xref: /titanic_50/usr/src/uts/common/io/mac/mac_client.c (revision d6114e2d100d9ec3b45f9968d45ac2e3a0827af0)
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 2010 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 void *
414 mac_driver(mac_handle_t mh)
415 {
416 	return (((mac_impl_t *)mh)->mi_driver);
417 }
418 
419 const char *
420 mac_name(mac_handle_t mh)
421 {
422 	return (((mac_impl_t *)mh)->mi_name);
423 }
424 
425 int
426 mac_type(mac_handle_t mh)
427 {
428 	return (((mac_impl_t *)mh)->mi_type->mt_type);
429 }
430 
431 int
432 mac_nativetype(mac_handle_t mh)
433 {
434 	return (((mac_impl_t *)mh)->mi_type->mt_nativetype);
435 }
436 
437 char *
438 mac_client_name(mac_client_handle_t mch)
439 {
440 	return (((mac_client_impl_t *)mch)->mci_name);
441 }
442 
443 minor_t
444 mac_minor(mac_handle_t mh)
445 {
446 	return (((mac_impl_t *)mh)->mi_minor);
447 }
448 
449 /*
450  * Return the VID associated with a MAC client. This function should
451  * be called for clients which are associated with only one VID.
452  */
453 uint16_t
454 mac_client_vid(mac_client_handle_t mch)
455 {
456 	uint16_t		vid = VLAN_ID_NONE;
457 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
458 	flow_desc_t		flow_desc;
459 
460 	if (mcip->mci_nflents == 0)
461 		return (vid);
462 
463 	ASSERT(MCIP_DATAPATH_SETUP(mcip) && mac_client_single_rcvr(mcip));
464 
465 	mac_flow_get_desc(mcip->mci_flent, &flow_desc);
466 	if ((flow_desc.fd_mask & FLOW_LINK_VID) != 0)
467 		vid = flow_desc.fd_vid;
468 
469 	return (vid);
470 }
471 
472 /*
473  * Return whether the specified MAC client corresponds to a VLAN VNIC.
474  */
475 boolean_t
476 mac_client_is_vlan_vnic(mac_client_handle_t mch)
477 {
478 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
479 
480 	return (((mcip->mci_state_flags & MCIS_IS_VNIC) != 0) &&
481 	    ((mcip->mci_flent->fe_type & FLOW_PRIMARY_MAC) != 0));
482 }
483 
484 /*
485  * Return the link speed associated with the specified MAC client.
486  *
487  * The link speed of a MAC client is equal to the smallest value of
488  * 1) the current link speed of the underlying NIC, or
489  * 2) the bandwidth limit set for the MAC client.
490  *
491  * Note that the bandwidth limit can be higher than the speed
492  * of the underlying NIC. This is allowed to avoid spurious
493  * administration action failures or artifically lowering the
494  * bandwidth limit of a link that may  have temporarily lowered
495  * its link speed due to hardware problem or administrator action.
496  */
497 static uint64_t
498 mac_client_ifspeed(mac_client_impl_t *mcip)
499 {
500 	mac_impl_t *mip = mcip->mci_mip;
501 	uint64_t nic_speed;
502 
503 	nic_speed = mac_stat_get((mac_handle_t)mip, MAC_STAT_IFSPEED);
504 
505 	if (nic_speed == 0) {
506 		return (0);
507 	} else {
508 		uint64_t policy_limit = (uint64_t)-1;
509 
510 		if (MCIP_RESOURCE_PROPS_MASK(mcip) & MRP_MAXBW)
511 			policy_limit = MCIP_RESOURCE_PROPS_MAXBW(mcip);
512 
513 		return (MIN(policy_limit, nic_speed));
514 	}
515 }
516 
517 /*
518  * Return the link state of the specified client. If here are more
519  * than one clients of the underying mac_impl_t, the link state
520  * will always be UP regardless of the link state of the underlying
521  * mac_impl_t. This is needed to allow the MAC clients to continue
522  * to communicate with each other even when the physical link of
523  * their mac_impl_t is down.
524  */
525 static uint64_t
526 mac_client_link_state(mac_client_impl_t *mcip)
527 {
528 	mac_impl_t *mip = mcip->mci_mip;
529 	uint16_t vid;
530 	mac_client_impl_t *mci_list;
531 	mac_unicast_impl_t *mui_list, *oth_mui_list;
532 
533 	/*
534 	 * Returns LINK_STATE_UP if there are other MAC clients defined on
535 	 * mac_impl_t which share same VLAN ID as that of mcip. Note that
536 	 * if 'mcip' has more than one VID's then we match ANY one of the
537 	 * VID's with other MAC client's VID's and return LINK_STATE_UP.
538 	 */
539 	rw_enter(&mcip->mci_rw_lock, RW_READER);
540 	for (mui_list = mcip->mci_unicast_list; mui_list != NULL;
541 	    mui_list = mui_list->mui_next) {
542 		vid = mui_list->mui_vid;
543 		for (mci_list = mip->mi_clients_list; mci_list != NULL;
544 		    mci_list = mci_list->mci_client_next) {
545 			if (mci_list == mcip)
546 				continue;
547 			for (oth_mui_list = mci_list->mci_unicast_list;
548 			    oth_mui_list != NULL; oth_mui_list = oth_mui_list->
549 			    mui_next) {
550 				if (vid == oth_mui_list->mui_vid) {
551 					rw_exit(&mcip->mci_rw_lock);
552 					return (LINK_STATE_UP);
553 				}
554 			}
555 		}
556 	}
557 	rw_exit(&mcip->mci_rw_lock);
558 
559 	return (mac_stat_get((mac_handle_t)mip, MAC_STAT_LINK_STATE));
560 }
561 
562 /*
563  * Return the statistics of a MAC client. These statistics are different
564  * then the statistics of the underlying MAC which are returned by
565  * mac_stat_get().
566  */
567 uint64_t
568 mac_client_stat_get(mac_client_handle_t mch, uint_t stat)
569 {
570 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
571 	mac_impl_t *mip = mcip->mci_mip;
572 	uint64_t val;
573 
574 	switch (stat) {
575 	case MAC_STAT_LINK_STATE:
576 		val = mac_client_link_state(mcip);
577 		break;
578 	case MAC_STAT_LINK_UP:
579 		val = (mac_client_link_state(mcip) == LINK_STATE_UP);
580 		break;
581 	case MAC_STAT_PROMISC:
582 		val = mac_stat_get((mac_handle_t)mip, MAC_STAT_PROMISC);
583 		break;
584 	case MAC_STAT_LOWLINK_STATE:
585 		val = mac_stat_get((mac_handle_t)mip, MAC_STAT_LOWLINK_STATE);
586 		break;
587 	case MAC_STAT_IFSPEED:
588 		val = mac_client_ifspeed(mcip);
589 		break;
590 	case MAC_STAT_MULTIRCV:
591 		val = mcip->mci_stat_multircv;
592 		break;
593 	case MAC_STAT_BRDCSTRCV:
594 		val = mcip->mci_stat_brdcstrcv;
595 		break;
596 	case MAC_STAT_MULTIXMT:
597 		val = mcip->mci_stat_multixmt;
598 		break;
599 	case MAC_STAT_BRDCSTXMT:
600 		val = mcip->mci_stat_brdcstxmt;
601 		break;
602 	case MAC_STAT_OBYTES:
603 		val = mcip->mci_stat_obytes;
604 		break;
605 	case MAC_STAT_OPACKETS:
606 		val = mcip->mci_stat_opackets;
607 		break;
608 	case MAC_STAT_OERRORS:
609 		val = mcip->mci_stat_oerrors;
610 		break;
611 	case MAC_STAT_IPACKETS:
612 		val = mcip->mci_stat_ipackets;
613 		break;
614 	case MAC_STAT_RBYTES:
615 		val = mcip->mci_stat_ibytes;
616 		break;
617 	case MAC_STAT_IERRORS:
618 		val = mcip->mci_stat_ierrors;
619 		break;
620 	default:
621 		val = mac_stat_default(mip, stat);
622 		break;
623 	}
624 
625 	return (val);
626 }
627 
628 /*
629  * Return the statistics of the specified MAC instance.
630  */
631 uint64_t
632 mac_stat_get(mac_handle_t mh, uint_t stat)
633 {
634 	mac_impl_t	*mip = (mac_impl_t *)mh;
635 	uint64_t	val;
636 	int		ret;
637 
638 	/*
639 	 * The range of stat determines where it is maintained.  Stat
640 	 * values from 0 up to (but not including) MAC_STAT_MIN are
641 	 * mainteined by the mac module itself.  Everything else is
642 	 * maintained by the driver.
643 	 *
644 	 * If the mac_impl_t being queried corresponds to a VNIC,
645 	 * the stats need to be queried from the lower MAC client
646 	 * corresponding to the VNIC. (The mac_link_update()
647 	 * invoked by the driver to the lower MAC causes the *lower
648 	 * MAC* to update its mi_linkstate, and send a notification
649 	 * to its MAC clients. Due to the VNIC passthrough,
650 	 * these notifications are sent to the upper MAC clients
651 	 * of the VNIC directly, and the upper mac_impl_t of the VNIC
652 	 * does not have a valid mi_linkstate.
653 	 */
654 	if (stat < MAC_STAT_MIN && !(mip->mi_state_flags & MIS_IS_VNIC)) {
655 		/* these stats are maintained by the mac module itself */
656 		switch (stat) {
657 		case MAC_STAT_LINK_STATE:
658 			return (mip->mi_linkstate);
659 		case MAC_STAT_LINK_UP:
660 			return (mip->mi_linkstate == LINK_STATE_UP);
661 		case MAC_STAT_PROMISC:
662 			return (mip->mi_devpromisc != 0);
663 		case MAC_STAT_LOWLINK_STATE:
664 			return (mip->mi_lowlinkstate);
665 		default:
666 			ASSERT(B_FALSE);
667 		}
668 	}
669 
670 	/*
671 	 * Call the driver to get the given statistic.
672 	 */
673 	ret = mip->mi_getstat(mip->mi_driver, stat, &val);
674 	if (ret != 0) {
675 		/*
676 		 * The driver doesn't support this statistic.  Get the
677 		 * statistic's default value.
678 		 */
679 		val = mac_stat_default(mip, stat);
680 	}
681 	return (val);
682 }
683 
684 /*
685  * Utility function which returns the VID associated with a flow entry.
686  */
687 uint16_t
688 i_mac_flow_vid(flow_entry_t *flent)
689 {
690 	flow_desc_t	flow_desc;
691 
692 	mac_flow_get_desc(flent, &flow_desc);
693 
694 	if ((flow_desc.fd_mask & FLOW_LINK_VID) != 0)
695 		return (flow_desc.fd_vid);
696 	return (VLAN_ID_NONE);
697 }
698 
699 /*
700  * Verify the validity of the specified unicast MAC address. Returns B_TRUE
701  * if the address is valid, B_FALSE otherwise (multicast address, or incorrect
702  * length.
703  */
704 boolean_t
705 mac_unicst_verify(mac_handle_t mh, const uint8_t *addr, uint_t len)
706 {
707 	mac_impl_t	*mip = (mac_impl_t *)mh;
708 
709 	/*
710 	 * Verify the address. No lock is needed since mi_type and plugin
711 	 * details don't change after mac_register().
712 	 */
713 	if ((len != mip->mi_type->mt_addr_length) ||
714 	    (mip->mi_type->mt_ops.mtops_unicst_verify(addr,
715 	    mip->mi_pdata)) != 0) {
716 		return (B_FALSE);
717 	} else {
718 		return (B_TRUE);
719 	}
720 }
721 
722 void
723 mac_sdu_get(mac_handle_t mh, uint_t *min_sdu, uint_t *max_sdu)
724 {
725 	mac_impl_t	*mip = (mac_impl_t *)mh;
726 
727 	if (min_sdu != NULL)
728 		*min_sdu = mip->mi_sdu_min;
729 	if (max_sdu != NULL)
730 		*max_sdu = mip->mi_sdu_max;
731 }
732 
733 /*
734  * Update the MAC unicast address of the specified client's flows. Currently
735  * only one unicast MAC unicast address is allowed per client.
736  */
737 static void
738 mac_unicast_update_client_flow(mac_client_impl_t *mcip)
739 {
740 	mac_impl_t *mip = mcip->mci_mip;
741 	flow_entry_t *flent = mcip->mci_flent;
742 	mac_address_t *map = mcip->mci_unicast;
743 	flow_desc_t flow_desc;
744 
745 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
746 	ASSERT(flent != NULL);
747 
748 	mac_flow_get_desc(flent, &flow_desc);
749 	ASSERT(flow_desc.fd_mask & FLOW_LINK_DST);
750 
751 	bcopy(map->ma_addr, flow_desc.fd_dst_mac, map->ma_len);
752 	mac_flow_set_desc(flent, &flow_desc);
753 
754 	/*
755 	 * A MAC client could have one MAC address but multiple
756 	 * VLANs. In that case update the flow entries corresponding
757 	 * to all VLANs of the MAC client.
758 	 */
759 	for (flent = mcip->mci_flent_list; flent != NULL;
760 	    flent = flent->fe_client_next) {
761 		mac_flow_get_desc(flent, &flow_desc);
762 		if (!(flent->fe_type & FLOW_PRIMARY_MAC ||
763 		    flent->fe_type & FLOW_VNIC_MAC))
764 			continue;
765 
766 		bcopy(map->ma_addr, flow_desc.fd_dst_mac, map->ma_len);
767 		mac_flow_set_desc(flent, &flow_desc);
768 	}
769 }
770 
771 /*
772  * Update all clients that share the same unicast address.
773  */
774 void
775 mac_unicast_update_clients(mac_impl_t *mip, mac_address_t *map)
776 {
777 	mac_client_impl_t *mcip;
778 
779 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
780 
781 	/*
782 	 * Find all clients that share the same unicast MAC address and update
783 	 * them appropriately.
784 	 */
785 	for (mcip = mip->mi_clients_list; mcip != NULL;
786 	    mcip = mcip->mci_client_next) {
787 		/*
788 		 * Ignore clients that don't share this MAC address.
789 		 */
790 		if (map != mcip->mci_unicast)
791 			continue;
792 
793 		/*
794 		 * Update those clients with same old unicast MAC address.
795 		 */
796 		mac_unicast_update_client_flow(mcip);
797 	}
798 }
799 
800 /*
801  * Update the unicast MAC address of the specified VNIC MAC client.
802  *
803  * Check whether the operation is valid. Any of following cases should fail:
804  *
805  * 1. It's a VLAN type of VNIC.
806  * 2. The new value is current "primary" MAC address.
807  * 3. The current MAC address is shared with other clients.
808  * 4. The new MAC address has been used. This case will be valid when
809  *    client migration is fully supported.
810  */
811 int
812 mac_vnic_unicast_set(mac_client_handle_t mch, const uint8_t *addr)
813 {
814 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
815 	mac_impl_t *mip = mcip->mci_mip;
816 	mac_address_t *map = mcip->mci_unicast;
817 	int err;
818 
819 	ASSERT(!(mip->mi_state_flags & MIS_IS_VNIC));
820 	ASSERT(mcip->mci_state_flags & MCIS_IS_VNIC);
821 	ASSERT(mcip->mci_flags != MAC_CLIENT_FLAGS_PRIMARY);
822 
823 	i_mac_perim_enter(mip);
824 
825 	/*
826 	 * If this is a VLAN type of VNIC, it's using "primary" MAC address
827 	 * of the underlying interface. Must fail here. Refer to case 1 above.
828 	 */
829 	if (bcmp(map->ma_addr, mip->mi_addr, map->ma_len) == 0) {
830 		i_mac_perim_exit(mip);
831 		return (ENOTSUP);
832 	}
833 
834 	/*
835 	 * If the new address is the "primary" one, must fail. Refer to
836 	 * case 2 above.
837 	 */
838 	if (bcmp(addr, mip->mi_addr, map->ma_len) == 0) {
839 		i_mac_perim_exit(mip);
840 		return (EACCES);
841 	}
842 
843 	/*
844 	 * If the address is shared by multiple clients, must fail. Refer
845 	 * to case 3 above.
846 	 */
847 	if (mac_check_macaddr_shared(map)) {
848 		i_mac_perim_exit(mip);
849 		return (EBUSY);
850 	}
851 
852 	/*
853 	 * If the new address has been used, must fail for now. Refer to
854 	 * case 4 above.
855 	 */
856 	if (mac_find_macaddr(mip, (uint8_t *)addr) != NULL) {
857 		i_mac_perim_exit(mip);
858 		return (ENOTSUP);
859 	}
860 
861 	/*
862 	 * Update the MAC address.
863 	 */
864 	err = mac_update_macaddr(map, (uint8_t *)addr);
865 
866 	if (err != 0) {
867 		i_mac_perim_exit(mip);
868 		return (err);
869 	}
870 
871 	/*
872 	 * Update all flows of this MAC client.
873 	 */
874 	mac_unicast_update_client_flow(mcip);
875 
876 	i_mac_perim_exit(mip);
877 	return (0);
878 }
879 
880 /*
881  * Program the new primary unicast address of the specified MAC.
882  *
883  * Function mac_update_macaddr() takes care different types of underlying
884  * MAC. If the underlying MAC is VNIC, the VNIC driver must have registerd
885  * mi_unicst() entry point, that indirectly calls mac_vnic_unicast_set()
886  * which will take care of updating the MAC address of the corresponding
887  * MAC client.
888  *
889  * This is the only interface that allow the client to update the "primary"
890  * MAC address of the underlying MAC. The new value must have not been
891  * used by other clients.
892  */
893 int
894 mac_unicast_primary_set(mac_handle_t mh, const uint8_t *addr)
895 {
896 	mac_impl_t *mip = (mac_impl_t *)mh;
897 	mac_address_t *map;
898 	int err;
899 
900 	/* verify the address validity */
901 	if (!mac_unicst_verify(mh, addr, mip->mi_type->mt_addr_length))
902 		return (EINVAL);
903 
904 	i_mac_perim_enter(mip);
905 
906 	/*
907 	 * If the new value is the same as the current primary address value,
908 	 * there's nothing to do.
909 	 */
910 	if (bcmp(addr, mip->mi_addr, mip->mi_type->mt_addr_length) == 0) {
911 		i_mac_perim_exit(mip);
912 		return (0);
913 	}
914 
915 	if (mac_find_macaddr(mip, (uint8_t *)addr) != 0) {
916 		i_mac_perim_exit(mip);
917 		return (EBUSY);
918 	}
919 
920 	map = mac_find_macaddr(mip, mip->mi_addr);
921 	ASSERT(map != NULL);
922 
923 	/*
924 	 * Update the MAC address.
925 	 */
926 	if (mip->mi_state_flags & MIS_IS_AGGR) {
927 		mac_capab_aggr_t aggr_cap;
928 
929 		/*
930 		 * If the mac is an aggregation, other than the unicast
931 		 * addresses programming, aggr must be informed about this
932 		 * primary unicst address change to change its mac address
933 		 * policy to be user-specified.
934 		 */
935 		ASSERT(map->ma_type == MAC_ADDRESS_TYPE_UNICAST_CLASSIFIED);
936 		VERIFY(i_mac_capab_get(mh, MAC_CAPAB_AGGR, &aggr_cap));
937 		err = aggr_cap.mca_unicst(mip->mi_driver, addr);
938 		if (err == 0)
939 			bcopy(addr, map->ma_addr, map->ma_len);
940 	} else {
941 		err = mac_update_macaddr(map, (uint8_t *)addr);
942 	}
943 
944 	if (err != 0) {
945 		i_mac_perim_exit(mip);
946 		return (err);
947 	}
948 
949 	mac_unicast_update_clients(mip, map);
950 
951 	/*
952 	 * Save the new primary MAC address in mac_impl_t.
953 	 */
954 	bcopy(addr, mip->mi_addr, mip->mi_type->mt_addr_length);
955 
956 	i_mac_perim_exit(mip);
957 
958 	if (err == 0)
959 		i_mac_notify(mip, MAC_NOTE_UNICST);
960 
961 	return (err);
962 }
963 
964 /*
965  * Return the current primary MAC address of the specified MAC.
966  */
967 void
968 mac_unicast_primary_get(mac_handle_t mh, uint8_t *addr)
969 {
970 	mac_impl_t *mip = (mac_impl_t *)mh;
971 
972 	rw_enter(&mip->mi_rw_lock, RW_READER);
973 	bcopy(mip->mi_addr, addr, mip->mi_type->mt_addr_length);
974 	rw_exit(&mip->mi_rw_lock);
975 }
976 
977 /*
978  * Return information about the use of the primary MAC address of the
979  * specified MAC instance:
980  *
981  * - if client_name is non-NULL, it must point to a string of at
982  *   least MAXNAMELEN bytes, and will be set to the name of the MAC
983  *   client which uses the primary MAC address.
984  *
985  * - if in_use is non-NULL, used to return whether the primary MAC
986  *   address is currently in use.
987  */
988 void
989 mac_unicast_primary_info(mac_handle_t mh, char *client_name, boolean_t *in_use)
990 {
991 	mac_impl_t *mip = (mac_impl_t *)mh;
992 	mac_client_impl_t *cur_client;
993 
994 	if (in_use != NULL)
995 		*in_use = B_FALSE;
996 	if (client_name != NULL)
997 		bzero(client_name, MAXNAMELEN);
998 
999 	/*
1000 	 * The mi_rw_lock is used to protect threads that don't hold the
1001 	 * mac perimeter to get a consistent view of the mi_clients_list.
1002 	 * Threads that modify the list must hold both the mac perimeter and
1003 	 * mi_rw_lock(RW_WRITER)
1004 	 */
1005 	rw_enter(&mip->mi_rw_lock, RW_READER);
1006 	for (cur_client = mip->mi_clients_list; cur_client != NULL;
1007 	    cur_client = cur_client->mci_client_next) {
1008 		if (mac_is_primary_client(cur_client) ||
1009 		    (mip->mi_state_flags & MIS_IS_VNIC)) {
1010 			rw_exit(&mip->mi_rw_lock);
1011 			if (in_use != NULL)
1012 				*in_use = B_TRUE;
1013 			if (client_name != NULL) {
1014 				bcopy(cur_client->mci_name, client_name,
1015 				    MAXNAMELEN);
1016 			}
1017 			return;
1018 		}
1019 	}
1020 	rw_exit(&mip->mi_rw_lock);
1021 }
1022 
1023 /*
1024  * Return the current destination MAC address of the specified MAC.
1025  */
1026 boolean_t
1027 mac_dst_get(mac_handle_t mh, uint8_t *addr)
1028 {
1029 	mac_impl_t *mip = (mac_impl_t *)mh;
1030 
1031 	rw_enter(&mip->mi_rw_lock, RW_READER);
1032 	if (mip->mi_dstaddr_set)
1033 		bcopy(mip->mi_dstaddr, addr, mip->mi_type->mt_addr_length);
1034 	rw_exit(&mip->mi_rw_lock);
1035 	return (mip->mi_dstaddr_set);
1036 }
1037 
1038 /*
1039  * Add the specified MAC client to the list of clients which opened
1040  * the specified MAC.
1041  */
1042 static void
1043 mac_client_add(mac_client_impl_t *mcip)
1044 {
1045 	mac_impl_t *mip = mcip->mci_mip;
1046 
1047 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1048 
1049 	/* add VNIC to the front of the list */
1050 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
1051 	mcip->mci_client_next = mip->mi_clients_list;
1052 	mip->mi_clients_list = mcip;
1053 	mip->mi_nclients++;
1054 	rw_exit(&mip->mi_rw_lock);
1055 }
1056 
1057 /*
1058  * Remove the specified MAC client from the list of clients which opened
1059  * the specified MAC.
1060  */
1061 static void
1062 mac_client_remove(mac_client_impl_t *mcip)
1063 {
1064 	mac_impl_t *mip = mcip->mci_mip;
1065 	mac_client_impl_t **prev, *cclient;
1066 
1067 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1068 
1069 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
1070 	prev = &mip->mi_clients_list;
1071 	cclient = *prev;
1072 	while (cclient != NULL && cclient != mcip) {
1073 		prev = &cclient->mci_client_next;
1074 		cclient = *prev;
1075 	}
1076 	ASSERT(cclient != NULL);
1077 	*prev = cclient->mci_client_next;
1078 	mip->mi_nclients--;
1079 	rw_exit(&mip->mi_rw_lock);
1080 }
1081 
1082 static mac_unicast_impl_t *
1083 mac_client_find_vid(mac_client_impl_t *mcip, uint16_t vid)
1084 {
1085 	mac_unicast_impl_t *muip = mcip->mci_unicast_list;
1086 
1087 	while ((muip != NULL) && (muip->mui_vid != vid))
1088 		muip = muip->mui_next;
1089 
1090 	return (muip);
1091 }
1092 
1093 /*
1094  * Return whether the specified (MAC address, VID) tuple is already used by
1095  * one of the MAC clients associated with the specified MAC.
1096  */
1097 static boolean_t
1098 mac_addr_in_use(mac_impl_t *mip, uint8_t *mac_addr, uint16_t vid)
1099 {
1100 	mac_client_impl_t *client;
1101 	mac_address_t *map;
1102 
1103 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1104 
1105 	for (client = mip->mi_clients_list; client != NULL;
1106 	    client = client->mci_client_next) {
1107 
1108 		/*
1109 		 * Ignore clients that don't have unicast address.
1110 		 */
1111 		if (client->mci_unicast_list == NULL)
1112 			continue;
1113 
1114 		map = client->mci_unicast;
1115 
1116 		if ((bcmp(mac_addr, map->ma_addr, map->ma_len) == 0) &&
1117 		    (mac_client_find_vid(client, vid) != NULL)) {
1118 			return (B_TRUE);
1119 		}
1120 	}
1121 
1122 	return (B_FALSE);
1123 }
1124 
1125 /*
1126  * Generate a random MAC address. The MAC address prefix is
1127  * stored in the array pointed to by mac_addr, and its length, in bytes,
1128  * is specified by prefix_len. The least significant bits
1129  * after prefix_len bytes are generated, and stored after the prefix
1130  * in the mac_addr array.
1131  */
1132 int
1133 mac_addr_random(mac_client_handle_t mch, uint_t prefix_len,
1134     uint8_t *mac_addr, mac_diag_t *diag)
1135 {
1136 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1137 	mac_impl_t *mip = mcip->mci_mip;
1138 	size_t addr_len = mip->mi_type->mt_addr_length;
1139 
1140 	if (prefix_len >= addr_len) {
1141 		*diag = MAC_DIAG_MACPREFIXLEN_INVALID;
1142 		return (EINVAL);
1143 	}
1144 
1145 	/* check the prefix value */
1146 	if (prefix_len > 0) {
1147 		bzero(mac_addr + prefix_len, addr_len - prefix_len);
1148 		if (!mac_unicst_verify((mac_handle_t)mip, mac_addr,
1149 		    addr_len)) {
1150 			*diag = MAC_DIAG_MACPREFIX_INVALID;
1151 			return (EINVAL);
1152 		}
1153 	}
1154 
1155 	/* generate the MAC address */
1156 	if (prefix_len < addr_len) {
1157 		(void) random_get_pseudo_bytes(mac_addr +
1158 		    prefix_len, addr_len - prefix_len);
1159 	}
1160 
1161 	*diag = 0;
1162 	return (0);
1163 }
1164 
1165 /*
1166  * Set the priority range for this MAC client. This will be used to
1167  * determine the absolute priority for the threads created for this
1168  * MAC client using the specified "low", "medium" and "high" level.
1169  * This will also be used for any subflows on this MAC client.
1170  */
1171 #define	MAC_CLIENT_SET_PRIORITY_RANGE(mcip, pri) {			\
1172 	(mcip)->mci_min_pri = FLOW_MIN_PRIORITY(MINCLSYSPRI,	\
1173 	    MAXCLSYSPRI, (pri));					\
1174 	(mcip)->mci_max_pri = FLOW_MAX_PRIORITY(MINCLSYSPRI,	\
1175 	    MAXCLSYSPRI, (mcip)->mci_min_pri);				\
1176 	}
1177 
1178 /*
1179  * MAC client open entry point. Return a new MAC client handle. Each
1180  * MAC client is associated with a name, specified through the 'name'
1181  * argument.
1182  */
1183 int
1184 mac_client_open(mac_handle_t mh, mac_client_handle_t *mchp, char *name,
1185     uint16_t flags)
1186 {
1187 	mac_impl_t *mip = (mac_impl_t *)mh;
1188 	mac_client_impl_t *mcip;
1189 	int err = 0;
1190 	boolean_t share_desired =
1191 	    ((flags & MAC_OPEN_FLAGS_SHARES_DESIRED) != 0);
1192 	boolean_t no_hwrings = ((flags & MAC_OPEN_FLAGS_NO_HWRINGS) != 0);
1193 	boolean_t req_hwrings = ((flags & MAC_OPEN_FLAGS_REQ_HWRINGS) != 0);
1194 	flow_entry_t	*flent = NULL;
1195 
1196 	*mchp = NULL;
1197 	if (share_desired && no_hwrings) {
1198 		/* can't have shares but no hardware rings */
1199 		return (EINVAL);
1200 	}
1201 
1202 	i_mac_perim_enter(mip);
1203 
1204 	if (mip->mi_state_flags & MIS_IS_VNIC) {
1205 		/*
1206 		 * The underlying MAC is a VNIC. Return the MAC client
1207 		 * handle of the lower MAC which was obtained by
1208 		 * the VNIC driver when it did its mac_client_open().
1209 		 */
1210 
1211 		mcip = mac_vnic_lower(mip);
1212 
1213 		/*
1214 		 * Note that multiple mac clients share the same mcip in
1215 		 * this case.
1216 		 */
1217 		if (flags & MAC_OPEN_FLAGS_EXCLUSIVE)
1218 			mcip->mci_state_flags |= MCIS_EXCLUSIVE;
1219 
1220 		if (flags & MAC_OPEN_FLAGS_MULTI_PRIMARY)
1221 			mcip->mci_flags |= MAC_CLIENT_FLAGS_MULTI_PRIMARY;
1222 
1223 		mip->mi_clients_list = mcip;
1224 		i_mac_perim_exit(mip);
1225 		*mchp = (mac_client_handle_t)mcip;
1226 		return (err);
1227 	}
1228 
1229 	mcip = kmem_cache_alloc(mac_client_impl_cache, KM_SLEEP);
1230 
1231 	mcip->mci_mip = mip;
1232 	mcip->mci_upper_mip = NULL;
1233 	mcip->mci_rx_fn = mac_pkt_drop;
1234 	mcip->mci_rx_arg = NULL;
1235 	mcip->mci_rx_p_fn = NULL;
1236 	mcip->mci_rx_p_arg = NULL;
1237 	mcip->mci_p_unicast_list = NULL;
1238 	mcip->mci_direct_rx_fn = NULL;
1239 	mcip->mci_direct_rx_arg = NULL;
1240 
1241 	mcip->mci_unicast_list = NULL;
1242 
1243 	if ((flags & MAC_OPEN_FLAGS_IS_VNIC) != 0)
1244 		mcip->mci_state_flags |= MCIS_IS_VNIC;
1245 
1246 	if ((flags & MAC_OPEN_FLAGS_EXCLUSIVE) != 0)
1247 		mcip->mci_state_flags |= MCIS_EXCLUSIVE;
1248 
1249 	if ((flags & MAC_OPEN_FLAGS_IS_AGGR_PORT) != 0)
1250 		mcip->mci_state_flags |= MCIS_IS_AGGR_PORT;
1251 
1252 	if ((flags & MAC_OPEN_FLAGS_USE_DATALINK_NAME) != 0) {
1253 		datalink_id_t	linkid;
1254 
1255 		ASSERT(name == NULL);
1256 		if ((err = dls_devnet_macname2linkid(mip->mi_name,
1257 		    &linkid)) != 0) {
1258 			goto done;
1259 		}
1260 		if ((err = dls_mgmt_get_linkinfo(linkid, mcip->mci_name, NULL,
1261 		    NULL, NULL)) != 0) {
1262 			/*
1263 			 * Use mac name if dlmgmtd is not available.
1264 			 */
1265 			if (err == EBADF) {
1266 				(void) strlcpy(mcip->mci_name, mip->mi_name,
1267 				    sizeof (mcip->mci_name));
1268 				err = 0;
1269 			} else {
1270 				goto done;
1271 			}
1272 		}
1273 		mcip->mci_state_flags |= MCIS_USE_DATALINK_NAME;
1274 	} else {
1275 		ASSERT(name != NULL);
1276 		if (strlen(name) > MAXNAMELEN) {
1277 			err = EINVAL;
1278 			goto done;
1279 		}
1280 		(void) strlcpy(mcip->mci_name, name, sizeof (mcip->mci_name));
1281 	}
1282 
1283 	if (flags & MAC_OPEN_FLAGS_MULTI_PRIMARY)
1284 		mcip->mci_flags |= MAC_CLIENT_FLAGS_MULTI_PRIMARY;
1285 
1286 	/* the subflow table will be created dynamically */
1287 	mcip->mci_subflow_tab = NULL;
1288 	mcip->mci_stat_multircv = 0;
1289 	mcip->mci_stat_brdcstrcv = 0;
1290 	mcip->mci_stat_multixmt = 0;
1291 	mcip->mci_stat_brdcstxmt = 0;
1292 
1293 	mcip->mci_stat_obytes = 0;
1294 	mcip->mci_stat_opackets = 0;
1295 	mcip->mci_stat_oerrors = 0;
1296 	mcip->mci_stat_ibytes = 0;
1297 	mcip->mci_stat_ipackets = 0;
1298 	mcip->mci_stat_ierrors = 0;
1299 
1300 	/* Create an initial flow */
1301 
1302 	err = mac_flow_create(NULL, NULL, mcip->mci_name, NULL,
1303 	    mcip->mci_state_flags & MCIS_IS_VNIC ? FLOW_VNIC_MAC :
1304 	    FLOW_PRIMARY_MAC, &flent);
1305 	if (err != 0)
1306 		goto done;
1307 	mcip->mci_flent = flent;
1308 	FLOW_MARK(flent, FE_MC_NO_DATAPATH);
1309 	flent->fe_mcip = mcip;
1310 	/*
1311 	 * Place initial creation reference on the flow. This reference
1312 	 * is released in the corresponding delete action viz.
1313 	 * mac_unicast_remove after waiting for all transient refs to
1314 	 * to go away. The wait happens in mac_flow_wait.
1315 	 */
1316 	FLOW_REFHOLD(flent);
1317 
1318 	/*
1319 	 * Do this ahead of the mac_bcast_add() below so that the mi_nclients
1320 	 * will have the right value for mac_rx_srs_setup().
1321 	 */
1322 	mac_client_add(mcip);
1323 
1324 	if (no_hwrings)
1325 		mcip->mci_state_flags |= MCIS_NO_HWRINGS;
1326 	if (req_hwrings)
1327 		mcip->mci_state_flags |= MCIS_REQ_HWRINGS;
1328 	mcip->mci_share = NULL;
1329 	if (share_desired) {
1330 		ASSERT(!no_hwrings);
1331 		i_mac_share_alloc(mcip);
1332 	}
1333 
1334 	DTRACE_PROBE2(mac__client__open__allocated, mac_impl_t *,
1335 	    mcip->mci_mip, mac_client_impl_t *, mcip);
1336 	*mchp = (mac_client_handle_t)mcip;
1337 
1338 	i_mac_perim_exit(mip);
1339 	return (0);
1340 
1341 done:
1342 	i_mac_perim_exit(mip);
1343 	mcip->mci_state_flags = 0;
1344 	mcip->mci_tx_flag = 0;
1345 	kmem_cache_free(mac_client_impl_cache, mcip);
1346 	return (err);
1347 }
1348 
1349 /*
1350  * Close the specified MAC client handle.
1351  */
1352 void
1353 mac_client_close(mac_client_handle_t mch, uint16_t flags)
1354 {
1355 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
1356 	mac_impl_t		*mip = mcip->mci_mip;
1357 	flow_entry_t		*flent;
1358 
1359 	i_mac_perim_enter(mip);
1360 
1361 	if (flags & MAC_CLOSE_FLAGS_EXCLUSIVE)
1362 		mcip->mci_state_flags &= ~MCIS_EXCLUSIVE;
1363 
1364 	if ((mcip->mci_state_flags & MCIS_IS_VNIC) &&
1365 	    !(flags & MAC_CLOSE_FLAGS_IS_VNIC)) {
1366 		/*
1367 		 * This is an upper VNIC client initiated operation.
1368 		 * The lower MAC client will be closed by the VNIC driver
1369 		 * when the VNIC is deleted.
1370 		 */
1371 
1372 		i_mac_perim_exit(mip);
1373 		return;
1374 	}
1375 
1376 	/*
1377 	 * Remove the flent associated with the MAC client
1378 	 */
1379 	flent = mcip->mci_flent;
1380 	mcip->mci_flent = NULL;
1381 	FLOW_FINAL_REFRELE(flent);
1382 
1383 	/*
1384 	 * MAC clients must remove the unicast addresses and promisc callbacks
1385 	 * they added before issuing a mac_client_close().
1386 	 */
1387 	ASSERT(mcip->mci_unicast_list == NULL);
1388 	ASSERT(mcip->mci_promisc_list == NULL);
1389 	ASSERT(mcip->mci_tx_notify_cb_list == NULL);
1390 
1391 	i_mac_share_free(mcip);
1392 
1393 	mac_client_remove(mcip);
1394 
1395 	i_mac_perim_exit(mip);
1396 	mcip->mci_subflow_tab = NULL;
1397 	mcip->mci_state_flags = 0;
1398 	mcip->mci_tx_flag = 0;
1399 	kmem_cache_free(mac_client_impl_cache, mch);
1400 }
1401 
1402 /*
1403  * Set the rx bypass receive callback.
1404  */
1405 boolean_t
1406 mac_rx_bypass_set(mac_client_handle_t mch, mac_direct_rx_t rx_fn, void *arg1)
1407 {
1408 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
1409 	mac_impl_t		*mip = mcip->mci_mip;
1410 
1411 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1412 
1413 	/*
1414 	 * If the mac_client is a VLAN, we should not do DLS bypass and
1415 	 * instead let the packets come up via mac_rx_deliver so the vlan
1416 	 * header can be stripped.
1417 	 */
1418 	if (mcip->mci_nvids > 0)
1419 		return (B_FALSE);
1420 
1421 	/*
1422 	 * These are not accessed directly in the data path, and hence
1423 	 * don't need any protection
1424 	 */
1425 	mcip->mci_direct_rx_fn = rx_fn;
1426 	mcip->mci_direct_rx_arg = arg1;
1427 	return (B_TRUE);
1428 }
1429 
1430 /*
1431  * Enable/Disable rx bypass. By default, bypass is assumed to be enabled.
1432  */
1433 void
1434 mac_rx_bypass_enable(mac_client_handle_t mch)
1435 {
1436 	((mac_client_impl_t *)mch)->mci_state_flags &= ~MCIS_RX_BYPASS_DISABLE;
1437 }
1438 
1439 void
1440 mac_rx_bypass_disable(mac_client_handle_t mch)
1441 {
1442 	((mac_client_impl_t *)mch)->mci_state_flags |= MCIS_RX_BYPASS_DISABLE;
1443 }
1444 
1445 /*
1446  * Set the receive callback for the specified MAC client. There can be
1447  * at most one such callback per MAC client.
1448  */
1449 void
1450 mac_rx_set(mac_client_handle_t mch, mac_rx_t rx_fn, void *arg)
1451 {
1452 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1453 	mac_impl_t	*mip = mcip->mci_mip;
1454 
1455 	/*
1456 	 * Instead of adding an extra set of locks and refcnts in
1457 	 * the datapath at the mac client boundary, we temporarily quiesce
1458 	 * the SRS and related entities. We then change the receive function
1459 	 * without interference from any receive data thread and then reenable
1460 	 * the data flow subsequently.
1461 	 */
1462 	i_mac_perim_enter(mip);
1463 	mac_rx_client_quiesce(mch);
1464 
1465 	mcip->mci_rx_fn = rx_fn;
1466 	mcip->mci_rx_arg = arg;
1467 	mac_rx_client_restart(mch);
1468 	i_mac_perim_exit(mip);
1469 }
1470 
1471 /*
1472  * Reset the receive callback for the specified MAC client.
1473  */
1474 void
1475 mac_rx_clear(mac_client_handle_t mch)
1476 {
1477 	mac_rx_set(mch, mac_pkt_drop, NULL);
1478 }
1479 
1480 /*
1481  * Walk the MAC client subflow table and updates their priority values.
1482  */
1483 static int
1484 mac_update_subflow_priority_cb(flow_entry_t *flent, void *arg)
1485 {
1486 	mac_flow_update_priority(arg, flent);
1487 	return (0);
1488 }
1489 
1490 void
1491 mac_update_subflow_priority(mac_client_impl_t *mcip)
1492 {
1493 	(void) mac_flow_walk(mcip->mci_subflow_tab,
1494 	    mac_update_subflow_priority_cb, mcip);
1495 }
1496 
1497 /*
1498  * When the MAC client is being brought up (i.e. we do a unicast_add) we need
1499  * to initialize the cpu and resource control structure in the
1500  * mac_client_impl_t from the mac_impl_t (i.e if there are any cached
1501  * properties before the flow entry for the unicast address was created).
1502  */
1503 int
1504 mac_resource_ctl_set(mac_client_handle_t mch, mac_resource_props_t *mrp)
1505 {
1506 	mac_client_impl_t 	*mcip = (mac_client_impl_t *)mch;
1507 	mac_impl_t		*mip = (mac_impl_t *)mcip->mci_mip;
1508 	int			err = 0;
1509 
1510 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1511 
1512 	err = mac_validate_props(mrp);
1513 	if (err != 0)
1514 		return (err);
1515 
1516 	mac_update_resources(mrp, MCIP_RESOURCE_PROPS(mcip), B_FALSE);
1517 	if (MCIP_DATAPATH_SETUP(mcip)) {
1518 		/*
1519 		 * We have to set this prior to calling mac_flow_modify.
1520 		 */
1521 		if (mrp->mrp_mask & MRP_PRIORITY) {
1522 			if (mrp->mrp_priority == MPL_RESET) {
1523 				MAC_CLIENT_SET_PRIORITY_RANGE(mcip,
1524 				    MPL_LINK_DEFAULT);
1525 			} else {
1526 				MAC_CLIENT_SET_PRIORITY_RANGE(mcip,
1527 				    mrp->mrp_priority);
1528 			}
1529 		}
1530 
1531 		mac_flow_modify(mip->mi_flow_tab, mcip->mci_flent, mrp);
1532 		if (mrp->mrp_mask & MRP_PRIORITY)
1533 			mac_update_subflow_priority(mcip);
1534 		return (0);
1535 	}
1536 	return (0);
1537 }
1538 
1539 void
1540 mac_resource_ctl_get(mac_client_handle_t mch, mac_resource_props_t *mrp)
1541 {
1542 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
1543 	mac_resource_props_t	*mcip_mrp = MCIP_RESOURCE_PROPS(mcip);
1544 
1545 	bcopy(mcip_mrp, mrp, sizeof (mac_resource_props_t));
1546 }
1547 
1548 static int
1549 mac_unicast_flow_create(mac_client_impl_t *mcip, uint8_t *mac_addr,
1550     uint16_t vid, boolean_t is_primary, boolean_t first_flow,
1551     flow_entry_t **flent, mac_resource_props_t *mrp)
1552 {
1553 	mac_impl_t	*mip = (mac_impl_t *)mcip->mci_mip;
1554 	flow_desc_t	flow_desc;
1555 	char		flowname[MAXFLOWNAMELEN];
1556 	int		err;
1557 	uint_t		flent_flags;
1558 
1559 	/*
1560 	 * First unicast address being added, create a new flow
1561 	 * for that MAC client.
1562 	 */
1563 	bzero(&flow_desc, sizeof (flow_desc));
1564 
1565 	flow_desc.fd_mac_len = mip->mi_type->mt_addr_length;
1566 	bcopy(mac_addr, flow_desc.fd_dst_mac, flow_desc.fd_mac_len);
1567 	flow_desc.fd_mask = FLOW_LINK_DST;
1568 	if (vid != 0) {
1569 		flow_desc.fd_vid = vid;
1570 		flow_desc.fd_mask |= FLOW_LINK_VID;
1571 	}
1572 
1573 	/*
1574 	 * XXX-nicolas. For now I'm keeping the FLOW_PRIMARY_MAC
1575 	 * and FLOW_VNIC. Even though they're a hack inherited
1576 	 * from the SRS code, we'll keep them for now. They're currently
1577 	 * consumed by mac_datapath_setup() to create the SRS.
1578 	 * That code should be eventually moved out of
1579 	 * mac_datapath_setup() and moved to a mac_srs_create()
1580 	 * function of some sort to keep things clean.
1581 	 *
1582 	 * Also, there's no reason why the SRS for the primary MAC
1583 	 * client should be different than any other MAC client. Until
1584 	 * this is cleaned-up, we support only one MAC unicast address
1585 	 * per client.
1586 	 *
1587 	 * We set FLOW_PRIMARY_MAC for the primary MAC address,
1588 	 * FLOW_VNIC for everything else.
1589 	 */
1590 	if (is_primary)
1591 		flent_flags = FLOW_PRIMARY_MAC;
1592 	else
1593 		flent_flags = FLOW_VNIC_MAC;
1594 
1595 	/*
1596 	 * For the first flow we use the mac client's name - mci_name, for
1597 	 * subsequent ones we just create a name with the vid. This is
1598 	 * so that we can add these flows to the same flow table. This is
1599 	 * fine as the flow name (except for the one with the mac client's
1600 	 * name) is not visible. When the first flow is removed, we just replace
1601 	 * its fdesc with another from the list, so we will still retain the
1602 	 * flent with the MAC client's flow name.
1603 	 */
1604 	if (first_flow) {
1605 		bcopy(mcip->mci_name, flowname, MAXFLOWNAMELEN);
1606 	} else {
1607 		(void) sprintf(flowname, "%s%u", mcip->mci_name, vid);
1608 		flent_flags = FLOW_NO_STATS;
1609 	}
1610 
1611 	if ((err = mac_flow_create(&flow_desc, mrp, flowname, NULL,
1612 	    flent_flags, flent)) != 0)
1613 		return (err);
1614 
1615 	FLOW_MARK(*flent, FE_INCIPIENT);
1616 	(*flent)->fe_mcip = mcip;
1617 
1618 	/*
1619 	 * Place initial creation reference on the flow. This reference
1620 	 * is released in the corresponding delete action viz.
1621 	 * mac_unicast_remove after waiting for all transient refs to
1622 	 * to go away. The wait happens in mac_flow_wait.
1623 	 * We have already held the reference in mac_client_open().
1624 	 */
1625 	if (!first_flow)
1626 		FLOW_REFHOLD(*flent);
1627 	return (0);
1628 }
1629 
1630 /* Refresh the multicast grouping for this VID. */
1631 int
1632 mac_client_update_mcast(void *arg, boolean_t add, const uint8_t *addrp)
1633 {
1634 	flow_entry_t		*flent = arg;
1635 	mac_client_impl_t	*mcip = flent->fe_mcip;
1636 	uint16_t		vid;
1637 	flow_desc_t		flow_desc;
1638 
1639 	mac_flow_get_desc(flent, &flow_desc);
1640 	vid = (flow_desc.fd_mask & FLOW_LINK_VID) != 0 ?
1641 	    flow_desc.fd_vid : VLAN_ID_NONE;
1642 
1643 	/*
1644 	 * We don't call mac_multicast_add()/mac_multicast_remove() as
1645 	 * we want to add/remove for this specific vid.
1646 	 */
1647 	if (add) {
1648 		return (mac_bcast_add(mcip, addrp, vid,
1649 		    MAC_ADDRTYPE_MULTICAST));
1650 	} else {
1651 		mac_bcast_delete(mcip, addrp, vid);
1652 		return (0);
1653 	}
1654 }
1655 
1656 static void
1657 mac_update_single_active_client(mac_impl_t *mip)
1658 {
1659 	mac_client_impl_t *client = NULL;
1660 
1661 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1662 
1663 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
1664 	if (mip->mi_nactiveclients == 1) {
1665 		/*
1666 		 * Find the one active MAC client from the list of MAC
1667 		 * clients. The active MAC client has at least one
1668 		 * unicast address.
1669 		 */
1670 		for (client = mip->mi_clients_list; client != NULL;
1671 		    client = client->mci_client_next) {
1672 			if (client->mci_unicast_list != NULL)
1673 				break;
1674 		}
1675 		ASSERT(client != NULL);
1676 	}
1677 
1678 	/*
1679 	 * mi_single_active_client is protected by the MAC impl's read/writer
1680 	 * lock, which allows mac_rx() to check the value of that pointer
1681 	 * as a reader.
1682 	 */
1683 	mip->mi_single_active_client = client;
1684 	rw_exit(&mip->mi_rw_lock);
1685 }
1686 
1687 /*
1688  * Set up the data path. Called from i_mac_unicast_add after having
1689  * done all the validations including making sure this is an active
1690  * client (i.e that is ready to process packets.)
1691  */
1692 static int
1693 mac_client_datapath_setup(mac_client_impl_t *mcip, uint16_t vid,
1694     uint8_t *mac_addr, mac_resource_props_t *mrp, boolean_t isprimary,
1695     mac_unicast_impl_t *muip)
1696 {
1697 	mac_impl_t	*mip = mcip->mci_mip;
1698 	boolean_t	mac_started = B_FALSE;
1699 	boolean_t	bcast_added = B_FALSE;
1700 	boolean_t	nactiveclients_added = B_FALSE;
1701 	flow_entry_t	*flent;
1702 	int		err = 0;
1703 
1704 	if ((err = mac_start((mac_handle_t)mip)) != 0)
1705 		goto bail;
1706 
1707 	mac_started = B_TRUE;
1708 
1709 	/* add the MAC client to the broadcast address group by default */
1710 	if (mip->mi_type->mt_brdcst_addr != NULL) {
1711 		err = mac_bcast_add(mcip, mip->mi_type->mt_brdcst_addr, vid,
1712 		    MAC_ADDRTYPE_BROADCAST);
1713 		if (err != 0)
1714 			goto bail;
1715 		bcast_added = B_TRUE;
1716 	}
1717 
1718 	/*
1719 	 * If this is the first unicast address addition for this
1720 	 * client, reuse the pre-allocated larval flow entry associated with
1721 	 * the MAC client.
1722 	 */
1723 	flent = (mcip->mci_nflents == 0) ? mcip->mci_flent : NULL;
1724 
1725 	/* We are configuring the unicast flow now */
1726 	if (!MCIP_DATAPATH_SETUP(mcip)) {
1727 
1728 		MAC_CLIENT_SET_PRIORITY_RANGE(mcip,
1729 		    (mrp->mrp_mask & MRP_PRIORITY) ? mrp->mrp_priority :
1730 		    MPL_LINK_DEFAULT);
1731 
1732 		if ((err = mac_unicast_flow_create(mcip, mac_addr, vid,
1733 		    isprimary, B_TRUE, &flent, mrp)) != 0)
1734 			goto bail;
1735 
1736 		mip->mi_nactiveclients++;
1737 		nactiveclients_added = B_TRUE;
1738 
1739 		/*
1740 		 * This will allocate the RX ring group if possible for the
1741 		 * flow and program the software classifier as needed.
1742 		 */
1743 		if ((err = mac_datapath_setup(mcip, flent, SRST_LINK)) != 0)
1744 			goto bail;
1745 
1746 		/*
1747 		 * The unicast MAC address must have been added successfully.
1748 		 */
1749 		ASSERT(mcip->mci_unicast != NULL);
1750 		/*
1751 		 * Push down the sub-flows that were defined on this link
1752 		 * hitherto. The flows are added to the active flow table
1753 		 * and SRS, softrings etc. are created as needed.
1754 		 */
1755 		mac_link_init_flows((mac_client_handle_t)mcip);
1756 	} else {
1757 		mac_address_t *map = mcip->mci_unicast;
1758 
1759 		/*
1760 		 * A unicast flow already exists for that MAC client,
1761 		 * this flow must be the same mac address but with
1762 		 * different VID. It has been checked by mac_addr_in_use().
1763 		 *
1764 		 * We will use the SRS etc. from the mci_flent. Note that
1765 		 * We don't need to create kstat for this as except for
1766 		 * the fdesc, everything will be used from in the 1st flent.
1767 		 */
1768 
1769 		if (bcmp(mac_addr, map->ma_addr, map->ma_len) != 0) {
1770 			err = EINVAL;
1771 			goto bail;
1772 		}
1773 
1774 		if ((err = mac_unicast_flow_create(mcip, mac_addr, vid,
1775 		    isprimary, B_FALSE, &flent, NULL)) != 0) {
1776 			goto bail;
1777 		}
1778 		if ((err = mac_flow_add(mip->mi_flow_tab, flent)) != 0) {
1779 			FLOW_FINAL_REFRELE(flent);
1780 			goto bail;
1781 		}
1782 
1783 		/* update the multicast group for this vid */
1784 		mac_client_bcast_refresh(mcip, mac_client_update_mcast,
1785 		    (void *)flent, B_TRUE);
1786 
1787 	}
1788 
1789 	/* populate the shared MAC address */
1790 	muip->mui_map = mcip->mci_unicast;
1791 
1792 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
1793 	muip->mui_next = mcip->mci_unicast_list;
1794 	mcip->mci_unicast_list = muip;
1795 	rw_exit(&mcip->mci_rw_lock);
1796 
1797 
1798 	/*
1799 	 * First add the flent to the flow list of this mcip. Then set
1800 	 * the mip's mi_single_active_client if needed. The Rx path assumes
1801 	 * that mip->mi_single_active_client will always have an associated
1802 	 * flent.
1803 	 */
1804 	mac_client_add_to_flow_list(mcip, flent);
1805 
1806 	if (nactiveclients_added)
1807 		mac_update_single_active_client(mip);
1808 	/*
1809 	 * Trigger a renegotiation of the capabilities when the number of
1810 	 * active clients changes from 1 to 2, since some of the capabilities
1811 	 * might have to be disabled. Also send a MAC_NOTE_LINK notification
1812 	 * to all the MAC clients whenever physical link is DOWN.
1813 	 */
1814 	if (mip->mi_nactiveclients == 2) {
1815 		mac_capab_update((mac_handle_t)mip);
1816 		mac_virtual_link_update(mip);
1817 	}
1818 	/*
1819 	 * Now that the setup is complete, clear the INCIPIENT flag.
1820 	 * The flag was set to avoid incoming packets seeing inconsistent
1821 	 * structures while the setup was in progress. Clear the mci_tx_flag
1822 	 * by calling mac_tx_client_block. It is possible that
1823 	 * mac_unicast_remove was called prior to this mac_unicast_add which
1824 	 * could have set the MCI_TX_QUIESCE flag.
1825 	 */
1826 	if (flent->fe_rx_ring_group != NULL)
1827 		mac_rx_group_unmark(flent->fe_rx_ring_group, MR_INCIPIENT);
1828 	FLOW_UNMARK(flent, FE_INCIPIENT);
1829 	FLOW_UNMARK(flent, FE_MC_NO_DATAPATH);
1830 	mac_tx_client_unblock(mcip);
1831 	return (0);
1832 bail:
1833 	if (bcast_added)
1834 		mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr, vid);
1835 
1836 	if (nactiveclients_added)
1837 		mip->mi_nactiveclients--;
1838 
1839 	if (mac_started)
1840 		mac_stop((mac_handle_t)mip);
1841 
1842 	return (err);
1843 }
1844 
1845 /*
1846  * Return the passive primary MAC client, if present. The passive client is
1847  * a stand-by client that has the same unicast address as another that is
1848  * currenly active. Once the active client goes away, the passive client
1849  * becomes active.
1850  */
1851 static mac_client_impl_t *
1852 mac_get_passive_primary_client(mac_impl_t *mip)
1853 {
1854 	mac_client_impl_t	*mcip;
1855 
1856 	for (mcip = mip->mi_clients_list; mcip != NULL;
1857 	    mcip = mcip->mci_client_next) {
1858 		if (mac_is_primary_client(mcip) &&
1859 		    (mcip->mci_flags & MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
1860 			return (mcip);
1861 		}
1862 	}
1863 	return (NULL);
1864 }
1865 
1866 /*
1867  * Add a new unicast address to the MAC client.
1868  *
1869  * The MAC address can be specified either by value, or the MAC client
1870  * can specify that it wants to use the primary MAC address of the
1871  * underlying MAC. See the introductory comments at the beginning
1872  * of this file for more more information on primary MAC addresses.
1873  *
1874  * Note also the tuple (MAC address, VID) must be unique
1875  * for the MAC clients defined on top of the same underlying MAC
1876  * instance, unless the MAC_UNICAST_NODUPCHECK is specified.
1877  *
1878  * In no case can a client use the PVID for the MAC, if the MAC has one set.
1879  */
1880 int
1881 i_mac_unicast_add(mac_client_handle_t mch, uint8_t *mac_addr, uint16_t flags,
1882     mac_unicast_handle_t *mah, uint16_t vid, mac_diag_t *diag)
1883 {
1884 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
1885 	mac_impl_t		*mip = mcip->mci_mip;
1886 	int			err;
1887 	uint_t			mac_len = mip->mi_type->mt_addr_length;
1888 	boolean_t		check_dups = !(flags & MAC_UNICAST_NODUPCHECK);
1889 	boolean_t		fastpath_disabled = B_FALSE;
1890 	boolean_t		is_primary = (flags & MAC_UNICAST_PRIMARY);
1891 	boolean_t		is_unicast_hw = (flags & MAC_UNICAST_HW);
1892 	mac_resource_props_t	mrp;
1893 	boolean_t		passive_client = B_FALSE;
1894 	mac_unicast_impl_t	*muip;
1895 	boolean_t		is_vnic_primary =
1896 	    (flags & MAC_UNICAST_VNIC_PRIMARY);
1897 
1898 	/* when VID is non-zero, the underlying MAC can not be VNIC */
1899 	ASSERT(!((mip->mi_state_flags & MIS_IS_VNIC) && (vid != 0)));
1900 
1901 	/*
1902 	 * Check for an attempted use of the current Port VLAN ID, if enabled.
1903 	 * No client may use it.
1904 	 */
1905 	if (mip->mi_pvid != 0 && vid == mip->mi_pvid)
1906 		return (EBUSY);
1907 
1908 	/*
1909 	 * Check whether it's the primary client and flag it.
1910 	 */
1911 	if (!(mcip->mci_state_flags & MCIS_IS_VNIC) && is_primary && vid == 0)
1912 		mcip->mci_flags |= MAC_CLIENT_FLAGS_PRIMARY;
1913 
1914 	/*
1915 	 * is_vnic_primary is true when we come here as a VLAN VNIC
1916 	 * which uses the primary mac client's address but with a non-zero
1917 	 * VID. In this case the MAC address is not specified by an upper
1918 	 * MAC client.
1919 	 */
1920 	if ((mcip->mci_state_flags & MCIS_IS_VNIC) && is_primary &&
1921 	    !is_vnic_primary) {
1922 		/*
1923 		 * The address is being set by the upper MAC client
1924 		 * of a VNIC. The MAC address was already set by the
1925 		 * VNIC driver during VNIC creation.
1926 		 *
1927 		 * Note: a VNIC has only one MAC address. We return
1928 		 * the MAC unicast address handle of the lower MAC client
1929 		 * corresponding to the VNIC. We allocate a new entry
1930 		 * which is flagged appropriately, so that mac_unicast_remove()
1931 		 * doesn't attempt to free the original entry that
1932 		 * was allocated by the VNIC driver.
1933 		 */
1934 		ASSERT(mcip->mci_unicast != NULL);
1935 
1936 		/* Check for VLAN flags, if present */
1937 		if ((flags & MAC_UNICAST_TAG_DISABLE) != 0)
1938 			mcip->mci_state_flags |= MCIS_TAG_DISABLE;
1939 
1940 		if ((flags & MAC_UNICAST_STRIP_DISABLE) != 0)
1941 			mcip->mci_state_flags |= MCIS_STRIP_DISABLE;
1942 
1943 		if ((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) != 0)
1944 			mcip->mci_state_flags |= MCIS_DISABLE_TX_VID_CHECK;
1945 
1946 		/*
1947 		 * Ensure that the primary unicast address of the VNIC
1948 		 * is added only once unless we have the
1949 		 * MAC_CLIENT_FLAGS_MULTI_PRIMARY set (and this is not
1950 		 * a passive MAC client).
1951 		 */
1952 		if ((mcip->mci_flags & MAC_CLIENT_FLAGS_VNIC_PRIMARY) != 0) {
1953 			if ((mcip->mci_flags &
1954 			    MAC_CLIENT_FLAGS_MULTI_PRIMARY) == 0 ||
1955 			    (mcip->mci_flags &
1956 			    MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
1957 				return (EBUSY);
1958 			}
1959 			mcip->mci_flags |= MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
1960 			passive_client = B_TRUE;
1961 		}
1962 
1963 		mcip->mci_flags |= MAC_CLIENT_FLAGS_VNIC_PRIMARY;
1964 
1965 		/*
1966 		 * Create a handle for vid 0.
1967 		 */
1968 		ASSERT(vid == 0);
1969 		muip = kmem_zalloc(sizeof (mac_unicast_impl_t), KM_SLEEP);
1970 		muip->mui_vid = vid;
1971 		*mah = (mac_unicast_handle_t)muip;
1972 		/*
1973 		 * This will be used by the caller to defer setting the
1974 		 * rx functions.
1975 		 */
1976 		if (passive_client)
1977 			return (EAGAIN);
1978 		return (0);
1979 	}
1980 
1981 	/* primary MAC clients cannot be opened on top of anchor VNICs */
1982 	if ((is_vnic_primary || is_primary) &&
1983 	    i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_ANCHOR_VNIC, NULL)) {
1984 		return (ENXIO);
1985 	}
1986 
1987 	/*
1988 	 * If this is a VNIC/VLAN, disable softmac fast-path.
1989 	 */
1990 	if (mcip->mci_state_flags & MCIS_IS_VNIC) {
1991 		err = mac_fastpath_disable((mac_handle_t)mip);
1992 		if (err != 0)
1993 			return (err);
1994 		fastpath_disabled = B_TRUE;
1995 	}
1996 
1997 	/*
1998 	 * Return EBUSY if:
1999 	 *  - there is an exclusively active mac client exists.
2000 	 *  - this is an exclusive active mac client but
2001 	 *	a. there is already active mac clients exist, or
2002 	 *	b. fastpath streams are already plumbed on this legacy device
2003 	 *  - the mac creator has disallowed active mac clients.
2004 	 */
2005 	if (mip->mi_state_flags & (MIS_EXCLUSIVE|MIS_NO_ACTIVE)) {
2006 		if (fastpath_disabled)
2007 			mac_fastpath_enable((mac_handle_t)mip);
2008 		return (EBUSY);
2009 	}
2010 
2011 	if (mcip->mci_state_flags & MCIS_EXCLUSIVE) {
2012 		ASSERT(!fastpath_disabled);
2013 		if (mip->mi_nactiveclients != 0)
2014 			return (EBUSY);
2015 
2016 		if ((mip->mi_state_flags & MIS_LEGACY) &&
2017 		    !(mip->mi_capab_legacy.ml_active_set(mip->mi_driver))) {
2018 			return (EBUSY);
2019 		}
2020 		mip->mi_state_flags |= MIS_EXCLUSIVE;
2021 	}
2022 
2023 	bzero(&mrp, sizeof (mac_resource_props_t));
2024 	if (is_primary && !(mcip->mci_state_flags & (MCIS_IS_VNIC |
2025 	    MCIS_IS_AGGR_PORT))) {
2026 		/*
2027 		 * Apply the property cached in the mac_impl_t to the primary
2028 		 * mac client. If the mac client is a VNIC or an aggregation
2029 		 * port, its property should be set in the mcip when the
2030 		 * VNIC/aggr was created.
2031 		 */
2032 		mac_get_resources((mac_handle_t)mip, &mrp);
2033 		(void) mac_client_set_resources(mch, &mrp);
2034 	} else if (mcip->mci_state_flags & MCIS_IS_VNIC) {
2035 		bcopy(MCIP_RESOURCE_PROPS(mcip), &mrp,
2036 		    sizeof (mac_resource_props_t));
2037 	}
2038 
2039 	muip = kmem_zalloc(sizeof (mac_unicast_impl_t), KM_SLEEP);
2040 	muip->mui_vid = vid;
2041 
2042 	if (is_primary || is_vnic_primary) {
2043 		mac_addr = mip->mi_addr;
2044 	} else {
2045 
2046 		/*
2047 		 * Verify the validity of the specified MAC addresses value.
2048 		 */
2049 		if (!mac_unicst_verify((mac_handle_t)mip, mac_addr, mac_len)) {
2050 			*diag = MAC_DIAG_MACADDR_INVALID;
2051 			err = EINVAL;
2052 			goto bail_out;
2053 		}
2054 
2055 		/*
2056 		 * Make sure that the specified MAC address is different
2057 		 * than the unicast MAC address of the underlying NIC.
2058 		 */
2059 		if (check_dups && bcmp(mip->mi_addr, mac_addr, mac_len) == 0) {
2060 			*diag = MAC_DIAG_MACADDR_NIC;
2061 			err = EINVAL;
2062 			goto bail_out;
2063 		}
2064 	}
2065 
2066 	/*
2067 	 * Set the flags here so that if this is a passive client, we
2068 	 * can return  and set it when we call mac_client_datapath_setup
2069 	 * when this becomes the active client. If we defer to using these
2070 	 * flags to mac_client_datapath_setup, then for a passive client,
2071 	 * we'd have to store the flags somewhere (probably fe_flags)
2072 	 * and then use it.
2073 	 */
2074 	if (!MCIP_DATAPATH_SETUP(mcip)) {
2075 		if (is_unicast_hw) {
2076 			/*
2077 			 * The client requires a hardware MAC address slot
2078 			 * for that unicast address. Since we support only
2079 			 * one unicast MAC address per client, flag the
2080 			 * MAC client itself.
2081 			 */
2082 			mcip->mci_state_flags |= MCIS_UNICAST_HW;
2083 		}
2084 
2085 		/* Check for VLAN flags, if present */
2086 		if ((flags & MAC_UNICAST_TAG_DISABLE) != 0)
2087 			mcip->mci_state_flags |= MCIS_TAG_DISABLE;
2088 
2089 		if ((flags & MAC_UNICAST_STRIP_DISABLE) != 0)
2090 			mcip->mci_state_flags |= MCIS_STRIP_DISABLE;
2091 
2092 		if ((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) != 0)
2093 			mcip->mci_state_flags |= MCIS_DISABLE_TX_VID_CHECK;
2094 	} else {
2095 		/*
2096 		 * Assert that the specified flags are consistent with the
2097 		 * flags specified by previous calls to mac_unicast_add().
2098 		 */
2099 		ASSERT(((flags & MAC_UNICAST_TAG_DISABLE) != 0 &&
2100 		    (mcip->mci_state_flags & MCIS_TAG_DISABLE) != 0) ||
2101 		    ((flags & MAC_UNICAST_TAG_DISABLE) == 0 &&
2102 		    (mcip->mci_state_flags & MCIS_TAG_DISABLE) == 0));
2103 
2104 		ASSERT(((flags & MAC_UNICAST_STRIP_DISABLE) != 0 &&
2105 		    (mcip->mci_state_flags & MCIS_STRIP_DISABLE) != 0) ||
2106 		    ((flags & MAC_UNICAST_STRIP_DISABLE) == 0 &&
2107 		    (mcip->mci_state_flags & MCIS_STRIP_DISABLE) == 0));
2108 
2109 		ASSERT(((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) != 0 &&
2110 		    (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK) != 0) ||
2111 		    ((flags & MAC_UNICAST_DISABLE_TX_VID_CHECK) == 0 &&
2112 		    (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK) == 0));
2113 
2114 		/*
2115 		 * Make sure the client is consistent about its requests
2116 		 * for MAC addresses. I.e. all requests from the clients
2117 		 * must have the MAC_UNICAST_HW flag set or clear.
2118 		 */
2119 		if ((mcip->mci_state_flags & MCIS_UNICAST_HW) != 0 &&
2120 		    !is_unicast_hw ||
2121 		    (mcip->mci_state_flags & MCIS_UNICAST_HW) == 0 &&
2122 		    is_unicast_hw) {
2123 			err = EINVAL;
2124 			goto bail_out;
2125 		}
2126 	}
2127 	/*
2128 	 * Make sure the MAC address is not already used by
2129 	 * another MAC client defined on top of the same
2130 	 * underlying NIC. Unless we have MAC_CLIENT_FLAGS_MULTI_PRIMARY
2131 	 * set when we allow a passive client to be present which will
2132 	 * be activated when the currently active client goes away - this
2133 	 * works only with primary addresses.
2134 	 */
2135 	if ((check_dups || is_primary || is_vnic_primary) &&
2136 	    mac_addr_in_use(mip, mac_addr, vid)) {
2137 		/*
2138 		 * Must have set the multiple primary address flag when
2139 		 * we did a mac_client_open AND this should be a primary
2140 		 * MAC client AND there should not already be a passive
2141 		 * primary. If all is true then we let this succeed
2142 		 * even if the address is a dup.
2143 		 */
2144 		if ((mcip->mci_flags & MAC_CLIENT_FLAGS_MULTI_PRIMARY) == 0 ||
2145 		    (mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY) == 0 ||
2146 		    mac_get_passive_primary_client(mip) != NULL) {
2147 			*diag = MAC_DIAG_MACADDR_INUSE;
2148 			err = EEXIST;
2149 			goto bail_out;
2150 		}
2151 		ASSERT((mcip->mci_flags &
2152 		    MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) == 0);
2153 		mcip->mci_flags |= MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
2154 
2155 		/*
2156 		 * Stash the unicast address handle, we will use it when
2157 		 * we set up the passive client.
2158 		 */
2159 		mcip->mci_p_unicast_list = muip;
2160 		*mah = (mac_unicast_handle_t)muip;
2161 		return (0);
2162 	}
2163 
2164 	err = mac_client_datapath_setup(mcip, vid, mac_addr, &mrp,
2165 	    is_primary || is_vnic_primary, muip);
2166 	if (err != 0)
2167 		goto bail_out;
2168 	*mah = (mac_unicast_handle_t)muip;
2169 	return (0);
2170 
2171 bail_out:
2172 	if (fastpath_disabled)
2173 		mac_fastpath_enable((mac_handle_t)mip);
2174 	if (mcip->mci_state_flags & MCIS_EXCLUSIVE) {
2175 		mip->mi_state_flags &= ~MIS_EXCLUSIVE;
2176 		if (mip->mi_state_flags & MIS_LEGACY) {
2177 			mip->mi_capab_legacy.ml_active_clear(
2178 			    mip->mi_driver);
2179 		}
2180 	}
2181 	kmem_free(muip, sizeof (mac_unicast_impl_t));
2182 	return (err);
2183 }
2184 
2185 /*
2186  * Wrapper function to mac_unicast_add when we want to have the same mac
2187  * client open for two instances, one that is currently active and another
2188  * that will become active when the current one is removed. In this case
2189  * mac_unicast_add will return EGAIN and we will save the rx function and
2190  * arg which will be used when we activate the passive client in
2191  * mac_unicast_remove.
2192  */
2193 int
2194 mac_unicast_add_set_rx(mac_client_handle_t mch, uint8_t *mac_addr,
2195     uint16_t flags, mac_unicast_handle_t *mah,  uint16_t vid, mac_diag_t *diag,
2196     mac_rx_t rx_fn, void *arg)
2197 {
2198 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2199 	uint_t			err;
2200 
2201 	err = mac_unicast_add(mch, mac_addr, flags, mah, vid, diag);
2202 	if (err != 0 && err != EAGAIN)
2203 		return (err);
2204 	if (err == EAGAIN) {
2205 		if (rx_fn != NULL) {
2206 			mcip->mci_rx_p_fn = rx_fn;
2207 			mcip->mci_rx_p_arg = arg;
2208 		}
2209 		return (0);
2210 	}
2211 	if (rx_fn != NULL)
2212 		mac_rx_set(mch, rx_fn, arg);
2213 	return (err);
2214 }
2215 
2216 int
2217 mac_unicast_add(mac_client_handle_t mch, uint8_t *mac_addr, uint16_t flags,
2218     mac_unicast_handle_t *mah, uint16_t vid, mac_diag_t *diag)
2219 {
2220 	mac_impl_t *mip = ((mac_client_impl_t *)mch)->mci_mip;
2221 	uint_t err;
2222 
2223 	i_mac_perim_enter(mip);
2224 	err = i_mac_unicast_add(mch, mac_addr, flags, mah, vid, diag);
2225 	i_mac_perim_exit(mip);
2226 
2227 	return (err);
2228 }
2229 
2230 void
2231 mac_client_datapath_teardown(mac_client_handle_t mch, mac_unicast_impl_t *muip,
2232     flow_entry_t *flent)
2233 {
2234 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2235 	mac_impl_t		*mip = mcip->mci_mip;
2236 
2237 	/*
2238 	 * We would have initialized subflows etc. only if we brought up
2239 	 * the primary client and set the unicast unicast address etc.
2240 	 * Deactivate the flows. The flow entry will be removed from the
2241 	 * active flow tables, and the associated SRS, softrings etc will
2242 	 * be deleted. But the flow entry itself won't be destroyed, instead
2243 	 * it will continue to be archived off the  the global flow hash
2244 	 * list, for a possible future activation when say IP is plumbed
2245 	 * again.
2246 	 */
2247 	mac_link_release_flows(mch);
2248 
2249 	mip->mi_nactiveclients--;
2250 	mac_update_single_active_client(mip);
2251 
2252 	/* Tear down the data path */
2253 	mac_datapath_teardown(mcip, mcip->mci_flent, SRST_LINK);
2254 
2255 	/*
2256 	 * Prevent any future access to the flow entry through the mci_flent
2257 	 * pointer by setting the mci_flent to NULL. Access to mci_flent in
2258 	 * mac_bcast_send is also under mi_rw_lock.
2259 	 */
2260 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
2261 	flent = mcip->mci_flent;
2262 	mac_client_remove_flow_from_list(mcip, flent);
2263 
2264 	if (mcip->mci_state_flags & MCIS_DESC_LOGGED)
2265 		mcip->mci_state_flags &= ~MCIS_DESC_LOGGED;
2266 
2267 	/*
2268 	 * This is the last unicast address being removed and there shouldn't
2269 	 * be any outbound data threads at this point coming down from mac
2270 	 * clients. We have waited for the data threads to finish before
2271 	 * starting dld_str_detach. Non-data threads must access TX SRS
2272 	 * under mi_rw_lock.
2273 	 */
2274 	rw_exit(&mip->mi_rw_lock);
2275 
2276 	/*
2277 	 * Don't use FLOW_MARK with FE_MC_NO_DATAPATH, as the flow might
2278 	 * contain other flags, such as FE_CONDEMNED, which we need to
2279 	 * cleared. We don't call mac_flow_cleanup() for this unicast
2280 	 * flow as we have a already cleaned up SRSs etc. (via the teadown
2281 	 * path). We just clear the stats and reset the initial callback
2282 	 * function, the rest will be set when we call mac_flow_create,
2283 	 * if at all.
2284 	 */
2285 	mutex_enter(&flent->fe_lock);
2286 	ASSERT(flent->fe_refcnt == 1 && flent->fe_mbg == NULL &&
2287 	    flent->fe_tx_srs == NULL && flent->fe_rx_srs_cnt == 0);
2288 	flent->fe_flags = FE_MC_NO_DATAPATH;
2289 	flow_stat_destroy(flent);
2290 
2291 	/* Initialize the receiver function to a safe routine */
2292 	flent->fe_cb_fn = (flow_fn_t)mac_pkt_drop;
2293 	flent->fe_cb_arg1 = NULL;
2294 	flent->fe_cb_arg2 = NULL;
2295 
2296 	flent->fe_index = -1;
2297 	mutex_exit(&flent->fe_lock);
2298 
2299 	if (mip->mi_type->mt_brdcst_addr != NULL) {
2300 		mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr,
2301 		    muip->mui_vid);
2302 	}
2303 
2304 	if (mip->mi_nactiveclients == 1) {
2305 		mac_capab_update((mac_handle_t)mip);
2306 		mac_virtual_link_update(mip);
2307 	}
2308 
2309 	if (mcip->mci_state_flags & MCIS_EXCLUSIVE) {
2310 		mip->mi_state_flags &= ~MIS_EXCLUSIVE;
2311 
2312 		if (mip->mi_state_flags & MIS_LEGACY)
2313 			mip->mi_capab_legacy.ml_active_clear(mip->mi_driver);
2314 	}
2315 
2316 	mcip->mci_state_flags &= ~MCIS_UNICAST_HW;
2317 
2318 	if (mcip->mci_state_flags & MCIS_TAG_DISABLE)
2319 		mcip->mci_state_flags &= ~MCIS_TAG_DISABLE;
2320 
2321 	if (mcip->mci_state_flags & MCIS_STRIP_DISABLE)
2322 		mcip->mci_state_flags &= ~MCIS_STRIP_DISABLE;
2323 
2324 	if (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK)
2325 		mcip->mci_state_flags &= ~MCIS_DISABLE_TX_VID_CHECK;
2326 
2327 	kmem_free(muip, sizeof (mac_unicast_impl_t));
2328 
2329 	/*
2330 	 * Disable fastpath if this is a VNIC or a VLAN.
2331 	 */
2332 	if (mcip->mci_state_flags & MCIS_IS_VNIC)
2333 		mac_fastpath_enable((mac_handle_t)mip);
2334 	mac_stop((mac_handle_t)mip);
2335 }
2336 
2337 /*
2338  * Remove a MAC address which was previously added by mac_unicast_add().
2339  */
2340 int
2341 mac_unicast_remove(mac_client_handle_t mch, mac_unicast_handle_t mah)
2342 {
2343 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
2344 	mac_unicast_impl_t *muip = (mac_unicast_impl_t *)mah;
2345 	mac_unicast_impl_t *pre;
2346 	mac_impl_t *mip = mcip->mci_mip;
2347 	flow_entry_t		*flent;
2348 	boolean_t		isprimary = B_FALSE;
2349 
2350 	i_mac_perim_enter(mip);
2351 	if (mcip->mci_flags & MAC_CLIENT_FLAGS_VNIC_PRIMARY) {
2352 		/*
2353 		 * Called made by the upper MAC client of a VNIC.
2354 		 * There's nothing much to do, the unicast address will
2355 		 * be removed by the VNIC driver when the VNIC is deleted,
2356 		 * but let's ensure that all our transmit is done before
2357 		 * the client does a mac_client_stop lest it trigger an
2358 		 * assert in the driver.
2359 		 */
2360 		ASSERT(muip->mui_vid == 0);
2361 
2362 		mac_tx_client_flush(mcip);
2363 
2364 		if ((mcip->mci_flags & MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
2365 			mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
2366 			if (mcip->mci_rx_p_fn != NULL) {
2367 				mac_rx_set(mch, mcip->mci_rx_p_fn,
2368 				    mcip->mci_rx_p_arg);
2369 				mcip->mci_rx_p_fn = NULL;
2370 				mcip->mci_rx_p_arg = NULL;
2371 			}
2372 			kmem_free(muip, sizeof (mac_unicast_impl_t));
2373 			i_mac_perim_exit(mip);
2374 			return (0);
2375 		}
2376 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_VNIC_PRIMARY;
2377 
2378 		if (mcip->mci_state_flags & MCIS_TAG_DISABLE)
2379 			mcip->mci_state_flags &= ~MCIS_TAG_DISABLE;
2380 
2381 		if (mcip->mci_state_flags & MCIS_STRIP_DISABLE)
2382 			mcip->mci_state_flags &= ~MCIS_STRIP_DISABLE;
2383 
2384 		if (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK)
2385 			mcip->mci_state_flags &= ~MCIS_DISABLE_TX_VID_CHECK;
2386 
2387 		kmem_free(muip, sizeof (mac_unicast_impl_t));
2388 		i_mac_perim_exit(mip);
2389 		return (0);
2390 	}
2391 
2392 	ASSERT(muip != NULL);
2393 
2394 	/*
2395 	 * We are removing a passive client, we haven't setup the datapath
2396 	 * for this yet, so nothing much to do.
2397 	 */
2398 	if ((mcip->mci_flags & MAC_CLIENT_FLAGS_PASSIVE_PRIMARY) != 0) {
2399 
2400 		ASSERT((mcip->mci_flent->fe_flags & FE_MC_NO_DATAPATH) != 0);
2401 		ASSERT(mcip->mci_p_unicast_list == muip);
2402 
2403 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
2404 
2405 		mcip->mci_p_unicast_list = NULL;
2406 		mcip->mci_rx_p_fn = NULL;
2407 		mcip->mci_rx_p_arg = NULL;
2408 
2409 		mcip->mci_state_flags &= ~MCIS_UNICAST_HW;
2410 
2411 		if (mcip->mci_state_flags & MCIS_TAG_DISABLE)
2412 			mcip->mci_state_flags &= ~MCIS_TAG_DISABLE;
2413 
2414 		if (mcip->mci_state_flags & MCIS_STRIP_DISABLE)
2415 			mcip->mci_state_flags &= ~MCIS_STRIP_DISABLE;
2416 
2417 		if (mcip->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK)
2418 			mcip->mci_state_flags &= ~MCIS_DISABLE_TX_VID_CHECK;
2419 
2420 		kmem_free(muip, sizeof (mac_unicast_impl_t));
2421 		i_mac_perim_exit(mip);
2422 		return (0);
2423 	}
2424 	/*
2425 	 * Remove the VID from the list of client's VIDs.
2426 	 */
2427 	pre = mcip->mci_unicast_list;
2428 	if (muip == pre) {
2429 		mcip->mci_unicast_list = muip->mui_next;
2430 	} else {
2431 		while ((pre->mui_next != NULL) && (pre->mui_next != muip))
2432 			pre = pre->mui_next;
2433 		ASSERT(pre->mui_next == muip);
2434 		rw_enter(&mcip->mci_rw_lock, RW_WRITER);
2435 		pre->mui_next = muip->mui_next;
2436 		rw_exit(&mcip->mci_rw_lock);
2437 	}
2438 
2439 	if ((mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY) &&
2440 	    muip->mui_vid == 0) {
2441 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PRIMARY;
2442 		isprimary = B_TRUE;
2443 	}
2444 	if (!mac_client_single_rcvr(mcip)) {
2445 		/*
2446 		 * This MAC client is shared by more than one unicast
2447 		 * addresses, so we will just remove the flent
2448 		 * corresponding to the address being removed. We don't invoke
2449 		 * mac_rx_classify_flow_rem() since the additional flow is
2450 		 * not associated with its own separate set of SRS and rings,
2451 		 * and these constructs are still needed for the remaining
2452 		 * flows.
2453 		 */
2454 		flent = mac_client_get_flow(mcip, muip);
2455 		ASSERT(flent != NULL);
2456 
2457 		/*
2458 		 * The first one is disappearing, need to make sure
2459 		 * we replace it with another from the list of
2460 		 * shared clients.
2461 		 */
2462 		if (flent == mcip->mci_flent)
2463 			flent = mac_client_swap_mciflent(mcip);
2464 		mac_client_remove_flow_from_list(mcip, flent);
2465 		mac_flow_remove(mip->mi_flow_tab, flent, B_FALSE);
2466 		mac_flow_wait(flent, FLOW_DRIVER_UPCALL);
2467 
2468 		/*
2469 		 * The multicast groups that were added by the client so
2470 		 * far must be removed from the brodcast domain corresponding
2471 		 * to the VID being removed.
2472 		 */
2473 		mac_client_bcast_refresh(mcip, mac_client_update_mcast,
2474 		    (void *)flent, B_FALSE);
2475 
2476 		if (mip->mi_type->mt_brdcst_addr != NULL) {
2477 			mac_bcast_delete(mcip, mip->mi_type->mt_brdcst_addr,
2478 			    muip->mui_vid);
2479 		}
2480 
2481 		FLOW_FINAL_REFRELE(flent);
2482 		ASSERT(!(mcip->mci_state_flags & MCIS_EXCLUSIVE));
2483 		/*
2484 		 * Enable fastpath if this is a VNIC or a VLAN.
2485 		 */
2486 		if (mcip->mci_state_flags & MCIS_IS_VNIC)
2487 			mac_fastpath_enable((mac_handle_t)mip);
2488 		mac_stop((mac_handle_t)mip);
2489 		i_mac_perim_exit(mip);
2490 		return (0);
2491 	}
2492 
2493 	mac_client_datapath_teardown(mch, muip, flent);
2494 
2495 	/*
2496 	 * If we are removing the primary, check if we have a passive primary
2497 	 * client that we need to activate now.
2498 	 */
2499 	if (!isprimary) {
2500 		i_mac_perim_exit(mip);
2501 		return (0);
2502 	}
2503 	mcip = mac_get_passive_primary_client(mip);
2504 	if (mcip != NULL) {
2505 		mac_resource_props_t	mrp;
2506 		mac_unicast_impl_t	*muip;
2507 
2508 		mcip->mci_flags &= ~MAC_CLIENT_FLAGS_PASSIVE_PRIMARY;
2509 		bzero(&mrp, sizeof (mac_resource_props_t));
2510 		/*
2511 		 * Apply the property cached in the mac_impl_t to the
2512 		 * primary mac client.
2513 		 */
2514 		mac_get_resources((mac_handle_t)mip, &mrp);
2515 		(void) mac_client_set_resources(mch, &mrp);
2516 		ASSERT(mcip->mci_p_unicast_list != NULL);
2517 		muip = mcip->mci_p_unicast_list;
2518 		mcip->mci_p_unicast_list = NULL;
2519 		if (mac_client_datapath_setup(mcip, VLAN_ID_NONE,
2520 		    mip->mi_addr, &mrp, B_TRUE, muip) == 0) {
2521 			if (mcip->mci_rx_p_fn != NULL) {
2522 				mac_rx_set(mch, mcip->mci_rx_p_fn,
2523 				    mcip->mci_rx_p_arg);
2524 				mcip->mci_rx_p_fn = NULL;
2525 				mcip->mci_rx_p_arg = NULL;
2526 			}
2527 		} else {
2528 			kmem_free(muip, sizeof (mac_unicast_impl_t));
2529 		}
2530 	}
2531 	i_mac_perim_exit(mip);
2532 	return (0);
2533 }
2534 
2535 /*
2536  * Multicast add function invoked by MAC clients.
2537  */
2538 int
2539 mac_multicast_add(mac_client_handle_t mch, const uint8_t *addr)
2540 {
2541 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2542 	mac_impl_t		*mip = mcip->mci_mip;
2543 	flow_entry_t		*flent = mcip->mci_flent_list;
2544 	flow_entry_t		*prev_fe = NULL;
2545 	uint16_t		vid;
2546 	int			err = 0;
2547 
2548 	/* Verify the address is a valid multicast address */
2549 	if ((err = mip->mi_type->mt_ops.mtops_multicst_verify(addr,
2550 	    mip->mi_pdata)) != 0)
2551 		return (err);
2552 
2553 	i_mac_perim_enter(mip);
2554 	while (flent != NULL) {
2555 		vid = i_mac_flow_vid(flent);
2556 
2557 		err = mac_bcast_add((mac_client_impl_t *)mch, addr, vid,
2558 		    MAC_ADDRTYPE_MULTICAST);
2559 		if (err != 0)
2560 			break;
2561 		prev_fe = flent;
2562 		flent = flent->fe_client_next;
2563 	}
2564 
2565 	/*
2566 	 * If we failed adding, then undo all, rather than partial
2567 	 * success.
2568 	 */
2569 	if (flent != NULL && prev_fe != NULL) {
2570 		flent = mcip->mci_flent_list;
2571 		while (flent != prev_fe->fe_client_next) {
2572 			vid = i_mac_flow_vid(flent);
2573 			mac_bcast_delete((mac_client_impl_t *)mch, addr, vid);
2574 			flent = flent->fe_client_next;
2575 		}
2576 	}
2577 	i_mac_perim_exit(mip);
2578 	return (err);
2579 }
2580 
2581 /*
2582  * Multicast delete function invoked by MAC clients.
2583  */
2584 void
2585 mac_multicast_remove(mac_client_handle_t mch, const uint8_t *addr)
2586 {
2587 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2588 	mac_impl_t		*mip = mcip->mci_mip;
2589 	flow_entry_t		*flent;
2590 	uint16_t		vid;
2591 
2592 	i_mac_perim_enter(mip);
2593 	for (flent = mcip->mci_flent_list; flent != NULL;
2594 	    flent = flent->fe_client_next) {
2595 		vid = i_mac_flow_vid(flent);
2596 		mac_bcast_delete((mac_client_impl_t *)mch, addr, vid);
2597 	}
2598 	i_mac_perim_exit(mip);
2599 }
2600 
2601 /*
2602  * When a MAC client desires to capture packets on an interface,
2603  * it registers a promiscuous call back with mac_promisc_add().
2604  * There are three types of promiscuous callbacks:
2605  *
2606  * * MAC_CLIENT_PROMISC_ALL
2607  *   Captures all packets sent and received by the MAC client,
2608  *   the physical interface, as well as all other MAC clients
2609  *   defined on top of the same MAC.
2610  *
2611  * * MAC_CLIENT_PROMISC_FILTERED
2612  *   Captures all packets sent and received by the MAC client,
2613  *   plus all multicast traffic sent and received by the phyisical
2614  *   interface and the other MAC clients.
2615  *
2616  * * MAC_CLIENT_PROMISC_MULTI
2617  *   Captures all broadcast and multicast packets sent and
2618  *   received by the MAC clients as well as the physical interface.
2619  *
2620  * In all cases, the underlying MAC is put in promiscuous mode.
2621  */
2622 int
2623 mac_promisc_add(mac_client_handle_t mch, mac_client_promisc_type_t type,
2624     mac_rx_t fn, void *arg, mac_promisc_handle_t *mphp, uint16_t flags)
2625 {
2626 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
2627 	mac_impl_t *mip = mcip->mci_mip;
2628 	mac_promisc_impl_t *mpip;
2629 	mac_cb_info_t	*mcbi;
2630 	int rc;
2631 
2632 	i_mac_perim_enter(mip);
2633 
2634 	if ((rc = mac_start((mac_handle_t)mip)) != 0) {
2635 		i_mac_perim_exit(mip);
2636 		return (rc);
2637 	}
2638 
2639 	if ((mcip->mci_state_flags & MCIS_IS_VNIC) &&
2640 	    type == MAC_CLIENT_PROMISC_ALL) {
2641 		/*
2642 		 * The function is being invoked by the upper MAC client
2643 		 * of a VNIC. The VNIC should only see the traffic
2644 		 * it is entitled to.
2645 		 */
2646 		type = MAC_CLIENT_PROMISC_FILTERED;
2647 	}
2648 
2649 
2650 	/*
2651 	 * Turn on promiscuous mode for the underlying NIC.
2652 	 * This is needed even for filtered callbacks which
2653 	 * expect to receive all multicast traffic on the wire.
2654 	 *
2655 	 * Physical promiscuous mode should not be turned on if
2656 	 * MAC_PROMISC_FLAGS_NO_PHYS is set.
2657 	 */
2658 	if ((flags & MAC_PROMISC_FLAGS_NO_PHYS) == 0) {
2659 		if ((rc = i_mac_promisc_set(mip, B_TRUE)) != 0) {
2660 			mac_stop((mac_handle_t)mip);
2661 			i_mac_perim_exit(mip);
2662 			return (rc);
2663 		}
2664 	}
2665 
2666 	mpip = kmem_cache_alloc(mac_promisc_impl_cache, KM_SLEEP);
2667 
2668 	mpip->mpi_type = type;
2669 	mpip->mpi_fn = fn;
2670 	mpip->mpi_arg = arg;
2671 	mpip->mpi_mcip = mcip;
2672 	mpip->mpi_no_tx_loop = ((flags & MAC_PROMISC_FLAGS_NO_TX_LOOP) != 0);
2673 	mpip->mpi_no_phys = ((flags & MAC_PROMISC_FLAGS_NO_PHYS) != 0);
2674 	mpip->mpi_strip_vlan_tag =
2675 	    ((flags & MAC_PROMISC_FLAGS_VLAN_TAG_STRIP) != 0);
2676 	mpip->mpi_no_copy = ((flags & MAC_PROMISC_FLAGS_NO_COPY) != 0);
2677 
2678 	mcbi = &mip->mi_promisc_cb_info;
2679 	mutex_enter(mcbi->mcbi_lockp);
2680 
2681 	mac_callback_add(&mip->mi_promisc_cb_info, &mcip->mci_promisc_list,
2682 	    &mpip->mpi_mci_link);
2683 	mac_callback_add(&mip->mi_promisc_cb_info, &mip->mi_promisc_list,
2684 	    &mpip->mpi_mi_link);
2685 
2686 	mutex_exit(mcbi->mcbi_lockp);
2687 
2688 	*mphp = (mac_promisc_handle_t)mpip;
2689 	i_mac_perim_exit(mip);
2690 	return (0);
2691 }
2692 
2693 /*
2694  * Remove a multicast address previously aded through mac_promisc_add().
2695  */
2696 void
2697 mac_promisc_remove(mac_promisc_handle_t mph)
2698 {
2699 	mac_promisc_impl_t *mpip = (mac_promisc_impl_t *)mph;
2700 	mac_client_impl_t *mcip = mpip->mpi_mcip;
2701 	mac_impl_t *mip = mcip->mci_mip;
2702 	mac_cb_info_t *mcbi;
2703 	int rv;
2704 
2705 	i_mac_perim_enter(mip);
2706 
2707 	/*
2708 	 * Even if the device can't be reset into normal mode, we still
2709 	 * need to clear the client promisc callbacks. The client may want
2710 	 * to close the mac end point and we can't have stale callbacks.
2711 	 */
2712 	if (!(mpip->mpi_no_phys)) {
2713 		if ((rv = i_mac_promisc_set(mip, B_FALSE)) != 0) {
2714 			cmn_err(CE_WARN, "%s: failed to switch OFF promiscuous"
2715 			    " mode because of error 0x%x", mip->mi_name, rv);
2716 		}
2717 	}
2718 	mcbi = &mip->mi_promisc_cb_info;
2719 	mutex_enter(mcbi->mcbi_lockp);
2720 	if (mac_callback_remove(mcbi, &mip->mi_promisc_list,
2721 	    &mpip->mpi_mi_link)) {
2722 		VERIFY(mac_callback_remove(&mip->mi_promisc_cb_info,
2723 		    &mcip->mci_promisc_list, &mpip->mpi_mci_link));
2724 		kmem_cache_free(mac_promisc_impl_cache, mpip);
2725 	} else {
2726 		mac_callback_remove_wait(&mip->mi_promisc_cb_info);
2727 	}
2728 	mutex_exit(mcbi->mcbi_lockp);
2729 	mac_stop((mac_handle_t)mip);
2730 
2731 	i_mac_perim_exit(mip);
2732 }
2733 
2734 /*
2735  * Reference count the number of active Tx threads. MCI_TX_QUIESCE indicates
2736  * that a control operation wants to quiesce the Tx data flow in which case
2737  * we return an error. Holding any of the per cpu locks ensures that the
2738  * mci_tx_flag won't change.
2739  *
2740  * 'CPU' must be accessed just once and used to compute the index into the
2741  * percpu array, and that index must be used for the entire duration of the
2742  * packet send operation. Note that the thread may be preempted and run on
2743  * another cpu any time and so we can't use 'CPU' more than once for the
2744  * operation.
2745  */
2746 #define	MAC_TX_TRY_HOLD(mcip, mytx, error)				\
2747 {									\
2748 	(error) = 0;							\
2749 	(mytx) = &(mcip)->mci_tx_pcpu[CPU->cpu_seqid & mac_tx_percpu_cnt]; \
2750 	mutex_enter(&(mytx)->pcpu_tx_lock);				\
2751 	if (!((mcip)->mci_tx_flag & MCI_TX_QUIESCE)) {			\
2752 		(mytx)->pcpu_tx_refcnt++;				\
2753 	} else {							\
2754 		(error) = -1;						\
2755 	}								\
2756 	mutex_exit(&(mytx)->pcpu_tx_lock);				\
2757 }
2758 
2759 /*
2760  * Release the reference. If needed, signal any control operation waiting
2761  * for Tx quiescence. The wait and signal are always done using the
2762  * mci_tx_pcpu[0]'s lock
2763  */
2764 #define	MAC_TX_RELE(mcip, mytx) {					\
2765 	mutex_enter(&(mytx)->pcpu_tx_lock);				\
2766 	if (--(mytx)->pcpu_tx_refcnt == 0 &&				\
2767 	    (mcip)->mci_tx_flag & MCI_TX_QUIESCE) {			\
2768 		mutex_exit(&(mytx)->pcpu_tx_lock);			\
2769 		mutex_enter(&(mcip)->mci_tx_pcpu[0].pcpu_tx_lock);	\
2770 		cv_signal(&(mcip)->mci_tx_cv);				\
2771 		mutex_exit(&(mcip)->mci_tx_pcpu[0].pcpu_tx_lock);	\
2772 	} else {							\
2773 		mutex_exit(&(mytx)->pcpu_tx_lock);			\
2774 	}								\
2775 }
2776 
2777 /*
2778  * Bump the count of the number of active Tx threads. This is maintained as
2779  * a per CPU counter. On (CMT kind of) machines with large number of CPUs,
2780  * a single mci_tx_lock may become contended. However a count of the total
2781  * number of Tx threads per client is needed in order to quiesce the Tx side
2782  * prior to reassigning a Tx ring dynamically to another client. The thread
2783  * that needs to quiesce the Tx traffic grabs all the percpu locks and checks
2784  * the sum of the individual percpu refcnts. Each Tx data thread only grabs
2785  * its own percpu lock and increments its own refcnt.
2786  */
2787 void *
2788 mac_tx_hold(mac_client_handle_t mch)
2789 {
2790 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
2791 	mac_tx_percpu_t	*mytx;
2792 	int error;
2793 
2794 	MAC_TX_TRY_HOLD(mcip, mytx, error);
2795 	return (error == 0 ? (void *)mytx : NULL);
2796 }
2797 
2798 void
2799 mac_tx_rele(mac_client_handle_t mch, void *mytx_handle)
2800 {
2801 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
2802 	mac_tx_percpu_t	*mytx = mytx_handle;
2803 
2804 	MAC_TX_RELE(mcip, mytx)
2805 }
2806 
2807 /*
2808  * Send function invoked by MAC clients.
2809  */
2810 mac_tx_cookie_t
2811 mac_tx(mac_client_handle_t mch, mblk_t *mp_chain, uintptr_t hint,
2812     uint16_t flag, mblk_t **ret_mp)
2813 {
2814 	mac_tx_cookie_t		cookie = NULL;
2815 	int			error;
2816 	mac_tx_percpu_t		*mytx;
2817 	mac_soft_ring_set_t	*srs;
2818 	flow_entry_t		*flent;
2819 	boolean_t		is_subflow = B_FALSE;
2820 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
2821 	mac_impl_t		*mip = mcip->mci_mip;
2822 	mac_srs_tx_t		*srs_tx;
2823 
2824 	/*
2825 	 * Check whether the active Tx threads count is bumped already.
2826 	 */
2827 	if (!(flag & MAC_TX_NO_HOLD)) {
2828 		MAC_TX_TRY_HOLD(mcip, mytx, error);
2829 		if (error != 0) {
2830 			freemsgchain(mp_chain);
2831 			return (NULL);
2832 		}
2833 	}
2834 
2835 	/*
2836 	 * If mac protection is enabled, only the permissible packets will be
2837 	 * returned by mac_protect_check().
2838 	 */
2839 	if ((mcip->mci_flent->
2840 	    fe_resource_props.mrp_mask & MRP_PROTECT) != 0 &&
2841 	    (mp_chain = mac_protect_check(mch, mp_chain)) == NULL)
2842 		goto done;
2843 
2844 	if (mcip->mci_subflow_tab != NULL &&
2845 	    mcip->mci_subflow_tab->ft_flow_count > 0 &&
2846 	    mac_flow_lookup(mcip->mci_subflow_tab, mp_chain,
2847 	    FLOW_OUTBOUND, &flent) == 0) {
2848 		/*
2849 		 * The main assumption here is that if in the event
2850 		 * we get a chain, all the packets will be classified
2851 		 * to the same Flow/SRS. If this changes for any
2852 		 * reason, the following logic should change as well.
2853 		 * I suppose the fanout_hint also assumes this .
2854 		 */
2855 		ASSERT(flent != NULL);
2856 		is_subflow = B_TRUE;
2857 	} else {
2858 		flent = mcip->mci_flent;
2859 	}
2860 
2861 	srs = flent->fe_tx_srs;
2862 	/*
2863 	 * This is to avoid panics with PF_PACKET that can call mac_tx()
2864 	 * against an interface that is not capable of sending. A rewrite
2865 	 * of the mac datapath is required to remove this limitation.
2866 	 */
2867 	if (srs == NULL) {
2868 		freemsgchain(mp_chain);
2869 		goto done;
2870 	}
2871 
2872 	srs_tx = &srs->srs_tx;
2873 	if (srs_tx->st_mode == SRS_TX_DEFAULT &&
2874 	    (srs->srs_state & SRS_ENQUEUED) == 0 &&
2875 	    mip->mi_nactiveclients == 1 && mip->mi_promisc_list == NULL &&
2876 	    mp_chain->b_next == NULL) {
2877 		uint64_t	obytes;
2878 
2879 		/*
2880 		 * Since dls always opens the underlying MAC, nclients equals
2881 		 * to 1 means that the only active client is dls itself acting
2882 		 * as a primary client of the MAC instance. Since dls will not
2883 		 * send tagged packets in that case, and dls is trusted to send
2884 		 * packets for its allowed VLAN(s), the VLAN tag insertion and
2885 		 * check is required only if nclients is greater than 1.
2886 		 */
2887 		if (mip->mi_nclients > 1) {
2888 			if (MAC_VID_CHECK_NEEDED(mcip)) {
2889 				int	err = 0;
2890 
2891 				MAC_VID_CHECK(mcip, mp_chain, err);
2892 				if (err != 0) {
2893 					freemsg(mp_chain);
2894 					mcip->mci_stat_oerrors++;
2895 					goto done;
2896 				}
2897 			}
2898 			if (MAC_TAG_NEEDED(mcip)) {
2899 				mp_chain = mac_add_vlan_tag(mp_chain, 0,
2900 				    mac_client_vid(mch));
2901 				if (mp_chain == NULL) {
2902 					mcip->mci_stat_oerrors++;
2903 					goto done;
2904 				}
2905 			}
2906 		}
2907 
2908 		obytes = (mp_chain->b_cont == NULL ? MBLKL(mp_chain) :
2909 		    msgdsize(mp_chain));
2910 
2911 		MAC_TX(mip, srs_tx->st_arg2, mp_chain,
2912 		    ((mcip->mci_state_flags & MCIS_SHARE_BOUND) != 0));
2913 
2914 		if (mp_chain == NULL) {
2915 			cookie = NULL;
2916 			mcip->mci_stat_obytes += obytes;
2917 			mcip->mci_stat_opackets += 1;
2918 			if ((srs->srs_type & SRST_FLOW) != 0) {
2919 				FLOW_STAT_UPDATE(flent, obytes, obytes);
2920 				FLOW_STAT_UPDATE(flent, opackets, 1);
2921 			}
2922 		} else {
2923 			mutex_enter(&srs->srs_lock);
2924 			cookie = mac_tx_srs_no_desc(srs, mp_chain,
2925 			    flag, ret_mp);
2926 			mutex_exit(&srs->srs_lock);
2927 		}
2928 	} else {
2929 		cookie = srs_tx->st_func(srs, mp_chain, hint, flag, ret_mp);
2930 	}
2931 
2932 done:
2933 	if (is_subflow)
2934 		FLOW_REFRELE(flent);
2935 
2936 	if (!(flag & MAC_TX_NO_HOLD))
2937 		MAC_TX_RELE(mcip, mytx);
2938 
2939 	return (cookie);
2940 }
2941 
2942 /*
2943  * mac_tx_is_blocked
2944  *
2945  * Given a cookie, it returns if the ring identified by the cookie is
2946  * flow-controlled or not. If NULL is passed in place of a cookie,
2947  * then it finds out if any of the underlying rings belonging to the
2948  * SRS is flow controlled or not and returns that status.
2949  */
2950 /* ARGSUSED */
2951 boolean_t
2952 mac_tx_is_flow_blocked(mac_client_handle_t mch, mac_tx_cookie_t cookie)
2953 {
2954 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
2955 	mac_soft_ring_set_t *mac_srs;
2956 	mac_soft_ring_t *sringp;
2957 	boolean_t blocked = B_FALSE;
2958 	mac_tx_percpu_t *mytx;
2959 	int err;
2960 	int i;
2961 
2962 	/*
2963 	 * Bump the reference count so that mac_srs won't be deleted.
2964 	 * If the client is currently quiesced and we failed to bump
2965 	 * the reference, return B_TRUE so that flow control stays
2966 	 * as enabled.
2967 	 *
2968 	 * Flow control will then be disabled once the client is no
2969 	 * longer quiesced.
2970 	 */
2971 	MAC_TX_TRY_HOLD(mcip, mytx, err);
2972 	if (err != 0)
2973 		return (B_TRUE);
2974 
2975 	if ((mac_srs = MCIP_TX_SRS(mcip)) == NULL) {
2976 		MAC_TX_RELE(mcip, mytx);
2977 		return (B_FALSE);
2978 	}
2979 
2980 	mutex_enter(&mac_srs->srs_lock);
2981 	if (mac_srs->srs_tx.st_mode == SRS_TX_FANOUT) {
2982 		if (cookie != NULL) {
2983 			sringp = (mac_soft_ring_t *)cookie;
2984 			mutex_enter(&sringp->s_ring_lock);
2985 			if (sringp->s_ring_state & S_RING_TX_HIWAT)
2986 				blocked = B_TRUE;
2987 			mutex_exit(&sringp->s_ring_lock);
2988 		} else {
2989 			for (i = 0; i < mac_srs->srs_oth_ring_count; i++) {
2990 				sringp = mac_srs->srs_oth_soft_rings[i];
2991 				mutex_enter(&sringp->s_ring_lock);
2992 				if (sringp->s_ring_state & S_RING_TX_HIWAT) {
2993 					blocked = B_TRUE;
2994 					mutex_exit(&sringp->s_ring_lock);
2995 					break;
2996 				}
2997 				mutex_exit(&sringp->s_ring_lock);
2998 			}
2999 		}
3000 	} else {
3001 		blocked = (mac_srs->srs_state & SRS_TX_HIWAT);
3002 	}
3003 	mutex_exit(&mac_srs->srs_lock);
3004 	MAC_TX_RELE(mcip, mytx);
3005 	return (blocked);
3006 }
3007 
3008 /*
3009  * Check if the MAC client is the primary MAC client.
3010  */
3011 boolean_t
3012 mac_is_primary_client(mac_client_impl_t *mcip)
3013 {
3014 	return (mcip->mci_flags & MAC_CLIENT_FLAGS_PRIMARY);
3015 }
3016 
3017 void
3018 mac_ioctl(mac_handle_t mh, queue_t *wq, mblk_t *bp)
3019 {
3020 	mac_impl_t	*mip = (mac_impl_t *)mh;
3021 	int cmd = ((struct iocblk *)bp->b_rptr)->ioc_cmd;
3022 
3023 	if ((cmd == ND_GET && (mip->mi_callbacks->mc_callbacks & MC_GETPROP)) ||
3024 	    (cmd == ND_SET && (mip->mi_callbacks->mc_callbacks & MC_SETPROP))) {
3025 		/*
3026 		 * If ndd props were registered, call them.
3027 		 * Note that ndd ioctls are Obsolete
3028 		 */
3029 		mac_ndd_ioctl(mip, wq, bp);
3030 		return;
3031 	}
3032 
3033 	/*
3034 	 * Call the driver to handle the ioctl.  The driver may not support
3035 	 * any ioctls, in which case we reply with a NAK on its behalf.
3036 	 */
3037 	if (mip->mi_callbacks->mc_callbacks & MC_IOCTL)
3038 		mip->mi_ioctl(mip->mi_driver, wq, bp);
3039 	else
3040 		miocnak(wq, bp, 0, EINVAL);
3041 }
3042 
3043 /*
3044  * Return the link state of the specified MAC instance.
3045  */
3046 link_state_t
3047 mac_link_get(mac_handle_t mh)
3048 {
3049 	return (((mac_impl_t *)mh)->mi_linkstate);
3050 }
3051 
3052 /*
3053  * Add a mac client specified notification callback. Please see the comments
3054  * above mac_callback_add() for general information about mac callback
3055  * addition/deletion in the presence of mac callback list walkers
3056  */
3057 mac_notify_handle_t
3058 mac_notify_add(mac_handle_t mh, mac_notify_t notify_fn, void *arg)
3059 {
3060 	mac_impl_t		*mip = (mac_impl_t *)mh;
3061 	mac_notify_cb_t		*mncb;
3062 	mac_cb_info_t		*mcbi;
3063 
3064 	/*
3065 	 * Allocate a notify callback structure, fill in the details and
3066 	 * use the mac callback list manipulation functions to chain into
3067 	 * the list of callbacks.
3068 	 */
3069 	mncb = kmem_zalloc(sizeof (mac_notify_cb_t), KM_SLEEP);
3070 	mncb->mncb_fn = notify_fn;
3071 	mncb->mncb_arg = arg;
3072 	mncb->mncb_mip = mip;
3073 	mncb->mncb_link.mcb_objp = mncb;
3074 	mncb->mncb_link.mcb_objsize = sizeof (mac_notify_cb_t);
3075 	mncb->mncb_link.mcb_flags = MCB_NOTIFY_CB_T;
3076 
3077 	mcbi = &mip->mi_notify_cb_info;
3078 
3079 	i_mac_perim_enter(mip);
3080 	mutex_enter(mcbi->mcbi_lockp);
3081 
3082 	mac_callback_add(&mip->mi_notify_cb_info, &mip->mi_notify_cb_list,
3083 	    &mncb->mncb_link);
3084 
3085 	mutex_exit(mcbi->mcbi_lockp);
3086 	i_mac_perim_exit(mip);
3087 	return ((mac_notify_handle_t)mncb);
3088 }
3089 
3090 void
3091 mac_notify_remove_wait(mac_handle_t mh)
3092 {
3093 	mac_impl_t	*mip = (mac_impl_t *)mh;
3094 	mac_cb_info_t	*mcbi = &mip->mi_notify_cb_info;
3095 
3096 	mutex_enter(mcbi->mcbi_lockp);
3097 	mac_callback_remove_wait(&mip->mi_notify_cb_info);
3098 	mutex_exit(mcbi->mcbi_lockp);
3099 }
3100 
3101 /*
3102  * Remove a mac client specified notification callback
3103  */
3104 int
3105 mac_notify_remove(mac_notify_handle_t mnh, boolean_t wait)
3106 {
3107 	mac_notify_cb_t	*mncb = (mac_notify_cb_t *)mnh;
3108 	mac_impl_t	*mip = mncb->mncb_mip;
3109 	mac_cb_info_t	*mcbi;
3110 	int		err = 0;
3111 
3112 	mcbi = &mip->mi_notify_cb_info;
3113 
3114 	i_mac_perim_enter(mip);
3115 	mutex_enter(mcbi->mcbi_lockp);
3116 
3117 	ASSERT(mncb->mncb_link.mcb_objp == mncb);
3118 	/*
3119 	 * If there aren't any list walkers, the remove would succeed
3120 	 * inline, else we wait for the deferred remove to complete
3121 	 */
3122 	if (mac_callback_remove(&mip->mi_notify_cb_info,
3123 	    &mip->mi_notify_cb_list, &mncb->mncb_link)) {
3124 		kmem_free(mncb, sizeof (mac_notify_cb_t));
3125 	} else {
3126 		err = EBUSY;
3127 	}
3128 
3129 	mutex_exit(mcbi->mcbi_lockp);
3130 	i_mac_perim_exit(mip);
3131 
3132 	/*
3133 	 * If we failed to remove the notification callback and "wait" is set
3134 	 * to be B_TRUE, wait for the callback to finish after we exit the
3135 	 * mac perimeter.
3136 	 */
3137 	if (err != 0 && wait) {
3138 		mac_notify_remove_wait((mac_handle_t)mip);
3139 		return (0);
3140 	}
3141 
3142 	return (err);
3143 }
3144 
3145 /*
3146  * Associate resource management callbacks with the specified MAC
3147  * clients.
3148  */
3149 
3150 void
3151 mac_resource_set_common(mac_client_handle_t mch, mac_resource_add_t add,
3152     mac_resource_remove_t remove, mac_resource_quiesce_t quiesce,
3153     mac_resource_restart_t restart, mac_resource_bind_t bind,
3154     void *arg)
3155 {
3156 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3157 
3158 	mcip->mci_resource_add = add;
3159 	mcip->mci_resource_remove = remove;
3160 	mcip->mci_resource_quiesce = quiesce;
3161 	mcip->mci_resource_restart = restart;
3162 	mcip->mci_resource_bind = bind;
3163 	mcip->mci_resource_arg = arg;
3164 }
3165 
3166 void
3167 mac_resource_set(mac_client_handle_t mch, mac_resource_add_t add, void *arg)
3168 {
3169 	/* update the 'resource_add' callback */
3170 	mac_resource_set_common(mch, add, NULL, NULL, NULL, NULL, arg);
3171 }
3172 
3173 /*
3174  * Sets up the client resources and enable the polling interface over all the
3175  * SRS's and the soft rings of the client
3176  */
3177 void
3178 mac_client_poll_enable(mac_client_handle_t mch)
3179 {
3180 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3181 	mac_soft_ring_set_t	*mac_srs;
3182 	flow_entry_t		*flent;
3183 	int			i;
3184 
3185 	flent = mcip->mci_flent;
3186 	ASSERT(flent != NULL);
3187 
3188 	mcip->mci_state_flags |= MCIS_CLIENT_POLL_CAPABLE;
3189 	for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
3190 		mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
3191 		ASSERT(mac_srs->srs_mcip == mcip);
3192 		mac_srs_client_poll_enable(mcip, mac_srs);
3193 	}
3194 }
3195 
3196 /*
3197  * Tears down the client resources and disable the polling interface over all
3198  * the SRS's and the soft rings of the client
3199  */
3200 void
3201 mac_client_poll_disable(mac_client_handle_t mch)
3202 {
3203 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3204 	mac_soft_ring_set_t	*mac_srs;
3205 	flow_entry_t		*flent;
3206 	int			i;
3207 
3208 	flent = mcip->mci_flent;
3209 	ASSERT(flent != NULL);
3210 
3211 	mcip->mci_state_flags &= ~MCIS_CLIENT_POLL_CAPABLE;
3212 	for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
3213 		mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
3214 		ASSERT(mac_srs->srs_mcip == mcip);
3215 		mac_srs_client_poll_disable(mcip, mac_srs);
3216 	}
3217 }
3218 
3219 /*
3220  * Associate the CPUs specified by the given property with a MAC client.
3221  */
3222 int
3223 mac_cpu_set(mac_client_handle_t mch, mac_resource_props_t *mrp)
3224 {
3225 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3226 	mac_impl_t *mip = mcip->mci_mip;
3227 	int err = 0;
3228 
3229 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
3230 
3231 	if ((err = mac_validate_props(mrp)) != 0)
3232 		return (err);
3233 
3234 	if (MCIP_DATAPATH_SETUP(mcip))
3235 		mac_flow_modify(mip->mi_flow_tab, mcip->mci_flent, mrp);
3236 
3237 	mac_update_resources(mrp, MCIP_RESOURCE_PROPS(mcip), B_FALSE);
3238 	return (0);
3239 }
3240 
3241 /*
3242  * Apply the specified properties to the specified MAC client.
3243  */
3244 int
3245 mac_client_set_resources(mac_client_handle_t mch, mac_resource_props_t *mrp)
3246 {
3247 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
3248 	mac_impl_t *mip = mcip->mci_mip;
3249 	int err = 0;
3250 
3251 	i_mac_perim_enter(mip);
3252 
3253 	if ((mrp->mrp_mask & MRP_MAXBW) || (mrp->mrp_mask & MRP_PRIORITY)) {
3254 		err = mac_resource_ctl_set(mch, mrp);
3255 		if (err != 0)
3256 			goto done;
3257 	}
3258 
3259 	if (mrp->mrp_mask & MRP_CPUS) {
3260 		err = mac_cpu_set(mch, mrp);
3261 		if (err != 0)
3262 			goto done;
3263 	}
3264 
3265 	if (mrp->mrp_mask & MRP_PROTECT)
3266 		err = mac_protect_set(mch, mrp);
3267 
3268 done:
3269 	i_mac_perim_exit(mip);
3270 	return (err);
3271 }
3272 
3273 /*
3274  * Return the properties currently associated with the specified MAC client.
3275  */
3276 void
3277 mac_client_get_resources(mac_client_handle_t mch, mac_resource_props_t *mrp)
3278 {
3279 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
3280 	mac_resource_props_t	*mcip_mrp = MCIP_RESOURCE_PROPS(mcip);
3281 
3282 	bcopy(mcip_mrp, mrp, sizeof (mac_resource_props_t));
3283 }
3284 
3285 /*
3286  * Pass a copy of the specified packet to the promiscuous callbacks
3287  * of the specified MAC.
3288  *
3289  * If sender is NULL, the function is being invoked for a packet chain
3290  * received from the wire. If sender is non-NULL, it points to
3291  * the MAC client from which the packet is being sent.
3292  *
3293  * The packets are distributed to the promiscuous callbacks as follows:
3294  *
3295  * - all packets are sent to the MAC_CLIENT_PROMISC_ALL callbacks
3296  * - all broadcast and multicast packets are sent to the
3297  *   MAC_CLIENT_PROMISC_FILTER and MAC_CLIENT_PROMISC_MULTI.
3298  *
3299  * The unicast packets of MAC_CLIENT_PROMISC_FILTER callbacks are dispatched
3300  * after classification by mac_rx_deliver().
3301  */
3302 
3303 static void
3304 mac_promisc_dispatch_one(mac_promisc_impl_t *mpip, mblk_t *mp,
3305     boolean_t loopback)
3306 {
3307 	mblk_t *mp_copy, *mp_next;
3308 
3309 	if (!mpip->mpi_no_copy || mpip->mpi_strip_vlan_tag) {
3310 		mp_copy = copymsg(mp);
3311 		if (mp_copy == NULL)
3312 			return;
3313 
3314 		if (mpip->mpi_strip_vlan_tag) {
3315 			mp_copy = mac_strip_vlan_tag_chain(mp_copy);
3316 			if (mp_copy == NULL)
3317 				return;
3318 		}
3319 		mp_next = NULL;
3320 	} else {
3321 		mp_copy = mp;
3322 		mp_next = mp->b_next;
3323 	}
3324 	mp_copy->b_next = NULL;
3325 
3326 	mpip->mpi_fn(mpip->mpi_arg, NULL, mp_copy, loopback);
3327 	if (mp_copy == mp)
3328 		mp->b_next = mp_next;
3329 }
3330 
3331 /*
3332  * Return the VID of a packet. Zero if the packet is not tagged.
3333  */
3334 static uint16_t
3335 mac_ether_vid(mblk_t *mp)
3336 {
3337 	struct ether_header *eth = (struct ether_header *)mp->b_rptr;
3338 
3339 	if (ntohs(eth->ether_type) == ETHERTYPE_VLAN) {
3340 		struct ether_vlan_header *t_evhp =
3341 		    (struct ether_vlan_header *)mp->b_rptr;
3342 		return (VLAN_ID(ntohs(t_evhp->ether_tci)));
3343 	}
3344 
3345 	return (0);
3346 }
3347 
3348 /*
3349  * Return whether the specified packet contains a multicast or broadcast
3350  * destination MAC address.
3351  */
3352 static boolean_t
3353 mac_is_mcast(mac_impl_t *mip, mblk_t *mp)
3354 {
3355 	mac_header_info_t hdr_info;
3356 
3357 	if (mac_header_info((mac_handle_t)mip, mp, &hdr_info) != 0)
3358 		return (B_FALSE);
3359 	return ((hdr_info.mhi_dsttype == MAC_ADDRTYPE_BROADCAST) ||
3360 	    (hdr_info.mhi_dsttype == MAC_ADDRTYPE_MULTICAST));
3361 }
3362 
3363 /*
3364  * Send a copy of an mblk chain to the MAC clients of the specified MAC.
3365  * "sender" points to the sender MAC client for outbound packets, and
3366  * is set to NULL for inbound packets.
3367  */
3368 void
3369 mac_promisc_dispatch(mac_impl_t *mip, mblk_t *mp_chain,
3370     mac_client_impl_t *sender)
3371 {
3372 	mac_promisc_impl_t *mpip;
3373 	mac_cb_t *mcb;
3374 	mblk_t *mp;
3375 	boolean_t is_mcast, is_sender;
3376 
3377 	MAC_PROMISC_WALKER_INC(mip);
3378 	for (mp = mp_chain; mp != NULL; mp = mp->b_next) {
3379 		is_mcast = mac_is_mcast(mip, mp);
3380 		/* send packet to interested callbacks */
3381 		for (mcb = mip->mi_promisc_list; mcb != NULL;
3382 		    mcb = mcb->mcb_nextp) {
3383 			mpip = (mac_promisc_impl_t *)mcb->mcb_objp;
3384 			is_sender = (mpip->mpi_mcip == sender);
3385 
3386 			if (is_sender && mpip->mpi_no_tx_loop)
3387 				/*
3388 				 * The sender doesn't want to receive
3389 				 * copies of the packets it sends.
3390 				 */
3391 				continue;
3392 
3393 			/* this client doesn't need any packets (bridge) */
3394 			if (mpip->mpi_fn == NULL)
3395 				continue;
3396 
3397 			/*
3398 			 * For an ethernet MAC, don't displatch a multicast
3399 			 * packet to a non-PROMISC_ALL callbacks unless the VID
3400 			 * of the packet matches the VID of the client.
3401 			 */
3402 			if (is_mcast &&
3403 			    mpip->mpi_type != MAC_CLIENT_PROMISC_ALL &&
3404 			    !mac_client_check_flow_vid(mpip->mpi_mcip,
3405 			    mac_ether_vid(mp)))
3406 				continue;
3407 
3408 			if (is_sender ||
3409 			    mpip->mpi_type == MAC_CLIENT_PROMISC_ALL ||
3410 			    is_mcast)
3411 				mac_promisc_dispatch_one(mpip, mp, is_sender);
3412 		}
3413 	}
3414 	MAC_PROMISC_WALKER_DCR(mip);
3415 }
3416 
3417 void
3418 mac_promisc_client_dispatch(mac_client_impl_t *mcip, mblk_t *mp_chain)
3419 {
3420 	mac_impl_t		*mip = mcip->mci_mip;
3421 	mac_promisc_impl_t	*mpip;
3422 	boolean_t		is_mcast;
3423 	mblk_t			*mp;
3424 	mac_cb_t		*mcb;
3425 
3426 	/*
3427 	 * The unicast packets for the MAC client still
3428 	 * need to be delivered to the MAC_CLIENT_PROMISC_FILTERED
3429 	 * promiscuous callbacks. The broadcast and multicast
3430 	 * packets were delivered from mac_rx().
3431 	 */
3432 	MAC_PROMISC_WALKER_INC(mip);
3433 	for (mp = mp_chain; mp != NULL; mp = mp->b_next) {
3434 		is_mcast = mac_is_mcast(mip, mp);
3435 		for (mcb = mcip->mci_promisc_list; mcb != NULL;
3436 		    mcb = mcb->mcb_nextp) {
3437 			mpip = (mac_promisc_impl_t *)mcb->mcb_objp;
3438 			if (mpip->mpi_type == MAC_CLIENT_PROMISC_FILTERED &&
3439 			    !is_mcast) {
3440 				mac_promisc_dispatch_one(mpip, mp, B_FALSE);
3441 			}
3442 		}
3443 	}
3444 	MAC_PROMISC_WALKER_DCR(mip);
3445 }
3446 
3447 /*
3448  * Return the margin value currently assigned to the specified MAC instance.
3449  */
3450 void
3451 mac_margin_get(mac_handle_t mh, uint32_t *marginp)
3452 {
3453 	mac_impl_t *mip = (mac_impl_t *)mh;
3454 
3455 	rw_enter(&(mip->mi_rw_lock), RW_READER);
3456 	*marginp = mip->mi_margin;
3457 	rw_exit(&(mip->mi_rw_lock));
3458 }
3459 
3460 /*
3461  * mac_info_get() is used for retrieving the mac_info when a DL_INFO_REQ is
3462  * issued before a DL_ATTACH_REQ. we walk the i_mac_impl_hash table and find
3463  * the first mac_impl_t with a matching driver name; then we copy its mac_info_t
3464  * to the caller. we do all this with i_mac_impl_lock held so the mac_impl_t
3465  * cannot disappear while we are accessing it.
3466  */
3467 typedef struct i_mac_info_state_s {
3468 	const char	*mi_name;
3469 	mac_info_t	*mi_infop;
3470 } i_mac_info_state_t;
3471 
3472 /*ARGSUSED*/
3473 static uint_t
3474 i_mac_info_walker(mod_hash_key_t key, mod_hash_val_t *val, void *arg)
3475 {
3476 	i_mac_info_state_t *statep = arg;
3477 	mac_impl_t *mip = (mac_impl_t *)val;
3478 
3479 	if (mip->mi_state_flags & MIS_DISABLED)
3480 		return (MH_WALK_CONTINUE);
3481 
3482 	if (strcmp(statep->mi_name,
3483 	    ddi_driver_name(mip->mi_dip)) != 0)
3484 		return (MH_WALK_CONTINUE);
3485 
3486 	statep->mi_infop = &mip->mi_info;
3487 	return (MH_WALK_TERMINATE);
3488 }
3489 
3490 boolean_t
3491 mac_info_get(const char *name, mac_info_t *minfop)
3492 {
3493 	i_mac_info_state_t state;
3494 
3495 	rw_enter(&i_mac_impl_lock, RW_READER);
3496 	state.mi_name = name;
3497 	state.mi_infop = NULL;
3498 	mod_hash_walk(i_mac_impl_hash, i_mac_info_walker, &state);
3499 	if (state.mi_infop == NULL) {
3500 		rw_exit(&i_mac_impl_lock);
3501 		return (B_FALSE);
3502 	}
3503 	*minfop = *state.mi_infop;
3504 	rw_exit(&i_mac_impl_lock);
3505 	return (B_TRUE);
3506 }
3507 
3508 /*
3509  * To get the capabilities that MAC layer cares about, such as rings, factory
3510  * mac address, vnic or not, it should directly invoke this function.  If the
3511  * link is part of a bridge, then the only "capability" it has is the inability
3512  * to do zero copy.
3513  */
3514 boolean_t
3515 i_mac_capab_get(mac_handle_t mh, mac_capab_t cap, void *cap_data)
3516 {
3517 	mac_impl_t *mip = (mac_impl_t *)mh;
3518 
3519 	if (mip->mi_bridge_link != NULL)
3520 		return (cap == MAC_CAPAB_NO_ZCOPY);
3521 	else if (mip->mi_callbacks->mc_callbacks & MC_GETCAPAB)
3522 		return (mip->mi_getcapab(mip->mi_driver, cap, cap_data));
3523 	else
3524 		return (B_FALSE);
3525 }
3526 
3527 /*
3528  * Capability query function. If number of active mac clients is greater than
3529  * 1, only limited capabilities can be advertised to the caller no matter the
3530  * driver has certain capability or not. Else, we query the driver to get the
3531  * capability.
3532  */
3533 boolean_t
3534 mac_capab_get(mac_handle_t mh, mac_capab_t cap, void *cap_data)
3535 {
3536 	mac_impl_t *mip = (mac_impl_t *)mh;
3537 
3538 	/*
3539 	 * if mi_nactiveclients > 1, only MAC_CAPAB_LEGACY, MAC_CAPAB_HCKSUM,
3540 	 * MAC_CAPAB_NO_NATIVEVLAN and MAC_CAPAB_NO_ZCOPY can be advertised.
3541 	 */
3542 	if (mip->mi_nactiveclients > 1) {
3543 		switch (cap) {
3544 		case MAC_CAPAB_NO_NATIVEVLAN:
3545 		case MAC_CAPAB_NO_ZCOPY:
3546 			return (B_TRUE);
3547 		case MAC_CAPAB_LEGACY:
3548 		case MAC_CAPAB_HCKSUM:
3549 			break;
3550 		default:
3551 			return (B_FALSE);
3552 		}
3553 	}
3554 
3555 	/* else get capab from driver */
3556 	return (i_mac_capab_get(mh, cap, cap_data));
3557 }
3558 
3559 boolean_t
3560 mac_sap_verify(mac_handle_t mh, uint32_t sap, uint32_t *bind_sap)
3561 {
3562 	mac_impl_t *mip = (mac_impl_t *)mh;
3563 
3564 	return (mip->mi_type->mt_ops.mtops_sap_verify(sap, bind_sap,
3565 	    mip->mi_pdata));
3566 }
3567 
3568 mblk_t *
3569 mac_header(mac_handle_t mh, const uint8_t *daddr, uint32_t sap, mblk_t *payload,
3570     size_t extra_len)
3571 {
3572 	mac_impl_t	*mip = (mac_impl_t *)mh;
3573 	const uint8_t	*hdr_daddr;
3574 
3575 	/*
3576 	 * If the MAC is point-to-point with a fixed destination address, then
3577 	 * we must always use that destination in the MAC header.
3578 	 */
3579 	hdr_daddr = (mip->mi_dstaddr_set ? mip->mi_dstaddr : daddr);
3580 	return (mip->mi_type->mt_ops.mtops_header(mip->mi_addr, hdr_daddr, sap,
3581 	    mip->mi_pdata, payload, extra_len));
3582 }
3583 
3584 int
3585 mac_header_info(mac_handle_t mh, mblk_t *mp, mac_header_info_t *mhip)
3586 {
3587 	mac_impl_t *mip = (mac_impl_t *)mh;
3588 
3589 	return (mip->mi_type->mt_ops.mtops_header_info(mp, mip->mi_pdata,
3590 	    mhip));
3591 }
3592 
3593 int
3594 mac_vlan_header_info(mac_handle_t mh, mblk_t *mp, mac_header_info_t *mhip)
3595 {
3596 	mac_impl_t	*mip = (mac_impl_t *)mh;
3597 	boolean_t	is_ethernet = (mip->mi_info.mi_media == DL_ETHER);
3598 	int		err = 0;
3599 
3600 	/*
3601 	 * Packets should always be at least 16 bit aligned.
3602 	 */
3603 	ASSERT(IS_P2ALIGNED(mp->b_rptr, sizeof (uint16_t)));
3604 
3605 	if ((err = mac_header_info(mh, mp, mhip)) != 0)
3606 		return (err);
3607 
3608 	/*
3609 	 * If this is a VLAN-tagged Ethernet packet, then the SAP in the
3610 	 * mac_header_info_t as returned by mac_header_info() is
3611 	 * ETHERTYPE_VLAN. We need to grab the ethertype from the VLAN header.
3612 	 */
3613 	if (is_ethernet && (mhip->mhi_bindsap == ETHERTYPE_VLAN)) {
3614 		struct ether_vlan_header *evhp;
3615 		uint16_t sap;
3616 		mblk_t *tmp = NULL;
3617 		size_t size;
3618 
3619 		size = sizeof (struct ether_vlan_header);
3620 		if (MBLKL(mp) < size) {
3621 			/*
3622 			 * Pullup the message in order to get the MAC header
3623 			 * infomation. Note that this is a read-only function,
3624 			 * we keep the input packet intact.
3625 			 */
3626 			if ((tmp = msgpullup(mp, size)) == NULL)
3627 				return (EINVAL);
3628 
3629 			mp = tmp;
3630 		}
3631 		evhp = (struct ether_vlan_header *)mp->b_rptr;
3632 		sap = ntohs(evhp->ether_type);
3633 		(void) mac_sap_verify(mh, sap, &mhip->mhi_bindsap);
3634 		mhip->mhi_hdrsize = sizeof (struct ether_vlan_header);
3635 		mhip->mhi_tci = ntohs(evhp->ether_tci);
3636 		mhip->mhi_istagged = B_TRUE;
3637 		freemsg(tmp);
3638 
3639 		if (VLAN_CFI(mhip->mhi_tci) != ETHER_CFI)
3640 			return (EINVAL);
3641 	} else {
3642 		mhip->mhi_istagged = B_FALSE;
3643 		mhip->mhi_tci = 0;
3644 	}
3645 
3646 	return (0);
3647 }
3648 
3649 mblk_t *
3650 mac_header_cook(mac_handle_t mh, mblk_t *mp)
3651 {
3652 	mac_impl_t *mip = (mac_impl_t *)mh;
3653 
3654 	if (mip->mi_type->mt_ops.mtops_ops & MTOPS_HEADER_COOK) {
3655 		if (DB_REF(mp) > 1) {
3656 			mblk_t *newmp = copymsg(mp);
3657 			if (newmp == NULL)
3658 				return (NULL);
3659 			freemsg(mp);
3660 			mp = newmp;
3661 		}
3662 		return (mip->mi_type->mt_ops.mtops_header_cook(mp,
3663 		    mip->mi_pdata));
3664 	}
3665 	return (mp);
3666 }
3667 
3668 mblk_t *
3669 mac_header_uncook(mac_handle_t mh, mblk_t *mp)
3670 {
3671 	mac_impl_t *mip = (mac_impl_t *)mh;
3672 
3673 	if (mip->mi_type->mt_ops.mtops_ops & MTOPS_HEADER_UNCOOK) {
3674 		if (DB_REF(mp) > 1) {
3675 			mblk_t *newmp = copymsg(mp);
3676 			if (newmp == NULL)
3677 				return (NULL);
3678 			freemsg(mp);
3679 			mp = newmp;
3680 		}
3681 		return (mip->mi_type->mt_ops.mtops_header_uncook(mp,
3682 		    mip->mi_pdata));
3683 	}
3684 	return (mp);
3685 }
3686 
3687 uint_t
3688 mac_addr_len(mac_handle_t mh)
3689 {
3690 	mac_impl_t *mip = (mac_impl_t *)mh;
3691 
3692 	return (mip->mi_type->mt_addr_length);
3693 }
3694 
3695 /* True if a MAC is a VNIC */
3696 boolean_t
3697 mac_is_vnic(mac_handle_t mh)
3698 {
3699 	return (((mac_impl_t *)mh)->mi_state_flags & MIS_IS_VNIC);
3700 }
3701 
3702 mac_handle_t
3703 mac_get_lower_mac_handle(mac_handle_t mh)
3704 {
3705 	mac_impl_t *mip = (mac_impl_t *)mh;
3706 
3707 	ASSERT(mac_is_vnic(mh));
3708 	return (((vnic_t *)mip->mi_driver)->vn_lower_mh);
3709 }
3710 
3711 void
3712 mac_update_resources(mac_resource_props_t *nmrp, mac_resource_props_t *cmrp,
3713     boolean_t is_user_flow)
3714 {
3715 	if (nmrp != NULL && cmrp != NULL) {
3716 		if (nmrp->mrp_mask & MRP_PRIORITY) {
3717 			if (nmrp->mrp_priority == MPL_RESET) {
3718 				cmrp->mrp_mask &= ~MRP_PRIORITY;
3719 				if (is_user_flow) {
3720 					cmrp->mrp_priority =
3721 					    MPL_SUBFLOW_DEFAULT;
3722 				} else {
3723 					cmrp->mrp_priority = MPL_LINK_DEFAULT;
3724 				}
3725 			} else {
3726 				cmrp->mrp_mask |= MRP_PRIORITY;
3727 				cmrp->mrp_priority = nmrp->mrp_priority;
3728 			}
3729 		}
3730 		if (nmrp->mrp_mask & MRP_MAXBW) {
3731 			cmrp->mrp_maxbw = nmrp->mrp_maxbw;
3732 			if (nmrp->mrp_maxbw == MRP_MAXBW_RESETVAL)
3733 				cmrp->mrp_mask &= ~MRP_MAXBW;
3734 			else
3735 				cmrp->mrp_mask |= MRP_MAXBW;
3736 		}
3737 		if (nmrp->mrp_mask & MRP_CPUS)
3738 			MAC_COPY_CPUS(nmrp, cmrp);
3739 
3740 		if (nmrp->mrp_mask & MRP_PROTECT)
3741 			mac_protect_update(nmrp, cmrp);
3742 	}
3743 }
3744 
3745 /*
3746  * i_mac_set_resources:
3747  *
3748  * This routine associates properties with the primary MAC client of
3749  * the specified MAC instance.
3750  * - Cache the properties in mac_impl_t
3751  * - Apply the properties to the primary MAC client if exists
3752  */
3753 int
3754 i_mac_set_resources(mac_handle_t mh, mac_resource_props_t *mrp)
3755 {
3756 	mac_impl_t		*mip = (mac_impl_t *)mh;
3757 	mac_client_impl_t	*mcip;
3758 	int			err = 0;
3759 	uint32_t		resmask, newresmask;
3760 	mac_resource_props_t	tmrp, umrp;
3761 
3762 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
3763 
3764 	err = mac_validate_props(mrp);
3765 	if (err != 0)
3766 		return (err);
3767 
3768 	bcopy(&mip->mi_resource_props, &umrp, sizeof (mac_resource_props_t));
3769 	resmask = umrp.mrp_mask;
3770 	mac_update_resources(mrp, &umrp, B_FALSE);
3771 	newresmask = umrp.mrp_mask;
3772 
3773 	if (resmask == 0 && newresmask != 0) {
3774 		/*
3775 		 * Bandwidth, priority or cpu link properties configured,
3776 		 * must disable fastpath.
3777 		 */
3778 		if ((err = mac_fastpath_disable((mac_handle_t)mip)) != 0)
3779 			return (err);
3780 	}
3781 
3782 	/*
3783 	 * Since bind_cpu may be modified by mac_client_set_resources()
3784 	 * we use a copy of bind_cpu and finally cache bind_cpu in mip.
3785 	 * This allows us to cache only user edits in mip.
3786 	 */
3787 	bcopy(mrp, &tmrp, sizeof (mac_resource_props_t));
3788 	mcip = mac_primary_client_handle(mip);
3789 	if (mcip != NULL && (mcip->mci_state_flags & MCIS_IS_AGGR_PORT) == 0) {
3790 		err =
3791 		    mac_client_set_resources((mac_client_handle_t)mcip, &tmrp);
3792 	}
3793 
3794 	/* Only update the values if mac_client_set_resources succeeded */
3795 	if (err == 0) {
3796 		bcopy(&umrp, &mip->mi_resource_props,
3797 		    sizeof (mac_resource_props_t));
3798 		/*
3799 		 * If bankwidth, priority or cpu link properties cleared,
3800 		 * renable fastpath.
3801 		 */
3802 		if (resmask != 0 && newresmask == 0)
3803 			mac_fastpath_enable((mac_handle_t)mip);
3804 	} else if (resmask == 0 && newresmask != 0) {
3805 		mac_fastpath_enable((mac_handle_t)mip);
3806 	}
3807 	return (err);
3808 }
3809 
3810 int
3811 mac_set_resources(mac_handle_t mh, mac_resource_props_t *mrp)
3812 {
3813 	int err;
3814 
3815 	i_mac_perim_enter((mac_impl_t *)mh);
3816 	err = i_mac_set_resources(mh, mrp);
3817 	i_mac_perim_exit((mac_impl_t *)mh);
3818 	return (err);
3819 }
3820 
3821 /*
3822  * Get the properties cached for the specified MAC instance.
3823  */
3824 void
3825 mac_get_resources(mac_handle_t mh, mac_resource_props_t *mrp)
3826 {
3827 	mac_impl_t 		*mip = (mac_impl_t *)mh;
3828 	mac_client_impl_t	*mcip;
3829 
3830 	if (mip->mi_state_flags & MIS_IS_VNIC) {
3831 		mcip = mac_primary_client_handle(mip);
3832 		if (mcip != NULL) {
3833 			mac_client_get_resources((mac_client_handle_t)mcip,
3834 			    mrp);
3835 			return;
3836 		}
3837 	}
3838 	bcopy(&mip->mi_resource_props, mrp, sizeof (mac_resource_props_t));
3839 }
3840 
3841 int
3842 mac_set_pvid(mac_handle_t mh, uint16_t pvid)
3843 {
3844 	mac_impl_t *mip = (mac_impl_t *)mh;
3845 	mac_client_impl_t *mcip;
3846 	mac_unicast_impl_t *muip;
3847 
3848 	i_mac_perim_enter(mip);
3849 	if (pvid != 0) {
3850 		for (mcip = mip->mi_clients_list; mcip != NULL;
3851 		    mcip = mcip->mci_client_next) {
3852 			for (muip = mcip->mci_unicast_list; muip != NULL;
3853 			    muip = muip->mui_next) {
3854 				if (muip->mui_vid == pvid) {
3855 					i_mac_perim_exit(mip);
3856 					return (EBUSY);
3857 				}
3858 			}
3859 		}
3860 	}
3861 	mip->mi_pvid = pvid;
3862 	i_mac_perim_exit(mip);
3863 	return (0);
3864 }
3865 
3866 uint16_t
3867 mac_get_pvid(mac_handle_t mh)
3868 {
3869 	mac_impl_t *mip = (mac_impl_t *)mh;
3870 
3871 	return (mip->mi_pvid);
3872 }
3873 
3874 uint32_t
3875 mac_get_llimit(mac_handle_t mh)
3876 {
3877 	mac_impl_t *mip = (mac_impl_t *)mh;
3878 
3879 	return (mip->mi_llimit);
3880 }
3881 
3882 uint32_t
3883 mac_get_ldecay(mac_handle_t mh)
3884 {
3885 	mac_impl_t *mip = (mac_impl_t *)mh;
3886 
3887 	return (mip->mi_ldecay);
3888 }
3889 
3890 /*
3891  * Rename a mac client, its flow, and the kstat.
3892  */
3893 int
3894 mac_rename_primary(mac_handle_t mh, const char *new_name)
3895 {
3896 	mac_impl_t		*mip = (mac_impl_t *)mh;
3897 	mac_client_impl_t	*cur_clnt = NULL;
3898 	flow_entry_t		*fep;
3899 
3900 	i_mac_perim_enter(mip);
3901 
3902 	/*
3903 	 * VNICs: we need to change the sys flow name and
3904 	 * the associated flow kstat.
3905 	 */
3906 	if (mip->mi_state_flags & MIS_IS_VNIC) {
3907 		ASSERT(new_name != NULL);
3908 		mac_rename_flow_names(mac_vnic_lower(mip), new_name);
3909 		goto done;
3910 	}
3911 	/*
3912 	 * This mac may itself be an aggr link, or it may have some client
3913 	 * which is an aggr port. For both cases, we need to change the
3914 	 * aggr port's mac client name, its flow name and the associated flow
3915 	 * kstat.
3916 	 */
3917 	if (mip->mi_state_flags & MIS_IS_AGGR) {
3918 		mac_capab_aggr_t aggr_cap;
3919 		mac_rename_fn_t rename_fn;
3920 		boolean_t ret;
3921 
3922 		ASSERT(new_name != NULL);
3923 		ret = i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_AGGR,
3924 		    (void *)(&aggr_cap));
3925 		ASSERT(ret == B_TRUE);
3926 		rename_fn = aggr_cap.mca_rename_fn;
3927 		rename_fn(new_name, mip->mi_driver);
3928 		/*
3929 		 * The aggr's client name and kstat flow name will be
3930 		 * updated below, i.e. via mac_rename_flow_names.
3931 		 */
3932 	}
3933 
3934 	for (cur_clnt = mip->mi_clients_list; cur_clnt != NULL;
3935 	    cur_clnt = cur_clnt->mci_client_next) {
3936 		if (cur_clnt->mci_state_flags & MCIS_IS_AGGR_PORT) {
3937 			if (new_name != NULL) {
3938 				char *str_st = cur_clnt->mci_name;
3939 				char *str_del = strchr(str_st, '-');
3940 
3941 				ASSERT(str_del != NULL);
3942 				bzero(str_del + 1, MAXNAMELEN -
3943 				    (str_del - str_st + 1));
3944 				bcopy(new_name, str_del + 1,
3945 				    strlen(new_name));
3946 			}
3947 			fep = cur_clnt->mci_flent;
3948 			mac_rename_flow(fep, cur_clnt->mci_name);
3949 			break;
3950 		} else if (new_name != NULL &&
3951 		    cur_clnt->mci_state_flags & MCIS_USE_DATALINK_NAME) {
3952 			mac_rename_flow_names(cur_clnt, new_name);
3953 			break;
3954 		}
3955 	}
3956 
3957 done:
3958 	i_mac_perim_exit(mip);
3959 	return (0);
3960 }
3961 
3962 /*
3963  * Rename the MAC client's flow names
3964  */
3965 static void
3966 mac_rename_flow_names(mac_client_impl_t *mcip, const char *new_name)
3967 {
3968 	flow_entry_t	*flent;
3969 	uint16_t	vid;
3970 	char		flowname[MAXFLOWNAMELEN];
3971 	mac_impl_t	*mip = mcip->mci_mip;
3972 
3973 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
3974 
3975 	/*
3976 	 * Use mi_rw_lock to ensure that threads not in the mac perimeter
3977 	 * see a self-consistent value for mci_name
3978 	 */
3979 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
3980 	(void) strlcpy(mcip->mci_name, new_name, sizeof (mcip->mci_name));
3981 	rw_exit(&mip->mi_rw_lock);
3982 
3983 	mac_rename_flow(mcip->mci_flent, new_name);
3984 
3985 	if (mcip->mci_nflents == 1)
3986 		return;
3987 
3988 	/*
3989 	 * We have to rename all the others too, no stats to destroy for
3990 	 * these.
3991 	 */
3992 	for (flent = mcip->mci_flent_list; flent != NULL;
3993 	    flent = flent->fe_client_next) {
3994 		if (flent != mcip->mci_flent) {
3995 			vid = i_mac_flow_vid(flent);
3996 			(void) sprintf(flowname, "%s%u", new_name, vid);
3997 			mac_flow_set_name(flent, flowname);
3998 		}
3999 	}
4000 }
4001 
4002 
4003 /*
4004  * Add a flow to the MAC client's flow list - i.e list of MAC/VID tuples
4005  * defined for the specified MAC client.
4006  */
4007 static void
4008 mac_client_add_to_flow_list(mac_client_impl_t *mcip, flow_entry_t *flent)
4009 {
4010 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
4011 	/*
4012 	 * The promisc Rx data path walks the mci_flent_list. Protect by
4013 	 * using mi_rw_lock
4014 	 */
4015 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
4016 
4017 	/* Add it to the head */
4018 	flent->fe_client_next = mcip->mci_flent_list;
4019 	mcip->mci_flent_list = flent;
4020 	mcip->mci_nflents++;
4021 
4022 	/*
4023 	 * Keep track of the number of non-zero VIDs addresses per MAC
4024 	 * client to avoid figuring it out in the data-path.
4025 	 */
4026 	if (i_mac_flow_vid(flent) != VLAN_ID_NONE)
4027 		mcip->mci_nvids++;
4028 
4029 	rw_exit(&mcip->mci_rw_lock);
4030 }
4031 
4032 /*
4033  * Remove a flow entry from the MAC client's list.
4034  */
4035 static void
4036 mac_client_remove_flow_from_list(mac_client_impl_t *mcip, flow_entry_t *flent)
4037 {
4038 	flow_entry_t	*fe = mcip->mci_flent_list;
4039 	flow_entry_t	*prev_fe = NULL;
4040 
4041 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
4042 	/*
4043 	 * The promisc Rx data path walks the mci_flent_list. Protect by
4044 	 * using mci_rw_lock
4045 	 */
4046 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
4047 	while ((fe != NULL) && (fe != flent)) {
4048 		prev_fe = fe;
4049 		fe = fe->fe_client_next;
4050 	}
4051 
4052 	ASSERT(fe != NULL);
4053 	if (prev_fe == NULL) {
4054 		/* Deleting the first node */
4055 		mcip->mci_flent_list = fe->fe_client_next;
4056 	} else {
4057 		prev_fe->fe_client_next = fe->fe_client_next;
4058 	}
4059 	mcip->mci_nflents--;
4060 
4061 	if (i_mac_flow_vid(flent) != VLAN_ID_NONE)
4062 		mcip->mci_nvids--;
4063 
4064 	rw_exit(&mcip->mci_rw_lock);
4065 }
4066 
4067 /*
4068  * Check if the given VID belongs to this MAC client.
4069  */
4070 boolean_t
4071 mac_client_check_flow_vid(mac_client_impl_t *mcip, uint16_t vid)
4072 {
4073 	flow_entry_t	*flent;
4074 	uint16_t	mci_vid;
4075 
4076 	/* The mci_flent_list is protected by mci_rw_lock */
4077 	rw_enter(&mcip->mci_rw_lock, RW_WRITER);
4078 	for (flent = mcip->mci_flent_list; flent != NULL;
4079 	    flent = flent->fe_client_next) {
4080 		mci_vid = i_mac_flow_vid(flent);
4081 		if (vid == mci_vid) {
4082 			rw_exit(&mcip->mci_rw_lock);
4083 			return (B_TRUE);
4084 		}
4085 	}
4086 	rw_exit(&mcip->mci_rw_lock);
4087 	return (B_FALSE);
4088 }
4089 
4090 /*
4091  * Get the flow entry for the specified <MAC addr, VID> tuple.
4092  */
4093 static flow_entry_t *
4094 mac_client_get_flow(mac_client_impl_t *mcip, mac_unicast_impl_t *muip)
4095 {
4096 	mac_address_t *map = mcip->mci_unicast;
4097 	flow_entry_t *flent;
4098 	uint16_t vid;
4099 	flow_desc_t flow_desc;
4100 
4101 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
4102 
4103 	mac_flow_get_desc(mcip->mci_flent, &flow_desc);
4104 	if (bcmp(flow_desc.fd_dst_mac, map->ma_addr, map->ma_len) != 0)
4105 		return (NULL);
4106 
4107 	for (flent = mcip->mci_flent_list; flent != NULL;
4108 	    flent = flent->fe_client_next) {
4109 		vid = i_mac_flow_vid(flent);
4110 		if (vid == muip->mui_vid) {
4111 			return (flent);
4112 		}
4113 	}
4114 
4115 	return (NULL);
4116 }
4117 
4118 /*
4119  * Since mci_flent has the SRSs, when we want to remove it, we replace
4120  * the flow_desc_t in mci_flent with that of an existing flent and then
4121  * remove that flent instead of mci_flent.
4122  */
4123 static flow_entry_t *
4124 mac_client_swap_mciflent(mac_client_impl_t *mcip)
4125 {
4126 	flow_entry_t	*flent = mcip->mci_flent;
4127 	flow_tab_t	*ft = flent->fe_flow_tab;
4128 	flow_entry_t	*flent1;
4129 	flow_desc_t	fl_desc;
4130 	char		fl_name[MAXFLOWNAMELEN];
4131 	int		err;
4132 
4133 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
4134 	ASSERT(mcip->mci_nflents > 1);
4135 
4136 	/* get the next flent following the primary flent  */
4137 	flent1 = mcip->mci_flent_list->fe_client_next;
4138 	ASSERT(flent1 != NULL && flent1->fe_flow_tab == ft);
4139 
4140 	/*
4141 	 * Remove the flent from the flow table before updating the
4142 	 * flow descriptor as the hash depends on the flow descriptor.
4143 	 * This also helps incoming packet classification avoid having
4144 	 * to grab fe_lock. Access to fe_flow_desc of a flent not in the
4145 	 * flow table is done under the fe_lock so that log or stat functions
4146 	 * see a self-consistent fe_flow_desc. The name and desc are specific
4147 	 * to a flow, the rest are shared by all the clients, including
4148 	 * resource control etc.
4149 	 */
4150 	mac_flow_remove(ft, flent, B_TRUE);
4151 	mac_flow_remove(ft, flent1, B_TRUE);
4152 
4153 	bcopy(&flent->fe_flow_desc, &fl_desc, sizeof (flow_desc_t));
4154 	bcopy(flent->fe_flow_name, fl_name, MAXFLOWNAMELEN);
4155 
4156 	/* update the primary flow entry */
4157 	mutex_enter(&flent->fe_lock);
4158 	bcopy(&flent1->fe_flow_desc, &flent->fe_flow_desc,
4159 	    sizeof (flow_desc_t));
4160 	bcopy(&flent1->fe_flow_name, &flent->fe_flow_name, MAXFLOWNAMELEN);
4161 	mutex_exit(&flent->fe_lock);
4162 
4163 	/* update the flow entry that is to be freed */
4164 	mutex_enter(&flent1->fe_lock);
4165 	bcopy(&fl_desc, &flent1->fe_flow_desc, sizeof (flow_desc_t));
4166 	bcopy(fl_name, &flent1->fe_flow_name, MAXFLOWNAMELEN);
4167 	mutex_exit(&flent1->fe_lock);
4168 
4169 	/* now reinsert the flow entries in the table */
4170 	err = mac_flow_add(ft, flent);
4171 	ASSERT(err == 0);
4172 
4173 	err = mac_flow_add(ft, flent1);
4174 	ASSERT(err == 0);
4175 
4176 	return (flent1);
4177 }
4178 
4179 /*
4180  * Return whether there is only one flow entry associated with this
4181  * MAC client.
4182  */
4183 static boolean_t
4184 mac_client_single_rcvr(mac_client_impl_t *mcip)
4185 {
4186 	return (mcip->mci_nflents == 1);
4187 }
4188 
4189 int
4190 mac_validate_props(mac_resource_props_t *mrp)
4191 {
4192 	if (mrp == NULL)
4193 		return (0);
4194 
4195 	if (mrp->mrp_mask & MRP_PRIORITY) {
4196 		mac_priority_level_t	pri = mrp->mrp_priority;
4197 
4198 		if (pri < MPL_LOW || pri > MPL_RESET)
4199 			return (EINVAL);
4200 	}
4201 
4202 	if (mrp->mrp_mask & MRP_MAXBW) {
4203 		uint64_t maxbw = mrp->mrp_maxbw;
4204 
4205 		if (maxbw < MRP_MAXBW_MINVAL && maxbw != 0)
4206 			return (EINVAL);
4207 	}
4208 	if (mrp->mrp_mask & MRP_CPUS) {
4209 		int i, j;
4210 		mac_cpu_mode_t	fanout;
4211 
4212 		if (mrp->mrp_ncpus > ncpus || mrp->mrp_ncpus > MAX_SR_FANOUT)
4213 			return (EINVAL);
4214 
4215 		for (i = 0; i < mrp->mrp_ncpus; i++) {
4216 			for (j = 0; j < mrp->mrp_ncpus; j++) {
4217 				if (i != j &&
4218 				    mrp->mrp_cpu[i] == mrp->mrp_cpu[j]) {
4219 					return (EINVAL);
4220 				}
4221 			}
4222 		}
4223 
4224 		for (i = 0; i < mrp->mrp_ncpus; i++) {
4225 			cpu_t *cp;
4226 			int rv;
4227 
4228 			mutex_enter(&cpu_lock);
4229 			cp = cpu_get(mrp->mrp_cpu[i]);
4230 			if (cp != NULL)
4231 				rv = cpu_is_online(cp);
4232 			else
4233 				rv = 0;
4234 			mutex_exit(&cpu_lock);
4235 			if (rv == 0)
4236 				return (EINVAL);
4237 		}
4238 
4239 		fanout = mrp->mrp_fanout_mode;
4240 		if (fanout < 0 || fanout > MCM_CPUS)
4241 			return (EINVAL);
4242 	}
4243 
4244 	if (mrp->mrp_mask & MRP_PROTECT) {
4245 		int err = mac_protect_validate(mrp);
4246 		if (err != 0)
4247 			return (err);
4248 	}
4249 	return (0);
4250 }
4251 
4252 /*
4253  * Send a MAC_NOTE_LINK notification to all the MAC clients whenever the
4254  * underlying physical link is down. This is to allow MAC clients to
4255  * communicate with other clients.
4256  */
4257 void
4258 mac_virtual_link_update(mac_impl_t *mip)
4259 {
4260 	if (mip->mi_linkstate != LINK_STATE_UP)
4261 		i_mac_notify(mip, MAC_NOTE_LINK);
4262 }
4263 
4264 /*
4265  * For clients that have a pass-thru MAC, e.g. VNIC, we set the VNIC's
4266  * mac handle in the client.
4267  */
4268 void
4269 mac_set_upper_mac(mac_client_handle_t mch, mac_handle_t mh)
4270 {
4271 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
4272 
4273 	mcip->mci_upper_mip = (mac_impl_t *)mh;
4274 }
4275 
4276 /*
4277  * Mark the mac as being used exclusively by the single mac client that is
4278  * doing some control operation on this mac. No further opens of this mac
4279  * will be allowed until this client calls mac_unmark_exclusive. The mac
4280  * client calling this function must already be in the mac perimeter
4281  */
4282 int
4283 mac_mark_exclusive(mac_handle_t mh)
4284 {
4285 	mac_impl_t	*mip = (mac_impl_t *)mh;
4286 
4287 	ASSERT(MAC_PERIM_HELD(mh));
4288 	/*
4289 	 * Look up its entry in the global hash table.
4290 	 */
4291 	rw_enter(&i_mac_impl_lock, RW_WRITER);
4292 	if (mip->mi_state_flags & MIS_DISABLED) {
4293 		rw_exit(&i_mac_impl_lock);
4294 		return (ENOENT);
4295 	}
4296 
4297 	/*
4298 	 * A reference to mac is held even if the link is not plumbed.
4299 	 * In i_dls_link_create() we open the MAC interface and hold the
4300 	 * reference. There is an additional reference for the mac_open
4301 	 * done in acquiring the mac perimeter
4302 	 */
4303 	if (mip->mi_ref != 2) {
4304 		rw_exit(&i_mac_impl_lock);
4305 		return (EBUSY);
4306 	}
4307 
4308 	ASSERT(!(mip->mi_state_flags & MIS_EXCLUSIVE_HELD));
4309 	mip->mi_state_flags |= MIS_EXCLUSIVE_HELD;
4310 	rw_exit(&i_mac_impl_lock);
4311 	return (0);
4312 }
4313 
4314 void
4315 mac_unmark_exclusive(mac_handle_t mh)
4316 {
4317 	mac_impl_t	*mip = (mac_impl_t *)mh;
4318 
4319 	ASSERT(MAC_PERIM_HELD(mh));
4320 
4321 	rw_enter(&i_mac_impl_lock, RW_WRITER);
4322 	/* 1 for the creation and another for the perimeter */
4323 	ASSERT(mip->mi_ref == 2 && (mip->mi_state_flags & MIS_EXCLUSIVE_HELD));
4324 	mip->mi_state_flags &= ~MIS_EXCLUSIVE_HELD;
4325 	rw_exit(&i_mac_impl_lock);
4326 }
4327 
4328 /*
4329  * Set the MTU for the specified MAC.  Note that this mechanism depends on
4330  * the driver calling mac_maxsdu_update() to update the link MTU if it was
4331  * successful in setting its MTU.
4332  *
4333  * Note that there is potential for improvement here.  A better model might be
4334  * to not require drivers to call mac_maxsdu_update(), but rather have this
4335  * function update mi_sdu_max and send notifications if the driver setprop
4336  * callback succeeds.  This would remove the burden and complexity from
4337  * drivers.
4338  */
4339 int
4340 mac_set_mtu(mac_handle_t mh, uint_t new_mtu, uint_t *old_mtu_arg)
4341 {
4342 	mac_impl_t *mip = (mac_impl_t *)mh;
4343 	uint_t old_mtu;
4344 	int rv = 0;
4345 
4346 	i_mac_perim_enter(mip);
4347 
4348 	if (!(mip->mi_callbacks->mc_callbacks & (MC_SETPROP|MC_GETPROP))) {
4349 		rv = ENOTSUP;
4350 		goto bail;
4351 	}
4352 
4353 	old_mtu = mip->mi_sdu_max;
4354 
4355 	if (old_mtu != new_mtu) {
4356 		rv = mip->mi_callbacks->mc_setprop(mip->mi_driver,
4357 		    "mtu", MAC_PROP_MTU, sizeof (uint_t), &new_mtu);
4358 	}
4359 
4360 bail:
4361 	i_mac_perim_exit(mip);
4362 
4363 	if (rv == 0 && old_mtu_arg != NULL)
4364 		*old_mtu_arg = old_mtu;
4365 	return (rv);
4366 }
4367 
4368 void
4369 mac_get_hwgrp_info(mac_handle_t mh, int grp_index, uint_t *grp_num,
4370     uint_t *n_rings, uint_t *type, uint_t *n_clnts, char *clnts_name)
4371 {
4372 	mac_impl_t *mip = (mac_impl_t *)mh;
4373 	mac_grp_client_t *mcip;
4374 	uint_t i = 0, index = 0;
4375 
4376 	/* Revisit when we implement fully dynamic group allocation */
4377 	ASSERT(grp_index >= 0 && grp_index < mip->mi_rx_group_count);
4378 
4379 	rw_enter(&mip->mi_rw_lock, RW_READER);
4380 	*grp_num = mip->mi_rx_groups[grp_index].mrg_index;
4381 	*type = mip->mi_rx_groups[grp_index].mrg_type;
4382 	*n_rings = mip->mi_rx_groups[grp_index].mrg_cur_count;
4383 	for (mcip = mip->mi_rx_groups[grp_index].mrg_clients; mcip != NULL;
4384 	    mcip = mcip->mgc_next) {
4385 		int name_len = strlen(mcip->mgc_client->mci_name);
4386 
4387 		/*
4388 		 * MAXCLIENTNAMELEN is the buffer size reserved for client
4389 		 * names.
4390 		 * XXXX Formating the client name string needs to be moved
4391 		 * to user land when fixing the size of dhi_clnts in
4392 		 * dld_hwgrpinfo_t. We should use n_clients * client_name for
4393 		 * dhi_clntsin instead of MAXCLIENTNAMELEN
4394 		 */
4395 		if (index + name_len >= MAXCLIENTNAMELEN) {
4396 			index = MAXCLIENTNAMELEN;
4397 			break;
4398 		}
4399 		bcopy(mcip->mgc_client->mci_name, &(clnts_name[index]),
4400 		    name_len);
4401 		index += name_len;
4402 		clnts_name[index++] = ',';
4403 		i++;
4404 	}
4405 
4406 	/* Get rid of the last , */
4407 	if (index > 0)
4408 		clnts_name[index - 1] = '\0';
4409 	*n_clnts = i;
4410 	rw_exit(&mip->mi_rw_lock);
4411 }
4412 
4413 uint_t
4414 mac_hwgrp_num(mac_handle_t mh)
4415 {
4416 	mac_impl_t *mip = (mac_impl_t *)mh;
4417 
4418 	return (mip->mi_rx_group_count);
4419 }
4420