xref: /illumos-gate/usr/src/uts/common/io/mac/mac_client.c (revision ddb365bfc9e868ad24ccdcb0dc91af18b10df082)
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 2019 Joyent, Inc.
25  * Copyright 2017 RackTop Systems.
26  * Copyright 2022 OmniOS Community Edition (OmniOSce) Association.
27  */
28 
29 /*
30  * - General Introduction:
31  *
32  * This file contains the implementation of the MAC client kernel
33  * API and related code. The MAC client API allows a kernel module
34  * to gain access to a MAC instance (physical NIC, link aggregation, etc).
35  * It allows a MAC client to associate itself with a MAC address,
36  * VLANs, callback functions for data traffic and for promiscuous mode.
37  * The MAC client API is also used to specify the properties associated
38  * with a MAC client, such as bandwidth limits, priority, CPUS, etc.
39  * These properties are further used to determine the hardware resources
40  * to allocate to the various MAC clients.
41  *
42  * - Primary MAC clients:
43  *
44  * The MAC client API refers to "primary MAC clients". A primary MAC
45  * client is a client which "owns" the primary MAC address of
46  * the underlying MAC instance. The primary MAC address is called out
47  * since it is associated with specific semantics: the primary MAC
48  * address is the MAC address which is assigned to the IP interface
49  * when it is plumbed, and the primary MAC address is assigned
50  * to VLAN data-links. The primary address of a MAC instance can
51  * also change dynamically from under the MAC client, for example
52  * as a result of a change of state of a link aggregation. In that
53  * case the MAC layer automatically updates all data-structures which
54  * refer to the current value of the primary MAC address. Typical
55  * primary MAC clients are dls, aggr, and xnb. A typical non-primary
56  * MAC client is the vnic driver.
57  *
58  * - Virtual Switching:
59  *
60  * The MAC layer implements a virtual switch between the MAC clients
61  * (primary and non-primary) defined on top of the same underlying
62  * NIC (physical, link aggregation, etc). The virtual switch is
63  * VLAN-aware, i.e. it allows multiple MAC clients to be member
64  * of one or more VLANs, and the virtual switch will distribute
65  * multicast tagged packets only to the member of the corresponding
66  * VLANs.
67  *
68  * - Upper vs Lower MAC:
69  *
70  * Creating a VNIC on top of a MAC instance effectively causes
71  * two MAC instances to be layered on top of each other, one for
72  * the VNIC(s), one for the underlying MAC instance (physical NIC,
73  * link aggregation, etc). In the code below we refer to the
74  * underlying NIC as the "lower MAC", and we refer to VNICs as
75  * the "upper MAC".
76  *
77  * - Pass-through for VNICs:
78  *
79  * When VNICs are created on top of an underlying MAC, this causes
80  * a layering of two MAC instances. Since the lower MAC already
81  * does the switching and demultiplexing to its MAC clients, the
82  * upper MAC would simply have to pass packets to the layer below
83  * or above it, which would introduce overhead. In order to avoid
84  * this overhead, the MAC layer implements a pass-through mechanism
85  * for VNICs. When a VNIC opens the lower MAC instance, it saves
86  * the MAC client handle it optains from the MAC layer. When a MAC
87  * client opens a VNIC (upper MAC), the MAC layer detects that
88  * the MAC being opened is a VNIC, and gets the MAC client handle
89  * that the VNIC driver obtained from the lower MAC. This exchange
90  * is done through a private capability between the MAC layer
91  * and the VNIC driver. The upper MAC then returns that handle
92  * directly to its MAC client. Any operation done by the upper
93  * MAC client is now done on the lower MAC client handle, which
94  * allows the VNIC driver to be completely bypassed for the
95  * performance sensitive data-path.
96  *
97  * - Secondary MACs for VNICs:
98  *
99  * VNICs support multiple upper mac clients to enable support for
100  * multiple MAC addresses on the VNIC. When the VNIC is created the
101  * initial mac client is the primary upper mac. Any additional mac
102  * clients are secondary macs. These are kept in sync with the primary
103  * (for things such as the rx function and resource control settings)
104  * using the same private capability interface between the MAC layer
105  * and the VNIC layer.
106  *
107  */
108 
109 #include <sys/types.h>
110 #include <sys/conf.h>
111 #include <sys/id_space.h>
112 #include <sys/esunddi.h>
113 #include <sys/stat.h>
114 #include <sys/mkdev.h>
115 #include <sys/stream.h>
116 #include <sys/strsun.h>
117 #include <sys/strsubr.h>
118 #include <sys/pattr.h>
119 #include <sys/dlpi.h>
120 #include <sys/modhash.h>
121 #include <sys/mac_impl.h>
122 #include <sys/mac_client_impl.h>
123 #include <sys/mac_soft_ring.h>
124 #include <sys/mac_stat.h>
125 #include <sys/dls.h>
126 #include <sys/dld.h>
127 #include <sys/modctl.h>
128 #include <sys/fs/dv_node.h>
129 #include <sys/thread.h>
130 #include <sys/proc.h>
131 #include <sys/callb.h>
132 #include <sys/cpuvar.h>
133 #include <sys/atomic.h>
134 #include <sys/sdt.h>
135 #include <sys/mac_flow.h>
136 #include <sys/ddi_intr_impl.h>
137 #include <sys/disp.h>
138 #include <sys/sdt.h>
139 #include <sys/vnic.h>
140 #include <sys/vnic_impl.h>
141 #include <sys/vlan.h>
142 #include <inet/ip.h>
143 #include <inet/ip6.h>
144 #include <sys/exacct.h>
145 #include <sys/exacct_impl.h>
146 #include <inet/nd.h>
147 #include <sys/ethernet.h>
148 
149 kmem_cache_t	*mac_client_impl_cache;
150 kmem_cache_t	*mac_promisc_impl_cache;
151 
152 static boolean_t mac_client_single_rcvr(mac_client_impl_t *);
153 static flow_entry_t *mac_client_swap_mciflent(mac_client_impl_t *);
154 static flow_entry_t *mac_client_get_flow(mac_client_impl_t *,
155     mac_unicast_impl_t *);
156 static void mac_client_remove_flow_from_list(mac_client_impl_t *,
157     flow_entry_t *);
158 static void mac_client_add_to_flow_list(mac_client_impl_t *, flow_entry_t *);
159 static void mac_rename_flow_names(mac_client_impl_t *, const char *);
160 static void mac_virtual_link_update(mac_impl_t *);
161 static int mac_client_datapath_setup(mac_client_impl_t *, uint16_t,
162     uint8_t *, mac_resource_props_t *, boolean_t, mac_unicast_impl_t *);
163 static void mac_client_datapath_teardown(mac_client_handle_t,
164     mac_unicast_impl_t *, flow_entry_t *);
165 static int mac_resource_ctl_set(mac_client_handle_t, mac_resource_props_t *);
166 
167 /* ARGSUSED */
168 static int
169 i_mac_client_impl_ctor(void *buf, void *arg, int kmflag)
170 {
171 	int	i;
172 	mac_client_impl_t	*mcip = buf;
173 
174 	bzero(buf, MAC_CLIENT_IMPL_SIZE);
175 	mutex_init(&mcip->mci_tx_cb_lock, NULL, MUTEX_DRIVER, NULL);
176 	mcip->mci_tx_notify_cb_info.mcbi_lockp = &mcip->mci_tx_cb_lock;
177 
178 	ASSERT(mac_tx_percpu_cnt >= 0);
179 	for (i = 0; i <= mac_tx_percpu_cnt; i++) {
180 		mutex_init(&mcip->mci_tx_pcpu[i].pcpu_tx_lock, NULL,
181 		    MUTEX_DRIVER, NULL);
182 	}
183 	cv_init(&mcip->mci_tx_cv, NULL, CV_DRIVER, NULL);
184 
185 	return (0);
186 }
187 
188 /* ARGSUSED */
189 static void
190 i_mac_client_impl_dtor(void *buf, void *arg)
191 {
192 	int	i;
193 	mac_client_impl_t *mcip = buf;
194 
195 	ASSERT(mcip->mci_promisc_list == NULL);
196 	ASSERT(mcip->mci_unicast_list == NULL);
197 	ASSERT(mcip->mci_state_flags == 0);
198 	ASSERT(mcip->mci_tx_flag == 0);
199 
200 	mutex_destroy(&mcip->mci_tx_cb_lock);
201 
202 	ASSERT(mac_tx_percpu_cnt >= 0);
203 	for (i = 0; i <= mac_tx_percpu_cnt; i++) {
204 		ASSERT(mcip->mci_tx_pcpu[i].pcpu_tx_refcnt == 0);
205 		mutex_destroy(&mcip->mci_tx_pcpu[i].pcpu_tx_lock);
206 	}
207 	cv_destroy(&mcip->mci_tx_cv);
208 }
209 
210 /* ARGSUSED */
211 static int
212 i_mac_promisc_impl_ctor(void *buf, void *arg, int kmflag)
213 {
214 	mac_promisc_impl_t	*mpip = buf;
215 
216 	bzero(buf, sizeof (mac_promisc_impl_t));
217 	mpip->mpi_mci_link.mcb_objp = buf;
218 	mpip->mpi_mci_link.mcb_objsize = sizeof (mac_promisc_impl_t);
219 	mpip->mpi_mi_link.mcb_objp = buf;
220 	mpip->mpi_mi_link.mcb_objsize = sizeof (mac_promisc_impl_t);
221 	return (0);
222 }
223 
224 /* ARGSUSED */
225 static void
226 i_mac_promisc_impl_dtor(void *buf, void *arg)
227 {
228 	mac_promisc_impl_t	*mpip = buf;
229 
230 	ASSERT(mpip->mpi_mci_link.mcb_objp != NULL);
231 	ASSERT(mpip->mpi_mci_link.mcb_objsize == sizeof (mac_promisc_impl_t));
232 	ASSERT(mpip->mpi_mi_link.mcb_objp == mpip->mpi_mci_link.mcb_objp);
233 	ASSERT(mpip->mpi_mi_link.mcb_objsize == sizeof (mac_promisc_impl_t));
234 
235 	mpip->mpi_mci_link.mcb_objp = NULL;
236 	mpip->mpi_mci_link.mcb_objsize = 0;
237 	mpip->mpi_mi_link.mcb_objp = NULL;
238 	mpip->mpi_mi_link.mcb_objsize = 0;
239 
240 	ASSERT(mpip->mpi_mci_link.mcb_flags == 0);
241 	mpip->mpi_mci_link.mcb_objsize = 0;
242 }
243 
244 void
245 mac_client_init(void)
246 {
247 	ASSERT(mac_tx_percpu_cnt >= 0);
248 
249 	mac_client_impl_cache = kmem_cache_create("mac_client_impl_cache",
250 	    MAC_CLIENT_IMPL_SIZE, 0, i_mac_client_impl_ctor,
251 	    i_mac_client_impl_dtor, NULL, NULL, NULL, 0);
252 	ASSERT(mac_client_impl_cache != NULL);
253 
254 	mac_promisc_impl_cache = kmem_cache_create("mac_promisc_impl_cache",
255 	    sizeof (mac_promisc_impl_t), 0, i_mac_promisc_impl_ctor,
256 	    i_mac_promisc_impl_dtor, NULL, NULL, NULL, 0);
257 	ASSERT(mac_promisc_impl_cache != NULL);
258 }
259 
260 void
261 mac_client_fini(void)
262 {
263 	kmem_cache_destroy(mac_client_impl_cache);
264 	kmem_cache_destroy(mac_promisc_impl_cache);
265 }
266 
267 /*
268  * Return the lower MAC client handle from the VNIC driver for the
269  * specified VNIC MAC instance.
270  */
271 mac_client_impl_t *
272 mac_vnic_lower(mac_impl_t *mip)
273 {
274 	mac_capab_vnic_t cap;
275 	mac_client_impl_t *mcip;
276 
277 	VERIFY(i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_VNIC, &cap));
278 	mcip = cap.mcv_mac_client_handle(cap.mcv_arg);
279 
280 	return (mcip);
281 }
282 
283 /*
284  * Update the secondary macs
285  */
286 void
287 mac_vnic_secondary_update(mac_impl_t *mip)
288 {
289 	mac_capab_vnic_t cap;
290 
291 	VERIFY(i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_VNIC, &cap));
292 	cap.mcv_mac_secondary_update(cap.mcv_arg);
293 }
294 
295 /*
296  * Return the MAC client handle of the primary MAC client for the
297  * specified MAC instance, or NULL otherwise.
298  */
299 mac_client_impl_t *
300 mac_primary_client_handle(mac_impl_t *mip)
301 {
302 	mac_client_impl_t *mcip;
303 
304 	if (mip->mi_state_flags & MIS_IS_VNIC)
305 		return (mac_vnic_lower(mip));
306 
307 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
308 
309 	for (mcip = mip->mi_clients_list; mcip != NULL;
310 	    mcip = mcip->mci_client_next) {
311 		if (MCIP_DATAPATH_SETUP(mcip) && mac_is_primary_client(mcip))
312 			return (mcip);
313 	}
314 	return (NULL);
315 }
316 
317 /*
318  * Open a MAC specified by its MAC name.
319  */
320 int
321 mac_open(const char *macname, mac_handle_t *mhp)
322 {
323 	mac_impl_t	*mip;
324 	int		err;
325 
326 	/*
327 	 * Look up its entry in the global hash table.
328 	 */
329 	if ((err = mac_hold(macname, &mip)) != 0)
330 		return (err);
331 
332 	/*
333 	 * Hold the dip associated to the MAC to prevent it from being
334 	 * detached. For a softmac, its underlying dip is held by the
335 	 * mi_open() callback.
336 	 *
337 	 * This is done to be more tolerant with some defective drivers,
338 	 * which incorrectly handle mac_unregister() failure in their
339 	 * xxx_detach() routine. For example, some drivers ignore the
340 	 * failure of mac_unregister() and free all resources that
341 	 * that are needed for data transmition.
342 	 */
343 	e_ddi_hold_devi(mip->mi_dip);
344 
345 	if (!(mip->mi_callbacks->mc_callbacks & MC_OPEN)) {
346 		*mhp = (mac_handle_t)mip;
347 		return (0);
348 	}
349 
350 	/*
351 	 * The mac perimeter is used in both mac_open and mac_close by the
352 	 * framework to single thread the MC_OPEN/MC_CLOSE of drivers.
353 	 */
354 	i_mac_perim_enter(mip);
355 	mip->mi_oref++;
356 	if (mip->mi_oref != 1 || ((err = mip->mi_open(mip->mi_driver)) == 0)) {
357 		*mhp = (mac_handle_t)mip;
358 		i_mac_perim_exit(mip);
359 		return (0);
360 	}
361 	mip->mi_oref--;
362 	ddi_release_devi(mip->mi_dip);
363 	mac_rele(mip);
364 	i_mac_perim_exit(mip);
365 	return (err);
366 }
367 
368 /*
369  * Open a MAC specified by its linkid.
370  */
371 int
372 mac_open_by_linkid(datalink_id_t linkid, mac_handle_t *mhp)
373 {
374 	dls_dl_handle_t	dlh;
375 	int		err;
376 
377 	if ((err = dls_devnet_hold_tmp(linkid, &dlh)) != 0)
378 		return (err);
379 
380 	dls_devnet_prop_task_wait(dlh);
381 
382 	err = mac_open(dls_devnet_mac(dlh), mhp);
383 
384 	dls_devnet_rele_tmp(dlh);
385 	return (err);
386 }
387 
388 /*
389  * Open a MAC specified by its link name.
390  */
391 int
392 mac_open_by_linkname(const char *link, mac_handle_t *mhp)
393 {
394 	datalink_id_t	linkid;
395 	int		err;
396 
397 	if ((err = dls_mgmt_get_linkid(link, &linkid)) != 0)
398 		return (err);
399 	return (mac_open_by_linkid(linkid, mhp));
400 }
401 
402 /*
403  * Close the specified MAC.
404  */
405 void
406 mac_close(mac_handle_t mh)
407 {
408 	mac_impl_t	*mip = (mac_impl_t *)mh;
409 
410 	i_mac_perim_enter(mip);
411 	/*
412 	 * The mac perimeter is used in both mac_open and mac_close by the
413 	 * framework to single thread the MC_OPEN/MC_CLOSE of drivers.
414 	 */
415 	if (mip->mi_callbacks->mc_callbacks & MC_OPEN) {
416 		ASSERT(mip->mi_oref != 0);
417 		if (--mip->mi_oref == 0) {
418 			if ((mip->mi_callbacks->mc_callbacks & MC_CLOSE))
419 				mip->mi_close(mip->mi_driver);
420 		}
421 	}
422 	i_mac_perim_exit(mip);
423 	ddi_release_devi(mip->mi_dip);
424 	mac_rele(mip);
425 }
426 
427 /*
428  * Misc utility functions to retrieve various information about a MAC
429  * instance or a MAC client.
430  */
431 
432 const mac_info_t *
433 mac_info(mac_handle_t mh)
434 {
435 	return (&((mac_impl_t *)mh)->mi_info);
436 }
437 
438 dev_info_t *
439 mac_devinfo_get(mac_handle_t mh)
440 {
441 	return (((mac_impl_t *)mh)->mi_dip);
442 }
443 
444 void *
445 mac_driver(mac_handle_t mh)
446 {
447 	return (((mac_impl_t *)mh)->mi_driver);
448 }
449 
450 const char *
451 mac_name(mac_handle_t mh)
452 {
453 	return (((mac_impl_t *)mh)->mi_name);
454 }
455 
456 int
457 mac_type(mac_handle_t mh)
458 {
459 	return (((mac_impl_t *)mh)->mi_type->mt_type);
460 }
461 
462 int
463 mac_nativetype(mac_handle_t mh)
464 {
465 	return (((mac_impl_t *)mh)->mi_type->mt_nativetype);
466 }
467 
468 char *
469 mac_client_name(mac_client_handle_t mch)
470 {
471 	return (((mac_client_impl_t *)mch)->mci_name);
472 }
473 
474 minor_t
475 mac_minor(mac_handle_t mh)
476 {
477 	return (((mac_impl_t *)mh)->mi_minor);
478 }
479 
480 /*
481  * Return the VID associated with a MAC client. This function should
482  * be called for clients which are associated with only one VID.
483  */
484 uint16_t
485 mac_client_vid(mac_client_handle_t mch)
486 {
487 	uint16_t		vid = VLAN_ID_NONE;
488 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
489 	flow_desc_t		flow_desc;
490 
491 	if (mcip->mci_nflents == 0)
492 		return (vid);
493 
494 	ASSERT(MCIP_DATAPATH_SETUP(mcip) && mac_client_single_rcvr(mcip));
495 
496 	mac_flow_get_desc(mcip->mci_flent, &flow_desc);
497 	if ((flow_desc.fd_mask & FLOW_LINK_VID) != 0)
498 		vid = flow_desc.fd_vid;
499 
500 	return (vid);
501 }
502 
503 /*
504  * Return whether the specified MAC client corresponds to a VLAN VNIC.
505  */
506 boolean_t
507 mac_client_is_vlan_vnic(mac_client_handle_t mch)
508 {
509 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
510 
511 	return (((mcip->mci_state_flags & MCIS_IS_VNIC) != 0) &&
512 	    ((mcip->mci_flent->fe_type & FLOW_PRIMARY_MAC) != 0));
513 }
514 
515 /*
516  * Return the link speed associated with the specified MAC client.
517  *
518  * The link speed of a MAC client is equal to the smallest value of
519  * 1) the current link speed of the underlying NIC, or
520  * 2) the bandwidth limit set for the MAC client.
521  *
522  * Note that the bandwidth limit can be higher than the speed
523  * of the underlying NIC. This is allowed to avoid spurious
524  * administration action failures or artifically lowering the
525  * bandwidth limit of a link that may  have temporarily lowered
526  * its link speed due to hardware problem or administrator action.
527  */
528 static uint64_t
529 mac_client_ifspeed(mac_client_impl_t *mcip)
530 {
531 	mac_impl_t *mip = mcip->mci_mip;
532 	uint64_t nic_speed;
533 
534 	nic_speed = mac_stat_get((mac_handle_t)mip, MAC_STAT_IFSPEED);
535 
536 	if (nic_speed == 0) {
537 		return (0);
538 	} else {
539 		uint64_t policy_limit = (uint64_t)-1;
540 
541 		if (MCIP_RESOURCE_PROPS_MASK(mcip) & MRP_MAXBW)
542 			policy_limit = MCIP_RESOURCE_PROPS_MAXBW(mcip);
543 
544 		return (MIN(policy_limit, nic_speed));
545 	}
546 }
547 
548 /*
549  * Return the link state of the specified client. If here are more
550  * than one clients of the underying mac_impl_t, the link state
551  * will always be UP regardless of the link state of the underlying
552  * mac_impl_t. This is needed to allow the MAC clients to continue
553  * to communicate with each other even when the physical link of
554  * their mac_impl_t is down.
555  */
556 static uint64_t
557 mac_client_link_state(mac_client_impl_t *mcip)
558 {
559 	mac_impl_t *mip = mcip->mci_mip;
560 	uint16_t vid;
561 	mac_client_impl_t *mci_list;
562 	mac_unicast_impl_t *mui_list, *oth_mui_list;
563 
564 	/*
565 	 * Returns LINK_STATE_UP if there are other MAC clients defined on
566 	 * mac_impl_t which share same VLAN ID as that of mcip. Note that
567 	 * if 'mcip' has more than one VID's then we match ANY one of the
568 	 * VID's with other MAC client's VID's and return LINK_STATE_UP.
569 	 */
570 	rw_enter(&mcip->mci_rw_lock, RW_READER);
571 	for (mui_list = mcip->mci_unicast_list; mui_list != NULL;
572 	    mui_list = mui_list->mui_next) {
573 		vid = mui_list->mui_vid;
574 		for (mci_list = mip->mi_clients_list; mci_list != NULL;
575 		    mci_list = mci_list->mci_client_next) {
576 			if (mci_list == mcip)
577 				continue;
578 			for (oth_mui_list = mci_list->mci_unicast_list;
579 			    oth_mui_list != NULL; oth_mui_list = oth_mui_list->
580 			    mui_next) {
581 				if (vid == oth_mui_list->mui_vid) {
582 					rw_exit(&mcip->mci_rw_lock);
583 					return (LINK_STATE_UP);
584 				}
585 			}
586 		}
587 	}
588 	rw_exit(&mcip->mci_rw_lock);
589 
590 	return (mac_stat_get((mac_handle_t)mip, MAC_STAT_LINK_STATE));
591 }
592 
593 /*
594  * These statistics are consumed by dladm show-link -s <vnic>,
595  * dladm show-vnic -s and netstat. With the introduction of dlstat,
596  * dladm show-link -s and dladm show-vnic -s witll be EOL'ed while
597  * netstat will consume from kstats introduced for dlstat. This code
598  * will be removed at that time.
599  */
600 
601 /*
602  * Return the statistics of a MAC client. These statistics are different
603  * then the statistics of the underlying MAC which are returned by
604  * mac_stat_get().
605  *
606  * Note that for things based on the tx and rx stats, mac will end up clobbering
607  * those stats when the underlying set of rings in the srs changes. As such, we
608  * need to source not only the current set, but also the historical set when
609  * returning to the client, lest our counters appear to go backwards.
610  */
611 uint64_t
612 mac_client_stat_get(mac_client_handle_t mch, uint_t stat)
613 {
614 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
615 	mac_impl_t		*mip = mcip->mci_mip;
616 	flow_entry_t		*flent = mcip->mci_flent;
617 	mac_soft_ring_set_t	*mac_srs;
618 	mac_rx_stats_t		*mac_rx_stat, *old_rx_stat;
619 	mac_tx_stats_t		*mac_tx_stat, *old_tx_stat;
620 	int i;
621 	uint64_t val = 0;
622 
623 	mac_srs = (mac_soft_ring_set_t *)(flent->fe_tx_srs);
624 	mac_tx_stat = &mac_srs->srs_tx.st_stat;
625 	old_rx_stat = &mcip->mci_misc_stat.mms_defunctrxlanestats;
626 	old_tx_stat = &mcip->mci_misc_stat.mms_defuncttxlanestats;
627 
628 	switch (stat) {
629 	case MAC_STAT_LINK_STATE:
630 		val = mac_client_link_state(mcip);
631 		break;
632 	case MAC_STAT_LINK_UP:
633 		val = (mac_client_link_state(mcip) == LINK_STATE_UP);
634 		break;
635 	case MAC_STAT_PROMISC:
636 		val = mac_stat_get((mac_handle_t)mip, MAC_STAT_PROMISC);
637 		break;
638 	case MAC_STAT_LOWLINK_STATE:
639 		val = mac_stat_get((mac_handle_t)mip, MAC_STAT_LOWLINK_STATE);
640 		break;
641 	case MAC_STAT_IFSPEED:
642 		val = mac_client_ifspeed(mcip);
643 		break;
644 	case MAC_STAT_MULTIRCV:
645 		val = mcip->mci_misc_stat.mms_multircv;
646 		break;
647 	case MAC_STAT_BRDCSTRCV:
648 		val = mcip->mci_misc_stat.mms_brdcstrcv;
649 		break;
650 	case MAC_STAT_MULTIXMT:
651 		val = mcip->mci_misc_stat.mms_multixmt;
652 		break;
653 	case MAC_STAT_BRDCSTXMT:
654 		val = mcip->mci_misc_stat.mms_brdcstxmt;
655 		break;
656 	case MAC_STAT_OBYTES:
657 		val = mac_tx_stat->mts_obytes;
658 		val += old_tx_stat->mts_obytes;
659 		break;
660 	case MAC_STAT_OPACKETS:
661 		val = mac_tx_stat->mts_opackets;
662 		val += old_tx_stat->mts_opackets;
663 		break;
664 	case MAC_STAT_OERRORS:
665 		val = mac_tx_stat->mts_oerrors;
666 		val += old_tx_stat->mts_oerrors;
667 		break;
668 	case MAC_STAT_IPACKETS:
669 		for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
670 			mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
671 			mac_rx_stat = &mac_srs->srs_rx.sr_stat;
672 			val += mac_rx_stat->mrs_intrcnt +
673 			    mac_rx_stat->mrs_pollcnt + mac_rx_stat->mrs_lclcnt;
674 		}
675 		val += old_rx_stat->mrs_intrcnt + old_rx_stat->mrs_pollcnt +
676 		    old_rx_stat->mrs_lclcnt;
677 		break;
678 	case MAC_STAT_RBYTES:
679 		for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
680 			mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
681 			mac_rx_stat = &mac_srs->srs_rx.sr_stat;
682 			val += mac_rx_stat->mrs_intrbytes +
683 			    mac_rx_stat->mrs_pollbytes +
684 			    mac_rx_stat->mrs_lclbytes;
685 		}
686 		val += old_rx_stat->mrs_intrbytes + old_rx_stat->mrs_pollbytes +
687 		    old_rx_stat->mrs_lclbytes;
688 		break;
689 	case MAC_STAT_IERRORS:
690 		for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
691 			mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
692 			mac_rx_stat = &mac_srs->srs_rx.sr_stat;
693 			val += mac_rx_stat->mrs_ierrors;
694 		}
695 		val += old_rx_stat->mrs_ierrors;
696 		break;
697 	default:
698 		val = mac_driver_stat_default(mip, stat);
699 		break;
700 	}
701 
702 	return (val);
703 }
704 
705 /*
706  * Return the statistics of the specified MAC instance.
707  */
708 uint64_t
709 mac_stat_get(mac_handle_t mh, uint_t stat)
710 {
711 	mac_impl_t	*mip = (mac_impl_t *)mh;
712 	uint64_t	val;
713 	int		ret;
714 
715 	/*
716 	 * The range of stat determines where it is maintained.  Stat
717 	 * values from 0 up to (but not including) MAC_STAT_MIN are
718 	 * mainteined by the mac module itself.  Everything else is
719 	 * maintained by the driver.
720 	 *
721 	 * If the mac_impl_t being queried corresponds to a VNIC,
722 	 * the stats need to be queried from the lower MAC client
723 	 * corresponding to the VNIC. (The mac_link_update()
724 	 * invoked by the driver to the lower MAC causes the *lower
725 	 * MAC* to update its mi_linkstate, and send a notification
726 	 * to its MAC clients. Due to the VNIC passthrough,
727 	 * these notifications are sent to the upper MAC clients
728 	 * of the VNIC directly, and the upper mac_impl_t of the VNIC
729 	 * does not have a valid mi_linkstate.
730 	 */
731 	if (stat < MAC_STAT_MIN && !(mip->mi_state_flags & MIS_IS_VNIC)) {
732 		/* these stats are maintained by the mac module itself */
733 		switch (stat) {
734 		case MAC_STAT_LINK_STATE:
735 			return (mip->mi_linkstate);
736 		case MAC_STAT_LINK_UP:
737 			return (mip->mi_linkstate == LINK_STATE_UP);
738 		case MAC_STAT_PROMISC:
739 			return (mip->mi_devpromisc != 0);
740 		case MAC_STAT_LOWLINK_STATE:
741 			return (mip->mi_lowlinkstate);
742 		default:
743 			ASSERT(B_FALSE);
744 		}
745 	}
746 
747 	/*
748 	 * Call the driver to get the given statistic.
749 	 */
750 	ret = mip->mi_getstat(mip->mi_driver, stat, &val);
751 	if (ret != 0) {
752 		/*
753 		 * The driver doesn't support this statistic.  Get the
754 		 * statistic's default value.
755 		 */
756 		val = mac_driver_stat_default(mip, stat);
757 	}
758 	return (val);
759 }
760 
761 /*
762  * Query hardware rx ring corresponding to the pseudo ring.
763  */
764 uint64_t
765 mac_pseudo_rx_ring_stat_get(mac_ring_handle_t handle, uint_t stat)
766 {
767 	return (mac_rx_ring_stat_get(handle, stat));
768 }
769 
770 /*
771  * Query hardware tx ring corresponding to the pseudo ring.
772  */
773 uint64_t
774 mac_pseudo_tx_ring_stat_get(mac_ring_handle_t handle, uint_t stat)
775 {
776 	return (mac_tx_ring_stat_get(handle, stat));
777 }
778 
779 /*
780  * Utility function which returns the VID associated with a flow entry.
781  */
782 uint16_t
783 i_mac_flow_vid(flow_entry_t *flent)
784 {
785 	flow_desc_t	flow_desc;
786 
787 	mac_flow_get_desc(flent, &flow_desc);
788 
789 	if ((flow_desc.fd_mask & FLOW_LINK_VID) != 0)
790 		return (flow_desc.fd_vid);
791 	return (VLAN_ID_NONE);
792 }
793 
794 /*
795  * Verify the validity of the specified unicast MAC address. Returns B_TRUE
796  * if the address is valid, B_FALSE otherwise (multicast address, or incorrect
797  * length.
798  */
799 boolean_t
800 mac_unicst_verify(mac_handle_t mh, const uint8_t *addr, uint_t len)
801 {
802 	mac_impl_t	*mip = (mac_impl_t *)mh;
803 
804 	/*
805 	 * Verify the address. No lock is needed since mi_type and plugin
806 	 * details don't change after mac_register().
807 	 */
808 	if ((len != mip->mi_type->mt_addr_length) ||
809 	    (mip->mi_type->mt_ops.mtops_unicst_verify(addr,
810 	    mip->mi_pdata)) != 0) {
811 		return (B_FALSE);
812 	} else {
813 		return (B_TRUE);
814 	}
815 }
816 
817 void
818 mac_sdu_get(mac_handle_t mh, uint_t *min_sdu, uint_t *max_sdu)
819 {
820 	mac_impl_t	*mip = (mac_impl_t *)mh;
821 
822 	if (min_sdu != NULL)
823 		*min_sdu = mip->mi_sdu_min;
824 	if (max_sdu != NULL)
825 		*max_sdu = mip->mi_sdu_max;
826 }
827 
828 void
829 mac_sdu_get2(mac_handle_t mh, uint_t *min_sdu, uint_t *max_sdu,
830     uint_t *multicast_sdu)
831 {
832 	mac_impl_t	*mip = (mac_impl_t *)mh;
833 
834 	if (min_sdu != NULL)
835 		*min_sdu = mip->mi_sdu_min;
836 	if (max_sdu != NULL)
837 		*max_sdu = mip->mi_sdu_max;
838 	if (multicast_sdu != NULL)
839 		*multicast_sdu = mip->mi_sdu_multicast;
840 }
841 
842 /*
843  * Update the MAC unicast address of the specified client's flows. Currently
844  * only one unicast MAC unicast address is allowed per client.
845  */
846 static void
847 mac_unicast_update_client_flow(mac_client_impl_t *mcip)
848 {
849 	mac_impl_t *mip = mcip->mci_mip;
850 	flow_entry_t *flent = mcip->mci_flent;
851 	mac_address_t *map = mcip->mci_unicast;
852 	flow_desc_t flow_desc;
853 
854 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
855 	ASSERT(flent != NULL);
856 
857 	mac_flow_get_desc(flent, &flow_desc);
858 	ASSERT(flow_desc.fd_mask & FLOW_LINK_DST);
859 
860 	bcopy(map->ma_addr, flow_desc.fd_dst_mac, map->ma_len);
861 	mac_flow_set_desc(flent, &flow_desc);
862 
863 	/*
864 	 * The v6 local and SLAAC addrs (used by mac protection) need to be
865 	 * regenerated because our mac address has changed.
866 	 */
867 	mac_protect_update_mac_token(mcip);
868 
869 	/*
870 	 * When there are multiple VLANs sharing the same MAC address,
871 	 * each gets its own MAC client, except when running on sun4v
872 	 * vsw. In that case the mci_flent_list is used to place
873 	 * multiple VLAN flows on one MAC client. If we ever get rid
874 	 * of vsw then this code can go, but until then we need to
875 	 * update all flow entries.
876 	 */
877 	for (flent = mcip->mci_flent_list; flent != NULL;
878 	    flent = flent->fe_client_next) {
879 		mac_flow_get_desc(flent, &flow_desc);
880 		if (!(flent->fe_type & FLOW_PRIMARY_MAC ||
881 		    flent->fe_type & FLOW_VNIC_MAC))
882 			continue;
883 
884 		bcopy(map->ma_addr, flow_desc.fd_dst_mac, map->ma_len);
885 		mac_flow_set_desc(flent, &flow_desc);
886 	}
887 }
888 
889 /*
890  * Update all clients that share the same unicast address.
891  */
892 void
893 mac_unicast_update_clients(mac_impl_t *mip, mac_address_t *map)
894 {
895 	mac_client_impl_t *mcip;
896 
897 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
898 
899 	/*
900 	 * Find all clients that share the same unicast MAC address and update
901 	 * them appropriately.
902 	 */
903 	for (mcip = mip->mi_clients_list; mcip != NULL;
904 	    mcip = mcip->mci_client_next) {
905 		/*
906 		 * Ignore clients that don't share this MAC address.
907 		 */
908 		if (map != mcip->mci_unicast)
909 			continue;
910 
911 		/*
912 		 * Update those clients with same old unicast MAC address.
913 		 */
914 		mac_unicast_update_client_flow(mcip);
915 	}
916 }
917 
918 /*
919  * Update the unicast MAC address of the specified VNIC MAC client.
920  *
921  * Check whether the operation is valid. Any of following cases should fail:
922  *
923  * 1. It's a VLAN type of VNIC.
924  * 2. The new value is current "primary" MAC address.
925  * 3. The current MAC address is shared with other clients.
926  * 4. The new MAC address has been used. This case will be valid when
927  *    client migration is fully supported.
928  */
929 int
930 mac_vnic_unicast_set(mac_client_handle_t mch, const uint8_t *addr)
931 {
932 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
933 	mac_impl_t *mip = mcip->mci_mip;
934 	mac_address_t *map = mcip->mci_unicast;
935 	int err;
936 
937 	ASSERT(!(mip->mi_state_flags & MIS_IS_VNIC));
938 	ASSERT(mcip->mci_state_flags & MCIS_IS_VNIC);
939 	ASSERT(mcip->mci_flags != MAC_CLIENT_FLAGS_PRIMARY);
940 
941 	i_mac_perim_enter(mip);
942 
943 	/*
944 	 * If this is a VLAN type of VNIC, it's using "primary" MAC address
945 	 * of the underlying interface. Must fail here. Refer to case 1 above.
946 	 */
947 	if (bcmp(map->ma_addr, mip->mi_addr, map->ma_len) == 0) {
948 		i_mac_perim_exit(mip);
949 		return (ENOTSUP);
950 	}
951 
952 	/*
953 	 * If the new address is the "primary" one, must fail. Refer to
954 	 * case 2 above.
955 	 */
956 	if (bcmp(addr, mip->mi_addr, map->ma_len) == 0) {
957 		i_mac_perim_exit(mip);
958 		return (EACCES);
959 	}
960 
961 	/*
962 	 * If the address is shared by multiple clients, must fail. Refer
963 	 * to case 3 above.
964 	 */
965 	if (mac_check_macaddr_shared(map)) {
966 		i_mac_perim_exit(mip);
967 		return (EBUSY);
968 	}
969 
970 	/*
971 	 * If the new address has been used, must fail for now. Refer to
972 	 * case 4 above.
973 	 */
974 	if (mac_find_macaddr(mip, (uint8_t *)addr) != NULL) {
975 		i_mac_perim_exit(mip);
976 		return (ENOTSUP);
977 	}
978 
979 	/*
980 	 * Update the MAC address.
981 	 */
982 	err = mac_update_macaddr(map, (uint8_t *)addr);
983 
984 	if (err != 0) {
985 		i_mac_perim_exit(mip);
986 		return (err);
987 	}
988 
989 	/*
990 	 * Update all flows of this MAC client.
991 	 */
992 	mac_unicast_update_client_flow(mcip);
993 
994 	i_mac_perim_exit(mip);
995 	return (0);
996 }
997 
998 /*
999  * Program the new primary unicast address of the specified MAC.
1000  *
1001  * Function mac_update_macaddr() takes care different types of underlying
1002  * MAC. If the underlying MAC is VNIC, the VNIC driver must have registerd
1003  * mi_unicst() entry point, that indirectly calls mac_vnic_unicast_set()
1004  * which will take care of updating the MAC address of the corresponding
1005  * MAC client.
1006  *
1007  * This is the only interface that allow the client to update the "primary"
1008  * MAC address of the underlying MAC. The new value must have not been
1009  * used by other clients.
1010  */
1011 int
1012 mac_unicast_primary_set(mac_handle_t mh, const uint8_t *addr)
1013 {
1014 	mac_impl_t *mip = (mac_impl_t *)mh;
1015 	mac_address_t *map;
1016 	int err;
1017 
1018 	/* verify the address validity */
1019 	if (!mac_unicst_verify(mh, addr, mip->mi_type->mt_addr_length))
1020 		return (EINVAL);
1021 
1022 	i_mac_perim_enter(mip);
1023 
1024 	/*
1025 	 * If the new value is the same as the current primary address value,
1026 	 * there's nothing to do.
1027 	 */
1028 	if (bcmp(addr, mip->mi_addr, mip->mi_type->mt_addr_length) == 0) {
1029 		i_mac_perim_exit(mip);
1030 		return (0);
1031 	}
1032 
1033 	if (mac_find_macaddr(mip, (uint8_t *)addr) != NULL) {
1034 		i_mac_perim_exit(mip);
1035 		return (EBUSY);
1036 	}
1037 
1038 	map = mac_find_macaddr(mip, mip->mi_addr);
1039 	ASSERT(map != NULL);
1040 
1041 	/*
1042 	 * Update the MAC address.
1043 	 */
1044 	if (mip->mi_state_flags & MIS_IS_AGGR) {
1045 		mac_capab_aggr_t aggr_cap;
1046 
1047 		/*
1048 		 * If the MAC is an aggregation, other than the unicast
1049 		 * addresses programming, aggr must be informed about this
1050 		 * primary unicst address change to change its MAC address
1051 		 * policy to be user-specified.
1052 		 */
1053 		ASSERT(map->ma_type == MAC_ADDRESS_TYPE_UNICAST_CLASSIFIED);
1054 		VERIFY(i_mac_capab_get(mh, MAC_CAPAB_AGGR, &aggr_cap));
1055 		err = aggr_cap.mca_unicst(mip->mi_driver, addr);
1056 		if (err == 0)
1057 			bcopy(addr, map->ma_addr, map->ma_len);
1058 	} else {
1059 		err = mac_update_macaddr(map, (uint8_t *)addr);
1060 	}
1061 
1062 	if (err != 0) {
1063 		i_mac_perim_exit(mip);
1064 		return (err);
1065 	}
1066 
1067 	mac_unicast_update_clients(mip, map);
1068 
1069 	/*
1070 	 * Save the new primary MAC address in mac_impl_t.
1071 	 */
1072 	bcopy(addr, mip->mi_addr, mip->mi_type->mt_addr_length);
1073 
1074 	i_mac_perim_exit(mip);
1075 
1076 	if (err == 0)
1077 		i_mac_notify(mip, MAC_NOTE_UNICST);
1078 
1079 	return (err);
1080 }
1081 
1082 /*
1083  * Return the current primary MAC address of the specified MAC.
1084  */
1085 void
1086 mac_unicast_primary_get(mac_handle_t mh, uint8_t *addr)
1087 {
1088 	mac_impl_t *mip = (mac_impl_t *)mh;
1089 
1090 	rw_enter(&mip->mi_rw_lock, RW_READER);
1091 	bcopy(mip->mi_addr, addr, mip->mi_type->mt_addr_length);
1092 	rw_exit(&mip->mi_rw_lock);
1093 }
1094 
1095 /*
1096  * Return the secondary MAC address for the specified handle
1097  */
1098 void
1099 mac_unicast_secondary_get(mac_client_handle_t mh, uint8_t *addr)
1100 {
1101 	mac_client_impl_t *mcip = (mac_client_impl_t *)mh;
1102 
1103 	ASSERT(mcip->mci_unicast != NULL);
1104 	bcopy(mcip->mci_unicast->ma_addr, addr, mcip->mci_unicast->ma_len);
1105 }
1106 
1107 /*
1108  * Return information about the use of the primary MAC address of the
1109  * specified MAC instance:
1110  *
1111  * - if client_name is non-NULL, it must point to a string of at
1112  *   least MAXNAMELEN bytes, and will be set to the name of the MAC
1113  *   client which uses the primary MAC address.
1114  *
1115  * - if in_use is non-NULL, used to return whether the primary MAC
1116  *   address is currently in use.
1117  */
1118 void
1119 mac_unicast_primary_info(mac_handle_t mh, char *client_name, boolean_t *in_use)
1120 {
1121 	mac_impl_t *mip = (mac_impl_t *)mh;
1122 	mac_client_impl_t *cur_client;
1123 
1124 	if (in_use != NULL)
1125 		*in_use = B_FALSE;
1126 	if (client_name != NULL)
1127 		bzero(client_name, MAXNAMELEN);
1128 
1129 	/*
1130 	 * The mi_rw_lock is used to protect threads that don't hold the
1131 	 * mac perimeter to get a consistent view of the mi_clients_list.
1132 	 * Threads that modify the list must hold both the mac perimeter and
1133 	 * mi_rw_lock(RW_WRITER)
1134 	 */
1135 	rw_enter(&mip->mi_rw_lock, RW_READER);
1136 	for (cur_client = mip->mi_clients_list; cur_client != NULL;
1137 	    cur_client = cur_client->mci_client_next) {
1138 		if (mac_is_primary_client(cur_client) ||
1139 		    (mip->mi_state_flags & MIS_IS_VNIC)) {
1140 			rw_exit(&mip->mi_rw_lock);
1141 			if (in_use != NULL)
1142 				*in_use = B_TRUE;
1143 			if (client_name != NULL) {
1144 				bcopy(cur_client->mci_name, client_name,
1145 				    MAXNAMELEN);
1146 			}
1147 			return;
1148 		}
1149 	}
1150 	rw_exit(&mip->mi_rw_lock);
1151 }
1152 
1153 /*
1154  * Return the current destination MAC address of the specified MAC.
1155  */
1156 boolean_t
1157 mac_dst_get(mac_handle_t mh, uint8_t *addr)
1158 {
1159 	mac_impl_t *mip = (mac_impl_t *)mh;
1160 
1161 	rw_enter(&mip->mi_rw_lock, RW_READER);
1162 	if (mip->mi_dstaddr_set)
1163 		bcopy(mip->mi_dstaddr, addr, mip->mi_type->mt_addr_length);
1164 	rw_exit(&mip->mi_rw_lock);
1165 	return (mip->mi_dstaddr_set);
1166 }
1167 
1168 /*
1169  * Add the specified MAC client to the list of clients which opened
1170  * the specified MAC.
1171  */
1172 static void
1173 mac_client_add(mac_client_impl_t *mcip)
1174 {
1175 	mac_impl_t *mip = mcip->mci_mip;
1176 
1177 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1178 
1179 	/* add VNIC to the front of the list */
1180 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
1181 	mcip->mci_client_next = mip->mi_clients_list;
1182 	mip->mi_clients_list = mcip;
1183 	mip->mi_nclients++;
1184 	rw_exit(&mip->mi_rw_lock);
1185 }
1186 
1187 /*
1188  * Remove the specified MAC client from the list of clients which opened
1189  * the specified MAC.
1190  */
1191 static void
1192 mac_client_remove(mac_client_impl_t *mcip)
1193 {
1194 	mac_impl_t *mip = mcip->mci_mip;
1195 	mac_client_impl_t **prev, *cclient;
1196 
1197 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1198 
1199 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
1200 	prev = &mip->mi_clients_list;
1201 	cclient = *prev;
1202 	while (cclient != NULL && cclient != mcip) {
1203 		prev = &cclient->mci_client_next;
1204 		cclient = *prev;
1205 	}
1206 	ASSERT(cclient != NULL);
1207 	*prev = cclient->mci_client_next;
1208 	mip->mi_nclients--;
1209 	rw_exit(&mip->mi_rw_lock);
1210 }
1211 
1212 static mac_unicast_impl_t *
1213 mac_client_find_vid(mac_client_impl_t *mcip, uint16_t vid)
1214 {
1215 	mac_unicast_impl_t *muip = mcip->mci_unicast_list;
1216 
1217 	while ((muip != NULL) && (muip->mui_vid != vid))
1218 		muip = muip->mui_next;
1219 
1220 	return (muip);
1221 }
1222 
1223 /*
1224  * Return whether the specified (MAC address, VID) tuple is already used by
1225  * one of the MAC clients associated with the specified MAC.
1226  */
1227 static boolean_t
1228 mac_addr_in_use(mac_impl_t *mip, uint8_t *mac_addr, uint16_t vid)
1229 {
1230 	mac_client_impl_t *client;
1231 	mac_address_t *map;
1232 
1233 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1234 
1235 	for (client = mip->mi_clients_list; client != NULL;
1236 	    client = client->mci_client_next) {
1237 
1238 		/*
1239 		 * Ignore clients that don't have unicast address.
1240 		 */
1241 		if (client->mci_unicast_list == NULL)
1242 			continue;
1243 
1244 		map = client->mci_unicast;
1245 
1246 		if ((bcmp(mac_addr, map->ma_addr, map->ma_len) == 0) &&
1247 		    (mac_client_find_vid(client, vid) != NULL)) {
1248 			return (B_TRUE);
1249 		}
1250 	}
1251 
1252 	return (B_FALSE);
1253 }
1254 
1255 /*
1256  * Generate a random MAC address. The MAC address prefix is
1257  * stored in the array pointed to by mac_addr, and its length, in bytes,
1258  * is specified by prefix_len. The least significant bits
1259  * after prefix_len bytes are generated, and stored after the prefix
1260  * in the mac_addr array.
1261  */
1262 int
1263 mac_addr_random(mac_client_handle_t mch, uint_t prefix_len,
1264     uint8_t *mac_addr, mac_diag_t *diag)
1265 {
1266 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1267 	mac_impl_t *mip = mcip->mci_mip;
1268 	size_t addr_len = mip->mi_type->mt_addr_length;
1269 
1270 	if (prefix_len >= addr_len) {
1271 		*diag = MAC_DIAG_MACPREFIXLEN_INVALID;
1272 		return (EINVAL);
1273 	}
1274 
1275 	/* check the prefix value */
1276 	if (prefix_len > 0) {
1277 		bzero(mac_addr + prefix_len, addr_len - prefix_len);
1278 		if (!mac_unicst_verify((mac_handle_t)mip, mac_addr,
1279 		    addr_len)) {
1280 			*diag = MAC_DIAG_MACPREFIX_INVALID;
1281 			return (EINVAL);
1282 		}
1283 	}
1284 
1285 	/* generate the MAC address */
1286 	if (prefix_len < addr_len) {
1287 		(void) random_get_pseudo_bytes(mac_addr +
1288 		    prefix_len, addr_len - prefix_len);
1289 	}
1290 
1291 	*diag = MAC_DIAG_NONE;
1292 	return (0);
1293 }
1294 
1295 /*
1296  * Set the priority range for this MAC client. This will be used to
1297  * determine the absolute priority for the threads created for this
1298  * MAC client using the specified "low", "medium" and "high" level.
1299  * This will also be used for any subflows on this MAC client.
1300  */
1301 #define	MAC_CLIENT_SET_PRIORITY_RANGE(mcip, pri) {			\
1302 	(mcip)->mci_min_pri = FLOW_MIN_PRIORITY(MINCLSYSPRI,	\
1303 	    MAXCLSYSPRI, (pri));					\
1304 	(mcip)->mci_max_pri = FLOW_MAX_PRIORITY(MINCLSYSPRI,	\
1305 	    MAXCLSYSPRI, (mcip)->mci_min_pri);				\
1306 	}
1307 
1308 /*
1309  * MAC client open entry point. Return a new MAC client handle. Each
1310  * MAC client is associated with a name, specified through the 'name'
1311  * argument.
1312  */
1313 int
1314 mac_client_open(mac_handle_t mh, mac_client_handle_t *mchp, char *name,
1315     uint16_t flags)
1316 {
1317 	mac_impl_t		*mip = (mac_impl_t *)mh;
1318 	mac_client_impl_t	*mcip;
1319 	int			err = 0;
1320 	boolean_t		share_desired;
1321 	flow_entry_t		*flent = NULL;
1322 
1323 	share_desired = (flags & MAC_OPEN_FLAGS_SHARES_DESIRED) != 0;
1324 	*mchp = NULL;
1325 
1326 	i_mac_perim_enter(mip);
1327 
1328 	if (mip->mi_state_flags & MIS_IS_VNIC) {
1329 		/*
1330 		 * The underlying MAC is a VNIC. Return the MAC client
1331 		 * handle of the lower MAC which was obtained by
1332 		 * the VNIC driver when it did its mac_client_open().
1333 		 */
1334 
1335 		mcip = mac_vnic_lower(mip);
1336 
1337 		/*
1338 		 * Note that multiple mac clients share the same mcip in
1339 		 * this case.
1340 		 */
1341 		if (flags & MAC_OPEN_FLAGS_EXCLUSIVE)
1342 			mcip->mci_state_flags |= MCIS_EXCLUSIVE;
1343 
1344 		if (flags & MAC_OPEN_FLAGS_MULTI_PRIMARY)
1345 			mcip->mci_flags |= MAC_CLIENT_FLAGS_MULTI_PRIMARY;
1346 
1347 		mip->mi_clients_list = mcip;
1348 		i_mac_perim_exit(mip);
1349 		*mchp = (mac_client_handle_t)mcip;
1350 
1351 		DTRACE_PROBE2(mac__client__open__nonallocated, mac_impl_t *,
1352 		    mcip->mci_mip, mac_client_impl_t *, mcip);
1353 
1354 		return (err);
1355 	}
1356 
1357 	mcip = kmem_cache_alloc(mac_client_impl_cache, KM_SLEEP);
1358 
1359 	mcip->mci_mip = mip;
1360 	mcip->mci_upper_mip = NULL;
1361 	mcip->mci_rx_fn = mac_rx_def;
1362 	mcip->mci_rx_arg = NULL;
1363 	mcip->mci_rx_p_fn = NULL;
1364 	mcip->mci_rx_p_arg = NULL;
1365 	mcip->mci_p_unicast_list = NULL;
1366 	mcip->mci_direct_rx_fn = NULL;
1367 	mcip->mci_direct_rx_arg = NULL;
1368 	mcip->mci_vidcache = MCIP_VIDCACHE_INVALID;
1369 
1370 	mcip->mci_unicast_list = NULL;
1371 
1372 	if ((flags & MAC_OPEN_FLAGS_IS_VNIC) != 0)
1373 		mcip->mci_state_flags |= MCIS_IS_VNIC;
1374 
1375 	if ((flags & MAC_OPEN_FLAGS_EXCLUSIVE) != 0)
1376 		mcip->mci_state_flags |= MCIS_EXCLUSIVE;
1377 
1378 	if ((flags & MAC_OPEN_FLAGS_IS_AGGR_PORT) != 0)
1379 		mcip->mci_state_flags |= MCIS_IS_AGGR_PORT;
1380 
1381 	if (mip->mi_state_flags & MIS_IS_AGGR)
1382 		mcip->mci_state_flags |= MCIS_IS_AGGR_CLIENT;
1383 
1384 	if ((flags & MAC_OPEN_FLAGS_USE_DATALINK_NAME) != 0) {
1385 		datalink_id_t	linkid;
1386 
1387 		ASSERT(name == NULL);
1388 		if ((err = dls_devnet_macname2linkid(mip->mi_name,
1389 		    &linkid)) != 0) {
1390 			goto done;
1391 		}
1392 		if ((err = dls_mgmt_get_linkinfo(linkid, mcip->mci_name, NULL,
1393 		    NULL, NULL)) != 0) {
1394 			/*
1395 			 * Use mac name if dlmgmtd is not available.
1396 			 */
1397 			if (err == EBADF) {
1398 				(void) strlcpy(mcip->mci_name, mip->mi_name,
1399 				    sizeof (mcip->mci_name));
1400 				err = 0;
1401 			} else {
1402 				goto done;
1403 			}
1404 		}
1405 		mcip->mci_state_flags |= MCIS_USE_DATALINK_NAME;
1406 	} else {
1407 		ASSERT(name != NULL);
1408 		if (strlen(name) > MAXNAMELEN) {
1409 			err = EINVAL;
1410 			goto done;
1411 		}
1412 		(void) strlcpy(mcip->mci_name, name, sizeof (mcip->mci_name));
1413 	}
1414 
1415 	if (flags & MAC_OPEN_FLAGS_MULTI_PRIMARY)
1416 		mcip->mci_flags |= MAC_CLIENT_FLAGS_MULTI_PRIMARY;
1417 
1418 	if (flags & MAC_OPEN_FLAGS_NO_UNICAST_ADDR)
1419 		mcip->mci_state_flags |= MCIS_NO_UNICAST_ADDR;
1420 
1421 	mac_protect_init(mcip);
1422 
1423 	/* the subflow table will be created dynamically */
1424 	mcip->mci_subflow_tab = NULL;
1425 
1426 	mcip->mci_misc_stat.mms_multircv = 0;
1427 	mcip->mci_misc_stat.mms_brdcstrcv = 0;
1428 	mcip->mci_misc_stat.mms_multixmt = 0;
1429 	mcip->mci_misc_stat.mms_brdcstxmt = 0;
1430 
1431 	/* Create an initial flow */
1432 
1433 	err = mac_flow_create(NULL, NULL, mcip->mci_name, NULL,
1434 	    mcip->mci_state_flags & MCIS_IS_VNIC ? FLOW_VNIC_MAC :
1435 	    FLOW_PRIMARY_MAC, &flent);
1436 	if (err != 0)
1437 		goto done;
1438 	mcip->mci_flent = flent;
1439 	FLOW_MARK(flent, FE_MC_NO_DATAPATH);
1440 	flent->fe_mcip = mcip;
1441 
1442 	/*
1443 	 * Place initial creation reference on the flow. This reference
1444 	 * is released in the corresponding delete action viz.
1445 	 * mac_unicast_remove after waiting for all transient refs to
1446 	 * to go away. The wait happens in mac_flow_wait.
1447 	 */
1448 	FLOW_REFHOLD(flent);
1449 
1450 	/*
1451 	 * Do this ahead of the mac_bcast_add() below so that the mi_nclients
1452 	 * will have the right value for mac_rx_srs_setup().
1453 	 */
1454 	mac_client_add(mcip);
1455 
1456 	mcip->mci_share = 0;
1457 	if (share_desired)
1458 		i_mac_share_alloc(mcip);
1459 
1460 	/*
1461 	 * We will do mimimal datapath setup to allow a MAC client to
1462 	 * transmit or receive non-unicast packets without waiting
1463 	 * for mac_unicast_add.
1464 	 */
1465 	if (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR) {
1466 		if ((err = mac_client_datapath_setup(mcip, VLAN_ID_NONE,
1467 		    NULL, NULL, B_TRUE, NULL)) != 0) {
1468 			goto done;
1469 		}
1470 	}
1471 
1472 	DTRACE_PROBE2(mac__client__open__allocated, mac_impl_t *,
1473 	    mcip->mci_mip, mac_client_impl_t *, mcip);
1474 
1475 	*mchp = (mac_client_handle_t)mcip;
1476 	i_mac_perim_exit(mip);
1477 	return (0);
1478 
1479 done:
1480 	i_mac_perim_exit(mip);
1481 	mcip->mci_state_flags = 0;
1482 	mcip->mci_tx_flag = 0;
1483 	kmem_cache_free(mac_client_impl_cache, mcip);
1484 	return (err);
1485 }
1486 
1487 /*
1488  * Close the specified MAC client handle.
1489  */
1490 void
1491 mac_client_close(mac_client_handle_t mch, uint16_t flags)
1492 {
1493 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
1494 	mac_impl_t		*mip = mcip->mci_mip;
1495 	flow_entry_t		*flent;
1496 
1497 	i_mac_perim_enter(mip);
1498 
1499 	if (flags & MAC_CLOSE_FLAGS_EXCLUSIVE)
1500 		mcip->mci_state_flags &= ~MCIS_EXCLUSIVE;
1501 
1502 	if ((mcip->mci_state_flags & MCIS_IS_VNIC) &&
1503 	    !(flags & MAC_CLOSE_FLAGS_IS_VNIC)) {
1504 		/*
1505 		 * This is an upper VNIC client initiated operation.
1506 		 * The lower MAC client will be closed by the VNIC driver
1507 		 * when the VNIC is deleted.
1508 		 */
1509 
1510 		i_mac_perim_exit(mip);
1511 		return;
1512 	}
1513 
1514 	/* If we have only setup up minimal datapth setup, tear it down */
1515 	if (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR) {
1516 		mac_client_datapath_teardown((mac_client_handle_t)mcip, NULL,
1517 		    mcip->mci_flent);
1518 		mcip->mci_state_flags &= ~MCIS_NO_UNICAST_ADDR;
1519 	}
1520 
1521 	/*
1522 	 * Remove the flent associated with the MAC client
1523 	 */
1524 	flent = mcip->mci_flent;
1525 	mcip->mci_flent = NULL;
1526 	FLOW_FINAL_REFRELE(flent);
1527 
1528 	/*
1529 	 * MAC clients must remove the unicast addresses and promisc callbacks
1530 	 * they added before issuing a mac_client_close().
1531 	 */
1532 	ASSERT(mcip->mci_unicast_list == NULL);
1533 	ASSERT(mcip->mci_promisc_list == NULL);
1534 	ASSERT(mcip->mci_tx_notify_cb_list == NULL);
1535 
1536 	i_mac_share_free(mcip);
1537 	mac_protect_fini(mcip);
1538 	mac_client_remove(mcip);
1539 
1540 	i_mac_perim_exit(mip);
1541 	mcip->mci_subflow_tab = NULL;
1542 	mcip->mci_state_flags = 0;
1543 	mcip->mci_tx_flag = 0;
1544 	kmem_cache_free(mac_client_impl_cache, mch);
1545 }
1546 
1547 /*
1548  * Set the Rx bypass receive callback and return B_TRUE. Return
1549  * B_FALSE if it's not possible to enable bypass.
1550  */
1551 boolean_t
1552 mac_rx_bypass_set(mac_client_handle_t mch, mac_direct_rx_t rx_fn, void *arg1)
1553 {
1554 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
1555 	mac_impl_t		*mip = mcip->mci_mip;
1556 
1557 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1558 
1559 	/*
1560 	 * If the client has more than one VLAN then process packets
1561 	 * through DLS. This should happen only when sun4v vsw is on
1562 	 * the scene.
1563 	 */
1564 	if (mcip->mci_nvids > 1)
1565 		return (B_FALSE);
1566 
1567 	/*
1568 	 * These are not accessed directly in the data path, and hence
1569 	 * don't need any protection
1570 	 */
1571 	mcip->mci_direct_rx_fn = rx_fn;
1572 	mcip->mci_direct_rx_arg = arg1;
1573 	return (B_TRUE);
1574 }
1575 
1576 /*
1577  * Enable/Disable rx bypass. By default, bypass is assumed to be enabled.
1578  */
1579 void
1580 mac_rx_bypass_enable(mac_client_handle_t mch)
1581 {
1582 	((mac_client_impl_t *)mch)->mci_state_flags &= ~MCIS_RX_BYPASS_DISABLE;
1583 }
1584 
1585 void
1586 mac_rx_bypass_disable(mac_client_handle_t mch)
1587 {
1588 	((mac_client_impl_t *)mch)->mci_state_flags |= MCIS_RX_BYPASS_DISABLE;
1589 }
1590 
1591 /*
1592  * Set the receive callback for the specified MAC client. There can be
1593  * at most one such callback per MAC client.
1594  */
1595 void
1596 mac_rx_set(mac_client_handle_t mch, mac_rx_t rx_fn, void *arg)
1597 {
1598 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1599 	mac_impl_t	*mip = mcip->mci_mip;
1600 	mac_impl_t	*umip = mcip->mci_upper_mip;
1601 
1602 	/*
1603 	 * Instead of adding an extra set of locks and refcnts in
1604 	 * the datapath at the mac client boundary, we temporarily quiesce
1605 	 * the SRS and related entities. We then change the receive function
1606 	 * without interference from any receive data thread and then reenable
1607 	 * the data flow subsequently.
1608 	 */
1609 	i_mac_perim_enter(mip);
1610 	mac_rx_client_quiesce(mch);
1611 
1612 	mcip->mci_rx_fn = rx_fn;
1613 	mcip->mci_rx_arg = arg;
1614 	mac_rx_client_restart(mch);
1615 
1616 	/*
1617 	 * If we're changing the Rx function on the primary MAC of a VNIC,
1618 	 * make sure any secondary addresses on the VNIC are updated as well.
1619 	 */
1620 	if (umip != NULL) {
1621 		ASSERT((umip->mi_state_flags & MIS_IS_VNIC) != 0);
1622 		mac_vnic_secondary_update(umip);
1623 	}
1624 
1625 	i_mac_perim_exit(mip);
1626 }
1627 
1628 /*
1629  * Reset the receive callback for the specified MAC client.
1630  */
1631 void
1632 mac_rx_clear(mac_client_handle_t mch)
1633 {
1634 	mac_rx_set(mch, mac_rx_def, NULL);
1635 }
1636 
1637 void
1638 mac_rx_barrier(mac_client_handle_t mch)
1639 {
1640 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1641 	mac_impl_t *mip = mcip->mci_mip;
1642 
1643 	i_mac_perim_enter(mip);
1644 
1645 	/* If a RX callback is set, quiesce and restart that datapath */
1646 	if (mcip->mci_rx_fn != mac_rx_def) {
1647 		mac_rx_client_quiesce(mch);
1648 		mac_rx_client_restart(mch);
1649 	}
1650 
1651 	/* If any promisc callbacks are registered, perform a barrier there */
1652 	if (mcip->mci_promisc_list != NULL || mip->mi_promisc_list != NULL) {
1653 		mac_cb_info_t *mcbi =  &mip->mi_promisc_cb_info;
1654 
1655 		mutex_enter(mcbi->mcbi_lockp);
1656 		mac_callback_barrier(mcbi);
1657 		mutex_exit(mcbi->mcbi_lockp);
1658 	}
1659 
1660 	i_mac_perim_exit(mip);
1661 }
1662 
1663 void
1664 mac_secondary_dup(mac_client_handle_t smch, mac_client_handle_t dmch)
1665 {
1666 	mac_client_impl_t *smcip = (mac_client_impl_t *)smch;
1667 	mac_client_impl_t *dmcip = (mac_client_impl_t *)dmch;
1668 	flow_entry_t *flent = dmcip->mci_flent;
1669 
1670 	/* This should only be called to setup secondary macs */
1671 	ASSERT((flent->fe_type & FLOW_PRIMARY_MAC) == 0);
1672 
1673 	mac_rx_set(dmch, smcip->mci_rx_fn, smcip->mci_rx_arg);
1674 	dmcip->mci_promisc_list = smcip->mci_promisc_list;
1675 
1676 	/*
1677 	 * Duplicate the primary mac resources to the secondary.
1678 	 * Since we already validated the resource controls when setting
1679 	 * them on the primary, we can ignore errors here.
1680 	 */
1681 	(void) mac_resource_ctl_set(dmch, MCIP_RESOURCE_PROPS(smcip));
1682 }
1683 
1684 /*
1685  * Called when removing a secondary MAC. Currently only clears the promisc_list
1686  * since we share the primary mac's promisc_list.
1687  */
1688 void
1689 mac_secondary_cleanup(mac_client_handle_t mch)
1690 {
1691 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1692 	flow_entry_t *flent = mcip->mci_flent;
1693 
1694 	/* This should only be called for secondary macs */
1695 	ASSERT((flent->fe_type & FLOW_PRIMARY_MAC) == 0);
1696 	mcip->mci_promisc_list = NULL;
1697 }
1698 
1699 /*
1700  * Walk the MAC client subflow table and updates their priority values.
1701  */
1702 static int
1703 mac_update_subflow_priority_cb(flow_entry_t *flent, void *arg)
1704 {
1705 	mac_flow_update_priority(arg, flent);
1706 	return (0);
1707 }
1708 
1709 void
1710 mac_update_subflow_priority(mac_client_impl_t *mcip)
1711 {
1712 	(void) mac_flow_walk(mcip->mci_subflow_tab,
1713 	    mac_update_subflow_priority_cb, mcip);
1714 }
1715 
1716 /*
1717  * Modify the TX or RX ring properties. We could either just move around
1718  * rings, i.e add/remove rings given to a client. Or this might cause the
1719  * client to move from hardware based to software or the other way around.
1720  * If we want to reset this property, then we clear the mask, additionally
1721  * if the client was given a non-default group we remove all rings except
1722  * for 1 and give it back to the default group.
1723  */
1724 int
1725 mac_client_set_rings_prop(mac_client_impl_t *mcip, mac_resource_props_t *mrp,
1726     mac_resource_props_t *tmrp)
1727 {
1728 	mac_impl_t		*mip = mcip->mci_mip;
1729 	flow_entry_t		*flent = mcip->mci_flent;
1730 	uint8_t			*mac_addr;
1731 	int			err = 0;
1732 	mac_group_t		*defgrp;
1733 	mac_group_t		*group;
1734 	mac_group_t		*ngrp;
1735 	mac_resource_props_t	*cmrp = MCIP_RESOURCE_PROPS(mcip);
1736 	uint_t			ringcnt;
1737 	boolean_t		unspec;
1738 
1739 	if (mcip->mci_share != 0)
1740 		return (EINVAL);
1741 
1742 	if (mrp->mrp_mask & MRP_RX_RINGS) {
1743 		unspec = mrp->mrp_mask & MRP_RXRINGS_UNSPEC;
1744 		group = flent->fe_rx_ring_group;
1745 		defgrp = MAC_DEFAULT_RX_GROUP(mip);
1746 		mac_addr = flent->fe_flow_desc.fd_dst_mac;
1747 
1748 		/*
1749 		 * No resulting change. If we are resetting on a client on
1750 		 * which there was no rx rings property. For dynamic group
1751 		 * if we are setting the same number of rings already set.
1752 		 * For static group if we are requesting a group again.
1753 		 */
1754 		if (mrp->mrp_mask & MRP_RINGS_RESET) {
1755 			if (!(tmrp->mrp_mask & MRP_RX_RINGS))
1756 				return (0);
1757 		} else {
1758 			if (unspec) {
1759 				if (tmrp->mrp_mask & MRP_RXRINGS_UNSPEC)
1760 					return (0);
1761 			} else if (mip->mi_rx_group_type ==
1762 			    MAC_GROUP_TYPE_DYNAMIC) {
1763 				if ((tmrp->mrp_mask & MRP_RX_RINGS) &&
1764 				    !(tmrp->mrp_mask & MRP_RXRINGS_UNSPEC) &&
1765 				    mrp->mrp_nrxrings == tmrp->mrp_nrxrings) {
1766 					return (0);
1767 				}
1768 			}
1769 		}
1770 		/* Resetting the prop */
1771 		if (mrp->mrp_mask & MRP_RINGS_RESET) {
1772 			/*
1773 			 * We will just keep one ring and give others back if
1774 			 * we are not the primary. For the primary we give
1775 			 * all the rings in the default group except the
1776 			 * default ring. If it is a static group, then
1777 			 * we don't do anything, but clear the MRP_RX_RINGS
1778 			 * flag.
1779 			 */
1780 			if (group != defgrp) {
1781 				if (mip->mi_rx_group_type ==
1782 				    MAC_GROUP_TYPE_DYNAMIC) {
1783 					/*
1784 					 * This group has reserved rings
1785 					 * that need to be released now,
1786 					 * so does the group.
1787 					 */
1788 					MAC_RX_RING_RELEASED(mip,
1789 					    group->mrg_cur_count);
1790 					MAC_RX_GRP_RELEASED(mip);
1791 					if ((flent->fe_type &
1792 					    FLOW_PRIMARY_MAC) != 0) {
1793 						if (mip->mi_nactiveclients ==
1794 						    1) {
1795 							(void)
1796 							    mac_rx_switch_group(
1797 							    mcip, group,
1798 							    defgrp);
1799 							return (0);
1800 						} else {
1801 							cmrp->mrp_nrxrings =
1802 							    group->
1803 							    mrg_cur_count +
1804 							    defgrp->
1805 							    mrg_cur_count - 1;
1806 						}
1807 					} else {
1808 						cmrp->mrp_nrxrings = 1;
1809 					}
1810 					(void) mac_group_ring_modify(mcip,
1811 					    group, defgrp);
1812 				} else {
1813 					/*
1814 					 * If this is a static group, we
1815 					 * need to release the group. The
1816 					 * client will remain in the same
1817 					 * group till some other client
1818 					 * needs this group.
1819 					 */
1820 					MAC_RX_GRP_RELEASED(mip);
1821 				}
1822 			/* Let check if we can give this an excl group */
1823 			} else if (group == defgrp) {
1824 				/*
1825 				 * If multiple clients share an
1826 				 * address then they must stay on the
1827 				 * default group.
1828 				 */
1829 				if (mac_check_macaddr_shared(mcip->mci_unicast))
1830 					return (0);
1831 
1832 				ngrp = mac_reserve_rx_group(mcip, mac_addr,
1833 				    B_TRUE);
1834 				/* Couldn't give it a group, that's fine */
1835 				if (ngrp == NULL)
1836 					return (0);
1837 				/* Switch to H/W */
1838 				if (mac_rx_switch_group(mcip, defgrp, ngrp) !=
1839 				    0) {
1840 					mac_stop_group(ngrp);
1841 					return (0);
1842 				}
1843 			}
1844 			/*
1845 			 * If the client is in the default group, we will
1846 			 * just clear the MRP_RX_RINGS and leave it as
1847 			 * it rather than look for an exclusive group
1848 			 * for it.
1849 			 */
1850 			return (0);
1851 		}
1852 
1853 		if (group == defgrp && ((mrp->mrp_nrxrings > 0) || unspec)) {
1854 			/*
1855 			 * We are requesting Rx rings. Try to reserve
1856 			 * a non-default group.
1857 			 *
1858 			 * If multiple clients share an address then
1859 			 * they must stay on the default group.
1860 			 */
1861 			if (mac_check_macaddr_shared(mcip->mci_unicast))
1862 				return (EINVAL);
1863 
1864 			ngrp = mac_reserve_rx_group(mcip, mac_addr, B_TRUE);
1865 			if (ngrp == NULL)
1866 				return (ENOSPC);
1867 
1868 			/* Switch to H/W */
1869 			if (mac_rx_switch_group(mcip, defgrp, ngrp) != 0) {
1870 				mac_release_rx_group(mcip, ngrp);
1871 				return (ENOSPC);
1872 			}
1873 			MAC_RX_GRP_RESERVED(mip);
1874 			if (mip->mi_rx_group_type == MAC_GROUP_TYPE_DYNAMIC)
1875 				MAC_RX_RING_RESERVED(mip, ngrp->mrg_cur_count);
1876 		} else if (group != defgrp && !unspec &&
1877 		    mrp->mrp_nrxrings == 0) {
1878 			/* Switch to S/W */
1879 			ringcnt = group->mrg_cur_count;
1880 			if (mac_rx_switch_group(mcip, group, defgrp) != 0)
1881 				return (ENOSPC);
1882 			if (tmrp->mrp_mask & MRP_RX_RINGS) {
1883 				MAC_RX_GRP_RELEASED(mip);
1884 				if (mip->mi_rx_group_type ==
1885 				    MAC_GROUP_TYPE_DYNAMIC) {
1886 					MAC_RX_RING_RELEASED(mip, ringcnt);
1887 				}
1888 			}
1889 		} else if (group != defgrp && mip->mi_rx_group_type ==
1890 		    MAC_GROUP_TYPE_DYNAMIC) {
1891 			ringcnt = group->mrg_cur_count;
1892 			err = mac_group_ring_modify(mcip, group, defgrp);
1893 			if (err != 0)
1894 				return (err);
1895 			/*
1896 			 * Update the accounting. If this group
1897 			 * already had explicitly reserved rings,
1898 			 * we need to update the rings based on
1899 			 * the new ring count. If this group
1900 			 * had not explicitly reserved rings,
1901 			 * then we just reserve the rings asked for
1902 			 * and reserve the group.
1903 			 */
1904 			if (tmrp->mrp_mask & MRP_RX_RINGS) {
1905 				if (ringcnt > group->mrg_cur_count) {
1906 					MAC_RX_RING_RELEASED(mip,
1907 					    ringcnt - group->mrg_cur_count);
1908 				} else {
1909 					MAC_RX_RING_RESERVED(mip,
1910 					    group->mrg_cur_count - ringcnt);
1911 				}
1912 			} else {
1913 				MAC_RX_RING_RESERVED(mip, group->mrg_cur_count);
1914 				MAC_RX_GRP_RESERVED(mip);
1915 			}
1916 		}
1917 	}
1918 	if (mrp->mrp_mask & MRP_TX_RINGS) {
1919 		unspec = mrp->mrp_mask & MRP_TXRINGS_UNSPEC;
1920 		group = flent->fe_tx_ring_group;
1921 		defgrp = MAC_DEFAULT_TX_GROUP(mip);
1922 
1923 		/*
1924 		 * For static groups we only allow rings=0 or resetting the
1925 		 * rings property.
1926 		 */
1927 		if (mrp->mrp_ntxrings > 0 &&
1928 		    mip->mi_tx_group_type != MAC_GROUP_TYPE_DYNAMIC) {
1929 			return (ENOTSUP);
1930 		}
1931 		if (mrp->mrp_mask & MRP_RINGS_RESET) {
1932 			if (!(tmrp->mrp_mask & MRP_TX_RINGS))
1933 				return (0);
1934 		} else {
1935 			if (unspec) {
1936 				if (tmrp->mrp_mask & MRP_TXRINGS_UNSPEC)
1937 					return (0);
1938 			} else if (mip->mi_tx_group_type ==
1939 			    MAC_GROUP_TYPE_DYNAMIC) {
1940 				if ((tmrp->mrp_mask & MRP_TX_RINGS) &&
1941 				    !(tmrp->mrp_mask & MRP_TXRINGS_UNSPEC) &&
1942 				    mrp->mrp_ntxrings == tmrp->mrp_ntxrings) {
1943 					return (0);
1944 				}
1945 			}
1946 		}
1947 		/* Resetting the prop */
1948 		if (mrp->mrp_mask & MRP_RINGS_RESET) {
1949 			if (group != defgrp) {
1950 				if (mip->mi_tx_group_type ==
1951 				    MAC_GROUP_TYPE_DYNAMIC) {
1952 					ringcnt = group->mrg_cur_count;
1953 					if ((flent->fe_type &
1954 					    FLOW_PRIMARY_MAC) != 0) {
1955 						mac_tx_client_quiesce(
1956 						    (mac_client_handle_t)
1957 						    mcip);
1958 						mac_tx_switch_group(mcip,
1959 						    group, defgrp);
1960 						mac_tx_client_restart(
1961 						    (mac_client_handle_t)
1962 						    mcip);
1963 						MAC_TX_GRP_RELEASED(mip);
1964 						MAC_TX_RING_RELEASED(mip,
1965 						    ringcnt);
1966 						return (0);
1967 					}
1968 					cmrp->mrp_ntxrings = 1;
1969 					(void) mac_group_ring_modify(mcip,
1970 					    group, defgrp);
1971 					/*
1972 					 * This group has reserved rings
1973 					 * that need to be released now.
1974 					 */
1975 					MAC_TX_RING_RELEASED(mip, ringcnt);
1976 				}
1977 				/*
1978 				 * If this is a static group, we
1979 				 * need to release the group. The
1980 				 * client will remain in the same
1981 				 * group till some other client
1982 				 * needs this group.
1983 				 */
1984 				MAC_TX_GRP_RELEASED(mip);
1985 			} else if (group == defgrp &&
1986 			    (flent->fe_type & FLOW_PRIMARY_MAC) == 0) {
1987 				ngrp = mac_reserve_tx_group(mcip, B_TRUE);
1988 				if (ngrp == NULL)
1989 					return (0);
1990 				mac_tx_client_quiesce(
1991 				    (mac_client_handle_t)mcip);
1992 				mac_tx_switch_group(mcip, defgrp, ngrp);
1993 				mac_tx_client_restart(
1994 				    (mac_client_handle_t)mcip);
1995 			}
1996 			/*
1997 			 * If the client is in the default group, we will
1998 			 * just clear the MRP_TX_RINGS and leave it as
1999 			 * it rather than look for an exclusive group
2000 			 * for it.
2001 			 */
2002 			return (0);
2003 		}
2004 
2005 		/* Switch to H/W */
2006 		if (group == defgrp && ((mrp->mrp_ntxrings > 0) || unspec)) {
2007 			ngrp = mac_reserve_tx_group(mcip, B_TRUE);
2008 			if (ngrp == NULL)
2009 				return (ENOSPC);
2010 			mac_tx_client_quiesce((mac_client_handle_t)mcip);
2011 			mac_tx_switch_group(mcip, defgrp, ngrp);
2012 			mac_tx_client_restart((mac_client_handle_t)mcip);
2013 			MAC_TX_GRP_RESERVED(mip);
2014 			if (mip->mi_tx_group_type == MAC_GROUP_TYPE_DYNAMIC)
2015 				MAC_TX_RING_RESERVED(mip, ngrp->mrg_cur_count);
2016 		/* Switch to S/W */
2017 		} else if (group != defgrp && !unspec &&
2018 		    mrp->mrp_ntxrings == 0) {
2019 			/* Switch to S/W */
2020 			ringcnt = group->mrg_cur_count;
2021 			mac_tx_client_quiesce((mac_client_handle_t)mcip);
2022 			mac_tx_switch_group(mcip, group, defgrp);
2023 			mac_tx_client_restart((mac_client_handle_t)mcip);
2024 			if (tmrp->mrp_mask & MRP_TX_RINGS) {
2025 				MAC_TX_GRP_RELEASED(mip);
2026 				if (mip->mi_tx_group_type ==
2027 				    MAC_GROUP_TYPE_DYNAMIC) {
2028 					MAC_TX_RING_RELEASED(mip, ringcnt);
2029 				}
2030 			}
2031 		} else if (group != defgrp && mip->mi_tx_group_type ==
2032 		    MAC_GROUP_TYPE_DYNAMIC) {
2033 			ringcnt = group->mrg_cur_count;
2034 			err = mac_group_ring_modify(mcip, group, defgrp);
2035 			if (err != 0)
2036 				return (err);
2037 			/*
2038 			 * Update the accounting. If this group
2039 			 * already had explicitly reserved rings,
2040 			 * we need to update the rings based on
2041 			 * the new ring count. If this group
2042 			 * had not explicitly reserved rings,
2043 			 * then we just reserve the rings asked for
2044 			 * and reserve the group.
2045 			 */
2046 			if (tmrp->mrp_mask & MRP_TX_RINGS) {
2047 				if (ringcnt > group->mrg_cur_count) {
2048 					MAC_TX_RING_RELEASED(mip,
2049 					    ringcnt - group->mrg_cur_count);
2050 				} else {
2051 					MAC_TX_RING_RESERVED(mip,
2052 					    group->mrg_cur_count - ringcnt);
2053 				}
2054 			} else {
2055 				MAC_TX_RING_RESERVED(mip, group->mrg_cur_count);
2056 				MAC_TX_GRP_RESERVED(mip);
2057 			}
2058 		}
2059 	}
2060 	return (0);
2061 }
2062 
2063 /*
2064  * When the MAC client is being brought up (i.e. we do a unicast_add) we need
2065  * to initialize the cpu and resource control structure in the
2066  * mac_client_impl_t from the mac_impl_t (i.e if there are any cached
2067  * properties before the flow entry for the unicast address was created).
2068  */
2069 static int
2070 mac_resource_ctl_set(mac_client_handle_t mch, mac_resource_props_t *mrp)
2071 {
2072 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2073 	mac_impl_t		*mip = (mac_impl_t *)mcip->mci_mip;
2074 	mac_impl_t		*umip = mcip->mci_upper_mip;
2075 	int			err = 0;
2076 	flow_entry_t		*flent = mcip->mci_flent;
2077 	mac_resource_props_t	*omrp, *nmrp = MCIP_RESOURCE_PROPS(mcip);
2078 
2079 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
2080 
2081 	err = mac_validate_props(mcip->mci_state_flags & MCIS_IS_VNIC ?
2082 	    mcip->mci_upper_mip : mip, mrp);
2083 	if (err != 0)
2084 		return (err);
2085 
2086 	/*
2087 	 * Copy over the existing properties since mac_update_resources
2088 	 * will modify the client's mrp. Currently, the saved property
2089 	 * is used to determine the difference between existing and
2090 	 * modified rings property.
2091 	 */
2092 	omrp = kmem_zalloc(sizeof (*omrp), KM_SLEEP);
2093 	bcopy(nmrp, omrp, sizeof (*omrp));
2094 	mac_update_resources(mrp, MCIP_RESOURCE_PROPS(mcip), B_FALSE);
2095 	if (MCIP_DATAPATH_SETUP(mcip)) {
2096 		/*
2097 		 * We support rings only for primary client when there are
2098 		 * multiple clients sharing the same MAC address (e.g. VLAN).
2099 		 */
2100 		if (mrp->mrp_mask & MRP_RX_RINGS ||
2101 		    mrp->mrp_mask & MRP_TX_RINGS) {
2102 
2103 			if ((err = mac_client_set_rings_prop(mcip, mrp,
2104 			    omrp)) != 0) {
2105 				if (omrp->mrp_mask & MRP_RX_RINGS) {
2106 					nmrp->mrp_mask |= MRP_RX_RINGS;
2107 					nmrp->mrp_nrxrings = omrp->mrp_nrxrings;
2108 				} else {
2109 					nmrp->mrp_mask &= ~MRP_RX_RINGS;
2110 					nmrp->mrp_nrxrings = 0;
2111 				}
2112 				if (omrp->mrp_mask & MRP_TX_RINGS) {
2113 					nmrp->mrp_mask |= MRP_TX_RINGS;
2114 					nmrp->mrp_ntxrings = omrp->mrp_ntxrings;
2115 				} else {
2116 					nmrp->mrp_mask &= ~MRP_TX_RINGS;
2117 					nmrp->mrp_ntxrings = 0;
2118 				}
2119 				if (omrp->mrp_mask & MRP_RXRINGS_UNSPEC)
2120 					omrp->mrp_mask |= MRP_RXRINGS_UNSPEC;
2121 				else
2122 					omrp->mrp_mask &= ~MRP_RXRINGS_UNSPEC;
2123 
2124 				if (omrp->mrp_mask & MRP_TXRINGS_UNSPEC)
2125 					omrp->mrp_mask |= MRP_TXRINGS_UNSPEC;
2126 				else
2127 					omrp->mrp_mask &= ~MRP_TXRINGS_UNSPEC;
2128 				kmem_free(omrp, sizeof (*omrp));
2129 				return (err);
2130 			}
2131 
2132 			/*
2133 			 * If we modified the rings property of the primary
2134 			 * we need to update the property fields of its
2135 			 * VLANs as they inherit the primary's properites.
2136 			 */
2137 			if (mac_is_primary_client(mcip)) {
2138 				mac_set_prim_vlan_rings(mip,
2139 				    MCIP_RESOURCE_PROPS(mcip));
2140 			}
2141 		}
2142 		/*
2143 		 * We have to set this prior to calling mac_flow_modify.
2144 		 */
2145 		if (mrp->mrp_mask & MRP_PRIORITY) {
2146 			if (mrp->mrp_priority == MPL_RESET) {
2147 				MAC_CLIENT_SET_PRIORITY_RANGE(mcip,
2148 				    MPL_LINK_DEFAULT);
2149 			} else {
2150 				MAC_CLIENT_SET_PRIORITY_RANGE(mcip,
2151 				    mrp->mrp_priority);
2152 			}
2153 		}
2154 
2155 		mac_flow_modify(mip->mi_flow_tab, flent, mrp);
2156 		if (mrp->mrp_mask & MRP_PRIORITY)
2157 			mac_update_subflow_priority(mcip);
2158 
2159 		/* Apply these resource settings to any secondary macs */
2160 		if (umip != NULL) {
2161 			ASSERT((umip->mi_state_flags & MIS_IS_VNIC) != 0);
2162 			mac_vnic_secondary_update(umip);
2163 		}
2164 	}
2165 	kmem_free(omrp, sizeof (*omrp));
2166 	return (0);
2167 }
2168 
2169 static int
2170 mac_unicast_flow_create(mac_client_impl_t *mcip, uint8_t *mac_addr,
2171     uint16_t vid, boolean_t is_primary, boolean_t first_flow,
2172     flow_entry_t **flent, mac_resource_props_t *mrp)
2173 {
2174 	mac_impl_t	*mip = (mac_impl_t *)mcip->mci_mip;
2175 	flow_desc_t	flow_desc;
2176 	char		flowname[MAXFLOWNAMELEN];
2177 	int		err;
2178 	uint_t		flent_flags;
2179 
2180 	/*
2181 	 * First unicast address being added, create a new flow
2182 	 * for that MAC client.
2183 	 */
2184 	bzero(&flow_desc, sizeof (flow_desc));
2185 
2186 	ASSERT(mac_addr != NULL ||
2187 	    (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR));
2188 	if (mac_addr != NULL) {
2189 		flow_desc.fd_mac_len = mip->mi_type->mt_addr_length;
2190 		bcopy(mac_addr, flow_desc.fd_dst_mac, flow_desc.fd_mac_len);
2191 	}
2192 	flow_desc.fd_mask = FLOW_LINK_DST;
2193 	if (vid != 0) {
2194 		flow_desc.fd_vid = vid;
2195 		flow_desc.fd_mask |= FLOW_LINK_VID;
2196 	}
2197 
2198 	/*
2199 	 * XXX-nicolas. For now I'm keeping the FLOW_PRIMARY_MAC
2200 	 * and FLOW_VNIC. Even though they're a hack inherited
2201 	 * from the SRS code, we'll keep them for now. They're currently
2202 	 * consumed by mac_datapath_setup() to create the SRS.
2203 	 * That code should be eventually moved out of
2204 	 * mac_datapath_setup() and moved to a mac_srs_create()
2205 	 * function of some sort to keep things clean.
2206 	 *
2207 	 * Also, there's no reason why the SRS for the primary MAC
2208 	 * client should be different than any other MAC client. Until
2209 	 * this is cleaned-up, we support only one MAC unicast address
2210 	 * per client.
2211 	 *
2212 	 * We set FLOW_PRIMARY_MAC for the primary MAC address,
2213 	 * FLOW_VNIC for everything else.
2214 	 */
2215 	if (is_primary)
2216 		flent_flags = FLOW_PRIMARY_MAC;
2217 	else
2218 		flent_flags = FLOW_VNIC_MAC;
2219 
2220 	/*
2221 	 * For the first flow we use the MAC client's name - mci_name, for
2222 	 * subsequent ones we just create a name with the VID. This is
2223 	 * so that we can add these flows to the same flow table. This is
2224 	 * fine as the flow name (except for the one with the MAC client's
2225 	 * name) is not visible. When the first flow is removed, we just replace
2226 	 * its fdesc with another from the list, so we will still retain the
2227 	 * flent with the MAC client's flow name.
2228 	 */
2229 	if (first_flow) {
2230 		bcopy(mcip->mci_name, flowname, MAXFLOWNAMELEN);
2231 	} else {
2232 		(void) sprintf(flowname, "%s%u", mcip->mci_name, vid);
2233 		flent_flags = FLOW_NO_STATS;
2234 	}
2235 
2236 	if ((err = mac_flow_create(&flow_desc, mrp, flowname, NULL,
2237 	    flent_flags, flent)) != 0)
2238 		return (err);
2239 
2240 	mac_misc_stat_create(*flent);
2241 	FLOW_MARK(*flent, FE_INCIPIENT);
2242 	(*flent)->fe_mcip = mcip;
2243 
2244 	/*
2245 	 * Place initial creation reference on the flow. This reference
2246 	 * is released in the corresponding delete action viz.
2247 	 * mac_unicast_remove after waiting for all transient refs to
2248 	 * to go away. The wait happens in mac_flow_wait.
2249 	 * We have already held the reference in mac_client_open().
2250 	 */
2251 	if (!first_flow)
2252 		FLOW_REFHOLD(*flent);
2253 	return (0);
2254 }
2255 
2256 /* Refresh the multicast grouping for this VID. */
2257 int
2258 mac_client_update_mcast(void *arg, boolean_t add, const uint8_t *addrp)
2259 {
2260 	flow_entry_t		*flent = arg;
2261 	mac_client_impl_t	*mcip = flent->fe_mcip;
2262 	uint16_t		vid;
2263 	flow_desc_t		flow_desc;
2264 
2265 	mac_flow_get_desc(flent, &flow_desc);
2266 	vid = (flow_desc.fd_mask & FLOW_LINK_VID) != 0 ?
2267 	    flow_desc.fd_vid : VLAN_ID_NONE;
2268 
2269 	/*
2270 	 * We don't call mac_multicast_add()/mac_multicast_remove() as
2271 	 * we want to add/remove for this specific vid.
2272 	 */
2273 	if (add) {
2274 		return (mac_bcast_add(mcip, addrp, vid,
2275 		    MAC_ADDRTYPE_MULTICAST));
2276 	} else {
2277 		mac_bcast_delete(mcip, addrp, vid);
2278 		return (0);
2279 	}
2280 }
2281 
2282 static void
2283 mac_update_single_active_client(mac_impl_t *mip)
2284 {
2285 	mac_client_impl_t *client = NULL;
2286 
2287 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
2288 
2289 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
2290 	if (mip->mi_nactiveclients == 1) {
2291 		/*
2292 		 * Find the one active MAC client from the list of MAC
2293 		 * clients. The active MAC client has at least one
2294 		 * unicast address.
2295 		 */
2296 		for (client = mip->mi_clients_list; client != NULL;
2297 		    client = client->mci_client_next) {
2298 			if (client->mci_unicast_list != NULL)
2299 				break;
2300 		}
2301 		ASSERT(client != NULL);
2302 	}
2303 
2304 	/*
2305 	 * mi_single_active_client is protected by the MAC impl's read/writer
2306 	 * lock, which allows mac_rx() to check the value of that pointer
2307 	 * as a reader.
2308 	 */
2309 	mip->mi_single_active_client = client;
2310 	rw_exit(&mip->mi_rw_lock);
2311 }
2312 
2313 /*
2314  * Set up the data path. Called from i_mac_unicast_add after having
2315  * done all the validations including making sure this is an active
2316  * client (i.e that is ready to process packets.)
2317  */
2318 static int
2319 mac_client_datapath_setup(mac_client_impl_t *mcip, uint16_t vid,
2320     uint8_t *mac_addr, mac_resource_props_t *mrp, boolean_t isprimary,
2321     mac_unicast_impl_t *muip)
2322 {
2323 	mac_impl_t	*mip = mcip->mci_mip;
2324 	boolean_t	mac_started = B_FALSE;
2325 	boolean_t	bcast_added = B_FALSE;
2326 	boolean_t	nactiveclients_added = B_FALSE;
2327 	flow_entry_t	*flent;
2328 	int		err = 0;
2329 	boolean_t	no_unicast;
2330 
2331 	no_unicast = mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR;
2332 
2333 	if ((err = mac_start((mac_handle_t)mip)) != 0)
2334 		goto bail;
2335 
2336 	mac_started = B_TRUE;
2337 
2338 	/* add the MAC client to the broadcast address group by default */
2339 	if (mip->mi_type->mt_brdcst_addr != NULL) {
2340 		err = mac_bcast_add(mcip, mip->mi_type->mt_brdcst_addr, vid,
2341 		    MAC_ADDRTYPE_BROADCAST);
2342 		if (err != 0)
2343 			goto bail;
2344 		bcast_added = B_TRUE;
2345 	}
2346 
2347 	/*
2348 	 * If this is the first unicast address addition for this
2349 	 * client, reuse the pre-allocated larval flow entry associated with
2350 	 * the MAC client.
2351 	 */
2352 	flent = (mcip->mci_nflents == 0) ? mcip->mci_flent : NULL;
2353 
2354 	/* We are configuring the unicast flow now */
2355 	if (!MCIP_DATAPATH_SETUP(mcip)) {
2356 
2357 		if (mrp != NULL) {
2358 			MAC_CLIENT_SET_PRIORITY_RANGE(mcip,
2359 			    (mrp->mrp_mask & MRP_PRIORITY) ? mrp->mrp_priority :
2360 			    MPL_LINK_DEFAULT);
2361 		}
2362 		if ((err = mac_unicast_flow_create(mcip, mac_addr, vid,
2363 		    isprimary, B_TRUE, &flent, mrp)) != 0)
2364 			goto bail;
2365 
2366 		mip->mi_nactiveclients++;
2367 		nactiveclients_added = B_TRUE;
2368 
2369 		/*
2370 		 * This will allocate the RX ring group if possible for the
2371 		 * flow and program the software classifier as needed.
2372 		 */
2373 		if ((err = mac_datapath_setup(mcip, flent, SRST_LINK)) != 0)
2374 			goto bail;
2375 
2376 		if (no_unicast)
2377 			goto done_setup;
2378 		/*
2379 		 * The unicast MAC address must have been added successfully.
2380 		 */
2381 		ASSERT(mcip->mci_unicast != NULL);
2382 
2383 		/*
2384 		 * Push down the sub-flows that were defined on this link
2385 		 * hitherto. The flows are added to the active flow table
2386 		 * and SRS, softrings etc. are created as needed.
2387 		 */
2388 		mac_link_init_flows((mac_client_handle_t)mcip);
2389 	} else {
2390 		mac_address_t *map = mcip->mci_unicast;
2391 
2392 		ASSERT(!no_unicast);
2393 		/*
2394 		 * A unicast flow already exists for that MAC client
2395 		 * so this flow must be the same MAC address but with
2396 		 * a different VID. It has been checked by
2397 		 * mac_addr_in_use().
2398 		 *
2399 		 * We will use the SRS etc. from the initial
2400 		 * mci_flent. We don't need to create a kstat for
2401 		 * this, as except for the fdesc, everything will be
2402 		 * used from the first flent.
2403 		 *
2404 		 * The only time we should see multiple flents on the
2405 		 * same MAC client is on the sun4v vsw. If we removed
2406 		 * that code we should be able to remove the entire
2407 		 * notion of multiple flents on a MAC client (this
2408 		 * doesn't affect sub/user flows because they have
2409 		 * their own list unrelated to mci_flent_list).
2410 		 */
2411 		if (bcmp(mac_addr, map->ma_addr, map->ma_len) != 0) {
2412 			err = EINVAL;
2413 			goto bail;
2414 		}
2415 
2416 		if ((err = mac_unicast_flow_create(mcip, mac_addr, vid,
2417 		    isprimary, B_FALSE, &flent, NULL)) != 0) {
2418 			goto bail;
2419 		}
2420 		if ((err = mac_flow_add(mip->mi_flow_tab, flent)) != 0) {
2421 			FLOW_FINAL_REFRELE(flent);
2422 			goto bail;
2423 		}
2424 
2425 		/* update the multicast group for this vid */
2426 		mac_client_bcast_refresh(mcip, mac_client_update_mcast,
2427 		    (void *)flent, B_TRUE);
2428 
2429 	}
2430 
2431 	/* populate the shared MAC address */
2432 	muip->mui_map = mcip->mci_unicast;
2433 
2434 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
2435 	muip->mui_next = mcip->mci_unicast_list;
2436 	mcip->mci_unicast_list = muip;
2437 	rw_exit(&mcip->mci_rw_lock);
2438 
2439 done_setup:
2440 	/*
2441 	 * First add the flent to the flow list of this mcip. Then set
2442 	 * the mip's mi_single_active_client if needed. The Rx path assumes
2443 	 * that mip->mi_single_active_client will always have an associated
2444 	 * flent.
2445 	 */
2446 	mac_client_add_to_flow_list(mcip, flent);
2447 	if (nactiveclients_added)
2448 		mac_update_single_active_client(mip);
2449 	/*
2450 	 * Trigger a renegotiation of the capabilities when the number of
2451 	 * active clients changes from 1 to 2, since some of the capabilities
2452 	 * might have to be disabled. Also send a MAC_NOTE_LINK notification
2453 	 * to all the MAC clients whenever physical link is DOWN.
2454 	 */
2455 	if (mip->mi_nactiveclients == 2) {
2456 		mac_capab_update((mac_handle_t)mip);
2457 		mac_virtual_link_update(mip);
2458 	}
2459 	/*
2460 	 * Now that the setup is complete, clear the INCIPIENT flag.
2461 	 * The flag was set to avoid incoming packets seeing inconsistent
2462 	 * structures while the setup was in progress. Clear the mci_tx_flag
2463 	 * by calling mac_tx_client_block. It is possible that
2464 	 * mac_unicast_remove was called prior to this mac_unicast_add which
2465 	 * could have set the MCI_TX_QUIESCE flag.
2466 	 */
2467 	if (flent->fe_rx_ring_group != NULL)
2468 		mac_rx_group_unmark(flent->fe_rx_ring_group, MR_INCIPIENT);
2469 	FLOW_UNMARK(flent, FE_INCIPIENT);
2470 
2471 	/*
2472 	 * If this is an aggr port client, don't enable the flow's
2473 	 * datapath at this stage. Otherwise, bcast traffic could
2474 	 * arrive while the aggr port is in the process of
2475 	 * initializing. Instead, the flow's datapath is started later
2476 	 * when mac_client_set_flow_cb() is called.
2477 	 */
2478 	if ((mcip->mci_state_flags & MCIS_IS_AGGR_PORT) == 0)
2479 		FLOW_UNMARK(flent, FE_MC_NO_DATAPATH);
2480 
2481 	mac_tx_client_unblock(mcip);
2482 	return (0);
2483 bail:
2484 	if (bcast_added)
2485 		mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr, vid);
2486 
2487 	if (nactiveclients_added)
2488 		mip->mi_nactiveclients--;
2489 
2490 	if (mac_started)
2491 		mac_stop((mac_handle_t)mip);
2492 
2493 	return (err);
2494 }
2495 
2496 /*
2497  * Return the passive primary MAC client, if present. The passive client is
2498  * a stand-by client that has the same unicast address as another that is
2499  * currenly active. Once the active client goes away, the passive client
2500  * becomes active.
2501  */
2502 static mac_client_impl_t *
2503 mac_get_passive_primary_client(mac_impl_t *mip)
2504 {
2505 	mac_client_impl_t	*mcip;
2506 
2507 	for (mcip = mip->mi_clients_list; mcip != NULL;
2508 	    mcip = mcip->mci_client_next) {
2509 		if (mac_is_primary_client(mcip) &&
2510 		    (mcip->mci_flags & MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
2511 			return (mcip);
2512 		}
2513 	}
2514 	return (NULL);
2515 }
2516 
2517 /*
2518  * Add a new unicast address to the MAC client.
2519  *
2520  * The MAC address can be specified either by value, or the MAC client
2521  * can specify that it wants to use the primary MAC address of the
2522  * underlying MAC. See the introductory comments at the beginning
2523  * of this file for more more information on primary MAC addresses.
2524  *
2525  * Note also the tuple (MAC address, VID) must be unique
2526  * for the MAC clients defined on top of the same underlying MAC
2527  * instance, unless the MAC_UNICAST_NODUPCHECK is specified.
2528  *
2529  * In no case can a client use the PVID for the MAC, if the MAC has one set.
2530  */
2531 int
2532 i_mac_unicast_add(mac_client_handle_t mch, uint8_t *mac_addr, uint16_t flags,
2533     mac_unicast_handle_t *mah, uint16_t vid, mac_diag_t *diag)
2534 {
2535 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2536 	mac_impl_t		*mip = mcip->mci_mip;
2537 	int			err;
2538 	uint_t			mac_len = mip->mi_type->mt_addr_length;
2539 	boolean_t		check_dups = !(flags & MAC_UNICAST_NODUPCHECK);
2540 	boolean_t		fastpath_disabled = B_FALSE;
2541 	boolean_t		is_primary = (flags & MAC_UNICAST_PRIMARY);
2542 	boolean_t		is_unicast_hw = (flags & MAC_UNICAST_HW);
2543 	mac_resource_props_t	*mrp;
2544 	boolean_t		passive_client = B_FALSE;
2545 	mac_unicast_impl_t	*muip;
2546 	boolean_t		is_vnic_primary =
2547 	    (flags & MAC_UNICAST_VNIC_PRIMARY);
2548 
2549 	/*
2550 	 * When the VID is non-zero the underlying MAC cannot be a
2551 	 * VNIC. I.e., dladm create-vlan cannot take a VNIC as
2552 	 * argument, only the primary MAC client.
2553 	 */
2554 	ASSERT(!((mip->mi_state_flags & MIS_IS_VNIC) && (vid != VLAN_ID_NONE)));
2555 
2556 	*diag = MAC_DIAG_NONE;
2557 
2558 	/*
2559 	 * Can't unicast add if the client asked only for minimal datapath
2560 	 * setup.
2561 	 */
2562 	if (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR)
2563 		return (ENOTSUP);
2564 
2565 	/*
2566 	 * Check for an attempted use of the current Port VLAN ID, if enabled.
2567 	 * No client may use it.
2568 	 */
2569 	if (mip->mi_pvid != VLAN_ID_NONE && vid == mip->mi_pvid)
2570 		return (EBUSY);
2571 
2572 	/*
2573 	 * Check whether it's the primary client and flag it.
2574 	 */
2575 	if (!(mcip->mci_state_flags & MCIS_IS_VNIC) && is_primary &&
2576 	    vid == VLAN_ID_NONE)
2577 		mcip->mci_flags |= MAC_CLIENT_FLAGS_PRIMARY;
2578 
2579 	/*
2580 	 * is_vnic_primary is true when we come here as a VLAN VNIC
2581 	 * which uses the primary MAC client's address but with a non-zero
2582 	 * VID. In this case the MAC address is not specified by an upper
2583 	 * MAC client.
2584 	 */
2585 	if ((mcip->mci_state_flags & MCIS_IS_VNIC) && is_primary &&
2586 	    !is_vnic_primary) {
2587 		/*
2588 		 * The address is being set by the upper MAC client
2589 		 * of a VNIC. The MAC address was already set by the
2590 		 * VNIC driver during VNIC creation.
2591 		 *
2592 		 * Note: a VNIC has only one MAC address. We return
2593 		 * the MAC unicast address handle of the lower MAC client
2594 		 * corresponding to the VNIC. We allocate a new entry
2595 		 * which is flagged appropriately, so that mac_unicast_remove()
2596 		 * doesn't attempt to free the original entry that
2597 		 * was allocated by the VNIC driver.
2598 		 */
2599 		ASSERT(mcip->mci_unicast != NULL);
2600 
2601 		/* Check for VLAN flags, if present */
2602 		if ((flags & MAC_UNICAST_TAG_DISABLE) != 0)
2603 			mcip->mci_state_flags |= MCIS_TAG_DISABLE;
2604 
2605 		if ((flags & MAC_UNICAST_STRIP_DISABLE) != 0)
2606 			mcip->mci_state_flags |= MCIS_STRIP_DISABLE;
2607 
2608 		if ((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) != 0)
2609 			mcip->mci_state_flags |= MCIS_DISABLE_TX_VID_CHECK;
2610 
2611 		/*
2612 		 * Ensure that the primary unicast address of the VNIC
2613 		 * is added only once unless we have the
2614 		 * MAC_CLIENT_FLAGS_MULTI_PRIMARY set (and this is not
2615 		 * a passive MAC client).
2616 		 */
2617 		if ((mcip->mci_flags & MAC_CLIENT_FLAGS_VNIC_PRIMARY) != 0) {
2618 			if ((mcip->mci_flags &
2619 			    MAC_CLIENT_FLAGS_MULTI_PRIMARY) == 0 ||
2620 			    (mcip->mci_flags &
2621 			    MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
2622 				return (EBUSY);
2623 			}
2624 			mcip->mci_flags |= MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
2625 			passive_client = B_TRUE;
2626 		}
2627 
2628 		mcip->mci_flags |= MAC_CLIENT_FLAGS_VNIC_PRIMARY;
2629 
2630 		/*
2631 		 * Create a handle for vid 0.
2632 		 */
2633 		ASSERT(vid == VLAN_ID_NONE);
2634 		muip = kmem_zalloc(sizeof (mac_unicast_impl_t), KM_SLEEP);
2635 		muip->mui_vid = vid;
2636 		*mah = (mac_unicast_handle_t)muip;
2637 		/*
2638 		 * This will be used by the caller to defer setting the
2639 		 * rx functions.
2640 		 */
2641 		if (passive_client)
2642 			return (EAGAIN);
2643 		return (0);
2644 	}
2645 
2646 	/* primary MAC clients cannot be opened on top of anchor VNICs */
2647 	if ((is_vnic_primary || is_primary) &&
2648 	    i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_ANCHOR_VNIC, NULL)) {
2649 		return (ENXIO);
2650 	}
2651 
2652 	/*
2653 	 * If this is a VNIC/VLAN, disable softmac fast-path. This is
2654 	 * only relevant to legacy devices which use softmac to
2655 	 * interface with GLDv3.
2656 	 */
2657 	if (mcip->mci_state_flags & MCIS_IS_VNIC) {
2658 		err = mac_fastpath_disable((mac_handle_t)mip);
2659 		if (err != 0)
2660 			return (err);
2661 		fastpath_disabled = B_TRUE;
2662 	}
2663 
2664 	/*
2665 	 * Return EBUSY if:
2666 	 *  - there is an exclusively active mac client exists.
2667 	 *  - this is an exclusive active mac client but
2668 	 *	a. there is already active mac clients exist, or
2669 	 *	b. fastpath streams are already plumbed on this legacy device
2670 	 *  - the mac creator has disallowed active mac clients.
2671 	 */
2672 	if (mip->mi_state_flags & (MIS_EXCLUSIVE|MIS_NO_ACTIVE)) {
2673 		if (fastpath_disabled)
2674 			mac_fastpath_enable((mac_handle_t)mip);
2675 		return (EBUSY);
2676 	}
2677 
2678 	if (mcip->mci_state_flags & MCIS_EXCLUSIVE) {
2679 		ASSERT(!fastpath_disabled);
2680 		if (mip->mi_nactiveclients != 0)
2681 			return (EBUSY);
2682 
2683 		if ((mip->mi_state_flags & MIS_LEGACY) &&
2684 		    !(mip->mi_capab_legacy.ml_active_set(mip->mi_driver))) {
2685 			return (EBUSY);
2686 		}
2687 		mip->mi_state_flags |= MIS_EXCLUSIVE;
2688 	}
2689 
2690 	mrp = kmem_zalloc(sizeof (*mrp), KM_SLEEP);
2691 	if (is_primary && !(mcip->mci_state_flags & (MCIS_IS_VNIC |
2692 	    MCIS_IS_AGGR_PORT))) {
2693 		/*
2694 		 * Apply the property cached in the mac_impl_t to the primary
2695 		 * mac client. If the mac client is a VNIC or an aggregation
2696 		 * port, its property should be set in the mcip when the
2697 		 * VNIC/aggr was created.
2698 		 */
2699 		mac_get_resources((mac_handle_t)mip, mrp);
2700 		(void) mac_client_set_resources(mch, mrp);
2701 	} else if (mcip->mci_state_flags & MCIS_IS_VNIC) {
2702 		/*
2703 		 * This is a VLAN client sharing the address of the
2704 		 * primary MAC client; i.e., one created via dladm
2705 		 * create-vlan. We don't support specifying ring
2706 		 * properties for this type of client as it inherits
2707 		 * these from the primary MAC client.
2708 		 */
2709 		if (is_vnic_primary) {
2710 			mac_resource_props_t	*vmrp;
2711 
2712 			vmrp = MCIP_RESOURCE_PROPS(mcip);
2713 			if (vmrp->mrp_mask & MRP_RX_RINGS ||
2714 			    vmrp->mrp_mask & MRP_TX_RINGS) {
2715 				if (fastpath_disabled)
2716 					mac_fastpath_enable((mac_handle_t)mip);
2717 				kmem_free(mrp, sizeof (*mrp));
2718 				return (ENOTSUP);
2719 			}
2720 			/*
2721 			 * Additionally we also need to inherit any
2722 			 * rings property from the MAC.
2723 			 */
2724 			mac_get_resources((mac_handle_t)mip, mrp);
2725 			if (mrp->mrp_mask & MRP_RX_RINGS) {
2726 				vmrp->mrp_mask |= MRP_RX_RINGS;
2727 				vmrp->mrp_nrxrings = mrp->mrp_nrxrings;
2728 			}
2729 			if (mrp->mrp_mask & MRP_TX_RINGS) {
2730 				vmrp->mrp_mask |= MRP_TX_RINGS;
2731 				vmrp->mrp_ntxrings = mrp->mrp_ntxrings;
2732 			}
2733 		}
2734 		bcopy(MCIP_RESOURCE_PROPS(mcip), mrp, sizeof (*mrp));
2735 	}
2736 
2737 	muip = kmem_zalloc(sizeof (mac_unicast_impl_t), KM_SLEEP);
2738 	muip->mui_vid = vid;
2739 
2740 	if (is_primary || is_vnic_primary) {
2741 		mac_addr = mip->mi_addr;
2742 	} else {
2743 
2744 		/*
2745 		 * Verify the validity of the specified MAC addresses value.
2746 		 */
2747 		if (!mac_unicst_verify((mac_handle_t)mip, mac_addr, mac_len)) {
2748 			*diag = MAC_DIAG_MACADDR_INVALID;
2749 			err = EINVAL;
2750 			goto bail_out;
2751 		}
2752 
2753 		/*
2754 		 * Make sure that the specified MAC address is different
2755 		 * than the unicast MAC address of the underlying NIC.
2756 		 */
2757 		if (check_dups && bcmp(mip->mi_addr, mac_addr, mac_len) == 0) {
2758 			*diag = MAC_DIAG_MACADDR_NIC;
2759 			err = EINVAL;
2760 			goto bail_out;
2761 		}
2762 	}
2763 
2764 	/*
2765 	 * Set the flags here so that if this is a passive client, we
2766 	 * can return and set it when we call mac_client_datapath_setup
2767 	 * when this becomes the active client. If we defer to using these
2768 	 * flags to mac_client_datapath_setup, then for a passive client,
2769 	 * we'd have to store the flags somewhere (probably fe_flags)
2770 	 * and then use it.
2771 	 */
2772 	if (!MCIP_DATAPATH_SETUP(mcip)) {
2773 		if (is_unicast_hw) {
2774 			/*
2775 			 * The client requires a hardware MAC address slot
2776 			 * for that unicast address. Since we support only
2777 			 * one unicast MAC address per client, flag the
2778 			 * MAC client itself.
2779 			 */
2780 			mcip->mci_state_flags |= MCIS_UNICAST_HW;
2781 		}
2782 
2783 		/* Check for VLAN flags, if present */
2784 		if ((flags & MAC_UNICAST_TAG_DISABLE) != 0)
2785 			mcip->mci_state_flags |= MCIS_TAG_DISABLE;
2786 
2787 		if ((flags & MAC_UNICAST_STRIP_DISABLE) != 0)
2788 			mcip->mci_state_flags |= MCIS_STRIP_DISABLE;
2789 
2790 		if ((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) != 0)
2791 			mcip->mci_state_flags |= MCIS_DISABLE_TX_VID_CHECK;
2792 	} else {
2793 		/*
2794 		 * Assert that the specified flags are consistent with the
2795 		 * flags specified by previous calls to mac_unicast_add().
2796 		 */
2797 		ASSERT(((flags & MAC_UNICAST_TAG_DISABLE) != 0 &&
2798 		    (mcip->mci_state_flags & MCIS_TAG_DISABLE) != 0) ||
2799 		    ((flags & MAC_UNICAST_TAG_DISABLE) == 0 &&
2800 		    (mcip->mci_state_flags & MCIS_TAG_DISABLE) == 0));
2801 
2802 		ASSERT(((flags & MAC_UNICAST_STRIP_DISABLE) != 0 &&
2803 		    (mcip->mci_state_flags & MCIS_STRIP_DISABLE) != 0) ||
2804 		    ((flags & MAC_UNICAST_STRIP_DISABLE) == 0 &&
2805 		    (mcip->mci_state_flags & MCIS_STRIP_DISABLE) == 0));
2806 
2807 		ASSERT(((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) != 0 &&
2808 		    (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK) != 0) ||
2809 		    ((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) == 0 &&
2810 		    (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK) == 0));
2811 
2812 		/*
2813 		 * Make sure the client is consistent about its requests
2814 		 * for MAC addresses. I.e. all requests from the clients
2815 		 * must have the MAC_UNICAST_HW flag set or clear.
2816 		 */
2817 		if (((mcip->mci_state_flags & MCIS_UNICAST_HW) != 0 &&
2818 		    !is_unicast_hw) ||
2819 		    ((mcip->mci_state_flags & MCIS_UNICAST_HW) == 0 &&
2820 		    is_unicast_hw)) {
2821 			err = EINVAL;
2822 			goto bail_out;
2823 		}
2824 	}
2825 	/*
2826 	 * Make sure the MAC address is not already used by
2827 	 * another MAC client defined on top of the same
2828 	 * underlying NIC. Unless we have MAC_CLIENT_FLAGS_MULTI_PRIMARY
2829 	 * set when we allow a passive client to be present which will
2830 	 * be activated when the currently active client goes away - this
2831 	 * works only with primary addresses.
2832 	 */
2833 	if ((check_dups || is_primary || is_vnic_primary) &&
2834 	    mac_addr_in_use(mip, mac_addr, vid)) {
2835 		/*
2836 		 * Must have set the multiple primary address flag when
2837 		 * we did a mac_client_open AND this should be a primary
2838 		 * MAC client AND there should not already be a passive
2839 		 * primary. If all is true then we let this succeed
2840 		 * even if the address is a dup.
2841 		 */
2842 		if ((mcip->mci_flags & MAC_CLIENT_FLAGS_MULTI_PRIMARY) == 0 ||
2843 		    (mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY) == 0 ||
2844 		    mac_get_passive_primary_client(mip) != NULL) {
2845 			*diag = MAC_DIAG_MACADDR_INUSE;
2846 			err = EEXIST;
2847 			goto bail_out;
2848 		}
2849 		ASSERT((mcip->mci_flags &
2850 		    MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) == 0);
2851 		mcip->mci_flags |= MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
2852 		kmem_free(mrp, sizeof (*mrp));
2853 
2854 		/*
2855 		 * Stash the unicast address handle, we will use it when
2856 		 * we set up the passive client.
2857 		 */
2858 		mcip->mci_p_unicast_list = muip;
2859 		*mah = (mac_unicast_handle_t)muip;
2860 		return (0);
2861 	}
2862 
2863 	err = mac_client_datapath_setup(mcip, vid, mac_addr, mrp,
2864 	    is_primary || is_vnic_primary, muip);
2865 	if (err != 0)
2866 		goto bail_out;
2867 
2868 	kmem_free(mrp, sizeof (*mrp));
2869 	*mah = (mac_unicast_handle_t)muip;
2870 	return (0);
2871 
2872 bail_out:
2873 	if (fastpath_disabled)
2874 		mac_fastpath_enable((mac_handle_t)mip);
2875 	if (mcip->mci_state_flags & MCIS_EXCLUSIVE) {
2876 		mip->mi_state_flags &= ~MIS_EXCLUSIVE;
2877 		if (mip->mi_state_flags & MIS_LEGACY) {
2878 			mip->mi_capab_legacy.ml_active_clear(
2879 			    mip->mi_driver);
2880 		}
2881 	}
2882 	kmem_free(mrp, sizeof (*mrp));
2883 	kmem_free(muip, sizeof (mac_unicast_impl_t));
2884 	return (err);
2885 }
2886 
2887 /*
2888  * Wrapper function to mac_unicast_add when we want to have the same mac
2889  * client open for two instances, one that is currently active and another
2890  * that will become active when the current one is removed. In this case
2891  * mac_unicast_add will return EGAIN and we will save the rx function and
2892  * arg which will be used when we activate the passive client in
2893  * mac_unicast_remove.
2894  */
2895 int
2896 mac_unicast_add_set_rx(mac_client_handle_t mch, uint8_t *mac_addr,
2897     uint16_t flags, mac_unicast_handle_t *mah,  uint16_t vid, mac_diag_t *diag,
2898     mac_rx_t rx_fn, void *arg)
2899 {
2900 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2901 	uint_t			err;
2902 
2903 	err = mac_unicast_add(mch, mac_addr, flags, mah, vid, diag);
2904 	if (err != 0 && err != EAGAIN)
2905 		return (err);
2906 	if (err == EAGAIN) {
2907 		if (rx_fn != NULL) {
2908 			mcip->mci_rx_p_fn = rx_fn;
2909 			mcip->mci_rx_p_arg = arg;
2910 		}
2911 		return (0);
2912 	}
2913 	if (rx_fn != NULL)
2914 		mac_rx_set(mch, rx_fn, arg);
2915 	return (err);
2916 }
2917 
2918 int
2919 mac_unicast_add(mac_client_handle_t mch, uint8_t *mac_addr, uint16_t flags,
2920     mac_unicast_handle_t *mah, uint16_t vid, mac_diag_t *diag)
2921 {
2922 	mac_impl_t *mip = ((mac_client_impl_t *)mch)->mci_mip;
2923 	uint_t err;
2924 
2925 	i_mac_perim_enter(mip);
2926 	err = i_mac_unicast_add(mch, mac_addr, flags, mah, vid, diag);
2927 	i_mac_perim_exit(mip);
2928 
2929 	return (err);
2930 }
2931 
2932 static void
2933 mac_client_datapath_teardown(mac_client_handle_t mch, mac_unicast_impl_t *muip,
2934     flow_entry_t *flent)
2935 {
2936 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2937 	mac_impl_t		*mip = mcip->mci_mip;
2938 	boolean_t		no_unicast;
2939 
2940 	/*
2941 	 * If we have not added a unicast address for this MAC client, just
2942 	 * teardown the datapath.
2943 	 */
2944 	no_unicast = mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR;
2945 
2946 	if (!no_unicast) {
2947 		/*
2948 		 * We would have initialized subflows etc. only if we brought
2949 		 * up the primary client and set the unicast unicast address
2950 		 * etc. Deactivate the flows. The flow entry will be removed
2951 		 * from the active flow tables, and the associated SRS,
2952 		 * softrings etc will be deleted. But the flow entry itself
2953 		 * won't be destroyed, instead it will continue to be archived
2954 		 * off the  the global flow hash list, for a possible future
2955 		 * activation when say IP is plumbed again.
2956 		 */
2957 		mac_link_release_flows(mch);
2958 	}
2959 	mip->mi_nactiveclients--;
2960 	mac_update_single_active_client(mip);
2961 
2962 	/* Tear down the data path */
2963 	mac_datapath_teardown(mcip, mcip->mci_flent, SRST_LINK);
2964 
2965 	/*
2966 	 * Prevent any future access to the flow entry through the mci_flent
2967 	 * pointer by setting the mci_flent to NULL. Access to mci_flent in
2968 	 * mac_bcast_send is also under mi_rw_lock.
2969 	 */
2970 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
2971 	flent = mcip->mci_flent;
2972 	mac_client_remove_flow_from_list(mcip, flent);
2973 
2974 	if (mcip->mci_state_flags & MCIS_DESC_LOGGED)
2975 		mcip->mci_state_flags &= ~MCIS_DESC_LOGGED;
2976 
2977 	/*
2978 	 * This is the last unicast address being removed and there shouldn't
2979 	 * be any outbound data threads at this point coming down from mac
2980 	 * clients. We have waited for the data threads to finish before
2981 	 * starting dld_str_detach. Non-data threads must access TX SRS
2982 	 * under mi_rw_lock.
2983 	 */
2984 	rw_exit(&mip->mi_rw_lock);
2985 
2986 	/*
2987 	 * Don't use FLOW_MARK with FE_MC_NO_DATAPATH, as the flow might
2988 	 * contain other flags, such as FE_CONDEMNED, which we need to
2989 	 * cleared. We don't call mac_flow_cleanup() for this unicast
2990 	 * flow as we have a already cleaned up SRSs etc. (via the teadown
2991 	 * path). We just clear the stats and reset the initial callback
2992 	 * function, the rest will be set when we call mac_flow_create,
2993 	 * if at all.
2994 	 */
2995 	mutex_enter(&flent->fe_lock);
2996 	ASSERT(flent->fe_refcnt == 1 && flent->fe_mbg == NULL &&
2997 	    flent->fe_tx_srs == NULL && flent->fe_rx_srs_cnt == 0);
2998 	flent->fe_flags = FE_MC_NO_DATAPATH;
2999 	flow_stat_destroy(flent);
3000 	mac_misc_stat_delete(flent);
3001 
3002 	/* Initialize the receiver function to a safe routine */
3003 	flent->fe_cb_fn = (flow_fn_t)mac_rx_def;
3004 	flent->fe_cb_arg1 = NULL;
3005 	flent->fe_cb_arg2 = NULL;
3006 
3007 	flent->fe_index = -1;
3008 	mutex_exit(&flent->fe_lock);
3009 
3010 	if (mip->mi_type->mt_brdcst_addr != NULL) {
3011 		ASSERT(muip != NULL || no_unicast);
3012 		mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr,
3013 		    muip != NULL ? muip->mui_vid : VLAN_ID_NONE);
3014 	}
3015 
3016 	if (mip->mi_nactiveclients == 1) {
3017 		mac_capab_update((mac_handle_t)mip);
3018 		mac_virtual_link_update(mip);
3019 	}
3020 
3021 	if (mcip->mci_state_flags & MCIS_EXCLUSIVE) {
3022 		mip->mi_state_flags &= ~MIS_EXCLUSIVE;
3023 
3024 		if (mip->mi_state_flags & MIS_LEGACY)
3025 			mip->mi_capab_legacy.ml_active_clear(mip->mi_driver);
3026 	}
3027 
3028 	mcip->mci_state_flags &= ~MCIS_UNICAST_HW;
3029 
3030 	if (mcip->mci_state_flags & MCIS_TAG_DISABLE)
3031 		mcip->mci_state_flags &= ~MCIS_TAG_DISABLE;
3032 
3033 	if (mcip->mci_state_flags & MCIS_STRIP_DISABLE)
3034 		mcip->mci_state_flags &= ~MCIS_STRIP_DISABLE;
3035 
3036 	if (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK)
3037 		mcip->mci_state_flags &= ~MCIS_DISABLE_TX_VID_CHECK;
3038 
3039 	if (muip != NULL)
3040 		kmem_free(muip, sizeof (mac_unicast_impl_t));
3041 	mac_protect_cancel_timer(mcip);
3042 	mac_protect_flush_dynamic(mcip);
3043 
3044 	bzero(&mcip->mci_misc_stat, sizeof (mcip->mci_misc_stat));
3045 	/*
3046 	 * Disable fastpath if this is a VNIC or a VLAN.
3047 	 */
3048 	if (mcip->mci_state_flags & MCIS_IS_VNIC)
3049 		mac_fastpath_enable((mac_handle_t)mip);
3050 	mac_stop((mac_handle_t)mip);
3051 }
3052 
3053 /*
3054  * Remove a MAC address which was previously added by mac_unicast_add().
3055  */
3056 int
3057 mac_unicast_remove(mac_client_handle_t mch, mac_unicast_handle_t mah)
3058 {
3059 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3060 	mac_unicast_impl_t *muip = (mac_unicast_impl_t *)mah;
3061 	mac_unicast_impl_t *pre;
3062 	mac_impl_t *mip = mcip->mci_mip;
3063 	flow_entry_t		*flent;
3064 	uint16_t mui_vid;
3065 
3066 	i_mac_perim_enter(mip);
3067 	if (mcip->mci_flags & MAC_CLIENT_FLAGS_VNIC_PRIMARY) {
3068 		/*
3069 		 * Call made by the upper MAC client of a VNIC.
3070 		 * There's nothing much to do, the unicast address will
3071 		 * be removed by the VNIC driver when the VNIC is deleted,
3072 		 * but let's ensure that all our transmit is done before
3073 		 * the client does a mac_client_stop lest it trigger an
3074 		 * assert in the driver.
3075 		 */
3076 		ASSERT(muip->mui_vid == VLAN_ID_NONE);
3077 
3078 		mac_tx_client_flush(mcip);
3079 
3080 		if ((mcip->mci_flags & MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
3081 			mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
3082 			if (mcip->mci_rx_p_fn != NULL) {
3083 				mac_rx_set(mch, mcip->mci_rx_p_fn,
3084 				    mcip->mci_rx_p_arg);
3085 				mcip->mci_rx_p_fn = NULL;
3086 				mcip->mci_rx_p_arg = NULL;
3087 			}
3088 			kmem_free(muip, sizeof (mac_unicast_impl_t));
3089 			i_mac_perim_exit(mip);
3090 			return (0);
3091 		}
3092 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_VNIC_PRIMARY;
3093 
3094 		if (mcip->mci_state_flags & MCIS_TAG_DISABLE)
3095 			mcip->mci_state_flags &= ~MCIS_TAG_DISABLE;
3096 
3097 		if (mcip->mci_state_flags & MCIS_STRIP_DISABLE)
3098 			mcip->mci_state_flags &= ~MCIS_STRIP_DISABLE;
3099 
3100 		if (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK)
3101 			mcip->mci_state_flags &= ~MCIS_DISABLE_TX_VID_CHECK;
3102 
3103 		kmem_free(muip, sizeof (mac_unicast_impl_t));
3104 		i_mac_perim_exit(mip);
3105 		return (0);
3106 	}
3107 
3108 	ASSERT(muip != NULL);
3109 
3110 	/*
3111 	 * We are removing a passive client, we haven't setup the datapath
3112 	 * for this yet, so nothing much to do.
3113 	 */
3114 	if ((mcip->mci_flags & MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
3115 
3116 		ASSERT((mcip->mci_flent->fe_flags & FE_MC_NO_DATAPATH) != 0);
3117 		ASSERT(mcip->mci_p_unicast_list == muip);
3118 
3119 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
3120 
3121 		mcip->mci_p_unicast_list = NULL;
3122 		mcip->mci_rx_p_fn = NULL;
3123 		mcip->mci_rx_p_arg = NULL;
3124 
3125 		mcip->mci_state_flags &= ~MCIS_UNICAST_HW;
3126 
3127 		if (mcip->mci_state_flags & MCIS_TAG_DISABLE)
3128 			mcip->mci_state_flags &= ~MCIS_TAG_DISABLE;
3129 
3130 		if (mcip->mci_state_flags & MCIS_STRIP_DISABLE)
3131 			mcip->mci_state_flags &= ~MCIS_STRIP_DISABLE;
3132 
3133 		if (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK)
3134 			mcip->mci_state_flags &= ~MCIS_DISABLE_TX_VID_CHECK;
3135 
3136 		kmem_free(muip, sizeof (mac_unicast_impl_t));
3137 		i_mac_perim_exit(mip);
3138 		return (0);
3139 	}
3140 
3141 	/*
3142 	 * Remove the VID from the list of client's VIDs.
3143 	 */
3144 	pre = mcip->mci_unicast_list;
3145 	if (muip == pre) {
3146 		mcip->mci_unicast_list = muip->mui_next;
3147 	} else {
3148 		while ((pre->mui_next != NULL) && (pre->mui_next != muip))
3149 			pre = pre->mui_next;
3150 		ASSERT(pre->mui_next == muip);
3151 		rw_enter(&mcip->mci_rw_lock, RW_WRITER);
3152 		pre->mui_next = muip->mui_next;
3153 		rw_exit(&mcip->mci_rw_lock);
3154 	}
3155 
3156 	if (!mac_client_single_rcvr(mcip)) {
3157 		/*
3158 		 * This MAC client is shared by more than one unicast
3159 		 * addresses, so we will just remove the flent
3160 		 * corresponding to the address being removed. We don't invoke
3161 		 * mac_rx_classify_flow_rem() since the additional flow is
3162 		 * not associated with its own separate set of SRS and rings,
3163 		 * and these constructs are still needed for the remaining
3164 		 * flows.
3165 		 */
3166 		flent = mac_client_get_flow(mcip, muip);
3167 		VERIFY3P(flent, !=, NULL);
3168 
3169 		/*
3170 		 * The first one is disappearing, need to make sure
3171 		 * we replace it with another from the list of
3172 		 * shared clients.
3173 		 */
3174 		if (flent == mcip->mci_flent)
3175 			flent = mac_client_swap_mciflent(mcip);
3176 		mac_client_remove_flow_from_list(mcip, flent);
3177 		mac_flow_remove(mip->mi_flow_tab, flent, B_FALSE);
3178 		mac_flow_wait(flent, FLOW_DRIVER_UPCALL);
3179 
3180 		/*
3181 		 * The multicast groups that were added by the client so
3182 		 * far must be removed from the brodcast domain corresponding
3183 		 * to the VID being removed.
3184 		 */
3185 		mac_client_bcast_refresh(mcip, mac_client_update_mcast,
3186 		    (void *)flent, B_FALSE);
3187 
3188 		if (mip->mi_type->mt_brdcst_addr != NULL) {
3189 			mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr,
3190 			    muip->mui_vid);
3191 		}
3192 
3193 		FLOW_FINAL_REFRELE(flent);
3194 		ASSERT(!(mcip->mci_state_flags & MCIS_EXCLUSIVE));
3195 
3196 		/*
3197 		 * Enable fastpath if this is a VNIC or a VLAN.
3198 		 */
3199 		if (mcip->mci_state_flags & MCIS_IS_VNIC)
3200 			mac_fastpath_enable((mac_handle_t)mip);
3201 		mac_stop((mac_handle_t)mip);
3202 		i_mac_perim_exit(mip);
3203 		return (0);
3204 	}
3205 
3206 	mui_vid = muip->mui_vid;
3207 	mac_client_datapath_teardown(mch, muip, flent);
3208 
3209 	if ((mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY) &&
3210 	    mui_vid == VLAN_ID_NONE) {
3211 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PRIMARY;
3212 	} else {
3213 		i_mac_perim_exit(mip);
3214 		return (0);
3215 	}
3216 
3217 	/*
3218 	 * If we are removing the primary, check if we have a passive primary
3219 	 * client that we need to activate now.
3220 	 */
3221 	mcip = mac_get_passive_primary_client(mip);
3222 	if (mcip != NULL) {
3223 		mac_resource_props_t	*mrp;
3224 		mac_unicast_impl_t	*muip;
3225 
3226 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
3227 		mrp = kmem_zalloc(sizeof (*mrp), KM_SLEEP);
3228 
3229 		/*
3230 		 * Apply the property cached in the mac_impl_t to the
3231 		 * primary mac client.
3232 		 */
3233 		mac_get_resources((mac_handle_t)mip, mrp);
3234 		(void) mac_client_set_resources(mch, mrp);
3235 		ASSERT(mcip->mci_p_unicast_list != NULL);
3236 		muip = mcip->mci_p_unicast_list;
3237 		mcip->mci_p_unicast_list = NULL;
3238 		if (mac_client_datapath_setup(mcip, VLAN_ID_NONE,
3239 		    mip->mi_addr, mrp, B_TRUE, muip) == 0) {
3240 			if (mcip->mci_rx_p_fn != NULL) {
3241 				mac_rx_set(mch, mcip->mci_rx_p_fn,
3242 				    mcip->mci_rx_p_arg);
3243 				mcip->mci_rx_p_fn = NULL;
3244 				mcip->mci_rx_p_arg = NULL;
3245 			}
3246 		} else {
3247 			kmem_free(muip, sizeof (mac_unicast_impl_t));
3248 		}
3249 		kmem_free(mrp, sizeof (*mrp));
3250 	}
3251 	i_mac_perim_exit(mip);
3252 	return (0);
3253 }
3254 
3255 /*
3256  * Multicast add function invoked by MAC clients.
3257  */
3258 int
3259 mac_multicast_add(mac_client_handle_t mch, const uint8_t *addr)
3260 {
3261 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3262 	mac_impl_t		*mip = mcip->mci_mip;
3263 	flow_entry_t		*flent = mcip->mci_flent_list;
3264 	flow_entry_t		*prev_fe = NULL;
3265 	uint16_t		vid;
3266 	int			err = 0;
3267 
3268 	/* Verify the address is a valid multicast address */
3269 	if ((err = mip->mi_type->mt_ops.mtops_multicst_verify(addr,
3270 	    mip->mi_pdata)) != 0)
3271 		return (err);
3272 
3273 	i_mac_perim_enter(mip);
3274 	while (flent != NULL) {
3275 		vid = i_mac_flow_vid(flent);
3276 
3277 		err = mac_bcast_add((mac_client_impl_t *)mch, addr, vid,
3278 		    MAC_ADDRTYPE_MULTICAST);
3279 		if (err != 0)
3280 			break;
3281 		prev_fe = flent;
3282 		flent = flent->fe_client_next;
3283 	}
3284 
3285 	/*
3286 	 * If we failed adding, then undo all, rather than partial
3287 	 * success.
3288 	 */
3289 	if (flent != NULL && prev_fe != NULL) {
3290 		flent = mcip->mci_flent_list;
3291 		while (flent != prev_fe->fe_client_next) {
3292 			vid = i_mac_flow_vid(flent);
3293 			mac_bcast_delete((mac_client_impl_t *)mch, addr, vid);
3294 			flent = flent->fe_client_next;
3295 		}
3296 	}
3297 	i_mac_perim_exit(mip);
3298 	return (err);
3299 }
3300 
3301 /*
3302  * Multicast delete function invoked by MAC clients.
3303  */
3304 void
3305 mac_multicast_remove(mac_client_handle_t mch, const uint8_t *addr)
3306 {
3307 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3308 	mac_impl_t		*mip = mcip->mci_mip;
3309 	flow_entry_t		*flent;
3310 	uint16_t		vid;
3311 
3312 	i_mac_perim_enter(mip);
3313 	for (flent = mcip->mci_flent_list; flent != NULL;
3314 	    flent = flent->fe_client_next) {
3315 		vid = i_mac_flow_vid(flent);
3316 		mac_bcast_delete((mac_client_impl_t *)mch, addr, vid);
3317 	}
3318 	i_mac_perim_exit(mip);
3319 }
3320 
3321 /*
3322  * When a MAC client desires to capture packets on an interface,
3323  * it registers a promiscuous call back with mac_promisc_add().
3324  * There are three types of promiscuous callbacks:
3325  *
3326  * * MAC_CLIENT_PROMISC_ALL
3327  *   Captures all packets sent and received by the MAC client,
3328  *   the physical interface, as well as all other MAC clients
3329  *   defined on top of the same MAC.
3330  *
3331  * * MAC_CLIENT_PROMISC_FILTERED
3332  *   Captures all packets sent and received by the MAC client,
3333  *   plus all multicast traffic sent and received by the phyisical
3334  *   interface and the other MAC clients.
3335  *
3336  * * MAC_CLIENT_PROMISC_MULTI
3337  *   Captures all broadcast and multicast packets sent and
3338  *   received by the MAC clients as well as the physical interface.
3339  *
3340  * In all cases, the underlying MAC is put in promiscuous mode.
3341  */
3342 int
3343 mac_promisc_add(mac_client_handle_t mch, mac_client_promisc_type_t type,
3344     mac_rx_t fn, void *arg, mac_promisc_handle_t *mphp, uint16_t flags)
3345 {
3346 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3347 	mac_impl_t *mip = mcip->mci_mip;
3348 	mac_promisc_impl_t *mpip;
3349 	mac_cb_info_t	*mcbi;
3350 	int rc;
3351 
3352 	i_mac_perim_enter(mip);
3353 
3354 	if ((rc = mac_start((mac_handle_t)mip)) != 0) {
3355 		i_mac_perim_exit(mip);
3356 		return (rc);
3357 	}
3358 
3359 	if ((mcip->mci_state_flags & MCIS_IS_VNIC) &&
3360 	    type == MAC_CLIENT_PROMISC_ALL &&
3361 	    (mcip->mci_protect_flags & MPT_FLAG_PROMISC_FILTERED)) {
3362 		/*
3363 		 * The function is being invoked by the upper MAC client
3364 		 * of a VNIC. The VNIC should only see the traffic
3365 		 * it is entitled to.
3366 		 */
3367 		type = MAC_CLIENT_PROMISC_FILTERED;
3368 	}
3369 
3370 
3371 	/*
3372 	 * Turn on promiscuous mode for the underlying NIC.
3373 	 * This is needed even for filtered callbacks which
3374 	 * expect to receive all multicast traffic on the wire.
3375 	 *
3376 	 * Physical promiscuous mode should not be turned on if
3377 	 * MAC_PROMISC_FLAGS_NO_PHYS is set.
3378 	 */
3379 	if ((flags & MAC_PROMISC_FLAGS_NO_PHYS) == 0) {
3380 		if ((rc = i_mac_promisc_set(mip, B_TRUE)) != 0) {
3381 			mac_stop((mac_handle_t)mip);
3382 			i_mac_perim_exit(mip);
3383 			return (rc);
3384 		}
3385 	}
3386 
3387 	mpip = kmem_cache_alloc(mac_promisc_impl_cache, KM_SLEEP);
3388 
3389 	mpip->mpi_type = type;
3390 	mpip->mpi_fn = fn;
3391 	mpip->mpi_arg = arg;
3392 	mpip->mpi_mcip = mcip;
3393 	mpip->mpi_no_tx_loop = ((flags & MAC_PROMISC_FLAGS_NO_TX_LOOP) != 0);
3394 	mpip->mpi_no_phys = ((flags & MAC_PROMISC_FLAGS_NO_PHYS) != 0);
3395 	mpip->mpi_strip_vlan_tag =
3396 	    ((flags & MAC_PROMISC_FLAGS_VLAN_TAG_STRIP) != 0);
3397 	mpip->mpi_no_copy = ((flags & MAC_PROMISC_FLAGS_NO_COPY) != 0);
3398 
3399 	mcbi = &mip->mi_promisc_cb_info;
3400 	mutex_enter(mcbi->mcbi_lockp);
3401 
3402 	mac_callback_add(&mip->mi_promisc_cb_info, &mcip->mci_promisc_list,
3403 	    &mpip->mpi_mci_link);
3404 	mac_callback_add(&mip->mi_promisc_cb_info, &mip->mi_promisc_list,
3405 	    &mpip->mpi_mi_link);
3406 
3407 	mutex_exit(mcbi->mcbi_lockp);
3408 
3409 	*mphp = (mac_promisc_handle_t)mpip;
3410 
3411 	if (mcip->mci_state_flags & MCIS_IS_VNIC) {
3412 		mac_impl_t *umip = mcip->mci_upper_mip;
3413 
3414 		ASSERT(umip != NULL);
3415 		mac_vnic_secondary_update(umip);
3416 	}
3417 
3418 	i_mac_perim_exit(mip);
3419 
3420 	return (0);
3421 }
3422 
3423 /*
3424  * Remove a multicast address previously aded through mac_promisc_add().
3425  */
3426 void
3427 mac_promisc_remove(mac_promisc_handle_t mph)
3428 {
3429 	mac_promisc_impl_t *mpip = (mac_promisc_impl_t *)mph;
3430 	mac_client_impl_t *mcip = mpip->mpi_mcip;
3431 	mac_impl_t *mip = mcip->mci_mip;
3432 	mac_cb_info_t *mcbi;
3433 	int rv;
3434 
3435 	i_mac_perim_enter(mip);
3436 
3437 	/*
3438 	 * Even if the device can't be reset into normal mode, we still
3439 	 * need to clear the client promisc callbacks. The client may want
3440 	 * to close the mac end point and we can't have stale callbacks.
3441 	 */
3442 	if (!(mpip->mpi_no_phys)) {
3443 		if ((rv = i_mac_promisc_set(mip, B_FALSE)) != 0) {
3444 			cmn_err(CE_WARN, "%s: failed to switch OFF promiscuous"
3445 			    " mode because of error 0x%x", mip->mi_name, rv);
3446 		}
3447 	}
3448 	mcbi = &mip->mi_promisc_cb_info;
3449 	mutex_enter(mcbi->mcbi_lockp);
3450 	if (mac_callback_remove(mcbi, &mip->mi_promisc_list,
3451 	    &mpip->mpi_mi_link)) {
3452 		VERIFY(mac_callback_remove(&mip->mi_promisc_cb_info,
3453 		    &mcip->mci_promisc_list, &mpip->mpi_mci_link));
3454 		kmem_cache_free(mac_promisc_impl_cache, mpip);
3455 	} else {
3456 		mac_callback_remove_wait(&mip->mi_promisc_cb_info);
3457 	}
3458 
3459 	if (mcip->mci_state_flags & MCIS_IS_VNIC) {
3460 		mac_impl_t *umip = mcip->mci_upper_mip;
3461 
3462 		ASSERT(umip != NULL);
3463 		mac_vnic_secondary_update(umip);
3464 	}
3465 
3466 	mutex_exit(mcbi->mcbi_lockp);
3467 	mac_stop((mac_handle_t)mip);
3468 
3469 	i_mac_perim_exit(mip);
3470 }
3471 
3472 /*
3473  * Reference count the number of active Tx threads. MCI_TX_QUIESCE indicates
3474  * that a control operation wants to quiesce the Tx data flow in which case
3475  * we return an error. Holding any of the per cpu locks ensures that the
3476  * mci_tx_flag won't change.
3477  *
3478  * 'CPU' must be accessed just once and used to compute the index into the
3479  * percpu array, and that index must be used for the entire duration of the
3480  * packet send operation. Note that the thread may be preempted and run on
3481  * another cpu any time and so we can't use 'CPU' more than once for the
3482  * operation.
3483  */
3484 #define	MAC_TX_TRY_HOLD(mcip, mytx, error)				\
3485 {									\
3486 	(error) = 0;							\
3487 	(mytx) = &(mcip)->mci_tx_pcpu[CPU->cpu_seqid & mac_tx_percpu_cnt]; \
3488 	mutex_enter(&(mytx)->pcpu_tx_lock);				\
3489 	if (!((mcip)->mci_tx_flag & MCI_TX_QUIESCE)) {			\
3490 		(mytx)->pcpu_tx_refcnt++;				\
3491 	} else {							\
3492 		(error) = -1;						\
3493 	}								\
3494 	mutex_exit(&(mytx)->pcpu_tx_lock);				\
3495 }
3496 
3497 /*
3498  * Release the reference. If needed, signal any control operation waiting
3499  * for Tx quiescence. The wait and signal are always done using the
3500  * mci_tx_pcpu[0]'s lock
3501  */
3502 #define	MAC_TX_RELE(mcip, mytx) {					\
3503 	mutex_enter(&(mytx)->pcpu_tx_lock);				\
3504 	if (--(mytx)->pcpu_tx_refcnt == 0 &&				\
3505 	    (mcip)->mci_tx_flag & MCI_TX_QUIESCE) {			\
3506 		mutex_exit(&(mytx)->pcpu_tx_lock);			\
3507 		mutex_enter(&(mcip)->mci_tx_pcpu[0].pcpu_tx_lock);	\
3508 		cv_signal(&(mcip)->mci_tx_cv);				\
3509 		mutex_exit(&(mcip)->mci_tx_pcpu[0].pcpu_tx_lock);	\
3510 	} else {							\
3511 		mutex_exit(&(mytx)->pcpu_tx_lock);			\
3512 	}								\
3513 }
3514 
3515 /*
3516  * Send function invoked by MAC clients.
3517  */
3518 mac_tx_cookie_t
3519 mac_tx(mac_client_handle_t mch, mblk_t *mp_chain, uintptr_t hint,
3520     uint16_t flag, mblk_t **ret_mp)
3521 {
3522 	mac_tx_cookie_t		cookie = 0;
3523 	int			error;
3524 	mac_tx_percpu_t		*mytx;
3525 	mac_soft_ring_set_t	*srs;
3526 	flow_entry_t		*flent;
3527 	boolean_t		is_subflow = B_FALSE;
3528 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3529 	mac_impl_t		*mip = mcip->mci_mip;
3530 	mac_srs_tx_t		*srs_tx;
3531 
3532 	/*
3533 	 * Check whether the active Tx threads count is bumped already.
3534 	 */
3535 	if (!(flag & MAC_TX_NO_HOLD)) {
3536 		MAC_TX_TRY_HOLD(mcip, mytx, error);
3537 		if (error != 0) {
3538 			freemsgchain(mp_chain);
3539 			return (0);
3540 		}
3541 	}
3542 
3543 	/*
3544 	 * If mac protection is enabled, only the permissible packets will be
3545 	 * returned by mac_protect_check().
3546 	 */
3547 	if ((mcip->mci_flent->
3548 	    fe_resource_props.mrp_mask & MRP_PROTECT) != 0 &&
3549 	    (mp_chain = mac_protect_check(mch, mp_chain)) == NULL)
3550 		goto done;
3551 
3552 	if (mcip->mci_subflow_tab != NULL &&
3553 	    mcip->mci_subflow_tab->ft_flow_count > 0 &&
3554 	    mac_flow_lookup(mcip->mci_subflow_tab, mp_chain,
3555 	    FLOW_OUTBOUND, &flent) == 0) {
3556 		/*
3557 		 * The main assumption here is that if in the event
3558 		 * we get a chain, all the packets will be classified
3559 		 * to the same Flow/SRS. If this changes for any
3560 		 * reason, the following logic should change as well.
3561 		 * I suppose the fanout_hint also assumes this .
3562 		 */
3563 		ASSERT(flent != NULL);
3564 		is_subflow = B_TRUE;
3565 	} else {
3566 		flent = mcip->mci_flent;
3567 	}
3568 
3569 	srs = flent->fe_tx_srs;
3570 	/*
3571 	 * This is to avoid panics with PF_PACKET that can call mac_tx()
3572 	 * against an interface that is not capable of sending. A rewrite
3573 	 * of the mac datapath is required to remove this limitation.
3574 	 */
3575 	if (srs == NULL) {
3576 		freemsgchain(mp_chain);
3577 		goto done;
3578 	}
3579 
3580 	srs_tx = &srs->srs_tx;
3581 	if (srs_tx->st_mode == SRS_TX_DEFAULT &&
3582 	    (srs->srs_state & SRS_ENQUEUED) == 0 &&
3583 	    mip->mi_nactiveclients == 1 &&
3584 	    mp_chain->b_next == NULL &&
3585 	    (DB_CKSUMFLAGS(mp_chain) & HW_LSO) == 0) {
3586 		uint64_t	obytes;
3587 
3588 		/*
3589 		 * Since dls always opens the underlying MAC, nclients equals
3590 		 * to 1 means that the only active client is dls itself acting
3591 		 * as a primary client of the MAC instance. Since dls will not
3592 		 * send tagged packets in that case, and dls is trusted to send
3593 		 * packets for its allowed VLAN(s), the VLAN tag insertion and
3594 		 * check is required only if nclients is greater than 1.
3595 		 */
3596 		if (mip->mi_nclients > 1) {
3597 			if (MAC_VID_CHECK_NEEDED(mcip)) {
3598 				int	err = 0;
3599 
3600 				MAC_VID_CHECK(mcip, mp_chain, err);
3601 				if (err != 0) {
3602 					freemsg(mp_chain);
3603 					mcip->mci_misc_stat.mms_txerrors++;
3604 					goto done;
3605 				}
3606 			}
3607 			if (MAC_TAG_NEEDED(mcip)) {
3608 				mp_chain = mac_add_vlan_tag(mp_chain, 0,
3609 				    mac_client_vid(mch));
3610 				if (mp_chain == NULL) {
3611 					mcip->mci_misc_stat.mms_txerrors++;
3612 					goto done;
3613 				}
3614 			}
3615 		}
3616 
3617 		obytes = (mp_chain->b_cont == NULL ? MBLKL(mp_chain) :
3618 		    msgdsize(mp_chain));
3619 
3620 		mp_chain = mac_provider_tx(mip, srs_tx->st_arg2, mp_chain,
3621 		    mcip);
3622 
3623 		if (mp_chain == NULL) {
3624 			cookie = 0;
3625 			SRS_TX_STAT_UPDATE(srs, opackets, 1);
3626 			SRS_TX_STAT_UPDATE(srs, obytes, obytes);
3627 		} else {
3628 			mutex_enter(&srs->srs_lock);
3629 			cookie = mac_tx_srs_no_desc(srs, mp_chain,
3630 			    flag, ret_mp);
3631 			mutex_exit(&srs->srs_lock);
3632 		}
3633 	} else {
3634 		mblk_t *mp = mp_chain;
3635 		mblk_t *new_head = NULL;
3636 		mblk_t *new_tail = NULL;
3637 
3638 		/*
3639 		 * There are occasions where the packets arriving here
3640 		 * may request hardware offloads that are not
3641 		 * available from the underlying MAC provider. This
3642 		 * currently only happens when a packet is sent across
3643 		 * the MAC-loopback path of one MAC and then forwarded
3644 		 * (via IP) to another MAC that lacks one or more of
3645 		 * the hardware offloads provided by the first one.
3646 		 * However, in the future, we may choose to pretend
3647 		 * all MAC providers support all offloads, performing
3648 		 * emulation on Tx as needed.
3649 		 *
3650 		 * We iterate each mblk in-turn, emulating hardware
3651 		 * offloads as required. From this process, we create
3652 		 * a new chain. The new chain may be the same as the
3653 		 * original chain (no hardware emulation needed), a
3654 		 * collection of new mblks (hardware emulation
3655 		 * needed), or a mix. At this point, the chain is safe
3656 		 * for consumption by the underlying MAC provider and
3657 		 * is passed down to the SRS.
3658 		 */
3659 		while (mp != NULL) {
3660 			mblk_t *next = mp->b_next;
3661 			mblk_t *tail = NULL;
3662 			const uint16_t needed =
3663 			    (DB_CKSUMFLAGS(mp) ^ mip->mi_tx_cksum_flags) &
3664 			    DB_CKSUMFLAGS(mp);
3665 
3666 			mp->b_next = NULL;
3667 
3668 			if ((needed & (HCK_TX_FLAGS | HW_LSO_FLAGS)) != 0) {
3669 				mac_emul_t emul = 0;
3670 
3671 				if (needed & HCK_IPV4_HDRCKSUM)
3672 					emul |= MAC_IPCKSUM_EMUL;
3673 				if (needed & (HCK_PARTIALCKSUM | HCK_FULLCKSUM))
3674 					emul |= MAC_HWCKSUM_EMUL;
3675 				if (needed & HW_LSO)
3676 					emul = MAC_LSO_EMUL;
3677 
3678 				mac_hw_emul(&mp, &tail, NULL, emul);
3679 
3680 				if (mp == NULL) {
3681 					mp = next;
3682 					continue;
3683 				}
3684 			}
3685 
3686 			if (new_head == NULL) {
3687 				new_head = mp;
3688 			} else {
3689 				new_tail->b_next = mp;
3690 			}
3691 
3692 			new_tail = (tail == NULL) ? mp : tail;
3693 			mp = next;
3694 		}
3695 
3696 		if (new_head == NULL) {
3697 			cookie = 0;
3698 			goto done;
3699 		}
3700 
3701 		cookie = srs_tx->st_func(srs, new_head, hint, flag, ret_mp);
3702 	}
3703 
3704 done:
3705 	if (is_subflow)
3706 		FLOW_REFRELE(flent);
3707 
3708 	if (!(flag & MAC_TX_NO_HOLD))
3709 		MAC_TX_RELE(mcip, mytx);
3710 
3711 	return (cookie);
3712 }
3713 
3714 /*
3715  * mac_tx_is_blocked
3716  *
3717  * Given a cookie, it returns if the ring identified by the cookie is
3718  * flow-controlled or not. If NULL is passed in place of a cookie,
3719  * then it finds out if any of the underlying rings belonging to the
3720  * SRS is flow controlled or not and returns that status.
3721  */
3722 /* ARGSUSED */
3723 boolean_t
3724 mac_tx_is_flow_blocked(mac_client_handle_t mch, mac_tx_cookie_t cookie)
3725 {
3726 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3727 	mac_soft_ring_set_t *mac_srs;
3728 	mac_soft_ring_t *sringp;
3729 	boolean_t blocked = B_FALSE;
3730 	mac_tx_percpu_t *mytx;
3731 	int err;
3732 	int i;
3733 
3734 	/*
3735 	 * Bump the reference count so that mac_srs won't be deleted.
3736 	 * If the client is currently quiesced and we failed to bump
3737 	 * the reference, return B_TRUE so that flow control stays
3738 	 * as enabled.
3739 	 *
3740 	 * Flow control will then be disabled once the client is no
3741 	 * longer quiesced.
3742 	 */
3743 	MAC_TX_TRY_HOLD(mcip, mytx, err);
3744 	if (err != 0)
3745 		return (B_TRUE);
3746 
3747 	if ((mac_srs = MCIP_TX_SRS(mcip)) == NULL) {
3748 		MAC_TX_RELE(mcip, mytx);
3749 		return (B_FALSE);
3750 	}
3751 
3752 	mutex_enter(&mac_srs->srs_lock);
3753 	/*
3754 	 * Only in the case of TX_FANOUT and TX_AGGR, the underlying
3755 	 * softring (s_ring_state) will have the HIWAT set. This is
3756 	 * the multiple Tx ring flow control case. For all other
3757 	 * case, SRS (srs_state) will store the condition.
3758 	 */
3759 	if (mac_srs->srs_tx.st_mode == SRS_TX_FANOUT ||
3760 	    mac_srs->srs_tx.st_mode == SRS_TX_AGGR) {
3761 		if (cookie != 0) {
3762 			sringp = (mac_soft_ring_t *)cookie;
3763 			mutex_enter(&sringp->s_ring_lock);
3764 			if (sringp->s_ring_state & S_RING_TX_HIWAT)
3765 				blocked = B_TRUE;
3766 			mutex_exit(&sringp->s_ring_lock);
3767 		} else {
3768 			for (i = 0; i < mac_srs->srs_tx_ring_count; i++) {
3769 				sringp = mac_srs->srs_tx_soft_rings[i];
3770 				mutex_enter(&sringp->s_ring_lock);
3771 				if (sringp->s_ring_state & S_RING_TX_HIWAT) {
3772 					blocked = B_TRUE;
3773 					mutex_exit(&sringp->s_ring_lock);
3774 					break;
3775 				}
3776 				mutex_exit(&sringp->s_ring_lock);
3777 			}
3778 		}
3779 	} else {
3780 		blocked = (mac_srs->srs_state & SRS_TX_HIWAT);
3781 	}
3782 	mutex_exit(&mac_srs->srs_lock);
3783 	MAC_TX_RELE(mcip, mytx);
3784 	return (blocked);
3785 }
3786 
3787 /*
3788  * Check if the MAC client is the primary MAC client.
3789  */
3790 boolean_t
3791 mac_is_primary_client(mac_client_impl_t *mcip)
3792 {
3793 	return (mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY);
3794 }
3795 
3796 void
3797 mac_ioctl(mac_handle_t mh, queue_t *wq, mblk_t *bp)
3798 {
3799 	mac_impl_t	*mip = (mac_impl_t *)mh;
3800 	int cmd = ((struct iocblk *)bp->b_rptr)->ioc_cmd;
3801 
3802 	if ((cmd == ND_GET && (mip->mi_callbacks->mc_callbacks & MC_GETPROP)) ||
3803 	    (cmd == ND_SET && (mip->mi_callbacks->mc_callbacks & MC_SETPROP))) {
3804 		/*
3805 		 * If ndd props were registered, call them.
3806 		 * Note that ndd ioctls are Obsolete
3807 		 */
3808 		mac_ndd_ioctl(mip, wq, bp);
3809 		return;
3810 	}
3811 
3812 	/*
3813 	 * Call the driver to handle the ioctl.  The driver may not support
3814 	 * any ioctls, in which case we reply with a NAK on its behalf.
3815 	 */
3816 	if (mip->mi_callbacks->mc_callbacks & MC_IOCTL)
3817 		mip->mi_ioctl(mip->mi_driver, wq, bp);
3818 	else
3819 		miocnak(wq, bp, 0, EINVAL);
3820 }
3821 
3822 /*
3823  * Return the link state of the specified MAC instance.
3824  */
3825 link_state_t
3826 mac_link_get(mac_handle_t mh)
3827 {
3828 	return (((mac_impl_t *)mh)->mi_linkstate);
3829 }
3830 
3831 /*
3832  * Add a mac client specified notification callback. Please see the comments
3833  * above mac_callback_add() for general information about mac callback
3834  * addition/deletion in the presence of mac callback list walkers
3835  */
3836 mac_notify_handle_t
3837 mac_notify_add(mac_handle_t mh, mac_notify_t notify_fn, void *arg)
3838 {
3839 	mac_impl_t		*mip = (mac_impl_t *)mh;
3840 	mac_notify_cb_t		*mncb;
3841 	mac_cb_info_t		*mcbi;
3842 
3843 	/*
3844 	 * Allocate a notify callback structure, fill in the details and
3845 	 * use the mac callback list manipulation functions to chain into
3846 	 * the list of callbacks.
3847 	 */
3848 	mncb = kmem_zalloc(sizeof (mac_notify_cb_t), KM_SLEEP);
3849 	mncb->mncb_fn = notify_fn;
3850 	mncb->mncb_arg = arg;
3851 	mncb->mncb_mip = mip;
3852 	mncb->mncb_link.mcb_objp = mncb;
3853 	mncb->mncb_link.mcb_objsize = sizeof (mac_notify_cb_t);
3854 	mncb->mncb_link.mcb_flags = MCB_NOTIFY_CB_T;
3855 
3856 	mcbi = &mip->mi_notify_cb_info;
3857 
3858 	i_mac_perim_enter(mip);
3859 	mutex_enter(mcbi->mcbi_lockp);
3860 
3861 	mac_callback_add(&mip->mi_notify_cb_info, &mip->mi_notify_cb_list,
3862 	    &mncb->mncb_link);
3863 
3864 	mutex_exit(mcbi->mcbi_lockp);
3865 	i_mac_perim_exit(mip);
3866 	return ((mac_notify_handle_t)mncb);
3867 }
3868 
3869 void
3870 mac_notify_remove_wait(mac_handle_t mh)
3871 {
3872 	mac_impl_t	*mip = (mac_impl_t *)mh;
3873 	mac_cb_info_t	*mcbi = &mip->mi_notify_cb_info;
3874 
3875 	mutex_enter(mcbi->mcbi_lockp);
3876 	mac_callback_remove_wait(&mip->mi_notify_cb_info);
3877 	mutex_exit(mcbi->mcbi_lockp);
3878 }
3879 
3880 /*
3881  * Remove a mac client specified notification callback
3882  */
3883 int
3884 mac_notify_remove(mac_notify_handle_t mnh, boolean_t wait)
3885 {
3886 	mac_notify_cb_t	*mncb = (mac_notify_cb_t *)mnh;
3887 	mac_impl_t	*mip = mncb->mncb_mip;
3888 	mac_cb_info_t	*mcbi;
3889 	int		err = 0;
3890 
3891 	mcbi = &mip->mi_notify_cb_info;
3892 
3893 	i_mac_perim_enter(mip);
3894 	mutex_enter(mcbi->mcbi_lockp);
3895 
3896 	ASSERT(mncb->mncb_link.mcb_objp == mncb);
3897 	/*
3898 	 * If there aren't any list walkers, the remove would succeed
3899 	 * inline, else we wait for the deferred remove to complete
3900 	 */
3901 	if (mac_callback_remove(&mip->mi_notify_cb_info,
3902 	    &mip->mi_notify_cb_list, &mncb->mncb_link)) {
3903 		kmem_free(mncb, sizeof (mac_notify_cb_t));
3904 	} else {
3905 		err = EBUSY;
3906 	}
3907 
3908 	mutex_exit(mcbi->mcbi_lockp);
3909 	i_mac_perim_exit(mip);
3910 
3911 	/*
3912 	 * If we failed to remove the notification callback and "wait" is set
3913 	 * to be B_TRUE, wait for the callback to finish after we exit the
3914 	 * mac perimeter.
3915 	 */
3916 	if (err != 0 && wait) {
3917 		mac_notify_remove_wait((mac_handle_t)mip);
3918 		return (0);
3919 	}
3920 
3921 	return (err);
3922 }
3923 
3924 /*
3925  * Associate resource management callbacks with the specified MAC
3926  * clients.
3927  */
3928 
3929 void
3930 mac_resource_set_common(mac_client_handle_t mch, mac_resource_add_t add,
3931     mac_resource_remove_t remove, mac_resource_quiesce_t quiesce,
3932     mac_resource_restart_t restart, mac_resource_bind_t bind,
3933     void *arg)
3934 {
3935 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3936 
3937 	mcip->mci_resource_add = add;
3938 	mcip->mci_resource_remove = remove;
3939 	mcip->mci_resource_quiesce = quiesce;
3940 	mcip->mci_resource_restart = restart;
3941 	mcip->mci_resource_bind = bind;
3942 	mcip->mci_resource_arg = arg;
3943 }
3944 
3945 void
3946 mac_resource_set(mac_client_handle_t mch, mac_resource_add_t add, void *arg)
3947 {
3948 	/* update the 'resource_add' callback */
3949 	mac_resource_set_common(mch, add, NULL, NULL, NULL, NULL, arg);
3950 }
3951 
3952 /*
3953  * Sets up the client resources and enable the polling interface over all the
3954  * SRS's and the soft rings of the client
3955  */
3956 void
3957 mac_client_poll_enable(mac_client_handle_t mch)
3958 {
3959 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3960 	mac_soft_ring_set_t	*mac_srs;
3961 	flow_entry_t		*flent;
3962 	int			i;
3963 
3964 	flent = mcip->mci_flent;
3965 	ASSERT(flent != NULL);
3966 
3967 	mcip->mci_state_flags |= MCIS_CLIENT_POLL_CAPABLE;
3968 	for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
3969 		mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
3970 		ASSERT(mac_srs->srs_mcip == mcip);
3971 		mac_srs_client_poll_enable(mcip, mac_srs);
3972 	}
3973 }
3974 
3975 /*
3976  * Tears down the client resources and disable the polling interface over all
3977  * the SRS's and the soft rings of the client
3978  */
3979 void
3980 mac_client_poll_disable(mac_client_handle_t mch)
3981 {
3982 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3983 	mac_soft_ring_set_t	*mac_srs;
3984 	flow_entry_t		*flent;
3985 	int			i;
3986 
3987 	flent = mcip->mci_flent;
3988 	ASSERT(flent != NULL);
3989 
3990 	mcip->mci_state_flags &= ~MCIS_CLIENT_POLL_CAPABLE;
3991 	for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
3992 		mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
3993 		ASSERT(mac_srs->srs_mcip == mcip);
3994 		mac_srs_client_poll_disable(mcip, mac_srs);
3995 	}
3996 }
3997 
3998 /*
3999  * Associate the CPUs specified by the given property with a MAC client.
4000  */
4001 int
4002 mac_cpu_set(mac_client_handle_t mch, mac_resource_props_t *mrp)
4003 {
4004 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
4005 	mac_impl_t *mip = mcip->mci_mip;
4006 	int err = 0;
4007 
4008 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
4009 
4010 	if ((err = mac_validate_props(mcip->mci_state_flags & MCIS_IS_VNIC ?
4011 	    mcip->mci_upper_mip : mip, mrp)) != 0) {
4012 		return (err);
4013 	}
4014 	if (MCIP_DATAPATH_SETUP(mcip))
4015 		mac_flow_modify(mip->mi_flow_tab, mcip->mci_flent, mrp);
4016 
4017 	mac_update_resources(mrp, MCIP_RESOURCE_PROPS(mcip), B_FALSE);
4018 	return (0);
4019 }
4020 
4021 /*
4022  * Apply the specified properties to the specified MAC client.
4023  */
4024 int
4025 mac_client_set_resources(mac_client_handle_t mch, mac_resource_props_t *mrp)
4026 {
4027 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
4028 	mac_impl_t *mip = mcip->mci_mip;
4029 	int err = 0;
4030 
4031 	i_mac_perim_enter(mip);
4032 
4033 	if ((mrp->mrp_mask & MRP_MAXBW) || (mrp->mrp_mask & MRP_PRIORITY)) {
4034 		err = mac_resource_ctl_set(mch, mrp);
4035 		if (err != 0)
4036 			goto done;
4037 	}
4038 
4039 	if (mrp->mrp_mask & (MRP_CPUS|MRP_POOL)) {
4040 		err = mac_cpu_set(mch, mrp);
4041 		if (err != 0)
4042 			goto done;
4043 	}
4044 
4045 	if (mrp->mrp_mask & MRP_PROTECT) {
4046 		err = mac_protect_set(mch, mrp);
4047 		if (err != 0)
4048 			goto done;
4049 	}
4050 
4051 	if ((mrp->mrp_mask & MRP_RX_RINGS) || (mrp->mrp_mask & MRP_TX_RINGS))
4052 		err = mac_resource_ctl_set(mch, mrp);
4053 
4054 done:
4055 	i_mac_perim_exit(mip);
4056 	return (err);
4057 }
4058 
4059 /*
4060  * Return the properties currently associated with the specified MAC client.
4061  */
4062 void
4063 mac_client_get_resources(mac_client_handle_t mch, mac_resource_props_t *mrp)
4064 {
4065 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
4066 	mac_resource_props_t	*mcip_mrp = MCIP_RESOURCE_PROPS(mcip);
4067 
4068 	bcopy(mcip_mrp, mrp, sizeof (mac_resource_props_t));
4069 }
4070 
4071 /*
4072  * Return the effective properties currently associated with the specified
4073  * MAC client.
4074  */
4075 void
4076 mac_client_get_effective_resources(mac_client_handle_t mch,
4077     mac_resource_props_t *mrp)
4078 {
4079 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
4080 	mac_resource_props_t	*mcip_mrp = MCIP_EFFECTIVE_PROPS(mcip);
4081 
4082 	bcopy(mcip_mrp, mrp, sizeof (mac_resource_props_t));
4083 }
4084 
4085 /*
4086  * Pass a copy of the specified packet to the promiscuous callbacks
4087  * of the specified MAC.
4088  *
4089  * If sender is NULL, the function is being invoked for a packet chain
4090  * received from the wire. If sender is non-NULL, it points to
4091  * the MAC client from which the packet is being sent.
4092  *
4093  * The packets are distributed to the promiscuous callbacks as follows:
4094  *
4095  * - all packets are sent to the MAC_CLIENT_PROMISC_ALL callbacks
4096  * - all broadcast and multicast packets are sent to the
4097  *   MAC_CLIENT_PROMISC_FILTER and MAC_CLIENT_PROMISC_MULTI.
4098  *
4099  * The unicast packets of MAC_CLIENT_PROMISC_FILTER callbacks are dispatched
4100  * after classification by mac_rx_deliver().
4101  */
4102 static void
4103 mac_promisc_dispatch_one(mac_promisc_impl_t *mpip, mblk_t *mp,
4104     boolean_t loopback, boolean_t local)
4105 {
4106 	mblk_t *mp_next;
4107 
4108 	if (!mpip->mpi_no_copy || mpip->mpi_strip_vlan_tag) {
4109 		mblk_t *mp_copy;
4110 
4111 		mp_copy = copymsg(mp);
4112 		if (mp_copy == NULL)
4113 			return;
4114 
4115 		if (mpip->mpi_strip_vlan_tag) {
4116 			mp_copy = mac_strip_vlan_tag_chain(mp_copy);
4117 			if (mp_copy == NULL)
4118 				return;
4119 		}
4120 
4121 		/*
4122 		 * There is code upstack that can't deal with message
4123 		 * chains.
4124 		 */
4125 		for (mblk_t *tmp = mp_copy; tmp != NULL; tmp = mp_next) {
4126 			mp_next = tmp->b_next;
4127 			tmp->b_next = NULL;
4128 			mpip->mpi_fn(mpip->mpi_arg, NULL, tmp, loopback);
4129 		}
4130 
4131 		return;
4132 	}
4133 
4134 	mp_next = mp->b_next;
4135 	mp->b_next = NULL;
4136 	mpip->mpi_fn(mpip->mpi_arg, NULL, mp, loopback);
4137 	mp->b_next = mp_next;
4138 }
4139 
4140 /*
4141  * Return the VID of a packet. Zero if the packet is not tagged.
4142  */
4143 static uint16_t
4144 mac_ether_vid(mblk_t *mp)
4145 {
4146 	struct ether_header *eth = (struct ether_header *)mp->b_rptr;
4147 
4148 	if (ntohs(eth->ether_type) == ETHERTYPE_VLAN) {
4149 		struct ether_vlan_header *t_evhp =
4150 		    (struct ether_vlan_header *)mp->b_rptr;
4151 		return (VLAN_ID(ntohs(t_evhp->ether_tci)));
4152 	}
4153 
4154 	return (0);
4155 }
4156 
4157 /*
4158  * Return whether the specified packet contains a multicast or broadcast
4159  * destination MAC address.
4160  */
4161 static boolean_t
4162 mac_is_mcast(mac_impl_t *mip, mblk_t *mp)
4163 {
4164 	mac_header_info_t hdr_info;
4165 
4166 	if (mac_header_info((mac_handle_t)mip, mp, &hdr_info) != 0)
4167 		return (B_FALSE);
4168 	return ((hdr_info.mhi_dsttype == MAC_ADDRTYPE_BROADCAST) ||
4169 	    (hdr_info.mhi_dsttype == MAC_ADDRTYPE_MULTICAST));
4170 }
4171 
4172 /*
4173  * Send a copy of an mblk chain to the MAC clients of the specified MAC.
4174  * "sender" points to the sender MAC client for outbound packets, and
4175  * is set to NULL for inbound packets.
4176  */
4177 void
4178 mac_promisc_dispatch(mac_impl_t *mip, mblk_t *mp_chain,
4179     mac_client_impl_t *sender, boolean_t local)
4180 {
4181 	mac_promisc_impl_t *mpip;
4182 	mac_cb_t *mcb;
4183 	mblk_t *mp;
4184 	boolean_t is_mcast, is_sender;
4185 
4186 	MAC_PROMISC_WALKER_INC(mip);
4187 	for (mp = mp_chain; mp != NULL; mp = mp->b_next) {
4188 		is_mcast = mac_is_mcast(mip, mp);
4189 		/* send packet to interested callbacks */
4190 		for (mcb = mip->mi_promisc_list; mcb != NULL;
4191 		    mcb = mcb->mcb_nextp) {
4192 			mpip = (mac_promisc_impl_t *)mcb->mcb_objp;
4193 			is_sender = (mpip->mpi_mcip == sender);
4194 
4195 			if (is_sender && mpip->mpi_no_tx_loop)
4196 				/*
4197 				 * The sender doesn't want to receive
4198 				 * copies of the packets it sends.
4199 				 */
4200 				continue;
4201 
4202 			/* this client doesn't need any packets (bridge) */
4203 			if (mpip->mpi_fn == NULL)
4204 				continue;
4205 
4206 			/*
4207 			 * For an ethernet MAC, don't displatch a multicast
4208 			 * packet to a non-PROMISC_ALL callbacks unless the VID
4209 			 * of the packet matches the VID of the client.
4210 			 */
4211 			if (is_mcast &&
4212 			    mpip->mpi_type != MAC_CLIENT_PROMISC_ALL &&
4213 			    !mac_client_check_flow_vid(mpip->mpi_mcip,
4214 			    mac_ether_vid(mp)))
4215 				continue;
4216 
4217 			if (is_sender ||
4218 			    mpip->mpi_type == MAC_CLIENT_PROMISC_ALL ||
4219 			    is_mcast) {
4220 				mac_promisc_dispatch_one(mpip, mp, is_sender,
4221 				    local);
4222 			}
4223 		}
4224 	}
4225 	MAC_PROMISC_WALKER_DCR(mip);
4226 }
4227 
4228 void
4229 mac_promisc_client_dispatch(mac_client_impl_t *mcip, mblk_t *mp_chain)
4230 {
4231 	mac_impl_t		*mip = mcip->mci_mip;
4232 	mac_promisc_impl_t	*mpip;
4233 	boolean_t		is_mcast;
4234 	mblk_t			*mp;
4235 	mac_cb_t		*mcb;
4236 
4237 	/*
4238 	 * The unicast packets for the MAC client still
4239 	 * need to be delivered to the MAC_CLIENT_PROMISC_FILTERED
4240 	 * promiscuous callbacks. The broadcast and multicast
4241 	 * packets were delivered from mac_rx().
4242 	 */
4243 	MAC_PROMISC_WALKER_INC(mip);
4244 	for (mp = mp_chain; mp != NULL; mp = mp->b_next) {
4245 		is_mcast = mac_is_mcast(mip, mp);
4246 		for (mcb = mcip->mci_promisc_list; mcb != NULL;
4247 		    mcb = mcb->mcb_nextp) {
4248 			mpip = (mac_promisc_impl_t *)mcb->mcb_objp;
4249 			if (mpip->mpi_type == MAC_CLIENT_PROMISC_FILTERED &&
4250 			    !is_mcast) {
4251 				mac_promisc_dispatch_one(mpip, mp, B_FALSE,
4252 				    B_FALSE);
4253 			}
4254 		}
4255 	}
4256 	MAC_PROMISC_WALKER_DCR(mip);
4257 }
4258 
4259 /*
4260  * Return the margin value currently assigned to the specified MAC instance.
4261  */
4262 void
4263 mac_margin_get(mac_handle_t mh, uint32_t *marginp)
4264 {
4265 	mac_impl_t *mip = (mac_impl_t *)mh;
4266 
4267 	rw_enter(&(mip->mi_rw_lock), RW_READER);
4268 	*marginp = mip->mi_margin;
4269 	rw_exit(&(mip->mi_rw_lock));
4270 }
4271 
4272 /*
4273  * mac_info_get() is used for retrieving the mac_info when a DL_INFO_REQ is
4274  * issued before a DL_ATTACH_REQ. we walk the i_mac_impl_hash table and find
4275  * the first mac_impl_t with a matching driver name; then we copy its mac_info_t
4276  * to the caller. we do all this with i_mac_impl_lock held so the mac_impl_t
4277  * cannot disappear while we are accessing it.
4278  */
4279 typedef struct i_mac_info_state_s {
4280 	const char	*mi_name;
4281 	mac_info_t	*mi_infop;
4282 } i_mac_info_state_t;
4283 
4284 /*ARGSUSED*/
4285 static uint_t
4286 i_mac_info_walker(mod_hash_key_t key, mod_hash_val_t *val, void *arg)
4287 {
4288 	i_mac_info_state_t *statep = arg;
4289 	mac_impl_t *mip = (mac_impl_t *)val;
4290 
4291 	if (mip->mi_state_flags & MIS_DISABLED)
4292 		return (MH_WALK_CONTINUE);
4293 
4294 	if (strcmp(statep->mi_name,
4295 	    ddi_driver_name(mip->mi_dip)) != 0)
4296 		return (MH_WALK_CONTINUE);
4297 
4298 	statep->mi_infop = &mip->mi_info;
4299 	return (MH_WALK_TERMINATE);
4300 }
4301 
4302 boolean_t
4303 mac_info_get(const char *name, mac_info_t *minfop)
4304 {
4305 	i_mac_info_state_t state;
4306 
4307 	rw_enter(&i_mac_impl_lock, RW_READER);
4308 	state.mi_name = name;
4309 	state.mi_infop = NULL;
4310 	mod_hash_walk(i_mac_impl_hash, i_mac_info_walker, &state);
4311 	if (state.mi_infop == NULL) {
4312 		rw_exit(&i_mac_impl_lock);
4313 		return (B_FALSE);
4314 	}
4315 	*minfop = *state.mi_infop;
4316 	rw_exit(&i_mac_impl_lock);
4317 	return (B_TRUE);
4318 }
4319 
4320 /*
4321  * To get the capabilities that MAC layer cares about, such as rings, factory
4322  * mac address, vnic or not, it should directly invoke this function.  If the
4323  * link is part of a bridge, then the only "capability" it has is the inability
4324  * to do zero copy.
4325  */
4326 boolean_t
4327 i_mac_capab_get(mac_handle_t mh, mac_capab_t cap, void *cap_data)
4328 {
4329 	mac_impl_t *mip = (mac_impl_t *)mh;
4330 
4331 	if (mip->mi_bridge_link != NULL) {
4332 		return (cap == MAC_CAPAB_NO_ZCOPY);
4333 	} else if (mip->mi_callbacks->mc_callbacks & MC_GETCAPAB) {
4334 		boolean_t res;
4335 
4336 		res = mip->mi_getcapab(mip->mi_driver, cap, cap_data);
4337 		/*
4338 		 * Until we have suppport for TSOv6 emulation in the MAC
4339 		 * loopback path, do not allow the TSOv6 capability to be
4340 		 * advertised to consumers.
4341 		 */
4342 		if (res && cap == MAC_CAPAB_LSO) {
4343 			mac_capab_lso_t *cap_lso = cap_data;
4344 
4345 			cap_lso->lso_flags &= ~LSO_TX_BASIC_TCP_IPV6;
4346 			cap_lso->lso_basic_tcp_ipv6.lso_max = 0;
4347 		}
4348 		return (res);
4349 	} else {
4350 		return (B_FALSE);
4351 	}
4352 }
4353 
4354 /*
4355  * Capability query function. If number of active mac clients is greater than
4356  * 1, only limited capabilities can be advertised to the caller no matter the
4357  * driver has certain capability or not. Else, we query the driver to get the
4358  * capability.
4359  */
4360 boolean_t
4361 mac_capab_get(mac_handle_t mh, mac_capab_t cap, void *cap_data)
4362 {
4363 	mac_impl_t *mip = (mac_impl_t *)mh;
4364 
4365 	/*
4366 	 * Some capabilities are restricted when there are more than one active
4367 	 * clients on the MAC resource.  The ones noted below are safe,
4368 	 * independent of that count.
4369 	 */
4370 	if (mip->mi_nactiveclients > 1) {
4371 		switch (cap) {
4372 		case MAC_CAPAB_NO_ZCOPY:
4373 			return (B_TRUE);
4374 		case MAC_CAPAB_LEGACY:
4375 		case MAC_CAPAB_HCKSUM:
4376 		case MAC_CAPAB_LSO:
4377 		case MAC_CAPAB_NO_NATIVEVLAN:
4378 			break;
4379 		default:
4380 			return (B_FALSE);
4381 		}
4382 	}
4383 
4384 	/* else get capab from driver */
4385 	return (i_mac_capab_get(mh, cap, cap_data));
4386 }
4387 
4388 boolean_t
4389 mac_sap_verify(mac_handle_t mh, uint32_t sap, uint32_t *bind_sap)
4390 {
4391 	mac_impl_t *mip = (mac_impl_t *)mh;
4392 
4393 	return (mip->mi_type->mt_ops.mtops_sap_verify(sap, bind_sap,
4394 	    mip->mi_pdata));
4395 }
4396 
4397 mblk_t *
4398 mac_header(mac_handle_t mh, const uint8_t *daddr, uint32_t sap, mblk_t *payload,
4399     size_t extra_len)
4400 {
4401 	mac_impl_t	*mip = (mac_impl_t *)mh;
4402 	const uint8_t	*hdr_daddr;
4403 
4404 	/*
4405 	 * If the MAC is point-to-point with a fixed destination address, then
4406 	 * we must always use that destination in the MAC header.
4407 	 */
4408 	hdr_daddr = (mip->mi_dstaddr_set ? mip->mi_dstaddr : daddr);
4409 	return (mip->mi_type->mt_ops.mtops_header(mip->mi_addr, hdr_daddr, sap,
4410 	    mip->mi_pdata, payload, extra_len));
4411 }
4412 
4413 int
4414 mac_header_info(mac_handle_t mh, mblk_t *mp, mac_header_info_t *mhip)
4415 {
4416 	mac_impl_t *mip = (mac_impl_t *)mh;
4417 
4418 	return (mip->mi_type->mt_ops.mtops_header_info(mp, mip->mi_pdata,
4419 	    mhip));
4420 }
4421 
4422 int
4423 mac_vlan_header_info(mac_handle_t mh, mblk_t *mp, mac_header_info_t *mhip)
4424 {
4425 	mac_impl_t	*mip = (mac_impl_t *)mh;
4426 	boolean_t	is_ethernet = (mip->mi_info.mi_media == DL_ETHER);
4427 	int		err = 0;
4428 
4429 	/*
4430 	 * Packets should always be at least 16 bit aligned.
4431 	 */
4432 	ASSERT(IS_P2ALIGNED(mp->b_rptr, sizeof (uint16_t)));
4433 
4434 	if ((err = mac_header_info(mh, mp, mhip)) != 0)
4435 		return (err);
4436 
4437 	/*
4438 	 * If this is a VLAN-tagged Ethernet packet, then the SAP in the
4439 	 * mac_header_info_t as returned by mac_header_info() is
4440 	 * ETHERTYPE_VLAN. We need to grab the ethertype from the VLAN header.
4441 	 */
4442 	if (is_ethernet && (mhip->mhi_bindsap == ETHERTYPE_VLAN)) {
4443 		struct ether_vlan_header *evhp;
4444 		uint16_t sap;
4445 		mblk_t *tmp = NULL;
4446 		size_t size;
4447 
4448 		size = sizeof (struct ether_vlan_header);
4449 		if (MBLKL(mp) < size) {
4450 			/*
4451 			 * Pullup the message in order to get the MAC header
4452 			 * infomation. Note that this is a read-only function,
4453 			 * we keep the input packet intact.
4454 			 */
4455 			if ((tmp = msgpullup(mp, size)) == NULL)
4456 				return (EINVAL);
4457 
4458 			mp = tmp;
4459 		}
4460 		evhp = (struct ether_vlan_header *)mp->b_rptr;
4461 		sap = ntohs(evhp->ether_type);
4462 		(void) mac_sap_verify(mh, sap, &mhip->mhi_bindsap);
4463 		mhip->mhi_hdrsize = sizeof (struct ether_vlan_header);
4464 		mhip->mhi_tci = ntohs(evhp->ether_tci);
4465 		mhip->mhi_istagged = B_TRUE;
4466 		freemsg(tmp);
4467 
4468 		if (VLAN_CFI(mhip->mhi_tci) != ETHER_CFI)
4469 			return (EINVAL);
4470 	} else {
4471 		mhip->mhi_istagged = B_FALSE;
4472 		mhip->mhi_tci = 0;
4473 	}
4474 
4475 	return (0);
4476 }
4477 
4478 mblk_t *
4479 mac_header_cook(mac_handle_t mh, mblk_t *mp)
4480 {
4481 	mac_impl_t *mip = (mac_impl_t *)mh;
4482 
4483 	if (mip->mi_type->mt_ops.mtops_ops & MTOPS_HEADER_COOK) {
4484 		if (DB_REF(mp) > 1) {
4485 			mblk_t *newmp = copymsg(mp);
4486 			if (newmp == NULL)
4487 				return (NULL);
4488 			freemsg(mp);
4489 			mp = newmp;
4490 		}
4491 		return (mip->mi_type->mt_ops.mtops_header_cook(mp,
4492 		    mip->mi_pdata));
4493 	}
4494 	return (mp);
4495 }
4496 
4497 mblk_t *
4498 mac_header_uncook(mac_handle_t mh, mblk_t *mp)
4499 {
4500 	mac_impl_t *mip = (mac_impl_t *)mh;
4501 
4502 	if (mip->mi_type->mt_ops.mtops_ops & MTOPS_HEADER_UNCOOK) {
4503 		if (DB_REF(mp) > 1) {
4504 			mblk_t *newmp = copymsg(mp);
4505 			if (newmp == NULL)
4506 				return (NULL);
4507 			freemsg(mp);
4508 			mp = newmp;
4509 		}
4510 		return (mip->mi_type->mt_ops.mtops_header_uncook(mp,
4511 		    mip->mi_pdata));
4512 	}
4513 	return (mp);
4514 }
4515 
4516 uint_t
4517 mac_addr_len(mac_handle_t mh)
4518 {
4519 	mac_impl_t *mip = (mac_impl_t *)mh;
4520 
4521 	return (mip->mi_type->mt_addr_length);
4522 }
4523 
4524 /* True if a MAC is a VNIC */
4525 boolean_t
4526 mac_is_vnic(mac_handle_t mh)
4527 {
4528 	return ((((mac_impl_t *)mh)->mi_state_flags & MIS_IS_VNIC) != 0);
4529 }
4530 
4531 boolean_t
4532 mac_is_overlay(mac_handle_t mh)
4533 {
4534 	return ((((mac_impl_t *)mh)->mi_state_flags & MIS_IS_OVERLAY) != 0);
4535 }
4536 
4537 mac_handle_t
4538 mac_get_lower_mac_handle(mac_handle_t mh)
4539 {
4540 	mac_impl_t *mip = (mac_impl_t *)mh;
4541 
4542 	ASSERT(mac_is_vnic(mh));
4543 	return (((vnic_t *)mip->mi_driver)->vn_lower_mh);
4544 }
4545 
4546 boolean_t
4547 mac_is_vnic_primary(mac_handle_t mh)
4548 {
4549 	mac_impl_t *mip = (mac_impl_t *)mh;
4550 
4551 	ASSERT(mac_is_vnic(mh));
4552 	return (((vnic_t *)mip->mi_driver)->vn_addr_type ==
4553 	    VNIC_MAC_ADDR_TYPE_PRIMARY);
4554 }
4555 
4556 void
4557 mac_update_resources(mac_resource_props_t *nmrp, mac_resource_props_t *cmrp,
4558     boolean_t is_user_flow)
4559 {
4560 	if (nmrp != NULL && cmrp != NULL) {
4561 		if (nmrp->mrp_mask & MRP_PRIORITY) {
4562 			if (nmrp->mrp_priority == MPL_RESET) {
4563 				cmrp->mrp_mask &= ~MRP_PRIORITY;
4564 				if (is_user_flow) {
4565 					cmrp->mrp_priority =
4566 					    MPL_SUBFLOW_DEFAULT;
4567 				} else {
4568 					cmrp->mrp_priority = MPL_LINK_DEFAULT;
4569 				}
4570 			} else {
4571 				cmrp->mrp_mask |= MRP_PRIORITY;
4572 				cmrp->mrp_priority = nmrp->mrp_priority;
4573 			}
4574 		}
4575 		if (nmrp->mrp_mask & MRP_MAXBW) {
4576 			if (nmrp->mrp_maxbw == MRP_MAXBW_RESETVAL) {
4577 				cmrp->mrp_mask &= ~MRP_MAXBW;
4578 				cmrp->mrp_maxbw = 0;
4579 			} else {
4580 				cmrp->mrp_mask |= MRP_MAXBW;
4581 				cmrp->mrp_maxbw = nmrp->mrp_maxbw;
4582 			}
4583 		}
4584 		if (nmrp->mrp_mask & MRP_CPUS)
4585 			MAC_COPY_CPUS(nmrp, cmrp);
4586 
4587 		if (nmrp->mrp_mask & MRP_POOL) {
4588 			if (strlen(nmrp->mrp_pool) == 0) {
4589 				cmrp->mrp_mask &= ~MRP_POOL;
4590 				bzero(cmrp->mrp_pool, sizeof (cmrp->mrp_pool));
4591 			} else {
4592 				cmrp->mrp_mask |= MRP_POOL;
4593 				(void) strncpy(cmrp->mrp_pool, nmrp->mrp_pool,
4594 				    sizeof (cmrp->mrp_pool));
4595 			}
4596 
4597 		}
4598 
4599 		if (nmrp->mrp_mask & MRP_PROTECT)
4600 			mac_protect_update(nmrp, cmrp);
4601 
4602 		/*
4603 		 * Update the rings specified.
4604 		 */
4605 		if (nmrp->mrp_mask & MRP_RX_RINGS) {
4606 			if (nmrp->mrp_mask & MRP_RINGS_RESET) {
4607 				cmrp->mrp_mask &= ~MRP_RX_RINGS;
4608 				if (cmrp->mrp_mask & MRP_RXRINGS_UNSPEC)
4609 					cmrp->mrp_mask &= ~MRP_RXRINGS_UNSPEC;
4610 				cmrp->mrp_nrxrings = 0;
4611 			} else {
4612 				cmrp->mrp_mask |= MRP_RX_RINGS;
4613 				cmrp->mrp_nrxrings = nmrp->mrp_nrxrings;
4614 			}
4615 		}
4616 		if (nmrp->mrp_mask & MRP_TX_RINGS) {
4617 			if (nmrp->mrp_mask & MRP_RINGS_RESET) {
4618 				cmrp->mrp_mask &= ~MRP_TX_RINGS;
4619 				if (cmrp->mrp_mask & MRP_TXRINGS_UNSPEC)
4620 					cmrp->mrp_mask &= ~MRP_TXRINGS_UNSPEC;
4621 				cmrp->mrp_ntxrings = 0;
4622 			} else {
4623 				cmrp->mrp_mask |= MRP_TX_RINGS;
4624 				cmrp->mrp_ntxrings = nmrp->mrp_ntxrings;
4625 			}
4626 		}
4627 		if (nmrp->mrp_mask & MRP_RXRINGS_UNSPEC)
4628 			cmrp->mrp_mask |= MRP_RXRINGS_UNSPEC;
4629 		else if (cmrp->mrp_mask & MRP_RXRINGS_UNSPEC)
4630 			cmrp->mrp_mask &= ~MRP_RXRINGS_UNSPEC;
4631 
4632 		if (nmrp->mrp_mask & MRP_TXRINGS_UNSPEC)
4633 			cmrp->mrp_mask |= MRP_TXRINGS_UNSPEC;
4634 		else if (cmrp->mrp_mask & MRP_TXRINGS_UNSPEC)
4635 			cmrp->mrp_mask &= ~MRP_TXRINGS_UNSPEC;
4636 	}
4637 }
4638 
4639 /*
4640  * i_mac_set_resources:
4641  *
4642  * This routine associates properties with the primary MAC client of
4643  * the specified MAC instance.
4644  * - Cache the properties in mac_impl_t
4645  * - Apply the properties to the primary MAC client if exists
4646  */
4647 int
4648 i_mac_set_resources(mac_handle_t mh, mac_resource_props_t *mrp)
4649 {
4650 	mac_impl_t		*mip = (mac_impl_t *)mh;
4651 	mac_client_impl_t	*mcip;
4652 	int			err = 0;
4653 	uint32_t		resmask, newresmask;
4654 	mac_resource_props_t	*tmrp, *umrp;
4655 
4656 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
4657 
4658 	err = mac_validate_props(mip, mrp);
4659 	if (err != 0)
4660 		return (err);
4661 
4662 	umrp = kmem_zalloc(sizeof (*umrp), KM_SLEEP);
4663 	bcopy(&mip->mi_resource_props, umrp, sizeof (*umrp));
4664 	resmask = umrp->mrp_mask;
4665 	mac_update_resources(mrp, umrp, B_FALSE);
4666 	newresmask = umrp->mrp_mask;
4667 
4668 	if (resmask == 0 && newresmask != 0) {
4669 		/*
4670 		 * Bandwidth, priority, cpu or pool link properties configured,
4671 		 * must disable fastpath.
4672 		 */
4673 		if ((err = mac_fastpath_disable((mac_handle_t)mip)) != 0) {
4674 			kmem_free(umrp, sizeof (*umrp));
4675 			return (err);
4676 		}
4677 	}
4678 
4679 	/*
4680 	 * Since bind_cpu may be modified by mac_client_set_resources()
4681 	 * we use a copy of bind_cpu and finally cache bind_cpu in mip.
4682 	 * This allows us to cache only user edits in mip.
4683 	 */
4684 	tmrp = kmem_zalloc(sizeof (*tmrp), KM_SLEEP);
4685 	bcopy(mrp, tmrp, sizeof (*tmrp));
4686 	mcip = mac_primary_client_handle(mip);
4687 	if (mcip != NULL && (mcip->mci_state_flags & MCIS_IS_AGGR_PORT) == 0) {
4688 		err = mac_client_set_resources((mac_client_handle_t)mcip, tmrp);
4689 	} else if ((mrp->mrp_mask & MRP_RX_RINGS ||
4690 	    mrp->mrp_mask & MRP_TX_RINGS)) {
4691 		mac_client_impl_t	*vmcip;
4692 
4693 		/*
4694 		 * If the primary is not up, we need to check if there
4695 		 * are any VLANs on this primary. If there are then
4696 		 * we need to set this property on the VLANs since
4697 		 * VLANs follow the primary they are based on. Just
4698 		 * look for the first VLAN and change its properties,
4699 		 * all the other VLANs should be in the same group.
4700 		 */
4701 		for (vmcip = mip->mi_clients_list; vmcip != NULL;
4702 		    vmcip = vmcip->mci_client_next) {
4703 			if ((vmcip->mci_flent->fe_type & FLOW_PRIMARY_MAC) &&
4704 			    mac_client_vid((mac_client_handle_t)vmcip) !=
4705 			    VLAN_ID_NONE) {
4706 				break;
4707 			}
4708 		}
4709 		if (vmcip != NULL) {
4710 			mac_resource_props_t	*omrp;
4711 			mac_resource_props_t	*vmrp;
4712 
4713 			omrp = kmem_zalloc(sizeof (*omrp), KM_SLEEP);
4714 			bcopy(MCIP_RESOURCE_PROPS(vmcip), omrp, sizeof (*omrp));
4715 			/*
4716 			 * We dont' call mac_update_resources since we
4717 			 * want to take only the ring properties and
4718 			 * not all the properties that may have changed.
4719 			 */
4720 			vmrp = MCIP_RESOURCE_PROPS(vmcip);
4721 			if (mrp->mrp_mask & MRP_RX_RINGS) {
4722 				if (mrp->mrp_mask & MRP_RINGS_RESET) {
4723 					vmrp->mrp_mask &= ~MRP_RX_RINGS;
4724 					if (vmrp->mrp_mask &
4725 					    MRP_RXRINGS_UNSPEC) {
4726 						vmrp->mrp_mask &=
4727 						    ~MRP_RXRINGS_UNSPEC;
4728 					}
4729 					vmrp->mrp_nrxrings = 0;
4730 				} else {
4731 					vmrp->mrp_mask |= MRP_RX_RINGS;
4732 					vmrp->mrp_nrxrings = mrp->mrp_nrxrings;
4733 				}
4734 			}
4735 			if (mrp->mrp_mask & MRP_TX_RINGS) {
4736 				if (mrp->mrp_mask & MRP_RINGS_RESET) {
4737 					vmrp->mrp_mask &= ~MRP_TX_RINGS;
4738 					if (vmrp->mrp_mask &
4739 					    MRP_TXRINGS_UNSPEC) {
4740 						vmrp->mrp_mask &=
4741 						    ~MRP_TXRINGS_UNSPEC;
4742 					}
4743 					vmrp->mrp_ntxrings = 0;
4744 				} else {
4745 					vmrp->mrp_mask |= MRP_TX_RINGS;
4746 					vmrp->mrp_ntxrings = mrp->mrp_ntxrings;
4747 				}
4748 			}
4749 			if (mrp->mrp_mask & MRP_RXRINGS_UNSPEC)
4750 				vmrp->mrp_mask |= MRP_RXRINGS_UNSPEC;
4751 
4752 			if (mrp->mrp_mask & MRP_TXRINGS_UNSPEC)
4753 				vmrp->mrp_mask |= MRP_TXRINGS_UNSPEC;
4754 
4755 			if ((err = mac_client_set_rings_prop(vmcip, mrp,
4756 			    omrp)) != 0) {
4757 				bcopy(omrp, MCIP_RESOURCE_PROPS(vmcip),
4758 				    sizeof (*omrp));
4759 			} else {
4760 				mac_set_prim_vlan_rings(mip, vmrp);
4761 			}
4762 			kmem_free(omrp, sizeof (*omrp));
4763 		}
4764 	}
4765 
4766 	/* Only update the values if mac_client_set_resources succeeded */
4767 	if (err == 0) {
4768 		bcopy(umrp, &mip->mi_resource_props, sizeof (*umrp));
4769 		/*
4770 		 * If bandwidth, priority or cpu link properties cleared,
4771 		 * renable fastpath.
4772 		 */
4773 		if (resmask != 0 && newresmask == 0)
4774 			mac_fastpath_enable((mac_handle_t)mip);
4775 	} else if (resmask == 0 && newresmask != 0) {
4776 		mac_fastpath_enable((mac_handle_t)mip);
4777 	}
4778 	kmem_free(tmrp, sizeof (*tmrp));
4779 	kmem_free(umrp, sizeof (*umrp));
4780 	return (err);
4781 }
4782 
4783 int
4784 mac_set_resources(mac_handle_t mh, mac_resource_props_t *mrp)
4785 {
4786 	int err;
4787 
4788 	i_mac_perim_enter((mac_impl_t *)mh);
4789 	err = i_mac_set_resources(mh, mrp);
4790 	i_mac_perim_exit((mac_impl_t *)mh);
4791 	return (err);
4792 }
4793 
4794 /*
4795  * Get the properties cached for the specified MAC instance.
4796  */
4797 void
4798 mac_get_resources(mac_handle_t mh, mac_resource_props_t *mrp)
4799 {
4800 	mac_impl_t		*mip = (mac_impl_t *)mh;
4801 	mac_client_impl_t	*mcip;
4802 
4803 	mcip = mac_primary_client_handle(mip);
4804 	if (mcip != NULL) {
4805 		mac_client_get_resources((mac_client_handle_t)mcip, mrp);
4806 		return;
4807 	}
4808 	bcopy(&mip->mi_resource_props, mrp, sizeof (mac_resource_props_t));
4809 }
4810 
4811 /*
4812  * Get the effective properties from the primary client of the
4813  * specified MAC instance.
4814  */
4815 void
4816 mac_get_effective_resources(mac_handle_t mh, mac_resource_props_t *mrp)
4817 {
4818 	mac_impl_t		*mip = (mac_impl_t *)mh;
4819 	mac_client_impl_t	*mcip;
4820 
4821 	mcip = mac_primary_client_handle(mip);
4822 	if (mcip != NULL) {
4823 		mac_client_get_effective_resources((mac_client_handle_t)mcip,
4824 		    mrp);
4825 		return;
4826 	}
4827 	bzero(mrp, sizeof (mac_resource_props_t));
4828 }
4829 
4830 int
4831 mac_set_pvid(mac_handle_t mh, uint16_t pvid)
4832 {
4833 	mac_impl_t *mip = (mac_impl_t *)mh;
4834 	mac_client_impl_t *mcip;
4835 	mac_unicast_impl_t *muip;
4836 
4837 	i_mac_perim_enter(mip);
4838 	if (pvid != 0) {
4839 		for (mcip = mip->mi_clients_list; mcip != NULL;
4840 		    mcip = mcip->mci_client_next) {
4841 			for (muip = mcip->mci_unicast_list; muip != NULL;
4842 			    muip = muip->mui_next) {
4843 				if (muip->mui_vid == pvid) {
4844 					i_mac_perim_exit(mip);
4845 					return (EBUSY);
4846 				}
4847 			}
4848 		}
4849 	}
4850 	mip->mi_pvid = pvid;
4851 	i_mac_perim_exit(mip);
4852 	return (0);
4853 }
4854 
4855 uint16_t
4856 mac_get_pvid(mac_handle_t mh)
4857 {
4858 	mac_impl_t *mip = (mac_impl_t *)mh;
4859 
4860 	return (mip->mi_pvid);
4861 }
4862 
4863 uint32_t
4864 mac_get_llimit(mac_handle_t mh)
4865 {
4866 	mac_impl_t *mip = (mac_impl_t *)mh;
4867 
4868 	return (mip->mi_llimit);
4869 }
4870 
4871 uint32_t
4872 mac_get_ldecay(mac_handle_t mh)
4873 {
4874 	mac_impl_t *mip = (mac_impl_t *)mh;
4875 
4876 	return (mip->mi_ldecay);
4877 }
4878 
4879 /*
4880  * Rename a mac client, its flow, and the kstat.
4881  */
4882 int
4883 mac_rename_primary(mac_handle_t mh, const char *new_name)
4884 {
4885 	mac_impl_t		*mip = (mac_impl_t *)mh;
4886 	mac_client_impl_t	*cur_clnt = NULL;
4887 	flow_entry_t		*fep;
4888 
4889 	i_mac_perim_enter(mip);
4890 
4891 	/*
4892 	 * VNICs: we need to change the sys flow name and
4893 	 * the associated flow kstat.
4894 	 */
4895 	if (mip->mi_state_flags & MIS_IS_VNIC) {
4896 		mac_client_impl_t *mcip = mac_vnic_lower(mip);
4897 		ASSERT(new_name != NULL);
4898 		mac_rename_flow_names(mcip, new_name);
4899 		mac_stat_rename(mcip);
4900 		goto done;
4901 	}
4902 	/*
4903 	 * This mac may itself be an aggr link, or it may have some client
4904 	 * which is an aggr port. For both cases, we need to change the
4905 	 * aggr port's mac client name, its flow name and the associated flow
4906 	 * kstat.
4907 	 */
4908 	if (mip->mi_state_flags & MIS_IS_AGGR) {
4909 		mac_capab_aggr_t aggr_cap;
4910 		mac_rename_fn_t rename_fn;
4911 		boolean_t ret;
4912 
4913 		ASSERT(new_name != NULL);
4914 		ret = i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_AGGR,
4915 		    (void *)(&aggr_cap));
4916 		ASSERT(ret == B_TRUE);
4917 		rename_fn = aggr_cap.mca_rename_fn;
4918 		rename_fn(new_name, mip->mi_driver);
4919 		/*
4920 		 * The aggr's client name and kstat flow name will be
4921 		 * updated below, i.e. via mac_rename_flow_names.
4922 		 */
4923 	}
4924 
4925 	for (cur_clnt = mip->mi_clients_list; cur_clnt != NULL;
4926 	    cur_clnt = cur_clnt->mci_client_next) {
4927 		if (cur_clnt->mci_state_flags & MCIS_IS_AGGR_PORT) {
4928 			if (new_name != NULL) {
4929 				char *str_st = cur_clnt->mci_name;
4930 				char *str_del = strchr(str_st, '-');
4931 
4932 				ASSERT(str_del != NULL);
4933 				bzero(str_del + 1, MAXNAMELEN -
4934 				    (str_del - str_st + 1));
4935 				bcopy(new_name, str_del + 1,
4936 				    strlen(new_name));
4937 			}
4938 			fep = cur_clnt->mci_flent;
4939 			mac_rename_flow(fep, cur_clnt->mci_name);
4940 			break;
4941 		} else if (new_name != NULL &&
4942 		    cur_clnt->mci_state_flags & MCIS_USE_DATALINK_NAME) {
4943 			mac_rename_flow_names(cur_clnt, new_name);
4944 			break;
4945 		}
4946 	}
4947 
4948 	/* Recreate kstats associated with aggr pseudo rings */
4949 	if (mip->mi_state_flags & MIS_IS_AGGR)
4950 		mac_pseudo_ring_stat_rename(mip);
4951 
4952 done:
4953 	i_mac_perim_exit(mip);
4954 	return (0);
4955 }
4956 
4957 /*
4958  * Rename the MAC client's flow names
4959  */
4960 static void
4961 mac_rename_flow_names(mac_client_impl_t *mcip, const char *new_name)
4962 {
4963 	flow_entry_t	*flent;
4964 	uint16_t	vid;
4965 	char		flowname[MAXFLOWNAMELEN];
4966 	mac_impl_t	*mip = mcip->mci_mip;
4967 
4968 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
4969 
4970 	/*
4971 	 * Use mi_rw_lock to ensure that threads not in the mac perimeter
4972 	 * see a self-consistent value for mci_name
4973 	 */
4974 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
4975 	(void) strlcpy(mcip->mci_name, new_name, sizeof (mcip->mci_name));
4976 	rw_exit(&mip->mi_rw_lock);
4977 
4978 	mac_rename_flow(mcip->mci_flent, new_name);
4979 
4980 	if (mcip->mci_nflents == 1)
4981 		return;
4982 
4983 	/*
4984 	 * We have to rename all the others too, no stats to destroy for
4985 	 * these.
4986 	 */
4987 	for (flent = mcip->mci_flent_list; flent != NULL;
4988 	    flent = flent->fe_client_next) {
4989 		if (flent != mcip->mci_flent) {
4990 			vid = i_mac_flow_vid(flent);
4991 			(void) sprintf(flowname, "%s%u", new_name, vid);
4992 			mac_flow_set_name(flent, flowname);
4993 		}
4994 	}
4995 }
4996 
4997 
4998 /*
4999  * Add a flow to the MAC client's flow list - i.e list of MAC/VID tuples
5000  * defined for the specified MAC client.
5001  */
5002 static void
5003 mac_client_add_to_flow_list(mac_client_impl_t *mcip, flow_entry_t *flent)
5004 {
5005 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
5006 	/*
5007 	 * The promisc Rx data path walks the mci_flent_list. Protect by
5008 	 * using mi_rw_lock
5009 	 */
5010 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
5011 
5012 	mcip->mci_vidcache = MCIP_VIDCACHE_INVALID;
5013 
5014 	/* Add it to the head */
5015 	flent->fe_client_next = mcip->mci_flent_list;
5016 	mcip->mci_flent_list = flent;
5017 	mcip->mci_nflents++;
5018 
5019 	/*
5020 	 * Keep track of the number of non-zero VIDs addresses per MAC
5021 	 * client to avoid figuring it out in the data-path.
5022 	 */
5023 	if (i_mac_flow_vid(flent) != VLAN_ID_NONE)
5024 		mcip->mci_nvids++;
5025 
5026 	rw_exit(&mcip->mci_rw_lock);
5027 }
5028 
5029 /*
5030  * Remove a flow entry from the MAC client's list.
5031  */
5032 static void
5033 mac_client_remove_flow_from_list(mac_client_impl_t *mcip, flow_entry_t *flent)
5034 {
5035 	flow_entry_t	*fe = mcip->mci_flent_list;
5036 	flow_entry_t	*prev_fe = NULL;
5037 
5038 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
5039 	/*
5040 	 * The promisc Rx data path walks the mci_flent_list. Protect by
5041 	 * using mci_rw_lock
5042 	 */
5043 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
5044 	mcip->mci_vidcache = MCIP_VIDCACHE_INVALID;
5045 
5046 	while ((fe != NULL) && (fe != flent)) {
5047 		prev_fe = fe;
5048 		fe = fe->fe_client_next;
5049 	}
5050 
5051 	ASSERT(fe != NULL);
5052 	if (prev_fe == NULL) {
5053 		/* Deleting the first node */
5054 		mcip->mci_flent_list = fe->fe_client_next;
5055 	} else {
5056 		prev_fe->fe_client_next = fe->fe_client_next;
5057 	}
5058 	mcip->mci_nflents--;
5059 
5060 	if (i_mac_flow_vid(flent) != VLAN_ID_NONE)
5061 		mcip->mci_nvids--;
5062 
5063 	rw_exit(&mcip->mci_rw_lock);
5064 }
5065 
5066 /*
5067  * Check if the given VID belongs to this MAC client.
5068  */
5069 boolean_t
5070 mac_client_check_flow_vid(mac_client_impl_t *mcip, uint16_t vid)
5071 {
5072 	flow_entry_t	*flent;
5073 	uint16_t	mci_vid;
5074 	uint32_t	cache = mcip->mci_vidcache;
5075 
5076 	/*
5077 	 * In hopes of not having to touch the mci_rw_lock, check to see if
5078 	 * this vid matches our cached result.
5079 	 */
5080 	if (MCIP_VIDCACHE_ISVALID(cache) && MCIP_VIDCACHE_VID(cache) == vid)
5081 		return (MCIP_VIDCACHE_BOOL(cache) ? B_TRUE : B_FALSE);
5082 
5083 	/* The mci_flent_list is protected by mci_rw_lock */
5084 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
5085 	for (flent = mcip->mci_flent_list; flent != NULL;
5086 	    flent = flent->fe_client_next) {
5087 		mci_vid = i_mac_flow_vid(flent);
5088 		if (vid == mci_vid) {
5089 			mcip->mci_vidcache = MCIP_VIDCACHE_CACHE(vid, B_TRUE);
5090 			rw_exit(&mcip->mci_rw_lock);
5091 			return (B_TRUE);
5092 		}
5093 	}
5094 
5095 	mcip->mci_vidcache = MCIP_VIDCACHE_CACHE(vid, B_FALSE);
5096 	rw_exit(&mcip->mci_rw_lock);
5097 	return (B_FALSE);
5098 }
5099 
5100 /*
5101  * Get the flow entry for the specified <MAC addr, VID> tuple.
5102  */
5103 static flow_entry_t *
5104 mac_client_get_flow(mac_client_impl_t *mcip, mac_unicast_impl_t *muip)
5105 {
5106 	mac_address_t *map = mcip->mci_unicast;
5107 	flow_entry_t *flent;
5108 	uint16_t vid;
5109 	flow_desc_t flow_desc;
5110 
5111 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
5112 
5113 	mac_flow_get_desc(mcip->mci_flent, &flow_desc);
5114 	if (bcmp(flow_desc.fd_dst_mac, map->ma_addr, map->ma_len) != 0)
5115 		return (NULL);
5116 
5117 	for (flent = mcip->mci_flent_list; flent != NULL;
5118 	    flent = flent->fe_client_next) {
5119 		vid = i_mac_flow_vid(flent);
5120 		if (vid == muip->mui_vid) {
5121 			return (flent);
5122 		}
5123 	}
5124 
5125 	return (NULL);
5126 }
5127 
5128 /*
5129  * Since mci_flent has the SRSs, when we want to remove it, we replace
5130  * the flow_desc_t in mci_flent with that of an existing flent and then
5131  * remove that flent instead of mci_flent.
5132  */
5133 static flow_entry_t *
5134 mac_client_swap_mciflent(mac_client_impl_t *mcip)
5135 {
5136 	flow_entry_t	*flent = mcip->mci_flent;
5137 	flow_tab_t	*ft = flent->fe_flow_tab;
5138 	flow_entry_t	*flent1;
5139 	flow_desc_t	fl_desc;
5140 	char		fl_name[MAXFLOWNAMELEN];
5141 	int		err;
5142 
5143 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
5144 	ASSERT(mcip->mci_nflents > 1);
5145 
5146 	/* get the next flent following the primary flent  */
5147 	flent1 = mcip->mci_flent_list->fe_client_next;
5148 	ASSERT(flent1 != NULL && flent1->fe_flow_tab == ft);
5149 
5150 	/*
5151 	 * Remove the flent from the flow table before updating the
5152 	 * flow descriptor as the hash depends on the flow descriptor.
5153 	 * This also helps incoming packet classification avoid having
5154 	 * to grab fe_lock. Access to fe_flow_desc of a flent not in the
5155 	 * flow table is done under the fe_lock so that log or stat functions
5156 	 * see a self-consistent fe_flow_desc. The name and desc are specific
5157 	 * to a flow, the rest are shared by all the clients, including
5158 	 * resource control etc.
5159 	 */
5160 	mac_flow_remove(ft, flent, B_TRUE);
5161 	mac_flow_remove(ft, flent1, B_TRUE);
5162 
5163 	bcopy(&flent->fe_flow_desc, &fl_desc, sizeof (flow_desc_t));
5164 	bcopy(flent->fe_flow_name, fl_name, MAXFLOWNAMELEN);
5165 
5166 	/* update the primary flow entry */
5167 	mutex_enter(&flent->fe_lock);
5168 	bcopy(&flent1->fe_flow_desc, &flent->fe_flow_desc,
5169 	    sizeof (flow_desc_t));
5170 	bcopy(&flent1->fe_flow_name, &flent->fe_flow_name, MAXFLOWNAMELEN);
5171 	mutex_exit(&flent->fe_lock);
5172 
5173 	/* update the flow entry that is to be freed */
5174 	mutex_enter(&flent1->fe_lock);
5175 	bcopy(&fl_desc, &flent1->fe_flow_desc, sizeof (flow_desc_t));
5176 	bcopy(fl_name, &flent1->fe_flow_name, MAXFLOWNAMELEN);
5177 	mutex_exit(&flent1->fe_lock);
5178 
5179 	/* now reinsert the flow entries in the table */
5180 	err = mac_flow_add(ft, flent);
5181 	ASSERT(err == 0);
5182 
5183 	err = mac_flow_add(ft, flent1);
5184 	ASSERT(err == 0);
5185 
5186 	return (flent1);
5187 }
5188 
5189 /*
5190  * Return whether there is only one flow entry associated with this
5191  * MAC client.
5192  */
5193 static boolean_t
5194 mac_client_single_rcvr(mac_client_impl_t *mcip)
5195 {
5196 	return (mcip->mci_nflents == 1);
5197 }
5198 
5199 int
5200 mac_validate_props(mac_impl_t *mip, mac_resource_props_t *mrp)
5201 {
5202 	boolean_t		reset;
5203 	uint32_t		rings_needed;
5204 	uint32_t		rings_avail;
5205 	mac_group_type_t	gtype;
5206 	mac_resource_props_t	*mip_mrp;
5207 
5208 	if (mrp == NULL)
5209 		return (0);
5210 
5211 	if (mrp->mrp_mask & MRP_PRIORITY) {
5212 		mac_priority_level_t	pri = mrp->mrp_priority;
5213 
5214 		if (pri < MPL_LOW || pri > MPL_RESET)
5215 			return (EINVAL);
5216 	}
5217 
5218 	if (mrp->mrp_mask & MRP_MAXBW) {
5219 		uint64_t maxbw = mrp->mrp_maxbw;
5220 
5221 		if (maxbw < MRP_MAXBW_MINVAL && maxbw != 0)
5222 			return (EINVAL);
5223 	}
5224 	if (mrp->mrp_mask & MRP_CPUS) {
5225 		int i, j;
5226 		mac_cpu_mode_t	fanout;
5227 
5228 		if (mrp->mrp_ncpus > ncpus)
5229 			return (EINVAL);
5230 
5231 		for (i = 0; i < mrp->mrp_ncpus; i++) {
5232 			for (j = 0; j < mrp->mrp_ncpus; j++) {
5233 				if (i != j &&
5234 				    mrp->mrp_cpu[i] == mrp->mrp_cpu[j]) {
5235 					return (EINVAL);
5236 				}
5237 			}
5238 		}
5239 
5240 		for (i = 0; i < mrp->mrp_ncpus; i++) {
5241 			cpu_t *cp;
5242 			int rv;
5243 
5244 			mutex_enter(&cpu_lock);
5245 			cp = cpu_get(mrp->mrp_cpu[i]);
5246 			if (cp != NULL)
5247 				rv = cpu_is_online(cp);
5248 			else
5249 				rv = 0;
5250 			mutex_exit(&cpu_lock);
5251 			if (rv == 0)
5252 				return (EINVAL);
5253 		}
5254 
5255 		fanout = mrp->mrp_fanout_mode;
5256 		if (fanout < 0 || fanout > MCM_CPUS)
5257 			return (EINVAL);
5258 	}
5259 
5260 	if (mrp->mrp_mask & MRP_PROTECT) {
5261 		int err = mac_protect_validate(mrp);
5262 		if (err != 0)
5263 			return (err);
5264 	}
5265 
5266 	if (!(mrp->mrp_mask & MRP_RX_RINGS) &&
5267 	    !(mrp->mrp_mask & MRP_TX_RINGS)) {
5268 		return (0);
5269 	}
5270 
5271 	/*
5272 	 * mip will be null when we come from mac_flow_create or
5273 	 * mac_link_flow_modify. In the latter case it is a user flow,
5274 	 * for which we don't support rings. In the former we would
5275 	 * have validated the props beforehand (i_mac_unicast_add ->
5276 	 * mac_client_set_resources -> validate for the primary and
5277 	 * vnic_dev_create -> mac_client_set_resources -> validate for
5278 	 * a vnic.
5279 	 */
5280 	if (mip == NULL)
5281 		return (0);
5282 
5283 	/*
5284 	 * We don't support setting rings property for a VNIC that is using a
5285 	 * primary address (VLAN)
5286 	 */
5287 	if ((mip->mi_state_flags & MIS_IS_VNIC) &&
5288 	    mac_is_vnic_primary((mac_handle_t)mip)) {
5289 		return (ENOTSUP);
5290 	}
5291 
5292 	mip_mrp = &mip->mi_resource_props;
5293 	/*
5294 	 * The rings property should be validated against the NICs
5295 	 * resources
5296 	 */
5297 	if (mip->mi_state_flags & MIS_IS_VNIC)
5298 		mip = (mac_impl_t *)mac_get_lower_mac_handle((mac_handle_t)mip);
5299 
5300 	reset = mrp->mrp_mask & MRP_RINGS_RESET;
5301 	/*
5302 	 * If groups are not supported, return error.
5303 	 */
5304 	if (((mrp->mrp_mask & MRP_RX_RINGS) && mip->mi_rx_groups == NULL) ||
5305 	    ((mrp->mrp_mask & MRP_TX_RINGS) && mip->mi_tx_groups == NULL)) {
5306 		return (EINVAL);
5307 	}
5308 	/*
5309 	 * If we are just resetting, there is no validation needed.
5310 	 */
5311 	if (reset)
5312 		return (0);
5313 
5314 	if (mrp->mrp_mask & MRP_RX_RINGS) {
5315 		rings_needed = mrp->mrp_nrxrings;
5316 		/*
5317 		 * We just want to check if the number of additional
5318 		 * rings requested is available.
5319 		 */
5320 		if (mip_mrp->mrp_mask & MRP_RX_RINGS) {
5321 			if (mrp->mrp_nrxrings > mip_mrp->mrp_nrxrings)
5322 				/* Just check for the additional rings */
5323 				rings_needed -= mip_mrp->mrp_nrxrings;
5324 			else
5325 				/* We are not asking for additional rings */
5326 				rings_needed = 0;
5327 		}
5328 		rings_avail = mip->mi_rxrings_avail;
5329 		gtype = mip->mi_rx_group_type;
5330 	} else {
5331 		rings_needed = mrp->mrp_ntxrings;
5332 		/* Similarly for the TX rings */
5333 		if (mip_mrp->mrp_mask & MRP_TX_RINGS) {
5334 			if (mrp->mrp_ntxrings > mip_mrp->mrp_ntxrings)
5335 				/* Just check for the additional rings */
5336 				rings_needed -= mip_mrp->mrp_ntxrings;
5337 			else
5338 				/* We are not asking for additional rings */
5339 				rings_needed = 0;
5340 		}
5341 		rings_avail = mip->mi_txrings_avail;
5342 		gtype = mip->mi_tx_group_type;
5343 	}
5344 
5345 	/* Error if the group is dynamic .. */
5346 	if (gtype == MAC_GROUP_TYPE_DYNAMIC) {
5347 		/*
5348 		 * .. and rings specified are more than available.
5349 		 */
5350 		if (rings_needed > rings_avail)
5351 			return (EINVAL);
5352 	} else {
5353 		/*
5354 		 * OR group is static and we have specified some rings.
5355 		 */
5356 		if (rings_needed > 0)
5357 			return (EINVAL);
5358 	}
5359 	return (0);
5360 }
5361 
5362 /*
5363  * Send a MAC_NOTE_LINK notification to all the MAC clients whenever the
5364  * underlying physical link is down. This is to allow MAC clients to
5365  * communicate with other clients.
5366  */
5367 void
5368 mac_virtual_link_update(mac_impl_t *mip)
5369 {
5370 	if (mip->mi_linkstate != LINK_STATE_UP)
5371 		i_mac_notify(mip, MAC_NOTE_LINK);
5372 }
5373 
5374 /*
5375  * For clients that have a pass-thru MAC, e.g. VNIC, we set the VNIC's
5376  * mac handle in the client.
5377  */
5378 void
5379 mac_set_upper_mac(mac_client_handle_t mch, mac_handle_t mh,
5380     mac_resource_props_t *mrp)
5381 {
5382 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
5383 	mac_impl_t		*mip = (mac_impl_t *)mh;
5384 
5385 	mcip->mci_upper_mip = mip;
5386 	/* If there are any properties, copy it over too */
5387 	if (mrp != NULL) {
5388 		bcopy(mrp, &mip->mi_resource_props,
5389 		    sizeof (mac_resource_props_t));
5390 	}
5391 }
5392 
5393 /*
5394  * Mark the mac as being used exclusively by the single mac client that is
5395  * doing some control operation on this mac. No further opens of this mac
5396  * will be allowed until this client calls mac_unmark_exclusive. The mac
5397  * client calling this function must already be in the mac perimeter
5398  */
5399 int
5400 mac_mark_exclusive(mac_handle_t mh)
5401 {
5402 	mac_impl_t	*mip = (mac_impl_t *)mh;
5403 
5404 	ASSERT(MAC_PERIM_HELD(mh));
5405 	/*
5406 	 * Look up its entry in the global hash table.
5407 	 */
5408 	rw_enter(&i_mac_impl_lock, RW_WRITER);
5409 	if (mip->mi_state_flags & MIS_DISABLED) {
5410 		rw_exit(&i_mac_impl_lock);
5411 		return (ENOENT);
5412 	}
5413 
5414 	/*
5415 	 * A reference to mac is held even if the link is not plumbed.
5416 	 * In i_dls_link_create() we open the MAC interface and hold the
5417 	 * reference. There is an additional reference for the mac_open
5418 	 * done in acquiring the mac perimeter
5419 	 */
5420 	if (mip->mi_ref != 2) {
5421 		rw_exit(&i_mac_impl_lock);
5422 		return (EBUSY);
5423 	}
5424 
5425 	ASSERT(!(mip->mi_state_flags & MIS_EXCLUSIVE_HELD));
5426 	mip->mi_state_flags |= MIS_EXCLUSIVE_HELD;
5427 	rw_exit(&i_mac_impl_lock);
5428 	return (0);
5429 }
5430 
5431 void
5432 mac_unmark_exclusive(mac_handle_t mh)
5433 {
5434 	mac_impl_t	*mip = (mac_impl_t *)mh;
5435 
5436 	ASSERT(MAC_PERIM_HELD(mh));
5437 
5438 	rw_enter(&i_mac_impl_lock, RW_WRITER);
5439 	/* 1 for the creation and another for the perimeter */
5440 	ASSERT(mip->mi_ref == 2 && (mip->mi_state_flags & MIS_EXCLUSIVE_HELD));
5441 	mip->mi_state_flags &= ~MIS_EXCLUSIVE_HELD;
5442 	rw_exit(&i_mac_impl_lock);
5443 }
5444 
5445 /*
5446  * Set the MTU for the specified MAC.
5447  */
5448 int
5449 mac_set_mtu(mac_handle_t mh, uint_t new_mtu, uint_t *old_mtu_arg)
5450 {
5451 	mac_impl_t *mip = (mac_impl_t *)mh;
5452 	uint_t old_mtu;
5453 	int rv = 0;
5454 
5455 	i_mac_perim_enter(mip);
5456 
5457 	if (!(mip->mi_callbacks->mc_callbacks & (MC_SETPROP|MC_GETPROP))) {
5458 		rv = ENOTSUP;
5459 		goto bail;
5460 	}
5461 
5462 	old_mtu = mip->mi_sdu_max;
5463 
5464 	if (new_mtu == 0 || new_mtu < mip->mi_sdu_min) {
5465 		rv = EINVAL;
5466 		goto bail;
5467 	}
5468 
5469 	rw_enter(&mip->mi_rw_lock, RW_READER);
5470 	if (mip->mi_mtrp != NULL && new_mtu < mip->mi_mtrp->mtr_mtu) {
5471 		rv = EBUSY;
5472 		rw_exit(&mip->mi_rw_lock);
5473 		goto bail;
5474 	}
5475 	rw_exit(&mip->mi_rw_lock);
5476 
5477 	if (old_mtu != new_mtu) {
5478 		rv = mip->mi_callbacks->mc_setprop(mip->mi_driver,
5479 		    "mtu", MAC_PROP_MTU, sizeof (uint_t), &new_mtu);
5480 		if (rv != 0)
5481 			goto bail;
5482 		rv = mac_maxsdu_update(mh, new_mtu);
5483 		ASSERT(rv == 0);
5484 	}
5485 
5486 bail:
5487 	i_mac_perim_exit(mip);
5488 
5489 	if (rv == 0 && old_mtu_arg != NULL)
5490 		*old_mtu_arg = old_mtu;
5491 	return (rv);
5492 }
5493 
5494 /*
5495  * Return the RX h/w information for the group indexed by grp_num.
5496  */
5497 void
5498 mac_get_hwrxgrp_info(mac_handle_t mh, int grp_index, uint_t *grp_num,
5499     uint_t *n_rings, uint_t *rings, uint_t *type, uint_t *n_clnts,
5500     char *clnts_name)
5501 {
5502 	mac_impl_t *mip = (mac_impl_t *)mh;
5503 	mac_grp_client_t *mcip;
5504 	uint_t i = 0, index = 0;
5505 	mac_ring_t	*ring;
5506 
5507 	/* Revisit when we implement fully dynamic group allocation */
5508 	ASSERT(grp_index >= 0 && grp_index < mip->mi_rx_group_count);
5509 
5510 	rw_enter(&mip->mi_rw_lock, RW_READER);
5511 	*grp_num = mip->mi_rx_groups[grp_index].mrg_index;
5512 	*type = mip->mi_rx_groups[grp_index].mrg_type;
5513 	*n_rings = mip->mi_rx_groups[grp_index].mrg_cur_count;
5514 	ring = mip->mi_rx_groups[grp_index].mrg_rings;
5515 	for (index = 0; index < mip->mi_rx_groups[grp_index].mrg_cur_count;
5516 	    index++) {
5517 		rings[index] = ring->mr_index;
5518 		ring = ring->mr_next;
5519 	}
5520 	/* Assuming the 1st is the default group */
5521 	index = 0;
5522 	if (grp_index == 0) {
5523 		(void) strlcpy(clnts_name, "<default,mcast>,",
5524 		    MAXCLIENTNAMELEN);
5525 		index += strlen("<default,mcast>,");
5526 	}
5527 	for (mcip = mip->mi_rx_groups[grp_index].mrg_clients; mcip != NULL;
5528 	    mcip = mcip->mgc_next) {
5529 		int name_len = strlen(mcip->mgc_client->mci_name);
5530 
5531 		/*
5532 		 * MAXCLIENTNAMELEN is the buffer size reserved for client
5533 		 * names.
5534 		 * XXXX Formating the client name string needs to be moved
5535 		 * to user land when fixing the size of dhi_clnts in
5536 		 * dld_hwgrpinfo_t. We should use n_clients * client_name for
5537 		 * dhi_clntsin instead of MAXCLIENTNAMELEN
5538 		 */
5539 		if (index + name_len >= MAXCLIENTNAMELEN) {
5540 			index = MAXCLIENTNAMELEN;
5541 			break;
5542 		}
5543 		bcopy(mcip->mgc_client->mci_name, &(clnts_name[index]),
5544 		    name_len);
5545 		index += name_len;
5546 		clnts_name[index++] = ',';
5547 		i++;
5548 	}
5549 
5550 	/* Get rid of the last , */
5551 	if (index > 0)
5552 		clnts_name[index - 1] = '\0';
5553 	*n_clnts = i;
5554 	rw_exit(&mip->mi_rw_lock);
5555 }
5556 
5557 /*
5558  * Return the TX h/w information for the group indexed by grp_num.
5559  */
5560 void
5561 mac_get_hwtxgrp_info(mac_handle_t mh, int grp_index, uint_t *grp_num,
5562     uint_t *n_rings, uint_t *rings, uint_t *type, uint_t *n_clnts,
5563     char *clnts_name)
5564 {
5565 	mac_impl_t *mip = (mac_impl_t *)mh;
5566 	mac_grp_client_t *mcip;
5567 	uint_t i = 0, index = 0;
5568 	mac_ring_t	*ring;
5569 
5570 	/* Revisit when we implement fully dynamic group allocation */
5571 	ASSERT(grp_index >= 0 && grp_index <= mip->mi_tx_group_count);
5572 
5573 	rw_enter(&mip->mi_rw_lock, RW_READER);
5574 	*grp_num = mip->mi_tx_groups[grp_index].mrg_index > 0 ?
5575 	    mip->mi_tx_groups[grp_index].mrg_index : grp_index;
5576 	*type = mip->mi_tx_groups[grp_index].mrg_type;
5577 	*n_rings = mip->mi_tx_groups[grp_index].mrg_cur_count;
5578 	ring = mip->mi_tx_groups[grp_index].mrg_rings;
5579 	for (index = 0; index < mip->mi_tx_groups[grp_index].mrg_cur_count;
5580 	    index++) {
5581 		rings[index] = ring->mr_index;
5582 		ring = ring->mr_next;
5583 	}
5584 	index = 0;
5585 	/* Default group has an index of -1 */
5586 	if (mip->mi_tx_groups[grp_index].mrg_index < 0) {
5587 		(void) strlcpy(clnts_name, "<default>,",
5588 		    MAXCLIENTNAMELEN);
5589 		index += strlen("<default>,");
5590 	}
5591 	for (mcip = mip->mi_tx_groups[grp_index].mrg_clients; mcip != NULL;
5592 	    mcip = mcip->mgc_next) {
5593 		int name_len = strlen(mcip->mgc_client->mci_name);
5594 
5595 		/*
5596 		 * MAXCLIENTNAMELEN is the buffer size reserved for client
5597 		 * names.
5598 		 * XXXX Formating the client name string needs to be moved
5599 		 * to user land when fixing the size of dhi_clnts in
5600 		 * dld_hwgrpinfo_t. We should use n_clients * client_name for
5601 		 * dhi_clntsin instead of MAXCLIENTNAMELEN
5602 		 */
5603 		if (index + name_len >= MAXCLIENTNAMELEN) {
5604 			index = MAXCLIENTNAMELEN;
5605 			break;
5606 		}
5607 		bcopy(mcip->mgc_client->mci_name, &(clnts_name[index]),
5608 		    name_len);
5609 		index += name_len;
5610 		clnts_name[index++] = ',';
5611 		i++;
5612 	}
5613 
5614 	/* Get rid of the last , */
5615 	if (index > 0)
5616 		clnts_name[index - 1] = '\0';
5617 	*n_clnts = i;
5618 	rw_exit(&mip->mi_rw_lock);
5619 }
5620 
5621 /*
5622  * Return the group count for RX or TX.
5623  */
5624 uint_t
5625 mac_hwgrp_num(mac_handle_t mh, int type)
5626 {
5627 	mac_impl_t *mip = (mac_impl_t *)mh;
5628 
5629 	/*
5630 	 * Return the Rx and Tx group count; for the Tx we need to
5631 	 * include the default too.
5632 	 */
5633 	return (type == MAC_RING_TYPE_RX ? mip->mi_rx_group_count :
5634 	    mip->mi_tx_groups != NULL ? mip->mi_tx_group_count + 1 : 0);
5635 }
5636 
5637 /*
5638  * The total number of free TX rings for this MAC.
5639  */
5640 uint_t
5641 mac_txavail_get(mac_handle_t mh)
5642 {
5643 	mac_impl_t	*mip = (mac_impl_t *)mh;
5644 
5645 	return (mip->mi_txrings_avail);
5646 }
5647 
5648 /*
5649  * The total number of free RX rings for this MAC.
5650  */
5651 uint_t
5652 mac_rxavail_get(mac_handle_t mh)
5653 {
5654 	mac_impl_t	*mip = (mac_impl_t *)mh;
5655 
5656 	return (mip->mi_rxrings_avail);
5657 }
5658 
5659 /*
5660  * The total number of reserved RX rings on this MAC.
5661  */
5662 uint_t
5663 mac_rxrsvd_get(mac_handle_t mh)
5664 {
5665 	mac_impl_t	*mip = (mac_impl_t *)mh;
5666 
5667 	return (mip->mi_rxrings_rsvd);
5668 }
5669 
5670 /*
5671  * The total number of reserved TX rings on this MAC.
5672  */
5673 uint_t
5674 mac_txrsvd_get(mac_handle_t mh)
5675 {
5676 	mac_impl_t	*mip = (mac_impl_t *)mh;
5677 
5678 	return (mip->mi_txrings_rsvd);
5679 }
5680 
5681 /*
5682  * Total number of free RX groups on this MAC.
5683  */
5684 uint_t
5685 mac_rxhwlnksavail_get(mac_handle_t mh)
5686 {
5687 	mac_impl_t	*mip = (mac_impl_t *)mh;
5688 
5689 	return (mip->mi_rxhwclnt_avail);
5690 }
5691 
5692 /*
5693  * Total number of RX groups reserved on this MAC.
5694  */
5695 uint_t
5696 mac_rxhwlnksrsvd_get(mac_handle_t mh)
5697 {
5698 	mac_impl_t	*mip = (mac_impl_t *)mh;
5699 
5700 	return (mip->mi_rxhwclnt_used);
5701 }
5702 
5703 /*
5704  * Total number of free TX groups on this MAC.
5705  */
5706 uint_t
5707 mac_txhwlnksavail_get(mac_handle_t mh)
5708 {
5709 	mac_impl_t	*mip = (mac_impl_t *)mh;
5710 
5711 	return (mip->mi_txhwclnt_avail);
5712 }
5713 
5714 /*
5715  * Total number of TX groups reserved on this MAC.
5716  */
5717 uint_t
5718 mac_txhwlnksrsvd_get(mac_handle_t mh)
5719 {
5720 	mac_impl_t	*mip = (mac_impl_t *)mh;
5721 
5722 	return (mip->mi_txhwclnt_used);
5723 }
5724 
5725 /*
5726  * Initialize the rings property for a mac client. A non-0 value for
5727  * rxring or txring specifies the number of rings required, a value
5728  * of MAC_RXRINGS_NONE/MAC_TXRINGS_NONE specifies that it doesn't need
5729  * any RX/TX rings and a value of MAC_RXRINGS_DONTCARE/MAC_TXRINGS_DONTCARE
5730  * means the system can decide whether it can give any rings or not.
5731  */
5732 void
5733 mac_client_set_rings(mac_client_handle_t mch, int rxrings, int txrings)
5734 {
5735 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
5736 	mac_resource_props_t	*mrp = MCIP_RESOURCE_PROPS(mcip);
5737 
5738 	if (rxrings != MAC_RXRINGS_DONTCARE) {
5739 		mrp->mrp_mask |= MRP_RX_RINGS;
5740 		mrp->mrp_nrxrings = rxrings;
5741 	}
5742 
5743 	if (txrings != MAC_TXRINGS_DONTCARE) {
5744 		mrp->mrp_mask |= MRP_TX_RINGS;
5745 		mrp->mrp_ntxrings = txrings;
5746 	}
5747 }
5748 
5749 boolean_t
5750 mac_get_promisc_filtered(mac_client_handle_t mch)
5751 {
5752 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
5753 
5754 	return (mcip->mci_protect_flags & MPT_FLAG_PROMISC_FILTERED);
5755 }
5756 
5757 void
5758 mac_set_promisc_filtered(mac_client_handle_t mch, boolean_t enable)
5759 {
5760 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
5761 
5762 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
5763 	if (enable)
5764 		mcip->mci_protect_flags |= MPT_FLAG_PROMISC_FILTERED;
5765 	else
5766 		mcip->mci_protect_flags &= ~MPT_FLAG_PROMISC_FILTERED;
5767 }
5768