xref: /titanic_52/usr/src/uts/common/io/dld/dld_proto.c (revision bd670b35a010421b6e1a5536c34453a827007c81)
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 #include <sys/sysmacros.h>
30 #include <sys/strsubr.h>
31 #include <sys/strsun.h>
32 #include <sys/vlan.h>
33 #include <sys/dld_impl.h>
34 #include <sys/mac_client.h>
35 #include <sys/mac_client_impl.h>
36 #include <sys/mac_client_priv.h>
37 
38 typedef void proto_reqfunc_t(dld_str_t *, mblk_t *);
39 
40 static proto_reqfunc_t proto_info_req, proto_attach_req, proto_detach_req,
41     proto_bind_req, proto_unbind_req, proto_promiscon_req, proto_promiscoff_req,
42     proto_enabmulti_req, proto_disabmulti_req, proto_physaddr_req,
43     proto_setphysaddr_req, proto_udqos_req, proto_req, proto_capability_req,
44     proto_notify_req, proto_passive_req;
45 
46 static void proto_capability_advertise(dld_str_t *, mblk_t *);
47 static int dld_capab_poll_disable(dld_str_t *, dld_capab_poll_t *);
48 static boolean_t check_mod_above(queue_t *, const char *);
49 
50 #define	DL_ACK_PENDING(state) \
51 	((state) == DL_ATTACH_PENDING || \
52 	(state) == DL_DETACH_PENDING || \
53 	(state) == DL_BIND_PENDING || \
54 	(state) == DL_UNBIND_PENDING)
55 
56 /*
57  * Process a DLPI protocol message.
58  * The primitives DL_BIND_REQ, DL_ENABMULTI_REQ, DL_PROMISCON_REQ,
59  * DL_SET_PHYS_ADDR_REQ put the data link below our dld_str_t into an
60  * 'active' state. The primitive DL_PASSIVE_REQ marks our dld_str_t
61  * as 'passive' and forbids it from being subsequently made 'active'
62  * by the above primitives.
63  */
64 void
65 dld_proto(dld_str_t *dsp, mblk_t *mp)
66 {
67 	t_uscalar_t		prim;
68 
69 	if (MBLKL(mp) < sizeof (t_uscalar_t)) {
70 		freemsg(mp);
71 		return;
72 	}
73 	prim = ((union DL_primitives *)mp->b_rptr)->dl_primitive;
74 
75 	switch (prim) {
76 	case DL_INFO_REQ:
77 		proto_info_req(dsp, mp);
78 		break;
79 	case DL_BIND_REQ:
80 		proto_bind_req(dsp, mp);
81 		break;
82 	case DL_UNBIND_REQ:
83 		proto_unbind_req(dsp, mp);
84 		break;
85 	case DL_UNITDATA_REQ:
86 		proto_unitdata_req(dsp, mp);
87 		break;
88 	case DL_UDQOS_REQ:
89 		proto_udqos_req(dsp, mp);
90 		break;
91 	case DL_ATTACH_REQ:
92 		proto_attach_req(dsp, mp);
93 		break;
94 	case DL_DETACH_REQ:
95 		proto_detach_req(dsp, mp);
96 		break;
97 	case DL_ENABMULTI_REQ:
98 		proto_enabmulti_req(dsp, mp);
99 		break;
100 	case DL_DISABMULTI_REQ:
101 		proto_disabmulti_req(dsp, mp);
102 		break;
103 	case DL_PROMISCON_REQ:
104 		proto_promiscon_req(dsp, mp);
105 		break;
106 	case DL_PROMISCOFF_REQ:
107 		proto_promiscoff_req(dsp, mp);
108 		break;
109 	case DL_PHYS_ADDR_REQ:
110 		proto_physaddr_req(dsp, mp);
111 		break;
112 	case DL_SET_PHYS_ADDR_REQ:
113 		proto_setphysaddr_req(dsp, mp);
114 		break;
115 	case DL_NOTIFY_REQ:
116 		proto_notify_req(dsp, mp);
117 		break;
118 	case DL_CAPABILITY_REQ:
119 		proto_capability_req(dsp, mp);
120 		break;
121 	case DL_PASSIVE_REQ:
122 		proto_passive_req(dsp, mp);
123 		break;
124 	default:
125 		proto_req(dsp, mp);
126 		break;
127 	}
128 }
129 
130 #define	NEG(x)	-(x)
131 typedef struct dl_info_ack_wrapper {
132 	dl_info_ack_t		dl_info;
133 	uint8_t			dl_addr[MAXMACADDRLEN + sizeof (uint16_t)];
134 	uint8_t			dl_brdcst_addr[MAXMACADDRLEN];
135 	dl_qos_cl_range1_t	dl_qos_range1;
136 	dl_qos_cl_sel1_t	dl_qos_sel1;
137 } dl_info_ack_wrapper_t;
138 
139 /*
140  * DL_INFO_REQ
141  */
142 static void
143 proto_info_req(dld_str_t *dsp, mblk_t *mp)
144 {
145 	dl_info_ack_wrapper_t	*dlwp;
146 	dl_info_ack_t		*dlp;
147 	dl_qos_cl_sel1_t	*selp;
148 	dl_qos_cl_range1_t	*rangep;
149 	uint8_t			*addr;
150 	uint8_t			*brdcst_addr;
151 	uint_t			addr_length;
152 	uint_t			sap_length;
153 	mac_info_t		minfo;
154 	mac_info_t		*minfop;
155 	queue_t			*q = dsp->ds_wq;
156 
157 	/*
158 	 * Swap the request message for one large enough to contain the
159 	 * wrapper structure defined above.
160 	 */
161 	if ((mp = mexchange(q, mp, sizeof (dl_info_ack_wrapper_t),
162 	    M_PCPROTO, 0)) == NULL)
163 		return;
164 
165 	bzero(mp->b_rptr, sizeof (dl_info_ack_wrapper_t));
166 	dlwp = (dl_info_ack_wrapper_t *)mp->b_rptr;
167 
168 	dlp = &(dlwp->dl_info);
169 	ASSERT(dlp == (dl_info_ack_t *)mp->b_rptr);
170 
171 	dlp->dl_primitive = DL_INFO_ACK;
172 
173 	/*
174 	 * Set up the sub-structure pointers.
175 	 */
176 	addr = dlwp->dl_addr;
177 	brdcst_addr = dlwp->dl_brdcst_addr;
178 	rangep = &(dlwp->dl_qos_range1);
179 	selp = &(dlwp->dl_qos_sel1);
180 
181 	/*
182 	 * This driver supports only version 2 connectionless DLPI provider
183 	 * nodes.
184 	 */
185 	dlp->dl_service_mode = DL_CLDLS;
186 	dlp->dl_version = DL_VERSION_2;
187 
188 	/*
189 	 * Set the style of the provider
190 	 */
191 	dlp->dl_provider_style = dsp->ds_style;
192 	ASSERT(dlp->dl_provider_style == DL_STYLE1 ||
193 	    dlp->dl_provider_style == DL_STYLE2);
194 
195 	/*
196 	 * Set the current DLPI state.
197 	 */
198 	dlp->dl_current_state = dsp->ds_dlstate;
199 
200 	/*
201 	 * Gratuitously set the media type. This is to deal with modules
202 	 * that assume the media type is known prior to DL_ATTACH_REQ
203 	 * being completed.
204 	 */
205 	dlp->dl_mac_type = DL_ETHER;
206 
207 	/*
208 	 * If the stream is not at least attached we try to retrieve the
209 	 * mac_info using mac_info_get()
210 	 */
211 	if (dsp->ds_dlstate == DL_UNATTACHED ||
212 	    dsp->ds_dlstate == DL_ATTACH_PENDING ||
213 	    dsp->ds_dlstate == DL_DETACH_PENDING) {
214 		if (!mac_info_get(ddi_major_to_name(dsp->ds_major), &minfo)) {
215 			/*
216 			 * Cannot find mac_info. giving up.
217 			 */
218 			goto done;
219 		}
220 		minfop = &minfo;
221 	} else {
222 		minfop = (mac_info_t *)dsp->ds_mip;
223 		/* We can only get the sdu if we're attached. */
224 		mac_sdu_get(dsp->ds_mh, &dlp->dl_min_sdu, &dlp->dl_max_sdu);
225 	}
226 
227 	/*
228 	 * Set the media type (properly this time).
229 	 */
230 	if (dsp->ds_native)
231 		dlp->dl_mac_type = minfop->mi_nativemedia;
232 	else
233 		dlp->dl_mac_type = minfop->mi_media;
234 
235 	/*
236 	 * Set the DLSAP length. We only support 16 bit values and they
237 	 * appear after the MAC address portion of DLSAP addresses.
238 	 */
239 	sap_length = sizeof (uint16_t);
240 	dlp->dl_sap_length = NEG(sap_length);
241 
242 	addr_length = minfop->mi_addr_length;
243 
244 	/*
245 	 * Copy in the media broadcast address.
246 	 */
247 	if (minfop->mi_brdcst_addr != NULL) {
248 		dlp->dl_brdcst_addr_offset =
249 		    (uintptr_t)brdcst_addr - (uintptr_t)dlp;
250 		bcopy(minfop->mi_brdcst_addr, brdcst_addr, addr_length);
251 		dlp->dl_brdcst_addr_length = addr_length;
252 	}
253 
254 	/* Only VLAN links and links that have a normal tag mode support QOS. */
255 	if ((dsp->ds_mch != NULL &&
256 	    mac_client_vid(dsp->ds_mch) != VLAN_ID_NONE) ||
257 	    (dsp->ds_dlp != NULL &&
258 	    dsp->ds_dlp->dl_tagmode == LINK_TAGMODE_NORMAL)) {
259 		dlp->dl_qos_range_offset = (uintptr_t)rangep - (uintptr_t)dlp;
260 		dlp->dl_qos_range_length = sizeof (dl_qos_cl_range1_t);
261 
262 		rangep->dl_qos_type = DL_QOS_CL_RANGE1;
263 		rangep->dl_trans_delay.dl_target_value = DL_UNKNOWN;
264 		rangep->dl_trans_delay.dl_accept_value = DL_UNKNOWN;
265 		rangep->dl_protection.dl_min = DL_UNKNOWN;
266 		rangep->dl_protection.dl_max = DL_UNKNOWN;
267 		rangep->dl_residual_error = DL_UNKNOWN;
268 
269 		/*
270 		 * Specify the supported range of priorities.
271 		 */
272 		rangep->dl_priority.dl_min = 0;
273 		rangep->dl_priority.dl_max = (1 << VLAN_PRI_SIZE) - 1;
274 
275 		dlp->dl_qos_offset = (uintptr_t)selp - (uintptr_t)dlp;
276 		dlp->dl_qos_length = sizeof (dl_qos_cl_sel1_t);
277 
278 		selp->dl_qos_type = DL_QOS_CL_SEL1;
279 		selp->dl_trans_delay = DL_UNKNOWN;
280 		selp->dl_protection = DL_UNKNOWN;
281 		selp->dl_residual_error = DL_UNKNOWN;
282 
283 		/*
284 		 * Specify the current priority (which can be changed by
285 		 * the DL_UDQOS_REQ primitive).
286 		 */
287 		selp->dl_priority = dsp->ds_pri;
288 	}
289 
290 	dlp->dl_addr_length = addr_length + sizeof (uint16_t);
291 	if (dsp->ds_dlstate == DL_IDLE) {
292 		/*
293 		 * The stream is bound. Therefore we can formulate a valid
294 		 * DLSAP address.
295 		 */
296 		dlp->dl_addr_offset = (uintptr_t)addr - (uintptr_t)dlp;
297 		if (addr_length > 0)
298 			mac_unicast_primary_get(dsp->ds_mh, addr);
299 
300 		*(uint16_t *)(addr + addr_length) = dsp->ds_sap;
301 	}
302 
303 done:
304 	ASSERT(IMPLY(dlp->dl_qos_offset != 0, dlp->dl_qos_length != 0));
305 	ASSERT(IMPLY(dlp->dl_qos_range_offset != 0,
306 	    dlp->dl_qos_range_length != 0));
307 	ASSERT(IMPLY(dlp->dl_addr_offset != 0, dlp->dl_addr_length != 0));
308 	ASSERT(IMPLY(dlp->dl_brdcst_addr_offset != 0,
309 	    dlp->dl_brdcst_addr_length != 0));
310 
311 	qreply(q, mp);
312 }
313 
314 /*
315  * DL_ATTACH_REQ
316  */
317 static void
318 proto_attach_req(dld_str_t *dsp, mblk_t *mp)
319 {
320 	dl_attach_req_t	*dlp = (dl_attach_req_t *)mp->b_rptr;
321 	int		err = 0;
322 	t_uscalar_t	dl_err;
323 	queue_t		*q = dsp->ds_wq;
324 
325 	if (MBLKL(mp) < sizeof (dl_attach_req_t) ||
326 	    dlp->dl_ppa < 0 || dsp->ds_style == DL_STYLE1) {
327 		dl_err = DL_BADPRIM;
328 		goto failed;
329 	}
330 
331 	if (dsp->ds_dlstate != DL_UNATTACHED) {
332 		dl_err = DL_OUTSTATE;
333 		goto failed;
334 	}
335 
336 	dsp->ds_dlstate = DL_ATTACH_PENDING;
337 
338 	err = dld_str_attach(dsp, dlp->dl_ppa);
339 	if (err != 0) {
340 		switch (err) {
341 		case ENOENT:
342 			dl_err = DL_BADPPA;
343 			err = 0;
344 			break;
345 		default:
346 			dl_err = DL_SYSERR;
347 			break;
348 		}
349 		dsp->ds_dlstate = DL_UNATTACHED;
350 		goto failed;
351 	}
352 	ASSERT(dsp->ds_dlstate == DL_UNBOUND);
353 	dlokack(q, mp, DL_ATTACH_REQ);
354 	return;
355 
356 failed:
357 	dlerrorack(q, mp, DL_ATTACH_REQ, dl_err, (t_uscalar_t)err);
358 }
359 
360 /*
361  * DL_DETACH_REQ
362  */
363 static void
364 proto_detach_req(dld_str_t *dsp, mblk_t *mp)
365 {
366 	queue_t		*q = dsp->ds_wq;
367 	t_uscalar_t	dl_err;
368 
369 	if (MBLKL(mp) < sizeof (dl_detach_req_t)) {
370 		dl_err = DL_BADPRIM;
371 		goto failed;
372 	}
373 
374 	if (dsp->ds_dlstate != DL_UNBOUND) {
375 		dl_err = DL_OUTSTATE;
376 		goto failed;
377 	}
378 
379 	if (dsp->ds_style == DL_STYLE1) {
380 		dl_err = DL_BADPRIM;
381 		goto failed;
382 	}
383 
384 	ASSERT(dsp->ds_datathr_cnt == 0);
385 	dsp->ds_dlstate = DL_DETACH_PENDING;
386 
387 	dld_str_detach(dsp);
388 	dlokack(dsp->ds_wq, mp, DL_DETACH_REQ);
389 	return;
390 
391 failed:
392 	dlerrorack(q, mp, DL_DETACH_REQ, dl_err, 0);
393 }
394 
395 /*
396  * DL_BIND_REQ
397  */
398 static void
399 proto_bind_req(dld_str_t *dsp, mblk_t *mp)
400 {
401 	dl_bind_req_t	*dlp = (dl_bind_req_t *)mp->b_rptr;
402 	int		err = 0;
403 	uint8_t		dlsap_addr[MAXMACADDRLEN + sizeof (uint16_t)];
404 	uint_t		dlsap_addr_length;
405 	t_uscalar_t	dl_err;
406 	t_scalar_t	sap;
407 	queue_t		*q = dsp->ds_wq;
408 	mac_perim_handle_t	mph;
409 	void		*mdip;
410 	int32_t		intr_cpu;
411 
412 	if (MBLKL(mp) < sizeof (dl_bind_req_t)) {
413 		dl_err = DL_BADPRIM;
414 		goto failed;
415 	}
416 
417 	if (dlp->dl_xidtest_flg != 0) {
418 		dl_err = DL_NOAUTO;
419 		goto failed;
420 	}
421 
422 	if (dlp->dl_service_mode != DL_CLDLS) {
423 		dl_err = DL_UNSUPPORTED;
424 		goto failed;
425 	}
426 
427 	if (dsp->ds_dlstate != DL_UNBOUND) {
428 		dl_err = DL_OUTSTATE;
429 		goto failed;
430 	}
431 
432 	mac_perim_enter_by_mh(dsp->ds_mh, &mph);
433 
434 	if ((err = dls_active_set(dsp)) != 0) {
435 		dl_err = DL_SYSERR;
436 		goto failed2;
437 	}
438 
439 	dsp->ds_dlstate = DL_BIND_PENDING;
440 	/*
441 	 * Set the receive callback.
442 	 */
443 	dls_rx_set(dsp, (dsp->ds_mode == DLD_RAW) ?
444 	    dld_str_rx_raw : dld_str_rx_unitdata, dsp);
445 
446 	/*
447 	 * Bind the channel such that it can receive packets.
448 	 */
449 	sap = dlp->dl_sap;
450 	dsp->ds_nonip = !check_mod_above(dsp->ds_rq, "ip") &&
451 	    !check_mod_above(dsp->ds_rq, "arp");
452 
453 	err = dls_bind(dsp, sap);
454 	if (err != 0) {
455 		switch (err) {
456 		case EINVAL:
457 			dl_err = DL_BADADDR;
458 			err = 0;
459 			break;
460 		default:
461 			dl_err = DL_SYSERR;
462 			break;
463 		}
464 
465 		dsp->ds_dlstate = DL_UNBOUND;
466 		dls_active_clear(dsp, B_FALSE);
467 		goto failed2;
468 	}
469 
470 	intr_cpu = mac_client_intr_cpu(dsp->ds_mch);
471 	mdip = mac_get_devinfo(dsp->ds_mh);
472 	mac_perim_exit(mph);
473 
474 	/*
475 	 * We do this after we get out of the perim to avoid deadlocks
476 	 * etc. since part of mac_client_retarget_intr is to walk the
477 	 * device tree in order to find and retarget the interrupts.
478 	 */
479 	mac_client_set_intr_cpu(mdip, dsp->ds_mch, intr_cpu);
480 
481 	/*
482 	 * Copy in MAC address.
483 	 */
484 	dlsap_addr_length = dsp->ds_mip->mi_addr_length;
485 	mac_unicast_primary_get(dsp->ds_mh, dlsap_addr);
486 
487 	/*
488 	 * Copy in the SAP.
489 	 */
490 	*(uint16_t *)(dlsap_addr + dlsap_addr_length) = sap;
491 	dlsap_addr_length += sizeof (uint16_t);
492 
493 	dsp->ds_dlstate = DL_IDLE;
494 	dlbindack(q, mp, sap, dlsap_addr, dlsap_addr_length, 0, 0);
495 	return;
496 
497 failed2:
498 	mac_perim_exit(mph);
499 failed:
500 	dlerrorack(q, mp, DL_BIND_REQ, dl_err, (t_uscalar_t)err);
501 }
502 
503 /*
504  * DL_UNBIND_REQ
505  */
506 static void
507 proto_unbind_req(dld_str_t *dsp, mblk_t *mp)
508 {
509 	queue_t		*q = dsp->ds_wq;
510 	t_uscalar_t	dl_err;
511 	mac_perim_handle_t	mph;
512 
513 	if (MBLKL(mp) < sizeof (dl_unbind_req_t)) {
514 		dl_err = DL_BADPRIM;
515 		goto failed;
516 	}
517 
518 	if (dsp->ds_dlstate != DL_IDLE) {
519 		dl_err = DL_OUTSTATE;
520 		goto failed;
521 	}
522 
523 	mutex_enter(&dsp->ds_lock);
524 	while (dsp->ds_datathr_cnt != 0)
525 		cv_wait(&dsp->ds_datathr_cv, &dsp->ds_lock);
526 
527 	dsp->ds_dlstate = DL_UNBIND_PENDING;
528 	mutex_exit(&dsp->ds_lock);
529 
530 	mac_perim_enter_by_mh(dsp->ds_mh, &mph);
531 	/*
532 	 * Unbind the channel to stop packets being received.
533 	 */
534 	dls_unbind(dsp);
535 
536 	/*
537 	 * Disable polling mode, if it is enabled.
538 	 */
539 	(void) dld_capab_poll_disable(dsp, NULL);
540 
541 	/*
542 	 * Clear LSO flags.
543 	 */
544 	dsp->ds_lso = B_FALSE;
545 	dsp->ds_lso_max = 0;
546 
547 	/*
548 	 * Clear the receive callback.
549 	 */
550 	dls_rx_set(dsp, NULL, NULL);
551 	dsp->ds_direct = B_FALSE;
552 
553 	/*
554 	 * Set the mode back to the default (unitdata).
555 	 */
556 	dsp->ds_mode = DLD_UNITDATA;
557 	dsp->ds_dlstate = DL_UNBOUND;
558 
559 	dls_active_clear(dsp, B_FALSE);
560 	mac_perim_exit(mph);
561 	dlokack(dsp->ds_wq, mp, DL_UNBIND_REQ);
562 	return;
563 failed:
564 	dlerrorack(q, mp, DL_UNBIND_REQ, dl_err, 0);
565 }
566 
567 /*
568  * DL_PROMISCON_REQ
569  */
570 static void
571 proto_promiscon_req(dld_str_t *dsp, mblk_t *mp)
572 {
573 	dl_promiscon_req_t *dlp = (dl_promiscon_req_t *)mp->b_rptr;
574 	int		err = 0;
575 	t_uscalar_t	dl_err;
576 	uint32_t	promisc_saved;
577 	queue_t		*q = dsp->ds_wq;
578 	mac_perim_handle_t	mph;
579 
580 	if (MBLKL(mp) < sizeof (dl_promiscon_req_t)) {
581 		dl_err = DL_BADPRIM;
582 		goto failed;
583 	}
584 
585 	if (dsp->ds_dlstate == DL_UNATTACHED ||
586 	    DL_ACK_PENDING(dsp->ds_dlstate)) {
587 		dl_err = DL_OUTSTATE;
588 		goto failed;
589 	}
590 
591 	promisc_saved = dsp->ds_promisc;
592 	switch (dlp->dl_level) {
593 	case DL_PROMISC_SAP:
594 		dsp->ds_promisc |= DLS_PROMISC_SAP;
595 		break;
596 
597 	case DL_PROMISC_MULTI:
598 		dsp->ds_promisc |= DLS_PROMISC_MULTI;
599 		break;
600 
601 	case DL_PROMISC_PHYS:
602 		dsp->ds_promisc |= DLS_PROMISC_PHYS;
603 		break;
604 
605 	default:
606 		dl_err = DL_NOTSUPPORTED;
607 		goto failed;
608 	}
609 
610 	mac_perim_enter_by_mh(dsp->ds_mh, &mph);
611 
612 	if ((promisc_saved == 0) && (err = dls_active_set(dsp)) != 0) {
613 		dsp->ds_promisc = promisc_saved;
614 		dl_err = DL_SYSERR;
615 		goto failed2;
616 	}
617 
618 	/*
619 	 * Adjust channel promiscuity.
620 	 */
621 	err = dls_promisc(dsp, promisc_saved);
622 
623 	if (err != 0) {
624 		dl_err = DL_SYSERR;
625 		dsp->ds_promisc = promisc_saved;
626 		if (promisc_saved == 0)
627 			dls_active_clear(dsp, B_FALSE);
628 		goto failed2;
629 	}
630 
631 	mac_perim_exit(mph);
632 
633 	dlokack(q, mp, DL_PROMISCON_REQ);
634 	return;
635 
636 failed2:
637 	mac_perim_exit(mph);
638 failed:
639 	dlerrorack(q, mp, DL_PROMISCON_REQ, dl_err, (t_uscalar_t)err);
640 }
641 
642 /*
643  * DL_PROMISCOFF_REQ
644  */
645 static void
646 proto_promiscoff_req(dld_str_t *dsp, mblk_t *mp)
647 {
648 	dl_promiscoff_req_t *dlp = (dl_promiscoff_req_t *)mp->b_rptr;
649 	int		err = 0;
650 	t_uscalar_t	dl_err;
651 	uint32_t	promisc_saved;
652 	queue_t		*q = dsp->ds_wq;
653 	mac_perim_handle_t	mph;
654 
655 	if (MBLKL(mp) < sizeof (dl_promiscoff_req_t)) {
656 		dl_err = DL_BADPRIM;
657 		goto failed;
658 	}
659 
660 	if (dsp->ds_dlstate == DL_UNATTACHED ||
661 	    DL_ACK_PENDING(dsp->ds_dlstate)) {
662 		dl_err = DL_OUTSTATE;
663 		goto failed;
664 	}
665 
666 	promisc_saved = dsp->ds_promisc;
667 	switch (dlp->dl_level) {
668 	case DL_PROMISC_SAP:
669 		if (!(dsp->ds_promisc & DLS_PROMISC_SAP)) {
670 			dl_err = DL_NOTENAB;
671 			goto failed;
672 		}
673 		dsp->ds_promisc &= ~DLS_PROMISC_SAP;
674 		break;
675 
676 	case DL_PROMISC_MULTI:
677 		if (!(dsp->ds_promisc & DLS_PROMISC_MULTI)) {
678 			dl_err = DL_NOTENAB;
679 			goto failed;
680 		}
681 		dsp->ds_promisc &= ~DLS_PROMISC_MULTI;
682 		break;
683 
684 	case DL_PROMISC_PHYS:
685 		if (!(dsp->ds_promisc & DLS_PROMISC_PHYS)) {
686 			dl_err = DL_NOTENAB;
687 			goto failed;
688 		}
689 		dsp->ds_promisc &= ~DLS_PROMISC_PHYS;
690 		break;
691 
692 	default:
693 		dl_err = DL_NOTSUPPORTED;
694 		goto failed;
695 	}
696 
697 	mac_perim_enter_by_mh(dsp->ds_mh, &mph);
698 	/*
699 	 * Adjust channel promiscuity.
700 	 */
701 	err = dls_promisc(dsp, promisc_saved);
702 
703 	if (err != 0) {
704 		mac_perim_exit(mph);
705 		dl_err = DL_SYSERR;
706 		goto failed;
707 	}
708 
709 	if (dsp->ds_promisc == 0)
710 		dls_active_clear(dsp, B_FALSE);
711 
712 	mac_perim_exit(mph);
713 
714 	dlokack(q, mp, DL_PROMISCOFF_REQ);
715 	return;
716 failed:
717 	dlerrorack(q, mp, DL_PROMISCOFF_REQ, dl_err, (t_uscalar_t)err);
718 }
719 
720 /*
721  * DL_ENABMULTI_REQ
722  */
723 static void
724 proto_enabmulti_req(dld_str_t *dsp, mblk_t *mp)
725 {
726 	dl_enabmulti_req_t *dlp = (dl_enabmulti_req_t *)mp->b_rptr;
727 	int		err = 0;
728 	t_uscalar_t	dl_err;
729 	queue_t		*q = dsp->ds_wq;
730 	mac_perim_handle_t	mph;
731 
732 	if (dsp->ds_dlstate == DL_UNATTACHED ||
733 	    DL_ACK_PENDING(dsp->ds_dlstate)) {
734 		dl_err = DL_OUTSTATE;
735 		goto failed;
736 	}
737 
738 	if (MBLKL(mp) < sizeof (dl_enabmulti_req_t) ||
739 	    !MBLKIN(mp, dlp->dl_addr_offset, dlp->dl_addr_length) ||
740 	    dlp->dl_addr_length != dsp->ds_mip->mi_addr_length) {
741 		dl_err = DL_BADPRIM;
742 		goto failed;
743 	}
744 
745 	mac_perim_enter_by_mh(dsp->ds_mh, &mph);
746 
747 	if ((dsp->ds_dmap == NULL) && (err = dls_active_set(dsp)) != 0) {
748 		dl_err = DL_SYSERR;
749 		goto failed2;
750 	}
751 
752 	err = dls_multicst_add(dsp, mp->b_rptr + dlp->dl_addr_offset);
753 	if (err != 0) {
754 		switch (err) {
755 		case EINVAL:
756 			dl_err = DL_BADADDR;
757 			err = 0;
758 			break;
759 		case ENOSPC:
760 			dl_err = DL_TOOMANY;
761 			err = 0;
762 			break;
763 		default:
764 			dl_err = DL_SYSERR;
765 			break;
766 		}
767 		if (dsp->ds_dmap == NULL)
768 			dls_active_clear(dsp, B_FALSE);
769 		goto failed2;
770 	}
771 
772 	mac_perim_exit(mph);
773 
774 	dlokack(q, mp, DL_ENABMULTI_REQ);
775 	return;
776 
777 failed2:
778 	mac_perim_exit(mph);
779 failed:
780 	dlerrorack(q, mp, DL_ENABMULTI_REQ, dl_err, (t_uscalar_t)err);
781 }
782 
783 /*
784  * DL_DISABMULTI_REQ
785  */
786 static void
787 proto_disabmulti_req(dld_str_t *dsp, mblk_t *mp)
788 {
789 	dl_disabmulti_req_t *dlp = (dl_disabmulti_req_t *)mp->b_rptr;
790 	int		err = 0;
791 	t_uscalar_t	dl_err;
792 	queue_t		*q = dsp->ds_wq;
793 	mac_perim_handle_t	mph;
794 
795 	if (dsp->ds_dlstate == DL_UNATTACHED ||
796 	    DL_ACK_PENDING(dsp->ds_dlstate)) {
797 		dl_err = DL_OUTSTATE;
798 		goto failed;
799 	}
800 
801 	if (MBLKL(mp) < sizeof (dl_disabmulti_req_t) ||
802 	    !MBLKIN(mp, dlp->dl_addr_offset, dlp->dl_addr_length) ||
803 	    dlp->dl_addr_length != dsp->ds_mip->mi_addr_length) {
804 		dl_err = DL_BADPRIM;
805 		goto failed;
806 	}
807 
808 	mac_perim_enter_by_mh(dsp->ds_mh, &mph);
809 	err = dls_multicst_remove(dsp, mp->b_rptr + dlp->dl_addr_offset);
810 	if ((err == 0) && (dsp->ds_dmap == NULL))
811 		dls_active_clear(dsp, B_FALSE);
812 	mac_perim_exit(mph);
813 
814 	if (err != 0) {
815 	switch (err) {
816 		case EINVAL:
817 			dl_err = DL_BADADDR;
818 			err = 0;
819 			break;
820 
821 		case ENOENT:
822 			dl_err = DL_NOTENAB;
823 			err = 0;
824 			break;
825 
826 		default:
827 			dl_err = DL_SYSERR;
828 			break;
829 		}
830 		goto failed;
831 	}
832 	dlokack(q, mp, DL_DISABMULTI_REQ);
833 	return;
834 failed:
835 	dlerrorack(q, mp, DL_DISABMULTI_REQ, dl_err, (t_uscalar_t)err);
836 }
837 
838 /*
839  * DL_PHYS_ADDR_REQ
840  */
841 static void
842 proto_physaddr_req(dld_str_t *dsp, mblk_t *mp)
843 {
844 	dl_phys_addr_req_t *dlp = (dl_phys_addr_req_t *)mp->b_rptr;
845 	queue_t		*q = dsp->ds_wq;
846 	t_uscalar_t	dl_err = 0;
847 	char		*addr = NULL;
848 	uint_t		addr_length;
849 
850 	if (MBLKL(mp) < sizeof (dl_phys_addr_req_t)) {
851 		dl_err = DL_BADPRIM;
852 		goto done;
853 	}
854 
855 	if (dsp->ds_dlstate == DL_UNATTACHED ||
856 	    DL_ACK_PENDING(dsp->ds_dlstate)) {
857 		dl_err = DL_OUTSTATE;
858 		goto done;
859 	}
860 
861 	addr_length = dsp->ds_mip->mi_addr_length;
862 	if (addr_length > 0) {
863 		addr = kmem_alloc(addr_length, KM_SLEEP);
864 		switch (dlp->dl_addr_type) {
865 		case DL_CURR_PHYS_ADDR:
866 			mac_unicast_primary_get(dsp->ds_mh, (uint8_t *)addr);
867 			break;
868 		case DL_FACT_PHYS_ADDR:
869 			bcopy(dsp->ds_mip->mi_unicst_addr, addr, addr_length);
870 			break;
871 		case DL_CURR_DEST_ADDR:
872 			if (!mac_dst_get(dsp->ds_mh, (uint8_t *)addr))
873 				dl_err = DL_NOTSUPPORTED;
874 			break;
875 		default:
876 			dl_err = DL_UNSUPPORTED;
877 		}
878 	}
879 done:
880 	if (dl_err == 0)
881 		dlphysaddrack(q, mp, addr, (t_uscalar_t)addr_length);
882 	else
883 		dlerrorack(q, mp, DL_PHYS_ADDR_REQ, dl_err, 0);
884 	if (addr != NULL)
885 		kmem_free(addr, addr_length);
886 }
887 
888 /*
889  * DL_SET_PHYS_ADDR_REQ
890  */
891 static void
892 proto_setphysaddr_req(dld_str_t *dsp, mblk_t *mp)
893 {
894 	dl_set_phys_addr_req_t *dlp = (dl_set_phys_addr_req_t *)mp->b_rptr;
895 	int		err = 0;
896 	t_uscalar_t	dl_err;
897 	queue_t		*q = dsp->ds_wq;
898 	mac_perim_handle_t	mph;
899 
900 	if (dsp->ds_dlstate == DL_UNATTACHED ||
901 	    DL_ACK_PENDING(dsp->ds_dlstate)) {
902 		dl_err = DL_OUTSTATE;
903 		goto failed;
904 	}
905 
906 	if (MBLKL(mp) < sizeof (dl_set_phys_addr_req_t) ||
907 	    !MBLKIN(mp, dlp->dl_addr_offset, dlp->dl_addr_length) ||
908 	    dlp->dl_addr_length != dsp->ds_mip->mi_addr_length) {
909 		dl_err = DL_BADPRIM;
910 		goto failed;
911 	}
912 
913 	mac_perim_enter_by_mh(dsp->ds_mh, &mph);
914 
915 	if ((err = dls_active_set(dsp)) != 0) {
916 		dl_err = DL_SYSERR;
917 		goto failed2;
918 	}
919 
920 	/*
921 	 * If mac-nospoof is enabled and the link is owned by a
922 	 * non-global zone, changing the mac address is not allowed.
923 	 */
924 	if (dsp->ds_dlp->dl_zid != GLOBAL_ZONEID &&
925 	    mac_protect_enabled(dsp->ds_mch, MPT_MACNOSPOOF)) {
926 		dls_active_clear(dsp, B_FALSE);
927 		err = EACCES;
928 		goto failed2;
929 	}
930 
931 	err = mac_unicast_primary_set(dsp->ds_mh,
932 	    mp->b_rptr + dlp->dl_addr_offset);
933 	if (err != 0) {
934 		switch (err) {
935 		case EINVAL:
936 			dl_err = DL_BADADDR;
937 			err = 0;
938 			break;
939 
940 		default:
941 			dl_err = DL_SYSERR;
942 			break;
943 		}
944 		dls_active_clear(dsp, B_FALSE);
945 		goto failed2;
946 
947 	}
948 
949 	mac_perim_exit(mph);
950 
951 	dlokack(q, mp, DL_SET_PHYS_ADDR_REQ);
952 	return;
953 
954 failed2:
955 	mac_perim_exit(mph);
956 failed:
957 	dlerrorack(q, mp, DL_SET_PHYS_ADDR_REQ, dl_err, (t_uscalar_t)err);
958 }
959 
960 /*
961  * DL_UDQOS_REQ
962  */
963 static void
964 proto_udqos_req(dld_str_t *dsp, mblk_t *mp)
965 {
966 	dl_udqos_req_t *dlp = (dl_udqos_req_t *)mp->b_rptr;
967 	dl_qos_cl_sel1_t *selp;
968 	int		off, len;
969 	t_uscalar_t	dl_err;
970 	queue_t		*q = dsp->ds_wq;
971 
972 	off = dlp->dl_qos_offset;
973 	len = dlp->dl_qos_length;
974 
975 	if (MBLKL(mp) < sizeof (dl_udqos_req_t) || !MBLKIN(mp, off, len)) {
976 		dl_err = DL_BADPRIM;
977 		goto failed;
978 	}
979 
980 	selp = (dl_qos_cl_sel1_t *)(mp->b_rptr + off);
981 	if (selp->dl_qos_type != DL_QOS_CL_SEL1) {
982 		dl_err = DL_BADQOSTYPE;
983 		goto failed;
984 	}
985 
986 	if (selp->dl_priority > (1 << VLAN_PRI_SIZE) - 1 ||
987 	    selp->dl_priority < 0) {
988 		dl_err = DL_BADQOSPARAM;
989 		goto failed;
990 	}
991 
992 	dsp->ds_pri = selp->dl_priority;
993 	dlokack(q, mp, DL_UDQOS_REQ);
994 	return;
995 failed:
996 	dlerrorack(q, mp, DL_UDQOS_REQ, dl_err, 0);
997 }
998 
999 static boolean_t
1000 check_mod_above(queue_t *q, const char *mod)
1001 {
1002 	queue_t		*next_q;
1003 	boolean_t	ret = B_TRUE;
1004 
1005 	claimstr(q);
1006 	next_q = q->q_next;
1007 	if (strcmp(next_q->q_qinfo->qi_minfo->mi_idname, mod) != 0)
1008 		ret = B_FALSE;
1009 	releasestr(q);
1010 	return (ret);
1011 }
1012 
1013 /*
1014  * DL_CAPABILITY_REQ
1015  */
1016 static void
1017 proto_capability_req(dld_str_t *dsp, mblk_t *mp)
1018 {
1019 	dl_capability_req_t *dlp = (dl_capability_req_t *)mp->b_rptr;
1020 	dl_capability_sub_t *sp;
1021 	size_t		size, len;
1022 	offset_t	off, end;
1023 	t_uscalar_t	dl_err;
1024 	queue_t		*q = dsp->ds_wq;
1025 
1026 	if (MBLKL(mp) < sizeof (dl_capability_req_t)) {
1027 		dl_err = DL_BADPRIM;
1028 		goto failed;
1029 	}
1030 
1031 	if (dsp->ds_dlstate == DL_UNATTACHED ||
1032 	    DL_ACK_PENDING(dsp->ds_dlstate)) {
1033 		dl_err = DL_OUTSTATE;
1034 		goto failed;
1035 	}
1036 
1037 	/*
1038 	 * This request is overloaded. If there are no requested capabilities
1039 	 * then we just want to acknowledge with all the capabilities we
1040 	 * support. Otherwise we enable the set of capabilities requested.
1041 	 */
1042 	if (dlp->dl_sub_length == 0) {
1043 		proto_capability_advertise(dsp, mp);
1044 		return;
1045 	}
1046 
1047 	if (!MBLKIN(mp, dlp->dl_sub_offset, dlp->dl_sub_length)) {
1048 		dl_err = DL_BADPRIM;
1049 		goto failed;
1050 	}
1051 
1052 	dlp->dl_primitive = DL_CAPABILITY_ACK;
1053 
1054 	off = dlp->dl_sub_offset;
1055 	len = dlp->dl_sub_length;
1056 
1057 	/*
1058 	 * Walk the list of capabilities to be enabled.
1059 	 */
1060 	for (end = off + len; off < end; ) {
1061 		sp = (dl_capability_sub_t *)(mp->b_rptr + off);
1062 		size = sizeof (dl_capability_sub_t) + sp->dl_length;
1063 
1064 		if (off + size > end ||
1065 		    !IS_P2ALIGNED(off, sizeof (uint32_t))) {
1066 			dl_err = DL_BADPRIM;
1067 			goto failed;
1068 		}
1069 
1070 		switch (sp->dl_cap) {
1071 		/*
1072 		 * TCP/IP checksum offload to hardware.
1073 		 */
1074 		case DL_CAPAB_HCKSUM: {
1075 			dl_capab_hcksum_t *hcksump;
1076 			dl_capab_hcksum_t hcksum;
1077 
1078 			hcksump = (dl_capab_hcksum_t *)&sp[1];
1079 			/*
1080 			 * Copy for alignment.
1081 			 */
1082 			bcopy(hcksump, &hcksum, sizeof (dl_capab_hcksum_t));
1083 			dlcapabsetqid(&(hcksum.hcksum_mid), dsp->ds_rq);
1084 			bcopy(&hcksum, hcksump, sizeof (dl_capab_hcksum_t));
1085 			break;
1086 		}
1087 
1088 		case DL_CAPAB_DLD: {
1089 			dl_capab_dld_t	*dldp;
1090 			dl_capab_dld_t	dld;
1091 
1092 			dldp = (dl_capab_dld_t *)&sp[1];
1093 			/*
1094 			 * Copy for alignment.
1095 			 */
1096 			bcopy(dldp, &dld, sizeof (dl_capab_dld_t));
1097 			dlcapabsetqid(&(dld.dld_mid), dsp->ds_rq);
1098 			bcopy(&dld, dldp, sizeof (dl_capab_dld_t));
1099 			break;
1100 		}
1101 		default:
1102 			break;
1103 		}
1104 		off += size;
1105 	}
1106 	qreply(q, mp);
1107 	return;
1108 failed:
1109 	dlerrorack(q, mp, DL_CAPABILITY_REQ, dl_err, 0);
1110 }
1111 
1112 /*
1113  * DL_NOTIFY_REQ
1114  */
1115 static void
1116 proto_notify_req(dld_str_t *dsp, mblk_t *mp)
1117 {
1118 	dl_notify_req_t	*dlp = (dl_notify_req_t *)mp->b_rptr;
1119 	t_uscalar_t	dl_err;
1120 	queue_t		*q = dsp->ds_wq;
1121 	uint_t		note =
1122 	    DL_NOTE_PROMISC_ON_PHYS |
1123 	    DL_NOTE_PROMISC_OFF_PHYS |
1124 	    DL_NOTE_PHYS_ADDR |
1125 	    DL_NOTE_LINK_UP |
1126 	    DL_NOTE_LINK_DOWN |
1127 	    DL_NOTE_CAPAB_RENEG |
1128 	    DL_NOTE_FASTPATH_FLUSH |
1129 	    DL_NOTE_SPEED |
1130 	    DL_NOTE_SDU_SIZE;
1131 
1132 	if (MBLKL(mp) < sizeof (dl_notify_req_t)) {
1133 		dl_err = DL_BADPRIM;
1134 		goto failed;
1135 	}
1136 
1137 	if (dsp->ds_dlstate == DL_UNATTACHED ||
1138 	    DL_ACK_PENDING(dsp->ds_dlstate)) {
1139 		dl_err = DL_OUTSTATE;
1140 		goto failed;
1141 	}
1142 
1143 	note &= ~(mac_no_notification(dsp->ds_mh));
1144 
1145 	/*
1146 	 * Cache the notifications that are being enabled.
1147 	 */
1148 	dsp->ds_notifications = dlp->dl_notifications & note;
1149 	/*
1150 	 * The ACK carries all notifications regardless of which set is
1151 	 * being enabled.
1152 	 */
1153 	dlnotifyack(q, mp, note);
1154 
1155 	/*
1156 	 * Generate DL_NOTIFY_IND messages for each enabled notification.
1157 	 */
1158 	if (dsp->ds_notifications != 0) {
1159 		dld_str_notify_ind(dsp);
1160 	}
1161 	return;
1162 failed:
1163 	dlerrorack(q, mp, DL_NOTIFY_REQ, dl_err, 0);
1164 }
1165 
1166 /*
1167  * DL_UINTDATA_REQ
1168  */
1169 void
1170 proto_unitdata_req(dld_str_t *dsp, mblk_t *mp)
1171 {
1172 	queue_t			*q = dsp->ds_wq;
1173 	dl_unitdata_req_t	*dlp = (dl_unitdata_req_t *)mp->b_rptr;
1174 	off_t			off;
1175 	size_t			len, size;
1176 	const uint8_t		*addr;
1177 	uint16_t		sap;
1178 	uint_t			addr_length;
1179 	mblk_t			*bp, *payload;
1180 	uint32_t		start, stuff, end, value, flags;
1181 	t_uscalar_t		dl_err;
1182 	uint_t			max_sdu;
1183 
1184 	if (MBLKL(mp) < sizeof (dl_unitdata_req_t) || mp->b_cont == NULL) {
1185 		dlerrorack(q, mp, DL_UNITDATA_REQ, DL_BADPRIM, 0);
1186 		return;
1187 	}
1188 
1189 	mutex_enter(&dsp->ds_lock);
1190 	if (dsp->ds_dlstate != DL_IDLE) {
1191 		mutex_exit(&dsp->ds_lock);
1192 		dlerrorack(q, mp, DL_UNITDATA_REQ, DL_OUTSTATE, 0);
1193 		return;
1194 	}
1195 	DLD_DATATHR_INC(dsp);
1196 	mutex_exit(&dsp->ds_lock);
1197 
1198 	addr_length = dsp->ds_mip->mi_addr_length;
1199 
1200 	off = dlp->dl_dest_addr_offset;
1201 	len = dlp->dl_dest_addr_length;
1202 
1203 	if (!MBLKIN(mp, off, len) || !IS_P2ALIGNED(off, sizeof (uint16_t))) {
1204 		dl_err = DL_BADPRIM;
1205 		goto failed;
1206 	}
1207 
1208 	if (len != addr_length + sizeof (uint16_t)) {
1209 		dl_err = DL_BADADDR;
1210 		goto failed;
1211 	}
1212 
1213 	addr = mp->b_rptr + off;
1214 	sap = *(uint16_t *)(mp->b_rptr + off + addr_length);
1215 
1216 	/*
1217 	 * Check the length of the packet and the block types.
1218 	 */
1219 	size = 0;
1220 	payload = mp->b_cont;
1221 	for (bp = payload; bp != NULL; bp = bp->b_cont) {
1222 		if (DB_TYPE(bp) != M_DATA)
1223 			goto baddata;
1224 
1225 		size += MBLKL(bp);
1226 	}
1227 
1228 	mac_sdu_get(dsp->ds_mh, NULL, &max_sdu);
1229 	if (size > max_sdu)
1230 		goto baddata;
1231 
1232 	/*
1233 	 * Build a packet header.
1234 	 */
1235 	if ((bp = dls_header(dsp, addr, sap, dlp->dl_priority.dl_max,
1236 	    &payload)) == NULL) {
1237 		dl_err = DL_BADADDR;
1238 		goto failed;
1239 	}
1240 
1241 	/*
1242 	 * We no longer need the M_PROTO header, so free it.
1243 	 */
1244 	freeb(mp);
1245 
1246 	/*
1247 	 * Transfer the checksum offload information if it is present.
1248 	 */
1249 	hcksum_retrieve(payload, NULL, NULL, &start, &stuff, &end, &value,
1250 	    &flags);
1251 	(void) hcksum_assoc(bp, NULL, NULL, start, stuff, end, value, flags, 0);
1252 
1253 	/*
1254 	 * Link the payload onto the new header.
1255 	 */
1256 	ASSERT(bp->b_cont == NULL);
1257 	bp->b_cont = payload;
1258 
1259 	/*
1260 	 * No lock can be held across modules and putnext()'s,
1261 	 * which can happen here with the call from DLD_TX().
1262 	 */
1263 	if (DLD_TX(dsp, bp, 0, 0) != NULL) {
1264 		/* flow-controlled */
1265 		DLD_SETQFULL(dsp);
1266 	}
1267 	DLD_DATATHR_DCR(dsp);
1268 	return;
1269 
1270 failed:
1271 	dlerrorack(q, mp, DL_UNITDATA_REQ, dl_err, 0);
1272 	DLD_DATATHR_DCR(dsp);
1273 	return;
1274 
1275 baddata:
1276 	dluderrorind(q, mp, (void *)addr, len, DL_BADDATA, 0);
1277 	DLD_DATATHR_DCR(dsp);
1278 }
1279 
1280 /*
1281  * DL_PASSIVE_REQ
1282  */
1283 static void
1284 proto_passive_req(dld_str_t *dsp, mblk_t *mp)
1285 {
1286 	t_uscalar_t dl_err;
1287 
1288 	/*
1289 	 * If we've already become active by issuing an active primitive,
1290 	 * then it's too late to try to become passive.
1291 	 */
1292 	if (dsp->ds_passivestate == DLD_ACTIVE) {
1293 		dl_err = DL_OUTSTATE;
1294 		goto failed;
1295 	}
1296 
1297 	if (MBLKL(mp) < sizeof (dl_passive_req_t)) {
1298 		dl_err = DL_BADPRIM;
1299 		goto failed;
1300 	}
1301 
1302 	dsp->ds_passivestate = DLD_PASSIVE;
1303 	dlokack(dsp->ds_wq, mp, DL_PASSIVE_REQ);
1304 	return;
1305 failed:
1306 	dlerrorack(dsp->ds_wq, mp, DL_PASSIVE_REQ, dl_err, 0);
1307 }
1308 
1309 
1310 /*
1311  * Catch-all handler.
1312  */
1313 static void
1314 proto_req(dld_str_t *dsp, mblk_t *mp)
1315 {
1316 	union DL_primitives	*dlp = (union DL_primitives *)mp->b_rptr;
1317 
1318 	dlerrorack(dsp->ds_wq, mp, dlp->dl_primitive, DL_UNSUPPORTED, 0);
1319 }
1320 
1321 static int
1322 dld_capab_perim(dld_str_t *dsp, void *data, uint_t flags)
1323 {
1324 	switch (flags) {
1325 	case DLD_ENABLE:
1326 		mac_perim_enter_by_mh(dsp->ds_mh, (mac_perim_handle_t *)data);
1327 		return (0);
1328 
1329 	case DLD_DISABLE:
1330 		mac_perim_exit((mac_perim_handle_t)data);
1331 		return (0);
1332 
1333 	case DLD_QUERY:
1334 		return (mac_perim_held(dsp->ds_mh));
1335 	}
1336 	return (0);
1337 }
1338 
1339 static int
1340 dld_capab_direct(dld_str_t *dsp, void *data, uint_t flags)
1341 {
1342 	dld_capab_direct_t	*direct = data;
1343 
1344 	ASSERT(MAC_PERIM_HELD(dsp->ds_mh));
1345 
1346 	switch (flags) {
1347 	case DLD_ENABLE:
1348 		dls_rx_set(dsp, (dls_rx_t)direct->di_rx_cf,
1349 		    direct->di_rx_ch);
1350 
1351 		direct->di_tx_df = (uintptr_t)str_mdata_fastpath_put;
1352 		direct->di_tx_dh = dsp;
1353 		direct->di_tx_cb_df = (uintptr_t)mac_client_tx_notify;
1354 		direct->di_tx_cb_dh = dsp->ds_mch;
1355 		direct->di_tx_fctl_df = (uintptr_t)mac_tx_is_flow_blocked;
1356 		direct->di_tx_fctl_dh = dsp->ds_mch;
1357 
1358 		dsp->ds_direct = B_TRUE;
1359 
1360 		return (0);
1361 
1362 	case DLD_DISABLE:
1363 		dls_rx_set(dsp, (dsp->ds_mode == DLD_FASTPATH) ?
1364 		    dld_str_rx_fastpath : dld_str_rx_unitdata, (void *)dsp);
1365 		dsp->ds_direct = B_FALSE;
1366 
1367 		return (0);
1368 	}
1369 	return (ENOTSUP);
1370 }
1371 
1372 /*
1373  * dld_capab_poll_enable()
1374  *
1375  * This function is misnamed. All polling  and fanouts are run out of the
1376  * lower mac (in case of VNIC and the only mac in case of NICs). The
1377  * availability of Rx ring and promiscous mode is all taken care between
1378  * the soft ring set (mac_srs), the Rx ring, and S/W classifier. Any
1379  * fanout necessary is done by the soft rings that are part of the
1380  * mac_srs (by default mac_srs sends the packets up via a TCP and
1381  * non TCP soft ring).
1382  *
1383  * The mac_srs (or its associated soft rings) always store the ill_rx_ring
1384  * (the cookie returned when they registered with IP during plumb) as their
1385  * 2nd argument which is passed up as mac_resource_handle_t. The upcall
1386  * function and 1st argument is what the caller registered when they
1387  * called mac_rx_classify_flow_add() to register the flow. For VNIC,
1388  * the function is vnic_rx and argument is vnic_t. For regular NIC
1389  * case, it mac_rx_default and mac_handle_t. As explained above, the
1390  * mac_srs (or its soft ring) will add the ill_rx_ring (mac_resource_handle_t)
1391  * from its stored 2nd argument.
1392  */
1393 static int
1394 dld_capab_poll_enable(dld_str_t *dsp, dld_capab_poll_t *poll)
1395 {
1396 	if (dsp->ds_polling)
1397 		return (EINVAL);
1398 
1399 	if ((dld_opt & DLD_OPT_NO_POLL) != 0 || dsp->ds_mode == DLD_RAW)
1400 		return (ENOTSUP);
1401 
1402 	/*
1403 	 * Enable client polling if and only if DLS bypass is possible.
1404 	 * Special cases like VLANs need DLS processing in the Rx data path.
1405 	 * In such a case we can neither allow the client (IP) to directly
1406 	 * poll the softring (since DLS processing hasn't been done) nor can
1407 	 * we allow DLS bypass.
1408 	 */
1409 	if (!mac_rx_bypass_set(dsp->ds_mch, dsp->ds_rx, dsp->ds_rx_arg))
1410 		return (ENOTSUP);
1411 
1412 	/*
1413 	 * Register soft ring resources. This will come in handy later if
1414 	 * the user decides to modify CPU bindings to use more CPUs for the
1415 	 * device in which case we will switch to fanout using soft rings.
1416 	 */
1417 	mac_resource_set_common(dsp->ds_mch,
1418 	    (mac_resource_add_t)poll->poll_ring_add_cf,
1419 	    (mac_resource_remove_t)poll->poll_ring_remove_cf,
1420 	    (mac_resource_quiesce_t)poll->poll_ring_quiesce_cf,
1421 	    (mac_resource_restart_t)poll->poll_ring_restart_cf,
1422 	    (mac_resource_bind_t)poll->poll_ring_bind_cf,
1423 	    poll->poll_ring_ch);
1424 
1425 	mac_client_poll_enable(dsp->ds_mch);
1426 
1427 	dsp->ds_polling = B_TRUE;
1428 	return (0);
1429 }
1430 
1431 /* ARGSUSED */
1432 static int
1433 dld_capab_poll_disable(dld_str_t *dsp, dld_capab_poll_t *poll)
1434 {
1435 	if (!dsp->ds_polling)
1436 		return (EINVAL);
1437 
1438 	mac_client_poll_disable(dsp->ds_mch);
1439 	mac_resource_set(dsp->ds_mch, NULL, NULL);
1440 
1441 	dsp->ds_polling = B_FALSE;
1442 	return (0);
1443 }
1444 
1445 static int
1446 dld_capab_poll(dld_str_t *dsp, void *data, uint_t flags)
1447 {
1448 	dld_capab_poll_t	*poll = data;
1449 
1450 	ASSERT(MAC_PERIM_HELD(dsp->ds_mh));
1451 
1452 	switch (flags) {
1453 	case DLD_ENABLE:
1454 		return (dld_capab_poll_enable(dsp, poll));
1455 	case DLD_DISABLE:
1456 		return (dld_capab_poll_disable(dsp, poll));
1457 	}
1458 	return (ENOTSUP);
1459 }
1460 
1461 static int
1462 dld_capab_lso(dld_str_t *dsp, void *data, uint_t flags)
1463 {
1464 	dld_capab_lso_t		*lso = data;
1465 
1466 	ASSERT(MAC_PERIM_HELD(dsp->ds_mh));
1467 
1468 	switch (flags) {
1469 	case DLD_ENABLE: {
1470 		mac_capab_lso_t		mac_lso;
1471 
1472 		/*
1473 		 * Check if LSO is supported on this MAC & enable LSO
1474 		 * accordingly.
1475 		 */
1476 		if (mac_capab_get(dsp->ds_mh, MAC_CAPAB_LSO, &mac_lso)) {
1477 			lso->lso_max = mac_lso.lso_basic_tcp_ipv4.lso_max;
1478 			lso->lso_flags = 0;
1479 			/* translate the flag for mac clients */
1480 			if ((mac_lso.lso_flags & LSO_TX_BASIC_TCP_IPV4) != 0)
1481 				lso->lso_flags |= DLD_LSO_BASIC_TCP_IPV4;
1482 			dsp->ds_lso = B_TRUE;
1483 			dsp->ds_lso_max = lso->lso_max;
1484 		} else {
1485 			dsp->ds_lso = B_FALSE;
1486 			dsp->ds_lso_max = 0;
1487 			return (ENOTSUP);
1488 		}
1489 		return (0);
1490 	}
1491 	case DLD_DISABLE: {
1492 		dsp->ds_lso = B_FALSE;
1493 		dsp->ds_lso_max = 0;
1494 		return (0);
1495 	}
1496 	}
1497 	return (ENOTSUP);
1498 }
1499 
1500 static int
1501 dld_capab(dld_str_t *dsp, uint_t type, void *data, uint_t flags)
1502 {
1503 	int	err;
1504 
1505 	/*
1506 	 * Don't enable direct callback capabilities unless the caller is
1507 	 * the IP client. When a module is inserted in a stream (_I_INSERT)
1508 	 * the stack initiates capability disable, but due to races, the
1509 	 * module insertion may complete before the capability disable
1510 	 * completes. So we limit the check to DLD_ENABLE case.
1511 	 */
1512 	if ((flags == DLD_ENABLE && type != DLD_CAPAB_PERIM) &&
1513 	    (dsp->ds_sap != ETHERTYPE_IP ||
1514 	    !check_mod_above(dsp->ds_rq, "ip"))) {
1515 		return (ENOTSUP);
1516 	}
1517 
1518 	switch (type) {
1519 	case DLD_CAPAB_DIRECT:
1520 		err = dld_capab_direct(dsp, data, flags);
1521 		break;
1522 
1523 	case DLD_CAPAB_POLL:
1524 		err =  dld_capab_poll(dsp, data, flags);
1525 		break;
1526 
1527 	case DLD_CAPAB_PERIM:
1528 		err = dld_capab_perim(dsp, data, flags);
1529 		break;
1530 
1531 	case DLD_CAPAB_LSO:
1532 		err = dld_capab_lso(dsp, data, flags);
1533 		break;
1534 
1535 	default:
1536 		err = ENOTSUP;
1537 		break;
1538 	}
1539 
1540 	return (err);
1541 }
1542 
1543 /*
1544  * DL_CAPABILITY_ACK/DL_ERROR_ACK
1545  */
1546 static void
1547 proto_capability_advertise(dld_str_t *dsp, mblk_t *mp)
1548 {
1549 	dl_capability_ack_t	*dlap;
1550 	dl_capability_sub_t	*dlsp;
1551 	size_t			subsize;
1552 	dl_capab_dld_t		dld;
1553 	dl_capab_hcksum_t	hcksum;
1554 	dl_capab_zerocopy_t	zcopy;
1555 	uint8_t			*ptr;
1556 	queue_t			*q = dsp->ds_wq;
1557 	mblk_t			*mp1;
1558 	boolean_t		is_vlan;
1559 	boolean_t		hcksum_capable = B_FALSE;
1560 	boolean_t		zcopy_capable = B_FALSE;
1561 	boolean_t		dld_capable = B_FALSE;
1562 
1563 	/*
1564 	 * Initially assume no capabilities.
1565 	 */
1566 	subsize = 0;
1567 	is_vlan = (mac_client_vid(dsp->ds_mch) != VLAN_ID_NONE);
1568 
1569 	/*
1570 	 * Check if checksum offload is supported on this MAC.  Don't
1571 	 * advertise DL_CAPAB_HCKSUM if the underlying MAC is VLAN incapable,
1572 	 * since it might not be able to do the hardware checksum offload
1573 	 * with the correct offset.
1574 	 */
1575 	bzero(&hcksum, sizeof (dl_capab_hcksum_t));
1576 	if ((!is_vlan || (!mac_capab_get(dsp->ds_mh, MAC_CAPAB_NO_NATIVEVLAN,
1577 	    NULL))) && mac_capab_get(dsp->ds_mh, MAC_CAPAB_HCKSUM,
1578 	    &hcksum.hcksum_txflags)) {
1579 		if (hcksum.hcksum_txflags != 0) {
1580 			hcksum_capable = B_TRUE;
1581 			subsize += sizeof (dl_capability_sub_t) +
1582 			    sizeof (dl_capab_hcksum_t);
1583 		}
1584 	}
1585 
1586 	/*
1587 	 * Check if zerocopy is supported on this interface.
1588 	 * If advertising DL_CAPAB_ZEROCOPY has not been explicitly disabled
1589 	 * then reserve space for that capability.
1590 	 */
1591 	if (!mac_capab_get(dsp->ds_mh, MAC_CAPAB_NO_ZCOPY, NULL) &&
1592 	    !(dld_opt & DLD_OPT_NO_ZEROCOPY)) {
1593 		zcopy_capable = B_TRUE;
1594 		subsize += sizeof (dl_capability_sub_t) +
1595 		    sizeof (dl_capab_zerocopy_t);
1596 	}
1597 
1598 	/*
1599 	 * Direct capability negotiation interface between IP and DLD
1600 	 */
1601 	if (dsp->ds_sap == ETHERTYPE_IP && check_mod_above(dsp->ds_rq, "ip")) {
1602 		dld_capable = B_TRUE;
1603 		subsize += sizeof (dl_capability_sub_t) +
1604 		    sizeof (dl_capab_dld_t);
1605 	}
1606 
1607 	/*
1608 	 * If there are no capabilities to advertise or if we
1609 	 * can't allocate a response, send a DL_ERROR_ACK.
1610 	 */
1611 	if ((mp1 = reallocb(mp,
1612 	    sizeof (dl_capability_ack_t) + subsize, 0)) == NULL) {
1613 		dlerrorack(q, mp, DL_CAPABILITY_REQ, DL_NOTSUPPORTED, 0);
1614 		return;
1615 	}
1616 
1617 	mp = mp1;
1618 	DB_TYPE(mp) = M_PROTO;
1619 	mp->b_wptr = mp->b_rptr + sizeof (dl_capability_ack_t) + subsize;
1620 	bzero(mp->b_rptr, MBLKL(mp));
1621 	dlap = (dl_capability_ack_t *)mp->b_rptr;
1622 	dlap->dl_primitive = DL_CAPABILITY_ACK;
1623 	dlap->dl_sub_offset = sizeof (dl_capability_ack_t);
1624 	dlap->dl_sub_length = subsize;
1625 	ptr = (uint8_t *)&dlap[1];
1626 
1627 	/*
1628 	 * TCP/IP checksum offload.
1629 	 */
1630 	if (hcksum_capable) {
1631 		dlsp = (dl_capability_sub_t *)ptr;
1632 
1633 		dlsp->dl_cap = DL_CAPAB_HCKSUM;
1634 		dlsp->dl_length = sizeof (dl_capab_hcksum_t);
1635 		ptr += sizeof (dl_capability_sub_t);
1636 
1637 		hcksum.hcksum_version = HCKSUM_VERSION_1;
1638 		dlcapabsetqid(&(hcksum.hcksum_mid), dsp->ds_rq);
1639 		bcopy(&hcksum, ptr, sizeof (dl_capab_hcksum_t));
1640 		ptr += sizeof (dl_capab_hcksum_t);
1641 	}
1642 
1643 	/*
1644 	 * Zero copy
1645 	 */
1646 	if (zcopy_capable) {
1647 		dlsp = (dl_capability_sub_t *)ptr;
1648 
1649 		dlsp->dl_cap = DL_CAPAB_ZEROCOPY;
1650 		dlsp->dl_length = sizeof (dl_capab_zerocopy_t);
1651 		ptr += sizeof (dl_capability_sub_t);
1652 
1653 		bzero(&zcopy, sizeof (dl_capab_zerocopy_t));
1654 		zcopy.zerocopy_version = ZEROCOPY_VERSION_1;
1655 		zcopy.zerocopy_flags = DL_CAPAB_VMSAFE_MEM;
1656 
1657 		dlcapabsetqid(&(zcopy.zerocopy_mid), dsp->ds_rq);
1658 		bcopy(&zcopy, ptr, sizeof (dl_capab_zerocopy_t));
1659 		ptr += sizeof (dl_capab_zerocopy_t);
1660 	}
1661 
1662 	/*
1663 	 * Direct capability negotiation interface between IP and DLD.
1664 	 * Refer to dld.h for details.
1665 	 */
1666 	if (dld_capable) {
1667 		dlsp = (dl_capability_sub_t *)ptr;
1668 		dlsp->dl_cap = DL_CAPAB_DLD;
1669 		dlsp->dl_length = sizeof (dl_capab_dld_t);
1670 		ptr += sizeof (dl_capability_sub_t);
1671 
1672 		bzero(&dld, sizeof (dl_capab_dld_t));
1673 		dld.dld_version = DLD_CURRENT_VERSION;
1674 		dld.dld_capab = (uintptr_t)dld_capab;
1675 		dld.dld_capab_handle = (uintptr_t)dsp;
1676 
1677 		dlcapabsetqid(&(dld.dld_mid), dsp->ds_rq);
1678 		bcopy(&dld, ptr, sizeof (dl_capab_dld_t));
1679 		ptr += sizeof (dl_capab_dld_t);
1680 	}
1681 
1682 	ASSERT(ptr == mp->b_rptr + sizeof (dl_capability_ack_t) + subsize);
1683 	qreply(q, mp);
1684 }
1685 
1686 /*
1687  * Disable any enabled capabilities.
1688  */
1689 void
1690 dld_capabilities_disable(dld_str_t *dsp)
1691 {
1692 	if (dsp->ds_polling)
1693 		(void) dld_capab_poll_disable(dsp, NULL);
1694 }
1695