xref: /illumos-gate/usr/src/uts/common/io/mac/mac_client.c (revision 8950e535f42dd006f8cfb2122c94f6b7557757e0)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright 2018 Joyent, Inc.
25  * Copyright 2017 RackTop Systems.
26  */
27 
28 /*
29  * - General Introduction:
30  *
31  * This file contains the implementation of the MAC client kernel
32  * API and related code. The MAC client API allows a kernel module
33  * to gain access to a MAC instance (physical NIC, link aggregation, etc).
34  * It allows a MAC client to associate itself with a MAC address,
35  * VLANs, callback functions for data traffic and for promiscuous mode.
36  * The MAC client API is also used to specify the properties associated
37  * with a MAC client, such as bandwidth limits, priority, CPUS, etc.
38  * These properties are further used to determine the hardware resources
39  * to allocate to the various MAC clients.
40  *
41  * - Primary MAC clients:
42  *
43  * The MAC client API refers to "primary MAC clients". A primary MAC
44  * client is a client which "owns" the primary MAC address of
45  * the underlying MAC instance. The primary MAC address is called out
46  * since it is associated with specific semantics: the primary MAC
47  * address is the MAC address which is assigned to the IP interface
48  * when it is plumbed, and the primary MAC address is assigned
49  * to VLAN data-links. The primary address of a MAC instance can
50  * also change dynamically from under the MAC client, for example
51  * as a result of a change of state of a link aggregation. In that
52  * case the MAC layer automatically updates all data-structures which
53  * refer to the current value of the primary MAC address. Typical
54  * primary MAC clients are dls, aggr, and xnb. A typical non-primary
55  * MAC client is the vnic driver.
56  *
57  * - Virtual Switching:
58  *
59  * The MAC layer implements a virtual switch between the MAC clients
60  * (primary and non-primary) defined on top of the same underlying
61  * NIC (physical, link aggregation, etc). The virtual switch is
62  * VLAN-aware, i.e. it allows multiple MAC clients to be member
63  * of one or more VLANs, and the virtual switch will distribute
64  * multicast tagged packets only to the member of the corresponding
65  * VLANs.
66  *
67  * - Upper vs Lower MAC:
68  *
69  * Creating a VNIC on top of a MAC instance effectively causes
70  * two MAC instances to be layered on top of each other, one for
71  * the VNIC(s), one for the underlying MAC instance (physical NIC,
72  * link aggregation, etc). In the code below we refer to the
73  * underlying NIC as the "lower MAC", and we refer to VNICs as
74  * the "upper MAC".
75  *
76  * - Pass-through for VNICs:
77  *
78  * When VNICs are created on top of an underlying MAC, this causes
79  * a layering of two MAC instances. Since the lower MAC already
80  * does the switching and demultiplexing to its MAC clients, the
81  * upper MAC would simply have to pass packets to the layer below
82  * or above it, which would introduce overhead. In order to avoid
83  * this overhead, the MAC layer implements a pass-through mechanism
84  * for VNICs. When a VNIC opens the lower MAC instance, it saves
85  * the MAC client handle it optains from the MAC layer. When a MAC
86  * client opens a VNIC (upper MAC), the MAC layer detects that
87  * the MAC being opened is a VNIC, and gets the MAC client handle
88  * that the VNIC driver obtained from the lower MAC. This exchange
89  * is done through a private capability between the MAC layer
90  * and the VNIC driver. The upper MAC then returns that handle
91  * directly to its MAC client. Any operation done by the upper
92  * MAC client is now done on the lower MAC client handle, which
93  * allows the VNIC driver to be completely bypassed for the
94  * performance sensitive data-path.
95  *
96  * - Secondary MACs for VNICs:
97  *
98  * VNICs support multiple upper mac clients to enable support for
99  * multiple MAC addresses on the VNIC. When the VNIC is created the
100  * initial mac client is the primary upper mac. Any additional mac
101  * clients are secondary macs. These are kept in sync with the primary
102  * (for things such as the rx function and resource control settings)
103  * using the same private capability interface between the MAC layer
104  * and the VNIC layer.
105  *
106  */
107 
108 #include <sys/types.h>
109 #include <sys/conf.h>
110 #include <sys/id_space.h>
111 #include <sys/esunddi.h>
112 #include <sys/stat.h>
113 #include <sys/mkdev.h>
114 #include <sys/stream.h>
115 #include <sys/strsun.h>
116 #include <sys/strsubr.h>
117 #include <sys/dlpi.h>
118 #include <sys/modhash.h>
119 #include <sys/mac_impl.h>
120 #include <sys/mac_client_impl.h>
121 #include <sys/mac_soft_ring.h>
122 #include <sys/mac_stat.h>
123 #include <sys/dls.h>
124 #include <sys/dld.h>
125 #include <sys/modctl.h>
126 #include <sys/fs/dv_node.h>
127 #include <sys/thread.h>
128 #include <sys/proc.h>
129 #include <sys/callb.h>
130 #include <sys/cpuvar.h>
131 #include <sys/atomic.h>
132 #include <sys/sdt.h>
133 #include <sys/mac_flow.h>
134 #include <sys/ddi_intr_impl.h>
135 #include <sys/disp.h>
136 #include <sys/sdt.h>
137 #include <sys/vnic.h>
138 #include <sys/vnic_impl.h>
139 #include <sys/vlan.h>
140 #include <inet/ip.h>
141 #include <inet/ip6.h>
142 #include <sys/exacct.h>
143 #include <sys/exacct_impl.h>
144 #include <inet/nd.h>
145 #include <sys/ethernet.h>
146 
147 kmem_cache_t	*mac_client_impl_cache;
148 kmem_cache_t	*mac_promisc_impl_cache;
149 
150 static boolean_t mac_client_single_rcvr(mac_client_impl_t *);
151 static flow_entry_t *mac_client_swap_mciflent(mac_client_impl_t *);
152 static flow_entry_t *mac_client_get_flow(mac_client_impl_t *,
153     mac_unicast_impl_t *);
154 static void mac_client_remove_flow_from_list(mac_client_impl_t *,
155     flow_entry_t *);
156 static void mac_client_add_to_flow_list(mac_client_impl_t *, flow_entry_t *);
157 static void mac_rename_flow_names(mac_client_impl_t *, const char *);
158 static void mac_virtual_link_update(mac_impl_t *);
159 static int mac_client_datapath_setup(mac_client_impl_t *, uint16_t,
160     uint8_t *, mac_resource_props_t *, boolean_t, mac_unicast_impl_t *);
161 static void mac_client_datapath_teardown(mac_client_handle_t,
162     mac_unicast_impl_t *, flow_entry_t *);
163 static int mac_resource_ctl_set(mac_client_handle_t, mac_resource_props_t *);
164 
165 /* ARGSUSED */
166 static int
167 i_mac_client_impl_ctor(void *buf, void *arg, int kmflag)
168 {
169 	int	i;
170 	mac_client_impl_t	*mcip = buf;
171 
172 	bzero(buf, MAC_CLIENT_IMPL_SIZE);
173 	mutex_init(&mcip->mci_tx_cb_lock, NULL, MUTEX_DRIVER, NULL);
174 	mcip->mci_tx_notify_cb_info.mcbi_lockp = &mcip->mci_tx_cb_lock;
175 
176 	ASSERT(mac_tx_percpu_cnt >= 0);
177 	for (i = 0; i <= mac_tx_percpu_cnt; i++) {
178 		mutex_init(&mcip->mci_tx_pcpu[i].pcpu_tx_lock, NULL,
179 		    MUTEX_DRIVER, NULL);
180 	}
181 	cv_init(&mcip->mci_tx_cv, NULL, CV_DRIVER, NULL);
182 
183 	return (0);
184 }
185 
186 /* ARGSUSED */
187 static void
188 i_mac_client_impl_dtor(void *buf, void *arg)
189 {
190 	int	i;
191 	mac_client_impl_t *mcip = buf;
192 
193 	ASSERT(mcip->mci_promisc_list == NULL);
194 	ASSERT(mcip->mci_unicast_list == NULL);
195 	ASSERT(mcip->mci_state_flags == 0);
196 	ASSERT(mcip->mci_tx_flag == 0);
197 
198 	mutex_destroy(&mcip->mci_tx_cb_lock);
199 
200 	ASSERT(mac_tx_percpu_cnt >= 0);
201 	for (i = 0; i <= mac_tx_percpu_cnt; i++) {
202 		ASSERT(mcip->mci_tx_pcpu[i].pcpu_tx_refcnt == 0);
203 		mutex_destroy(&mcip->mci_tx_pcpu[i].pcpu_tx_lock);
204 	}
205 	cv_destroy(&mcip->mci_tx_cv);
206 }
207 
208 /* ARGSUSED */
209 static int
210 i_mac_promisc_impl_ctor(void *buf, void *arg, int kmflag)
211 {
212 	mac_promisc_impl_t	*mpip = buf;
213 
214 	bzero(buf, sizeof (mac_promisc_impl_t));
215 	mpip->mpi_mci_link.mcb_objp = buf;
216 	mpip->mpi_mci_link.mcb_objsize = sizeof (mac_promisc_impl_t);
217 	mpip->mpi_mi_link.mcb_objp = buf;
218 	mpip->mpi_mi_link.mcb_objsize = sizeof (mac_promisc_impl_t);
219 	return (0);
220 }
221 
222 /* ARGSUSED */
223 static void
224 i_mac_promisc_impl_dtor(void *buf, void *arg)
225 {
226 	mac_promisc_impl_t	*mpip = buf;
227 
228 	ASSERT(mpip->mpi_mci_link.mcb_objp != NULL);
229 	ASSERT(mpip->mpi_mci_link.mcb_objsize == sizeof (mac_promisc_impl_t));
230 	ASSERT(mpip->mpi_mi_link.mcb_objp == mpip->mpi_mci_link.mcb_objp);
231 	ASSERT(mpip->mpi_mi_link.mcb_objsize == sizeof (mac_promisc_impl_t));
232 
233 	mpip->mpi_mci_link.mcb_objp = NULL;
234 	mpip->mpi_mci_link.mcb_objsize = 0;
235 	mpip->mpi_mi_link.mcb_objp = NULL;
236 	mpip->mpi_mi_link.mcb_objsize = 0;
237 
238 	ASSERT(mpip->mpi_mci_link.mcb_flags == 0);
239 	mpip->mpi_mci_link.mcb_objsize = 0;
240 }
241 
242 void
243 mac_client_init(void)
244 {
245 	ASSERT(mac_tx_percpu_cnt >= 0);
246 
247 	mac_client_impl_cache = kmem_cache_create("mac_client_impl_cache",
248 	    MAC_CLIENT_IMPL_SIZE, 0, i_mac_client_impl_ctor,
249 	    i_mac_client_impl_dtor, NULL, NULL, NULL, 0);
250 	ASSERT(mac_client_impl_cache != NULL);
251 
252 	mac_promisc_impl_cache = kmem_cache_create("mac_promisc_impl_cache",
253 	    sizeof (mac_promisc_impl_t), 0, i_mac_promisc_impl_ctor,
254 	    i_mac_promisc_impl_dtor, NULL, NULL, NULL, 0);
255 	ASSERT(mac_promisc_impl_cache != NULL);
256 }
257 
258 void
259 mac_client_fini(void)
260 {
261 	kmem_cache_destroy(mac_client_impl_cache);
262 	kmem_cache_destroy(mac_promisc_impl_cache);
263 }
264 
265 /*
266  * Return the lower MAC client handle from the VNIC driver for the
267  * specified VNIC MAC instance.
268  */
269 mac_client_impl_t *
270 mac_vnic_lower(mac_impl_t *mip)
271 {
272 	mac_capab_vnic_t cap;
273 	mac_client_impl_t *mcip;
274 
275 	VERIFY(i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_VNIC, &cap));
276 	mcip = cap.mcv_mac_client_handle(cap.mcv_arg);
277 
278 	return (mcip);
279 }
280 
281 /*
282  * Update the secondary macs
283  */
284 void
285 mac_vnic_secondary_update(mac_impl_t *mip)
286 {
287 	mac_capab_vnic_t cap;
288 
289 	VERIFY(i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_VNIC, &cap));
290 	cap.mcv_mac_secondary_update(cap.mcv_arg);
291 }
292 
293 /*
294  * Return the MAC client handle of the primary MAC client for the
295  * specified MAC instance, or NULL otherwise.
296  */
297 mac_client_impl_t *
298 mac_primary_client_handle(mac_impl_t *mip)
299 {
300 	mac_client_impl_t *mcip;
301 
302 	if (mip->mi_state_flags & MIS_IS_VNIC)
303 		return (mac_vnic_lower(mip));
304 
305 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
306 
307 	for (mcip = mip->mi_clients_list; mcip != NULL;
308 	    mcip = mcip->mci_client_next) {
309 		if (MCIP_DATAPATH_SETUP(mcip) && mac_is_primary_client(mcip))
310 			return (mcip);
311 	}
312 	return (NULL);
313 }
314 
315 /*
316  * Open a MAC specified by its MAC name.
317  */
318 int
319 mac_open(const char *macname, mac_handle_t *mhp)
320 {
321 	mac_impl_t	*mip;
322 	int		err;
323 
324 	/*
325 	 * Look up its entry in the global hash table.
326 	 */
327 	if ((err = mac_hold(macname, &mip)) != 0)
328 		return (err);
329 
330 	/*
331 	 * Hold the dip associated to the MAC to prevent it from being
332 	 * detached. For a softmac, its underlying dip is held by the
333 	 * mi_open() callback.
334 	 *
335 	 * This is done to be more tolerant with some defective drivers,
336 	 * which incorrectly handle mac_unregister() failure in their
337 	 * xxx_detach() routine. For example, some drivers ignore the
338 	 * failure of mac_unregister() and free all resources that
339 	 * that are needed for data transmition.
340 	 */
341 	e_ddi_hold_devi(mip->mi_dip);
342 
343 	if (!(mip->mi_callbacks->mc_callbacks & MC_OPEN)) {
344 		*mhp = (mac_handle_t)mip;
345 		return (0);
346 	}
347 
348 	/*
349 	 * The mac perimeter is used in both mac_open and mac_close by the
350 	 * framework to single thread the MC_OPEN/MC_CLOSE of drivers.
351 	 */
352 	i_mac_perim_enter(mip);
353 	mip->mi_oref++;
354 	if (mip->mi_oref != 1 || ((err = mip->mi_open(mip->mi_driver)) == 0)) {
355 		*mhp = (mac_handle_t)mip;
356 		i_mac_perim_exit(mip);
357 		return (0);
358 	}
359 	mip->mi_oref--;
360 	ddi_release_devi(mip->mi_dip);
361 	mac_rele(mip);
362 	i_mac_perim_exit(mip);
363 	return (err);
364 }
365 
366 /*
367  * Open a MAC specified by its linkid.
368  */
369 int
370 mac_open_by_linkid(datalink_id_t linkid, mac_handle_t *mhp)
371 {
372 	dls_dl_handle_t	dlh;
373 	int		err;
374 
375 	if ((err = dls_devnet_hold_tmp(linkid, &dlh)) != 0)
376 		return (err);
377 
378 	dls_devnet_prop_task_wait(dlh);
379 
380 	err = mac_open(dls_devnet_mac(dlh), mhp);
381 
382 	dls_devnet_rele_tmp(dlh);
383 	return (err);
384 }
385 
386 /*
387  * Open a MAC specified by its link name.
388  */
389 int
390 mac_open_by_linkname(const char *link, mac_handle_t *mhp)
391 {
392 	datalink_id_t	linkid;
393 	int		err;
394 
395 	if ((err = dls_mgmt_get_linkid(link, &linkid)) != 0)
396 		return (err);
397 	return (mac_open_by_linkid(linkid, mhp));
398 }
399 
400 /*
401  * Close the specified MAC.
402  */
403 void
404 mac_close(mac_handle_t mh)
405 {
406 	mac_impl_t	*mip = (mac_impl_t *)mh;
407 
408 	i_mac_perim_enter(mip);
409 	/*
410 	 * The mac perimeter is used in both mac_open and mac_close by the
411 	 * framework to single thread the MC_OPEN/MC_CLOSE of drivers.
412 	 */
413 	if (mip->mi_callbacks->mc_callbacks & MC_OPEN) {
414 		ASSERT(mip->mi_oref != 0);
415 		if (--mip->mi_oref == 0) {
416 			if ((mip->mi_callbacks->mc_callbacks & MC_CLOSE))
417 				mip->mi_close(mip->mi_driver);
418 		}
419 	}
420 	i_mac_perim_exit(mip);
421 	ddi_release_devi(mip->mi_dip);
422 	mac_rele(mip);
423 }
424 
425 /*
426  * Misc utility functions to retrieve various information about a MAC
427  * instance or a MAC client.
428  */
429 
430 const mac_info_t *
431 mac_info(mac_handle_t mh)
432 {
433 	return (&((mac_impl_t *)mh)->mi_info);
434 }
435 
436 dev_info_t *
437 mac_devinfo_get(mac_handle_t mh)
438 {
439 	return (((mac_impl_t *)mh)->mi_dip);
440 }
441 
442 void *
443 mac_driver(mac_handle_t mh)
444 {
445 	return (((mac_impl_t *)mh)->mi_driver);
446 }
447 
448 const char *
449 mac_name(mac_handle_t mh)
450 {
451 	return (((mac_impl_t *)mh)->mi_name);
452 }
453 
454 int
455 mac_type(mac_handle_t mh)
456 {
457 	return (((mac_impl_t *)mh)->mi_type->mt_type);
458 }
459 
460 int
461 mac_nativetype(mac_handle_t mh)
462 {
463 	return (((mac_impl_t *)mh)->mi_type->mt_nativetype);
464 }
465 
466 char *
467 mac_client_name(mac_client_handle_t mch)
468 {
469 	return (((mac_client_impl_t *)mch)->mci_name);
470 }
471 
472 minor_t
473 mac_minor(mac_handle_t mh)
474 {
475 	return (((mac_impl_t *)mh)->mi_minor);
476 }
477 
478 /*
479  * Return the VID associated with a MAC client. This function should
480  * be called for clients which are associated with only one VID.
481  */
482 uint16_t
483 mac_client_vid(mac_client_handle_t mch)
484 {
485 	uint16_t		vid = VLAN_ID_NONE;
486 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
487 	flow_desc_t		flow_desc;
488 
489 	if (mcip->mci_nflents == 0)
490 		return (vid);
491 
492 	ASSERT(MCIP_DATAPATH_SETUP(mcip) && mac_client_single_rcvr(mcip));
493 
494 	mac_flow_get_desc(mcip->mci_flent, &flow_desc);
495 	if ((flow_desc.fd_mask & FLOW_LINK_VID) != 0)
496 		vid = flow_desc.fd_vid;
497 
498 	return (vid);
499 }
500 
501 /*
502  * Return whether the specified MAC client corresponds to a VLAN VNIC.
503  */
504 boolean_t
505 mac_client_is_vlan_vnic(mac_client_handle_t mch)
506 {
507 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
508 
509 	return (((mcip->mci_state_flags & MCIS_IS_VNIC) != 0) &&
510 	    ((mcip->mci_flent->fe_type & FLOW_PRIMARY_MAC) != 0));
511 }
512 
513 /*
514  * Return the link speed associated with the specified MAC client.
515  *
516  * The link speed of a MAC client is equal to the smallest value of
517  * 1) the current link speed of the underlying NIC, or
518  * 2) the bandwidth limit set for the MAC client.
519  *
520  * Note that the bandwidth limit can be higher than the speed
521  * of the underlying NIC. This is allowed to avoid spurious
522  * administration action failures or artifically lowering the
523  * bandwidth limit of a link that may  have temporarily lowered
524  * its link speed due to hardware problem or administrator action.
525  */
526 static uint64_t
527 mac_client_ifspeed(mac_client_impl_t *mcip)
528 {
529 	mac_impl_t *mip = mcip->mci_mip;
530 	uint64_t nic_speed;
531 
532 	nic_speed = mac_stat_get((mac_handle_t)mip, MAC_STAT_IFSPEED);
533 
534 	if (nic_speed == 0) {
535 		return (0);
536 	} else {
537 		uint64_t policy_limit = (uint64_t)-1;
538 
539 		if (MCIP_RESOURCE_PROPS_MASK(mcip) & MRP_MAXBW)
540 			policy_limit = MCIP_RESOURCE_PROPS_MAXBW(mcip);
541 
542 		return (MIN(policy_limit, nic_speed));
543 	}
544 }
545 
546 /*
547  * Return the link state of the specified client. If here are more
548  * than one clients of the underying mac_impl_t, the link state
549  * will always be UP regardless of the link state of the underlying
550  * mac_impl_t. This is needed to allow the MAC clients to continue
551  * to communicate with each other even when the physical link of
552  * their mac_impl_t is down.
553  */
554 static uint64_t
555 mac_client_link_state(mac_client_impl_t *mcip)
556 {
557 	mac_impl_t *mip = mcip->mci_mip;
558 	uint16_t vid;
559 	mac_client_impl_t *mci_list;
560 	mac_unicast_impl_t *mui_list, *oth_mui_list;
561 
562 	/*
563 	 * Returns LINK_STATE_UP if there are other MAC clients defined on
564 	 * mac_impl_t which share same VLAN ID as that of mcip. Note that
565 	 * if 'mcip' has more than one VID's then we match ANY one of the
566 	 * VID's with other MAC client's VID's and return LINK_STATE_UP.
567 	 */
568 	rw_enter(&mcip->mci_rw_lock, RW_READER);
569 	for (mui_list = mcip->mci_unicast_list; mui_list != NULL;
570 	    mui_list = mui_list->mui_next) {
571 		vid = mui_list->mui_vid;
572 		for (mci_list = mip->mi_clients_list; mci_list != NULL;
573 		    mci_list = mci_list->mci_client_next) {
574 			if (mci_list == mcip)
575 				continue;
576 			for (oth_mui_list = mci_list->mci_unicast_list;
577 			    oth_mui_list != NULL; oth_mui_list = oth_mui_list->
578 			    mui_next) {
579 				if (vid == oth_mui_list->mui_vid) {
580 					rw_exit(&mcip->mci_rw_lock);
581 					return (LINK_STATE_UP);
582 				}
583 			}
584 		}
585 	}
586 	rw_exit(&mcip->mci_rw_lock);
587 
588 	return (mac_stat_get((mac_handle_t)mip, MAC_STAT_LINK_STATE));
589 }
590 
591 /*
592  * These statistics are consumed by dladm show-link -s <vnic>,
593  * dladm show-vnic -s and netstat. With the introduction of dlstat,
594  * dladm show-link -s and dladm show-vnic -s witll be EOL'ed while
595  * netstat will consume from kstats introduced for dlstat. This code
596  * will be removed at that time.
597  */
598 
599 /*
600  * Return the statistics of a MAC client. These statistics are different
601  * then the statistics of the underlying MAC which are returned by
602  * mac_stat_get().
603  *
604  * Note that for things based on the tx and rx stats, mac will end up clobbering
605  * those stats when the underlying set of rings in the srs changes. As such, we
606  * need to source not only the current set, but also the historical set when
607  * returning to the client, lest our counters appear to go backwards.
608  */
609 uint64_t
610 mac_client_stat_get(mac_client_handle_t mch, uint_t stat)
611 {
612 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
613 	mac_impl_t		*mip = mcip->mci_mip;
614 	flow_entry_t		*flent = mcip->mci_flent;
615 	mac_soft_ring_set_t	*mac_srs;
616 	mac_rx_stats_t		*mac_rx_stat, *old_rx_stat;
617 	mac_tx_stats_t		*mac_tx_stat, *old_tx_stat;
618 	int i;
619 	uint64_t val = 0;
620 
621 	mac_srs = (mac_soft_ring_set_t *)(flent->fe_tx_srs);
622 	mac_tx_stat = &mac_srs->srs_tx.st_stat;
623 	old_rx_stat = &mcip->mci_misc_stat.mms_defunctrxlanestats;
624 	old_tx_stat = &mcip->mci_misc_stat.mms_defuncttxlanestats;
625 
626 	switch (stat) {
627 	case MAC_STAT_LINK_STATE:
628 		val = mac_client_link_state(mcip);
629 		break;
630 	case MAC_STAT_LINK_UP:
631 		val = (mac_client_link_state(mcip) == LINK_STATE_UP);
632 		break;
633 	case MAC_STAT_PROMISC:
634 		val = mac_stat_get((mac_handle_t)mip, MAC_STAT_PROMISC);
635 		break;
636 	case MAC_STAT_LOWLINK_STATE:
637 		val = mac_stat_get((mac_handle_t)mip, MAC_STAT_LOWLINK_STATE);
638 		break;
639 	case MAC_STAT_IFSPEED:
640 		val = mac_client_ifspeed(mcip);
641 		break;
642 	case MAC_STAT_MULTIRCV:
643 		val = mcip->mci_misc_stat.mms_multircv;
644 		break;
645 	case MAC_STAT_BRDCSTRCV:
646 		val = mcip->mci_misc_stat.mms_brdcstrcv;
647 		break;
648 	case MAC_STAT_MULTIXMT:
649 		val = mcip->mci_misc_stat.mms_multixmt;
650 		break;
651 	case MAC_STAT_BRDCSTXMT:
652 		val = mcip->mci_misc_stat.mms_brdcstxmt;
653 		break;
654 	case MAC_STAT_OBYTES:
655 		val = mac_tx_stat->mts_obytes;
656 		val += old_tx_stat->mts_obytes;
657 		break;
658 	case MAC_STAT_OPACKETS:
659 		val = mac_tx_stat->mts_opackets;
660 		val += old_tx_stat->mts_opackets;
661 		break;
662 	case MAC_STAT_OERRORS:
663 		val = mac_tx_stat->mts_oerrors;
664 		val += old_tx_stat->mts_oerrors;
665 		break;
666 	case MAC_STAT_IPACKETS:
667 		for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
668 			mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
669 			mac_rx_stat = &mac_srs->srs_rx.sr_stat;
670 			val += mac_rx_stat->mrs_intrcnt +
671 			    mac_rx_stat->mrs_pollcnt + mac_rx_stat->mrs_lclcnt;
672 		}
673 		val += old_rx_stat->mrs_intrcnt + old_rx_stat->mrs_pollcnt +
674 		    old_rx_stat->mrs_lclcnt;
675 		break;
676 	case MAC_STAT_RBYTES:
677 		for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
678 			mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
679 			mac_rx_stat = &mac_srs->srs_rx.sr_stat;
680 			val += mac_rx_stat->mrs_intrbytes +
681 			    mac_rx_stat->mrs_pollbytes +
682 			    mac_rx_stat->mrs_lclbytes;
683 		}
684 		val += old_rx_stat->mrs_intrbytes + old_rx_stat->mrs_pollbytes +
685 		    old_rx_stat->mrs_lclbytes;
686 		break;
687 	case MAC_STAT_IERRORS:
688 		for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
689 			mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
690 			mac_rx_stat = &mac_srs->srs_rx.sr_stat;
691 			val += mac_rx_stat->mrs_ierrors;
692 		}
693 		val += old_rx_stat->mrs_ierrors;
694 		break;
695 	default:
696 		val = mac_driver_stat_default(mip, stat);
697 		break;
698 	}
699 
700 	return (val);
701 }
702 
703 /*
704  * Return the statistics of the specified MAC instance.
705  */
706 uint64_t
707 mac_stat_get(mac_handle_t mh, uint_t stat)
708 {
709 	mac_impl_t	*mip = (mac_impl_t *)mh;
710 	uint64_t	val;
711 	int		ret;
712 
713 	/*
714 	 * The range of stat determines where it is maintained.  Stat
715 	 * values from 0 up to (but not including) MAC_STAT_MIN are
716 	 * mainteined by the mac module itself.  Everything else is
717 	 * maintained by the driver.
718 	 *
719 	 * If the mac_impl_t being queried corresponds to a VNIC,
720 	 * the stats need to be queried from the lower MAC client
721 	 * corresponding to the VNIC. (The mac_link_update()
722 	 * invoked by the driver to the lower MAC causes the *lower
723 	 * MAC* to update its mi_linkstate, and send a notification
724 	 * to its MAC clients. Due to the VNIC passthrough,
725 	 * these notifications are sent to the upper MAC clients
726 	 * of the VNIC directly, and the upper mac_impl_t of the VNIC
727 	 * does not have a valid mi_linkstate.
728 	 */
729 	if (stat < MAC_STAT_MIN && !(mip->mi_state_flags & MIS_IS_VNIC)) {
730 		/* these stats are maintained by the mac module itself */
731 		switch (stat) {
732 		case MAC_STAT_LINK_STATE:
733 			return (mip->mi_linkstate);
734 		case MAC_STAT_LINK_UP:
735 			return (mip->mi_linkstate == LINK_STATE_UP);
736 		case MAC_STAT_PROMISC:
737 			return (mip->mi_devpromisc != 0);
738 		case MAC_STAT_LOWLINK_STATE:
739 			return (mip->mi_lowlinkstate);
740 		default:
741 			ASSERT(B_FALSE);
742 		}
743 	}
744 
745 	/*
746 	 * Call the driver to get the given statistic.
747 	 */
748 	ret = mip->mi_getstat(mip->mi_driver, stat, &val);
749 	if (ret != 0) {
750 		/*
751 		 * The driver doesn't support this statistic.  Get the
752 		 * statistic's default value.
753 		 */
754 		val = mac_driver_stat_default(mip, stat);
755 	}
756 	return (val);
757 }
758 
759 /*
760  * Query hardware rx ring corresponding to the pseudo ring.
761  */
762 uint64_t
763 mac_pseudo_rx_ring_stat_get(mac_ring_handle_t handle, uint_t stat)
764 {
765 	return (mac_rx_ring_stat_get(handle, stat));
766 }
767 
768 /*
769  * Query hardware tx ring corresponding to the pseudo ring.
770  */
771 uint64_t
772 mac_pseudo_tx_ring_stat_get(mac_ring_handle_t handle, uint_t stat)
773 {
774 	return (mac_tx_ring_stat_get(handle, stat));
775 }
776 
777 /*
778  * Utility function which returns the VID associated with a flow entry.
779  */
780 uint16_t
781 i_mac_flow_vid(flow_entry_t *flent)
782 {
783 	flow_desc_t	flow_desc;
784 
785 	mac_flow_get_desc(flent, &flow_desc);
786 
787 	if ((flow_desc.fd_mask & FLOW_LINK_VID) != 0)
788 		return (flow_desc.fd_vid);
789 	return (VLAN_ID_NONE);
790 }
791 
792 /*
793  * Verify the validity of the specified unicast MAC address. Returns B_TRUE
794  * if the address is valid, B_FALSE otherwise (multicast address, or incorrect
795  * length.
796  */
797 boolean_t
798 mac_unicst_verify(mac_handle_t mh, const uint8_t *addr, uint_t len)
799 {
800 	mac_impl_t	*mip = (mac_impl_t *)mh;
801 
802 	/*
803 	 * Verify the address. No lock is needed since mi_type and plugin
804 	 * details don't change after mac_register().
805 	 */
806 	if ((len != mip->mi_type->mt_addr_length) ||
807 	    (mip->mi_type->mt_ops.mtops_unicst_verify(addr,
808 	    mip->mi_pdata)) != 0) {
809 		return (B_FALSE);
810 	} else {
811 		return (B_TRUE);
812 	}
813 }
814 
815 void
816 mac_sdu_get(mac_handle_t mh, uint_t *min_sdu, uint_t *max_sdu)
817 {
818 	mac_impl_t	*mip = (mac_impl_t *)mh;
819 
820 	if (min_sdu != NULL)
821 		*min_sdu = mip->mi_sdu_min;
822 	if (max_sdu != NULL)
823 		*max_sdu = mip->mi_sdu_max;
824 }
825 
826 void
827 mac_sdu_get2(mac_handle_t mh, uint_t *min_sdu, uint_t *max_sdu,
828     uint_t *multicast_sdu)
829 {
830 	mac_impl_t	*mip = (mac_impl_t *)mh;
831 
832 	if (min_sdu != NULL)
833 		*min_sdu = mip->mi_sdu_min;
834 	if (max_sdu != NULL)
835 		*max_sdu = mip->mi_sdu_max;
836 	if (multicast_sdu != NULL)
837 		*multicast_sdu = mip->mi_sdu_multicast;
838 }
839 
840 /*
841  * Update the MAC unicast address of the specified client's flows. Currently
842  * only one unicast MAC unicast address is allowed per client.
843  */
844 static void
845 mac_unicast_update_client_flow(mac_client_impl_t *mcip)
846 {
847 	mac_impl_t *mip = mcip->mci_mip;
848 	flow_entry_t *flent = mcip->mci_flent;
849 	mac_address_t *map = mcip->mci_unicast;
850 	flow_desc_t flow_desc;
851 
852 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
853 	ASSERT(flent != NULL);
854 
855 	mac_flow_get_desc(flent, &flow_desc);
856 	ASSERT(flow_desc.fd_mask & FLOW_LINK_DST);
857 
858 	bcopy(map->ma_addr, flow_desc.fd_dst_mac, map->ma_len);
859 	mac_flow_set_desc(flent, &flow_desc);
860 
861 	/*
862 	 * The v6 local and SLAAC addrs (used by mac protection) need to be
863 	 * regenerated because our mac address has changed.
864 	 */
865 	mac_protect_update_mac_token(mcip);
866 
867 	/*
868 	 * When there are multiple VLANs sharing the same MAC address,
869 	 * each gets its own MAC client, except when running on sun4v
870 	 * vsw. In that case the mci_flent_list is used to place
871 	 * multiple VLAN flows on one MAC client. If we ever get rid
872 	 * of vsw then this code can go, but until then we need to
873 	 * update all flow entries.
874 	 */
875 	for (flent = mcip->mci_flent_list; flent != NULL;
876 	    flent = flent->fe_client_next) {
877 		mac_flow_get_desc(flent, &flow_desc);
878 		if (!(flent->fe_type & FLOW_PRIMARY_MAC ||
879 		    flent->fe_type & FLOW_VNIC_MAC))
880 			continue;
881 
882 		bcopy(map->ma_addr, flow_desc.fd_dst_mac, map->ma_len);
883 		mac_flow_set_desc(flent, &flow_desc);
884 	}
885 }
886 
887 /*
888  * Update all clients that share the same unicast address.
889  */
890 void
891 mac_unicast_update_clients(mac_impl_t *mip, mac_address_t *map)
892 {
893 	mac_client_impl_t *mcip;
894 
895 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
896 
897 	/*
898 	 * Find all clients that share the same unicast MAC address and update
899 	 * them appropriately.
900 	 */
901 	for (mcip = mip->mi_clients_list; mcip != NULL;
902 	    mcip = mcip->mci_client_next) {
903 		/*
904 		 * Ignore clients that don't share this MAC address.
905 		 */
906 		if (map != mcip->mci_unicast)
907 			continue;
908 
909 		/*
910 		 * Update those clients with same old unicast MAC address.
911 		 */
912 		mac_unicast_update_client_flow(mcip);
913 	}
914 }
915 
916 /*
917  * Update the unicast MAC address of the specified VNIC MAC client.
918  *
919  * Check whether the operation is valid. Any of following cases should fail:
920  *
921  * 1. It's a VLAN type of VNIC.
922  * 2. The new value is current "primary" MAC address.
923  * 3. The current MAC address is shared with other clients.
924  * 4. The new MAC address has been used. This case will be valid when
925  *    client migration is fully supported.
926  */
927 int
928 mac_vnic_unicast_set(mac_client_handle_t mch, const uint8_t *addr)
929 {
930 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
931 	mac_impl_t *mip = mcip->mci_mip;
932 	mac_address_t *map = mcip->mci_unicast;
933 	int err;
934 
935 	ASSERT(!(mip->mi_state_flags & MIS_IS_VNIC));
936 	ASSERT(mcip->mci_state_flags & MCIS_IS_VNIC);
937 	ASSERT(mcip->mci_flags != MAC_CLIENT_FLAGS_PRIMARY);
938 
939 	i_mac_perim_enter(mip);
940 
941 	/*
942 	 * If this is a VLAN type of VNIC, it's using "primary" MAC address
943 	 * of the underlying interface. Must fail here. Refer to case 1 above.
944 	 */
945 	if (bcmp(map->ma_addr, mip->mi_addr, map->ma_len) == 0) {
946 		i_mac_perim_exit(mip);
947 		return (ENOTSUP);
948 	}
949 
950 	/*
951 	 * If the new address is the "primary" one, must fail. Refer to
952 	 * case 2 above.
953 	 */
954 	if (bcmp(addr, mip->mi_addr, map->ma_len) == 0) {
955 		i_mac_perim_exit(mip);
956 		return (EACCES);
957 	}
958 
959 	/*
960 	 * If the address is shared by multiple clients, must fail. Refer
961 	 * to case 3 above.
962 	 */
963 	if (mac_check_macaddr_shared(map)) {
964 		i_mac_perim_exit(mip);
965 		return (EBUSY);
966 	}
967 
968 	/*
969 	 * If the new address has been used, must fail for now. Refer to
970 	 * case 4 above.
971 	 */
972 	if (mac_find_macaddr(mip, (uint8_t *)addr) != NULL) {
973 		i_mac_perim_exit(mip);
974 		return (ENOTSUP);
975 	}
976 
977 	/*
978 	 * Update the MAC address.
979 	 */
980 	err = mac_update_macaddr(map, (uint8_t *)addr);
981 
982 	if (err != 0) {
983 		i_mac_perim_exit(mip);
984 		return (err);
985 	}
986 
987 	/*
988 	 * Update all flows of this MAC client.
989 	 */
990 	mac_unicast_update_client_flow(mcip);
991 
992 	i_mac_perim_exit(mip);
993 	return (0);
994 }
995 
996 /*
997  * Program the new primary unicast address of the specified MAC.
998  *
999  * Function mac_update_macaddr() takes care different types of underlying
1000  * MAC. If the underlying MAC is VNIC, the VNIC driver must have registerd
1001  * mi_unicst() entry point, that indirectly calls mac_vnic_unicast_set()
1002  * which will take care of updating the MAC address of the corresponding
1003  * MAC client.
1004  *
1005  * This is the only interface that allow the client to update the "primary"
1006  * MAC address of the underlying MAC. The new value must have not been
1007  * used by other clients.
1008  */
1009 int
1010 mac_unicast_primary_set(mac_handle_t mh, const uint8_t *addr)
1011 {
1012 	mac_impl_t *mip = (mac_impl_t *)mh;
1013 	mac_address_t *map;
1014 	int err;
1015 
1016 	/* verify the address validity */
1017 	if (!mac_unicst_verify(mh, addr, mip->mi_type->mt_addr_length))
1018 		return (EINVAL);
1019 
1020 	i_mac_perim_enter(mip);
1021 
1022 	/*
1023 	 * If the new value is the same as the current primary address value,
1024 	 * there's nothing to do.
1025 	 */
1026 	if (bcmp(addr, mip->mi_addr, mip->mi_type->mt_addr_length) == 0) {
1027 		i_mac_perim_exit(mip);
1028 		return (0);
1029 	}
1030 
1031 	if (mac_find_macaddr(mip, (uint8_t *)addr) != NULL) {
1032 		i_mac_perim_exit(mip);
1033 		return (EBUSY);
1034 	}
1035 
1036 	map = mac_find_macaddr(mip, mip->mi_addr);
1037 	ASSERT(map != NULL);
1038 
1039 	/*
1040 	 * Update the MAC address.
1041 	 */
1042 	if (mip->mi_state_flags & MIS_IS_AGGR) {
1043 		mac_capab_aggr_t aggr_cap;
1044 
1045 		/*
1046 		 * If the MAC is an aggregation, other than the unicast
1047 		 * addresses programming, aggr must be informed about this
1048 		 * primary unicst address change to change its MAC address
1049 		 * policy to be user-specified.
1050 		 */
1051 		ASSERT(map->ma_type == MAC_ADDRESS_TYPE_UNICAST_CLASSIFIED);
1052 		VERIFY(i_mac_capab_get(mh, MAC_CAPAB_AGGR, &aggr_cap));
1053 		err = aggr_cap.mca_unicst(mip->mi_driver, addr);
1054 		if (err == 0)
1055 			bcopy(addr, map->ma_addr, map->ma_len);
1056 	} else {
1057 		err = mac_update_macaddr(map, (uint8_t *)addr);
1058 	}
1059 
1060 	if (err != 0) {
1061 		i_mac_perim_exit(mip);
1062 		return (err);
1063 	}
1064 
1065 	mac_unicast_update_clients(mip, map);
1066 
1067 	/*
1068 	 * Save the new primary MAC address in mac_impl_t.
1069 	 */
1070 	bcopy(addr, mip->mi_addr, mip->mi_type->mt_addr_length);
1071 
1072 	i_mac_perim_exit(mip);
1073 
1074 	if (err == 0)
1075 		i_mac_notify(mip, MAC_NOTE_UNICST);
1076 
1077 	return (err);
1078 }
1079 
1080 /*
1081  * Return the current primary MAC address of the specified MAC.
1082  */
1083 void
1084 mac_unicast_primary_get(mac_handle_t mh, uint8_t *addr)
1085 {
1086 	mac_impl_t *mip = (mac_impl_t *)mh;
1087 
1088 	rw_enter(&mip->mi_rw_lock, RW_READER);
1089 	bcopy(mip->mi_addr, addr, mip->mi_type->mt_addr_length);
1090 	rw_exit(&mip->mi_rw_lock);
1091 }
1092 
1093 /*
1094  * Return the secondary MAC address for the specified handle
1095  */
1096 void
1097 mac_unicast_secondary_get(mac_client_handle_t mh, uint8_t *addr)
1098 {
1099 	mac_client_impl_t *mcip = (mac_client_impl_t *)mh;
1100 
1101 	ASSERT(mcip->mci_unicast != NULL);
1102 	bcopy(mcip->mci_unicast->ma_addr, addr, mcip->mci_unicast->ma_len);
1103 }
1104 
1105 /*
1106  * Return information about the use of the primary MAC address of the
1107  * specified MAC instance:
1108  *
1109  * - if client_name is non-NULL, it must point to a string of at
1110  *   least MAXNAMELEN bytes, and will be set to the name of the MAC
1111  *   client which uses the primary MAC address.
1112  *
1113  * - if in_use is non-NULL, used to return whether the primary MAC
1114  *   address is currently in use.
1115  */
1116 void
1117 mac_unicast_primary_info(mac_handle_t mh, char *client_name, boolean_t *in_use)
1118 {
1119 	mac_impl_t *mip = (mac_impl_t *)mh;
1120 	mac_client_impl_t *cur_client;
1121 
1122 	if (in_use != NULL)
1123 		*in_use = B_FALSE;
1124 	if (client_name != NULL)
1125 		bzero(client_name, MAXNAMELEN);
1126 
1127 	/*
1128 	 * The mi_rw_lock is used to protect threads that don't hold the
1129 	 * mac perimeter to get a consistent view of the mi_clients_list.
1130 	 * Threads that modify the list must hold both the mac perimeter and
1131 	 * mi_rw_lock(RW_WRITER)
1132 	 */
1133 	rw_enter(&mip->mi_rw_lock, RW_READER);
1134 	for (cur_client = mip->mi_clients_list; cur_client != NULL;
1135 	    cur_client = cur_client->mci_client_next) {
1136 		if (mac_is_primary_client(cur_client) ||
1137 		    (mip->mi_state_flags & MIS_IS_VNIC)) {
1138 			rw_exit(&mip->mi_rw_lock);
1139 			if (in_use != NULL)
1140 				*in_use = B_TRUE;
1141 			if (client_name != NULL) {
1142 				bcopy(cur_client->mci_name, client_name,
1143 				    MAXNAMELEN);
1144 			}
1145 			return;
1146 		}
1147 	}
1148 	rw_exit(&mip->mi_rw_lock);
1149 }
1150 
1151 /*
1152  * Return the current destination MAC address of the specified MAC.
1153  */
1154 boolean_t
1155 mac_dst_get(mac_handle_t mh, uint8_t *addr)
1156 {
1157 	mac_impl_t *mip = (mac_impl_t *)mh;
1158 
1159 	rw_enter(&mip->mi_rw_lock, RW_READER);
1160 	if (mip->mi_dstaddr_set)
1161 		bcopy(mip->mi_dstaddr, addr, mip->mi_type->mt_addr_length);
1162 	rw_exit(&mip->mi_rw_lock);
1163 	return (mip->mi_dstaddr_set);
1164 }
1165 
1166 /*
1167  * Add the specified MAC client to the list of clients which opened
1168  * the specified MAC.
1169  */
1170 static void
1171 mac_client_add(mac_client_impl_t *mcip)
1172 {
1173 	mac_impl_t *mip = mcip->mci_mip;
1174 
1175 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1176 
1177 	/* add VNIC to the front of the list */
1178 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
1179 	mcip->mci_client_next = mip->mi_clients_list;
1180 	mip->mi_clients_list = mcip;
1181 	mip->mi_nclients++;
1182 	rw_exit(&mip->mi_rw_lock);
1183 }
1184 
1185 /*
1186  * Remove the specified MAC client from the list of clients which opened
1187  * the specified MAC.
1188  */
1189 static void
1190 mac_client_remove(mac_client_impl_t *mcip)
1191 {
1192 	mac_impl_t *mip = mcip->mci_mip;
1193 	mac_client_impl_t **prev, *cclient;
1194 
1195 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1196 
1197 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
1198 	prev = &mip->mi_clients_list;
1199 	cclient = *prev;
1200 	while (cclient != NULL && cclient != mcip) {
1201 		prev = &cclient->mci_client_next;
1202 		cclient = *prev;
1203 	}
1204 	ASSERT(cclient != NULL);
1205 	*prev = cclient->mci_client_next;
1206 	mip->mi_nclients--;
1207 	rw_exit(&mip->mi_rw_lock);
1208 }
1209 
1210 static mac_unicast_impl_t *
1211 mac_client_find_vid(mac_client_impl_t *mcip, uint16_t vid)
1212 {
1213 	mac_unicast_impl_t *muip = mcip->mci_unicast_list;
1214 
1215 	while ((muip != NULL) && (muip->mui_vid != vid))
1216 		muip = muip->mui_next;
1217 
1218 	return (muip);
1219 }
1220 
1221 /*
1222  * Return whether the specified (MAC address, VID) tuple is already used by
1223  * one of the MAC clients associated with the specified MAC.
1224  */
1225 static boolean_t
1226 mac_addr_in_use(mac_impl_t *mip, uint8_t *mac_addr, uint16_t vid)
1227 {
1228 	mac_client_impl_t *client;
1229 	mac_address_t *map;
1230 
1231 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1232 
1233 	for (client = mip->mi_clients_list; client != NULL;
1234 	    client = client->mci_client_next) {
1235 
1236 		/*
1237 		 * Ignore clients that don't have unicast address.
1238 		 */
1239 		if (client->mci_unicast_list == NULL)
1240 			continue;
1241 
1242 		map = client->mci_unicast;
1243 
1244 		if ((bcmp(mac_addr, map->ma_addr, map->ma_len) == 0) &&
1245 		    (mac_client_find_vid(client, vid) != NULL)) {
1246 			return (B_TRUE);
1247 		}
1248 	}
1249 
1250 	return (B_FALSE);
1251 }
1252 
1253 /*
1254  * Generate a random MAC address. The MAC address prefix is
1255  * stored in the array pointed to by mac_addr, and its length, in bytes,
1256  * is specified by prefix_len. The least significant bits
1257  * after prefix_len bytes are generated, and stored after the prefix
1258  * in the mac_addr array.
1259  */
1260 int
1261 mac_addr_random(mac_client_handle_t mch, uint_t prefix_len,
1262     uint8_t *mac_addr, mac_diag_t *diag)
1263 {
1264 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1265 	mac_impl_t *mip = mcip->mci_mip;
1266 	size_t addr_len = mip->mi_type->mt_addr_length;
1267 
1268 	if (prefix_len >= addr_len) {
1269 		*diag = MAC_DIAG_MACPREFIXLEN_INVALID;
1270 		return (EINVAL);
1271 	}
1272 
1273 	/* check the prefix value */
1274 	if (prefix_len > 0) {
1275 		bzero(mac_addr + prefix_len, addr_len - prefix_len);
1276 		if (!mac_unicst_verify((mac_handle_t)mip, mac_addr,
1277 		    addr_len)) {
1278 			*diag = MAC_DIAG_MACPREFIX_INVALID;
1279 			return (EINVAL);
1280 		}
1281 	}
1282 
1283 	/* generate the MAC address */
1284 	if (prefix_len < addr_len) {
1285 		(void) random_get_pseudo_bytes(mac_addr +
1286 		    prefix_len, addr_len - prefix_len);
1287 	}
1288 
1289 	*diag = 0;
1290 	return (0);
1291 }
1292 
1293 /*
1294  * Set the priority range for this MAC client. This will be used to
1295  * determine the absolute priority for the threads created for this
1296  * MAC client using the specified "low", "medium" and "high" level.
1297  * This will also be used for any subflows on this MAC client.
1298  */
1299 #define	MAC_CLIENT_SET_PRIORITY_RANGE(mcip, pri) {			\
1300 	(mcip)->mci_min_pri = FLOW_MIN_PRIORITY(MINCLSYSPRI,	\
1301 	    MAXCLSYSPRI, (pri));					\
1302 	(mcip)->mci_max_pri = FLOW_MAX_PRIORITY(MINCLSYSPRI,	\
1303 	    MAXCLSYSPRI, (mcip)->mci_min_pri);				\
1304 	}
1305 
1306 /*
1307  * MAC client open entry point. Return a new MAC client handle. Each
1308  * MAC client is associated with a name, specified through the 'name'
1309  * argument.
1310  */
1311 int
1312 mac_client_open(mac_handle_t mh, mac_client_handle_t *mchp, char *name,
1313     uint16_t flags)
1314 {
1315 	mac_impl_t		*mip = (mac_impl_t *)mh;
1316 	mac_client_impl_t	*mcip;
1317 	int			err = 0;
1318 	boolean_t		share_desired;
1319 	flow_entry_t		*flent = NULL;
1320 
1321 	share_desired = (flags & MAC_OPEN_FLAGS_SHARES_DESIRED) != 0;
1322 	*mchp = NULL;
1323 
1324 	i_mac_perim_enter(mip);
1325 
1326 	if (mip->mi_state_flags & MIS_IS_VNIC) {
1327 		/*
1328 		 * The underlying MAC is a VNIC. Return the MAC client
1329 		 * handle of the lower MAC which was obtained by
1330 		 * the VNIC driver when it did its mac_client_open().
1331 		 */
1332 
1333 		mcip = mac_vnic_lower(mip);
1334 
1335 		/*
1336 		 * Note that multiple mac clients share the same mcip in
1337 		 * this case.
1338 		 */
1339 		if (flags & MAC_OPEN_FLAGS_EXCLUSIVE)
1340 			mcip->mci_state_flags |= MCIS_EXCLUSIVE;
1341 
1342 		if (flags & MAC_OPEN_FLAGS_MULTI_PRIMARY)
1343 			mcip->mci_flags |= MAC_CLIENT_FLAGS_MULTI_PRIMARY;
1344 
1345 		mip->mi_clients_list = mcip;
1346 		i_mac_perim_exit(mip);
1347 		*mchp = (mac_client_handle_t)mcip;
1348 
1349 		DTRACE_PROBE2(mac__client__open__nonallocated, mac_impl_t *,
1350 		    mcip->mci_mip, mac_client_impl_t *, mcip);
1351 
1352 		return (err);
1353 	}
1354 
1355 	mcip = kmem_cache_alloc(mac_client_impl_cache, KM_SLEEP);
1356 
1357 	mcip->mci_mip = mip;
1358 	mcip->mci_upper_mip = NULL;
1359 	mcip->mci_rx_fn = mac_pkt_drop;
1360 	mcip->mci_rx_arg = NULL;
1361 	mcip->mci_rx_p_fn = NULL;
1362 	mcip->mci_rx_p_arg = NULL;
1363 	mcip->mci_p_unicast_list = NULL;
1364 	mcip->mci_direct_rx_fn = NULL;
1365 	mcip->mci_direct_rx_arg = NULL;
1366 	mcip->mci_vidcache = MCIP_VIDCACHE_INVALID;
1367 
1368 	mcip->mci_unicast_list = NULL;
1369 
1370 	if ((flags & MAC_OPEN_FLAGS_IS_VNIC) != 0)
1371 		mcip->mci_state_flags |= MCIS_IS_VNIC;
1372 
1373 	if ((flags & MAC_OPEN_FLAGS_EXCLUSIVE) != 0)
1374 		mcip->mci_state_flags |= MCIS_EXCLUSIVE;
1375 
1376 	if ((flags & MAC_OPEN_FLAGS_IS_AGGR_PORT) != 0)
1377 		mcip->mci_state_flags |= MCIS_IS_AGGR_PORT;
1378 
1379 	if (mip->mi_state_flags & MIS_IS_AGGR)
1380 		mcip->mci_state_flags |= MCIS_IS_AGGR_CLIENT;
1381 
1382 	if ((flags & MAC_OPEN_FLAGS_USE_DATALINK_NAME) != 0) {
1383 		datalink_id_t	linkid;
1384 
1385 		ASSERT(name == NULL);
1386 		if ((err = dls_devnet_macname2linkid(mip->mi_name,
1387 		    &linkid)) != 0) {
1388 			goto done;
1389 		}
1390 		if ((err = dls_mgmt_get_linkinfo(linkid, mcip->mci_name, NULL,
1391 		    NULL, NULL)) != 0) {
1392 			/*
1393 			 * Use mac name if dlmgmtd is not available.
1394 			 */
1395 			if (err == EBADF) {
1396 				(void) strlcpy(mcip->mci_name, mip->mi_name,
1397 				    sizeof (mcip->mci_name));
1398 				err = 0;
1399 			} else {
1400 				goto done;
1401 			}
1402 		}
1403 		mcip->mci_state_flags |= MCIS_USE_DATALINK_NAME;
1404 	} else {
1405 		ASSERT(name != NULL);
1406 		if (strlen(name) > MAXNAMELEN) {
1407 			err = EINVAL;
1408 			goto done;
1409 		}
1410 		(void) strlcpy(mcip->mci_name, name, sizeof (mcip->mci_name));
1411 	}
1412 
1413 	if (flags & MAC_OPEN_FLAGS_MULTI_PRIMARY)
1414 		mcip->mci_flags |= MAC_CLIENT_FLAGS_MULTI_PRIMARY;
1415 
1416 	if (flags & MAC_OPEN_FLAGS_NO_UNICAST_ADDR)
1417 		mcip->mci_state_flags |= MCIS_NO_UNICAST_ADDR;
1418 
1419 	mac_protect_init(mcip);
1420 
1421 	/* the subflow table will be created dynamically */
1422 	mcip->mci_subflow_tab = NULL;
1423 
1424 	mcip->mci_misc_stat.mms_multircv = 0;
1425 	mcip->mci_misc_stat.mms_brdcstrcv = 0;
1426 	mcip->mci_misc_stat.mms_multixmt = 0;
1427 	mcip->mci_misc_stat.mms_brdcstxmt = 0;
1428 
1429 	/* Create an initial flow */
1430 
1431 	err = mac_flow_create(NULL, NULL, mcip->mci_name, NULL,
1432 	    mcip->mci_state_flags & MCIS_IS_VNIC ? FLOW_VNIC_MAC :
1433 	    FLOW_PRIMARY_MAC, &flent);
1434 	if (err != 0)
1435 		goto done;
1436 	mcip->mci_flent = flent;
1437 	FLOW_MARK(flent, FE_MC_NO_DATAPATH);
1438 	flent->fe_mcip = mcip;
1439 
1440 	/*
1441 	 * Place initial creation reference on the flow. This reference
1442 	 * is released in the corresponding delete action viz.
1443 	 * mac_unicast_remove after waiting for all transient refs to
1444 	 * to go away. The wait happens in mac_flow_wait.
1445 	 */
1446 	FLOW_REFHOLD(flent);
1447 
1448 	/*
1449 	 * Do this ahead of the mac_bcast_add() below so that the mi_nclients
1450 	 * will have the right value for mac_rx_srs_setup().
1451 	 */
1452 	mac_client_add(mcip);
1453 
1454 	mcip->mci_share = 0;
1455 	if (share_desired)
1456 		i_mac_share_alloc(mcip);
1457 
1458 	/*
1459 	 * We will do mimimal datapath setup to allow a MAC client to
1460 	 * transmit or receive non-unicast packets without waiting
1461 	 * for mac_unicast_add.
1462 	 */
1463 	if (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR) {
1464 		if ((err = mac_client_datapath_setup(mcip, VLAN_ID_NONE,
1465 		    NULL, NULL, B_TRUE, NULL)) != 0) {
1466 			goto done;
1467 		}
1468 	}
1469 
1470 	DTRACE_PROBE2(mac__client__open__allocated, mac_impl_t *,
1471 	    mcip->mci_mip, mac_client_impl_t *, mcip);
1472 
1473 	*mchp = (mac_client_handle_t)mcip;
1474 	i_mac_perim_exit(mip);
1475 	return (0);
1476 
1477 done:
1478 	i_mac_perim_exit(mip);
1479 	mcip->mci_state_flags = 0;
1480 	mcip->mci_tx_flag = 0;
1481 	kmem_cache_free(mac_client_impl_cache, mcip);
1482 	return (err);
1483 }
1484 
1485 /*
1486  * Close the specified MAC client handle.
1487  */
1488 void
1489 mac_client_close(mac_client_handle_t mch, uint16_t flags)
1490 {
1491 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
1492 	mac_impl_t		*mip = mcip->mci_mip;
1493 	flow_entry_t		*flent;
1494 
1495 	i_mac_perim_enter(mip);
1496 
1497 	if (flags & MAC_CLOSE_FLAGS_EXCLUSIVE)
1498 		mcip->mci_state_flags &= ~MCIS_EXCLUSIVE;
1499 
1500 	if ((mcip->mci_state_flags & MCIS_IS_VNIC) &&
1501 	    !(flags & MAC_CLOSE_FLAGS_IS_VNIC)) {
1502 		/*
1503 		 * This is an upper VNIC client initiated operation.
1504 		 * The lower MAC client will be closed by the VNIC driver
1505 		 * when the VNIC is deleted.
1506 		 */
1507 
1508 		i_mac_perim_exit(mip);
1509 		return;
1510 	}
1511 
1512 	/* If we have only setup up minimal datapth setup, tear it down */
1513 	if (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR) {
1514 		mac_client_datapath_teardown((mac_client_handle_t)mcip, NULL,
1515 		    mcip->mci_flent);
1516 		mcip->mci_state_flags &= ~MCIS_NO_UNICAST_ADDR;
1517 	}
1518 
1519 	/*
1520 	 * Remove the flent associated with the MAC client
1521 	 */
1522 	flent = mcip->mci_flent;
1523 	mcip->mci_flent = NULL;
1524 	FLOW_FINAL_REFRELE(flent);
1525 
1526 	/*
1527 	 * MAC clients must remove the unicast addresses and promisc callbacks
1528 	 * they added before issuing a mac_client_close().
1529 	 */
1530 	ASSERT(mcip->mci_unicast_list == NULL);
1531 	ASSERT(mcip->mci_promisc_list == NULL);
1532 	ASSERT(mcip->mci_tx_notify_cb_list == NULL);
1533 
1534 	i_mac_share_free(mcip);
1535 	mac_protect_fini(mcip);
1536 	mac_client_remove(mcip);
1537 
1538 	i_mac_perim_exit(mip);
1539 	mcip->mci_subflow_tab = NULL;
1540 	mcip->mci_state_flags = 0;
1541 	mcip->mci_tx_flag = 0;
1542 	kmem_cache_free(mac_client_impl_cache, mch);
1543 }
1544 
1545 /*
1546  * Set the Rx bypass receive callback and return B_TRUE. Return
1547  * B_FALSE if it's not possible to enable bypass.
1548  */
1549 boolean_t
1550 mac_rx_bypass_set(mac_client_handle_t mch, mac_direct_rx_t rx_fn, void *arg1)
1551 {
1552 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
1553 	mac_impl_t		*mip = mcip->mci_mip;
1554 
1555 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1556 
1557 	/*
1558 	 * If the client has more than one VLAN then process packets
1559 	 * through DLS. This should happen only when sun4v vsw is on
1560 	 * the scene.
1561 	 */
1562 	if (mcip->mci_nvids > 1)
1563 		return (B_FALSE);
1564 
1565 	/*
1566 	 * These are not accessed directly in the data path, and hence
1567 	 * don't need any protection
1568 	 */
1569 	mcip->mci_direct_rx_fn = rx_fn;
1570 	mcip->mci_direct_rx_arg = arg1;
1571 	return (B_TRUE);
1572 }
1573 
1574 /*
1575  * Enable/Disable rx bypass. By default, bypass is assumed to be enabled.
1576  */
1577 void
1578 mac_rx_bypass_enable(mac_client_handle_t mch)
1579 {
1580 	((mac_client_impl_t *)mch)->mci_state_flags &= ~MCIS_RX_BYPASS_DISABLE;
1581 }
1582 
1583 void
1584 mac_rx_bypass_disable(mac_client_handle_t mch)
1585 {
1586 	((mac_client_impl_t *)mch)->mci_state_flags |= MCIS_RX_BYPASS_DISABLE;
1587 }
1588 
1589 /*
1590  * Set the receive callback for the specified MAC client. There can be
1591  * at most one such callback per MAC client.
1592  */
1593 void
1594 mac_rx_set(mac_client_handle_t mch, mac_rx_t rx_fn, void *arg)
1595 {
1596 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1597 	mac_impl_t	*mip = mcip->mci_mip;
1598 	mac_impl_t	*umip = mcip->mci_upper_mip;
1599 
1600 	/*
1601 	 * Instead of adding an extra set of locks and refcnts in
1602 	 * the datapath at the mac client boundary, we temporarily quiesce
1603 	 * the SRS and related entities. We then change the receive function
1604 	 * without interference from any receive data thread and then reenable
1605 	 * the data flow subsequently.
1606 	 */
1607 	i_mac_perim_enter(mip);
1608 	mac_rx_client_quiesce(mch);
1609 
1610 	mcip->mci_rx_fn = rx_fn;
1611 	mcip->mci_rx_arg = arg;
1612 	mac_rx_client_restart(mch);
1613 	i_mac_perim_exit(mip);
1614 
1615 	/*
1616 	 * If we're changing the Rx function on the primary MAC of a VNIC,
1617 	 * make sure any secondary addresses on the VNIC are updated as well.
1618 	 */
1619 	if (umip != NULL) {
1620 		ASSERT((umip->mi_state_flags & MIS_IS_VNIC) != 0);
1621 		mac_vnic_secondary_update(umip);
1622 	}
1623 }
1624 
1625 /*
1626  * Reset the receive callback for the specified MAC client.
1627  */
1628 void
1629 mac_rx_clear(mac_client_handle_t mch)
1630 {
1631 	mac_rx_set(mch, mac_pkt_drop, NULL);
1632 }
1633 
1634 void
1635 mac_secondary_dup(mac_client_handle_t smch, mac_client_handle_t dmch)
1636 {
1637 	mac_client_impl_t *smcip = (mac_client_impl_t *)smch;
1638 	mac_client_impl_t *dmcip = (mac_client_impl_t *)dmch;
1639 	flow_entry_t *flent = dmcip->mci_flent;
1640 
1641 	/* This should only be called to setup secondary macs */
1642 	ASSERT((flent->fe_type & FLOW_PRIMARY_MAC) == 0);
1643 
1644 	mac_rx_set(dmch, smcip->mci_rx_fn, smcip->mci_rx_arg);
1645 	dmcip->mci_promisc_list = smcip->mci_promisc_list;
1646 
1647 	/*
1648 	 * Duplicate the primary mac resources to the secondary.
1649 	 * Since we already validated the resource controls when setting
1650 	 * them on the primary, we can ignore errors here.
1651 	 */
1652 	(void) mac_resource_ctl_set(dmch, MCIP_RESOURCE_PROPS(smcip));
1653 }
1654 
1655 /*
1656  * Called when removing a secondary MAC. Currently only clears the promisc_list
1657  * since we share the primary mac's promisc_list.
1658  */
1659 void
1660 mac_secondary_cleanup(mac_client_handle_t mch)
1661 {
1662 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1663 	flow_entry_t *flent = mcip->mci_flent;
1664 
1665 	/* This should only be called for secondary macs */
1666 	ASSERT((flent->fe_type & FLOW_PRIMARY_MAC) == 0);
1667 	mcip->mci_promisc_list = NULL;
1668 }
1669 
1670 /*
1671  * Walk the MAC client subflow table and updates their priority values.
1672  */
1673 static int
1674 mac_update_subflow_priority_cb(flow_entry_t *flent, void *arg)
1675 {
1676 	mac_flow_update_priority(arg, flent);
1677 	return (0);
1678 }
1679 
1680 void
1681 mac_update_subflow_priority(mac_client_impl_t *mcip)
1682 {
1683 	(void) mac_flow_walk(mcip->mci_subflow_tab,
1684 	    mac_update_subflow_priority_cb, mcip);
1685 }
1686 
1687 /*
1688  * Modify the TX or RX ring properties. We could either just move around
1689  * rings, i.e add/remove rings given to a client. Or this might cause the
1690  * client to move from hardware based to software or the other way around.
1691  * If we want to reset this property, then we clear the mask, additionally
1692  * if the client was given a non-default group we remove all rings except
1693  * for 1 and give it back to the default group.
1694  */
1695 int
1696 mac_client_set_rings_prop(mac_client_impl_t *mcip, mac_resource_props_t *mrp,
1697     mac_resource_props_t *tmrp)
1698 {
1699 	mac_impl_t		*mip = mcip->mci_mip;
1700 	flow_entry_t		*flent = mcip->mci_flent;
1701 	uint8_t			*mac_addr;
1702 	int			err = 0;
1703 	mac_group_t		*defgrp;
1704 	mac_group_t		*group;
1705 	mac_group_t		*ngrp;
1706 	mac_resource_props_t	*cmrp = MCIP_RESOURCE_PROPS(mcip);
1707 	uint_t			ringcnt;
1708 	boolean_t		unspec;
1709 
1710 	if (mcip->mci_share != 0)
1711 		return (EINVAL);
1712 
1713 	if (mrp->mrp_mask & MRP_RX_RINGS) {
1714 		unspec = mrp->mrp_mask & MRP_RXRINGS_UNSPEC;
1715 		group = flent->fe_rx_ring_group;
1716 		defgrp = MAC_DEFAULT_RX_GROUP(mip);
1717 		mac_addr = flent->fe_flow_desc.fd_dst_mac;
1718 
1719 		/*
1720 		 * No resulting change. If we are resetting on a client on
1721 		 * which there was no rx rings property. For dynamic group
1722 		 * if we are setting the same number of rings already set.
1723 		 * For static group if we are requesting a group again.
1724 		 */
1725 		if (mrp->mrp_mask & MRP_RINGS_RESET) {
1726 			if (!(tmrp->mrp_mask & MRP_RX_RINGS))
1727 				return (0);
1728 		} else {
1729 			if (unspec) {
1730 				if (tmrp->mrp_mask & MRP_RXRINGS_UNSPEC)
1731 					return (0);
1732 			} else if (mip->mi_rx_group_type ==
1733 			    MAC_GROUP_TYPE_DYNAMIC) {
1734 				if ((tmrp->mrp_mask & MRP_RX_RINGS) &&
1735 				    !(tmrp->mrp_mask & MRP_RXRINGS_UNSPEC) &&
1736 				    mrp->mrp_nrxrings == tmrp->mrp_nrxrings) {
1737 					return (0);
1738 				}
1739 			}
1740 		}
1741 		/* Resetting the prop */
1742 		if (mrp->mrp_mask & MRP_RINGS_RESET) {
1743 			/*
1744 			 * We will just keep one ring and give others back if
1745 			 * we are not the primary. For the primary we give
1746 			 * all the rings in the default group except the
1747 			 * default ring. If it is a static group, then
1748 			 * we don't do anything, but clear the MRP_RX_RINGS
1749 			 * flag.
1750 			 */
1751 			if (group != defgrp) {
1752 				if (mip->mi_rx_group_type ==
1753 				    MAC_GROUP_TYPE_DYNAMIC) {
1754 					/*
1755 					 * This group has reserved rings
1756 					 * that need to be released now,
1757 					 * so does the group.
1758 					 */
1759 					MAC_RX_RING_RELEASED(mip,
1760 					    group->mrg_cur_count);
1761 					MAC_RX_GRP_RELEASED(mip);
1762 					if ((flent->fe_type &
1763 					    FLOW_PRIMARY_MAC) != 0) {
1764 						if (mip->mi_nactiveclients ==
1765 						    1) {
1766 							(void)
1767 							    mac_rx_switch_group(
1768 							    mcip, group,
1769 							    defgrp);
1770 							return (0);
1771 						} else {
1772 							cmrp->mrp_nrxrings =
1773 							    group->
1774 							    mrg_cur_count +
1775 							    defgrp->
1776 							    mrg_cur_count - 1;
1777 						}
1778 					} else {
1779 						cmrp->mrp_nrxrings = 1;
1780 					}
1781 					(void) mac_group_ring_modify(mcip,
1782 					    group, defgrp);
1783 				} else {
1784 					/*
1785 					 * If this is a static group, we
1786 					 * need to release the group. The
1787 					 * client will remain in the same
1788 					 * group till some other client
1789 					 * needs this group.
1790 					 */
1791 					MAC_RX_GRP_RELEASED(mip);
1792 				}
1793 			/* Let check if we can give this an excl group */
1794 			} else if (group == defgrp) {
1795 				/*
1796 				 * If multiple clients share an
1797 				 * address then they must stay on the
1798 				 * default group.
1799 				 */
1800 				if (mac_check_macaddr_shared(mcip->mci_unicast))
1801 					return (0);
1802 
1803 				ngrp = mac_reserve_rx_group(mcip, mac_addr,
1804 				    B_TRUE);
1805 				/* Couldn't give it a group, that's fine */
1806 				if (ngrp == NULL)
1807 					return (0);
1808 				/* Switch to H/W */
1809 				if (mac_rx_switch_group(mcip, defgrp, ngrp) !=
1810 				    0) {
1811 					mac_stop_group(ngrp);
1812 					return (0);
1813 				}
1814 			}
1815 			/*
1816 			 * If the client is in the default group, we will
1817 			 * just clear the MRP_RX_RINGS and leave it as
1818 			 * it rather than look for an exclusive group
1819 			 * for it.
1820 			 */
1821 			return (0);
1822 		}
1823 
1824 		if (group == defgrp && ((mrp->mrp_nrxrings > 0) || unspec)) {
1825 			/*
1826 			 * We are requesting Rx rings. Try to reserve
1827 			 * a non-default group.
1828 			 *
1829 			 * If multiple clients share an address then
1830 			 * they must stay on the default group.
1831 			 */
1832 			if (mac_check_macaddr_shared(mcip->mci_unicast))
1833 				return (EINVAL);
1834 
1835 			ngrp = mac_reserve_rx_group(mcip, mac_addr, B_TRUE);
1836 			if (ngrp == NULL)
1837 				return (ENOSPC);
1838 
1839 			/* Switch to H/W */
1840 			if (mac_rx_switch_group(mcip, defgrp, ngrp) != 0) {
1841 				mac_release_rx_group(mcip, ngrp);
1842 				return (ENOSPC);
1843 			}
1844 			MAC_RX_GRP_RESERVED(mip);
1845 			if (mip->mi_rx_group_type == MAC_GROUP_TYPE_DYNAMIC)
1846 				MAC_RX_RING_RESERVED(mip, ngrp->mrg_cur_count);
1847 		} else if (group != defgrp && !unspec &&
1848 		    mrp->mrp_nrxrings == 0) {
1849 			/* Switch to S/W */
1850 			ringcnt = group->mrg_cur_count;
1851 			if (mac_rx_switch_group(mcip, group, defgrp) != 0)
1852 				return (ENOSPC);
1853 			if (tmrp->mrp_mask & MRP_RX_RINGS) {
1854 				MAC_RX_GRP_RELEASED(mip);
1855 				if (mip->mi_rx_group_type ==
1856 				    MAC_GROUP_TYPE_DYNAMIC) {
1857 					MAC_RX_RING_RELEASED(mip, ringcnt);
1858 				}
1859 			}
1860 		} else if (group != defgrp && mip->mi_rx_group_type ==
1861 		    MAC_GROUP_TYPE_DYNAMIC) {
1862 			ringcnt = group->mrg_cur_count;
1863 			err = mac_group_ring_modify(mcip, group, defgrp);
1864 			if (err != 0)
1865 				return (err);
1866 			/*
1867 			 * Update the accounting. If this group
1868 			 * already had explicitly reserved rings,
1869 			 * we need to update the rings based on
1870 			 * the new ring count. If this group
1871 			 * had not explicitly reserved rings,
1872 			 * then we just reserve the rings asked for
1873 			 * and reserve the group.
1874 			 */
1875 			if (tmrp->mrp_mask & MRP_RX_RINGS) {
1876 				if (ringcnt > group->mrg_cur_count) {
1877 					MAC_RX_RING_RELEASED(mip,
1878 					    ringcnt - group->mrg_cur_count);
1879 				} else {
1880 					MAC_RX_RING_RESERVED(mip,
1881 					    group->mrg_cur_count - ringcnt);
1882 				}
1883 			} else {
1884 				MAC_RX_RING_RESERVED(mip, group->mrg_cur_count);
1885 				MAC_RX_GRP_RESERVED(mip);
1886 			}
1887 		}
1888 	}
1889 	if (mrp->mrp_mask & MRP_TX_RINGS) {
1890 		unspec = mrp->mrp_mask & MRP_TXRINGS_UNSPEC;
1891 		group = flent->fe_tx_ring_group;
1892 		defgrp = MAC_DEFAULT_TX_GROUP(mip);
1893 
1894 		/*
1895 		 * For static groups we only allow rings=0 or resetting the
1896 		 * rings property.
1897 		 */
1898 		if (mrp->mrp_ntxrings > 0 &&
1899 		    mip->mi_tx_group_type != MAC_GROUP_TYPE_DYNAMIC) {
1900 			return (ENOTSUP);
1901 		}
1902 		if (mrp->mrp_mask & MRP_RINGS_RESET) {
1903 			if (!(tmrp->mrp_mask & MRP_TX_RINGS))
1904 				return (0);
1905 		} else {
1906 			if (unspec) {
1907 				if (tmrp->mrp_mask & MRP_TXRINGS_UNSPEC)
1908 					return (0);
1909 			} else if (mip->mi_tx_group_type ==
1910 			    MAC_GROUP_TYPE_DYNAMIC) {
1911 				if ((tmrp->mrp_mask & MRP_TX_RINGS) &&
1912 				    !(tmrp->mrp_mask & MRP_TXRINGS_UNSPEC) &&
1913 				    mrp->mrp_ntxrings == tmrp->mrp_ntxrings) {
1914 					return (0);
1915 				}
1916 			}
1917 		}
1918 		/* Resetting the prop */
1919 		if (mrp->mrp_mask & MRP_RINGS_RESET) {
1920 			if (group != defgrp) {
1921 				if (mip->mi_tx_group_type ==
1922 				    MAC_GROUP_TYPE_DYNAMIC) {
1923 					ringcnt = group->mrg_cur_count;
1924 					if ((flent->fe_type &
1925 					    FLOW_PRIMARY_MAC) != 0) {
1926 						mac_tx_client_quiesce(
1927 						    (mac_client_handle_t)
1928 						    mcip);
1929 						mac_tx_switch_group(mcip,
1930 						    group, defgrp);
1931 						mac_tx_client_restart(
1932 						    (mac_client_handle_t)
1933 						    mcip);
1934 						MAC_TX_GRP_RELEASED(mip);
1935 						MAC_TX_RING_RELEASED(mip,
1936 						    ringcnt);
1937 						return (0);
1938 					}
1939 					cmrp->mrp_ntxrings = 1;
1940 					(void) mac_group_ring_modify(mcip,
1941 					    group, defgrp);
1942 					/*
1943 					 * This group has reserved rings
1944 					 * that need to be released now.
1945 					 */
1946 					MAC_TX_RING_RELEASED(mip, ringcnt);
1947 				}
1948 				/*
1949 				 * If this is a static group, we
1950 				 * need to release the group. The
1951 				 * client will remain in the same
1952 				 * group till some other client
1953 				 * needs this group.
1954 				 */
1955 				MAC_TX_GRP_RELEASED(mip);
1956 			} else if (group == defgrp &&
1957 			    (flent->fe_type & FLOW_PRIMARY_MAC) == 0) {
1958 				ngrp = mac_reserve_tx_group(mcip, B_TRUE);
1959 				if (ngrp == NULL)
1960 					return (0);
1961 				mac_tx_client_quiesce(
1962 				    (mac_client_handle_t)mcip);
1963 				mac_tx_switch_group(mcip, defgrp, ngrp);
1964 				mac_tx_client_restart(
1965 				    (mac_client_handle_t)mcip);
1966 			}
1967 			/*
1968 			 * If the client is in the default group, we will
1969 			 * just clear the MRP_TX_RINGS and leave it as
1970 			 * it rather than look for an exclusive group
1971 			 * for it.
1972 			 */
1973 			return (0);
1974 		}
1975 
1976 		/* Switch to H/W */
1977 		if (group == defgrp && ((mrp->mrp_ntxrings > 0) || unspec)) {
1978 			ngrp = mac_reserve_tx_group(mcip, B_TRUE);
1979 			if (ngrp == NULL)
1980 				return (ENOSPC);
1981 			mac_tx_client_quiesce((mac_client_handle_t)mcip);
1982 			mac_tx_switch_group(mcip, defgrp, ngrp);
1983 			mac_tx_client_restart((mac_client_handle_t)mcip);
1984 			MAC_TX_GRP_RESERVED(mip);
1985 			if (mip->mi_tx_group_type == MAC_GROUP_TYPE_DYNAMIC)
1986 				MAC_TX_RING_RESERVED(mip, ngrp->mrg_cur_count);
1987 		/* Switch to S/W */
1988 		} else if (group != defgrp && !unspec &&
1989 		    mrp->mrp_ntxrings == 0) {
1990 			/* Switch to S/W */
1991 			ringcnt = group->mrg_cur_count;
1992 			mac_tx_client_quiesce((mac_client_handle_t)mcip);
1993 			mac_tx_switch_group(mcip, group, defgrp);
1994 			mac_tx_client_restart((mac_client_handle_t)mcip);
1995 			if (tmrp->mrp_mask & MRP_TX_RINGS) {
1996 				MAC_TX_GRP_RELEASED(mip);
1997 				if (mip->mi_tx_group_type ==
1998 				    MAC_GROUP_TYPE_DYNAMIC) {
1999 					MAC_TX_RING_RELEASED(mip, ringcnt);
2000 				}
2001 			}
2002 		} else if (group != defgrp && mip->mi_tx_group_type ==
2003 		    MAC_GROUP_TYPE_DYNAMIC) {
2004 			ringcnt = group->mrg_cur_count;
2005 			err = mac_group_ring_modify(mcip, group, defgrp);
2006 			if (err != 0)
2007 				return (err);
2008 			/*
2009 			 * Update the accounting. If this group
2010 			 * already had explicitly reserved rings,
2011 			 * we need to update the rings based on
2012 			 * the new ring count. If this group
2013 			 * had not explicitly reserved rings,
2014 			 * then we just reserve the rings asked for
2015 			 * and reserve the group.
2016 			 */
2017 			if (tmrp->mrp_mask & MRP_TX_RINGS) {
2018 				if (ringcnt > group->mrg_cur_count) {
2019 					MAC_TX_RING_RELEASED(mip,
2020 					    ringcnt - group->mrg_cur_count);
2021 				} else {
2022 					MAC_TX_RING_RESERVED(mip,
2023 					    group->mrg_cur_count - ringcnt);
2024 				}
2025 			} else {
2026 				MAC_TX_RING_RESERVED(mip, group->mrg_cur_count);
2027 				MAC_TX_GRP_RESERVED(mip);
2028 			}
2029 		}
2030 	}
2031 	return (0);
2032 }
2033 
2034 /*
2035  * When the MAC client is being brought up (i.e. we do a unicast_add) we need
2036  * to initialize the cpu and resource control structure in the
2037  * mac_client_impl_t from the mac_impl_t (i.e if there are any cached
2038  * properties before the flow entry for the unicast address was created).
2039  */
2040 static int
2041 mac_resource_ctl_set(mac_client_handle_t mch, mac_resource_props_t *mrp)
2042 {
2043 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2044 	mac_impl_t		*mip = (mac_impl_t *)mcip->mci_mip;
2045 	mac_impl_t		*umip = mcip->mci_upper_mip;
2046 	int			err = 0;
2047 	flow_entry_t		*flent = mcip->mci_flent;
2048 	mac_resource_props_t	*omrp, *nmrp = MCIP_RESOURCE_PROPS(mcip);
2049 
2050 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
2051 
2052 	err = mac_validate_props(mcip->mci_state_flags & MCIS_IS_VNIC ?
2053 	    mcip->mci_upper_mip : mip, mrp);
2054 	if (err != 0)
2055 		return (err);
2056 
2057 	/*
2058 	 * Copy over the existing properties since mac_update_resources
2059 	 * will modify the client's mrp. Currently, the saved property
2060 	 * is used to determine the difference between existing and
2061 	 * modified rings property.
2062 	 */
2063 	omrp = kmem_zalloc(sizeof (*omrp), KM_SLEEP);
2064 	bcopy(nmrp, omrp, sizeof (*omrp));
2065 	mac_update_resources(mrp, MCIP_RESOURCE_PROPS(mcip), B_FALSE);
2066 	if (MCIP_DATAPATH_SETUP(mcip)) {
2067 		/*
2068 		 * We support rings only for primary client when there are
2069 		 * multiple clients sharing the same MAC address (e.g. VLAN).
2070 		 */
2071 		if (mrp->mrp_mask & MRP_RX_RINGS ||
2072 		    mrp->mrp_mask & MRP_TX_RINGS) {
2073 
2074 			if ((err = mac_client_set_rings_prop(mcip, mrp,
2075 			    omrp)) != 0) {
2076 				if (omrp->mrp_mask & MRP_RX_RINGS) {
2077 					nmrp->mrp_mask |= MRP_RX_RINGS;
2078 					nmrp->mrp_nrxrings = omrp->mrp_nrxrings;
2079 				} else {
2080 					nmrp->mrp_mask &= ~MRP_RX_RINGS;
2081 					nmrp->mrp_nrxrings = 0;
2082 				}
2083 				if (omrp->mrp_mask & MRP_TX_RINGS) {
2084 					nmrp->mrp_mask |= MRP_TX_RINGS;
2085 					nmrp->mrp_ntxrings = omrp->mrp_ntxrings;
2086 				} else {
2087 					nmrp->mrp_mask &= ~MRP_TX_RINGS;
2088 					nmrp->mrp_ntxrings = 0;
2089 				}
2090 				if (omrp->mrp_mask & MRP_RXRINGS_UNSPEC)
2091 					omrp->mrp_mask |= MRP_RXRINGS_UNSPEC;
2092 				else
2093 					omrp->mrp_mask &= ~MRP_RXRINGS_UNSPEC;
2094 
2095 				if (omrp->mrp_mask & MRP_TXRINGS_UNSPEC)
2096 					omrp->mrp_mask |= MRP_TXRINGS_UNSPEC;
2097 				else
2098 					omrp->mrp_mask &= ~MRP_TXRINGS_UNSPEC;
2099 				kmem_free(omrp, sizeof (*omrp));
2100 				return (err);
2101 			}
2102 
2103 			/*
2104 			 * If we modified the rings property of the primary
2105 			 * we need to update the property fields of its
2106 			 * VLANs as they inherit the primary's properites.
2107 			 */
2108 			if (mac_is_primary_client(mcip)) {
2109 				mac_set_prim_vlan_rings(mip,
2110 				    MCIP_RESOURCE_PROPS(mcip));
2111 			}
2112 		}
2113 		/*
2114 		 * We have to set this prior to calling mac_flow_modify.
2115 		 */
2116 		if (mrp->mrp_mask & MRP_PRIORITY) {
2117 			if (mrp->mrp_priority == MPL_RESET) {
2118 				MAC_CLIENT_SET_PRIORITY_RANGE(mcip,
2119 				    MPL_LINK_DEFAULT);
2120 			} else {
2121 				MAC_CLIENT_SET_PRIORITY_RANGE(mcip,
2122 				    mrp->mrp_priority);
2123 			}
2124 		}
2125 
2126 		mac_flow_modify(mip->mi_flow_tab, flent, mrp);
2127 		if (mrp->mrp_mask & MRP_PRIORITY)
2128 			mac_update_subflow_priority(mcip);
2129 
2130 		/* Apply these resource settings to any secondary macs */
2131 		if (umip != NULL) {
2132 			ASSERT((umip->mi_state_flags & MIS_IS_VNIC) != 0);
2133 			mac_vnic_secondary_update(umip);
2134 		}
2135 	}
2136 	kmem_free(omrp, sizeof (*omrp));
2137 	return (0);
2138 }
2139 
2140 static int
2141 mac_unicast_flow_create(mac_client_impl_t *mcip, uint8_t *mac_addr,
2142     uint16_t vid, boolean_t is_primary, boolean_t first_flow,
2143     flow_entry_t **flent, mac_resource_props_t *mrp)
2144 {
2145 	mac_impl_t	*mip = (mac_impl_t *)mcip->mci_mip;
2146 	flow_desc_t	flow_desc;
2147 	char		flowname[MAXFLOWNAMELEN];
2148 	int		err;
2149 	uint_t		flent_flags;
2150 
2151 	/*
2152 	 * First unicast address being added, create a new flow
2153 	 * for that MAC client.
2154 	 */
2155 	bzero(&flow_desc, sizeof (flow_desc));
2156 
2157 	ASSERT(mac_addr != NULL ||
2158 	    (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR));
2159 	if (mac_addr != NULL) {
2160 		flow_desc.fd_mac_len = mip->mi_type->mt_addr_length;
2161 		bcopy(mac_addr, flow_desc.fd_dst_mac, flow_desc.fd_mac_len);
2162 	}
2163 	flow_desc.fd_mask = FLOW_LINK_DST;
2164 	if (vid != 0) {
2165 		flow_desc.fd_vid = vid;
2166 		flow_desc.fd_mask |= FLOW_LINK_VID;
2167 	}
2168 
2169 	/*
2170 	 * XXX-nicolas. For now I'm keeping the FLOW_PRIMARY_MAC
2171 	 * and FLOW_VNIC. Even though they're a hack inherited
2172 	 * from the SRS code, we'll keep them for now. They're currently
2173 	 * consumed by mac_datapath_setup() to create the SRS.
2174 	 * That code should be eventually moved out of
2175 	 * mac_datapath_setup() and moved to a mac_srs_create()
2176 	 * function of some sort to keep things clean.
2177 	 *
2178 	 * Also, there's no reason why the SRS for the primary MAC
2179 	 * client should be different than any other MAC client. Until
2180 	 * this is cleaned-up, we support only one MAC unicast address
2181 	 * per client.
2182 	 *
2183 	 * We set FLOW_PRIMARY_MAC for the primary MAC address,
2184 	 * FLOW_VNIC for everything else.
2185 	 */
2186 	if (is_primary)
2187 		flent_flags = FLOW_PRIMARY_MAC;
2188 	else
2189 		flent_flags = FLOW_VNIC_MAC;
2190 
2191 	/*
2192 	 * For the first flow we use the MAC client's name - mci_name, for
2193 	 * subsequent ones we just create a name with the VID. This is
2194 	 * so that we can add these flows to the same flow table. This is
2195 	 * fine as the flow name (except for the one with the MAC client's
2196 	 * name) is not visible. When the first flow is removed, we just replace
2197 	 * its fdesc with another from the list, so we will still retain the
2198 	 * flent with the MAC client's flow name.
2199 	 */
2200 	if (first_flow) {
2201 		bcopy(mcip->mci_name, flowname, MAXFLOWNAMELEN);
2202 	} else {
2203 		(void) sprintf(flowname, "%s%u", mcip->mci_name, vid);
2204 		flent_flags = FLOW_NO_STATS;
2205 	}
2206 
2207 	if ((err = mac_flow_create(&flow_desc, mrp, flowname, NULL,
2208 	    flent_flags, flent)) != 0)
2209 		return (err);
2210 
2211 	mac_misc_stat_create(*flent);
2212 	FLOW_MARK(*flent, FE_INCIPIENT);
2213 	(*flent)->fe_mcip = mcip;
2214 
2215 	/*
2216 	 * Place initial creation reference on the flow. This reference
2217 	 * is released in the corresponding delete action viz.
2218 	 * mac_unicast_remove after waiting for all transient refs to
2219 	 * to go away. The wait happens in mac_flow_wait.
2220 	 * We have already held the reference in mac_client_open().
2221 	 */
2222 	if (!first_flow)
2223 		FLOW_REFHOLD(*flent);
2224 	return (0);
2225 }
2226 
2227 /* Refresh the multicast grouping for this VID. */
2228 int
2229 mac_client_update_mcast(void *arg, boolean_t add, const uint8_t *addrp)
2230 {
2231 	flow_entry_t		*flent = arg;
2232 	mac_client_impl_t	*mcip = flent->fe_mcip;
2233 	uint16_t		vid;
2234 	flow_desc_t		flow_desc;
2235 
2236 	mac_flow_get_desc(flent, &flow_desc);
2237 	vid = (flow_desc.fd_mask & FLOW_LINK_VID) != 0 ?
2238 	    flow_desc.fd_vid : VLAN_ID_NONE;
2239 
2240 	/*
2241 	 * We don't call mac_multicast_add()/mac_multicast_remove() as
2242 	 * we want to add/remove for this specific vid.
2243 	 */
2244 	if (add) {
2245 		return (mac_bcast_add(mcip, addrp, vid,
2246 		    MAC_ADDRTYPE_MULTICAST));
2247 	} else {
2248 		mac_bcast_delete(mcip, addrp, vid);
2249 		return (0);
2250 	}
2251 }
2252 
2253 static void
2254 mac_update_single_active_client(mac_impl_t *mip)
2255 {
2256 	mac_client_impl_t *client = NULL;
2257 
2258 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
2259 
2260 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
2261 	if (mip->mi_nactiveclients == 1) {
2262 		/*
2263 		 * Find the one active MAC client from the list of MAC
2264 		 * clients. The active MAC client has at least one
2265 		 * unicast address.
2266 		 */
2267 		for (client = mip->mi_clients_list; client != NULL;
2268 		    client = client->mci_client_next) {
2269 			if (client->mci_unicast_list != NULL)
2270 				break;
2271 		}
2272 		ASSERT(client != NULL);
2273 	}
2274 
2275 	/*
2276 	 * mi_single_active_client is protected by the MAC impl's read/writer
2277 	 * lock, which allows mac_rx() to check the value of that pointer
2278 	 * as a reader.
2279 	 */
2280 	mip->mi_single_active_client = client;
2281 	rw_exit(&mip->mi_rw_lock);
2282 }
2283 
2284 /*
2285  * Set up the data path. Called from i_mac_unicast_add after having
2286  * done all the validations including making sure this is an active
2287  * client (i.e that is ready to process packets.)
2288  */
2289 static int
2290 mac_client_datapath_setup(mac_client_impl_t *mcip, uint16_t vid,
2291     uint8_t *mac_addr, mac_resource_props_t *mrp, boolean_t isprimary,
2292     mac_unicast_impl_t *muip)
2293 {
2294 	mac_impl_t	*mip = mcip->mci_mip;
2295 	boolean_t	mac_started = B_FALSE;
2296 	boolean_t	bcast_added = B_FALSE;
2297 	boolean_t	nactiveclients_added = B_FALSE;
2298 	flow_entry_t	*flent;
2299 	int		err = 0;
2300 	boolean_t	no_unicast;
2301 
2302 	no_unicast = mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR;
2303 
2304 	if ((err = mac_start((mac_handle_t)mip)) != 0)
2305 		goto bail;
2306 
2307 	mac_started = B_TRUE;
2308 
2309 	/* add the MAC client to the broadcast address group by default */
2310 	if (mip->mi_type->mt_brdcst_addr != NULL) {
2311 		err = mac_bcast_add(mcip, mip->mi_type->mt_brdcst_addr, vid,
2312 		    MAC_ADDRTYPE_BROADCAST);
2313 		if (err != 0)
2314 			goto bail;
2315 		bcast_added = B_TRUE;
2316 	}
2317 
2318 	/*
2319 	 * If this is the first unicast address addition for this
2320 	 * client, reuse the pre-allocated larval flow entry associated with
2321 	 * the MAC client.
2322 	 */
2323 	flent = (mcip->mci_nflents == 0) ? mcip->mci_flent : NULL;
2324 
2325 	/* We are configuring the unicast flow now */
2326 	if (!MCIP_DATAPATH_SETUP(mcip)) {
2327 
2328 		if (mrp != NULL) {
2329 			MAC_CLIENT_SET_PRIORITY_RANGE(mcip,
2330 			    (mrp->mrp_mask & MRP_PRIORITY) ? mrp->mrp_priority :
2331 			    MPL_LINK_DEFAULT);
2332 		}
2333 		if ((err = mac_unicast_flow_create(mcip, mac_addr, vid,
2334 		    isprimary, B_TRUE, &flent, mrp)) != 0)
2335 			goto bail;
2336 
2337 		mip->mi_nactiveclients++;
2338 		nactiveclients_added = B_TRUE;
2339 
2340 		/*
2341 		 * This will allocate the RX ring group if possible for the
2342 		 * flow and program the software classifier as needed.
2343 		 */
2344 		if ((err = mac_datapath_setup(mcip, flent, SRST_LINK)) != 0)
2345 			goto bail;
2346 
2347 		if (no_unicast)
2348 			goto done_setup;
2349 		/*
2350 		 * The unicast MAC address must have been added successfully.
2351 		 */
2352 		ASSERT(mcip->mci_unicast != NULL);
2353 
2354 		/*
2355 		 * Push down the sub-flows that were defined on this link
2356 		 * hitherto. The flows are added to the active flow table
2357 		 * and SRS, softrings etc. are created as needed.
2358 		 */
2359 		mac_link_init_flows((mac_client_handle_t)mcip);
2360 	} else {
2361 		mac_address_t *map = mcip->mci_unicast;
2362 
2363 		ASSERT(!no_unicast);
2364 		/*
2365 		 * A unicast flow already exists for that MAC client
2366 		 * so this flow must be the same MAC address but with
2367 		 * a different VID. It has been checked by
2368 		 * mac_addr_in_use().
2369 		 *
2370 		 * We will use the SRS etc. from the initial
2371 		 * mci_flent. We don't need to create a kstat for
2372 		 * this, as except for the fdesc, everything will be
2373 		 * used from the first flent.
2374 		 *
2375 		 * The only time we should see multiple flents on the
2376 		 * same MAC client is on the sun4v vsw. If we removed
2377 		 * that code we should be able to remove the entire
2378 		 * notion of multiple flents on a MAC client (this
2379 		 * doesn't affect sub/user flows because they have
2380 		 * their own list unrelated to mci_flent_list).
2381 		 */
2382 		if (bcmp(mac_addr, map->ma_addr, map->ma_len) != 0) {
2383 			err = EINVAL;
2384 			goto bail;
2385 		}
2386 
2387 		if ((err = mac_unicast_flow_create(mcip, mac_addr, vid,
2388 		    isprimary, B_FALSE, &flent, NULL)) != 0) {
2389 			goto bail;
2390 		}
2391 		if ((err = mac_flow_add(mip->mi_flow_tab, flent)) != 0) {
2392 			FLOW_FINAL_REFRELE(flent);
2393 			goto bail;
2394 		}
2395 
2396 		/* update the multicast group for this vid */
2397 		mac_client_bcast_refresh(mcip, mac_client_update_mcast,
2398 		    (void *)flent, B_TRUE);
2399 
2400 	}
2401 
2402 	/* populate the shared MAC address */
2403 	muip->mui_map = mcip->mci_unicast;
2404 
2405 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
2406 	muip->mui_next = mcip->mci_unicast_list;
2407 	mcip->mci_unicast_list = muip;
2408 	rw_exit(&mcip->mci_rw_lock);
2409 
2410 done_setup:
2411 	/*
2412 	 * First add the flent to the flow list of this mcip. Then set
2413 	 * the mip's mi_single_active_client if needed. The Rx path assumes
2414 	 * that mip->mi_single_active_client will always have an associated
2415 	 * flent.
2416 	 */
2417 	mac_client_add_to_flow_list(mcip, flent);
2418 	if (nactiveclients_added)
2419 		mac_update_single_active_client(mip);
2420 	/*
2421 	 * Trigger a renegotiation of the capabilities when the number of
2422 	 * active clients changes from 1 to 2, since some of the capabilities
2423 	 * might have to be disabled. Also send a MAC_NOTE_LINK notification
2424 	 * to all the MAC clients whenever physical link is DOWN.
2425 	 */
2426 	if (mip->mi_nactiveclients == 2) {
2427 		mac_capab_update((mac_handle_t)mip);
2428 		mac_virtual_link_update(mip);
2429 	}
2430 	/*
2431 	 * Now that the setup is complete, clear the INCIPIENT flag.
2432 	 * The flag was set to avoid incoming packets seeing inconsistent
2433 	 * structures while the setup was in progress. Clear the mci_tx_flag
2434 	 * by calling mac_tx_client_block. It is possible that
2435 	 * mac_unicast_remove was called prior to this mac_unicast_add which
2436 	 * could have set the MCI_TX_QUIESCE flag.
2437 	 */
2438 	if (flent->fe_rx_ring_group != NULL)
2439 		mac_rx_group_unmark(flent->fe_rx_ring_group, MR_INCIPIENT);
2440 	FLOW_UNMARK(flent, FE_INCIPIENT);
2441 
2442 	/*
2443 	 * If this is an aggr port client, don't enable the flow's
2444 	 * datapath at this stage. Otherwise, bcast traffic could
2445 	 * arrive while the aggr port is in the process of
2446 	 * initializing. Instead, the flow's datapath is started later
2447 	 * when mac_client_set_flow_cb() is called.
2448 	 */
2449 	if ((mcip->mci_state_flags & MCIS_IS_AGGR_PORT) == 0)
2450 		FLOW_UNMARK(flent, FE_MC_NO_DATAPATH);
2451 
2452 	mac_tx_client_unblock(mcip);
2453 	return (0);
2454 bail:
2455 	if (bcast_added)
2456 		mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr, vid);
2457 
2458 	if (nactiveclients_added)
2459 		mip->mi_nactiveclients--;
2460 
2461 	if (mac_started)
2462 		mac_stop((mac_handle_t)mip);
2463 
2464 	return (err);
2465 }
2466 
2467 /*
2468  * Return the passive primary MAC client, if present. The passive client is
2469  * a stand-by client that has the same unicast address as another that is
2470  * currenly active. Once the active client goes away, the passive client
2471  * becomes active.
2472  */
2473 static mac_client_impl_t *
2474 mac_get_passive_primary_client(mac_impl_t *mip)
2475 {
2476 	mac_client_impl_t	*mcip;
2477 
2478 	for (mcip = mip->mi_clients_list; mcip != NULL;
2479 	    mcip = mcip->mci_client_next) {
2480 		if (mac_is_primary_client(mcip) &&
2481 		    (mcip->mci_flags & MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
2482 			return (mcip);
2483 		}
2484 	}
2485 	return (NULL);
2486 }
2487 
2488 /*
2489  * Add a new unicast address to the MAC client.
2490  *
2491  * The MAC address can be specified either by value, or the MAC client
2492  * can specify that it wants to use the primary MAC address of the
2493  * underlying MAC. See the introductory comments at the beginning
2494  * of this file for more more information on primary MAC addresses.
2495  *
2496  * Note also the tuple (MAC address, VID) must be unique
2497  * for the MAC clients defined on top of the same underlying MAC
2498  * instance, unless the MAC_UNICAST_NODUPCHECK is specified.
2499  *
2500  * In no case can a client use the PVID for the MAC, if the MAC has one set.
2501  */
2502 int
2503 i_mac_unicast_add(mac_client_handle_t mch, uint8_t *mac_addr, uint16_t flags,
2504     mac_unicast_handle_t *mah, uint16_t vid, mac_diag_t *diag)
2505 {
2506 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2507 	mac_impl_t		*mip = mcip->mci_mip;
2508 	int			err;
2509 	uint_t			mac_len = mip->mi_type->mt_addr_length;
2510 	boolean_t		check_dups = !(flags & MAC_UNICAST_NODUPCHECK);
2511 	boolean_t		fastpath_disabled = B_FALSE;
2512 	boolean_t		is_primary = (flags & MAC_UNICAST_PRIMARY);
2513 	boolean_t		is_unicast_hw = (flags & MAC_UNICAST_HW);
2514 	mac_resource_props_t	*mrp;
2515 	boolean_t		passive_client = B_FALSE;
2516 	mac_unicast_impl_t	*muip;
2517 	boolean_t		is_vnic_primary =
2518 	    (flags & MAC_UNICAST_VNIC_PRIMARY);
2519 
2520 	/*
2521 	 * When the VID is non-zero the underlying MAC cannot be a
2522 	 * VNIC. I.e., dladm create-vlan cannot take a VNIC as
2523 	 * argument, only the primary MAC client.
2524 	 */
2525 	ASSERT(!((mip->mi_state_flags & MIS_IS_VNIC) && (vid != VLAN_ID_NONE)));
2526 
2527 	/*
2528 	 * Can't unicast add if the client asked only for minimal datapath
2529 	 * setup.
2530 	 */
2531 	if (mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR)
2532 		return (ENOTSUP);
2533 
2534 	/*
2535 	 * Check for an attempted use of the current Port VLAN ID, if enabled.
2536 	 * No client may use it.
2537 	 */
2538 	if (mip->mi_pvid != VLAN_ID_NONE && vid == mip->mi_pvid)
2539 		return (EBUSY);
2540 
2541 	/*
2542 	 * Check whether it's the primary client and flag it.
2543 	 */
2544 	if (!(mcip->mci_state_flags & MCIS_IS_VNIC) && is_primary &&
2545 	    vid == VLAN_ID_NONE)
2546 		mcip->mci_flags |= MAC_CLIENT_FLAGS_PRIMARY;
2547 
2548 	/*
2549 	 * is_vnic_primary is true when we come here as a VLAN VNIC
2550 	 * which uses the primary MAC client's address but with a non-zero
2551 	 * VID. In this case the MAC address is not specified by an upper
2552 	 * MAC client.
2553 	 */
2554 	if ((mcip->mci_state_flags & MCIS_IS_VNIC) && is_primary &&
2555 	    !is_vnic_primary) {
2556 		/*
2557 		 * The address is being set by the upper MAC client
2558 		 * of a VNIC. The MAC address was already set by the
2559 		 * VNIC driver during VNIC creation.
2560 		 *
2561 		 * Note: a VNIC has only one MAC address. We return
2562 		 * the MAC unicast address handle of the lower MAC client
2563 		 * corresponding to the VNIC. We allocate a new entry
2564 		 * which is flagged appropriately, so that mac_unicast_remove()
2565 		 * doesn't attempt to free the original entry that
2566 		 * was allocated by the VNIC driver.
2567 		 */
2568 		ASSERT(mcip->mci_unicast != NULL);
2569 
2570 		/* Check for VLAN flags, if present */
2571 		if ((flags & MAC_UNICAST_TAG_DISABLE) != 0)
2572 			mcip->mci_state_flags |= MCIS_TAG_DISABLE;
2573 
2574 		if ((flags & MAC_UNICAST_STRIP_DISABLE) != 0)
2575 			mcip->mci_state_flags |= MCIS_STRIP_DISABLE;
2576 
2577 		if ((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) != 0)
2578 			mcip->mci_state_flags |= MCIS_DISABLE_TX_VID_CHECK;
2579 
2580 		/*
2581 		 * Ensure that the primary unicast address of the VNIC
2582 		 * is added only once unless we have the
2583 		 * MAC_CLIENT_FLAGS_MULTI_PRIMARY set (and this is not
2584 		 * a passive MAC client).
2585 		 */
2586 		if ((mcip->mci_flags & MAC_CLIENT_FLAGS_VNIC_PRIMARY) != 0) {
2587 			if ((mcip->mci_flags &
2588 			    MAC_CLIENT_FLAGS_MULTI_PRIMARY) == 0 ||
2589 			    (mcip->mci_flags &
2590 			    MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
2591 				return (EBUSY);
2592 			}
2593 			mcip->mci_flags |= MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
2594 			passive_client = B_TRUE;
2595 		}
2596 
2597 		mcip->mci_flags |= MAC_CLIENT_FLAGS_VNIC_PRIMARY;
2598 
2599 		/*
2600 		 * Create a handle for vid 0.
2601 		 */
2602 		ASSERT(vid == VLAN_ID_NONE);
2603 		muip = kmem_zalloc(sizeof (mac_unicast_impl_t), KM_SLEEP);
2604 		muip->mui_vid = vid;
2605 		*mah = (mac_unicast_handle_t)muip;
2606 		/*
2607 		 * This will be used by the caller to defer setting the
2608 		 * rx functions.
2609 		 */
2610 		if (passive_client)
2611 			return (EAGAIN);
2612 		return (0);
2613 	}
2614 
2615 	/* primary MAC clients cannot be opened on top of anchor VNICs */
2616 	if ((is_vnic_primary || is_primary) &&
2617 	    i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_ANCHOR_VNIC, NULL)) {
2618 		return (ENXIO);
2619 	}
2620 
2621 	/*
2622 	 * If this is a VNIC/VLAN, disable softmac fast-path. This is
2623 	 * only relevant to legacy devices which use softmac to
2624 	 * interface with GLDv3.
2625 	 */
2626 	if (mcip->mci_state_flags & MCIS_IS_VNIC) {
2627 		err = mac_fastpath_disable((mac_handle_t)mip);
2628 		if (err != 0)
2629 			return (err);
2630 		fastpath_disabled = B_TRUE;
2631 	}
2632 
2633 	/*
2634 	 * Return EBUSY if:
2635 	 *  - there is an exclusively active mac client exists.
2636 	 *  - this is an exclusive active mac client but
2637 	 *	a. there is already active mac clients exist, or
2638 	 *	b. fastpath streams are already plumbed on this legacy device
2639 	 *  - the mac creator has disallowed active mac clients.
2640 	 */
2641 	if (mip->mi_state_flags & (MIS_EXCLUSIVE|MIS_NO_ACTIVE)) {
2642 		if (fastpath_disabled)
2643 			mac_fastpath_enable((mac_handle_t)mip);
2644 		return (EBUSY);
2645 	}
2646 
2647 	if (mcip->mci_state_flags & MCIS_EXCLUSIVE) {
2648 		ASSERT(!fastpath_disabled);
2649 		if (mip->mi_nactiveclients != 0)
2650 			return (EBUSY);
2651 
2652 		if ((mip->mi_state_flags & MIS_LEGACY) &&
2653 		    !(mip->mi_capab_legacy.ml_active_set(mip->mi_driver))) {
2654 			return (EBUSY);
2655 		}
2656 		mip->mi_state_flags |= MIS_EXCLUSIVE;
2657 	}
2658 
2659 	mrp = kmem_zalloc(sizeof (*mrp), KM_SLEEP);
2660 	if (is_primary && !(mcip->mci_state_flags & (MCIS_IS_VNIC |
2661 	    MCIS_IS_AGGR_PORT))) {
2662 		/*
2663 		 * Apply the property cached in the mac_impl_t to the primary
2664 		 * mac client. If the mac client is a VNIC or an aggregation
2665 		 * port, its property should be set in the mcip when the
2666 		 * VNIC/aggr was created.
2667 		 */
2668 		mac_get_resources((mac_handle_t)mip, mrp);
2669 		(void) mac_client_set_resources(mch, mrp);
2670 	} else if (mcip->mci_state_flags & MCIS_IS_VNIC) {
2671 		/*
2672 		 * This is a VLAN client sharing the address of the
2673 		 * primary MAC client; i.e., one created via dladm
2674 		 * create-vlan. We don't support specifying ring
2675 		 * properties for this type of client as it inherits
2676 		 * these from the primary MAC client.
2677 		 */
2678 		if (is_vnic_primary) {
2679 			mac_resource_props_t	*vmrp;
2680 
2681 			vmrp = MCIP_RESOURCE_PROPS(mcip);
2682 			if (vmrp->mrp_mask & MRP_RX_RINGS ||
2683 			    vmrp->mrp_mask & MRP_TX_RINGS) {
2684 				if (fastpath_disabled)
2685 					mac_fastpath_enable((mac_handle_t)mip);
2686 				kmem_free(mrp, sizeof (*mrp));
2687 				return (ENOTSUP);
2688 			}
2689 			/*
2690 			 * Additionally we also need to inherit any
2691 			 * rings property from the MAC.
2692 			 */
2693 			mac_get_resources((mac_handle_t)mip, mrp);
2694 			if (mrp->mrp_mask & MRP_RX_RINGS) {
2695 				vmrp->mrp_mask |= MRP_RX_RINGS;
2696 				vmrp->mrp_nrxrings = mrp->mrp_nrxrings;
2697 			}
2698 			if (mrp->mrp_mask & MRP_TX_RINGS) {
2699 				vmrp->mrp_mask |= MRP_TX_RINGS;
2700 				vmrp->mrp_ntxrings = mrp->mrp_ntxrings;
2701 			}
2702 		}
2703 		bcopy(MCIP_RESOURCE_PROPS(mcip), mrp, sizeof (*mrp));
2704 	}
2705 
2706 	muip = kmem_zalloc(sizeof (mac_unicast_impl_t), KM_SLEEP);
2707 	muip->mui_vid = vid;
2708 
2709 	if (is_primary || is_vnic_primary) {
2710 		mac_addr = mip->mi_addr;
2711 	} else {
2712 
2713 		/*
2714 		 * Verify the validity of the specified MAC addresses value.
2715 		 */
2716 		if (!mac_unicst_verify((mac_handle_t)mip, mac_addr, mac_len)) {
2717 			*diag = MAC_DIAG_MACADDR_INVALID;
2718 			err = EINVAL;
2719 			goto bail_out;
2720 		}
2721 
2722 		/*
2723 		 * Make sure that the specified MAC address is different
2724 		 * than the unicast MAC address of the underlying NIC.
2725 		 */
2726 		if (check_dups && bcmp(mip->mi_addr, mac_addr, mac_len) == 0) {
2727 			*diag = MAC_DIAG_MACADDR_NIC;
2728 			err = EINVAL;
2729 			goto bail_out;
2730 		}
2731 	}
2732 
2733 	/*
2734 	 * Set the flags here so that if this is a passive client, we
2735 	 * can return and set it when we call mac_client_datapath_setup
2736 	 * when this becomes the active client. If we defer to using these
2737 	 * flags to mac_client_datapath_setup, then for a passive client,
2738 	 * we'd have to store the flags somewhere (probably fe_flags)
2739 	 * and then use it.
2740 	 */
2741 	if (!MCIP_DATAPATH_SETUP(mcip)) {
2742 		if (is_unicast_hw) {
2743 			/*
2744 			 * The client requires a hardware MAC address slot
2745 			 * for that unicast address. Since we support only
2746 			 * one unicast MAC address per client, flag the
2747 			 * MAC client itself.
2748 			 */
2749 			mcip->mci_state_flags |= MCIS_UNICAST_HW;
2750 		}
2751 
2752 		/* Check for VLAN flags, if present */
2753 		if ((flags & MAC_UNICAST_TAG_DISABLE) != 0)
2754 			mcip->mci_state_flags |= MCIS_TAG_DISABLE;
2755 
2756 		if ((flags & MAC_UNICAST_STRIP_DISABLE) != 0)
2757 			mcip->mci_state_flags |= MCIS_STRIP_DISABLE;
2758 
2759 		if ((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) != 0)
2760 			mcip->mci_state_flags |= MCIS_DISABLE_TX_VID_CHECK;
2761 	} else {
2762 		/*
2763 		 * Assert that the specified flags are consistent with the
2764 		 * flags specified by previous calls to mac_unicast_add().
2765 		 */
2766 		ASSERT(((flags & MAC_UNICAST_TAG_DISABLE) != 0 &&
2767 		    (mcip->mci_state_flags & MCIS_TAG_DISABLE) != 0) ||
2768 		    ((flags & MAC_UNICAST_TAG_DISABLE) == 0 &&
2769 		    (mcip->mci_state_flags & MCIS_TAG_DISABLE) == 0));
2770 
2771 		ASSERT(((flags & MAC_UNICAST_STRIP_DISABLE) != 0 &&
2772 		    (mcip->mci_state_flags & MCIS_STRIP_DISABLE) != 0) ||
2773 		    ((flags & MAC_UNICAST_STRIP_DISABLE) == 0 &&
2774 		    (mcip->mci_state_flags & MCIS_STRIP_DISABLE) == 0));
2775 
2776 		ASSERT(((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) != 0 &&
2777 		    (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK) != 0) ||
2778 		    ((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) == 0 &&
2779 		    (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK) == 0));
2780 
2781 		/*
2782 		 * Make sure the client is consistent about its requests
2783 		 * for MAC addresses. I.e. all requests from the clients
2784 		 * must have the MAC_UNICAST_HW flag set or clear.
2785 		 */
2786 		if (((mcip->mci_state_flags & MCIS_UNICAST_HW) != 0 &&
2787 		    !is_unicast_hw) ||
2788 		    ((mcip->mci_state_flags & MCIS_UNICAST_HW) == 0 &&
2789 		    is_unicast_hw)) {
2790 			err = EINVAL;
2791 			goto bail_out;
2792 		}
2793 	}
2794 	/*
2795 	 * Make sure the MAC address is not already used by
2796 	 * another MAC client defined on top of the same
2797 	 * underlying NIC. Unless we have MAC_CLIENT_FLAGS_MULTI_PRIMARY
2798 	 * set when we allow a passive client to be present which will
2799 	 * be activated when the currently active client goes away - this
2800 	 * works only with primary addresses.
2801 	 */
2802 	if ((check_dups || is_primary || is_vnic_primary) &&
2803 	    mac_addr_in_use(mip, mac_addr, vid)) {
2804 		/*
2805 		 * Must have set the multiple primary address flag when
2806 		 * we did a mac_client_open AND this should be a primary
2807 		 * MAC client AND there should not already be a passive
2808 		 * primary. If all is true then we let this succeed
2809 		 * even if the address is a dup.
2810 		 */
2811 		if ((mcip->mci_flags & MAC_CLIENT_FLAGS_MULTI_PRIMARY) == 0 ||
2812 		    (mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY) == 0 ||
2813 		    mac_get_passive_primary_client(mip) != NULL) {
2814 			*diag = MAC_DIAG_MACADDR_INUSE;
2815 			err = EEXIST;
2816 			goto bail_out;
2817 		}
2818 		ASSERT((mcip->mci_flags &
2819 		    MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) == 0);
2820 		mcip->mci_flags |= MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
2821 		kmem_free(mrp, sizeof (*mrp));
2822 
2823 		/*
2824 		 * Stash the unicast address handle, we will use it when
2825 		 * we set up the passive client.
2826 		 */
2827 		mcip->mci_p_unicast_list = muip;
2828 		*mah = (mac_unicast_handle_t)muip;
2829 		return (0);
2830 	}
2831 
2832 	err = mac_client_datapath_setup(mcip, vid, mac_addr, mrp,
2833 	    is_primary || is_vnic_primary, muip);
2834 	if (err != 0)
2835 		goto bail_out;
2836 
2837 	kmem_free(mrp, sizeof (*mrp));
2838 	*mah = (mac_unicast_handle_t)muip;
2839 	return (0);
2840 
2841 bail_out:
2842 	if (fastpath_disabled)
2843 		mac_fastpath_enable((mac_handle_t)mip);
2844 	if (mcip->mci_state_flags & MCIS_EXCLUSIVE) {
2845 		mip->mi_state_flags &= ~MIS_EXCLUSIVE;
2846 		if (mip->mi_state_flags & MIS_LEGACY) {
2847 			mip->mi_capab_legacy.ml_active_clear(
2848 			    mip->mi_driver);
2849 		}
2850 	}
2851 	kmem_free(mrp, sizeof (*mrp));
2852 	kmem_free(muip, sizeof (mac_unicast_impl_t));
2853 	return (err);
2854 }
2855 
2856 /*
2857  * Wrapper function to mac_unicast_add when we want to have the same mac
2858  * client open for two instances, one that is currently active and another
2859  * that will become active when the current one is removed. In this case
2860  * mac_unicast_add will return EGAIN and we will save the rx function and
2861  * arg which will be used when we activate the passive client in
2862  * mac_unicast_remove.
2863  */
2864 int
2865 mac_unicast_add_set_rx(mac_client_handle_t mch, uint8_t *mac_addr,
2866     uint16_t flags, mac_unicast_handle_t *mah,  uint16_t vid, mac_diag_t *diag,
2867     mac_rx_t rx_fn, void *arg)
2868 {
2869 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2870 	uint_t			err;
2871 
2872 	err = mac_unicast_add(mch, mac_addr, flags, mah, vid, diag);
2873 	if (err != 0 && err != EAGAIN)
2874 		return (err);
2875 	if (err == EAGAIN) {
2876 		if (rx_fn != NULL) {
2877 			mcip->mci_rx_p_fn = rx_fn;
2878 			mcip->mci_rx_p_arg = arg;
2879 		}
2880 		return (0);
2881 	}
2882 	if (rx_fn != NULL)
2883 		mac_rx_set(mch, rx_fn, arg);
2884 	return (err);
2885 }
2886 
2887 int
2888 mac_unicast_add(mac_client_handle_t mch, uint8_t *mac_addr, uint16_t flags,
2889     mac_unicast_handle_t *mah, uint16_t vid, mac_diag_t *diag)
2890 {
2891 	mac_impl_t *mip = ((mac_client_impl_t *)mch)->mci_mip;
2892 	uint_t err;
2893 
2894 	i_mac_perim_enter(mip);
2895 	err = i_mac_unicast_add(mch, mac_addr, flags, mah, vid, diag);
2896 	i_mac_perim_exit(mip);
2897 
2898 	return (err);
2899 }
2900 
2901 static void
2902 mac_client_datapath_teardown(mac_client_handle_t mch, mac_unicast_impl_t *muip,
2903     flow_entry_t *flent)
2904 {
2905 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2906 	mac_impl_t		*mip = mcip->mci_mip;
2907 	boolean_t		no_unicast;
2908 
2909 	/*
2910 	 * If we have not added a unicast address for this MAC client, just
2911 	 * teardown the datapath.
2912 	 */
2913 	no_unicast = mcip->mci_state_flags & MCIS_NO_UNICAST_ADDR;
2914 
2915 	if (!no_unicast) {
2916 		/*
2917 		 * We would have initialized subflows etc. only if we brought
2918 		 * up the primary client and set the unicast unicast address
2919 		 * etc. Deactivate the flows. The flow entry will be removed
2920 		 * from the active flow tables, and the associated SRS,
2921 		 * softrings etc will be deleted. But the flow entry itself
2922 		 * won't be destroyed, instead it will continue to be archived
2923 		 * off the  the global flow hash list, for a possible future
2924 		 * activation when say IP is plumbed again.
2925 		 */
2926 		mac_link_release_flows(mch);
2927 	}
2928 	mip->mi_nactiveclients--;
2929 	mac_update_single_active_client(mip);
2930 
2931 	/* Tear down the data path */
2932 	mac_datapath_teardown(mcip, mcip->mci_flent, SRST_LINK);
2933 
2934 	/*
2935 	 * Prevent any future access to the flow entry through the mci_flent
2936 	 * pointer by setting the mci_flent to NULL. Access to mci_flent in
2937 	 * mac_bcast_send is also under mi_rw_lock.
2938 	 */
2939 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
2940 	flent = mcip->mci_flent;
2941 	mac_client_remove_flow_from_list(mcip, flent);
2942 
2943 	if (mcip->mci_state_flags & MCIS_DESC_LOGGED)
2944 		mcip->mci_state_flags &= ~MCIS_DESC_LOGGED;
2945 
2946 	/*
2947 	 * This is the last unicast address being removed and there shouldn't
2948 	 * be any outbound data threads at this point coming down from mac
2949 	 * clients. We have waited for the data threads to finish before
2950 	 * starting dld_str_detach. Non-data threads must access TX SRS
2951 	 * under mi_rw_lock.
2952 	 */
2953 	rw_exit(&mip->mi_rw_lock);
2954 
2955 	/*
2956 	 * Don't use FLOW_MARK with FE_MC_NO_DATAPATH, as the flow might
2957 	 * contain other flags, such as FE_CONDEMNED, which we need to
2958 	 * cleared. We don't call mac_flow_cleanup() for this unicast
2959 	 * flow as we have a already cleaned up SRSs etc. (via the teadown
2960 	 * path). We just clear the stats and reset the initial callback
2961 	 * function, the rest will be set when we call mac_flow_create,
2962 	 * if at all.
2963 	 */
2964 	mutex_enter(&flent->fe_lock);
2965 	ASSERT(flent->fe_refcnt == 1 && flent->fe_mbg == NULL &&
2966 	    flent->fe_tx_srs == NULL && flent->fe_rx_srs_cnt == 0);
2967 	flent->fe_flags = FE_MC_NO_DATAPATH;
2968 	flow_stat_destroy(flent);
2969 	mac_misc_stat_delete(flent);
2970 
2971 	/* Initialize the receiver function to a safe routine */
2972 	flent->fe_cb_fn = (flow_fn_t)mac_pkt_drop;
2973 	flent->fe_cb_arg1 = NULL;
2974 	flent->fe_cb_arg2 = NULL;
2975 
2976 	flent->fe_index = -1;
2977 	mutex_exit(&flent->fe_lock);
2978 
2979 	if (mip->mi_type->mt_brdcst_addr != NULL) {
2980 		ASSERT(muip != NULL || no_unicast);
2981 		mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr,
2982 		    muip != NULL ? muip->mui_vid : VLAN_ID_NONE);
2983 	}
2984 
2985 	if (mip->mi_nactiveclients == 1) {
2986 		mac_capab_update((mac_handle_t)mip);
2987 		mac_virtual_link_update(mip);
2988 	}
2989 
2990 	if (mcip->mci_state_flags & MCIS_EXCLUSIVE) {
2991 		mip->mi_state_flags &= ~MIS_EXCLUSIVE;
2992 
2993 		if (mip->mi_state_flags & MIS_LEGACY)
2994 			mip->mi_capab_legacy.ml_active_clear(mip->mi_driver);
2995 	}
2996 
2997 	mcip->mci_state_flags &= ~MCIS_UNICAST_HW;
2998 
2999 	if (mcip->mci_state_flags & MCIS_TAG_DISABLE)
3000 		mcip->mci_state_flags &= ~MCIS_TAG_DISABLE;
3001 
3002 	if (mcip->mci_state_flags & MCIS_STRIP_DISABLE)
3003 		mcip->mci_state_flags &= ~MCIS_STRIP_DISABLE;
3004 
3005 	if (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK)
3006 		mcip->mci_state_flags &= ~MCIS_DISABLE_TX_VID_CHECK;
3007 
3008 	if (muip != NULL)
3009 		kmem_free(muip, sizeof (mac_unicast_impl_t));
3010 	mac_protect_cancel_timer(mcip);
3011 	mac_protect_flush_dynamic(mcip);
3012 
3013 	bzero(&mcip->mci_misc_stat, sizeof (mcip->mci_misc_stat));
3014 	/*
3015 	 * Disable fastpath if this is a VNIC or a VLAN.
3016 	 */
3017 	if (mcip->mci_state_flags & MCIS_IS_VNIC)
3018 		mac_fastpath_enable((mac_handle_t)mip);
3019 	mac_stop((mac_handle_t)mip);
3020 }
3021 
3022 /*
3023  * Remove a MAC address which was previously added by mac_unicast_add().
3024  */
3025 int
3026 mac_unicast_remove(mac_client_handle_t mch, mac_unicast_handle_t mah)
3027 {
3028 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3029 	mac_unicast_impl_t *muip = (mac_unicast_impl_t *)mah;
3030 	mac_unicast_impl_t *pre;
3031 	mac_impl_t *mip = mcip->mci_mip;
3032 	flow_entry_t		*flent;
3033 	uint16_t mui_vid;
3034 
3035 	i_mac_perim_enter(mip);
3036 	if (mcip->mci_flags & MAC_CLIENT_FLAGS_VNIC_PRIMARY) {
3037 		/*
3038 		 * Call made by the upper MAC client of a VNIC.
3039 		 * There's nothing much to do, the unicast address will
3040 		 * be removed by the VNIC driver when the VNIC is deleted,
3041 		 * but let's ensure that all our transmit is done before
3042 		 * the client does a mac_client_stop lest it trigger an
3043 		 * assert in the driver.
3044 		 */
3045 		ASSERT(muip->mui_vid == VLAN_ID_NONE);
3046 
3047 		mac_tx_client_flush(mcip);
3048 
3049 		if ((mcip->mci_flags & MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
3050 			mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
3051 			if (mcip->mci_rx_p_fn != NULL) {
3052 				mac_rx_set(mch, mcip->mci_rx_p_fn,
3053 				    mcip->mci_rx_p_arg);
3054 				mcip->mci_rx_p_fn = NULL;
3055 				mcip->mci_rx_p_arg = NULL;
3056 			}
3057 			kmem_free(muip, sizeof (mac_unicast_impl_t));
3058 			i_mac_perim_exit(mip);
3059 			return (0);
3060 		}
3061 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_VNIC_PRIMARY;
3062 
3063 		if (mcip->mci_state_flags & MCIS_TAG_DISABLE)
3064 			mcip->mci_state_flags &= ~MCIS_TAG_DISABLE;
3065 
3066 		if (mcip->mci_state_flags & MCIS_STRIP_DISABLE)
3067 			mcip->mci_state_flags &= ~MCIS_STRIP_DISABLE;
3068 
3069 		if (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK)
3070 			mcip->mci_state_flags &= ~MCIS_DISABLE_TX_VID_CHECK;
3071 
3072 		kmem_free(muip, sizeof (mac_unicast_impl_t));
3073 		i_mac_perim_exit(mip);
3074 		return (0);
3075 	}
3076 
3077 	ASSERT(muip != NULL);
3078 
3079 	/*
3080 	 * We are removing a passive client, we haven't setup the datapath
3081 	 * for this yet, so nothing much to do.
3082 	 */
3083 	if ((mcip->mci_flags & MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
3084 
3085 		ASSERT((mcip->mci_flent->fe_flags & FE_MC_NO_DATAPATH) != 0);
3086 		ASSERT(mcip->mci_p_unicast_list == muip);
3087 
3088 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
3089 
3090 		mcip->mci_p_unicast_list = NULL;
3091 		mcip->mci_rx_p_fn = NULL;
3092 		mcip->mci_rx_p_arg = NULL;
3093 
3094 		mcip->mci_state_flags &= ~MCIS_UNICAST_HW;
3095 
3096 		if (mcip->mci_state_flags & MCIS_TAG_DISABLE)
3097 			mcip->mci_state_flags &= ~MCIS_TAG_DISABLE;
3098 
3099 		if (mcip->mci_state_flags & MCIS_STRIP_DISABLE)
3100 			mcip->mci_state_flags &= ~MCIS_STRIP_DISABLE;
3101 
3102 		if (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK)
3103 			mcip->mci_state_flags &= ~MCIS_DISABLE_TX_VID_CHECK;
3104 
3105 		kmem_free(muip, sizeof (mac_unicast_impl_t));
3106 		i_mac_perim_exit(mip);
3107 		return (0);
3108 	}
3109 
3110 	/*
3111 	 * Remove the VID from the list of client's VIDs.
3112 	 */
3113 	pre = mcip->mci_unicast_list;
3114 	if (muip == pre) {
3115 		mcip->mci_unicast_list = muip->mui_next;
3116 	} else {
3117 		while ((pre->mui_next != NULL) && (pre->mui_next != muip))
3118 			pre = pre->mui_next;
3119 		ASSERT(pre->mui_next == muip);
3120 		rw_enter(&mcip->mci_rw_lock, RW_WRITER);
3121 		pre->mui_next = muip->mui_next;
3122 		rw_exit(&mcip->mci_rw_lock);
3123 	}
3124 
3125 	if (!mac_client_single_rcvr(mcip)) {
3126 		/*
3127 		 * This MAC client is shared by more than one unicast
3128 		 * addresses, so we will just remove the flent
3129 		 * corresponding to the address being removed. We don't invoke
3130 		 * mac_rx_classify_flow_rem() since the additional flow is
3131 		 * not associated with its own separate set of SRS and rings,
3132 		 * and these constructs are still needed for the remaining
3133 		 * flows.
3134 		 */
3135 		flent = mac_client_get_flow(mcip, muip);
3136 		VERIFY3P(flent, !=, NULL);
3137 
3138 		/*
3139 		 * The first one is disappearing, need to make sure
3140 		 * we replace it with another from the list of
3141 		 * shared clients.
3142 		 */
3143 		if (flent == mcip->mci_flent)
3144 			flent = mac_client_swap_mciflent(mcip);
3145 		mac_client_remove_flow_from_list(mcip, flent);
3146 		mac_flow_remove(mip->mi_flow_tab, flent, B_FALSE);
3147 		mac_flow_wait(flent, FLOW_DRIVER_UPCALL);
3148 
3149 		/*
3150 		 * The multicast groups that were added by the client so
3151 		 * far must be removed from the brodcast domain corresponding
3152 		 * to the VID being removed.
3153 		 */
3154 		mac_client_bcast_refresh(mcip, mac_client_update_mcast,
3155 		    (void *)flent, B_FALSE);
3156 
3157 		if (mip->mi_type->mt_brdcst_addr != NULL) {
3158 			mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr,
3159 			    muip->mui_vid);
3160 		}
3161 
3162 		FLOW_FINAL_REFRELE(flent);
3163 		ASSERT(!(mcip->mci_state_flags & MCIS_EXCLUSIVE));
3164 
3165 		/*
3166 		 * Enable fastpath if this is a VNIC or a VLAN.
3167 		 */
3168 		if (mcip->mci_state_flags & MCIS_IS_VNIC)
3169 			mac_fastpath_enable((mac_handle_t)mip);
3170 		mac_stop((mac_handle_t)mip);
3171 		i_mac_perim_exit(mip);
3172 		return (0);
3173 	}
3174 
3175 	mui_vid = muip->mui_vid;
3176 	mac_client_datapath_teardown(mch, muip, flent);
3177 
3178 	if ((mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY) &&
3179 	    mui_vid == VLAN_ID_NONE) {
3180 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PRIMARY;
3181 	} else {
3182 		i_mac_perim_exit(mip);
3183 		return (0);
3184 	}
3185 
3186 	/*
3187 	 * If we are removing the primary, check if we have a passive primary
3188 	 * client that we need to activate now.
3189 	 */
3190 	mcip = mac_get_passive_primary_client(mip);
3191 	if (mcip != NULL) {
3192 		mac_resource_props_t	*mrp;
3193 		mac_unicast_impl_t	*muip;
3194 
3195 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
3196 		mrp = kmem_zalloc(sizeof (*mrp), KM_SLEEP);
3197 
3198 		/*
3199 		 * Apply the property cached in the mac_impl_t to the
3200 		 * primary mac client.
3201 		 */
3202 		mac_get_resources((mac_handle_t)mip, mrp);
3203 		(void) mac_client_set_resources(mch, mrp);
3204 		ASSERT(mcip->mci_p_unicast_list != NULL);
3205 		muip = mcip->mci_p_unicast_list;
3206 		mcip->mci_p_unicast_list = NULL;
3207 		if (mac_client_datapath_setup(mcip, VLAN_ID_NONE,
3208 		    mip->mi_addr, mrp, B_TRUE, muip) == 0) {
3209 			if (mcip->mci_rx_p_fn != NULL) {
3210 				mac_rx_set(mch, mcip->mci_rx_p_fn,
3211 				    mcip->mci_rx_p_arg);
3212 				mcip->mci_rx_p_fn = NULL;
3213 				mcip->mci_rx_p_arg = NULL;
3214 			}
3215 		} else {
3216 			kmem_free(muip, sizeof (mac_unicast_impl_t));
3217 		}
3218 		kmem_free(mrp, sizeof (*mrp));
3219 	}
3220 	i_mac_perim_exit(mip);
3221 	return (0);
3222 }
3223 
3224 /*
3225  * Multicast add function invoked by MAC clients.
3226  */
3227 int
3228 mac_multicast_add(mac_client_handle_t mch, const uint8_t *addr)
3229 {
3230 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3231 	mac_impl_t		*mip = mcip->mci_mip;
3232 	flow_entry_t		*flent = mcip->mci_flent_list;
3233 	flow_entry_t		*prev_fe = NULL;
3234 	uint16_t		vid;
3235 	int			err = 0;
3236 
3237 	/* Verify the address is a valid multicast address */
3238 	if ((err = mip->mi_type->mt_ops.mtops_multicst_verify(addr,
3239 	    mip->mi_pdata)) != 0)
3240 		return (err);
3241 
3242 	i_mac_perim_enter(mip);
3243 	while (flent != NULL) {
3244 		vid = i_mac_flow_vid(flent);
3245 
3246 		err = mac_bcast_add((mac_client_impl_t *)mch, addr, vid,
3247 		    MAC_ADDRTYPE_MULTICAST);
3248 		if (err != 0)
3249 			break;
3250 		prev_fe = flent;
3251 		flent = flent->fe_client_next;
3252 	}
3253 
3254 	/*
3255 	 * If we failed adding, then undo all, rather than partial
3256 	 * success.
3257 	 */
3258 	if (flent != NULL && prev_fe != NULL) {
3259 		flent = mcip->mci_flent_list;
3260 		while (flent != prev_fe->fe_client_next) {
3261 			vid = i_mac_flow_vid(flent);
3262 			mac_bcast_delete((mac_client_impl_t *)mch, addr, vid);
3263 			flent = flent->fe_client_next;
3264 		}
3265 	}
3266 	i_mac_perim_exit(mip);
3267 	return (err);
3268 }
3269 
3270 /*
3271  * Multicast delete function invoked by MAC clients.
3272  */
3273 void
3274 mac_multicast_remove(mac_client_handle_t mch, const uint8_t *addr)
3275 {
3276 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3277 	mac_impl_t		*mip = mcip->mci_mip;
3278 	flow_entry_t		*flent;
3279 	uint16_t		vid;
3280 
3281 	i_mac_perim_enter(mip);
3282 	for (flent = mcip->mci_flent_list; flent != NULL;
3283 	    flent = flent->fe_client_next) {
3284 		vid = i_mac_flow_vid(flent);
3285 		mac_bcast_delete((mac_client_impl_t *)mch, addr, vid);
3286 	}
3287 	i_mac_perim_exit(mip);
3288 }
3289 
3290 /*
3291  * When a MAC client desires to capture packets on an interface,
3292  * it registers a promiscuous call back with mac_promisc_add().
3293  * There are three types of promiscuous callbacks:
3294  *
3295  * * MAC_CLIENT_PROMISC_ALL
3296  *   Captures all packets sent and received by the MAC client,
3297  *   the physical interface, as well as all other MAC clients
3298  *   defined on top of the same MAC.
3299  *
3300  * * MAC_CLIENT_PROMISC_FILTERED
3301  *   Captures all packets sent and received by the MAC client,
3302  *   plus all multicast traffic sent and received by the phyisical
3303  *   interface and the other MAC clients.
3304  *
3305  * * MAC_CLIENT_PROMISC_MULTI
3306  *   Captures all broadcast and multicast packets sent and
3307  *   received by the MAC clients as well as the physical interface.
3308  *
3309  * In all cases, the underlying MAC is put in promiscuous mode.
3310  */
3311 int
3312 mac_promisc_add(mac_client_handle_t mch, mac_client_promisc_type_t type,
3313     mac_rx_t fn, void *arg, mac_promisc_handle_t *mphp, uint16_t flags)
3314 {
3315 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3316 	mac_impl_t *mip = mcip->mci_mip;
3317 	mac_promisc_impl_t *mpip;
3318 	mac_cb_info_t	*mcbi;
3319 	int rc;
3320 
3321 	i_mac_perim_enter(mip);
3322 
3323 	if ((rc = mac_start((mac_handle_t)mip)) != 0) {
3324 		i_mac_perim_exit(mip);
3325 		return (rc);
3326 	}
3327 
3328 	if ((mcip->mci_state_flags & MCIS_IS_VNIC) &&
3329 	    type == MAC_CLIENT_PROMISC_ALL &&
3330 	    (mcip->mci_protect_flags & MPT_FLAG_PROMISC_FILTERED)) {
3331 		/*
3332 		 * The function is being invoked by the upper MAC client
3333 		 * of a VNIC. The VNIC should only see the traffic
3334 		 * it is entitled to.
3335 		 */
3336 		type = MAC_CLIENT_PROMISC_FILTERED;
3337 	}
3338 
3339 
3340 	/*
3341 	 * Turn on promiscuous mode for the underlying NIC.
3342 	 * This is needed even for filtered callbacks which
3343 	 * expect to receive all multicast traffic on the wire.
3344 	 *
3345 	 * Physical promiscuous mode should not be turned on if
3346 	 * MAC_PROMISC_FLAGS_NO_PHYS is set.
3347 	 */
3348 	if ((flags & MAC_PROMISC_FLAGS_NO_PHYS) == 0) {
3349 		if ((rc = i_mac_promisc_set(mip, B_TRUE)) != 0) {
3350 			mac_stop((mac_handle_t)mip);
3351 			i_mac_perim_exit(mip);
3352 			return (rc);
3353 		}
3354 	}
3355 
3356 	mpip = kmem_cache_alloc(mac_promisc_impl_cache, KM_SLEEP);
3357 
3358 	mpip->mpi_type = type;
3359 	mpip->mpi_fn = fn;
3360 	mpip->mpi_arg = arg;
3361 	mpip->mpi_mcip = mcip;
3362 	mpip->mpi_no_tx_loop = ((flags & MAC_PROMISC_FLAGS_NO_TX_LOOP) != 0);
3363 	mpip->mpi_no_phys = ((flags & MAC_PROMISC_FLAGS_NO_PHYS) != 0);
3364 	mpip->mpi_strip_vlan_tag =
3365 	    ((flags & MAC_PROMISC_FLAGS_VLAN_TAG_STRIP) != 0);
3366 	mpip->mpi_no_copy = ((flags & MAC_PROMISC_FLAGS_NO_COPY) != 0);
3367 
3368 	mcbi = &mip->mi_promisc_cb_info;
3369 	mutex_enter(mcbi->mcbi_lockp);
3370 
3371 	mac_callback_add(&mip->mi_promisc_cb_info, &mcip->mci_promisc_list,
3372 	    &mpip->mpi_mci_link);
3373 	mac_callback_add(&mip->mi_promisc_cb_info, &mip->mi_promisc_list,
3374 	    &mpip->mpi_mi_link);
3375 
3376 	mutex_exit(mcbi->mcbi_lockp);
3377 
3378 	*mphp = (mac_promisc_handle_t)mpip;
3379 
3380 	if (mcip->mci_state_flags & MCIS_IS_VNIC) {
3381 		mac_impl_t *umip = mcip->mci_upper_mip;
3382 
3383 		ASSERT(umip != NULL);
3384 		mac_vnic_secondary_update(umip);
3385 	}
3386 
3387 	i_mac_perim_exit(mip);
3388 
3389 	return (0);
3390 }
3391 
3392 /*
3393  * Remove a multicast address previously aded through mac_promisc_add().
3394  */
3395 void
3396 mac_promisc_remove(mac_promisc_handle_t mph)
3397 {
3398 	mac_promisc_impl_t *mpip = (mac_promisc_impl_t *)mph;
3399 	mac_client_impl_t *mcip = mpip->mpi_mcip;
3400 	mac_impl_t *mip = mcip->mci_mip;
3401 	mac_cb_info_t *mcbi;
3402 	int rv;
3403 
3404 	i_mac_perim_enter(mip);
3405 
3406 	/*
3407 	 * Even if the device can't be reset into normal mode, we still
3408 	 * need to clear the client promisc callbacks. The client may want
3409 	 * to close the mac end point and we can't have stale callbacks.
3410 	 */
3411 	if (!(mpip->mpi_no_phys)) {
3412 		if ((rv = i_mac_promisc_set(mip, B_FALSE)) != 0) {
3413 			cmn_err(CE_WARN, "%s: failed to switch OFF promiscuous"
3414 			    " mode because of error 0x%x", mip->mi_name, rv);
3415 		}
3416 	}
3417 	mcbi = &mip->mi_promisc_cb_info;
3418 	mutex_enter(mcbi->mcbi_lockp);
3419 	if (mac_callback_remove(mcbi, &mip->mi_promisc_list,
3420 	    &mpip->mpi_mi_link)) {
3421 		VERIFY(mac_callback_remove(&mip->mi_promisc_cb_info,
3422 		    &mcip->mci_promisc_list, &mpip->mpi_mci_link));
3423 		kmem_cache_free(mac_promisc_impl_cache, mpip);
3424 	} else {
3425 		mac_callback_remove_wait(&mip->mi_promisc_cb_info);
3426 	}
3427 
3428 	if (mcip->mci_state_flags & MCIS_IS_VNIC) {
3429 		mac_impl_t *umip = mcip->mci_upper_mip;
3430 
3431 		ASSERT(umip != NULL);
3432 		mac_vnic_secondary_update(umip);
3433 	}
3434 
3435 	mutex_exit(mcbi->mcbi_lockp);
3436 	mac_stop((mac_handle_t)mip);
3437 
3438 	i_mac_perim_exit(mip);
3439 }
3440 
3441 /*
3442  * Reference count the number of active Tx threads. MCI_TX_QUIESCE indicates
3443  * that a control operation wants to quiesce the Tx data flow in which case
3444  * we return an error. Holding any of the per cpu locks ensures that the
3445  * mci_tx_flag won't change.
3446  *
3447  * 'CPU' must be accessed just once and used to compute the index into the
3448  * percpu array, and that index must be used for the entire duration of the
3449  * packet send operation. Note that the thread may be preempted and run on
3450  * another cpu any time and so we can't use 'CPU' more than once for the
3451  * operation.
3452  */
3453 #define	MAC_TX_TRY_HOLD(mcip, mytx, error)				\
3454 {									\
3455 	(error) = 0;							\
3456 	(mytx) = &(mcip)->mci_tx_pcpu[CPU->cpu_seqid & mac_tx_percpu_cnt]; \
3457 	mutex_enter(&(mytx)->pcpu_tx_lock);				\
3458 	if (!((mcip)->mci_tx_flag & MCI_TX_QUIESCE)) {			\
3459 		(mytx)->pcpu_tx_refcnt++;				\
3460 	} else {							\
3461 		(error) = -1;						\
3462 	}								\
3463 	mutex_exit(&(mytx)->pcpu_tx_lock);				\
3464 }
3465 
3466 /*
3467  * Release the reference. If needed, signal any control operation waiting
3468  * for Tx quiescence. The wait and signal are always done using the
3469  * mci_tx_pcpu[0]'s lock
3470  */
3471 #define	MAC_TX_RELE(mcip, mytx) {					\
3472 	mutex_enter(&(mytx)->pcpu_tx_lock);				\
3473 	if (--(mytx)->pcpu_tx_refcnt == 0 &&				\
3474 	    (mcip)->mci_tx_flag & MCI_TX_QUIESCE) {			\
3475 		mutex_exit(&(mytx)->pcpu_tx_lock);			\
3476 		mutex_enter(&(mcip)->mci_tx_pcpu[0].pcpu_tx_lock);	\
3477 		cv_signal(&(mcip)->mci_tx_cv);				\
3478 		mutex_exit(&(mcip)->mci_tx_pcpu[0].pcpu_tx_lock);	\
3479 	} else {							\
3480 		mutex_exit(&(mytx)->pcpu_tx_lock);			\
3481 	}								\
3482 }
3483 
3484 /*
3485  * Send function invoked by MAC clients.
3486  */
3487 mac_tx_cookie_t
3488 mac_tx(mac_client_handle_t mch, mblk_t *mp_chain, uintptr_t hint,
3489     uint16_t flag, mblk_t **ret_mp)
3490 {
3491 	mac_tx_cookie_t		cookie = 0;
3492 	int			error;
3493 	mac_tx_percpu_t		*mytx;
3494 	mac_soft_ring_set_t	*srs;
3495 	flow_entry_t		*flent;
3496 	boolean_t		is_subflow = B_FALSE;
3497 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3498 	mac_impl_t		*mip = mcip->mci_mip;
3499 	mac_srs_tx_t		*srs_tx;
3500 
3501 	/*
3502 	 * Check whether the active Tx threads count is bumped already.
3503 	 */
3504 	if (!(flag & MAC_TX_NO_HOLD)) {
3505 		MAC_TX_TRY_HOLD(mcip, mytx, error);
3506 		if (error != 0) {
3507 			freemsgchain(mp_chain);
3508 			return (0);
3509 		}
3510 	}
3511 
3512 	/*
3513 	 * If mac protection is enabled, only the permissible packets will be
3514 	 * returned by mac_protect_check().
3515 	 */
3516 	if ((mcip->mci_flent->
3517 	    fe_resource_props.mrp_mask & MRP_PROTECT) != 0 &&
3518 	    (mp_chain = mac_protect_check(mch, mp_chain)) == NULL)
3519 		goto done;
3520 
3521 	if (mcip->mci_subflow_tab != NULL &&
3522 	    mcip->mci_subflow_tab->ft_flow_count > 0 &&
3523 	    mac_flow_lookup(mcip->mci_subflow_tab, mp_chain,
3524 	    FLOW_OUTBOUND, &flent) == 0) {
3525 		/*
3526 		 * The main assumption here is that if in the event
3527 		 * we get a chain, all the packets will be classified
3528 		 * to the same Flow/SRS. If this changes for any
3529 		 * reason, the following logic should change as well.
3530 		 * I suppose the fanout_hint also assumes this .
3531 		 */
3532 		ASSERT(flent != NULL);
3533 		is_subflow = B_TRUE;
3534 	} else {
3535 		flent = mcip->mci_flent;
3536 	}
3537 
3538 	srs = flent->fe_tx_srs;
3539 	/*
3540 	 * This is to avoid panics with PF_PACKET that can call mac_tx()
3541 	 * against an interface that is not capable of sending. A rewrite
3542 	 * of the mac datapath is required to remove this limitation.
3543 	 */
3544 	if (srs == NULL) {
3545 		freemsgchain(mp_chain);
3546 		goto done;
3547 	}
3548 
3549 	srs_tx = &srs->srs_tx;
3550 	if (srs_tx->st_mode == SRS_TX_DEFAULT &&
3551 	    (srs->srs_state & SRS_ENQUEUED) == 0 &&
3552 	    mip->mi_nactiveclients == 1 && mp_chain->b_next == NULL) {
3553 		uint64_t	obytes;
3554 
3555 		/*
3556 		 * Since dls always opens the underlying MAC, nclients equals
3557 		 * to 1 means that the only active client is dls itself acting
3558 		 * as a primary client of the MAC instance. Since dls will not
3559 		 * send tagged packets in that case, and dls is trusted to send
3560 		 * packets for its allowed VLAN(s), the VLAN tag insertion and
3561 		 * check is required only if nclients is greater than 1.
3562 		 */
3563 		if (mip->mi_nclients > 1) {
3564 			if (MAC_VID_CHECK_NEEDED(mcip)) {
3565 				int	err = 0;
3566 
3567 				MAC_VID_CHECK(mcip, mp_chain, err);
3568 				if (err != 0) {
3569 					freemsg(mp_chain);
3570 					mcip->mci_misc_stat.mms_txerrors++;
3571 					goto done;
3572 				}
3573 			}
3574 			if (MAC_TAG_NEEDED(mcip)) {
3575 				mp_chain = mac_add_vlan_tag(mp_chain, 0,
3576 				    mac_client_vid(mch));
3577 				if (mp_chain == NULL) {
3578 					mcip->mci_misc_stat.mms_txerrors++;
3579 					goto done;
3580 				}
3581 			}
3582 		}
3583 
3584 		obytes = (mp_chain->b_cont == NULL ? MBLKL(mp_chain) :
3585 		    msgdsize(mp_chain));
3586 
3587 		MAC_TX(mip, srs_tx->st_arg2, mp_chain, mcip);
3588 		if (mp_chain == NULL) {
3589 			cookie = 0;
3590 			SRS_TX_STAT_UPDATE(srs, opackets, 1);
3591 			SRS_TX_STAT_UPDATE(srs, obytes, obytes);
3592 		} else {
3593 			mutex_enter(&srs->srs_lock);
3594 			cookie = mac_tx_srs_no_desc(srs, mp_chain,
3595 			    flag, ret_mp);
3596 			mutex_exit(&srs->srs_lock);
3597 		}
3598 	} else {
3599 		cookie = srs_tx->st_func(srs, mp_chain, hint, flag, ret_mp);
3600 	}
3601 
3602 done:
3603 	if (is_subflow)
3604 		FLOW_REFRELE(flent);
3605 
3606 	if (!(flag & MAC_TX_NO_HOLD))
3607 		MAC_TX_RELE(mcip, mytx);
3608 
3609 	return (cookie);
3610 }
3611 
3612 /*
3613  * mac_tx_is_blocked
3614  *
3615  * Given a cookie, it returns if the ring identified by the cookie is
3616  * flow-controlled or not. If NULL is passed in place of a cookie,
3617  * then it finds out if any of the underlying rings belonging to the
3618  * SRS is flow controlled or not and returns that status.
3619  */
3620 /* ARGSUSED */
3621 boolean_t
3622 mac_tx_is_flow_blocked(mac_client_handle_t mch, mac_tx_cookie_t cookie)
3623 {
3624 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3625 	mac_soft_ring_set_t *mac_srs;
3626 	mac_soft_ring_t *sringp;
3627 	boolean_t blocked = B_FALSE;
3628 	mac_tx_percpu_t *mytx;
3629 	int err;
3630 	int i;
3631 
3632 	/*
3633 	 * Bump the reference count so that mac_srs won't be deleted.
3634 	 * If the client is currently quiesced and we failed to bump
3635 	 * the reference, return B_TRUE so that flow control stays
3636 	 * as enabled.
3637 	 *
3638 	 * Flow control will then be disabled once the client is no
3639 	 * longer quiesced.
3640 	 */
3641 	MAC_TX_TRY_HOLD(mcip, mytx, err);
3642 	if (err != 0)
3643 		return (B_TRUE);
3644 
3645 	if ((mac_srs = MCIP_TX_SRS(mcip)) == NULL) {
3646 		MAC_TX_RELE(mcip, mytx);
3647 		return (B_FALSE);
3648 	}
3649 
3650 	mutex_enter(&mac_srs->srs_lock);
3651 	/*
3652 	 * Only in the case of TX_FANOUT and TX_AGGR, the underlying
3653 	 * softring (s_ring_state) will have the HIWAT set. This is
3654 	 * the multiple Tx ring flow control case. For all other
3655 	 * case, SRS (srs_state) will store the condition.
3656 	 */
3657 	if (mac_srs->srs_tx.st_mode == SRS_TX_FANOUT ||
3658 	    mac_srs->srs_tx.st_mode == SRS_TX_AGGR) {
3659 		if (cookie != 0) {
3660 			sringp = (mac_soft_ring_t *)cookie;
3661 			mutex_enter(&sringp->s_ring_lock);
3662 			if (sringp->s_ring_state & S_RING_TX_HIWAT)
3663 				blocked = B_TRUE;
3664 			mutex_exit(&sringp->s_ring_lock);
3665 		} else {
3666 			for (i = 0; i < mac_srs->srs_tx_ring_count; i++) {
3667 				sringp = mac_srs->srs_tx_soft_rings[i];
3668 				mutex_enter(&sringp->s_ring_lock);
3669 				if (sringp->s_ring_state & S_RING_TX_HIWAT) {
3670 					blocked = B_TRUE;
3671 					mutex_exit(&sringp->s_ring_lock);
3672 					break;
3673 				}
3674 				mutex_exit(&sringp->s_ring_lock);
3675 			}
3676 		}
3677 	} else {
3678 		blocked = (mac_srs->srs_state & SRS_TX_HIWAT);
3679 	}
3680 	mutex_exit(&mac_srs->srs_lock);
3681 	MAC_TX_RELE(mcip, mytx);
3682 	return (blocked);
3683 }
3684 
3685 /*
3686  * Check if the MAC client is the primary MAC client.
3687  */
3688 boolean_t
3689 mac_is_primary_client(mac_client_impl_t *mcip)
3690 {
3691 	return (mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY);
3692 }
3693 
3694 void
3695 mac_ioctl(mac_handle_t mh, queue_t *wq, mblk_t *bp)
3696 {
3697 	mac_impl_t	*mip = (mac_impl_t *)mh;
3698 	int cmd = ((struct iocblk *)bp->b_rptr)->ioc_cmd;
3699 
3700 	if ((cmd == ND_GET && (mip->mi_callbacks->mc_callbacks & MC_GETPROP)) ||
3701 	    (cmd == ND_SET && (mip->mi_callbacks->mc_callbacks & MC_SETPROP))) {
3702 		/*
3703 		 * If ndd props were registered, call them.
3704 		 * Note that ndd ioctls are Obsolete
3705 		 */
3706 		mac_ndd_ioctl(mip, wq, bp);
3707 		return;
3708 	}
3709 
3710 	/*
3711 	 * Call the driver to handle the ioctl.  The driver may not support
3712 	 * any ioctls, in which case we reply with a NAK on its behalf.
3713 	 */
3714 	if (mip->mi_callbacks->mc_callbacks & MC_IOCTL)
3715 		mip->mi_ioctl(mip->mi_driver, wq, bp);
3716 	else
3717 		miocnak(wq, bp, 0, EINVAL);
3718 }
3719 
3720 /*
3721  * Return the link state of the specified MAC instance.
3722  */
3723 link_state_t
3724 mac_link_get(mac_handle_t mh)
3725 {
3726 	return (((mac_impl_t *)mh)->mi_linkstate);
3727 }
3728 
3729 /*
3730  * Add a mac client specified notification callback. Please see the comments
3731  * above mac_callback_add() for general information about mac callback
3732  * addition/deletion in the presence of mac callback list walkers
3733  */
3734 mac_notify_handle_t
3735 mac_notify_add(mac_handle_t mh, mac_notify_t notify_fn, void *arg)
3736 {
3737 	mac_impl_t		*mip = (mac_impl_t *)mh;
3738 	mac_notify_cb_t		*mncb;
3739 	mac_cb_info_t		*mcbi;
3740 
3741 	/*
3742 	 * Allocate a notify callback structure, fill in the details and
3743 	 * use the mac callback list manipulation functions to chain into
3744 	 * the list of callbacks.
3745 	 */
3746 	mncb = kmem_zalloc(sizeof (mac_notify_cb_t), KM_SLEEP);
3747 	mncb->mncb_fn = notify_fn;
3748 	mncb->mncb_arg = arg;
3749 	mncb->mncb_mip = mip;
3750 	mncb->mncb_link.mcb_objp = mncb;
3751 	mncb->mncb_link.mcb_objsize = sizeof (mac_notify_cb_t);
3752 	mncb->mncb_link.mcb_flags = MCB_NOTIFY_CB_T;
3753 
3754 	mcbi = &mip->mi_notify_cb_info;
3755 
3756 	i_mac_perim_enter(mip);
3757 	mutex_enter(mcbi->mcbi_lockp);
3758 
3759 	mac_callback_add(&mip->mi_notify_cb_info, &mip->mi_notify_cb_list,
3760 	    &mncb->mncb_link);
3761 
3762 	mutex_exit(mcbi->mcbi_lockp);
3763 	i_mac_perim_exit(mip);
3764 	return ((mac_notify_handle_t)mncb);
3765 }
3766 
3767 void
3768 mac_notify_remove_wait(mac_handle_t mh)
3769 {
3770 	mac_impl_t	*mip = (mac_impl_t *)mh;
3771 	mac_cb_info_t	*mcbi = &mip->mi_notify_cb_info;
3772 
3773 	mutex_enter(mcbi->mcbi_lockp);
3774 	mac_callback_remove_wait(&mip->mi_notify_cb_info);
3775 	mutex_exit(mcbi->mcbi_lockp);
3776 }
3777 
3778 /*
3779  * Remove a mac client specified notification callback
3780  */
3781 int
3782 mac_notify_remove(mac_notify_handle_t mnh, boolean_t wait)
3783 {
3784 	mac_notify_cb_t	*mncb = (mac_notify_cb_t *)mnh;
3785 	mac_impl_t	*mip = mncb->mncb_mip;
3786 	mac_cb_info_t	*mcbi;
3787 	int		err = 0;
3788 
3789 	mcbi = &mip->mi_notify_cb_info;
3790 
3791 	i_mac_perim_enter(mip);
3792 	mutex_enter(mcbi->mcbi_lockp);
3793 
3794 	ASSERT(mncb->mncb_link.mcb_objp == mncb);
3795 	/*
3796 	 * If there aren't any list walkers, the remove would succeed
3797 	 * inline, else we wait for the deferred remove to complete
3798 	 */
3799 	if (mac_callback_remove(&mip->mi_notify_cb_info,
3800 	    &mip->mi_notify_cb_list, &mncb->mncb_link)) {
3801 		kmem_free(mncb, sizeof (mac_notify_cb_t));
3802 	} else {
3803 		err = EBUSY;
3804 	}
3805 
3806 	mutex_exit(mcbi->mcbi_lockp);
3807 	i_mac_perim_exit(mip);
3808 
3809 	/*
3810 	 * If we failed to remove the notification callback and "wait" is set
3811 	 * to be B_TRUE, wait for the callback to finish after we exit the
3812 	 * mac perimeter.
3813 	 */
3814 	if (err != 0 && wait) {
3815 		mac_notify_remove_wait((mac_handle_t)mip);
3816 		return (0);
3817 	}
3818 
3819 	return (err);
3820 }
3821 
3822 /*
3823  * Associate resource management callbacks with the specified MAC
3824  * clients.
3825  */
3826 
3827 void
3828 mac_resource_set_common(mac_client_handle_t mch, mac_resource_add_t add,
3829     mac_resource_remove_t remove, mac_resource_quiesce_t quiesce,
3830     mac_resource_restart_t restart, mac_resource_bind_t bind,
3831     void *arg)
3832 {
3833 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3834 
3835 	mcip->mci_resource_add = add;
3836 	mcip->mci_resource_remove = remove;
3837 	mcip->mci_resource_quiesce = quiesce;
3838 	mcip->mci_resource_restart = restart;
3839 	mcip->mci_resource_bind = bind;
3840 	mcip->mci_resource_arg = arg;
3841 }
3842 
3843 void
3844 mac_resource_set(mac_client_handle_t mch, mac_resource_add_t add, void *arg)
3845 {
3846 	/* update the 'resource_add' callback */
3847 	mac_resource_set_common(mch, add, NULL, NULL, NULL, NULL, arg);
3848 }
3849 
3850 /*
3851  * Sets up the client resources and enable the polling interface over all the
3852  * SRS's and the soft rings of the client
3853  */
3854 void
3855 mac_client_poll_enable(mac_client_handle_t mch)
3856 {
3857 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3858 	mac_soft_ring_set_t	*mac_srs;
3859 	flow_entry_t		*flent;
3860 	int			i;
3861 
3862 	flent = mcip->mci_flent;
3863 	ASSERT(flent != NULL);
3864 
3865 	mcip->mci_state_flags |= MCIS_CLIENT_POLL_CAPABLE;
3866 	for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
3867 		mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
3868 		ASSERT(mac_srs->srs_mcip == mcip);
3869 		mac_srs_client_poll_enable(mcip, mac_srs);
3870 	}
3871 }
3872 
3873 /*
3874  * Tears down the client resources and disable the polling interface over all
3875  * the SRS's and the soft rings of the client
3876  */
3877 void
3878 mac_client_poll_disable(mac_client_handle_t mch)
3879 {
3880 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3881 	mac_soft_ring_set_t	*mac_srs;
3882 	flow_entry_t		*flent;
3883 	int			i;
3884 
3885 	flent = mcip->mci_flent;
3886 	ASSERT(flent != NULL);
3887 
3888 	mcip->mci_state_flags &= ~MCIS_CLIENT_POLL_CAPABLE;
3889 	for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
3890 		mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
3891 		ASSERT(mac_srs->srs_mcip == mcip);
3892 		mac_srs_client_poll_disable(mcip, mac_srs);
3893 	}
3894 }
3895 
3896 /*
3897  * Associate the CPUs specified by the given property with a MAC client.
3898  */
3899 int
3900 mac_cpu_set(mac_client_handle_t mch, mac_resource_props_t *mrp)
3901 {
3902 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3903 	mac_impl_t *mip = mcip->mci_mip;
3904 	int err = 0;
3905 
3906 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
3907 
3908 	if ((err = mac_validate_props(mcip->mci_state_flags & MCIS_IS_VNIC ?
3909 	    mcip->mci_upper_mip : mip, mrp)) != 0) {
3910 		return (err);
3911 	}
3912 	if (MCIP_DATAPATH_SETUP(mcip))
3913 		mac_flow_modify(mip->mi_flow_tab, mcip->mci_flent, mrp);
3914 
3915 	mac_update_resources(mrp, MCIP_RESOURCE_PROPS(mcip), B_FALSE);
3916 	return (0);
3917 }
3918 
3919 /*
3920  * Apply the specified properties to the specified MAC client.
3921  */
3922 int
3923 mac_client_set_resources(mac_client_handle_t mch, mac_resource_props_t *mrp)
3924 {
3925 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3926 	mac_impl_t *mip = mcip->mci_mip;
3927 	int err = 0;
3928 
3929 	i_mac_perim_enter(mip);
3930 
3931 	if ((mrp->mrp_mask & MRP_MAXBW) || (mrp->mrp_mask & MRP_PRIORITY)) {
3932 		err = mac_resource_ctl_set(mch, mrp);
3933 		if (err != 0)
3934 			goto done;
3935 	}
3936 
3937 	if (mrp->mrp_mask & (MRP_CPUS|MRP_POOL)) {
3938 		err = mac_cpu_set(mch, mrp);
3939 		if (err != 0)
3940 			goto done;
3941 	}
3942 
3943 	if (mrp->mrp_mask & MRP_PROTECT) {
3944 		err = mac_protect_set(mch, mrp);
3945 		if (err != 0)
3946 			goto done;
3947 	}
3948 
3949 	if ((mrp->mrp_mask & MRP_RX_RINGS) || (mrp->mrp_mask & MRP_TX_RINGS))
3950 		err = mac_resource_ctl_set(mch, mrp);
3951 
3952 done:
3953 	i_mac_perim_exit(mip);
3954 	return (err);
3955 }
3956 
3957 /*
3958  * Return the properties currently associated with the specified MAC client.
3959  */
3960 void
3961 mac_client_get_resources(mac_client_handle_t mch, mac_resource_props_t *mrp)
3962 {
3963 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3964 	mac_resource_props_t	*mcip_mrp = MCIP_RESOURCE_PROPS(mcip);
3965 
3966 	bcopy(mcip_mrp, mrp, sizeof (mac_resource_props_t));
3967 }
3968 
3969 /*
3970  * Return the effective properties currently associated with the specified
3971  * MAC client.
3972  */
3973 void
3974 mac_client_get_effective_resources(mac_client_handle_t mch,
3975     mac_resource_props_t *mrp)
3976 {
3977 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3978 	mac_resource_props_t	*mcip_mrp = MCIP_EFFECTIVE_PROPS(mcip);
3979 
3980 	bcopy(mcip_mrp, mrp, sizeof (mac_resource_props_t));
3981 }
3982 
3983 /*
3984  * Pass a copy of the specified packet to the promiscuous callbacks
3985  * of the specified MAC.
3986  *
3987  * If sender is NULL, the function is being invoked for a packet chain
3988  * received from the wire. If sender is non-NULL, it points to
3989  * the MAC client from which the packet is being sent.
3990  *
3991  * The packets are distributed to the promiscuous callbacks as follows:
3992  *
3993  * - all packets are sent to the MAC_CLIENT_PROMISC_ALL callbacks
3994  * - all broadcast and multicast packets are sent to the
3995  *   MAC_CLIENT_PROMISC_FILTER and MAC_CLIENT_PROMISC_MULTI.
3996  *
3997  * The unicast packets of MAC_CLIENT_PROMISC_FILTER callbacks are dispatched
3998  * after classification by mac_rx_deliver().
3999  */
4000 
4001 static void
4002 mac_promisc_dispatch_one(mac_promisc_impl_t *mpip, mblk_t *mp,
4003     boolean_t loopback)
4004 {
4005 	mblk_t *mp_copy, *mp_next;
4006 
4007 	if (!mpip->mpi_no_copy || mpip->mpi_strip_vlan_tag) {
4008 		mp_copy = copymsg(mp);
4009 		if (mp_copy == NULL)
4010 			return;
4011 
4012 		if (mpip->mpi_strip_vlan_tag) {
4013 			mp_copy = mac_strip_vlan_tag_chain(mp_copy);
4014 			if (mp_copy == NULL)
4015 				return;
4016 		}
4017 		mp_next = NULL;
4018 	} else {
4019 		mp_copy = mp;
4020 		mp_next = mp->b_next;
4021 	}
4022 	mp_copy->b_next = NULL;
4023 
4024 	mpip->mpi_fn(mpip->mpi_arg, NULL, mp_copy, loopback);
4025 	if (mp_copy == mp)
4026 		mp->b_next = mp_next;
4027 }
4028 
4029 /*
4030  * Return the VID of a packet. Zero if the packet is not tagged.
4031  */
4032 static uint16_t
4033 mac_ether_vid(mblk_t *mp)
4034 {
4035 	struct ether_header *eth = (struct ether_header *)mp->b_rptr;
4036 
4037 	if (ntohs(eth->ether_type) == ETHERTYPE_VLAN) {
4038 		struct ether_vlan_header *t_evhp =
4039 		    (struct ether_vlan_header *)mp->b_rptr;
4040 		return (VLAN_ID(ntohs(t_evhp->ether_tci)));
4041 	}
4042 
4043 	return (0);
4044 }
4045 
4046 /*
4047  * Return whether the specified packet contains a multicast or broadcast
4048  * destination MAC address.
4049  */
4050 static boolean_t
4051 mac_is_mcast(mac_impl_t *mip, mblk_t *mp)
4052 {
4053 	mac_header_info_t hdr_info;
4054 
4055 	if (mac_header_info((mac_handle_t)mip, mp, &hdr_info) != 0)
4056 		return (B_FALSE);
4057 	return ((hdr_info.mhi_dsttype == MAC_ADDRTYPE_BROADCAST) ||
4058 	    (hdr_info.mhi_dsttype == MAC_ADDRTYPE_MULTICAST));
4059 }
4060 
4061 /*
4062  * Send a copy of an mblk chain to the MAC clients of the specified MAC.
4063  * "sender" points to the sender MAC client for outbound packets, and
4064  * is set to NULL for inbound packets.
4065  */
4066 void
4067 mac_promisc_dispatch(mac_impl_t *mip, mblk_t *mp_chain,
4068     mac_client_impl_t *sender)
4069 {
4070 	mac_promisc_impl_t *mpip;
4071 	mac_cb_t *mcb;
4072 	mblk_t *mp;
4073 	boolean_t is_mcast, is_sender;
4074 
4075 	MAC_PROMISC_WALKER_INC(mip);
4076 	for (mp = mp_chain; mp != NULL; mp = mp->b_next) {
4077 		is_mcast = mac_is_mcast(mip, mp);
4078 		/* send packet to interested callbacks */
4079 		for (mcb = mip->mi_promisc_list; mcb != NULL;
4080 		    mcb = mcb->mcb_nextp) {
4081 			mpip = (mac_promisc_impl_t *)mcb->mcb_objp;
4082 			is_sender = (mpip->mpi_mcip == sender);
4083 
4084 			if (is_sender && mpip->mpi_no_tx_loop)
4085 				/*
4086 				 * The sender doesn't want to receive
4087 				 * copies of the packets it sends.
4088 				 */
4089 				continue;
4090 
4091 			/* this client doesn't need any packets (bridge) */
4092 			if (mpip->mpi_fn == NULL)
4093 				continue;
4094 
4095 			/*
4096 			 * For an ethernet MAC, don't displatch a multicast
4097 			 * packet to a non-PROMISC_ALL callbacks unless the VID
4098 			 * of the packet matches the VID of the client.
4099 			 */
4100 			if (is_mcast &&
4101 			    mpip->mpi_type != MAC_CLIENT_PROMISC_ALL &&
4102 			    !mac_client_check_flow_vid(mpip->mpi_mcip,
4103 			    mac_ether_vid(mp)))
4104 				continue;
4105 
4106 			if (is_sender ||
4107 			    mpip->mpi_type == MAC_CLIENT_PROMISC_ALL ||
4108 			    is_mcast)
4109 				mac_promisc_dispatch_one(mpip, mp, is_sender);
4110 		}
4111 	}
4112 	MAC_PROMISC_WALKER_DCR(mip);
4113 }
4114 
4115 void
4116 mac_promisc_client_dispatch(mac_client_impl_t *mcip, mblk_t *mp_chain)
4117 {
4118 	mac_impl_t		*mip = mcip->mci_mip;
4119 	mac_promisc_impl_t	*mpip;
4120 	boolean_t		is_mcast;
4121 	mblk_t			*mp;
4122 	mac_cb_t		*mcb;
4123 
4124 	/*
4125 	 * The unicast packets for the MAC client still
4126 	 * need to be delivered to the MAC_CLIENT_PROMISC_FILTERED
4127 	 * promiscuous callbacks. The broadcast and multicast
4128 	 * packets were delivered from mac_rx().
4129 	 */
4130 	MAC_PROMISC_WALKER_INC(mip);
4131 	for (mp = mp_chain; mp != NULL; mp = mp->b_next) {
4132 		is_mcast = mac_is_mcast(mip, mp);
4133 		for (mcb = mcip->mci_promisc_list; mcb != NULL;
4134 		    mcb = mcb->mcb_nextp) {
4135 			mpip = (mac_promisc_impl_t *)mcb->mcb_objp;
4136 			if (mpip->mpi_type == MAC_CLIENT_PROMISC_FILTERED &&
4137 			    !is_mcast) {
4138 				mac_promisc_dispatch_one(mpip, mp, B_FALSE);
4139 			}
4140 		}
4141 	}
4142 	MAC_PROMISC_WALKER_DCR(mip);
4143 }
4144 
4145 /*
4146  * Return the margin value currently assigned to the specified MAC instance.
4147  */
4148 void
4149 mac_margin_get(mac_handle_t mh, uint32_t *marginp)
4150 {
4151 	mac_impl_t *mip = (mac_impl_t *)mh;
4152 
4153 	rw_enter(&(mip->mi_rw_lock), RW_READER);
4154 	*marginp = mip->mi_margin;
4155 	rw_exit(&(mip->mi_rw_lock));
4156 }
4157 
4158 /*
4159  * mac_info_get() is used for retrieving the mac_info when a DL_INFO_REQ is
4160  * issued before a DL_ATTACH_REQ. we walk the i_mac_impl_hash table and find
4161  * the first mac_impl_t with a matching driver name; then we copy its mac_info_t
4162  * to the caller. we do all this with i_mac_impl_lock held so the mac_impl_t
4163  * cannot disappear while we are accessing it.
4164  */
4165 typedef struct i_mac_info_state_s {
4166 	const char	*mi_name;
4167 	mac_info_t	*mi_infop;
4168 } i_mac_info_state_t;
4169 
4170 /*ARGSUSED*/
4171 static uint_t
4172 i_mac_info_walker(mod_hash_key_t key, mod_hash_val_t *val, void *arg)
4173 {
4174 	i_mac_info_state_t *statep = arg;
4175 	mac_impl_t *mip = (mac_impl_t *)val;
4176 
4177 	if (mip->mi_state_flags & MIS_DISABLED)
4178 		return (MH_WALK_CONTINUE);
4179 
4180 	if (strcmp(statep->mi_name,
4181 	    ddi_driver_name(mip->mi_dip)) != 0)
4182 		return (MH_WALK_CONTINUE);
4183 
4184 	statep->mi_infop = &mip->mi_info;
4185 	return (MH_WALK_TERMINATE);
4186 }
4187 
4188 boolean_t
4189 mac_info_get(const char *name, mac_info_t *minfop)
4190 {
4191 	i_mac_info_state_t state;
4192 
4193 	rw_enter(&i_mac_impl_lock, RW_READER);
4194 	state.mi_name = name;
4195 	state.mi_infop = NULL;
4196 	mod_hash_walk(i_mac_impl_hash, i_mac_info_walker, &state);
4197 	if (state.mi_infop == NULL) {
4198 		rw_exit(&i_mac_impl_lock);
4199 		return (B_FALSE);
4200 	}
4201 	*minfop = *state.mi_infop;
4202 	rw_exit(&i_mac_impl_lock);
4203 	return (B_TRUE);
4204 }
4205 
4206 /*
4207  * To get the capabilities that MAC layer cares about, such as rings, factory
4208  * mac address, vnic or not, it should directly invoke this function.  If the
4209  * link is part of a bridge, then the only "capability" it has is the inability
4210  * to do zero copy.
4211  */
4212 boolean_t
4213 i_mac_capab_get(mac_handle_t mh, mac_capab_t cap, void *cap_data)
4214 {
4215 	mac_impl_t *mip = (mac_impl_t *)mh;
4216 
4217 	if (mip->mi_bridge_link != NULL)
4218 		return (cap == MAC_CAPAB_NO_ZCOPY);
4219 	else if (mip->mi_callbacks->mc_callbacks & MC_GETCAPAB)
4220 		return (mip->mi_getcapab(mip->mi_driver, cap, cap_data));
4221 	else
4222 		return (B_FALSE);
4223 }
4224 
4225 /*
4226  * Capability query function. If number of active mac clients is greater than
4227  * 1, only limited capabilities can be advertised to the caller no matter the
4228  * driver has certain capability or not. Else, we query the driver to get the
4229  * capability.
4230  */
4231 boolean_t
4232 mac_capab_get(mac_handle_t mh, mac_capab_t cap, void *cap_data)
4233 {
4234 	mac_impl_t *mip = (mac_impl_t *)mh;
4235 
4236 	/*
4237 	 * if mi_nactiveclients > 1, only MAC_CAPAB_LEGACY, MAC_CAPAB_HCKSUM,
4238 	 * MAC_CAPAB_NO_NATIVEVLAN and MAC_CAPAB_NO_ZCOPY can be advertised.
4239 	 */
4240 	if (mip->mi_nactiveclients > 1) {
4241 		switch (cap) {
4242 		case MAC_CAPAB_NO_ZCOPY:
4243 			return (B_TRUE);
4244 		case MAC_CAPAB_LEGACY:
4245 		case MAC_CAPAB_HCKSUM:
4246 		case MAC_CAPAB_NO_NATIVEVLAN:
4247 			break;
4248 		default:
4249 			return (B_FALSE);
4250 		}
4251 	}
4252 
4253 	/* else get capab from driver */
4254 	return (i_mac_capab_get(mh, cap, cap_data));
4255 }
4256 
4257 boolean_t
4258 mac_sap_verify(mac_handle_t mh, uint32_t sap, uint32_t *bind_sap)
4259 {
4260 	mac_impl_t *mip = (mac_impl_t *)mh;
4261 
4262 	return (mip->mi_type->mt_ops.mtops_sap_verify(sap, bind_sap,
4263 	    mip->mi_pdata));
4264 }
4265 
4266 mblk_t *
4267 mac_header(mac_handle_t mh, const uint8_t *daddr, uint32_t sap, mblk_t *payload,
4268     size_t extra_len)
4269 {
4270 	mac_impl_t	*mip = (mac_impl_t *)mh;
4271 	const uint8_t	*hdr_daddr;
4272 
4273 	/*
4274 	 * If the MAC is point-to-point with a fixed destination address, then
4275 	 * we must always use that destination in the MAC header.
4276 	 */
4277 	hdr_daddr = (mip->mi_dstaddr_set ? mip->mi_dstaddr : daddr);
4278 	return (mip->mi_type->mt_ops.mtops_header(mip->mi_addr, hdr_daddr, sap,
4279 	    mip->mi_pdata, payload, extra_len));
4280 }
4281 
4282 int
4283 mac_header_info(mac_handle_t mh, mblk_t *mp, mac_header_info_t *mhip)
4284 {
4285 	mac_impl_t *mip = (mac_impl_t *)mh;
4286 
4287 	return (mip->mi_type->mt_ops.mtops_header_info(mp, mip->mi_pdata,
4288 	    mhip));
4289 }
4290 
4291 int
4292 mac_vlan_header_info(mac_handle_t mh, mblk_t *mp, mac_header_info_t *mhip)
4293 {
4294 	mac_impl_t	*mip = (mac_impl_t *)mh;
4295 	boolean_t	is_ethernet = (mip->mi_info.mi_media == DL_ETHER);
4296 	int		err = 0;
4297 
4298 	/*
4299 	 * Packets should always be at least 16 bit aligned.
4300 	 */
4301 	ASSERT(IS_P2ALIGNED(mp->b_rptr, sizeof (uint16_t)));
4302 
4303 	if ((err = mac_header_info(mh, mp, mhip)) != 0)
4304 		return (err);
4305 
4306 	/*
4307 	 * If this is a VLAN-tagged Ethernet packet, then the SAP in the
4308 	 * mac_header_info_t as returned by mac_header_info() is
4309 	 * ETHERTYPE_VLAN. We need to grab the ethertype from the VLAN header.
4310 	 */
4311 	if (is_ethernet && (mhip->mhi_bindsap == ETHERTYPE_VLAN)) {
4312 		struct ether_vlan_header *evhp;
4313 		uint16_t sap;
4314 		mblk_t *tmp = NULL;
4315 		size_t size;
4316 
4317 		size = sizeof (struct ether_vlan_header);
4318 		if (MBLKL(mp) < size) {
4319 			/*
4320 			 * Pullup the message in order to get the MAC header
4321 			 * infomation. Note that this is a read-only function,
4322 			 * we keep the input packet intact.
4323 			 */
4324 			if ((tmp = msgpullup(mp, size)) == NULL)
4325 				return (EINVAL);
4326 
4327 			mp = tmp;
4328 		}
4329 		evhp = (struct ether_vlan_header *)mp->b_rptr;
4330 		sap = ntohs(evhp->ether_type);
4331 		(void) mac_sap_verify(mh, sap, &mhip->mhi_bindsap);
4332 		mhip->mhi_hdrsize = sizeof (struct ether_vlan_header);
4333 		mhip->mhi_tci = ntohs(evhp->ether_tci);
4334 		mhip->mhi_istagged = B_TRUE;
4335 		freemsg(tmp);
4336 
4337 		if (VLAN_CFI(mhip->mhi_tci) != ETHER_CFI)
4338 			return (EINVAL);
4339 	} else {
4340 		mhip->mhi_istagged = B_FALSE;
4341 		mhip->mhi_tci = 0;
4342 	}
4343 
4344 	return (0);
4345 }
4346 
4347 mblk_t *
4348 mac_header_cook(mac_handle_t mh, mblk_t *mp)
4349 {
4350 	mac_impl_t *mip = (mac_impl_t *)mh;
4351 
4352 	if (mip->mi_type->mt_ops.mtops_ops & MTOPS_HEADER_COOK) {
4353 		if (DB_REF(mp) > 1) {
4354 			mblk_t *newmp = copymsg(mp);
4355 			if (newmp == NULL)
4356 				return (NULL);
4357 			freemsg(mp);
4358 			mp = newmp;
4359 		}
4360 		return (mip->mi_type->mt_ops.mtops_header_cook(mp,
4361 		    mip->mi_pdata));
4362 	}
4363 	return (mp);
4364 }
4365 
4366 mblk_t *
4367 mac_header_uncook(mac_handle_t mh, mblk_t *mp)
4368 {
4369 	mac_impl_t *mip = (mac_impl_t *)mh;
4370 
4371 	if (mip->mi_type->mt_ops.mtops_ops & MTOPS_HEADER_UNCOOK) {
4372 		if (DB_REF(mp) > 1) {
4373 			mblk_t *newmp = copymsg(mp);
4374 			if (newmp == NULL)
4375 				return (NULL);
4376 			freemsg(mp);
4377 			mp = newmp;
4378 		}
4379 		return (mip->mi_type->mt_ops.mtops_header_uncook(mp,
4380 		    mip->mi_pdata));
4381 	}
4382 	return (mp);
4383 }
4384 
4385 uint_t
4386 mac_addr_len(mac_handle_t mh)
4387 {
4388 	mac_impl_t *mip = (mac_impl_t *)mh;
4389 
4390 	return (mip->mi_type->mt_addr_length);
4391 }
4392 
4393 /* True if a MAC is a VNIC */
4394 boolean_t
4395 mac_is_vnic(mac_handle_t mh)
4396 {
4397 	return (((mac_impl_t *)mh)->mi_state_flags & MIS_IS_VNIC);
4398 }
4399 
4400 mac_handle_t
4401 mac_get_lower_mac_handle(mac_handle_t mh)
4402 {
4403 	mac_impl_t *mip = (mac_impl_t *)mh;
4404 
4405 	ASSERT(mac_is_vnic(mh));
4406 	return (((vnic_t *)mip->mi_driver)->vn_lower_mh);
4407 }
4408 
4409 boolean_t
4410 mac_is_vnic_primary(mac_handle_t mh)
4411 {
4412 	mac_impl_t *mip = (mac_impl_t *)mh;
4413 
4414 	ASSERT(mac_is_vnic(mh));
4415 	return (((vnic_t *)mip->mi_driver)->vn_addr_type ==
4416 	    VNIC_MAC_ADDR_TYPE_PRIMARY);
4417 }
4418 
4419 void
4420 mac_update_resources(mac_resource_props_t *nmrp, mac_resource_props_t *cmrp,
4421     boolean_t is_user_flow)
4422 {
4423 	if (nmrp != NULL && cmrp != NULL) {
4424 		if (nmrp->mrp_mask & MRP_PRIORITY) {
4425 			if (nmrp->mrp_priority == MPL_RESET) {
4426 				cmrp->mrp_mask &= ~MRP_PRIORITY;
4427 				if (is_user_flow) {
4428 					cmrp->mrp_priority =
4429 					    MPL_SUBFLOW_DEFAULT;
4430 				} else {
4431 					cmrp->mrp_priority = MPL_LINK_DEFAULT;
4432 				}
4433 			} else {
4434 				cmrp->mrp_mask |= MRP_PRIORITY;
4435 				cmrp->mrp_priority = nmrp->mrp_priority;
4436 			}
4437 		}
4438 		if (nmrp->mrp_mask & MRP_MAXBW) {
4439 			if (nmrp->mrp_maxbw == MRP_MAXBW_RESETVAL) {
4440 				cmrp->mrp_mask &= ~MRP_MAXBW;
4441 				cmrp->mrp_maxbw = 0;
4442 			} else {
4443 				cmrp->mrp_mask |= MRP_MAXBW;
4444 				cmrp->mrp_maxbw = nmrp->mrp_maxbw;
4445 			}
4446 		}
4447 		if (nmrp->mrp_mask & MRP_CPUS)
4448 			MAC_COPY_CPUS(nmrp, cmrp);
4449 
4450 		if (nmrp->mrp_mask & MRP_POOL) {
4451 			if (strlen(nmrp->mrp_pool) == 0) {
4452 				cmrp->mrp_mask &= ~MRP_POOL;
4453 				bzero(cmrp->mrp_pool, sizeof (cmrp->mrp_pool));
4454 			} else {
4455 				cmrp->mrp_mask |= MRP_POOL;
4456 				(void) strncpy(cmrp->mrp_pool, nmrp->mrp_pool,
4457 				    sizeof (cmrp->mrp_pool));
4458 			}
4459 
4460 		}
4461 
4462 		if (nmrp->mrp_mask & MRP_PROTECT)
4463 			mac_protect_update(nmrp, cmrp);
4464 
4465 		/*
4466 		 * Update the rings specified.
4467 		 */
4468 		if (nmrp->mrp_mask & MRP_RX_RINGS) {
4469 			if (nmrp->mrp_mask & MRP_RINGS_RESET) {
4470 				cmrp->mrp_mask &= ~MRP_RX_RINGS;
4471 				if (cmrp->mrp_mask & MRP_RXRINGS_UNSPEC)
4472 					cmrp->mrp_mask &= ~MRP_RXRINGS_UNSPEC;
4473 				cmrp->mrp_nrxrings = 0;
4474 			} else {
4475 				cmrp->mrp_mask |= MRP_RX_RINGS;
4476 				cmrp->mrp_nrxrings = nmrp->mrp_nrxrings;
4477 			}
4478 		}
4479 		if (nmrp->mrp_mask & MRP_TX_RINGS) {
4480 			if (nmrp->mrp_mask & MRP_RINGS_RESET) {
4481 				cmrp->mrp_mask &= ~MRP_TX_RINGS;
4482 				if (cmrp->mrp_mask & MRP_TXRINGS_UNSPEC)
4483 					cmrp->mrp_mask &= ~MRP_TXRINGS_UNSPEC;
4484 				cmrp->mrp_ntxrings = 0;
4485 			} else {
4486 				cmrp->mrp_mask |= MRP_TX_RINGS;
4487 				cmrp->mrp_ntxrings = nmrp->mrp_ntxrings;
4488 			}
4489 		}
4490 		if (nmrp->mrp_mask & MRP_RXRINGS_UNSPEC)
4491 			cmrp->mrp_mask |= MRP_RXRINGS_UNSPEC;
4492 		else if (cmrp->mrp_mask & MRP_RXRINGS_UNSPEC)
4493 			cmrp->mrp_mask &= ~MRP_RXRINGS_UNSPEC;
4494 
4495 		if (nmrp->mrp_mask & MRP_TXRINGS_UNSPEC)
4496 			cmrp->mrp_mask |= MRP_TXRINGS_UNSPEC;
4497 		else if (cmrp->mrp_mask & MRP_TXRINGS_UNSPEC)
4498 			cmrp->mrp_mask &= ~MRP_TXRINGS_UNSPEC;
4499 	}
4500 }
4501 
4502 /*
4503  * i_mac_set_resources:
4504  *
4505  * This routine associates properties with the primary MAC client of
4506  * the specified MAC instance.
4507  * - Cache the properties in mac_impl_t
4508  * - Apply the properties to the primary MAC client if exists
4509  */
4510 int
4511 i_mac_set_resources(mac_handle_t mh, mac_resource_props_t *mrp)
4512 {
4513 	mac_impl_t		*mip = (mac_impl_t *)mh;
4514 	mac_client_impl_t	*mcip;
4515 	int			err = 0;
4516 	uint32_t		resmask, newresmask;
4517 	mac_resource_props_t	*tmrp, *umrp;
4518 
4519 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
4520 
4521 	err = mac_validate_props(mip, mrp);
4522 	if (err != 0)
4523 		return (err);
4524 
4525 	umrp = kmem_zalloc(sizeof (*umrp), KM_SLEEP);
4526 	bcopy(&mip->mi_resource_props, umrp, sizeof (*umrp));
4527 	resmask = umrp->mrp_mask;
4528 	mac_update_resources(mrp, umrp, B_FALSE);
4529 	newresmask = umrp->mrp_mask;
4530 
4531 	if (resmask == 0 && newresmask != 0) {
4532 		/*
4533 		 * Bandwidth, priority, cpu or pool link properties configured,
4534 		 * must disable fastpath.
4535 		 */
4536 		if ((err = mac_fastpath_disable((mac_handle_t)mip)) != 0) {
4537 			kmem_free(umrp, sizeof (*umrp));
4538 			return (err);
4539 		}
4540 	}
4541 
4542 	/*
4543 	 * Since bind_cpu may be modified by mac_client_set_resources()
4544 	 * we use a copy of bind_cpu and finally cache bind_cpu in mip.
4545 	 * This allows us to cache only user edits in mip.
4546 	 */
4547 	tmrp = kmem_zalloc(sizeof (*tmrp), KM_SLEEP);
4548 	bcopy(mrp, tmrp, sizeof (*tmrp));
4549 	mcip = mac_primary_client_handle(mip);
4550 	if (mcip != NULL && (mcip->mci_state_flags & MCIS_IS_AGGR_PORT) == 0) {
4551 		err = mac_client_set_resources((mac_client_handle_t)mcip, tmrp);
4552 	} else if ((mrp->mrp_mask & MRP_RX_RINGS ||
4553 	    mrp->mrp_mask & MRP_TX_RINGS)) {
4554 		mac_client_impl_t	*vmcip;
4555 
4556 		/*
4557 		 * If the primary is not up, we need to check if there
4558 		 * are any VLANs on this primary. If there are then
4559 		 * we need to set this property on the VLANs since
4560 		 * VLANs follow the primary they are based on. Just
4561 		 * look for the first VLAN and change its properties,
4562 		 * all the other VLANs should be in the same group.
4563 		 */
4564 		for (vmcip = mip->mi_clients_list; vmcip != NULL;
4565 		    vmcip = vmcip->mci_client_next) {
4566 			if ((vmcip->mci_flent->fe_type & FLOW_PRIMARY_MAC) &&
4567 			    mac_client_vid((mac_client_handle_t)vmcip) !=
4568 			    VLAN_ID_NONE) {
4569 				break;
4570 			}
4571 		}
4572 		if (vmcip != NULL) {
4573 			mac_resource_props_t	*omrp;
4574 			mac_resource_props_t	*vmrp;
4575 
4576 			omrp = kmem_zalloc(sizeof (*omrp), KM_SLEEP);
4577 			bcopy(MCIP_RESOURCE_PROPS(vmcip), omrp, sizeof (*omrp));
4578 			/*
4579 			 * We dont' call mac_update_resources since we
4580 			 * want to take only the ring properties and
4581 			 * not all the properties that may have changed.
4582 			 */
4583 			vmrp = MCIP_RESOURCE_PROPS(vmcip);
4584 			if (mrp->mrp_mask & MRP_RX_RINGS) {
4585 				if (mrp->mrp_mask & MRP_RINGS_RESET) {
4586 					vmrp->mrp_mask &= ~MRP_RX_RINGS;
4587 					if (vmrp->mrp_mask &
4588 					    MRP_RXRINGS_UNSPEC) {
4589 						vmrp->mrp_mask &=
4590 						    ~MRP_RXRINGS_UNSPEC;
4591 					}
4592 					vmrp->mrp_nrxrings = 0;
4593 				} else {
4594 					vmrp->mrp_mask |= MRP_RX_RINGS;
4595 					vmrp->mrp_nrxrings = mrp->mrp_nrxrings;
4596 				}
4597 			}
4598 			if (mrp->mrp_mask & MRP_TX_RINGS) {
4599 				if (mrp->mrp_mask & MRP_RINGS_RESET) {
4600 					vmrp->mrp_mask &= ~MRP_TX_RINGS;
4601 					if (vmrp->mrp_mask &
4602 					    MRP_TXRINGS_UNSPEC) {
4603 						vmrp->mrp_mask &=
4604 						    ~MRP_TXRINGS_UNSPEC;
4605 					}
4606 					vmrp->mrp_ntxrings = 0;
4607 				} else {
4608 					vmrp->mrp_mask |= MRP_TX_RINGS;
4609 					vmrp->mrp_ntxrings = mrp->mrp_ntxrings;
4610 				}
4611 			}
4612 			if (mrp->mrp_mask & MRP_RXRINGS_UNSPEC)
4613 				vmrp->mrp_mask |= MRP_RXRINGS_UNSPEC;
4614 
4615 			if (mrp->mrp_mask & MRP_TXRINGS_UNSPEC)
4616 				vmrp->mrp_mask |= MRP_TXRINGS_UNSPEC;
4617 
4618 			if ((err = mac_client_set_rings_prop(vmcip, mrp,
4619 			    omrp)) != 0) {
4620 				bcopy(omrp, MCIP_RESOURCE_PROPS(vmcip),
4621 				    sizeof (*omrp));
4622 			} else {
4623 				mac_set_prim_vlan_rings(mip, vmrp);
4624 			}
4625 			kmem_free(omrp, sizeof (*omrp));
4626 		}
4627 	}
4628 
4629 	/* Only update the values if mac_client_set_resources succeeded */
4630 	if (err == 0) {
4631 		bcopy(umrp, &mip->mi_resource_props, sizeof (*umrp));
4632 		/*
4633 		 * If bandwidth, priority or cpu link properties cleared,
4634 		 * renable fastpath.
4635 		 */
4636 		if (resmask != 0 && newresmask == 0)
4637 			mac_fastpath_enable((mac_handle_t)mip);
4638 	} else if (resmask == 0 && newresmask != 0) {
4639 		mac_fastpath_enable((mac_handle_t)mip);
4640 	}
4641 	kmem_free(tmrp, sizeof (*tmrp));
4642 	kmem_free(umrp, sizeof (*umrp));
4643 	return (err);
4644 }
4645 
4646 int
4647 mac_set_resources(mac_handle_t mh, mac_resource_props_t *mrp)
4648 {
4649 	int err;
4650 
4651 	i_mac_perim_enter((mac_impl_t *)mh);
4652 	err = i_mac_set_resources(mh, mrp);
4653 	i_mac_perim_exit((mac_impl_t *)mh);
4654 	return (err);
4655 }
4656 
4657 /*
4658  * Get the properties cached for the specified MAC instance.
4659  */
4660 void
4661 mac_get_resources(mac_handle_t mh, mac_resource_props_t *mrp)
4662 {
4663 	mac_impl_t		*mip = (mac_impl_t *)mh;
4664 	mac_client_impl_t	*mcip;
4665 
4666 	mcip = mac_primary_client_handle(mip);
4667 	if (mcip != NULL) {
4668 		mac_client_get_resources((mac_client_handle_t)mcip, mrp);
4669 		return;
4670 	}
4671 	bcopy(&mip->mi_resource_props, mrp, sizeof (mac_resource_props_t));
4672 }
4673 
4674 /*
4675  * Get the effective properties from the primary client of the
4676  * specified MAC instance.
4677  */
4678 void
4679 mac_get_effective_resources(mac_handle_t mh, mac_resource_props_t *mrp)
4680 {
4681 	mac_impl_t		*mip = (mac_impl_t *)mh;
4682 	mac_client_impl_t	*mcip;
4683 
4684 	mcip = mac_primary_client_handle(mip);
4685 	if (mcip != NULL) {
4686 		mac_client_get_effective_resources((mac_client_handle_t)mcip,
4687 		    mrp);
4688 		return;
4689 	}
4690 	bzero(mrp, sizeof (mac_resource_props_t));
4691 }
4692 
4693 int
4694 mac_set_pvid(mac_handle_t mh, uint16_t pvid)
4695 {
4696 	mac_impl_t *mip = (mac_impl_t *)mh;
4697 	mac_client_impl_t *mcip;
4698 	mac_unicast_impl_t *muip;
4699 
4700 	i_mac_perim_enter(mip);
4701 	if (pvid != 0) {
4702 		for (mcip = mip->mi_clients_list; mcip != NULL;
4703 		    mcip = mcip->mci_client_next) {
4704 			for (muip = mcip->mci_unicast_list; muip != NULL;
4705 			    muip = muip->mui_next) {
4706 				if (muip->mui_vid == pvid) {
4707 					i_mac_perim_exit(mip);
4708 					return (EBUSY);
4709 				}
4710 			}
4711 		}
4712 	}
4713 	mip->mi_pvid = pvid;
4714 	i_mac_perim_exit(mip);
4715 	return (0);
4716 }
4717 
4718 uint16_t
4719 mac_get_pvid(mac_handle_t mh)
4720 {
4721 	mac_impl_t *mip = (mac_impl_t *)mh;
4722 
4723 	return (mip->mi_pvid);
4724 }
4725 
4726 uint32_t
4727 mac_get_llimit(mac_handle_t mh)
4728 {
4729 	mac_impl_t *mip = (mac_impl_t *)mh;
4730 
4731 	return (mip->mi_llimit);
4732 }
4733 
4734 uint32_t
4735 mac_get_ldecay(mac_handle_t mh)
4736 {
4737 	mac_impl_t *mip = (mac_impl_t *)mh;
4738 
4739 	return (mip->mi_ldecay);
4740 }
4741 
4742 /*
4743  * Rename a mac client, its flow, and the kstat.
4744  */
4745 int
4746 mac_rename_primary(mac_handle_t mh, const char *new_name)
4747 {
4748 	mac_impl_t		*mip = (mac_impl_t *)mh;
4749 	mac_client_impl_t	*cur_clnt = NULL;
4750 	flow_entry_t		*fep;
4751 
4752 	i_mac_perim_enter(mip);
4753 
4754 	/*
4755 	 * VNICs: we need to change the sys flow name and
4756 	 * the associated flow kstat.
4757 	 */
4758 	if (mip->mi_state_flags & MIS_IS_VNIC) {
4759 		mac_client_impl_t *mcip = mac_vnic_lower(mip);
4760 		ASSERT(new_name != NULL);
4761 		mac_rename_flow_names(mcip, new_name);
4762 		mac_stat_rename(mcip);
4763 		goto done;
4764 	}
4765 	/*
4766 	 * This mac may itself be an aggr link, or it may have some client
4767 	 * which is an aggr port. For both cases, we need to change the
4768 	 * aggr port's mac client name, its flow name and the associated flow
4769 	 * kstat.
4770 	 */
4771 	if (mip->mi_state_flags & MIS_IS_AGGR) {
4772 		mac_capab_aggr_t aggr_cap;
4773 		mac_rename_fn_t rename_fn;
4774 		boolean_t ret;
4775 
4776 		ASSERT(new_name != NULL);
4777 		ret = i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_AGGR,
4778 		    (void *)(&aggr_cap));
4779 		ASSERT(ret == B_TRUE);
4780 		rename_fn = aggr_cap.mca_rename_fn;
4781 		rename_fn(new_name, mip->mi_driver);
4782 		/*
4783 		 * The aggr's client name and kstat flow name will be
4784 		 * updated below, i.e. via mac_rename_flow_names.
4785 		 */
4786 	}
4787 
4788 	for (cur_clnt = mip->mi_clients_list; cur_clnt != NULL;
4789 	    cur_clnt = cur_clnt->mci_client_next) {
4790 		if (cur_clnt->mci_state_flags & MCIS_IS_AGGR_PORT) {
4791 			if (new_name != NULL) {
4792 				char *str_st = cur_clnt->mci_name;
4793 				char *str_del = strchr(str_st, '-');
4794 
4795 				ASSERT(str_del != NULL);
4796 				bzero(str_del + 1, MAXNAMELEN -
4797 				    (str_del - str_st + 1));
4798 				bcopy(new_name, str_del + 1,
4799 				    strlen(new_name));
4800 			}
4801 			fep = cur_clnt->mci_flent;
4802 			mac_rename_flow(fep, cur_clnt->mci_name);
4803 			break;
4804 		} else if (new_name != NULL &&
4805 		    cur_clnt->mci_state_flags & MCIS_USE_DATALINK_NAME) {
4806 			mac_rename_flow_names(cur_clnt, new_name);
4807 			break;
4808 		}
4809 	}
4810 
4811 	/* Recreate kstats associated with aggr pseudo rings */
4812 	if (mip->mi_state_flags & MIS_IS_AGGR)
4813 		mac_pseudo_ring_stat_rename(mip);
4814 
4815 done:
4816 	i_mac_perim_exit(mip);
4817 	return (0);
4818 }
4819 
4820 /*
4821  * Rename the MAC client's flow names
4822  */
4823 static void
4824 mac_rename_flow_names(mac_client_impl_t *mcip, const char *new_name)
4825 {
4826 	flow_entry_t	*flent;
4827 	uint16_t	vid;
4828 	char		flowname[MAXFLOWNAMELEN];
4829 	mac_impl_t	*mip = mcip->mci_mip;
4830 
4831 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
4832 
4833 	/*
4834 	 * Use mi_rw_lock to ensure that threads not in the mac perimeter
4835 	 * see a self-consistent value for mci_name
4836 	 */
4837 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
4838 	(void) strlcpy(mcip->mci_name, new_name, sizeof (mcip->mci_name));
4839 	rw_exit(&mip->mi_rw_lock);
4840 
4841 	mac_rename_flow(mcip->mci_flent, new_name);
4842 
4843 	if (mcip->mci_nflents == 1)
4844 		return;
4845 
4846 	/*
4847 	 * We have to rename all the others too, no stats to destroy for
4848 	 * these.
4849 	 */
4850 	for (flent = mcip->mci_flent_list; flent != NULL;
4851 	    flent = flent->fe_client_next) {
4852 		if (flent != mcip->mci_flent) {
4853 			vid = i_mac_flow_vid(flent);
4854 			(void) sprintf(flowname, "%s%u", new_name, vid);
4855 			mac_flow_set_name(flent, flowname);
4856 		}
4857 	}
4858 }
4859 
4860 
4861 /*
4862  * Add a flow to the MAC client's flow list - i.e list of MAC/VID tuples
4863  * defined for the specified MAC client.
4864  */
4865 static void
4866 mac_client_add_to_flow_list(mac_client_impl_t *mcip, flow_entry_t *flent)
4867 {
4868 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
4869 	/*
4870 	 * The promisc Rx data path walks the mci_flent_list. Protect by
4871 	 * using mi_rw_lock
4872 	 */
4873 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
4874 
4875 	mcip->mci_vidcache = MCIP_VIDCACHE_INVALID;
4876 
4877 	/* Add it to the head */
4878 	flent->fe_client_next = mcip->mci_flent_list;
4879 	mcip->mci_flent_list = flent;
4880 	mcip->mci_nflents++;
4881 
4882 	/*
4883 	 * Keep track of the number of non-zero VIDs addresses per MAC
4884 	 * client to avoid figuring it out in the data-path.
4885 	 */
4886 	if (i_mac_flow_vid(flent) != VLAN_ID_NONE)
4887 		mcip->mci_nvids++;
4888 
4889 	rw_exit(&mcip->mci_rw_lock);
4890 }
4891 
4892 /*
4893  * Remove a flow entry from the MAC client's list.
4894  */
4895 static void
4896 mac_client_remove_flow_from_list(mac_client_impl_t *mcip, flow_entry_t *flent)
4897 {
4898 	flow_entry_t	*fe = mcip->mci_flent_list;
4899 	flow_entry_t	*prev_fe = NULL;
4900 
4901 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
4902 	/*
4903 	 * The promisc Rx data path walks the mci_flent_list. Protect by
4904 	 * using mci_rw_lock
4905 	 */
4906 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
4907 	mcip->mci_vidcache = MCIP_VIDCACHE_INVALID;
4908 
4909 	while ((fe != NULL) && (fe != flent)) {
4910 		prev_fe = fe;
4911 		fe = fe->fe_client_next;
4912 	}
4913 
4914 	ASSERT(fe != NULL);
4915 	if (prev_fe == NULL) {
4916 		/* Deleting the first node */
4917 		mcip->mci_flent_list = fe->fe_client_next;
4918 	} else {
4919 		prev_fe->fe_client_next = fe->fe_client_next;
4920 	}
4921 	mcip->mci_nflents--;
4922 
4923 	if (i_mac_flow_vid(flent) != VLAN_ID_NONE)
4924 		mcip->mci_nvids--;
4925 
4926 	rw_exit(&mcip->mci_rw_lock);
4927 }
4928 
4929 /*
4930  * Check if the given VID belongs to this MAC client.
4931  */
4932 boolean_t
4933 mac_client_check_flow_vid(mac_client_impl_t *mcip, uint16_t vid)
4934 {
4935 	flow_entry_t	*flent;
4936 	uint16_t	mci_vid;
4937 	uint32_t	cache = mcip->mci_vidcache;
4938 
4939 	/*
4940 	 * In hopes of not having to touch the mci_rw_lock, check to see if
4941 	 * this vid matches our cached result.
4942 	 */
4943 	if (MCIP_VIDCACHE_ISVALID(cache) && MCIP_VIDCACHE_VID(cache) == vid)
4944 		return (MCIP_VIDCACHE_BOOL(cache) ? B_TRUE : B_FALSE);
4945 
4946 	/* The mci_flent_list is protected by mci_rw_lock */
4947 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
4948 	for (flent = mcip->mci_flent_list; flent != NULL;
4949 	    flent = flent->fe_client_next) {
4950 		mci_vid = i_mac_flow_vid(flent);
4951 		if (vid == mci_vid) {
4952 			mcip->mci_vidcache = MCIP_VIDCACHE_CACHE(vid, B_TRUE);
4953 			rw_exit(&mcip->mci_rw_lock);
4954 			return (B_TRUE);
4955 		}
4956 	}
4957 
4958 	mcip->mci_vidcache = MCIP_VIDCACHE_CACHE(vid, B_FALSE);
4959 	rw_exit(&mcip->mci_rw_lock);
4960 	return (B_FALSE);
4961 }
4962 
4963 /*
4964  * Get the flow entry for the specified <MAC addr, VID> tuple.
4965  */
4966 static flow_entry_t *
4967 mac_client_get_flow(mac_client_impl_t *mcip, mac_unicast_impl_t *muip)
4968 {
4969 	mac_address_t *map = mcip->mci_unicast;
4970 	flow_entry_t *flent;
4971 	uint16_t vid;
4972 	flow_desc_t flow_desc;
4973 
4974 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
4975 
4976 	mac_flow_get_desc(mcip->mci_flent, &flow_desc);
4977 	if (bcmp(flow_desc.fd_dst_mac, map->ma_addr, map->ma_len) != 0)
4978 		return (NULL);
4979 
4980 	for (flent = mcip->mci_flent_list; flent != NULL;
4981 	    flent = flent->fe_client_next) {
4982 		vid = i_mac_flow_vid(flent);
4983 		if (vid == muip->mui_vid) {
4984 			return (flent);
4985 		}
4986 	}
4987 
4988 	return (NULL);
4989 }
4990 
4991 /*
4992  * Since mci_flent has the SRSs, when we want to remove it, we replace
4993  * the flow_desc_t in mci_flent with that of an existing flent and then
4994  * remove that flent instead of mci_flent.
4995  */
4996 static flow_entry_t *
4997 mac_client_swap_mciflent(mac_client_impl_t *mcip)
4998 {
4999 	flow_entry_t	*flent = mcip->mci_flent;
5000 	flow_tab_t	*ft = flent->fe_flow_tab;
5001 	flow_entry_t	*flent1;
5002 	flow_desc_t	fl_desc;
5003 	char		fl_name[MAXFLOWNAMELEN];
5004 	int		err;
5005 
5006 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
5007 	ASSERT(mcip->mci_nflents > 1);
5008 
5009 	/* get the next flent following the primary flent  */
5010 	flent1 = mcip->mci_flent_list->fe_client_next;
5011 	ASSERT(flent1 != NULL && flent1->fe_flow_tab == ft);
5012 
5013 	/*
5014 	 * Remove the flent from the flow table before updating the
5015 	 * flow descriptor as the hash depends on the flow descriptor.
5016 	 * This also helps incoming packet classification avoid having
5017 	 * to grab fe_lock. Access to fe_flow_desc of a flent not in the
5018 	 * flow table is done under the fe_lock so that log or stat functions
5019 	 * see a self-consistent fe_flow_desc. The name and desc are specific
5020 	 * to a flow, the rest are shared by all the clients, including
5021 	 * resource control etc.
5022 	 */
5023 	mac_flow_remove(ft, flent, B_TRUE);
5024 	mac_flow_remove(ft, flent1, B_TRUE);
5025 
5026 	bcopy(&flent->fe_flow_desc, &fl_desc, sizeof (flow_desc_t));
5027 	bcopy(flent->fe_flow_name, fl_name, MAXFLOWNAMELEN);
5028 
5029 	/* update the primary flow entry */
5030 	mutex_enter(&flent->fe_lock);
5031 	bcopy(&flent1->fe_flow_desc, &flent->fe_flow_desc,
5032 	    sizeof (flow_desc_t));
5033 	bcopy(&flent1->fe_flow_name, &flent->fe_flow_name, MAXFLOWNAMELEN);
5034 	mutex_exit(&flent->fe_lock);
5035 
5036 	/* update the flow entry that is to be freed */
5037 	mutex_enter(&flent1->fe_lock);
5038 	bcopy(&fl_desc, &flent1->fe_flow_desc, sizeof (flow_desc_t));
5039 	bcopy(fl_name, &flent1->fe_flow_name, MAXFLOWNAMELEN);
5040 	mutex_exit(&flent1->fe_lock);
5041 
5042 	/* now reinsert the flow entries in the table */
5043 	err = mac_flow_add(ft, flent);
5044 	ASSERT(err == 0);
5045 
5046 	err = mac_flow_add(ft, flent1);
5047 	ASSERT(err == 0);
5048 
5049 	return (flent1);
5050 }
5051 
5052 /*
5053  * Return whether there is only one flow entry associated with this
5054  * MAC client.
5055  */
5056 static boolean_t
5057 mac_client_single_rcvr(mac_client_impl_t *mcip)
5058 {
5059 	return (mcip->mci_nflents == 1);
5060 }
5061 
5062 int
5063 mac_validate_props(mac_impl_t *mip, mac_resource_props_t *mrp)
5064 {
5065 	boolean_t		reset;
5066 	uint32_t		rings_needed;
5067 	uint32_t		rings_avail;
5068 	mac_group_type_t	gtype;
5069 	mac_resource_props_t	*mip_mrp;
5070 
5071 	if (mrp == NULL)
5072 		return (0);
5073 
5074 	if (mrp->mrp_mask & MRP_PRIORITY) {
5075 		mac_priority_level_t	pri = mrp->mrp_priority;
5076 
5077 		if (pri < MPL_LOW || pri > MPL_RESET)
5078 			return (EINVAL);
5079 	}
5080 
5081 	if (mrp->mrp_mask & MRP_MAXBW) {
5082 		uint64_t maxbw = mrp->mrp_maxbw;
5083 
5084 		if (maxbw < MRP_MAXBW_MINVAL && maxbw != 0)
5085 			return (EINVAL);
5086 	}
5087 	if (mrp->mrp_mask & MRP_CPUS) {
5088 		int i, j;
5089 		mac_cpu_mode_t	fanout;
5090 
5091 		if (mrp->mrp_ncpus > ncpus)
5092 			return (EINVAL);
5093 
5094 		for (i = 0; i < mrp->mrp_ncpus; i++) {
5095 			for (j = 0; j < mrp->mrp_ncpus; j++) {
5096 				if (i != j &&
5097 				    mrp->mrp_cpu[i] == mrp->mrp_cpu[j]) {
5098 					return (EINVAL);
5099 				}
5100 			}
5101 		}
5102 
5103 		for (i = 0; i < mrp->mrp_ncpus; i++) {
5104 			cpu_t *cp;
5105 			int rv;
5106 
5107 			mutex_enter(&cpu_lock);
5108 			cp = cpu_get(mrp->mrp_cpu[i]);
5109 			if (cp != NULL)
5110 				rv = cpu_is_online(cp);
5111 			else
5112 				rv = 0;
5113 			mutex_exit(&cpu_lock);
5114 			if (rv == 0)
5115 				return (EINVAL);
5116 		}
5117 
5118 		fanout = mrp->mrp_fanout_mode;
5119 		if (fanout < 0 || fanout > MCM_CPUS)
5120 			return (EINVAL);
5121 	}
5122 
5123 	if (mrp->mrp_mask & MRP_PROTECT) {
5124 		int err = mac_protect_validate(mrp);
5125 		if (err != 0)
5126 			return (err);
5127 	}
5128 
5129 	if (!(mrp->mrp_mask & MRP_RX_RINGS) &&
5130 	    !(mrp->mrp_mask & MRP_TX_RINGS)) {
5131 		return (0);
5132 	}
5133 
5134 	/*
5135 	 * mip will be null when we come from mac_flow_create or
5136 	 * mac_link_flow_modify. In the latter case it is a user flow,
5137 	 * for which we don't support rings. In the former we would
5138 	 * have validated the props beforehand (i_mac_unicast_add ->
5139 	 * mac_client_set_resources -> validate for the primary and
5140 	 * vnic_dev_create -> mac_client_set_resources -> validate for
5141 	 * a vnic.
5142 	 */
5143 	if (mip == NULL)
5144 		return (0);
5145 
5146 	/*
5147 	 * We don't support setting rings property for a VNIC that is using a
5148 	 * primary address (VLAN)
5149 	 */
5150 	if ((mip->mi_state_flags & MIS_IS_VNIC) &&
5151 	    mac_is_vnic_primary((mac_handle_t)mip)) {
5152 		return (ENOTSUP);
5153 	}
5154 
5155 	mip_mrp = &mip->mi_resource_props;
5156 	/*
5157 	 * The rings property should be validated against the NICs
5158 	 * resources
5159 	 */
5160 	if (mip->mi_state_flags & MIS_IS_VNIC)
5161 		mip = (mac_impl_t *)mac_get_lower_mac_handle((mac_handle_t)mip);
5162 
5163 	reset = mrp->mrp_mask & MRP_RINGS_RESET;
5164 	/*
5165 	 * If groups are not supported, return error.
5166 	 */
5167 	if (((mrp->mrp_mask & MRP_RX_RINGS) && mip->mi_rx_groups == NULL) ||
5168 	    ((mrp->mrp_mask & MRP_TX_RINGS) && mip->mi_tx_groups == NULL)) {
5169 		return (EINVAL);
5170 	}
5171 	/*
5172 	 * If we are just resetting, there is no validation needed.
5173 	 */
5174 	if (reset)
5175 		return (0);
5176 
5177 	if (mrp->mrp_mask & MRP_RX_RINGS) {
5178 		rings_needed = mrp->mrp_nrxrings;
5179 		/*
5180 		 * We just want to check if the number of additional
5181 		 * rings requested is available.
5182 		 */
5183 		if (mip_mrp->mrp_mask & MRP_RX_RINGS) {
5184 			if (mrp->mrp_nrxrings > mip_mrp->mrp_nrxrings)
5185 				/* Just check for the additional rings */
5186 				rings_needed -= mip_mrp->mrp_nrxrings;
5187 			else
5188 				/* We are not asking for additional rings */
5189 				rings_needed = 0;
5190 		}
5191 		rings_avail = mip->mi_rxrings_avail;
5192 		gtype = mip->mi_rx_group_type;
5193 	} else {
5194 		rings_needed = mrp->mrp_ntxrings;
5195 		/* Similarly for the TX rings */
5196 		if (mip_mrp->mrp_mask & MRP_TX_RINGS) {
5197 			if (mrp->mrp_ntxrings > mip_mrp->mrp_ntxrings)
5198 				/* Just check for the additional rings */
5199 				rings_needed -= mip_mrp->mrp_ntxrings;
5200 			else
5201 				/* We are not asking for additional rings */
5202 				rings_needed = 0;
5203 		}
5204 		rings_avail = mip->mi_txrings_avail;
5205 		gtype = mip->mi_tx_group_type;
5206 	}
5207 
5208 	/* Error if the group is dynamic .. */
5209 	if (gtype == MAC_GROUP_TYPE_DYNAMIC) {
5210 		/*
5211 		 * .. and rings specified are more than available.
5212 		 */
5213 		if (rings_needed > rings_avail)
5214 			return (EINVAL);
5215 	} else {
5216 		/*
5217 		 * OR group is static and we have specified some rings.
5218 		 */
5219 		if (rings_needed > 0)
5220 			return (EINVAL);
5221 	}
5222 	return (0);
5223 }
5224 
5225 /*
5226  * Send a MAC_NOTE_LINK notification to all the MAC clients whenever the
5227  * underlying physical link is down. This is to allow MAC clients to
5228  * communicate with other clients.
5229  */
5230 void
5231 mac_virtual_link_update(mac_impl_t *mip)
5232 {
5233 	if (mip->mi_linkstate != LINK_STATE_UP)
5234 		i_mac_notify(mip, MAC_NOTE_LINK);
5235 }
5236 
5237 /*
5238  * For clients that have a pass-thru MAC, e.g. VNIC, we set the VNIC's
5239  * mac handle in the client.
5240  */
5241 void
5242 mac_set_upper_mac(mac_client_handle_t mch, mac_handle_t mh,
5243     mac_resource_props_t *mrp)
5244 {
5245 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
5246 	mac_impl_t		*mip = (mac_impl_t *)mh;
5247 
5248 	mcip->mci_upper_mip = mip;
5249 	/* If there are any properties, copy it over too */
5250 	if (mrp != NULL) {
5251 		bcopy(mrp, &mip->mi_resource_props,
5252 		    sizeof (mac_resource_props_t));
5253 	}
5254 }
5255 
5256 /*
5257  * Mark the mac as being used exclusively by the single mac client that is
5258  * doing some control operation on this mac. No further opens of this mac
5259  * will be allowed until this client calls mac_unmark_exclusive. The mac
5260  * client calling this function must already be in the mac perimeter
5261  */
5262 int
5263 mac_mark_exclusive(mac_handle_t mh)
5264 {
5265 	mac_impl_t	*mip = (mac_impl_t *)mh;
5266 
5267 	ASSERT(MAC_PERIM_HELD(mh));
5268 	/*
5269 	 * Look up its entry in the global hash table.
5270 	 */
5271 	rw_enter(&i_mac_impl_lock, RW_WRITER);
5272 	if (mip->mi_state_flags & MIS_DISABLED) {
5273 		rw_exit(&i_mac_impl_lock);
5274 		return (ENOENT);
5275 	}
5276 
5277 	/*
5278 	 * A reference to mac is held even if the link is not plumbed.
5279 	 * In i_dls_link_create() we open the MAC interface and hold the
5280 	 * reference. There is an additional reference for the mac_open
5281 	 * done in acquiring the mac perimeter
5282 	 */
5283 	if (mip->mi_ref != 2) {
5284 		rw_exit(&i_mac_impl_lock);
5285 		return (EBUSY);
5286 	}
5287 
5288 	ASSERT(!(mip->mi_state_flags & MIS_EXCLUSIVE_HELD));
5289 	mip->mi_state_flags |= MIS_EXCLUSIVE_HELD;
5290 	rw_exit(&i_mac_impl_lock);
5291 	return (0);
5292 }
5293 
5294 void
5295 mac_unmark_exclusive(mac_handle_t mh)
5296 {
5297 	mac_impl_t	*mip = (mac_impl_t *)mh;
5298 
5299 	ASSERT(MAC_PERIM_HELD(mh));
5300 
5301 	rw_enter(&i_mac_impl_lock, RW_WRITER);
5302 	/* 1 for the creation and another for the perimeter */
5303 	ASSERT(mip->mi_ref == 2 && (mip->mi_state_flags & MIS_EXCLUSIVE_HELD));
5304 	mip->mi_state_flags &= ~MIS_EXCLUSIVE_HELD;
5305 	rw_exit(&i_mac_impl_lock);
5306 }
5307 
5308 /*
5309  * Set the MTU for the specified MAC.
5310  */
5311 int
5312 mac_set_mtu(mac_handle_t mh, uint_t new_mtu, uint_t *old_mtu_arg)
5313 {
5314 	mac_impl_t *mip = (mac_impl_t *)mh;
5315 	uint_t old_mtu;
5316 	int rv = 0;
5317 
5318 	i_mac_perim_enter(mip);
5319 
5320 	if (!(mip->mi_callbacks->mc_callbacks & (MC_SETPROP|MC_GETPROP))) {
5321 		rv = ENOTSUP;
5322 		goto bail;
5323 	}
5324 
5325 	old_mtu = mip->mi_sdu_max;
5326 
5327 	if (new_mtu == 0 || new_mtu < mip->mi_sdu_min) {
5328 		rv = EINVAL;
5329 		goto bail;
5330 	}
5331 
5332 	rw_enter(&mip->mi_rw_lock, RW_READER);
5333 	if (mip->mi_mtrp != NULL && new_mtu < mip->mi_mtrp->mtr_mtu) {
5334 		rv = EBUSY;
5335 		rw_exit(&mip->mi_rw_lock);
5336 		goto bail;
5337 	}
5338 	rw_exit(&mip->mi_rw_lock);
5339 
5340 	if (old_mtu != new_mtu) {
5341 		rv = mip->mi_callbacks->mc_setprop(mip->mi_driver,
5342 		    "mtu", MAC_PROP_MTU, sizeof (uint_t), &new_mtu);
5343 		if (rv != 0)
5344 			goto bail;
5345 		rv = mac_maxsdu_update(mh, new_mtu);
5346 		ASSERT(rv == 0);
5347 	}
5348 
5349 bail:
5350 	i_mac_perim_exit(mip);
5351 
5352 	if (rv == 0 && old_mtu_arg != NULL)
5353 		*old_mtu_arg = old_mtu;
5354 	return (rv);
5355 }
5356 
5357 /*
5358  * Return the RX h/w information for the group indexed by grp_num.
5359  */
5360 void
5361 mac_get_hwrxgrp_info(mac_handle_t mh, int grp_index, uint_t *grp_num,
5362     uint_t *n_rings, uint_t *rings, uint_t *type, uint_t *n_clnts,
5363     char *clnts_name)
5364 {
5365 	mac_impl_t *mip = (mac_impl_t *)mh;
5366 	mac_grp_client_t *mcip;
5367 	uint_t i = 0, index = 0;
5368 	mac_ring_t	*ring;
5369 
5370 	/* Revisit when we implement fully dynamic group allocation */
5371 	ASSERT(grp_index >= 0 && grp_index < mip->mi_rx_group_count);
5372 
5373 	rw_enter(&mip->mi_rw_lock, RW_READER);
5374 	*grp_num = mip->mi_rx_groups[grp_index].mrg_index;
5375 	*type = mip->mi_rx_groups[grp_index].mrg_type;
5376 	*n_rings = mip->mi_rx_groups[grp_index].mrg_cur_count;
5377 	ring = mip->mi_rx_groups[grp_index].mrg_rings;
5378 	for (index = 0; index < mip->mi_rx_groups[grp_index].mrg_cur_count;
5379 	    index++) {
5380 		rings[index] = ring->mr_index;
5381 		ring = ring->mr_next;
5382 	}
5383 	/* Assuming the 1st is the default group */
5384 	index = 0;
5385 	if (grp_index == 0) {
5386 		(void) strlcpy(clnts_name, "<default,mcast>,",
5387 		    MAXCLIENTNAMELEN);
5388 		index += strlen("<default,mcast>,");
5389 	}
5390 	for (mcip = mip->mi_rx_groups[grp_index].mrg_clients; mcip != NULL;
5391 	    mcip = mcip->mgc_next) {
5392 		int name_len = strlen(mcip->mgc_client->mci_name);
5393 
5394 		/*
5395 		 * MAXCLIENTNAMELEN is the buffer size reserved for client
5396 		 * names.
5397 		 * XXXX Formating the client name string needs to be moved
5398 		 * to user land when fixing the size of dhi_clnts in
5399 		 * dld_hwgrpinfo_t. We should use n_clients * client_name for
5400 		 * dhi_clntsin instead of MAXCLIENTNAMELEN
5401 		 */
5402 		if (index + name_len >= MAXCLIENTNAMELEN) {
5403 			index = MAXCLIENTNAMELEN;
5404 			break;
5405 		}
5406 		bcopy(mcip->mgc_client->mci_name, &(clnts_name[index]),
5407 		    name_len);
5408 		index += name_len;
5409 		clnts_name[index++] = ',';
5410 		i++;
5411 	}
5412 
5413 	/* Get rid of the last , */
5414 	if (index > 0)
5415 		clnts_name[index - 1] = '\0';
5416 	*n_clnts = i;
5417 	rw_exit(&mip->mi_rw_lock);
5418 }
5419 
5420 /*
5421  * Return the TX h/w information for the group indexed by grp_num.
5422  */
5423 void
5424 mac_get_hwtxgrp_info(mac_handle_t mh, int grp_index, uint_t *grp_num,
5425     uint_t *n_rings, uint_t *rings, uint_t *type, uint_t *n_clnts,
5426     char *clnts_name)
5427 {
5428 	mac_impl_t *mip = (mac_impl_t *)mh;
5429 	mac_grp_client_t *mcip;
5430 	uint_t i = 0, index = 0;
5431 	mac_ring_t	*ring;
5432 
5433 	/* Revisit when we implement fully dynamic group allocation */
5434 	ASSERT(grp_index >= 0 && grp_index <= mip->mi_tx_group_count);
5435 
5436 	rw_enter(&mip->mi_rw_lock, RW_READER);
5437 	*grp_num = mip->mi_tx_groups[grp_index].mrg_index > 0 ?
5438 	    mip->mi_tx_groups[grp_index].mrg_index : grp_index;
5439 	*type = mip->mi_tx_groups[grp_index].mrg_type;
5440 	*n_rings = mip->mi_tx_groups[grp_index].mrg_cur_count;
5441 	ring = mip->mi_tx_groups[grp_index].mrg_rings;
5442 	for (index = 0; index < mip->mi_tx_groups[grp_index].mrg_cur_count;
5443 	    index++) {
5444 		rings[index] = ring->mr_index;
5445 		ring = ring->mr_next;
5446 	}
5447 	index = 0;
5448 	/* Default group has an index of -1 */
5449 	if (mip->mi_tx_groups[grp_index].mrg_index < 0) {
5450 		(void) strlcpy(clnts_name, "<default>,",
5451 		    MAXCLIENTNAMELEN);
5452 		index += strlen("<default>,");
5453 	}
5454 	for (mcip = mip->mi_tx_groups[grp_index].mrg_clients; mcip != NULL;
5455 	    mcip = mcip->mgc_next) {
5456 		int name_len = strlen(mcip->mgc_client->mci_name);
5457 
5458 		/*
5459 		 * MAXCLIENTNAMELEN is the buffer size reserved for client
5460 		 * names.
5461 		 * XXXX Formating the client name string needs to be moved
5462 		 * to user land when fixing the size of dhi_clnts in
5463 		 * dld_hwgrpinfo_t. We should use n_clients * client_name for
5464 		 * dhi_clntsin instead of MAXCLIENTNAMELEN
5465 		 */
5466 		if (index + name_len >= MAXCLIENTNAMELEN) {
5467 			index = MAXCLIENTNAMELEN;
5468 			break;
5469 		}
5470 		bcopy(mcip->mgc_client->mci_name, &(clnts_name[index]),
5471 		    name_len);
5472 		index += name_len;
5473 		clnts_name[index++] = ',';
5474 		i++;
5475 	}
5476 
5477 	/* Get rid of the last , */
5478 	if (index > 0)
5479 		clnts_name[index - 1] = '\0';
5480 	*n_clnts = i;
5481 	rw_exit(&mip->mi_rw_lock);
5482 }
5483 
5484 /*
5485  * Return the group count for RX or TX.
5486  */
5487 uint_t
5488 mac_hwgrp_num(mac_handle_t mh, int type)
5489 {
5490 	mac_impl_t *mip = (mac_impl_t *)mh;
5491 
5492 	/*
5493 	 * Return the Rx and Tx group count; for the Tx we need to
5494 	 * include the default too.
5495 	 */
5496 	return (type == MAC_RING_TYPE_RX ? mip->mi_rx_group_count :
5497 	    mip->mi_tx_groups != NULL ? mip->mi_tx_group_count + 1 : 0);
5498 }
5499 
5500 /*
5501  * The total number of free TX rings for this MAC.
5502  */
5503 uint_t
5504 mac_txavail_get(mac_handle_t mh)
5505 {
5506 	mac_impl_t	*mip = (mac_impl_t *)mh;
5507 
5508 	return (mip->mi_txrings_avail);
5509 }
5510 
5511 /*
5512  * The total number of free RX rings for this MAC.
5513  */
5514 uint_t
5515 mac_rxavail_get(mac_handle_t mh)
5516 {
5517 	mac_impl_t	*mip = (mac_impl_t *)mh;
5518 
5519 	return (mip->mi_rxrings_avail);
5520 }
5521 
5522 /*
5523  * The total number of reserved RX rings on this MAC.
5524  */
5525 uint_t
5526 mac_rxrsvd_get(mac_handle_t mh)
5527 {
5528 	mac_impl_t	*mip = (mac_impl_t *)mh;
5529 
5530 	return (mip->mi_rxrings_rsvd);
5531 }
5532 
5533 /*
5534  * The total number of reserved TX rings on this MAC.
5535  */
5536 uint_t
5537 mac_txrsvd_get(mac_handle_t mh)
5538 {
5539 	mac_impl_t	*mip = (mac_impl_t *)mh;
5540 
5541 	return (mip->mi_txrings_rsvd);
5542 }
5543 
5544 /*
5545  * Total number of free RX groups on this MAC.
5546  */
5547 uint_t
5548 mac_rxhwlnksavail_get(mac_handle_t mh)
5549 {
5550 	mac_impl_t	*mip = (mac_impl_t *)mh;
5551 
5552 	return (mip->mi_rxhwclnt_avail);
5553 }
5554 
5555 /*
5556  * Total number of RX groups reserved on this MAC.
5557  */
5558 uint_t
5559 mac_rxhwlnksrsvd_get(mac_handle_t mh)
5560 {
5561 	mac_impl_t	*mip = (mac_impl_t *)mh;
5562 
5563 	return (mip->mi_rxhwclnt_used);
5564 }
5565 
5566 /*
5567  * Total number of free TX groups on this MAC.
5568  */
5569 uint_t
5570 mac_txhwlnksavail_get(mac_handle_t mh)
5571 {
5572 	mac_impl_t	*mip = (mac_impl_t *)mh;
5573 
5574 	return (mip->mi_txhwclnt_avail);
5575 }
5576 
5577 /*
5578  * Total number of TX groups reserved on this MAC.
5579  */
5580 uint_t
5581 mac_txhwlnksrsvd_get(mac_handle_t mh)
5582 {
5583 	mac_impl_t	*mip = (mac_impl_t *)mh;
5584 
5585 	return (mip->mi_txhwclnt_used);
5586 }
5587 
5588 /*
5589  * Initialize the rings property for a mac client. A non-0 value for
5590  * rxring or txring specifies the number of rings required, a value
5591  * of MAC_RXRINGS_NONE/MAC_TXRINGS_NONE specifies that it doesn't need
5592  * any RX/TX rings and a value of MAC_RXRINGS_DONTCARE/MAC_TXRINGS_DONTCARE
5593  * means the system can decide whether it can give any rings or not.
5594  */
5595 void
5596 mac_client_set_rings(mac_client_handle_t mch, int rxrings, int txrings)
5597 {
5598 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
5599 	mac_resource_props_t	*mrp = MCIP_RESOURCE_PROPS(mcip);
5600 
5601 	if (rxrings != MAC_RXRINGS_DONTCARE) {
5602 		mrp->mrp_mask |= MRP_RX_RINGS;
5603 		mrp->mrp_nrxrings = rxrings;
5604 	}
5605 
5606 	if (txrings != MAC_TXRINGS_DONTCARE) {
5607 		mrp->mrp_mask |= MRP_TX_RINGS;
5608 		mrp->mrp_ntxrings = txrings;
5609 	}
5610 }
5611 
5612 boolean_t
5613 mac_get_promisc_filtered(mac_client_handle_t mch)
5614 {
5615 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
5616 
5617 	return (mcip->mci_protect_flags & MPT_FLAG_PROMISC_FILTERED);
5618 }
5619 
5620 void
5621 mac_set_promisc_filtered(mac_client_handle_t mch, boolean_t enable)
5622 {
5623 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
5624 
5625 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
5626 	if (enable)
5627 		mcip->mci_protect_flags |= MPT_FLAG_PROMISC_FILTERED;
5628 	else
5629 		mcip->mci_protect_flags &= ~MPT_FLAG_PROMISC_FILTERED;
5630 }
5631