xref: /titanic_50/usr/src/uts/common/io/dld/dld_str.c (revision 08556f6c986d17ed3d96f39954c1fa8e6e302a81)
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  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*
27  * Data-Link Driver
28  */
29 
30 #include	<inet/common.h>
31 #include	<sys/strsubr.h>
32 #include	<sys/stropts.h>
33 #include	<sys/strsun.h>
34 #include	<sys/vlan.h>
35 #include	<sys/dld_impl.h>
36 #include	<sys/cpuvar.h>
37 #include	<sys/callb.h>
38 #include	<sys/list.h>
39 #include	<sys/mac_client.h>
40 #include	<sys/mac_client_priv.h>
41 
42 static int	str_constructor(void *, void *, int);
43 static void	str_destructor(void *, void *);
44 static mblk_t	*str_unitdata_ind(dld_str_t *, mblk_t *, boolean_t);
45 static void	str_notify_promisc_on_phys(dld_str_t *);
46 static void	str_notify_promisc_off_phys(dld_str_t *);
47 static void	str_notify_phys_addr(dld_str_t *, const uint8_t *);
48 static void	str_notify_link_up(dld_str_t *);
49 static void	str_notify_link_down(dld_str_t *);
50 static void	str_notify_capab_reneg(dld_str_t *);
51 static void	str_notify_speed(dld_str_t *, uint32_t);
52 
53 static void	ioc_native(dld_str_t *,  mblk_t *);
54 static void	ioc_margin(dld_str_t *, mblk_t *);
55 static void	ioc_raw(dld_str_t *, mblk_t *);
56 static void	ioc_fast(dld_str_t *,  mblk_t *);
57 static void	ioc(dld_str_t *, mblk_t *);
58 static void	dld_ioc(dld_str_t *, mblk_t *);
59 static void	dld_wput_nondata(dld_str_t *, mblk_t *);
60 
61 static void	str_mdata_raw_put(dld_str_t *, mblk_t *);
62 static mblk_t	*i_dld_ether_header_update_tag(mblk_t *, uint_t, uint16_t,
63     link_tagmode_t);
64 static mblk_t	*i_dld_ether_header_strip_tag(mblk_t *);
65 
66 static uint32_t		str_count;
67 static kmem_cache_t	*str_cachep;
68 static mod_hash_t	*str_hashp;
69 
70 #define	STR_HASHSZ		64
71 #define	STR_HASH_KEY(key)	((mod_hash_key_t)(uintptr_t)(key))
72 
73 #define	dld_taskq	system_taskq
74 
75 static kmutex_t		dld_taskq_lock;
76 static kcondvar_t	dld_taskq_cv;
77 static list_t		dld_taskq_list;		/* List of dld_str_t */
78 boolean_t		dld_taskq_quit;
79 boolean_t		dld_taskq_done;
80 
81 static void		dld_taskq_dispatch(void);
82 
83 /*
84  * Some notes on entry points, flow-control, queueing.
85  *
86  * This driver exports the traditional STREAMS put entry point as well as
87  * the non-STREAMS fast-path transmit routine which is provided to IP via
88  * the DL_CAPAB_POLL negotiation.  The put procedure handles all control
89  * and data operations, while the fast-path routine deals only with M_DATA
90  * fast-path packets.  Regardless of the entry point, all outbound packets
91  * will end up in DLD_TX(), where they will be delivered to the MAC layer.
92  *
93  * The transmit logic operates in the following way: All packets coming
94  * into DLD will be sent to the MAC layer through DLD_TX(). Flow-control
95  * happens when the MAC layer indicates the packets couldn't be
96  * transmitted due to 1) lack of resources (e.g. running out of
97  * descriptors),  or 2) reaching the allowed bandwidth limit for this
98  * particular flow. The indication comes in the form of a Tx cookie that
99  * identifies the blocked ring. In such case, DLD will place a
100  * dummy message on its write-side STREAMS queue so that the queue is
101  * marked as "full". Any subsequent packets arriving at the driver will
102  * still be sent to the MAC layer where it either gets queued in the Tx
103  * SRS or discarded it if queue limit is exceeded. The write-side STREAMS
104  * queue gets enabled when MAC layer notifies DLD through MAC_NOTE_TX.
105  * When the write service procedure runs, it will remove the dummy
106  * message from the write-side STREAMS queue; in effect this will trigger
107  * backenabling. The sizes of q_hiwat and q_lowat are set to 1 and 0,
108  * respectively, due to the above reasons.
109  *
110  * All non-data operations, both DLPI and ioctls are single threaded on a per
111  * dld_str_t endpoint. This is done using a taskq so that the control operation
112  * has kernel context and can cv_wait for resources. In addition all set type
113  * operations that involve mac level state modification are serialized on a
114  * per mac end point using the perimeter mechanism provided by the mac layer.
115  * This serializes all mac clients trying to modify a single mac end point over
116  * the entire sequence of mac calls made by that client as an atomic unit. The
117  * mac framework locking is described in mac.c. A critical element is that
118  * DLD/DLS does not hold any locks across the mac perimeter.
119  *
120  * dld_finddevinfo() returns the dev_info_t * corresponding to a particular
121  * dev_t. It searches str_hashp (a table of dld_str_t's) for streams that
122  * match dev_t. If a stream is found and it is attached, its dev_info_t *
123  * is returned. If the mac handle is non-null, it can be safely accessed
124  * below. The mac handle won't be freed until the mac_unregister which
125  * won't happen until the driver detaches. The DDI framework ensures that
126  * the detach won't happen while a getinfo is in progress.
127  */
128 typedef struct i_dld_str_state_s {
129 	major_t		ds_major;
130 	minor_t		ds_minor;
131 	dev_info_t	*ds_dip;
132 } i_dld_str_state_t;
133 
134 /* ARGSUSED */
135 static uint_t
136 i_dld_str_walker(mod_hash_key_t key, mod_hash_val_t *val, void *arg)
137 {
138 	i_dld_str_state_t	*statep = arg;
139 	dld_str_t		*dsp = (dld_str_t *)val;
140 	mac_handle_t		mh;
141 
142 	if (statep->ds_major != dsp->ds_major)
143 		return (MH_WALK_CONTINUE);
144 
145 	ASSERT(statep->ds_minor != 0);
146 	mh = dsp->ds_mh;
147 
148 	if (statep->ds_minor == dsp->ds_minor) {
149 		/*
150 		 * Clone: a clone minor is unique. we can terminate the
151 		 * walk if we find a matching stream -- even if we fail
152 		 * to obtain the devinfo.
153 		 */
154 		if (mh != NULL)
155 			statep->ds_dip = mac_devinfo_get(mh);
156 		return (MH_WALK_TERMINATE);
157 	}
158 	return (MH_WALK_CONTINUE);
159 }
160 
161 static dev_info_t *
162 dld_finddevinfo(dev_t dev)
163 {
164 	dev_info_t		*dip;
165 	i_dld_str_state_t	state;
166 
167 	if (getminor(dev) == 0)
168 		return (NULL);
169 
170 	/*
171 	 * See if it's a minor node of a link
172 	 */
173 	if ((dip = dls_link_devinfo(dev)) != NULL)
174 		return (dip);
175 
176 	state.ds_minor = getminor(dev);
177 	state.ds_major = getmajor(dev);
178 	state.ds_dip = NULL;
179 
180 	mod_hash_walk(str_hashp, i_dld_str_walker, &state);
181 	return (state.ds_dip);
182 }
183 
184 /*
185  * devo_getinfo: getinfo(9e)
186  */
187 /*ARGSUSED*/
188 int
189 dld_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg, void **resp)
190 {
191 	dev_info_t	*devinfo;
192 	minor_t		minor = getminor((dev_t)arg);
193 	int		rc = DDI_FAILURE;
194 
195 	switch (cmd) {
196 	case DDI_INFO_DEVT2DEVINFO:
197 		if ((devinfo = dld_finddevinfo((dev_t)arg)) != NULL) {
198 			*(dev_info_t **)resp = devinfo;
199 			rc = DDI_SUCCESS;
200 		}
201 		break;
202 	case DDI_INFO_DEVT2INSTANCE:
203 		if (minor > 0 && minor <= DLS_MAX_MINOR) {
204 			*resp = (void *)(uintptr_t)DLS_MINOR2INST(minor);
205 			rc = DDI_SUCCESS;
206 		} else if (minor > DLS_MAX_MINOR &&
207 		    (devinfo = dld_finddevinfo((dev_t)arg)) != NULL) {
208 			*resp = (void *)(uintptr_t)ddi_get_instance(devinfo);
209 			rc = DDI_SUCCESS;
210 		}
211 		break;
212 	}
213 	return (rc);
214 }
215 
216 /*
217  * qi_qopen: open(9e)
218  */
219 /*ARGSUSED*/
220 int
221 dld_open(queue_t *rq, dev_t *devp, int flag, int sflag, cred_t *credp)
222 {
223 	dld_str_t	*dsp;
224 	major_t		major;
225 	minor_t		minor;
226 	int		err;
227 
228 	if (sflag == MODOPEN)
229 		return (ENOTSUP);
230 
231 	/*
232 	 * This is a cloning driver and therefore each queue should only
233 	 * ever get opened once.
234 	 */
235 	if (rq->q_ptr != NULL)
236 		return (EBUSY);
237 
238 	major = getmajor(*devp);
239 	minor = getminor(*devp);
240 
241 	/*
242 	 * Create a new dld_str_t for the stream. This will grab a new minor
243 	 * number that will be handed back in the cloned dev_t.  Creation may
244 	 * fail if we can't allocate the dummy mblk used for flow-control.
245 	 */
246 	dsp = dld_str_create(rq, DLD_DLPI, major,
247 	    ((minor == 0) ? DL_STYLE2 : DL_STYLE1));
248 	if (dsp == NULL)
249 		return (ENOSR);
250 
251 	ASSERT(dsp->ds_dlstate == DL_UNATTACHED);
252 	if (minor != 0) {
253 		/*
254 		 * Style 1 open
255 		 */
256 		if ((err = dld_str_attach(dsp, (t_uscalar_t)minor - 1)) != 0)
257 			goto failed;
258 		ASSERT(dsp->ds_dlstate == DL_UNBOUND);
259 	} else {
260 		(void) qassociate(rq, -1);
261 	}
262 
263 	/*
264 	 * Enable the queue srv(9e) routine.
265 	 */
266 	qprocson(rq);
267 
268 	/*
269 	 * Construct a cloned dev_t to hand back.
270 	 */
271 	*devp = makedevice(getmajor(*devp), dsp->ds_minor);
272 	return (0);
273 
274 failed:
275 	dld_str_destroy(dsp);
276 	return (err);
277 }
278 
279 /*
280  * qi_qclose: close(9e)
281  */
282 int
283 dld_close(queue_t *rq)
284 {
285 	dld_str_t	*dsp = rq->q_ptr;
286 
287 	/*
288 	 * All modules on top have been popped off. So there can't be any
289 	 * threads from the top.
290 	 */
291 	ASSERT(dsp->ds_datathr_cnt == 0);
292 
293 	/*
294 	 * Wait until pending DLPI requests are processed.
295 	 */
296 	mutex_enter(&dsp->ds_lock);
297 	while (dsp->ds_dlpi_pending)
298 		cv_wait(&dsp->ds_dlpi_pending_cv, &dsp->ds_lock);
299 	mutex_exit(&dsp->ds_lock);
300 
301 	/*
302 	 * Disable the queue srv(9e) routine.
303 	 */
304 	qprocsoff(rq);
305 
306 
307 	/*
308 	 * This stream was open to a provider node. Check to see
309 	 * if it has been cleanly shut down.
310 	 */
311 	if (dsp->ds_dlstate != DL_UNATTACHED) {
312 		/*
313 		 * The stream is either open to a style 1 provider or
314 		 * this is not clean shutdown. Detach from the PPA.
315 		 * (This is still ok even in the style 1 case).
316 		 */
317 		dld_str_detach(dsp);
318 	}
319 
320 	dld_str_destroy(dsp);
321 	return (0);
322 }
323 
324 /*
325  * qi_qputp: put(9e)
326  */
327 void
328 dld_wput(queue_t *wq, mblk_t *mp)
329 {
330 	dld_str_t *dsp = (dld_str_t *)wq->q_ptr;
331 	dld_str_mode_t	mode;
332 
333 	switch (DB_TYPE(mp)) {
334 	case M_DATA:
335 		mutex_enter(&dsp->ds_lock);
336 		if (dsp->ds_dlstate == DL_IDLE) {
337 			mode = dsp->ds_mode;
338 			if (mode == DLD_FASTPATH || mode == DLD_RAW) {
339 				DLD_DATATHR_INC(dsp);
340 				mutex_exit(&dsp->ds_lock);
341 				if (mode == DLD_FASTPATH) {
342 					(void) str_mdata_fastpath_put(dsp, mp,
343 					    0, 0);
344 				} else {
345 					str_mdata_raw_put(dsp, mp);
346 				}
347 				DLD_DATATHR_DCR(dsp);
348 				break;
349 			}
350 		}
351 		mutex_exit(&dsp->ds_lock);
352 		freemsg(mp);
353 		break;
354 
355 	case M_PROTO:
356 	case M_PCPROTO: {
357 		t_uscalar_t	prim;
358 
359 		if (MBLKL(mp) < sizeof (t_uscalar_t))
360 			break;
361 
362 		prim = ((union DL_primitives *)mp->b_rptr)->dl_primitive;
363 
364 		if (prim == DL_UNITDATA_REQ) {
365 			proto_unitdata_req(dsp, mp);
366 		} else {
367 			dld_wput_nondata(dsp, mp);
368 		}
369 		break;
370 	}
371 
372 	case M_IOCTL:
373 		dld_wput_nondata(dsp, mp);
374 		break;
375 
376 	case M_FLUSH:
377 		if (*mp->b_rptr & FLUSHW) {
378 			DLD_CLRQFULL(dsp);
379 			*mp->b_rptr &= ~FLUSHW;
380 		}
381 
382 		if (*mp->b_rptr & FLUSHR) {
383 			qreply(wq, mp);
384 		} else {
385 			freemsg(mp);
386 		}
387 		break;
388 
389 	default:
390 		freemsg(mp);
391 		break;
392 	}
393 }
394 
395 /*
396  * qi_srvp: srv(9e)
397  */
398 void
399 dld_wsrv(queue_t *wq)
400 {
401 	dld_str_t	*dsp = wq->q_ptr;
402 
403 	DLD_CLRQFULL(dsp);
404 }
405 
406 void
407 dld_init_ops(struct dev_ops *ops, const char *name)
408 {
409 	struct streamtab *stream;
410 	struct qinit *rq, *wq;
411 	struct module_info *modinfo;
412 
413 	modinfo = kmem_zalloc(sizeof (struct module_info), KM_SLEEP);
414 	modinfo->mi_idname = kmem_zalloc(FMNAMESZ, KM_SLEEP);
415 	(void) snprintf(modinfo->mi_idname, FMNAMESZ, "%s", name);
416 	modinfo->mi_minpsz = 0;
417 	modinfo->mi_maxpsz = 64*1024;
418 	modinfo->mi_hiwat  = 1;
419 	modinfo->mi_lowat = 0;
420 
421 	rq = kmem_zalloc(sizeof (struct qinit), KM_SLEEP);
422 	rq->qi_qopen = dld_open;
423 	rq->qi_qclose = dld_close;
424 	rq->qi_minfo = modinfo;
425 
426 	wq = kmem_zalloc(sizeof (struct qinit), KM_SLEEP);
427 	wq->qi_putp = (pfi_t)dld_wput;
428 	wq->qi_srvp = (pfi_t)dld_wsrv;
429 	wq->qi_minfo = modinfo;
430 
431 	stream = kmem_zalloc(sizeof (struct streamtab), KM_SLEEP);
432 	stream->st_rdinit = rq;
433 	stream->st_wrinit = wq;
434 	ops->devo_cb_ops->cb_str = stream;
435 
436 	if (ops->devo_getinfo == NULL)
437 		ops->devo_getinfo = &dld_getinfo;
438 }
439 
440 void
441 dld_fini_ops(struct dev_ops *ops)
442 {
443 	struct streamtab *stream;
444 	struct qinit *rq, *wq;
445 	struct module_info *modinfo;
446 
447 	stream = ops->devo_cb_ops->cb_str;
448 	rq = stream->st_rdinit;
449 	wq = stream->st_wrinit;
450 	modinfo = rq->qi_minfo;
451 	ASSERT(wq->qi_minfo == modinfo);
452 
453 	kmem_free(stream, sizeof (struct streamtab));
454 	kmem_free(wq, sizeof (struct qinit));
455 	kmem_free(rq, sizeof (struct qinit));
456 	kmem_free(modinfo->mi_idname, FMNAMESZ);
457 	kmem_free(modinfo, sizeof (struct module_info));
458 }
459 
460 /*
461  * Initialize this module's data structures.
462  */
463 void
464 dld_str_init(void)
465 {
466 	/*
467 	 * Create dld_str_t object cache.
468 	 */
469 	str_cachep = kmem_cache_create("dld_str_cache", sizeof (dld_str_t),
470 	    0, str_constructor, str_destructor, NULL, NULL, NULL, 0);
471 	ASSERT(str_cachep != NULL);
472 
473 	/*
474 	 * Create a hash table for maintaining dld_str_t's.
475 	 * The ds_minor field (the clone minor number) of a dld_str_t
476 	 * is used as a key for this hash table because this number is
477 	 * globally unique (allocated from "dls_minor_arena").
478 	 */
479 	str_hashp = mod_hash_create_idhash("dld_str_hash", STR_HASHSZ,
480 	    mod_hash_null_valdtor);
481 
482 	mutex_init(&dld_taskq_lock, NULL, MUTEX_DRIVER, NULL);
483 	cv_init(&dld_taskq_cv, NULL, CV_DRIVER, NULL);
484 
485 	dld_taskq_quit = B_FALSE;
486 	dld_taskq_done = B_FALSE;
487 	list_create(&dld_taskq_list, sizeof (dld_str_t),
488 	    offsetof(dld_str_t, ds_tqlist));
489 	(void) thread_create(NULL, 0, dld_taskq_dispatch, NULL, 0,
490 	    &p0, TS_RUN, minclsyspri);
491 }
492 
493 /*
494  * Tear down this module's data structures.
495  */
496 int
497 dld_str_fini(void)
498 {
499 	/*
500 	 * Make sure that there are no objects in use.
501 	 */
502 	if (str_count != 0)
503 		return (EBUSY);
504 
505 	/*
506 	 * Ask the dld_taskq thread to quit and wait for it to be done
507 	 */
508 	mutex_enter(&dld_taskq_lock);
509 	dld_taskq_quit = B_TRUE;
510 	cv_signal(&dld_taskq_cv);
511 	while (!dld_taskq_done)
512 		cv_wait(&dld_taskq_cv, &dld_taskq_lock);
513 	mutex_exit(&dld_taskq_lock);
514 	list_destroy(&dld_taskq_list);
515 	/*
516 	 * Destroy object cache.
517 	 */
518 	kmem_cache_destroy(str_cachep);
519 	mod_hash_destroy_idhash(str_hashp);
520 	return (0);
521 }
522 
523 /*
524  * Create a new dld_str_t object.
525  */
526 dld_str_t *
527 dld_str_create(queue_t *rq, uint_t type, major_t major, t_uscalar_t style)
528 {
529 	dld_str_t	*dsp;
530 	int		err;
531 
532 	/*
533 	 * Allocate an object from the cache.
534 	 */
535 	atomic_add_32(&str_count, 1);
536 	dsp = kmem_cache_alloc(str_cachep, KM_SLEEP);
537 
538 	/*
539 	 * Allocate the dummy mblk for flow-control.
540 	 */
541 	dsp->ds_tx_flow_mp = allocb(1, BPRI_HI);
542 	if (dsp->ds_tx_flow_mp == NULL) {
543 		kmem_cache_free(str_cachep, dsp);
544 		atomic_add_32(&str_count, -1);
545 		return (NULL);
546 	}
547 	dsp->ds_type = type;
548 	dsp->ds_major = major;
549 	dsp->ds_style = style;
550 
551 	/*
552 	 * Initialize the queue pointers.
553 	 */
554 	ASSERT(RD(rq) == rq);
555 	dsp->ds_rq = rq;
556 	dsp->ds_wq = WR(rq);
557 	rq->q_ptr = WR(rq)->q_ptr = (void *)dsp;
558 
559 	/*
560 	 * We want explicit control over our write-side STREAMS queue
561 	 * where the dummy mblk gets added/removed for flow-control.
562 	 */
563 	noenable(WR(rq));
564 
565 	err = mod_hash_insert(str_hashp, STR_HASH_KEY(dsp->ds_minor),
566 	    (mod_hash_val_t)dsp);
567 	ASSERT(err == 0);
568 	return (dsp);
569 }
570 
571 /*
572  * Destroy a dld_str_t object.
573  */
574 void
575 dld_str_destroy(dld_str_t *dsp)
576 {
577 	queue_t		*rq;
578 	queue_t		*wq;
579 	mod_hash_val_t	val;
580 
581 	/*
582 	 * Clear the queue pointers.
583 	 */
584 	rq = dsp->ds_rq;
585 	wq = dsp->ds_wq;
586 	ASSERT(wq == WR(rq));
587 	rq->q_ptr = wq->q_ptr = NULL;
588 	dsp->ds_rq = dsp->ds_wq = NULL;
589 
590 	ASSERT(dsp->ds_dlstate == DL_UNATTACHED);
591 	ASSERT(dsp->ds_sap == 0);
592 	ASSERT(dsp->ds_mh == NULL);
593 	ASSERT(dsp->ds_mch == NULL);
594 	ASSERT(dsp->ds_promisc == 0);
595 	ASSERT(dsp->ds_mph == NULL);
596 	ASSERT(dsp->ds_mip == NULL);
597 	ASSERT(dsp->ds_mnh == NULL);
598 
599 	ASSERT(dsp->ds_polling == B_FALSE);
600 	ASSERT(dsp->ds_direct == B_FALSE);
601 	ASSERT(dsp->ds_lso == B_FALSE);
602 	ASSERT(dsp->ds_lso_max == 0);
603 
604 	/*
605 	 * Reinitialize all the flags.
606 	 */
607 	dsp->ds_notifications = 0;
608 	dsp->ds_passivestate = DLD_UNINITIALIZED;
609 	dsp->ds_mode = DLD_UNITDATA;
610 	dsp->ds_native = B_FALSE;
611 
612 	ASSERT(dsp->ds_datathr_cnt == 0);
613 	ASSERT(dsp->ds_pending_head == NULL);
614 	ASSERT(dsp->ds_pending_tail == NULL);
615 	ASSERT(!dsp->ds_dlpi_pending);
616 
617 	ASSERT(dsp->ds_dlp == NULL);
618 	ASSERT(dsp->ds_dmap == NULL);
619 	ASSERT(dsp->ds_rx == NULL);
620 	ASSERT(dsp->ds_rx_arg == NULL);
621 	ASSERT(dsp->ds_next == NULL);
622 	ASSERT(dsp->ds_head == NULL);
623 
624 	/*
625 	 * Free the dummy mblk if exists.
626 	 */
627 	if (dsp->ds_tx_flow_mp != NULL) {
628 		freeb(dsp->ds_tx_flow_mp);
629 		dsp->ds_tx_flow_mp = NULL;
630 	}
631 
632 	(void) mod_hash_remove(str_hashp, STR_HASH_KEY(dsp->ds_minor), &val);
633 	ASSERT(dsp == (dld_str_t *)val);
634 
635 	/*
636 	 * Free the object back to the cache.
637 	 */
638 	kmem_cache_free(str_cachep, dsp);
639 	atomic_add_32(&str_count, -1);
640 }
641 
642 /*
643  * kmem_cache contructor function: see kmem_cache_create(9f).
644  */
645 /*ARGSUSED*/
646 static int
647 str_constructor(void *buf, void *cdrarg, int kmflags)
648 {
649 	dld_str_t	*dsp = buf;
650 
651 	bzero(buf, sizeof (dld_str_t));
652 
653 	/*
654 	 * Allocate a new minor number.
655 	 */
656 	if ((dsp->ds_minor = mac_minor_hold(kmflags == KM_SLEEP)) == 0)
657 		return (-1);
658 
659 	/*
660 	 * Initialize the DLPI state machine.
661 	 */
662 	dsp->ds_dlstate = DL_UNATTACHED;
663 
664 	mutex_init(&dsp->ds_lock, NULL, MUTEX_DRIVER, NULL);
665 	cv_init(&dsp->ds_datathr_cv, NULL, CV_DRIVER, NULL);
666 	cv_init(&dsp->ds_dlpi_pending_cv, NULL, CV_DRIVER, NULL);
667 
668 	return (0);
669 }
670 
671 /*
672  * kmem_cache destructor function.
673  */
674 /*ARGSUSED*/
675 static void
676 str_destructor(void *buf, void *cdrarg)
677 {
678 	dld_str_t	*dsp = buf;
679 
680 	/*
681 	 * Release the minor number.
682 	 */
683 	mac_minor_rele(dsp->ds_minor);
684 
685 	ASSERT(dsp->ds_tx_flow_mp == NULL);
686 
687 	mutex_destroy(&dsp->ds_lock);
688 	cv_destroy(&dsp->ds_datathr_cv);
689 	cv_destroy(&dsp->ds_dlpi_pending_cv);
690 }
691 
692 /*
693  * Update the priority bits and VID (may need to insert tag if mp points
694  * to an untagged packet.
695  * If vid is VLAN_ID_NONE, use the VID encoded in the packet.
696  */
697 static mblk_t *
698 i_dld_ether_header_update_tag(mblk_t *mp, uint_t pri, uint16_t vid,
699     link_tagmode_t tagmode)
700 {
701 	mblk_t *hmp;
702 	struct ether_vlan_header *evhp;
703 	struct ether_header *ehp;
704 	uint16_t old_tci = 0;
705 	size_t len;
706 
707 	ASSERT(pri != 0 || vid != VLAN_ID_NONE);
708 
709 	evhp = (struct ether_vlan_header *)mp->b_rptr;
710 	if (ntohs(evhp->ether_tpid) == ETHERTYPE_VLAN) {
711 		/*
712 		 * Tagged packet, update the priority bits.
713 		 */
714 		old_tci = ntohs(evhp->ether_tci);
715 		len = sizeof (struct ether_vlan_header);
716 
717 		if ((DB_REF(mp) > 1) || (MBLKL(mp) < len)) {
718 			/*
719 			 * In case some drivers only check the db_ref
720 			 * count of the first mblk, we pullup the
721 			 * message into a single mblk.
722 			 */
723 			hmp = msgpullup(mp, -1);
724 			if ((hmp == NULL) || (MBLKL(hmp) < len)) {
725 				freemsg(hmp);
726 				return (NULL);
727 			} else {
728 				freemsg(mp);
729 				mp = hmp;
730 			}
731 		}
732 
733 		evhp = (struct ether_vlan_header *)mp->b_rptr;
734 	} else {
735 		/*
736 		 * Untagged packet.  Two factors will cause us to insert a
737 		 * VLAN header:
738 		 * - This is a VLAN link (vid is specified)
739 		 * - The link supports user priority tagging and the priority
740 		 *   is non-zero.
741 		 */
742 		if (vid == VLAN_ID_NONE && tagmode == LINK_TAGMODE_VLANONLY)
743 			return (mp);
744 
745 		hmp = allocb(sizeof (struct ether_vlan_header), BPRI_MED);
746 		if (hmp == NULL)
747 			return (NULL);
748 
749 		evhp = (struct ether_vlan_header *)hmp->b_rptr;
750 		ehp = (struct ether_header *)mp->b_rptr;
751 
752 		/*
753 		 * Copy the MAC addresses and typelen
754 		 */
755 		bcopy(ehp, evhp, (ETHERADDRL * 2));
756 		evhp->ether_type = ehp->ether_type;
757 		evhp->ether_tpid = htons(ETHERTYPE_VLAN);
758 
759 		hmp->b_wptr += sizeof (struct ether_vlan_header);
760 		mp->b_rptr += sizeof (struct ether_header);
761 
762 		/*
763 		 * Free the original message if it's now empty. Link the
764 		 * rest of the messages to the header message.
765 		 */
766 		if (MBLKL(mp) == 0) {
767 			hmp->b_cont = mp->b_cont;
768 			freeb(mp);
769 		} else {
770 			hmp->b_cont = mp;
771 		}
772 		mp = hmp;
773 	}
774 
775 	if (pri == 0)
776 		pri = VLAN_PRI(old_tci);
777 	if (vid == VLAN_ID_NONE)
778 		vid = VLAN_ID(old_tci);
779 	evhp->ether_tci = htons(VLAN_TCI(pri, VLAN_CFI(old_tci), vid));
780 	return (mp);
781 }
782 
783 /*
784  * M_DATA put (IP fast-path mode)
785  */
786 mac_tx_cookie_t
787 str_mdata_fastpath_put(dld_str_t *dsp, mblk_t *mp, uintptr_t f_hint,
788     uint16_t flag)
789 {
790 	boolean_t is_ethernet = (dsp->ds_mip->mi_media == DL_ETHER);
791 	mblk_t *newmp;
792 	uint_t pri;
793 	mac_tx_cookie_t cookie;
794 
795 	if (is_ethernet) {
796 		/*
797 		 * Update the priority bits to the assigned priority.
798 		 */
799 		pri = (VLAN_MBLKPRI(mp) == 0) ? dsp->ds_pri : VLAN_MBLKPRI(mp);
800 
801 		if (pri != 0) {
802 			newmp = i_dld_ether_header_update_tag(mp, pri,
803 			    VLAN_ID_NONE, dsp->ds_dlp->dl_tagmode);
804 			if (newmp == NULL)
805 				goto discard;
806 			mp = newmp;
807 		}
808 	}
809 
810 	if ((cookie = DLD_TX(dsp, mp, f_hint, flag)) != NULL) {
811 		DLD_SETQFULL(dsp);
812 	}
813 	return (cookie);
814 
815 discard:
816 	/* TODO: bump kstat? */
817 	freemsg(mp);
818 	return (NULL);
819 }
820 
821 /*
822  * M_DATA put (DLIOCRAW mode)
823  */
824 static void
825 str_mdata_raw_put(dld_str_t *dsp, mblk_t *mp)
826 {
827 	boolean_t is_ethernet = (dsp->ds_mip->mi_media == DL_ETHER);
828 	mblk_t *bp, *newmp;
829 	size_t size;
830 	mac_header_info_t mhi;
831 	uint_t pri, vid, dvid;
832 	uint_t max_sdu;
833 
834 	/*
835 	 * Certain MAC type plugins provide an illusion for raw DLPI
836 	 * consumers.  They pretend that the MAC layer is something that
837 	 * it's not for the benefit of observability tools.  For example,
838 	 * mac_wifi pretends that it's Ethernet for such consumers.
839 	 * Here, unless native mode is enabled, we call into the MAC layer so
840 	 * that this illusion can be maintained.  The plugin will optionally
841 	 * transform the MAC header here into something that can be passed
842 	 * down.  The header goes from raw mode to "cooked" mode.
843 	 */
844 	if (!dsp->ds_native) {
845 		if ((newmp = mac_header_cook(dsp->ds_mh, mp)) == NULL)
846 			goto discard;
847 		mp = newmp;
848 	}
849 
850 	size = MBLKL(mp);
851 
852 	/*
853 	 * Check the packet is not too big and that any remaining
854 	 * fragment list is composed entirely of M_DATA messages. (We
855 	 * know the first fragment was M_DATA otherwise we could not
856 	 * have got here).
857 	 */
858 	for (bp = mp->b_cont; bp != NULL; bp = bp->b_cont) {
859 		if (DB_TYPE(bp) != M_DATA)
860 			goto discard;
861 		size += MBLKL(bp);
862 	}
863 
864 	if (dls_link_header_info(dsp->ds_dlp, mp, &mhi) != 0)
865 		goto discard;
866 
867 	mac_sdu_get(dsp->ds_mh, NULL, &max_sdu);
868 	/*
869 	 * If LSO is enabled, check the size against lso_max. Otherwise,
870 	 * compare the packet size with max_sdu.
871 	 */
872 	max_sdu = dsp->ds_lso ? dsp->ds_lso_max : max_sdu;
873 	if (size > max_sdu + mhi.mhi_hdrsize)
874 		goto discard;
875 
876 	if (is_ethernet) {
877 		dvid = mac_client_vid(dsp->ds_mch);
878 
879 		/*
880 		 * Discard the packet if this is a VLAN stream but the VID in
881 		 * the packet is not correct.
882 		 */
883 		vid = VLAN_ID(mhi.mhi_tci);
884 		if ((dvid != VLAN_ID_NONE) && (vid != VLAN_ID_NONE))
885 			goto discard;
886 
887 		/*
888 		 * Discard the packet if this packet is a tagged packet
889 		 * but both pri and VID are 0.
890 		 */
891 		pri = VLAN_PRI(mhi.mhi_tci);
892 		if (mhi.mhi_istagged && (pri == 0) && (vid == VLAN_ID_NONE))
893 			goto discard;
894 
895 		/*
896 		 * Update the priority bits to the per-stream priority if
897 		 * priority is not set in the packet. Update the VID for
898 		 * packets on a VLAN stream.
899 		 */
900 		pri = (pri == 0) ? dsp->ds_pri : 0;
901 		if ((pri != 0) || (dvid != VLAN_ID_NONE)) {
902 			if ((newmp = i_dld_ether_header_update_tag(mp, pri,
903 			    dvid, dsp->ds_dlp->dl_tagmode)) == NULL) {
904 				goto discard;
905 			}
906 			mp = newmp;
907 		}
908 	}
909 
910 	if (DLD_TX(dsp, mp, 0, 0) != NULL) {
911 		/* Turn on flow-control for dld */
912 		DLD_SETQFULL(dsp);
913 	}
914 	return;
915 
916 discard:
917 	/* TODO: bump kstat? */
918 	freemsg(mp);
919 }
920 
921 /*
922  * Process DL_ATTACH_REQ (style 2) or open(2) (style 1).
923  */
924 int
925 dld_str_attach(dld_str_t *dsp, t_uscalar_t ppa)
926 {
927 	dev_t			dev;
928 	int			err;
929 	const char		*drvname;
930 	mac_perim_handle_t	mph;
931 	boolean_t		qassociated = B_FALSE;
932 	dls_link_t		*dlp = NULL;
933 	dls_dl_handle_t		ddp = NULL;
934 	boolean_t		entered_perim = B_FALSE;
935 
936 	if ((drvname = ddi_major_to_name(dsp->ds_major)) == NULL)
937 		return (EINVAL);
938 
939 	if (dsp->ds_style == DL_STYLE2 && ppa > DLS_MAX_PPA)
940 		return (ENOTSUP);
941 
942 	/*
943 	 * /dev node access. This will still be supported for backward
944 	 * compatibility reason.
945 	 */
946 	if ((dsp->ds_style == DL_STYLE2) && (strcmp(drvname, "aggr") != 0) &&
947 	    (strcmp(drvname, "vnic") != 0)) {
948 		if (qassociate(dsp->ds_wq, DLS_PPA2INST(ppa)) != 0)
949 			return (EINVAL);
950 		qassociated = B_TRUE;
951 	}
952 
953 	dev = makedevice(dsp->ds_major, (minor_t)ppa + 1);
954 	if ((err = dls_devnet_hold_by_dev(dev, &ddp)) != 0)
955 		goto failed;
956 
957 	if ((err = mac_perim_enter_by_macname(dls_devnet_mac(ddp), &mph)) != 0)
958 		goto failed;
959 	entered_perim = B_TRUE;
960 
961 	/*
962 	 * Open a channel.
963 	 */
964 	if ((err = dls_link_hold(dls_devnet_mac(ddp), &dlp)) != 0)
965 		goto failed;
966 
967 	if ((err = dls_open(dlp, ddp, dsp)) != 0)
968 		goto failed;
969 
970 	/*
971 	 * Set the default packet priority.
972 	 */
973 	dsp->ds_pri = 0;
974 
975 	/*
976 	 * Add a notify function so that the we get updates from the MAC.
977 	 */
978 	dsp->ds_mnh = mac_notify_add(dsp->ds_mh, str_notify, dsp);
979 	dsp->ds_dlstate = DL_UNBOUND;
980 	mac_perim_exit(mph);
981 	return (0);
982 
983 failed:
984 	if (dlp != NULL)
985 		dls_link_rele(dlp);
986 	if (entered_perim)
987 		mac_perim_exit(mph);
988 	if (ddp != NULL)
989 		dls_devnet_rele(ddp);
990 	if (qassociated)
991 		(void) qassociate(dsp->ds_wq, -1);
992 
993 	return (err);
994 }
995 
996 /*
997  * Process DL_DETACH_REQ (style 2) or close(2) (style 1). Can also be called
998  * from close(2) for style 2.
999  */
1000 void
1001 dld_str_detach(dld_str_t *dsp)
1002 {
1003 	mac_perim_handle_t	mph;
1004 	int			err;
1005 
1006 	ASSERT(dsp->ds_datathr_cnt == 0);
1007 
1008 	mac_perim_enter_by_mh(dsp->ds_mh, &mph);
1009 	/*
1010 	 * Remove the notify function.
1011 	 *
1012 	 * Note that we cannot wait for the notification callback to be removed
1013 	 * since it could cause the deadlock with str_notify() since they both
1014 	 * need the mac perimeter. Continue if we cannot remove the
1015 	 * notification callback right now and wait after we leave the
1016 	 * perimeter.
1017 	 */
1018 	err = mac_notify_remove(dsp->ds_mnh, B_FALSE);
1019 	dsp->ds_mnh = NULL;
1020 
1021 	/*
1022 	 * Disable the capabilities
1023 	 */
1024 	dld_capabilities_disable(dsp);
1025 
1026 	/*
1027 	 * Clear LSO flags.
1028 	 */
1029 	dsp->ds_lso = B_FALSE;
1030 	dsp->ds_lso_max = 0;
1031 
1032 	dls_close(dsp);
1033 	mac_perim_exit(mph);
1034 
1035 	/*
1036 	 * Now we leave the mac perimeter. If mac_notify_remove() failed
1037 	 * because the notification callback was in progress, wait for
1038 	 * it to finish before we proceed.
1039 	 */
1040 	if (err != 0)
1041 		mac_notify_remove_wait(dsp->ds_mh);
1042 
1043 	/*
1044 	 * An unreferenced tagged (non-persistent) vlan gets destroyed
1045 	 * automatically in the call to dls_devnet_rele.
1046 	 */
1047 	dls_devnet_rele(dsp->ds_ddh);
1048 
1049 	dsp->ds_sap = 0;
1050 	dsp->ds_mh = NULL;
1051 	dsp->ds_mch = NULL;
1052 	dsp->ds_mip = NULL;
1053 
1054 	if (dsp->ds_style == DL_STYLE2)
1055 		(void) qassociate(dsp->ds_wq, -1);
1056 
1057 	/*
1058 	 * Re-initialize the DLPI state machine.
1059 	 */
1060 	dsp->ds_dlstate = DL_UNATTACHED;
1061 }
1062 
1063 /*
1064  * This function is only called for VLAN streams. In raw mode, we strip VLAN
1065  * tags before sending packets up to the DLS clients, with the exception of
1066  * special priority tagged packets, in that case, we set the VID to 0.
1067  * mp must be a VLAN tagged packet.
1068  */
1069 static mblk_t *
1070 i_dld_ether_header_strip_tag(mblk_t *mp)
1071 {
1072 	mblk_t *newmp;
1073 	struct ether_vlan_header *evhp;
1074 	uint16_t tci, new_tci;
1075 
1076 	ASSERT(MBLKL(mp) >= sizeof (struct ether_vlan_header));
1077 	if (DB_REF(mp) > 1) {
1078 		newmp = copymsg(mp);
1079 		if (newmp == NULL)
1080 			return (NULL);
1081 		freemsg(mp);
1082 		mp = newmp;
1083 	}
1084 	evhp = (struct ether_vlan_header *)mp->b_rptr;
1085 
1086 	tci = ntohs(evhp->ether_tci);
1087 	if (VLAN_PRI(tci) == 0) {
1088 		/*
1089 		 * Priority is 0, strip the tag.
1090 		 */
1091 		ovbcopy(mp->b_rptr, mp->b_rptr + VLAN_TAGSZ, 2 * ETHERADDRL);
1092 		mp->b_rptr += VLAN_TAGSZ;
1093 	} else {
1094 		/*
1095 		 * Priority is not 0, update the VID to 0.
1096 		 */
1097 		new_tci = VLAN_TCI(VLAN_PRI(tci), VLAN_CFI(tci), VLAN_ID_NONE);
1098 		evhp->ether_tci = htons(new_tci);
1099 	}
1100 	return (mp);
1101 }
1102 
1103 /*
1104  * Raw mode receive function.
1105  */
1106 /*ARGSUSED*/
1107 void
1108 dld_str_rx_raw(void *arg, mac_resource_handle_t mrh, mblk_t *mp,
1109     mac_header_info_t *mhip)
1110 {
1111 	dld_str_t *dsp = (dld_str_t *)arg;
1112 	boolean_t is_ethernet = (dsp->ds_mip->mi_media == DL_ETHER);
1113 	mblk_t *next, *newmp;
1114 
1115 	ASSERT(mp != NULL);
1116 	do {
1117 		/*
1118 		 * Get the pointer to the next packet in the chain and then
1119 		 * clear b_next before the packet gets passed on.
1120 		 */
1121 		next = mp->b_next;
1122 		mp->b_next = NULL;
1123 
1124 		/*
1125 		 * Wind back b_rptr to point at the MAC header.
1126 		 */
1127 		ASSERT(mp->b_rptr >= DB_BASE(mp) + mhip->mhi_hdrsize);
1128 		mp->b_rptr -= mhip->mhi_hdrsize;
1129 
1130 		/*
1131 		 * Certain MAC type plugins provide an illusion for raw
1132 		 * DLPI consumers.  They pretend that the MAC layer is
1133 		 * something that it's not for the benefit of observability
1134 		 * tools.  For example, mac_wifi pretends that it's Ethernet
1135 		 * for such consumers.	Here, unless native mode is enabled,
1136 		 * we call into the MAC layer so that this illusion can be
1137 		 * maintained.	The plugin will optionally transform the MAC
1138 		 * header here into something that can be passed up to raw
1139 		 * consumers.  The header goes from "cooked" mode to raw mode.
1140 		 */
1141 		if (!dsp->ds_native) {
1142 			newmp = mac_header_uncook(dsp->ds_mh, mp);
1143 			if (newmp == NULL) {
1144 				freemsg(mp);
1145 				goto next;
1146 			}
1147 			mp = newmp;
1148 		}
1149 
1150 		/*
1151 		 * Strip the VLAN tag for VLAN streams.
1152 		 */
1153 		if (is_ethernet &&
1154 		    mac_client_vid(dsp->ds_mch) != VLAN_ID_NONE) {
1155 			newmp = i_dld_ether_header_strip_tag(mp);
1156 			if (newmp == NULL) {
1157 				freemsg(mp);
1158 				goto next;
1159 			}
1160 			mp = newmp;
1161 		}
1162 
1163 		/*
1164 		 * Pass the packet on.
1165 		 */
1166 		if (canputnext(dsp->ds_rq))
1167 			putnext(dsp->ds_rq, mp);
1168 		else
1169 			freemsg(mp);
1170 
1171 next:
1172 		/*
1173 		 * Move on to the next packet in the chain.
1174 		 */
1175 		mp = next;
1176 	} while (mp != NULL);
1177 }
1178 
1179 /*
1180  * Fast-path receive function.
1181  */
1182 /*ARGSUSED*/
1183 void
1184 dld_str_rx_fastpath(void *arg, mac_resource_handle_t mrh, mblk_t *mp,
1185     mac_header_info_t *mhip)
1186 {
1187 	dld_str_t *dsp = (dld_str_t *)arg;
1188 	mblk_t *next;
1189 	size_t offset = 0;
1190 
1191 	/*
1192 	 * MAC header stripping rules:
1193 	 *    - Tagged packets:
1194 	 *	a. VLAN streams. Strip the whole VLAN header including the tag.
1195 	 *	b. Physical streams
1196 	 *	- VLAN packets (non-zero VID). The stream must be either a
1197 	 *	  DL_PROMISC_SAP listener or a ETHERTYPE_VLAN listener.
1198 	 *	  Strip the Ethernet header but keep the VLAN header.
1199 	 *	- Special tagged packets (zero VID)
1200 	 *	  * The stream is either a DL_PROMISC_SAP listener or a
1201 	 *	    ETHERTYPE_VLAN listener, strip the Ethernet header but
1202 	 *	    keep the VLAN header.
1203 	 *	  * Otherwise, strip the whole VLAN header.
1204 	 *    - Untagged packets. Strip the whole MAC header.
1205 	 */
1206 	if (mhip->mhi_istagged &&
1207 	    (mac_client_vid(dsp->ds_mch) == VLAN_ID_NONE) &&
1208 	    ((dsp->ds_sap == ETHERTYPE_VLAN) ||
1209 	    (dsp->ds_promisc & DLS_PROMISC_SAP))) {
1210 		offset = VLAN_TAGSZ;
1211 	}
1212 
1213 	ASSERT(mp != NULL);
1214 	do {
1215 		/*
1216 		 * Get the pointer to the next packet in the chain and then
1217 		 * clear b_next before the packet gets passed on.
1218 		 */
1219 		next = mp->b_next;
1220 		mp->b_next = NULL;
1221 
1222 		/*
1223 		 * Wind back b_rptr to point at the VLAN header.
1224 		 */
1225 		ASSERT(mp->b_rptr >= DB_BASE(mp) + offset);
1226 		mp->b_rptr -= offset;
1227 
1228 		/*
1229 		 * Pass the packet on.
1230 		 */
1231 		if (canputnext(dsp->ds_rq))
1232 			putnext(dsp->ds_rq, mp);
1233 		else
1234 			freemsg(mp);
1235 		/*
1236 		 * Move on to the next packet in the chain.
1237 		 */
1238 		mp = next;
1239 	} while (mp != NULL);
1240 }
1241 
1242 /*
1243  * Default receive function (send DL_UNITDATA_IND messages).
1244  */
1245 /*ARGSUSED*/
1246 void
1247 dld_str_rx_unitdata(void *arg, mac_resource_handle_t mrh, mblk_t *mp,
1248     mac_header_info_t *mhip)
1249 {
1250 	dld_str_t		*dsp = (dld_str_t *)arg;
1251 	mblk_t			*ud_mp;
1252 	mblk_t			*next;
1253 	size_t			offset = 0;
1254 	boolean_t		strip_vlan = B_TRUE;
1255 
1256 	/*
1257 	 * See MAC header stripping rules in the dld_str_rx_fastpath() function.
1258 	 */
1259 	if (mhip->mhi_istagged &&
1260 	    (mac_client_vid(dsp->ds_mch) == VLAN_ID_NONE) &&
1261 	    ((dsp->ds_sap == ETHERTYPE_VLAN) ||
1262 	    (dsp->ds_promisc & DLS_PROMISC_SAP))) {
1263 		offset = VLAN_TAGSZ;
1264 		strip_vlan = B_FALSE;
1265 	}
1266 
1267 	ASSERT(mp != NULL);
1268 	do {
1269 		/*
1270 		 * Get the pointer to the next packet in the chain and then
1271 		 * clear b_next before the packet gets passed on.
1272 		 */
1273 		next = mp->b_next;
1274 		mp->b_next = NULL;
1275 
1276 		/*
1277 		 * Wind back b_rptr to point at the MAC header.
1278 		 */
1279 		ASSERT(mp->b_rptr >= DB_BASE(mp) + mhip->mhi_hdrsize);
1280 		mp->b_rptr -= mhip->mhi_hdrsize;
1281 
1282 		/*
1283 		 * Create the DL_UNITDATA_IND M_PROTO.
1284 		 */
1285 		if ((ud_mp = str_unitdata_ind(dsp, mp, strip_vlan)) == NULL) {
1286 			freemsgchain(mp);
1287 			return;
1288 		}
1289 
1290 		/*
1291 		 * Advance b_rptr to point at the payload (or the VLAN header).
1292 		 */
1293 		mp->b_rptr += (mhip->mhi_hdrsize - offset);
1294 
1295 		/*
1296 		 * Prepend the DL_UNITDATA_IND.
1297 		 */
1298 		ud_mp->b_cont = mp;
1299 
1300 		/*
1301 		 * Send the message.
1302 		 */
1303 		if (canputnext(dsp->ds_rq))
1304 			putnext(dsp->ds_rq, ud_mp);
1305 		else
1306 			freemsg(ud_mp);
1307 
1308 		/*
1309 		 * Move on to the next packet in the chain.
1310 		 */
1311 		mp = next;
1312 	} while (mp != NULL);
1313 }
1314 
1315 /*
1316  * DL_NOTIFY_IND: DL_NOTE_SDU_SIZE
1317  */
1318 static void
1319 str_notify_sdu_size(dld_str_t *dsp, uint_t max_sdu)
1320 {
1321 	mblk_t		*mp;
1322 	dl_notify_ind_t *dlip;
1323 
1324 	if (!(dsp->ds_notifications & DL_NOTE_SDU_SIZE))
1325 		return;
1326 
1327 	if ((mp = mexchange(dsp->ds_wq, NULL, sizeof (dl_notify_ind_t),
1328 	    M_PROTO, 0)) == NULL)
1329 		return;
1330 
1331 	bzero(mp->b_rptr, sizeof (dl_notify_ind_t));
1332 	dlip = (dl_notify_ind_t *)mp->b_rptr;
1333 	dlip->dl_primitive = DL_NOTIFY_IND;
1334 	dlip->dl_notification = DL_NOTE_SDU_SIZE;
1335 	dlip->dl_data = max_sdu;
1336 
1337 	qreply(dsp->ds_wq, mp);
1338 }
1339 
1340 /*
1341  * Generate DL_NOTIFY_IND messages to notify the DLPI consumer of the
1342  * current state of the interface.
1343  */
1344 void
1345 dld_str_notify_ind(dld_str_t *dsp)
1346 {
1347 	mac_notify_type_t	type;
1348 
1349 	for (type = 0; type < MAC_NNOTE; type++)
1350 		str_notify(dsp, type);
1351 }
1352 
1353 typedef struct dl_unitdata_ind_wrapper {
1354 	dl_unitdata_ind_t	dl_unitdata;
1355 	uint8_t			dl_dest_addr[MAXMACADDRLEN + sizeof (uint16_t)];
1356 	uint8_t			dl_src_addr[MAXMACADDRLEN + sizeof (uint16_t)];
1357 } dl_unitdata_ind_wrapper_t;
1358 
1359 /*
1360  * Create a DL_UNITDATA_IND M_PROTO message.
1361  */
1362 static mblk_t *
1363 str_unitdata_ind(dld_str_t *dsp, mblk_t *mp, boolean_t strip_vlan)
1364 {
1365 	mblk_t				*nmp;
1366 	dl_unitdata_ind_wrapper_t	*dlwp;
1367 	dl_unitdata_ind_t		*dlp;
1368 	mac_header_info_t		mhi;
1369 	uint_t				addr_length;
1370 	uint8_t				*daddr;
1371 	uint8_t				*saddr;
1372 
1373 	/*
1374 	 * Get the packet header information.
1375 	 */
1376 	if (dls_link_header_info(dsp->ds_dlp, mp, &mhi) != 0)
1377 		return (NULL);
1378 
1379 	/*
1380 	 * Allocate a message large enough to contain the wrapper structure
1381 	 * defined above.
1382 	 */
1383 	if ((nmp = mexchange(dsp->ds_wq, NULL,
1384 	    sizeof (dl_unitdata_ind_wrapper_t), M_PROTO,
1385 	    DL_UNITDATA_IND)) == NULL)
1386 		return (NULL);
1387 
1388 	dlwp = (dl_unitdata_ind_wrapper_t *)nmp->b_rptr;
1389 
1390 	dlp = &(dlwp->dl_unitdata);
1391 	ASSERT(dlp == (dl_unitdata_ind_t *)nmp->b_rptr);
1392 	ASSERT(dlp->dl_primitive == DL_UNITDATA_IND);
1393 
1394 	/*
1395 	 * Copy in the destination address.
1396 	 */
1397 	addr_length = dsp->ds_mip->mi_addr_length;
1398 	daddr = dlwp->dl_dest_addr;
1399 	dlp->dl_dest_addr_offset = (uintptr_t)daddr - (uintptr_t)dlp;
1400 	bcopy(mhi.mhi_daddr, daddr, addr_length);
1401 
1402 	/*
1403 	 * Set the destination DLSAP to the SAP value encoded in the packet.
1404 	 */
1405 	if (mhi.mhi_istagged && !strip_vlan)
1406 		*(uint16_t *)(daddr + addr_length) = ETHERTYPE_VLAN;
1407 	else
1408 		*(uint16_t *)(daddr + addr_length) = mhi.mhi_bindsap;
1409 	dlp->dl_dest_addr_length = addr_length + sizeof (uint16_t);
1410 
1411 	/*
1412 	 * If the destination address was multicast or broadcast then the
1413 	 * dl_group_address field should be non-zero.
1414 	 */
1415 	dlp->dl_group_address = (mhi.mhi_dsttype == MAC_ADDRTYPE_MULTICAST) ||
1416 	    (mhi.mhi_dsttype == MAC_ADDRTYPE_BROADCAST);
1417 
1418 	/*
1419 	 * Copy in the source address if one exists.  Some MAC types (DL_IB
1420 	 * for example) may not have access to source information.
1421 	 */
1422 	if (mhi.mhi_saddr == NULL) {
1423 		dlp->dl_src_addr_offset = dlp->dl_src_addr_length = 0;
1424 	} else {
1425 		saddr = dlwp->dl_src_addr;
1426 		dlp->dl_src_addr_offset = (uintptr_t)saddr - (uintptr_t)dlp;
1427 		bcopy(mhi.mhi_saddr, saddr, addr_length);
1428 
1429 		/*
1430 		 * Set the source DLSAP to the packet ethertype.
1431 		 */
1432 		*(uint16_t *)(saddr + addr_length) = mhi.mhi_origsap;
1433 		dlp->dl_src_addr_length = addr_length + sizeof (uint16_t);
1434 	}
1435 
1436 	return (nmp);
1437 }
1438 
1439 /*
1440  * DL_NOTIFY_IND: DL_NOTE_PROMISC_ON_PHYS
1441  */
1442 static void
1443 str_notify_promisc_on_phys(dld_str_t *dsp)
1444 {
1445 	mblk_t		*mp;
1446 	dl_notify_ind_t	*dlip;
1447 
1448 	if (!(dsp->ds_notifications & DL_NOTE_PROMISC_ON_PHYS))
1449 		return;
1450 
1451 	if ((mp = mexchange(dsp->ds_wq, NULL, sizeof (dl_notify_ind_t),
1452 	    M_PROTO, 0)) == NULL)
1453 		return;
1454 
1455 	bzero(mp->b_rptr, sizeof (dl_notify_ind_t));
1456 	dlip = (dl_notify_ind_t *)mp->b_rptr;
1457 	dlip->dl_primitive = DL_NOTIFY_IND;
1458 	dlip->dl_notification = DL_NOTE_PROMISC_ON_PHYS;
1459 
1460 	qreply(dsp->ds_wq, mp);
1461 }
1462 
1463 /*
1464  * DL_NOTIFY_IND: DL_NOTE_PROMISC_OFF_PHYS
1465  */
1466 static void
1467 str_notify_promisc_off_phys(dld_str_t *dsp)
1468 {
1469 	mblk_t		*mp;
1470 	dl_notify_ind_t	*dlip;
1471 
1472 	if (!(dsp->ds_notifications & DL_NOTE_PROMISC_OFF_PHYS))
1473 		return;
1474 
1475 	if ((mp = mexchange(dsp->ds_wq, NULL, sizeof (dl_notify_ind_t),
1476 	    M_PROTO, 0)) == NULL)
1477 		return;
1478 
1479 	bzero(mp->b_rptr, sizeof (dl_notify_ind_t));
1480 	dlip = (dl_notify_ind_t *)mp->b_rptr;
1481 	dlip->dl_primitive = DL_NOTIFY_IND;
1482 	dlip->dl_notification = DL_NOTE_PROMISC_OFF_PHYS;
1483 
1484 	qreply(dsp->ds_wq, mp);
1485 }
1486 
1487 /*
1488  * DL_NOTIFY_IND: DL_NOTE_PHYS_ADDR
1489  */
1490 static void
1491 str_notify_phys_addr(dld_str_t *dsp, const uint8_t *addr)
1492 {
1493 	mblk_t		*mp;
1494 	dl_notify_ind_t	*dlip;
1495 	uint_t		addr_length;
1496 	uint16_t	ethertype;
1497 
1498 	if (!(dsp->ds_notifications & DL_NOTE_PHYS_ADDR))
1499 		return;
1500 
1501 	addr_length = dsp->ds_mip->mi_addr_length;
1502 	if ((mp = mexchange(dsp->ds_wq, NULL,
1503 	    sizeof (dl_notify_ind_t) + addr_length + sizeof (uint16_t),
1504 	    M_PROTO, 0)) == NULL)
1505 		return;
1506 
1507 	bzero(mp->b_rptr, sizeof (dl_notify_ind_t));
1508 	dlip = (dl_notify_ind_t *)mp->b_rptr;
1509 	dlip->dl_primitive = DL_NOTIFY_IND;
1510 	dlip->dl_notification = DL_NOTE_PHYS_ADDR;
1511 	dlip->dl_data = DL_CURR_PHYS_ADDR;
1512 	dlip->dl_addr_offset = sizeof (dl_notify_ind_t);
1513 	dlip->dl_addr_length = addr_length + sizeof (uint16_t);
1514 
1515 	bcopy(addr, &dlip[1], addr_length);
1516 
1517 	ethertype = (dsp->ds_sap < ETHERTYPE_802_MIN) ? 0 : dsp->ds_sap;
1518 	*(uint16_t *)((uchar_t *)(dlip + 1) + addr_length) = ethertype;
1519 
1520 	qreply(dsp->ds_wq, mp);
1521 }
1522 
1523 /*
1524  * DL_NOTIFY_IND: DL_NOTE_LINK_UP
1525  */
1526 static void
1527 str_notify_link_up(dld_str_t *dsp)
1528 {
1529 	mblk_t		*mp;
1530 	dl_notify_ind_t	*dlip;
1531 
1532 	if (!(dsp->ds_notifications & DL_NOTE_LINK_UP))
1533 		return;
1534 
1535 	if ((mp = mexchange(dsp->ds_wq, NULL, sizeof (dl_notify_ind_t),
1536 	    M_PROTO, 0)) == NULL)
1537 		return;
1538 
1539 	bzero(mp->b_rptr, sizeof (dl_notify_ind_t));
1540 	dlip = (dl_notify_ind_t *)mp->b_rptr;
1541 	dlip->dl_primitive = DL_NOTIFY_IND;
1542 	dlip->dl_notification = DL_NOTE_LINK_UP;
1543 
1544 	qreply(dsp->ds_wq, mp);
1545 }
1546 
1547 /*
1548  * DL_NOTIFY_IND: DL_NOTE_LINK_DOWN
1549  */
1550 static void
1551 str_notify_link_down(dld_str_t *dsp)
1552 {
1553 	mblk_t		*mp;
1554 	dl_notify_ind_t	*dlip;
1555 
1556 	if (!(dsp->ds_notifications & DL_NOTE_LINK_DOWN))
1557 		return;
1558 
1559 	if ((mp = mexchange(dsp->ds_wq, NULL, sizeof (dl_notify_ind_t),
1560 	    M_PROTO, 0)) == NULL)
1561 		return;
1562 
1563 	bzero(mp->b_rptr, sizeof (dl_notify_ind_t));
1564 	dlip = (dl_notify_ind_t *)mp->b_rptr;
1565 	dlip->dl_primitive = DL_NOTIFY_IND;
1566 	dlip->dl_notification = DL_NOTE_LINK_DOWN;
1567 
1568 	qreply(dsp->ds_wq, mp);
1569 }
1570 
1571 /*
1572  * DL_NOTIFY_IND: DL_NOTE_SPEED
1573  */
1574 static void
1575 str_notify_speed(dld_str_t *dsp, uint32_t speed)
1576 {
1577 	mblk_t		*mp;
1578 	dl_notify_ind_t	*dlip;
1579 
1580 	if (!(dsp->ds_notifications & DL_NOTE_SPEED))
1581 		return;
1582 
1583 	if ((mp = mexchange(dsp->ds_wq, NULL, sizeof (dl_notify_ind_t),
1584 	    M_PROTO, 0)) == NULL)
1585 		return;
1586 
1587 	bzero(mp->b_rptr, sizeof (dl_notify_ind_t));
1588 	dlip = (dl_notify_ind_t *)mp->b_rptr;
1589 	dlip->dl_primitive = DL_NOTIFY_IND;
1590 	dlip->dl_notification = DL_NOTE_SPEED;
1591 	dlip->dl_data = speed;
1592 
1593 	qreply(dsp->ds_wq, mp);
1594 }
1595 
1596 /*
1597  * DL_NOTIFY_IND: DL_NOTE_CAPAB_RENEG
1598  */
1599 static void
1600 str_notify_capab_reneg(dld_str_t *dsp)
1601 {
1602 	mblk_t		*mp;
1603 	dl_notify_ind_t	*dlip;
1604 
1605 	if (!(dsp->ds_notifications & DL_NOTE_CAPAB_RENEG))
1606 		return;
1607 
1608 	if ((mp = mexchange(dsp->ds_wq, NULL, sizeof (dl_notify_ind_t),
1609 	    M_PROTO, 0)) == NULL)
1610 		return;
1611 
1612 	bzero(mp->b_rptr, sizeof (dl_notify_ind_t));
1613 	dlip = (dl_notify_ind_t *)mp->b_rptr;
1614 	dlip->dl_primitive = DL_NOTIFY_IND;
1615 	dlip->dl_notification = DL_NOTE_CAPAB_RENEG;
1616 
1617 	qreply(dsp->ds_wq, mp);
1618 }
1619 
1620 /*
1621  * DL_NOTIFY_IND: DL_NOTE_FASTPATH_FLUSH
1622  */
1623 static void
1624 str_notify_fastpath_flush(dld_str_t *dsp)
1625 {
1626 	mblk_t		*mp;
1627 	dl_notify_ind_t	*dlip;
1628 
1629 	if (!(dsp->ds_notifications & DL_NOTE_FASTPATH_FLUSH))
1630 		return;
1631 
1632 	if ((mp = mexchange(dsp->ds_wq, NULL, sizeof (dl_notify_ind_t),
1633 	    M_PROTO, 0)) == NULL)
1634 		return;
1635 
1636 	bzero(mp->b_rptr, sizeof (dl_notify_ind_t));
1637 	dlip = (dl_notify_ind_t *)mp->b_rptr;
1638 	dlip->dl_primitive = DL_NOTIFY_IND;
1639 	dlip->dl_notification = DL_NOTE_FASTPATH_FLUSH;
1640 
1641 	qreply(dsp->ds_wq, mp);
1642 }
1643 
1644 /*
1645  * MAC notification callback.
1646  */
1647 void
1648 str_notify(void *arg, mac_notify_type_t type)
1649 {
1650 	dld_str_t		*dsp = (dld_str_t *)arg;
1651 	queue_t			*q = dsp->ds_wq;
1652 	mac_handle_t		mh = dsp->ds_mh;
1653 	mac_client_handle_t	mch = dsp->ds_mch;
1654 	uint8_t			addr[MAXMACADDRLEN];
1655 
1656 	switch (type) {
1657 	case MAC_NOTE_TX:
1658 		qenable(q);
1659 		break;
1660 
1661 	case MAC_NOTE_DEVPROMISC:
1662 		/*
1663 		 * Send the appropriate DL_NOTIFY_IND.
1664 		 */
1665 		if (mac_promisc_get(mh, MAC_DEVPROMISC))
1666 			str_notify_promisc_on_phys(dsp);
1667 		else
1668 			str_notify_promisc_off_phys(dsp);
1669 		break;
1670 
1671 	case MAC_NOTE_UNICST:
1672 		/*
1673 		 * This notification is sent whenever the MAC unicast
1674 		 * address changes.
1675 		 */
1676 		mac_unicast_primary_get(mh, addr);
1677 
1678 		/*
1679 		 * Send the appropriate DL_NOTIFY_IND.
1680 		 */
1681 		str_notify_phys_addr(dsp, addr);
1682 		break;
1683 
1684 	case MAC_NOTE_LINK:
1685 		/*
1686 		 * This notification is sent every time the MAC driver
1687 		 * updates the link state.
1688 		 */
1689 		switch (mac_client_stat_get(mch, MAC_STAT_LINK_STATE)) {
1690 		case LINK_STATE_UP: {
1691 			uint64_t speed;
1692 			/*
1693 			 * The link is up so send the appropriate
1694 			 * DL_NOTIFY_IND.
1695 			 */
1696 			str_notify_link_up(dsp);
1697 
1698 			speed = mac_stat_get(mh, MAC_STAT_IFSPEED);
1699 			str_notify_speed(dsp, (uint32_t)(speed / 1000ull));
1700 			break;
1701 		}
1702 		case LINK_STATE_DOWN:
1703 			/*
1704 			 * The link is down so send the appropriate
1705 			 * DL_NOTIFY_IND.
1706 			 */
1707 			str_notify_link_down(dsp);
1708 			break;
1709 
1710 		default:
1711 			break;
1712 		}
1713 		break;
1714 
1715 	case MAC_NOTE_RESOURCE:
1716 	case MAC_NOTE_CAPAB_CHG:
1717 		/*
1718 		 * This notification is sent whenever the MAC resources
1719 		 * change or capabilities change. We need to renegotiate
1720 		 * the capabilities. Send the appropriate DL_NOTIFY_IND.
1721 		 */
1722 		str_notify_capab_reneg(dsp);
1723 		break;
1724 
1725 	case MAC_NOTE_SDU_SIZE: {
1726 		uint_t  max_sdu;
1727 		mac_sdu_get(dsp->ds_mh, NULL, &max_sdu);
1728 		str_notify_sdu_size(dsp, max_sdu);
1729 		break;
1730 	}
1731 
1732 	case MAC_NOTE_FASTPATH_FLUSH:
1733 		str_notify_fastpath_flush(dsp);
1734 		break;
1735 
1736 	case MAC_NOTE_MARGIN:
1737 		break;
1738 
1739 	case MAC_NOTE_PROMISC:
1740 		break;
1741 
1742 	default:
1743 		ASSERT(B_FALSE);
1744 		break;
1745 	}
1746 }
1747 
1748 /*
1749  * This function is called via a taskq mechansim to process all control
1750  * messages on a per 'dsp' end point.
1751  */
1752 static void
1753 dld_wput_nondata_task(void *arg)
1754 {
1755 	dld_str_t	*dsp = arg;
1756 	mblk_t		*mp;
1757 
1758 	mutex_enter(&dsp->ds_lock);
1759 	while (dsp->ds_pending_head != NULL) {
1760 		mp = dsp->ds_pending_head;
1761 		dsp->ds_pending_head = mp->b_next;
1762 		mp->b_next = NULL;
1763 		if (dsp->ds_pending_head == NULL)
1764 			dsp->ds_pending_tail = NULL;
1765 		mutex_exit(&dsp->ds_lock);
1766 
1767 		switch (DB_TYPE(mp)) {
1768 		case M_PROTO:
1769 		case M_PCPROTO:
1770 			dld_proto(dsp, mp);
1771 			break;
1772 		case M_IOCTL:
1773 			dld_ioc(dsp, mp);
1774 			break;
1775 		default:
1776 			ASSERT(0);
1777 		}
1778 
1779 		mutex_enter(&dsp->ds_lock);
1780 	}
1781 	ASSERT(dsp->ds_pending_tail == NULL);
1782 	dsp->ds_dlpi_pending = 0;
1783 	cv_broadcast(&dsp->ds_dlpi_pending_cv);
1784 	mutex_exit(&dsp->ds_lock);
1785 }
1786 
1787 /*
1788  * Kernel thread to handle taskq dispatch failures in dld_wput_data. This
1789  * thread is started at boot time.
1790  */
1791 static void
1792 dld_taskq_dispatch(void)
1793 {
1794 	callb_cpr_t	cprinfo;
1795 	dld_str_t	*dsp;
1796 
1797 	CALLB_CPR_INIT(&cprinfo, &dld_taskq_lock, callb_generic_cpr,
1798 	    "dld_taskq_dispatch");
1799 	mutex_enter(&dld_taskq_lock);
1800 
1801 	while (!dld_taskq_quit) {
1802 		dsp = list_head(&dld_taskq_list);
1803 		while (dsp != NULL) {
1804 			list_remove(&dld_taskq_list, dsp);
1805 			mutex_exit(&dld_taskq_lock);
1806 			VERIFY(taskq_dispatch(dld_taskq, dld_wput_nondata_task,
1807 			    dsp, TQ_SLEEP) != 0);
1808 			mutex_enter(&dld_taskq_lock);
1809 			dsp = list_head(&dld_taskq_list);
1810 		}
1811 
1812 		CALLB_CPR_SAFE_BEGIN(&cprinfo);
1813 		cv_wait(&dld_taskq_cv, &dld_taskq_lock);
1814 		CALLB_CPR_SAFE_END(&cprinfo, &dld_taskq_lock);
1815 	}
1816 
1817 	dld_taskq_done = B_TRUE;
1818 	cv_signal(&dld_taskq_cv);
1819 	CALLB_CPR_EXIT(&cprinfo);
1820 	thread_exit();
1821 }
1822 
1823 /*
1824  * All control operations are serialized on the 'dsp' and are also funneled
1825  * through a taskq mechanism to ensure that subsequent processing has kernel
1826  * context and can safely use cv_wait.
1827  *
1828  * Mechanisms to handle taskq dispatch failures
1829  *
1830  * The only way to be sure that taskq dispatch does not fail is to either
1831  * specify TQ_SLEEP or to use a static taskq and prepopulate it with
1832  * some number of entries and make sure that the number of outstanding requests
1833  * are less than that number. We can't use TQ_SLEEP since we don't know the
1834  * context. Nor can we bound the total number of 'dsp' end points. So we are
1835  * unable to use either of the above schemes, and are forced to deal with
1836  * taskq dispatch failures. Note that even dynamic taskq could fail in
1837  * dispatch if TQ_NOSLEEP is specified, since this flag is translated
1838  * eventually to KM_NOSLEEP and kmem allocations could fail in the taskq
1839  * framework.
1840  *
1841  * We maintain a queue of 'dsp's that encountered taskq dispatch failure.
1842  * We also have a single global thread to retry the taskq dispatch. This
1843  * thread loops in 'dld_taskq_dispatch' and retries the taskq dispatch, but
1844  * uses TQ_SLEEP to ensure eventual success of the dispatch operation.
1845  */
1846 static void
1847 dld_wput_nondata(dld_str_t *dsp, mblk_t *mp)
1848 {
1849 	ASSERT(mp->b_next == NULL);
1850 	mutex_enter(&dsp->ds_lock);
1851 	if (dsp->ds_pending_head != NULL) {
1852 		ASSERT(dsp->ds_dlpi_pending);
1853 		dsp->ds_pending_tail->b_next = mp;
1854 		dsp->ds_pending_tail = mp;
1855 		mutex_exit(&dsp->ds_lock);
1856 		return;
1857 	}
1858 	ASSERT(dsp->ds_pending_tail == NULL);
1859 	dsp->ds_pending_head = dsp->ds_pending_tail = mp;
1860 	/*
1861 	 * At this point if ds_dlpi_pending is set, it implies that the taskq
1862 	 * thread is still active and is processing the last message, though
1863 	 * the pending queue has been emptied.
1864 	 */
1865 	if (dsp->ds_dlpi_pending) {
1866 		mutex_exit(&dsp->ds_lock);
1867 		return;
1868 	}
1869 
1870 	dsp->ds_dlpi_pending = 1;
1871 	mutex_exit(&dsp->ds_lock);
1872 
1873 	if (taskq_dispatch(dld_taskq, dld_wput_nondata_task, dsp,
1874 	    TQ_NOSLEEP) != 0)
1875 		return;
1876 
1877 	mutex_enter(&dld_taskq_lock);
1878 	list_insert_tail(&dld_taskq_list, dsp);
1879 	cv_signal(&dld_taskq_cv);
1880 	mutex_exit(&dld_taskq_lock);
1881 }
1882 
1883 /*
1884  * Process an M_IOCTL message.
1885  */
1886 static void
1887 dld_ioc(dld_str_t *dsp, mblk_t *mp)
1888 {
1889 	uint_t			cmd;
1890 
1891 	cmd = ((struct iocblk *)mp->b_rptr)->ioc_cmd;
1892 	ASSERT(dsp->ds_type == DLD_DLPI);
1893 
1894 	switch (cmd) {
1895 	case DLIOCNATIVE:
1896 		ioc_native(dsp, mp);
1897 		break;
1898 	case DLIOCMARGININFO:
1899 		ioc_margin(dsp, mp);
1900 		break;
1901 	case DLIOCRAW:
1902 		ioc_raw(dsp, mp);
1903 		break;
1904 	case DLIOCHDRINFO:
1905 		ioc_fast(dsp, mp);
1906 		break;
1907 	default:
1908 		ioc(dsp, mp);
1909 	}
1910 }
1911 
1912 /*
1913  * DLIOCNATIVE
1914  */
1915 static void
1916 ioc_native(dld_str_t *dsp, mblk_t *mp)
1917 {
1918 	queue_t *q = dsp->ds_wq;
1919 	const mac_info_t *mip = dsp->ds_mip;
1920 
1921 	/*
1922 	 * Native mode can be enabled if it's disabled and if the
1923 	 * native media type is different.
1924 	 */
1925 	if (!dsp->ds_native && mip->mi_media != mip->mi_nativemedia)
1926 		dsp->ds_native = B_TRUE;
1927 
1928 	if (dsp->ds_native)
1929 		miocack(q, mp, 0, mip->mi_nativemedia);
1930 	else
1931 		miocnak(q, mp, 0, ENOTSUP);
1932 }
1933 
1934 /*
1935  * DLIOCMARGININFO
1936  */
1937 static void
1938 ioc_margin(dld_str_t *dsp, mblk_t *mp)
1939 {
1940 	queue_t *q = dsp->ds_wq;
1941 	uint32_t margin;
1942 	int err;
1943 
1944 	if (dsp->ds_dlstate == DL_UNATTACHED) {
1945 		err = EINVAL;
1946 		goto failed;
1947 	}
1948 	if ((err = miocpullup(mp, sizeof (uint32_t))) != 0)
1949 		goto failed;
1950 
1951 	mac_margin_get(dsp->ds_mh, &margin);
1952 	*((uint32_t *)mp->b_cont->b_rptr) = margin;
1953 	miocack(q, mp, sizeof (uint32_t), 0);
1954 	return;
1955 
1956 failed:
1957 	miocnak(q, mp, 0, err);
1958 }
1959 
1960 /*
1961  * DLIOCRAW
1962  */
1963 static void
1964 ioc_raw(dld_str_t *dsp, mblk_t *mp)
1965 {
1966 	queue_t *q = dsp->ds_wq;
1967 	mac_perim_handle_t	mph;
1968 
1969 	if (dsp->ds_mh == NULL) {
1970 		dsp->ds_mode = DLD_RAW;
1971 		miocack(q, mp, 0, 0);
1972 		return;
1973 	}
1974 
1975 	mac_perim_enter_by_mh(dsp->ds_mh, &mph);
1976 	if (dsp->ds_polling || dsp->ds_direct) {
1977 		mac_perim_exit(mph);
1978 		miocnak(q, mp, 0, EPROTO);
1979 		return;
1980 	}
1981 
1982 	if (dsp->ds_mode != DLD_RAW && dsp->ds_dlstate == DL_IDLE) {
1983 		/*
1984 		 * Set the receive callback.
1985 		 */
1986 		dls_rx_set(dsp, dld_str_rx_raw, dsp);
1987 	}
1988 
1989 	/*
1990 	 * Note that raw mode is enabled.
1991 	 */
1992 	dsp->ds_mode = DLD_RAW;
1993 	mac_perim_exit(mph);
1994 
1995 	miocack(q, mp, 0, 0);
1996 }
1997 
1998 /*
1999  * DLIOCHDRINFO
2000  */
2001 static void
2002 ioc_fast(dld_str_t *dsp, mblk_t *mp)
2003 {
2004 	dl_unitdata_req_t *dlp;
2005 	off_t		off;
2006 	size_t		len;
2007 	const uint8_t	*addr;
2008 	uint16_t	sap;
2009 	mblk_t		*nmp;
2010 	mblk_t		*hmp;
2011 	uint_t		addr_length;
2012 	queue_t		*q = dsp->ds_wq;
2013 	int		err;
2014 	mac_perim_handle_t	mph;
2015 
2016 	if (dld_opt & DLD_OPT_NO_FASTPATH) {
2017 		err = ENOTSUP;
2018 		goto failed;
2019 	}
2020 
2021 	/*
2022 	 * DLIOCHDRINFO should only come from IP. The one initiated from
2023 	 * user-land should not be allowed.
2024 	 */
2025 	if (((struct iocblk *)mp->b_rptr)->ioc_cr != kcred) {
2026 		err = EINVAL;
2027 		goto failed;
2028 	}
2029 
2030 	nmp = mp->b_cont;
2031 	if (nmp == NULL || MBLKL(nmp) < sizeof (dl_unitdata_req_t) ||
2032 	    (dlp = (dl_unitdata_req_t *)nmp->b_rptr,
2033 	    dlp->dl_primitive != DL_UNITDATA_REQ)) {
2034 		err = EINVAL;
2035 		goto failed;
2036 	}
2037 
2038 	off = dlp->dl_dest_addr_offset;
2039 	len = dlp->dl_dest_addr_length;
2040 
2041 	if (!MBLKIN(nmp, off, len)) {
2042 		err = EINVAL;
2043 		goto failed;
2044 	}
2045 
2046 	if (dsp->ds_dlstate != DL_IDLE) {
2047 		err = ENOTSUP;
2048 		goto failed;
2049 	}
2050 
2051 	addr_length = dsp->ds_mip->mi_addr_length;
2052 	if (len != addr_length + sizeof (uint16_t)) {
2053 		err = EINVAL;
2054 		goto failed;
2055 	}
2056 
2057 	addr = nmp->b_rptr + off;
2058 	sap = *(uint16_t *)(nmp->b_rptr + off + addr_length);
2059 
2060 	if ((hmp = dls_header(dsp, addr, sap, 0, NULL)) == NULL) {
2061 		err = ENOMEM;
2062 		goto failed;
2063 	}
2064 
2065 	/*
2066 	 * This ioctl might happen concurrently with a direct call to dld_capab
2067 	 * that tries to enable direct and/or poll capabilities. Since the
2068 	 * stack does not serialize them, we do so here to avoid mixing
2069 	 * the callbacks.
2070 	 */
2071 	mac_perim_enter_by_mh(dsp->ds_mh, &mph);
2072 	if (dsp->ds_mode != DLD_FASTPATH) {
2073 		/*
2074 		 * Set the receive callback (unless polling is enabled).
2075 		 */
2076 		if (!dsp->ds_polling && !dsp->ds_direct)
2077 			dls_rx_set(dsp, dld_str_rx_fastpath, dsp);
2078 
2079 		/*
2080 		 * Note that fast-path mode is enabled.
2081 		 */
2082 		dsp->ds_mode = DLD_FASTPATH;
2083 	}
2084 	mac_perim_exit(mph);
2085 
2086 	freemsg(nmp->b_cont);
2087 	nmp->b_cont = hmp;
2088 
2089 	miocack(q, mp, MBLKL(nmp) + MBLKL(hmp), 0);
2090 	return;
2091 failed:
2092 	miocnak(q, mp, 0, err);
2093 }
2094 
2095 /*
2096  * Catch-all handler.
2097  */
2098 static void
2099 ioc(dld_str_t *dsp, mblk_t *mp)
2100 {
2101 	queue_t	*q = dsp->ds_wq;
2102 
2103 	if (dsp->ds_dlstate == DL_UNATTACHED) {
2104 		miocnak(q, mp, 0, EINVAL);
2105 		return;
2106 	}
2107 	mac_ioctl(dsp->ds_mh, q, mp);
2108 }
2109