xref: /illumos-gate/usr/src/uts/common/io/mac/mac_client.c (revision e07d85f87c3920e032adb855fdc500e4616c7718)
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  * Copyright 2017 RackTop Systems.
26  */
27 
28 /*
29  * - General Introduction:
30  *
31  * This file contains the implementation of the MAC client kernel
32  * API and related code. The MAC client API allows a kernel module
33  * to gain access to a MAC instance (physical NIC, link aggregation, etc).
34  * It allows a MAC client to associate itself with a MAC address,
35  * VLANs, callback functions for data traffic and for promiscuous mode.
36  * The MAC client API is also used to specify the properties associated
37  * with a MAC client, such as bandwidth limits, priority, CPUS, etc.
38  * These properties are further used to determine the hardware resources
39  * to allocate to the various MAC clients.
40  *
41  * - Primary MAC clients:
42  *
43  * The MAC client API refers to "primary MAC clients". A primary MAC
44  * client is a client which "owns" the primary MAC address of
45  * the underlying MAC instance. The primary MAC address is called out
46  * since it is associated with specific semantics: the primary MAC
47  * address is the MAC address which is assigned to the IP interface
48  * when it is plumbed, and the primary MAC address is assigned
49  * to VLAN data-links. The primary address of a MAC instance can
50  * also change dynamically from under the MAC client, for example
51  * as a result of a change of state of a link aggregation. In that
52  * case the MAC layer automatically updates all data-structures which
53  * refer to the current value of the primary MAC address. Typical
54  * primary MAC clients are dls, aggr, and xnb. A typical non-primary
55  * MAC client is the vnic driver.
56  *
57  * - Virtual Switching:
58  *
59  * The MAC layer implements a virtual switch between the MAC clients
60  * (primary and non-primary) defined on top of the same underlying
61  * NIC (physical, link aggregation, etc). The virtual switch is
62  * VLAN-aware, i.e. it allows multiple MAC clients to be member
63  * of one or more VLANs, and the virtual switch will distribute
64  * multicast tagged packets only to the member of the corresponding
65  * VLANs.
66  *
67  * - Upper vs Lower MAC:
68  *
69  * Creating a VNIC on top of a MAC instance effectively causes
70  * two MAC instances to be layered on top of each other, one for
71  * the VNIC(s), one for the underlying MAC instance (physical NIC,
72  * link aggregation, etc). In the code below we refer to the
73  * underlying NIC as the "lower MAC", and we refer to VNICs as
74  * the "upper MAC".
75  *
76  * - Pass-through for VNICs:
77  *
78  * When VNICs are created on top of an underlying MAC, this causes
79  * a layering of two MAC instances. Since the lower MAC already
80  * does the switching and demultiplexing to its MAC clients, the
81  * upper MAC would simply have to pass packets to the layer below
82  * or above it, which would introduce overhead. In order to avoid
83  * this overhead, the MAC layer implements a pass-through mechanism
84  * for VNICs. When a VNIC opens the lower MAC instance, it saves
85  * the MAC client handle it optains from the MAC layer. When a MAC
86  * client opens a VNIC (upper MAC), the MAC layer detects that
87  * the MAC being opened is a VNIC, and gets the MAC client handle
88  * that the VNIC driver obtained from the lower MAC. This exchange
89  * is done through a private capability between the MAC layer
90  * and the VNIC driver. The upper MAC then returns that handle
91  * directly to its MAC client. Any operation done by the upper
92  * MAC client is now done on the lower MAC client handle, which
93  * allows the VNIC driver to be completely bypassed for the
94  * performance sensitive data-path.
95  *
96  * - Secondary MACs for VNICs:
97  *
98  * VNICs support multiple upper mac clients to enable support for
99  * multiple MAC addresses on the VNIC. When the VNIC is created the
100  * initial mac client is the primary upper mac. Any additional mac
101  * clients are secondary macs. These are kept in sync with the primary
102  * (for things such as the rx function and resource control settings)
103  * using the same private capability interface between the MAC layer
104  * and the VNIC layer.
105  *
106  */
107 
108 #include <sys/types.h>
109 #include <sys/conf.h>
110 #include <sys/id_space.h>
111 #include <sys/esunddi.h>
112 #include <sys/stat.h>
113 #include <sys/mkdev.h>
114 #include <sys/stream.h>
115 #include <sys/strsun.h>
116 #include <sys/strsubr.h>
117 #include <sys/dlpi.h>
118 #include <sys/modhash.h>
119 #include <sys/mac_impl.h>
120 #include <sys/mac_client_impl.h>
121 #include <sys/mac_soft_ring.h>
122 #include <sys/mac_stat.h>
123 #include <sys/dls.h>
124 #include <sys/dld.h>
125 #include <sys/modctl.h>
126 #include <sys/fs/dv_node.h>
127 #include <sys/thread.h>
128 #include <sys/proc.h>
129 #include <sys/callb.h>
130 #include <sys/cpuvar.h>
131 #include <sys/atomic.h>
132 #include <sys/sdt.h>
133 #include <sys/mac_flow.h>
134 #include <sys/ddi_intr_impl.h>
135 #include <sys/disp.h>
136 #include <sys/sdt.h>
137 #include <sys/vnic.h>
138 #include <sys/vnic_impl.h>
139 #include <sys/vlan.h>
140 #include <inet/ip.h>
141 #include <inet/ip6.h>
142 #include <sys/exacct.h>
143 #include <sys/exacct_impl.h>
144 #include <inet/nd.h>
145 #include <sys/ethernet.h>
146 
147 kmem_cache_t	*mac_client_impl_cache;
148 kmem_cache_t	*mac_promisc_impl_cache;
149 
150 static boolean_t mac_client_single_rcvr(mac_client_impl_t *);
151 static flow_entry_t *mac_client_swap_mciflent(mac_client_impl_t *);
152 static flow_entry_t *mac_client_get_flow(mac_client_impl_t *,
153     mac_unicast_impl_t *);
154 static void mac_client_remove_flow_from_list(mac_client_impl_t *,
155     flow_entry_t *);
156 static void mac_client_add_to_flow_list(mac_client_impl_t *, flow_entry_t *);
157 static void mac_rename_flow_names(mac_client_impl_t *, const char *);
158 static void mac_virtual_link_update(mac_impl_t *);
159 static int mac_client_datapath_setup(mac_client_impl_t *, uint16_t,
160     uint8_t *, mac_resource_props_t *, boolean_t, mac_unicast_impl_t *);
161 static void mac_client_datapath_teardown(mac_client_handle_t,
162     mac_unicast_impl_t *, flow_entry_t *);
163 static int mac_resource_ctl_set(mac_client_handle_t, mac_resource_props_t *);
164 
165 /* ARGSUSED */
166 static int
167 i_mac_client_impl_ctor(void *buf, void *arg, int kmflag)
168 {
169 	int	i;
170 	mac_client_impl_t	*mcip = buf;
171 
172 	bzero(buf, MAC_CLIENT_IMPL_SIZE);
173 	mutex_init(&mcip->mci_tx_cb_lock, NULL, MUTEX_DRIVER, NULL);
174 	mcip->mci_tx_notify_cb_info.mcbi_lockp = &mcip->mci_tx_cb_lock;
175 
176 	ASSERT(mac_tx_percpu_cnt >= 0);
177 	for (i = 0; i <= mac_tx_percpu_cnt; i++) {
178 		mutex_init(&mcip->mci_tx_pcpu[i].pcpu_tx_lock, NULL,
179 		    MUTEX_DRIVER, NULL);
180 	}
181 	cv_init(&mcip->mci_tx_cv, NULL, CV_DRIVER, NULL);
182 
183 	return (0);
184 }
185 
186 /* ARGSUSED */
187 static void
188 i_mac_client_impl_dtor(void *buf, void *arg)
189 {
190 	int	i;
191 	mac_client_impl_t *mcip = buf;
192 
193 	ASSERT(mcip->mci_promisc_list == NULL);
194 	ASSERT(mcip->mci_unicast_list == NULL);
195 	ASSERT(mcip->mci_state_flags == 0);
196 	ASSERT(mcip->mci_tx_flag == 0);
197 
198 	mutex_destroy(&mcip->mci_tx_cb_lock);
199 
200 	ASSERT(mac_tx_percpu_cnt >= 0);
201 	for (i = 0; i <= mac_tx_percpu_cnt; i++) {
202 		ASSERT(mcip->mci_tx_pcpu[i].pcpu_tx_refcnt == 0);
203 		mutex_destroy(&mcip->mci_tx_pcpu[i].pcpu_tx_lock);
204 	}
205 	cv_destroy(&mcip->mci_tx_cv);
206 }
207 
208 /* ARGSUSED */
209 static int
210 i_mac_promisc_impl_ctor(void *buf, void *arg, int kmflag)
211 {
212 	mac_promisc_impl_t	*mpip = buf;
213 
214 	bzero(buf, sizeof (mac_promisc_impl_t));
215 	mpip->mpi_mci_link.mcb_objp = buf;
216 	mpip->mpi_mci_link.mcb_objsize = sizeof (mac_promisc_impl_t);
217 	mpip->mpi_mi_link.mcb_objp = buf;
218 	mpip->mpi_mi_link.mcb_objsize = sizeof (mac_promisc_impl_t);
219 	return (0);
220 }
221 
222 /* ARGSUSED */
223 static void
224 i_mac_promisc_impl_dtor(void *buf, void *arg)
225 {
226 	mac_promisc_impl_t	*mpip = buf;
227 
228 	ASSERT(mpip->mpi_mci_link.mcb_objp != NULL);
229 	ASSERT(mpip->mpi_mci_link.mcb_objsize == sizeof (mac_promisc_impl_t));
230 	ASSERT(mpip->mpi_mi_link.mcb_objp == mpip->mpi_mci_link.mcb_objp);
231 	ASSERT(mpip->mpi_mi_link.mcb_objsize == sizeof (mac_promisc_impl_t));
232 
233 	mpip->mpi_mci_link.mcb_objp = NULL;
234 	mpip->mpi_mci_link.mcb_objsize = 0;
235 	mpip->mpi_mi_link.mcb_objp = NULL;
236 	mpip->mpi_mi_link.mcb_objsize = 0;
237 
238 	ASSERT(mpip->mpi_mci_link.mcb_flags == 0);
239 	mpip->mpi_mci_link.mcb_objsize = 0;
240 }
241 
242 void
243 mac_client_init(void)
244 {
245 	ASSERT(mac_tx_percpu_cnt >= 0);
246 
247 	mac_client_impl_cache = kmem_cache_create("mac_client_impl_cache",
248 	    MAC_CLIENT_IMPL_SIZE, 0, i_mac_client_impl_ctor,
249 	    i_mac_client_impl_dtor, NULL, NULL, NULL, 0);
250 	ASSERT(mac_client_impl_cache != NULL);
251 
252 	mac_promisc_impl_cache = kmem_cache_create("mac_promisc_impl_cache",
253 	    sizeof (mac_promisc_impl_t), 0, i_mac_promisc_impl_ctor,
254 	    i_mac_promisc_impl_dtor, NULL, NULL, NULL, 0);
255 	ASSERT(mac_promisc_impl_cache != NULL);
256 }
257 
258 void
259 mac_client_fini(void)
260 {
261 	kmem_cache_destroy(mac_client_impl_cache);
262 	kmem_cache_destroy(mac_promisc_impl_cache);
263 }
264 
265 /*
266  * Return the lower MAC client handle from the VNIC driver for the
267  * specified VNIC MAC instance.
268  */
269 mac_client_impl_t *
270 mac_vnic_lower(mac_impl_t *mip)
271 {
272 	mac_capab_vnic_t cap;
273 	mac_client_impl_t *mcip;
274 
275 	VERIFY(i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_VNIC, &cap));
276 	mcip = cap.mcv_mac_client_handle(cap.mcv_arg);
277 
278 	return (mcip);
279 }
280 
281 /*
282  * Update the secondary macs
283  */
284 void
285 mac_vnic_secondary_update(mac_impl_t *mip)
286 {
287 	mac_capab_vnic_t cap;
288 
289 	VERIFY(i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_VNIC, &cap));
290 	cap.mcv_mac_secondary_update(cap.mcv_arg);
291 }
292 
293 /*
294  * Return the MAC client handle of the primary MAC client for the
295  * specified MAC instance, or NULL otherwise.
296  */
297 mac_client_impl_t *
298 mac_primary_client_handle(mac_impl_t *mip)
299 {
300 	mac_client_impl_t *mcip;
301 
302 	if (mip->mi_state_flags & MIS_IS_VNIC)
303 		return (mac_vnic_lower(mip));
304 
305 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
306 
307 	for (mcip = mip->mi_clients_list; mcip != NULL;
308 	    mcip = mcip->mci_client_next) {
309 		if (MCIP_DATAPATH_SETUP(mcip) && mac_is_primary_client(mcip))
310 			return (mcip);
311 	}
312 	return (NULL);
313 }
314 
315 /*
316  * Open a MAC specified by its MAC name.
317  */
318 int
319 mac_open(const char *macname, mac_handle_t *mhp)
320 {
321 	mac_impl_t	*mip;
322 	int		err;
323 
324 	/*
325 	 * Look up its entry in the global hash table.
326 	 */
327 	if ((err = mac_hold(macname, &mip)) != 0)
328 		return (err);
329 
330 	/*
331 	 * Hold the dip associated to the MAC to prevent it from being
332 	 * detached. For a softmac, its underlying dip is held by the
333 	 * mi_open() callback.
334 	 *
335 	 * This is done to be more tolerant with some defective drivers,
336 	 * which incorrectly handle mac_unregister() failure in their
337 	 * xxx_detach() routine. For example, some drivers ignore the
338 	 * failure of mac_unregister() and free all resources that
339 	 * that are needed for data transmition.
340 	 */
341 	e_ddi_hold_devi(mip->mi_dip);
342 
343 	if (!(mip->mi_callbacks->mc_callbacks & MC_OPEN)) {
344 		*mhp = (mac_handle_t)mip;
345 		return (0);
346 	}
347 
348 	/*
349 	 * The mac perimeter is used in both mac_open and mac_close by the
350 	 * framework to single thread the MC_OPEN/MC_CLOSE of drivers.
351 	 */
352 	i_mac_perim_enter(mip);
353 	mip->mi_oref++;
354 	if (mip->mi_oref != 1 || ((err = mip->mi_open(mip->mi_driver)) == 0)) {
355 		*mhp = (mac_handle_t)mip;
356 		i_mac_perim_exit(mip);
357 		return (0);
358 	}
359 	mip->mi_oref--;
360 	ddi_release_devi(mip->mi_dip);
361 	mac_rele(mip);
362 	i_mac_perim_exit(mip);
363 	return (err);
364 }
365 
366 /*
367  * Open a MAC specified by its linkid.
368  */
369 int
370 mac_open_by_linkid(datalink_id_t linkid, mac_handle_t *mhp)
371 {
372 	dls_dl_handle_t	dlh;
373 	int		err;
374 
375 	if ((err = dls_devnet_hold_tmp(linkid, &dlh)) != 0)
376 		return (err);
377 
378 	dls_devnet_prop_task_wait(dlh);
379 
380 	err = mac_open(dls_devnet_mac(dlh), mhp);
381 
382 	dls_devnet_rele_tmp(dlh);
383 	return (err);
384 }
385 
386 /*
387  * Open a MAC specified by its link name.
388  */
389 int
390 mac_open_by_linkname(const char *link, mac_handle_t *mhp)
391 {
392 	datalink_id_t	linkid;
393 	int		err;
394 
395 	if ((err = dls_mgmt_get_linkid(link, &linkid)) != 0)
396 		return (err);
397 	return (mac_open_by_linkid(linkid, mhp));
398 }
399 
400 /*
401  * Close the specified MAC.
402  */
403 void
404 mac_close(mac_handle_t mh)
405 {
406 	mac_impl_t	*mip = (mac_impl_t *)mh;
407 
408 	i_mac_perim_enter(mip);
409 	/*
410 	 * The mac perimeter is used in both mac_open and mac_close by the
411 	 * framework to single thread the MC_OPEN/MC_CLOSE of drivers.
412 	 */
413 	if (mip->mi_callbacks->mc_callbacks & MC_OPEN) {
414 		ASSERT(mip->mi_oref != 0);
415 		if (--mip->mi_oref == 0) {
416 			if ((mip->mi_callbacks->mc_callbacks & MC_CLOSE))
417 				mip->mi_close(mip->mi_driver);
418 		}
419 	}
420 	i_mac_perim_exit(mip);
421 	ddi_release_devi(mip->mi_dip);
422 	mac_rele(mip);
423 }
424 
425 /*
426  * Misc utility functions to retrieve various information about a MAC
427  * instance or a MAC client.
428  */
429 
430 const mac_info_t *
431 mac_info(mac_handle_t mh)
432 {
433 	return (&((mac_impl_t *)mh)->mi_info);
434 }
435 
436 dev_info_t *
437 mac_devinfo_get(mac_handle_t mh)
438 {
439 	return (((mac_impl_t *)mh)->mi_dip);
440 }
441 
442 void *
443 mac_driver(mac_handle_t mh)
444 {
445 	return (((mac_impl_t *)mh)->mi_driver);
446 }
447 
448 const char *
449 mac_name(mac_handle_t mh)
450 {
451 	return (((mac_impl_t *)mh)->mi_name);
452 }
453 
454 int
455 mac_type(mac_handle_t mh)
456 {
457 	return (((mac_impl_t *)mh)->mi_type->mt_type);
458 }
459 
460 int
461 mac_nativetype(mac_handle_t mh)
462 {
463 	return (((mac_impl_t *)mh)->mi_type->mt_nativetype);
464 }
465 
466 char *
467 mac_client_name(mac_client_handle_t mch)
468 {
469 	return (((mac_client_impl_t *)mch)->mci_name);
470 }
471 
472 minor_t
473 mac_minor(mac_handle_t mh)
474 {
475 	return (((mac_impl_t *)mh)->mi_minor);
476 }
477 
478 /*
479  * Return the VID associated with a MAC client. This function should
480  * be called for clients which are associated with only one VID.
481  */
482 uint16_t
483 mac_client_vid(mac_client_handle_t mch)
484 {
485 	uint16_t		vid = VLAN_ID_NONE;
486 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
487 	flow_desc_t		flow_desc;
488 
489 	if (mcip->mci_nflents == 0)
490 		return (vid);
491 
492 	ASSERT(MCIP_DATAPATH_SETUP(mcip) && mac_client_single_rcvr(mcip));
493 
494 	mac_flow_get_desc(mcip->mci_flent, &flow_desc);
495 	if ((flow_desc.fd_mask & FLOW_LINK_VID) != 0)
496 		vid = flow_desc.fd_vid;
497 
498 	return (vid);
499 }
500 
501 /*
502  * Return whether the specified MAC client corresponds to a VLAN VNIC.
503  */
504 boolean_t
505 mac_client_is_vlan_vnic(mac_client_handle_t mch)
506 {
507 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
508 
509 	return (((mcip->mci_state_flags & MCIS_IS_VNIC) != 0) &&
510 	    ((mcip->mci_flent->fe_type & FLOW_PRIMARY_MAC) != 0));
511 }
512 
513 /*
514  * Return the link speed associated with the specified MAC client.
515  *
516  * The link speed of a MAC client is equal to the smallest value of
517  * 1) the current link speed of the underlying NIC, or
518  * 2) the bandwidth limit set for the MAC client.
519  *
520  * Note that the bandwidth limit can be higher than the speed
521  * of the underlying NIC. This is allowed to avoid spurious
522  * administration action failures or artifically lowering the
523  * bandwidth limit of a link that may  have temporarily lowered
524  * its link speed due to hardware problem or administrator action.
525  */
526 static uint64_t
527 mac_client_ifspeed(mac_client_impl_t *mcip)
528 {
529 	mac_impl_t *mip = mcip->mci_mip;
530 	uint64_t nic_speed;
531 
532 	nic_speed = mac_stat_get((mac_handle_t)mip, MAC_STAT_IFSPEED);
533 
534 	if (nic_speed == 0) {
535 		return (0);
536 	} else {
537 		uint64_t policy_limit = (uint64_t)-1;
538 
539 		if (MCIP_RESOURCE_PROPS_MASK(mcip) & MRP_MAXBW)
540 			policy_limit = MCIP_RESOURCE_PROPS_MAXBW(mcip);
541 
542 		return (MIN(policy_limit, nic_speed));
543 	}
544 }
545 
546 /*
547  * Return the link state of the specified client. If here are more
548  * than one clients of the underying mac_impl_t, the link state
549  * will always be UP regardless of the link state of the underlying
550  * mac_impl_t. This is needed to allow the MAC clients to continue
551  * to communicate with each other even when the physical link of
552  * their mac_impl_t is down.
553  */
554 static uint64_t
555 mac_client_link_state(mac_client_impl_t *mcip)
556 {
557 	mac_impl_t *mip = mcip->mci_mip;
558 	uint16_t vid;
559 	mac_client_impl_t *mci_list;
560 	mac_unicast_impl_t *mui_list, *oth_mui_list;
561 
562 	/*
563 	 * Returns LINK_STATE_UP if there are other MAC clients defined on
564 	 * mac_impl_t which share same VLAN ID as that of mcip. Note that
565 	 * if 'mcip' has more than one VID's then we match ANY one of the
566 	 * VID's with other MAC client's VID's and return LINK_STATE_UP.
567 	 */
568 	rw_enter(&mcip->mci_rw_lock, RW_READER);
569 	for (mui_list = mcip->mci_unicast_list; mui_list != NULL;
570 	    mui_list = mui_list->mui_next) {
571 		vid = mui_list->mui_vid;
572 		for (mci_list = mip->mi_clients_list; mci_list != NULL;
573 		    mci_list = mci_list->mci_client_next) {
574 			if (mci_list == mcip)
575 				continue;
576 			for (oth_mui_list = mci_list->mci_unicast_list;
577 			    oth_mui_list != NULL; oth_mui_list = oth_mui_list->
578 			    mui_next) {
579 				if (vid == oth_mui_list->mui_vid) {
580 					rw_exit(&mcip->mci_rw_lock);
581 					return (LINK_STATE_UP);
582 				}
583 			}
584 		}
585 	}
586 	rw_exit(&mcip->mci_rw_lock);
587 
588 	return (mac_stat_get((mac_handle_t)mip, MAC_STAT_LINK_STATE));
589 }
590 
591 /*
592  * These statistics are consumed by dladm show-link -s <vnic>,
593  * dladm show-vnic -s and netstat. With the introduction of dlstat,
594  * dladm show-link -s and dladm show-vnic -s witll be EOL'ed while
595  * netstat will consume from kstats introduced for dlstat. This code
596  * will be removed at that time.
597  */
598 
599 /*
600  * Return the statistics of a MAC client. These statistics are different
601  * then the statistics of the underlying MAC which are returned by
602  * mac_stat_get().
603  *
604  * Note that for things based on the tx and rx stats, mac will end up clobbering
605  * those stats when the underlying set of rings in the srs changes. As such, we
606  * need to source not only the current set, but also the historical set when
607  * returning to the client, lest our counters appear to go backwards.
608  */
609 uint64_t
610 mac_client_stat_get(mac_client_handle_t mch, uint_t stat)
611 {
612 	mac_client_impl_t 	*mcip = (mac_client_impl_t *)mch;
613 	mac_impl_t 		*mip = mcip->mci_mip;
614 	flow_entry_t 		*flent = mcip->mci_flent;
615 	mac_soft_ring_set_t 	*mac_srs;
616 	mac_rx_stats_t		*mac_rx_stat, *old_rx_stat;
617 	mac_tx_stats_t		*mac_tx_stat, *old_tx_stat;
618 	int i;
619 	uint64_t val = 0;
620 
621 	mac_srs = (mac_soft_ring_set_t *)(flent->fe_tx_srs);
622 	mac_tx_stat = &mac_srs->srs_tx.st_stat;
623 	old_rx_stat = &mcip->mci_misc_stat.mms_defunctrxlanestats;
624 	old_tx_stat = &mcip->mci_misc_stat.mms_defuncttxlanestats;
625 
626 	switch (stat) {
627 	case MAC_STAT_LINK_STATE:
628 		val = mac_client_link_state(mcip);
629 		break;
630 	case MAC_STAT_LINK_UP:
631 		val = (mac_client_link_state(mcip) == LINK_STATE_UP);
632 		break;
633 	case MAC_STAT_PROMISC:
634 		val = mac_stat_get((mac_handle_t)mip, MAC_STAT_PROMISC);
635 		break;
636 	case MAC_STAT_LOWLINK_STATE:
637 		val = mac_stat_get((mac_handle_t)mip, MAC_STAT_LOWLINK_STATE);
638 		break;
639 	case MAC_STAT_IFSPEED:
640 		val = mac_client_ifspeed(mcip);
641 		break;
642 	case MAC_STAT_MULTIRCV:
643 		val = mcip->mci_misc_stat.mms_multircv;
644 		break;
645 	case MAC_STAT_BRDCSTRCV:
646 		val = mcip->mci_misc_stat.mms_brdcstrcv;
647 		break;
648 	case MAC_STAT_MULTIXMT:
649 		val = mcip->mci_misc_stat.mms_multixmt;
650 		break;
651 	case MAC_STAT_BRDCSTXMT:
652 		val = mcip->mci_misc_stat.mms_brdcstxmt;
653 		break;
654 	case MAC_STAT_OBYTES:
655 		val = mac_tx_stat->mts_obytes;
656 		val += old_tx_stat->mts_obytes;
657 		break;
658 	case MAC_STAT_OPACKETS:
659 		val = mac_tx_stat->mts_opackets;
660 		val += old_tx_stat->mts_opackets;
661 		break;
662 	case MAC_STAT_OERRORS:
663 		val = mac_tx_stat->mts_oerrors;
664 		val += old_tx_stat->mts_oerrors;
665 		break;
666 	case MAC_STAT_IPACKETS:
667 		for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
668 			mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
669 			mac_rx_stat = &mac_srs->srs_rx.sr_stat;
670 			val += mac_rx_stat->mrs_intrcnt +
671 			    mac_rx_stat->mrs_pollcnt + mac_rx_stat->mrs_lclcnt;
672 		}
673 		val += old_rx_stat->mrs_intrcnt + old_rx_stat->mrs_pollcnt +
674 		    old_rx_stat->mrs_lclcnt;
675 		break;
676 	case MAC_STAT_RBYTES:
677 		for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
678 			mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
679 			mac_rx_stat = &mac_srs->srs_rx.sr_stat;
680 			val += mac_rx_stat->mrs_intrbytes +
681 			    mac_rx_stat->mrs_pollbytes +
682 			    mac_rx_stat->mrs_lclbytes;
683 		}
684 		val += old_rx_stat->mrs_intrbytes + old_rx_stat->mrs_pollbytes +
685 		    old_rx_stat->mrs_lclbytes;
686 		break;
687 	case MAC_STAT_IERRORS:
688 		for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
689 			mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
690 			mac_rx_stat = &mac_srs->srs_rx.sr_stat;
691 			val += mac_rx_stat->mrs_ierrors;
692 		}
693 		val += old_rx_stat->mrs_ierrors;
694 		break;
695 	default:
696 		val = mac_driver_stat_default(mip, stat);
697 		break;
698 	}
699 
700 	return (val);
701 }
702 
703 /*
704  * Return the statistics of the specified MAC instance.
705  */
706 uint64_t
707 mac_stat_get(mac_handle_t mh, uint_t stat)
708 {
709 	mac_impl_t	*mip = (mac_impl_t *)mh;
710 	uint64_t	val;
711 	int		ret;
712 
713 	/*
714 	 * The range of stat determines where it is maintained.  Stat
715 	 * values from 0 up to (but not including) MAC_STAT_MIN are
716 	 * mainteined by the mac module itself.  Everything else is
717 	 * maintained by the driver.
718 	 *
719 	 * If the mac_impl_t being queried corresponds to a VNIC,
720 	 * the stats need to be queried from the lower MAC client
721 	 * corresponding to the VNIC. (The mac_link_update()
722 	 * invoked by the driver to the lower MAC causes the *lower
723 	 * MAC* to update its mi_linkstate, and send a notification
724 	 * to its MAC clients. Due to the VNIC passthrough,
725 	 * these notifications are sent to the upper MAC clients
726 	 * of the VNIC directly, and the upper mac_impl_t of the VNIC
727 	 * does not have a valid mi_linkstate.
728 	 */
729 	if (stat < MAC_STAT_MIN && !(mip->mi_state_flags & MIS_IS_VNIC)) {
730 		/* these stats are maintained by the mac module itself */
731 		switch (stat) {
732 		case MAC_STAT_LINK_STATE:
733 			return (mip->mi_linkstate);
734 		case MAC_STAT_LINK_UP:
735 			return (mip->mi_linkstate == LINK_STATE_UP);
736 		case MAC_STAT_PROMISC:
737 			return (mip->mi_devpromisc != 0);
738 		case MAC_STAT_LOWLINK_STATE:
739 			return (mip->mi_lowlinkstate);
740 		default:
741 			ASSERT(B_FALSE);
742 		}
743 	}
744 
745 	/*
746 	 * Call the driver to get the given statistic.
747 	 */
748 	ret = mip->mi_getstat(mip->mi_driver, stat, &val);
749 	if (ret != 0) {
750 		/*
751 		 * The driver doesn't support this statistic.  Get the
752 		 * statistic's default value.
753 		 */
754 		val = mac_driver_stat_default(mip, stat);
755 	}
756 	return (val);
757 }
758 
759 /*
760  * Query hardware rx ring corresponding to the pseudo ring.
761  */
762 uint64_t
763 mac_pseudo_rx_ring_stat_get(mac_ring_handle_t handle, uint_t stat)
764 {
765 	return (mac_rx_ring_stat_get(handle, stat));
766 }
767 
768 /*
769  * Query hardware tx ring corresponding to the pseudo ring.
770  */
771 uint64_t
772 mac_pseudo_tx_ring_stat_get(mac_ring_handle_t handle, uint_t stat)
773 {
774 	return (mac_tx_ring_stat_get(handle, stat));
775 }
776 
777 /*
778  * Utility function which returns the VID associated with a flow entry.
779  */
780 uint16_t
781 i_mac_flow_vid(flow_entry_t *flent)
782 {
783 	flow_desc_t	flow_desc;
784 
785 	mac_flow_get_desc(flent, &flow_desc);
786 
787 	if ((flow_desc.fd_mask & FLOW_LINK_VID) != 0)
788 		return (flow_desc.fd_vid);
789 	return (VLAN_ID_NONE);
790 }
791 
792 /*
793  * Verify the validity of the specified unicast MAC address. Returns B_TRUE
794  * if the address is valid, B_FALSE otherwise (multicast address, or incorrect
795  * length.
796  */
797 boolean_t
798 mac_unicst_verify(mac_handle_t mh, const uint8_t *addr, uint_t len)
799 {
800 	mac_impl_t	*mip = (mac_impl_t *)mh;
801 
802 	/*
803 	 * Verify the address. No lock is needed since mi_type and plugin
804 	 * details don't change after mac_register().
805 	 */
806 	if ((len != mip->mi_type->mt_addr_length) ||
807 	    (mip->mi_type->mt_ops.mtops_unicst_verify(addr,
808 	    mip->mi_pdata)) != 0) {
809 		return (B_FALSE);
810 	} else {
811 		return (B_TRUE);
812 	}
813 }
814 
815 void
816 mac_sdu_get(mac_handle_t mh, uint_t *min_sdu, uint_t *max_sdu)
817 {
818 	mac_impl_t	*mip = (mac_impl_t *)mh;
819 
820 	if (min_sdu != NULL)
821 		*min_sdu = mip->mi_sdu_min;
822 	if (max_sdu != NULL)
823 		*max_sdu = mip->mi_sdu_max;
824 }
825 
826 void
827 mac_sdu_get2(mac_handle_t mh, uint_t *min_sdu, uint_t *max_sdu,
828     uint_t *multicast_sdu)
829 {
830 	mac_impl_t	*mip = (mac_impl_t *)mh;
831 
832 	if (min_sdu != NULL)
833 		*min_sdu = mip->mi_sdu_min;
834 	if (max_sdu != NULL)
835 		*max_sdu = mip->mi_sdu_max;
836 	if (multicast_sdu != NULL)
837 		*multicast_sdu = mip->mi_sdu_multicast;
838 }
839 
840 /*
841  * Update the MAC unicast address of the specified client's flows. Currently
842  * only one unicast MAC unicast address is allowed per client.
843  */
844 static void
845 mac_unicast_update_client_flow(mac_client_impl_t *mcip)
846 {
847 	mac_impl_t *mip = mcip->mci_mip;
848 	flow_entry_t *flent = mcip->mci_flent;
849 	mac_address_t *map = mcip->mci_unicast;
850 	flow_desc_t flow_desc;
851 
852 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
853 	ASSERT(flent != NULL);
854 
855 	mac_flow_get_desc(flent, &flow_desc);
856 	ASSERT(flow_desc.fd_mask & FLOW_LINK_DST);
857 
858 	bcopy(map->ma_addr, flow_desc.fd_dst_mac, map->ma_len);
859 	mac_flow_set_desc(flent, &flow_desc);
860 
861 	/*
862 	 * The v6 local and SLAAC addrs (used by mac protection) need to be
863 	 * regenerated because our mac address has changed.
864 	 */
865 	mac_protect_update_mac_token(mcip);
866 
867 	/*
868 	 * A MAC client could have one MAC address but multiple
869 	 * VLANs. In that case update the flow entries corresponding
870 	 * to all VLANs of the MAC client.
871 	 */
872 	for (flent = mcip->mci_flent_list; flent != NULL;
873 	    flent = flent->fe_client_next) {
874 		mac_flow_get_desc(flent, &flow_desc);
875 		if (!(flent->fe_type & FLOW_PRIMARY_MAC ||
876 		    flent->fe_type & FLOW_VNIC_MAC))
877 			continue;
878 
879 		bcopy(map->ma_addr, flow_desc.fd_dst_mac, map->ma_len);
880 		mac_flow_set_desc(flent, &flow_desc);
881 	}
882 }
883 
884 /*
885  * Update all clients that share the same unicast address.
886  */
887 void
888 mac_unicast_update_clients(mac_impl_t *mip, mac_address_t *map)
889 {
890 	mac_client_impl_t *mcip;
891 
892 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
893 
894 	/*
895 	 * Find all clients that share the same unicast MAC address and update
896 	 * them appropriately.
897 	 */
898 	for (mcip = mip->mi_clients_list; mcip != NULL;
899 	    mcip = mcip->mci_client_next) {
900 		/*
901 		 * Ignore clients that don't share this MAC address.
902 		 */
903 		if (map != mcip->mci_unicast)
904 			continue;
905 
906 		/*
907 		 * Update those clients with same old unicast MAC address.
908 		 */
909 		mac_unicast_update_client_flow(mcip);
910 	}
911 }
912 
913 /*
914  * Update the unicast MAC address of the specified VNIC MAC client.
915  *
916  * Check whether the operation is valid. Any of following cases should fail:
917  *
918  * 1. It's a VLAN type of VNIC.
919  * 2. The new value is current "primary" MAC address.
920  * 3. The current MAC address is shared with other clients.
921  * 4. The new MAC address has been used. This case will be valid when
922  *    client migration is fully supported.
923  */
924 int
925 mac_vnic_unicast_set(mac_client_handle_t mch, const uint8_t *addr)
926 {
927 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
928 	mac_impl_t *mip = mcip->mci_mip;
929 	mac_address_t *map = mcip->mci_unicast;
930 	int err;
931 
932 	ASSERT(!(mip->mi_state_flags & MIS_IS_VNIC));
933 	ASSERT(mcip->mci_state_flags & MCIS_IS_VNIC);
934 	ASSERT(mcip->mci_flags != MAC_CLIENT_FLAGS_PRIMARY);
935 
936 	i_mac_perim_enter(mip);
937 
938 	/*
939 	 * If this is a VLAN type of VNIC, it's using "primary" MAC address
940 	 * of the underlying interface. Must fail here. Refer to case 1 above.
941 	 */
942 	if (bcmp(map->ma_addr, mip->mi_addr, map->ma_len) == 0) {
943 		i_mac_perim_exit(mip);
944 		return (ENOTSUP);
945 	}
946 
947 	/*
948 	 * If the new address is the "primary" one, must fail. Refer to
949 	 * case 2 above.
950 	 */
951 	if (bcmp(addr, mip->mi_addr, map->ma_len) == 0) {
952 		i_mac_perim_exit(mip);
953 		return (EACCES);
954 	}
955 
956 	/*
957 	 * If the address is shared by multiple clients, must fail. Refer
958 	 * to case 3 above.
959 	 */
960 	if (mac_check_macaddr_shared(map)) {
961 		i_mac_perim_exit(mip);
962 		return (EBUSY);
963 	}
964 
965 	/*
966 	 * If the new address has been used, must fail for now. Refer to
967 	 * case 4 above.
968 	 */
969 	if (mac_find_macaddr(mip, (uint8_t *)addr) != NULL) {
970 		i_mac_perim_exit(mip);
971 		return (ENOTSUP);
972 	}
973 
974 	/*
975 	 * Update the MAC address.
976 	 */
977 	err = mac_update_macaddr(map, (uint8_t *)addr);
978 
979 	if (err != 0) {
980 		i_mac_perim_exit(mip);
981 		return (err);
982 	}
983 
984 	/*
985 	 * Update all flows of this MAC client.
986 	 */
987 	mac_unicast_update_client_flow(mcip);
988 
989 	i_mac_perim_exit(mip);
990 	return (0);
991 }
992 
993 /*
994  * Program the new primary unicast address of the specified MAC.
995  *
996  * Function mac_update_macaddr() takes care different types of underlying
997  * MAC. If the underlying MAC is VNIC, the VNIC driver must have registerd
998  * mi_unicst() entry point, that indirectly calls mac_vnic_unicast_set()
999  * which will take care of updating the MAC address of the corresponding
1000  * MAC client.
1001  *
1002  * This is the only interface that allow the client to update the "primary"
1003  * MAC address of the underlying MAC. The new value must have not been
1004  * used by other clients.
1005  */
1006 int
1007 mac_unicast_primary_set(mac_handle_t mh, const uint8_t *addr)
1008 {
1009 	mac_impl_t *mip = (mac_impl_t *)mh;
1010 	mac_address_t *map;
1011 	int err;
1012 
1013 	/* verify the address validity */
1014 	if (!mac_unicst_verify(mh, addr, mip->mi_type->mt_addr_length))
1015 		return (EINVAL);
1016 
1017 	i_mac_perim_enter(mip);
1018 
1019 	/*
1020 	 * If the new value is the same as the current primary address value,
1021 	 * there's nothing to do.
1022 	 */
1023 	if (bcmp(addr, mip->mi_addr, mip->mi_type->mt_addr_length) == 0) {
1024 		i_mac_perim_exit(mip);
1025 		return (0);
1026 	}
1027 
1028 	if (mac_find_macaddr(mip, (uint8_t *)addr) != 0) {
1029 		i_mac_perim_exit(mip);
1030 		return (EBUSY);
1031 	}
1032 
1033 	map = mac_find_macaddr(mip, mip->mi_addr);
1034 	ASSERT(map != NULL);
1035 
1036 	/*
1037 	 * Update the MAC address.
1038 	 */
1039 	if (mip->mi_state_flags & MIS_IS_AGGR) {
1040 		mac_capab_aggr_t aggr_cap;
1041 
1042 		/*
1043 		 * If the mac is an aggregation, other than the unicast
1044 		 * addresses programming, aggr must be informed about this
1045 		 * primary unicst address change to change its mac address
1046 		 * policy to be user-specified.
1047 		 */
1048 		ASSERT(map->ma_type == MAC_ADDRESS_TYPE_UNICAST_CLASSIFIED);
1049 		VERIFY(i_mac_capab_get(mh, MAC_CAPAB_AGGR, &aggr_cap));
1050 		err = aggr_cap.mca_unicst(mip->mi_driver, addr);
1051 		if (err == 0)
1052 			bcopy(addr, map->ma_addr, map->ma_len);
1053 	} else {
1054 		err = mac_update_macaddr(map, (uint8_t *)addr);
1055 	}
1056 
1057 	if (err != 0) {
1058 		i_mac_perim_exit(mip);
1059 		return (err);
1060 	}
1061 
1062 	mac_unicast_update_clients(mip, map);
1063 
1064 	/*
1065 	 * Save the new primary MAC address in mac_impl_t.
1066 	 */
1067 	bcopy(addr, mip->mi_addr, mip->mi_type->mt_addr_length);
1068 
1069 	i_mac_perim_exit(mip);
1070 
1071 	if (err == 0)
1072 		i_mac_notify(mip, MAC_NOTE_UNICST);
1073 
1074 	return (err);
1075 }
1076 
1077 /*
1078  * Return the current primary MAC address of the specified MAC.
1079  */
1080 void
1081 mac_unicast_primary_get(mac_handle_t mh, uint8_t *addr)
1082 {
1083 	mac_impl_t *mip = (mac_impl_t *)mh;
1084 
1085 	rw_enter(&mip->mi_rw_lock, RW_READER);
1086 	bcopy(mip->mi_addr, addr, mip->mi_type->mt_addr_length);
1087 	rw_exit(&mip->mi_rw_lock);
1088 }
1089 
1090 /*
1091  * Return the secondary MAC address for the specified handle
1092  */
1093 void
1094 mac_unicast_secondary_get(mac_client_handle_t mh, uint8_t *addr)
1095 {
1096 	mac_client_impl_t *mcip = (mac_client_impl_t *)mh;
1097 
1098 	ASSERT(mcip->mci_unicast != NULL);
1099 	bcopy(mcip->mci_unicast->ma_addr, addr, mcip->mci_unicast->ma_len);
1100 }
1101 
1102 /*
1103  * Return information about the use of the primary MAC address of the
1104  * specified MAC instance:
1105  *
1106  * - if client_name is non-NULL, it must point to a string of at
1107  *   least MAXNAMELEN bytes, and will be set to the name of the MAC
1108  *   client which uses the primary MAC address.
1109  *
1110  * - if in_use is non-NULL, used to return whether the primary MAC
1111  *   address is currently in use.
1112  */
1113 void
1114 mac_unicast_primary_info(mac_handle_t mh, char *client_name, boolean_t *in_use)
1115 {
1116 	mac_impl_t *mip = (mac_impl_t *)mh;
1117 	mac_client_impl_t *cur_client;
1118 
1119 	if (in_use != NULL)
1120 		*in_use = B_FALSE;
1121 	if (client_name != NULL)
1122 		bzero(client_name, MAXNAMELEN);
1123 
1124 	/*
1125 	 * The mi_rw_lock is used to protect threads that don't hold the
1126 	 * mac perimeter to get a consistent view of the mi_clients_list.
1127 	 * Threads that modify the list must hold both the mac perimeter and
1128 	 * mi_rw_lock(RW_WRITER)
1129 	 */
1130 	rw_enter(&mip->mi_rw_lock, RW_READER);
1131 	for (cur_client = mip->mi_clients_list; cur_client != NULL;
1132 	    cur_client = cur_client->mci_client_next) {
1133 		if (mac_is_primary_client(cur_client) ||
1134 		    (mip->mi_state_flags & MIS_IS_VNIC)) {
1135 			rw_exit(&mip->mi_rw_lock);
1136 			if (in_use != NULL)
1137 				*in_use = B_TRUE;
1138 			if (client_name != NULL) {
1139 				bcopy(cur_client->mci_name, client_name,
1140 				    MAXNAMELEN);
1141 			}
1142 			return;
1143 		}
1144 	}
1145 	rw_exit(&mip->mi_rw_lock);
1146 }
1147 
1148 /*
1149  * Return the current destination MAC address of the specified MAC.
1150  */
1151 boolean_t
1152 mac_dst_get(mac_handle_t mh, uint8_t *addr)
1153 {
1154 	mac_impl_t *mip = (mac_impl_t *)mh;
1155 
1156 	rw_enter(&mip->mi_rw_lock, RW_READER);
1157 	if (mip->mi_dstaddr_set)
1158 		bcopy(mip->mi_dstaddr, addr, mip->mi_type->mt_addr_length);
1159 	rw_exit(&mip->mi_rw_lock);
1160 	return (mip->mi_dstaddr_set);
1161 }
1162 
1163 /*
1164  * Add the specified MAC client to the list of clients which opened
1165  * the specified MAC.
1166  */
1167 static void
1168 mac_client_add(mac_client_impl_t *mcip)
1169 {
1170 	mac_impl_t *mip = mcip->mci_mip;
1171 
1172 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1173 
1174 	/* add VNIC to the front of the list */
1175 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
1176 	mcip->mci_client_next = mip->mi_clients_list;
1177 	mip->mi_clients_list = mcip;
1178 	mip->mi_nclients++;
1179 	rw_exit(&mip->mi_rw_lock);
1180 }
1181 
1182 /*
1183  * Remove the specified MAC client from the list of clients which opened
1184  * the specified MAC.
1185  */
1186 static void
1187 mac_client_remove(mac_client_impl_t *mcip)
1188 {
1189 	mac_impl_t *mip = mcip->mci_mip;
1190 	mac_client_impl_t **prev, *cclient;
1191 
1192 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1193 
1194 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
1195 	prev = &mip->mi_clients_list;
1196 	cclient = *prev;
1197 	while (cclient != NULL && cclient != mcip) {
1198 		prev = &cclient->mci_client_next;
1199 		cclient = *prev;
1200 	}
1201 	ASSERT(cclient != NULL);
1202 	*prev = cclient->mci_client_next;
1203 	mip->mi_nclients--;
1204 	rw_exit(&mip->mi_rw_lock);
1205 }
1206 
1207 static mac_unicast_impl_t *
1208 mac_client_find_vid(mac_client_impl_t *mcip, uint16_t vid)
1209 {
1210 	mac_unicast_impl_t *muip = mcip->mci_unicast_list;
1211 
1212 	while ((muip != NULL) && (muip->mui_vid != vid))
1213 		muip = muip->mui_next;
1214 
1215 	return (muip);
1216 }
1217 
1218 /*
1219  * Return whether the specified (MAC address, VID) tuple is already used by
1220  * one of the MAC clients associated with the specified MAC.
1221  */
1222 static boolean_t
1223 mac_addr_in_use(mac_impl_t *mip, uint8_t *mac_addr, uint16_t vid)
1224 {
1225 	mac_client_impl_t *client;
1226 	mac_address_t *map;
1227 
1228 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1229 
1230 	for (client = mip->mi_clients_list; client != NULL;
1231 	    client = client->mci_client_next) {
1232 
1233 		/*
1234 		 * Ignore clients that don't have unicast address.
1235 		 */
1236 		if (client->mci_unicast_list == NULL)
1237 			continue;
1238 
1239 		map = client->mci_unicast;
1240 
1241 		if ((bcmp(mac_addr, map->ma_addr, map->ma_len) == 0) &&
1242 		    (mac_client_find_vid(client, vid) != NULL)) {
1243 			return (B_TRUE);
1244 		}
1245 	}
1246 
1247 	return (B_FALSE);
1248 }
1249 
1250 /*
1251  * Generate a random MAC address. The MAC address prefix is
1252  * stored in the array pointed to by mac_addr, and its length, in bytes,
1253  * is specified by prefix_len. The least significant bits
1254  * after prefix_len bytes are generated, and stored after the prefix
1255  * in the mac_addr array.
1256  */
1257 int
1258 mac_addr_random(mac_client_handle_t mch, uint_t prefix_len,
1259     uint8_t *mac_addr, mac_diag_t *diag)
1260 {
1261 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1262 	mac_impl_t *mip = mcip->mci_mip;
1263 	size_t addr_len = mip->mi_type->mt_addr_length;
1264 
1265 	if (prefix_len >= addr_len) {
1266 		*diag = MAC_DIAG_MACPREFIXLEN_INVALID;
1267 		return (EINVAL);
1268 	}
1269 
1270 	/* check the prefix value */
1271 	if (prefix_len > 0) {
1272 		bzero(mac_addr + prefix_len, addr_len - prefix_len);
1273 		if (!mac_unicst_verify((mac_handle_t)mip, mac_addr,
1274 		    addr_len)) {
1275 			*diag = MAC_DIAG_MACPREFIX_INVALID;
1276 			return (EINVAL);
1277 		}
1278 	}
1279 
1280 	/* generate the MAC address */
1281 	if (prefix_len < addr_len) {
1282 		(void) random_get_pseudo_bytes(mac_addr +
1283 		    prefix_len, addr_len - prefix_len);
1284 	}
1285 
1286 	*diag = 0;
1287 	return (0);
1288 }
1289 
1290 /*
1291  * Set the priority range for this MAC client. This will be used to
1292  * determine the absolute priority for the threads created for this
1293  * MAC client using the specified "low", "medium" and "high" level.
1294  * This will also be used for any subflows on this MAC client.
1295  */
1296 #define	MAC_CLIENT_SET_PRIORITY_RANGE(mcip, pri) {			\
1297 	(mcip)->mci_min_pri = FLOW_MIN_PRIORITY(MINCLSYSPRI,	\
1298 	    MAXCLSYSPRI, (pri));					\
1299 	(mcip)->mci_max_pri = FLOW_MAX_PRIORITY(MINCLSYSPRI,	\
1300 	    MAXCLSYSPRI, (mcip)->mci_min_pri);				\
1301 	}
1302 
1303 /*
1304  * MAC client open entry point. Return a new MAC client handle. Each
1305  * MAC client is associated with a name, specified through the 'name'
1306  * argument.
1307  */
1308 int
1309 mac_client_open(mac_handle_t mh, mac_client_handle_t *mchp, char *name,
1310     uint16_t flags)
1311 {
1312 	mac_impl_t		*mip = (mac_impl_t *)mh;
1313 	mac_client_impl_t	*mcip;
1314 	int			err = 0;
1315 	boolean_t		share_desired;
1316 	flow_entry_t		*flent = NULL;
1317 
1318 	share_desired = (flags & MAC_OPEN_FLAGS_SHARES_DESIRED) != 0;
1319 	*mchp = NULL;
1320 
1321 	i_mac_perim_enter(mip);
1322 
1323 	if (mip->mi_state_flags & MIS_IS_VNIC) {
1324 		/*
1325 		 * The underlying MAC is a VNIC. Return the MAC client
1326 		 * handle of the lower MAC which was obtained by
1327 		 * the VNIC driver when it did its mac_client_open().
1328 		 */
1329 
1330 		mcip = mac_vnic_lower(mip);
1331 
1332 		/*
1333 		 * Note that multiple mac clients share the same mcip in
1334 		 * this case.
1335 		 */
1336 		if (flags & MAC_OPEN_FLAGS_EXCLUSIVE)
1337 			mcip->mci_state_flags |= MCIS_EXCLUSIVE;
1338 
1339 		if (flags & MAC_OPEN_FLAGS_MULTI_PRIMARY)
1340 			mcip->mci_flags |= MAC_CLIENT_FLAGS_MULTI_PRIMARY;
1341 
1342 		mip->mi_clients_list = mcip;
1343 		i_mac_perim_exit(mip);
1344 		*mchp = (mac_client_handle_t)mcip;
1345 
1346 		DTRACE_PROBE2(mac__client__open__nonallocated, mac_impl_t *,
1347 		    mcip->mci_mip, mac_client_impl_t *, mcip);
1348 
1349 		return (err);
1350 	}
1351 
1352 	mcip = kmem_cache_alloc(mac_client_impl_cache, KM_SLEEP);
1353 
1354 	mcip->mci_mip = mip;
1355 	mcip->mci_upper_mip = NULL;
1356 	mcip->mci_rx_fn = mac_pkt_drop;
1357 	mcip->mci_rx_arg = NULL;
1358 	mcip->mci_rx_p_fn = NULL;
1359 	mcip->mci_rx_p_arg = NULL;
1360 	mcip->mci_p_unicast_list = NULL;
1361 	mcip->mci_direct_rx_fn = NULL;
1362 	mcip->mci_direct_rx_arg = NULL;
1363 	mcip->mci_vidcache = MCIP_VIDCACHE_INVALID;
1364 
1365 	mcip->mci_unicast_list = NULL;
1366 
1367 	if ((flags & MAC_OPEN_FLAGS_IS_VNIC) != 0)
1368 		mcip->mci_state_flags |= MCIS_IS_VNIC;
1369 
1370 	if ((flags & MAC_OPEN_FLAGS_EXCLUSIVE) != 0)
1371 		mcip->mci_state_flags |= MCIS_EXCLUSIVE;
1372 
1373 	if ((flags & MAC_OPEN_FLAGS_IS_AGGR_PORT) != 0)
1374 		mcip->mci_state_flags |= MCIS_IS_AGGR_PORT;
1375 
1376 	if (mip->mi_state_flags & MIS_IS_AGGR)
1377 		mcip->mci_state_flags |= MCIS_IS_AGGR;
1378 
1379 	if ((flags & MAC_OPEN_FLAGS_USE_DATALINK_NAME) != 0) {
1380 		datalink_id_t	linkid;
1381 
1382 		ASSERT(name == NULL);
1383 		if ((err = dls_devnet_macname2linkid(mip->mi_name,
1384 		    &linkid)) != 0) {
1385 			goto done;
1386 		}
1387 		if ((err = dls_mgmt_get_linkinfo(linkid, mcip->mci_name, NULL,
1388 		    NULL, NULL)) != 0) {
1389 			/*
1390 			 * Use mac name if dlmgmtd is not available.
1391 			 */
1392 			if (err == EBADF) {
1393 				(void) strlcpy(mcip->mci_name, mip->mi_name,
1394 				    sizeof (mcip->mci_name));
1395 				err = 0;
1396 			} else {
1397 				goto done;
1398 			}
1399 		}
1400 		mcip->mci_state_flags |= MCIS_USE_DATALINK_NAME;
1401 	} else {
1402 		ASSERT(name != NULL);
1403 		if (strlen(name) > MAXNAMELEN) {
1404 			err = EINVAL;
1405 			goto done;
1406 		}
1407 		(void) strlcpy(mcip->mci_name, name, sizeof (mcip->mci_name));
1408 	}
1409 
1410 	if (flags & MAC_OPEN_FLAGS_MULTI_PRIMARY)
1411 		mcip->mci_flags |= MAC_CLIENT_FLAGS_MULTI_PRIMARY;
1412 
1413 	if (flags & MAC_OPEN_FLAGS_NO_UNICAST_ADDR)
1414 		mcip->mci_state_flags |= MCIS_NO_UNICAST_ADDR;
1415 
1416 	mac_protect_init(mcip);
1417 
1418 	/* the subflow table will be created dynamically */
1419 	mcip->mci_subflow_tab = NULL;
1420 
1421 	mcip->mci_misc_stat.mms_multircv = 0;
1422 	mcip->mci_misc_stat.mms_brdcstrcv = 0;
1423 	mcip->mci_misc_stat.mms_multixmt = 0;
1424 	mcip->mci_misc_stat.mms_brdcstxmt = 0;
1425 
1426 	/* Create an initial flow */
1427 
1428 	err = mac_flow_create(NULL, NULL, mcip->mci_name, NULL,
1429 	    mcip->mci_state_flags & MCIS_IS_VNIC ? FLOW_VNIC_MAC :
1430 	    FLOW_PRIMARY_MAC, &flent);
1431 	if (err != 0)
1432 		goto done;
1433 	mcip->mci_flent = flent;
1434 	FLOW_MARK(flent, FE_MC_NO_DATAPATH);
1435 	flent->fe_mcip = mcip;
1436 	/*
1437 	 * Place initial creation reference on the flow. This reference
1438 	 * is released in the corresponding delete action viz.
1439 	 * mac_unicast_remove after waiting for all transient refs to
1440 	 * to go away. The wait happens in mac_flow_wait.
1441 	 */
1442 	FLOW_REFHOLD(flent);
1443 
1444 	/*
1445 	 * Do this ahead of the mac_bcast_add() below so that the mi_nclients
1446 	 * will have the right value for mac_rx_srs_setup().
1447 	 */
1448 	mac_client_add(mcip);
1449 
1450 	mcip->mci_share = NULL;
1451 	if (share_desired)
1452 		i_mac_share_alloc(mcip);
1453 
1454 	/*
1455 	 * We will do mimimal datapath setup to allow a MAC client to
1456 	 * transmit or receive non-unicast packets without waiting
1457 	 * for mac_unicast_add.
1458 	 */
1459 	if (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR) {
1460 		if ((err = mac_client_datapath_setup(mcip, VLAN_ID_NONE,
1461 		    NULL, NULL, B_TRUE, NULL)) != 0) {
1462 			goto done;
1463 		}
1464 	}
1465 
1466 	DTRACE_PROBE2(mac__client__open__allocated, mac_impl_t *,
1467 	    mcip->mci_mip, mac_client_impl_t *, mcip);
1468 
1469 	*mchp = (mac_client_handle_t)mcip;
1470 	i_mac_perim_exit(mip);
1471 	return (0);
1472 
1473 done:
1474 	i_mac_perim_exit(mip);
1475 	mcip->mci_state_flags = 0;
1476 	mcip->mci_tx_flag = 0;
1477 	kmem_cache_free(mac_client_impl_cache, mcip);
1478 	return (err);
1479 }
1480 
1481 /*
1482  * Close the specified MAC client handle.
1483  */
1484 void
1485 mac_client_close(mac_client_handle_t mch, uint16_t flags)
1486 {
1487 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
1488 	mac_impl_t		*mip = mcip->mci_mip;
1489 	flow_entry_t		*flent;
1490 
1491 	i_mac_perim_enter(mip);
1492 
1493 	if (flags & MAC_CLOSE_FLAGS_EXCLUSIVE)
1494 		mcip->mci_state_flags &= ~MCIS_EXCLUSIVE;
1495 
1496 	if ((mcip->mci_state_flags & MCIS_IS_VNIC) &&
1497 	    !(flags & MAC_CLOSE_FLAGS_IS_VNIC)) {
1498 		/*
1499 		 * This is an upper VNIC client initiated operation.
1500 		 * The lower MAC client will be closed by the VNIC driver
1501 		 * when the VNIC is deleted.
1502 		 */
1503 
1504 		i_mac_perim_exit(mip);
1505 		return;
1506 	}
1507 
1508 	/* If we have only setup up minimal datapth setup, tear it down */
1509 	if (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR) {
1510 		mac_client_datapath_teardown((mac_client_handle_t)mcip, NULL,
1511 		    mcip->mci_flent);
1512 		mcip->mci_state_flags &= ~MCIS_NO_UNICAST_ADDR;
1513 	}
1514 
1515 	/*
1516 	 * Remove the flent associated with the MAC client
1517 	 */
1518 	flent = mcip->mci_flent;
1519 	mcip->mci_flent = NULL;
1520 	FLOW_FINAL_REFRELE(flent);
1521 
1522 	/*
1523 	 * MAC clients must remove the unicast addresses and promisc callbacks
1524 	 * they added before issuing a mac_client_close().
1525 	 */
1526 	ASSERT(mcip->mci_unicast_list == NULL);
1527 	ASSERT(mcip->mci_promisc_list == NULL);
1528 	ASSERT(mcip->mci_tx_notify_cb_list == NULL);
1529 
1530 	i_mac_share_free(mcip);
1531 	mac_protect_fini(mcip);
1532 	mac_client_remove(mcip);
1533 
1534 	i_mac_perim_exit(mip);
1535 	mcip->mci_subflow_tab = NULL;
1536 	mcip->mci_state_flags = 0;
1537 	mcip->mci_tx_flag = 0;
1538 	kmem_cache_free(mac_client_impl_cache, mch);
1539 }
1540 
1541 /*
1542  * Set the rx bypass receive callback.
1543  */
1544 boolean_t
1545 mac_rx_bypass_set(mac_client_handle_t mch, mac_direct_rx_t rx_fn, void *arg1)
1546 {
1547 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
1548 	mac_impl_t		*mip = mcip->mci_mip;
1549 
1550 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1551 
1552 	/*
1553 	 * If the mac_client is a VLAN, we should not do DLS bypass and
1554 	 * instead let the packets come up via mac_rx_deliver so the vlan
1555 	 * header can be stripped.
1556 	 */
1557 	if (mcip->mci_nvids > 0)
1558 		return (B_FALSE);
1559 
1560 	/*
1561 	 * These are not accessed directly in the data path, and hence
1562 	 * don't need any protection
1563 	 */
1564 	mcip->mci_direct_rx_fn = rx_fn;
1565 	mcip->mci_direct_rx_arg = arg1;
1566 	return (B_TRUE);
1567 }
1568 
1569 /*
1570  * Enable/Disable rx bypass. By default, bypass is assumed to be enabled.
1571  */
1572 void
1573 mac_rx_bypass_enable(mac_client_handle_t mch)
1574 {
1575 	((mac_client_impl_t *)mch)->mci_state_flags &= ~MCIS_RX_BYPASS_DISABLE;
1576 }
1577 
1578 void
1579 mac_rx_bypass_disable(mac_client_handle_t mch)
1580 {
1581 	((mac_client_impl_t *)mch)->mci_state_flags |= MCIS_RX_BYPASS_DISABLE;
1582 }
1583 
1584 /*
1585  * Set the receive callback for the specified MAC client. There can be
1586  * at most one such callback per MAC client.
1587  */
1588 void
1589 mac_rx_set(mac_client_handle_t mch, mac_rx_t rx_fn, void *arg)
1590 {
1591 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1592 	mac_impl_t	*mip = mcip->mci_mip;
1593 	mac_impl_t	*umip = mcip->mci_upper_mip;
1594 
1595 	/*
1596 	 * Instead of adding an extra set of locks and refcnts in
1597 	 * the datapath at the mac client boundary, we temporarily quiesce
1598 	 * the SRS and related entities. We then change the receive function
1599 	 * without interference from any receive data thread and then reenable
1600 	 * the data flow subsequently.
1601 	 */
1602 	i_mac_perim_enter(mip);
1603 	mac_rx_client_quiesce(mch);
1604 
1605 	mcip->mci_rx_fn = rx_fn;
1606 	mcip->mci_rx_arg = arg;
1607 	mac_rx_client_restart(mch);
1608 	i_mac_perim_exit(mip);
1609 
1610 	/*
1611 	 * If we're changing the rx function on the primary mac of a vnic,
1612 	 * make sure any secondary macs on the vnic are updated as well.
1613 	 */
1614 	if (umip != NULL) {
1615 		ASSERT((umip->mi_state_flags & MIS_IS_VNIC) != 0);
1616 		mac_vnic_secondary_update(umip);
1617 	}
1618 }
1619 
1620 /*
1621  * Reset the receive callback for the specified MAC client.
1622  */
1623 void
1624 mac_rx_clear(mac_client_handle_t mch)
1625 {
1626 	mac_rx_set(mch, mac_pkt_drop, NULL);
1627 }
1628 
1629 void
1630 mac_secondary_dup(mac_client_handle_t smch, mac_client_handle_t dmch)
1631 {
1632 	mac_client_impl_t *smcip = (mac_client_impl_t *)smch;
1633 	mac_client_impl_t *dmcip = (mac_client_impl_t *)dmch;
1634 	flow_entry_t *flent = dmcip->mci_flent;
1635 
1636 	/* This should only be called to setup secondary macs */
1637 	ASSERT((flent->fe_type & FLOW_PRIMARY_MAC) == 0);
1638 
1639 	mac_rx_set(dmch, smcip->mci_rx_fn, smcip->mci_rx_arg);
1640 	dmcip->mci_promisc_list = smcip->mci_promisc_list;
1641 
1642 	/*
1643 	 * Duplicate the primary mac resources to the secondary.
1644 	 * Since we already validated the resource controls when setting
1645 	 * them on the primary, we can ignore errors here.
1646 	 */
1647 	(void) mac_resource_ctl_set(dmch, MCIP_RESOURCE_PROPS(smcip));
1648 }
1649 
1650 /*
1651  * Called when removing a secondary MAC. Currently only clears the promisc_list
1652  * since we share the primary mac's promisc_list.
1653  */
1654 void
1655 mac_secondary_cleanup(mac_client_handle_t mch)
1656 {
1657 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1658 	flow_entry_t *flent = mcip->mci_flent;
1659 
1660 	/* This should only be called for secondary macs */
1661 	ASSERT((flent->fe_type & FLOW_PRIMARY_MAC) == 0);
1662 	mcip->mci_promisc_list = NULL;
1663 }
1664 
1665 /*
1666  * Walk the MAC client subflow table and updates their priority values.
1667  */
1668 static int
1669 mac_update_subflow_priority_cb(flow_entry_t *flent, void *arg)
1670 {
1671 	mac_flow_update_priority(arg, flent);
1672 	return (0);
1673 }
1674 
1675 void
1676 mac_update_subflow_priority(mac_client_impl_t *mcip)
1677 {
1678 	(void) mac_flow_walk(mcip->mci_subflow_tab,
1679 	    mac_update_subflow_priority_cb, mcip);
1680 }
1681 
1682 /*
1683  * Modify the TX or RX ring properties. We could either just move around
1684  * rings, i.e add/remove rings given to a client. Or this might cause the
1685  * client to move from hardware based to software or the other way around.
1686  * If we want to reset this property, then we clear the mask, additionally
1687  * if the client was given a non-default group we remove all rings except
1688  * for 1 and give it back to the default group.
1689  */
1690 int
1691 mac_client_set_rings_prop(mac_client_impl_t *mcip, mac_resource_props_t *mrp,
1692     mac_resource_props_t *tmrp)
1693 {
1694 	mac_impl_t		*mip = mcip->mci_mip;
1695 	flow_entry_t		*flent = mcip->mci_flent;
1696 	uint8_t			*mac_addr;
1697 	int			err = 0;
1698 	mac_group_t		*defgrp;
1699 	mac_group_t		*group;
1700 	mac_group_t		*ngrp;
1701 	mac_resource_props_t	*cmrp = MCIP_RESOURCE_PROPS(mcip);
1702 	uint_t			ringcnt;
1703 	boolean_t		unspec;
1704 
1705 	if (mcip->mci_share != NULL)
1706 		return (EINVAL);
1707 
1708 	if (mrp->mrp_mask & MRP_RX_RINGS) {
1709 		unspec = mrp->mrp_mask & MRP_RXRINGS_UNSPEC;
1710 		group = flent->fe_rx_ring_group;
1711 		defgrp = MAC_DEFAULT_RX_GROUP(mip);
1712 		mac_addr = flent->fe_flow_desc.fd_dst_mac;
1713 
1714 		/*
1715 		 * No resulting change. If we are resetting on a client on
1716 		 * which there was no rx rings property. For dynamic group
1717 		 * if we are setting the same number of rings already set.
1718 		 * For static group if we are requesting a group again.
1719 		 */
1720 		if (mrp->mrp_mask & MRP_RINGS_RESET) {
1721 			if (!(tmrp->mrp_mask & MRP_RX_RINGS))
1722 				return (0);
1723 		} else {
1724 			if (unspec) {
1725 				if (tmrp->mrp_mask & MRP_RXRINGS_UNSPEC)
1726 					return (0);
1727 			} else if (mip->mi_rx_group_type ==
1728 			    MAC_GROUP_TYPE_DYNAMIC) {
1729 				if ((tmrp->mrp_mask & MRP_RX_RINGS) &&
1730 				    !(tmrp->mrp_mask & MRP_RXRINGS_UNSPEC) &&
1731 				    mrp->mrp_nrxrings == tmrp->mrp_nrxrings) {
1732 					return (0);
1733 				}
1734 			}
1735 		}
1736 		/* Resetting the prop */
1737 		if (mrp->mrp_mask & MRP_RINGS_RESET) {
1738 			/*
1739 			 * We will just keep one ring and give others back if
1740 			 * we are not the primary. For the primary we give
1741 			 * all the rings in the default group except the
1742 			 * default ring. If it is a static group, then
1743 			 * we don't do anything, but clear the MRP_RX_RINGS
1744 			 * flag.
1745 			 */
1746 			if (group != defgrp) {
1747 				if (mip->mi_rx_group_type ==
1748 				    MAC_GROUP_TYPE_DYNAMIC) {
1749 					/*
1750 					 * This group has reserved rings
1751 					 * that need to be released now,
1752 					 * so does the group.
1753 					 */
1754 					MAC_RX_RING_RELEASED(mip,
1755 					    group->mrg_cur_count);
1756 					MAC_RX_GRP_RELEASED(mip);
1757 					if ((flent->fe_type &
1758 					    FLOW_PRIMARY_MAC) != 0) {
1759 						if (mip->mi_nactiveclients ==
1760 						    1) {
1761 							(void)
1762 							    mac_rx_switch_group(
1763 							    mcip, group,
1764 							    defgrp);
1765 							return (0);
1766 						} else {
1767 							cmrp->mrp_nrxrings =
1768 							    group->
1769 							    mrg_cur_count +
1770 							    defgrp->
1771 							    mrg_cur_count - 1;
1772 						}
1773 					} else {
1774 						cmrp->mrp_nrxrings = 1;
1775 					}
1776 					(void) mac_group_ring_modify(mcip,
1777 					    group, defgrp);
1778 				} else {
1779 					/*
1780 					 * If this is a static group, we
1781 					 * need to release the group. The
1782 					 * client will remain in the same
1783 					 * group till some other client
1784 					 * needs this group.
1785 					 */
1786 					MAC_RX_GRP_RELEASED(mip);
1787 				}
1788 			/* Let check if we can give this an excl group */
1789 			} else if (group == defgrp) {
1790 				ngrp = 	mac_reserve_rx_group(mcip, mac_addr,
1791 				    B_TRUE);
1792 				/* Couldn't give it a group, that's fine */
1793 				if (ngrp == NULL)
1794 					return (0);
1795 				/* Switch to H/W */
1796 				if (mac_rx_switch_group(mcip, defgrp, ngrp) !=
1797 				    0) {
1798 					mac_stop_group(ngrp);
1799 					return (0);
1800 				}
1801 			}
1802 			/*
1803 			 * If the client is in the default group, we will
1804 			 * just clear the MRP_RX_RINGS and leave it as
1805 			 * it rather than look for an exclusive group
1806 			 * for it.
1807 			 */
1808 			return (0);
1809 		}
1810 
1811 		if (group == defgrp && ((mrp->mrp_nrxrings > 0) || unspec)) {
1812 			ngrp = 	mac_reserve_rx_group(mcip, mac_addr, B_TRUE);
1813 			if (ngrp == NULL)
1814 				return (ENOSPC);
1815 
1816 			/* Switch to H/W */
1817 			if (mac_rx_switch_group(mcip, defgrp, ngrp) != 0) {
1818 				mac_release_rx_group(mcip, ngrp);
1819 				return (ENOSPC);
1820 			}
1821 			MAC_RX_GRP_RESERVED(mip);
1822 			if (mip->mi_rx_group_type == MAC_GROUP_TYPE_DYNAMIC)
1823 				MAC_RX_RING_RESERVED(mip, ngrp->mrg_cur_count);
1824 		} else if (group != defgrp && !unspec &&
1825 		    mrp->mrp_nrxrings == 0) {
1826 			/* Switch to S/W */
1827 			ringcnt = group->mrg_cur_count;
1828 			if (mac_rx_switch_group(mcip, group, defgrp) != 0)
1829 				return (ENOSPC);
1830 			if (tmrp->mrp_mask & MRP_RX_RINGS) {
1831 				MAC_RX_GRP_RELEASED(mip);
1832 				if (mip->mi_rx_group_type ==
1833 				    MAC_GROUP_TYPE_DYNAMIC) {
1834 					MAC_RX_RING_RELEASED(mip, ringcnt);
1835 				}
1836 			}
1837 		} else if (group != defgrp && mip->mi_rx_group_type ==
1838 		    MAC_GROUP_TYPE_DYNAMIC) {
1839 			ringcnt = group->mrg_cur_count;
1840 			err = mac_group_ring_modify(mcip, group, defgrp);
1841 			if (err != 0)
1842 				return (err);
1843 			/*
1844 			 * Update the accounting. If this group
1845 			 * already had explicitly reserved rings,
1846 			 * we need to update the rings based on
1847 			 * the new ring count. If this group
1848 			 * had not explicitly reserved rings,
1849 			 * then we just reserve the rings asked for
1850 			 * and reserve the group.
1851 			 */
1852 			if (tmrp->mrp_mask & MRP_RX_RINGS) {
1853 				if (ringcnt > group->mrg_cur_count) {
1854 					MAC_RX_RING_RELEASED(mip,
1855 					    ringcnt - group->mrg_cur_count);
1856 				} else {
1857 					MAC_RX_RING_RESERVED(mip,
1858 					    group->mrg_cur_count - ringcnt);
1859 				}
1860 			} else {
1861 				MAC_RX_RING_RESERVED(mip, group->mrg_cur_count);
1862 				MAC_RX_GRP_RESERVED(mip);
1863 			}
1864 		}
1865 	}
1866 	if (mrp->mrp_mask & MRP_TX_RINGS) {
1867 		unspec = mrp->mrp_mask & MRP_TXRINGS_UNSPEC;
1868 		group = flent->fe_tx_ring_group;
1869 		defgrp = MAC_DEFAULT_TX_GROUP(mip);
1870 
1871 		/*
1872 		 * For static groups we only allow rings=0 or resetting the
1873 		 * rings property.
1874 		 */
1875 		if (mrp->mrp_ntxrings > 0 &&
1876 		    mip->mi_tx_group_type != MAC_GROUP_TYPE_DYNAMIC) {
1877 			return (ENOTSUP);
1878 		}
1879 		if (mrp->mrp_mask & MRP_RINGS_RESET) {
1880 			if (!(tmrp->mrp_mask & MRP_TX_RINGS))
1881 				return (0);
1882 		} else {
1883 			if (unspec) {
1884 				if (tmrp->mrp_mask & MRP_TXRINGS_UNSPEC)
1885 					return (0);
1886 			} else if (mip->mi_tx_group_type ==
1887 			    MAC_GROUP_TYPE_DYNAMIC) {
1888 				if ((tmrp->mrp_mask & MRP_TX_RINGS) &&
1889 				    !(tmrp->mrp_mask & MRP_TXRINGS_UNSPEC) &&
1890 				    mrp->mrp_ntxrings == tmrp->mrp_ntxrings) {
1891 					return (0);
1892 				}
1893 			}
1894 		}
1895 		/* Resetting the prop */
1896 		if (mrp->mrp_mask & MRP_RINGS_RESET) {
1897 			if (group != defgrp) {
1898 				if (mip->mi_tx_group_type ==
1899 				    MAC_GROUP_TYPE_DYNAMIC) {
1900 					ringcnt = group->mrg_cur_count;
1901 					if ((flent->fe_type &
1902 					    FLOW_PRIMARY_MAC) != 0) {
1903 						mac_tx_client_quiesce(
1904 						    (mac_client_handle_t)
1905 						    mcip);
1906 						mac_tx_switch_group(mcip,
1907 						    group, defgrp);
1908 						mac_tx_client_restart(
1909 						    (mac_client_handle_t)
1910 						    mcip);
1911 						MAC_TX_GRP_RELEASED(mip);
1912 						MAC_TX_RING_RELEASED(mip,
1913 						    ringcnt);
1914 						return (0);
1915 					}
1916 					cmrp->mrp_ntxrings = 1;
1917 					(void) mac_group_ring_modify(mcip,
1918 					    group, defgrp);
1919 					/*
1920 					 * This group has reserved rings
1921 					 * that need to be released now.
1922 					 */
1923 					MAC_TX_RING_RELEASED(mip, ringcnt);
1924 				}
1925 				/*
1926 				 * If this is a static group, we
1927 				 * need to release the group. The
1928 				 * client will remain in the same
1929 				 * group till some other client
1930 				 * needs this group.
1931 				 */
1932 				MAC_TX_GRP_RELEASED(mip);
1933 			} else if (group == defgrp &&
1934 			    (flent->fe_type & FLOW_PRIMARY_MAC) == 0) {
1935 				ngrp = mac_reserve_tx_group(mcip, B_TRUE);
1936 				if (ngrp == NULL)
1937 					return (0);
1938 				mac_tx_client_quiesce(
1939 				    (mac_client_handle_t)mcip);
1940 				mac_tx_switch_group(mcip, defgrp, ngrp);
1941 				mac_tx_client_restart(
1942 				    (mac_client_handle_t)mcip);
1943 			}
1944 			/*
1945 			 * If the client is in the default group, we will
1946 			 * just clear the MRP_TX_RINGS and leave it as
1947 			 * it rather than look for an exclusive group
1948 			 * for it.
1949 			 */
1950 			return (0);
1951 		}
1952 
1953 		/* Switch to H/W */
1954 		if (group == defgrp && ((mrp->mrp_ntxrings > 0) || unspec)) {
1955 			ngrp = 	mac_reserve_tx_group(mcip, B_TRUE);
1956 			if (ngrp == NULL)
1957 				return (ENOSPC);
1958 			mac_tx_client_quiesce((mac_client_handle_t)mcip);
1959 			mac_tx_switch_group(mcip, defgrp, ngrp);
1960 			mac_tx_client_restart((mac_client_handle_t)mcip);
1961 			MAC_TX_GRP_RESERVED(mip);
1962 			if (mip->mi_tx_group_type == MAC_GROUP_TYPE_DYNAMIC)
1963 				MAC_TX_RING_RESERVED(mip, ngrp->mrg_cur_count);
1964 		/* Switch to S/W */
1965 		} else if (group != defgrp && !unspec &&
1966 		    mrp->mrp_ntxrings == 0) {
1967 			/* Switch to S/W */
1968 			ringcnt = group->mrg_cur_count;
1969 			mac_tx_client_quiesce((mac_client_handle_t)mcip);
1970 			mac_tx_switch_group(mcip, group, defgrp);
1971 			mac_tx_client_restart((mac_client_handle_t)mcip);
1972 			if (tmrp->mrp_mask & MRP_TX_RINGS) {
1973 				MAC_TX_GRP_RELEASED(mip);
1974 				if (mip->mi_tx_group_type ==
1975 				    MAC_GROUP_TYPE_DYNAMIC) {
1976 					MAC_TX_RING_RELEASED(mip, ringcnt);
1977 				}
1978 			}
1979 		} else if (group != defgrp && mip->mi_tx_group_type ==
1980 		    MAC_GROUP_TYPE_DYNAMIC) {
1981 			ringcnt = group->mrg_cur_count;
1982 			err = mac_group_ring_modify(mcip, group, defgrp);
1983 			if (err != 0)
1984 				return (err);
1985 			/*
1986 			 * Update the accounting. If this group
1987 			 * already had explicitly reserved rings,
1988 			 * we need to update the rings based on
1989 			 * the new ring count. If this group
1990 			 * had not explicitly reserved rings,
1991 			 * then we just reserve the rings asked for
1992 			 * and reserve the group.
1993 			 */
1994 			if (tmrp->mrp_mask & MRP_TX_RINGS) {
1995 				if (ringcnt > group->mrg_cur_count) {
1996 					MAC_TX_RING_RELEASED(mip,
1997 					    ringcnt - group->mrg_cur_count);
1998 				} else {
1999 					MAC_TX_RING_RESERVED(mip,
2000 					    group->mrg_cur_count - ringcnt);
2001 				}
2002 			} else {
2003 				MAC_TX_RING_RESERVED(mip, group->mrg_cur_count);
2004 				MAC_TX_GRP_RESERVED(mip);
2005 			}
2006 		}
2007 	}
2008 	return (0);
2009 }
2010 
2011 /*
2012  * When the MAC client is being brought up (i.e. we do a unicast_add) we need
2013  * to initialize the cpu and resource control structure in the
2014  * mac_client_impl_t from the mac_impl_t (i.e if there are any cached
2015  * properties before the flow entry for the unicast address was created).
2016  */
2017 static int
2018 mac_resource_ctl_set(mac_client_handle_t mch, mac_resource_props_t *mrp)
2019 {
2020 	mac_client_impl_t 	*mcip = (mac_client_impl_t *)mch;
2021 	mac_impl_t		*mip = (mac_impl_t *)mcip->mci_mip;
2022 	mac_impl_t		*umip = mcip->mci_upper_mip;
2023 	int			err = 0;
2024 	flow_entry_t		*flent = mcip->mci_flent;
2025 	mac_resource_props_t	*omrp, *nmrp = MCIP_RESOURCE_PROPS(mcip);
2026 
2027 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
2028 
2029 	err = mac_validate_props(mcip->mci_state_flags & MCIS_IS_VNIC ?
2030 	    mcip->mci_upper_mip : mip, mrp);
2031 	if (err != 0)
2032 		return (err);
2033 
2034 	/*
2035 	 * Copy over the existing properties since mac_update_resources
2036 	 * will modify the client's mrp. Currently, the saved property
2037 	 * is used to determine the difference between existing and
2038 	 * modified rings property.
2039 	 */
2040 	omrp = kmem_zalloc(sizeof (*omrp), KM_SLEEP);
2041 	bcopy(nmrp, omrp, sizeof (*omrp));
2042 	mac_update_resources(mrp, MCIP_RESOURCE_PROPS(mcip), B_FALSE);
2043 	if (MCIP_DATAPATH_SETUP(mcip)) {
2044 		/*
2045 		 * We support rings only for primary client when there are
2046 		 * multiple clients sharing the same MAC address (e.g. VLAN).
2047 		 */
2048 		if (mrp->mrp_mask & MRP_RX_RINGS ||
2049 		    mrp->mrp_mask & MRP_TX_RINGS) {
2050 
2051 			if ((err = mac_client_set_rings_prop(mcip, mrp,
2052 			    omrp)) != 0) {
2053 				if (omrp->mrp_mask & MRP_RX_RINGS) {
2054 					nmrp->mrp_mask |= MRP_RX_RINGS;
2055 					nmrp->mrp_nrxrings = omrp->mrp_nrxrings;
2056 				} else {
2057 					nmrp->mrp_mask &= ~MRP_RX_RINGS;
2058 					nmrp->mrp_nrxrings = 0;
2059 				}
2060 				if (omrp->mrp_mask & MRP_TX_RINGS) {
2061 					nmrp->mrp_mask |= MRP_TX_RINGS;
2062 					nmrp->mrp_ntxrings = omrp->mrp_ntxrings;
2063 				} else {
2064 					nmrp->mrp_mask &= ~MRP_TX_RINGS;
2065 					nmrp->mrp_ntxrings = 0;
2066 				}
2067 				if (omrp->mrp_mask & MRP_RXRINGS_UNSPEC)
2068 					omrp->mrp_mask |= MRP_RXRINGS_UNSPEC;
2069 				else
2070 					omrp->mrp_mask &= ~MRP_RXRINGS_UNSPEC;
2071 
2072 				if (omrp->mrp_mask & MRP_TXRINGS_UNSPEC)
2073 					omrp->mrp_mask |= MRP_TXRINGS_UNSPEC;
2074 				else
2075 					omrp->mrp_mask &= ~MRP_TXRINGS_UNSPEC;
2076 				kmem_free(omrp, sizeof (*omrp));
2077 				return (err);
2078 			}
2079 
2080 			/*
2081 			 * If we modified the rings property of the primary
2082 			 * we need to update the property fields of its
2083 			 * VLANs as they inherit the primary's properites.
2084 			 */
2085 			if (mac_is_primary_client(mcip)) {
2086 				mac_set_prim_vlan_rings(mip,
2087 				    MCIP_RESOURCE_PROPS(mcip));
2088 			}
2089 		}
2090 		/*
2091 		 * We have to set this prior to calling mac_flow_modify.
2092 		 */
2093 		if (mrp->mrp_mask & MRP_PRIORITY) {
2094 			if (mrp->mrp_priority == MPL_RESET) {
2095 				MAC_CLIENT_SET_PRIORITY_RANGE(mcip,
2096 				    MPL_LINK_DEFAULT);
2097 			} else {
2098 				MAC_CLIENT_SET_PRIORITY_RANGE(mcip,
2099 				    mrp->mrp_priority);
2100 			}
2101 		}
2102 
2103 		mac_flow_modify(mip->mi_flow_tab, flent, mrp);
2104 		if (mrp->mrp_mask & MRP_PRIORITY)
2105 			mac_update_subflow_priority(mcip);
2106 
2107 		/* Apply these resource settings to any secondary macs */
2108 		if (umip != NULL) {
2109 			ASSERT((umip->mi_state_flags & MIS_IS_VNIC) != 0);
2110 			mac_vnic_secondary_update(umip);
2111 		}
2112 	}
2113 	kmem_free(omrp, sizeof (*omrp));
2114 	return (0);
2115 }
2116 
2117 static int
2118 mac_unicast_flow_create(mac_client_impl_t *mcip, uint8_t *mac_addr,
2119     uint16_t vid, boolean_t is_primary, boolean_t first_flow,
2120     flow_entry_t **flent, mac_resource_props_t *mrp)
2121 {
2122 	mac_impl_t	*mip = (mac_impl_t *)mcip->mci_mip;
2123 	flow_desc_t	flow_desc;
2124 	char		flowname[MAXFLOWNAMELEN];
2125 	int		err;
2126 	uint_t		flent_flags;
2127 
2128 	/*
2129 	 * First unicast address being added, create a new flow
2130 	 * for that MAC client.
2131 	 */
2132 	bzero(&flow_desc, sizeof (flow_desc));
2133 
2134 	ASSERT(mac_addr != NULL ||
2135 	    (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR));
2136 	if (mac_addr != NULL) {
2137 		flow_desc.fd_mac_len = mip->mi_type->mt_addr_length;
2138 		bcopy(mac_addr, flow_desc.fd_dst_mac, flow_desc.fd_mac_len);
2139 	}
2140 	flow_desc.fd_mask = FLOW_LINK_DST;
2141 	if (vid != 0) {
2142 		flow_desc.fd_vid = vid;
2143 		flow_desc.fd_mask |= FLOW_LINK_VID;
2144 	}
2145 
2146 	/*
2147 	 * XXX-nicolas. For now I'm keeping the FLOW_PRIMARY_MAC
2148 	 * and FLOW_VNIC. Even though they're a hack inherited
2149 	 * from the SRS code, we'll keep them for now. They're currently
2150 	 * consumed by mac_datapath_setup() to create the SRS.
2151 	 * That code should be eventually moved out of
2152 	 * mac_datapath_setup() and moved to a mac_srs_create()
2153 	 * function of some sort to keep things clean.
2154 	 *
2155 	 * Also, there's no reason why the SRS for the primary MAC
2156 	 * client should be different than any other MAC client. Until
2157 	 * this is cleaned-up, we support only one MAC unicast address
2158 	 * per client.
2159 	 *
2160 	 * We set FLOW_PRIMARY_MAC for the primary MAC address,
2161 	 * FLOW_VNIC for everything else.
2162 	 */
2163 	if (is_primary)
2164 		flent_flags = FLOW_PRIMARY_MAC;
2165 	else
2166 		flent_flags = FLOW_VNIC_MAC;
2167 
2168 	/*
2169 	 * For the first flow we use the mac client's name - mci_name, for
2170 	 * subsequent ones we just create a name with the vid. This is
2171 	 * so that we can add these flows to the same flow table. This is
2172 	 * fine as the flow name (except for the one with the mac client's
2173 	 * name) is not visible. When the first flow is removed, we just replace
2174 	 * its fdesc with another from the list, so we will still retain the
2175 	 * flent with the MAC client's flow name.
2176 	 */
2177 	if (first_flow) {
2178 		bcopy(mcip->mci_name, flowname, MAXFLOWNAMELEN);
2179 	} else {
2180 		(void) sprintf(flowname, "%s%u", mcip->mci_name, vid);
2181 		flent_flags = FLOW_NO_STATS;
2182 	}
2183 
2184 	if ((err = mac_flow_create(&flow_desc, mrp, flowname, NULL,
2185 	    flent_flags, flent)) != 0)
2186 		return (err);
2187 
2188 	mac_misc_stat_create(*flent);
2189 	FLOW_MARK(*flent, FE_INCIPIENT);
2190 	(*flent)->fe_mcip = mcip;
2191 
2192 	/*
2193 	 * Place initial creation reference on the flow. This reference
2194 	 * is released in the corresponding delete action viz.
2195 	 * mac_unicast_remove after waiting for all transient refs to
2196 	 * to go away. The wait happens in mac_flow_wait.
2197 	 * We have already held the reference in mac_client_open().
2198 	 */
2199 	if (!first_flow)
2200 		FLOW_REFHOLD(*flent);
2201 	return (0);
2202 }
2203 
2204 /* Refresh the multicast grouping for this VID. */
2205 int
2206 mac_client_update_mcast(void *arg, boolean_t add, const uint8_t *addrp)
2207 {
2208 	flow_entry_t		*flent = arg;
2209 	mac_client_impl_t	*mcip = flent->fe_mcip;
2210 	uint16_t		vid;
2211 	flow_desc_t		flow_desc;
2212 
2213 	mac_flow_get_desc(flent, &flow_desc);
2214 	vid = (flow_desc.fd_mask & FLOW_LINK_VID) != 0 ?
2215 	    flow_desc.fd_vid : VLAN_ID_NONE;
2216 
2217 	/*
2218 	 * We don't call mac_multicast_add()/mac_multicast_remove() as
2219 	 * we want to add/remove for this specific vid.
2220 	 */
2221 	if (add) {
2222 		return (mac_bcast_add(mcip, addrp, vid,
2223 		    MAC_ADDRTYPE_MULTICAST));
2224 	} else {
2225 		mac_bcast_delete(mcip, addrp, vid);
2226 		return (0);
2227 	}
2228 }
2229 
2230 static void
2231 mac_update_single_active_client(mac_impl_t *mip)
2232 {
2233 	mac_client_impl_t *client = NULL;
2234 
2235 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
2236 
2237 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
2238 	if (mip->mi_nactiveclients == 1) {
2239 		/*
2240 		 * Find the one active MAC client from the list of MAC
2241 		 * clients. The active MAC client has at least one
2242 		 * unicast address.
2243 		 */
2244 		for (client = mip->mi_clients_list; client != NULL;
2245 		    client = client->mci_client_next) {
2246 			if (client->mci_unicast_list != NULL)
2247 				break;
2248 		}
2249 		ASSERT(client != NULL);
2250 	}
2251 
2252 	/*
2253 	 * mi_single_active_client is protected by the MAC impl's read/writer
2254 	 * lock, which allows mac_rx() to check the value of that pointer
2255 	 * as a reader.
2256 	 */
2257 	mip->mi_single_active_client = client;
2258 	rw_exit(&mip->mi_rw_lock);
2259 }
2260 
2261 /*
2262  * Set up the data path. Called from i_mac_unicast_add after having
2263  * done all the validations including making sure this is an active
2264  * client (i.e that is ready to process packets.)
2265  */
2266 static int
2267 mac_client_datapath_setup(mac_client_impl_t *mcip, uint16_t vid,
2268     uint8_t *mac_addr, mac_resource_props_t *mrp, boolean_t isprimary,
2269     mac_unicast_impl_t *muip)
2270 {
2271 	mac_impl_t	*mip = mcip->mci_mip;
2272 	boolean_t	mac_started = B_FALSE;
2273 	boolean_t	bcast_added = B_FALSE;
2274 	boolean_t	nactiveclients_added = B_FALSE;
2275 	flow_entry_t	*flent;
2276 	int		err = 0;
2277 	boolean_t	no_unicast;
2278 
2279 	no_unicast = mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR;
2280 
2281 	if ((err = mac_start((mac_handle_t)mip)) != 0)
2282 		goto bail;
2283 
2284 	mac_started = B_TRUE;
2285 
2286 	/* add the MAC client to the broadcast address group by default */
2287 	if (mip->mi_type->mt_brdcst_addr != NULL) {
2288 		err = mac_bcast_add(mcip, mip->mi_type->mt_brdcst_addr, vid,
2289 		    MAC_ADDRTYPE_BROADCAST);
2290 		if (err != 0)
2291 			goto bail;
2292 		bcast_added = B_TRUE;
2293 	}
2294 
2295 	/*
2296 	 * If this is the first unicast address addition for this
2297 	 * client, reuse the pre-allocated larval flow entry associated with
2298 	 * the MAC client.
2299 	 */
2300 	flent = (mcip->mci_nflents == 0) ? mcip->mci_flent : NULL;
2301 
2302 	/* We are configuring the unicast flow now */
2303 	if (!MCIP_DATAPATH_SETUP(mcip)) {
2304 
2305 		if (mrp != NULL) {
2306 			MAC_CLIENT_SET_PRIORITY_RANGE(mcip,
2307 			    (mrp->mrp_mask & MRP_PRIORITY) ? mrp->mrp_priority :
2308 			    MPL_LINK_DEFAULT);
2309 		}
2310 		if ((err = mac_unicast_flow_create(mcip, mac_addr, vid,
2311 		    isprimary, B_TRUE, &flent, mrp)) != 0)
2312 			goto bail;
2313 
2314 		mip->mi_nactiveclients++;
2315 		nactiveclients_added = B_TRUE;
2316 
2317 		/*
2318 		 * This will allocate the RX ring group if possible for the
2319 		 * flow and program the software classifier as needed.
2320 		 */
2321 		if ((err = mac_datapath_setup(mcip, flent, SRST_LINK)) != 0)
2322 			goto bail;
2323 
2324 		if (no_unicast)
2325 			goto done_setup;
2326 		/*
2327 		 * The unicast MAC address must have been added successfully.
2328 		 */
2329 		ASSERT(mcip->mci_unicast != NULL);
2330 		/*
2331 		 * Push down the sub-flows that were defined on this link
2332 		 * hitherto. The flows are added to the active flow table
2333 		 * and SRS, softrings etc. are created as needed.
2334 		 */
2335 		mac_link_init_flows((mac_client_handle_t)mcip);
2336 	} else {
2337 		mac_address_t *map = mcip->mci_unicast;
2338 
2339 		ASSERT(!no_unicast);
2340 		/*
2341 		 * A unicast flow already exists for that MAC client,
2342 		 * this flow must be the same mac address but with
2343 		 * different VID. It has been checked by mac_addr_in_use().
2344 		 *
2345 		 * We will use the SRS etc. from the mci_flent. Note that
2346 		 * We don't need to create kstat for this as except for
2347 		 * the fdesc, everything will be used from in the 1st flent.
2348 		 */
2349 
2350 		if (bcmp(mac_addr, map->ma_addr, map->ma_len) != 0) {
2351 			err = EINVAL;
2352 			goto bail;
2353 		}
2354 
2355 		if ((err = mac_unicast_flow_create(mcip, mac_addr, vid,
2356 		    isprimary, B_FALSE, &flent, NULL)) != 0) {
2357 			goto bail;
2358 		}
2359 		if ((err = mac_flow_add(mip->mi_flow_tab, flent)) != 0) {
2360 			FLOW_FINAL_REFRELE(flent);
2361 			goto bail;
2362 		}
2363 
2364 		/* update the multicast group for this vid */
2365 		mac_client_bcast_refresh(mcip, mac_client_update_mcast,
2366 		    (void *)flent, B_TRUE);
2367 
2368 	}
2369 
2370 	/* populate the shared MAC address */
2371 	muip->mui_map = mcip->mci_unicast;
2372 
2373 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
2374 	muip->mui_next = mcip->mci_unicast_list;
2375 	mcip->mci_unicast_list = muip;
2376 	rw_exit(&mcip->mci_rw_lock);
2377 
2378 done_setup:
2379 	/*
2380 	 * First add the flent to the flow list of this mcip. Then set
2381 	 * the mip's mi_single_active_client if needed. The Rx path assumes
2382 	 * that mip->mi_single_active_client will always have an associated
2383 	 * flent.
2384 	 */
2385 	mac_client_add_to_flow_list(mcip, flent);
2386 	if (nactiveclients_added)
2387 		mac_update_single_active_client(mip);
2388 	/*
2389 	 * Trigger a renegotiation of the capabilities when the number of
2390 	 * active clients changes from 1 to 2, since some of the capabilities
2391 	 * might have to be disabled. Also send a MAC_NOTE_LINK notification
2392 	 * to all the MAC clients whenever physical link is DOWN.
2393 	 */
2394 	if (mip->mi_nactiveclients == 2) {
2395 		mac_capab_update((mac_handle_t)mip);
2396 		mac_virtual_link_update(mip);
2397 	}
2398 	/*
2399 	 * Now that the setup is complete, clear the INCIPIENT flag.
2400 	 * The flag was set to avoid incoming packets seeing inconsistent
2401 	 * structures while the setup was in progress. Clear the mci_tx_flag
2402 	 * by calling mac_tx_client_block. It is possible that
2403 	 * mac_unicast_remove was called prior to this mac_unicast_add which
2404 	 * could have set the MCI_TX_QUIESCE flag.
2405 	 */
2406 	if (flent->fe_rx_ring_group != NULL)
2407 		mac_rx_group_unmark(flent->fe_rx_ring_group, MR_INCIPIENT);
2408 	FLOW_UNMARK(flent, FE_INCIPIENT);
2409 	FLOW_UNMARK(flent, FE_MC_NO_DATAPATH);
2410 	mac_tx_client_unblock(mcip);
2411 	return (0);
2412 bail:
2413 	if (bcast_added)
2414 		mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr, vid);
2415 
2416 	if (nactiveclients_added)
2417 		mip->mi_nactiveclients--;
2418 
2419 	if (mac_started)
2420 		mac_stop((mac_handle_t)mip);
2421 
2422 	return (err);
2423 }
2424 
2425 /*
2426  * Return the passive primary MAC client, if present. The passive client is
2427  * a stand-by client that has the same unicast address as another that is
2428  * currenly active. Once the active client goes away, the passive client
2429  * becomes active.
2430  */
2431 static mac_client_impl_t *
2432 mac_get_passive_primary_client(mac_impl_t *mip)
2433 {
2434 	mac_client_impl_t	*mcip;
2435 
2436 	for (mcip = mip->mi_clients_list; mcip != NULL;
2437 	    mcip = mcip->mci_client_next) {
2438 		if (mac_is_primary_client(mcip) &&
2439 		    (mcip->mci_flags & MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
2440 			return (mcip);
2441 		}
2442 	}
2443 	return (NULL);
2444 }
2445 
2446 /*
2447  * Add a new unicast address to the MAC client.
2448  *
2449  * The MAC address can be specified either by value, or the MAC client
2450  * can specify that it wants to use the primary MAC address of the
2451  * underlying MAC. See the introductory comments at the beginning
2452  * of this file for more more information on primary MAC addresses.
2453  *
2454  * Note also the tuple (MAC address, VID) must be unique
2455  * for the MAC clients defined on top of the same underlying MAC
2456  * instance, unless the MAC_UNICAST_NODUPCHECK is specified.
2457  *
2458  * In no case can a client use the PVID for the MAC, if the MAC has one set.
2459  */
2460 int
2461 i_mac_unicast_add(mac_client_handle_t mch, uint8_t *mac_addr, uint16_t flags,
2462     mac_unicast_handle_t *mah, uint16_t vid, mac_diag_t *diag)
2463 {
2464 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2465 	mac_impl_t		*mip = mcip->mci_mip;
2466 	int			err;
2467 	uint_t			mac_len = mip->mi_type->mt_addr_length;
2468 	boolean_t		check_dups = !(flags & MAC_UNICAST_NODUPCHECK);
2469 	boolean_t		fastpath_disabled = B_FALSE;
2470 	boolean_t		is_primary = (flags & MAC_UNICAST_PRIMARY);
2471 	boolean_t		is_unicast_hw = (flags & MAC_UNICAST_HW);
2472 	mac_resource_props_t	*mrp;
2473 	boolean_t		passive_client = B_FALSE;
2474 	mac_unicast_impl_t	*muip;
2475 	boolean_t		is_vnic_primary =
2476 	    (flags & MAC_UNICAST_VNIC_PRIMARY);
2477 
2478 	/* when VID is non-zero, the underlying MAC can not be VNIC */
2479 	ASSERT(!((mip->mi_state_flags & MIS_IS_VNIC) && (vid != 0)));
2480 
2481 	/*
2482 	 * Can't unicast add if the client asked only for minimal datapath
2483 	 * setup.
2484 	 */
2485 	if (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR)
2486 		return (ENOTSUP);
2487 
2488 	/*
2489 	 * Check for an attempted use of the current Port VLAN ID, if enabled.
2490 	 * No client may use it.
2491 	 */
2492 	if (mip->mi_pvid != 0 && vid == mip->mi_pvid)
2493 		return (EBUSY);
2494 
2495 	/*
2496 	 * Check whether it's the primary client and flag it.
2497 	 */
2498 	if (!(mcip->mci_state_flags & MCIS_IS_VNIC) && is_primary && vid == 0)
2499 		mcip->mci_flags |= MAC_CLIENT_FLAGS_PRIMARY;
2500 
2501 	/*
2502 	 * is_vnic_primary is true when we come here as a VLAN VNIC
2503 	 * which uses the primary mac client's address but with a non-zero
2504 	 * VID. In this case the MAC address is not specified by an upper
2505 	 * MAC client.
2506 	 */
2507 	if ((mcip->mci_state_flags & MCIS_IS_VNIC) && is_primary &&
2508 	    !is_vnic_primary) {
2509 		/*
2510 		 * The address is being set by the upper MAC client
2511 		 * of a VNIC. The MAC address was already set by the
2512 		 * VNIC driver during VNIC creation.
2513 		 *
2514 		 * Note: a VNIC has only one MAC address. We return
2515 		 * the MAC unicast address handle of the lower MAC client
2516 		 * corresponding to the VNIC. We allocate a new entry
2517 		 * which is flagged appropriately, so that mac_unicast_remove()
2518 		 * doesn't attempt to free the original entry that
2519 		 * was allocated by the VNIC driver.
2520 		 */
2521 		ASSERT(mcip->mci_unicast != NULL);
2522 
2523 		/* Check for VLAN flags, if present */
2524 		if ((flags & MAC_UNICAST_TAG_DISABLE) != 0)
2525 			mcip->mci_state_flags |= MCIS_TAG_DISABLE;
2526 
2527 		if ((flags & MAC_UNICAST_STRIP_DISABLE) != 0)
2528 			mcip->mci_state_flags |= MCIS_STRIP_DISABLE;
2529 
2530 		if ((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) != 0)
2531 			mcip->mci_state_flags |= MCIS_DISABLE_TX_VID_CHECK;
2532 
2533 		/*
2534 		 * Ensure that the primary unicast address of the VNIC
2535 		 * is added only once unless we have the
2536 		 * MAC_CLIENT_FLAGS_MULTI_PRIMARY set (and this is not
2537 		 * a passive MAC client).
2538 		 */
2539 		if ((mcip->mci_flags & MAC_CLIENT_FLAGS_VNIC_PRIMARY) != 0) {
2540 			if ((mcip->mci_flags &
2541 			    MAC_CLIENT_FLAGS_MULTI_PRIMARY) == 0 ||
2542 			    (mcip->mci_flags &
2543 			    MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
2544 				return (EBUSY);
2545 			}
2546 			mcip->mci_flags |= MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
2547 			passive_client = B_TRUE;
2548 		}
2549 
2550 		mcip->mci_flags |= MAC_CLIENT_FLAGS_VNIC_PRIMARY;
2551 
2552 		/*
2553 		 * Create a handle for vid 0.
2554 		 */
2555 		ASSERT(vid == 0);
2556 		muip = kmem_zalloc(sizeof (mac_unicast_impl_t), KM_SLEEP);
2557 		muip->mui_vid = vid;
2558 		*mah = (mac_unicast_handle_t)muip;
2559 		/*
2560 		 * This will be used by the caller to defer setting the
2561 		 * rx functions.
2562 		 */
2563 		if (passive_client)
2564 			return (EAGAIN);
2565 		return (0);
2566 	}
2567 
2568 	/* primary MAC clients cannot be opened on top of anchor VNICs */
2569 	if ((is_vnic_primary || is_primary) &&
2570 	    i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_ANCHOR_VNIC, NULL)) {
2571 		return (ENXIO);
2572 	}
2573 
2574 	/*
2575 	 * If this is a VNIC/VLAN, disable softmac fast-path.
2576 	 */
2577 	if (mcip->mci_state_flags & MCIS_IS_VNIC) {
2578 		err = mac_fastpath_disable((mac_handle_t)mip);
2579 		if (err != 0)
2580 			return (err);
2581 		fastpath_disabled = B_TRUE;
2582 	}
2583 
2584 	/*
2585 	 * Return EBUSY if:
2586 	 *  - there is an exclusively active mac client exists.
2587 	 *  - this is an exclusive active mac client but
2588 	 *	a. there is already active mac clients exist, or
2589 	 *	b. fastpath streams are already plumbed on this legacy device
2590 	 *  - the mac creator has disallowed active mac clients.
2591 	 */
2592 	if (mip->mi_state_flags & (MIS_EXCLUSIVE|MIS_NO_ACTIVE)) {
2593 		if (fastpath_disabled)
2594 			mac_fastpath_enable((mac_handle_t)mip);
2595 		return (EBUSY);
2596 	}
2597 
2598 	if (mcip->mci_state_flags & MCIS_EXCLUSIVE) {
2599 		ASSERT(!fastpath_disabled);
2600 		if (mip->mi_nactiveclients != 0)
2601 			return (EBUSY);
2602 
2603 		if ((mip->mi_state_flags & MIS_LEGACY) &&
2604 		    !(mip->mi_capab_legacy.ml_active_set(mip->mi_driver))) {
2605 			return (EBUSY);
2606 		}
2607 		mip->mi_state_flags |= MIS_EXCLUSIVE;
2608 	}
2609 
2610 	mrp = kmem_zalloc(sizeof (*mrp), KM_SLEEP);
2611 	if (is_primary && !(mcip->mci_state_flags & (MCIS_IS_VNIC |
2612 	    MCIS_IS_AGGR_PORT))) {
2613 		/*
2614 		 * Apply the property cached in the mac_impl_t to the primary
2615 		 * mac client. If the mac client is a VNIC or an aggregation
2616 		 * port, its property should be set in the mcip when the
2617 		 * VNIC/aggr was created.
2618 		 */
2619 		mac_get_resources((mac_handle_t)mip, mrp);
2620 		(void) mac_client_set_resources(mch, mrp);
2621 	} else if (mcip->mci_state_flags & MCIS_IS_VNIC) {
2622 		/*
2623 		 * This is a primary VLAN client, we don't support
2624 		 * specifying rings property for this as it inherits the
2625 		 * rings property from its MAC.
2626 		 */
2627 		if (is_vnic_primary) {
2628 			mac_resource_props_t	*vmrp;
2629 
2630 			vmrp = MCIP_RESOURCE_PROPS(mcip);
2631 			if (vmrp->mrp_mask & MRP_RX_RINGS ||
2632 			    vmrp->mrp_mask & MRP_TX_RINGS) {
2633 				if (fastpath_disabled)
2634 					mac_fastpath_enable((mac_handle_t)mip);
2635 				kmem_free(mrp, sizeof (*mrp));
2636 				return (ENOTSUP);
2637 			}
2638 			/*
2639 			 * Additionally we also need to inherit any
2640 			 * rings property from the MAC.
2641 			 */
2642 			mac_get_resources((mac_handle_t)mip, mrp);
2643 			if (mrp->mrp_mask & MRP_RX_RINGS) {
2644 				vmrp->mrp_mask |= MRP_RX_RINGS;
2645 				vmrp->mrp_nrxrings = mrp->mrp_nrxrings;
2646 			}
2647 			if (mrp->mrp_mask & MRP_TX_RINGS) {
2648 				vmrp->mrp_mask |= MRP_TX_RINGS;
2649 				vmrp->mrp_ntxrings = mrp->mrp_ntxrings;
2650 			}
2651 		}
2652 		bcopy(MCIP_RESOURCE_PROPS(mcip), mrp, sizeof (*mrp));
2653 	}
2654 
2655 	muip = kmem_zalloc(sizeof (mac_unicast_impl_t), KM_SLEEP);
2656 	muip->mui_vid = vid;
2657 
2658 	if (is_primary || is_vnic_primary) {
2659 		mac_addr = mip->mi_addr;
2660 	} else {
2661 
2662 		/*
2663 		 * Verify the validity of the specified MAC addresses value.
2664 		 */
2665 		if (!mac_unicst_verify((mac_handle_t)mip, mac_addr, mac_len)) {
2666 			*diag = MAC_DIAG_MACADDR_INVALID;
2667 			err = EINVAL;
2668 			goto bail_out;
2669 		}
2670 
2671 		/*
2672 		 * Make sure that the specified MAC address is different
2673 		 * than the unicast MAC address of the underlying NIC.
2674 		 */
2675 		if (check_dups && bcmp(mip->mi_addr, mac_addr, mac_len) == 0) {
2676 			*diag = MAC_DIAG_MACADDR_NIC;
2677 			err = EINVAL;
2678 			goto bail_out;
2679 		}
2680 	}
2681 
2682 	/*
2683 	 * Set the flags here so that if this is a passive client, we
2684 	 * can return  and set it when we call mac_client_datapath_setup
2685 	 * when this becomes the active client. If we defer to using these
2686 	 * flags to mac_client_datapath_setup, then for a passive client,
2687 	 * we'd have to store the flags somewhere (probably fe_flags)
2688 	 * and then use it.
2689 	 */
2690 	if (!MCIP_DATAPATH_SETUP(mcip)) {
2691 		if (is_unicast_hw) {
2692 			/*
2693 			 * The client requires a hardware MAC address slot
2694 			 * for that unicast address. Since we support only
2695 			 * one unicast MAC address per client, flag the
2696 			 * MAC client itself.
2697 			 */
2698 			mcip->mci_state_flags |= MCIS_UNICAST_HW;
2699 		}
2700 
2701 		/* Check for VLAN flags, if present */
2702 		if ((flags & MAC_UNICAST_TAG_DISABLE) != 0)
2703 			mcip->mci_state_flags |= MCIS_TAG_DISABLE;
2704 
2705 		if ((flags & MAC_UNICAST_STRIP_DISABLE) != 0)
2706 			mcip->mci_state_flags |= MCIS_STRIP_DISABLE;
2707 
2708 		if ((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) != 0)
2709 			mcip->mci_state_flags |= MCIS_DISABLE_TX_VID_CHECK;
2710 	} else {
2711 		/*
2712 		 * Assert that the specified flags are consistent with the
2713 		 * flags specified by previous calls to mac_unicast_add().
2714 		 */
2715 		ASSERT(((flags & MAC_UNICAST_TAG_DISABLE) != 0 &&
2716 		    (mcip->mci_state_flags & MCIS_TAG_DISABLE) != 0) ||
2717 		    ((flags & MAC_UNICAST_TAG_DISABLE) == 0 &&
2718 		    (mcip->mci_state_flags & MCIS_TAG_DISABLE) == 0));
2719 
2720 		ASSERT(((flags & MAC_UNICAST_STRIP_DISABLE) != 0 &&
2721 		    (mcip->mci_state_flags & MCIS_STRIP_DISABLE) != 0) ||
2722 		    ((flags & MAC_UNICAST_STRIP_DISABLE) == 0 &&
2723 		    (mcip->mci_state_flags & MCIS_STRIP_DISABLE) == 0));
2724 
2725 		ASSERT(((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) != 0 &&
2726 		    (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK) != 0) ||
2727 		    ((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) == 0 &&
2728 		    (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK) == 0));
2729 
2730 		/*
2731 		 * Make sure the client is consistent about its requests
2732 		 * for MAC addresses. I.e. all requests from the clients
2733 		 * must have the MAC_UNICAST_HW flag set or clear.
2734 		 */
2735 		if (((mcip->mci_state_flags & MCIS_UNICAST_HW) != 0 &&
2736 		    !is_unicast_hw) ||
2737 		    ((mcip->mci_state_flags & MCIS_UNICAST_HW) == 0 &&
2738 		    is_unicast_hw)) {
2739 			err = EINVAL;
2740 			goto bail_out;
2741 		}
2742 	}
2743 	/*
2744 	 * Make sure the MAC address is not already used by
2745 	 * another MAC client defined on top of the same
2746 	 * underlying NIC. Unless we have MAC_CLIENT_FLAGS_MULTI_PRIMARY
2747 	 * set when we allow a passive client to be present which will
2748 	 * be activated when the currently active client goes away - this
2749 	 * works only with primary addresses.
2750 	 */
2751 	if ((check_dups || is_primary || is_vnic_primary) &&
2752 	    mac_addr_in_use(mip, mac_addr, vid)) {
2753 		/*
2754 		 * Must have set the multiple primary address flag when
2755 		 * we did a mac_client_open AND this should be a primary
2756 		 * MAC client AND there should not already be a passive
2757 		 * primary. If all is true then we let this succeed
2758 		 * even if the address is a dup.
2759 		 */
2760 		if ((mcip->mci_flags & MAC_CLIENT_FLAGS_MULTI_PRIMARY) == 0 ||
2761 		    (mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY) == 0 ||
2762 		    mac_get_passive_primary_client(mip) != NULL) {
2763 			*diag = MAC_DIAG_MACADDR_INUSE;
2764 			err = EEXIST;
2765 			goto bail_out;
2766 		}
2767 		ASSERT((mcip->mci_flags &
2768 		    MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) == 0);
2769 		mcip->mci_flags |= MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
2770 		kmem_free(mrp, sizeof (*mrp));
2771 
2772 		/*
2773 		 * Stash the unicast address handle, we will use it when
2774 		 * we set up the passive client.
2775 		 */
2776 		mcip->mci_p_unicast_list = muip;
2777 		*mah = (mac_unicast_handle_t)muip;
2778 		return (0);
2779 	}
2780 
2781 	err = mac_client_datapath_setup(mcip, vid, mac_addr, mrp,
2782 	    is_primary || is_vnic_primary, muip);
2783 	if (err != 0)
2784 		goto bail_out;
2785 
2786 	kmem_free(mrp, sizeof (*mrp));
2787 	*mah = (mac_unicast_handle_t)muip;
2788 	return (0);
2789 
2790 bail_out:
2791 	if (fastpath_disabled)
2792 		mac_fastpath_enable((mac_handle_t)mip);
2793 	if (mcip->mci_state_flags & MCIS_EXCLUSIVE) {
2794 		mip->mi_state_flags &= ~MIS_EXCLUSIVE;
2795 		if (mip->mi_state_flags & MIS_LEGACY) {
2796 			mip->mi_capab_legacy.ml_active_clear(
2797 			    mip->mi_driver);
2798 		}
2799 	}
2800 	kmem_free(mrp, sizeof (*mrp));
2801 	kmem_free(muip, sizeof (mac_unicast_impl_t));
2802 	return (err);
2803 }
2804 
2805 /*
2806  * Wrapper function to mac_unicast_add when we want to have the same mac
2807  * client open for two instances, one that is currently active and another
2808  * that will become active when the current one is removed. In this case
2809  * mac_unicast_add will return EGAIN and we will save the rx function and
2810  * arg which will be used when we activate the passive client in
2811  * mac_unicast_remove.
2812  */
2813 int
2814 mac_unicast_add_set_rx(mac_client_handle_t mch, uint8_t *mac_addr,
2815     uint16_t flags, mac_unicast_handle_t *mah,  uint16_t vid, mac_diag_t *diag,
2816     mac_rx_t rx_fn, void *arg)
2817 {
2818 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2819 	uint_t			err;
2820 
2821 	err = mac_unicast_add(mch, mac_addr, flags, mah, vid, diag);
2822 	if (err != 0 && err != EAGAIN)
2823 		return (err);
2824 	if (err == EAGAIN) {
2825 		if (rx_fn != NULL) {
2826 			mcip->mci_rx_p_fn = rx_fn;
2827 			mcip->mci_rx_p_arg = arg;
2828 		}
2829 		return (0);
2830 	}
2831 	if (rx_fn != NULL)
2832 		mac_rx_set(mch, rx_fn, arg);
2833 	return (err);
2834 }
2835 
2836 int
2837 mac_unicast_add(mac_client_handle_t mch, uint8_t *mac_addr, uint16_t flags,
2838     mac_unicast_handle_t *mah, uint16_t vid, mac_diag_t *diag)
2839 {
2840 	mac_impl_t *mip = ((mac_client_impl_t *)mch)->mci_mip;
2841 	uint_t err;
2842 
2843 	i_mac_perim_enter(mip);
2844 	err = i_mac_unicast_add(mch, mac_addr, flags, mah, vid, diag);
2845 	i_mac_perim_exit(mip);
2846 
2847 	return (err);
2848 }
2849 
2850 static void
2851 mac_client_datapath_teardown(mac_client_handle_t mch, mac_unicast_impl_t *muip,
2852     flow_entry_t *flent)
2853 {
2854 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2855 	mac_impl_t		*mip = mcip->mci_mip;
2856 	boolean_t		no_unicast;
2857 
2858 	/*
2859 	 * If we have not added a unicast address for this MAC client, just
2860 	 * teardown the datapath.
2861 	 */
2862 	no_unicast = mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR;
2863 
2864 	if (!no_unicast) {
2865 		/*
2866 		 * We would have initialized subflows etc. only if we brought
2867 		 * up the primary client and set the unicast unicast address
2868 		 * etc. Deactivate the flows. The flow entry will be removed
2869 		 * from the active flow tables, and the associated SRS,
2870 		 * softrings etc will be deleted. But the flow entry itself
2871 		 * won't be destroyed, instead it will continue to be archived
2872 		 * off the  the global flow hash list, for a possible future
2873 		 * activation when say IP is plumbed again.
2874 		 */
2875 		mac_link_release_flows(mch);
2876 	}
2877 	mip->mi_nactiveclients--;
2878 	mac_update_single_active_client(mip);
2879 
2880 	/* Tear down the data path */
2881 	mac_datapath_teardown(mcip, mcip->mci_flent, SRST_LINK);
2882 
2883 	/*
2884 	 * Prevent any future access to the flow entry through the mci_flent
2885 	 * pointer by setting the mci_flent to NULL. Access to mci_flent in
2886 	 * mac_bcast_send is also under mi_rw_lock.
2887 	 */
2888 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
2889 	flent = mcip->mci_flent;
2890 	mac_client_remove_flow_from_list(mcip, flent);
2891 
2892 	if (mcip->mci_state_flags & MCIS_DESC_LOGGED)
2893 		mcip->mci_state_flags &= ~MCIS_DESC_LOGGED;
2894 
2895 	/*
2896 	 * This is the last unicast address being removed and there shouldn't
2897 	 * be any outbound data threads at this point coming down from mac
2898 	 * clients. We have waited for the data threads to finish before
2899 	 * starting dld_str_detach. Non-data threads must access TX SRS
2900 	 * under mi_rw_lock.
2901 	 */
2902 	rw_exit(&mip->mi_rw_lock);
2903 
2904 	/*
2905 	 * Don't use FLOW_MARK with FE_MC_NO_DATAPATH, as the flow might
2906 	 * contain other flags, such as FE_CONDEMNED, which we need to
2907 	 * cleared. We don't call mac_flow_cleanup() for this unicast
2908 	 * flow as we have a already cleaned up SRSs etc. (via the teadown
2909 	 * path). We just clear the stats and reset the initial callback
2910 	 * function, the rest will be set when we call mac_flow_create,
2911 	 * if at all.
2912 	 */
2913 	mutex_enter(&flent->fe_lock);
2914 	ASSERT(flent->fe_refcnt == 1 && flent->fe_mbg == NULL &&
2915 	    flent->fe_tx_srs == NULL && flent->fe_rx_srs_cnt == 0);
2916 	flent->fe_flags = FE_MC_NO_DATAPATH;
2917 	flow_stat_destroy(flent);
2918 	mac_misc_stat_delete(flent);
2919 
2920 	/* Initialize the receiver function to a safe routine */
2921 	flent->fe_cb_fn = (flow_fn_t)mac_pkt_drop;
2922 	flent->fe_cb_arg1 = NULL;
2923 	flent->fe_cb_arg2 = NULL;
2924 
2925 	flent->fe_index = -1;
2926 	mutex_exit(&flent->fe_lock);
2927 
2928 	if (mip->mi_type->mt_brdcst_addr != NULL) {
2929 		ASSERT(muip != NULL || no_unicast);
2930 		mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr,
2931 		    muip != NULL ? muip->mui_vid : VLAN_ID_NONE);
2932 	}
2933 
2934 	if (mip->mi_nactiveclients == 1) {
2935 		mac_capab_update((mac_handle_t)mip);
2936 		mac_virtual_link_update(mip);
2937 	}
2938 
2939 	if (mcip->mci_state_flags & MCIS_EXCLUSIVE) {
2940 		mip->mi_state_flags &= ~MIS_EXCLUSIVE;
2941 
2942 		if (mip->mi_state_flags & MIS_LEGACY)
2943 			mip->mi_capab_legacy.ml_active_clear(mip->mi_driver);
2944 	}
2945 
2946 	mcip->mci_state_flags &= ~MCIS_UNICAST_HW;
2947 
2948 	if (mcip->mci_state_flags & MCIS_TAG_DISABLE)
2949 		mcip->mci_state_flags &= ~MCIS_TAG_DISABLE;
2950 
2951 	if (mcip->mci_state_flags & MCIS_STRIP_DISABLE)
2952 		mcip->mci_state_flags &= ~MCIS_STRIP_DISABLE;
2953 
2954 	if (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK)
2955 		mcip->mci_state_flags &= ~MCIS_DISABLE_TX_VID_CHECK;
2956 
2957 	if (muip != NULL)
2958 		kmem_free(muip, sizeof (mac_unicast_impl_t));
2959 	mac_protect_cancel_timer(mcip);
2960 	mac_protect_flush_dynamic(mcip);
2961 
2962 	bzero(&mcip->mci_misc_stat, sizeof (mcip->mci_misc_stat));
2963 	/*
2964 	 * Disable fastpath if this is a VNIC or a VLAN.
2965 	 */
2966 	if (mcip->mci_state_flags & MCIS_IS_VNIC)
2967 		mac_fastpath_enable((mac_handle_t)mip);
2968 	mac_stop((mac_handle_t)mip);
2969 }
2970 
2971 /*
2972  * Remove a MAC address which was previously added by mac_unicast_add().
2973  */
2974 int
2975 mac_unicast_remove(mac_client_handle_t mch, mac_unicast_handle_t mah)
2976 {
2977 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
2978 	mac_unicast_impl_t *muip = (mac_unicast_impl_t *)mah;
2979 	mac_unicast_impl_t *pre;
2980 	mac_impl_t *mip = mcip->mci_mip;
2981 	flow_entry_t		*flent;
2982 	uint16_t mui_vid;
2983 
2984 	i_mac_perim_enter(mip);
2985 	if (mcip->mci_flags & MAC_CLIENT_FLAGS_VNIC_PRIMARY) {
2986 		/*
2987 		 * Called made by the upper MAC client of a VNIC.
2988 		 * There's nothing much to do, the unicast address will
2989 		 * be removed by the VNIC driver when the VNIC is deleted,
2990 		 * but let's ensure that all our transmit is done before
2991 		 * the client does a mac_client_stop lest it trigger an
2992 		 * assert in the driver.
2993 		 */
2994 		ASSERT(muip->mui_vid == 0);
2995 
2996 		mac_tx_client_flush(mcip);
2997 
2998 		if ((mcip->mci_flags & MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
2999 			mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
3000 			if (mcip->mci_rx_p_fn != NULL) {
3001 				mac_rx_set(mch, mcip->mci_rx_p_fn,
3002 				    mcip->mci_rx_p_arg);
3003 				mcip->mci_rx_p_fn = NULL;
3004 				mcip->mci_rx_p_arg = NULL;
3005 			}
3006 			kmem_free(muip, sizeof (mac_unicast_impl_t));
3007 			i_mac_perim_exit(mip);
3008 			return (0);
3009 		}
3010 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_VNIC_PRIMARY;
3011 
3012 		if (mcip->mci_state_flags & MCIS_TAG_DISABLE)
3013 			mcip->mci_state_flags &= ~MCIS_TAG_DISABLE;
3014 
3015 		if (mcip->mci_state_flags & MCIS_STRIP_DISABLE)
3016 			mcip->mci_state_flags &= ~MCIS_STRIP_DISABLE;
3017 
3018 		if (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK)
3019 			mcip->mci_state_flags &= ~MCIS_DISABLE_TX_VID_CHECK;
3020 
3021 		kmem_free(muip, sizeof (mac_unicast_impl_t));
3022 		i_mac_perim_exit(mip);
3023 		return (0);
3024 	}
3025 
3026 	ASSERT(muip != NULL);
3027 
3028 	/*
3029 	 * We are removing a passive client, we haven't setup the datapath
3030 	 * for this yet, so nothing much to do.
3031 	 */
3032 	if ((mcip->mci_flags & MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
3033 
3034 		ASSERT((mcip->mci_flent->fe_flags & FE_MC_NO_DATAPATH) != 0);
3035 		ASSERT(mcip->mci_p_unicast_list == muip);
3036 
3037 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
3038 
3039 		mcip->mci_p_unicast_list = NULL;
3040 		mcip->mci_rx_p_fn = NULL;
3041 		mcip->mci_rx_p_arg = NULL;
3042 
3043 		mcip->mci_state_flags &= ~MCIS_UNICAST_HW;
3044 
3045 		if (mcip->mci_state_flags & MCIS_TAG_DISABLE)
3046 			mcip->mci_state_flags &= ~MCIS_TAG_DISABLE;
3047 
3048 		if (mcip->mci_state_flags & MCIS_STRIP_DISABLE)
3049 			mcip->mci_state_flags &= ~MCIS_STRIP_DISABLE;
3050 
3051 		if (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK)
3052 			mcip->mci_state_flags &= ~MCIS_DISABLE_TX_VID_CHECK;
3053 
3054 		kmem_free(muip, sizeof (mac_unicast_impl_t));
3055 		i_mac_perim_exit(mip);
3056 		return (0);
3057 	}
3058 	/*
3059 	 * Remove the VID from the list of client's VIDs.
3060 	 */
3061 	pre = mcip->mci_unicast_list;
3062 	if (muip == pre) {
3063 		mcip->mci_unicast_list = muip->mui_next;
3064 	} else {
3065 		while ((pre->mui_next != NULL) && (pre->mui_next != muip))
3066 			pre = pre->mui_next;
3067 		ASSERT(pre->mui_next == muip);
3068 		rw_enter(&mcip->mci_rw_lock, RW_WRITER);
3069 		pre->mui_next = muip->mui_next;
3070 		rw_exit(&mcip->mci_rw_lock);
3071 	}
3072 
3073 	if (!mac_client_single_rcvr(mcip)) {
3074 		/*
3075 		 * This MAC client is shared by more than one unicast
3076 		 * addresses, so we will just remove the flent
3077 		 * corresponding to the address being removed. We don't invoke
3078 		 * mac_rx_classify_flow_rem() since the additional flow is
3079 		 * not associated with its own separate set of SRS and rings,
3080 		 * and these constructs are still needed for the remaining
3081 		 * flows.
3082 		 */
3083 		flent = mac_client_get_flow(mcip, muip);
3084 		ASSERT(flent != NULL);
3085 
3086 		/*
3087 		 * The first one is disappearing, need to make sure
3088 		 * we replace it with another from the list of
3089 		 * shared clients.
3090 		 */
3091 		if (flent == mcip->mci_flent)
3092 			flent = mac_client_swap_mciflent(mcip);
3093 		mac_client_remove_flow_from_list(mcip, flent);
3094 		mac_flow_remove(mip->mi_flow_tab, flent, B_FALSE);
3095 		mac_flow_wait(flent, FLOW_DRIVER_UPCALL);
3096 
3097 		/*
3098 		 * The multicast groups that were added by the client so
3099 		 * far must be removed from the brodcast domain corresponding
3100 		 * to the VID being removed.
3101 		 */
3102 		mac_client_bcast_refresh(mcip, mac_client_update_mcast,
3103 		    (void *)flent, B_FALSE);
3104 
3105 		if (mip->mi_type->mt_brdcst_addr != NULL) {
3106 			mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr,
3107 			    muip->mui_vid);
3108 		}
3109 
3110 		FLOW_FINAL_REFRELE(flent);
3111 		ASSERT(!(mcip->mci_state_flags & MCIS_EXCLUSIVE));
3112 		/*
3113 		 * Enable fastpath if this is a VNIC or a VLAN.
3114 		 */
3115 		if (mcip->mci_state_flags & MCIS_IS_VNIC)
3116 			mac_fastpath_enable((mac_handle_t)mip);
3117 		mac_stop((mac_handle_t)mip);
3118 		i_mac_perim_exit(mip);
3119 		return (0);
3120 	}
3121 
3122 	mui_vid = muip->mui_vid;
3123 	mac_client_datapath_teardown(mch, muip, flent);
3124 
3125 	if ((mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY) && mui_vid == 0) {
3126 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PRIMARY;
3127 	} else {
3128 		i_mac_perim_exit(mip);
3129 		return (0);
3130 	}
3131 
3132 	/*
3133 	 * If we are removing the primary, check if we have a passive primary
3134 	 * client that we need to activate now.
3135 	 */
3136 	mcip = mac_get_passive_primary_client(mip);
3137 	if (mcip != NULL) {
3138 		mac_resource_props_t	*mrp;
3139 		mac_unicast_impl_t	*muip;
3140 
3141 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
3142 		mrp = kmem_zalloc(sizeof (*mrp), KM_SLEEP);
3143 
3144 		/*
3145 		 * Apply the property cached in the mac_impl_t to the
3146 		 * primary mac client.
3147 		 */
3148 		mac_get_resources((mac_handle_t)mip, mrp);
3149 		(void) mac_client_set_resources(mch, mrp);
3150 		ASSERT(mcip->mci_p_unicast_list != NULL);
3151 		muip = mcip->mci_p_unicast_list;
3152 		mcip->mci_p_unicast_list = NULL;
3153 		if (mac_client_datapath_setup(mcip, VLAN_ID_NONE,
3154 		    mip->mi_addr, mrp, B_TRUE, muip) == 0) {
3155 			if (mcip->mci_rx_p_fn != NULL) {
3156 				mac_rx_set(mch, mcip->mci_rx_p_fn,
3157 				    mcip->mci_rx_p_arg);
3158 				mcip->mci_rx_p_fn = NULL;
3159 				mcip->mci_rx_p_arg = NULL;
3160 			}
3161 		} else {
3162 			kmem_free(muip, sizeof (mac_unicast_impl_t));
3163 		}
3164 		kmem_free(mrp, sizeof (*mrp));
3165 	}
3166 	i_mac_perim_exit(mip);
3167 	return (0);
3168 }
3169 
3170 /*
3171  * Multicast add function invoked by MAC clients.
3172  */
3173 int
3174 mac_multicast_add(mac_client_handle_t mch, const uint8_t *addr)
3175 {
3176 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3177 	mac_impl_t		*mip = mcip->mci_mip;
3178 	flow_entry_t		*flent = mcip->mci_flent_list;
3179 	flow_entry_t		*prev_fe = NULL;
3180 	uint16_t		vid;
3181 	int			err = 0;
3182 
3183 	/* Verify the address is a valid multicast address */
3184 	if ((err = mip->mi_type->mt_ops.mtops_multicst_verify(addr,
3185 	    mip->mi_pdata)) != 0)
3186 		return (err);
3187 
3188 	i_mac_perim_enter(mip);
3189 	while (flent != NULL) {
3190 		vid = i_mac_flow_vid(flent);
3191 
3192 		err = mac_bcast_add((mac_client_impl_t *)mch, addr, vid,
3193 		    MAC_ADDRTYPE_MULTICAST);
3194 		if (err != 0)
3195 			break;
3196 		prev_fe = flent;
3197 		flent = flent->fe_client_next;
3198 	}
3199 
3200 	/*
3201 	 * If we failed adding, then undo all, rather than partial
3202 	 * success.
3203 	 */
3204 	if (flent != NULL && prev_fe != NULL) {
3205 		flent = mcip->mci_flent_list;
3206 		while (flent != prev_fe->fe_client_next) {
3207 			vid = i_mac_flow_vid(flent);
3208 			mac_bcast_delete((mac_client_impl_t *)mch, addr, vid);
3209 			flent = flent->fe_client_next;
3210 		}
3211 	}
3212 	i_mac_perim_exit(mip);
3213 	return (err);
3214 }
3215 
3216 /*
3217  * Multicast delete function invoked by MAC clients.
3218  */
3219 void
3220 mac_multicast_remove(mac_client_handle_t mch, const uint8_t *addr)
3221 {
3222 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3223 	mac_impl_t		*mip = mcip->mci_mip;
3224 	flow_entry_t		*flent;
3225 	uint16_t		vid;
3226 
3227 	i_mac_perim_enter(mip);
3228 	for (flent = mcip->mci_flent_list; flent != NULL;
3229 	    flent = flent->fe_client_next) {
3230 		vid = i_mac_flow_vid(flent);
3231 		mac_bcast_delete((mac_client_impl_t *)mch, addr, vid);
3232 	}
3233 	i_mac_perim_exit(mip);
3234 }
3235 
3236 /*
3237  * When a MAC client desires to capture packets on an interface,
3238  * it registers a promiscuous call back with mac_promisc_add().
3239  * There are three types of promiscuous callbacks:
3240  *
3241  * * MAC_CLIENT_PROMISC_ALL
3242  *   Captures all packets sent and received by the MAC client,
3243  *   the physical interface, as well as all other MAC clients
3244  *   defined on top of the same MAC.
3245  *
3246  * * MAC_CLIENT_PROMISC_FILTERED
3247  *   Captures all packets sent and received by the MAC client,
3248  *   plus all multicast traffic sent and received by the phyisical
3249  *   interface and the other MAC clients.
3250  *
3251  * * MAC_CLIENT_PROMISC_MULTI
3252  *   Captures all broadcast and multicast packets sent and
3253  *   received by the MAC clients as well as the physical interface.
3254  *
3255  * In all cases, the underlying MAC is put in promiscuous mode.
3256  */
3257 int
3258 mac_promisc_add(mac_client_handle_t mch, mac_client_promisc_type_t type,
3259     mac_rx_t fn, void *arg, mac_promisc_handle_t *mphp, uint16_t flags)
3260 {
3261 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3262 	mac_impl_t *mip = mcip->mci_mip;
3263 	mac_promisc_impl_t *mpip;
3264 	mac_cb_info_t	*mcbi;
3265 	int rc;
3266 
3267 	i_mac_perim_enter(mip);
3268 
3269 	if ((rc = mac_start((mac_handle_t)mip)) != 0) {
3270 		i_mac_perim_exit(mip);
3271 		return (rc);
3272 	}
3273 
3274 	if ((mcip->mci_state_flags & MCIS_IS_VNIC) &&
3275 	    type == MAC_CLIENT_PROMISC_ALL &&
3276 	    (mcip->mci_protect_flags & MPT_FLAG_PROMISC_FILTERED)) {
3277 		/*
3278 		 * The function is being invoked by the upper MAC client
3279 		 * of a VNIC. The VNIC should only see the traffic
3280 		 * it is entitled to.
3281 		 */
3282 		type = MAC_CLIENT_PROMISC_FILTERED;
3283 	}
3284 
3285 
3286 	/*
3287 	 * Turn on promiscuous mode for the underlying NIC.
3288 	 * This is needed even for filtered callbacks which
3289 	 * expect to receive all multicast traffic on the wire.
3290 	 *
3291 	 * Physical promiscuous mode should not be turned on if
3292 	 * MAC_PROMISC_FLAGS_NO_PHYS is set.
3293 	 */
3294 	if ((flags & MAC_PROMISC_FLAGS_NO_PHYS) == 0) {
3295 		if ((rc = i_mac_promisc_set(mip, B_TRUE)) != 0) {
3296 			mac_stop((mac_handle_t)mip);
3297 			i_mac_perim_exit(mip);
3298 			return (rc);
3299 		}
3300 	}
3301 
3302 	mpip = kmem_cache_alloc(mac_promisc_impl_cache, KM_SLEEP);
3303 
3304 	mpip->mpi_type = type;
3305 	mpip->mpi_fn = fn;
3306 	mpip->mpi_arg = arg;
3307 	mpip->mpi_mcip = mcip;
3308 	mpip->mpi_no_tx_loop = ((flags & MAC_PROMISC_FLAGS_NO_TX_LOOP) != 0);
3309 	mpip->mpi_no_phys = ((flags & MAC_PROMISC_FLAGS_NO_PHYS) != 0);
3310 	mpip->mpi_strip_vlan_tag =
3311 	    ((flags & MAC_PROMISC_FLAGS_VLAN_TAG_STRIP) != 0);
3312 	mpip->mpi_no_copy = ((flags & MAC_PROMISC_FLAGS_NO_COPY) != 0);
3313 
3314 	mcbi = &mip->mi_promisc_cb_info;
3315 	mutex_enter(mcbi->mcbi_lockp);
3316 
3317 	mac_callback_add(&mip->mi_promisc_cb_info, &mcip->mci_promisc_list,
3318 	    &mpip->mpi_mci_link);
3319 	mac_callback_add(&mip->mi_promisc_cb_info, &mip->mi_promisc_list,
3320 	    &mpip->mpi_mi_link);
3321 
3322 	mutex_exit(mcbi->mcbi_lockp);
3323 
3324 	*mphp = (mac_promisc_handle_t)mpip;
3325 
3326 	if (mcip->mci_state_flags & MCIS_IS_VNIC) {
3327 		mac_impl_t *umip = mcip->mci_upper_mip;
3328 
3329 		ASSERT(umip != NULL);
3330 		mac_vnic_secondary_update(umip);
3331 	}
3332 
3333 	i_mac_perim_exit(mip);
3334 
3335 	return (0);
3336 }
3337 
3338 /*
3339  * Remove a multicast address previously aded through mac_promisc_add().
3340  */
3341 void
3342 mac_promisc_remove(mac_promisc_handle_t mph)
3343 {
3344 	mac_promisc_impl_t *mpip = (mac_promisc_impl_t *)mph;
3345 	mac_client_impl_t *mcip = mpip->mpi_mcip;
3346 	mac_impl_t *mip = mcip->mci_mip;
3347 	mac_cb_info_t *mcbi;
3348 	int rv;
3349 
3350 	i_mac_perim_enter(mip);
3351 
3352 	/*
3353 	 * Even if the device can't be reset into normal mode, we still
3354 	 * need to clear the client promisc callbacks. The client may want
3355 	 * to close the mac end point and we can't have stale callbacks.
3356 	 */
3357 	if (!(mpip->mpi_no_phys)) {
3358 		if ((rv = i_mac_promisc_set(mip, B_FALSE)) != 0) {
3359 			cmn_err(CE_WARN, "%s: failed to switch OFF promiscuous"
3360 			    " mode because of error 0x%x", mip->mi_name, rv);
3361 		}
3362 	}
3363 	mcbi = &mip->mi_promisc_cb_info;
3364 	mutex_enter(mcbi->mcbi_lockp);
3365 	if (mac_callback_remove(mcbi, &mip->mi_promisc_list,
3366 	    &mpip->mpi_mi_link)) {
3367 		VERIFY(mac_callback_remove(&mip->mi_promisc_cb_info,
3368 		    &mcip->mci_promisc_list, &mpip->mpi_mci_link));
3369 		kmem_cache_free(mac_promisc_impl_cache, mpip);
3370 	} else {
3371 		mac_callback_remove_wait(&mip->mi_promisc_cb_info);
3372 	}
3373 
3374 	if (mcip->mci_state_flags & MCIS_IS_VNIC) {
3375 		mac_impl_t *umip = mcip->mci_upper_mip;
3376 
3377 		ASSERT(umip != NULL);
3378 		mac_vnic_secondary_update(umip);
3379 	}
3380 
3381 	mutex_exit(mcbi->mcbi_lockp);
3382 	mac_stop((mac_handle_t)mip);
3383 
3384 	i_mac_perim_exit(mip);
3385 }
3386 
3387 /*
3388  * Reference count the number of active Tx threads. MCI_TX_QUIESCE indicates
3389  * that a control operation wants to quiesce the Tx data flow in which case
3390  * we return an error. Holding any of the per cpu locks ensures that the
3391  * mci_tx_flag won't change.
3392  *
3393  * 'CPU' must be accessed just once and used to compute the index into the
3394  * percpu array, and that index must be used for the entire duration of the
3395  * packet send operation. Note that the thread may be preempted and run on
3396  * another cpu any time and so we can't use 'CPU' more than once for the
3397  * operation.
3398  */
3399 #define	MAC_TX_TRY_HOLD(mcip, mytx, error)				\
3400 {									\
3401 	(error) = 0;							\
3402 	(mytx) = &(mcip)->mci_tx_pcpu[CPU->cpu_seqid & mac_tx_percpu_cnt]; \
3403 	mutex_enter(&(mytx)->pcpu_tx_lock);				\
3404 	if (!((mcip)->mci_tx_flag & MCI_TX_QUIESCE)) {			\
3405 		(mytx)->pcpu_tx_refcnt++;				\
3406 	} else {							\
3407 		(error) = -1;						\
3408 	}								\
3409 	mutex_exit(&(mytx)->pcpu_tx_lock);				\
3410 }
3411 
3412 /*
3413  * Release the reference. If needed, signal any control operation waiting
3414  * for Tx quiescence. The wait and signal are always done using the
3415  * mci_tx_pcpu[0]'s lock
3416  */
3417 #define	MAC_TX_RELE(mcip, mytx) {					\
3418 	mutex_enter(&(mytx)->pcpu_tx_lock);				\
3419 	if (--(mytx)->pcpu_tx_refcnt == 0 &&				\
3420 	    (mcip)->mci_tx_flag & MCI_TX_QUIESCE) {			\
3421 		mutex_exit(&(mytx)->pcpu_tx_lock);			\
3422 		mutex_enter(&(mcip)->mci_tx_pcpu[0].pcpu_tx_lock);	\
3423 		cv_signal(&(mcip)->mci_tx_cv);				\
3424 		mutex_exit(&(mcip)->mci_tx_pcpu[0].pcpu_tx_lock);	\
3425 	} else {							\
3426 		mutex_exit(&(mytx)->pcpu_tx_lock);			\
3427 	}								\
3428 }
3429 
3430 /*
3431  * Send function invoked by MAC clients.
3432  */
3433 mac_tx_cookie_t
3434 mac_tx(mac_client_handle_t mch, mblk_t *mp_chain, uintptr_t hint,
3435     uint16_t flag, mblk_t **ret_mp)
3436 {
3437 	mac_tx_cookie_t		cookie = NULL;
3438 	int			error;
3439 	mac_tx_percpu_t		*mytx;
3440 	mac_soft_ring_set_t	*srs;
3441 	flow_entry_t		*flent;
3442 	boolean_t		is_subflow = B_FALSE;
3443 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3444 	mac_impl_t		*mip = mcip->mci_mip;
3445 	mac_srs_tx_t		*srs_tx;
3446 
3447 	/*
3448 	 * Check whether the active Tx threads count is bumped already.
3449 	 */
3450 	if (!(flag & MAC_TX_NO_HOLD)) {
3451 		MAC_TX_TRY_HOLD(mcip, mytx, error);
3452 		if (error != 0) {
3453 			freemsgchain(mp_chain);
3454 			return (NULL);
3455 		}
3456 	}
3457 
3458 	/*
3459 	 * If mac protection is enabled, only the permissible packets will be
3460 	 * returned by mac_protect_check().
3461 	 */
3462 	if ((mcip->mci_flent->
3463 	    fe_resource_props.mrp_mask & MRP_PROTECT) != 0 &&
3464 	    (mp_chain = mac_protect_check(mch, mp_chain)) == NULL)
3465 		goto done;
3466 
3467 	if (mcip->mci_subflow_tab != NULL &&
3468 	    mcip->mci_subflow_tab->ft_flow_count > 0 &&
3469 	    mac_flow_lookup(mcip->mci_subflow_tab, mp_chain,
3470 	    FLOW_OUTBOUND, &flent) == 0) {
3471 		/*
3472 		 * The main assumption here is that if in the event
3473 		 * we get a chain, all the packets will be classified
3474 		 * to the same Flow/SRS. If this changes for any
3475 		 * reason, the following logic should change as well.
3476 		 * I suppose the fanout_hint also assumes this .
3477 		 */
3478 		ASSERT(flent != NULL);
3479 		is_subflow = B_TRUE;
3480 	} else {
3481 		flent = mcip->mci_flent;
3482 	}
3483 
3484 	srs = flent->fe_tx_srs;
3485 	/*
3486 	 * This is to avoid panics with PF_PACKET that can call mac_tx()
3487 	 * against an interface that is not capable of sending. A rewrite
3488 	 * of the mac datapath is required to remove this limitation.
3489 	 */
3490 	if (srs == NULL) {
3491 		freemsgchain(mp_chain);
3492 		goto done;
3493 	}
3494 
3495 	srs_tx = &srs->srs_tx;
3496 	if (srs_tx->st_mode == SRS_TX_DEFAULT &&
3497 	    (srs->srs_state & SRS_ENQUEUED) == 0 &&
3498 	    mip->mi_nactiveclients == 1 && mp_chain->b_next == NULL) {
3499 		uint64_t	obytes;
3500 
3501 		/*
3502 		 * Since dls always opens the underlying MAC, nclients equals
3503 		 * to 1 means that the only active client is dls itself acting
3504 		 * as a primary client of the MAC instance. Since dls will not
3505 		 * send tagged packets in that case, and dls is trusted to send
3506 		 * packets for its allowed VLAN(s), the VLAN tag insertion and
3507 		 * check is required only if nclients is greater than 1.
3508 		 */
3509 		if (mip->mi_nclients > 1) {
3510 			if (MAC_VID_CHECK_NEEDED(mcip)) {
3511 				int	err = 0;
3512 
3513 				MAC_VID_CHECK(mcip, mp_chain, err);
3514 				if (err != 0) {
3515 					freemsg(mp_chain);
3516 					mcip->mci_misc_stat.mms_txerrors++;
3517 					goto done;
3518 				}
3519 			}
3520 			if (MAC_TAG_NEEDED(mcip)) {
3521 				mp_chain = mac_add_vlan_tag(mp_chain, 0,
3522 				    mac_client_vid(mch));
3523 				if (mp_chain == NULL) {
3524 					mcip->mci_misc_stat.mms_txerrors++;
3525 					goto done;
3526 				}
3527 			}
3528 		}
3529 
3530 		obytes = (mp_chain->b_cont == NULL ? MBLKL(mp_chain) :
3531 		    msgdsize(mp_chain));
3532 
3533 		MAC_TX(mip, srs_tx->st_arg2, mp_chain, mcip);
3534 		if (mp_chain == NULL) {
3535 			cookie = NULL;
3536 			SRS_TX_STAT_UPDATE(srs, opackets, 1);
3537 			SRS_TX_STAT_UPDATE(srs, obytes, obytes);
3538 		} else {
3539 			mutex_enter(&srs->srs_lock);
3540 			cookie = mac_tx_srs_no_desc(srs, mp_chain,
3541 			    flag, ret_mp);
3542 			mutex_exit(&srs->srs_lock);
3543 		}
3544 	} else {
3545 		cookie = srs_tx->st_func(srs, mp_chain, hint, flag, ret_mp);
3546 	}
3547 
3548 done:
3549 	if (is_subflow)
3550 		FLOW_REFRELE(flent);
3551 
3552 	if (!(flag & MAC_TX_NO_HOLD))
3553 		MAC_TX_RELE(mcip, mytx);
3554 
3555 	return (cookie);
3556 }
3557 
3558 /*
3559  * mac_tx_is_blocked
3560  *
3561  * Given a cookie, it returns if the ring identified by the cookie is
3562  * flow-controlled or not. If NULL is passed in place of a cookie,
3563  * then it finds out if any of the underlying rings belonging to the
3564  * SRS is flow controlled or not and returns that status.
3565  */
3566 /* ARGSUSED */
3567 boolean_t
3568 mac_tx_is_flow_blocked(mac_client_handle_t mch, mac_tx_cookie_t cookie)
3569 {
3570 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3571 	mac_soft_ring_set_t *mac_srs;
3572 	mac_soft_ring_t *sringp;
3573 	boolean_t blocked = B_FALSE;
3574 	mac_tx_percpu_t *mytx;
3575 	int err;
3576 	int i;
3577 
3578 	/*
3579 	 * Bump the reference count so that mac_srs won't be deleted.
3580 	 * If the client is currently quiesced and we failed to bump
3581 	 * the reference, return B_TRUE so that flow control stays
3582 	 * as enabled.
3583 	 *
3584 	 * Flow control will then be disabled once the client is no
3585 	 * longer quiesced.
3586 	 */
3587 	MAC_TX_TRY_HOLD(mcip, mytx, err);
3588 	if (err != 0)
3589 		return (B_TRUE);
3590 
3591 	if ((mac_srs = MCIP_TX_SRS(mcip)) == NULL) {
3592 		MAC_TX_RELE(mcip, mytx);
3593 		return (B_FALSE);
3594 	}
3595 
3596 	mutex_enter(&mac_srs->srs_lock);
3597 	/*
3598 	 * Only in the case of TX_FANOUT and TX_AGGR, the underlying
3599 	 * softring (s_ring_state) will have the HIWAT set. This is
3600 	 * the multiple Tx ring flow control case. For all other
3601 	 * case, SRS (srs_state) will store the condition.
3602 	 */
3603 	if (mac_srs->srs_tx.st_mode == SRS_TX_FANOUT ||
3604 	    mac_srs->srs_tx.st_mode == SRS_TX_AGGR) {
3605 		if (cookie != NULL) {
3606 			sringp = (mac_soft_ring_t *)cookie;
3607 			mutex_enter(&sringp->s_ring_lock);
3608 			if (sringp->s_ring_state & S_RING_TX_HIWAT)
3609 				blocked = B_TRUE;
3610 			mutex_exit(&sringp->s_ring_lock);
3611 		} else {
3612 			for (i = 0; i < mac_srs->srs_tx_ring_count; i++) {
3613 				sringp = mac_srs->srs_tx_soft_rings[i];
3614 				mutex_enter(&sringp->s_ring_lock);
3615 				if (sringp->s_ring_state & S_RING_TX_HIWAT) {
3616 					blocked = B_TRUE;
3617 					mutex_exit(&sringp->s_ring_lock);
3618 					break;
3619 				}
3620 				mutex_exit(&sringp->s_ring_lock);
3621 			}
3622 		}
3623 	} else {
3624 		blocked = (mac_srs->srs_state & SRS_TX_HIWAT);
3625 	}
3626 	mutex_exit(&mac_srs->srs_lock);
3627 	MAC_TX_RELE(mcip, mytx);
3628 	return (blocked);
3629 }
3630 
3631 /*
3632  * Check if the MAC client is the primary MAC client.
3633  */
3634 boolean_t
3635 mac_is_primary_client(mac_client_impl_t *mcip)
3636 {
3637 	return (mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY);
3638 }
3639 
3640 void
3641 mac_ioctl(mac_handle_t mh, queue_t *wq, mblk_t *bp)
3642 {
3643 	mac_impl_t	*mip = (mac_impl_t *)mh;
3644 	int cmd = ((struct iocblk *)bp->b_rptr)->ioc_cmd;
3645 
3646 	if ((cmd == ND_GET && (mip->mi_callbacks->mc_callbacks & MC_GETPROP)) ||
3647 	    (cmd == ND_SET && (mip->mi_callbacks->mc_callbacks & MC_SETPROP))) {
3648 		/*
3649 		 * If ndd props were registered, call them.
3650 		 * Note that ndd ioctls are Obsolete
3651 		 */
3652 		mac_ndd_ioctl(mip, wq, bp);
3653 		return;
3654 	}
3655 
3656 	/*
3657 	 * Call the driver to handle the ioctl.  The driver may not support
3658 	 * any ioctls, in which case we reply with a NAK on its behalf.
3659 	 */
3660 	if (mip->mi_callbacks->mc_callbacks & MC_IOCTL)
3661 		mip->mi_ioctl(mip->mi_driver, wq, bp);
3662 	else
3663 		miocnak(wq, bp, 0, EINVAL);
3664 }
3665 
3666 /*
3667  * Return the link state of the specified MAC instance.
3668  */
3669 link_state_t
3670 mac_link_get(mac_handle_t mh)
3671 {
3672 	return (((mac_impl_t *)mh)->mi_linkstate);
3673 }
3674 
3675 /*
3676  * Add a mac client specified notification callback. Please see the comments
3677  * above mac_callback_add() for general information about mac callback
3678  * addition/deletion in the presence of mac callback list walkers
3679  */
3680 mac_notify_handle_t
3681 mac_notify_add(mac_handle_t mh, mac_notify_t notify_fn, void *arg)
3682 {
3683 	mac_impl_t		*mip = (mac_impl_t *)mh;
3684 	mac_notify_cb_t		*mncb;
3685 	mac_cb_info_t		*mcbi;
3686 
3687 	/*
3688 	 * Allocate a notify callback structure, fill in the details and
3689 	 * use the mac callback list manipulation functions to chain into
3690 	 * the list of callbacks.
3691 	 */
3692 	mncb = kmem_zalloc(sizeof (mac_notify_cb_t), KM_SLEEP);
3693 	mncb->mncb_fn = notify_fn;
3694 	mncb->mncb_arg = arg;
3695 	mncb->mncb_mip = mip;
3696 	mncb->mncb_link.mcb_objp = mncb;
3697 	mncb->mncb_link.mcb_objsize = sizeof (mac_notify_cb_t);
3698 	mncb->mncb_link.mcb_flags = MCB_NOTIFY_CB_T;
3699 
3700 	mcbi = &mip->mi_notify_cb_info;
3701 
3702 	i_mac_perim_enter(mip);
3703 	mutex_enter(mcbi->mcbi_lockp);
3704 
3705 	mac_callback_add(&mip->mi_notify_cb_info, &mip->mi_notify_cb_list,
3706 	    &mncb->mncb_link);
3707 
3708 	mutex_exit(mcbi->mcbi_lockp);
3709 	i_mac_perim_exit(mip);
3710 	return ((mac_notify_handle_t)mncb);
3711 }
3712 
3713 void
3714 mac_notify_remove_wait(mac_handle_t mh)
3715 {
3716 	mac_impl_t	*mip = (mac_impl_t *)mh;
3717 	mac_cb_info_t	*mcbi = &mip->mi_notify_cb_info;
3718 
3719 	mutex_enter(mcbi->mcbi_lockp);
3720 	mac_callback_remove_wait(&mip->mi_notify_cb_info);
3721 	mutex_exit(mcbi->mcbi_lockp);
3722 }
3723 
3724 /*
3725  * Remove a mac client specified notification callback
3726  */
3727 int
3728 mac_notify_remove(mac_notify_handle_t mnh, boolean_t wait)
3729 {
3730 	mac_notify_cb_t	*mncb = (mac_notify_cb_t *)mnh;
3731 	mac_impl_t	*mip = mncb->mncb_mip;
3732 	mac_cb_info_t	*mcbi;
3733 	int		err = 0;
3734 
3735 	mcbi = &mip->mi_notify_cb_info;
3736 
3737 	i_mac_perim_enter(mip);
3738 	mutex_enter(mcbi->mcbi_lockp);
3739 
3740 	ASSERT(mncb->mncb_link.mcb_objp == mncb);
3741 	/*
3742 	 * If there aren't any list walkers, the remove would succeed
3743 	 * inline, else we wait for the deferred remove to complete
3744 	 */
3745 	if (mac_callback_remove(&mip->mi_notify_cb_info,
3746 	    &mip->mi_notify_cb_list, &mncb->mncb_link)) {
3747 		kmem_free(mncb, sizeof (mac_notify_cb_t));
3748 	} else {
3749 		err = EBUSY;
3750 	}
3751 
3752 	mutex_exit(mcbi->mcbi_lockp);
3753 	i_mac_perim_exit(mip);
3754 
3755 	/*
3756 	 * If we failed to remove the notification callback and "wait" is set
3757 	 * to be B_TRUE, wait for the callback to finish after we exit the
3758 	 * mac perimeter.
3759 	 */
3760 	if (err != 0 && wait) {
3761 		mac_notify_remove_wait((mac_handle_t)mip);
3762 		return (0);
3763 	}
3764 
3765 	return (err);
3766 }
3767 
3768 /*
3769  * Associate resource management callbacks with the specified MAC
3770  * clients.
3771  */
3772 
3773 void
3774 mac_resource_set_common(mac_client_handle_t mch, mac_resource_add_t add,
3775     mac_resource_remove_t remove, mac_resource_quiesce_t quiesce,
3776     mac_resource_restart_t restart, mac_resource_bind_t bind,
3777     void *arg)
3778 {
3779 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3780 
3781 	mcip->mci_resource_add = add;
3782 	mcip->mci_resource_remove = remove;
3783 	mcip->mci_resource_quiesce = quiesce;
3784 	mcip->mci_resource_restart = restart;
3785 	mcip->mci_resource_bind = bind;
3786 	mcip->mci_resource_arg = arg;
3787 }
3788 
3789 void
3790 mac_resource_set(mac_client_handle_t mch, mac_resource_add_t add, void *arg)
3791 {
3792 	/* update the 'resource_add' callback */
3793 	mac_resource_set_common(mch, add, NULL, NULL, NULL, NULL, arg);
3794 }
3795 
3796 /*
3797  * Sets up the client resources and enable the polling interface over all the
3798  * SRS's and the soft rings of the client
3799  */
3800 void
3801 mac_client_poll_enable(mac_client_handle_t mch)
3802 {
3803 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3804 	mac_soft_ring_set_t	*mac_srs;
3805 	flow_entry_t		*flent;
3806 	int			i;
3807 
3808 	flent = mcip->mci_flent;
3809 	ASSERT(flent != NULL);
3810 
3811 	mcip->mci_state_flags |= MCIS_CLIENT_POLL_CAPABLE;
3812 	for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
3813 		mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
3814 		ASSERT(mac_srs->srs_mcip == mcip);
3815 		mac_srs_client_poll_enable(mcip, mac_srs);
3816 	}
3817 }
3818 
3819 /*
3820  * Tears down the client resources and disable the polling interface over all
3821  * the SRS's and the soft rings of the client
3822  */
3823 void
3824 mac_client_poll_disable(mac_client_handle_t mch)
3825 {
3826 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3827 	mac_soft_ring_set_t	*mac_srs;
3828 	flow_entry_t		*flent;
3829 	int			i;
3830 
3831 	flent = mcip->mci_flent;
3832 	ASSERT(flent != NULL);
3833 
3834 	mcip->mci_state_flags &= ~MCIS_CLIENT_POLL_CAPABLE;
3835 	for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
3836 		mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
3837 		ASSERT(mac_srs->srs_mcip == mcip);
3838 		mac_srs_client_poll_disable(mcip, mac_srs);
3839 	}
3840 }
3841 
3842 /*
3843  * Associate the CPUs specified by the given property with a MAC client.
3844  */
3845 int
3846 mac_cpu_set(mac_client_handle_t mch, mac_resource_props_t *mrp)
3847 {
3848 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3849 	mac_impl_t *mip = mcip->mci_mip;
3850 	int err = 0;
3851 
3852 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
3853 
3854 	if ((err = mac_validate_props(mcip->mci_state_flags & MCIS_IS_VNIC ?
3855 	    mcip->mci_upper_mip : mip, mrp)) != 0) {
3856 		return (err);
3857 	}
3858 	if (MCIP_DATAPATH_SETUP(mcip))
3859 		mac_flow_modify(mip->mi_flow_tab, mcip->mci_flent, mrp);
3860 
3861 	mac_update_resources(mrp, MCIP_RESOURCE_PROPS(mcip), B_FALSE);
3862 	return (0);
3863 }
3864 
3865 /*
3866  * Apply the specified properties to the specified MAC client.
3867  */
3868 int
3869 mac_client_set_resources(mac_client_handle_t mch, mac_resource_props_t *mrp)
3870 {
3871 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3872 	mac_impl_t *mip = mcip->mci_mip;
3873 	int err = 0;
3874 
3875 	i_mac_perim_enter(mip);
3876 
3877 	if ((mrp->mrp_mask & MRP_MAXBW) || (mrp->mrp_mask & MRP_PRIORITY)) {
3878 		err = mac_resource_ctl_set(mch, mrp);
3879 		if (err != 0)
3880 			goto done;
3881 	}
3882 
3883 	if (mrp->mrp_mask & (MRP_CPUS|MRP_POOL)) {
3884 		err = mac_cpu_set(mch, mrp);
3885 		if (err != 0)
3886 			goto done;
3887 	}
3888 
3889 	if (mrp->mrp_mask & MRP_PROTECT) {
3890 		err = mac_protect_set(mch, mrp);
3891 		if (err != 0)
3892 			goto done;
3893 	}
3894 
3895 	if ((mrp->mrp_mask & MRP_RX_RINGS) || (mrp->mrp_mask & MRP_TX_RINGS))
3896 		err = mac_resource_ctl_set(mch, mrp);
3897 
3898 done:
3899 	i_mac_perim_exit(mip);
3900 	return (err);
3901 }
3902 
3903 /*
3904  * Return the properties currently associated with the specified MAC client.
3905  */
3906 void
3907 mac_client_get_resources(mac_client_handle_t mch, mac_resource_props_t *mrp)
3908 {
3909 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3910 	mac_resource_props_t	*mcip_mrp = MCIP_RESOURCE_PROPS(mcip);
3911 
3912 	bcopy(mcip_mrp, mrp, sizeof (mac_resource_props_t));
3913 }
3914 
3915 /*
3916  * Return the effective properties currently associated with the specified
3917  * MAC client.
3918  */
3919 void
3920 mac_client_get_effective_resources(mac_client_handle_t mch,
3921     mac_resource_props_t *mrp)
3922 {
3923 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3924 	mac_resource_props_t	*mcip_mrp = MCIP_EFFECTIVE_PROPS(mcip);
3925 
3926 	bcopy(mcip_mrp, mrp, sizeof (mac_resource_props_t));
3927 }
3928 
3929 /*
3930  * Pass a copy of the specified packet to the promiscuous callbacks
3931  * of the specified MAC.
3932  *
3933  * If sender is NULL, the function is being invoked for a packet chain
3934  * received from the wire. If sender is non-NULL, it points to
3935  * the MAC client from which the packet is being sent.
3936  *
3937  * The packets are distributed to the promiscuous callbacks as follows:
3938  *
3939  * - all packets are sent to the MAC_CLIENT_PROMISC_ALL callbacks
3940  * - all broadcast and multicast packets are sent to the
3941  *   MAC_CLIENT_PROMISC_FILTER and MAC_CLIENT_PROMISC_MULTI.
3942  *
3943  * The unicast packets of MAC_CLIENT_PROMISC_FILTER callbacks are dispatched
3944  * after classification by mac_rx_deliver().
3945  */
3946 
3947 static void
3948 mac_promisc_dispatch_one(mac_promisc_impl_t *mpip, mblk_t *mp,
3949     boolean_t loopback)
3950 {
3951 	mblk_t *mp_copy, *mp_next;
3952 
3953 	if (!mpip->mpi_no_copy || mpip->mpi_strip_vlan_tag) {
3954 		mp_copy = copymsg(mp);
3955 		if (mp_copy == NULL)
3956 			return;
3957 
3958 		if (mpip->mpi_strip_vlan_tag) {
3959 			mp_copy = mac_strip_vlan_tag_chain(mp_copy);
3960 			if (mp_copy == NULL)
3961 				return;
3962 		}
3963 		mp_next = NULL;
3964 	} else {
3965 		mp_copy = mp;
3966 		mp_next = mp->b_next;
3967 	}
3968 	mp_copy->b_next = NULL;
3969 
3970 	mpip->mpi_fn(mpip->mpi_arg, NULL, mp_copy, loopback);
3971 	if (mp_copy == mp)
3972 		mp->b_next = mp_next;
3973 }
3974 
3975 /*
3976  * Return the VID of a packet. Zero if the packet is not tagged.
3977  */
3978 static uint16_t
3979 mac_ether_vid(mblk_t *mp)
3980 {
3981 	struct ether_header *eth = (struct ether_header *)mp->b_rptr;
3982 
3983 	if (ntohs(eth->ether_type) == ETHERTYPE_VLAN) {
3984 		struct ether_vlan_header *t_evhp =
3985 		    (struct ether_vlan_header *)mp->b_rptr;
3986 		return (VLAN_ID(ntohs(t_evhp->ether_tci)));
3987 	}
3988 
3989 	return (0);
3990 }
3991 
3992 /*
3993  * Return whether the specified packet contains a multicast or broadcast
3994  * destination MAC address.
3995  */
3996 static boolean_t
3997 mac_is_mcast(mac_impl_t *mip, mblk_t *mp)
3998 {
3999 	mac_header_info_t hdr_info;
4000 
4001 	if (mac_header_info((mac_handle_t)mip, mp, &hdr_info) != 0)
4002 		return (B_FALSE);
4003 	return ((hdr_info.mhi_dsttype == MAC_ADDRTYPE_BROADCAST) ||
4004 	    (hdr_info.mhi_dsttype == MAC_ADDRTYPE_MULTICAST));
4005 }
4006 
4007 /*
4008  * Send a copy of an mblk chain to the MAC clients of the specified MAC.
4009  * "sender" points to the sender MAC client for outbound packets, and
4010  * is set to NULL for inbound packets.
4011  */
4012 void
4013 mac_promisc_dispatch(mac_impl_t *mip, mblk_t *mp_chain,
4014     mac_client_impl_t *sender)
4015 {
4016 	mac_promisc_impl_t *mpip;
4017 	mac_cb_t *mcb;
4018 	mblk_t *mp;
4019 	boolean_t is_mcast, is_sender;
4020 
4021 	MAC_PROMISC_WALKER_INC(mip);
4022 	for (mp = mp_chain; mp != NULL; mp = mp->b_next) {
4023 		is_mcast = mac_is_mcast(mip, mp);
4024 		/* send packet to interested callbacks */
4025 		for (mcb = mip->mi_promisc_list; mcb != NULL;
4026 		    mcb = mcb->mcb_nextp) {
4027 			mpip = (mac_promisc_impl_t *)mcb->mcb_objp;
4028 			is_sender = (mpip->mpi_mcip == sender);
4029 
4030 			if (is_sender && mpip->mpi_no_tx_loop)
4031 				/*
4032 				 * The sender doesn't want to receive
4033 				 * copies of the packets it sends.
4034 				 */
4035 				continue;
4036 
4037 			/* this client doesn't need any packets (bridge) */
4038 			if (mpip->mpi_fn == NULL)
4039 				continue;
4040 
4041 			/*
4042 			 * For an ethernet MAC, don't displatch a multicast
4043 			 * packet to a non-PROMISC_ALL callbacks unless the VID
4044 			 * of the packet matches the VID of the client.
4045 			 */
4046 			if (is_mcast &&
4047 			    mpip->mpi_type != MAC_CLIENT_PROMISC_ALL &&
4048 			    !mac_client_check_flow_vid(mpip->mpi_mcip,
4049 			    mac_ether_vid(mp)))
4050 				continue;
4051 
4052 			if (is_sender ||
4053 			    mpip->mpi_type == MAC_CLIENT_PROMISC_ALL ||
4054 			    is_mcast)
4055 				mac_promisc_dispatch_one(mpip, mp, is_sender);
4056 		}
4057 	}
4058 	MAC_PROMISC_WALKER_DCR(mip);
4059 }
4060 
4061 void
4062 mac_promisc_client_dispatch(mac_client_impl_t *mcip, mblk_t *mp_chain)
4063 {
4064 	mac_impl_t		*mip = mcip->mci_mip;
4065 	mac_promisc_impl_t	*mpip;
4066 	boolean_t		is_mcast;
4067 	mblk_t			*mp;
4068 	mac_cb_t		*mcb;
4069 
4070 	/*
4071 	 * The unicast packets for the MAC client still
4072 	 * need to be delivered to the MAC_CLIENT_PROMISC_FILTERED
4073 	 * promiscuous callbacks. The broadcast and multicast
4074 	 * packets were delivered from mac_rx().
4075 	 */
4076 	MAC_PROMISC_WALKER_INC(mip);
4077 	for (mp = mp_chain; mp != NULL; mp = mp->b_next) {
4078 		is_mcast = mac_is_mcast(mip, mp);
4079 		for (mcb = mcip->mci_promisc_list; mcb != NULL;
4080 		    mcb = mcb->mcb_nextp) {
4081 			mpip = (mac_promisc_impl_t *)mcb->mcb_objp;
4082 			if (mpip->mpi_type == MAC_CLIENT_PROMISC_FILTERED &&
4083 			    !is_mcast) {
4084 				mac_promisc_dispatch_one(mpip, mp, B_FALSE);
4085 			}
4086 		}
4087 	}
4088 	MAC_PROMISC_WALKER_DCR(mip);
4089 }
4090 
4091 /*
4092  * Return the margin value currently assigned to the specified MAC instance.
4093  */
4094 void
4095 mac_margin_get(mac_handle_t mh, uint32_t *marginp)
4096 {
4097 	mac_impl_t *mip = (mac_impl_t *)mh;
4098 
4099 	rw_enter(&(mip->mi_rw_lock), RW_READER);
4100 	*marginp = mip->mi_margin;
4101 	rw_exit(&(mip->mi_rw_lock));
4102 }
4103 
4104 /*
4105  * mac_info_get() is used for retrieving the mac_info when a DL_INFO_REQ is
4106  * issued before a DL_ATTACH_REQ. we walk the i_mac_impl_hash table and find
4107  * the first mac_impl_t with a matching driver name; then we copy its mac_info_t
4108  * to the caller. we do all this with i_mac_impl_lock held so the mac_impl_t
4109  * cannot disappear while we are accessing it.
4110  */
4111 typedef struct i_mac_info_state_s {
4112 	const char	*mi_name;
4113 	mac_info_t	*mi_infop;
4114 } i_mac_info_state_t;
4115 
4116 /*ARGSUSED*/
4117 static uint_t
4118 i_mac_info_walker(mod_hash_key_t key, mod_hash_val_t *val, void *arg)
4119 {
4120 	i_mac_info_state_t *statep = arg;
4121 	mac_impl_t *mip = (mac_impl_t *)val;
4122 
4123 	if (mip->mi_state_flags & MIS_DISABLED)
4124 		return (MH_WALK_CONTINUE);
4125 
4126 	if (strcmp(statep->mi_name,
4127 	    ddi_driver_name(mip->mi_dip)) != 0)
4128 		return (MH_WALK_CONTINUE);
4129 
4130 	statep->mi_infop = &mip->mi_info;
4131 	return (MH_WALK_TERMINATE);
4132 }
4133 
4134 boolean_t
4135 mac_info_get(const char *name, mac_info_t *minfop)
4136 {
4137 	i_mac_info_state_t state;
4138 
4139 	rw_enter(&i_mac_impl_lock, RW_READER);
4140 	state.mi_name = name;
4141 	state.mi_infop = NULL;
4142 	mod_hash_walk(i_mac_impl_hash, i_mac_info_walker, &state);
4143 	if (state.mi_infop == NULL) {
4144 		rw_exit(&i_mac_impl_lock);
4145 		return (B_FALSE);
4146 	}
4147 	*minfop = *state.mi_infop;
4148 	rw_exit(&i_mac_impl_lock);
4149 	return (B_TRUE);
4150 }
4151 
4152 /*
4153  * To get the capabilities that MAC layer cares about, such as rings, factory
4154  * mac address, vnic or not, it should directly invoke this function.  If the
4155  * link is part of a bridge, then the only "capability" it has is the inability
4156  * to do zero copy.
4157  */
4158 boolean_t
4159 i_mac_capab_get(mac_handle_t mh, mac_capab_t cap, void *cap_data)
4160 {
4161 	mac_impl_t *mip = (mac_impl_t *)mh;
4162 
4163 	if (mip->mi_bridge_link != NULL)
4164 		return (cap == MAC_CAPAB_NO_ZCOPY);
4165 	else if (mip->mi_callbacks->mc_callbacks & MC_GETCAPAB)
4166 		return (mip->mi_getcapab(mip->mi_driver, cap, cap_data));
4167 	else
4168 		return (B_FALSE);
4169 }
4170 
4171 /*
4172  * Capability query function. If number of active mac clients is greater than
4173  * 1, only limited capabilities can be advertised to the caller no matter the
4174  * driver has certain capability or not. Else, we query the driver to get the
4175  * capability.
4176  */
4177 boolean_t
4178 mac_capab_get(mac_handle_t mh, mac_capab_t cap, void *cap_data)
4179 {
4180 	mac_impl_t *mip = (mac_impl_t *)mh;
4181 
4182 	/*
4183 	 * if mi_nactiveclients > 1, only MAC_CAPAB_LEGACY, MAC_CAPAB_HCKSUM,
4184 	 * MAC_CAPAB_NO_NATIVEVLAN and MAC_CAPAB_NO_ZCOPY can be advertised.
4185 	 */
4186 	if (mip->mi_nactiveclients > 1) {
4187 		switch (cap) {
4188 		case MAC_CAPAB_NO_ZCOPY:
4189 			return (B_TRUE);
4190 		case MAC_CAPAB_LEGACY:
4191 		case MAC_CAPAB_HCKSUM:
4192 		case MAC_CAPAB_NO_NATIVEVLAN:
4193 			break;
4194 		default:
4195 			return (B_FALSE);
4196 		}
4197 	}
4198 
4199 	/* else get capab from driver */
4200 	return (i_mac_capab_get(mh, cap, cap_data));
4201 }
4202 
4203 boolean_t
4204 mac_sap_verify(mac_handle_t mh, uint32_t sap, uint32_t *bind_sap)
4205 {
4206 	mac_impl_t *mip = (mac_impl_t *)mh;
4207 
4208 	return (mip->mi_type->mt_ops.mtops_sap_verify(sap, bind_sap,
4209 	    mip->mi_pdata));
4210 }
4211 
4212 mblk_t *
4213 mac_header(mac_handle_t mh, const uint8_t *daddr, uint32_t sap, mblk_t *payload,
4214     size_t extra_len)
4215 {
4216 	mac_impl_t	*mip = (mac_impl_t *)mh;
4217 	const uint8_t	*hdr_daddr;
4218 
4219 	/*
4220 	 * If the MAC is point-to-point with a fixed destination address, then
4221 	 * we must always use that destination in the MAC header.
4222 	 */
4223 	hdr_daddr = (mip->mi_dstaddr_set ? mip->mi_dstaddr : daddr);
4224 	return (mip->mi_type->mt_ops.mtops_header(mip->mi_addr, hdr_daddr, sap,
4225 	    mip->mi_pdata, payload, extra_len));
4226 }
4227 
4228 int
4229 mac_header_info(mac_handle_t mh, mblk_t *mp, mac_header_info_t *mhip)
4230 {
4231 	mac_impl_t *mip = (mac_impl_t *)mh;
4232 
4233 	return (mip->mi_type->mt_ops.mtops_header_info(mp, mip->mi_pdata,
4234 	    mhip));
4235 }
4236 
4237 int
4238 mac_vlan_header_info(mac_handle_t mh, mblk_t *mp, mac_header_info_t *mhip)
4239 {
4240 	mac_impl_t	*mip = (mac_impl_t *)mh;
4241 	boolean_t	is_ethernet = (mip->mi_info.mi_media == DL_ETHER);
4242 	int		err = 0;
4243 
4244 	/*
4245 	 * Packets should always be at least 16 bit aligned.
4246 	 */
4247 	ASSERT(IS_P2ALIGNED(mp->b_rptr, sizeof (uint16_t)));
4248 
4249 	if ((err = mac_header_info(mh, mp, mhip)) != 0)
4250 		return (err);
4251 
4252 	/*
4253 	 * If this is a VLAN-tagged Ethernet packet, then the SAP in the
4254 	 * mac_header_info_t as returned by mac_header_info() is
4255 	 * ETHERTYPE_VLAN. We need to grab the ethertype from the VLAN header.
4256 	 */
4257 	if (is_ethernet && (mhip->mhi_bindsap == ETHERTYPE_VLAN)) {
4258 		struct ether_vlan_header *evhp;
4259 		uint16_t sap;
4260 		mblk_t *tmp = NULL;
4261 		size_t size;
4262 
4263 		size = sizeof (struct ether_vlan_header);
4264 		if (MBLKL(mp) < size) {
4265 			/*
4266 			 * Pullup the message in order to get the MAC header
4267 			 * infomation. Note that this is a read-only function,
4268 			 * we keep the input packet intact.
4269 			 */
4270 			if ((tmp = msgpullup(mp, size)) == NULL)
4271 				return (EINVAL);
4272 
4273 			mp = tmp;
4274 		}
4275 		evhp = (struct ether_vlan_header *)mp->b_rptr;
4276 		sap = ntohs(evhp->ether_type);
4277 		(void) mac_sap_verify(mh, sap, &mhip->mhi_bindsap);
4278 		mhip->mhi_hdrsize = sizeof (struct ether_vlan_header);
4279 		mhip->mhi_tci = ntohs(evhp->ether_tci);
4280 		mhip->mhi_istagged = B_TRUE;
4281 		freemsg(tmp);
4282 
4283 		if (VLAN_CFI(mhip->mhi_tci) != ETHER_CFI)
4284 			return (EINVAL);
4285 	} else {
4286 		mhip->mhi_istagged = B_FALSE;
4287 		mhip->mhi_tci = 0;
4288 	}
4289 
4290 	return (0);
4291 }
4292 
4293 mblk_t *
4294 mac_header_cook(mac_handle_t mh, mblk_t *mp)
4295 {
4296 	mac_impl_t *mip = (mac_impl_t *)mh;
4297 
4298 	if (mip->mi_type->mt_ops.mtops_ops & MTOPS_HEADER_COOK) {
4299 		if (DB_REF(mp) > 1) {
4300 			mblk_t *newmp = copymsg(mp);
4301 			if (newmp == NULL)
4302 				return (NULL);
4303 			freemsg(mp);
4304 			mp = newmp;
4305 		}
4306 		return (mip->mi_type->mt_ops.mtops_header_cook(mp,
4307 		    mip->mi_pdata));
4308 	}
4309 	return (mp);
4310 }
4311 
4312 mblk_t *
4313 mac_header_uncook(mac_handle_t mh, mblk_t *mp)
4314 {
4315 	mac_impl_t *mip = (mac_impl_t *)mh;
4316 
4317 	if (mip->mi_type->mt_ops.mtops_ops & MTOPS_HEADER_UNCOOK) {
4318 		if (DB_REF(mp) > 1) {
4319 			mblk_t *newmp = copymsg(mp);
4320 			if (newmp == NULL)
4321 				return (NULL);
4322 			freemsg(mp);
4323 			mp = newmp;
4324 		}
4325 		return (mip->mi_type->mt_ops.mtops_header_uncook(mp,
4326 		    mip->mi_pdata));
4327 	}
4328 	return (mp);
4329 }
4330 
4331 uint_t
4332 mac_addr_len(mac_handle_t mh)
4333 {
4334 	mac_impl_t *mip = (mac_impl_t *)mh;
4335 
4336 	return (mip->mi_type->mt_addr_length);
4337 }
4338 
4339 /* True if a MAC is a VNIC */
4340 boolean_t
4341 mac_is_vnic(mac_handle_t mh)
4342 {
4343 	return (((mac_impl_t *)mh)->mi_state_flags & MIS_IS_VNIC);
4344 }
4345 
4346 mac_handle_t
4347 mac_get_lower_mac_handle(mac_handle_t mh)
4348 {
4349 	mac_impl_t *mip = (mac_impl_t *)mh;
4350 
4351 	ASSERT(mac_is_vnic(mh));
4352 	return (((vnic_t *)mip->mi_driver)->vn_lower_mh);
4353 }
4354 
4355 boolean_t
4356 mac_is_vnic_primary(mac_handle_t mh)
4357 {
4358 	mac_impl_t *mip = (mac_impl_t *)mh;
4359 
4360 	ASSERT(mac_is_vnic(mh));
4361 	return (((vnic_t *)mip->mi_driver)->vn_addr_type ==
4362 	    VNIC_MAC_ADDR_TYPE_PRIMARY);
4363 }
4364 
4365 void
4366 mac_update_resources(mac_resource_props_t *nmrp, mac_resource_props_t *cmrp,
4367     boolean_t is_user_flow)
4368 {
4369 	if (nmrp != NULL && cmrp != NULL) {
4370 		if (nmrp->mrp_mask & MRP_PRIORITY) {
4371 			if (nmrp->mrp_priority == MPL_RESET) {
4372 				cmrp->mrp_mask &= ~MRP_PRIORITY;
4373 				if (is_user_flow) {
4374 					cmrp->mrp_priority =
4375 					    MPL_SUBFLOW_DEFAULT;
4376 				} else {
4377 					cmrp->mrp_priority = MPL_LINK_DEFAULT;
4378 				}
4379 			} else {
4380 				cmrp->mrp_mask |= MRP_PRIORITY;
4381 				cmrp->mrp_priority = nmrp->mrp_priority;
4382 			}
4383 		}
4384 		if (nmrp->mrp_mask & MRP_MAXBW) {
4385 			if (nmrp->mrp_maxbw == MRP_MAXBW_RESETVAL) {
4386 				cmrp->mrp_mask &= ~MRP_MAXBW;
4387 				cmrp->mrp_maxbw = 0;
4388 			} else {
4389 				cmrp->mrp_mask |= MRP_MAXBW;
4390 				cmrp->mrp_maxbw = nmrp->mrp_maxbw;
4391 			}
4392 		}
4393 		if (nmrp->mrp_mask & MRP_CPUS)
4394 			MAC_COPY_CPUS(nmrp, cmrp);
4395 
4396 		if (nmrp->mrp_mask & MRP_POOL) {
4397 			if (strlen(nmrp->mrp_pool) == 0) {
4398 				cmrp->mrp_mask &= ~MRP_POOL;
4399 				bzero(cmrp->mrp_pool, sizeof (cmrp->mrp_pool));
4400 			} else {
4401 				cmrp->mrp_mask |= MRP_POOL;
4402 				(void) strncpy(cmrp->mrp_pool, nmrp->mrp_pool,
4403 				    sizeof (cmrp->mrp_pool));
4404 			}
4405 
4406 		}
4407 
4408 		if (nmrp->mrp_mask & MRP_PROTECT)
4409 			mac_protect_update(nmrp, cmrp);
4410 
4411 		/*
4412 		 * Update the rings specified.
4413 		 */
4414 		if (nmrp->mrp_mask & MRP_RX_RINGS) {
4415 			if (nmrp->mrp_mask & MRP_RINGS_RESET) {
4416 				cmrp->mrp_mask &= ~MRP_RX_RINGS;
4417 				if (cmrp->mrp_mask & MRP_RXRINGS_UNSPEC)
4418 					cmrp->mrp_mask &= ~MRP_RXRINGS_UNSPEC;
4419 				cmrp->mrp_nrxrings = 0;
4420 			} else {
4421 				cmrp->mrp_mask |= MRP_RX_RINGS;
4422 				cmrp->mrp_nrxrings = nmrp->mrp_nrxrings;
4423 			}
4424 		}
4425 		if (nmrp->mrp_mask & MRP_TX_RINGS) {
4426 			if (nmrp->mrp_mask & MRP_RINGS_RESET) {
4427 				cmrp->mrp_mask &= ~MRP_TX_RINGS;
4428 				if (cmrp->mrp_mask & MRP_TXRINGS_UNSPEC)
4429 					cmrp->mrp_mask &= ~MRP_TXRINGS_UNSPEC;
4430 				cmrp->mrp_ntxrings = 0;
4431 			} else {
4432 				cmrp->mrp_mask |= MRP_TX_RINGS;
4433 				cmrp->mrp_ntxrings = nmrp->mrp_ntxrings;
4434 			}
4435 		}
4436 		if (nmrp->mrp_mask & MRP_RXRINGS_UNSPEC)
4437 			cmrp->mrp_mask |= MRP_RXRINGS_UNSPEC;
4438 		else if (cmrp->mrp_mask & MRP_RXRINGS_UNSPEC)
4439 			cmrp->mrp_mask &= ~MRP_RXRINGS_UNSPEC;
4440 
4441 		if (nmrp->mrp_mask & MRP_TXRINGS_UNSPEC)
4442 			cmrp->mrp_mask |= MRP_TXRINGS_UNSPEC;
4443 		else if (cmrp->mrp_mask & MRP_TXRINGS_UNSPEC)
4444 			cmrp->mrp_mask &= ~MRP_TXRINGS_UNSPEC;
4445 	}
4446 }
4447 
4448 /*
4449  * i_mac_set_resources:
4450  *
4451  * This routine associates properties with the primary MAC client of
4452  * the specified MAC instance.
4453  * - Cache the properties in mac_impl_t
4454  * - Apply the properties to the primary MAC client if exists
4455  */
4456 int
4457 i_mac_set_resources(mac_handle_t mh, mac_resource_props_t *mrp)
4458 {
4459 	mac_impl_t		*mip = (mac_impl_t *)mh;
4460 	mac_client_impl_t	*mcip;
4461 	int			err = 0;
4462 	uint32_t		resmask, newresmask;
4463 	mac_resource_props_t	*tmrp, *umrp;
4464 
4465 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
4466 
4467 	err = mac_validate_props(mip, mrp);
4468 	if (err != 0)
4469 		return (err);
4470 
4471 	umrp = kmem_zalloc(sizeof (*umrp), KM_SLEEP);
4472 	bcopy(&mip->mi_resource_props, umrp, sizeof (*umrp));
4473 	resmask = umrp->mrp_mask;
4474 	mac_update_resources(mrp, umrp, B_FALSE);
4475 	newresmask = umrp->mrp_mask;
4476 
4477 	if (resmask == 0 && newresmask != 0) {
4478 		/*
4479 		 * Bandwidth, priority, cpu or pool link properties configured,
4480 		 * must disable fastpath.
4481 		 */
4482 		if ((err = mac_fastpath_disable((mac_handle_t)mip)) != 0) {
4483 			kmem_free(umrp, sizeof (*umrp));
4484 			return (err);
4485 		}
4486 	}
4487 
4488 	/*
4489 	 * Since bind_cpu may be modified by mac_client_set_resources()
4490 	 * we use a copy of bind_cpu and finally cache bind_cpu in mip.
4491 	 * This allows us to cache only user edits in mip.
4492 	 */
4493 	tmrp = kmem_zalloc(sizeof (*tmrp), KM_SLEEP);
4494 	bcopy(mrp, tmrp, sizeof (*tmrp));
4495 	mcip = mac_primary_client_handle(mip);
4496 	if (mcip != NULL && (mcip->mci_state_flags & MCIS_IS_AGGR_PORT) == 0) {
4497 		err = mac_client_set_resources((mac_client_handle_t)mcip, tmrp);
4498 	} else if ((mrp->mrp_mask & MRP_RX_RINGS ||
4499 	    mrp->mrp_mask & MRP_TX_RINGS)) {
4500 		mac_client_impl_t	*vmcip;
4501 
4502 		/*
4503 		 * If the primary is not up, we need to check if there
4504 		 * are any VLANs on this primary. If there are then
4505 		 * we need to set this property on the VLANs since
4506 		 * VLANs follow the primary they are based on. Just
4507 		 * look for the first VLAN and change its properties,
4508 		 * all the other VLANs should be in the same group.
4509 		 */
4510 		for (vmcip = mip->mi_clients_list; vmcip != NULL;
4511 		    vmcip = vmcip->mci_client_next) {
4512 			if ((vmcip->mci_flent->fe_type & FLOW_PRIMARY_MAC) &&
4513 			    mac_client_vid((mac_client_handle_t)vmcip) !=
4514 			    VLAN_ID_NONE) {
4515 				break;
4516 			}
4517 		}
4518 		if (vmcip != NULL) {
4519 			mac_resource_props_t	*omrp;
4520 			mac_resource_props_t	*vmrp;
4521 
4522 			omrp = kmem_zalloc(sizeof (*omrp), KM_SLEEP);
4523 			bcopy(MCIP_RESOURCE_PROPS(vmcip), omrp, sizeof (*omrp));
4524 			/*
4525 			 * We dont' call mac_update_resources since we
4526 			 * want to take only the ring properties and
4527 			 * not all the properties that may have changed.
4528 			 */
4529 			vmrp = MCIP_RESOURCE_PROPS(vmcip);
4530 			if (mrp->mrp_mask & MRP_RX_RINGS) {
4531 				if (mrp->mrp_mask & MRP_RINGS_RESET) {
4532 					vmrp->mrp_mask &= ~MRP_RX_RINGS;
4533 					if (vmrp->mrp_mask &
4534 					    MRP_RXRINGS_UNSPEC) {
4535 						vmrp->mrp_mask &=
4536 						    ~MRP_RXRINGS_UNSPEC;
4537 					}
4538 					vmrp->mrp_nrxrings = 0;
4539 				} else {
4540 					vmrp->mrp_mask |= MRP_RX_RINGS;
4541 					vmrp->mrp_nrxrings = mrp->mrp_nrxrings;
4542 				}
4543 			}
4544 			if (mrp->mrp_mask & MRP_TX_RINGS) {
4545 				if (mrp->mrp_mask & MRP_RINGS_RESET) {
4546 					vmrp->mrp_mask &= ~MRP_TX_RINGS;
4547 					if (vmrp->mrp_mask &
4548 					    MRP_TXRINGS_UNSPEC) {
4549 						vmrp->mrp_mask &=
4550 						    ~MRP_TXRINGS_UNSPEC;
4551 					}
4552 					vmrp->mrp_ntxrings = 0;
4553 				} else {
4554 					vmrp->mrp_mask |= MRP_TX_RINGS;
4555 					vmrp->mrp_ntxrings = mrp->mrp_ntxrings;
4556 				}
4557 			}
4558 			if (mrp->mrp_mask & MRP_RXRINGS_UNSPEC)
4559 				vmrp->mrp_mask |= MRP_RXRINGS_UNSPEC;
4560 
4561 			if (mrp->mrp_mask & MRP_TXRINGS_UNSPEC)
4562 				vmrp->mrp_mask |= MRP_TXRINGS_UNSPEC;
4563 
4564 			if ((err = mac_client_set_rings_prop(vmcip, mrp,
4565 			    omrp)) != 0) {
4566 				bcopy(omrp, MCIP_RESOURCE_PROPS(vmcip),
4567 				    sizeof (*omrp));
4568 			} else {
4569 				mac_set_prim_vlan_rings(mip, vmrp);
4570 			}
4571 			kmem_free(omrp, sizeof (*omrp));
4572 		}
4573 	}
4574 
4575 	/* Only update the values if mac_client_set_resources succeeded */
4576 	if (err == 0) {
4577 		bcopy(umrp, &mip->mi_resource_props, sizeof (*umrp));
4578 		/*
4579 		 * If bandwidth, priority or cpu link properties cleared,
4580 		 * renable fastpath.
4581 		 */
4582 		if (resmask != 0 && newresmask == 0)
4583 			mac_fastpath_enable((mac_handle_t)mip);
4584 	} else if (resmask == 0 && newresmask != 0) {
4585 		mac_fastpath_enable((mac_handle_t)mip);
4586 	}
4587 	kmem_free(tmrp, sizeof (*tmrp));
4588 	kmem_free(umrp, sizeof (*umrp));
4589 	return (err);
4590 }
4591 
4592 int
4593 mac_set_resources(mac_handle_t mh, mac_resource_props_t *mrp)
4594 {
4595 	int err;
4596 
4597 	i_mac_perim_enter((mac_impl_t *)mh);
4598 	err = i_mac_set_resources(mh, mrp);
4599 	i_mac_perim_exit((mac_impl_t *)mh);
4600 	return (err);
4601 }
4602 
4603 /*
4604  * Get the properties cached for the specified MAC instance.
4605  */
4606 void
4607 mac_get_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_resources((mac_client_handle_t)mcip, mrp);
4615 		return;
4616 	}
4617 	bcopy(&mip->mi_resource_props, mrp, sizeof (mac_resource_props_t));
4618 }
4619 
4620 /*
4621  * Get the effective properties from the primary client of the
4622  * specified MAC instance.
4623  */
4624 void
4625 mac_get_effective_resources(mac_handle_t mh, mac_resource_props_t *mrp)
4626 {
4627 	mac_impl_t 		*mip = (mac_impl_t *)mh;
4628 	mac_client_impl_t	*mcip;
4629 
4630 	mcip = mac_primary_client_handle(mip);
4631 	if (mcip != NULL) {
4632 		mac_client_get_effective_resources((mac_client_handle_t)mcip,
4633 		    mrp);
4634 		return;
4635 	}
4636 	bzero(mrp, sizeof (mac_resource_props_t));
4637 }
4638 
4639 int
4640 mac_set_pvid(mac_handle_t mh, uint16_t pvid)
4641 {
4642 	mac_impl_t *mip = (mac_impl_t *)mh;
4643 	mac_client_impl_t *mcip;
4644 	mac_unicast_impl_t *muip;
4645 
4646 	i_mac_perim_enter(mip);
4647 	if (pvid != 0) {
4648 		for (mcip = mip->mi_clients_list; mcip != NULL;
4649 		    mcip = mcip->mci_client_next) {
4650 			for (muip = mcip->mci_unicast_list; muip != NULL;
4651 			    muip = muip->mui_next) {
4652 				if (muip->mui_vid == pvid) {
4653 					i_mac_perim_exit(mip);
4654 					return (EBUSY);
4655 				}
4656 			}
4657 		}
4658 	}
4659 	mip->mi_pvid = pvid;
4660 	i_mac_perim_exit(mip);
4661 	return (0);
4662 }
4663 
4664 uint16_t
4665 mac_get_pvid(mac_handle_t mh)
4666 {
4667 	mac_impl_t *mip = (mac_impl_t *)mh;
4668 
4669 	return (mip->mi_pvid);
4670 }
4671 
4672 uint32_t
4673 mac_get_llimit(mac_handle_t mh)
4674 {
4675 	mac_impl_t *mip = (mac_impl_t *)mh;
4676 
4677 	return (mip->mi_llimit);
4678 }
4679 
4680 uint32_t
4681 mac_get_ldecay(mac_handle_t mh)
4682 {
4683 	mac_impl_t *mip = (mac_impl_t *)mh;
4684 
4685 	return (mip->mi_ldecay);
4686 }
4687 
4688 /*
4689  * Rename a mac client, its flow, and the kstat.
4690  */
4691 int
4692 mac_rename_primary(mac_handle_t mh, const char *new_name)
4693 {
4694 	mac_impl_t		*mip = (mac_impl_t *)mh;
4695 	mac_client_impl_t	*cur_clnt = NULL;
4696 	flow_entry_t		*fep;
4697 
4698 	i_mac_perim_enter(mip);
4699 
4700 	/*
4701 	 * VNICs: we need to change the sys flow name and
4702 	 * the associated flow kstat.
4703 	 */
4704 	if (mip->mi_state_flags & MIS_IS_VNIC) {
4705 		mac_client_impl_t *mcip = mac_vnic_lower(mip);
4706 		ASSERT(new_name != NULL);
4707 		mac_rename_flow_names(mcip, new_name);
4708 		mac_stat_rename(mcip);
4709 		goto done;
4710 	}
4711 	/*
4712 	 * This mac may itself be an aggr link, or it may have some client
4713 	 * which is an aggr port. For both cases, we need to change the
4714 	 * aggr port's mac client name, its flow name and the associated flow
4715 	 * kstat.
4716 	 */
4717 	if (mip->mi_state_flags & MIS_IS_AGGR) {
4718 		mac_capab_aggr_t aggr_cap;
4719 		mac_rename_fn_t rename_fn;
4720 		boolean_t ret;
4721 
4722 		ASSERT(new_name != NULL);
4723 		ret = i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_AGGR,
4724 		    (void *)(&aggr_cap));
4725 		ASSERT(ret == B_TRUE);
4726 		rename_fn = aggr_cap.mca_rename_fn;
4727 		rename_fn(new_name, mip->mi_driver);
4728 		/*
4729 		 * The aggr's client name and kstat flow name will be
4730 		 * updated below, i.e. via mac_rename_flow_names.
4731 		 */
4732 	}
4733 
4734 	for (cur_clnt = mip->mi_clients_list; cur_clnt != NULL;
4735 	    cur_clnt = cur_clnt->mci_client_next) {
4736 		if (cur_clnt->mci_state_flags & MCIS_IS_AGGR_PORT) {
4737 			if (new_name != NULL) {
4738 				char *str_st = cur_clnt->mci_name;
4739 				char *str_del = strchr(str_st, '-');
4740 
4741 				ASSERT(str_del != NULL);
4742 				bzero(str_del + 1, MAXNAMELEN -
4743 				    (str_del - str_st + 1));
4744 				bcopy(new_name, str_del + 1,
4745 				    strlen(new_name));
4746 			}
4747 			fep = cur_clnt->mci_flent;
4748 			mac_rename_flow(fep, cur_clnt->mci_name);
4749 			break;
4750 		} else if (new_name != NULL &&
4751 		    cur_clnt->mci_state_flags & MCIS_USE_DATALINK_NAME) {
4752 			mac_rename_flow_names(cur_clnt, new_name);
4753 			break;
4754 		}
4755 	}
4756 
4757 	/* Recreate kstats associated with aggr pseudo rings */
4758 	if (mip->mi_state_flags & MIS_IS_AGGR)
4759 		mac_pseudo_ring_stat_rename(mip);
4760 
4761 done:
4762 	i_mac_perim_exit(mip);
4763 	return (0);
4764 }
4765 
4766 /*
4767  * Rename the MAC client's flow names
4768  */
4769 static void
4770 mac_rename_flow_names(mac_client_impl_t *mcip, const char *new_name)
4771 {
4772 	flow_entry_t	*flent;
4773 	uint16_t	vid;
4774 	char		flowname[MAXFLOWNAMELEN];
4775 	mac_impl_t	*mip = mcip->mci_mip;
4776 
4777 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
4778 
4779 	/*
4780 	 * Use mi_rw_lock to ensure that threads not in the mac perimeter
4781 	 * see a self-consistent value for mci_name
4782 	 */
4783 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
4784 	(void) strlcpy(mcip->mci_name, new_name, sizeof (mcip->mci_name));
4785 	rw_exit(&mip->mi_rw_lock);
4786 
4787 	mac_rename_flow(mcip->mci_flent, new_name);
4788 
4789 	if (mcip->mci_nflents == 1)
4790 		return;
4791 
4792 	/*
4793 	 * We have to rename all the others too, no stats to destroy for
4794 	 * these.
4795 	 */
4796 	for (flent = mcip->mci_flent_list; flent != NULL;
4797 	    flent = flent->fe_client_next) {
4798 		if (flent != mcip->mci_flent) {
4799 			vid = i_mac_flow_vid(flent);
4800 			(void) sprintf(flowname, "%s%u", new_name, vid);
4801 			mac_flow_set_name(flent, flowname);
4802 		}
4803 	}
4804 }
4805 
4806 
4807 /*
4808  * Add a flow to the MAC client's flow list - i.e list of MAC/VID tuples
4809  * defined for the specified MAC client.
4810  */
4811 static void
4812 mac_client_add_to_flow_list(mac_client_impl_t *mcip, flow_entry_t *flent)
4813 {
4814 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
4815 	/*
4816 	 * The promisc Rx data path walks the mci_flent_list. Protect by
4817 	 * using mi_rw_lock
4818 	 */
4819 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
4820 
4821 	mcip->mci_vidcache = MCIP_VIDCACHE_INVALID;
4822 
4823 	/* Add it to the head */
4824 	flent->fe_client_next = mcip->mci_flent_list;
4825 	mcip->mci_flent_list = flent;
4826 	mcip->mci_nflents++;
4827 
4828 	/*
4829 	 * Keep track of the number of non-zero VIDs addresses per MAC
4830 	 * client to avoid figuring it out in the data-path.
4831 	 */
4832 	if (i_mac_flow_vid(flent) != VLAN_ID_NONE)
4833 		mcip->mci_nvids++;
4834 
4835 	rw_exit(&mcip->mci_rw_lock);
4836 }
4837 
4838 /*
4839  * Remove a flow entry from the MAC client's list.
4840  */
4841 static void
4842 mac_client_remove_flow_from_list(mac_client_impl_t *mcip, flow_entry_t *flent)
4843 {
4844 	flow_entry_t	*fe = mcip->mci_flent_list;
4845 	flow_entry_t	*prev_fe = NULL;
4846 
4847 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
4848 	/*
4849 	 * The promisc Rx data path walks the mci_flent_list. Protect by
4850 	 * using mci_rw_lock
4851 	 */
4852 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
4853 	mcip->mci_vidcache = MCIP_VIDCACHE_INVALID;
4854 
4855 	while ((fe != NULL) && (fe != flent)) {
4856 		prev_fe = fe;
4857 		fe = fe->fe_client_next;
4858 	}
4859 
4860 	ASSERT(fe != NULL);
4861 	if (prev_fe == NULL) {
4862 		/* Deleting the first node */
4863 		mcip->mci_flent_list = fe->fe_client_next;
4864 	} else {
4865 		prev_fe->fe_client_next = fe->fe_client_next;
4866 	}
4867 	mcip->mci_nflents--;
4868 
4869 	if (i_mac_flow_vid(flent) != VLAN_ID_NONE)
4870 		mcip->mci_nvids--;
4871 
4872 	rw_exit(&mcip->mci_rw_lock);
4873 }
4874 
4875 /*
4876  * Check if the given VID belongs to this MAC client.
4877  */
4878 boolean_t
4879 mac_client_check_flow_vid(mac_client_impl_t *mcip, uint16_t vid)
4880 {
4881 	flow_entry_t	*flent;
4882 	uint16_t	mci_vid;
4883 	uint32_t	cache = mcip->mci_vidcache;
4884 
4885 	/*
4886 	 * In hopes of not having to touch the mci_rw_lock, check to see if
4887 	 * this vid matches our cached result.
4888 	 */
4889 	if (MCIP_VIDCACHE_ISVALID(cache) && MCIP_VIDCACHE_VID(cache) == vid)
4890 		return (MCIP_VIDCACHE_BOOL(cache) ? B_TRUE : B_FALSE);
4891 
4892 	/* The mci_flent_list is protected by mci_rw_lock */
4893 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
4894 	for (flent = mcip->mci_flent_list; flent != NULL;
4895 	    flent = flent->fe_client_next) {
4896 		mci_vid = i_mac_flow_vid(flent);
4897 		if (vid == mci_vid) {
4898 			mcip->mci_vidcache = MCIP_VIDCACHE_CACHE(vid, B_TRUE);
4899 			rw_exit(&mcip->mci_rw_lock);
4900 			return (B_TRUE);
4901 		}
4902 	}
4903 
4904 	mcip->mci_vidcache = MCIP_VIDCACHE_CACHE(vid, B_FALSE);
4905 	rw_exit(&mcip->mci_rw_lock);
4906 	return (B_FALSE);
4907 }
4908 
4909 /*
4910  * Get the flow entry for the specified <MAC addr, VID> tuple.
4911  */
4912 static flow_entry_t *
4913 mac_client_get_flow(mac_client_impl_t *mcip, mac_unicast_impl_t *muip)
4914 {
4915 	mac_address_t *map = mcip->mci_unicast;
4916 	flow_entry_t *flent;
4917 	uint16_t vid;
4918 	flow_desc_t flow_desc;
4919 
4920 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
4921 
4922 	mac_flow_get_desc(mcip->mci_flent, &flow_desc);
4923 	if (bcmp(flow_desc.fd_dst_mac, map->ma_addr, map->ma_len) != 0)
4924 		return (NULL);
4925 
4926 	for (flent = mcip->mci_flent_list; flent != NULL;
4927 	    flent = flent->fe_client_next) {
4928 		vid = i_mac_flow_vid(flent);
4929 		if (vid == muip->mui_vid) {
4930 			return (flent);
4931 		}
4932 	}
4933 
4934 	return (NULL);
4935 }
4936 
4937 /*
4938  * Since mci_flent has the SRSs, when we want to remove it, we replace
4939  * the flow_desc_t in mci_flent with that of an existing flent and then
4940  * remove that flent instead of mci_flent.
4941  */
4942 static flow_entry_t *
4943 mac_client_swap_mciflent(mac_client_impl_t *mcip)
4944 {
4945 	flow_entry_t	*flent = mcip->mci_flent;
4946 	flow_tab_t	*ft = flent->fe_flow_tab;
4947 	flow_entry_t	*flent1;
4948 	flow_desc_t	fl_desc;
4949 	char		fl_name[MAXFLOWNAMELEN];
4950 	int		err;
4951 
4952 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
4953 	ASSERT(mcip->mci_nflents > 1);
4954 
4955 	/* get the next flent following the primary flent  */
4956 	flent1 = mcip->mci_flent_list->fe_client_next;
4957 	ASSERT(flent1 != NULL && flent1->fe_flow_tab == ft);
4958 
4959 	/*
4960 	 * Remove the flent from the flow table before updating the
4961 	 * flow descriptor as the hash depends on the flow descriptor.
4962 	 * This also helps incoming packet classification avoid having
4963 	 * to grab fe_lock. Access to fe_flow_desc of a flent not in the
4964 	 * flow table is done under the fe_lock so that log or stat functions
4965 	 * see a self-consistent fe_flow_desc. The name and desc are specific
4966 	 * to a flow, the rest are shared by all the clients, including
4967 	 * resource control etc.
4968 	 */
4969 	mac_flow_remove(ft, flent, B_TRUE);
4970 	mac_flow_remove(ft, flent1, B_TRUE);
4971 
4972 	bcopy(&flent->fe_flow_desc, &fl_desc, sizeof (flow_desc_t));
4973 	bcopy(flent->fe_flow_name, fl_name, MAXFLOWNAMELEN);
4974 
4975 	/* update the primary flow entry */
4976 	mutex_enter(&flent->fe_lock);
4977 	bcopy(&flent1->fe_flow_desc, &flent->fe_flow_desc,
4978 	    sizeof (flow_desc_t));
4979 	bcopy(&flent1->fe_flow_name, &flent->fe_flow_name, MAXFLOWNAMELEN);
4980 	mutex_exit(&flent->fe_lock);
4981 
4982 	/* update the flow entry that is to be freed */
4983 	mutex_enter(&flent1->fe_lock);
4984 	bcopy(&fl_desc, &flent1->fe_flow_desc, sizeof (flow_desc_t));
4985 	bcopy(fl_name, &flent1->fe_flow_name, MAXFLOWNAMELEN);
4986 	mutex_exit(&flent1->fe_lock);
4987 
4988 	/* now reinsert the flow entries in the table */
4989 	err = mac_flow_add(ft, flent);
4990 	ASSERT(err == 0);
4991 
4992 	err = mac_flow_add(ft, flent1);
4993 	ASSERT(err == 0);
4994 
4995 	return (flent1);
4996 }
4997 
4998 /*
4999  * Return whether there is only one flow entry associated with this
5000  * MAC client.
5001  */
5002 static boolean_t
5003 mac_client_single_rcvr(mac_client_impl_t *mcip)
5004 {
5005 	return (mcip->mci_nflents == 1);
5006 }
5007 
5008 int
5009 mac_validate_props(mac_impl_t *mip, mac_resource_props_t *mrp)
5010 {
5011 	boolean_t		reset;
5012 	uint32_t		rings_needed;
5013 	uint32_t		rings_avail;
5014 	mac_group_type_t	gtype;
5015 	mac_resource_props_t	*mip_mrp;
5016 
5017 	if (mrp == NULL)
5018 		return (0);
5019 
5020 	if (mrp->mrp_mask & MRP_PRIORITY) {
5021 		mac_priority_level_t	pri = mrp->mrp_priority;
5022 
5023 		if (pri < MPL_LOW || pri > MPL_RESET)
5024 			return (EINVAL);
5025 	}
5026 
5027 	if (mrp->mrp_mask & MRP_MAXBW) {
5028 		uint64_t maxbw = mrp->mrp_maxbw;
5029 
5030 		if (maxbw < MRP_MAXBW_MINVAL && maxbw != 0)
5031 			return (EINVAL);
5032 	}
5033 	if (mrp->mrp_mask & MRP_CPUS) {
5034 		int i, j;
5035 		mac_cpu_mode_t	fanout;
5036 
5037 		if (mrp->mrp_ncpus > ncpus)
5038 			return (EINVAL);
5039 
5040 		for (i = 0; i < mrp->mrp_ncpus; i++) {
5041 			for (j = 0; j < mrp->mrp_ncpus; j++) {
5042 				if (i != j &&
5043 				    mrp->mrp_cpu[i] == mrp->mrp_cpu[j]) {
5044 					return (EINVAL);
5045 				}
5046 			}
5047 		}
5048 
5049 		for (i = 0; i < mrp->mrp_ncpus; i++) {
5050 			cpu_t *cp;
5051 			int rv;
5052 
5053 			mutex_enter(&cpu_lock);
5054 			cp = cpu_get(mrp->mrp_cpu[i]);
5055 			if (cp != NULL)
5056 				rv = cpu_is_online(cp);
5057 			else
5058 				rv = 0;
5059 			mutex_exit(&cpu_lock);
5060 			if (rv == 0)
5061 				return (EINVAL);
5062 		}
5063 
5064 		fanout = mrp->mrp_fanout_mode;
5065 		if (fanout < 0 || fanout > MCM_CPUS)
5066 			return (EINVAL);
5067 	}
5068 
5069 	if (mrp->mrp_mask & MRP_PROTECT) {
5070 		int err = mac_protect_validate(mrp);
5071 		if (err != 0)
5072 			return (err);
5073 	}
5074 
5075 	if (!(mrp->mrp_mask & MRP_RX_RINGS) &&
5076 	    !(mrp->mrp_mask & MRP_TX_RINGS)) {
5077 		return (0);
5078 	}
5079 
5080 	/*
5081 	 * mip will be null when we come from mac_flow_create or
5082 	 * mac_link_flow_modify. In the latter case it is a user flow,
5083 	 * for which we don't support rings. In the former we would
5084 	 * have validated the props beforehand (i_mac_unicast_add ->
5085 	 * mac_client_set_resources -> validate for the primary and
5086 	 * vnic_dev_create -> mac_client_set_resources -> validate for
5087 	 * a vnic.
5088 	 */
5089 	if (mip == NULL)
5090 		return (0);
5091 
5092 	/*
5093 	 * We don't support setting rings property for a VNIC that is using a
5094 	 * primary address (VLAN)
5095 	 */
5096 	if ((mip->mi_state_flags & MIS_IS_VNIC) &&
5097 	    mac_is_vnic_primary((mac_handle_t)mip)) {
5098 		return (ENOTSUP);
5099 	}
5100 
5101 	mip_mrp = &mip->mi_resource_props;
5102 	/*
5103 	 * The rings property should be validated against the NICs
5104 	 * resources
5105 	 */
5106 	if (mip->mi_state_flags & MIS_IS_VNIC)
5107 		mip = (mac_impl_t *)mac_get_lower_mac_handle((mac_handle_t)mip);
5108 
5109 	reset = mrp->mrp_mask & MRP_RINGS_RESET;
5110 	/*
5111 	 * If groups are not supported, return error.
5112 	 */
5113 	if (((mrp->mrp_mask & MRP_RX_RINGS) && mip->mi_rx_groups == NULL) ||
5114 	    ((mrp->mrp_mask & MRP_TX_RINGS) && mip->mi_tx_groups == NULL)) {
5115 		return (EINVAL);
5116 	}
5117 	/*
5118 	 * If we are just resetting, there is no validation needed.
5119 	 */
5120 	if (reset)
5121 		return (0);
5122 
5123 	if (mrp->mrp_mask & MRP_RX_RINGS) {
5124 		rings_needed = mrp->mrp_nrxrings;
5125 		/*
5126 		 * We just want to check if the number of additional
5127 		 * rings requested is available.
5128 		 */
5129 		if (mip_mrp->mrp_mask & MRP_RX_RINGS) {
5130 			if (mrp->mrp_nrxrings > mip_mrp->mrp_nrxrings)
5131 				/* Just check for the additional rings */
5132 				rings_needed -= mip_mrp->mrp_nrxrings;
5133 			else
5134 				/* We are not asking for additional rings */
5135 				rings_needed = 0;
5136 		}
5137 		rings_avail = mip->mi_rxrings_avail;
5138 		gtype = mip->mi_rx_group_type;
5139 	} else {
5140 		rings_needed = mrp->mrp_ntxrings;
5141 		/* Similarly for the TX rings */
5142 		if (mip_mrp->mrp_mask & MRP_TX_RINGS) {
5143 			if (mrp->mrp_ntxrings > mip_mrp->mrp_ntxrings)
5144 				/* Just check for the additional rings */
5145 				rings_needed -= mip_mrp->mrp_ntxrings;
5146 			else
5147 				/* We are not asking for additional rings */
5148 				rings_needed = 0;
5149 		}
5150 		rings_avail = mip->mi_txrings_avail;
5151 		gtype = mip->mi_tx_group_type;
5152 	}
5153 
5154 	/* Error if the group is dynamic .. */
5155 	if (gtype == MAC_GROUP_TYPE_DYNAMIC) {
5156 		/*
5157 		 * .. and rings specified are more than available.
5158 		 */
5159 		if (rings_needed > rings_avail)
5160 			return (EINVAL);
5161 	} else {
5162 		/*
5163 		 * OR group is static and we have specified some rings.
5164 		 */
5165 		if (rings_needed > 0)
5166 			return (EINVAL);
5167 	}
5168 	return (0);
5169 }
5170 
5171 /*
5172  * Send a MAC_NOTE_LINK notification to all the MAC clients whenever the
5173  * underlying physical link is down. This is to allow MAC clients to
5174  * communicate with other clients.
5175  */
5176 void
5177 mac_virtual_link_update(mac_impl_t *mip)
5178 {
5179 	if (mip->mi_linkstate != LINK_STATE_UP)
5180 		i_mac_notify(mip, MAC_NOTE_LINK);
5181 }
5182 
5183 /*
5184  * For clients that have a pass-thru MAC, e.g. VNIC, we set the VNIC's
5185  * mac handle in the client.
5186  */
5187 void
5188 mac_set_upper_mac(mac_client_handle_t mch, mac_handle_t mh,
5189     mac_resource_props_t *mrp)
5190 {
5191 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
5192 	mac_impl_t		*mip = (mac_impl_t *)mh;
5193 
5194 	mcip->mci_upper_mip = mip;
5195 	/* If there are any properties, copy it over too */
5196 	if (mrp != NULL) {
5197 		bcopy(mrp, &mip->mi_resource_props,
5198 		    sizeof (mac_resource_props_t));
5199 	}
5200 }
5201 
5202 /*
5203  * Mark the mac as being used exclusively by the single mac client that is
5204  * doing some control operation on this mac. No further opens of this mac
5205  * will be allowed until this client calls mac_unmark_exclusive. The mac
5206  * client calling this function must already be in the mac perimeter
5207  */
5208 int
5209 mac_mark_exclusive(mac_handle_t mh)
5210 {
5211 	mac_impl_t	*mip = (mac_impl_t *)mh;
5212 
5213 	ASSERT(MAC_PERIM_HELD(mh));
5214 	/*
5215 	 * Look up its entry in the global hash table.
5216 	 */
5217 	rw_enter(&i_mac_impl_lock, RW_WRITER);
5218 	if (mip->mi_state_flags & MIS_DISABLED) {
5219 		rw_exit(&i_mac_impl_lock);
5220 		return (ENOENT);
5221 	}
5222 
5223 	/*
5224 	 * A reference to mac is held even if the link is not plumbed.
5225 	 * In i_dls_link_create() we open the MAC interface and hold the
5226 	 * reference. There is an additional reference for the mac_open
5227 	 * done in acquiring the mac perimeter
5228 	 */
5229 	if (mip->mi_ref != 2) {
5230 		rw_exit(&i_mac_impl_lock);
5231 		return (EBUSY);
5232 	}
5233 
5234 	ASSERT(!(mip->mi_state_flags & MIS_EXCLUSIVE_HELD));
5235 	mip->mi_state_flags |= MIS_EXCLUSIVE_HELD;
5236 	rw_exit(&i_mac_impl_lock);
5237 	return (0);
5238 }
5239 
5240 void
5241 mac_unmark_exclusive(mac_handle_t mh)
5242 {
5243 	mac_impl_t	*mip = (mac_impl_t *)mh;
5244 
5245 	ASSERT(MAC_PERIM_HELD(mh));
5246 
5247 	rw_enter(&i_mac_impl_lock, RW_WRITER);
5248 	/* 1 for the creation and another for the perimeter */
5249 	ASSERT(mip->mi_ref == 2 && (mip->mi_state_flags & MIS_EXCLUSIVE_HELD));
5250 	mip->mi_state_flags &= ~MIS_EXCLUSIVE_HELD;
5251 	rw_exit(&i_mac_impl_lock);
5252 }
5253 
5254 /*
5255  * Set the MTU for the specified MAC.
5256  */
5257 int
5258 mac_set_mtu(mac_handle_t mh, uint_t new_mtu, uint_t *old_mtu_arg)
5259 {
5260 	mac_impl_t *mip = (mac_impl_t *)mh;
5261 	uint_t old_mtu;
5262 	int rv = 0;
5263 
5264 	i_mac_perim_enter(mip);
5265 
5266 	if (!(mip->mi_callbacks->mc_callbacks & (MC_SETPROP|MC_GETPROP))) {
5267 		rv = ENOTSUP;
5268 		goto bail;
5269 	}
5270 
5271 	old_mtu = mip->mi_sdu_max;
5272 
5273 	if (new_mtu == 0 || new_mtu < mip->mi_sdu_min) {
5274 		rv = EINVAL;
5275 		goto bail;
5276 	}
5277 
5278 	rw_enter(&mip->mi_rw_lock, RW_READER);
5279 	if (mip->mi_mtrp != NULL && new_mtu < mip->mi_mtrp->mtr_mtu) {
5280 		rv = EBUSY;
5281 		rw_exit(&mip->mi_rw_lock);
5282 		goto bail;
5283 	}
5284 	rw_exit(&mip->mi_rw_lock);
5285 
5286 	if (old_mtu != new_mtu) {
5287 		rv = mip->mi_callbacks->mc_setprop(mip->mi_driver,
5288 		    "mtu", MAC_PROP_MTU, sizeof (uint_t), &new_mtu);
5289 		if (rv != 0)
5290 			goto bail;
5291 		rv = mac_maxsdu_update(mh, new_mtu);
5292 		ASSERT(rv == 0);
5293 	}
5294 
5295 bail:
5296 	i_mac_perim_exit(mip);
5297 
5298 	if (rv == 0 && old_mtu_arg != NULL)
5299 		*old_mtu_arg = old_mtu;
5300 	return (rv);
5301 }
5302 
5303 /*
5304  * Return the RX h/w information for the group indexed by grp_num.
5305  */
5306 void
5307 mac_get_hwrxgrp_info(mac_handle_t mh, int grp_index, uint_t *grp_num,
5308     uint_t *n_rings, uint_t *rings, uint_t *type, uint_t *n_clnts,
5309     char *clnts_name)
5310 {
5311 	mac_impl_t *mip = (mac_impl_t *)mh;
5312 	mac_grp_client_t *mcip;
5313 	uint_t i = 0, index = 0;
5314 	mac_ring_t	*ring;
5315 
5316 	/* Revisit when we implement fully dynamic group allocation */
5317 	ASSERT(grp_index >= 0 && grp_index < mip->mi_rx_group_count);
5318 
5319 	rw_enter(&mip->mi_rw_lock, RW_READER);
5320 	*grp_num = mip->mi_rx_groups[grp_index].mrg_index;
5321 	*type = mip->mi_rx_groups[grp_index].mrg_type;
5322 	*n_rings = mip->mi_rx_groups[grp_index].mrg_cur_count;
5323 	ring = mip->mi_rx_groups[grp_index].mrg_rings;
5324 	for (index = 0; index < mip->mi_rx_groups[grp_index].mrg_cur_count;
5325 	    index++) {
5326 		rings[index] = ring->mr_index;
5327 		ring = ring->mr_next;
5328 	}
5329 	/* Assuming the 1st is the default group */
5330 	index = 0;
5331 	if (grp_index == 0) {
5332 		(void) strlcpy(clnts_name, "<default,mcast>,",
5333 		    MAXCLIENTNAMELEN);
5334 		index += strlen("<default,mcast>,");
5335 	}
5336 	for (mcip = mip->mi_rx_groups[grp_index].mrg_clients; mcip != NULL;
5337 	    mcip = mcip->mgc_next) {
5338 		int name_len = strlen(mcip->mgc_client->mci_name);
5339 
5340 		/*
5341 		 * MAXCLIENTNAMELEN is the buffer size reserved for client
5342 		 * names.
5343 		 * XXXX Formating the client name string needs to be moved
5344 		 * to user land when fixing the size of dhi_clnts in
5345 		 * dld_hwgrpinfo_t. We should use n_clients * client_name for
5346 		 * dhi_clntsin instead of MAXCLIENTNAMELEN
5347 		 */
5348 		if (index + name_len >= MAXCLIENTNAMELEN) {
5349 			index = MAXCLIENTNAMELEN;
5350 			break;
5351 		}
5352 		bcopy(mcip->mgc_client->mci_name, &(clnts_name[index]),
5353 		    name_len);
5354 		index += name_len;
5355 		clnts_name[index++] = ',';
5356 		i++;
5357 	}
5358 
5359 	/* Get rid of the last , */
5360 	if (index > 0)
5361 		clnts_name[index - 1] = '\0';
5362 	*n_clnts = i;
5363 	rw_exit(&mip->mi_rw_lock);
5364 }
5365 
5366 /*
5367  * Return the TX h/w information for the group indexed by grp_num.
5368  */
5369 void
5370 mac_get_hwtxgrp_info(mac_handle_t mh, int grp_index, uint_t *grp_num,
5371     uint_t *n_rings, uint_t *rings, uint_t *type, uint_t *n_clnts,
5372     char *clnts_name)
5373 {
5374 	mac_impl_t *mip = (mac_impl_t *)mh;
5375 	mac_grp_client_t *mcip;
5376 	uint_t i = 0, index = 0;
5377 	mac_ring_t	*ring;
5378 
5379 	/* Revisit when we implement fully dynamic group allocation */
5380 	ASSERT(grp_index >= 0 && grp_index <= mip->mi_tx_group_count);
5381 
5382 	rw_enter(&mip->mi_rw_lock, RW_READER);
5383 	*grp_num = mip->mi_tx_groups[grp_index].mrg_index > 0 ?
5384 	    mip->mi_tx_groups[grp_index].mrg_index : grp_index;
5385 	*type = mip->mi_tx_groups[grp_index].mrg_type;
5386 	*n_rings = mip->mi_tx_groups[grp_index].mrg_cur_count;
5387 	ring = mip->mi_tx_groups[grp_index].mrg_rings;
5388 	for (index = 0; index < mip->mi_tx_groups[grp_index].mrg_cur_count;
5389 	    index++) {
5390 		rings[index] = ring->mr_index;
5391 		ring = ring->mr_next;
5392 	}
5393 	index = 0;
5394 	/* Default group has an index of -1 */
5395 	if (mip->mi_tx_groups[grp_index].mrg_index < 0) {
5396 		(void) strlcpy(clnts_name, "<default>,",
5397 		    MAXCLIENTNAMELEN);
5398 		index += strlen("<default>,");
5399 	}
5400 	for (mcip = mip->mi_tx_groups[grp_index].mrg_clients; mcip != NULL;
5401 	    mcip = mcip->mgc_next) {
5402 		int name_len = strlen(mcip->mgc_client->mci_name);
5403 
5404 		/*
5405 		 * MAXCLIENTNAMELEN is the buffer size reserved for client
5406 		 * names.
5407 		 * XXXX Formating the client name string needs to be moved
5408 		 * to user land when fixing the size of dhi_clnts in
5409 		 * dld_hwgrpinfo_t. We should use n_clients * client_name for
5410 		 * dhi_clntsin instead of MAXCLIENTNAMELEN
5411 		 */
5412 		if (index + name_len >= MAXCLIENTNAMELEN) {
5413 			index = MAXCLIENTNAMELEN;
5414 			break;
5415 		}
5416 		bcopy(mcip->mgc_client->mci_name, &(clnts_name[index]),
5417 		    name_len);
5418 		index += name_len;
5419 		clnts_name[index++] = ',';
5420 		i++;
5421 	}
5422 
5423 	/* Get rid of the last , */
5424 	if (index > 0)
5425 		clnts_name[index - 1] = '\0';
5426 	*n_clnts = i;
5427 	rw_exit(&mip->mi_rw_lock);
5428 }
5429 
5430 /*
5431  * Return the group count for RX or TX.
5432  */
5433 uint_t
5434 mac_hwgrp_num(mac_handle_t mh, int type)
5435 {
5436 	mac_impl_t *mip = (mac_impl_t *)mh;
5437 
5438 	/*
5439 	 * Return the Rx and Tx group count; for the Tx we need to
5440 	 * include the default too.
5441 	 */
5442 	return (type == MAC_RING_TYPE_RX ? mip->mi_rx_group_count :
5443 	    mip->mi_tx_groups != NULL ? mip->mi_tx_group_count + 1 : 0);
5444 }
5445 
5446 /*
5447  * The total number of free TX rings for this MAC.
5448  */
5449 uint_t
5450 mac_txavail_get(mac_handle_t mh)
5451 {
5452 	mac_impl_t	*mip = (mac_impl_t *)mh;
5453 
5454 	return (mip->mi_txrings_avail);
5455 }
5456 
5457 /*
5458  * The total number of free RX rings for this MAC.
5459  */
5460 uint_t
5461 mac_rxavail_get(mac_handle_t mh)
5462 {
5463 	mac_impl_t	*mip = (mac_impl_t *)mh;
5464 
5465 	return (mip->mi_rxrings_avail);
5466 }
5467 
5468 /*
5469  * The total number of reserved RX rings on this MAC.
5470  */
5471 uint_t
5472 mac_rxrsvd_get(mac_handle_t mh)
5473 {
5474 	mac_impl_t	*mip = (mac_impl_t *)mh;
5475 
5476 	return (mip->mi_rxrings_rsvd);
5477 }
5478 
5479 /*
5480  * The total number of reserved TX rings on this MAC.
5481  */
5482 uint_t
5483 mac_txrsvd_get(mac_handle_t mh)
5484 {
5485 	mac_impl_t	*mip = (mac_impl_t *)mh;
5486 
5487 	return (mip->mi_txrings_rsvd);
5488 }
5489 
5490 /*
5491  * Total number of free RX groups on this MAC.
5492  */
5493 uint_t
5494 mac_rxhwlnksavail_get(mac_handle_t mh)
5495 {
5496 	mac_impl_t	*mip = (mac_impl_t *)mh;
5497 
5498 	return (mip->mi_rxhwclnt_avail);
5499 }
5500 
5501 /*
5502  * Total number of RX groups reserved on this MAC.
5503  */
5504 uint_t
5505 mac_rxhwlnksrsvd_get(mac_handle_t mh)
5506 {
5507 	mac_impl_t	*mip = (mac_impl_t *)mh;
5508 
5509 	return (mip->mi_rxhwclnt_used);
5510 }
5511 
5512 /*
5513  * Total number of free TX groups on this MAC.
5514  */
5515 uint_t
5516 mac_txhwlnksavail_get(mac_handle_t mh)
5517 {
5518 	mac_impl_t	*mip = (mac_impl_t *)mh;
5519 
5520 	return (mip->mi_txhwclnt_avail);
5521 }
5522 
5523 /*
5524  * Total number of TX groups reserved on this MAC.
5525  */
5526 uint_t
5527 mac_txhwlnksrsvd_get(mac_handle_t mh)
5528 {
5529 	mac_impl_t	*mip = (mac_impl_t *)mh;
5530 
5531 	return (mip->mi_txhwclnt_used);
5532 }
5533 
5534 /*
5535  * Initialize the rings property for a mac client. A non-0 value for
5536  * rxring or txring specifies the number of rings required, a value
5537  * of MAC_RXRINGS_NONE/MAC_TXRINGS_NONE specifies that it doesn't need
5538  * any RX/TX rings and a value of MAC_RXRINGS_DONTCARE/MAC_TXRINGS_DONTCARE
5539  * means the system can decide whether it can give any rings or not.
5540  */
5541 void
5542 mac_client_set_rings(mac_client_handle_t mch, int rxrings, int txrings)
5543 {
5544 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
5545 	mac_resource_props_t	*mrp = MCIP_RESOURCE_PROPS(mcip);
5546 
5547 	if (rxrings != MAC_RXRINGS_DONTCARE) {
5548 		mrp->mrp_mask |= MRP_RX_RINGS;
5549 		mrp->mrp_nrxrings = rxrings;
5550 	}
5551 
5552 	if (txrings != MAC_TXRINGS_DONTCARE) {
5553 		mrp->mrp_mask |= MRP_TX_RINGS;
5554 		mrp->mrp_ntxrings = txrings;
5555 	}
5556 }
5557 
5558 boolean_t
5559 mac_get_promisc_filtered(mac_client_handle_t mch)
5560 {
5561 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
5562 
5563 	return (mcip->mci_protect_flags & MPT_FLAG_PROMISC_FILTERED);
5564 }
5565 
5566 void
5567 mac_set_promisc_filtered(mac_client_handle_t mch, boolean_t enable)
5568 {
5569 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
5570 
5571 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
5572 	if (enable)
5573 		mcip->mci_protect_flags |= MPT_FLAG_PROMISC_FILTERED;
5574 	else
5575 		mcip->mci_protect_flags &= ~MPT_FLAG_PROMISC_FILTERED;
5576 }
5577