resource_props_t	*omrp;
4563 			mac_resource_props_t	*vmrp;
4564 
4565 			omrp = kmem_zalloc(sizeof (*omrp), KM_SLEEP);
4566 			bcopy(MCIP_RESOURCE_PROPS(vmcip), omrp, sizeof (*omrp));
4567 			/*
4568 			 * We dont' call mac_update_resources since we
4569 			 * want to take only the ring properties and
4570 			 * not all the properties that may have changed.
4571 			 */
4572 			vmrp = MCIP_RESOURCE_PROPS(vmcip);
4573 			if (mrp->mrp_mask & MRP_RX_RINGS) {
4574 				if (mrp->mrp_mask & MRP_RINGS_RESET) {
4575 					vmrp->mrp_mask &= ~MRP_RX_RINGS;
4576 					if (vmrp->mrp_mask &
4577 					    MRP_RXRINGS_UNSPEC) {
4578 						vmrp->mrp_mask &=
4579 						    ~MRP_RXRINGS_UNSPEC;
4580 					}
4581 					vmrp->mrp_nrxrings = 0;
4582 				} else {
4583 					vmrp->mrp_mask |= MRP_RX_RINGS;
4584 					vmrp->mrp_nrxrings = mrp->mrp_nrxrings;
4585 				}
4586 			}
4587 			if (mrp->mrp_mask & MRP_TX_RINGS) {
4588 				if (mrp->mrp_mask & MRP_RINGS_RESET) {
4589 					vmrp->mrp_mask &= ~MRP_TX_RINGS;
4590 					if (vmrp->mrp_mask &
4591 					    MRP_TXRINGS_UNSPEC) {
4592 						vmrp->mrp_mask &=
4593 						    ~MRP_TXRINGS_UNSPEC;
4594 					}
4595 					vmrp->mrp_ntxrings = 0;
4596 				} else {
4597 					vmrp->mrp_mask |= MRP_TX_RINGS;
4598 					vmrp->mrp_ntxrings = mrp->mrp_ntxrings;
4599 				}
4600 			}
4601 			if (mrp->mrp_mask & MRP_RXRINGS_UNSPEC)
4602 				vmrp->mrp_mask |= MRP_RXRINGS_UNSPEC;
4603 
4604 			if (mrp->mrp_mask & MRP_TXRINGS_UNSPEC)
4605 				vmrp->mrp_mask |= MRP_TXRINGS_UNSPEC;
4606 
4607 			if ((err = mac_client_set_rings_prop(vmcip, mrp,
4608 			    omrp)) != 0) {
4609 				bcopy(omrp, MCIP_RESOURCE_PROPS(vmcip),
4610 				    sizeof (*omrp));
4611 			} else {
4612 				mac_set_prim_vlan_rings(mip, vmrp);
4613 			}
4614 			kmem_free(omrp, sizeof (*omrp));
4615 		}
4616 	}
4617 
4618 	/* Only update the values if mac_client_set_resources succeeded */
4619 	if (err == 0) {
4620 		bcopy(umrp, &mip->mi_resource_props, sizeof (*umrp));
4621 		/*
4622 		 * If bandwidth, priority or cpu link properties cleared,
4623 		 * renable fastpath.
4624 		 */
4625 		if (resmask != 0 && newresmask == 0)
4626 			mac_fastpath_enable((mac_handle_t)mip);
4627 	} else if (resmask == 0 && newresmask != 0) {
4628 		mac_fastpath_enable((mac_handle_t)mip);
4629 	}
4630 	kmem_free(tmrp, sizeof (*tmrp));
4631 	kmem_free(umrp, sizeof (*umrp));
4632 	return (err);
4633 }
4634 
4635 int
4636 mac_set_resources(mac_handle_t mh, mac_resource_props_t *mrp)
4637 {
4638 	int err;
4639 
4640 	i_mac_perim_enter((mac_impl_t *)mh);
4641 	err = i_mac_set_resources(mh, mrp);
4642 	i_mac_perim_exit((mac_impl_t *)mh);
4643 	return (err);
4644 }
4645 
4646 /*
4647  * Get the properties cached for the specified MAC instance.
4648  */
4649 void
4650 mac_get_resources(mac_handle_t mh, mac_resource_props_t *mrp)
4651 {
4652 	mac_impl_t		*mip = (mac_impl_t *)mh;
4653 	mac_client_impl_t	*mcip;
4654 
4655 	mcip = mac_primary_client_handle(mip);
4656 	if (mcip != NULL) {
4657 		mac_client_get_resources((mac_client_handle_t)mcip, mrp);
4658 		return;
4659 	}
4660 	bcopy(&mip->mi_resource_props, mrp, sizeof (mac_resource_props_t));
4661 }
4662 
4663 /*
4664  * Get the effective properties from the primary client of the
4665  * specified MAC instance.
4666  */
4667 void
4668 mac_get_effective_resources(mac_handle_t mh, mac_resource_props_t *mrp)
4669 {
4670 	mac_impl_t		*mip = (mac_impl_t *)mh;
4671 	mac_client_impl_t	*mcip;
4672 
4673 	mcip = mac_primary_client_handle(mip);
4674 	if (mcip != NULL) {
4675 		mac_client_get_effective_resources((mac_client_handle_t)mcip,
4676 		    mrp);
4677 		return;
4678 	}
4679 	bzero(mrp, sizeof (mac_resource_props_t));
4680 }
4681 
4682 int
4683 mac_set_pvid(mac_handle_t mh, uint16_t pvid)
4684 {
4685 	mac_impl_t *mip = (mac_impl_t *)mh;
4686 	mac_client_impl_t *mcip;
4687 	mac_unicast_impl_t *muip;
4688 
4689 	i_mac_perim_enter(mip);
4690 	if (pvid != 0) {
4691 		for (mcip = mip->mi_clients_list; mcip != NULL;
4692 		    mcip = mcip->mci_client_next) {
4693 			for (muip = mcip->mci_unicast_list; muip != NULL;
4694 			    muip = muip->mui_next) {
4695 				if (muip->mui_vid == pvid) {
4696 					i_mac_perim_exit(mip);
4697 					return (EBUSY);
4698 				}
4699 			}
4700 		}
4701 	}
4702 	mip->mi_pvid = pvid;
4703 	i_mac_perim_exit(mip);
4704 	return (0);
4705 }
4706 
4707 uint16_t
4708 mac_get_pvid(mac_handle_t mh)
4709 {
4710 	mac_impl_t *mip = (mac_impl_t *)mh;
4711 
4712 	return (mip->mi_pvid);
4713 }
4714 
4715 uint32_t
4716 mac_get_llimit(mac_handle_t mh)
4717 {
4718 	mac_impl_t *mip = (mac_impl_t *)mh;
4719 
4720 	return (mip->mi_llimit);
4721 }
4722 
4723 uint32_t
4724 mac_get_ldecay(mac_handle_t mh)
4725 {
4726 	mac_impl_t *mip = (mac_impl_t *)mh;
4727 
4728 	return (mip->mi_ldecay);
4729 }
4730 
4731 /*
4732  * Rename a mac client, its flow, and the kstat.
4733  */
4734 int
4735 mac_rename_primary(mac_handle_t mh, const char *new_name)
4736 {
4737 	mac_impl_t		*mip = (mac_impl_t *)mh;
4738 	mac_client_impl_t	*cur_clnt = NULL;
4739 	flow_entry_t		*fep;
4740 
4741 	i_mac_perim_enter(mip);
4742 
4743 	/*
4744 	 * VNICs: we need to change the sys flow name and
4745 	 * the associated flow kstat.
4746 	 */
4747 	if (mip->mi_state_flags & MIS_IS_VNIC) {
4748 		mac_client_impl_t *mcip = mac_vnic_lower(mip);
4749 		ASSERT(new_name != NULL);
4750 		mac_rename_flow_names(mcip, new_name);
4751 		mac_stat_rename(mcip);
4752 		goto done;
4753 	}
4754 	/*
4755 	 * This mac may itself be an aggr link, or it may have some client
4756 	 * which is an aggr port. For both cases, we need to change the
4757 	 * aggr port's mac client name, its flow name and the associated flow
4758 	 * kstat.
4759 	 */
4760 	if (mip->mi_state_flags & MIS_IS_AGGR) {
4761 		mac_capab_aggr_t aggr_cap;
4762 		mac_rename_fn_t rename_fn;
4763 		boolean_t ret;
4764 
4765 		ASSERT(new_name != NULL);
4766 		ret = i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_AGGR,
4767 		    (void *)(&aggr_cap));
4768 		ASSERT(ret == B_TRUE);
4769 		rename_fn = aggr_cap.mca_rename_fn;
4770 		rename_fn(new_name, mip->mi_driver);
4771 		/*
4772 		 * The aggr's client name and kstat flow name will be
4773 		 * updated below, i.e. via mac_rename_flow_names.
4774 		 */
4775 	}
4776 
4777 	for (cur_clnt = mip->mi_clients_list; cur_clnt != NULL;
4778 	    cur_clnt = cur_clnt->mci_client_next) {
4779 		if (cur_clnt->mci_state_flags & MCIS_IS_AGGR_PORT) {
4780 			if (new_name != NULL) {
4781 				char *str_st = cur_clnt->mci_name;
4782 				char *str_del = strchr(str_st, '-');
4783 
4784 				ASSERT(str_del != NULL);
4785 				bzero(str_del + 1, MAXNAMELEN -
4786 				    (str_del - str_st + 1));
4787 				bcopy(new_name, str_del + 1,
4788 				    strlen(new_name));
4789 			}
4790 			fep = cur_clnt->mci_flent;
4791 			mac_rename_flow(fep, cur_clnt->mci_name);
4792 			break;
4793 		} else if (new_name != NULL &&
4794 		    cur_clnt->mci_state_flags & MCIS_USE_DATALINK_NAME) {
4795 			mac_rename_flow_names(cur_clnt, new_name);
4796 			break;
4797 		}
4798 	}
4799 
4800 	/* Recreate kstats associated with aggr pseudo rings */
4801 	if (mip->mi_state_flags & MIS_IS_AGGR)
4802 		mac_pseudo_ring_stat_rename(mip);
4803 
4804 done:
4805 	i_mac_perim_exit(mip);
4806 	return (0);
4807 }
4808 
4809 /*
4810  * Rename the MAC client's flow names
4811  */
4812 static void
4813 mac_rename_flow_names(mac_client_impl_t *mcip, const char *new_name)
4814 {
4815 	flow_entry_t	*flent;
4816 	uint16_t	vid;
4817 	char		flowname[MAXFLOWNAMELEN];
4818 	mac_impl_t	*mip = mcip->mci_mip;
4819 
4820 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
4821 
4822 	/*
4823 	 * Use mi_rw_lock to ensure that threads not in the mac perimeter
4824 	 * see a self-consistent value for mci_name
4825 	 */
4826 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
4827 	(void) strlcpy(mcip->mci_name, new_name, sizeof (mcip->mci_name));
4828 	rw_exit(&mip->mi_rw_lock);
4829 
4830 	mac_rename_flow(mcip->mci_flent, new_name);
4831 
4832 	if (mcip->mci_nflents == 1)
4833 		return;
4834 
4835 	/*
4836 	 * We have to rename all the others too, no stats to destroy for
4837 	 * these.
4838 	 */
4839 	for (flent = mcip->mci_flent_list; flent != NULL;
4840 	    flent = flent->fe_client_next) {
4841 		if (flent != mcip->mci_flent) {
4842 			vid = i_mac_flow_vid(flent);
4843 			(void) sprintf(flowname, "%s%u", new_name, vid);
4844 			mac_flow_set_name(flent, flowname);
4845 		}
4846 	}
4847 }
4848 
4849 
4850 /*
4851  * Add a flow to the MAC client's flow list - i.e list of MAC/VID tuples
4852  * defined for the specified MAC client.
4853  */
4854 static void
4855 mac_client_add_to_flow_list(mac_client_impl_t *mcip, flow_entry_t *flent)
4856 {
4857 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
4858 	/*
4859 	 * The promisc Rx data path walks the mci_flent_list. Protect by
4860 	 * using mi_rw_lock
4861 	 */
4862 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
4863 
4864 	mcip->mci_vidcache = MCIP_VIDCACHE_INVALID;
4865 
4866 	/* Add it to the head */
4867 	flent->fe_client_next = mcip->mci_flent_list;
4868 	mcip->mci_flent_list = flent;
4869 	mcip->mci_nflents++;
4870 
4871 	/*
4872 	 * Keep track of the number of non-zero VIDs addresses per MAC
4873 	 * client to avoid figuring it out in the data-path.
4874 	 */
4875 	if (i_mac_flow_vid(flent) != VLAN_ID_NONE)
4876 		mcip->mci_nvids++;
4877 
4878 	rw_exit(&mcip->mci_rw_lock);
4879 }
4880 
4881 /*
4882  * Remove a flow entry from the MAC client's list.
4883  */
4884 static void
4885 mac_client_remove_flow_from_list(mac_client_impl_t *mcip, flow_entry_t *flent)
4886 {
4887 	flow_entry_t	*fe = mcip->mci_flent_list;
4888 	flow_entry_t	*prev_fe = NULL;
4889 
4890 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
4891 	/*
4892 	 * The promisc Rx data path walks the mci_flent_list. Protect by
4893 	 * using mci_rw_lock
4894 	 */
4895 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
4896 	mcip->mci_vidcache = MCIP_VIDCACHE_INVALID;
4897 
4898 	while ((fe != NULL) && (fe != flent)) {
4899 		prev_fe = fe;
4900 		fe = fe->fe_client_next;
4901 	}
4902 
4903 	ASSERT(fe != NULL);
4904 	if (prev_fe == NULL) {
4905 		/* Deleting the first node */
4906 		mcip->mci_flent_list = fe->fe_client_next;
4907 	} else {
4908 		prev_fe->fe_client_next = fe->fe_client_next;
4909 	}
4910 	mcip->mci_nflents--;
4911 
4912 	if (i_mac_flow_vid(flent) != VLAN_ID_NONE)
4913 		mcip->mci_nvids--;
4914 
4915 	rw_exit(&mcip->mci_rw_lock);
4916 }
4917 
4918 /*
4919  * Check if the given VID belongs to this MAC client.
4920  */
4921 boolean_t
4922 mac_client_check_flow_vid(mac_client_impl_t *mcip, uint16_t vid)
4923 {
4924 	flow_entry_t	*flent;
4925 	uint16_t	mci_vid;
4926 	uint32_t	cache = mcip->mci_vidcache;
4927 
4928 	/*
4929 	 * In hopes of not having to touch the mci_rw_lock, check to see if
4930 	 * this vid matches our cached result.
4931 	 */
4932 	if (MCIP_VIDCACHE_ISVALID(cache) && MCIP_VIDCACHE_VID(cache) == vid)
4933 		return (MCIP_VIDCACHE_BOOL(cache) ? B_TRUE : B_FALSE);
4934 
4935 	/* The mci_flent_list is protected by mci_rw_lock */
4936 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
4937 	for (flent = mcip->mci_flent_list; flent != NULL;
4938 	    flent = flent->fe_client_next) {
4939 		mci_vid = i_mac_flow_vid(flent);
4940 		if (vid == mci_vid) {
4941 			mcip->mci_vidcache = MCIP_VIDCACHE_CACHE(vid, B_TRUE);
4942 			rw_exit(&mcip->mci_rw_lock);
4943 			return (B_TRUE);
4944 		}
4945 	}
4946 
4947 	mcip->mci_vidcache = MCIP_VIDCACHE_CACHE(vid, B_FALSE);
4948 	rw_exit(&mcip->mci_rw_lock);
4949 	return (B_FALSE);
4950 }
4951 
4952 /*
4953  * Get the flow entry for the specified <MAC addr, VID> tuple.
4954  */
4955 static flow_entry_t *
4956 mac_client_get_flow(mac_client_impl_t *mcip, mac_unicast_impl_t *muip)
4957 {
4958 	mac_address_t *map = mcip->mci_unicast;
4959 	flow_entry_t *flent;
4960 	uint16_t vid;
4961 	flow_desc_t flow_desc;
4962 
4963 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
4964 
4965 	mac_flow_get_desc(mcip->mci_flent, &flow_desc);
4966 	if (bcmp(flow_desc.fd_dst_mac, map->ma_addr, map->ma_len) != 0)
4967 		return (NULL);
4968 
4969 	for (flent = mcip->mci_flent_list; flent != NULL;
4970 	    flent = flent->fe_client_next) {
4971 		vid = i_mac_flow_vid(flent);
4972 		if (vid == muip->mui_vid) {
4973 			return (flent);
4974 		}
4975 	}
4976 
4977 	return (NULL);
4978 }
4979 
4980 /*
4981  * Since mci_flent has the SRSs, when we want to remove it, we replace
4982  * the flow_desc_t in mci_flent with that of an existing flent and then
4983  * remove that flent instead of mci_flent.
4984  */
4985 static flow_entry_t *
4986 mac_client_swap_mciflent(mac_client_impl_t *mcip)
4987 {
4988 	flow_entry_t	*flent = mcip->mci_flent;
4989 	flow_tab_t	*ft = flent->fe_flow_tab;
4990 	flow_entry_t	*flent1;
4991 	flow_desc_t	fl_desc;
4992 	char		fl_name[MAXFLOWNAMELEN];
4993 	int		err;
4994 
4995 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
4996 	ASSERT(mcip->mci_nflents > 1);
4997 
4998 	/* get the next flent following the primary flent  */
4999 	flent1 = mcip->mci_flent_list->fe_client_next;
5000 	ASSERT(flent1 != NULL && flent1->fe_flow_tab == ft);
5001 
5002 	/*
5003 	 * Remove the flent from the flow table before updating the
5004 	 * flow descriptor as the hash depends on the flow descriptor.
5005 	 * This also helps incoming packet classification avoid having
5006 	 * to grab fe_lock. Access to fe_flow_desc of a flent not in the
5007 	 * flow table is done under the fe_lock so that log or stat functions
5008 	 * see a self-consistent fe_flow_desc. The name and desc are specific
5009 	 * to a flow, the rest are shared by all the clients, including
5010 	 * resource control etc.
5011 	 */
5012 	mac_flow_remove(ft, flent, B_TRUE);
5013 	mac_flow_remove(ft, flent1, B_TRUE);
5014 
5015 	bcopy(&flent->fe_flow_desc, &fl_desc, sizeof (flow_desc_t));
5016 	bcopy(flent->fe_flow_name, fl_name, MAXFLOWNAMELEN);
5017 
5018 	/* update the primary flow entry */
5019 	mutex_enter(&flent->fe_lock);
5020 	bcopy(&flent1->fe_flow_desc, &flent->fe_flow_desc,
5021 	    sizeof (flow_desc_t));
5022 	bcopy(&flent1->fe_flow_name, &flent->fe_flow_name, MAXFLOWNAMELEN);
5023 	mutex_exit(&flent->fe_lock);
5024 
5025 	/* update the flow entry that is to be freed */
5026 	mutex_enter(&flent1->fe_lock);
5027 	bcopy(&fl_desc, &flent1->fe_flow_desc, sizeof (flow_desc_t));
5028 	bcopy(fl_name, &flent1->fe_flow_name, MAXFLOWNAMELEN);
5029 	mutex_exit(&flent1->fe_lock);
5030 
5031 	/* now reinsert the flow entries in the table */
5032 	err = mac_flow_add(ft, flent);
5033 	ASSERT(err == 0);
5034 
5035 	err = mac_flow_add(ft, flent1);
5036 	ASSERT(err == 0);
5037 
5038 	return (flent1);
5039 }
5040 
5041 /*
5042  * Return whether there is only one flow entry associated with this
5043  * MAC client.
5044  */
5045 static boolean_t
5046 mac_client_single_rcvr(mac_client_impl_t *mcip)
5047 {
5048 	return (mcip->mci_nflents == 1);
5049 }
5050 
5051 int
5052 mac_validate_props(mac_impl_t *mip, mac_resource_props_t *mrp)
5053 {
5054 	boolean_t		reset;
5055 	uint32_t		rings_needed;
5056 	uint32_t		rings_avail;
5057 	mac_group_type_t	gtype;
5058 	mac_resource_props_t	*mip_mrp;
5059 
5060 	if (mrp == NULL)
5061 		return (0);
5062 
5063 	if (mrp->mrp_mask & MRP_PRIORITY) {
5064 		mac_priority_level_t	pri = mrp->mrp_priority;
5065 
5066 		if (pri < MPL_LOW || pri > MPL_RESET)
5067 			return (EINVAL);
5068 	}
5069 
5070 	if (mrp->mrp_mask & MRP_MAXBW) {
5071 		uint64_t maxbw = mrp->mrp_maxbw;
5072 
5073 		if (maxbw < MRP_MAXBW_MINVAL && maxbw != 0)
5074 			return (EINVAL);
5075 	}
5076 	if (mrp->mrp_mask & MRP_CPUS) {
5077 		int i, j;
5078 		mac_cpu_mode_t	fanout;
5079 
5080 		if (mrp->mrp_ncpus > ncpus)
5081 			return (EINVAL);
5082 
5083 		for (i = 0; i < mrp->mrp_ncpus; i++) {
5084 			for (j = 0; j < mrp->mrp_ncpus; j++) {
5085 				if (i != j &&
5086 				    mrp->mrp_cpu[i] == mrp->mrp_cpu[j]) {
5087 					return (EINVAL);
5088 				}
5089 			}
5090 		}
5091 
5092 		for (i = 0; i < mrp->mrp_ncpus; i++) {
5093 			cpu_t *cp;
5094 			int rv;
5095 
5096 			mutex_enter(&cpu_lock);
5097 			cp = cpu_get(mrp->mrp_cpu[i]);
5098 			if (cp != NULL)
5099 				rv = cpu_is_online(cp);
5100 			else
5101 				rv = 0;
5102 			mutex_exit(&cpu_lock);
5103 			if (rv == 0)
5104 				return (EINVAL);
5105 		}
5106 
5107 		fanout = mrp->mrp_fanout_mode;
5108 		if (fanout < 0 || fanout > MCM_CPUS)
5109 			return (EINVAL);
5110 	}
5111 
5112 	if (mrp->mrp_mask & MRP_PROTECT) {
5113 		int err = mac_protect_validate(mrp);
5114 		if (err != 0)
5115 			return (err);
5116 	}
5117 
5118 	if (!(mrp->mrp_mask & MRP_RX_RINGS) &&
5119 	    !(mrp->mrp_mask & MRP_TX_RINGS)) {
5120 		return (0);
5121 	}
5122 
5123 	/*
5124 	 * mip will be null when we come from mac_flow_create or
5125 	 * mac_link_flow_modify. In the latter case it is a user flow,
5126 	 * for which we don't support rings. In the former we would
5127 	 * have validated the props beforehand (i_mac_unicast_add ->
5128 	 * mac_client_set_resources -> validate for the primary and
5129 	 * vnic_dev_create -> mac_client_set_resources -> validate for
5130 	 * a vnic.
5131 	 */
5132 	if (mip == NULL)
5133 		return (0);
5134 
5135 	/*
5136 	 * We don't support setting rings property for a VNIC that is using a
5137 	 * primary address (VLAN)
5138 	 */
5139 	if ((mip->mi_state_flags & MIS_IS_VNIC) &&
5140 	    mac_is_vnic_primary((mac_handle_t)mip)) {
5141 		return (ENOTSUP);
5142 	}
5143 
5144 	mip_mrp = &mip->mi_resource_props;
5145 	/*
5146 	 * The rings property should be validated against the NICs
5147 	 * resources
5148 	 */
5149 	if (mip->mi_state_flags & MIS_IS_VNIC)
5150 		mip = (mac_impl_t *)mac_get_lower_mac_handle((mac_handle_t)mip);
5151 
5152 	reset = mrp->mrp_mask & MRP_RINGS_RESET;
5153 	/*
5154 	 * If groups are not supported, return error.
5155 	 */
5156 	if (((mrp->mrp_mask & MRP_RX_RINGS) && mip->mi_rx_groups == NULL) ||
5157 	    ((mrp->mrp_mask & MRP_TX_RINGS) && mip->mi_tx_groups == NULL)) {
5158 		return (EINVAL);
5159 	}
5160 	/*
5161 	 * If we are just resetting, there is no validation needed.
5162 	 */
5163 	if (reset)
5164 		return (0);
5165 
5166 	if (mrp->mrp_mask & MRP_RX_RINGS) {
5167 		rings_needed = mrp->mrp_nrxrings;
5168 		/*
5169 		 * We just want to check if the number of additional
5170 		 * rings requested is available.
5171 		 */
5172 		if (mip_mrp->mrp_mask & MRP_RX_RINGS) {
5173 			if (mrp->mrp_nrxrings > mip_mrp->mrp_nrxrings)
5174 				/* Just check for the additional rings */
5175 				rings_needed -= mip_mrp->mrp_nrxrings;
5176 			else
5177 				/* We are not asking for additional rings */
5178 				rings_needed = 0;
5179 		}
5180 		rings_avail = mip->mi_rxrings_avail;
5181 		gtype = mip->mi_rx_group_type;
5182 	} else {
5183 		rings_needed = mrp->mrp_ntxrings;
5184 		/* Similarly for the TX rings */
5185 		if (mip_mrp->mrp_mask & MRP_TX_RINGS) {
5186 			if (mrp->mrp_ntxrings > mip_mrp->mrp_ntxrings)
5187 				/* Just check for the additional rings */
5188 				rings_needed -= mip_mrp->mrp_ntxrings;
5189 			else
5190 				/* We are not asking for additional rings */
5191 				rings_needed = 0;
5192 		}
5193 		rings_avail = mip->mi_txrings_avail;
5194 		gtype = mip->mi_tx_group_type;
5195 	}
5196 
5197 	/* Error if the group is dynamic .. */
5198 	if (gtype == MAC_GROUP_TYPE_DYNAMIC) {
5199 		/*
5200 		 * .. and rings specified are more than available.
5201 		 */
5202 		if (rings_needed > rings_avail)
5203 			return (EINVAL);
5204 	} else {
5205 		/*
5206 		 * OR group is static and we have specified some rings.
5207 		 */
5208 		if (rings_needed > 0)
5209 			return (EINVAL);
5210 	}
5211 	return (0);
5212 }
5213 
5214 /*
5215  * Send a MAC_NOTE_LINK notification to all the MAC clients whenever the
5216  * underlying physical link is down. This is to allow MAC clients to
5217  * communicate with other clients.
5218  */
5219 void
5220 mac_virtual_link_update(mac_impl_t *mip)
5221 {
5222 	if (mip->mi_linkstate != LINK_STATE_UP)
5223 		i_mac_notify(mip, MAC_NOTE_LINK);
5224 }
5225 
5226 /*
5227  * For clients that have a pass-thru MAC, e.g. VNIC, we set the VNIC's
5228  * mac handle in the client.
5229  */
5230 void
5231 mac_set_upper_mac(mac_client_handle_t mch, mac_handle_t mh,
5232     mac_resource_props_t *mrp)
5233 {
5234 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
5235 	mac_impl_t		*mip = (mac_impl_t *)mh;
5236 
5237 	mcip->mci_upper_mip = mip;
5238 	/* If there are any properties, copy it over too */
5239 	if (mrp != NULL) {
5240 		bcopy(mrp, &mip->mi_resource_props,
5241 		    sizeof (mac_resource_props_t));
5242 	}
5243 }
5244 
5245 /*
5246  * Mark the mac as being used exclusively by the single mac client that is
5247  * doing some control operation on this mac. No further opens of this mac
5248  * will be allowed until this client calls mac_unmark_exclusive. The mac
5249  * client calling this function must already be in the mac perimeter
5250  */
5251 int
5252 mac_mark_exclusive(mac_handle_t mh)
5253 {
5254 	mac_impl_t	*mip = (mac_impl_t *)mh;
5255 
5256 	ASSERT(MAC_PERIM_HELD(mh));
5257 	/*
5258 	 * Look up its entry in the global hash table.
5259 	 */
5260 	rw_enter(&i_mac_impl_lock, RW_WRITER);
5261 	if (mip->mi_state_flags & MIS_DISABLED) {
5262 		rw_exit(&i_mac_impl_lock);
5263 		return (ENOENT);
5264 	}
5265 
5266 	/*
5267 	 * A reference to mac is held even if the link is not plumbed.
5268 	 * In i_dls_link_create() we open the MAC interface and hold the
5269 	 * reference. There is an additional reference for the mac_open
5270 	 * done in acquiring the mac perimeter
5271 	 */
5272 	if (mip->mi_ref != 2) {
5273 		rw_exit(&i_mac_impl_lock);
5274 		return (EBUSY);
5275 	}
5276 
5277 	ASSERT(!(mip->mi_state_flags & MIS_EXCLUSIVE_HELD));
5278 	mip->mi_state_flags |= MIS_EXCLUSIVE_HELD;
5279 	rw_exit(&i_mac_impl_lock);
5280 	return (0);
5281 }
5282 
5283 void
5284 mac_unmark_exclusive(mac_handle_t mh)
5285 {
5286 	mac_impl_t	*mip = (mac_impl_t *)mh;
5287 
5288 	ASSERT(MAC_PERIM_HELD(mh));
5289 
5290 	rw_enter(&i_mac_impl_lock, RW_WRITER);
5291 	/* 1 for the creation and another for the perimeter */
5292 	ASSERT(mip->mi_ref == 2 && (mip->mi_state_flags & MIS_EXCLUSIVE_HELD));
5293 	mip->mi_state_flags &= ~MIS_EXCLUSIVE_HELD;
5294 	rw_exit(&i_mac_impl_lock);
5295 }
5296 
5297 /*
5298  * Set the MTU for the specified MAC.
5299  */
5300 int
5301 mac_set_mtu(mac_handle_t mh, uint_t new_mtu, uint_t *old_mtu_arg)
5302 {
5303 	mac_impl_t *mip = (mac_impl_t *)mh;
5304 	uint_t old_mtu;
5305 	int rv = 0;
5306 
5307 	i_mac_perim_enter(mip);
5308 
5309 	if (!(mip->mi_callbacks->mc_callbacks & (MC_SETPROP|MC_GETPROP))) {
5310 		rv = ENOTSUP;
5311 		goto bail;
5312 	}
5313 
5314 	old_mtu = mip->mi_sdu_max;
5315 
5316 	if (new_mtu == 0 || new_mtu < mip->mi_sdu_min) {
5317 		rv = EINVAL;
5318 		goto bail;
5319 	}
5320 
5321 	rw_enter(&mip->mi_rw_lock, RW_READER);
5322 	if (mip->mi_mtrp != NULL && new_mtu < mip->mi_mtrp->mtr_mtu) {
5323 		rv = EBUSY;
5324 		rw_exit(&mip->mi_rw_lock);
5325 		goto bail;
5326 	}
5327 	rw_exit(&mip->mi_rw_lock);
5328 
5329 	if (old_mtu != new_mtu) {
5330 		rv = mip->mi_callbacks->mc_setprop(mip->mi_driver,
5331 		    "mtu", MAC_PROP_MTU, sizeof (uint_t), &new_mtu);
5332 		if (rv != 0)
5333 			goto bail;
5334 		rv = mac_maxsdu_update(mh, new_mtu);
5335 		ASSERT(rv == 0);
5336 	}
5337 
5338 bail:
5339 	i_mac_perim_exit(mip);
5340 
5341 	if (rv == 0 && old_mtu_arg != NULL)
5342 		*old_mtu_arg = old_mtu;
5343 	return (rv);
5344 }
5345 
5346 /*
5347  * Return the RX h/w information for the group indexed by grp_num.
5348  */
5349 void
5350 mac_get_hwrxgrp_info(mac_handle_t mh, int grp_index, uint_t *grp_num,
5351     uint_t *n_rings, uint_t *rings, uint_t *type, uint_t *n_clnts,
5352     char *clnts_name)
5353 {
5354 	mac_impl_t *mip = (mac_impl_t *)mh;
5355 	mac_grp_client_t *mcip;
5356 	uint_t i = 0, index = 0;
5357 	mac_ring_t	*ring;
5358 
5359 	/* Revisit when we implement fully dynamic group allocation */
5360 	ASSERT(grp_index >= 0 && grp_index < mip->mi_rx_group_count);
5361 
5362 	rw_enter(&mip->mi_rw_lock, RW_READER);
5363 	*grp_num = mip->mi_rx_groups[grp_index].mrg_index;
5364 	*type = mip->mi_rx_groups[grp_index].mrg_type;
5365 	*n_rings = mip->mi_rx_groups[grp_index].mrg_cur_count;
5366 	ring = mip->mi_rx_groups[grp_index].mrg_rings;
5367 	for (index = 0; index < mip->mi_rx_groups[grp_index].mrg_cur_count;
5368 	    index++) {
5369 		rings[index] = ring->mr_index;
5370 		ring = ring->mr_next;
5371 	}
5372 	/* Assuming the 1st is the default group */
5373 	index = 0;
5374 	if (grp_index == 0) {
5375 		(void) strlcpy(clnts_name, "<default,mcast>,",
5376 		    MAXCLIENTNAMELEN);
5377 		index += strlen("<default,mcast>,");
5378 	}
5379 	for (mcip = mip->mi_rx_groups[grp_index].mrg_clients; mcip != NULL;
5380 	    mcip = mcip->mgc_next) {
5381 		int name_len = strlen(mcip->mgc_client->mci_name);
5382 
5383 		/*
5384 		 * MAXCLIENTNAMELEN is the buffer size reserved for client
5385 		 * names.
5386 		 * XXXX Formating the client name string needs to be moved
5387 		 * to user land when fixing the size of dhi_clnts in
5388 		 * dld_hwgrpinfo_t. We should use n_clients * client_name for
5389 		 * dhi_clntsin instead of MAXCLIENTNAMELEN
5390 		 */
5391 		if (index + name_len >= MAXCLIENTNAMELEN) {
5392 			index = MAXCLIENTNAMELEN;
5393 			break;
5394 		}
5395 		bcopy(mcip->mgc_client->mci_name, &(clnts_name[index]),
5396 		    name_len);
5397 		index += name_len;
5398 		clnts_name[index++] = ',';
5399 		i++;
5400 	}
5401 
5402 	/* Get rid of the last , */
5403 	if (index > 0)
5404 		clnts_name[index - 1] = '\0';
5405 	*n_clnts = i;
5406 	rw_exit(&mip->mi_rw_lock);
5407 }
5408 
5409 /*
5410  * Return the TX h/w information for the group indexed by grp_num.
5411  */
5412 void
5413 mac_get_hwtxgrp_info(mac_handle_t mh, int grp_index, uint_t *grp_num,
5414     uint_t *n_rings, uint_t *rings, uint_t *type, uint_t *n_clnts,
5415     char *clnts_name)
5416 {
5417 	mac_impl_t *mip = (mac_impl_t *)mh;
5418 	mac_grp_client_t *mcip;
5419 	uint_t i = 0, index = 0;
5420 	mac_ring_t	*ring;
5421 
5422 	/* Revisit when we implement fully dynamic group allocation */
5423 	ASSERT(grp_index >= 0 && grp_index <= mip->mi_tx_group_count);
5424 
5425 	rw_enter(&mip->mi_rw_lock, RW_READER);
5426 	*grp_num = mip->mi_tx_groups[grp_index].mrg_index > 0 ?
5427 	    mip->mi_tx_groups[grp_index].mrg_index : grp_index;
5428 	*type = mip->mi_tx_groups[grp_index].mrg_type;
5429 	*n_rings = mip->mi_tx_groups[grp_index].mrg_cur_count;
5430 	ring = mip->mi_tx_groups[grp_index].mrg_rings;
5431 	for (index = 0; index < mip->mi_tx_groups[grp_index].mrg_cur_count;
5432 	    index++) {
5433 		rings[index] = ring->mr_index;
5434 		ring = ring->mr_next;
5435 	}
5436 	index = 0;
5437 	/* Default group has an index of -1 */
5438 	if (mip->mi_tx_groups[grp_index].mrg_index < 0) {
5439 		(void) strlcpy(clnts_name, "<default>,",
5440 		    MAXCLIENTNAMELEN);
5441 		index += strlen("<default>,");
5442 	}
5443 	for (mcip = mip->mi_tx_groups[grp_index].mrg_clients; mcip != NULL;
5444 	    mcip = mcip->mgc_next) {
5445 		int name_len = strlen(mcip->mgc_client->mci_name);
5446 
5447 		/*
5448 		 * MAXCLIENTNAMELEN is the buffer size reserved for client
5449 		 * names.
5450 		 * XXXX Formating the client name string needs to be moved
5451 		 * to user land when fixing the size of dhi_clnts in
5452 		 * dld_hwgrpinfo_t. We should use n_clients * client_name for
5453 		 * dhi_clntsin instead of MAXCLIENTNAMELEN
5454 		 */
5455 		if (index + name_len >= MAXCLIENTNAMELEN) {
5456 			index = MAXCLIENTNAMELEN;
5457 			break;
5458 		}
5459 		bcopy(mcip->mgc_client->mci_name, &(clnts_name[index]),
5460 		    name_len);
5461 		index += name_len;
5462 		clnts_name[index++] = ',';
5463 		i++;
5464 	}
5465 
5466 	/* Get rid of the last , */
5467 	if (index > 0)
5468 		clnts_name[index - 1] = '\0';
5469 	*n_clnts = i;
5470 	rw_exit(&mip->mi_rw_lock);
5471 }
5472 
5473 /*
5474  * Return the group count for RX or TX.
5475  */
5476 uint_t
5477 mac_hwgrp_num(mac_handle_t mh, int type)
5478 {
5479 	mac_impl_t *mip = (mac_impl_t *)mh;
5480 
5481 	/*
5482 	 * Return the Rx and Tx group count; for the Tx we need to
5483 	 * include the default too.
5484 	 */
5485 	return (type == MAC_RING_TYPE_RX ? mip->mi_rx_group_count :
5486 	    mip->mi_tx_groups != NULL ? mip->mi_tx_group_count + 1 : 0);
5487 }
5488 
5489 /*
5490  * The total number of free TX rings for this MAC.
5491  */
5492 uint_t
5493 mac_txavail_get(mac_handle_t mh)
5494 {
5495 	mac_impl_t	*mip = (mac_impl_t *)mh;
5496 
5497 	return (mip->mi_txrings_avail);
5498 }
5499 
5500 /*
5501  * The total number of free RX rings for this MAC.
5502  */
5503 uint_t
5504 mac_rxavail_get(mac_handle_t mh)
5505 {
5506 	mac_impl_t	*mip = (mac_impl_t *)mh;
5507 
5508 	return (mip->mi_rxrings_avail);
5509 }
5510 
5511 /*
5512  * The total number of reserved RX rings on this MAC.
5513  */
5514 uint_t
5515 mac_rxrsvd_get(mac_handle_t mh)
5516 {
5517 	mac_impl_t	*mip = (mac_impl_t *)mh;
5518 
5519 	return (mip->mi_rxrings_rsvd);
5520 }
5521 
5522 /*
5523  * The total number of reserved TX rings on this MAC.
5524  */
5525 uint_t
5526 mac_txrsvd_get(mac_handle_t mh)
5527 {
5528 	mac_impl_t	*mip = (mac_impl_t *)mh;
5529 
5530 	return (mip->mi_txrings_rsvd);
5531 }
5532 
5533 /*
5534  * Total number of free RX groups on this MAC.
5535  */
5536 uint_t
5537 mac_rxhwlnksavail_get(mac_handle_t mh)
5538 {
5539 	mac_impl_t	*mip = (mac_impl_t *)mh;
5540 
5541 	return (mip->mi_rxhwclnt_avail);
5542 }
5543 
5544 /*
5545  * Total number of RX groups reserved on this MAC.
5546  */
5547 uint_t
5548 mac_rxhwlnksrsvd_get(mac_handle_t mh)
5549 {
5550 	mac_impl_t	*mip = (mac_impl_t *)mh;
5551 
5552 	return (mip->mi_rxhwclnt_used);
5553 }
5554 
5555 /*
5556  * Total number of free TX groups on this MAC.
5557  */
5558 uint_t
5559 mac_txhwlnksavail_get(mac_handle_t mh)
5560 {
5561 	mac_impl_t	*mip = (mac_impl_t *)mh;
5562 
5563 	return (mip->mi_txhwclnt_avail);
5564 }
5565 
5566 /*
5567  * Total number of TX groups reserved on this MAC.
5568  */
5569 uint_t
5570 mac_txhwlnksrsvd_get(mac_handle_t mh)
5571 {
5572 	mac_impl_t	*mip = (mac_impl_t *)mh;
5573 
5574 	return (mip->mi_txhwclnt_used);
5575 }
5576 
5577 /*
5578  * Initialize the rings property for a mac client. A non-0 value for
5579  * rxring or txring specifies the number of rings required, a value
5580  * of MAC_RXRINGS_NONE/MAC_TXRINGS_NONE specifies that it doesn't need
5581  * any RX/TX rings and a value of MAC_RXRINGS_DONTCARE/MAC_TXRINGS_DONTCARE
5582  * means the system can decide whether it can give any rings or not.
5583  */
5584 void
5585 mac_client_set_rings(mac_client_handle_t mch, int rxrings, int txrings)
5586 {
5587 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
5588 	mac_resource_props_t	*mrp = MCIP_RESOURCE_PROPS(mcip);
5589 
5590 	if (rxrings != MAC_RXRINGS_DONTCARE) {
5591 		mrp->mrp_mask |= MRP_RX_RINGS;
5592 		mrp->mrp_nrxrings = rxrings;
5593 	}
5594 
5595 	if (txrings != MAC_TXRINGS_DONTCARE) {
5596 		mrp->mrp_mask |= MRP_TX_RINGS;
5597 		mrp->mrp_ntxrings = txrings;
5598 	}
5599 }
5600 
5601 boolean_t
5602 mac_get_promisc_filtered(mac_client_handle_t mch)
5603 {
5604 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
5605 
5606 	return (mcip->mci_protect_flags & MPT_FLAG_PROMISC_FILTERED);
5607 }
5608 
5609 void
5610 mac_set_promisc_filtered(mac_client_handle_t mch, boolean_t enable)
5611 {
5612 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
5613 
5614 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
5615 	if (enable)
5616 		mcip->mci_protect_flags |= MPT_FLAG_PROMISC_FILTERED;
5617 	else
5618 		mcip->mci_protect_flags &= ~MPT_FLAG_PROMISC_FILTERED;
5619 }
5620