xref: /titanic_44/usr/src/uts/common/io/mac/mac_client.c (revision 5fd03bc0f2e00e7ba02316c2e08f45d52aab15db)
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 (c) 2014, Joyent, Inc.  All rights reserved.
25  */
26 
27 /*
28  * - General Introduction:
29  *
30  * This file contains the implementation of the MAC client kernel
31  * API and related code. The MAC client API allows a kernel module
32  * to gain access to a MAC instance (physical NIC, link aggregation, etc).
33  * It allows a MAC client to associate itself with a MAC address,
34  * VLANs, callback functions for data traffic and for promiscuous mode.
35  * The MAC client API is also used to specify the properties associated
36  * with a MAC client, such as bandwidth limits, priority, CPUS, etc.
37  * These properties are further used to determine the hardware resources
38  * to allocate to the various MAC clients.
39  *
40  * - Primary MAC clients:
41  *
42  * The MAC client API refers to "primary MAC clients". A primary MAC
43  * client is a client which "owns" the primary MAC address of
44  * the underlying MAC instance. The primary MAC address is called out
45  * since it is associated with specific semantics: the primary MAC
46  * address is the MAC address which is assigned to the IP interface
47  * when it is plumbed, and the primary MAC address is assigned
48  * to VLAN data-links. The primary address of a MAC instance can
49  * also change dynamically from under the MAC client, for example
50  * as a result of a change of state of a link aggregation. In that
51  * case the MAC layer automatically updates all data-structures which
52  * refer to the current value of the primary MAC address. Typical
53  * primary MAC clients are dls, aggr, and xnb. A typical non-primary
54  * MAC client is the vnic driver.
55  *
56  * - Virtual Switching:
57  *
58  * The MAC layer implements a virtual switch between the MAC clients
59  * (primary and non-primary) defined on top of the same underlying
60  * NIC (physical, link aggregation, etc). The virtual switch is
61  * VLAN-aware, i.e. it allows multiple MAC clients to be member
62  * of one or more VLANs, and the virtual switch will distribute
63  * multicast tagged packets only to the member of the corresponding
64  * VLANs.
65  *
66  * - Upper vs Lower MAC:
67  *
68  * Creating a VNIC on top of a MAC instance effectively causes
69  * two MAC instances to be layered on top of each other, one for
70  * the VNIC(s), one for the underlying MAC instance (physical NIC,
71  * link aggregation, etc). In the code below we refer to the
72  * underlying NIC as the "lower MAC", and we refer to VNICs as
73  * the "upper MAC".
74  *
75  * - Pass-through for VNICs:
76  *
77  * When VNICs are created on top of an underlying MAC, this causes
78  * a layering of two MAC instances. Since the lower MAC already
79  * does the switching and demultiplexing to its MAC clients, the
80  * upper MAC would simply have to pass packets to the layer below
81  * or above it, which would introduce overhead. In order to avoid
82  * this overhead, the MAC layer implements a pass-through mechanism
83  * for VNICs. When a VNIC opens the lower MAC instance, it saves
84  * the MAC client handle it optains from the MAC layer. When a MAC
85  * client opens a VNIC (upper MAC), the MAC layer detects that
86  * the MAC being opened is a VNIC, and gets the MAC client handle
87  * that the VNIC driver obtained from the lower MAC. This exchange
88  * is done through a private capability between the MAC layer
89  * and the VNIC driver. The upper MAC then returns that handle
90  * directly to its MAC client. Any operation done by the upper
91  * MAC client is now done on the lower MAC client handle, which
92  * allows the VNIC driver to be completely bypassed for the
93  * performance sensitive data-path.
94  *
95  * - Secondary MACs for VNICs:
96  *
97  * VNICs support multiple upper mac clients to enable support for
98  * multiple MAC addresses on the VNIC. When the VNIC is created the
99  * initial mac client is the primary upper mac. Any additional mac
100  * clients are secondary macs. These are kept in sync with the primary
101  * (for things such as the rx function and resource control settings)
102  * using the same private capability interface between the MAC layer
103  * and the VNIC layer.
104  *
105  */
106 
107 #include <sys/types.h>
108 #include <sys/conf.h>
109 #include <sys/id_space.h>
110 #include <sys/esunddi.h>
111 #include <sys/stat.h>
112 #include <sys/mkdev.h>
113 #include <sys/stream.h>
114 #include <sys/strsun.h>
115 #include <sys/strsubr.h>
116 #include <sys/dlpi.h>
117 #include <sys/modhash.h>
118 #include <sys/mac_impl.h>
119 #include <sys/mac_client_impl.h>
120 #include <sys/mac_soft_ring.h>
121 #include <sys/mac_stat.h>
122 #include <sys/dls.h>
123 #include <sys/dld.h>
124 #include <sys/modctl.h>
125 #include <sys/fs/dv_node.h>
126 #include <sys/thread.h>
127 #include <sys/proc.h>
128 #include <sys/callb.h>
129 #include <sys/cpuvar.h>
130 #include <sys/atomic.h>
131 #include <sys/sdt.h>
132 #include <sys/mac_flow.h>
133 #include <sys/ddi_intr_impl.h>
134 #include <sys/disp.h>
135 #include <sys/sdt.h>
136 #include <sys/vnic.h>
137 #include <sys/vnic_impl.h>
138 #include <sys/vlan.h>
139 #include <inet/ip.h>
140 #include <inet/ip6.h>
141 #include <sys/exacct.h>
142 #include <sys/exacct_impl.h>
143 #include <inet/nd.h>
144 #include <sys/ethernet.h>
145 
146 kmem_cache_t	*mac_client_impl_cache;
147 kmem_cache_t	*mac_promisc_impl_cache;
148 
149 static boolean_t mac_client_single_rcvr(mac_client_impl_t *);
150 static flow_entry_t *mac_client_swap_mciflent(mac_client_impl_t *);
151 static flow_entry_t *mac_client_get_flow(mac_client_impl_t *,
152     mac_unicast_impl_t *);
153 static void mac_client_remove_flow_from_list(mac_client_impl_t *,
154     flow_entry_t *);
155 static void mac_client_add_to_flow_list(mac_client_impl_t *, flow_entry_t *);
156 static void mac_rename_flow_names(mac_client_impl_t *, const char *);
157 static void mac_virtual_link_update(mac_impl_t *);
158 static int mac_client_datapath_setup(mac_client_impl_t *, uint16_t,
159     uint8_t *, mac_resource_props_t *, boolean_t, mac_unicast_impl_t *);
160 static void mac_client_datapath_teardown(mac_client_handle_t,
161     mac_unicast_impl_t *, flow_entry_t *);
162 static int mac_resource_ctl_set(mac_client_handle_t, mac_resource_props_t *);
163 
164 /* ARGSUSED */
165 static int
166 i_mac_client_impl_ctor(void *buf, void *arg, int kmflag)
167 {
168 	int	i;
169 	mac_client_impl_t	*mcip = buf;
170 
171 	bzero(buf, MAC_CLIENT_IMPL_SIZE);
172 	mutex_init(&mcip->mci_tx_cb_lock, NULL, MUTEX_DRIVER, NULL);
173 	mcip->mci_tx_notify_cb_info.mcbi_lockp = &mcip->mci_tx_cb_lock;
174 
175 	ASSERT(mac_tx_percpu_cnt >= 0);
176 	for (i = 0; i <= mac_tx_percpu_cnt; i++) {
177 		mutex_init(&mcip->mci_tx_pcpu[i].pcpu_tx_lock, NULL,
178 		    MUTEX_DRIVER, NULL);
179 	}
180 	cv_init(&mcip->mci_tx_cv, NULL, CV_DRIVER, NULL);
181 
182 	return (0);
183 }
184 
185 /* ARGSUSED */
186 static void
187 i_mac_client_impl_dtor(void *buf, void *arg)
188 {
189 	int	i;
190 	mac_client_impl_t *mcip = buf;
191 
192 	ASSERT(mcip->mci_promisc_list == NULL);
193 	ASSERT(mcip->mci_unicast_list == NULL);
194 	ASSERT(mcip->mci_state_flags == 0);
195 	ASSERT(mcip->mci_tx_flag == 0);
196 
197 	mutex_destroy(&mcip->mci_tx_cb_lock);
198 
199 	ASSERT(mac_tx_percpu_cnt >= 0);
200 	for (i = 0; i <= mac_tx_percpu_cnt; i++) {
201 		ASSERT(mcip->mci_tx_pcpu[i].pcpu_tx_refcnt == 0);
202 		mutex_destroy(&mcip->mci_tx_pcpu[i].pcpu_tx_lock);
203 	}
204 	cv_destroy(&mcip->mci_tx_cv);
205 }
206 
207 /* ARGSUSED */
208 static int
209 i_mac_promisc_impl_ctor(void *buf, void *arg, int kmflag)
210 {
211 	mac_promisc_impl_t	*mpip = buf;
212 
213 	bzero(buf, sizeof (mac_promisc_impl_t));
214 	mpip->mpi_mci_link.mcb_objp = buf;
215 	mpip->mpi_mci_link.mcb_objsize = sizeof (mac_promisc_impl_t);
216 	mpip->mpi_mi_link.mcb_objp = buf;
217 	mpip->mpi_mi_link.mcb_objsize = sizeof (mac_promisc_impl_t);
218 	return (0);
219 }
220 
221 /* ARGSUSED */
222 static void
223 i_mac_promisc_impl_dtor(void *buf, void *arg)
224 {
225 	mac_promisc_impl_t	*mpip = buf;
226 
227 	ASSERT(mpip->mpi_mci_link.mcb_objp != NULL);
228 	ASSERT(mpip->mpi_mci_link.mcb_objsize == sizeof (mac_promisc_impl_t));
229 	ASSERT(mpip->mpi_mi_link.mcb_objp == mpip->mpi_mci_link.mcb_objp);
230 	ASSERT(mpip->mpi_mi_link.mcb_objsize == sizeof (mac_promisc_impl_t));
231 
232 	mpip->mpi_mci_link.mcb_objp = NULL;
233 	mpip->mpi_mci_link.mcb_objsize = 0;
234 	mpip->mpi_mi_link.mcb_objp = NULL;
235 	mpip->mpi_mi_link.mcb_objsize = 0;
236 
237 	ASSERT(mpip->mpi_mci_link.mcb_flags == 0);
238 	mpip->mpi_mci_link.mcb_objsize = 0;
239 }
240 
241 void
242 mac_client_init(void)
243 {
244 	ASSERT(mac_tx_percpu_cnt >= 0);
245 
246 	mac_client_impl_cache = kmem_cache_create("mac_client_impl_cache",
247 	    MAC_CLIENT_IMPL_SIZE, 0, i_mac_client_impl_ctor,
248 	    i_mac_client_impl_dtor, NULL, NULL, NULL, 0);
249 	ASSERT(mac_client_impl_cache != NULL);
250 
251 	mac_promisc_impl_cache = kmem_cache_create("mac_promisc_impl_cache",
252 	    sizeof (mac_promisc_impl_t), 0, i_mac_promisc_impl_ctor,
253 	    i_mac_promisc_impl_dtor, NULL, NULL, NULL, 0);
254 	ASSERT(mac_promisc_impl_cache != NULL);
255 }
256 
257 void
258 mac_client_fini(void)
259 {
260 	kmem_cache_destroy(mac_client_impl_cache);
261 	kmem_cache_destroy(mac_promisc_impl_cache);
262 }
263 
264 /*
265  * Return the lower MAC client handle from the VNIC driver for the
266  * specified VNIC MAC instance.
267  */
268 mac_client_impl_t *
269 mac_vnic_lower(mac_impl_t *mip)
270 {
271 	mac_capab_vnic_t cap;
272 	mac_client_impl_t *mcip;
273 
274 	VERIFY(i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_VNIC, &cap));
275 	mcip = cap.mcv_mac_client_handle(cap.mcv_arg);
276 
277 	return (mcip);
278 }
279 
280 /*
281  * Update the secondary macs
282  */
283 void
284 mac_vnic_secondary_update(mac_impl_t *mip)
285 {
286 	mac_capab_vnic_t cap;
287 
288 	VERIFY(i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_VNIC, &cap));
289 	cap.mcv_mac_secondary_update(cap.mcv_arg);
290 }
291 
292 /*
293  * Return the MAC client handle of the primary MAC client for the
294  * specified MAC instance, or NULL otherwise.
295  */
296 mac_client_impl_t *
297 mac_primary_client_handle(mac_impl_t *mip)
298 {
299 	mac_client_impl_t *mcip;
300 
301 	if (mip->mi_state_flags & MIS_IS_VNIC)
302 		return (mac_vnic_lower(mip));
303 
304 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
305 
306 	for (mcip = mip->mi_clients_list; mcip != NULL;
307 	    mcip = mcip->mci_client_next) {
308 		if (MCIP_DATAPATH_SETUP(mcip) && mac_is_primary_client(mcip))
309 			return (mcip);
310 	}
311 	return (NULL);
312 }
313 
314 /*
315  * Open a MAC specified by its MAC name.
316  */
317 int
318 mac_open(const char *macname, mac_handle_t *mhp)
319 {
320 	mac_impl_t	*mip;
321 	int		err;
322 
323 	/*
324 	 * Look up its entry in the global hash table.
325 	 */
326 	if ((err = mac_hold(macname, &mip)) != 0)
327 		return (err);
328 
329 	/*
330 	 * Hold the dip associated to the MAC to prevent it from being
331 	 * detached. For a softmac, its underlying dip is held by the
332 	 * mi_open() callback.
333 	 *
334 	 * This is done to be more tolerant with some defective drivers,
335 	 * which incorrectly handle mac_unregister() failure in their
336 	 * xxx_detach() routine. For example, some drivers ignore the
337 	 * failure of mac_unregister() and free all resources that
338 	 * that are needed for data transmition.
339 	 */
340 	e_ddi_hold_devi(mip->mi_dip);
341 
342 	if (!(mip->mi_callbacks->mc_callbacks & MC_OPEN)) {
343 		*mhp = (mac_handle_t)mip;
344 		return (0);
345 	}
346 
347 	/*
348 	 * The mac perimeter is used in both mac_open and mac_close by the
349 	 * framework to single thread the MC_OPEN/MC_CLOSE of drivers.
350 	 */
351 	i_mac_perim_enter(mip);
352 	mip->mi_oref++;
353 	if (mip->mi_oref != 1 || ((err = mip->mi_open(mip->mi_driver)) == 0)) {
354 		*mhp = (mac_handle_t)mip;
355 		i_mac_perim_exit(mip);
356 		return (0);
357 	}
358 	mip->mi_oref--;
359 	ddi_release_devi(mip->mi_dip);
360 	mac_rele(mip);
361 	i_mac_perim_exit(mip);
362 	return (err);
363 }
364 
365 /*
366  * Open a MAC specified by its linkid.
367  */
368 int
369 mac_open_by_linkid(datalink_id_t linkid, mac_handle_t *mhp)
370 {
371 	dls_dl_handle_t	dlh;
372 	int		err;
373 
374 	if ((err = dls_devnet_hold_tmp(linkid, &dlh)) != 0)
375 		return (err);
376 
377 	dls_devnet_prop_task_wait(dlh);
378 
379 	err = mac_open(dls_devnet_mac(dlh), mhp);
380 
381 	dls_devnet_rele_tmp(dlh);
382 	return (err);
383 }
384 
385 /*
386  * Open a MAC specified by its link name.
387  */
388 int
389 mac_open_by_linkname(const char *link, mac_handle_t *mhp)
390 {
391 	datalink_id_t	linkid;
392 	int		err;
393 
394 	if ((err = dls_mgmt_get_linkid(link, &linkid)) != 0)
395 		return (err);
396 	return (mac_open_by_linkid(linkid, mhp));
397 }
398 
399 /*
400  * Close the specified MAC.
401  */
402 void
403 mac_close(mac_handle_t mh)
404 {
405 	mac_impl_t	*mip = (mac_impl_t *)mh;
406 
407 	i_mac_perim_enter(mip);
408 	/*
409 	 * The mac perimeter is used in both mac_open and mac_close by the
410 	 * framework to single thread the MC_OPEN/MC_CLOSE of drivers.
411 	 */
412 	if (mip->mi_callbacks->mc_callbacks & MC_OPEN) {
413 		ASSERT(mip->mi_oref != 0);
414 		if (--mip->mi_oref == 0) {
415 			if ((mip->mi_callbacks->mc_callbacks & MC_CLOSE))
416 				mip->mi_close(mip->mi_driver);
417 		}
418 	}
419 	i_mac_perim_exit(mip);
420 	ddi_release_devi(mip->mi_dip);
421 	mac_rele(mip);
422 }
423 
424 /*
425  * Misc utility functions to retrieve various information about a MAC
426  * instance or a MAC client.
427  */
428 
429 const mac_info_t *
430 mac_info(mac_handle_t mh)
431 {
432 	return (&((mac_impl_t *)mh)->mi_info);
433 }
434 
435 dev_info_t *
436 mac_devinfo_get(mac_handle_t mh)
437 {
438 	return (((mac_impl_t *)mh)->mi_dip);
439 }
440 
441 void *
442 mac_driver(mac_handle_t mh)
443 {
444 	return (((mac_impl_t *)mh)->mi_driver);
445 }
446 
447 const char *
448 mac_name(mac_handle_t mh)
449 {
450 	return (((mac_impl_t *)mh)->mi_name);
451 }
452 
453 int
454 mac_type(mac_handle_t mh)
455 {
456 	return (((mac_impl_t *)mh)->mi_type->mt_type);
457 }
458 
459 int
460 mac_nativetype(mac_handle_t mh)
461 {
462 	return (((mac_impl_t *)mh)->mi_type->mt_nativetype);
463 }
464 
465 char *
466 mac_client_name(mac_client_handle_t mch)
467 {
468 	return (((mac_client_impl_t *)mch)->mci_name);
469 }
470 
471 minor_t
472 mac_minor(mac_handle_t mh)
473 {
474 	return (((mac_impl_t *)mh)->mi_minor);
475 }
476 
477 /*
478  * Return the VID associated with a MAC client. This function should
479  * be called for clients which are associated with only one VID.
480  */
481 uint16_t
482 mac_client_vid(mac_client_handle_t mch)
483 {
484 	uint16_t		vid = VLAN_ID_NONE;
485 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
486 	flow_desc_t		flow_desc;
487 
488 	if (mcip->mci_nflents == 0)
489 		return (vid);
490 
491 	ASSERT(MCIP_DATAPATH_SETUP(mcip) && mac_client_single_rcvr(mcip));
492 
493 	mac_flow_get_desc(mcip->mci_flent, &flow_desc);
494 	if ((flow_desc.fd_mask & FLOW_LINK_VID) != 0)
495 		vid = flow_desc.fd_vid;
496 
497 	return (vid);
498 }
499 
500 /*
501  * Return whether the specified MAC client corresponds to a VLAN VNIC.
502  */
503 boolean_t
504 mac_client_is_vlan_vnic(mac_client_handle_t mch)
505 {
506 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
507 
508 	return (((mcip->mci_state_flags & MCIS_IS_VNIC) != 0) &&
509 	    ((mcip->mci_flent->fe_type & FLOW_PRIMARY_MAC) != 0));
510 }
511 
512 /*
513  * Return the link speed associated with the specified MAC client.
514  *
515  * The link speed of a MAC client is equal to the smallest value of
516  * 1) the current link speed of the underlying NIC, or
517  * 2) the bandwidth limit set for the MAC client.
518  *
519  * Note that the bandwidth limit can be higher than the speed
520  * of the underlying NIC. This is allowed to avoid spurious
521  * administration action failures or artifically lowering the
522  * bandwidth limit of a link that may  have temporarily lowered
523  * its link speed due to hardware problem or administrator action.
524  */
525 static uint64_t
526 mac_client_ifspeed(mac_client_impl_t *mcip)
527 {
528 	mac_impl_t *mip = mcip->mci_mip;
529 	uint64_t nic_speed;
530 
531 	nic_speed = mac_stat_get((mac_handle_t)mip, MAC_STAT_IFSPEED);
532 
533 	if (nic_speed == 0) {
534 		return (0);
535 	} else {
536 		uint64_t policy_limit = (uint64_t)-1;
537 
538 		if (MCIP_RESOURCE_PROPS_MASK(mcip) & MRP_MAXBW)
539 			policy_limit = MCIP_RESOURCE_PROPS_MAXBW(mcip);
540 
541 		return (MIN(policy_limit, nic_speed));
542 	}
543 }
544 
545 /*
546  * Return the link state of the specified client. If here are more
547  * than one clients of the underying mac_impl_t, the link state
548  * will always be UP regardless of the link state of the underlying
549  * mac_impl_t. This is needed to allow the MAC clients to continue
550  * to communicate with each other even when the physical link of
551  * their mac_impl_t is down.
552  */
553 static uint64_t
554 mac_client_link_state(mac_client_impl_t *mcip)
555 {
556 	mac_impl_t *mip = mcip->mci_mip;
557 	uint16_t vid;
558 	mac_client_impl_t *mci_list;
559 	mac_unicast_impl_t *mui_list, *oth_mui_list;
560 
561 	/*
562 	 * Returns LINK_STATE_UP if there are other MAC clients defined on
563 	 * mac_impl_t which share same VLAN ID as that of mcip. Note that
564 	 * if 'mcip' has more than one VID's then we match ANY one of the
565 	 * VID's with other MAC client's VID's and return LINK_STATE_UP.
566 	 */
567 	rw_enter(&mcip->mci_rw_lock, RW_READER);
568 	for (mui_list = mcip->mci_unicast_list; mui_list != NULL;
569 	    mui_list = mui_list->mui_next) {
570 		vid = mui_list->mui_vid;
571 		for (mci_list = mip->mi_clients_list; mci_list != NULL;
572 		    mci_list = mci_list->mci_client_next) {
573 			if (mci_list == mcip)
574 				continue;
575 			for (oth_mui_list = mci_list->mci_unicast_list;
576 			    oth_mui_list != NULL; oth_mui_list = oth_mui_list->
577 			    mui_next) {
578 				if (vid == oth_mui_list->mui_vid) {
579 					rw_exit(&mcip->mci_rw_lock);
580 					return (LINK_STATE_UP);
581 				}
582 			}
583 		}
584 	}
585 	rw_exit(&mcip->mci_rw_lock);
586 
587 	return (mac_stat_get((mac_handle_t)mip, MAC_STAT_LINK_STATE));
588 }
589 
590 /*
591  * These statistics are consumed by dladm show-link -s <vnic>,
592  * dladm show-vnic -s and netstat. With the introduction of dlstat,
593  * dladm show-link -s and dladm show-vnic -s witll be EOL'ed while
594  * netstat will consume from kstats introduced for dlstat. This code
595  * will be removed at that time.
596  */
597 
598 /*
599  * Return the statistics of a MAC client. These statistics are different
600  * then the statistics of the underlying MAC which are returned by
601  * mac_stat_get().
602  */
603 uint64_t
604 mac_client_stat_get(mac_client_handle_t mch, uint_t stat)
605 {
606 	mac_client_impl_t 	*mcip = (mac_client_impl_t *)mch;
607 	mac_impl_t 		*mip = mcip->mci_mip;
608 	flow_entry_t 		*flent = mcip->mci_flent;
609 	mac_soft_ring_set_t 	*mac_srs;
610 	mac_rx_stats_t		*mac_rx_stat;
611 	mac_tx_stats_t		*mac_tx_stat;
612 	int i;
613 	uint64_t val = 0;
614 
615 	mac_srs = (mac_soft_ring_set_t *)(flent->fe_tx_srs);
616 	mac_tx_stat = &mac_srs->srs_tx.st_stat;
617 
618 	switch (stat) {
619 	case MAC_STAT_LINK_STATE:
620 		val = mac_client_link_state(mcip);
621 		break;
622 	case MAC_STAT_LINK_UP:
623 		val = (mac_client_link_state(mcip) == LINK_STATE_UP);
624 		break;
625 	case MAC_STAT_PROMISC:
626 		val = mac_stat_get((mac_handle_t)mip, MAC_STAT_PROMISC);
627 		break;
628 	case MAC_STAT_LOWLINK_STATE:
629 		val = mac_stat_get((mac_handle_t)mip, MAC_STAT_LOWLINK_STATE);
630 		break;
631 	case MAC_STAT_IFSPEED:
632 		val = mac_client_ifspeed(mcip);
633 		break;
634 	case MAC_STAT_MULTIRCV:
635 		val = mcip->mci_misc_stat.mms_multircv;
636 		break;
637 	case MAC_STAT_BRDCSTRCV:
638 		val = mcip->mci_misc_stat.mms_brdcstrcv;
639 		break;
640 	case MAC_STAT_MULTIXMT:
641 		val = mcip->mci_misc_stat.mms_multixmt;
642 		break;
643 	case MAC_STAT_BRDCSTXMT:
644 		val = mcip->mci_misc_stat.mms_brdcstxmt;
645 		break;
646 	case MAC_STAT_OBYTES:
647 		val = mac_tx_stat->mts_obytes;
648 		break;
649 	case MAC_STAT_OPACKETS:
650 		val = mac_tx_stat->mts_opackets;
651 		break;
652 	case MAC_STAT_OERRORS:
653 		val = mac_tx_stat->mts_oerrors;
654 		break;
655 	case MAC_STAT_IPACKETS:
656 		for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
657 			mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
658 			mac_rx_stat = &mac_srs->srs_rx.sr_stat;
659 			val += mac_rx_stat->mrs_intrcnt +
660 			    mac_rx_stat->mrs_pollcnt + mac_rx_stat->mrs_lclcnt;
661 		}
662 		break;
663 	case MAC_STAT_RBYTES:
664 		for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
665 			mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
666 			mac_rx_stat = &mac_srs->srs_rx.sr_stat;
667 			val += mac_rx_stat->mrs_intrbytes +
668 			    mac_rx_stat->mrs_pollbytes +
669 			    mac_rx_stat->mrs_lclbytes;
670 		}
671 		break;
672 	case MAC_STAT_IERRORS:
673 		for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
674 			mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
675 			mac_rx_stat = &mac_srs->srs_rx.sr_stat;
676 			val += mac_rx_stat->mrs_ierrors;
677 		}
678 		break;
679 	default:
680 		val = mac_driver_stat_default(mip, stat);
681 		break;
682 	}
683 
684 	return (val);
685 }
686 
687 /*
688  * Return the statistics of the specified MAC instance.
689  */
690 uint64_t
691 mac_stat_get(mac_handle_t mh, uint_t stat)
692 {
693 	mac_impl_t	*mip = (mac_impl_t *)mh;
694 	uint64_t	val;
695 	int		ret;
696 
697 	/*
698 	 * The range of stat determines where it is maintained.  Stat
699 	 * values from 0 up to (but not including) MAC_STAT_MIN are
700 	 * mainteined by the mac module itself.  Everything else is
701 	 * maintained by the driver.
702 	 *
703 	 * If the mac_impl_t being queried corresponds to a VNIC,
704 	 * the stats need to be queried from the lower MAC client
705 	 * corresponding to the VNIC. (The mac_link_update()
706 	 * invoked by the driver to the lower MAC causes the *lower
707 	 * MAC* to update its mi_linkstate, and send a notification
708 	 * to its MAC clients. Due to the VNIC passthrough,
709 	 * these notifications are sent to the upper MAC clients
710 	 * of the VNIC directly, and the upper mac_impl_t of the VNIC
711 	 * does not have a valid mi_linkstate.
712 	 */
713 	if (stat < MAC_STAT_MIN && !(mip->mi_state_flags & MIS_IS_VNIC)) {
714 		/* these stats are maintained by the mac module itself */
715 		switch (stat) {
716 		case MAC_STAT_LINK_STATE:
717 			return (mip->mi_linkstate);
718 		case MAC_STAT_LINK_UP:
719 			return (mip->mi_linkstate == LINK_STATE_UP);
720 		case MAC_STAT_PROMISC:
721 			return (mip->mi_devpromisc != 0);
722 		case MAC_STAT_LOWLINK_STATE:
723 			return (mip->mi_lowlinkstate);
724 		default:
725 			ASSERT(B_FALSE);
726 		}
727 	}
728 
729 	/*
730 	 * Call the driver to get the given statistic.
731 	 */
732 	ret = mip->mi_getstat(mip->mi_driver, stat, &val);
733 	if (ret != 0) {
734 		/*
735 		 * The driver doesn't support this statistic.  Get the
736 		 * statistic's default value.
737 		 */
738 		val = mac_driver_stat_default(mip, stat);
739 	}
740 	return (val);
741 }
742 
743 /*
744  * Query hardware rx ring corresponding to the pseudo ring.
745  */
746 uint64_t
747 mac_pseudo_rx_ring_stat_get(mac_ring_handle_t handle, uint_t stat)
748 {
749 	return (mac_rx_ring_stat_get(handle, stat));
750 }
751 
752 /*
753  * Query hardware tx ring corresponding to the pseudo ring.
754  */
755 uint64_t
756 mac_pseudo_tx_ring_stat_get(mac_ring_handle_t handle, uint_t stat)
757 {
758 	return (mac_tx_ring_stat_get(handle, stat));
759 }
760 
761 /*
762  * Utility function which returns the VID associated with a flow entry.
763  */
764 uint16_t
765 i_mac_flow_vid(flow_entry_t *flent)
766 {
767 	flow_desc_t	flow_desc;
768 
769 	mac_flow_get_desc(flent, &flow_desc);
770 
771 	if ((flow_desc.fd_mask & FLOW_LINK_VID) != 0)
772 		return (flow_desc.fd_vid);
773 	return (VLAN_ID_NONE);
774 }
775 
776 /*
777  * Verify the validity of the specified unicast MAC address. Returns B_TRUE
778  * if the address is valid, B_FALSE otherwise (multicast address, or incorrect
779  * length.
780  */
781 boolean_t
782 mac_unicst_verify(mac_handle_t mh, const uint8_t *addr, uint_t len)
783 {
784 	mac_impl_t	*mip = (mac_impl_t *)mh;
785 
786 	/*
787 	 * Verify the address. No lock is needed since mi_type and plugin
788 	 * details don't change after mac_register().
789 	 */
790 	if ((len != mip->mi_type->mt_addr_length) ||
791 	    (mip->mi_type->mt_ops.mtops_unicst_verify(addr,
792 	    mip->mi_pdata)) != 0) {
793 		return (B_FALSE);
794 	} else {
795 		return (B_TRUE);
796 	}
797 }
798 
799 void
800 mac_sdu_get(mac_handle_t mh, uint_t *min_sdu, uint_t *max_sdu)
801 {
802 	mac_impl_t	*mip = (mac_impl_t *)mh;
803 
804 	if (min_sdu != NULL)
805 		*min_sdu = mip->mi_sdu_min;
806 	if (max_sdu != NULL)
807 		*max_sdu = mip->mi_sdu_max;
808 }
809 
810 void
811 mac_sdu_get2(mac_handle_t mh, uint_t *min_sdu, uint_t *max_sdu,
812     uint_t *multicast_sdu)
813 {
814 	mac_impl_t	*mip = (mac_impl_t *)mh;
815 
816 	if (min_sdu != NULL)
817 		*min_sdu = mip->mi_sdu_min;
818 	if (max_sdu != NULL)
819 		*max_sdu = mip->mi_sdu_max;
820 	if (multicast_sdu != NULL)
821 		*multicast_sdu = mip->mi_sdu_multicast;
822 }
823 
824 /*
825  * Update the MAC unicast address of the specified client's flows. Currently
826  * only one unicast MAC unicast address is allowed per client.
827  */
828 static void
829 mac_unicast_update_client_flow(mac_client_impl_t *mcip)
830 {
831 	mac_impl_t *mip = mcip->mci_mip;
832 	flow_entry_t *flent = mcip->mci_flent;
833 	mac_address_t *map = mcip->mci_unicast;
834 	flow_desc_t flow_desc;
835 
836 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
837 	ASSERT(flent != NULL);
838 
839 	mac_flow_get_desc(flent, &flow_desc);
840 	ASSERT(flow_desc.fd_mask & FLOW_LINK_DST);
841 
842 	bcopy(map->ma_addr, flow_desc.fd_dst_mac, map->ma_len);
843 	mac_flow_set_desc(flent, &flow_desc);
844 
845 	/*
846 	 * The v6 local addr (used by mac protection) needs to be
847 	 * regenerated because our mac address has changed.
848 	 */
849 	mac_protect_update_v6_local_addr(mcip);
850 
851 	/*
852 	 * A MAC client could have one MAC address but multiple
853 	 * VLANs. In that case update the flow entries corresponding
854 	 * to all VLANs of the MAC client.
855 	 */
856 	for (flent = mcip->mci_flent_list; flent != NULL;
857 	    flent = flent->fe_client_next) {
858 		mac_flow_get_desc(flent, &flow_desc);
859 		if (!(flent->fe_type & FLOW_PRIMARY_MAC ||
860 		    flent->fe_type & FLOW_VNIC_MAC))
861 			continue;
862 
863 		bcopy(map->ma_addr, flow_desc.fd_dst_mac, map->ma_len);
864 		mac_flow_set_desc(flent, &flow_desc);
865 	}
866 }
867 
868 /*
869  * Update all clients that share the same unicast address.
870  */
871 void
872 mac_unicast_update_clients(mac_impl_t *mip, mac_address_t *map)
873 {
874 	mac_client_impl_t *mcip;
875 
876 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
877 
878 	/*
879 	 * Find all clients that share the same unicast MAC address and update
880 	 * them appropriately.
881 	 */
882 	for (mcip = mip->mi_clients_list; mcip != NULL;
883 	    mcip = mcip->mci_client_next) {
884 		/*
885 		 * Ignore clients that don't share this MAC address.
886 		 */
887 		if (map != mcip->mci_unicast)
888 			continue;
889 
890 		/*
891 		 * Update those clients with same old unicast MAC address.
892 		 */
893 		mac_unicast_update_client_flow(mcip);
894 	}
895 }
896 
897 /*
898  * Update the unicast MAC address of the specified VNIC MAC client.
899  *
900  * Check whether the operation is valid. Any of following cases should fail:
901  *
902  * 1. It's a VLAN type of VNIC.
903  * 2. The new value is current "primary" MAC address.
904  * 3. The current MAC address is shared with other clients.
905  * 4. The new MAC address has been used. This case will be valid when
906  *    client migration is fully supported.
907  */
908 int
909 mac_vnic_unicast_set(mac_client_handle_t mch, const uint8_t *addr)
910 {
911 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
912 	mac_impl_t *mip = mcip->mci_mip;
913 	mac_address_t *map = mcip->mci_unicast;
914 	int err;
915 
916 	ASSERT(!(mip->mi_state_flags & MIS_IS_VNIC));
917 	ASSERT(mcip->mci_state_flags & MCIS_IS_VNIC);
918 	ASSERT(mcip->mci_flags != MAC_CLIENT_FLAGS_PRIMARY);
919 
920 	i_mac_perim_enter(mip);
921 
922 	/*
923 	 * If this is a VLAN type of VNIC, it's using "primary" MAC address
924 	 * of the underlying interface. Must fail here. Refer to case 1 above.
925 	 */
926 	if (bcmp(map->ma_addr, mip->mi_addr, map->ma_len) == 0) {
927 		i_mac_perim_exit(mip);
928 		return (ENOTSUP);
929 	}
930 
931 	/*
932 	 * If the new address is the "primary" one, must fail. Refer to
933 	 * case 2 above.
934 	 */
935 	if (bcmp(addr, mip->mi_addr, map->ma_len) == 0) {
936 		i_mac_perim_exit(mip);
937 		return (EACCES);
938 	}
939 
940 	/*
941 	 * If the address is shared by multiple clients, must fail. Refer
942 	 * to case 3 above.
943 	 */
944 	if (mac_check_macaddr_shared(map)) {
945 		i_mac_perim_exit(mip);
946 		return (EBUSY);
947 	}
948 
949 	/*
950 	 * If the new address has been used, must fail for now. Refer to
951 	 * case 4 above.
952 	 */
953 	if (mac_find_macaddr(mip, (uint8_t *)addr) != NULL) {
954 		i_mac_perim_exit(mip);
955 		return (ENOTSUP);
956 	}
957 
958 	/*
959 	 * Update the MAC address.
960 	 */
961 	err = mac_update_macaddr(map, (uint8_t *)addr);
962 
963 	if (err != 0) {
964 		i_mac_perim_exit(mip);
965 		return (err);
966 	}
967 
968 	/*
969 	 * Update all flows of this MAC client.
970 	 */
971 	mac_unicast_update_client_flow(mcip);
972 
973 	i_mac_perim_exit(mip);
974 	return (0);
975 }
976 
977 /*
978  * Program the new primary unicast address of the specified MAC.
979  *
980  * Function mac_update_macaddr() takes care different types of underlying
981  * MAC. If the underlying MAC is VNIC, the VNIC driver must have registerd
982  * mi_unicst() entry point, that indirectly calls mac_vnic_unicast_set()
983  * which will take care of updating the MAC address of the corresponding
984  * MAC client.
985  *
986  * This is the only interface that allow the client to update the "primary"
987  * MAC address of the underlying MAC. The new value must have not been
988  * used by other clients.
989  */
990 int
991 mac_unicast_primary_set(mac_handle_t mh, const uint8_t *addr)
992 {
993 	mac_impl_t *mip = (mac_impl_t *)mh;
994 	mac_address_t *map;
995 	int err;
996 
997 	/* verify the address validity */
998 	if (!mac_unicst_verify(mh, addr, mip->mi_type->mt_addr_length))
999 		return (EINVAL);
1000 
1001 	i_mac_perim_enter(mip);
1002 
1003 	/*
1004 	 * If the new value is the same as the current primary address value,
1005 	 * there's nothing to do.
1006 	 */
1007 	if (bcmp(addr, mip->mi_addr, mip->mi_type->mt_addr_length) == 0) {
1008 		i_mac_perim_exit(mip);
1009 		return (0);
1010 	}
1011 
1012 	if (mac_find_macaddr(mip, (uint8_t *)addr) != 0) {
1013 		i_mac_perim_exit(mip);
1014 		return (EBUSY);
1015 	}
1016 
1017 	map = mac_find_macaddr(mip, mip->mi_addr);
1018 	ASSERT(map != NULL);
1019 
1020 	/*
1021 	 * Update the MAC address.
1022 	 */
1023 	if (mip->mi_state_flags & MIS_IS_AGGR) {
1024 		mac_capab_aggr_t aggr_cap;
1025 
1026 		/*
1027 		 * If the mac is an aggregation, other than the unicast
1028 		 * addresses programming, aggr must be informed about this
1029 		 * primary unicst address change to change its mac address
1030 		 * policy to be user-specified.
1031 		 */
1032 		ASSERT(map->ma_type == MAC_ADDRESS_TYPE_UNICAST_CLASSIFIED);
1033 		VERIFY(i_mac_capab_get(mh, MAC_CAPAB_AGGR, &aggr_cap));
1034 		err = aggr_cap.mca_unicst(mip->mi_driver, addr);
1035 		if (err == 0)
1036 			bcopy(addr, map->ma_addr, map->ma_len);
1037 	} else {
1038 		err = mac_update_macaddr(map, (uint8_t *)addr);
1039 	}
1040 
1041 	if (err != 0) {
1042 		i_mac_perim_exit(mip);
1043 		return (err);
1044 	}
1045 
1046 	mac_unicast_update_clients(mip, map);
1047 
1048 	/*
1049 	 * Save the new primary MAC address in mac_impl_t.
1050 	 */
1051 	bcopy(addr, mip->mi_addr, mip->mi_type->mt_addr_length);
1052 
1053 	i_mac_perim_exit(mip);
1054 
1055 	if (err == 0)
1056 		i_mac_notify(mip, MAC_NOTE_UNICST);
1057 
1058 	return (err);
1059 }
1060 
1061 /*
1062  * Return the current primary MAC address of the specified MAC.
1063  */
1064 void
1065 mac_unicast_primary_get(mac_handle_t mh, uint8_t *addr)
1066 {
1067 	mac_impl_t *mip = (mac_impl_t *)mh;
1068 
1069 	rw_enter(&mip->mi_rw_lock, RW_READER);
1070 	bcopy(mip->mi_addr, addr, mip->mi_type->mt_addr_length);
1071 	rw_exit(&mip->mi_rw_lock);
1072 }
1073 
1074 /*
1075  * Return the secondary MAC address for the specified handle
1076  */
1077 void
1078 mac_unicast_secondary_get(mac_client_handle_t mh, uint8_t *addr)
1079 {
1080 	mac_client_impl_t *mcip = (mac_client_impl_t *)mh;
1081 
1082 	ASSERT(mcip->mci_unicast != NULL);
1083 	bcopy(mcip->mci_unicast->ma_addr, addr, mcip->mci_unicast->ma_len);
1084 }
1085 
1086 /*
1087  * Return information about the use of the primary MAC address of the
1088  * specified MAC instance:
1089  *
1090  * - if client_name is non-NULL, it must point to a string of at
1091  *   least MAXNAMELEN bytes, and will be set to the name of the MAC
1092  *   client which uses the primary MAC address.
1093  *
1094  * - if in_use is non-NULL, used to return whether the primary MAC
1095  *   address is currently in use.
1096  */
1097 void
1098 mac_unicast_primary_info(mac_handle_t mh, char *client_name, boolean_t *in_use)
1099 {
1100 	mac_impl_t *mip = (mac_impl_t *)mh;
1101 	mac_client_impl_t *cur_client;
1102 
1103 	if (in_use != NULL)
1104 		*in_use = B_FALSE;
1105 	if (client_name != NULL)
1106 		bzero(client_name, MAXNAMELEN);
1107 
1108 	/*
1109 	 * The mi_rw_lock is used to protect threads that don't hold the
1110 	 * mac perimeter to get a consistent view of the mi_clients_list.
1111 	 * Threads that modify the list must hold both the mac perimeter and
1112 	 * mi_rw_lock(RW_WRITER)
1113 	 */
1114 	rw_enter(&mip->mi_rw_lock, RW_READER);
1115 	for (cur_client = mip->mi_clients_list; cur_client != NULL;
1116 	    cur_client = cur_client->mci_client_next) {
1117 		if (mac_is_primary_client(cur_client) ||
1118 		    (mip->mi_state_flags & MIS_IS_VNIC)) {
1119 			rw_exit(&mip->mi_rw_lock);
1120 			if (in_use != NULL)
1121 				*in_use = B_TRUE;
1122 			if (client_name != NULL) {
1123 				bcopy(cur_client->mci_name, client_name,
1124 				    MAXNAMELEN);
1125 			}
1126 			return;
1127 		}
1128 	}
1129 	rw_exit(&mip->mi_rw_lock);
1130 }
1131 
1132 /*
1133  * Return the current destination MAC address of the specified MAC.
1134  */
1135 boolean_t
1136 mac_dst_get(mac_handle_t mh, uint8_t *addr)
1137 {
1138 	mac_impl_t *mip = (mac_impl_t *)mh;
1139 
1140 	rw_enter(&mip->mi_rw_lock, RW_READER);
1141 	if (mip->mi_dstaddr_set)
1142 		bcopy(mip->mi_dstaddr, addr, mip->mi_type->mt_addr_length);
1143 	rw_exit(&mip->mi_rw_lock);
1144 	return (mip->mi_dstaddr_set);
1145 }
1146 
1147 /*
1148  * Add the specified MAC client to the list of clients which opened
1149  * the specified MAC.
1150  */
1151 static void
1152 mac_client_add(mac_client_impl_t *mcip)
1153 {
1154 	mac_impl_t *mip = mcip->mci_mip;
1155 
1156 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1157 
1158 	/* add VNIC to the front of the list */
1159 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
1160 	mcip->mci_client_next = mip->mi_clients_list;
1161 	mip->mi_clients_list = mcip;
1162 	mip->mi_nclients++;
1163 	rw_exit(&mip->mi_rw_lock);
1164 }
1165 
1166 /*
1167  * Remove the specified MAC client from the list of clients which opened
1168  * the specified MAC.
1169  */
1170 static void
1171 mac_client_remove(mac_client_impl_t *mcip)
1172 {
1173 	mac_impl_t *mip = mcip->mci_mip;
1174 	mac_client_impl_t **prev, *cclient;
1175 
1176 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1177 
1178 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
1179 	prev = &mip->mi_clients_list;
1180 	cclient = *prev;
1181 	while (cclient != NULL && cclient != mcip) {
1182 		prev = &cclient->mci_client_next;
1183 		cclient = *prev;
1184 	}
1185 	ASSERT(cclient != NULL);
1186 	*prev = cclient->mci_client_next;
1187 	mip->mi_nclients--;
1188 	rw_exit(&mip->mi_rw_lock);
1189 }
1190 
1191 static mac_unicast_impl_t *
1192 mac_client_find_vid(mac_client_impl_t *mcip, uint16_t vid)
1193 {
1194 	mac_unicast_impl_t *muip = mcip->mci_unicast_list;
1195 
1196 	while ((muip != NULL) && (muip->mui_vid != vid))
1197 		muip = muip->mui_next;
1198 
1199 	return (muip);
1200 }
1201 
1202 /*
1203  * Return whether the specified (MAC address, VID) tuple is already used by
1204  * one of the MAC clients associated with the specified MAC.
1205  */
1206 static boolean_t
1207 mac_addr_in_use(mac_impl_t *mip, uint8_t *mac_addr, uint16_t vid)
1208 {
1209 	mac_client_impl_t *client;
1210 	mac_address_t *map;
1211 
1212 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1213 
1214 	for (client = mip->mi_clients_list; client != NULL;
1215 	    client = client->mci_client_next) {
1216 
1217 		/*
1218 		 * Ignore clients that don't have unicast address.
1219 		 */
1220 		if (client->mci_unicast_list == NULL)
1221 			continue;
1222 
1223 		map = client->mci_unicast;
1224 
1225 		if ((bcmp(mac_addr, map->ma_addr, map->ma_len) == 0) &&
1226 		    (mac_client_find_vid(client, vid) != NULL)) {
1227 			return (B_TRUE);
1228 		}
1229 	}
1230 
1231 	return (B_FALSE);
1232 }
1233 
1234 /*
1235  * Generate a random MAC address. The MAC address prefix is
1236  * stored in the array pointed to by mac_addr, and its length, in bytes,
1237  * is specified by prefix_len. The least significant bits
1238  * after prefix_len bytes are generated, and stored after the prefix
1239  * in the mac_addr array.
1240  */
1241 int
1242 mac_addr_random(mac_client_handle_t mch, uint_t prefix_len,
1243     uint8_t *mac_addr, mac_diag_t *diag)
1244 {
1245 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1246 	mac_impl_t *mip = mcip->mci_mip;
1247 	size_t addr_len = mip->mi_type->mt_addr_length;
1248 
1249 	if (prefix_len >= addr_len) {
1250 		*diag = MAC_DIAG_MACPREFIXLEN_INVALID;
1251 		return (EINVAL);
1252 	}
1253 
1254 	/* check the prefix value */
1255 	if (prefix_len > 0) {
1256 		bzero(mac_addr + prefix_len, addr_len - prefix_len);
1257 		if (!mac_unicst_verify((mac_handle_t)mip, mac_addr,
1258 		    addr_len)) {
1259 			*diag = MAC_DIAG_MACPREFIX_INVALID;
1260 			return (EINVAL);
1261 		}
1262 	}
1263 
1264 	/* generate the MAC address */
1265 	if (prefix_len < addr_len) {
1266 		(void) random_get_pseudo_bytes(mac_addr +
1267 		    prefix_len, addr_len - prefix_len);
1268 	}
1269 
1270 	*diag = 0;
1271 	return (0);
1272 }
1273 
1274 /*
1275  * Set the priority range for this MAC client. This will be used to
1276  * determine the absolute priority for the threads created for this
1277  * MAC client using the specified "low", "medium" and "high" level.
1278  * This will also be used for any subflows on this MAC client.
1279  */
1280 #define	MAC_CLIENT_SET_PRIORITY_RANGE(mcip, pri) {			\
1281 	(mcip)->mci_min_pri = FLOW_MIN_PRIORITY(MINCLSYSPRI,	\
1282 	    MAXCLSYSPRI, (pri));					\
1283 	(mcip)->mci_max_pri = FLOW_MAX_PRIORITY(MINCLSYSPRI,	\
1284 	    MAXCLSYSPRI, (mcip)->mci_min_pri);				\
1285 	}
1286 
1287 /*
1288  * MAC client open entry point. Return a new MAC client handle. Each
1289  * MAC client is associated with a name, specified through the 'name'
1290  * argument.
1291  */
1292 int
1293 mac_client_open(mac_handle_t mh, mac_client_handle_t *mchp, char *name,
1294     uint16_t flags)
1295 {
1296 	mac_impl_t		*mip = (mac_impl_t *)mh;
1297 	mac_client_impl_t	*mcip;
1298 	int			err = 0;
1299 	boolean_t		share_desired;
1300 	flow_entry_t		*flent = NULL;
1301 
1302 	share_desired = (flags & MAC_OPEN_FLAGS_SHARES_DESIRED) != 0;
1303 	*mchp = NULL;
1304 
1305 	i_mac_perim_enter(mip);
1306 
1307 	if (mip->mi_state_flags & MIS_IS_VNIC) {
1308 		/*
1309 		 * The underlying MAC is a VNIC. Return the MAC client
1310 		 * handle of the lower MAC which was obtained by
1311 		 * the VNIC driver when it did its mac_client_open().
1312 		 */
1313 
1314 		mcip = mac_vnic_lower(mip);
1315 
1316 		/*
1317 		 * Note that multiple mac clients share the same mcip in
1318 		 * this case.
1319 		 */
1320 		if (flags & MAC_OPEN_FLAGS_EXCLUSIVE)
1321 			mcip->mci_state_flags |= MCIS_EXCLUSIVE;
1322 
1323 		if (flags & MAC_OPEN_FLAGS_MULTI_PRIMARY)
1324 			mcip->mci_flags |= MAC_CLIENT_FLAGS_MULTI_PRIMARY;
1325 
1326 		mip->mi_clients_list = mcip;
1327 		i_mac_perim_exit(mip);
1328 		*mchp = (mac_client_handle_t)mcip;
1329 
1330 		DTRACE_PROBE2(mac__client__open__nonallocated, mac_impl_t *,
1331 		    mcip->mci_mip, mac_client_impl_t *, mcip);
1332 
1333 		return (err);
1334 	}
1335 
1336 	mcip = kmem_cache_alloc(mac_client_impl_cache, KM_SLEEP);
1337 
1338 	mcip->mci_mip = mip;
1339 	mcip->mci_upper_mip = NULL;
1340 	mcip->mci_rx_fn = mac_pkt_drop;
1341 	mcip->mci_rx_arg = NULL;
1342 	mcip->mci_rx_p_fn = NULL;
1343 	mcip->mci_rx_p_arg = NULL;
1344 	mcip->mci_p_unicast_list = NULL;
1345 	mcip->mci_direct_rx_fn = NULL;
1346 	mcip->mci_direct_rx_arg = NULL;
1347 
1348 	mcip->mci_unicast_list = NULL;
1349 
1350 	if ((flags & MAC_OPEN_FLAGS_IS_VNIC) != 0)
1351 		mcip->mci_state_flags |= MCIS_IS_VNIC;
1352 
1353 	if ((flags & MAC_OPEN_FLAGS_EXCLUSIVE) != 0)
1354 		mcip->mci_state_flags |= MCIS_EXCLUSIVE;
1355 
1356 	if ((flags & MAC_OPEN_FLAGS_IS_AGGR_PORT) != 0)
1357 		mcip->mci_state_flags |= MCIS_IS_AGGR_PORT;
1358 
1359 	if (mip->mi_state_flags & MIS_IS_AGGR)
1360 		mcip->mci_state_flags |= MCIS_IS_AGGR;
1361 
1362 	if ((flags & MAC_OPEN_FLAGS_USE_DATALINK_NAME) != 0) {
1363 		datalink_id_t	linkid;
1364 
1365 		ASSERT(name == NULL);
1366 		if ((err = dls_devnet_macname2linkid(mip->mi_name,
1367 		    &linkid)) != 0) {
1368 			goto done;
1369 		}
1370 		if ((err = dls_mgmt_get_linkinfo(linkid, mcip->mci_name, NULL,
1371 		    NULL, NULL)) != 0) {
1372 			/*
1373 			 * Use mac name if dlmgmtd is not available.
1374 			 */
1375 			if (err == EBADF) {
1376 				(void) strlcpy(mcip->mci_name, mip->mi_name,
1377 				    sizeof (mcip->mci_name));
1378 				err = 0;
1379 			} else {
1380 				goto done;
1381 			}
1382 		}
1383 		mcip->mci_state_flags |= MCIS_USE_DATALINK_NAME;
1384 	} else {
1385 		ASSERT(name != NULL);
1386 		if (strlen(name) > MAXNAMELEN) {
1387 			err = EINVAL;
1388 			goto done;
1389 		}
1390 		(void) strlcpy(mcip->mci_name, name, sizeof (mcip->mci_name));
1391 	}
1392 
1393 	if (flags & MAC_OPEN_FLAGS_MULTI_PRIMARY)
1394 		mcip->mci_flags |= MAC_CLIENT_FLAGS_MULTI_PRIMARY;
1395 
1396 	if (flags & MAC_OPEN_FLAGS_NO_UNICAST_ADDR)
1397 		mcip->mci_state_flags |= MCIS_NO_UNICAST_ADDR;
1398 
1399 	mac_protect_init(mcip);
1400 
1401 	/* the subflow table will be created dynamically */
1402 	mcip->mci_subflow_tab = NULL;
1403 
1404 	mcip->mci_misc_stat.mms_multircv = 0;
1405 	mcip->mci_misc_stat.mms_brdcstrcv = 0;
1406 	mcip->mci_misc_stat.mms_multixmt = 0;
1407 	mcip->mci_misc_stat.mms_brdcstxmt = 0;
1408 
1409 	/* Create an initial flow */
1410 
1411 	err = mac_flow_create(NULL, NULL, mcip->mci_name, NULL,
1412 	    mcip->mci_state_flags & MCIS_IS_VNIC ? FLOW_VNIC_MAC :
1413 	    FLOW_PRIMARY_MAC, &flent);
1414 	if (err != 0)
1415 		goto done;
1416 	mcip->mci_flent = flent;
1417 	FLOW_MARK(flent, FE_MC_NO_DATAPATH);
1418 	flent->fe_mcip = mcip;
1419 	/*
1420 	 * Place initial creation reference on the flow. This reference
1421 	 * is released in the corresponding delete action viz.
1422 	 * mac_unicast_remove after waiting for all transient refs to
1423 	 * to go away. The wait happens in mac_flow_wait.
1424 	 */
1425 	FLOW_REFHOLD(flent);
1426 
1427 	/*
1428 	 * Do this ahead of the mac_bcast_add() below so that the mi_nclients
1429 	 * will have the right value for mac_rx_srs_setup().
1430 	 */
1431 	mac_client_add(mcip);
1432 
1433 	mcip->mci_share = NULL;
1434 	if (share_desired)
1435 		i_mac_share_alloc(mcip);
1436 
1437 	/*
1438 	 * We will do mimimal datapath setup to allow a MAC client to
1439 	 * transmit or receive non-unicast packets without waiting
1440 	 * for mac_unicast_add.
1441 	 */
1442 	if (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR) {
1443 		if ((err = mac_client_datapath_setup(mcip, VLAN_ID_NONE,
1444 		    NULL, NULL, B_TRUE, NULL)) != 0) {
1445 			goto done;
1446 		}
1447 	}
1448 
1449 	DTRACE_PROBE2(mac__client__open__allocated, mac_impl_t *,
1450 	    mcip->mci_mip, mac_client_impl_t *, mcip);
1451 
1452 	*mchp = (mac_client_handle_t)mcip;
1453 	i_mac_perim_exit(mip);
1454 	return (0);
1455 
1456 done:
1457 	i_mac_perim_exit(mip);
1458 	mcip->mci_state_flags = 0;
1459 	mcip->mci_tx_flag = 0;
1460 	kmem_cache_free(mac_client_impl_cache, mcip);
1461 	return (err);
1462 }
1463 
1464 /*
1465  * Close the specified MAC client handle.
1466  */
1467 void
1468 mac_client_close(mac_client_handle_t mch, uint16_t flags)
1469 {
1470 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
1471 	mac_impl_t		*mip = mcip->mci_mip;
1472 	flow_entry_t		*flent;
1473 
1474 	i_mac_perim_enter(mip);
1475 
1476 	if (flags & MAC_CLOSE_FLAGS_EXCLUSIVE)
1477 		mcip->mci_state_flags &= ~MCIS_EXCLUSIVE;
1478 
1479 	if ((mcip->mci_state_flags & MCIS_IS_VNIC) &&
1480 	    !(flags & MAC_CLOSE_FLAGS_IS_VNIC)) {
1481 		/*
1482 		 * This is an upper VNIC client initiated operation.
1483 		 * The lower MAC client will be closed by the VNIC driver
1484 		 * when the VNIC is deleted.
1485 		 */
1486 
1487 		i_mac_perim_exit(mip);
1488 		return;
1489 	}
1490 
1491 	/* If we have only setup up minimal datapth setup, tear it down */
1492 	if (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR) {
1493 		mac_client_datapath_teardown((mac_client_handle_t)mcip, NULL,
1494 		    mcip->mci_flent);
1495 		mcip->mci_state_flags &= ~MCIS_NO_UNICAST_ADDR;
1496 	}
1497 
1498 	/*
1499 	 * Remove the flent associated with the MAC client
1500 	 */
1501 	flent = mcip->mci_flent;
1502 	mcip->mci_flent = NULL;
1503 	FLOW_FINAL_REFRELE(flent);
1504 
1505 	/*
1506 	 * MAC clients must remove the unicast addresses and promisc callbacks
1507 	 * they added before issuing a mac_client_close().
1508 	 */
1509 	ASSERT(mcip->mci_unicast_list == NULL);
1510 	ASSERT(mcip->mci_promisc_list == NULL);
1511 	ASSERT(mcip->mci_tx_notify_cb_list == NULL);
1512 
1513 	i_mac_share_free(mcip);
1514 	mac_protect_fini(mcip);
1515 	mac_client_remove(mcip);
1516 
1517 	i_mac_perim_exit(mip);
1518 	mcip->mci_subflow_tab = NULL;
1519 	mcip->mci_state_flags = 0;
1520 	mcip->mci_tx_flag = 0;
1521 	kmem_cache_free(mac_client_impl_cache, mch);
1522 }
1523 
1524 /*
1525  * Set the rx bypass receive callback.
1526  */
1527 boolean_t
1528 mac_rx_bypass_set(mac_client_handle_t mch, mac_direct_rx_t rx_fn, void *arg1)
1529 {
1530 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
1531 	mac_impl_t		*mip = mcip->mci_mip;
1532 
1533 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1534 
1535 	/*
1536 	 * If the mac_client is a VLAN, we should not do DLS bypass and
1537 	 * instead let the packets come up via mac_rx_deliver so the vlan
1538 	 * header can be stripped.
1539 	 */
1540 	if (mcip->mci_nvids > 0)
1541 		return (B_FALSE);
1542 
1543 	/*
1544 	 * These are not accessed directly in the data path, and hence
1545 	 * don't need any protection
1546 	 */
1547 	mcip->mci_direct_rx_fn = rx_fn;
1548 	mcip->mci_direct_rx_arg = arg1;
1549 	return (B_TRUE);
1550 }
1551 
1552 /*
1553  * Enable/Disable rx bypass. By default, bypass is assumed to be enabled.
1554  */
1555 void
1556 mac_rx_bypass_enable(mac_client_handle_t mch)
1557 {
1558 	((mac_client_impl_t *)mch)->mci_state_flags &= ~MCIS_RX_BYPASS_DISABLE;
1559 }
1560 
1561 void
1562 mac_rx_bypass_disable(mac_client_handle_t mch)
1563 {
1564 	((mac_client_impl_t *)mch)->mci_state_flags |= MCIS_RX_BYPASS_DISABLE;
1565 }
1566 
1567 /*
1568  * Set the receive callback for the specified MAC client. There can be
1569  * at most one such callback per MAC client.
1570  */
1571 void
1572 mac_rx_set(mac_client_handle_t mch, mac_rx_t rx_fn, void *arg)
1573 {
1574 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1575 	mac_impl_t	*mip = mcip->mci_mip;
1576 	mac_impl_t	*umip = mcip->mci_upper_mip;
1577 
1578 	/*
1579 	 * Instead of adding an extra set of locks and refcnts in
1580 	 * the datapath at the mac client boundary, we temporarily quiesce
1581 	 * the SRS and related entities. We then change the receive function
1582 	 * without interference from any receive data thread and then reenable
1583 	 * the data flow subsequently.
1584 	 */
1585 	i_mac_perim_enter(mip);
1586 	mac_rx_client_quiesce(mch);
1587 
1588 	mcip->mci_rx_fn = rx_fn;
1589 	mcip->mci_rx_arg = arg;
1590 	mac_rx_client_restart(mch);
1591 	i_mac_perim_exit(mip);
1592 
1593 	/*
1594 	 * If we're changing the rx function on the primary mac of a vnic,
1595 	 * make sure any secondary macs on the vnic are updated as well.
1596 	 */
1597 	if (umip != NULL) {
1598 		ASSERT((umip->mi_state_flags & MIS_IS_VNIC) != 0);
1599 		mac_vnic_secondary_update(umip);
1600 	}
1601 }
1602 
1603 /*
1604  * Reset the receive callback for the specified MAC client.
1605  */
1606 void
1607 mac_rx_clear(mac_client_handle_t mch)
1608 {
1609 	mac_rx_set(mch, mac_pkt_drop, NULL);
1610 }
1611 
1612 void
1613 mac_secondary_dup(mac_client_handle_t smch, mac_client_handle_t dmch)
1614 {
1615 	mac_client_impl_t *smcip = (mac_client_impl_t *)smch;
1616 	mac_client_impl_t *dmcip = (mac_client_impl_t *)dmch;
1617 	flow_entry_t *flent = dmcip->mci_flent;
1618 
1619 	/* This should only be called to setup secondary macs */
1620 	ASSERT((flent->fe_type & FLOW_PRIMARY_MAC) == 0);
1621 
1622 	mac_rx_set(dmch, smcip->mci_rx_fn, smcip->mci_rx_arg);
1623 	dmcip->mci_promisc_list = smcip->mci_promisc_list;
1624 
1625 	/*
1626 	 * Duplicate the primary mac resources to the secondary.
1627 	 * Since we already validated the resource controls when setting
1628 	 * them on the primary, we can ignore errors here.
1629 	 */
1630 	(void) mac_resource_ctl_set(dmch, MCIP_RESOURCE_PROPS(smcip));
1631 }
1632 
1633 /*
1634  * Called when removing a secondary MAC. Currently only clears the promisc_list
1635  * since we share the primary mac's promisc_list.
1636  */
1637 void
1638 mac_secondary_cleanup(mac_client_handle_t mch)
1639 {
1640 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1641 	flow_entry_t *flent = mcip->mci_flent;
1642 
1643 	/* This should only be called for secondary macs */
1644 	ASSERT((flent->fe_type & FLOW_PRIMARY_MAC) == 0);
1645 	mcip->mci_promisc_list = NULL;
1646 }
1647 
1648 /*
1649  * Walk the MAC client subflow table and updates their priority values.
1650  */
1651 static int
1652 mac_update_subflow_priority_cb(flow_entry_t *flent, void *arg)
1653 {
1654 	mac_flow_update_priority(arg, flent);
1655 	return (0);
1656 }
1657 
1658 void
1659 mac_update_subflow_priority(mac_client_impl_t *mcip)
1660 {
1661 	(void) mac_flow_walk(mcip->mci_subflow_tab,
1662 	    mac_update_subflow_priority_cb, mcip);
1663 }
1664 
1665 /*
1666  * Modify the TX or RX ring properties. We could either just move around
1667  * rings, i.e add/remove rings given to a client. Or this might cause the
1668  * client to move from hardware based to software or the other way around.
1669  * If we want to reset this property, then we clear the mask, additionally
1670  * if the client was given a non-default group we remove all rings except
1671  * for 1 and give it back to the default group.
1672  */
1673 int
1674 mac_client_set_rings_prop(mac_client_impl_t *mcip, mac_resource_props_t *mrp,
1675     mac_resource_props_t *tmrp)
1676 {
1677 	mac_impl_t		*mip = mcip->mci_mip;
1678 	flow_entry_t		*flent = mcip->mci_flent;
1679 	uint8_t			*mac_addr;
1680 	int			err = 0;
1681 	mac_group_t		*defgrp;
1682 	mac_group_t		*group;
1683 	mac_group_t		*ngrp;
1684 	mac_resource_props_t	*cmrp = MCIP_RESOURCE_PROPS(mcip);
1685 	uint_t			ringcnt;
1686 	boolean_t		unspec;
1687 
1688 	if (mcip->mci_share != NULL)
1689 		return (EINVAL);
1690 
1691 	if (mrp->mrp_mask & MRP_RX_RINGS) {
1692 		unspec = mrp->mrp_mask & MRP_RXRINGS_UNSPEC;
1693 		group = flent->fe_rx_ring_group;
1694 		defgrp = MAC_DEFAULT_RX_GROUP(mip);
1695 		mac_addr = flent->fe_flow_desc.fd_dst_mac;
1696 
1697 		/*
1698 		 * No resulting change. If we are resetting on a client on
1699 		 * which there was no rx rings property. For dynamic group
1700 		 * if we are setting the same number of rings already set.
1701 		 * For static group if we are requesting a group again.
1702 		 */
1703 		if (mrp->mrp_mask & MRP_RINGS_RESET) {
1704 			if (!(tmrp->mrp_mask & MRP_RX_RINGS))
1705 				return (0);
1706 		} else {
1707 			if (unspec) {
1708 				if (tmrp->mrp_mask & MRP_RXRINGS_UNSPEC)
1709 					return (0);
1710 			} else if (mip->mi_rx_group_type ==
1711 			    MAC_GROUP_TYPE_DYNAMIC) {
1712 				if ((tmrp->mrp_mask & MRP_RX_RINGS) &&
1713 				    !(tmrp->mrp_mask & MRP_RXRINGS_UNSPEC) &&
1714 				    mrp->mrp_nrxrings == tmrp->mrp_nrxrings) {
1715 					return (0);
1716 				}
1717 			}
1718 		}
1719 		/* Resetting the prop */
1720 		if (mrp->mrp_mask & MRP_RINGS_RESET) {
1721 			/*
1722 			 * We will just keep one ring and give others back if
1723 			 * we are not the primary. For the primary we give
1724 			 * all the rings in the default group except the
1725 			 * default ring. If it is a static group, then
1726 			 * we don't do anything, but clear the MRP_RX_RINGS
1727 			 * flag.
1728 			 */
1729 			if (group != defgrp) {
1730 				if (mip->mi_rx_group_type ==
1731 				    MAC_GROUP_TYPE_DYNAMIC) {
1732 					/*
1733 					 * This group has reserved rings
1734 					 * that need to be released now,
1735 					 * so does the group.
1736 					 */
1737 					MAC_RX_RING_RELEASED(mip,
1738 					    group->mrg_cur_count);
1739 					MAC_RX_GRP_RELEASED(mip);
1740 					if ((flent->fe_type &
1741 					    FLOW_PRIMARY_MAC) != 0) {
1742 						if (mip->mi_nactiveclients ==
1743 						    1) {
1744 							(void)
1745 							    mac_rx_switch_group(
1746 							    mcip, group,
1747 							    defgrp);
1748 							return (0);
1749 						} else {
1750 							cmrp->mrp_nrxrings =
1751 							    group->
1752 							    mrg_cur_count +
1753 							    defgrp->
1754 							    mrg_cur_count - 1;
1755 						}
1756 					} else {
1757 						cmrp->mrp_nrxrings = 1;
1758 					}
1759 					(void) mac_group_ring_modify(mcip,
1760 					    group, defgrp);
1761 				} else {
1762 					/*
1763 					 * If this is a static group, we
1764 					 * need to release the group. The
1765 					 * client will remain in the same
1766 					 * group till some other client
1767 					 * needs this group.
1768 					 */
1769 					MAC_RX_GRP_RELEASED(mip);
1770 				}
1771 			/* Let check if we can give this an excl group */
1772 			} else if (group == defgrp) {
1773 				ngrp = 	mac_reserve_rx_group(mcip, mac_addr,
1774 				    B_TRUE);
1775 				/* Couldn't give it a group, that's fine */
1776 				if (ngrp == NULL)
1777 					return (0);
1778 				/* Switch to H/W */
1779 				if (mac_rx_switch_group(mcip, defgrp, ngrp) !=
1780 				    0) {
1781 					mac_stop_group(ngrp);
1782 					return (0);
1783 				}
1784 			}
1785 			/*
1786 			 * If the client is in the default group, we will
1787 			 * just clear the MRP_RX_RINGS and leave it as
1788 			 * it rather than look for an exclusive group
1789 			 * for it.
1790 			 */
1791 			return (0);
1792 		}
1793 
1794 		if (group == defgrp && ((mrp->mrp_nrxrings > 0) || unspec)) {
1795 			ngrp = 	mac_reserve_rx_group(mcip, mac_addr, B_TRUE);
1796 			if (ngrp == NULL)
1797 				return (ENOSPC);
1798 
1799 			/* Switch to H/W */
1800 			if (mac_rx_switch_group(mcip, defgrp, ngrp) != 0) {
1801 				mac_release_rx_group(mcip, ngrp);
1802 				return (ENOSPC);
1803 			}
1804 			MAC_RX_GRP_RESERVED(mip);
1805 			if (mip->mi_rx_group_type == MAC_GROUP_TYPE_DYNAMIC)
1806 				MAC_RX_RING_RESERVED(mip, ngrp->mrg_cur_count);
1807 		} else if (group != defgrp && !unspec &&
1808 		    mrp->mrp_nrxrings == 0) {
1809 			/* Switch to S/W */
1810 			ringcnt = group->mrg_cur_count;
1811 			if (mac_rx_switch_group(mcip, group, defgrp) != 0)
1812 				return (ENOSPC);
1813 			if (tmrp->mrp_mask & MRP_RX_RINGS) {
1814 				MAC_RX_GRP_RELEASED(mip);
1815 				if (mip->mi_rx_group_type ==
1816 				    MAC_GROUP_TYPE_DYNAMIC) {
1817 					MAC_RX_RING_RELEASED(mip, ringcnt);
1818 				}
1819 			}
1820 		} else if (group != defgrp && mip->mi_rx_group_type ==
1821 		    MAC_GROUP_TYPE_DYNAMIC) {
1822 			ringcnt = group->mrg_cur_count;
1823 			err = mac_group_ring_modify(mcip, group, defgrp);
1824 			if (err != 0)
1825 				return (err);
1826 			/*
1827 			 * Update the accounting. If this group
1828 			 * already had explicitly reserved rings,
1829 			 * we need to update the rings based on
1830 			 * the new ring count. If this group
1831 			 * had not explicitly reserved rings,
1832 			 * then we just reserve the rings asked for
1833 			 * and reserve the group.
1834 			 */
1835 			if (tmrp->mrp_mask & MRP_RX_RINGS) {
1836 				if (ringcnt > group->mrg_cur_count) {
1837 					MAC_RX_RING_RELEASED(mip,
1838 					    ringcnt - group->mrg_cur_count);
1839 				} else {
1840 					MAC_RX_RING_RESERVED(mip,
1841 					    group->mrg_cur_count - ringcnt);
1842 				}
1843 			} else {
1844 				MAC_RX_RING_RESERVED(mip, group->mrg_cur_count);
1845 				MAC_RX_GRP_RESERVED(mip);
1846 			}
1847 		}
1848 	}
1849 	if (mrp->mrp_mask & MRP_TX_RINGS) {
1850 		unspec = mrp->mrp_mask & MRP_TXRINGS_UNSPEC;
1851 		group = flent->fe_tx_ring_group;
1852 		defgrp = MAC_DEFAULT_TX_GROUP(mip);
1853 
1854 		/*
1855 		 * For static groups we only allow rings=0 or resetting the
1856 		 * rings property.
1857 		 */
1858 		if (mrp->mrp_ntxrings > 0 &&
1859 		    mip->mi_tx_group_type != MAC_GROUP_TYPE_DYNAMIC) {
1860 			return (ENOTSUP);
1861 		}
1862 		if (mrp->mrp_mask & MRP_RINGS_RESET) {
1863 			if (!(tmrp->mrp_mask & MRP_TX_RINGS))
1864 				return (0);
1865 		} else {
1866 			if (unspec) {
1867 				if (tmrp->mrp_mask & MRP_TXRINGS_UNSPEC)
1868 					return (0);
1869 			} else if (mip->mi_tx_group_type ==
1870 			    MAC_GROUP_TYPE_DYNAMIC) {
1871 				if ((tmrp->mrp_mask & MRP_TX_RINGS) &&
1872 				    !(tmrp->mrp_mask & MRP_TXRINGS_UNSPEC) &&
1873 				    mrp->mrp_ntxrings == tmrp->mrp_ntxrings) {
1874 					return (0);
1875 				}
1876 			}
1877 		}
1878 		/* Resetting the prop */
1879 		if (mrp->mrp_mask & MRP_RINGS_RESET) {
1880 			if (group != defgrp) {
1881 				if (mip->mi_tx_group_type ==
1882 				    MAC_GROUP_TYPE_DYNAMIC) {
1883 					ringcnt = group->mrg_cur_count;
1884 					if ((flent->fe_type &
1885 					    FLOW_PRIMARY_MAC) != 0) {
1886 						mac_tx_client_quiesce(
1887 						    (mac_client_handle_t)
1888 						    mcip);
1889 						mac_tx_switch_group(mcip,
1890 						    group, defgrp);
1891 						mac_tx_client_restart(
1892 						    (mac_client_handle_t)
1893 						    mcip);
1894 						MAC_TX_GRP_RELEASED(mip);
1895 						MAC_TX_RING_RELEASED(mip,
1896 						    ringcnt);
1897 						return (0);
1898 					}
1899 					cmrp->mrp_ntxrings = 1;
1900 					(void) mac_group_ring_modify(mcip,
1901 					    group, defgrp);
1902 					/*
1903 					 * This group has reserved rings
1904 					 * that need to be released now.
1905 					 */
1906 					MAC_TX_RING_RELEASED(mip, ringcnt);
1907 				}
1908 				/*
1909 				 * If this is a static group, we
1910 				 * need to release the group. The
1911 				 * client will remain in the same
1912 				 * group till some other client
1913 				 * needs this group.
1914 				 */
1915 				MAC_TX_GRP_RELEASED(mip);
1916 			} else if (group == defgrp &&
1917 			    (flent->fe_type & FLOW_PRIMARY_MAC) == 0) {
1918 				ngrp = mac_reserve_tx_group(mcip, B_TRUE);
1919 				if (ngrp == NULL)
1920 					return (0);
1921 				mac_tx_client_quiesce(
1922 				    (mac_client_handle_t)mcip);
1923 				mac_tx_switch_group(mcip, defgrp, ngrp);
1924 				mac_tx_client_restart(
1925 				    (mac_client_handle_t)mcip);
1926 			}
1927 			/*
1928 			 * If the client is in the default group, we will
1929 			 * just clear the MRP_TX_RINGS and leave it as
1930 			 * it rather than look for an exclusive group
1931 			 * for it.
1932 			 */
1933 			return (0);
1934 		}
1935 
1936 		/* Switch to H/W */
1937 		if (group == defgrp && ((mrp->mrp_ntxrings > 0) || unspec)) {
1938 			ngrp = 	mac_reserve_tx_group(mcip, B_TRUE);
1939 			if (ngrp == NULL)
1940 				return (ENOSPC);
1941 			mac_tx_client_quiesce((mac_client_handle_t)mcip);
1942 			mac_tx_switch_group(mcip, defgrp, ngrp);
1943 			mac_tx_client_restart((mac_client_handle_t)mcip);
1944 			MAC_TX_GRP_RESERVED(mip);
1945 			if (mip->mi_tx_group_type == MAC_GROUP_TYPE_DYNAMIC)
1946 				MAC_TX_RING_RESERVED(mip, ngrp->mrg_cur_count);
1947 		/* Switch to S/W */
1948 		} else if (group != defgrp && !unspec &&
1949 		    mrp->mrp_ntxrings == 0) {
1950 			/* Switch to S/W */
1951 			ringcnt = group->mrg_cur_count;
1952 			mac_tx_client_quiesce((mac_client_handle_t)mcip);
1953 			mac_tx_switch_group(mcip, group, defgrp);
1954 			mac_tx_client_restart((mac_client_handle_t)mcip);
1955 			if (tmrp->mrp_mask & MRP_TX_RINGS) {
1956 				MAC_TX_GRP_RELEASED(mip);
1957 				if (mip->mi_tx_group_type ==
1958 				    MAC_GROUP_TYPE_DYNAMIC) {
1959 					MAC_TX_RING_RELEASED(mip, ringcnt);
1960 				}
1961 			}
1962 		} else if (group != defgrp && mip->mi_tx_group_type ==
1963 		    MAC_GROUP_TYPE_DYNAMIC) {
1964 			ringcnt = group->mrg_cur_count;
1965 			err = mac_group_ring_modify(mcip, group, defgrp);
1966 			if (err != 0)
1967 				return (err);
1968 			/*
1969 			 * Update the accounting. If this group
1970 			 * already had explicitly reserved rings,
1971 			 * we need to update the rings based on
1972 			 * the new ring count. If this group
1973 			 * had not explicitly reserved rings,
1974 			 * then we just reserve the rings asked for
1975 			 * and reserve the group.
1976 			 */
1977 			if (tmrp->mrp_mask & MRP_TX_RINGS) {
1978 				if (ringcnt > group->mrg_cur_count) {
1979 					MAC_TX_RING_RELEASED(mip,
1980 					    ringcnt - group->mrg_cur_count);
1981 				} else {
1982 					MAC_TX_RING_RESERVED(mip,
1983 					    group->mrg_cur_count - ringcnt);
1984 				}
1985 			} else {
1986 				MAC_TX_RING_RESERVED(mip, group->mrg_cur_count);
1987 				MAC_TX_GRP_RESERVED(mip);
1988 			}
1989 		}
1990 	}
1991 	return (0);
1992 }
1993 
1994 /*
1995  * When the MAC client is being brought up (i.e. we do a unicast_add) we need
1996  * to initialize the cpu and resource control structure in the
1997  * mac_client_impl_t from the mac_impl_t (i.e if there are any cached
1998  * properties before the flow entry for the unicast address was created).
1999  */
2000 static int
2001 mac_resource_ctl_set(mac_client_handle_t mch, mac_resource_props_t *mrp)
2002 {
2003 	mac_client_impl_t 	*mcip = (mac_client_impl_t *)mch;
2004 	mac_impl_t		*mip = (mac_impl_t *)mcip->mci_mip;
2005 	mac_impl_t		*umip = mcip->mci_upper_mip;
2006 	int			err = 0;
2007 	flow_entry_t		*flent = mcip->mci_flent;
2008 	mac_resource_props_t	*omrp, *nmrp = MCIP_RESOURCE_PROPS(mcip);
2009 
2010 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
2011 
2012 	err = mac_validate_props(mcip->mci_state_flags & MCIS_IS_VNIC ?
2013 	    mcip->mci_upper_mip : mip, mrp);
2014 	if (err != 0)
2015 		return (err);
2016 
2017 	/*
2018 	 * Copy over the existing properties since mac_update_resources
2019 	 * will modify the client's mrp. Currently, the saved property
2020 	 * is used to determine the difference between existing and
2021 	 * modified rings property.
2022 	 */
2023 	omrp = kmem_zalloc(sizeof (*omrp), KM_SLEEP);
2024 	bcopy(nmrp, omrp, sizeof (*omrp));
2025 	mac_update_resources(mrp, MCIP_RESOURCE_PROPS(mcip), B_FALSE);
2026 	if (MCIP_DATAPATH_SETUP(mcip)) {
2027 		/*
2028 		 * We support rings only for primary client when there are
2029 		 * multiple clients sharing the same MAC address (e.g. VLAN).
2030 		 */
2031 		if (mrp->mrp_mask & MRP_RX_RINGS ||
2032 		    mrp->mrp_mask & MRP_TX_RINGS) {
2033 
2034 			if ((err = mac_client_set_rings_prop(mcip, mrp,
2035 			    omrp)) != 0) {
2036 				if (omrp->mrp_mask & MRP_RX_RINGS) {
2037 					nmrp->mrp_mask |= MRP_RX_RINGS;
2038 					nmrp->mrp_nrxrings = omrp->mrp_nrxrings;
2039 				} else {
2040 					nmrp->mrp_mask &= ~MRP_RX_RINGS;
2041 					nmrp->mrp_nrxrings = 0;
2042 				}
2043 				if (omrp->mrp_mask & MRP_TX_RINGS) {
2044 					nmrp->mrp_mask |= MRP_TX_RINGS;
2045 					nmrp->mrp_ntxrings = omrp->mrp_ntxrings;
2046 				} else {
2047 					nmrp->mrp_mask &= ~MRP_TX_RINGS;
2048 					nmrp->mrp_ntxrings = 0;
2049 				}
2050 				if (omrp->mrp_mask & MRP_RXRINGS_UNSPEC)
2051 					omrp->mrp_mask |= MRP_RXRINGS_UNSPEC;
2052 				else
2053 					omrp->mrp_mask &= ~MRP_RXRINGS_UNSPEC;
2054 
2055 				if (omrp->mrp_mask & MRP_TXRINGS_UNSPEC)
2056 					omrp->mrp_mask |= MRP_TXRINGS_UNSPEC;
2057 				else
2058 					omrp->mrp_mask &= ~MRP_TXRINGS_UNSPEC;
2059 				kmem_free(omrp, sizeof (*omrp));
2060 				return (err);
2061 			}
2062 
2063 			/*
2064 			 * If we modified the rings property of the primary
2065 			 * we need to update the property fields of its
2066 			 * VLANs as they inherit the primary's properites.
2067 			 */
2068 			if (mac_is_primary_client(mcip)) {
2069 				mac_set_prim_vlan_rings(mip,
2070 				    MCIP_RESOURCE_PROPS(mcip));
2071 			}
2072 		}
2073 		/*
2074 		 * We have to set this prior to calling mac_flow_modify.
2075 		 */
2076 		if (mrp->mrp_mask & MRP_PRIORITY) {
2077 			if (mrp->mrp_priority == MPL_RESET) {
2078 				MAC_CLIENT_SET_PRIORITY_RANGE(mcip,
2079 				    MPL_LINK_DEFAULT);
2080 			} else {
2081 				MAC_CLIENT_SET_PRIORITY_RANGE(mcip,
2082 				    mrp->mrp_priority);
2083 			}
2084 		}
2085 
2086 		mac_flow_modify(mip->mi_flow_tab, flent, mrp);
2087 		if (mrp->mrp_mask & MRP_PRIORITY)
2088 			mac_update_subflow_priority(mcip);
2089 
2090 		/* Apply these resource settings to any secondary macs */
2091 		if (umip != NULL) {
2092 			ASSERT((umip->mi_state_flags & MIS_IS_VNIC) != 0);
2093 			mac_vnic_secondary_update(umip);
2094 		}
2095 	}
2096 	kmem_free(omrp, sizeof (*omrp));
2097 	return (0);
2098 }
2099 
2100 static int
2101 mac_unicast_flow_create(mac_client_impl_t *mcip, uint8_t *mac_addr,
2102     uint16_t vid, boolean_t is_primary, boolean_t first_flow,
2103     flow_entry_t **flent, mac_resource_props_t *mrp)
2104 {
2105 	mac_impl_t	*mip = (mac_impl_t *)mcip->mci_mip;
2106 	flow_desc_t	flow_desc;
2107 	char		flowname[MAXFLOWNAMELEN];
2108 	int		err;
2109 	uint_t		flent_flags;
2110 
2111 	/*
2112 	 * First unicast address being added, create a new flow
2113 	 * for that MAC client.
2114 	 */
2115 	bzero(&flow_desc, sizeof (flow_desc));
2116 
2117 	ASSERT(mac_addr != NULL ||
2118 	    (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR));
2119 	if (mac_addr != NULL) {
2120 		flow_desc.fd_mac_len = mip->mi_type->mt_addr_length;
2121 		bcopy(mac_addr, flow_desc.fd_dst_mac, flow_desc.fd_mac_len);
2122 	}
2123 	flow_desc.fd_mask = FLOW_LINK_DST;
2124 	if (vid != 0) {
2125 		flow_desc.fd_vid = vid;
2126 		flow_desc.fd_mask |= FLOW_LINK_VID;
2127 	}
2128 
2129 	/*
2130 	 * XXX-nicolas. For now I'm keeping the FLOW_PRIMARY_MAC
2131 	 * and FLOW_VNIC. Even though they're a hack inherited
2132 	 * from the SRS code, we'll keep them for now. They're currently
2133 	 * consumed by mac_datapath_setup() to create the SRS.
2134 	 * That code should be eventually moved out of
2135 	 * mac_datapath_setup() and moved to a mac_srs_create()
2136 	 * function of some sort to keep things clean.
2137 	 *
2138 	 * Also, there's no reason why the SRS for the primary MAC
2139 	 * client should be different than any other MAC client. Until
2140 	 * this is cleaned-up, we support only one MAC unicast address
2141 	 * per client.
2142 	 *
2143 	 * We set FLOW_PRIMARY_MAC for the primary MAC address,
2144 	 * FLOW_VNIC for everything else.
2145 	 */
2146 	if (is_primary)
2147 		flent_flags = FLOW_PRIMARY_MAC;
2148 	else
2149 		flent_flags = FLOW_VNIC_MAC;
2150 
2151 	/*
2152 	 * For the first flow we use the mac client's name - mci_name, for
2153 	 * subsequent ones we just create a name with the vid. This is
2154 	 * so that we can add these flows to the same flow table. This is
2155 	 * fine as the flow name (except for the one with the mac client's
2156 	 * name) is not visible. When the first flow is removed, we just replace
2157 	 * its fdesc with another from the list, so we will still retain the
2158 	 * flent with the MAC client's flow name.
2159 	 */
2160 	if (first_flow) {
2161 		bcopy(mcip->mci_name, flowname, MAXFLOWNAMELEN);
2162 	} else {
2163 		(void) sprintf(flowname, "%s%u", mcip->mci_name, vid);
2164 		flent_flags = FLOW_NO_STATS;
2165 	}
2166 
2167 	if ((err = mac_flow_create(&flow_desc, mrp, flowname, NULL,
2168 	    flent_flags, flent)) != 0)
2169 		return (err);
2170 
2171 	mac_misc_stat_create(*flent);
2172 	FLOW_MARK(*flent, FE_INCIPIENT);
2173 	(*flent)->fe_mcip = mcip;
2174 
2175 	/*
2176 	 * Place initial creation reference on the flow. This reference
2177 	 * is released in the corresponding delete action viz.
2178 	 * mac_unicast_remove after waiting for all transient refs to
2179 	 * to go away. The wait happens in mac_flow_wait.
2180 	 * We have already held the reference in mac_client_open().
2181 	 */
2182 	if (!first_flow)
2183 		FLOW_REFHOLD(*flent);
2184 	return (0);
2185 }
2186 
2187 /* Refresh the multicast grouping for this VID. */
2188 int
2189 mac_client_update_mcast(void *arg, boolean_t add, const uint8_t *addrp)
2190 {
2191 	flow_entry_t		*flent = arg;
2192 	mac_client_impl_t	*mcip = flent->fe_mcip;
2193 	uint16_t		vid;
2194 	flow_desc_t		flow_desc;
2195 
2196 	mac_flow_get_desc(flent, &flow_desc);
2197 	vid = (flow_desc.fd_mask & FLOW_LINK_VID) != 0 ?
2198 	    flow_desc.fd_vid : VLAN_ID_NONE;
2199 
2200 	/*
2201 	 * We don't call mac_multicast_add()/mac_multicast_remove() as
2202 	 * we want to add/remove for this specific vid.
2203 	 */
2204 	if (add) {
2205 		return (mac_bcast_add(mcip, addrp, vid,
2206 		    MAC_ADDRTYPE_MULTICAST));
2207 	} else {
2208 		mac_bcast_delete(mcip, addrp, vid);
2209 		return (0);
2210 	}
2211 }
2212 
2213 static void
2214 mac_update_single_active_client(mac_impl_t *mip)
2215 {
2216 	mac_client_impl_t *client = NULL;
2217 
2218 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
2219 
2220 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
2221 	if (mip->mi_nactiveclients == 1) {
2222 		/*
2223 		 * Find the one active MAC client from the list of MAC
2224 		 * clients. The active MAC client has at least one
2225 		 * unicast address.
2226 		 */
2227 		for (client = mip->mi_clients_list; client != NULL;
2228 		    client = client->mci_client_next) {
2229 			if (client->mci_unicast_list != NULL)
2230 				break;
2231 		}
2232 		ASSERT(client != NULL);
2233 	}
2234 
2235 	/*
2236 	 * mi_single_active_client is protected by the MAC impl's read/writer
2237 	 * lock, which allows mac_rx() to check the value of that pointer
2238 	 * as a reader.
2239 	 */
2240 	mip->mi_single_active_client = client;
2241 	rw_exit(&mip->mi_rw_lock);
2242 }
2243 
2244 /*
2245  * Set up the data path. Called from i_mac_unicast_add after having
2246  * done all the validations including making sure this is an active
2247  * client (i.e that is ready to process packets.)
2248  */
2249 static int
2250 mac_client_datapath_setup(mac_client_impl_t *mcip, uint16_t vid,
2251     uint8_t *mac_addr, mac_resource_props_t *mrp, boolean_t isprimary,
2252     mac_unicast_impl_t *muip)
2253 {
2254 	mac_impl_t	*mip = mcip->mci_mip;
2255 	boolean_t	mac_started = B_FALSE;
2256 	boolean_t	bcast_added = B_FALSE;
2257 	boolean_t	nactiveclients_added = B_FALSE;
2258 	flow_entry_t	*flent;
2259 	int		err = 0;
2260 	boolean_t	no_unicast;
2261 
2262 	no_unicast = mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR;
2263 
2264 	if ((err = mac_start((mac_handle_t)mip)) != 0)
2265 		goto bail;
2266 
2267 	mac_started = B_TRUE;
2268 
2269 	/* add the MAC client to the broadcast address group by default */
2270 	if (mip->mi_type->mt_brdcst_addr != NULL) {
2271 		err = mac_bcast_add(mcip, mip->mi_type->mt_brdcst_addr, vid,
2272 		    MAC_ADDRTYPE_BROADCAST);
2273 		if (err != 0)
2274 			goto bail;
2275 		bcast_added = B_TRUE;
2276 	}
2277 
2278 	/*
2279 	 * If this is the first unicast address addition for this
2280 	 * client, reuse the pre-allocated larval flow entry associated with
2281 	 * the MAC client.
2282 	 */
2283 	flent = (mcip->mci_nflents == 0) ? mcip->mci_flent : NULL;
2284 
2285 	/* We are configuring the unicast flow now */
2286 	if (!MCIP_DATAPATH_SETUP(mcip)) {
2287 
2288 		if (mrp != NULL) {
2289 			MAC_CLIENT_SET_PRIORITY_RANGE(mcip,
2290 			    (mrp->mrp_mask & MRP_PRIORITY) ? mrp->mrp_priority :
2291 			    MPL_LINK_DEFAULT);
2292 		}
2293 		if ((err = mac_unicast_flow_create(mcip, mac_addr, vid,
2294 		    isprimary, B_TRUE, &flent, mrp)) != 0)
2295 			goto bail;
2296 
2297 		mip->mi_nactiveclients++;
2298 		nactiveclients_added = B_TRUE;
2299 
2300 		/*
2301 		 * This will allocate the RX ring group if possible for the
2302 		 * flow and program the software classifier as needed.
2303 		 */
2304 		if ((err = mac_datapath_setup(mcip, flent, SRST_LINK)) != 0)
2305 			goto bail;
2306 
2307 		if (no_unicast)
2308 			goto done_setup;
2309 		/*
2310 		 * The unicast MAC address must have been added successfully.
2311 		 */
2312 		ASSERT(mcip->mci_unicast != NULL);
2313 		/*
2314 		 * Push down the sub-flows that were defined on this link
2315 		 * hitherto. The flows are added to the active flow table
2316 		 * and SRS, softrings etc. are created as needed.
2317 		 */
2318 		mac_link_init_flows((mac_client_handle_t)mcip);
2319 	} else {
2320 		mac_address_t *map = mcip->mci_unicast;
2321 
2322 		ASSERT(!no_unicast);
2323 		/*
2324 		 * A unicast flow already exists for that MAC client,
2325 		 * this flow must be the same mac address but with
2326 		 * different VID. It has been checked by mac_addr_in_use().
2327 		 *
2328 		 * We will use the SRS etc. from the mci_flent. Note that
2329 		 * We don't need to create kstat for this as except for
2330 		 * the fdesc, everything will be used from in the 1st flent.
2331 		 */
2332 
2333 		if (bcmp(mac_addr, map->ma_addr, map->ma_len) != 0) {
2334 			err = EINVAL;
2335 			goto bail;
2336 		}
2337 
2338 		if ((err = mac_unicast_flow_create(mcip, mac_addr, vid,
2339 		    isprimary, B_FALSE, &flent, NULL)) != 0) {
2340 			goto bail;
2341 		}
2342 		if ((err = mac_flow_add(mip->mi_flow_tab, flent)) != 0) {
2343 			FLOW_FINAL_REFRELE(flent);
2344 			goto bail;
2345 		}
2346 
2347 		/* update the multicast group for this vid */
2348 		mac_client_bcast_refresh(mcip, mac_client_update_mcast,
2349 		    (void *)flent, B_TRUE);
2350 
2351 	}
2352 
2353 	/* populate the shared MAC address */
2354 	muip->mui_map = mcip->mci_unicast;
2355 
2356 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
2357 	muip->mui_next = mcip->mci_unicast_list;
2358 	mcip->mci_unicast_list = muip;
2359 	rw_exit(&mcip->mci_rw_lock);
2360 
2361 done_setup:
2362 	/*
2363 	 * First add the flent to the flow list of this mcip. Then set
2364 	 * the mip's mi_single_active_client if needed. The Rx path assumes
2365 	 * that mip->mi_single_active_client will always have an associated
2366 	 * flent.
2367 	 */
2368 	mac_client_add_to_flow_list(mcip, flent);
2369 	if (nactiveclients_added)
2370 		mac_update_single_active_client(mip);
2371 	/*
2372 	 * Trigger a renegotiation of the capabilities when the number of
2373 	 * active clients changes from 1 to 2, since some of the capabilities
2374 	 * might have to be disabled. Also send a MAC_NOTE_LINK notification
2375 	 * to all the MAC clients whenever physical link is DOWN.
2376 	 */
2377 	if (mip->mi_nactiveclients == 2) {
2378 		mac_capab_update((mac_handle_t)mip);
2379 		mac_virtual_link_update(mip);
2380 	}
2381 	/*
2382 	 * Now that the setup is complete, clear the INCIPIENT flag.
2383 	 * The flag was set to avoid incoming packets seeing inconsistent
2384 	 * structures while the setup was in progress. Clear the mci_tx_flag
2385 	 * by calling mac_tx_client_block. It is possible that
2386 	 * mac_unicast_remove was called prior to this mac_unicast_add which
2387 	 * could have set the MCI_TX_QUIESCE flag.
2388 	 */
2389 	if (flent->fe_rx_ring_group != NULL)
2390 		mac_rx_group_unmark(flent->fe_rx_ring_group, MR_INCIPIENT);
2391 	FLOW_UNMARK(flent, FE_INCIPIENT);
2392 	FLOW_UNMARK(flent, FE_MC_NO_DATAPATH);
2393 	mac_tx_client_unblock(mcip);
2394 	return (0);
2395 bail:
2396 	if (bcast_added)
2397 		mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr, vid);
2398 
2399 	if (nactiveclients_added)
2400 		mip->mi_nactiveclients--;
2401 
2402 	if (mac_started)
2403 		mac_stop((mac_handle_t)mip);
2404 
2405 	return (err);
2406 }
2407 
2408 /*
2409  * Return the passive primary MAC client, if present. The passive client is
2410  * a stand-by client that has the same unicast address as another that is
2411  * currenly active. Once the active client goes away, the passive client
2412  * becomes active.
2413  */
2414 static mac_client_impl_t *
2415 mac_get_passive_primary_client(mac_impl_t *mip)
2416 {
2417 	mac_client_impl_t	*mcip;
2418 
2419 	for (mcip = mip->mi_clients_list; mcip != NULL;
2420 	    mcip = mcip->mci_client_next) {
2421 		if (mac_is_primary_client(mcip) &&
2422 		    (mcip->mci_flags & MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
2423 			return (mcip);
2424 		}
2425 	}
2426 	return (NULL);
2427 }
2428 
2429 /*
2430  * Add a new unicast address to the MAC client.
2431  *
2432  * The MAC address can be specified either by value, or the MAC client
2433  * can specify that it wants to use the primary MAC address of the
2434  * underlying MAC. See the introductory comments at the beginning
2435  * of this file for more more information on primary MAC addresses.
2436  *
2437  * Note also the tuple (MAC address, VID) must be unique
2438  * for the MAC clients defined on top of the same underlying MAC
2439  * instance, unless the MAC_UNICAST_NODUPCHECK is specified.
2440  *
2441  * In no case can a client use the PVID for the MAC, if the MAC has one set.
2442  */
2443 int
2444 i_mac_unicast_add(mac_client_handle_t mch, uint8_t *mac_addr, uint16_t flags,
2445     mac_unicast_handle_t *mah, uint16_t vid, mac_diag_t *diag)
2446 {
2447 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2448 	mac_impl_t		*mip = mcip->mci_mip;
2449 	int			err;
2450 	uint_t			mac_len = mip->mi_type->mt_addr_length;
2451 	boolean_t		check_dups = !(flags & MAC_UNICAST_NODUPCHECK);
2452 	boolean_t		fastpath_disabled = B_FALSE;
2453 	boolean_t		is_primary = (flags & MAC_UNICAST_PRIMARY);
2454 	boolean_t		is_unicast_hw = (flags & MAC_UNICAST_HW);
2455 	mac_resource_props_t	*mrp;
2456 	boolean_t		passive_client = B_FALSE;
2457 	mac_unicast_impl_t	*muip;
2458 	boolean_t		is_vnic_primary =
2459 	    (flags & MAC_UNICAST_VNIC_PRIMARY);
2460 
2461 	/* when VID is non-zero, the underlying MAC can not be VNIC */
2462 	ASSERT(!((mip->mi_state_flags & MIS_IS_VNIC) && (vid != 0)));
2463 
2464 	/*
2465 	 * Can't unicast add if the client asked only for minimal datapath
2466 	 * setup.
2467 	 */
2468 	if (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR)
2469 		return (ENOTSUP);
2470 
2471 	/*
2472 	 * Check for an attempted use of the current Port VLAN ID, if enabled.
2473 	 * No client may use it.
2474 	 */
2475 	if (mip->mi_pvid != 0 && vid == mip->mi_pvid)
2476 		return (EBUSY);
2477 
2478 	/*
2479 	 * Check whether it's the primary client and flag it.
2480 	 */
2481 	if (!(mcip->mci_state_flags & MCIS_IS_VNIC) && is_primary && vid == 0)
2482 		mcip->mci_flags |= MAC_CLIENT_FLAGS_PRIMARY;
2483 
2484 	/*
2485 	 * is_vnic_primary is true when we come here as a VLAN VNIC
2486 	 * which uses the primary mac client's address but with a non-zero
2487 	 * VID. In this case the MAC address is not specified by an upper
2488 	 * MAC client.
2489 	 */
2490 	if ((mcip->mci_state_flags & MCIS_IS_VNIC) && is_primary &&
2491 	    !is_vnic_primary) {
2492 		/*
2493 		 * The address is being set by the upper MAC client
2494 		 * of a VNIC. The MAC address was already set by the
2495 		 * VNIC driver during VNIC creation.
2496 		 *
2497 		 * Note: a VNIC has only one MAC address. We return
2498 		 * the MAC unicast address handle of the lower MAC client
2499 		 * corresponding to the VNIC. We allocate a new entry
2500 		 * which is flagged appropriately, so that mac_unicast_remove()
2501 		 * doesn't attempt to free the original entry that
2502 		 * was allocated by the VNIC driver.
2503 		 */
2504 		ASSERT(mcip->mci_unicast != NULL);
2505 
2506 		/* Check for VLAN flags, if present */
2507 		if ((flags & MAC_UNICAST_TAG_DISABLE) != 0)
2508 			mcip->mci_state_flags |= MCIS_TAG_DISABLE;
2509 
2510 		if ((flags & MAC_UNICAST_STRIP_DISABLE) != 0)
2511 			mcip->mci_state_flags |= MCIS_STRIP_DISABLE;
2512 
2513 		if ((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) != 0)
2514 			mcip->mci_state_flags |= MCIS_DISABLE_TX_VID_CHECK;
2515 
2516 		/*
2517 		 * Ensure that the primary unicast address of the VNIC
2518 		 * is added only once unless we have the
2519 		 * MAC_CLIENT_FLAGS_MULTI_PRIMARY set (and this is not
2520 		 * a passive MAC client).
2521 		 */
2522 		if ((mcip->mci_flags & MAC_CLIENT_FLAGS_VNIC_PRIMARY) != 0) {
2523 			if ((mcip->mci_flags &
2524 			    MAC_CLIENT_FLAGS_MULTI_PRIMARY) == 0 ||
2525 			    (mcip->mci_flags &
2526 			    MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
2527 				return (EBUSY);
2528 			}
2529 			mcip->mci_flags |= MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
2530 			passive_client = B_TRUE;
2531 		}
2532 
2533 		mcip->mci_flags |= MAC_CLIENT_FLAGS_VNIC_PRIMARY;
2534 
2535 		/*
2536 		 * Create a handle for vid 0.
2537 		 */
2538 		ASSERT(vid == 0);
2539 		muip = kmem_zalloc(sizeof (mac_unicast_impl_t), KM_SLEEP);
2540 		muip->mui_vid = vid;
2541 		*mah = (mac_unicast_handle_t)muip;
2542 		/*
2543 		 * This will be used by the caller to defer setting the
2544 		 * rx functions.
2545 		 */
2546 		if (passive_client)
2547 			return (EAGAIN);
2548 		return (0);
2549 	}
2550 
2551 	/* primary MAC clients cannot be opened on top of anchor VNICs */
2552 	if ((is_vnic_primary || is_primary) &&
2553 	    i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_ANCHOR_VNIC, NULL)) {
2554 		return (ENXIO);
2555 	}
2556 
2557 	/*
2558 	 * If this is a VNIC/VLAN, disable softmac fast-path.
2559 	 */
2560 	if (mcip->mci_state_flags & MCIS_IS_VNIC) {
2561 		err = mac_fastpath_disable((mac_handle_t)mip);
2562 		if (err != 0)
2563 			return (err);
2564 		fastpath_disabled = B_TRUE;
2565 	}
2566 
2567 	/*
2568 	 * Return EBUSY if:
2569 	 *  - there is an exclusively active mac client exists.
2570 	 *  - this is an exclusive active mac client but
2571 	 *	a. there is already active mac clients exist, or
2572 	 *	b. fastpath streams are already plumbed on this legacy device
2573 	 *  - the mac creator has disallowed active mac clients.
2574 	 */
2575 	if (mip->mi_state_flags & (MIS_EXCLUSIVE|MIS_NO_ACTIVE)) {
2576 		if (fastpath_disabled)
2577 			mac_fastpath_enable((mac_handle_t)mip);
2578 		return (EBUSY);
2579 	}
2580 
2581 	if (mcip->mci_state_flags & MCIS_EXCLUSIVE) {
2582 		ASSERT(!fastpath_disabled);
2583 		if (mip->mi_nactiveclients != 0)
2584 			return (EBUSY);
2585 
2586 		if ((mip->mi_state_flags & MIS_LEGACY) &&
2587 		    !(mip->mi_capab_legacy.ml_active_set(mip->mi_driver))) {
2588 			return (EBUSY);
2589 		}
2590 		mip->mi_state_flags |= MIS_EXCLUSIVE;
2591 	}
2592 
2593 	mrp = kmem_zalloc(sizeof (*mrp), KM_SLEEP);
2594 	if (is_primary && !(mcip->mci_state_flags & (MCIS_IS_VNIC |
2595 	    MCIS_IS_AGGR_PORT))) {
2596 		/*
2597 		 * Apply the property cached in the mac_impl_t to the primary
2598 		 * mac client. If the mac client is a VNIC or an aggregation
2599 		 * port, its property should be set in the mcip when the
2600 		 * VNIC/aggr was created.
2601 		 */
2602 		mac_get_resources((mac_handle_t)mip, mrp);
2603 		(void) mac_client_set_resources(mch, mrp);
2604 	} else if (mcip->mci_state_flags & MCIS_IS_VNIC) {
2605 		/*
2606 		 * This is a primary VLAN client, we don't support
2607 		 * specifying rings property for this as it inherits the
2608 		 * rings property from its MAC.
2609 		 */
2610 		if (is_vnic_primary) {
2611 			mac_resource_props_t	*vmrp;
2612 
2613 			vmrp = MCIP_RESOURCE_PROPS(mcip);
2614 			if (vmrp->mrp_mask & MRP_RX_RINGS ||
2615 			    vmrp->mrp_mask & MRP_TX_RINGS) {
2616 				if (fastpath_disabled)
2617 					mac_fastpath_enable((mac_handle_t)mip);
2618 				kmem_free(mrp, sizeof (*mrp));
2619 				return (ENOTSUP);
2620 			}
2621 			/*
2622 			 * Additionally we also need to inherit any
2623 			 * rings property from the MAC.
2624 			 */
2625 			mac_get_resources((mac_handle_t)mip, mrp);
2626 			if (mrp->mrp_mask & MRP_RX_RINGS) {
2627 				vmrp->mrp_mask |= MRP_RX_RINGS;
2628 				vmrp->mrp_nrxrings = mrp->mrp_nrxrings;
2629 			}
2630 			if (mrp->mrp_mask & MRP_TX_RINGS) {
2631 				vmrp->mrp_mask |= MRP_TX_RINGS;
2632 				vmrp->mrp_ntxrings = mrp->mrp_ntxrings;
2633 			}
2634 		}
2635 		bcopy(MCIP_RESOURCE_PROPS(mcip), mrp, sizeof (*mrp));
2636 	}
2637 
2638 	muip = kmem_zalloc(sizeof (mac_unicast_impl_t), KM_SLEEP);
2639 	muip->mui_vid = vid;
2640 
2641 	if (is_primary || is_vnic_primary) {
2642 		mac_addr = mip->mi_addr;
2643 	} else {
2644 
2645 		/*
2646 		 * Verify the validity of the specified MAC addresses value.
2647 		 */
2648 		if (!mac_unicst_verify((mac_handle_t)mip, mac_addr, mac_len)) {
2649 			*diag = MAC_DIAG_MACADDR_INVALID;
2650 			err = EINVAL;
2651 			goto bail_out;
2652 		}
2653 
2654 		/*
2655 		 * Make sure that the specified MAC address is different
2656 		 * than the unicast MAC address of the underlying NIC.
2657 		 */
2658 		if (check_dups && bcmp(mip->mi_addr, mac_addr, mac_len) == 0) {
2659 			*diag = MAC_DIAG_MACADDR_NIC;
2660 			err = EINVAL;
2661 			goto bail_out;
2662 		}
2663 	}
2664 
2665 	/*
2666 	 * Set the flags here so that if this is a passive client, we
2667 	 * can return  and set it when we call mac_client_datapath_setup
2668 	 * when this becomes the active client. If we defer to using these
2669 	 * flags to mac_client_datapath_setup, then for a passive client,
2670 	 * we'd have to store the flags somewhere (probably fe_flags)
2671 	 * and then use it.
2672 	 */
2673 	if (!MCIP_DATAPATH_SETUP(mcip)) {
2674 		if (is_unicast_hw) {
2675 			/*
2676 			 * The client requires a hardware MAC address slot
2677 			 * for that unicast address. Since we support only
2678 			 * one unicast MAC address per client, flag the
2679 			 * MAC client itself.
2680 			 */
2681 			mcip->mci_state_flags |= MCIS_UNICAST_HW;
2682 		}
2683 
2684 		/* Check for VLAN flags, if present */
2685 		if ((flags & MAC_UNICAST_TAG_DISABLE) != 0)
2686 			mcip->mci_state_flags |= MCIS_TAG_DISABLE;
2687 
2688 		if ((flags & MAC_UNICAST_STRIP_DISABLE) != 0)
2689 			mcip->mci_state_flags |= MCIS_STRIP_DISABLE;
2690 
2691 		if ((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) != 0)
2692 			mcip->mci_state_flags |= MCIS_DISABLE_TX_VID_CHECK;
2693 	} else {
2694 		/*
2695 		 * Assert that the specified flags are consistent with the
2696 		 * flags specified by previous calls to mac_unicast_add().
2697 		 */
2698 		ASSERT(((flags & MAC_UNICAST_TAG_DISABLE) != 0 &&
2699 		    (mcip->mci_state_flags & MCIS_TAG_DISABLE) != 0) ||
2700 		    ((flags & MAC_UNICAST_TAG_DISABLE) == 0 &&
2701 		    (mcip->mci_state_flags & MCIS_TAG_DISABLE) == 0));
2702 
2703 		ASSERT(((flags & MAC_UNICAST_STRIP_DISABLE) != 0 &&
2704 		    (mcip->mci_state_flags & MCIS_STRIP_DISABLE) != 0) ||
2705 		    ((flags & MAC_UNICAST_STRIP_DISABLE) == 0 &&
2706 		    (mcip->mci_state_flags & MCIS_STRIP_DISABLE) == 0));
2707 
2708 		ASSERT(((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) != 0 &&
2709 		    (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK) != 0) ||
2710 		    ((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) == 0 &&
2711 		    (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK) == 0));
2712 
2713 		/*
2714 		 * Make sure the client is consistent about its requests
2715 		 * for MAC addresses. I.e. all requests from the clients
2716 		 * must have the MAC_UNICAST_HW flag set or clear.
2717 		 */
2718 		if ((mcip->mci_state_flags & MCIS_UNICAST_HW) != 0 &&
2719 		    !is_unicast_hw ||
2720 		    (mcip->mci_state_flags & MCIS_UNICAST_HW) == 0 &&
2721 		    is_unicast_hw) {
2722 			err = EINVAL;
2723 			goto bail_out;
2724 		}
2725 	}
2726 	/*
2727 	 * Make sure the MAC address is not already used by
2728 	 * another MAC client defined on top of the same
2729 	 * underlying NIC. Unless we have MAC_CLIENT_FLAGS_MULTI_PRIMARY
2730 	 * set when we allow a passive client to be present which will
2731 	 * be activated when the currently active client goes away - this
2732 	 * works only with primary addresses.
2733 	 */
2734 	if ((check_dups || is_primary || is_vnic_primary) &&
2735 	    mac_addr_in_use(mip, mac_addr, vid)) {
2736 		/*
2737 		 * Must have set the multiple primary address flag when
2738 		 * we did a mac_client_open AND this should be a primary
2739 		 * MAC client AND there should not already be a passive
2740 		 * primary. If all is true then we let this succeed
2741 		 * even if the address is a dup.
2742 		 */
2743 		if ((mcip->mci_flags & MAC_CLIENT_FLAGS_MULTI_PRIMARY) == 0 ||
2744 		    (mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY) == 0 ||
2745 		    mac_get_passive_primary_client(mip) != NULL) {
2746 			*diag = MAC_DIAG_MACADDR_INUSE;
2747 			err = EEXIST;
2748 			goto bail_out;
2749 		}
2750 		ASSERT((mcip->mci_flags &
2751 		    MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) == 0);
2752 		mcip->mci_flags |= MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
2753 		kmem_free(mrp, sizeof (*mrp));
2754 
2755 		/*
2756 		 * Stash the unicast address handle, we will use it when
2757 		 * we set up the passive client.
2758 		 */
2759 		mcip->mci_p_unicast_list = muip;
2760 		*mah = (mac_unicast_handle_t)muip;
2761 		return (0);
2762 	}
2763 
2764 	err = mac_client_datapath_setup(mcip, vid, mac_addr, mrp,
2765 	    is_primary || is_vnic_primary, muip);
2766 	if (err != 0)
2767 		goto bail_out;
2768 
2769 	kmem_free(mrp, sizeof (*mrp));
2770 	*mah = (mac_unicast_handle_t)muip;
2771 	return (0);
2772 
2773 bail_out:
2774 	if (fastpath_disabled)
2775 		mac_fastpath_enable((mac_handle_t)mip);
2776 	if (mcip->mci_state_flags & MCIS_EXCLUSIVE) {
2777 		mip->mi_state_flags &= ~MIS_EXCLUSIVE;
2778 		if (mip->mi_state_flags & MIS_LEGACY) {
2779 			mip->mi_capab_legacy.ml_active_clear(
2780 			    mip->mi_driver);
2781 		}
2782 	}
2783 	kmem_free(mrp, sizeof (*mrp));
2784 	kmem_free(muip, sizeof (mac_unicast_impl_t));
2785 	return (err);
2786 }
2787 
2788 /*
2789  * Wrapper function to mac_unicast_add when we want to have the same mac
2790  * client open for two instances, one that is currently active and another
2791  * that will become active when the current one is removed. In this case
2792  * mac_unicast_add will return EGAIN and we will save the rx function and
2793  * arg which will be used when we activate the passive client in
2794  * mac_unicast_remove.
2795  */
2796 int
2797 mac_unicast_add_set_rx(mac_client_handle_t mch, uint8_t *mac_addr,
2798     uint16_t flags, mac_unicast_handle_t *mah,  uint16_t vid, mac_diag_t *diag,
2799     mac_rx_t rx_fn, void *arg)
2800 {
2801 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2802 	uint_t			err;
2803 
2804 	err = mac_unicast_add(mch, mac_addr, flags, mah, vid, diag);
2805 	if (err != 0 && err != EAGAIN)
2806 		return (err);
2807 	if (err == EAGAIN) {
2808 		if (rx_fn != NULL) {
2809 			mcip->mci_rx_p_fn = rx_fn;
2810 			mcip->mci_rx_p_arg = arg;
2811 		}
2812 		return (0);
2813 	}
2814 	if (rx_fn != NULL)
2815 		mac_rx_set(mch, rx_fn, arg);
2816 	return (err);
2817 }
2818 
2819 int
2820 mac_unicast_add(mac_client_handle_t mch, uint8_t *mac_addr, uint16_t flags,
2821     mac_unicast_handle_t *mah, uint16_t vid, mac_diag_t *diag)
2822 {
2823 	mac_impl_t *mip = ((mac_client_impl_t *)mch)->mci_mip;
2824 	uint_t err;
2825 
2826 	i_mac_perim_enter(mip);
2827 	err = i_mac_unicast_add(mch, mac_addr, flags, mah, vid, diag);
2828 	i_mac_perim_exit(mip);
2829 
2830 	return (err);
2831 }
2832 
2833 static void
2834 mac_client_datapath_teardown(mac_client_handle_t mch, mac_unicast_impl_t *muip,
2835     flow_entry_t *flent)
2836 {
2837 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2838 	mac_impl_t		*mip = mcip->mci_mip;
2839 	boolean_t		no_unicast;
2840 
2841 	/*
2842 	 * If we have not added a unicast address for this MAC client, just
2843 	 * teardown the datapath.
2844 	 */
2845 	no_unicast = mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR;
2846 
2847 	if (!no_unicast) {
2848 		/*
2849 		 * We would have initialized subflows etc. only if we brought
2850 		 * up the primary client and set the unicast unicast address
2851 		 * etc. Deactivate the flows. The flow entry will be removed
2852 		 * from the active flow tables, and the associated SRS,
2853 		 * softrings etc will be deleted. But the flow entry itself
2854 		 * won't be destroyed, instead it will continue to be archived
2855 		 * off the  the global flow hash list, for a possible future
2856 		 * activation when say IP is plumbed again.
2857 		 */
2858 		mac_link_release_flows(mch);
2859 	}
2860 	mip->mi_nactiveclients--;
2861 	mac_update_single_active_client(mip);
2862 
2863 	/* Tear down the data path */
2864 	mac_datapath_teardown(mcip, mcip->mci_flent, SRST_LINK);
2865 
2866 	/*
2867 	 * Prevent any future access to the flow entry through the mci_flent
2868 	 * pointer by setting the mci_flent to NULL. Access to mci_flent in
2869 	 * mac_bcast_send is also under mi_rw_lock.
2870 	 */
2871 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
2872 	flent = mcip->mci_flent;
2873 	mac_client_remove_flow_from_list(mcip, flent);
2874 
2875 	if (mcip->mci_state_flags & MCIS_DESC_LOGGED)
2876 		mcip->mci_state_flags &= ~MCIS_DESC_LOGGED;
2877 
2878 	/*
2879 	 * This is the last unicast address being removed and there shouldn't
2880 	 * be any outbound data threads at this point coming down from mac
2881 	 * clients. We have waited for the data threads to finish before
2882 	 * starting dld_str_detach. Non-data threads must access TX SRS
2883 	 * under mi_rw_lock.
2884 	 */
2885 	rw_exit(&mip->mi_rw_lock);
2886 
2887 	/*
2888 	 * Don't use FLOW_MARK with FE_MC_NO_DATAPATH, as the flow might
2889 	 * contain other flags, such as FE_CONDEMNED, which we need to
2890 	 * cleared. We don't call mac_flow_cleanup() for this unicast
2891 	 * flow as we have a already cleaned up SRSs etc. (via the teadown
2892 	 * path). We just clear the stats and reset the initial callback
2893 	 * function, the rest will be set when we call mac_flow_create,
2894 	 * if at all.
2895 	 */
2896 	mutex_enter(&flent->fe_lock);
2897 	ASSERT(flent->fe_refcnt == 1 && flent->fe_mbg == NULL &&
2898 	    flent->fe_tx_srs == NULL && flent->fe_rx_srs_cnt == 0);
2899 	flent->fe_flags = FE_MC_NO_DATAPATH;
2900 	flow_stat_destroy(flent);
2901 	mac_misc_stat_delete(flent);
2902 
2903 	/* Initialize the receiver function to a safe routine */
2904 	flent->fe_cb_fn = (flow_fn_t)mac_pkt_drop;
2905 	flent->fe_cb_arg1 = NULL;
2906 	flent->fe_cb_arg2 = NULL;
2907 
2908 	flent->fe_index = -1;
2909 	mutex_exit(&flent->fe_lock);
2910 
2911 	if (mip->mi_type->mt_brdcst_addr != NULL) {
2912 		ASSERT(muip != NULL || no_unicast);
2913 		mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr,
2914 		    muip != NULL ? muip->mui_vid : VLAN_ID_NONE);
2915 	}
2916 
2917 	if (mip->mi_nactiveclients == 1) {
2918 		mac_capab_update((mac_handle_t)mip);
2919 		mac_virtual_link_update(mip);
2920 	}
2921 
2922 	if (mcip->mci_state_flags & MCIS_EXCLUSIVE) {
2923 		mip->mi_state_flags &= ~MIS_EXCLUSIVE;
2924 
2925 		if (mip->mi_state_flags & MIS_LEGACY)
2926 			mip->mi_capab_legacy.ml_active_clear(mip->mi_driver);
2927 	}
2928 
2929 	mcip->mci_state_flags &= ~MCIS_UNICAST_HW;
2930 
2931 	if (mcip->mci_state_flags & MCIS_TAG_DISABLE)
2932 		mcip->mci_state_flags &= ~MCIS_TAG_DISABLE;
2933 
2934 	if (mcip->mci_state_flags & MCIS_STRIP_DISABLE)
2935 		mcip->mci_state_flags &= ~MCIS_STRIP_DISABLE;
2936 
2937 	if (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK)
2938 		mcip->mci_state_flags &= ~MCIS_DISABLE_TX_VID_CHECK;
2939 
2940 	if (muip != NULL)
2941 		kmem_free(muip, sizeof (mac_unicast_impl_t));
2942 	mac_protect_cancel_timer(mcip);
2943 	mac_protect_flush_dhcp(mcip);
2944 
2945 	bzero(&mcip->mci_misc_stat, sizeof (mcip->mci_misc_stat));
2946 	/*
2947 	 * Disable fastpath if this is a VNIC or a VLAN.
2948 	 */
2949 	if (mcip->mci_state_flags & MCIS_IS_VNIC)
2950 		mac_fastpath_enable((mac_handle_t)mip);
2951 	mac_stop((mac_handle_t)mip);
2952 }
2953 
2954 /*
2955  * Remove a MAC address which was previously added by mac_unicast_add().
2956  */
2957 int
2958 mac_unicast_remove(mac_client_handle_t mch, mac_unicast_handle_t mah)
2959 {
2960 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
2961 	mac_unicast_impl_t *muip = (mac_unicast_impl_t *)mah;
2962 	mac_unicast_impl_t *pre;
2963 	mac_impl_t *mip = mcip->mci_mip;
2964 	flow_entry_t		*flent;
2965 	uint16_t mui_vid;
2966 
2967 	i_mac_perim_enter(mip);
2968 	if (mcip->mci_flags & MAC_CLIENT_FLAGS_VNIC_PRIMARY) {
2969 		/*
2970 		 * Called made by the upper MAC client of a VNIC.
2971 		 * There's nothing much to do, the unicast address will
2972 		 * be removed by the VNIC driver when the VNIC is deleted,
2973 		 * but let's ensure that all our transmit is done before
2974 		 * the client does a mac_client_stop lest it trigger an
2975 		 * assert in the driver.
2976 		 */
2977 		ASSERT(muip->mui_vid == 0);
2978 
2979 		mac_tx_client_flush(mcip);
2980 
2981 		if ((mcip->mci_flags & MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
2982 			mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
2983 			if (mcip->mci_rx_p_fn != NULL) {
2984 				mac_rx_set(mch, mcip->mci_rx_p_fn,
2985 				    mcip->mci_rx_p_arg);
2986 				mcip->mci_rx_p_fn = NULL;
2987 				mcip->mci_rx_p_arg = NULL;
2988 			}
2989 			kmem_free(muip, sizeof (mac_unicast_impl_t));
2990 			i_mac_perim_exit(mip);
2991 			return (0);
2992 		}
2993 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_VNIC_PRIMARY;
2994 
2995 		if (mcip->mci_state_flags & MCIS_TAG_DISABLE)
2996 			mcip->mci_state_flags &= ~MCIS_TAG_DISABLE;
2997 
2998 		if (mcip->mci_state_flags & MCIS_STRIP_DISABLE)
2999 			mcip->mci_state_flags &= ~MCIS_STRIP_DISABLE;
3000 
3001 		if (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK)
3002 			mcip->mci_state_flags &= ~MCIS_DISABLE_TX_VID_CHECK;
3003 
3004 		kmem_free(muip, sizeof (mac_unicast_impl_t));
3005 		i_mac_perim_exit(mip);
3006 		return (0);
3007 	}
3008 
3009 	ASSERT(muip != NULL);
3010 
3011 	/*
3012 	 * We are removing a passive client, we haven't setup the datapath
3013 	 * for this yet, so nothing much to do.
3014 	 */
3015 	if ((mcip->mci_flags & MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
3016 
3017 		ASSERT((mcip->mci_flent->fe_flags & FE_MC_NO_DATAPATH) != 0);
3018 		ASSERT(mcip->mci_p_unicast_list == muip);
3019 
3020 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
3021 
3022 		mcip->mci_p_unicast_list = NULL;
3023 		mcip->mci_rx_p_fn = NULL;
3024 		mcip->mci_rx_p_arg = NULL;
3025 
3026 		mcip->mci_state_flags &= ~MCIS_UNICAST_HW;
3027 
3028 		if (mcip->mci_state_flags & MCIS_TAG_DISABLE)
3029 			mcip->mci_state_flags &= ~MCIS_TAG_DISABLE;
3030 
3031 		if (mcip->mci_state_flags & MCIS_STRIP_DISABLE)
3032 			mcip->mci_state_flags &= ~MCIS_STRIP_DISABLE;
3033 
3034 		if (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK)
3035 			mcip->mci_state_flags &= ~MCIS_DISABLE_TX_VID_CHECK;
3036 
3037 		kmem_free(muip, sizeof (mac_unicast_impl_t));
3038 		i_mac_perim_exit(mip);
3039 		return (0);
3040 	}
3041 	/*
3042 	 * Remove the VID from the list of client's VIDs.
3043 	 */
3044 	pre = mcip->mci_unicast_list;
3045 	if (muip == pre) {
3046 		mcip->mci_unicast_list = muip->mui_next;
3047 	} else {
3048 		while ((pre->mui_next != NULL) && (pre->mui_next != muip))
3049 			pre = pre->mui_next;
3050 		ASSERT(pre->mui_next == muip);
3051 		rw_enter(&mcip->mci_rw_lock, RW_WRITER);
3052 		pre->mui_next = muip->mui_next;
3053 		rw_exit(&mcip->mci_rw_lock);
3054 	}
3055 
3056 	if (!mac_client_single_rcvr(mcip)) {
3057 		/*
3058 		 * This MAC client is shared by more than one unicast
3059 		 * addresses, so we will just remove the flent
3060 		 * corresponding to the address being removed. We don't invoke
3061 		 * mac_rx_classify_flow_rem() since the additional flow is
3062 		 * not associated with its own separate set of SRS and rings,
3063 		 * and these constructs are still needed for the remaining
3064 		 * flows.
3065 		 */
3066 		flent = mac_client_get_flow(mcip, muip);
3067 		ASSERT(flent != NULL);
3068 
3069 		/*
3070 		 * The first one is disappearing, need to make sure
3071 		 * we replace it with another from the list of
3072 		 * shared clients.
3073 		 */
3074 		if (flent == mcip->mci_flent)
3075 			flent = mac_client_swap_mciflent(mcip);
3076 		mac_client_remove_flow_from_list(mcip, flent);
3077 		mac_flow_remove(mip->mi_flow_tab, flent, B_FALSE);
3078 		mac_flow_wait(flent, FLOW_DRIVER_UPCALL);
3079 
3080 		/*
3081 		 * The multicast groups that were added by the client so
3082 		 * far must be removed from the brodcast domain corresponding
3083 		 * to the VID being removed.
3084 		 */
3085 		mac_client_bcast_refresh(mcip, mac_client_update_mcast,
3086 		    (void *)flent, B_FALSE);
3087 
3088 		if (mip->mi_type->mt_brdcst_addr != NULL) {
3089 			mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr,
3090 			    muip->mui_vid);
3091 		}
3092 
3093 		FLOW_FINAL_REFRELE(flent);
3094 		ASSERT(!(mcip->mci_state_flags & MCIS_EXCLUSIVE));
3095 		/*
3096 		 * Enable fastpath if this is a VNIC or a VLAN.
3097 		 */
3098 		if (mcip->mci_state_flags & MCIS_IS_VNIC)
3099 			mac_fastpath_enable((mac_handle_t)mip);
3100 		mac_stop((mac_handle_t)mip);
3101 		i_mac_perim_exit(mip);
3102 		return (0);
3103 	}
3104 
3105 	mui_vid = muip->mui_vid;
3106 	mac_client_datapath_teardown(mch, muip, flent);
3107 
3108 	if ((mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY) && mui_vid == 0) {
3109 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PRIMARY;
3110 	} else {
3111 		i_mac_perim_exit(mip);
3112 		return (0);
3113 	}
3114 
3115 	/*
3116 	 * If we are removing the primary, check if we have a passive primary
3117 	 * client that we need to activate now.
3118 	 */
3119 	mcip = mac_get_passive_primary_client(mip);
3120 	if (mcip != NULL) {
3121 		mac_resource_props_t	*mrp;
3122 		mac_unicast_impl_t	*muip;
3123 
3124 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
3125 		mrp = kmem_zalloc(sizeof (*mrp), KM_SLEEP);
3126 
3127 		/*
3128 		 * Apply the property cached in the mac_impl_t to the
3129 		 * primary mac client.
3130 		 */
3131 		mac_get_resources((mac_handle_t)mip, mrp);
3132 		(void) mac_client_set_resources(mch, mrp);
3133 		ASSERT(mcip->mci_p_unicast_list != NULL);
3134 		muip = mcip->mci_p_unicast_list;
3135 		mcip->mci_p_unicast_list = NULL;
3136 		if (mac_client_datapath_setup(mcip, VLAN_ID_NONE,
3137 		    mip->mi_addr, mrp, B_TRUE, muip) == 0) {
3138 			if (mcip->mci_rx_p_fn != NULL) {
3139 				mac_rx_set(mch, mcip->mci_rx_p_fn,
3140 				    mcip->mci_rx_p_arg);
3141 				mcip->mci_rx_p_fn = NULL;
3142 				mcip->mci_rx_p_arg = NULL;
3143 			}
3144 		} else {
3145 			kmem_free(muip, sizeof (mac_unicast_impl_t));
3146 		}
3147 		kmem_free(mrp, sizeof (*mrp));
3148 	}
3149 	i_mac_perim_exit(mip);
3150 	return (0);
3151 }
3152 
3153 /*
3154  * Multicast add function invoked by MAC clients.
3155  */
3156 int
3157 mac_multicast_add(mac_client_handle_t mch, const uint8_t *addr)
3158 {
3159 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3160 	mac_impl_t		*mip = mcip->mci_mip;
3161 	flow_entry_t		*flent = mcip->mci_flent_list;
3162 	flow_entry_t		*prev_fe = NULL;
3163 	uint16_t		vid;
3164 	int			err = 0;
3165 
3166 	/* Verify the address is a valid multicast address */
3167 	if ((err = mip->mi_type->mt_ops.mtops_multicst_verify(addr,
3168 	    mip->mi_pdata)) != 0)
3169 		return (err);
3170 
3171 	i_mac_perim_enter(mip);
3172 	while (flent != NULL) {
3173 		vid = i_mac_flow_vid(flent);
3174 
3175 		err = mac_bcast_add((mac_client_impl_t *)mch, addr, vid,
3176 		    MAC_ADDRTYPE_MULTICAST);
3177 		if (err != 0)
3178 			break;
3179 		prev_fe = flent;
3180 		flent = flent->fe_client_next;
3181 	}
3182 
3183 	/*
3184 	 * If we failed adding, then undo all, rather than partial
3185 	 * success.
3186 	 */
3187 	if (flent != NULL && prev_fe != NULL) {
3188 		flent = mcip->mci_flent_list;
3189 		while (flent != prev_fe->fe_client_next) {
3190 			vid = i_mac_flow_vid(flent);
3191 			mac_bcast_delete((mac_client_impl_t *)mch, addr, vid);
3192 			flent = flent->fe_client_next;
3193 		}
3194 	}
3195 	i_mac_perim_exit(mip);
3196 	return (err);
3197 }
3198 
3199 /*
3200  * Multicast delete function invoked by MAC clients.
3201  */
3202 void
3203 mac_multicast_remove(mac_client_handle_t mch, const uint8_t *addr)
3204 {
3205 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3206 	mac_impl_t		*mip = mcip->mci_mip;
3207 	flow_entry_t		*flent;
3208 	uint16_t		vid;
3209 
3210 	i_mac_perim_enter(mip);
3211 	for (flent = mcip->mci_flent_list; flent != NULL;
3212 	    flent = flent->fe_client_next) {
3213 		vid = i_mac_flow_vid(flent);
3214 		mac_bcast_delete((mac_client_impl_t *)mch, addr, vid);
3215 	}
3216 	i_mac_perim_exit(mip);
3217 }
3218 
3219 /*
3220  * When a MAC client desires to capture packets on an interface,
3221  * it registers a promiscuous call back with mac_promisc_add().
3222  * There are three types of promiscuous callbacks:
3223  *
3224  * * MAC_CLIENT_PROMISC_ALL
3225  *   Captures all packets sent and received by the MAC client,
3226  *   the physical interface, as well as all other MAC clients
3227  *   defined on top of the same MAC.
3228  *
3229  * * MAC_CLIENT_PROMISC_FILTERED
3230  *   Captures all packets sent and received by the MAC client,
3231  *   plus all multicast traffic sent and received by the phyisical
3232  *   interface and the other MAC clients.
3233  *
3234  * * MAC_CLIENT_PROMISC_MULTI
3235  *   Captures all broadcast and multicast packets sent and
3236  *   received by the MAC clients as well as the physical interface.
3237  *
3238  * In all cases, the underlying MAC is put in promiscuous mode.
3239  */
3240 int
3241 mac_promisc_add(mac_client_handle_t mch, mac_client_promisc_type_t type,
3242     mac_rx_t fn, void *arg, mac_promisc_handle_t *mphp, uint16_t flags)
3243 {
3244 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3245 	mac_impl_t *mip = mcip->mci_mip;
3246 	mac_promisc_impl_t *mpip;
3247 	mac_cb_info_t	*mcbi;
3248 	int rc;
3249 
3250 	i_mac_perim_enter(mip);
3251 
3252 	if ((rc = mac_start((mac_handle_t)mip)) != 0) {
3253 		i_mac_perim_exit(mip);
3254 		return (rc);
3255 	}
3256 
3257 	if ((mcip->mci_state_flags & MCIS_IS_VNIC) &&
3258 	    type == MAC_CLIENT_PROMISC_ALL) {
3259 		/*
3260 		 * The function is being invoked by the upper MAC client
3261 		 * of a VNIC. The VNIC should only see the traffic
3262 		 * it is entitled to.
3263 		 */
3264 		type = MAC_CLIENT_PROMISC_FILTERED;
3265 	}
3266 
3267 
3268 	/*
3269 	 * Turn on promiscuous mode for the underlying NIC.
3270 	 * This is needed even for filtered callbacks which
3271 	 * expect to receive all multicast traffic on the wire.
3272 	 *
3273 	 * Physical promiscuous mode should not be turned on if
3274 	 * MAC_PROMISC_FLAGS_NO_PHYS is set.
3275 	 */
3276 	if ((flags & MAC_PROMISC_FLAGS_NO_PHYS) == 0) {
3277 		if ((rc = i_mac_promisc_set(mip, B_TRUE)) != 0) {
3278 			mac_stop((mac_handle_t)mip);
3279 			i_mac_perim_exit(mip);
3280 			return (rc);
3281 		}
3282 	}
3283 
3284 	mpip = kmem_cache_alloc(mac_promisc_impl_cache, KM_SLEEP);
3285 
3286 	mpip->mpi_type = type;
3287 	mpip->mpi_fn = fn;
3288 	mpip->mpi_arg = arg;
3289 	mpip->mpi_mcip = mcip;
3290 	mpip->mpi_no_tx_loop = ((flags & MAC_PROMISC_FLAGS_NO_TX_LOOP) != 0);
3291 	mpip->mpi_no_phys = ((flags & MAC_PROMISC_FLAGS_NO_PHYS) != 0);
3292 	mpip->mpi_strip_vlan_tag =
3293 	    ((flags & MAC_PROMISC_FLAGS_VLAN_TAG_STRIP) != 0);
3294 	mpip->mpi_no_copy = ((flags & MAC_PROMISC_FLAGS_NO_COPY) != 0);
3295 
3296 	mcbi = &mip->mi_promisc_cb_info;
3297 	mutex_enter(mcbi->mcbi_lockp);
3298 
3299 	mac_callback_add(&mip->mi_promisc_cb_info, &mcip->mci_promisc_list,
3300 	    &mpip->mpi_mci_link);
3301 	mac_callback_add(&mip->mi_promisc_cb_info, &mip->mi_promisc_list,
3302 	    &mpip->mpi_mi_link);
3303 
3304 	mutex_exit(mcbi->mcbi_lockp);
3305 
3306 	*mphp = (mac_promisc_handle_t)mpip;
3307 
3308 	if (mcip->mci_state_flags & MCIS_IS_VNIC) {
3309 		mac_impl_t *umip = mcip->mci_upper_mip;
3310 
3311 		ASSERT(umip != NULL);
3312 		mac_vnic_secondary_update(umip);
3313 	}
3314 
3315 	i_mac_perim_exit(mip);
3316 
3317 	return (0);
3318 }
3319 
3320 /*
3321  * Remove a multicast address previously aded through mac_promisc_add().
3322  */
3323 void
3324 mac_promisc_remove(mac_promisc_handle_t mph)
3325 {
3326 	mac_promisc_impl_t *mpip = (mac_promisc_impl_t *)mph;
3327 	mac_client_impl_t *mcip = mpip->mpi_mcip;
3328 	mac_impl_t *mip = mcip->mci_mip;
3329 	mac_cb_info_t *mcbi;
3330 	int rv;
3331 
3332 	i_mac_perim_enter(mip);
3333 
3334 	/*
3335 	 * Even if the device can't be reset into normal mode, we still
3336 	 * need to clear the client promisc callbacks. The client may want
3337 	 * to close the mac end point and we can't have stale callbacks.
3338 	 */
3339 	if (!(mpip->mpi_no_phys)) {
3340 		if ((rv = i_mac_promisc_set(mip, B_FALSE)) != 0) {
3341 			cmn_err(CE_WARN, "%s: failed to switch OFF promiscuous"
3342 			    " mode because of error 0x%x", mip->mi_name, rv);
3343 		}
3344 	}
3345 	mcbi = &mip->mi_promisc_cb_info;
3346 	mutex_enter(mcbi->mcbi_lockp);
3347 	if (mac_callback_remove(mcbi, &mip->mi_promisc_list,
3348 	    &mpip->mpi_mi_link)) {
3349 		VERIFY(mac_callback_remove(&mip->mi_promisc_cb_info,
3350 		    &mcip->mci_promisc_list, &mpip->mpi_mci_link));
3351 		kmem_cache_free(mac_promisc_impl_cache, mpip);
3352 	} else {
3353 		mac_callback_remove_wait(&mip->mi_promisc_cb_info);
3354 	}
3355 
3356 	if (mcip->mci_state_flags & MCIS_IS_VNIC) {
3357 		mac_impl_t *umip = mcip->mci_upper_mip;
3358 
3359 		ASSERT(umip != NULL);
3360 		mac_vnic_secondary_update(umip);
3361 	}
3362 
3363 	mutex_exit(mcbi->mcbi_lockp);
3364 	mac_stop((mac_handle_t)mip);
3365 
3366 	i_mac_perim_exit(mip);
3367 }
3368 
3369 /*
3370  * Reference count the number of active Tx threads. MCI_TX_QUIESCE indicates
3371  * that a control operation wants to quiesce the Tx data flow in which case
3372  * we return an error. Holding any of the per cpu locks ensures that the
3373  * mci_tx_flag won't change.
3374  *
3375  * 'CPU' must be accessed just once and used to compute the index into the
3376  * percpu array, and that index must be used for the entire duration of the
3377  * packet send operation. Note that the thread may be preempted and run on
3378  * another cpu any time and so we can't use 'CPU' more than once for the
3379  * operation.
3380  */
3381 #define	MAC_TX_TRY_HOLD(mcip, mytx, error)				\
3382 {									\
3383 	(error) = 0;							\
3384 	(mytx) = &(mcip)->mci_tx_pcpu[CPU->cpu_seqid & mac_tx_percpu_cnt]; \
3385 	mutex_enter(&(mytx)->pcpu_tx_lock);				\
3386 	if (!((mcip)->mci_tx_flag & MCI_TX_QUIESCE)) {			\
3387 		(mytx)->pcpu_tx_refcnt++;				\
3388 	} else {							\
3389 		(error) = -1;						\
3390 	}								\
3391 	mutex_exit(&(mytx)->pcpu_tx_lock);				\
3392 }
3393 
3394 /*
3395  * Release the reference. If needed, signal any control operation waiting
3396  * for Tx quiescence. The wait and signal are always done using the
3397  * mci_tx_pcpu[0]'s lock
3398  */
3399 #define	MAC_TX_RELE(mcip, mytx) {					\
3400 	mutex_enter(&(mytx)->pcpu_tx_lock);				\
3401 	if (--(mytx)->pcpu_tx_refcnt == 0 &&				\
3402 	    (mcip)->mci_tx_flag & MCI_TX_QUIESCE) {			\
3403 		mutex_exit(&(mytx)->pcpu_tx_lock);			\
3404 		mutex_enter(&(mcip)->mci_tx_pcpu[0].pcpu_tx_lock);	\
3405 		cv_signal(&(mcip)->mci_tx_cv);				\
3406 		mutex_exit(&(mcip)->mci_tx_pcpu[0].pcpu_tx_lock);	\
3407 	} else {							\
3408 		mutex_exit(&(mytx)->pcpu_tx_lock);			\
3409 	}								\
3410 }
3411 
3412 /*
3413  * Send function invoked by MAC clients.
3414  */
3415 mac_tx_cookie_t
3416 mac_tx(mac_client_handle_t mch, mblk_t *mp_chain, uintptr_t hint,
3417     uint16_t flag, mblk_t **ret_mp)
3418 {
3419 	mac_tx_cookie_t		cookie = NULL;
3420 	int			error;
3421 	mac_tx_percpu_t		*mytx;
3422 	mac_soft_ring_set_t	*srs;
3423 	flow_entry_t		*flent;
3424 	boolean_t		is_subflow = B_FALSE;
3425 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3426 	mac_impl_t		*mip = mcip->mci_mip;
3427 	mac_srs_tx_t		*srs_tx;
3428 
3429 	/*
3430 	 * Check whether the active Tx threads count is bumped already.
3431 	 */
3432 	if (!(flag & MAC_TX_NO_HOLD)) {
3433 		MAC_TX_TRY_HOLD(mcip, mytx, error);
3434 		if (error != 0) {
3435 			freemsgchain(mp_chain);
3436 			return (NULL);
3437 		}
3438 	}
3439 
3440 	/*
3441 	 * If mac protection is enabled, only the permissible packets will be
3442 	 * returned by mac_protect_check().
3443 	 */
3444 	if ((mcip->mci_flent->
3445 	    fe_resource_props.mrp_mask & MRP_PROTECT) != 0 &&
3446 	    (mp_chain = mac_protect_check(mch, mp_chain)) == NULL)
3447 		goto done;
3448 
3449 	if (mcip->mci_subflow_tab != NULL &&
3450 	    mcip->mci_subflow_tab->ft_flow_count > 0 &&
3451 	    mac_flow_lookup(mcip->mci_subflow_tab, mp_chain,
3452 	    FLOW_OUTBOUND, &flent) == 0) {
3453 		/*
3454 		 * The main assumption here is that if in the event
3455 		 * we get a chain, all the packets will be classified
3456 		 * to the same Flow/SRS. If this changes for any
3457 		 * reason, the following logic should change as well.
3458 		 * I suppose the fanout_hint also assumes this .
3459 		 */
3460 		ASSERT(flent != NULL);
3461 		is_subflow = B_TRUE;
3462 	} else {
3463 		flent = mcip->mci_flent;
3464 	}
3465 
3466 	srs = flent->fe_tx_srs;
3467 	/*
3468 	 * This is to avoid panics with PF_PACKET that can call mac_tx()
3469 	 * against an interface that is not capable of sending. A rewrite
3470 	 * of the mac datapath is required to remove this limitation.
3471 	 */
3472 	if (srs == NULL) {
3473 		freemsgchain(mp_chain);
3474 		goto done;
3475 	}
3476 
3477 	srs_tx = &srs->srs_tx;
3478 	if (srs_tx->st_mode == SRS_TX_DEFAULT &&
3479 	    (srs->srs_state & SRS_ENQUEUED) == 0 &&
3480 	    mip->mi_nactiveclients == 1 && mp_chain->b_next == NULL) {
3481 		uint64_t	obytes;
3482 
3483 		/*
3484 		 * Since dls always opens the underlying MAC, nclients equals
3485 		 * to 1 means that the only active client is dls itself acting
3486 		 * as a primary client of the MAC instance. Since dls will not
3487 		 * send tagged packets in that case, and dls is trusted to send
3488 		 * packets for its allowed VLAN(s), the VLAN tag insertion and
3489 		 * check is required only if nclients is greater than 1.
3490 		 */
3491 		if (mip->mi_nclients > 1) {
3492 			if (MAC_VID_CHECK_NEEDED(mcip)) {
3493 				int	err = 0;
3494 
3495 				MAC_VID_CHECK(mcip, mp_chain, err);
3496 				if (err != 0) {
3497 					freemsg(mp_chain);
3498 					mcip->mci_misc_stat.mms_txerrors++;
3499 					goto done;
3500 				}
3501 			}
3502 			if (MAC_TAG_NEEDED(mcip)) {
3503 				mp_chain = mac_add_vlan_tag(mp_chain, 0,
3504 				    mac_client_vid(mch));
3505 				if (mp_chain == NULL) {
3506 					mcip->mci_misc_stat.mms_txerrors++;
3507 					goto done;
3508 				}
3509 			}
3510 		}
3511 
3512 		obytes = (mp_chain->b_cont == NULL ? MBLKL(mp_chain) :
3513 		    msgdsize(mp_chain));
3514 
3515 		MAC_TX(mip, srs_tx->st_arg2, mp_chain, mcip);
3516 		if (mp_chain == NULL) {
3517 			cookie = NULL;
3518 			SRS_TX_STAT_UPDATE(srs, opackets, 1);
3519 			SRS_TX_STAT_UPDATE(srs, obytes, obytes);
3520 		} else {
3521 			mutex_enter(&srs->srs_lock);
3522 			cookie = mac_tx_srs_no_desc(srs, mp_chain,
3523 			    flag, ret_mp);
3524 			mutex_exit(&srs->srs_lock);
3525 		}
3526 	} else {
3527 		cookie = srs_tx->st_func(srs, mp_chain, hint, flag, ret_mp);
3528 	}
3529 
3530 done:
3531 	if (is_subflow)
3532 		FLOW_REFRELE(flent);
3533 
3534 	if (!(flag & MAC_TX_NO_HOLD))
3535 		MAC_TX_RELE(mcip, mytx);
3536 
3537 	return (cookie);
3538 }
3539 
3540 /*
3541  * mac_tx_is_blocked
3542  *
3543  * Given a cookie, it returns if the ring identified by the cookie is
3544  * flow-controlled or not. If NULL is passed in place of a cookie,
3545  * then it finds out if any of the underlying rings belonging to the
3546  * SRS is flow controlled or not and returns that status.
3547  */
3548 /* ARGSUSED */
3549 boolean_t
3550 mac_tx_is_flow_blocked(mac_client_handle_t mch, mac_tx_cookie_t cookie)
3551 {
3552 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3553 	mac_soft_ring_set_t *mac_srs;
3554 	mac_soft_ring_t *sringp;
3555 	boolean_t blocked = B_FALSE;
3556 	mac_tx_percpu_t *mytx;
3557 	int err;
3558 	int i;
3559 
3560 	/*
3561 	 * Bump the reference count so that mac_srs won't be deleted.
3562 	 * If the client is currently quiesced and we failed to bump
3563 	 * the reference, return B_TRUE so that flow control stays
3564 	 * as enabled.
3565 	 *
3566 	 * Flow control will then be disabled once the client is no
3567 	 * longer quiesced.
3568 	 */
3569 	MAC_TX_TRY_HOLD(mcip, mytx, err);
3570 	if (err != 0)
3571 		return (B_TRUE);
3572 
3573 	if ((mac_srs = MCIP_TX_SRS(mcip)) == NULL) {
3574 		MAC_TX_RELE(mcip, mytx);
3575 		return (B_FALSE);
3576 	}
3577 
3578 	mutex_enter(&mac_srs->srs_lock);
3579 	/*
3580 	 * Only in the case of TX_FANOUT and TX_AGGR, the underlying
3581 	 * softring (s_ring_state) will have the HIWAT set. This is
3582 	 * the multiple Tx ring flow control case. For all other
3583 	 * case, SRS (srs_state) will store the condition.
3584 	 */
3585 	if (mac_srs->srs_tx.st_mode == SRS_TX_FANOUT ||
3586 	    mac_srs->srs_tx.st_mode == SRS_TX_AGGR) {
3587 		if (cookie != NULL) {
3588 			sringp = (mac_soft_ring_t *)cookie;
3589 			mutex_enter(&sringp->s_ring_lock);
3590 			if (sringp->s_ring_state & S_RING_TX_HIWAT)
3591 				blocked = B_TRUE;
3592 			mutex_exit(&sringp->s_ring_lock);
3593 		} else {
3594 			for (i = 0; i < mac_srs->srs_tx_ring_count; i++) {
3595 				sringp = mac_srs->srs_tx_soft_rings[i];
3596 				mutex_enter(&sringp->s_ring_lock);
3597 				if (sringp->s_ring_state & S_RING_TX_HIWAT) {
3598 					blocked = B_TRUE;
3599 					mutex_exit(&sringp->s_ring_lock);
3600 					break;
3601 				}
3602 				mutex_exit(&sringp->s_ring_lock);
3603 			}
3604 		}
3605 	} else {
3606 		blocked = (mac_srs->srs_state & SRS_TX_HIWAT);
3607 	}
3608 	mutex_exit(&mac_srs->srs_lock);
3609 	MAC_TX_RELE(mcip, mytx);
3610 	return (blocked);
3611 }
3612 
3613 /*
3614  * Check if the MAC client is the primary MAC client.
3615  */
3616 boolean_t
3617 mac_is_primary_client(mac_client_impl_t *mcip)
3618 {
3619 	return (mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY);
3620 }
3621 
3622 void
3623 mac_ioctl(mac_handle_t mh, queue_t *wq, mblk_t *bp)
3624 {
3625 	mac_impl_t	*mip = (mac_impl_t *)mh;
3626 	int cmd = ((struct iocblk *)bp->b_rptr)->ioc_cmd;
3627 
3628 	if ((cmd == ND_GET && (mip->mi_callbacks->mc_callbacks & MC_GETPROP)) ||
3629 	    (cmd == ND_SET && (mip->mi_callbacks->mc_callbacks & MC_SETPROP))) {
3630 		/*
3631 		 * If ndd props were registered, call them.
3632 		 * Note that ndd ioctls are Obsolete
3633 		 */
3634 		mac_ndd_ioctl(mip, wq, bp);
3635 		return;
3636 	}
3637 
3638 	/*
3639 	 * Call the driver to handle the ioctl.  The driver may not support
3640 	 * any ioctls, in which case we reply with a NAK on its behalf.
3641 	 */
3642 	if (mip->mi_callbacks->mc_callbacks & MC_IOCTL)
3643 		mip->mi_ioctl(mip->mi_driver, wq, bp);
3644 	else
3645 		miocnak(wq, bp, 0, EINVAL);
3646 }
3647 
3648 /*
3649  * Return the link state of the specified MAC instance.
3650  */
3651 link_state_t
3652 mac_link_get(mac_handle_t mh)
3653 {
3654 	return (((mac_impl_t *)mh)->mi_linkstate);
3655 }
3656 
3657 /*
3658  * Add a mac client specified notification callback. Please see the comments
3659  * above mac_callback_add() for general information about mac callback
3660  * addition/deletion in the presence of mac callback list walkers
3661  */
3662 mac_notify_handle_t
3663 mac_notify_add(mac_handle_t mh, mac_notify_t notify_fn, void *arg)
3664 {
3665 	mac_impl_t		*mip = (mac_impl_t *)mh;
3666 	mac_notify_cb_t		*mncb;
3667 	mac_cb_info_t		*mcbi;
3668 
3669 	/*
3670 	 * Allocate a notify callback structure, fill in the details and
3671 	 * use the mac callback list manipulation functions to chain into
3672 	 * the list of callbacks.
3673 	 */
3674 	mncb = kmem_zalloc(sizeof (mac_notify_cb_t), KM_SLEEP);
3675 	mncb->mncb_fn = notify_fn;
3676 	mncb->mncb_arg = arg;
3677 	mncb->mncb_mip = mip;
3678 	mncb->mncb_link.mcb_objp = mncb;
3679 	mncb->mncb_link.mcb_objsize = sizeof (mac_notify_cb_t);
3680 	mncb->mncb_link.mcb_flags = MCB_NOTIFY_CB_T;
3681 
3682 	mcbi = &mip->mi_notify_cb_info;
3683 
3684 	i_mac_perim_enter(mip);
3685 	mutex_enter(mcbi->mcbi_lockp);
3686 
3687 	mac_callback_add(&mip->mi_notify_cb_info, &mip->mi_notify_cb_list,
3688 	    &mncb->mncb_link);
3689 
3690 	mutex_exit(mcbi->mcbi_lockp);
3691 	i_mac_perim_exit(mip);
3692 	return ((mac_notify_handle_t)mncb);
3693 }
3694 
3695 void
3696 mac_notify_remove_wait(mac_handle_t mh)
3697 {
3698 	mac_impl_t	*mip = (mac_impl_t *)mh;
3699 	mac_cb_info_t	*mcbi = &mip->mi_notify_cb_info;
3700 
3701 	mutex_enter(mcbi->mcbi_lockp);
3702 	mac_callback_remove_wait(&mip->mi_notify_cb_info);
3703 	mutex_exit(mcbi->mcbi_lockp);
3704 }
3705 
3706 /*
3707  * Remove a mac client specified notification callback
3708  */
3709 int
3710 mac_notify_remove(mac_notify_handle_t mnh, boolean_t wait)
3711 {
3712 	mac_notify_cb_t	*mncb = (mac_notify_cb_t *)mnh;
3713 	mac_impl_t	*mip = mncb->mncb_mip;
3714 	mac_cb_info_t	*mcbi;
3715 	int		err = 0;
3716 
3717 	mcbi = &mip->mi_notify_cb_info;
3718 
3719 	i_mac_perim_enter(mip);
3720 	mutex_enter(mcbi->mcbi_lockp);
3721 
3722 	ASSERT(mncb->mncb_link.mcb_objp == mncb);
3723 	/*
3724 	 * If there aren't any list walkers, the remove would succeed
3725 	 * inline, else we wait for the deferred remove to complete
3726 	 */
3727 	if (mac_callback_remove(&mip->mi_notify_cb_info,
3728 	    &mip->mi_notify_cb_list, &mncb->mncb_link)) {
3729 		kmem_free(mncb, sizeof (mac_notify_cb_t));
3730 	} else {
3731 		err = EBUSY;
3732 	}
3733 
3734 	mutex_exit(mcbi->mcbi_lockp);
3735 	i_mac_perim_exit(mip);
3736 
3737 	/*
3738 	 * If we failed to remove the notification callback and "wait" is set
3739 	 * to be B_TRUE, wait for the callback to finish after we exit the
3740 	 * mac perimeter.
3741 	 */
3742 	if (err != 0 && wait) {
3743 		mac_notify_remove_wait((mac_handle_t)mip);
3744 		return (0);
3745 	}
3746 
3747 	return (err);
3748 }
3749 
3750 /*
3751  * Associate resource management callbacks with the specified MAC
3752  * clients.
3753  */
3754 
3755 void
3756 mac_resource_set_common(mac_client_handle_t mch, mac_resource_add_t add,
3757     mac_resource_remove_t remove, mac_resource_quiesce_t quiesce,
3758     mac_resource_restart_t restart, mac_resource_bind_t bind,
3759     void *arg)
3760 {
3761 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3762 
3763 	mcip->mci_resource_add = add;
3764 	mcip->mci_resource_remove = remove;
3765 	mcip->mci_resource_quiesce = quiesce;
3766 	mcip->mci_resource_restart = restart;
3767 	mcip->mci_resource_bind = bind;
3768 	mcip->mci_resource_arg = arg;
3769 }
3770 
3771 void
3772 mac_resource_set(mac_client_handle_t mch, mac_resource_add_t add, void *arg)
3773 {
3774 	/* update the 'resource_add' callback */
3775 	mac_resource_set_common(mch, add, NULL, NULL, NULL, NULL, arg);
3776 }
3777 
3778 /*
3779  * Sets up the client resources and enable the polling interface over all the
3780  * SRS's and the soft rings of the client
3781  */
3782 void
3783 mac_client_poll_enable(mac_client_handle_t mch)
3784 {
3785 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3786 	mac_soft_ring_set_t	*mac_srs;
3787 	flow_entry_t		*flent;
3788 	int			i;
3789 
3790 	flent = mcip->mci_flent;
3791 	ASSERT(flent != NULL);
3792 
3793 	mcip->mci_state_flags |= MCIS_CLIENT_POLL_CAPABLE;
3794 	for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
3795 		mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
3796 		ASSERT(mac_srs->srs_mcip == mcip);
3797 		mac_srs_client_poll_enable(mcip, mac_srs);
3798 	}
3799 }
3800 
3801 /*
3802  * Tears down the client resources and disable the polling interface over all
3803  * the SRS's and the soft rings of the client
3804  */
3805 void
3806 mac_client_poll_disable(mac_client_handle_t mch)
3807 {
3808 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3809 	mac_soft_ring_set_t	*mac_srs;
3810 	flow_entry_t		*flent;
3811 	int			i;
3812 
3813 	flent = mcip->mci_flent;
3814 	ASSERT(flent != NULL);
3815 
3816 	mcip->mci_state_flags &= ~MCIS_CLIENT_POLL_CAPABLE;
3817 	for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
3818 		mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
3819 		ASSERT(mac_srs->srs_mcip == mcip);
3820 		mac_srs_client_poll_disable(mcip, mac_srs);
3821 	}
3822 }
3823 
3824 /*
3825  * Associate the CPUs specified by the given property with a MAC client.
3826  */
3827 int
3828 mac_cpu_set(mac_client_handle_t mch, mac_resource_props_t *mrp)
3829 {
3830 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3831 	mac_impl_t *mip = mcip->mci_mip;
3832 	int err = 0;
3833 
3834 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
3835 
3836 	if ((err = mac_validate_props(mcip->mci_state_flags & MCIS_IS_VNIC ?
3837 	    mcip->mci_upper_mip : mip, mrp)) != 0) {
3838 		return (err);
3839 	}
3840 	if (MCIP_DATAPATH_SETUP(mcip))
3841 		mac_flow_modify(mip->mi_flow_tab, mcip->mci_flent, mrp);
3842 
3843 	mac_update_resources(mrp, MCIP_RESOURCE_PROPS(mcip), B_FALSE);
3844 	return (0);
3845 }
3846 
3847 /*
3848  * Apply the specified properties to the specified MAC client.
3849  */
3850 int
3851 mac_client_set_resources(mac_client_handle_t mch, mac_resource_props_t *mrp)
3852 {
3853 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3854 	mac_impl_t *mip = mcip->mci_mip;
3855 	int err = 0;
3856 
3857 	i_mac_perim_enter(mip);
3858 
3859 	if ((mrp->mrp_mask & MRP_MAXBW) || (mrp->mrp_mask & MRP_PRIORITY)) {
3860 		err = mac_resource_ctl_set(mch, mrp);
3861 		if (err != 0)
3862 			goto done;
3863 	}
3864 
3865 	if (mrp->mrp_mask & (MRP_CPUS|MRP_POOL)) {
3866 		err = mac_cpu_set(mch, mrp);
3867 		if (err != 0)
3868 			goto done;
3869 	}
3870 
3871 	if (mrp->mrp_mask & MRP_PROTECT) {
3872 		err = mac_protect_set(mch, mrp);
3873 		if (err != 0)
3874 			goto done;
3875 	}
3876 
3877 	if ((mrp->mrp_mask & MRP_RX_RINGS) || (mrp->mrp_mask & MRP_TX_RINGS))
3878 		err = mac_resource_ctl_set(mch, mrp);
3879 
3880 done:
3881 	i_mac_perim_exit(mip);
3882 	return (err);
3883 }
3884 
3885 /*
3886  * Return the properties currently associated with the specified MAC client.
3887  */
3888 void
3889 mac_client_get_resources(mac_client_handle_t mch, mac_resource_props_t *mrp)
3890 {
3891 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3892 	mac_resource_props_t	*mcip_mrp = MCIP_RESOURCE_PROPS(mcip);
3893 
3894 	bcopy(mcip_mrp, mrp, sizeof (mac_resource_props_t));
3895 }
3896 
3897 /*
3898  * Return the effective properties currently associated with the specified
3899  * MAC client.
3900  */
3901 void
3902 mac_client_get_effective_resources(mac_client_handle_t mch,
3903     mac_resource_props_t *mrp)
3904 {
3905 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3906 	mac_resource_props_t	*mcip_mrp = MCIP_EFFECTIVE_PROPS(mcip);
3907 
3908 	bcopy(mcip_mrp, mrp, sizeof (mac_resource_props_t));
3909 }
3910 
3911 /*
3912  * Pass a copy of the specified packet to the promiscuous callbacks
3913  * of the specified MAC.
3914  *
3915  * If sender is NULL, the function is being invoked for a packet chain
3916  * received from the wire. If sender is non-NULL, it points to
3917  * the MAC client from which the packet is being sent.
3918  *
3919  * The packets are distributed to the promiscuous callbacks as follows:
3920  *
3921  * - all packets are sent to the MAC_CLIENT_PROMISC_ALL callbacks
3922  * - all broadcast and multicast packets are sent to the
3923  *   MAC_CLIENT_PROMISC_FILTER and MAC_CLIENT_PROMISC_MULTI.
3924  *
3925  * The unicast packets of MAC_CLIENT_PROMISC_FILTER callbacks are dispatched
3926  * after classification by mac_rx_deliver().
3927  */
3928 
3929 static void
3930 mac_promisc_dispatch_one(mac_promisc_impl_t *mpip, mblk_t *mp,
3931     boolean_t loopback)
3932 {
3933 	mblk_t *mp_copy, *mp_next;
3934 
3935 	if (!mpip->mpi_no_copy || mpip->mpi_strip_vlan_tag) {
3936 		mp_copy = copymsg(mp);
3937 		if (mp_copy == NULL)
3938 			return;
3939 
3940 		if (mpip->mpi_strip_vlan_tag) {
3941 			mp_copy = mac_strip_vlan_tag_chain(mp_copy);
3942 			if (mp_copy == NULL)
3943 				return;
3944 		}
3945 		mp_next = NULL;
3946 	} else {
3947 		mp_copy = mp;
3948 		mp_next = mp->b_next;
3949 	}
3950 	mp_copy->b_next = NULL;
3951 
3952 	mpip->mpi_fn(mpip->mpi_arg, NULL, mp_copy, loopback);
3953 	if (mp_copy == mp)
3954 		mp->b_next = mp_next;
3955 }
3956 
3957 /*
3958  * Return the VID of a packet. Zero if the packet is not tagged.
3959  */
3960 static uint16_t
3961 mac_ether_vid(mblk_t *mp)
3962 {
3963 	struct ether_header *eth = (struct ether_header *)mp->b_rptr;
3964 
3965 	if (ntohs(eth->ether_type) == ETHERTYPE_VLAN) {
3966 		struct ether_vlan_header *t_evhp =
3967 		    (struct ether_vlan_header *)mp->b_rptr;
3968 		return (VLAN_ID(ntohs(t_evhp->ether_tci)));
3969 	}
3970 
3971 	return (0);
3972 }
3973 
3974 /*
3975  * Return whether the specified packet contains a multicast or broadcast
3976  * destination MAC address.
3977  */
3978 static boolean_t
3979 mac_is_mcast(mac_impl_t *mip, mblk_t *mp)
3980 {
3981 	mac_header_info_t hdr_info;
3982 
3983 	if (mac_header_info((mac_handle_t)mip, mp, &hdr_info) != 0)
3984 		return (B_FALSE);
3985 	return ((hdr_info.mhi_dsttype == MAC_ADDRTYPE_BROADCAST) ||
3986 	    (hdr_info.mhi_dsttype == MAC_ADDRTYPE_MULTICAST));
3987 }
3988 
3989 /*
3990  * Send a copy of an mblk chain to the MAC clients of the specified MAC.
3991  * "sender" points to the sender MAC client for outbound packets, and
3992  * is set to NULL for inbound packets.
3993  */
3994 void
3995 mac_promisc_dispatch(mac_impl_t *mip, mblk_t *mp_chain,
3996     mac_client_impl_t *sender)
3997 {
3998 	mac_promisc_impl_t *mpip;
3999 	mac_cb_t *mcb;
4000 	mblk_t *mp;
4001 	boolean_t is_mcast, is_sender;
4002 
4003 	MAC_PROMISC_WALKER_INC(mip);
4004 	for (mp = mp_chain; mp != NULL; mp = mp->b_next) {
4005 		is_mcast = mac_is_mcast(mip, mp);
4006 		/* send packet to interested callbacks */
4007 		for (mcb = mip->mi_promisc_list; mcb != NULL;
4008 		    mcb = mcb->mcb_nextp) {
4009 			mpip = (mac_promisc_impl_t *)mcb->mcb_objp;
4010 			is_sender = (mpip->mpi_mcip == sender);
4011 
4012 			if (is_sender && mpip->mpi_no_tx_loop)
4013 				/*
4014 				 * The sender doesn't want to receive
4015 				 * copies of the packets it sends.
4016 				 */
4017 				continue;
4018 
4019 			/* this client doesn't need any packets (bridge) */
4020 			if (mpip->mpi_fn == NULL)
4021 				continue;
4022 
4023 			/*
4024 			 * For an ethernet MAC, don't displatch a multicast
4025 			 * packet to a non-PROMISC_ALL callbacks unless the VID
4026 			 * of the packet matches the VID of the client.
4027 			 */
4028 			if (is_mcast &&
4029 			    mpip->mpi_type != MAC_CLIENT_PROMISC_ALL &&
4030 			    !mac_client_check_flow_vid(mpip->mpi_mcip,
4031 			    mac_ether_vid(mp)))
4032 				continue;
4033 
4034 			if (is_sender ||
4035 			    mpip->mpi_type == MAC_CLIENT_PROMISC_ALL ||
4036 			    is_mcast)
4037 				mac_promisc_dispatch_one(mpip, mp, is_sender);
4038 		}
4039 	}
4040 	MAC_PROMISC_WALKER_DCR(mip);
4041 }
4042 
4043 void
4044 mac_promisc_client_dispatch(mac_client_impl_t *mcip, mblk_t *mp_chain)
4045 {
4046 	mac_impl_t		*mip = mcip->mci_mip;
4047 	mac_promisc_impl_t	*mpip;
4048 	boolean_t		is_mcast;
4049 	mblk_t			*mp;
4050 	mac_cb_t		*mcb;
4051 
4052 	/*
4053 	 * The unicast packets for the MAC client still
4054 	 * need to be delivered to the MAC_CLIENT_PROMISC_FILTERED
4055 	 * promiscuous callbacks. The broadcast and multicast
4056 	 * packets were delivered from mac_rx().
4057 	 */
4058 	MAC_PROMISC_WALKER_INC(mip);
4059 	for (mp = mp_chain; mp != NULL; mp = mp->b_next) {
4060 		is_mcast = mac_is_mcast(mip, mp);
4061 		for (mcb = mcip->mci_promisc_list; mcb != NULL;
4062 		    mcb = mcb->mcb_nextp) {
4063 			mpip = (mac_promisc_impl_t *)mcb->mcb_objp;
4064 			if (mpip->mpi_type == MAC_CLIENT_PROMISC_FILTERED &&
4065 			    !is_mcast) {
4066 				mac_promisc_dispatch_one(mpip, mp, B_FALSE);
4067 			}
4068 		}
4069 	}
4070 	MAC_PROMISC_WALKER_DCR(mip);
4071 }
4072 
4073 /*
4074  * Return the margin value currently assigned to the specified MAC instance.
4075  */
4076 void
4077 mac_margin_get(mac_handle_t mh, uint32_t *marginp)
4078 {
4079 	mac_impl_t *mip = (mac_impl_t *)mh;
4080 
4081 	rw_enter(&(mip->mi_rw_lock), RW_READER);
4082 	*marginp = mip->mi_margin;
4083 	rw_exit(&(mip->mi_rw_lock));
4084 }
4085 
4086 /*
4087  * mac_info_get() is used for retrieving the mac_info when a DL_INFO_REQ is
4088  * issued before a DL_ATTACH_REQ. we walk the i_mac_impl_hash table and find
4089  * the first mac_impl_t with a matching driver name; then we copy its mac_info_t
4090  * to the caller. we do all this with i_mac_impl_lock held so the mac_impl_t
4091  * cannot disappear while we are accessing it.
4092  */
4093 typedef struct i_mac_info_state_s {
4094 	const char	*mi_name;
4095 	mac_info_t	*mi_infop;
4096 } i_mac_info_state_t;
4097 
4098 /*ARGSUSED*/
4099 static uint_t
4100 i_mac_info_walker(mod_hash_key_t key, mod_hash_val_t *val, void *arg)
4101 {
4102 	i_mac_info_state_t *statep = arg;
4103 	mac_impl_t *mip = (mac_impl_t *)val;
4104 
4105 	if (mip->mi_state_flags & MIS_DISABLED)
4106 		return (MH_WALK_CONTINUE);
4107 
4108 	if (strcmp(statep->mi_name,
4109 	    ddi_driver_name(mip->mi_dip)) != 0)
4110 		return (MH_WALK_CONTINUE);
4111 
4112 	statep->mi_infop = &mip->mi_info;
4113 	return (MH_WALK_TERMINATE);
4114 }
4115 
4116 boolean_t
4117 mac_info_get(const char *name, mac_info_t *minfop)
4118 {
4119 	i_mac_info_state_t state;
4120 
4121 	rw_enter(&i_mac_impl_lock, RW_READER);
4122 	state.mi_name = name;
4123 	state.mi_infop = NULL;
4124 	mod_hash_walk(i_mac_impl_hash, i_mac_info_walker, &state);
4125 	if (state.mi_infop == NULL) {
4126 		rw_exit(&i_mac_impl_lock);
4127 		return (B_FALSE);
4128 	}
4129 	*minfop = *state.mi_infop;
4130 	rw_exit(&i_mac_impl_lock);
4131 	return (B_TRUE);
4132 }
4133 
4134 /*
4135  * To get the capabilities that MAC layer cares about, such as rings, factory
4136  * mac address, vnic or not, it should directly invoke this function.  If the
4137  * link is part of a bridge, then the only "capability" it has is the inability
4138  * to do zero copy.
4139  */
4140 boolean_t
4141 i_mac_capab_get(mac_handle_t mh, mac_capab_t cap, void *cap_data)
4142 {
4143 	mac_impl_t *mip = (mac_impl_t *)mh;
4144 
4145 	if (mip->mi_bridge_link != NULL)
4146 		return (cap == MAC_CAPAB_NO_ZCOPY);
4147 	else if (mip->mi_callbacks->mc_callbacks & MC_GETCAPAB)
4148 		return (mip->mi_getcapab(mip->mi_driver, cap, cap_data));
4149 	else
4150 		return (B_FALSE);
4151 }
4152 
4153 /*
4154  * Capability query function. If number of active mac clients is greater than
4155  * 1, only limited capabilities can be advertised to the caller no matter the
4156  * driver has certain capability or not. Else, we query the driver to get the
4157  * capability.
4158  */
4159 boolean_t
4160 mac_capab_get(mac_handle_t mh, mac_capab_t cap, void *cap_data)
4161 {
4162 	mac_impl_t *mip = (mac_impl_t *)mh;
4163 
4164 	/*
4165 	 * if mi_nactiveclients > 1, only MAC_CAPAB_LEGACY, MAC_CAPAB_HCKSUM,
4166 	 * MAC_CAPAB_NO_NATIVEVLAN and MAC_CAPAB_NO_ZCOPY can be advertised.
4167 	 */
4168 	if (mip->mi_nactiveclients > 1) {
4169 		switch (cap) {
4170 		case MAC_CAPAB_NO_ZCOPY:
4171 			return (B_TRUE);
4172 		case MAC_CAPAB_LEGACY:
4173 		case MAC_CAPAB_HCKSUM:
4174 		case MAC_CAPAB_NO_NATIVEVLAN:
4175 			break;
4176 		default:
4177 			return (B_FALSE);
4178 		}
4179 	}
4180 
4181 	/* else get capab from driver */
4182 	return (i_mac_capab_get(mh, cap, cap_data));
4183 }
4184 
4185 boolean_t
4186 mac_sap_verify(mac_handle_t mh, uint32_t sap, uint32_t *bind_sap)
4187 {
4188 	mac_impl_t *mip = (mac_impl_t *)mh;
4189 
4190 	return (mip->mi_type->mt_ops.mtops_sap_verify(sap, bind_sap,
4191 	    mip->mi_pdata));
4192 }
4193 
4194 mblk_t *
4195 mac_header(mac_handle_t mh, const uint8_t *daddr, uint32_t sap, mblk_t *payload,
4196     size_t extra_len)
4197 {
4198 	mac_impl_t	*mip = (mac_impl_t *)mh;
4199 	const uint8_t	*hdr_daddr;
4200 
4201 	/*
4202 	 * If the MAC is point-to-point with a fixed destination address, then
4203 	 * we must always use that destination in the MAC header.
4204 	 */
4205 	hdr_daddr = (mip->mi_dstaddr_set ? mip->mi_dstaddr : daddr);
4206 	return (mip->mi_type->mt_ops.mtops_header(mip->mi_addr, hdr_daddr, sap,
4207 	    mip->mi_pdata, payload, extra_len));
4208 }
4209 
4210 int
4211 mac_header_info(mac_handle_t mh, mblk_t *mp, mac_header_info_t *mhip)
4212 {
4213 	mac_impl_t *mip = (mac_impl_t *)mh;
4214 
4215 	return (mip->mi_type->mt_ops.mtops_header_info(mp, mip->mi_pdata,
4216 	    mhip));
4217 }
4218 
4219 int
4220 mac_vlan_header_info(mac_handle_t mh, mblk_t *mp, mac_header_info_t *mhip)
4221 {
4222 	mac_impl_t	*mip = (mac_impl_t *)mh;
4223 	boolean_t	is_ethernet = (mip->mi_info.mi_media == DL_ETHER);
4224 	int		err = 0;
4225 
4226 	/*
4227 	 * Packets should always be at least 16 bit aligned.
4228 	 */
4229 	ASSERT(IS_P2ALIGNED(mp->b_rptr, sizeof (uint16_t)));
4230 
4231 	if ((err = mac_header_info(mh, mp, mhip)) != 0)
4232 		return (err);
4233 
4234 	/*
4235 	 * If this is a VLAN-tagged Ethernet packet, then the SAP in the
4236 	 * mac_header_info_t as returned by mac_header_info() is
4237 	 * ETHERTYPE_VLAN. We need to grab the ethertype from the VLAN header.
4238 	 */
4239 	if (is_ethernet && (mhip->mhi_bindsap == ETHERTYPE_VLAN)) {
4240 		struct ether_vlan_header *evhp;
4241 		uint16_t sap;
4242 		mblk_t *tmp = NULL;
4243 		size_t size;
4244 
4245 		size = sizeof (struct ether_vlan_header);
4246 		if (MBLKL(mp) < size) {
4247 			/*
4248 			 * Pullup the message in order to get the MAC header
4249 			 * infomation. Note that this is a read-only function,
4250 			 * we keep the input packet intact.
4251 			 */
4252 			if ((tmp = msgpullup(mp, size)) == NULL)
4253 				return (EINVAL);
4254 
4255 			mp = tmp;
4256 		}
4257 		evhp = (struct ether_vlan_header *)mp->b_rptr;
4258 		sap = ntohs(evhp->ether_type);
4259 		(void) mac_sap_verify(mh, sap, &mhip->mhi_bindsap);
4260 		mhip->mhi_hdrsize = sizeof (struct ether_vlan_header);
4261 		mhip->mhi_tci = ntohs(evhp->ether_tci);
4262 		mhip->mhi_istagged = B_TRUE;
4263 		freemsg(tmp);
4264 
4265 		if (VLAN_CFI(mhip->mhi_tci) != ETHER_CFI)
4266 			return (EINVAL);
4267 	} else {
4268 		mhip->mhi_istagged = B_FALSE;
4269 		mhip->mhi_tci = 0;
4270 	}
4271 
4272 	return (0);
4273 }
4274 
4275 mblk_t *
4276 mac_header_cook(mac_handle_t mh, mblk_t *mp)
4277 {
4278 	mac_impl_t *mip = (mac_impl_t *)mh;
4279 
4280 	if (mip->mi_type->mt_ops.mtops_ops & MTOPS_HEADER_COOK) {
4281 		if (DB_REF(mp) > 1) {
4282 			mblk_t *newmp = copymsg(mp);
4283 			if (newmp == NULL)
4284 				return (NULL);
4285 			freemsg(mp);
4286 			mp = newmp;
4287 		}
4288 		return (mip->mi_type->mt_ops.mtops_header_cook(mp,
4289 		    mip->mi_pdata));
4290 	}
4291 	return (mp);
4292 }
4293 
4294 mblk_t *
4295 mac_header_uncook(mac_handle_t mh, mblk_t *mp)
4296 {
4297 	mac_impl_t *mip = (mac_impl_t *)mh;
4298 
4299 	if (mip->mi_type->mt_ops.mtops_ops & MTOPS_HEADER_UNCOOK) {
4300 		if (DB_REF(mp) > 1) {
4301 			mblk_t *newmp = copymsg(mp);
4302 			if (newmp == NULL)
4303 				return (NULL);
4304 			freemsg(mp);
4305 			mp = newmp;
4306 		}
4307 		return (mip->mi_type->mt_ops.mtops_header_uncook(mp,
4308 		    mip->mi_pdata));
4309 	}
4310 	return (mp);
4311 }
4312 
4313 uint_t
4314 mac_addr_len(mac_handle_t mh)
4315 {
4316 	mac_impl_t *mip = (mac_impl_t *)mh;
4317 
4318 	return (mip->mi_type->mt_addr_length);
4319 }
4320 
4321 /* True if a MAC is a VNIC */
4322 boolean_t
4323 mac_is_vnic(mac_handle_t mh)
4324 {
4325 	return (((mac_impl_t *)mh)->mi_state_flags & MIS_IS_VNIC);
4326 }
4327 
4328 mac_handle_t
4329 mac_get_lower_mac_handle(mac_handle_t mh)
4330 {
4331 	mac_impl_t *mip = (mac_impl_t *)mh;
4332 
4333 	ASSERT(mac_is_vnic(mh));
4334 	return (((vnic_t *)mip->mi_driver)->vn_lower_mh);
4335 }
4336 
4337 boolean_t
4338 mac_is_vnic_primary(mac_handle_t mh)
4339 {
4340 	mac_impl_t *mip = (mac_impl_t *)mh;
4341 
4342 	ASSERT(mac_is_vnic(mh));
4343 	return (((vnic_t *)mip->mi_driver)->vn_addr_type ==
4344 	    VNIC_MAC_ADDR_TYPE_PRIMARY);
4345 }
4346 
4347 void
4348 mac_update_resources(mac_resource_props_t *nmrp, mac_resource_props_t *cmrp,
4349     boolean_t is_user_flow)
4350 {
4351 	if (nmrp != NULL && cmrp != NULL) {
4352 		if (nmrp->mrp_mask & MRP_PRIORITY) {
4353 			if (nmrp->mrp_priority == MPL_RESET) {
4354 				cmrp->mrp_mask &= ~MRP_PRIORITY;
4355 				if (is_user_flow) {
4356 					cmrp->mrp_priority =
4357 					    MPL_SUBFLOW_DEFAULT;
4358 				} else {
4359 					cmrp->mrp_priority = MPL_LINK_DEFAULT;
4360 				}
4361 			} else {
4362 				cmrp->mrp_mask |= MRP_PRIORITY;
4363 				cmrp->mrp_priority = nmrp->mrp_priority;
4364 			}
4365 		}
4366 		if (nmrp->mrp_mask & MRP_MAXBW) {
4367 			if (nmrp->mrp_maxbw == MRP_MAXBW_RESETVAL) {
4368 				cmrp->mrp_mask &= ~MRP_MAXBW;
4369 				cmrp->mrp_maxbw = 0;
4370 			} else {
4371 				cmrp->mrp_mask |= MRP_MAXBW;
4372 				cmrp->mrp_maxbw = nmrp->mrp_maxbw;
4373 			}
4374 		}
4375 		if (nmrp->mrp_mask & MRP_CPUS)
4376 			MAC_COPY_CPUS(nmrp, cmrp);
4377 
4378 		if (nmrp->mrp_mask & MRP_POOL) {
4379 			if (strlen(nmrp->mrp_pool) == 0) {
4380 				cmrp->mrp_mask &= ~MRP_POOL;
4381 				bzero(cmrp->mrp_pool, sizeof (cmrp->mrp_pool));
4382 			} else {
4383 				cmrp->mrp_mask |= MRP_POOL;
4384 				(void) strncpy(cmrp->mrp_pool, nmrp->mrp_pool,
4385 				    sizeof (cmrp->mrp_pool));
4386 			}
4387 
4388 		}
4389 
4390 		if (nmrp->mrp_mask & MRP_PROTECT)
4391 			mac_protect_update(nmrp, cmrp);
4392 
4393 		/*
4394 		 * Update the rings specified.
4395 		 */
4396 		if (nmrp->mrp_mask & MRP_RX_RINGS) {
4397 			if (nmrp->mrp_mask & MRP_RINGS_RESET) {
4398 				cmrp->mrp_mask &= ~MRP_RX_RINGS;
4399 				if (cmrp->mrp_mask & MRP_RXRINGS_UNSPEC)
4400 					cmrp->mrp_mask &= ~MRP_RXRINGS_UNSPEC;
4401 				cmrp->mrp_nrxrings = 0;
4402 			} else {
4403 				cmrp->mrp_mask |= MRP_RX_RINGS;
4404 				cmrp->mrp_nrxrings = nmrp->mrp_nrxrings;
4405 			}
4406 		}
4407 		if (nmrp->mrp_mask & MRP_TX_RINGS) {
4408 			if (nmrp->mrp_mask & MRP_RINGS_RESET) {
4409 				cmrp->mrp_mask &= ~MRP_TX_RINGS;
4410 				if (cmrp->mrp_mask & MRP_TXRINGS_UNSPEC)
4411 					cmrp->mrp_mask &= ~MRP_TXRINGS_UNSPEC;
4412 				cmrp->mrp_ntxrings = 0;
4413 			} else {
4414 				cmrp->mrp_mask |= MRP_TX_RINGS;
4415 				cmrp->mrp_ntxrings = nmrp->mrp_ntxrings;
4416 			}
4417 		}
4418 		if (nmrp->mrp_mask & MRP_RXRINGS_UNSPEC)
4419 			cmrp->mrp_mask |= MRP_RXRINGS_UNSPEC;
4420 		else if (cmrp->mrp_mask & MRP_RXRINGS_UNSPEC)
4421 			cmrp->mrp_mask &= ~MRP_RXRINGS_UNSPEC;
4422 
4423 		if (nmrp->mrp_mask & MRP_TXRINGS_UNSPEC)
4424 			cmrp->mrp_mask |= MRP_TXRINGS_UNSPEC;
4425 		else if (cmrp->mrp_mask & MRP_TXRINGS_UNSPEC)
4426 			cmrp->mrp_mask &= ~MRP_TXRINGS_UNSPEC;
4427 	}
4428 }
4429 
4430 /*
4431  * i_mac_set_resources:
4432  *
4433  * This routine associates properties with the primary MAC client of
4434  * the specified MAC instance.
4435  * - Cache the properties in mac_impl_t
4436  * - Apply the properties to the primary MAC client if exists
4437  */
4438 int
4439 i_mac_set_resources(mac_handle_t mh, mac_resource_props_t *mrp)
4440 {
4441 	mac_impl_t		*mip = (mac_impl_t *)mh;
4442 	mac_client_impl_t	*mcip;
4443 	int			err = 0;
4444 	uint32_t		resmask, newresmask;
4445 	mac_resource_props_t	*tmrp, *umrp;
4446 
4447 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
4448 
4449 	err = mac_validate_props(mip, mrp);
4450 	if (err != 0)
4451 		return (err);
4452 
4453 	umrp = kmem_zalloc(sizeof (*umrp), KM_SLEEP);
4454 	bcopy(&mip->mi_resource_props, umrp, sizeof (*umrp));
4455 	resmask = umrp->mrp_mask;
4456 	mac_update_resources(mrp, umrp, B_FALSE);
4457 	newresmask = umrp->mrp_mask;
4458 
4459 	if (resmask == 0 && newresmask != 0) {
4460 		/*
4461 		 * Bandwidth, priority, cpu or pool link properties configured,
4462 		 * must disable fastpath.
4463 		 */
4464 		if ((err = mac_fastpath_disable((mac_handle_t)mip)) != 0) {
4465 			kmem_free(umrp, sizeof (*umrp));
4466 			return (err);
4467 		}
4468 	}
4469 
4470 	/*
4471 	 * Since bind_cpu may be modified by mac_client_set_resources()
4472 	 * we use a copy of bind_cpu and finally cache bind_cpu in mip.
4473 	 * This allows us to cache only user edits in mip.
4474 	 */
4475 	tmrp = kmem_zalloc(sizeof (*tmrp), KM_SLEEP);
4476 	bcopy(mrp, tmrp, sizeof (*tmrp));
4477 	mcip = mac_primary_client_handle(mip);
4478 	if (mcip != NULL && (mcip->mci_state_flags & MCIS_IS_AGGR_PORT) == 0) {
4479 		err = mac_client_set_resources((mac_client_handle_t)mcip, tmrp);
4480 	} else if ((mrp->mrp_mask & MRP_RX_RINGS ||
4481 	    mrp->mrp_mask & MRP_TX_RINGS)) {
4482 		mac_client_impl_t	*vmcip;
4483 
4484 		/*
4485 		 * If the primary is not up, we need to check if there
4486 		 * are any VLANs on this primary. If there are then
4487 		 * we need to set this property on the VLANs since
4488 		 * VLANs follow the primary they are based on. Just
4489 		 * look for the first VLAN and change its properties,
4490 		 * all the other VLANs should be in the same group.
4491 		 */
4492 		for (vmcip = mip->mi_clients_list; vmcip != NULL;
4493 		    vmcip = vmcip->mci_client_next) {
4494 			if ((vmcip->mci_flent->fe_type & FLOW_PRIMARY_MAC) &&
4495 			    mac_client_vid((mac_client_handle_t)vmcip) !=
4496 			    VLAN_ID_NONE) {
4497 				break;
4498 			}
4499 		}
4500 		if (vmcip != NULL) {
4501 			mac_resource_props_t	*omrp;
4502 			mac_resource_props_t	*vmrp;
4503 
4504 			omrp = kmem_zalloc(sizeof (*omrp), KM_SLEEP);
4505 			bcopy(MCIP_RESOURCE_PROPS(vmcip), omrp, sizeof (*omrp));
4506 			/*
4507 			 * We dont' call mac_update_resources since we
4508 			 * want to take only the ring properties and
4509 			 * not all the properties that may have changed.
4510 			 */
4511 			vmrp = MCIP_RESOURCE_PROPS(vmcip);
4512 			if (mrp->mrp_mask & MRP_RX_RINGS) {
4513 				if (mrp->mrp_mask & MRP_RINGS_RESET) {
4514 					vmrp->mrp_mask &= ~MRP_RX_RINGS;
4515 					if (vmrp->mrp_mask &
4516 					    MRP_RXRINGS_UNSPEC) {
4517 						vmrp->mrp_mask &=
4518 						    ~MRP_RXRINGS_UNSPEC;
4519 					}
4520 					vmrp->mrp_nrxrings = 0;
4521 				} else {
4522 					vmrp->mrp_mask |= MRP_RX_RINGS;
4523 					vmrp->mrp_nrxrings = mrp->mrp_nrxrings;
4524 				}
4525 			}
4526 			if (mrp->mrp_mask & MRP_TX_RINGS) {
4527 				if (mrp->mrp_mask & MRP_RINGS_RESET) {
4528 					vmrp->mrp_mask &= ~MRP_TX_RINGS;
4529 					if (vmrp->mrp_mask &
4530 					    MRP_TXRINGS_UNSPEC) {
4531 						vmrp->mrp_mask &=
4532 						    ~MRP_TXRINGS_UNSPEC;
4533 					}
4534 					vmrp->mrp_ntxrings = 0;
4535 				} else {
4536 					vmrp->mrp_mask |= MRP_TX_RINGS;
4537 					vmrp->mrp_ntxrings = mrp->mrp_ntxrings;
4538 				}
4539 			}
4540 			if (mrp->mrp_mask & MRP_RXRINGS_UNSPEC)
4541 				vmrp->mrp_mask |= MRP_RXRINGS_UNSPEC;
4542 
4543 			if (mrp->mrp_mask & MRP_TXRINGS_UNSPEC)
4544 				vmrp->mrp_mask |= MRP_TXRINGS_UNSPEC;
4545 
4546 			if ((err = mac_client_set_rings_prop(vmcip, mrp,
4547 			    omrp)) != 0) {
4548 				bcopy(omrp, MCIP_RESOURCE_PROPS(vmcip),
4549 				    sizeof (*omrp));
4550 			} else {
4551 				mac_set_prim_vlan_rings(mip, vmrp);
4552 			}
4553 			kmem_free(omrp, sizeof (*omrp));
4554 		}
4555 	}
4556 
4557 	/* Only update the values if mac_client_set_resources succeeded */
4558 	if (err == 0) {
4559 		bcopy(umrp, &mip->mi_resource_props, sizeof (*umrp));
4560 		/*
4561 		 * If bandwidth, priority or cpu link properties cleared,
4562 		 * renable fastpath.
4563 		 */
4564 		if (resmask != 0 && newresmask == 0)
4565 			mac_fastpath_enable((mac_handle_t)mip);
4566 	} else if (resmask == 0 && newresmask != 0) {
4567 		mac_fastpath_enable((mac_handle_t)mip);
4568 	}
4569 	kmem_free(tmrp, sizeof (*tmrp));
4570 	kmem_free(umrp, sizeof (*umrp));
4571 	return (err);
4572 }
4573 
4574 int
4575 mac_set_resources(mac_handle_t mh, mac_resource_props_t *mrp)
4576 {
4577 	int err;
4578 
4579 	i_mac_perim_enter((mac_impl_t *)mh);
4580 	err = i_mac_set_resources(mh, mrp);
4581 	i_mac_perim_exit((mac_impl_t *)mh);
4582 	return (err);
4583 }
4584 
4585 /*
4586  * Get the properties cached for the specified MAC instance.
4587  */
4588 void
4589 mac_get_resources(mac_handle_t mh, mac_resource_props_t *mrp)
4590 {
4591 	mac_impl_t 		*mip = (mac_impl_t *)mh;
4592 	mac_client_impl_t	*mcip;
4593 
4594 	mcip = mac_primary_client_handle(mip);
4595 	if (mcip != NULL) {
4596 		mac_client_get_resources((mac_client_handle_t)mcip, mrp);
4597 		return;
4598 	}
4599 	bcopy(&mip->mi_resource_props, mrp, sizeof (mac_resource_props_t));
4600 }
4601 
4602 /*
4603  * Get the effective properties from the primary client of the
4604  * specified MAC instance.
4605  */
4606 void
4607 mac_get_effective_resources(mac_handle_t mh, mac_resource_props_t *mrp)
4608 {
4609 	mac_impl_t 		*mip = (mac_impl_t *)mh;
4610 	mac_client_impl_t	*mcip;
4611 
4612 	mcip = mac_primary_client_handle(mip);
4613 	if (mcip != NULL) {
4614 		mac_client_get_effective_resources((mac_client_handle_t)mcip,
4615 		    mrp);
4616 		return;
4617 	}
4618 	bzero(mrp, sizeof (mac_resource_props_t));
4619 }
4620 
4621 int
4622 mac_set_pvid(mac_handle_t mh, uint16_t pvid)
4623 {
4624 	mac_impl_t *mip = (mac_impl_t *)mh;
4625 	mac_client_impl_t *mcip;
4626 	mac_unicast_impl_t *muip;
4627 
4628 	i_mac_perim_enter(mip);
4629 	if (pvid != 0) {
4630 		for (mcip = mip->mi_clients_list; mcip != NULL;
4631 		    mcip = mcip->mci_client_next) {
4632 			for (muip = mcip->mci_unicast_list; muip != NULL;
4633 			    muip = muip->mui_next) {
4634 				if (muip->mui_vid == pvid) {
4635 					i_mac_perim_exit(mip);
4636 					return (EBUSY);
4637 				}
4638 			}
4639 		}
4640 	}
4641 	mip->mi_pvid = pvid;
4642 	i_mac_perim_exit(mip);
4643 	return (0);
4644 }
4645 
4646 uint16_t
4647 mac_get_pvid(mac_handle_t mh)
4648 {
4649 	mac_impl_t *mip = (mac_impl_t *)mh;
4650 
4651 	return (mip->mi_pvid);
4652 }
4653 
4654 uint32_t
4655 mac_get_llimit(mac_handle_t mh)
4656 {
4657 	mac_impl_t *mip = (mac_impl_t *)mh;
4658 
4659 	return (mip->mi_llimit);
4660 }
4661 
4662 uint32_t
4663 mac_get_ldecay(mac_handle_t mh)
4664 {
4665 	mac_impl_t *mip = (mac_impl_t *)mh;
4666 
4667 	return (mip->mi_ldecay);
4668 }
4669 
4670 /*
4671  * Rename a mac client, its flow, and the kstat.
4672  */
4673 int
4674 mac_rename_primary(mac_handle_t mh, const char *new_name)
4675 {
4676 	mac_impl_t		*mip = (mac_impl_t *)mh;
4677 	mac_client_impl_t	*cur_clnt = NULL;
4678 	flow_entry_t		*fep;
4679 
4680 	i_mac_perim_enter(mip);
4681 
4682 	/*
4683 	 * VNICs: we need to change the sys flow name and
4684 	 * the associated flow kstat.
4685 	 */
4686 	if (mip->mi_state_flags & MIS_IS_VNIC) {
4687 		mac_client_impl_t *mcip = mac_vnic_lower(mip);
4688 		ASSERT(new_name != NULL);
4689 		mac_rename_flow_names(mcip, new_name);
4690 		mac_stat_rename(mcip);
4691 		goto done;
4692 	}
4693 	/*
4694 	 * This mac may itself be an aggr link, or it may have some client
4695 	 * which is an aggr port. For both cases, we need to change the
4696 	 * aggr port's mac client name, its flow name and the associated flow
4697 	 * kstat.
4698 	 */
4699 	if (mip->mi_state_flags & MIS_IS_AGGR) {
4700 		mac_capab_aggr_t aggr_cap;
4701 		mac_rename_fn_t rename_fn;
4702 		boolean_t ret;
4703 
4704 		ASSERT(new_name != NULL);
4705 		ret = i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_AGGR,
4706 		    (void *)(&aggr_cap));
4707 		ASSERT(ret == B_TRUE);
4708 		rename_fn = aggr_cap.mca_rename_fn;
4709 		rename_fn(new_name, mip->mi_driver);
4710 		/*
4711 		 * The aggr's client name and kstat flow name will be
4712 		 * updated below, i.e. via mac_rename_flow_names.
4713 		 */
4714 	}
4715 
4716 	for (cur_clnt = mip->mi_clients_list; cur_clnt != NULL;
4717 	    cur_clnt = cur_clnt->mci_client_next) {
4718 		if (cur_clnt->mci_state_flags & MCIS_IS_AGGR_PORT) {
4719 			if (new_name != NULL) {
4720 				char *str_st = cur_clnt->mci_name;
4721 				char *str_del = strchr(str_st, '-');
4722 
4723 				ASSERT(str_del != NULL);
4724 				bzero(str_del + 1, MAXNAMELEN -
4725 				    (str_del - str_st + 1));
4726 				bcopy(new_name, str_del + 1,
4727 				    strlen(new_name));
4728 			}
4729 			fep = cur_clnt->mci_flent;
4730 			mac_rename_flow(fep, cur_clnt->mci_name);
4731 			break;
4732 		} else if (new_name != NULL &&
4733 		    cur_clnt->mci_state_flags & MCIS_USE_DATALINK_NAME) {
4734 			mac_rename_flow_names(cur_clnt, new_name);
4735 			break;
4736 		}
4737 	}
4738 
4739 	/* Recreate kstats associated with aggr pseudo rings */
4740 	if (mip->mi_state_flags & MIS_IS_AGGR)
4741 		mac_pseudo_ring_stat_rename(mip);
4742 
4743 done:
4744 	i_mac_perim_exit(mip);
4745 	return (0);
4746 }
4747 
4748 /*
4749  * Rename the MAC client's flow names
4750  */
4751 static void
4752 mac_rename_flow_names(mac_client_impl_t *mcip, const char *new_name)
4753 {
4754 	flow_entry_t	*flent;
4755 	uint16_t	vid;
4756 	char		flowname[MAXFLOWNAMELEN];
4757 	mac_impl_t	*mip = mcip->mci_mip;
4758 
4759 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
4760 
4761 	/*
4762 	 * Use mi_rw_lock to ensure that threads not in the mac perimeter
4763 	 * see a self-consistent value for mci_name
4764 	 */
4765 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
4766 	(void) strlcpy(mcip->mci_name, new_name, sizeof (mcip->mci_name));
4767 	rw_exit(&mip->mi_rw_lock);
4768 
4769 	mac_rename_flow(mcip->mci_flent, new_name);
4770 
4771 	if (mcip->mci_nflents == 1)
4772 		return;
4773 
4774 	/*
4775 	 * We have to rename all the others too, no stats to destroy for
4776 	 * these.
4777 	 */
4778 	for (flent = mcip->mci_flent_list; flent != NULL;
4779 	    flent = flent->fe_client_next) {
4780 		if (flent != mcip->mci_flent) {
4781 			vid = i_mac_flow_vid(flent);
4782 			(void) sprintf(flowname, "%s%u", new_name, vid);
4783 			mac_flow_set_name(flent, flowname);
4784 		}
4785 	}
4786 }
4787 
4788 
4789 /*
4790  * Add a flow to the MAC client's flow list - i.e list of MAC/VID tuples
4791  * defined for the specified MAC client.
4792  */
4793 static void
4794 mac_client_add_to_flow_list(mac_client_impl_t *mcip, flow_entry_t *flent)
4795 {
4796 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
4797 	/*
4798 	 * The promisc Rx data path walks the mci_flent_list. Protect by
4799 	 * using mi_rw_lock
4800 	 */
4801 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
4802 
4803 	/* Add it to the head */
4804 	flent->fe_client_next = mcip->mci_flent_list;
4805 	mcip->mci_flent_list = flent;
4806 	mcip->mci_nflents++;
4807 
4808 	/*
4809 	 * Keep track of the number of non-zero VIDs addresses per MAC
4810 	 * client to avoid figuring it out in the data-path.
4811 	 */
4812 	if (i_mac_flow_vid(flent) != VLAN_ID_NONE)
4813 		mcip->mci_nvids++;
4814 
4815 	rw_exit(&mcip->mci_rw_lock);
4816 }
4817 
4818 /*
4819  * Remove a flow entry from the MAC client's list.
4820  */
4821 static void
4822 mac_client_remove_flow_from_list(mac_client_impl_t *mcip, flow_entry_t *flent)
4823 {
4824 	flow_entry_t	*fe = mcip->mci_flent_list;
4825 	flow_entry_t	*prev_fe = NULL;
4826 
4827 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
4828 	/*
4829 	 * The promisc Rx data path walks the mci_flent_list. Protect by
4830 	 * using mci_rw_lock
4831 	 */
4832 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
4833 	while ((fe != NULL) && (fe != flent)) {
4834 		prev_fe = fe;
4835 		fe = fe->fe_client_next;
4836 	}
4837 
4838 	ASSERT(fe != NULL);
4839 	if (prev_fe == NULL) {
4840 		/* Deleting the first node */
4841 		mcip->mci_flent_list = fe->fe_client_next;
4842 	} else {
4843 		prev_fe->fe_client_next = fe->fe_client_next;
4844 	}
4845 	mcip->mci_nflents--;
4846 
4847 	if (i_mac_flow_vid(flent) != VLAN_ID_NONE)
4848 		mcip->mci_nvids--;
4849 
4850 	rw_exit(&mcip->mci_rw_lock);
4851 }
4852 
4853 /*
4854  * Check if the given VID belongs to this MAC client.
4855  */
4856 boolean_t
4857 mac_client_check_flow_vid(mac_client_impl_t *mcip, uint16_t vid)
4858 {
4859 	flow_entry_t	*flent;
4860 	uint16_t	mci_vid;
4861 
4862 	/* The mci_flent_list is protected by mci_rw_lock */
4863 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
4864 	for (flent = mcip->mci_flent_list; flent != NULL;
4865 	    flent = flent->fe_client_next) {
4866 		mci_vid = i_mac_flow_vid(flent);
4867 		if (vid == mci_vid) {
4868 			rw_exit(&mcip->mci_rw_lock);
4869 			return (B_TRUE);
4870 		}
4871 	}
4872 	rw_exit(&mcip->mci_rw_lock);
4873 	return (B_FALSE);
4874 }
4875 
4876 /*
4877  * Get the flow entry for the specified <MAC addr, VID> tuple.
4878  */
4879 static flow_entry_t *
4880 mac_client_get_flow(mac_client_impl_t *mcip, mac_unicast_impl_t *muip)
4881 {
4882 	mac_address_t *map = mcip->mci_unicast;
4883 	flow_entry_t *flent;
4884 	uint16_t vid;
4885 	flow_desc_t flow_desc;
4886 
4887 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
4888 
4889 	mac_flow_get_desc(mcip->mci_flent, &flow_desc);
4890 	if (bcmp(flow_desc.fd_dst_mac, map->ma_addr, map->ma_len) != 0)
4891 		return (NULL);
4892 
4893 	for (flent = mcip->mci_flent_list; flent != NULL;
4894 	    flent = flent->fe_client_next) {
4895 		vid = i_mac_flow_vid(flent);
4896 		if (vid == muip->mui_vid) {
4897 			return (flent);
4898 		}
4899 	}
4900 
4901 	return (NULL);
4902 }
4903 
4904 /*
4905  * Since mci_flent has the SRSs, when we want to remove it, we replace
4906  * the flow_desc_t in mci_flent with that of an existing flent and then
4907  * remove that flent instead of mci_flent.
4908  */
4909 static flow_entry_t *
4910 mac_client_swap_mciflent(mac_client_impl_t *mcip)
4911 {
4912 	flow_entry_t	*flent = mcip->mci_flent;
4913 	flow_tab_t	*ft = flent->fe_flow_tab;
4914 	flow_entry_t	*flent1;
4915 	flow_desc_t	fl_desc;
4916 	char		fl_name[MAXFLOWNAMELEN];
4917 	int		err;
4918 
4919 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
4920 	ASSERT(mcip->mci_nflents > 1);
4921 
4922 	/* get the next flent following the primary flent  */
4923 	flent1 = mcip->mci_flent_list->fe_client_next;
4924 	ASSERT(flent1 != NULL && flent1->fe_flow_tab == ft);
4925 
4926 	/*
4927 	 * Remove the flent from the flow table before updating the
4928 	 * flow descriptor as the hash depends on the flow descriptor.
4929 	 * This also helps incoming packet classification avoid having
4930 	 * to grab fe_lock. Access to fe_flow_desc of a flent not in the
4931 	 * flow table is done under the fe_lock so that log or stat functions
4932 	 * see a self-consistent fe_flow_desc. The name and desc are specific
4933 	 * to a flow, the rest are shared by all the clients, including
4934 	 * resource control etc.
4935 	 */
4936 	mac_flow_remove(ft, flent, B_TRUE);
4937 	mac_flow_remove(ft, flent1, B_TRUE);
4938 
4939 	bcopy(&flent->fe_flow_desc, &fl_desc, sizeof (flow_desc_t));
4940 	bcopy(flent->fe_flow_name, fl_name, MAXFLOWNAMELEN);
4941 
4942 	/* update the primary flow entry */
4943 	mutex_enter(&flent->fe_lock);
4944 	bcopy(&flent1->fe_flow_desc, &flent->fe_flow_desc,
4945 	    sizeof (flow_desc_t));
4946 	bcopy(&flent1->fe_flow_name, &flent->fe_flow_name, MAXFLOWNAMELEN);
4947 	mutex_exit(&flent->fe_lock);
4948 
4949 	/* update the flow entry that is to be freed */
4950 	mutex_enter(&flent1->fe_lock);
4951 	bcopy(&fl_desc, &flent1->fe_flow_desc, sizeof (flow_desc_t));
4952 	bcopy(fl_name, &flent1->fe_flow_name, MAXFLOWNAMELEN);
4953 	mutex_exit(&flent1->fe_lock);
4954 
4955 	/* now reinsert the flow entries in the table */
4956 	err = mac_flow_add(ft, flent);
4957 	ASSERT(err == 0);
4958 
4959 	err = mac_flow_add(ft, flent1);
4960 	ASSERT(err == 0);
4961 
4962 	return (flent1);
4963 }
4964 
4965 /*
4966  * Return whether there is only one flow entry associated with this
4967  * MAC client.
4968  */
4969 static boolean_t
4970 mac_client_single_rcvr(mac_client_impl_t *mcip)
4971 {
4972 	return (mcip->mci_nflents == 1);
4973 }
4974 
4975 int
4976 mac_validate_props(mac_impl_t *mip, mac_resource_props_t *mrp)
4977 {
4978 	boolean_t		reset;
4979 	uint32_t		rings_needed;
4980 	uint32_t		rings_avail;
4981 	mac_group_type_t	gtype;
4982 	mac_resource_props_t	*mip_mrp;
4983 
4984 	if (mrp == NULL)
4985 		return (0);
4986 
4987 	if (mrp->mrp_mask & MRP_PRIORITY) {
4988 		mac_priority_level_t	pri = mrp->mrp_priority;
4989 
4990 		if (pri < MPL_LOW || pri > MPL_RESET)
4991 			return (EINVAL);
4992 	}
4993 
4994 	if (mrp->mrp_mask & MRP_MAXBW) {
4995 		uint64_t maxbw = mrp->mrp_maxbw;
4996 
4997 		if (maxbw < MRP_MAXBW_MINVAL && maxbw != 0)
4998 			return (EINVAL);
4999 	}
5000 	if (mrp->mrp_mask & MRP_CPUS) {
5001 		int i, j;
5002 		mac_cpu_mode_t	fanout;
5003 
5004 		if (mrp->mrp_ncpus > ncpus)
5005 			return (EINVAL);
5006 
5007 		for (i = 0; i < mrp->mrp_ncpus; i++) {
5008 			for (j = 0; j < mrp->mrp_ncpus; j++) {
5009 				if (i != j &&
5010 				    mrp->mrp_cpu[i] == mrp->mrp_cpu[j]) {
5011 					return (EINVAL);
5012 				}
5013 			}
5014 		}
5015 
5016 		for (i = 0; i < mrp->mrp_ncpus; i++) {
5017 			cpu_t *cp;
5018 			int rv;
5019 
5020 			mutex_enter(&cpu_lock);
5021 			cp = cpu_get(mrp->mrp_cpu[i]);
5022 			if (cp != NULL)
5023 				rv = cpu_is_online(cp);
5024 			else
5025 				rv = 0;
5026 			mutex_exit(&cpu_lock);
5027 			if (rv == 0)
5028 				return (EINVAL);
5029 		}
5030 
5031 		fanout = mrp->mrp_fanout_mode;
5032 		if (fanout < 0 || fanout > MCM_CPUS)
5033 			return (EINVAL);
5034 	}
5035 
5036 	if (mrp->mrp_mask & MRP_PROTECT) {
5037 		int err = mac_protect_validate(mrp);
5038 		if (err != 0)
5039 			return (err);
5040 	}
5041 
5042 	if (!(mrp->mrp_mask & MRP_RX_RINGS) &&
5043 	    !(mrp->mrp_mask & MRP_TX_RINGS)) {
5044 		return (0);
5045 	}
5046 
5047 	/*
5048 	 * mip will be null when we come from mac_flow_create or
5049 	 * mac_link_flow_modify. In the latter case it is a user flow,
5050 	 * for which we don't support rings. In the former we would
5051 	 * have validated the props beforehand (i_mac_unicast_add ->
5052 	 * mac_client_set_resources -> validate for the primary and
5053 	 * vnic_dev_create -> mac_client_set_resources -> validate for
5054 	 * a vnic.
5055 	 */
5056 	if (mip == NULL)
5057 		return (0);
5058 
5059 	/*
5060 	 * We don't support setting rings property for a VNIC that is using a
5061 	 * primary address (VLAN)
5062 	 */
5063 	if ((mip->mi_state_flags & MIS_IS_VNIC) &&
5064 	    mac_is_vnic_primary((mac_handle_t)mip)) {
5065 		return (ENOTSUP);
5066 	}
5067 
5068 	mip_mrp = &mip->mi_resource_props;
5069 	/*
5070 	 * The rings property should be validated against the NICs
5071 	 * resources
5072 	 */
5073 	if (mip->mi_state_flags & MIS_IS_VNIC)
5074 		mip = (mac_impl_t *)mac_get_lower_mac_handle((mac_handle_t)mip);
5075 
5076 	reset = mrp->mrp_mask & MRP_RINGS_RESET;
5077 	/*
5078 	 * If groups are not supported, return error.
5079 	 */
5080 	if (((mrp->mrp_mask & MRP_RX_RINGS) && mip->mi_rx_groups == NULL) ||
5081 	    ((mrp->mrp_mask & MRP_TX_RINGS) && mip->mi_tx_groups == NULL)) {
5082 		return (EINVAL);
5083 	}
5084 	/*
5085 	 * If we are just resetting, there is no validation needed.
5086 	 */
5087 	if (reset)
5088 		return (0);
5089 
5090 	if (mrp->mrp_mask & MRP_RX_RINGS) {
5091 		rings_needed = mrp->mrp_nrxrings;
5092 		/*
5093 		 * We just want to check if the number of additional
5094 		 * rings requested is available.
5095 		 */
5096 		if (mip_mrp->mrp_mask & MRP_RX_RINGS) {
5097 			if (mrp->mrp_nrxrings > mip_mrp->mrp_nrxrings)
5098 				/* Just check for the additional rings */
5099 				rings_needed -= mip_mrp->mrp_nrxrings;
5100 			else
5101 				/* We are not asking for additional rings */
5102 				rings_needed = 0;
5103 		}
5104 		rings_avail = mip->mi_rxrings_avail;
5105 		gtype = mip->mi_rx_group_type;
5106 	} else {
5107 		rings_needed = mrp->mrp_ntxrings;
5108 		/* Similarly for the TX rings */
5109 		if (mip_mrp->mrp_mask & MRP_TX_RINGS) {
5110 			if (mrp->mrp_ntxrings > mip_mrp->mrp_ntxrings)
5111 				/* Just check for the additional rings */
5112 				rings_needed -= mip_mrp->mrp_ntxrings;
5113 			else
5114 				/* We are not asking for additional rings */
5115 				rings_needed = 0;
5116 		}
5117 		rings_avail = mip->mi_txrings_avail;
5118 		gtype = mip->mi_tx_group_type;
5119 	}
5120 
5121 	/* Error if the group is dynamic .. */
5122 	if (gtype == MAC_GROUP_TYPE_DYNAMIC) {
5123 		/*
5124 		 * .. and rings specified are more than available.
5125 		 */
5126 		if (rings_needed > rings_avail)
5127 			return (EINVAL);
5128 	} else {
5129 		/*
5130 		 * OR group is static and we have specified some rings.
5131 		 */
5132 		if (rings_needed > 0)
5133 			return (EINVAL);
5134 	}
5135 	return (0);
5136 }
5137 
5138 /*
5139  * Send a MAC_NOTE_LINK notification to all the MAC clients whenever the
5140  * underlying physical link is down. This is to allow MAC clients to
5141  * communicate with other clients.
5142  */
5143 void
5144 mac_virtual_link_update(mac_impl_t *mip)
5145 {
5146 	if (mip->mi_linkstate != LINK_STATE_UP)
5147 		i_mac_notify(mip, MAC_NOTE_LINK);
5148 }
5149 
5150 /*
5151  * For clients that have a pass-thru MAC, e.g. VNIC, we set the VNIC's
5152  * mac handle in the client.
5153  */
5154 void
5155 mac_set_upper_mac(mac_client_handle_t mch, mac_handle_t mh,
5156     mac_resource_props_t *mrp)
5157 {
5158 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
5159 	mac_impl_t		*mip = (mac_impl_t *)mh;
5160 
5161 	mcip->mci_upper_mip = mip;
5162 	/* If there are any properties, copy it over too */
5163 	if (mrp != NULL) {
5164 		bcopy(mrp, &mip->mi_resource_props,
5165 		    sizeof (mac_resource_props_t));
5166 	}
5167 }
5168 
5169 /*
5170  * Mark the mac as being used exclusively by the single mac client that is
5171  * doing some control operation on this mac. No further opens of this mac
5172  * will be allowed until this client calls mac_unmark_exclusive. The mac
5173  * client calling this function must already be in the mac perimeter
5174  */
5175 int
5176 mac_mark_exclusive(mac_handle_t mh)
5177 {
5178 	mac_impl_t	*mip = (mac_impl_t *)mh;
5179 
5180 	ASSERT(MAC_PERIM_HELD(mh));
5181 	/*
5182 	 * Look up its entry in the global hash table.
5183 	 */
5184 	rw_enter(&i_mac_impl_lock, RW_WRITER);
5185 	if (mip->mi_state_flags & MIS_DISABLED) {
5186 		rw_exit(&i_mac_impl_lock);
5187 		return (ENOENT);
5188 	}
5189 
5190 	/*
5191 	 * A reference to mac is held even if the link is not plumbed.
5192 	 * In i_dls_link_create() we open the MAC interface and hold the
5193 	 * reference. There is an additional reference for the mac_open
5194 	 * done in acquiring the mac perimeter
5195 	 */
5196 	if (mip->mi_ref != 2) {
5197 		rw_exit(&i_mac_impl_lock);
5198 		return (EBUSY);
5199 	}
5200 
5201 	ASSERT(!(mip->mi_state_flags & MIS_EXCLUSIVE_HELD));
5202 	mip->mi_state_flags |= MIS_EXCLUSIVE_HELD;
5203 	rw_exit(&i_mac_impl_lock);
5204 	return (0);
5205 }
5206 
5207 void
5208 mac_unmark_exclusive(mac_handle_t mh)
5209 {
5210 	mac_impl_t	*mip = (mac_impl_t *)mh;
5211 
5212 	ASSERT(MAC_PERIM_HELD(mh));
5213 
5214 	rw_enter(&i_mac_impl_lock, RW_WRITER);
5215 	/* 1 for the creation and another for the perimeter */
5216 	ASSERT(mip->mi_ref == 2 && (mip->mi_state_flags & MIS_EXCLUSIVE_HELD));
5217 	mip->mi_state_flags &= ~MIS_EXCLUSIVE_HELD;
5218 	rw_exit(&i_mac_impl_lock);
5219 }
5220 
5221 /*
5222  * Set the MTU for the specified MAC.
5223  */
5224 int
5225 mac_set_mtu(mac_handle_t mh, uint_t new_mtu, uint_t *old_mtu_arg)
5226 {
5227 	mac_impl_t *mip = (mac_impl_t *)mh;
5228 	uint_t old_mtu;
5229 	int rv = 0;
5230 
5231 	i_mac_perim_enter(mip);
5232 
5233 	if (!(mip->mi_callbacks->mc_callbacks & (MC_SETPROP|MC_GETPROP))) {
5234 		rv = ENOTSUP;
5235 		goto bail;
5236 	}
5237 
5238 	old_mtu = mip->mi_sdu_max;
5239 
5240 	if (new_mtu == 0 || new_mtu < mip->mi_sdu_min) {
5241 		rv = EINVAL;
5242 		goto bail;
5243 	}
5244 
5245 	rw_enter(&mip->mi_rw_lock, RW_READER);
5246 	if (mip->mi_mtrp != NULL && new_mtu < mip->mi_mtrp->mtr_mtu) {
5247 		rv = EBUSY;
5248 		rw_exit(&mip->mi_rw_lock);
5249 		goto bail;
5250 	}
5251 	rw_exit(&mip->mi_rw_lock);
5252 
5253 	if (old_mtu != new_mtu) {
5254 		rv = mip->mi_callbacks->mc_setprop(mip->mi_driver,
5255 		    "mtu", MAC_PROP_MTU, sizeof (uint_t), &new_mtu);
5256 		if (rv != 0)
5257 			goto bail;
5258 		rv = mac_maxsdu_update(mh, new_mtu);
5259 		ASSERT(rv == 0);
5260 	}
5261 
5262 bail:
5263 	i_mac_perim_exit(mip);
5264 
5265 	if (rv == 0 && old_mtu_arg != NULL)
5266 		*old_mtu_arg = old_mtu;
5267 	return (rv);
5268 }
5269 
5270 /*
5271  * Return the RX h/w information for the group indexed by grp_num.
5272  */
5273 void
5274 mac_get_hwrxgrp_info(mac_handle_t mh, int grp_index, uint_t *grp_num,
5275     uint_t *n_rings, uint_t *rings, uint_t *type, uint_t *n_clnts,
5276     char *clnts_name)
5277 {
5278 	mac_impl_t *mip = (mac_impl_t *)mh;
5279 	mac_grp_client_t *mcip;
5280 	uint_t i = 0, index = 0;
5281 	mac_ring_t	*ring;
5282 
5283 	/* Revisit when we implement fully dynamic group allocation */
5284 	ASSERT(grp_index >= 0 && grp_index < mip->mi_rx_group_count);
5285 
5286 	rw_enter(&mip->mi_rw_lock, RW_READER);
5287 	*grp_num = mip->mi_rx_groups[grp_index].mrg_index;
5288 	*type = mip->mi_rx_groups[grp_index].mrg_type;
5289 	*n_rings = mip->mi_rx_groups[grp_index].mrg_cur_count;
5290 	ring = mip->mi_rx_groups[grp_index].mrg_rings;
5291 	for (index = 0; index < mip->mi_rx_groups[grp_index].mrg_cur_count;
5292 	    index++) {
5293 		rings[index] = ring->mr_index;
5294 		ring = ring->mr_next;
5295 	}
5296 	/* Assuming the 1st is the default group */
5297 	index = 0;
5298 	if (grp_index == 0) {
5299 		(void) strlcpy(clnts_name, "<default,mcast>,",
5300 		    MAXCLIENTNAMELEN);
5301 		index += strlen("<default,mcast>,");
5302 	}
5303 	for (mcip = mip->mi_rx_groups[grp_index].mrg_clients; mcip != NULL;
5304 	    mcip = mcip->mgc_next) {
5305 		int name_len = strlen(mcip->mgc_client->mci_name);
5306 
5307 		/*
5308 		 * MAXCLIENTNAMELEN is the buffer size reserved for client
5309 		 * names.
5310 		 * XXXX Formating the client name string needs to be moved
5311 		 * to user land when fixing the size of dhi_clnts in
5312 		 * dld_hwgrpinfo_t. We should use n_clients * client_name for
5313 		 * dhi_clntsin instead of MAXCLIENTNAMELEN
5314 		 */
5315 		if (index + name_len >= MAXCLIENTNAMELEN) {
5316 			index = MAXCLIENTNAMELEN;
5317 			break;
5318 		}
5319 		bcopy(mcip->mgc_client->mci_name, &(clnts_name[index]),
5320 		    name_len);
5321 		index += name_len;
5322 		clnts_name[index++] = ',';
5323 		i++;
5324 	}
5325 
5326 	/* Get rid of the last , */
5327 	if (index > 0)
5328 		clnts_name[index - 1] = '\0';
5329 	*n_clnts = i;
5330 	rw_exit(&mip->mi_rw_lock);
5331 }
5332 
5333 /*
5334  * Return the TX h/w information for the group indexed by grp_num.
5335  */
5336 void
5337 mac_get_hwtxgrp_info(mac_handle_t mh, int grp_index, uint_t *grp_num,
5338     uint_t *n_rings, uint_t *rings, uint_t *type, uint_t *n_clnts,
5339     char *clnts_name)
5340 {
5341 	mac_impl_t *mip = (mac_impl_t *)mh;
5342 	mac_grp_client_t *mcip;
5343 	uint_t i = 0, index = 0;
5344 	mac_ring_t	*ring;
5345 
5346 	/* Revisit when we implement fully dynamic group allocation */
5347 	ASSERT(grp_index >= 0 && grp_index <= mip->mi_tx_group_count);
5348 
5349 	rw_enter(&mip->mi_rw_lock, RW_READER);
5350 	*grp_num = mip->mi_tx_groups[grp_index].mrg_index > 0 ?
5351 	    mip->mi_tx_groups[grp_index].mrg_index : grp_index;
5352 	*type = mip->mi_tx_groups[grp_index].mrg_type;
5353 	*n_rings = mip->mi_tx_groups[grp_index].mrg_cur_count;
5354 	ring = mip->mi_tx_groups[grp_index].mrg_rings;
5355 	for (index = 0; index < mip->mi_tx_groups[grp_index].mrg_cur_count;
5356 	    index++) {
5357 		rings[index] = ring->mr_index;
5358 		ring = ring->mr_next;
5359 	}
5360 	index = 0;
5361 	/* Default group has an index of -1 */
5362 	if (mip->mi_tx_groups[grp_index].mrg_index < 0) {
5363 		(void) strlcpy(clnts_name, "<default>,",
5364 		    MAXCLIENTNAMELEN);
5365 		index += strlen("<default>,");
5366 	}
5367 	for (mcip = mip->mi_tx_groups[grp_index].mrg_clients; mcip != NULL;
5368 	    mcip = mcip->mgc_next) {
5369 		int name_len = strlen(mcip->mgc_client->mci_name);
5370 
5371 		/*
5372 		 * MAXCLIENTNAMELEN is the buffer size reserved for client
5373 		 * names.
5374 		 * XXXX Formating the client name string needs to be moved
5375 		 * to user land when fixing the size of dhi_clnts in
5376 		 * dld_hwgrpinfo_t. We should use n_clients * client_name for
5377 		 * dhi_clntsin instead of MAXCLIENTNAMELEN
5378 		 */
5379 		if (index + name_len >= MAXCLIENTNAMELEN) {
5380 			index = MAXCLIENTNAMELEN;
5381 			break;
5382 		}
5383 		bcopy(mcip->mgc_client->mci_name, &(clnts_name[index]),
5384 		    name_len);
5385 		index += name_len;
5386 		clnts_name[index++] = ',';
5387 		i++;
5388 	}
5389 
5390 	/* Get rid of the last , */
5391 	if (index > 0)
5392 		clnts_name[index - 1] = '\0';
5393 	*n_clnts = i;
5394 	rw_exit(&mip->mi_rw_lock);
5395 }
5396 
5397 /*
5398  * Return the group count for RX or TX.
5399  */
5400 uint_t
5401 mac_hwgrp_num(mac_handle_t mh, int type)
5402 {
5403 	mac_impl_t *mip = (mac_impl_t *)mh;
5404 
5405 	/*
5406 	 * Return the Rx and Tx group count; for the Tx we need to
5407 	 * include the default too.
5408 	 */
5409 	return (type == MAC_RING_TYPE_RX ? mip->mi_rx_group_count :
5410 	    mip->mi_tx_groups != NULL ? mip->mi_tx_group_count + 1 : 0);
5411 }
5412 
5413 /*
5414  * The total number of free TX rings for this MAC.
5415  */
5416 uint_t
5417 mac_txavail_get(mac_handle_t mh)
5418 {
5419 	mac_impl_t	*mip = (mac_impl_t *)mh;
5420 
5421 	return (mip->mi_txrings_avail);
5422 }
5423 
5424 /*
5425  * The total number of free RX rings for this MAC.
5426  */
5427 uint_t
5428 mac_rxavail_get(mac_handle_t mh)
5429 {
5430 	mac_impl_t	*mip = (mac_impl_t *)mh;
5431 
5432 	return (mip->mi_rxrings_avail);
5433 }
5434 
5435 /*
5436  * The total number of reserved RX rings on this MAC.
5437  */
5438 uint_t
5439 mac_rxrsvd_get(mac_handle_t mh)
5440 {
5441 	mac_impl_t	*mip = (mac_impl_t *)mh;
5442 
5443 	return (mip->mi_rxrings_rsvd);
5444 }
5445 
5446 /*
5447  * The total number of reserved TX rings on this MAC.
5448  */
5449 uint_t
5450 mac_txrsvd_get(mac_handle_t mh)
5451 {
5452 	mac_impl_t	*mip = (mac_impl_t *)mh;
5453 
5454 	return (mip->mi_txrings_rsvd);
5455 }
5456 
5457 /*
5458  * Total number of free RX groups on this MAC.
5459  */
5460 uint_t
5461 mac_rxhwlnksavail_get(mac_handle_t mh)
5462 {
5463 	mac_impl_t	*mip = (mac_impl_t *)mh;
5464 
5465 	return (mip->mi_rxhwclnt_avail);
5466 }
5467 
5468 /*
5469  * Total number of RX groups reserved on this MAC.
5470  */
5471 uint_t
5472 mac_rxhwlnksrsvd_get(mac_handle_t mh)
5473 {
5474 	mac_impl_t	*mip = (mac_impl_t *)mh;
5475 
5476 	return (mip->mi_rxhwclnt_used);
5477 }
5478 
5479 /*
5480  * Total number of free TX groups on this MAC.
5481  */
5482 uint_t
5483 mac_txhwlnksavail_get(mac_handle_t mh)
5484 {
5485 	mac_impl_t	*mip = (mac_impl_t *)mh;
5486 
5487 	return (mip->mi_txhwclnt_avail);
5488 }
5489 
5490 /*
5491  * Total number of TX groups reserved on this MAC.
5492  */
5493 uint_t
5494 mac_txhwlnksrsvd_get(mac_handle_t mh)
5495 {
5496 	mac_impl_t	*mip = (mac_impl_t *)mh;
5497 
5498 	return (mip->mi_txhwclnt_used);
5499 }
5500 
5501 /*
5502  * Initialize the rings property for a mac client. A non-0 value for
5503  * rxring or txring specifies the number of rings required, a value
5504  * of MAC_RXRINGS_NONE/MAC_TXRINGS_NONE specifies that it doesn't need
5505  * any RX/TX rings and a value of MAC_RXRINGS_DONTCARE/MAC_TXRINGS_DONTCARE
5506  * means the system can decide whether it can give any rings or not.
5507  */
5508 void
5509 mac_client_set_rings(mac_client_handle_t mch, int rxrings, int txrings)
5510 {
5511 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
5512 	mac_resource_props_t	*mrp = MCIP_RESOURCE_PROPS(mcip);
5513 
5514 	if (rxrings != MAC_RXRINGS_DONTCARE) {
5515 		mrp->mrp_mask |= MRP_RX_RINGS;
5516 		mrp->mrp_nrxrings = rxrings;
5517 	}
5518 
5519 	if (txrings != MAC_TXRINGS_DONTCARE) {
5520 		mrp->mrp_mask |= MRP_TX_RINGS;
5521 		mrp->mrp_ntxrings = txrings;
5522 	}
5523 }
5524