xref: /illumos-gate/usr/src/uts/common/io/mac/mac_client.c (revision 2eef1f2b3c0d57d3f401f917b9f38f01456fd554)
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 2008 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*
28  * - General Introduction:
29  *
30  * This file contains the implementation of the MAC client kernel
31  * API and related code. The MAC client API allows a kernel module
32  * to gain access to a MAC instance (physical NIC, link aggregation, etc).
33  * It allows a MAC client to associate itself with a MAC address,
34  * VLANs, callback functions for data traffic and for promiscuous mode.
35  * The MAC client API is also used to specify the properties associated
36  * with a MAC client, such as bandwidth limits, priority, CPUS, etc.
37  * These properties are further used to determine the hardware resources
38  * to allocate to the various MAC clients.
39  *
40  * - Primary MAC clients:
41  *
42  * The MAC client API refers to "primary MAC clients". A primary MAC
43  * client is a client which "owns" the primary MAC address of
44  * the underlying MAC instance. The primary MAC address is called out
45  * since it is associated with specific semantics: the primary MAC
46  * address is the MAC address which is assigned to the IP interface
47  * when it is plumbed, and the primary MAC address is assigned
48  * to VLAN data-links. The primary address of a MAC instance can
49  * also change dynamically from under the MAC client, for example
50  * as a result of a change of state of a link aggregation. In that
51  * case the MAC layer automatically updates all data-structures which
52  * refer to the current value of the primary MAC address. Typical
53  * primary MAC clients are dls, aggr, and xnb. A typical non-primary
54  * MAC client is the vnic driver.
55  *
56  * - Virtual Switching:
57  *
58  * The MAC layer implements a virtual switch between the MAC clients
59  * (primary and non-primary) defined on top of the same underlying
60  * NIC (physical, link aggregation, etc). The virtual switch is
61  * VLAN-aware, i.e. it allows multiple MAC clients to be member
62  * of one or more VLANs, and the virtual switch will distribute
63  * multicast tagged packets only to the member of the corresponding
64  * VLANs.
65  *
66  * - Upper vs Lower MAC:
67  *
68  * Creating a VNIC on top of a MAC instance effectively causes
69  * two MAC instances to be layered on top of each other, one for
70  * the VNIC(s), one for the underlying MAC instance (physical NIC,
71  * link aggregation, etc). In the code below we refer to the
72  * underlying NIC as the "lower MAC", and we refer to VNICs as
73  * the "upper MAC".
74  *
75  * - Pass-through for VNICs:
76  *
77  * When VNICs are created on top of an underlying MAC, this causes
78  * a layering of two MAC instances. Since the lower MAC already
79  * does the switching and demultiplexing to its MAC clients, the
80  * upper MAC would simply have to pass packets to the layer below
81  * or above it, which would introduce overhead. In order to avoid
82  * this overhead, the MAC layer implements a pass-through mechanism
83  * for VNICs. When a VNIC opens the lower MAC instance, it saves
84  * the MAC client handle it optains from the MAC layer. When a MAC
85  * client opens a VNIC (upper MAC), the MAC layer detects that
86  * the MAC being opened is a VNIC, and gets the MAC client handle
87  * that the VNIC driver obtained from the lower MAC. This exchange
88  * is doing through a private capability between the MAC layer
89  * and the VNIC driver. The upper MAC then returns that handle
90  * directly to its MAC client. Any operation done by the upper
91  * MAC client is now done on the lower MAC client handle, which
92  * allows the VNIC driver to be completely bypassed for the
93  * performance sensitive data-path.
94  *
95  */
96 
97 #include <sys/types.h>
98 #include <sys/conf.h>
99 #include <sys/id_space.h>
100 #include <sys/esunddi.h>
101 #include <sys/stat.h>
102 #include <sys/mkdev.h>
103 #include <sys/stream.h>
104 #include <sys/strsun.h>
105 #include <sys/strsubr.h>
106 #include <sys/dlpi.h>
107 #include <sys/modhash.h>
108 #include <sys/mac_impl.h>
109 #include <sys/mac_client_impl.h>
110 #include <sys/mac_soft_ring.h>
111 #include <sys/dls.h>
112 #include <sys/dld.h>
113 #include <sys/modctl.h>
114 #include <sys/fs/dv_node.h>
115 #include <sys/thread.h>
116 #include <sys/proc.h>
117 #include <sys/callb.h>
118 #include <sys/cpuvar.h>
119 #include <sys/atomic.h>
120 #include <sys/sdt.h>
121 #include <sys/mac_flow.h>
122 #include <sys/ddi_intr_impl.h>
123 #include <sys/disp.h>
124 #include <sys/sdt.h>
125 #include <sys/vnic.h>
126 #include <sys/vnic_impl.h>
127 #include <sys/vlan.h>
128 #include <inet/ip.h>
129 #include <inet/ip6.h>
130 #include <sys/exacct.h>
131 #include <sys/exacct_impl.h>
132 #include <inet/nd.h>
133 #include <sys/ethernet.h>
134 
135 kmem_cache_t	*mac_client_impl_cache;
136 kmem_cache_t	*mac_promisc_impl_cache;
137 
138 static boolean_t mac_client_single_rcvr(mac_client_impl_t *);
139 static flow_entry_t *mac_client_swap_mciflent(mac_client_impl_t *);
140 static flow_entry_t *mac_client_get_flow(mac_client_impl_t *,
141     mac_unicast_impl_t *);
142 static void mac_client_remove_flow_from_list(mac_client_impl_t *,
143     flow_entry_t *);
144 static void mac_client_add_to_flow_list(mac_client_impl_t *, flow_entry_t *);
145 static void mac_rename_flow_names(mac_client_impl_t *, const char *);
146 static void mac_virtual_link_update(mac_impl_t *);
147 
148 /* ARGSUSED */
149 static int
150 i_mac_client_impl_ctor(void *buf, void *arg, int kmflag)
151 {
152 	int	i;
153 	mac_client_impl_t	*mcip = buf;
154 
155 	bzero(buf, MAC_CLIENT_IMPL_SIZE);
156 	mutex_init(&mcip->mci_tx_cb_lock, NULL, MUTEX_DRIVER, NULL);
157 	mcip->mci_tx_notify_cb_info.mcbi_lockp = &mcip->mci_tx_cb_lock;
158 
159 	ASSERT(mac_tx_percpu_cnt >= 0);
160 	for (i = 0; i <= mac_tx_percpu_cnt; i++) {
161 		mutex_init(&mcip->mci_tx_pcpu[i].pcpu_tx_lock, NULL,
162 		    MUTEX_DRIVER, NULL);
163 	}
164 	cv_init(&mcip->mci_tx_cv, NULL, CV_DRIVER, NULL);
165 
166 	return (0);
167 }
168 
169 /* ARGSUSED */
170 static void
171 i_mac_client_impl_dtor(void *buf, void *arg)
172 {
173 	int	i;
174 	mac_client_impl_t *mcip = buf;
175 
176 	ASSERT(mcip->mci_promisc_list == NULL);
177 	ASSERT(mcip->mci_unicast_list == NULL);
178 	ASSERT(mcip->mci_state_flags == 0);
179 	ASSERT(mcip->mci_tx_flag == 0);
180 
181 	mutex_destroy(&mcip->mci_tx_cb_lock);
182 
183 	ASSERT(mac_tx_percpu_cnt >= 0);
184 	for (i = 0; i <= mac_tx_percpu_cnt; i++) {
185 		ASSERT(mcip->mci_tx_pcpu[i].pcpu_tx_refcnt == 0);
186 		mutex_destroy(&mcip->mci_tx_pcpu[i].pcpu_tx_lock);
187 	}
188 	cv_destroy(&mcip->mci_tx_cv);
189 }
190 
191 /* ARGSUSED */
192 static int
193 i_mac_promisc_impl_ctor(void *buf, void *arg, int kmflag)
194 {
195 	mac_promisc_impl_t	*mpip = buf;
196 
197 	bzero(buf, sizeof (mac_promisc_impl_t));
198 	mpip->mpi_mci_link.mcb_objp = buf;
199 	mpip->mpi_mci_link.mcb_objsize = sizeof (mac_promisc_impl_t);
200 	mpip->mpi_mi_link.mcb_objp = buf;
201 	mpip->mpi_mi_link.mcb_objsize = sizeof (mac_promisc_impl_t);
202 	return (0);
203 }
204 
205 /* ARGSUSED */
206 static void
207 i_mac_promisc_impl_dtor(void *buf, void *arg)
208 {
209 	mac_promisc_impl_t	*mpip = buf;
210 
211 	ASSERT(mpip->mpi_mci_link.mcb_objp != NULL);
212 	ASSERT(mpip->mpi_mci_link.mcb_objsize == sizeof (mac_promisc_impl_t));
213 	ASSERT(mpip->mpi_mi_link.mcb_objp == mpip->mpi_mci_link.mcb_objp);
214 	ASSERT(mpip->mpi_mi_link.mcb_objsize == sizeof (mac_promisc_impl_t));
215 
216 	mpip->mpi_mci_link.mcb_objp = NULL;
217 	mpip->mpi_mci_link.mcb_objsize = 0;
218 	mpip->mpi_mi_link.mcb_objp = NULL;
219 	mpip->mpi_mi_link.mcb_objsize = 0;
220 
221 	ASSERT(mpip->mpi_mci_link.mcb_flags == 0);
222 	mpip->mpi_mci_link.mcb_objsize = 0;
223 }
224 
225 void
226 mac_client_init(void)
227 {
228 	ASSERT(mac_tx_percpu_cnt >= 0);
229 
230 	mac_client_impl_cache = kmem_cache_create("mac_client_impl_cache",
231 	    MAC_CLIENT_IMPL_SIZE, 0, i_mac_client_impl_ctor,
232 	    i_mac_client_impl_dtor, NULL, NULL, NULL, 0);
233 	ASSERT(mac_client_impl_cache != NULL);
234 
235 	mac_promisc_impl_cache = kmem_cache_create("mac_promisc_impl_cache",
236 	    sizeof (mac_promisc_impl_t), 0, i_mac_promisc_impl_ctor,
237 	    i_mac_promisc_impl_dtor, NULL, NULL, NULL, 0);
238 	ASSERT(mac_promisc_impl_cache != NULL);
239 }
240 
241 void
242 mac_client_fini(void)
243 {
244 	kmem_cache_destroy(mac_client_impl_cache);
245 	kmem_cache_destroy(mac_promisc_impl_cache);
246 }
247 
248 /*
249  * Return the lower MAC client handle from the VNIC driver for the
250  * specified VNIC MAC instance.
251  */
252 mac_client_impl_t *
253 mac_vnic_lower(mac_impl_t *mip)
254 {
255 	mac_capab_vnic_t cap;
256 	mac_client_impl_t *mcip;
257 
258 	VERIFY(i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_VNIC, &cap));
259 	mcip = cap.mcv_mac_client_handle(cap.mcv_arg);
260 
261 	return (mcip);
262 }
263 
264 /*
265  * Return the MAC client handle of the primary MAC client for the
266  * specified MAC instance, or NULL otherwise.
267  */
268 mac_client_impl_t *
269 mac_primary_client_handle(mac_impl_t *mip)
270 {
271 	mac_client_impl_t *mcip;
272 
273 	if (mip->mi_state_flags & MIS_IS_VNIC)
274 		return (mac_vnic_lower(mip));
275 
276 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
277 
278 	for (mcip = mip->mi_clients_list; mcip != NULL;
279 	    mcip = mcip->mci_client_next) {
280 		if (MCIP_DATAPATH_SETUP(mcip) && mac_is_primary_client(mcip))
281 			return (mcip);
282 	}
283 	return (NULL);
284 }
285 
286 /*
287  * Open a MAC specified by its MAC name.
288  */
289 int
290 mac_open(const char *macname, mac_handle_t *mhp)
291 {
292 	mac_impl_t	*mip;
293 	int		err;
294 
295 	/*
296 	 * Look up its entry in the global hash table.
297 	 */
298 	if ((err = mac_hold(macname, &mip)) != 0)
299 		return (err);
300 
301 	/*
302 	 * Hold the dip associated to the MAC to prevent it from being
303 	 * detached. For a softmac, its underlying dip is held by the
304 	 * mi_open() callback.
305 	 *
306 	 * This is done to be more tolerant with some defective drivers,
307 	 * which incorrectly handle mac_unregister() failure in their
308 	 * xxx_detach() routine. For example, some drivers ignore the
309 	 * failure of mac_unregister() and free all resources that
310 	 * that are needed for data transmition.
311 	 */
312 	e_ddi_hold_devi(mip->mi_dip);
313 
314 	if (!(mip->mi_callbacks->mc_callbacks & MC_OPEN)) {
315 		*mhp = (mac_handle_t)mip;
316 		return (0);
317 	}
318 
319 	/*
320 	 * The mac perimeter is used in both mac_open and mac_close by the
321 	 * framework to single thread the MC_OPEN/MC_CLOSE of drivers.
322 	 */
323 	i_mac_perim_enter(mip);
324 	mip->mi_oref++;
325 	if (mip->mi_oref != 1 || ((err = mip->mi_open(mip->mi_driver)) == 0)) {
326 		*mhp = (mac_handle_t)mip;
327 		i_mac_perim_exit(mip);
328 		return (0);
329 	}
330 	mip->mi_oref--;
331 	ddi_release_devi(mip->mi_dip);
332 	mac_rele(mip);
333 	i_mac_perim_exit(mip);
334 	return (err);
335 }
336 
337 /*
338  * Open a MAC specified by its linkid.
339  */
340 int
341 mac_open_by_linkid(datalink_id_t linkid, mac_handle_t *mhp)
342 {
343 	dls_dl_handle_t	dlh;
344 	int		err;
345 
346 	if ((err = dls_devnet_hold_tmp(linkid, &dlh)) != 0)
347 		return (err);
348 
349 	dls_devnet_prop_task_wait(dlh);
350 
351 	err = mac_open(dls_devnet_mac(dlh), mhp);
352 
353 	dls_devnet_rele_tmp(dlh);
354 	return (err);
355 }
356 
357 /*
358  * Open a MAC specified by its link name.
359  */
360 int
361 mac_open_by_linkname(const char *link, mac_handle_t *mhp)
362 {
363 	datalink_id_t	linkid;
364 	int		err;
365 
366 	if ((err = dls_mgmt_get_linkid(link, &linkid)) != 0)
367 		return (err);
368 	return (mac_open_by_linkid(linkid, mhp));
369 }
370 
371 /*
372  * Close the specified MAC.
373  */
374 void
375 mac_close(mac_handle_t mh)
376 {
377 	mac_impl_t	*mip = (mac_impl_t *)mh;
378 
379 	i_mac_perim_enter(mip);
380 	/*
381 	 * The mac perimeter is used in both mac_open and mac_close by the
382 	 * framework to single thread the MC_OPEN/MC_CLOSE of drivers.
383 	 */
384 	if (mip->mi_callbacks->mc_callbacks & MC_OPEN) {
385 		ASSERT(mip->mi_oref != 0);
386 		if (--mip->mi_oref == 0) {
387 			if ((mip->mi_callbacks->mc_callbacks & MC_CLOSE))
388 				mip->mi_close(mip->mi_driver);
389 		}
390 	}
391 	i_mac_perim_exit(mip);
392 	ddi_release_devi(mip->mi_dip);
393 	mac_rele(mip);
394 }
395 
396 /*
397  * Misc utility functions to retrieve various information about a MAC
398  * instance or a MAC client.
399  */
400 
401 const mac_info_t *
402 mac_info(mac_handle_t mh)
403 {
404 	return (&((mac_impl_t *)mh)->mi_info);
405 }
406 
407 dev_info_t *
408 mac_devinfo_get(mac_handle_t mh)
409 {
410 	return (((mac_impl_t *)mh)->mi_dip);
411 }
412 
413 const char *
414 mac_name(mac_handle_t mh)
415 {
416 	return (((mac_impl_t *)mh)->mi_name);
417 }
418 
419 char *
420 mac_client_name(mac_client_handle_t mch)
421 {
422 	return (((mac_client_impl_t *)mch)->mci_name);
423 }
424 
425 minor_t
426 mac_minor(mac_handle_t mh)
427 {
428 	return (((mac_impl_t *)mh)->mi_minor);
429 }
430 
431 /*
432  * Return the VID associated with a MAC client. This function should
433  * be called for clients which are associated with only one VID.
434  */
435 uint16_t
436 mac_client_vid(mac_client_handle_t mch)
437 {
438 	uint16_t		vid = VLAN_ID_NONE;
439 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
440 	flow_desc_t		flow_desc;
441 
442 	if (mcip->mci_nflents == 0)
443 		return (vid);
444 
445 	ASSERT(MCIP_DATAPATH_SETUP(mcip) && mac_client_single_rcvr(mcip));
446 
447 	mac_flow_get_desc(mcip->mci_flent, &flow_desc);
448 	if ((flow_desc.fd_mask & FLOW_LINK_VID) != 0)
449 		vid = flow_desc.fd_vid;
450 
451 	return (vid);
452 }
453 
454 /*
455  * Return the link speed associated with the specified MAC client.
456  *
457  * The link speed of a MAC client is equal to the smallest value of
458  * 1) the current link speed of the underlying NIC, or
459  * 2) the bandwidth limit set for the MAC client.
460  *
461  * Note that the bandwidth limit can be higher than the speed
462  * of the underlying NIC. This is allowed to avoid spurious
463  * administration action failures or artifically lowering the
464  * bandwidth limit of a link that may  have temporarily lowered
465  * its link speed due to hardware problem or administrator action.
466  */
467 static uint64_t
468 mac_client_ifspeed(mac_client_impl_t *mcip)
469 {
470 	mac_impl_t *mip = mcip->mci_mip;
471 	uint64_t nic_speed;
472 
473 	nic_speed = mac_stat_get((mac_handle_t)mip, MAC_STAT_IFSPEED);
474 
475 	if (nic_speed == 0) {
476 		return (0);
477 	} else {
478 		uint64_t policy_limit = (uint64_t)-1;
479 
480 		if (MCIP_RESOURCE_PROPS_MASK(mcip) & MRP_MAXBW)
481 			policy_limit = MCIP_RESOURCE_PROPS_MAXBW(mcip);
482 
483 		return (MIN(policy_limit, nic_speed));
484 	}
485 }
486 
487 /*
488  * Return the link state of the specified client. If here are more
489  * than one clients of the underying mac_impl_t, the link state
490  * will always be UP regardless of the link state of the underlying
491  * mac_impl_t. This is needed to allow the MAC clients to continue
492  * to communicate with each other even when the physical link of
493  * their mac_impl_t is down.
494  */
495 static uint64_t
496 mac_client_link_state(mac_client_impl_t *mcip)
497 {
498 	mac_impl_t *mip = mcip->mci_mip;
499 	uint16_t vid;
500 	mac_client_impl_t *mci_list;
501 	mac_unicast_impl_t *mui_list, *oth_mui_list;
502 
503 	/*
504 	 * Returns LINK_STATE_UP if there are other MAC clients defined on
505 	 * mac_impl_t which share same VLAN ID as that of mcip. Note that
506 	 * if 'mcip' has more than one VID's then we match ANY one of the
507 	 * VID's with other MAC client's VID's and return LINK_STATE_UP.
508 	 */
509 	rw_enter(&mcip->mci_rw_lock, RW_READER);
510 	for (mui_list = mcip->mci_unicast_list; mui_list != NULL;
511 	    mui_list = mui_list->mui_next) {
512 		vid = mui_list->mui_vid;
513 		for (mci_list = mip->mi_clients_list; mci_list != NULL;
514 		    mci_list = mci_list->mci_client_next) {
515 			if (mci_list == mcip)
516 				continue;
517 			for (oth_mui_list = mci_list->mci_unicast_list;
518 			    oth_mui_list != NULL; oth_mui_list = oth_mui_list->
519 			    mui_next) {
520 				if (vid == oth_mui_list->mui_vid) {
521 					rw_exit(&mcip->mci_rw_lock);
522 					return (LINK_STATE_UP);
523 				}
524 			}
525 		}
526 	}
527 	rw_exit(&mcip->mci_rw_lock);
528 
529 	return (mac_stat_get((mac_handle_t)mip, MAC_STAT_LINK_STATE));
530 }
531 
532 /*
533  * Return the statistics of a MAC client. These statistics are different
534  * then the statistics of the underlying MAC which are returned by
535  * mac_stat_get().
536  */
537 uint64_t
538 mac_client_stat_get(mac_client_handle_t mch, uint_t stat)
539 {
540 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
541 	mac_impl_t *mip = mcip->mci_mip;
542 	uint64_t val;
543 
544 	switch (stat) {
545 	case MAC_STAT_LINK_STATE:
546 		val = mac_client_link_state(mcip);
547 		break;
548 	case MAC_STAT_LINK_UP:
549 		val = (mac_client_link_state(mcip) == LINK_STATE_UP);
550 		break;
551 	case MAC_STAT_PROMISC:
552 		val = mac_stat_get((mac_handle_t)mip, MAC_STAT_PROMISC);
553 		break;
554 	case MAC_STAT_IFSPEED:
555 		val = mac_client_ifspeed(mcip);
556 		break;
557 	case MAC_STAT_MULTIRCV:
558 		val = mcip->mci_stat_multircv;
559 		break;
560 	case MAC_STAT_BRDCSTRCV:
561 		val = mcip->mci_stat_brdcstrcv;
562 		break;
563 	case MAC_STAT_MULTIXMT:
564 		val = mcip->mci_stat_multixmt;
565 		break;
566 	case MAC_STAT_BRDCSTXMT:
567 		val = mcip->mci_stat_brdcstxmt;
568 		break;
569 	case MAC_STAT_OBYTES:
570 		val = mcip->mci_stat_obytes;
571 		break;
572 	case MAC_STAT_OPACKETS:
573 		val = mcip->mci_stat_opackets;
574 		break;
575 	case MAC_STAT_OERRORS:
576 		val = mcip->mci_stat_oerrors;
577 		break;
578 	case MAC_STAT_IPACKETS:
579 		val = mcip->mci_stat_ipackets;
580 		break;
581 	case MAC_STAT_RBYTES:
582 		val = mcip->mci_stat_ibytes;
583 		break;
584 	case MAC_STAT_IERRORS:
585 		val = mcip->mci_stat_ierrors;
586 		break;
587 	default:
588 		val = mac_stat_default(mip, stat);
589 		break;
590 	}
591 
592 	return (val);
593 }
594 
595 /*
596  * Return the statistics of the specified MAC instance.
597  */
598 uint64_t
599 mac_stat_get(mac_handle_t mh, uint_t stat)
600 {
601 	mac_impl_t	*mip = (mac_impl_t *)mh;
602 	uint64_t	val;
603 	int		ret;
604 
605 	/*
606 	 * The range of stat determines where it is maintained.  Stat
607 	 * values from 0 up to (but not including) MAC_STAT_MIN are
608 	 * mainteined by the mac module itself.  Everything else is
609 	 * maintained by the driver.
610 	 *
611 	 * If the mac_impl_t being queried corresponds to a VNIC,
612 	 * the stats need to be queried from the lower MAC client
613 	 * corresponding to the VNIC. (The mac_link_update()
614 	 * invoked by the driver to the lower MAC causes the *lower
615 	 * MAC* to update its mi_linkstate, and send a notification
616 	 * to its MAC clients. Due to the VNIC passthrough,
617 	 * these notifications are sent to the upper MAC clients
618 	 * of the VNIC directly, and the upper mac_impl_t of the VNIC
619 	 * does not have a valid mi_linkstate.
620 	 */
621 	if (stat < MAC_STAT_MIN && !(mip->mi_state_flags & MIS_IS_VNIC)) {
622 		/* these stats are maintained by the mac module itself */
623 		switch (stat) {
624 		case MAC_STAT_LINK_STATE:
625 			return (mip->mi_linkstate);
626 		case MAC_STAT_LINK_UP:
627 			return (mip->mi_linkstate == LINK_STATE_UP);
628 		case MAC_STAT_PROMISC:
629 			return (mip->mi_devpromisc != 0);
630 		default:
631 			ASSERT(B_FALSE);
632 		}
633 	}
634 
635 	/*
636 	 * Call the driver to get the given statistic.
637 	 */
638 	ret = mip->mi_getstat(mip->mi_driver, stat, &val);
639 	if (ret != 0) {
640 		/*
641 		 * The driver doesn't support this statistic.  Get the
642 		 * statistic's default value.
643 		 */
644 		val = mac_stat_default(mip, stat);
645 	}
646 	return (val);
647 }
648 
649 /*
650  * Utility function which returns the VID associated with a flow entry.
651  */
652 uint16_t
653 i_mac_flow_vid(flow_entry_t *flent)
654 {
655 	flow_desc_t	flow_desc;
656 
657 	mac_flow_get_desc(flent, &flow_desc);
658 
659 	if ((flow_desc.fd_mask & FLOW_LINK_VID) != 0)
660 		return (flow_desc.fd_vid);
661 	return (VLAN_ID_NONE);
662 }
663 
664 /*
665  * Verify the validity of the specified unicast MAC address. Returns B_TRUE
666  * if the address is valid, B_FALSE otherwise (multicast address, or incorrect
667  * length.
668  */
669 boolean_t
670 mac_unicst_verify(mac_handle_t mh, const uint8_t *addr, uint_t len)
671 {
672 	mac_impl_t	*mip = (mac_impl_t *)mh;
673 
674 	/*
675 	 * Verify the address. No lock is needed since mi_type and plugin
676 	 * details don't change after mac_register().
677 	 */
678 	if ((len != mip->mi_type->mt_addr_length) ||
679 	    (mip->mi_type->mt_ops.mtops_unicst_verify(addr,
680 	    mip->mi_pdata)) != 0) {
681 		return (B_FALSE);
682 	} else {
683 		return (B_TRUE);
684 	}
685 }
686 
687 void
688 mac_sdu_get(mac_handle_t mh, uint_t *min_sdu, uint_t *max_sdu)
689 {
690 	mac_impl_t	*mip = (mac_impl_t *)mh;
691 
692 	if (min_sdu != NULL)
693 		*min_sdu = mip->mi_sdu_min;
694 	if (max_sdu != NULL)
695 		*max_sdu = mip->mi_sdu_max;
696 }
697 
698 /*
699  * Update the MAC unicast address of the specified client's flows. Currently
700  * only one unicast MAC unicast address is allowed per client.
701  */
702 static void
703 mac_unicast_update_client_flow(mac_client_impl_t *mcip)
704 {
705 	mac_impl_t *mip = mcip->mci_mip;
706 	flow_entry_t *flent = mcip->mci_flent;
707 	mac_address_t *map = mcip->mci_unicast;
708 	flow_desc_t flow_desc;
709 
710 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
711 	ASSERT(flent != NULL);
712 
713 	mac_flow_get_desc(flent, &flow_desc);
714 	ASSERT(flow_desc.fd_mask & FLOW_LINK_DST);
715 
716 	bcopy(map->ma_addr, flow_desc.fd_dst_mac, map->ma_len);
717 	mac_flow_set_desc(flent, &flow_desc);
718 
719 	/*
720 	 * A MAC client could have one MAC address but multiple
721 	 * VLANs. In that case update the flow entries corresponding
722 	 * to all VLANs of the MAC client.
723 	 */
724 	for (flent = mcip->mci_flent_list; flent != NULL;
725 	    flent = flent->fe_client_next) {
726 		mac_flow_get_desc(flent, &flow_desc);
727 		if (!(flent->fe_type & FLOW_PRIMARY_MAC ||
728 		    flent->fe_type & FLOW_VNIC_MAC))
729 			continue;
730 
731 		bcopy(map->ma_addr, flow_desc.fd_dst_mac, map->ma_len);
732 		mac_flow_set_desc(flent, &flow_desc);
733 	}
734 }
735 
736 /*
737  * Update all clients that share the same unicast address.
738  */
739 void
740 mac_unicast_update_clients(mac_impl_t *mip, mac_address_t *map)
741 {
742 	mac_client_impl_t *mcip;
743 
744 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
745 
746 	/*
747 	 * Find all clients that share the same unicast MAC address and update
748 	 * them appropriately.
749 	 */
750 	for (mcip = mip->mi_clients_list; mcip != NULL;
751 	    mcip = mcip->mci_client_next) {
752 		/*
753 		 * Ignore clients that don't share this MAC address.
754 		 */
755 		if (map != mcip->mci_unicast)
756 			continue;
757 
758 		/*
759 		 * Update those clients with same old unicast MAC address.
760 		 */
761 		mac_unicast_update_client_flow(mcip);
762 	}
763 }
764 
765 /*
766  * Update the unicast MAC address of the specified VNIC MAC client.
767  *
768  * Check whether the operation is valid. Any of following cases should fail:
769  *
770  * 1. It's a VLAN type of VNIC.
771  * 2. The new value is current "primary" MAC address.
772  * 3. The current MAC address is shared with other clients.
773  * 4. The new MAC address has been used. This case will be valid when
774  *    client migration is fully supported.
775  */
776 int
777 mac_vnic_unicast_set(mac_client_handle_t mch, const uint8_t *addr)
778 {
779 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
780 	mac_impl_t *mip = mcip->mci_mip;
781 	mac_address_t *map = mcip->mci_unicast;
782 	int err;
783 
784 	ASSERT(!(mip->mi_state_flags & MIS_IS_VNIC));
785 	ASSERT(mcip->mci_state_flags & MCIS_IS_VNIC);
786 	ASSERT(mcip->mci_flags != MAC_CLIENT_FLAGS_PRIMARY);
787 
788 	i_mac_perim_enter(mip);
789 
790 	/*
791 	 * If this is a VLAN type of VNIC, it's using "primary" MAC address
792 	 * of the underlying interface. Must fail here. Refer to case 1 above.
793 	 */
794 	if (bcmp(map->ma_addr, mip->mi_addr, map->ma_len) == 0) {
795 		i_mac_perim_exit(mip);
796 		return (ENOTSUP);
797 	}
798 
799 	/*
800 	 * If the new address is the "primary" one, must fail. Refer to
801 	 * case 2 above.
802 	 */
803 	if (bcmp(addr, mip->mi_addr, map->ma_len) == 0) {
804 		i_mac_perim_exit(mip);
805 		return (EACCES);
806 	}
807 
808 	/*
809 	 * If the address is shared by multiple clients, must fail. Refer
810 	 * to case 3 above.
811 	 */
812 	if (mac_check_macaddr_shared(map)) {
813 		i_mac_perim_exit(mip);
814 		return (EBUSY);
815 	}
816 
817 	/*
818 	 * If the new address has been used, must fail for now. Refer to
819 	 * case 4 above.
820 	 */
821 	if (mac_find_macaddr(mip, (uint8_t *)addr) != NULL) {
822 		i_mac_perim_exit(mip);
823 		return (ENOTSUP);
824 	}
825 
826 	/*
827 	 * Update the MAC address.
828 	 */
829 	err = mac_update_macaddr(map, (uint8_t *)addr);
830 
831 	if (err != 0) {
832 		i_mac_perim_exit(mip);
833 		return (err);
834 	}
835 
836 	/*
837 	 * Update all flows of this MAC client.
838 	 */
839 	mac_unicast_update_client_flow(mcip);
840 
841 	i_mac_perim_exit(mip);
842 	return (0);
843 }
844 
845 /*
846  * Program the new primary unicast address of the specified MAC.
847  *
848  * Function mac_update_macaddr() takes care different types of underlying
849  * MAC. If the underlying MAC is VNIC, the VNIC driver must have registerd
850  * mi_unicst() entry point, that indirectly calls mac_vnic_unicast_set()
851  * which will take care of updating the MAC address of the corresponding
852  * MAC client.
853  *
854  * This is the only interface that allow the client to update the "primary"
855  * MAC address of the underlying MAC. The new value must have not been
856  * used by other clients.
857  */
858 int
859 mac_unicast_primary_set(mac_handle_t mh, const uint8_t *addr)
860 {
861 	mac_impl_t *mip = (mac_impl_t *)mh;
862 	mac_address_t *map;
863 	int err;
864 
865 	/* verify the address validity */
866 	if (!mac_unicst_verify(mh, addr, mip->mi_type->mt_addr_length))
867 		return (EINVAL);
868 
869 	i_mac_perim_enter(mip);
870 
871 	/*
872 	 * If the new value is the same as the current primary address value,
873 	 * there's nothing to do.
874 	 */
875 	if (bcmp(addr, mip->mi_addr, mip->mi_type->mt_addr_length) == 0) {
876 		i_mac_perim_exit(mip);
877 		return (0);
878 	}
879 
880 	if (mac_find_macaddr(mip, (uint8_t *)addr) != 0) {
881 		i_mac_perim_exit(mip);
882 		return (EBUSY);
883 	}
884 
885 	map = mac_find_macaddr(mip, mip->mi_addr);
886 	ASSERT(map != NULL);
887 
888 	/*
889 	 * Update the MAC address.
890 	 */
891 	if (mip->mi_state_flags & MIS_IS_AGGR) {
892 		mac_capab_aggr_t aggr_cap;
893 
894 		/*
895 		 * If the mac is an aggregation, other than the unicast
896 		 * addresses programming, aggr must be informed about this
897 		 * primary unicst address change to change its mac address
898 		 * policy to be user-specified.
899 		 */
900 		ASSERT(map->ma_type == MAC_ADDRESS_TYPE_UNICAST_CLASSIFIED);
901 		VERIFY(i_mac_capab_get(mh, MAC_CAPAB_AGGR, &aggr_cap));
902 		err = aggr_cap.mca_unicst(mip->mi_driver, addr);
903 		if (err == 0)
904 			bcopy(addr, map->ma_addr, map->ma_len);
905 	} else {
906 		err = mac_update_macaddr(map, (uint8_t *)addr);
907 	}
908 
909 	if (err != 0) {
910 		i_mac_perim_exit(mip);
911 		return (err);
912 	}
913 
914 	mac_unicast_update_clients(mip, map);
915 
916 	/*
917 	 * Save the new primary MAC address in mac_impl_t.
918 	 */
919 	bcopy(addr, mip->mi_addr, mip->mi_type->mt_addr_length);
920 
921 	i_mac_perim_exit(mip);
922 
923 	if (err == 0)
924 		i_mac_notify(mip, MAC_NOTE_UNICST);
925 
926 	return (err);
927 }
928 
929 /*
930  * Return the current primary MAC address of the specified MAC.
931  */
932 void
933 mac_unicast_primary_get(mac_handle_t mh, uint8_t *addr)
934 {
935 	mac_impl_t *mip = (mac_impl_t *)mh;
936 
937 	rw_enter(&mip->mi_rw_lock, RW_READER);
938 	bcopy(mip->mi_addr, addr, mip->mi_type->mt_addr_length);
939 	rw_exit(&mip->mi_rw_lock);
940 }
941 
942 /*
943  * Return information about the use of the primary MAC address of the
944  * specified MAC instance:
945  *
946  * - if client_name is non-NULL, it must point to a string of at
947  *   least MAXNAMELEN bytes, and will be set to the name of the MAC
948  *   client which uses the primary MAC address.
949  *
950  * - if in_use is non-NULL, used to return whether the primary MAC
951  *   address is currently in use.
952  */
953 void
954 mac_unicast_primary_info(mac_handle_t mh, char *client_name, boolean_t *in_use)
955 {
956 	mac_impl_t *mip = (mac_impl_t *)mh;
957 	mac_client_impl_t *cur_client;
958 
959 	if (in_use != NULL)
960 		*in_use = B_FALSE;
961 	if (client_name != NULL)
962 		bzero(client_name, MAXNAMELEN);
963 
964 	/*
965 	 * The mi_rw_lock is used to protect threads that don't hold the
966 	 * mac perimeter to get a consistent view of the mi_clients_list.
967 	 * Threads that modify the list must hold both the mac perimeter and
968 	 * mi_rw_lock(RW_WRITER)
969 	 */
970 	rw_enter(&mip->mi_rw_lock, RW_READER);
971 	for (cur_client = mip->mi_clients_list; cur_client != NULL;
972 	    cur_client = cur_client->mci_client_next) {
973 		if (mac_is_primary_client(cur_client) ||
974 		    (mip->mi_state_flags & MIS_IS_VNIC)) {
975 			rw_exit(&mip->mi_rw_lock);
976 			if (in_use != NULL)
977 				*in_use = B_TRUE;
978 			if (client_name != NULL) {
979 				bcopy(cur_client->mci_name, client_name,
980 				    MAXNAMELEN);
981 			}
982 			return;
983 		}
984 	}
985 	rw_exit(&mip->mi_rw_lock);
986 }
987 
988 /*
989  * Add the specified MAC client to the list of clients which opened
990  * the specified MAC.
991  */
992 static void
993 mac_client_add(mac_client_impl_t *mcip)
994 {
995 	mac_impl_t *mip = mcip->mci_mip;
996 
997 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
998 
999 	/* add VNIC to the front of the list */
1000 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
1001 	mcip->mci_client_next = mip->mi_clients_list;
1002 	mip->mi_clients_list = mcip;
1003 	mip->mi_nclients++;
1004 	rw_exit(&mip->mi_rw_lock);
1005 }
1006 
1007 /*
1008  * Remove the specified MAC client from the list of clients which opened
1009  * the specified MAC.
1010  */
1011 static void
1012 mac_client_remove(mac_client_impl_t *mcip)
1013 {
1014 	mac_impl_t *mip = mcip->mci_mip;
1015 	mac_client_impl_t **prev, *cclient;
1016 
1017 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1018 
1019 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
1020 	prev = &mip->mi_clients_list;
1021 	cclient = *prev;
1022 	while (cclient != NULL && cclient != mcip) {
1023 		prev = &cclient->mci_client_next;
1024 		cclient = *prev;
1025 	}
1026 	ASSERT(cclient != NULL);
1027 	*prev = cclient->mci_client_next;
1028 	mip->mi_nclients--;
1029 	rw_exit(&mip->mi_rw_lock);
1030 }
1031 
1032 static mac_unicast_impl_t *
1033 mac_client_find_vid(mac_client_impl_t *mcip, uint16_t vid)
1034 {
1035 	mac_unicast_impl_t *muip = mcip->mci_unicast_list;
1036 
1037 	while ((muip != NULL) && (muip->mui_vid != vid))
1038 		muip = muip->mui_next;
1039 
1040 	return (muip);
1041 }
1042 
1043 /*
1044  * Return whether the specified (MAC address, VID) tuple is already used by
1045  * one of the MAC clients associated with the specified MAC.
1046  */
1047 static boolean_t
1048 mac_addr_in_use(mac_impl_t *mip, uint8_t *mac_addr, uint16_t vid)
1049 {
1050 	mac_client_impl_t *client;
1051 	mac_address_t *map;
1052 
1053 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1054 
1055 	for (client = mip->mi_clients_list; client != NULL;
1056 	    client = client->mci_client_next) {
1057 
1058 		/*
1059 		 * Ignore clients that don't have unicast address.
1060 		 */
1061 		if (client->mci_unicast_list == NULL)
1062 			continue;
1063 
1064 		map = client->mci_unicast;
1065 
1066 		if ((bcmp(mac_addr, map->ma_addr, map->ma_len) == 0) &&
1067 		    (mac_client_find_vid(client, vid) != NULL)) {
1068 			return (B_TRUE);
1069 		}
1070 	}
1071 
1072 	return (B_FALSE);
1073 }
1074 
1075 /*
1076  * Generate a random MAC address. The MAC address prefix is
1077  * stored in the array pointed to by mac_addr, and its length, in bytes,
1078  * is specified by prefix_len. The least significant bits
1079  * after prefix_len bytes are generated, and stored after the prefix
1080  * in the mac_addr array.
1081  */
1082 int
1083 mac_addr_random(mac_client_handle_t mch, uint_t prefix_len,
1084     uint8_t *mac_addr, mac_diag_t *diag)
1085 {
1086 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1087 	mac_impl_t *mip = mcip->mci_mip;
1088 	size_t addr_len = mip->mi_type->mt_addr_length;
1089 
1090 	if (prefix_len >= addr_len) {
1091 		*diag = MAC_DIAG_MACPREFIXLEN_INVALID;
1092 		return (EINVAL);
1093 	}
1094 
1095 	/* check the prefix value */
1096 	if (prefix_len > 0) {
1097 		bzero(mac_addr + prefix_len, addr_len - prefix_len);
1098 		if (!mac_unicst_verify((mac_handle_t)mip, mac_addr,
1099 		    addr_len)) {
1100 			*diag = MAC_DIAG_MACPREFIX_INVALID;
1101 			return (EINVAL);
1102 		}
1103 	}
1104 
1105 	/* generate the MAC address */
1106 	if (prefix_len < addr_len) {
1107 		(void) random_get_pseudo_bytes(mac_addr +
1108 		    prefix_len, addr_len - prefix_len);
1109 	}
1110 
1111 	*diag = 0;
1112 	return (0);
1113 }
1114 
1115 /*
1116  * Set the priority range for this MAC client. This will be used to
1117  * determine the absolute priority for the threads created for this
1118  * MAC client using the specified "low", "medium" and "high" level.
1119  * This will also be used for any subflows on this MAC client.
1120  */
1121 #define	MAC_CLIENT_SET_PRIORITY_RANGE(mcip, pri) {			\
1122 	(mcip)->mci_min_pri = FLOW_MIN_PRIORITY(MINCLSYSPRI,	\
1123 	    MAXCLSYSPRI, (pri));					\
1124 	(mcip)->mci_max_pri = FLOW_MAX_PRIORITY(MINCLSYSPRI,	\
1125 	    MAXCLSYSPRI, (mcip)->mci_min_pri);				\
1126 	}
1127 
1128 /*
1129  * MAC client open entry point. Return a new MAC client handle. Each
1130  * MAC client is associated with a name, specified through the 'name'
1131  * argument.
1132  */
1133 int
1134 mac_client_open(mac_handle_t mh, mac_client_handle_t *mchp, char *name,
1135     uint16_t flags)
1136 {
1137 	mac_impl_t *mip = (mac_impl_t *)mh;
1138 	mac_client_impl_t *mcip;
1139 	int err = 0;
1140 	boolean_t share_desired =
1141 	    ((flags & MAC_OPEN_FLAGS_SHARES_DESIRED) != 0);
1142 	boolean_t no_hwrings = ((flags & MAC_OPEN_FLAGS_NO_HWRINGS) != 0);
1143 	boolean_t req_hwrings = ((flags & MAC_OPEN_FLAGS_REQ_HWRINGS) != 0);
1144 	flow_entry_t	*flent = NULL;
1145 
1146 	*mchp = NULL;
1147 	if (share_desired && no_hwrings) {
1148 		/* can't have shares but no hardware rings */
1149 		return (EINVAL);
1150 	}
1151 
1152 	i_mac_perim_enter(mip);
1153 
1154 	if (mip->mi_state_flags & MIS_IS_VNIC) {
1155 		/*
1156 		 * The underlying MAC is a VNIC. Return the MAC client
1157 		 * handle of the lower MAC which was obtained by
1158 		 * the VNIC driver when it did its mac_client_open().
1159 		 */
1160 
1161 		mcip = mac_vnic_lower(mip);
1162 		/*
1163 		 * If there are multiple MAC clients of the VNIC, they
1164 		 * all share the same underlying MAC client handle.
1165 		 */
1166 		if ((flags & MAC_OPEN_FLAGS_TAG_DISABLE) != 0)
1167 			mcip->mci_state_flags |= MCIS_TAG_DISABLE;
1168 
1169 		if ((flags & MAC_OPEN_FLAGS_STRIP_DISABLE) != 0)
1170 			mcip->mci_state_flags |= MCIS_STRIP_DISABLE;
1171 
1172 		if ((flags & MAC_OPEN_FLAGS_DISABLE_TX_VID_CHECK) != 0)
1173 			mcip->mci_state_flags |= MCIS_DISABLE_TX_VID_CHECK;
1174 
1175 		/*
1176 		 * Note that multiple mac clients share the same mcip in
1177 		 * this case.
1178 		 */
1179 		if (flags & MAC_OPEN_FLAGS_EXCLUSIVE)
1180 			mcip->mci_state_flags |= MCIS_EXCLUSIVE;
1181 
1182 		mip->mi_clients_list = mcip;
1183 		i_mac_perim_exit(mip);
1184 		*mchp = (mac_client_handle_t)mcip;
1185 		return (err);
1186 	}
1187 
1188 	mcip = kmem_cache_alloc(mac_client_impl_cache, KM_SLEEP);
1189 
1190 	mcip->mci_mip = mip;
1191 	mcip->mci_upper_mip = NULL;
1192 	mcip->mci_rx_fn = mac_pkt_drop;
1193 	mcip->mci_rx_arg = NULL;
1194 	mcip->mci_direct_rx_fn = NULL;
1195 	mcip->mci_direct_rx_arg = NULL;
1196 
1197 	if ((flags & MAC_OPEN_FLAGS_IS_VNIC) != 0)
1198 		mcip->mci_state_flags |= MCIS_IS_VNIC;
1199 
1200 	if ((flags & MAC_OPEN_FLAGS_EXCLUSIVE) != 0)
1201 		mcip->mci_state_flags |= MCIS_EXCLUSIVE;
1202 
1203 	if ((flags & MAC_OPEN_FLAGS_IS_AGGR_PORT) != 0)
1204 		mcip->mci_state_flags |= MCIS_IS_AGGR_PORT;
1205 
1206 	if ((flags & MAC_OPEN_FLAGS_TAG_DISABLE) != 0)
1207 		mcip->mci_state_flags |= MCIS_TAG_DISABLE;
1208 
1209 	if ((flags & MAC_OPEN_FLAGS_STRIP_DISABLE) != 0)
1210 		mcip->mci_state_flags |= MCIS_STRIP_DISABLE;
1211 
1212 	if ((flags & MAC_OPEN_FLAGS_DISABLE_TX_VID_CHECK) != 0)
1213 		mcip->mci_state_flags |= MCIS_DISABLE_TX_VID_CHECK;
1214 
1215 	if ((flags & MAC_OPEN_FLAGS_USE_DATALINK_NAME) != 0) {
1216 		datalink_id_t	linkid;
1217 
1218 		ASSERT(name == NULL);
1219 		if ((err = dls_devnet_macname2linkid(mip->mi_name,
1220 		    &linkid)) != 0) {
1221 			goto done;
1222 		}
1223 		if ((err = dls_mgmt_get_linkinfo(linkid, mcip->mci_name, NULL,
1224 		    NULL, NULL)) != 0) {
1225 			/*
1226 			 * Use mac name if dlmgmtd is not available.
1227 			 */
1228 			if (err == EBADF) {
1229 				(void) strlcpy(mcip->mci_name, mip->mi_name,
1230 				    sizeof (mcip->mci_name));
1231 				err = 0;
1232 			} else {
1233 				goto done;
1234 			}
1235 		}
1236 		mcip->mci_state_flags |= MCIS_USE_DATALINK_NAME;
1237 	} else {
1238 		ASSERT(name != NULL);
1239 		if (strlen(name) > MAXNAMELEN) {
1240 			err = EINVAL;
1241 			goto done;
1242 		}
1243 		(void) strlcpy(mcip->mci_name, name, sizeof (mcip->mci_name));
1244 	}
1245 	/* the subflow table will be created dynamically */
1246 	mcip->mci_subflow_tab = NULL;
1247 	mcip->mci_stat_multircv = 0;
1248 	mcip->mci_stat_brdcstrcv = 0;
1249 	mcip->mci_stat_multixmt = 0;
1250 	mcip->mci_stat_brdcstxmt = 0;
1251 
1252 	mcip->mci_stat_obytes = 0;
1253 	mcip->mci_stat_opackets = 0;
1254 	mcip->mci_stat_oerrors = 0;
1255 	mcip->mci_stat_ibytes = 0;
1256 	mcip->mci_stat_ipackets = 0;
1257 	mcip->mci_stat_ierrors = 0;
1258 
1259 	/* Create an initial flow */
1260 
1261 	err = mac_flow_create(NULL, NULL, mcip->mci_name, NULL,
1262 	    mcip->mci_state_flags & MCIS_IS_VNIC ? FLOW_VNIC_MAC :
1263 	    FLOW_PRIMARY_MAC, &flent);
1264 	if (err != 0)
1265 		goto done;
1266 	mcip->mci_flent = flent;
1267 	FLOW_MARK(flent, FE_MC_NO_DATAPATH);
1268 	flent->fe_mcip = mcip;
1269 	/*
1270 	 * Place initial creation reference on the flow. This reference
1271 	 * is released in the corresponding delete action viz.
1272 	 * mac_unicast_remove after waiting for all transient refs to
1273 	 * to go away. The wait happens in mac_flow_wait.
1274 	 */
1275 	FLOW_REFHOLD(flent);
1276 
1277 	/*
1278 	 * Do this ahead of the mac_bcast_add() below so that the mi_nclients
1279 	 * will have the right value for mac_rx_srs_setup().
1280 	 */
1281 	mac_client_add(mcip);
1282 
1283 	if (no_hwrings)
1284 		mcip->mci_state_flags |= MCIS_NO_HWRINGS;
1285 	if (req_hwrings)
1286 		mcip->mci_state_flags |= MCIS_REQ_HWRINGS;
1287 	mcip->mci_share = NULL;
1288 	if (share_desired) {
1289 		ASSERT(!no_hwrings);
1290 		i_mac_share_alloc(mcip);
1291 	}
1292 
1293 	DTRACE_PROBE2(mac__client__open__allocated, mac_impl_t *,
1294 	    mcip->mci_mip, mac_client_impl_t *, mcip);
1295 	*mchp = (mac_client_handle_t)mcip;
1296 
1297 	i_mac_perim_exit(mip);
1298 	return (0);
1299 
1300 done:
1301 	i_mac_perim_exit(mip);
1302 	mcip->mci_state_flags = 0;
1303 	mcip->mci_tx_flag = 0;
1304 	kmem_cache_free(mac_client_impl_cache, mcip);
1305 	return (err);
1306 }
1307 
1308 /*
1309  * Close the specified MAC client handle.
1310  */
1311 void
1312 mac_client_close(mac_client_handle_t mch, uint16_t flags)
1313 {
1314 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
1315 	mac_impl_t		*mip = mcip->mci_mip;
1316 	flow_entry_t		*flent;
1317 
1318 	i_mac_perim_enter(mip);
1319 
1320 	if (flags & MAC_CLOSE_FLAGS_EXCLUSIVE)
1321 		mcip->mci_state_flags &= ~MCIS_EXCLUSIVE;
1322 
1323 	if ((mcip->mci_state_flags & MCIS_IS_VNIC) &&
1324 	    !(flags & MAC_CLOSE_FLAGS_IS_VNIC)) {
1325 		/*
1326 		 * This is an upper VNIC client initiated operation.
1327 		 * The lower MAC client will be closed by the VNIC driver
1328 		 * when the VNIC is deleted.
1329 		 */
1330 
1331 		/*
1332 		 * Clear the flags set when the upper client initiated
1333 		 * open.
1334 		 */
1335 		mcip->mci_state_flags &= ~(MCIS_TAG_DISABLE |
1336 		    MCIS_STRIP_DISABLE | MCIS_DISABLE_TX_VID_CHECK);
1337 
1338 		i_mac_perim_exit(mip);
1339 		return;
1340 	}
1341 
1342 	/*
1343 	 * Remove the flent associated with the MAC client
1344 	 */
1345 	flent = mcip->mci_flent;
1346 	mcip->mci_flent = NULL;
1347 	FLOW_FINAL_REFRELE(flent);
1348 
1349 	/*
1350 	 * MAC clients must remove the unicast addresses and promisc callbacks
1351 	 * they added before issuing a mac_client_close().
1352 	 */
1353 	ASSERT(mcip->mci_unicast_list == NULL);
1354 	ASSERT(mcip->mci_promisc_list == NULL);
1355 	ASSERT(mcip->mci_tx_notify_cb_list == NULL);
1356 
1357 	i_mac_share_free(mcip);
1358 
1359 	mac_client_remove(mcip);
1360 
1361 	i_mac_perim_exit(mip);
1362 	mcip->mci_subflow_tab = NULL;
1363 	mcip->mci_state_flags = 0;
1364 	mcip->mci_tx_flag = 0;
1365 	kmem_cache_free(mac_client_impl_cache, mch);
1366 }
1367 
1368 /*
1369  * Enable bypass for the specified MAC client.
1370  */
1371 boolean_t
1372 mac_rx_bypass_set(mac_client_handle_t mch, mac_direct_rx_t rx_fn, void *arg1)
1373 {
1374 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
1375 	mac_impl_t		*mip = mcip->mci_mip;
1376 
1377 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1378 
1379 	/*
1380 	 * If the mac_client is a VLAN or native media is non ethernet, we
1381 	 * should not do DLS bypass and instead let the packets go via the
1382 	 * default mac_rx_deliver route so vlan header can be stripped etc.
1383 	 */
1384 	if (mcip->mci_nvids > 0 ||
1385 	    mip->mi_info.mi_nativemedia != DL_ETHER)
1386 		return (B_FALSE);
1387 
1388 	/*
1389 	 * These are not accessed directly in the data path, and hence
1390 	 * don't need any protection
1391 	 */
1392 	mcip->mci_direct_rx_fn = rx_fn;
1393 	mcip->mci_direct_rx_arg = arg1;
1394 	mcip->mci_state_flags |= MCIS_CLIENT_POLL_CAPABLE;
1395 	return (B_TRUE);
1396 }
1397 
1398 /*
1399  * Set the receive callback for the specified MAC client. There can be
1400  * at most one such callback per MAC client.
1401  */
1402 void
1403 mac_rx_set(mac_client_handle_t mch, mac_rx_t rx_fn, void *arg)
1404 {
1405 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1406 	mac_impl_t	*mip = mcip->mci_mip;
1407 
1408 	/*
1409 	 * Instead of adding an extra set of locks and refcnts in
1410 	 * the datapath at the mac client boundary, we temporarily quiesce
1411 	 * the SRS and related entities. We then change the receive function
1412 	 * without interference from any receive data thread and then reenable
1413 	 * the data flow subsequently.
1414 	 */
1415 	i_mac_perim_enter(mip);
1416 	mac_rx_client_quiesce(mch);
1417 
1418 	mcip->mci_rx_fn = rx_fn;
1419 	mcip->mci_rx_arg = arg;
1420 	mac_rx_client_restart(mch);
1421 	i_mac_perim_exit(mip);
1422 }
1423 
1424 /*
1425  * Reset the receive callback for the specified MAC client.
1426  */
1427 void
1428 mac_rx_clear(mac_client_handle_t mch)
1429 {
1430 	mac_rx_set(mch, mac_pkt_drop, NULL);
1431 }
1432 
1433 /*
1434  * Walk the MAC client subflow table and updates their priority values.
1435  */
1436 static int
1437 mac_update_subflow_priority_cb(flow_entry_t *flent, void *arg)
1438 {
1439 	mac_flow_update_priority(arg, flent);
1440 	return (0);
1441 }
1442 
1443 void
1444 mac_update_subflow_priority(mac_client_impl_t *mcip)
1445 {
1446 	(void) mac_flow_walk(mcip->mci_subflow_tab,
1447 	    mac_update_subflow_priority_cb, mcip);
1448 }
1449 
1450 /*
1451  * When the MAC client is being brought up (i.e. we do a unicast_add) we need
1452  * to initialize the cpu and resource control structure in the
1453  * mac_client_impl_t from the mac_impl_t (i.e if there are any cached
1454  * properties before the flow entry for the unicast address was created).
1455  */
1456 int
1457 mac_resource_ctl_set(mac_client_handle_t mch, mac_resource_props_t *mrp)
1458 {
1459 	mac_client_impl_t 	*mcip = (mac_client_impl_t *)mch;
1460 	mac_impl_t		*mip = (mac_impl_t *)mcip->mci_mip;
1461 	int			err = 0;
1462 
1463 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1464 
1465 	err = mac_validate_props(mrp);
1466 	if (err != 0)
1467 		return (err);
1468 
1469 	mac_update_resources(mrp, MCIP_RESOURCE_PROPS(mcip), B_FALSE);
1470 	if (MCIP_DATAPATH_SETUP(mcip)) {
1471 		/*
1472 		 * We have to set this prior to calling mac_flow_modify.
1473 		 */
1474 		if (mrp->mrp_mask & MRP_PRIORITY) {
1475 			if (mrp->mrp_priority == MPL_RESET) {
1476 				MAC_CLIENT_SET_PRIORITY_RANGE(mcip,
1477 				    MPL_LINK_DEFAULT);
1478 			} else {
1479 				MAC_CLIENT_SET_PRIORITY_RANGE(mcip,
1480 				    mrp->mrp_priority);
1481 			}
1482 		}
1483 
1484 		mac_flow_modify(mip->mi_flow_tab, mcip->mci_flent, mrp);
1485 		if (mrp->mrp_mask & MRP_PRIORITY)
1486 			mac_update_subflow_priority(mcip);
1487 		return (0);
1488 	}
1489 	return (0);
1490 }
1491 
1492 void
1493 mac_resource_ctl_get(mac_client_handle_t mch, mac_resource_props_t *mrp)
1494 {
1495 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
1496 	mac_resource_props_t	*mcip_mrp = MCIP_RESOURCE_PROPS(mcip);
1497 
1498 	bcopy(mcip_mrp, mrp, sizeof (mac_resource_props_t));
1499 }
1500 
1501 static int
1502 mac_unicast_flow_create(mac_client_impl_t *mcip, uint8_t *mac_addr,
1503     uint16_t vid, boolean_t is_primary, boolean_t first_flow,
1504     flow_entry_t **flent, mac_resource_props_t *mrp)
1505 {
1506 	mac_impl_t	*mip = (mac_impl_t *)mcip->mci_mip;
1507 	flow_desc_t	flow_desc;
1508 	char		flowname[MAXFLOWNAME];
1509 	int		err;
1510 	uint_t		flent_flags;
1511 
1512 	/*
1513 	 * First unicast address being added, create a new flow
1514 	 * for that MAC client.
1515 	 */
1516 	bzero(&flow_desc, sizeof (flow_desc));
1517 
1518 	flow_desc.fd_mac_len = mip->mi_type->mt_addr_length;
1519 	bcopy(mac_addr, flow_desc.fd_dst_mac, flow_desc.fd_mac_len);
1520 	flow_desc.fd_mask = FLOW_LINK_DST;
1521 	if (vid != 0) {
1522 		flow_desc.fd_vid = vid;
1523 		flow_desc.fd_mask |= FLOW_LINK_VID;
1524 	}
1525 
1526 	/*
1527 	 * XXX-nicolas. For now I'm keeping the FLOW_PRIMARY_MAC
1528 	 * and FLOW_VNIC. Even though they're a hack inherited
1529 	 * from the SRS code, we'll keep them for now. They're currently
1530 	 * consumed by mac_datapath_setup() to create the SRS.
1531 	 * That code should be eventually moved out of
1532 	 * mac_datapath_setup() and moved to a mac_srs_create()
1533 	 * function of some sort to keep things clean.
1534 	 *
1535 	 * Also, there's no reason why the SRS for the primary MAC
1536 	 * client should be different than any other MAC client. Until
1537 	 * this is cleaned-up, we support only one MAC unicast address
1538 	 * per client.
1539 	 *
1540 	 * We set FLOW_PRIMARY_MAC for the primary MAC address,
1541 	 * FLOW_VNIC for everything else.
1542 	 */
1543 	if (is_primary)
1544 		flent_flags = FLOW_PRIMARY_MAC;
1545 	else
1546 		flent_flags = FLOW_VNIC_MAC;
1547 
1548 	/*
1549 	 * For the first flow we use the mac client's name - mci_name, for
1550 	 * subsequent ones we just create a name with the vid. This is
1551 	 * so that we can add these flows to the same flow table. This is
1552 	 * fine as the flow name (except for the one with the mac client's
1553 	 * name) is not visible. When the first flow is removed, we just replace
1554 	 * its fdesc with another from the list, so we will still retain the
1555 	 * flent with the MAC client's flow name.
1556 	 */
1557 	if (first_flow) {
1558 		bcopy(mcip->mci_name, flowname, MAXFLOWNAME);
1559 	} else {
1560 		(void) sprintf(flowname, "%s%u", mcip->mci_name, vid);
1561 		flent_flags = FLOW_NO_STATS;
1562 	}
1563 
1564 	if ((err = mac_flow_create(&flow_desc, mrp, flowname, NULL,
1565 	    flent_flags, flent)) != 0)
1566 		return (err);
1567 
1568 	FLOW_MARK(*flent, FE_INCIPIENT);
1569 	(*flent)->fe_mcip = mcip;
1570 
1571 	/*
1572 	 * Place initial creation reference on the flow. This reference
1573 	 * is released in the corresponding delete action viz.
1574 	 * mac_unicast_remove after waiting for all transient refs to
1575 	 * to go away. The wait happens in mac_flow_wait.
1576 	 * We have already held the reference in mac_client_open().
1577 	 */
1578 	if (!first_flow)
1579 		FLOW_REFHOLD(*flent);
1580 	return (0);
1581 }
1582 
1583 /* Refresh the multicast grouping for this VID. */
1584 int
1585 mac_client_update_mcast(void *arg, boolean_t add, const uint8_t *addrp)
1586 {
1587 	flow_entry_t		*flent = arg;
1588 	mac_client_impl_t	*mcip = flent->fe_mcip;
1589 	uint16_t		vid;
1590 	flow_desc_t		flow_desc;
1591 
1592 	mac_flow_get_desc(flent, &flow_desc);
1593 	vid = (flow_desc.fd_mask & FLOW_LINK_VID) != 0 ?
1594 	    flow_desc.fd_vid : VLAN_ID_NONE;
1595 
1596 	/*
1597 	 * We don't call mac_multicast_add()/mac_multicast_remove() as
1598 	 * we want to add/remove for this specific vid.
1599 	 */
1600 	if (add) {
1601 		return (mac_bcast_add(mcip, addrp, vid,
1602 		    MAC_ADDRTYPE_MULTICAST));
1603 	} else {
1604 		mac_bcast_delete(mcip, addrp, vid);
1605 		return (0);
1606 	}
1607 }
1608 
1609 /*
1610  * Add a new unicast address to the MAC client.
1611  *
1612  * The MAC address can be specified either by value, or the MAC client
1613  * can specify that it wants to use the primary MAC address of the
1614  * underlying MAC. See the introductory comments at the beginning
1615  * of this file for more more information on primary MAC addresses.
1616  *
1617  * Note also the tuple (MAC address, VID) must be unique
1618  * for the MAC clients defined on top of the same underlying MAC
1619  * instance, unless the MAC_UNICAST_NODUPCHECK is specified.
1620  */
1621 
1622 int
1623 i_mac_unicast_add(mac_client_handle_t mch, uint8_t *mac_addr, uint16_t flags,
1624     mac_unicast_handle_t *mah, uint16_t vid, mac_diag_t *diag)
1625 {
1626 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1627 	mac_impl_t *mip = mcip->mci_mip;
1628 	mac_unicast_impl_t *muip;
1629 	flow_entry_t *flent;
1630 	int err;
1631 	uint_t mac_len = mip->mi_type->mt_addr_length;
1632 	boolean_t check_dups = !(flags & MAC_UNICAST_NODUPCHECK);
1633 	boolean_t is_primary = (flags & MAC_UNICAST_PRIMARY);
1634 	boolean_t is_vnic_primary = (flags & MAC_UNICAST_VNIC_PRIMARY);
1635 	boolean_t is_unicast_hw = (flags & MAC_UNICAST_HW);
1636 	boolean_t bcast_added = B_FALSE;
1637 	boolean_t nactiveclients_added = B_FALSE;
1638 	boolean_t mac_started = B_FALSE;
1639 	mac_resource_props_t mrp;
1640 
1641 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1642 
1643 	/* when VID is non-zero, the underlying MAC can not be VNIC */
1644 	ASSERT(!((mip->mi_state_flags & MIS_IS_VNIC) && (vid != 0)));
1645 
1646 	/*
1647 	 * Check whether it's the primary client and flag it.
1648 	 */
1649 	if (!(mcip->mci_state_flags & MCIS_IS_VNIC) && is_primary && vid == 0)
1650 		mcip->mci_flags |= MAC_CLIENT_FLAGS_PRIMARY;
1651 
1652 	/*
1653 	 * is_vnic_primary is true when we come here as a VLAN VNIC
1654 	 * which uses the primary mac client's address but with a non-zero
1655 	 * VID. In this case the MAC address is not specified by an upper
1656 	 * MAC client.
1657 	 */
1658 	if ((mcip->mci_state_flags & MCIS_IS_VNIC) && is_primary &&
1659 	    !is_vnic_primary) {
1660 		/*
1661 		 * The address is being set by the upper MAC client
1662 		 * of a VNIC. The MAC address was already set by the
1663 		 * VNIC driver during VNIC creation.
1664 		 *
1665 		 * Note: a VNIC has only one MAC address. We return
1666 		 * the MAC unicast address handle of the lower MAC client
1667 		 * corresponding to the VNIC. We allocate a new entry
1668 		 * which is flagged appropriately, so that mac_unicast_remove()
1669 		 * doesn't attempt to free the original entry that
1670 		 * was allocated by the VNIC driver.
1671 		 */
1672 		ASSERT(mcip->mci_unicast != NULL);
1673 
1674 		/*
1675 		 * Ensure that the primary unicast address of the VNIC
1676 		 * is added only once.
1677 		 */
1678 		if (mcip->mci_flags & MAC_CLIENT_FLAGS_VNIC_PRIMARY)
1679 			return (EBUSY);
1680 
1681 		mcip->mci_flags |= MAC_CLIENT_FLAGS_VNIC_PRIMARY;
1682 
1683 		/*
1684 		 * Create a handle for vid 0.
1685 		 */
1686 		ASSERT(vid == 0);
1687 		muip = kmem_zalloc(sizeof (mac_unicast_impl_t), KM_SLEEP);
1688 		muip->mui_vid = vid;
1689 		*mah = (mac_unicast_handle_t)muip;
1690 		return (0);
1691 	}
1692 
1693 	/* primary MAC clients cannot be opened on top of anchor VNICs */
1694 	if ((is_vnic_primary || is_primary) &&
1695 	    i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_ANCHOR_VNIC, NULL)) {
1696 		return (ENXIO);
1697 	}
1698 
1699 	/*
1700 	 * Return EBUSY if:
1701 	 *  - this is an exclusive active mac client and there already exist
1702 	 *    active mac clients, or
1703 	 *  - there already exist an exclusively active mac client.
1704 	 */
1705 	if ((mcip->mci_state_flags & MCIS_EXCLUSIVE) &&
1706 	    (mip->mi_nactiveclients != 0) || (mip->mi_state_flags &
1707 	    MIS_EXCLUSIVE)) {
1708 		return (EBUSY);
1709 	}
1710 
1711 	if (mcip->mci_state_flags & MCIS_EXCLUSIVE)
1712 		mip->mi_state_flags |= MIS_EXCLUSIVE;
1713 
1714 	bzero(&mrp, sizeof (mac_resource_props_t));
1715 	if (is_primary && !(mcip->mci_state_flags & MCIS_IS_VNIC)) {
1716 		/*
1717 		 * Apply the property cached in the mac_impl_t to the primary
1718 		 * mac client. If the mac client is a VNIC, its property were
1719 		 * already set in the mcip when the VNIC was created.
1720 		 */
1721 		mac_get_resources((mac_handle_t)mip, &mrp);
1722 		(void) mac_client_set_resources(mch, &mrp);
1723 	} else if (mcip->mci_state_flags & MCIS_IS_VNIC) {
1724 		bcopy(MCIP_RESOURCE_PROPS(mcip), &mrp,
1725 		    sizeof (mac_resource_props_t));
1726 	}
1727 
1728 	muip = kmem_zalloc(sizeof (mac_unicast_impl_t), KM_SLEEP);
1729 	muip->mui_vid = vid;
1730 
1731 	if (is_primary || is_vnic_primary) {
1732 		mac_addr = mip->mi_addr;
1733 		check_dups = B_TRUE;
1734 	} else {
1735 
1736 		/*
1737 		 * Verify the validity of the specified MAC addresses value.
1738 		 */
1739 		if (!mac_unicst_verify((mac_handle_t)mip, mac_addr, mac_len)) {
1740 			*diag = MAC_DIAG_MACADDR_INVALID;
1741 			err = EINVAL;
1742 			goto bail;
1743 		}
1744 
1745 		/*
1746 		 * Make sure that the specified MAC address is different
1747 		 * than the unicast MAC address of the underlying NIC.
1748 		 */
1749 		if (check_dups && bcmp(mip->mi_addr, mac_addr, mac_len) == 0) {
1750 			*diag = MAC_DIAG_MACADDR_NIC;
1751 			err = EINVAL;
1752 			goto bail;
1753 		}
1754 	}
1755 
1756 	/*
1757 	 * Make sure the MAC address is not already used by
1758 	 * another MAC client defined on top of the same
1759 	 * underlying NIC.
1760 	 * xxx-venu mac_unicast_add doesnt' seem to be called
1761 	 * with MAC_UNICAST_NODUPCHECK currently, if it does
1762 	 * get called we need to do mac_addr_in_use() just
1763 	 * to check for addr_in_use till 6697876 is fixed.
1764 	 */
1765 	if (check_dups && mac_addr_in_use(mip, mac_addr, vid)) {
1766 		*diag = MAC_DIAG_MACADDR_INUSE;
1767 		err = EEXIST;
1768 		goto bail;
1769 	}
1770 
1771 	if ((err = mac_start(mip)) != 0)
1772 		goto bail;
1773 
1774 	mac_started = B_TRUE;
1775 
1776 	/* add the MAC client to the broadcast address group by default */
1777 	if (mip->mi_type->mt_brdcst_addr != NULL) {
1778 		err = mac_bcast_add(mcip, mip->mi_type->mt_brdcst_addr, vid,
1779 		    MAC_ADDRTYPE_BROADCAST);
1780 		if (err != 0)
1781 			goto bail;
1782 		bcast_added = B_TRUE;
1783 	}
1784 	flent = mcip->mci_flent;
1785 	ASSERT(flent != NULL);
1786 
1787 	/* We are configuring the unicast flow now */
1788 	if (!MCIP_DATAPATH_SETUP(mcip)) {
1789 		if (is_unicast_hw) {
1790 			/*
1791 			 * The client requires a hardware MAC address slot
1792 			 * for that unicast address. Since we support only
1793 			 * one unicast MAC address per client, flag the
1794 			 * MAC client itself.
1795 			 */
1796 			mcip->mci_state_flags |= MCIS_UNICAST_HW;
1797 		}
1798 
1799 		MAC_CLIENT_SET_PRIORITY_RANGE(mcip,
1800 		    (mrp.mrp_mask & MRP_PRIORITY) ? mrp.mrp_priority :
1801 		    MPL_LINK_DEFAULT);
1802 
1803 		if ((err = mac_unicast_flow_create(mcip, mac_addr, vid,
1804 		    is_primary || is_vnic_primary, B_TRUE, &flent, &mrp)) != 0)
1805 			goto bail;
1806 
1807 		mip->mi_nactiveclients++;
1808 		nactiveclients_added = B_TRUE;
1809 		/*
1810 		 * This will allocate the RX ring group if possible for the
1811 		 * flow and program the software classifier as needed.
1812 		 */
1813 		if ((err = mac_datapath_setup(mcip, flent, SRST_LINK)) != 0)
1814 			goto bail;
1815 
1816 		/*
1817 		 * The unicast MAC address must have been added successfully.
1818 		 */
1819 		ASSERT(mcip->mci_unicast != NULL);
1820 	} else {
1821 		mac_address_t *map = mcip->mci_unicast;
1822 
1823 		/*
1824 		 * A unicast flow already exists for that MAC client,
1825 		 * this flow must be the same mac address but with
1826 		 * different VID. It has been checked by mac_addr_in_use().
1827 		 *
1828 		 * We will use the SRS etc. from the mci_flent. Note that
1829 		 * We don't need to create kstat for this as except for
1830 		 * the fdesc, everything will be used from in the 1st flent.
1831 		 */
1832 
1833 		if (bcmp(mac_addr, map->ma_addr, map->ma_len) != 0) {
1834 			err = EINVAL;
1835 			goto bail;
1836 		}
1837 
1838 		/*
1839 		 * Make sure the client is consistent about its requests
1840 		 * for MAC addresses. I.e. all requests from the clients
1841 		 * must have the MAC_UNICAST_HW flag set or clear.
1842 		 */
1843 		if ((mcip->mci_state_flags & MCIS_UNICAST_HW) != 0 &&
1844 		    !is_unicast_hw ||
1845 		    (mcip->mci_state_flags & MCIS_UNICAST_HW) == 0 &&
1846 		    is_unicast_hw) {
1847 			err = EINVAL;
1848 			goto bail;
1849 		}
1850 
1851 		if ((err = mac_unicast_flow_create(mcip, mac_addr, vid,
1852 		    is_primary || is_vnic_primary, B_FALSE, &flent, NULL)) != 0)
1853 			goto bail;
1854 
1855 		if ((err = mac_flow_add(mip->mi_flow_tab, flent)) != 0) {
1856 			FLOW_FINAL_REFRELE(flent);
1857 			goto bail;
1858 		}
1859 
1860 		/* update the multicast group for this vid */
1861 		mac_client_bcast_refresh(mcip, mac_client_update_mcast,
1862 		    (void *)flent, B_TRUE);
1863 
1864 	}
1865 
1866 	/* populate the shared MAC address */
1867 	muip->mui_map = mcip->mci_unicast;
1868 
1869 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
1870 	muip->mui_next = mcip->mci_unicast_list;
1871 	mcip->mci_unicast_list = muip;
1872 	rw_exit(&mcip->mci_rw_lock);
1873 
1874 	*mah = (mac_unicast_handle_t)muip;
1875 
1876 	/* add it to the flow list of this mcip */
1877 	mac_client_add_to_flow_list(mcip, flent);
1878 
1879 	/*
1880 	 * Trigger a renegotiation of the capabilities when the number of
1881 	 * active clients changes from 1 to 2, since some of the capabilities
1882 	 * might have to be disabled. Also send a MAC_NOTE_LINK notification
1883 	 * to all the MAC clients whenever physical link is DOWN.
1884 	 */
1885 	if (mip->mi_nactiveclients == 2) {
1886 		mac_capab_update((mac_handle_t)mip);
1887 		mac_virtual_link_update(mip);
1888 	}
1889 	/*
1890 	 * Now that the setup is complete, clear the INCIPIENT flag.
1891 	 * The flag was set to avoid incoming packets seeing inconsistent
1892 	 * structures while the setup was in progress. Clear the mci_tx_flag
1893 	 * by calling mac_tx_client_block. It is possible that
1894 	 * mac_unicast_remove was called prior to this mac_unicast_add which
1895 	 * could have set the MCI_TX_QUIESCE flag.
1896 	 */
1897 	if (flent->fe_rx_ring_group != NULL)
1898 		mac_rx_group_unmark(flent->fe_rx_ring_group, MR_INCIPIENT);
1899 	FLOW_UNMARK(flent, FE_INCIPIENT);
1900 	FLOW_UNMARK(flent, FE_MC_NO_DATAPATH);
1901 	mac_tx_client_unblock(mcip);
1902 	return (0);
1903 bail:
1904 	if (bcast_added)
1905 		mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr, vid);
1906 	if (mac_started)
1907 		mac_stop(mip);
1908 
1909 	if (nactiveclients_added)
1910 		mip->mi_nactiveclients--;
1911 	if (mcip->mci_state_flags & MCIS_EXCLUSIVE)
1912 		mip->mi_state_flags &= ~MIS_EXCLUSIVE;
1913 	kmem_free(muip, sizeof (mac_unicast_impl_t));
1914 	return (err);
1915 }
1916 
1917 int
1918 mac_unicast_add(mac_client_handle_t mch, uint8_t *mac_addr, uint16_t flags,
1919     mac_unicast_handle_t *mah, uint16_t vid, mac_diag_t *diag)
1920 {
1921 	mac_impl_t *mip = ((mac_client_impl_t *)mch)->mci_mip;
1922 	uint_t err;
1923 
1924 	i_mac_perim_enter(mip);
1925 	err = i_mac_unicast_add(mch, mac_addr, flags, mah, vid, diag);
1926 	i_mac_perim_exit(mip);
1927 
1928 	return (err);
1929 }
1930 
1931 /*
1932  * Add the primary MAC address to the MAC client. This is a convenience
1933  * function which can be called by primary MAC clients which do not
1934  * need to specify any other additional flags.
1935  *
1936  * It's called in one of following situations:
1937  *   * dls as the primary MAC client
1938  *   * aggr as an exclusive client
1939  *   * by VNIC's client
1940  */
1941 int
1942 mac_unicast_primary_add(mac_client_handle_t mch, mac_unicast_handle_t *mah,
1943     mac_diag_t *diag)
1944 {
1945 	return (mac_unicast_add(mch, NULL, MAC_UNICAST_PRIMARY, mah, 0, diag));
1946 }
1947 
1948 /*
1949  * Remove a MAC address which was previously added by mac_unicast_add().
1950  */
1951 int
1952 mac_unicast_remove(mac_client_handle_t mch, mac_unicast_handle_t mah)
1953 {
1954 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1955 	mac_unicast_impl_t *muip = (mac_unicast_impl_t *)mah;
1956 	mac_unicast_impl_t *pre;
1957 	mac_impl_t *mip = mcip->mci_mip;
1958 	flow_entry_t *flent;
1959 
1960 	i_mac_perim_enter(mip);
1961 	if (mcip->mci_flags & MAC_CLIENT_FLAGS_VNIC_PRIMARY) {
1962 		/*
1963 		 * Called made by the upper MAC client of a VNIC.
1964 		 * There's nothing much to do, the unicast address will
1965 		 * be removed by the VNIC driver when the VNIC is deleted,
1966 		 * but let's ensure that all our transmit is done before
1967 		 * the client does a mac_client_stop lest it trigger an
1968 		 * assert in the driver.
1969 		 */
1970 		ASSERT(muip->mui_vid == 0);
1971 
1972 		mac_tx_client_flush(mcip);
1973 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_VNIC_PRIMARY;
1974 
1975 		kmem_free(muip, sizeof (mac_unicast_impl_t));
1976 		i_mac_perim_exit(mip);
1977 		return (0);
1978 	}
1979 
1980 	ASSERT(muip != NULL);
1981 
1982 	/*
1983 	 * Remove the VID from the list of client's VIDs.
1984 	 */
1985 	pre = mcip->mci_unicast_list;
1986 	if (muip == pre)
1987 		mcip->mci_unicast_list = muip->mui_next;
1988 	else {
1989 		while ((pre->mui_next != NULL) && (pre->mui_next != muip))
1990 			pre = pre->mui_next;
1991 		ASSERT(pre->mui_next == muip);
1992 		rw_enter(&mcip->mci_rw_lock, RW_WRITER);
1993 		pre->mui_next = muip->mui_next;
1994 		rw_exit(&mcip->mci_rw_lock);
1995 	}
1996 
1997 	if ((mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY) && muip->mui_vid == 0)
1998 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PRIMARY;
1999 
2000 	/*
2001 	 * This MAC client is shared, so we will just remove the flent
2002 	 * corresponding to the address being removed. We don't invoke
2003 	 * mac_rx_classify_flow_rem() since the additional flow is
2004 	 * not associated with its own separate set of SRS and rings,
2005 	 * and these constructs are still needed for the remaining flows.
2006 	 */
2007 	if (!mac_client_single_rcvr(mcip)) {
2008 		flent = mac_client_get_flow(mcip, muip);
2009 		ASSERT(flent != NULL);
2010 
2011 		/*
2012 		 * The first one is disappearing, need to make sure
2013 		 * we replace it with another from the list of
2014 		 * shared clients.
2015 		 */
2016 		if (flent == mcip->mci_flent)
2017 			flent = mac_client_swap_mciflent(mcip);
2018 		mac_client_remove_flow_from_list(mcip, flent);
2019 		mac_flow_remove(mip->mi_flow_tab, flent, B_FALSE);
2020 		mac_flow_wait(flent, FLOW_DRIVER_UPCALL);
2021 
2022 		/*
2023 		 * The multicast groups that were added by the client so
2024 		 * far must be removed from the brodcast domain corresponding
2025 		 * to the VID being removed.
2026 		 */
2027 		mac_client_bcast_refresh(mcip, mac_client_update_mcast,
2028 		    (void *)flent, B_FALSE);
2029 
2030 		if (mip->mi_type->mt_brdcst_addr != NULL) {
2031 			mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr,
2032 			    muip->mui_vid);
2033 		}
2034 		mac_stop(mip);
2035 		FLOW_FINAL_REFRELE(flent);
2036 		i_mac_perim_exit(mip);
2037 		return (0);
2038 	}
2039 
2040 	mip->mi_nactiveclients--;
2041 
2042 	/* Tear down the Data path */
2043 	mac_datapath_teardown(mcip, mcip->mci_flent, SRST_LINK);
2044 
2045 	/*
2046 	 * Prevent any future access to the flow entry through the mci_flent
2047 	 * pointer by setting the mci_flent to NULL. Access to mci_flent in
2048 	 * mac_bcast_send is also under mi_rw_lock.
2049 	 */
2050 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
2051 	flent = mcip->mci_flent;
2052 	mac_client_remove_flow_from_list(mcip, flent);
2053 
2054 	if (mcip->mci_state_flags & MCIS_DESC_LOGGED)
2055 		mcip->mci_state_flags &= ~MCIS_DESC_LOGGED;
2056 
2057 	/*
2058 	 * This is the last unicast address being removed and there shouldn't
2059 	 * be any outbound data threads at this point coming down from mac
2060 	 * clients. We have waited for the data threads to finish before
2061 	 * starting dld_str_detach. Non-data threads must access TX SRS
2062 	 * under mi_rw_lock.
2063 	 */
2064 	rw_exit(&mip->mi_rw_lock);
2065 
2066 	/*
2067 	 * Update the multicast group for this vid.
2068 	 */
2069 	mac_client_bcast_refresh(mcip, mac_client_update_mcast, (void *)flent,
2070 	    B_FALSE);
2071 
2072 	/*
2073 	 * Don't use FLOW_MARK with FE_MC_NO_DATAPATH, as the flow might
2074 	 * contain other flags, such as FE_CONDEMNED, which we need to
2075 	 * cleared. We don't call mac_flow_cleanup() for this unicast
2076 	 * flow as we have a already cleaned up SRSs etc. (via the teadown
2077 	 * path). We just clear the stats and reset the initial callback
2078 	 * function, the rest will be set when we call mac_flow_create,
2079 	 * if at all.
2080 	 */
2081 	mutex_enter(&flent->fe_lock);
2082 	ASSERT(flent->fe_refcnt == 1 && flent->fe_mbg == NULL &&
2083 	    flent->fe_tx_srs == NULL && flent->fe_rx_srs_cnt == 0);
2084 	flent->fe_flags = FE_MC_NO_DATAPATH;
2085 	flow_stat_destroy(flent);
2086 
2087 	/* Initialize the receiver function to a safe routine */
2088 	flent->fe_cb_fn = (flow_fn_t)mac_pkt_drop;
2089 	flent->fe_cb_arg1 = NULL;
2090 	flent->fe_cb_arg2 = NULL;
2091 
2092 	flent->fe_index = -1;
2093 	mutex_exit(&flent->fe_lock);
2094 
2095 	if (mip->mi_type->mt_brdcst_addr != NULL) {
2096 		mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr,
2097 		    muip->mui_vid);
2098 	}
2099 
2100 	if (mip->mi_nactiveclients == 1) {
2101 		mac_capab_update((mac_handle_t)mip);
2102 		mac_virtual_link_update(mip);
2103 	}
2104 	if (mcip->mci_state_flags & MCIS_EXCLUSIVE)
2105 		mip->mi_state_flags &= ~MIS_EXCLUSIVE;
2106 	mcip->mci_state_flags &= ~MCIS_UNICAST_HW;
2107 
2108 	mac_stop(mip);
2109 
2110 	i_mac_perim_exit(mip);
2111 	kmem_free(muip, sizeof (mac_unicast_impl_t));
2112 	return (0);
2113 }
2114 
2115 /*
2116  * Multicast add function invoked by MAC clients.
2117  */
2118 int
2119 mac_multicast_add(mac_client_handle_t mch, const uint8_t *addr)
2120 {
2121 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2122 	mac_impl_t		*mip = mcip->mci_mip;
2123 	flow_entry_t		*flent = mcip->mci_flent_list;
2124 	flow_entry_t		*prev_fe = NULL;
2125 	uint16_t		vid;
2126 	int			err = 0;
2127 
2128 	/* Verify the address is a valid multicast address */
2129 	if ((err = mip->mi_type->mt_ops.mtops_multicst_verify(addr,
2130 	    mip->mi_pdata)) != 0)
2131 		return (err);
2132 
2133 	i_mac_perim_enter(mip);
2134 	while (flent != NULL) {
2135 		vid = i_mac_flow_vid(flent);
2136 
2137 		err = mac_bcast_add((mac_client_impl_t *)mch, addr, vid,
2138 		    MAC_ADDRTYPE_MULTICAST);
2139 		if (err != 0)
2140 			break;
2141 		prev_fe = flent;
2142 		flent = flent->fe_client_next;
2143 	}
2144 
2145 	/*
2146 	 * If we failed adding, then undo all, rather than partial
2147 	 * success.
2148 	 */
2149 	if (flent != NULL && prev_fe != NULL) {
2150 		flent = mcip->mci_flent_list;
2151 		while (flent != prev_fe->fe_client_next) {
2152 			vid = i_mac_flow_vid(flent);
2153 			mac_bcast_delete((mac_client_impl_t *)mch, addr, vid);
2154 			flent = flent->fe_client_next;
2155 		}
2156 	}
2157 	i_mac_perim_exit(mip);
2158 	return (err);
2159 }
2160 
2161 /*
2162  * Multicast delete function invoked by MAC clients.
2163  */
2164 void
2165 mac_multicast_remove(mac_client_handle_t mch, const uint8_t *addr)
2166 {
2167 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2168 	mac_impl_t		*mip = mcip->mci_mip;
2169 	flow_entry_t		*flent;
2170 	uint16_t		vid;
2171 
2172 	i_mac_perim_enter(mip);
2173 	for (flent = mcip->mci_flent_list; flent != NULL;
2174 	    flent = flent->fe_client_next) {
2175 		vid = i_mac_flow_vid(flent);
2176 		mac_bcast_delete((mac_client_impl_t *)mch, addr, vid);
2177 	}
2178 	i_mac_perim_exit(mip);
2179 }
2180 
2181 /*
2182  * When a MAC client desires to capture packets on an interface,
2183  * it registers a promiscuous call back with mac_promisc_add().
2184  * There are three types of promiscuous callbacks:
2185  *
2186  * * MAC_CLIENT_PROMISC_ALL
2187  *   Captures all packets sent and received by the MAC client,
2188  *   the physical interface, as well as all other MAC clients
2189  *   defined on top of the same MAC.
2190  *
2191  * * MAC_CLIENT_PROMISC_FILTERED
2192  *   Captures all packets sent and received by the MAC client,
2193  *   plus all multicast traffic sent and received by the phyisical
2194  *   interface and the other MAC clients.
2195  *
2196  * * MAC_CLIENT_PROMISC_MULTI
2197  *   Captures all broadcast and multicast packets sent and
2198  *   received by the MAC clients as well as the physical interface.
2199  *
2200  * In all cases, the underlying MAC is put in promiscuous mode.
2201  */
2202 int
2203 mac_promisc_add(mac_client_handle_t mch, mac_client_promisc_type_t type,
2204     mac_rx_t fn, void *arg, mac_promisc_handle_t *mphp, uint16_t flags)
2205 {
2206 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
2207 	mac_impl_t *mip = mcip->mci_mip;
2208 	mac_promisc_impl_t *mpip;
2209 	mac_cb_info_t	*mcbi;
2210 	int rc;
2211 
2212 	i_mac_perim_enter(mip);
2213 
2214 	if ((rc = mac_start(mip)) != 0) {
2215 		i_mac_perim_exit(mip);
2216 		return (rc);
2217 	}
2218 
2219 	if ((mcip->mci_state_flags & MCIS_IS_VNIC) &&
2220 	    type == MAC_CLIENT_PROMISC_ALL) {
2221 		/*
2222 		 * The function is being invoked by the upper MAC client
2223 		 * of a VNIC. The VNIC should only see the traffic
2224 		 * it is entitled to.
2225 		 */
2226 		type = MAC_CLIENT_PROMISC_FILTERED;
2227 	}
2228 
2229 
2230 	/*
2231 	 * Turn on promiscuous mode for the underlying NIC.
2232 	 * This is needed even for filtered callbacks which
2233 	 * expect to receive all multicast traffic on the wire.
2234 	 *
2235 	 * Physical promiscuous mode should not be turned on if
2236 	 * MAC_PROMISC_FLAGS_NO_PHYS is set.
2237 	 */
2238 	if ((flags & MAC_PROMISC_FLAGS_NO_PHYS) == 0) {
2239 		if ((rc = i_mac_promisc_set(mip, B_TRUE, MAC_DEVPROMISC))
2240 		    != 0) {
2241 			mac_stop(mip);
2242 			i_mac_perim_exit(mip);
2243 			return (rc);
2244 		}
2245 	}
2246 
2247 	mpip = kmem_cache_alloc(mac_promisc_impl_cache, KM_SLEEP);
2248 
2249 	mpip->mpi_type = type;
2250 	mpip->mpi_fn = fn;
2251 	mpip->mpi_arg = arg;
2252 	mpip->mpi_mcip = mcip;
2253 	mpip->mpi_no_tx_loop = ((flags & MAC_PROMISC_FLAGS_NO_TX_LOOP) != 0);
2254 	mpip->mpi_no_phys = ((flags & MAC_PROMISC_FLAGS_NO_PHYS) != 0);
2255 
2256 	mcbi = &mip->mi_promisc_cb_info;
2257 	mutex_enter(mcbi->mcbi_lockp);
2258 
2259 	mac_callback_add(&mip->mi_promisc_cb_info, &mcip->mci_promisc_list,
2260 	    &mpip->mpi_mci_link);
2261 	mac_callback_add(&mip->mi_promisc_cb_info, &mip->mi_promisc_list,
2262 	    &mpip->mpi_mi_link);
2263 
2264 	mutex_exit(mcbi->mcbi_lockp);
2265 
2266 	*mphp = (mac_promisc_handle_t)mpip;
2267 	i_mac_perim_exit(mip);
2268 	return (0);
2269 }
2270 
2271 /*
2272  * Remove a multicast address previously aded through mac_promisc_add().
2273  */
2274 int
2275 mac_promisc_remove(mac_promisc_handle_t mph)
2276 {
2277 	mac_promisc_impl_t *mpip = (mac_promisc_impl_t *)mph;
2278 	mac_client_impl_t *mcip = mpip->mpi_mcip;
2279 	mac_impl_t *mip = mcip->mci_mip;
2280 	mac_cb_info_t *mcbi;
2281 	int rc = 0;
2282 
2283 	i_mac_perim_enter(mip);
2284 
2285 	/*
2286 	 * Even if the device can't be reset into normal mode, we still
2287 	 * need to clear the client promisc callbacks. The client may want
2288 	 * to close the mac end point and we can't have stale callbacks.
2289 	 */
2290 	if (!(mpip->mpi_no_phys)) {
2291 		rc = mac_promisc_set((mac_handle_t)mip, B_FALSE,
2292 		    MAC_DEVPROMISC);
2293 		if (rc != 0)
2294 			goto done;
2295 	}
2296 	mcbi = &mip->mi_promisc_cb_info;
2297 	mutex_enter(mcbi->mcbi_lockp);
2298 	if (mac_callback_remove(mcbi, &mip->mi_promisc_list,
2299 	    &mpip->mpi_mi_link)) {
2300 		VERIFY(mac_callback_remove(&mip->mi_promisc_cb_info,
2301 		    &mcip->mci_promisc_list, &mpip->mpi_mci_link));
2302 		kmem_cache_free(mac_promisc_impl_cache, mpip);
2303 	} else {
2304 		mac_callback_remove_wait(&mip->mi_promisc_cb_info);
2305 	}
2306 	mutex_exit(mcbi->mcbi_lockp);
2307 	mac_stop(mip);
2308 
2309 done:
2310 	i_mac_perim_exit(mip);
2311 	return (rc);
2312 }
2313 
2314 /*
2315  * Reference count the number of active Tx threads. MCI_TX_QUIESCE indicates
2316  * that a control operation wants to quiesce the Tx data flow in which case
2317  * we return an error. Holding any of the per cpu locks ensures that the
2318  * mci_tx_flag won't change.
2319  *
2320  * 'CPU' must be accessed just once and used to compute the index into the
2321  * percpu array, and that index must be used for the entire duration of the
2322  * packet send operation. Note that the thread may be preempted and run on
2323  * another cpu any time and so we can't use 'CPU' more than once for the
2324  * operation.
2325  */
2326 #define	MAC_TX_TRY_HOLD(mcip, mytx, error)				\
2327 {									\
2328 	(error) = 0;							\
2329 	(mytx) = &(mcip)->mci_tx_pcpu[CPU->cpu_seqid & mac_tx_percpu_cnt]; \
2330 	mutex_enter(&(mytx)->pcpu_tx_lock);				\
2331 	if (!((mcip)->mci_tx_flag & MCI_TX_QUIESCE)) {			\
2332 		(mytx)->pcpu_tx_refcnt++;				\
2333 	} else {							\
2334 		(error) = -1;						\
2335 	}								\
2336 	mutex_exit(&(mytx)->pcpu_tx_lock);				\
2337 }
2338 
2339 /*
2340  * Release the reference. If needed, signal any control operation waiting
2341  * for Tx quiescence. The wait and signal are always done using the
2342  * mci_tx_pcpu[0]'s lock
2343  */
2344 #define	MAC_TX_RELE(mcip, mytx) {					\
2345 	mutex_enter(&(mytx)->pcpu_tx_lock);				\
2346 	if (--(mytx)->pcpu_tx_refcnt == 0 &&				\
2347 	    (mcip)->mci_tx_flag & MCI_TX_QUIESCE) {			\
2348 		mutex_exit(&(mytx)->pcpu_tx_lock);			\
2349 		mutex_enter(&(mcip)->mci_tx_pcpu[0].pcpu_tx_lock);	\
2350 		cv_signal(&(mcip)->mci_tx_cv);				\
2351 		mutex_exit(&(mcip)->mci_tx_pcpu[0].pcpu_tx_lock);	\
2352 	} else {							\
2353 		mutex_exit(&(mytx)->pcpu_tx_lock);			\
2354 	}								\
2355 }
2356 
2357 /*
2358  * Bump the count of the number of active Tx threads. This is maintained as
2359  * a per CPU counter. On (CMT kind of) machines with large number of CPUs,
2360  * a single mci_tx_lock may become contended. However a count of the total
2361  * number of Tx threads per client is needed in order to quiesce the Tx side
2362  * prior to reassigning a Tx ring dynamically to another client. The thread
2363  * that needs to quiesce the Tx traffic grabs all the percpu locks and checks
2364  * the sum of the individual percpu refcnts. Each Tx data thread only grabs
2365  * its own percpu lock and increments its own refcnt.
2366  */
2367 void *
2368 mac_tx_hold(mac_client_handle_t mch)
2369 {
2370 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
2371 	mac_tx_percpu_t	*mytx;
2372 	int error;
2373 
2374 	MAC_TX_TRY_HOLD(mcip, mytx, error);
2375 	return (error == 0 ? (void *)mytx : NULL);
2376 }
2377 
2378 void
2379 mac_tx_rele(mac_client_handle_t mch, void *mytx_handle)
2380 {
2381 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
2382 	mac_tx_percpu_t	*mytx = mytx_handle;
2383 
2384 	MAC_TX_RELE(mcip, mytx)
2385 }
2386 
2387 /*
2388  * Send function invoked by MAC clients.
2389  */
2390 mac_tx_cookie_t
2391 mac_tx(mac_client_handle_t mch, mblk_t *mp_chain, uintptr_t hint,
2392     uint16_t flag, mblk_t **ret_mp)
2393 {
2394 	mac_tx_cookie_t		cookie;
2395 	int			error;
2396 	mac_tx_percpu_t		*mytx;
2397 	mac_soft_ring_set_t	*srs;
2398 	flow_entry_t		*flent;
2399 	boolean_t		is_subflow = B_FALSE;
2400 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2401 	mac_impl_t		*mip = mcip->mci_mip;
2402 	mac_srs_tx_t		*srs_tx;
2403 
2404 	/*
2405 	 * Check whether the active Tx threads count is bumped already.
2406 	 */
2407 	if (!(flag & MAC_TX_NO_HOLD)) {
2408 		MAC_TX_TRY_HOLD(mcip, mytx, error);
2409 		if (error != 0) {
2410 			freemsgchain(mp_chain);
2411 			return (NULL);
2412 		}
2413 	}
2414 
2415 	if (mcip->mci_subflow_tab != NULL &&
2416 	    mcip->mci_subflow_tab->ft_flow_count > 0 &&
2417 	    mac_flow_lookup(mcip->mci_subflow_tab, mp_chain,
2418 	    FLOW_OUTBOUND, &flent) == 0) {
2419 		/*
2420 		 * The main assumption here is that if in the event
2421 		 * we get a chain, all the packets will be classified
2422 		 * to the same Flow/SRS. If this changes for any
2423 		 * reason, the following logic should change as well.
2424 		 * I suppose the fanout_hint also assumes this .
2425 		 */
2426 		ASSERT(flent != NULL);
2427 		is_subflow = B_TRUE;
2428 	} else {
2429 		flent = mcip->mci_flent;
2430 	}
2431 
2432 	srs = flent->fe_tx_srs;
2433 	srs_tx = &srs->srs_tx;
2434 	if (srs_tx->st_mode == SRS_TX_DEFAULT &&
2435 	    (srs->srs_state & SRS_ENQUEUED) == 0 &&
2436 	    mip->mi_nactiveclients == 1 && mip->mi_promisc_list == NULL &&
2437 	    mp_chain->b_next == NULL) {
2438 		uint64_t	obytes;
2439 
2440 		/*
2441 		 * Since dls always opens the underlying MAC, nclients equals
2442 		 * to 1 means that the only active client is dls itself acting
2443 		 * as a primary client of the MAC instance. Since dls will not
2444 		 * send tagged packets in that case, and dls is trusted to send
2445 		 * packets for its allowed VLAN(s), the VLAN tag insertion and
2446 		 * check is required only if nclients is greater than 1.
2447 		 */
2448 		if (mip->mi_nclients > 1) {
2449 			if (MAC_VID_CHECK_NEEDED(mcip)) {
2450 				int	err = 0;
2451 
2452 				MAC_VID_CHECK(mcip, mp_chain, err);
2453 				if (err != 0) {
2454 					freemsg(mp_chain);
2455 					mcip->mci_stat_oerrors++;
2456 					goto done;
2457 				}
2458 			}
2459 			if (MAC_TAG_NEEDED(mcip)) {
2460 				mp_chain = mac_add_vlan_tag(mp_chain, 0,
2461 				    mac_client_vid(mch));
2462 				if (mp_chain == NULL) {
2463 					mcip->mci_stat_oerrors++;
2464 					goto done;
2465 				}
2466 			}
2467 		}
2468 
2469 		obytes = (mp_chain->b_cont == NULL ? MBLKL(mp_chain) :
2470 		    msgdsize(mp_chain));
2471 
2472 		MAC_TX(mip, srs_tx->st_arg2, mp_chain, mcip);
2473 
2474 		if (mp_chain == NULL) {
2475 			cookie = NULL;
2476 			mcip->mci_stat_obytes += obytes;
2477 			mcip->mci_stat_opackets += 1;
2478 			if ((srs->srs_type & SRST_FLOW) != 0) {
2479 				FLOW_STAT_UPDATE(flent, obytes, obytes);
2480 				FLOW_STAT_UPDATE(flent, opackets, 1);
2481 			}
2482 		} else {
2483 			mutex_enter(&srs->srs_lock);
2484 			cookie = mac_tx_srs_no_desc(srs, mp_chain,
2485 			    flag, ret_mp);
2486 			mutex_exit(&srs->srs_lock);
2487 		}
2488 	} else {
2489 		cookie = srs_tx->st_func(srs, mp_chain, hint, flag, ret_mp);
2490 	}
2491 
2492 done:
2493 	if (is_subflow)
2494 		FLOW_REFRELE(flent);
2495 
2496 	if (!(flag & MAC_TX_NO_HOLD))
2497 		MAC_TX_RELE(mcip, mytx);
2498 
2499 	return (cookie);
2500 }
2501 
2502 /*
2503  * mac_tx_is_blocked
2504  *
2505  * Given a cookie, it returns if the ring identified by the cookie is
2506  * flow-controlled or not (this is not implemented yet). If NULL is
2507  * passed in place of a cookie, then it finds out if any of the
2508  * underlying rings belonging to the SRS is flow controlled or not
2509  * and returns that status.
2510  */
2511 /* ARGSUSED */
2512 boolean_t
2513 mac_tx_is_flow_blocked(mac_client_handle_t mch, mac_tx_cookie_t cookie)
2514 {
2515 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
2516 	mac_soft_ring_set_t *mac_srs = MCIP_TX_SRS(mcip);
2517 	mac_soft_ring_t *sringp;
2518 	boolean_t blocked = B_FALSE;
2519 	int i;
2520 
2521 	/*
2522 	 * On etherstubs, there won't be a Tx SRS or an Rx
2523 	 * SRS. Infact there won't even be a flow_entry.
2524 	 */
2525 	if (mac_srs == NULL)
2526 		return (B_FALSE);
2527 
2528 	mutex_enter(&mac_srs->srs_lock);
2529 	if (mac_srs->srs_tx.st_mode == SRS_TX_FANOUT) {
2530 		for (i = 0; i < mac_srs->srs_oth_ring_count; i++) {
2531 			sringp = mac_srs->srs_oth_soft_rings[i];
2532 			mutex_enter(&sringp->s_ring_lock);
2533 			if (sringp->s_ring_state & S_RING_TX_HIWAT) {
2534 				blocked = B_TRUE;
2535 				mutex_exit(&sringp->s_ring_lock);
2536 				break;
2537 			}
2538 			mutex_exit(&sringp->s_ring_lock);
2539 		}
2540 	} else {
2541 		blocked = (mac_srs->srs_state & SRS_TX_HIWAT);
2542 	}
2543 	mutex_exit(&mac_srs->srs_lock);
2544 	return (blocked);
2545 }
2546 
2547 /*
2548  * Check if the MAC client is the primary MAC client.
2549  */
2550 boolean_t
2551 mac_is_primary_client(mac_client_impl_t *mcip)
2552 {
2553 	return (mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY);
2554 }
2555 
2556 void
2557 mac_ioctl(mac_handle_t mh, queue_t *wq, mblk_t *bp)
2558 {
2559 	mac_impl_t	*mip = (mac_impl_t *)mh;
2560 	int cmd = ((struct iocblk *)bp->b_rptr)->ioc_cmd;
2561 
2562 	if ((cmd == ND_GET && (mip->mi_callbacks->mc_callbacks & MC_GETPROP)) ||
2563 	    (cmd == ND_SET && (mip->mi_callbacks->mc_callbacks & MC_SETPROP))) {
2564 		/*
2565 		 * If ndd props were registered, call them.
2566 		 * Note that ndd ioctls are Obsolete
2567 		 */
2568 		mac_ndd_ioctl(mip, wq, bp);
2569 		return;
2570 	}
2571 
2572 	/*
2573 	 * Call the driver to handle the ioctl.  The driver may not support
2574 	 * any ioctls, in which case we reply with a NAK on its behalf.
2575 	 */
2576 	if (mip->mi_callbacks->mc_callbacks & MC_IOCTL)
2577 		mip->mi_ioctl(mip->mi_driver, wq, bp);
2578 	else
2579 		miocnak(wq, bp, 0, EINVAL);
2580 }
2581 
2582 /*
2583  * Return the link state of the specified MAC instance.
2584  */
2585 link_state_t
2586 mac_link_get(mac_handle_t mh)
2587 {
2588 	return (((mac_impl_t *)mh)->mi_linkstate);
2589 }
2590 
2591 /*
2592  * Add a mac client specified notification callback. Please see the comments
2593  * above mac_callback_add() for general information about mac callback
2594  * addition/deletion in the presence of mac callback list walkers
2595  */
2596 mac_notify_handle_t
2597 mac_notify_add(mac_handle_t mh, mac_notify_t notify_fn, void *arg)
2598 {
2599 	mac_impl_t		*mip = (mac_impl_t *)mh;
2600 	mac_notify_cb_t		*mncb;
2601 	mac_cb_info_t		*mcbi;
2602 
2603 	/*
2604 	 * Allocate a notify callback structure, fill in the details and
2605 	 * use the mac callback list manipulation functions to chain into
2606 	 * the list of callbacks.
2607 	 */
2608 	mncb = kmem_zalloc(sizeof (mac_notify_cb_t), KM_SLEEP);
2609 	mncb->mncb_fn = notify_fn;
2610 	mncb->mncb_arg = arg;
2611 	mncb->mncb_mip = mip;
2612 	mncb->mncb_link.mcb_objp = mncb;
2613 	mncb->mncb_link.mcb_objsize = sizeof (mac_notify_cb_t);
2614 	mncb->mncb_link.mcb_flags = MCB_NOTIFY_CB_T;
2615 
2616 	mcbi = &mip->mi_notify_cb_info;
2617 
2618 	i_mac_perim_enter(mip);
2619 	mutex_enter(mcbi->mcbi_lockp);
2620 
2621 	mac_callback_add(&mip->mi_notify_cb_info, &mip->mi_notify_cb_list,
2622 	    &mncb->mncb_link);
2623 
2624 	mutex_exit(mcbi->mcbi_lockp);
2625 	i_mac_perim_exit(mip);
2626 	return ((mac_notify_handle_t)mncb);
2627 }
2628 
2629 void
2630 mac_notify_remove_wait(mac_handle_t mh)
2631 {
2632 	mac_impl_t	*mip = (mac_impl_t *)mh;
2633 	mac_cb_info_t	*mcbi = &mip->mi_notify_cb_info;
2634 
2635 	mutex_enter(mcbi->mcbi_lockp);
2636 	mac_callback_remove_wait(&mip->mi_notify_cb_info);
2637 	mutex_exit(mcbi->mcbi_lockp);
2638 }
2639 
2640 /*
2641  * Remove a mac client specified notification callback
2642  */
2643 int
2644 mac_notify_remove(mac_notify_handle_t mnh, boolean_t wait)
2645 {
2646 	mac_notify_cb_t	*mncb = (mac_notify_cb_t *)mnh;
2647 	mac_impl_t	*mip = mncb->mncb_mip;
2648 	mac_cb_info_t	*mcbi;
2649 	int		err = 0;
2650 
2651 	mcbi = &mip->mi_notify_cb_info;
2652 
2653 	i_mac_perim_enter(mip);
2654 	mutex_enter(mcbi->mcbi_lockp);
2655 
2656 	ASSERT(mncb->mncb_link.mcb_objp == mncb);
2657 	/*
2658 	 * If there aren't any list walkers, the remove would succeed
2659 	 * inline, else we wait for the deferred remove to complete
2660 	 */
2661 	if (mac_callback_remove(&mip->mi_notify_cb_info,
2662 	    &mip->mi_notify_cb_list, &mncb->mncb_link)) {
2663 		kmem_free(mncb, sizeof (mac_notify_cb_t));
2664 	} else {
2665 		err = EBUSY;
2666 	}
2667 
2668 	mutex_exit(mcbi->mcbi_lockp);
2669 	i_mac_perim_exit(mip);
2670 
2671 	/*
2672 	 * If we failed to remove the notification callback and "wait" is set
2673 	 * to be B_TRUE, wait for the callback to finish after we exit the
2674 	 * mac perimeter.
2675 	 */
2676 	if (err != 0 && wait) {
2677 		mac_notify_remove_wait((mac_handle_t)mip);
2678 		return (0);
2679 	}
2680 
2681 	return (err);
2682 }
2683 
2684 /*
2685  * Associate resource management callbacks with the specified MAC
2686  * clients.
2687  */
2688 
2689 void
2690 mac_resource_set_common(mac_client_handle_t mch, mac_resource_add_t add,
2691     mac_resource_remove_t remove, mac_resource_quiesce_t quiesce,
2692     mac_resource_restart_t restart, mac_resource_bind_t bind,
2693     void *arg)
2694 {
2695 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
2696 
2697 	mcip->mci_resource_add = add;
2698 	mcip->mci_resource_remove = remove;
2699 	mcip->mci_resource_quiesce = quiesce;
2700 	mcip->mci_resource_restart = restart;
2701 	mcip->mci_resource_bind = bind;
2702 	mcip->mci_resource_arg = arg;
2703 
2704 	if (arg == NULL)
2705 		mcip->mci_state_flags &= ~MCIS_CLIENT_POLL_CAPABLE;
2706 }
2707 
2708 void
2709 mac_resource_set(mac_client_handle_t mch, mac_resource_add_t add, void *arg)
2710 {
2711 	/* update the 'resource_add' callback */
2712 	mac_resource_set_common(mch, add, NULL, NULL, NULL, NULL, arg);
2713 }
2714 
2715 /*
2716  * Sets up the client resources and enable the polling interface over all the
2717  * SRS's and the soft rings of the client
2718  */
2719 void
2720 mac_client_poll_enable(mac_client_handle_t mch)
2721 {
2722 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2723 	mac_soft_ring_set_t	*mac_srs;
2724 	flow_entry_t		*flent;
2725 	int			i;
2726 
2727 	flent = mcip->mci_flent;
2728 	ASSERT(flent != NULL);
2729 
2730 	for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
2731 		mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
2732 		ASSERT(mac_srs->srs_mcip == mcip);
2733 		mac_srs_client_poll_enable(mcip, mac_srs);
2734 	}
2735 }
2736 
2737 /*
2738  * Tears down the client resources and disable the polling interface over all
2739  * the SRS's and the soft rings of the client
2740  */
2741 void
2742 mac_client_poll_disable(mac_client_handle_t mch)
2743 {
2744 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2745 	mac_soft_ring_set_t	*mac_srs;
2746 	flow_entry_t		*flent;
2747 	int			i;
2748 
2749 	flent = mcip->mci_flent;
2750 	ASSERT(flent != NULL);
2751 
2752 	for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
2753 		mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
2754 		ASSERT(mac_srs->srs_mcip == mcip);
2755 		mac_srs_client_poll_disable(mcip, mac_srs);
2756 	}
2757 }
2758 
2759 /*
2760  * Associate the CPUs specified by the given property with a MAC client.
2761  */
2762 int
2763 mac_cpu_set(mac_client_handle_t mch, mac_resource_props_t *mrp)
2764 {
2765 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
2766 	mac_impl_t *mip = mcip->mci_mip;
2767 	int err = 0;
2768 
2769 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
2770 
2771 	if ((err = mac_validate_props(mrp)) != 0)
2772 		return (err);
2773 
2774 	if (MCIP_DATAPATH_SETUP(mcip))
2775 		mac_flow_modify(mip->mi_flow_tab, mcip->mci_flent, mrp);
2776 
2777 	mac_update_resources(mrp, MCIP_RESOURCE_PROPS(mcip), B_FALSE);
2778 	return (0);
2779 }
2780 
2781 /*
2782  * Apply the specified properties to the specified MAC client.
2783  */
2784 int
2785 mac_client_set_resources(mac_client_handle_t mch, mac_resource_props_t *mrp)
2786 {
2787 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
2788 	mac_impl_t *mip = mcip->mci_mip;
2789 	int err = 0;
2790 
2791 	i_mac_perim_enter(mip);
2792 
2793 	if ((mrp->mrp_mask & MRP_MAXBW) || (mrp->mrp_mask & MRP_PRIORITY)) {
2794 		err = mac_resource_ctl_set(mch, mrp);
2795 		if (err != 0) {
2796 			i_mac_perim_exit(mip);
2797 			return (err);
2798 		}
2799 	}
2800 
2801 	if (mrp->mrp_mask & MRP_CPUS)
2802 		err = mac_cpu_set(mch, mrp);
2803 
2804 	i_mac_perim_exit(mip);
2805 	return (err);
2806 }
2807 
2808 /*
2809  * Return the properties currently associated with the specified MAC client.
2810  */
2811 void
2812 mac_client_get_resources(mac_client_handle_t mch, mac_resource_props_t *mrp)
2813 {
2814 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2815 	mac_resource_props_t	*mcip_mrp = MCIP_RESOURCE_PROPS(mcip);
2816 
2817 	bcopy(mcip_mrp, mrp, sizeof (mac_resource_props_t));
2818 }
2819 
2820 /*
2821  * Pass a copy of the specified packet to the promiscuous callbacks
2822  * of the specified MAC.
2823  *
2824  * If sender is NULL, the function is being invoked for a packet chain
2825  * received from the wire. If sender is non-NULL, it points to
2826  * the MAC client from which the packet is being sent.
2827  *
2828  * The packets are distributed to the promiscuous callbacks as follows:
2829  *
2830  * - all packets are sent to the MAC_CLIENT_PROMISC_ALL callbacks
2831  * - all broadcast and multicast packets are sent to the
2832  *   MAC_CLIENT_PROMISC_FILTER and MAC_CLIENT_PROMISC_MULTI.
2833  *
2834  * The unicast packets of MAC_CLIENT_PROMISC_FILTER callbacks are dispatched
2835  * after classification by mac_rx_deliver().
2836  */
2837 
2838 static void
2839 mac_promisc_dispatch_one(mac_promisc_impl_t *mpip, mblk_t *mp,
2840     boolean_t loopback)
2841 {
2842 	mblk_t *mp_copy;
2843 
2844 	mp_copy = copymsg(mp);
2845 	if (mp_copy == NULL)
2846 		return;
2847 	mp_copy->b_next = NULL;
2848 
2849 	mpip->mpi_fn(mpip->mpi_arg, NULL, mp_copy, loopback);
2850 }
2851 
2852 /*
2853  * Return the VID of a packet. Zero if the packet is not tagged.
2854  */
2855 static uint16_t
2856 mac_ether_vid(mblk_t *mp)
2857 {
2858 	struct ether_header *eth = (struct ether_header *)mp->b_rptr;
2859 
2860 	if (ntohs(eth->ether_type) == ETHERTYPE_VLAN) {
2861 		struct ether_vlan_header *t_evhp =
2862 		    (struct ether_vlan_header *)mp->b_rptr;
2863 		return (VLAN_ID(ntohs(t_evhp->ether_tci)));
2864 	}
2865 
2866 	return (0);
2867 }
2868 
2869 /*
2870  * Return whether the specified packet contains a multicast or broadcast
2871  * destination MAC address.
2872  */
2873 static boolean_t
2874 mac_is_mcast(mac_impl_t *mip, mblk_t *mp)
2875 {
2876 	mac_header_info_t hdr_info;
2877 
2878 	if (mac_header_info((mac_handle_t)mip, mp, &hdr_info) != 0)
2879 		return (B_FALSE);
2880 	return ((hdr_info.mhi_dsttype == MAC_ADDRTYPE_BROADCAST) ||
2881 	    (hdr_info.mhi_dsttype == MAC_ADDRTYPE_MULTICAST));
2882 }
2883 
2884 /*
2885  * Send a copy of an mblk chain to the MAC clients of the specified MAC.
2886  * "sender" points to the sender MAC client for outbound packets, and
2887  * is set to NULL for inbound packets.
2888  */
2889 void
2890 mac_promisc_dispatch(mac_impl_t *mip, mblk_t *mp_chain,
2891     mac_client_impl_t *sender)
2892 {
2893 	mac_promisc_impl_t *mpip;
2894 	mac_cb_t *mcb;
2895 	mblk_t *mp;
2896 	boolean_t is_mcast, is_sender;
2897 
2898 	MAC_PROMISC_WALKER_INC(mip);
2899 	for (mp = mp_chain; mp != NULL; mp = mp->b_next) {
2900 		is_mcast = mac_is_mcast(mip, mp);
2901 		/* send packet to interested callbacks */
2902 		for (mcb = mip->mi_promisc_list; mcb != NULL;
2903 		    mcb = mcb->mcb_nextp) {
2904 			mpip = (mac_promisc_impl_t *)mcb->mcb_objp;
2905 			is_sender = (mpip->mpi_mcip == sender);
2906 
2907 			if (is_sender && mpip->mpi_no_tx_loop)
2908 				/*
2909 				 * The sender doesn't want to receive
2910 				 * copies of the packets it sends.
2911 				 */
2912 				continue;
2913 
2914 			/*
2915 			 * For an ethernet MAC, don't displatch a multicast
2916 			 * packet to a non-PROMISC_ALL callbacks unless the VID
2917 			 * of the packet matches the VID of the client.
2918 			 */
2919 			if (is_mcast &&
2920 			    mpip->mpi_type != MAC_CLIENT_PROMISC_ALL &&
2921 			    !mac_client_check_flow_vid(mpip->mpi_mcip,
2922 			    mac_ether_vid(mp)))
2923 				continue;
2924 
2925 			if (is_sender ||
2926 			    mpip->mpi_type == MAC_CLIENT_PROMISC_ALL ||
2927 			    is_mcast)
2928 				mac_promisc_dispatch_one(mpip, mp, is_sender);
2929 		}
2930 	}
2931 	MAC_PROMISC_WALKER_DCR(mip);
2932 }
2933 
2934 void
2935 mac_promisc_client_dispatch(mac_client_impl_t *mcip, mblk_t *mp_chain)
2936 {
2937 	mac_impl_t		*mip = mcip->mci_mip;
2938 	mac_promisc_impl_t	*mpip;
2939 	boolean_t		is_mcast;
2940 	mblk_t			*mp;
2941 	mac_cb_t		*mcb;
2942 
2943 	/*
2944 	 * The unicast packets for the MAC client still
2945 	 * need to be delivered to the MAC_CLIENT_PROMISC_FILTERED
2946 	 * promiscuous callbacks. The broadcast and multicast
2947 	 * packets were delivered from mac_rx().
2948 	 */
2949 	MAC_PROMISC_WALKER_INC(mip);
2950 	for (mp = mp_chain; mp != NULL; mp = mp->b_next) {
2951 		is_mcast = mac_is_mcast(mip, mp);
2952 		for (mcb = mcip->mci_promisc_list; mcb != NULL;
2953 		    mcb = mcb->mcb_nextp) {
2954 			mpip = (mac_promisc_impl_t *)mcb->mcb_objp;
2955 			if (mpip->mpi_type == MAC_CLIENT_PROMISC_FILTERED &&
2956 			    !is_mcast) {
2957 				mac_promisc_dispatch_one(mpip, mp, B_FALSE);
2958 			}
2959 		}
2960 	}
2961 	MAC_PROMISC_WALKER_DCR(mip);
2962 }
2963 
2964 /*
2965  * Return the margin value currently assigned to the specified MAC instance.
2966  */
2967 void
2968 mac_margin_get(mac_handle_t mh, uint32_t *marginp)
2969 {
2970 	mac_impl_t *mip = (mac_impl_t *)mh;
2971 
2972 	rw_enter(&(mip->mi_rw_lock), RW_READER);
2973 	*marginp = mip->mi_margin;
2974 	rw_exit(&(mip->mi_rw_lock));
2975 }
2976 
2977 /*
2978  * mac_info_get() is used for retrieving the mac_info when a DL_INFO_REQ is
2979  * issued before a DL_ATTACH_REQ. we walk the i_mac_impl_hash table and find
2980  * the first mac_impl_t with a matching driver name; then we copy its mac_info_t
2981  * to the caller. we do all this with i_mac_impl_lock held so the mac_impl_t
2982  * cannot disappear while we are accessing it.
2983  */
2984 typedef struct i_mac_info_state_s {
2985 	const char	*mi_name;
2986 	mac_info_t	*mi_infop;
2987 } i_mac_info_state_t;
2988 
2989 /*ARGSUSED*/
2990 static uint_t
2991 i_mac_info_walker(mod_hash_key_t key, mod_hash_val_t *val, void *arg)
2992 {
2993 	i_mac_info_state_t *statep = arg;
2994 	mac_impl_t *mip = (mac_impl_t *)val;
2995 
2996 	if (mip->mi_state_flags & MIS_DISABLED)
2997 		return (MH_WALK_CONTINUE);
2998 
2999 	if (strcmp(statep->mi_name,
3000 	    ddi_driver_name(mip->mi_dip)) != 0)
3001 		return (MH_WALK_CONTINUE);
3002 
3003 	statep->mi_infop = &mip->mi_info;
3004 	return (MH_WALK_TERMINATE);
3005 }
3006 
3007 boolean_t
3008 mac_info_get(const char *name, mac_info_t *minfop)
3009 {
3010 	i_mac_info_state_t state;
3011 
3012 	rw_enter(&i_mac_impl_lock, RW_READER);
3013 	state.mi_name = name;
3014 	state.mi_infop = NULL;
3015 	mod_hash_walk(i_mac_impl_hash, i_mac_info_walker, &state);
3016 	if (state.mi_infop == NULL) {
3017 		rw_exit(&i_mac_impl_lock);
3018 		return (B_FALSE);
3019 	}
3020 	*minfop = *state.mi_infop;
3021 	rw_exit(&i_mac_impl_lock);
3022 	return (B_TRUE);
3023 }
3024 
3025 /*
3026  * To get the capabilities that MAC layer cares about, such as rings, factory
3027  * mac address, vnic or not, it should directly invoke this function
3028  */
3029 boolean_t
3030 i_mac_capab_get(mac_handle_t mh, mac_capab_t cap, void *cap_data)
3031 {
3032 	mac_impl_t *mip = (mac_impl_t *)mh;
3033 
3034 	if (mip->mi_callbacks->mc_callbacks & MC_GETCAPAB)
3035 		return (mip->mi_getcapab(mip->mi_driver, cap, cap_data));
3036 	else
3037 		return (B_FALSE);
3038 }
3039 
3040 /*
3041  * Capability query function. If number of active mac clients is greater than
3042  * 1, only limited capabilities can be advertised to the caller no matter the
3043  * driver has certain capability or not. Else, we query the driver to get the
3044  * capability.
3045  */
3046 boolean_t
3047 mac_capab_get(mac_handle_t mh, mac_capab_t cap, void *cap_data)
3048 {
3049 	mac_impl_t *mip = (mac_impl_t *)mh;
3050 
3051 	/*
3052 	 * if mi_nactiveclients > 1, only MAC_CAPAB_HCKSUM,
3053 	 * MAC_CAPAB_NO_NATIVEVLAN, MAC_CAPAB_NO_ZCOPY can be advertised.
3054 	 */
3055 	if (mip->mi_nactiveclients > 1) {
3056 		switch (cap) {
3057 		case MAC_CAPAB_HCKSUM:
3058 			return (i_mac_capab_get(mh, cap, cap_data));
3059 		case MAC_CAPAB_NO_NATIVEVLAN:
3060 		case MAC_CAPAB_NO_ZCOPY:
3061 			return (B_TRUE);
3062 		default:
3063 			return (B_FALSE);
3064 		}
3065 	}
3066 
3067 	/* else get capab from driver */
3068 	return (i_mac_capab_get(mh, cap, cap_data));
3069 }
3070 
3071 boolean_t
3072 mac_sap_verify(mac_handle_t mh, uint32_t sap, uint32_t *bind_sap)
3073 {
3074 	mac_impl_t *mip = (mac_impl_t *)mh;
3075 
3076 	return (mip->mi_type->mt_ops.mtops_sap_verify(sap, bind_sap,
3077 	    mip->mi_pdata));
3078 }
3079 
3080 mblk_t *
3081 mac_header(mac_handle_t mh, const uint8_t *daddr, uint32_t sap, mblk_t *payload,
3082     size_t extra_len)
3083 {
3084 	mac_impl_t *mip = (mac_impl_t *)mh;
3085 
3086 	return (mip->mi_type->mt_ops.mtops_header(mip->mi_addr, daddr, sap,
3087 	    mip->mi_pdata, payload, extra_len));
3088 }
3089 
3090 int
3091 mac_header_info(mac_handle_t mh, mblk_t *mp, mac_header_info_t *mhip)
3092 {
3093 	mac_impl_t *mip = (mac_impl_t *)mh;
3094 
3095 	return (mip->mi_type->mt_ops.mtops_header_info(mp, mip->mi_pdata,
3096 	    mhip));
3097 }
3098 
3099 mblk_t *
3100 mac_header_cook(mac_handle_t mh, mblk_t *mp)
3101 {
3102 	mac_impl_t *mip = (mac_impl_t *)mh;
3103 
3104 	if (mip->mi_type->mt_ops.mtops_ops & MTOPS_HEADER_COOK) {
3105 		if (DB_REF(mp) > 1) {
3106 			mblk_t *newmp = copymsg(mp);
3107 			if (newmp == NULL)
3108 				return (NULL);
3109 			freemsg(mp);
3110 			mp = newmp;
3111 		}
3112 		return (mip->mi_type->mt_ops.mtops_header_cook(mp,
3113 		    mip->mi_pdata));
3114 	}
3115 	return (mp);
3116 }
3117 
3118 mblk_t *
3119 mac_header_uncook(mac_handle_t mh, mblk_t *mp)
3120 {
3121 	mac_impl_t *mip = (mac_impl_t *)mh;
3122 
3123 	if (mip->mi_type->mt_ops.mtops_ops & MTOPS_HEADER_UNCOOK) {
3124 		if (DB_REF(mp) > 1) {
3125 			mblk_t *newmp = copymsg(mp);
3126 			if (newmp == NULL)
3127 				return (NULL);
3128 			freemsg(mp);
3129 			mp = newmp;
3130 		}
3131 		return (mip->mi_type->mt_ops.mtops_header_uncook(mp,
3132 		    mip->mi_pdata));
3133 	}
3134 	return (mp);
3135 }
3136 
3137 uint_t
3138 mac_addr_len(mac_handle_t mh)
3139 {
3140 	mac_impl_t *mip = (mac_impl_t *)mh;
3141 
3142 	return (mip->mi_type->mt_addr_length);
3143 }
3144 
3145 /* True if a MAC is a VNIC */
3146 boolean_t
3147 mac_is_vnic(mac_handle_t mh)
3148 {
3149 	return (((mac_impl_t *)mh)->mi_state_flags & MIS_IS_VNIC);
3150 }
3151 
3152 mac_handle_t
3153 mac_get_lower_mac_handle(mac_handle_t mh)
3154 {
3155 	mac_impl_t *mip = (mac_impl_t *)mh;
3156 
3157 	ASSERT(mac_is_vnic(mh));
3158 	return (((vnic_t *)mip->mi_driver)->vn_lower_mh);
3159 }
3160 
3161 void
3162 mac_update_resources(mac_resource_props_t *nmrp, mac_resource_props_t *cmrp,
3163     boolean_t is_user_flow)
3164 {
3165 	if (nmrp != NULL && cmrp != NULL) {
3166 		if (nmrp->mrp_mask & MRP_PRIORITY) {
3167 			if (nmrp->mrp_priority == MPL_RESET) {
3168 				cmrp->mrp_mask &= ~MRP_PRIORITY;
3169 				if (is_user_flow) {
3170 					cmrp->mrp_priority =
3171 					    MPL_SUBFLOW_DEFAULT;
3172 				} else {
3173 					cmrp->mrp_priority = MPL_LINK_DEFAULT;
3174 				}
3175 			} else {
3176 				cmrp->mrp_mask |= MRP_PRIORITY;
3177 				cmrp->mrp_priority = nmrp->mrp_priority;
3178 			}
3179 		}
3180 		if (nmrp->mrp_mask & MRP_MAXBW) {
3181 			cmrp->mrp_maxbw = nmrp->mrp_maxbw;
3182 			if (nmrp->mrp_maxbw == MRP_MAXBW_RESETVAL)
3183 				cmrp->mrp_mask &= ~MRP_MAXBW;
3184 			else
3185 				cmrp->mrp_mask |= MRP_MAXBW;
3186 		}
3187 		if (nmrp->mrp_mask & MRP_CPUS)
3188 			MAC_COPY_CPUS(nmrp, cmrp);
3189 	}
3190 }
3191 
3192 /*
3193  * i_mac_set_resources:
3194  *
3195  * This routine associates properties with the primary MAC client of
3196  * the specified MAC instance.
3197  * - Cache the properties in mac_impl_t
3198  * - Apply the properties to the primary MAC client if exists
3199  */
3200 int
3201 i_mac_set_resources(mac_handle_t mh, mac_resource_props_t *mrp)
3202 {
3203 	mac_impl_t		*mip = (mac_impl_t *)mh;
3204 	mac_client_impl_t	*mcip;
3205 	int			err = 0;
3206 	mac_resource_props_t	tmrp;
3207 
3208 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
3209 
3210 	err = mac_validate_props(mrp);
3211 	if (err != 0)
3212 		return (err);
3213 
3214 	/*
3215 	 * Since bind_cpu may be modified by mac_client_set_resources()
3216 	 * we use a copy of bind_cpu and finally cache bind_cpu in mip.
3217 	 * This allows us to cache only user edits in mip.
3218 	 */
3219 	bcopy(mrp, &tmrp, sizeof (mac_resource_props_t));
3220 	mcip = mac_primary_client_handle(mip);
3221 	if (mcip != NULL) {
3222 		err =
3223 		    mac_client_set_resources((mac_client_handle_t)mcip, &tmrp);
3224 	}
3225 	/* if mac_client_set_resources failed, do not update the values */
3226 	if (err == 0)
3227 		mac_update_resources(mrp, &mip->mi_resource_props, B_FALSE);
3228 	return (err);
3229 }
3230 
3231 int
3232 mac_set_resources(mac_handle_t mh, mac_resource_props_t *mrp)
3233 {
3234 	int err;
3235 
3236 	i_mac_perim_enter((mac_impl_t *)mh);
3237 	err = i_mac_set_resources(mh, mrp);
3238 	i_mac_perim_exit((mac_impl_t *)mh);
3239 	return (err);
3240 }
3241 
3242 /*
3243  * Get the properties cached for the specified MAC instance.
3244  */
3245 void
3246 mac_get_resources(mac_handle_t mh, mac_resource_props_t *mrp)
3247 {
3248 	mac_impl_t 		*mip = (mac_impl_t *)mh;
3249 	mac_client_impl_t	*mcip;
3250 
3251 	if (mip->mi_state_flags & MIS_IS_VNIC) {
3252 		mcip = mac_primary_client_handle(mip);
3253 		if (mcip != NULL) {
3254 			mac_client_get_resources((mac_client_handle_t)mcip,
3255 			    mrp);
3256 			return;
3257 		}
3258 	}
3259 	bcopy(&mip->mi_resource_props, mrp, sizeof (mac_resource_props_t));
3260 }
3261 
3262 /*
3263  * Rename a mac client, its flow, and the kstat.
3264  */
3265 int
3266 mac_rename_primary(mac_handle_t mh, const char *new_name)
3267 {
3268 	mac_impl_t		*mip = (mac_impl_t *)mh;
3269 	mac_client_impl_t	*cur_clnt = NULL;
3270 	flow_entry_t		*fep;
3271 
3272 	i_mac_perim_enter(mip);
3273 
3274 	/*
3275 	 * VNICs: we need to change the sys flow name and
3276 	 * the associated flow kstat.
3277 	 */
3278 	if (mip->mi_state_flags & MIS_IS_VNIC) {
3279 		ASSERT(new_name != NULL);
3280 		mac_rename_flow_names(mac_vnic_lower(mip), new_name);
3281 		goto done;
3282 	}
3283 	/*
3284 	 * This mac may itself be an aggr link, or it may have some client
3285 	 * which is an aggr port. For both cases, we need to change the
3286 	 * aggr port's mac client name, its flow name and the associated flow
3287 	 * kstat.
3288 	 */
3289 	if (mip->mi_state_flags & MIS_IS_AGGR) {
3290 		mac_capab_aggr_t aggr_cap;
3291 		mac_rename_fn_t rename_fn;
3292 		boolean_t ret;
3293 
3294 		ASSERT(new_name != NULL);
3295 		ret = i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_AGGR,
3296 		    (void *)(&aggr_cap));
3297 		ASSERT(ret == B_TRUE);
3298 		rename_fn = aggr_cap.mca_rename_fn;
3299 		rename_fn(new_name, mip->mi_driver);
3300 		/*
3301 		 * The aggr's client name and kstat flow name will be
3302 		 * updated below, i.e. via mac_rename_flow_names.
3303 		 */
3304 	}
3305 
3306 	for (cur_clnt = mip->mi_clients_list; cur_clnt != NULL;
3307 	    cur_clnt = cur_clnt->mci_client_next) {
3308 		if (cur_clnt->mci_state_flags & MCIS_IS_AGGR_PORT) {
3309 			if (new_name != NULL) {
3310 				char *str_st = cur_clnt->mci_name;
3311 				char *str_del = strchr(str_st, '-');
3312 
3313 				ASSERT(str_del != NULL);
3314 				bzero(str_del + 1, MAXNAMELEN -
3315 				    (str_del - str_st + 1));
3316 				bcopy(new_name, str_del + 1,
3317 				    strlen(new_name));
3318 			}
3319 			fep = cur_clnt->mci_flent;
3320 			mac_rename_flow(fep, cur_clnt->mci_name);
3321 			break;
3322 		} else if (new_name != NULL &&
3323 		    cur_clnt->mci_state_flags & MCIS_USE_DATALINK_NAME) {
3324 			mac_rename_flow_names(cur_clnt, new_name);
3325 			break;
3326 		}
3327 	}
3328 
3329 done:
3330 	i_mac_perim_exit(mip);
3331 	return (0);
3332 }
3333 
3334 /*
3335  * Rename the MAC client's flow names
3336  */
3337 static void
3338 mac_rename_flow_names(mac_client_impl_t *mcip, const char *new_name)
3339 {
3340 	flow_entry_t	*flent;
3341 	uint16_t	vid;
3342 	char		flowname[MAXFLOWNAME];
3343 	mac_impl_t	*mip = mcip->mci_mip;
3344 
3345 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
3346 
3347 	/*
3348 	 * Use mi_rw_lock to ensure that threads not in the mac perimeter
3349 	 * see a self-consistent value for mci_name
3350 	 */
3351 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
3352 	(void) strlcpy(mcip->mci_name, new_name, sizeof (mcip->mci_name));
3353 	rw_exit(&mip->mi_rw_lock);
3354 
3355 	mac_rename_flow(mcip->mci_flent, new_name);
3356 
3357 	if (mcip->mci_nflents == 1)
3358 		return;
3359 
3360 	/*
3361 	 * We have to rename all the others too, no stats to destroy for
3362 	 * these.
3363 	 */
3364 	for (flent = mcip->mci_flent_list; flent != NULL;
3365 	    flent = flent->fe_client_next) {
3366 		if (flent != mcip->mci_flent) {
3367 			vid = i_mac_flow_vid(flent);
3368 			(void) sprintf(flowname, "%s%u", new_name, vid);
3369 			mac_flow_set_name(flent, flowname);
3370 		}
3371 	}
3372 }
3373 
3374 
3375 /*
3376  * Add a flow to the MAC client's flow list - i.e list of MAC/VID tuples
3377  * defined for the specified MAC client.
3378  */
3379 static void
3380 mac_client_add_to_flow_list(mac_client_impl_t *mcip, flow_entry_t *flent)
3381 {
3382 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
3383 	/*
3384 	 * The promisc Rx data path walks the mci_flent_list. Protect by
3385 	 * using mi_rw_lock
3386 	 */
3387 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
3388 
3389 	/* Add it to the head */
3390 	flent->fe_client_next = mcip->mci_flent_list;
3391 	mcip->mci_flent_list = flent;
3392 	mcip->mci_nflents++;
3393 
3394 	/*
3395 	 * Keep track of the number of non-zero VIDs addresses per MAC
3396 	 * client to avoid figuring it out in the data-path.
3397 	 */
3398 	if (i_mac_flow_vid(flent) != VLAN_ID_NONE)
3399 		mcip->mci_nvids++;
3400 
3401 	rw_exit(&mcip->mci_rw_lock);
3402 }
3403 
3404 /*
3405  * Remove a flow entry from the MAC client's list.
3406  */
3407 static void
3408 mac_client_remove_flow_from_list(mac_client_impl_t *mcip, flow_entry_t *flent)
3409 {
3410 	flow_entry_t	*fe = mcip->mci_flent_list;
3411 	flow_entry_t	*prev_fe = NULL;
3412 
3413 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
3414 	/*
3415 	 * The promisc Rx data path walks the mci_flent_list. Protect by
3416 	 * using mci_rw_lock
3417 	 */
3418 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
3419 	while ((fe != NULL) && (fe != flent)) {
3420 		prev_fe = fe;
3421 		fe = fe->fe_client_next;
3422 	}
3423 
3424 	/* XXX should be an ASSERT */
3425 	if (fe != NULL) {
3426 		if (prev_fe == NULL) {
3427 			/* Deleting the first node */
3428 			mcip->mci_flent_list = fe->fe_client_next;
3429 		} else {
3430 			prev_fe->fe_client_next = fe->fe_client_next;
3431 		}
3432 		mcip->mci_nflents--;
3433 
3434 		if (i_mac_flow_vid(flent) != VLAN_ID_NONE)
3435 			mcip->mci_nvids--;
3436 	}
3437 	rw_exit(&mcip->mci_rw_lock);
3438 }
3439 
3440 /*
3441  * Check if the given VID belongs to this MAC client.
3442  */
3443 boolean_t
3444 mac_client_check_flow_vid(mac_client_impl_t *mcip, uint16_t vid)
3445 {
3446 	flow_entry_t	*flent;
3447 	uint16_t	mci_vid;
3448 
3449 	/* The mci_flent_list is protected by mci_rw_lock */
3450 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
3451 	for (flent = mcip->mci_flent_list; flent != NULL;
3452 	    flent = flent->fe_client_next) {
3453 		mci_vid = i_mac_flow_vid(flent);
3454 		if (vid == mci_vid) {
3455 			rw_exit(&mcip->mci_rw_lock);
3456 			return (B_TRUE);
3457 		}
3458 	}
3459 	rw_exit(&mcip->mci_rw_lock);
3460 	return (B_FALSE);
3461 }
3462 
3463 /*
3464  * Get the flow entry for the specified <MAC addr, VID> tuple.
3465  */
3466 static flow_entry_t *
3467 mac_client_get_flow(mac_client_impl_t *mcip, mac_unicast_impl_t *muip)
3468 {
3469 	mac_address_t *map = mcip->mci_unicast;
3470 	flow_entry_t *flent;
3471 	uint16_t vid;
3472 	flow_desc_t flow_desc;
3473 
3474 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
3475 
3476 	mac_flow_get_desc(mcip->mci_flent, &flow_desc);
3477 	if (bcmp(flow_desc.fd_dst_mac, map->ma_addr, map->ma_len) != 0)
3478 		return (NULL);
3479 
3480 	for (flent = mcip->mci_flent_list; flent != NULL;
3481 	    flent = flent->fe_client_next) {
3482 		vid = i_mac_flow_vid(flent);
3483 		if (vid == muip->mui_vid) {
3484 			return (flent);
3485 		}
3486 	}
3487 
3488 	return (NULL);
3489 }
3490 
3491 /*
3492  * Since mci_flent has the SRSs, when we want to remove it, we replace
3493  * the flow_desc_t in mci_flent with that of an existing flent and then
3494  * remove that flent instead of mci_flent.
3495  */
3496 static flow_entry_t *
3497 mac_client_swap_mciflent(mac_client_impl_t *mcip)
3498 {
3499 	flow_entry_t	*flent = mcip->mci_flent;
3500 	flow_tab_t	*ft = flent->fe_flow_tab;
3501 	flow_entry_t	*flent1;
3502 	flow_desc_t	fl_desc;
3503 	char		fl_name[MAXFLOWNAME];
3504 	int		err;
3505 
3506 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
3507 	ASSERT(mcip->mci_nflents > 1);
3508 
3509 	/* get the next flent following the primary flent  */
3510 	flent1 = mcip->mci_flent_list->fe_client_next;
3511 	ASSERT(flent1 != NULL && flent1->fe_flow_tab == ft);
3512 
3513 	/*
3514 	 * Remove the flent from the flow table before updating the
3515 	 * flow descriptor as the hash depends on the flow descriptor.
3516 	 * This also helps incoming packet classification avoid having
3517 	 * to grab fe_lock. Access to fe_flow_desc of a flent not in the
3518 	 * flow table is done under the fe_lock so that log or stat functions
3519 	 * see a self-consistent fe_flow_desc. The name and desc are specific
3520 	 * to a flow, the rest are shared by all the clients, including
3521 	 * resource control etc.
3522 	 */
3523 	mac_flow_remove(ft, flent, B_TRUE);
3524 	mac_flow_remove(ft, flent1, B_TRUE);
3525 
3526 	bcopy(&flent->fe_flow_desc, &fl_desc, sizeof (flow_desc_t));
3527 	bcopy(flent->fe_flow_name, fl_name, MAXFLOWNAME);
3528 
3529 	/* update the primary flow entry */
3530 	mutex_enter(&flent->fe_lock);
3531 	bcopy(&flent1->fe_flow_desc, &flent->fe_flow_desc,
3532 	    sizeof (flow_desc_t));
3533 	bcopy(&flent1->fe_flow_name, &flent->fe_flow_name, MAXFLOWNAME);
3534 	mutex_exit(&flent->fe_lock);
3535 
3536 	/* update the flow entry that is to be freed */
3537 	mutex_enter(&flent1->fe_lock);
3538 	bcopy(&fl_desc, &flent1->fe_flow_desc, sizeof (flow_desc_t));
3539 	bcopy(fl_name, &flent1->fe_flow_name, MAXFLOWNAME);
3540 	mutex_exit(&flent1->fe_lock);
3541 
3542 	/* now reinsert the flow entries in the table */
3543 	err = mac_flow_add(ft, flent);
3544 	ASSERT(err == 0);
3545 
3546 	err = mac_flow_add(ft, flent1);
3547 	ASSERT(err == 0);
3548 
3549 	return (flent1);
3550 }
3551 
3552 /*
3553  * Return whether there is only one flow entry associated with this
3554  * MAC client.
3555  */
3556 static boolean_t
3557 mac_client_single_rcvr(mac_client_impl_t *mcip)
3558 {
3559 	return (mcip->mci_nflents == 1);
3560 }
3561 
3562 int
3563 mac_validate_props(mac_resource_props_t *mrp)
3564 {
3565 	if (mrp == NULL)
3566 		return (0);
3567 
3568 	if (mrp->mrp_mask & MRP_PRIORITY) {
3569 		mac_priority_level_t	pri = mrp->mrp_priority;
3570 
3571 		if (pri < MPL_LOW || pri > MPL_RESET)
3572 			return (EINVAL);
3573 	}
3574 
3575 	if (mrp->mrp_mask & MRP_MAXBW) {
3576 		uint64_t maxbw = mrp->mrp_maxbw;
3577 
3578 		if (maxbw < MRP_MAXBW_MINVAL && maxbw != 0)
3579 			return (EINVAL);
3580 	}
3581 	if (mrp->mrp_mask & MRP_CPUS) {
3582 		int i;
3583 		mac_cpu_mode_t	fanout;
3584 
3585 		if (mrp->mrp_ncpus > ncpus || mrp->mrp_ncpus > MAX_SR_FANOUT)
3586 			return (EINVAL);
3587 
3588 		for (i = 0; i < mrp->mrp_ncpus; i++) {
3589 			cpu_t *cp;
3590 			int rv;
3591 
3592 			mutex_enter(&cpu_lock);
3593 			cp = cpu_get(mrp->mrp_cpu[i]);
3594 			if (cp != NULL)
3595 				rv = cpu_is_online(cp);
3596 			else
3597 				rv = 0;
3598 			mutex_exit(&cpu_lock);
3599 			if (rv == 0)
3600 				return (EINVAL);
3601 		}
3602 
3603 		fanout = mrp->mrp_fanout_mode;
3604 		if (fanout < 0 || fanout > MCM_CPUS)
3605 			return (EINVAL);
3606 	}
3607 	return (0);
3608 }
3609 
3610 /*
3611  * Send a MAC_NOTE_LINK notification to all the MAC clients whenever the
3612  * underlying physical link is down. This is to allow MAC clients to
3613  * communicate with other clients.
3614  */
3615 void
3616 mac_virtual_link_update(mac_impl_t *mip)
3617 {
3618 	if (mip->mi_linkstate != LINK_STATE_UP)
3619 		i_mac_notify(mip, MAC_NOTE_LINK);
3620 }
3621 
3622 /*
3623  * For clients that have a pass-thru MAC, e.g. VNIC, we set the VNIC's
3624  * mac handle in the client.
3625  */
3626 void
3627 mac_set_upper_mac(mac_client_handle_t mch, mac_handle_t mh)
3628 {
3629 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3630 
3631 	mcip->mci_upper_mip = (mac_impl_t *)mh;
3632 }
3633 
3634 /*
3635  * Mark the mac as being used exclusively by the single mac client that is
3636  * doing some control operation on this mac. No further opens of this mac
3637  * will be allowed until this client calls mac_unmark_exclusive. The mac
3638  * client calling this function must already be in the mac perimeter
3639  */
3640 int
3641 mac_mark_exclusive(mac_handle_t mh)
3642 {
3643 	mac_impl_t	*mip = (mac_impl_t *)mh;
3644 
3645 	ASSERT(MAC_PERIM_HELD(mh));
3646 	/*
3647 	 * Look up its entry in the global hash table.
3648 	 */
3649 	rw_enter(&i_mac_impl_lock, RW_WRITER);
3650 	if (mip->mi_state_flags & MIS_DISABLED) {
3651 		rw_exit(&i_mac_impl_lock);
3652 		return (ENOENT);
3653 	}
3654 
3655 	/*
3656 	 * A reference to mac is held even if the link is not plumbed.
3657 	 * In i_dls_link_create() we open the MAC interface and hold the
3658 	 * reference. There is an additional reference for the mac_open
3659 	 * done in acquiring the mac perimeter
3660 	 */
3661 	if (mip->mi_ref != 2) {
3662 		rw_exit(&i_mac_impl_lock);
3663 		return (EBUSY);
3664 	}
3665 
3666 	ASSERT(!(mip->mi_state_flags & MIS_EXCLUSIVE_HELD));
3667 	mip->mi_state_flags |= MIS_EXCLUSIVE_HELD;
3668 	rw_exit(&i_mac_impl_lock);
3669 	return (0);
3670 }
3671 
3672 void
3673 mac_unmark_exclusive(mac_handle_t mh)
3674 {
3675 	mac_impl_t	*mip = (mac_impl_t *)mh;
3676 
3677 	ASSERT(MAC_PERIM_HELD(mh));
3678 
3679 	rw_enter(&i_mac_impl_lock, RW_WRITER);
3680 	/* 1 for the creation and another for the perimeter */
3681 	ASSERT(mip->mi_ref == 2 && (mip->mi_state_flags & MIS_EXCLUSIVE_HELD));
3682 	mip->mi_state_flags &= ~MIS_EXCLUSIVE_HELD;
3683 	rw_exit(&i_mac_impl_lock);
3684 }
3685 
3686 /*
3687  * Set the MTU for the specified device. The function returns EBUSY if
3688  * another MAC client prevents the caller to become the exclusive client.
3689  * Returns EAGAIN if the client is started.
3690  */
3691 int
3692 mac_set_mtu(mac_handle_t mh, uint_t new_mtu, uint_t *old_mtu_arg)
3693 {
3694 	mac_impl_t *mip = (mac_impl_t *)mh;
3695 	uint_t old_mtu;
3696 	int rv;
3697 	boolean_t exclusive = B_FALSE;
3698 
3699 	i_mac_perim_enter(mip);
3700 
3701 	if ((mip->mi_callbacks->mc_callbacks & MC_SETPROP) == 0 ||
3702 	    (mip->mi_callbacks->mc_callbacks & MC_GETPROP) == 0) {
3703 		rv = ENOTSUP;
3704 		goto bail;
3705 	}
3706 
3707 	if ((rv = mac_mark_exclusive(mh)) != 0)
3708 		goto bail;
3709 	exclusive = B_TRUE;
3710 
3711 	if (mip->mi_active > 0) {
3712 		/*
3713 		 * The MAC instance is started, for example due to the
3714 		 * presence of a promiscuous clients. Fail the operation
3715 		 * since the MAC's MTU cannot be changed while the NIC
3716 		 * is started.
3717 		 */
3718 		rv = EAGAIN;
3719 		goto bail;
3720 	}
3721 
3722 	mac_sdu_get(mh, NULL, &old_mtu);
3723 
3724 	if (old_mtu != new_mtu) {
3725 		rv = mip->mi_callbacks->mc_setprop(mip->mi_driver,
3726 		    "mtu", MAC_PROP_MTU, sizeof (uint_t), &new_mtu);
3727 	}
3728 
3729 bail:
3730 	if (exclusive)
3731 		mac_unmark_exclusive(mh);
3732 	i_mac_perim_exit(mip);
3733 
3734 	if (rv == 0 && old_mtu_arg != NULL)
3735 		*old_mtu_arg = old_mtu;
3736 	return (rv);
3737 }
3738 
3739 void
3740 mac_get_hwgrp_info(mac_handle_t mh, int grp_index, uint_t *grp_num,
3741     uint_t *n_rings, uint_t *type, uint_t *n_clnts, char *clnts_name)
3742 {
3743 	mac_impl_t *mip = (mac_impl_t *)mh;
3744 	mac_grp_client_t *mcip;
3745 	uint_t i = 0, index = 0;
3746 
3747 	/* Revisit when we implement fully dynamic group allocation */
3748 	ASSERT(grp_index >= 0 && grp_index < mip->mi_rx_group_count);
3749 
3750 	rw_enter(&mip->mi_rw_lock, RW_READER);
3751 	*grp_num = mip->mi_rx_groups[grp_index].mrg_index;
3752 	*type = mip->mi_rx_groups[grp_index].mrg_type;
3753 	*n_rings = mip->mi_rx_groups[grp_index].mrg_cur_count;
3754 	for (mcip = mip->mi_rx_groups[grp_index].mrg_clients; mcip != NULL;
3755 	    mcip = mcip->mgc_next) {
3756 		int name_len = strlen(mcip->mgc_client->mci_name);
3757 
3758 		/*
3759 		 * MAXCLIENTNAMELEN is the buffer size reserved for client
3760 		 * names.
3761 		 * XXXX Formating the client name string needs to be moved
3762 		 * to user land when fixing the size of dhi_clnts in
3763 		 * dld_hwgrpinfo_t. We should use n_clients * client_name for
3764 		 * dhi_clntsin instead of MAXCLIENTNAMELEN
3765 		 */
3766 		if (index + name_len >= MAXCLIENTNAMELEN) {
3767 			index = MAXCLIENTNAMELEN;
3768 			break;
3769 		}
3770 		bcopy(mcip->mgc_client->mci_name, &(clnts_name[index]),
3771 		    name_len);
3772 		index += name_len;
3773 		clnts_name[index++] = ',';
3774 		i++;
3775 	}
3776 
3777 	/* Get rid of the last , */
3778 	if (index > 0)
3779 		clnts_name[index - 1] = '\0';
3780 	*n_clnts = i;
3781 	rw_exit(&mip->mi_rw_lock);
3782 }
3783 
3784 uint_t
3785 mac_hwgrp_num(mac_handle_t mh)
3786 {
3787 	mac_impl_t *mip = (mac_impl_t *)mh;
3788 
3789 	return (mip->mi_rx_group_count);
3790 }
3791