xref: /illumos-gate/usr/src/uts/common/io/mac/mac_util.c (revision ae6aa22afeb444ae208c287e7227a4a7c877f17a)
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  * MAC Services Module - misc utilities
28  */
29 
30 #include <sys/types.h>
31 #include <sys/mac.h>
32 #include <sys/mac_impl.h>
33 #include <sys/mac_client_priv.h>
34 #include <sys/mac_client_impl.h>
35 #include <sys/mac_soft_ring.h>
36 #include <sys/strsubr.h>
37 #include <sys/strsun.h>
38 #include <sys/vlan.h>
39 #include <sys/pattr.h>
40 #include <sys/pci_tools.h>
41 #include <inet/ip.h>
42 #include <inet/ip_impl.h>
43 #include <inet/ip6.h>
44 #include <sys/vtrace.h>
45 #include <sys/dlpi.h>
46 #include <sys/sunndi.h>
47 #include <inet/ipsec_impl.h>
48 #include <inet/sadb.h>
49 #include <inet/ipsecesp.h>
50 #include <inet/ipsecah.h>
51 
52 /*
53  * Copy an mblk, preserving its hardware checksum flags.
54  */
55 static mblk_t *
56 mac_copymsg_cksum(mblk_t *mp)
57 {
58 	mblk_t *mp1;
59 	uint32_t start, stuff, end, value, flags;
60 
61 	mp1 = copymsg(mp);
62 	if (mp1 == NULL)
63 		return (NULL);
64 
65 	hcksum_retrieve(mp, NULL, NULL, &start, &stuff, &end, &value, &flags);
66 	(void) hcksum_assoc(mp1, NULL, NULL, start, stuff, end, value,
67 	    flags, KM_NOSLEEP);
68 
69 	return (mp1);
70 }
71 
72 /*
73  * Copy an mblk chain, presenting the hardware checksum flags of the
74  * individual mblks.
75  */
76 mblk_t *
77 mac_copymsgchain_cksum(mblk_t *mp)
78 {
79 	mblk_t *nmp = NULL;
80 	mblk_t **nmpp = &nmp;
81 
82 	for (; mp != NULL; mp = mp->b_next) {
83 		if ((*nmpp = mac_copymsg_cksum(mp)) == NULL) {
84 			freemsgchain(nmp);
85 			return (NULL);
86 		}
87 
88 		nmpp = &((*nmpp)->b_next);
89 	}
90 
91 	return (nmp);
92 }
93 
94 /*
95  * Process the specified mblk chain for proper handling of hardware
96  * checksum offload. This routine is invoked for loopback traffic
97  * between MAC clients.
98  * The function handles a NULL mblk chain passed as argument.
99  */
100 mblk_t *
101 mac_fix_cksum(mblk_t *mp_chain)
102 {
103 	mblk_t *mp, *prev = NULL, *new_chain = mp_chain, *mp1;
104 	uint32_t flags, start, stuff, end, value;
105 
106 	for (mp = mp_chain; mp != NULL; prev = mp, mp = mp->b_next) {
107 		uint16_t len;
108 		uint32_t offset;
109 		struct ether_header *ehp;
110 		uint16_t sap;
111 
112 		hcksum_retrieve(mp, NULL, NULL, &start, &stuff, &end, &value,
113 		    &flags);
114 		if (flags == 0)
115 			continue;
116 
117 		/*
118 		 * Since the processing of checksum offload for loopback
119 		 * traffic requires modification of the packet contents,
120 		 * ensure sure that we are always modifying our own copy.
121 		 */
122 		if (DB_REF(mp) > 1) {
123 			mp1 = copymsg(mp);
124 			if (mp1 == NULL)
125 				continue;
126 			mp1->b_next = mp->b_next;
127 			mp->b_next = NULL;
128 			freemsg(mp);
129 			if (prev != NULL)
130 				prev->b_next = mp1;
131 			else
132 				new_chain = mp1;
133 			mp = mp1;
134 		}
135 
136 		/*
137 		 * Ethernet, and optionally VLAN header.
138 		 */
139 		/* LINTED: improper alignment cast */
140 		ehp = (struct ether_header *)mp->b_rptr;
141 		if (ntohs(ehp->ether_type) == VLAN_TPID) {
142 			struct ether_vlan_header *evhp;
143 
144 			ASSERT(MBLKL(mp) >= sizeof (struct ether_vlan_header));
145 			/* LINTED: improper alignment cast */
146 			evhp = (struct ether_vlan_header *)mp->b_rptr;
147 			sap = ntohs(evhp->ether_type);
148 			offset = sizeof (struct ether_vlan_header);
149 		} else {
150 			sap = ntohs(ehp->ether_type);
151 			offset = sizeof (struct ether_header);
152 		}
153 
154 		if (MBLKL(mp) <= offset) {
155 			offset -= MBLKL(mp);
156 			if (mp->b_cont == NULL) {
157 				/* corrupted packet, skip it */
158 				if (prev != NULL)
159 					prev->b_next = mp->b_next;
160 				else
161 					new_chain = mp->b_next;
162 				mp1 = mp->b_next;
163 				mp->b_next = NULL;
164 				freemsg(mp);
165 				mp = mp1;
166 				continue;
167 			}
168 			mp = mp->b_cont;
169 		}
170 
171 		if (flags & (HCK_FULLCKSUM | HCK_IPV4_HDRCKSUM)) {
172 			ipha_t *ipha = NULL;
173 
174 			/*
175 			 * In order to compute the full and header
176 			 * checksums, we need to find and parse
177 			 * the IP and/or ULP headers.
178 			 */
179 
180 			sap = (sap < ETHERTYPE_802_MIN) ? 0 : sap;
181 
182 			/*
183 			 * IP header.
184 			 */
185 			if (sap != ETHERTYPE_IP)
186 				continue;
187 
188 			ASSERT(MBLKL(mp) >= offset + sizeof (ipha_t));
189 			/* LINTED: improper alignment cast */
190 			ipha = (ipha_t *)(mp->b_rptr + offset);
191 
192 			if (flags & HCK_FULLCKSUM) {
193 				ipaddr_t src, dst;
194 				uint32_t cksum;
195 				uint16_t *up;
196 				uint8_t proto;
197 
198 				/*
199 				 * Pointer to checksum field in ULP header.
200 				 */
201 				proto = ipha->ipha_protocol;
202 				ASSERT(ipha->ipha_version_and_hdr_length ==
203 				    IP_SIMPLE_HDR_VERSION);
204 				if (proto == IPPROTO_TCP) {
205 					/* LINTED: improper alignment cast */
206 					up = IPH_TCPH_CHECKSUMP(ipha,
207 					    IP_SIMPLE_HDR_LENGTH);
208 				} else {
209 					ASSERT(proto == IPPROTO_UDP);
210 					/* LINTED: improper alignment cast */
211 					up = IPH_UDPH_CHECKSUMP(ipha,
212 					    IP_SIMPLE_HDR_LENGTH);
213 				}
214 
215 				/*
216 				 * Pseudo-header checksum.
217 				 */
218 				src = ipha->ipha_src;
219 				dst = ipha->ipha_dst;
220 				len = ntohs(ipha->ipha_length) -
221 				    IP_SIMPLE_HDR_LENGTH;
222 
223 				cksum = (dst >> 16) + (dst & 0xFFFF) +
224 				    (src >> 16) + (src & 0xFFFF);
225 				cksum += htons(len);
226 
227 				/*
228 				 * The checksum value stored in the packet needs
229 				 * to be correct. Compute it here.
230 				 */
231 				*up = 0;
232 				cksum += (((proto) == IPPROTO_UDP) ?
233 				    IP_UDP_CSUM_COMP : IP_TCP_CSUM_COMP);
234 				cksum = IP_CSUM(mp, IP_SIMPLE_HDR_LENGTH +
235 				    offset, cksum);
236 				*(up) = (uint16_t)(cksum ? cksum : ~cksum);
237 
238 				flags |= HCK_FULLCKSUM_OK;
239 				value = 0xffff;
240 			}
241 
242 			if (flags & HCK_IPV4_HDRCKSUM) {
243 				ASSERT(ipha != NULL);
244 				ipha->ipha_hdr_checksum =
245 				    (uint16_t)ip_csum_hdr(ipha);
246 			}
247 		}
248 
249 		if (flags & HCK_PARTIALCKSUM) {
250 			uint16_t *up, partial, cksum;
251 			uchar_t *ipp; /* ptr to beginning of IP header */
252 
253 			if (mp->b_cont != NULL) {
254 				mblk_t *mp1;
255 
256 				mp1 = msgpullup(mp, offset + end);
257 				if (mp1 == NULL)
258 					continue;
259 				mp1->b_next = mp->b_next;
260 				mp->b_next = NULL;
261 				freemsg(mp);
262 				if (prev != NULL)
263 					prev->b_next = mp1;
264 				else
265 					new_chain = mp1;
266 				mp = mp1;
267 			}
268 
269 			ipp = mp->b_rptr + offset;
270 			/* LINTED: cast may result in improper alignment */
271 			up = (uint16_t *)((uchar_t *)ipp + stuff);
272 			partial = *up;
273 			*up = 0;
274 
275 			cksum = IP_BCSUM_PARTIAL(mp->b_rptr + offset + start,
276 			    end - start, partial);
277 			cksum = ~cksum;
278 			*up = cksum ? cksum : ~cksum;
279 
280 			/*
281 			 * Since we already computed the whole checksum,
282 			 * indicate to the stack that it has already
283 			 * been verified by the hardware.
284 			 */
285 			flags &= ~HCK_PARTIALCKSUM;
286 			flags |= (HCK_FULLCKSUM | HCK_FULLCKSUM_OK);
287 			value = 0xffff;
288 		}
289 
290 		(void) hcksum_assoc(mp, NULL, NULL, start, stuff, end,
291 		    value, flags, KM_NOSLEEP);
292 	}
293 
294 	return (new_chain);
295 }
296 
297 /*
298  * Add VLAN tag to the specified mblk.
299  */
300 mblk_t *
301 mac_add_vlan_tag(mblk_t *mp, uint_t pri, uint16_t vid)
302 {
303 	mblk_t *hmp;
304 	struct ether_vlan_header *evhp;
305 	struct ether_header *ehp;
306 	uint32_t start, stuff, end, value, flags;
307 
308 	ASSERT(pri != 0 || vid != 0);
309 
310 	/*
311 	 * Allocate an mblk for the new tagged ethernet header,
312 	 * and copy the MAC addresses and ethertype from the
313 	 * original header.
314 	 */
315 
316 	hmp = allocb(sizeof (struct ether_vlan_header), BPRI_MED);
317 	if (hmp == NULL) {
318 		freemsg(mp);
319 		return (NULL);
320 	}
321 
322 	evhp = (struct ether_vlan_header *)hmp->b_rptr;
323 	ehp = (struct ether_header *)mp->b_rptr;
324 
325 	bcopy(ehp, evhp, (ETHERADDRL * 2));
326 	evhp->ether_type = ehp->ether_type;
327 	evhp->ether_tpid = htons(ETHERTYPE_VLAN);
328 
329 	hmp->b_wptr += sizeof (struct ether_vlan_header);
330 	mp->b_rptr += sizeof (struct ether_header);
331 
332 	/*
333 	 * Free the original message if it's now empty. Link the
334 	 * rest of messages to the header message.
335 	 */
336 	hcksum_retrieve(mp, NULL, NULL, &start, &stuff, &end, &value, &flags);
337 	(void) hcksum_assoc(hmp, NULL, NULL, start, stuff, end, value, flags,
338 	    KM_NOSLEEP);
339 	if (MBLKL(mp) == 0) {
340 		hmp->b_cont = mp->b_cont;
341 		freeb(mp);
342 	} else {
343 		hmp->b_cont = mp;
344 	}
345 	ASSERT(MBLKL(hmp) >= sizeof (struct ether_vlan_header));
346 
347 	/*
348 	 * Initialize the new TCI (Tag Control Information).
349 	 */
350 	evhp->ether_tci = htons(VLAN_TCI(pri, 0, vid));
351 
352 	return (hmp);
353 }
354 
355 /*
356  * Adds a VLAN tag with the specified VID and priority to each mblk of
357  * the specified chain.
358  */
359 mblk_t *
360 mac_add_vlan_tag_chain(mblk_t *mp_chain, uint_t pri, uint16_t vid)
361 {
362 	mblk_t *next_mp, **prev, *mp;
363 
364 	mp = mp_chain;
365 	prev = &mp_chain;
366 
367 	while (mp != NULL) {
368 		next_mp = mp->b_next;
369 		mp->b_next = NULL;
370 		if ((mp = mac_add_vlan_tag(mp, pri, vid)) == NULL) {
371 			freemsgchain(next_mp);
372 			break;
373 		}
374 		*prev = mp;
375 		prev = &mp->b_next;
376 		mp = mp->b_next = next_mp;
377 	}
378 
379 	return (mp_chain);
380 }
381 
382 /*
383  * Strip VLAN tag
384  */
385 mblk_t *
386 mac_strip_vlan_tag(mblk_t *mp)
387 {
388 	mblk_t *newmp;
389 	struct ether_vlan_header *evhp;
390 
391 	evhp = (struct ether_vlan_header *)mp->b_rptr;
392 	if (ntohs(evhp->ether_tpid) == ETHERTYPE_VLAN) {
393 		ASSERT(MBLKL(mp) >= sizeof (struct ether_vlan_header));
394 
395 		if (DB_REF(mp) > 1) {
396 			newmp = copymsg(mp);
397 			if (newmp == NULL)
398 				return (NULL);
399 			freemsg(mp);
400 			mp = newmp;
401 		}
402 
403 		evhp = (struct ether_vlan_header *)mp->b_rptr;
404 
405 		ovbcopy(mp->b_rptr, mp->b_rptr + VLAN_TAGSZ, 2 * ETHERADDRL);
406 		mp->b_rptr += VLAN_TAGSZ;
407 	}
408 	return (mp);
409 }
410 
411 /*
412  * Strip VLAN tag from each mblk of the chain.
413  */
414 mblk_t *
415 mac_strip_vlan_tag_chain(mblk_t *mp_chain)
416 {
417 	mblk_t *mp, *next_mp, **prev;
418 
419 	mp = mp_chain;
420 	prev = &mp_chain;
421 
422 	while (mp != NULL) {
423 		next_mp = mp->b_next;
424 		mp->b_next = NULL;
425 		if ((mp = mac_strip_vlan_tag(mp)) == NULL) {
426 			freemsgchain(next_mp);
427 			break;
428 		}
429 		*prev = mp;
430 		prev = &mp->b_next;
431 		mp = mp->b_next = next_mp;
432 	}
433 
434 	return (mp_chain);
435 }
436 
437 /*
438  * Default callback function. Used when the datapath is not yet initialized.
439  */
440 /* ARGSUSED */
441 void
442 mac_pkt_drop(void *arg, mac_resource_handle_t resource, mblk_t *mp,
443     boolean_t loopback)
444 {
445 	mblk_t	*mp1 = mp;
446 
447 	while (mp1 != NULL) {
448 		mp1->b_prev = NULL;
449 		mp1->b_queue = NULL;
450 		mp1 = mp1->b_next;
451 	}
452 	freemsgchain(mp);
453 }
454 
455 /*
456  * Determines the IPv6 header length accounting for all the optional IPv6
457  * headers (hop-by-hop, destination, routing and fragment). The header length
458  * and next header value (a transport header) is captured.
459  *
460  * Returns B_FALSE if all the IP headers are not in the same mblk otherwise
461  * returns B_TRUE.
462  */
463 boolean_t
464 mac_ip_hdr_length_v6(mblk_t *mp, ip6_t *ip6h, uint16_t *hdr_length,
465     uint8_t *next_hdr)
466 {
467 	uint16_t length;
468 	uint_t	ehdrlen;
469 	uint8_t *whereptr;
470 	uint8_t *endptr;
471 	uint8_t *nexthdrp;
472 	ip6_dest_t *desthdr;
473 	ip6_rthdr_t *rthdr;
474 	ip6_frag_t *fraghdr;
475 
476 	endptr = mp->b_wptr;
477 	if (((uchar_t *)ip6h + IPV6_HDR_LEN) > endptr)
478 		return (B_FALSE);
479 	ASSERT((IPH_HDR_VERSION(ip6h) & ~IP_FORWARD_PROG_BIT) == IPV6_VERSION);
480 	length = IPV6_HDR_LEN;
481 	whereptr = ((uint8_t *)&ip6h[1]); /* point to next hdr */
482 
483 	nexthdrp = &ip6h->ip6_nxt;
484 	while (whereptr < endptr) {
485 		/* Is there enough left for len + nexthdr? */
486 		if (whereptr + MIN_EHDR_LEN > endptr)
487 			break;
488 
489 		switch (*nexthdrp) {
490 		case IPPROTO_HOPOPTS:
491 		case IPPROTO_DSTOPTS:
492 			/* Assumes the headers are identical for hbh and dst */
493 			desthdr = (ip6_dest_t *)whereptr;
494 			ehdrlen = 8 * (desthdr->ip6d_len + 1);
495 			if ((uchar_t *)desthdr +  ehdrlen > endptr)
496 				return (B_FALSE);
497 			nexthdrp = &desthdr->ip6d_nxt;
498 			break;
499 		case IPPROTO_ROUTING:
500 			rthdr = (ip6_rthdr_t *)whereptr;
501 			ehdrlen =  8 * (rthdr->ip6r_len + 1);
502 			if ((uchar_t *)rthdr +  ehdrlen > endptr)
503 				return (B_FALSE);
504 			nexthdrp = &rthdr->ip6r_nxt;
505 			break;
506 		case IPPROTO_FRAGMENT:
507 			fraghdr = (ip6_frag_t *)whereptr;
508 			ehdrlen = sizeof (ip6_frag_t);
509 			if ((uchar_t *)&fraghdr[1] > endptr)
510 				return (B_FALSE);
511 			nexthdrp = &fraghdr->ip6f_nxt;
512 			break;
513 		case IPPROTO_NONE:
514 			/* No next header means we're finished */
515 		default:
516 			*hdr_length = length;
517 			*next_hdr = *nexthdrp;
518 			return (B_TRUE);
519 		}
520 		length += ehdrlen;
521 		whereptr += ehdrlen;
522 		*hdr_length = length;
523 		*next_hdr = *nexthdrp;
524 	}
525 	switch (*nexthdrp) {
526 	case IPPROTO_HOPOPTS:
527 	case IPPROTO_DSTOPTS:
528 	case IPPROTO_ROUTING:
529 	case IPPROTO_FRAGMENT:
530 		/*
531 		 * If any know extension headers are still to be processed,
532 		 * the packet's malformed (or at least all the IP header(s) are
533 		 * not in the same mblk - and that should never happen.
534 		 */
535 		return (B_FALSE);
536 
537 	default:
538 		/*
539 		 * If we get here, we know that all of the IP headers were in
540 		 * the same mblk, even if the ULP header is in the next mblk.
541 		 */
542 		*hdr_length = length;
543 		*next_hdr = *nexthdrp;
544 		return (B_TRUE);
545 	}
546 }
547 
548 typedef struct mac_dladm_intr {
549 	int	ino;
550 	int	cpu_id;
551 	char	driver_path[MAXPATHLEN];
552 	char	nexus_path[MAXPATHLEN];
553 } mac_dladm_intr_t;
554 
555 /* Bind the interrupt to cpu_num */
556 static int
557 mac_set_intr(ldi_handle_t lh, processorid_t cpu_num, int ino)
558 {
559 	pcitool_intr_set_t	iset;
560 	int			err;
561 
562 	iset.ino = ino;
563 	iset.cpu_id = cpu_num;
564 	iset.user_version = PCITOOL_VERSION;
565 	err = ldi_ioctl(lh, PCITOOL_DEVICE_SET_INTR, (intptr_t)&iset, FKIOCTL,
566 	    kcred, NULL);
567 
568 	return (err);
569 }
570 
571 /*
572  * Search interrupt information. iget is filled in with the info to search
573  */
574 static boolean_t
575 mac_search_intrinfo(pcitool_intr_get_t *iget_p, mac_dladm_intr_t *dln)
576 {
577 	int	i;
578 	char	driver_path[2 * MAXPATHLEN];
579 
580 	for (i = 0; i < iget_p->num_devs; i++) {
581 		(void) strlcpy(driver_path, iget_p->dev[i].path, MAXPATHLEN);
582 		(void) snprintf(&driver_path[strlen(driver_path)], MAXPATHLEN,
583 		    ":%s%d", iget_p->dev[i].driver_name,
584 		    iget_p->dev[i].dev_inst);
585 		/* Match the device path for the device path */
586 		if (strcmp(driver_path, dln->driver_path) == 0) {
587 			dln->ino = iget_p->ino;
588 			dln->cpu_id = iget_p->cpu_id;
589 			return (B_TRUE);
590 		}
591 	}
592 	return (B_FALSE);
593 }
594 
595 /*
596  * Get information about ino, i.e. if this is the interrupt for our
597  * device and where it is bound etc.
598  */
599 static boolean_t
600 mac_get_single_intr(ldi_handle_t lh, int ino, mac_dladm_intr_t *dln)
601 {
602 	pcitool_intr_get_t	*iget_p;
603 	int			ipsz;
604 	int			nipsz;
605 	int			err;
606 	uint8_t			inum;
607 
608 	/*
609 	 * Check if SLEEP is OK, i.e if could come here in response to
610 	 * changing the fanout due to some callback from the driver, say
611 	 * link speed changes.
612 	 */
613 	ipsz = PCITOOL_IGET_SIZE(0);
614 	iget_p = kmem_zalloc(ipsz, KM_SLEEP);
615 
616 	iget_p->num_devs_ret = 0;
617 	iget_p->user_version = PCITOOL_VERSION;
618 	iget_p->ino = ino;
619 
620 	err = ldi_ioctl(lh, PCITOOL_DEVICE_GET_INTR, (intptr_t)iget_p,
621 	    FKIOCTL, kcred, NULL);
622 	if (err != 0) {
623 		kmem_free(iget_p, ipsz);
624 		return (B_FALSE);
625 	}
626 	if (iget_p->num_devs == 0) {
627 		kmem_free(iget_p, ipsz);
628 		return (B_FALSE);
629 	}
630 	inum = iget_p->num_devs;
631 	if (iget_p->num_devs_ret < iget_p->num_devs) {
632 		/* Reallocate */
633 		nipsz = PCITOOL_IGET_SIZE(iget_p->num_devs);
634 
635 		kmem_free(iget_p, ipsz);
636 		ipsz = nipsz;
637 		iget_p = kmem_zalloc(ipsz, KM_SLEEP);
638 
639 		iget_p->num_devs_ret = inum;
640 		iget_p->ino = ino;
641 		iget_p->user_version = PCITOOL_VERSION;
642 		err = ldi_ioctl(lh, PCITOOL_DEVICE_GET_INTR, (intptr_t)iget_p,
643 		    FKIOCTL, kcred, NULL);
644 		if (err != 0) {
645 			kmem_free(iget_p, ipsz);
646 			return (B_FALSE);
647 		}
648 		/* defensive */
649 		if (iget_p->num_devs != iget_p->num_devs_ret) {
650 			kmem_free(iget_p, ipsz);
651 			return (B_FALSE);
652 		}
653 	}
654 
655 	if (mac_search_intrinfo(iget_p, dln)) {
656 		kmem_free(iget_p, ipsz);
657 		return (B_TRUE);
658 	}
659 	kmem_free(iget_p, ipsz);
660 	return (B_FALSE);
661 }
662 
663 /*
664  * Get the interrupts and check each one to see if it is for our device.
665  */
666 static int
667 mac_validate_intr(ldi_handle_t lh, mac_dladm_intr_t *dln, processorid_t cpuid)
668 {
669 	pcitool_intr_info_t	intr_info;
670 	int			err;
671 	int			ino;
672 
673 	err = ldi_ioctl(lh, PCITOOL_SYSTEM_INTR_INFO, (intptr_t)&intr_info,
674 	    FKIOCTL, kcred, NULL);
675 	if (err != 0)
676 		return (-1);
677 
678 	for (ino = 0; ino < intr_info.num_intr; ino++) {
679 		if (mac_get_single_intr(lh, ino, dln)) {
680 			if (dln->cpu_id == cpuid)
681 				return (0);
682 			return (1);
683 		}
684 	}
685 	return (-1);
686 }
687 
688 /*
689  * Obtain the nexus parent node info. for mdip.
690  */
691 static dev_info_t *
692 mac_get_nexus_node(dev_info_t *mdip, mac_dladm_intr_t *dln)
693 {
694 	struct dev_info		*tdip = (struct dev_info *)mdip;
695 	struct ddi_minor_data	*minordata;
696 	int			circ;
697 	dev_info_t		*pdip;
698 	char			pathname[MAXPATHLEN];
699 
700 	while (tdip != NULL) {
701 		ndi_devi_enter((dev_info_t *)tdip, &circ);
702 		for (minordata = tdip->devi_minor; minordata != NULL;
703 		    minordata = minordata->next) {
704 			if (strncmp(minordata->ddm_node_type, DDI_NT_INTRCTL,
705 			    strlen(DDI_NT_INTRCTL)) == 0) {
706 				pdip = minordata->dip;
707 				(void) ddi_pathname(pdip, pathname);
708 				(void) snprintf(dln->nexus_path, MAXPATHLEN,
709 				    "/devices%s:intr", pathname);
710 				(void) ddi_pathname_minor(minordata, pathname);
711 				ndi_devi_exit((dev_info_t *)tdip, circ);
712 				return (pdip);
713 			}
714 		}
715 		ndi_devi_exit((dev_info_t *)tdip, circ);
716 		tdip = tdip->devi_parent;
717 	}
718 	return (NULL);
719 }
720 
721 /*
722  * For a primary MAC client, if the user has set a list or CPUs or
723  * we have obtained it implicitly, we try to retarget the interrupt
724  * for that device on one of the CPUs in the list.
725  * We assign the interrupt to the same CPU as the poll thread.
726  */
727 static boolean_t
728 mac_check_interrupt_binding(dev_info_t *mdip, int32_t cpuid)
729 {
730 	ldi_handle_t		lh = NULL;
731 	ldi_ident_t		li = NULL;
732 	int			err;
733 	int			ret;
734 	mac_dladm_intr_t	dln;
735 	dev_info_t		*dip;
736 	struct ddi_minor_data	*minordata;
737 
738 	dln.nexus_path[0] = '\0';
739 	dln.driver_path[0] = '\0';
740 
741 	minordata = ((struct dev_info *)mdip)->devi_minor;
742 	while (minordata != NULL) {
743 		if (minordata->type == DDM_MINOR)
744 			break;
745 		minordata = minordata->next;
746 	}
747 	if (minordata == NULL)
748 		return (B_FALSE);
749 
750 	(void) ddi_pathname_minor(minordata, dln.driver_path);
751 
752 	dip = mac_get_nexus_node(mdip, &dln);
753 	/* defensive */
754 	if (dip == NULL)
755 		return (B_FALSE);
756 
757 	err = ldi_ident_from_major(ddi_driver_major(dip), &li);
758 	if (err != 0)
759 		return (B_FALSE);
760 
761 	err = ldi_open_by_name(dln.nexus_path, FREAD|FWRITE, kcred, &lh, li);
762 	if (err != 0)
763 		return (B_FALSE);
764 
765 	ret = mac_validate_intr(lh, &dln, cpuid);
766 	if (ret < 0) {
767 		(void) ldi_close(lh, FREAD|FWRITE, kcred);
768 		return (B_FALSE);
769 	}
770 	/* cmn_note? */
771 	if (ret != 0)
772 		if ((err = (mac_set_intr(lh, cpuid, dln.ino))) != 0) {
773 			(void) ldi_close(lh, FREAD|FWRITE, kcred);
774 			return (B_FALSE);
775 		}
776 	(void) ldi_close(lh, FREAD|FWRITE, kcred);
777 	return (B_TRUE);
778 }
779 
780 void
781 mac_client_set_intr_cpu(void *arg, mac_client_handle_t mch, int32_t cpuid)
782 {
783 	dev_info_t		*mdip = (dev_info_t *)arg;
784 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
785 	mac_resource_props_t	*mrp;
786 	mac_perim_handle_t	mph;
787 
788 	if (cpuid == -1 || !mac_check_interrupt_binding(mdip, cpuid))
789 		return;
790 
791 	mac_perim_enter_by_mh((mac_handle_t)mcip->mci_mip, &mph);
792 	mrp = MCIP_RESOURCE_PROPS(mcip);
793 	mrp->mrp_intr_cpu = cpuid;
794 	mac_perim_exit(mph);
795 }
796 
797 int32_t
798 mac_client_intr_cpu(mac_client_handle_t mch)
799 {
800 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
801 	mac_cpus_t		*srs_cpu;
802 	mac_soft_ring_set_t	*rx_srs;
803 	flow_entry_t		*flent = mcip->mci_flent;
804 	mac_resource_props_t	*mrp = MCIP_RESOURCE_PROPS(mcip);
805 
806 	/*
807 	 * Check if we need to retarget the interrupt. We do this only
808 	 * for the primary MAC client. We do this if we have the only
809 	 *  exclusive ring in the group.
810 	 */
811 	if (mac_is_primary_client(mcip) && flent->fe_rx_srs_cnt == 2) {
812 		rx_srs = flent->fe_rx_srs[1];
813 		srs_cpu = &rx_srs->srs_cpu;
814 		if (mrp->mrp_intr_cpu == srs_cpu->mc_pollid)
815 			return (-1);
816 		return (srs_cpu->mc_pollid);
817 	}
818 	return (-1);
819 }
820 
821 void *
822 mac_get_devinfo(mac_handle_t mh)
823 {
824 	mac_impl_t	*mip = (mac_impl_t *)mh;
825 
826 	return ((void *)mip->mi_dip);
827 }
828 
829 #define	PKT_HASH_4BYTES(x) ((x)[0] ^ (x)[1] ^ (x)[2] ^ (x)[3])
830 #define	PKT_HASH_MAC(x) ((x)[0] ^ (x)[1] ^ (x)[2] ^ (x)[3] ^ (x)[4] ^ (x)[5])
831 
832 uint64_t
833 mac_pkt_hash(uint_t media, mblk_t *mp, uint8_t policy, boolean_t is_outbound)
834 {
835 	struct ether_header *ehp;
836 	uint64_t hash = 0;
837 	uint16_t sap;
838 	uint_t skip_len;
839 	uint8_t proto;
840 
841 	/*
842 	 * We may want to have one of these per MAC type plugin in the
843 	 * future. For now supports only ethernet.
844 	 */
845 	if (media != DL_ETHER)
846 		return (0L);
847 
848 	/* for now we support only outbound packets */
849 	ASSERT(is_outbound);
850 	ASSERT(IS_P2ALIGNED(mp->b_rptr, sizeof (uint16_t)));
851 	ASSERT(MBLKL(mp) >= sizeof (struct ether_header));
852 
853 	/* compute L2 hash */
854 
855 	ehp = (struct ether_header *)mp->b_rptr;
856 
857 	if ((policy & MAC_PKT_HASH_L2) != 0) {
858 		uchar_t *mac_src = ehp->ether_shost.ether_addr_octet;
859 		uchar_t *mac_dst = ehp->ether_dhost.ether_addr_octet;
860 		hash = PKT_HASH_MAC(mac_src) ^ PKT_HASH_MAC(mac_dst);
861 		policy &= ~MAC_PKT_HASH_L2;
862 	}
863 
864 	if (policy == 0)
865 		goto done;
866 
867 	/* skip ethernet header */
868 
869 	sap = ntohs(ehp->ether_type);
870 	if (sap == ETHERTYPE_VLAN) {
871 		struct ether_vlan_header *evhp;
872 		mblk_t *newmp = NULL;
873 
874 		skip_len = sizeof (struct ether_vlan_header);
875 		if (MBLKL(mp) < skip_len) {
876 			/* the vlan tag is the payload, pull up first */
877 			newmp = msgpullup(mp, -1);
878 			if ((newmp == NULL) || (MBLKL(newmp) < skip_len)) {
879 				goto done;
880 			}
881 			evhp = (struct ether_vlan_header *)newmp->b_rptr;
882 		} else {
883 			evhp = (struct ether_vlan_header *)mp->b_rptr;
884 		}
885 
886 		sap = ntohs(evhp->ether_type);
887 		freemsg(newmp);
888 	} else {
889 		skip_len = sizeof (struct ether_header);
890 	}
891 
892 	/* if ethernet header is in its own mblk, skip it */
893 	if (MBLKL(mp) <= skip_len) {
894 		skip_len -= MBLKL(mp);
895 		mp = mp->b_cont;
896 		if (mp == NULL)
897 			goto done;
898 	}
899 
900 	sap = (sap < ETHERTYPE_802_MIN) ? 0 : sap;
901 
902 	/* compute IP src/dst addresses hash and skip IPv{4,6} header */
903 
904 	switch (sap) {
905 	case ETHERTYPE_IP: {
906 		ipha_t *iphp;
907 
908 		/*
909 		 * If the header is not aligned or the header doesn't fit
910 		 * in the mblk, bail now. Note that this may cause packets
911 		 * reordering.
912 		 */
913 		iphp = (ipha_t *)(mp->b_rptr + skip_len);
914 		if (((unsigned char *)iphp + sizeof (ipha_t) > mp->b_wptr) ||
915 		    !OK_32PTR((char *)iphp))
916 			goto done;
917 
918 		proto = iphp->ipha_protocol;
919 		skip_len += IPH_HDR_LENGTH(iphp);
920 
921 		if ((policy & MAC_PKT_HASH_L3) != 0) {
922 			uint8_t *ip_src = (uint8_t *)&(iphp->ipha_src);
923 			uint8_t *ip_dst = (uint8_t *)&(iphp->ipha_dst);
924 
925 			hash ^= (PKT_HASH_4BYTES(ip_src) ^
926 			    PKT_HASH_4BYTES(ip_dst));
927 			policy &= ~MAC_PKT_HASH_L3;
928 		}
929 		break;
930 	}
931 	case ETHERTYPE_IPV6: {
932 		ip6_t *ip6hp;
933 		uint16_t hdr_length;
934 
935 		/*
936 		 * If the header is not aligned or the header doesn't fit
937 		 * in the mblk, bail now. Note that this may cause packets
938 		 * reordering.
939 		 */
940 
941 		ip6hp = (ip6_t *)(mp->b_rptr + skip_len);
942 		if (((unsigned char *)ip6hp + IPV6_HDR_LEN > mp->b_wptr) ||
943 		    !OK_32PTR((char *)ip6hp))
944 			goto done;
945 
946 		if (!mac_ip_hdr_length_v6(mp, ip6hp, &hdr_length, &proto))
947 			goto done;
948 		skip_len += hdr_length;
949 
950 		if ((policy & MAC_PKT_HASH_L3) != 0) {
951 			uint8_t *ip_src = &(ip6hp->ip6_src.s6_addr8[12]);
952 			uint8_t *ip_dst = &(ip6hp->ip6_dst.s6_addr8[12]);
953 
954 			hash ^= (PKT_HASH_4BYTES(ip_src) ^
955 			    PKT_HASH_4BYTES(ip_dst));
956 			policy &= ~MAC_PKT_HASH_L3;
957 		}
958 		break;
959 	}
960 	default:
961 		goto done;
962 	}
963 
964 	if (policy == 0)
965 		goto done;
966 
967 	/* if ip header is in its own mblk, skip it */
968 	if (MBLKL(mp) <= skip_len) {
969 		skip_len -= MBLKL(mp);
970 		mp = mp->b_cont;
971 		if (mp == NULL)
972 			goto done;
973 	}
974 
975 	/* parse ULP header */
976 again:
977 	switch (proto) {
978 	case IPPROTO_TCP:
979 	case IPPROTO_UDP:
980 	case IPPROTO_ESP:
981 	case IPPROTO_SCTP:
982 		/*
983 		 * These Internet Protocols are intentionally designed
984 		 * for hashing from the git-go.  Port numbers are in the first
985 		 * word for transports, SPI is first for ESP.
986 		 */
987 		if (mp->b_rptr + skip_len + 4 > mp->b_wptr)
988 			goto done;
989 		hash ^= PKT_HASH_4BYTES((mp->b_rptr + skip_len));
990 		break;
991 
992 	case IPPROTO_AH: {
993 		ah_t *ah = (ah_t *)(mp->b_rptr + skip_len);
994 		uint_t ah_length = AH_TOTAL_LEN(ah);
995 
996 		if ((unsigned char *)ah + sizeof (ah_t) > mp->b_wptr)
997 			goto done;
998 
999 		proto = ah->ah_nexthdr;
1000 		skip_len += ah_length;
1001 
1002 		/* if AH header is in its own mblk, skip it */
1003 		if (MBLKL(mp) <= skip_len) {
1004 			skip_len -= MBLKL(mp);
1005 			mp = mp->b_cont;
1006 			if (mp == NULL)
1007 				goto done;
1008 		}
1009 
1010 		goto again;
1011 	}
1012 	}
1013 
1014 done:
1015 	return (hash);
1016 }
1017