xref: /freebsd/sys/ofed/drivers/infiniband/ulp/ipoib/ipoib_main.c (revision 1f4bcc459a76b7aa664f3fd557684cd0ba6da352)
1 /*
2  * Copyright (c) 2004 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4  * Copyright (c) 2004 Voltaire, Inc. All rights reserved.
5  *
6  * This software is available to you under a choice of one of two
7  * licenses.  You may choose to be licensed under the terms of the GNU
8  * General Public License (GPL) Version 2, available from the file
9  * COPYING in the main directory of this source tree, or the
10  * OpenIB.org BSD license below:
11  *
12  *     Redistribution and use in source and binary forms, with or
13  *     without modification, are permitted provided that the following
14  *     conditions are met:
15  *
16  *      - Redistributions of source code must retain the above
17  *        copyright notice, this list of conditions and the following
18  *        disclaimer.
19  *
20  *      - Redistributions in binary form must reproduce the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer in the documentation and/or other materials
23  *        provided with the distribution.
24  *
25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32  * SOFTWARE.
33  */
34 
35 #include "ipoib.h"
36 
37 static	int ipoib_resolvemulti(struct ifnet *, struct sockaddr **,
38 		struct sockaddr *);
39 
40 
41 #include <linux/module.h>
42 
43 #include <linux/slab.h>
44 #include <linux/kernel.h>
45 #include <linux/vmalloc.h>
46 
47 #include <linux/if_arp.h>	/* For ARPHRD_xxx */
48 #include <linux/if_vlan.h>
49 #include <net/ip.h>
50 #include <net/ipv6.h>
51 
52 MODULE_AUTHOR("Roland Dreier");
53 MODULE_DESCRIPTION("IP-over-InfiniBand net driver");
54 MODULE_LICENSE("Dual BSD/GPL");
55 
56 int ipoib_sendq_size = IPOIB_TX_RING_SIZE;
57 int ipoib_recvq_size = IPOIB_RX_RING_SIZE;
58 
59 module_param_named(send_queue_size, ipoib_sendq_size, int, 0444);
60 MODULE_PARM_DESC(send_queue_size, "Number of descriptors in send queue");
61 module_param_named(recv_queue_size, ipoib_recvq_size, int, 0444);
62 MODULE_PARM_DESC(recv_queue_size, "Number of descriptors in receive queue");
63 
64 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
65 int ipoib_debug_level = 1;
66 
67 module_param_named(debug_level, ipoib_debug_level, int, 0644);
68 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
69 #endif
70 
71 struct ipoib_path_iter {
72 	struct ipoib_dev_priv *priv;
73 	struct ipoib_path  path;
74 };
75 
76 static const u8 ipv4_bcast_addr[] = {
77 	0x00, 0xff, 0xff, 0xff,
78 	0xff, 0x12, 0x40, 0x1b,	0x00, 0x00, 0x00, 0x00,
79 	0x00, 0x00, 0x00, 0x00,	0xff, 0xff, 0xff, 0xff
80 };
81 
82 struct workqueue_struct *ipoib_workqueue;
83 
84 struct ib_sa_client ipoib_sa_client;
85 
86 static void ipoib_add_one(struct ib_device *device);
87 static void ipoib_remove_one(struct ib_device *device);
88 static void ipoib_start(struct ifnet *dev);
89 static int ipoib_output(struct ifnet *ifp, struct mbuf *m,
90 	    const struct sockaddr *dst, struct route *ro);
91 static int ipoib_ioctl(struct ifnet *ifp, u_long command, caddr_t data);
92 static void ipoib_input(struct ifnet *ifp, struct mbuf *m);
93 
94 #define	IPOIB_MTAP(_ifp, _m)					\
95 do {								\
96 	if (bpf_peers_present((_ifp)->if_bpf)) {		\
97 		M_ASSERTVALID(_m);				\
98 		ipoib_mtap_mb((_ifp), (_m));			\
99 	}							\
100 } while (0)
101 
102 /*
103  * This is for clients that have an ipoib_header in the mbuf.
104  */
105 static void
106 ipoib_mtap_mb(struct ifnet *ifp, struct mbuf *mb)
107 {
108 	struct ipoib_header *ih;
109 	struct ether_header eh;
110 
111 	ih = mtod(mb, struct ipoib_header *);
112 	eh.ether_type = ih->proto;
113 	bcopy(ih->hwaddr, &eh.ether_dhost, ETHER_ADDR_LEN);
114 	bzero(&eh.ether_shost, ETHER_ADDR_LEN);
115 	mb->m_data += sizeof(struct ipoib_header);
116 	mb->m_len -= sizeof(struct ipoib_header);
117 	bpf_mtap2(ifp->if_bpf, &eh, sizeof(eh), mb);
118 	mb->m_data -= sizeof(struct ipoib_header);
119 	mb->m_len += sizeof(struct ipoib_header);
120 }
121 
122 void
123 ipoib_mtap_proto(struct ifnet *ifp, struct mbuf *mb, uint16_t proto)
124 {
125 	struct ether_header eh;
126 
127 	eh.ether_type = proto;
128 	bzero(&eh.ether_shost, ETHER_ADDR_LEN);
129 	bzero(&eh.ether_dhost, ETHER_ADDR_LEN);
130 	bpf_mtap2(ifp->if_bpf, &eh, sizeof(eh), mb);
131 }
132 
133 static struct ib_client ipoib_client = {
134 	.name   = "ipoib",
135 	.add    = ipoib_add_one,
136 	.remove = ipoib_remove_one
137 };
138 
139 int
140 ipoib_open(struct ipoib_dev_priv *priv)
141 {
142 	struct ifnet *dev = priv->dev;
143 
144 	ipoib_dbg(priv, "bringing up interface\n");
145 
146 	set_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
147 
148 	if (ipoib_pkey_dev_delay_open(priv))
149 		return 0;
150 
151 	if (ipoib_ib_dev_open(priv))
152 		goto err_disable;
153 
154 	if (ipoib_ib_dev_up(priv))
155 		goto err_stop;
156 
157 	if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
158 		struct ipoib_dev_priv *cpriv;
159 
160 		/* Bring up any child interfaces too */
161 		mutex_lock(&priv->vlan_mutex);
162 		list_for_each_entry(cpriv, &priv->child_intfs, list)
163 			if ((cpriv->dev->if_drv_flags & IFF_DRV_RUNNING) == 0)
164 				ipoib_open(cpriv);
165 		mutex_unlock(&priv->vlan_mutex);
166 	}
167 	dev->if_drv_flags |= IFF_DRV_RUNNING;
168 	dev->if_drv_flags &= ~IFF_DRV_OACTIVE;
169 
170 	return 0;
171 
172 err_stop:
173 	ipoib_ib_dev_stop(priv, 1);
174 
175 err_disable:
176 	clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
177 
178 	return -EINVAL;
179 }
180 
181 static void
182 ipoib_init(void *arg)
183 {
184 	struct ifnet *dev;
185 	struct ipoib_dev_priv *priv;
186 
187 	priv = arg;
188 	dev = priv->dev;
189 	if ((dev->if_drv_flags & IFF_DRV_RUNNING) == 0)
190 		ipoib_open(priv);
191 	queue_work(ipoib_workqueue, &priv->flush_light);
192 }
193 
194 
195 static int
196 ipoib_stop(struct ipoib_dev_priv *priv)
197 {
198 	struct ifnet *dev = priv->dev;
199 
200 	ipoib_dbg(priv, "stopping interface\n");
201 
202 	clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
203 
204 	dev->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
205 
206 	ipoib_ib_dev_down(priv, 0);
207 	ipoib_ib_dev_stop(priv, 0);
208 
209 	if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
210 		struct ipoib_dev_priv *cpriv;
211 
212 		/* Bring down any child interfaces too */
213 		mutex_lock(&priv->vlan_mutex);
214 		list_for_each_entry(cpriv, &priv->child_intfs, list)
215 			if ((cpriv->dev->if_drv_flags & IFF_DRV_RUNNING) != 0)
216 				ipoib_stop(cpriv);
217 		mutex_unlock(&priv->vlan_mutex);
218 	}
219 
220 	return 0;
221 }
222 
223 int
224 ipoib_change_mtu(struct ipoib_dev_priv *priv, int new_mtu)
225 {
226 	struct ifnet *dev = priv->dev;
227 
228 	/* dev->if_mtu > 2K ==> connected mode */
229 	if (ipoib_cm_admin_enabled(priv)) {
230 		if (new_mtu > IPOIB_CM_MTU(ipoib_cm_max_mtu(priv)))
231 			return -EINVAL;
232 
233 		if (new_mtu > priv->mcast_mtu)
234 			ipoib_warn(priv, "mtu > %d will cause multicast packet drops.\n",
235 				   priv->mcast_mtu);
236 
237 		dev->if_mtu = new_mtu;
238 		return 0;
239 	}
240 
241 	if (new_mtu > IPOIB_UD_MTU(priv->max_ib_mtu))
242 		return -EINVAL;
243 
244 	priv->admin_mtu = new_mtu;
245 
246 	dev->if_mtu = min(priv->mcast_mtu, priv->admin_mtu);
247 
248 	queue_work(ipoib_workqueue, &priv->flush_light);
249 
250 	return 0;
251 }
252 
253 static int
254 ipoib_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
255 {
256 	struct ipoib_dev_priv *priv = ifp->if_softc;
257 	struct ifaddr *ifa = (struct ifaddr *) data;
258 	struct ifreq *ifr = (struct ifreq *) data;
259 	int error = 0;
260 
261 	switch (command) {
262 	case SIOCSIFFLAGS:
263 		if (ifp->if_flags & IFF_UP) {
264 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
265 				error = -ipoib_open(priv);
266 		} else
267 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
268 				ipoib_stop(priv);
269 		break;
270 	case SIOCADDMULTI:
271 	case SIOCDELMULTI:
272 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
273 			queue_work(ipoib_workqueue, &priv->restart_task);
274 		break;
275 	case SIOCSIFADDR:
276 		ifp->if_flags |= IFF_UP;
277 
278 		switch (ifa->ifa_addr->sa_family) {
279 #ifdef INET
280 		case AF_INET:
281 			ifp->if_init(ifp->if_softc);	/* before arpwhohas */
282 			arp_ifinit(ifp, ifa);
283 			break;
284 #endif
285 		default:
286 			ifp->if_init(ifp->if_softc);
287 			break;
288 		}
289 		break;
290 
291 	case SIOCGIFADDR:
292 		{
293 			struct sockaddr *sa;
294 
295 			sa = (struct sockaddr *) & ifr->ifr_data;
296 			bcopy(IF_LLADDR(ifp),
297 			      (caddr_t) sa->sa_data, INFINIBAND_ALEN);
298 		}
299 		break;
300 
301 	case SIOCSIFMTU:
302 		/*
303 		 * Set the interface MTU.
304 		 */
305 		error = -ipoib_change_mtu(priv, ifr->ifr_mtu);
306 		break;
307 	default:
308 		error = EINVAL;
309 		break;
310 	}
311 	return (error);
312 }
313 
314 
315 static struct ipoib_path *
316 __path_find(struct ipoib_dev_priv *priv, void *gid)
317 {
318 	struct rb_node *n = priv->path_tree.rb_node;
319 	struct ipoib_path *path;
320 	int ret;
321 
322 	while (n) {
323 		path = rb_entry(n, struct ipoib_path, rb_node);
324 
325 		ret = memcmp(gid, path->pathrec.dgid.raw,
326 			     sizeof (union ib_gid));
327 
328 		if (ret < 0)
329 			n = n->rb_left;
330 		else if (ret > 0)
331 			n = n->rb_right;
332 		else
333 			return path;
334 	}
335 
336 	return NULL;
337 }
338 
339 static int
340 __path_add(struct ipoib_dev_priv *priv, struct ipoib_path *path)
341 {
342 	struct rb_node **n = &priv->path_tree.rb_node;
343 	struct rb_node *pn = NULL;
344 	struct ipoib_path *tpath;
345 	int ret;
346 
347 	while (*n) {
348 		pn = *n;
349 		tpath = rb_entry(pn, struct ipoib_path, rb_node);
350 
351 		ret = memcmp(path->pathrec.dgid.raw, tpath->pathrec.dgid.raw,
352 			     sizeof (union ib_gid));
353 		if (ret < 0)
354 			n = &pn->rb_left;
355 		else if (ret > 0)
356 			n = &pn->rb_right;
357 		else
358 			return -EEXIST;
359 	}
360 
361 	rb_link_node(&path->rb_node, pn, n);
362 	rb_insert_color(&path->rb_node, &priv->path_tree);
363 
364 	list_add_tail(&path->list, &priv->path_list);
365 
366 	return 0;
367 }
368 
369 void
370 ipoib_path_free(struct ipoib_dev_priv *priv, struct ipoib_path *path)
371 {
372 
373 	_IF_DRAIN(&path->queue);
374 
375 	if (path->ah)
376 		ipoib_put_ah(path->ah);
377 	if (ipoib_cm_get(path))
378 		ipoib_cm_destroy_tx(ipoib_cm_get(path));
379 
380 	kfree(path);
381 }
382 
383 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
384 
385 struct ipoib_path_iter *
386 ipoib_path_iter_init(struct ipoib_dev_priv *priv)
387 {
388 	struct ipoib_path_iter *iter;
389 
390 	iter = kmalloc(sizeof *iter, GFP_KERNEL);
391 	if (!iter)
392 		return NULL;
393 
394 	iter->priv = priv;
395 	memset(iter->path.pathrec.dgid.raw, 0, 16);
396 
397 	if (ipoib_path_iter_next(iter)) {
398 		kfree(iter);
399 		return NULL;
400 	}
401 
402 	return iter;
403 }
404 
405 int
406 ipoib_path_iter_next(struct ipoib_path_iter *iter)
407 {
408 	struct ipoib_dev_priv *priv = iter->priv;
409 	struct rb_node *n;
410 	struct ipoib_path *path;
411 	int ret = 1;
412 
413 	spin_lock_irq(&priv->lock);
414 
415 	n = rb_first(&priv->path_tree);
416 
417 	while (n) {
418 		path = rb_entry(n, struct ipoib_path, rb_node);
419 
420 		if (memcmp(iter->path.pathrec.dgid.raw, path->pathrec.dgid.raw,
421 			   sizeof (union ib_gid)) < 0) {
422 			iter->path = *path;
423 			ret = 0;
424 			break;
425 		}
426 
427 		n = rb_next(n);
428 	}
429 
430 	spin_unlock_irq(&priv->lock);
431 
432 	return ret;
433 }
434 
435 void
436 ipoib_path_iter_read(struct ipoib_path_iter *iter, struct ipoib_path *path)
437 {
438 	*path = iter->path;
439 }
440 
441 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
442 
443 void
444 ipoib_mark_paths_invalid(struct ipoib_dev_priv *priv)
445 {
446 	struct ipoib_path *path, *tp;
447 
448 	spin_lock_irq(&priv->lock);
449 
450 	list_for_each_entry_safe(path, tp, &priv->path_list, list) {
451 		ipoib_dbg(priv, "mark path LID 0x%04x GID %16D invalid\n",
452 			be16_to_cpu(path->pathrec.dlid),
453 			path->pathrec.dgid.raw, ":");
454 		path->valid =  0;
455 	}
456 
457 	spin_unlock_irq(&priv->lock);
458 }
459 
460 void
461 ipoib_flush_paths(struct ipoib_dev_priv *priv)
462 {
463 	struct ipoib_path *path, *tp;
464 	LIST_HEAD(remove_list);
465 	unsigned long flags;
466 
467 	spin_lock_irqsave(&priv->lock, flags);
468 
469 	list_splice_init(&priv->path_list, &remove_list);
470 
471 	list_for_each_entry(path, &remove_list, list)
472 		rb_erase(&path->rb_node, &priv->path_tree);
473 
474 	list_for_each_entry_safe(path, tp, &remove_list, list) {
475 		if (path->query)
476 			ib_sa_cancel_query(path->query_id, path->query);
477 		spin_unlock_irqrestore(&priv->lock, flags);
478 		wait_for_completion(&path->done);
479 		ipoib_path_free(priv, path);
480 		spin_lock_irqsave(&priv->lock, flags);
481 	}
482 
483 	spin_unlock_irqrestore(&priv->lock, flags);
484 }
485 
486 static void
487 path_rec_completion(int status, struct ib_sa_path_rec *pathrec, void *path_ptr)
488 {
489 	struct ipoib_path *path = path_ptr;
490 	struct ipoib_dev_priv *priv = path->priv;
491 	struct ifnet *dev = priv->dev;
492 	struct ipoib_ah *ah = NULL;
493 	struct ipoib_ah *old_ah = NULL;
494 	struct ifqueue mbqueue;
495 	struct mbuf *mb;
496 	unsigned long flags;
497 
498 	if (!status)
499 		ipoib_dbg(priv, "PathRec LID 0x%04x for GID %16D\n",
500 			  be16_to_cpu(pathrec->dlid), pathrec->dgid.raw, ":");
501 	else
502 		ipoib_dbg(priv, "PathRec status %d for GID %16D\n",
503 			  status, path->pathrec.dgid.raw, ":");
504 
505 	bzero(&mbqueue, sizeof(mbqueue));
506 
507 	if (!status) {
508 		struct ib_ah_attr av;
509 
510 		if (!ib_init_ah_from_path(priv->ca, priv->port, pathrec, &av))
511 			ah = ipoib_create_ah(priv, priv->pd, &av);
512 	}
513 
514 	spin_lock_irqsave(&priv->lock, flags);
515 
516 	if (ah) {
517 		path->pathrec = *pathrec;
518 
519 		old_ah   = path->ah;
520 		path->ah = ah;
521 
522 		ipoib_dbg(priv, "created address handle %p for LID 0x%04x, SL %d\n",
523 			  ah, be16_to_cpu(pathrec->dlid), pathrec->sl);
524 
525 		for (;;) {
526 			_IF_DEQUEUE(&path->queue, mb);
527 			if (mb == NULL)
528 				break;
529 			_IF_ENQUEUE(&mbqueue, mb);
530 		}
531 
532 #ifdef CONFIG_INFINIBAND_IPOIB_CM
533 		if (ipoib_cm_enabled(priv, path->hwaddr) && !ipoib_cm_get(path))
534 			ipoib_cm_set(path, ipoib_cm_create_tx(priv, path));
535 #endif
536 
537 		path->valid = 1;
538 	}
539 
540 	path->query = NULL;
541 	complete(&path->done);
542 
543 	spin_unlock_irqrestore(&priv->lock, flags);
544 
545 	if (old_ah)
546 		ipoib_put_ah(old_ah);
547 
548 	for (;;) {
549 		_IF_DEQUEUE(&mbqueue, mb);
550 		if (mb == NULL)
551 			break;
552 		mb->m_pkthdr.rcvif = dev;
553 		if (dev->if_transmit(dev, mb))
554 			ipoib_warn(priv, "dev_queue_xmit failed "
555 				   "to requeue packet\n");
556 	}
557 }
558 
559 static struct ipoib_path *
560 path_rec_create(struct ipoib_dev_priv *priv, uint8_t *hwaddr)
561 {
562 	struct ipoib_path *path;
563 
564 	if (!priv->broadcast)
565 		return NULL;
566 
567 	path = kzalloc(sizeof *path, GFP_ATOMIC);
568 	if (!path)
569 		return NULL;
570 
571 	path->priv = priv;
572 
573 	bzero(&path->queue, sizeof(path->queue));
574 
575 #ifdef CONFIG_INFINIBAND_IPOIB_CM
576 	memcpy(&path->hwaddr, hwaddr, INFINIBAND_ALEN);
577 #endif
578 	memcpy(path->pathrec.dgid.raw, &hwaddr[4], sizeof (union ib_gid));
579 	path->pathrec.sgid	    = priv->local_gid;
580 	path->pathrec.pkey	    = cpu_to_be16(priv->pkey);
581 	path->pathrec.numb_path     = 1;
582 	path->pathrec.traffic_class = priv->broadcast->mcmember.traffic_class;
583 
584 	return path;
585 }
586 
587 static int
588 path_rec_start(struct ipoib_dev_priv *priv, struct ipoib_path *path)
589 {
590 	struct ifnet *dev = priv->dev;
591 
592 	ib_sa_comp_mask comp_mask = IB_SA_PATH_REC_MTU_SELECTOR | IB_SA_PATH_REC_MTU;
593 	struct ib_sa_path_rec p_rec;
594 
595 	p_rec = path->pathrec;
596 	p_rec.mtu_selector = IB_SA_GT;
597 
598 	switch (roundup_pow_of_two(dev->if_mtu + IPOIB_ENCAP_LEN)) {
599 	case 512:
600 		p_rec.mtu = IB_MTU_256;
601 		break;
602 	case 1024:
603 		p_rec.mtu = IB_MTU_512;
604 		break;
605 	case 2048:
606 		p_rec.mtu = IB_MTU_1024;
607 		break;
608 	case 4096:
609 		p_rec.mtu = IB_MTU_2048;
610 		break;
611 	default:
612 		/* Wildcard everything */
613 		comp_mask = 0;
614 		p_rec.mtu = 0;
615 		p_rec.mtu_selector = 0;
616 	}
617 
618 	ipoib_dbg(priv, "Start path record lookup for %16D MTU > %d\n",
619 		  p_rec.dgid.raw, ":",
620 		  comp_mask ? ib_mtu_enum_to_int(p_rec.mtu) : 0);
621 
622 	init_completion(&path->done);
623 
624 	path->query_id =
625 		ib_sa_path_rec_get(&ipoib_sa_client, priv->ca, priv->port,
626 				   &p_rec, comp_mask		|
627 				   IB_SA_PATH_REC_DGID		|
628 				   IB_SA_PATH_REC_SGID		|
629 				   IB_SA_PATH_REC_NUMB_PATH	|
630 				   IB_SA_PATH_REC_TRAFFIC_CLASS |
631 				   IB_SA_PATH_REC_PKEY,
632 				   1000, GFP_ATOMIC,
633 				   path_rec_completion,
634 				   path, &path->query);
635 	if (path->query_id < 0) {
636 		ipoib_warn(priv, "ib_sa_path_rec_get failed: %d\n", path->query_id);
637 		path->query = NULL;
638 		complete(&path->done);
639 		return path->query_id;
640 	}
641 
642 	return 0;
643 }
644 
645 static void
646 ipoib_unicast_send(struct mbuf *mb, struct ipoib_dev_priv *priv, struct ipoib_header *eh)
647 {
648 	struct ipoib_path *path;
649 
650 	path = __path_find(priv, eh->hwaddr + 4);
651 	if (!path || !path->valid) {
652 		int new_path = 0;
653 
654 		if (!path) {
655 			path = path_rec_create(priv, eh->hwaddr);
656 			new_path = 1;
657 		}
658 		if (path) {
659 			_IF_ENQUEUE(&path->queue, mb);
660 			if (!path->query && path_rec_start(priv, path)) {
661 				spin_unlock_irqrestore(&priv->lock, flags);
662 				if (new_path)
663 					ipoib_path_free(priv, path);
664 				return;
665 			} else
666 				__path_add(priv, path);
667 		} else {
668 			if_inc_counter(priv->dev, IFCOUNTER_OERRORS, 1);
669 			m_freem(mb);
670 		}
671 
672 		return;
673 	}
674 
675 	if (ipoib_cm_get(path) && ipoib_cm_up(path)) {
676 		ipoib_cm_send(priv, mb, ipoib_cm_get(path));
677 	} else if (path->ah) {
678 		ipoib_send(priv, mb, path->ah, IPOIB_QPN(eh->hwaddr));
679 	} else if ((path->query || !path_rec_start(priv, path)) &&
680 		    path->queue.ifq_len < IPOIB_MAX_PATH_REC_QUEUE) {
681 		_IF_ENQUEUE(&path->queue, mb);
682 	} else {
683 		if_inc_counter(priv->dev, IFCOUNTER_OERRORS, 1);
684 		m_freem(mb);
685 	}
686 }
687 
688 static int
689 ipoib_send_one(struct ipoib_dev_priv *priv, struct mbuf *mb)
690 {
691 	struct ipoib_header *eh;
692 
693 	eh = mtod(mb, struct ipoib_header *);
694 	if (IPOIB_IS_MULTICAST(eh->hwaddr)) {
695 		/* Add in the P_Key for multicast*/
696 		eh->hwaddr[8] = (priv->pkey >> 8) & 0xff;
697 		eh->hwaddr[9] = priv->pkey & 0xff;
698 
699 		ipoib_mcast_send(priv, eh->hwaddr + 4, mb);
700 	} else
701 		ipoib_unicast_send(mb, priv, eh);
702 
703 	return 0;
704 }
705 
706 
707 static void
708 _ipoib_start(struct ifnet *dev, struct ipoib_dev_priv *priv)
709 {
710 	struct mbuf *mb;
711 
712 	if ((dev->if_drv_flags & (IFF_DRV_RUNNING|IFF_DRV_OACTIVE)) !=
713 	    IFF_DRV_RUNNING)
714 		return;
715 
716 	spin_lock(&priv->lock);
717 	while (!IFQ_DRV_IS_EMPTY(&dev->if_snd) &&
718 	    (dev->if_drv_flags & IFF_DRV_OACTIVE) == 0) {
719 		IFQ_DRV_DEQUEUE(&dev->if_snd, mb);
720 		if (mb == NULL)
721 			break;
722 		IPOIB_MTAP(dev, mb);
723 		ipoib_send_one(priv, mb);
724 	}
725 	spin_unlock(&priv->lock);
726 }
727 
728 static void
729 ipoib_start(struct ifnet *dev)
730 {
731 	_ipoib_start(dev, dev->if_softc);
732 }
733 
734 static void
735 ipoib_vlan_start(struct ifnet *dev)
736 {
737 	struct ipoib_dev_priv *priv;
738 	struct mbuf *mb;
739 
740 	priv = VLAN_COOKIE(dev);
741 	if (priv != NULL)
742 		return _ipoib_start(dev, priv);
743 	while (!IFQ_DRV_IS_EMPTY(&dev->if_snd)) {
744 		IFQ_DRV_DEQUEUE(&dev->if_snd, mb);
745 		if (mb == NULL)
746 			break;
747 		m_freem(mb);
748 		if_inc_counter(dev, IFCOUNTER_OERRORS, 1);
749 	}
750 }
751 
752 int
753 ipoib_dev_init(struct ipoib_dev_priv *priv, struct ib_device *ca, int port)
754 {
755 
756 	/* Allocate RX/TX "rings" to hold queued mbs */
757 	priv->rx_ring =	kzalloc(ipoib_recvq_size * sizeof *priv->rx_ring,
758 				GFP_KERNEL);
759 	if (!priv->rx_ring) {
760 		printk(KERN_WARNING "%s: failed to allocate RX ring (%d entries)\n",
761 		       ca->name, ipoib_recvq_size);
762 		goto out;
763 	}
764 
765 	priv->tx_ring = kzalloc(ipoib_sendq_size * sizeof *priv->tx_ring, GFP_KERNEL);
766 	if (!priv->tx_ring) {
767 		printk(KERN_WARNING "%s: failed to allocate TX ring (%d entries)\n",
768 		       ca->name, ipoib_sendq_size);
769 		goto out_rx_ring_cleanup;
770 	}
771 	memset(priv->tx_ring, 0, ipoib_sendq_size * sizeof *priv->tx_ring);
772 
773 	/* priv->tx_head, tx_tail & tx_outstanding are already 0 */
774 
775 	if (ipoib_ib_dev_init(priv, ca, port))
776 		goto out_tx_ring_cleanup;
777 
778 	return 0;
779 
780 out_tx_ring_cleanup:
781 	kfree(priv->tx_ring);
782 
783 out_rx_ring_cleanup:
784 	kfree(priv->rx_ring);
785 
786 out:
787 	return -ENOMEM;
788 }
789 
790 static void
791 ipoib_detach(struct ipoib_dev_priv *priv)
792 {
793 	struct ifnet *dev;
794 
795 	dev = priv->dev;
796 	if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
797 		bpfdetach(dev);
798 		if_detach(dev);
799 		if_free(dev);
800 	} else
801 		VLAN_SETCOOKIE(priv->dev, NULL);
802 
803 	free(priv, M_TEMP);
804 }
805 
806 void
807 ipoib_dev_cleanup(struct ipoib_dev_priv *priv)
808 {
809 	struct ipoib_dev_priv *cpriv, *tcpriv;
810 
811 	/* Delete any child interfaces first */
812 	list_for_each_entry_safe(cpriv, tcpriv, &priv->child_intfs, list) {
813 		ipoib_dev_cleanup(cpriv);
814 		ipoib_detach(cpriv);
815 	}
816 
817 	ipoib_ib_dev_cleanup(priv);
818 
819 	kfree(priv->rx_ring);
820 	kfree(priv->tx_ring);
821 
822 	priv->rx_ring = NULL;
823 	priv->tx_ring = NULL;
824 }
825 
826 static volatile int ipoib_unit;
827 
828 static struct ipoib_dev_priv *
829 ipoib_priv_alloc(void)
830 {
831 	struct ipoib_dev_priv *priv;
832 
833 	priv = malloc(sizeof(struct ipoib_dev_priv), M_TEMP, M_ZERO|M_WAITOK);
834 	spin_lock_init(&priv->lock);
835 	spin_lock_init(&priv->drain_lock);
836 	mutex_init(&priv->vlan_mutex);
837 	INIT_LIST_HEAD(&priv->path_list);
838 	INIT_LIST_HEAD(&priv->child_intfs);
839 	INIT_LIST_HEAD(&priv->dead_ahs);
840 	INIT_LIST_HEAD(&priv->multicast_list);
841 	INIT_DELAYED_WORK(&priv->pkey_poll_task, ipoib_pkey_poll);
842 	INIT_DELAYED_WORK(&priv->mcast_task,   ipoib_mcast_join_task);
843 	INIT_WORK(&priv->carrier_on_task, ipoib_mcast_carrier_on_task);
844 	INIT_WORK(&priv->flush_light,   ipoib_ib_dev_flush_light);
845 	INIT_WORK(&priv->flush_normal,   ipoib_ib_dev_flush_normal);
846 	INIT_WORK(&priv->flush_heavy,   ipoib_ib_dev_flush_heavy);
847 	INIT_WORK(&priv->restart_task, ipoib_mcast_restart_task);
848 	INIT_DELAYED_WORK(&priv->ah_reap_task, ipoib_reap_ah);
849 	memcpy(priv->broadcastaddr, ipv4_bcast_addr, INFINIBAND_ALEN);
850 
851 	return (priv);
852 }
853 
854 struct ipoib_dev_priv *
855 ipoib_intf_alloc(const char *name)
856 {
857 	struct ipoib_dev_priv *priv;
858 	struct sockaddr_dl *sdl;
859 	struct ifnet *dev;
860 
861 	priv = ipoib_priv_alloc();
862 	dev = priv->dev = if_alloc(IFT_INFINIBAND);
863 	if (!dev) {
864 		free(priv, M_TEMP);
865 		return NULL;
866 	}
867 	dev->if_softc = priv;
868 	if_initname(dev, name, atomic_fetchadd_int(&ipoib_unit, 1));
869 	dev->if_flags = IFF_BROADCAST | IFF_MULTICAST;
870 	dev->if_addrlen = INFINIBAND_ALEN;
871 	dev->if_hdrlen = IPOIB_HEADER_LEN;
872 	if_attach(dev);
873 	dev->if_init = ipoib_init;
874 	dev->if_ioctl = ipoib_ioctl;
875 	dev->if_start = ipoib_start;
876 	dev->if_output = ipoib_output;
877 	dev->if_input = ipoib_input;
878 	dev->if_resolvemulti = ipoib_resolvemulti;
879 	dev->if_baudrate = IF_Gbps(10);
880 	dev->if_broadcastaddr = priv->broadcastaddr;
881 	dev->if_snd.ifq_maxlen = ipoib_sendq_size * 2;
882 	sdl = (struct sockaddr_dl *)dev->if_addr->ifa_addr;
883 	sdl->sdl_type = IFT_INFINIBAND;
884 	sdl->sdl_alen = dev->if_addrlen;
885 	priv->dev = dev;
886 	if_link_state_change(dev, LINK_STATE_DOWN);
887 	bpfattach(dev, DLT_EN10MB, ETHER_HDR_LEN);
888 
889 	return dev->if_softc;
890 }
891 
892 int
893 ipoib_set_dev_features(struct ipoib_dev_priv *priv, struct ib_device *hca)
894 {
895 	struct ib_device_attr *device_attr;
896 	int result = -ENOMEM;
897 
898 	device_attr = kmalloc(sizeof *device_attr, GFP_KERNEL);
899 	if (!device_attr) {
900 		printk(KERN_WARNING "%s: allocation of %zu bytes failed\n",
901 		       hca->name, sizeof *device_attr);
902 		return result;
903 	}
904 
905 	result = ib_query_device(hca, device_attr);
906 	if (result) {
907 		printk(KERN_WARNING "%s: ib_query_device failed (ret = %d)\n",
908 		       hca->name, result);
909 		kfree(device_attr);
910 		return result;
911 	}
912 	priv->hca_caps = device_attr->device_cap_flags;
913 
914 	kfree(device_attr);
915 
916 	priv->dev->if_hwassist = 0;
917 	priv->dev->if_capabilities = 0;
918 
919 #ifndef CONFIG_INFINIBAND_IPOIB_CM
920 	if (priv->hca_caps & IB_DEVICE_UD_IP_CSUM) {
921 		set_bit(IPOIB_FLAG_CSUM, &priv->flags);
922 		priv->dev->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP;
923 		priv->dev->if_capabilities = IFCAP_HWCSUM | IFCAP_VLAN_HWCSUM;
924 	}
925 
926 #if 0
927 	if (priv->dev->features & NETIF_F_SG && priv->hca_caps & IB_DEVICE_UD_TSO) {
928 		priv->dev->if_capabilities |= IFCAP_TSO4;
929 		priv->dev->if_hwassist |= CSUM_TSO;
930 	}
931 #endif
932 #endif
933 	priv->dev->if_capabilities |=
934 	    IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_LINKSTATE;
935 	priv->dev->if_capenable = priv->dev->if_capabilities;
936 
937 	return 0;
938 }
939 
940 
941 static struct ifnet *
942 ipoib_add_port(const char *format, struct ib_device *hca, u8 port)
943 {
944 	struct ipoib_dev_priv *priv;
945 	struct ib_port_attr attr;
946 	int result = -ENOMEM;
947 
948 	priv = ipoib_intf_alloc(format);
949 	if (!priv)
950 		goto alloc_mem_failed;
951 
952 	if (!ib_query_port(hca, port, &attr))
953 		priv->max_ib_mtu = ib_mtu_enum_to_int(attr.max_mtu);
954 	else {
955 		printk(KERN_WARNING "%s: ib_query_port %d failed\n",
956 		       hca->name, port);
957 		goto device_init_failed;
958 	}
959 
960 	/* MTU will be reset when mcast join happens */
961 	priv->dev->if_mtu = IPOIB_UD_MTU(priv->max_ib_mtu);
962 	priv->mcast_mtu = priv->admin_mtu = priv->dev->if_mtu;
963 
964 	result = ib_query_pkey(hca, port, 0, &priv->pkey);
965 	if (result) {
966 		printk(KERN_WARNING "%s: ib_query_pkey port %d failed (ret = %d)\n",
967 		       hca->name, port, result);
968 		goto device_init_failed;
969 	}
970 
971 	if (ipoib_set_dev_features(priv, hca))
972 		goto device_init_failed;
973 
974 	/*
975 	 * Set the full membership bit, so that we join the right
976 	 * broadcast group, etc.
977 	 */
978 	priv->pkey |= 0x8000;
979 
980 	priv->broadcastaddr[8] = priv->pkey >> 8;
981 	priv->broadcastaddr[9] = priv->pkey & 0xff;
982 
983 	result = ib_query_gid(hca, port, 0, &priv->local_gid);
984 	if (result) {
985 		printk(KERN_WARNING "%s: ib_query_gid port %d failed (ret = %d)\n",
986 		       hca->name, port, result);
987 		goto device_init_failed;
988 	}
989 	memcpy(IF_LLADDR(priv->dev) + 4, priv->local_gid.raw, sizeof (union ib_gid));
990 
991 	result = ipoib_dev_init(priv, hca, port);
992 	if (result < 0) {
993 		printk(KERN_WARNING "%s: failed to initialize port %d (ret = %d)\n",
994 		       hca->name, port, result);
995 		goto device_init_failed;
996 	}
997 	if (ipoib_cm_admin_enabled(priv))
998 		priv->dev->if_mtu = IPOIB_CM_MTU(ipoib_cm_max_mtu(priv));
999 
1000 	INIT_IB_EVENT_HANDLER(&priv->event_handler,
1001 			      priv->ca, ipoib_event);
1002 	result = ib_register_event_handler(&priv->event_handler);
1003 	if (result < 0) {
1004 		printk(KERN_WARNING "%s: ib_register_event_handler failed for "
1005 		       "port %d (ret = %d)\n",
1006 		       hca->name, port, result);
1007 		goto event_failed;
1008 	}
1009 	if_printf(priv->dev, "Attached to %s port %d\n", hca->name, port);
1010 
1011 	return priv->dev;
1012 
1013 event_failed:
1014 	ipoib_dev_cleanup(priv);
1015 
1016 device_init_failed:
1017 	ipoib_detach(priv);
1018 
1019 alloc_mem_failed:
1020 	return ERR_PTR(result);
1021 }
1022 
1023 static void
1024 ipoib_add_one(struct ib_device *device)
1025 {
1026 	struct list_head *dev_list;
1027 	struct ifnet *dev;
1028 	struct ipoib_dev_priv *priv;
1029 	int s, e, p;
1030 
1031 	if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
1032 		return;
1033 
1034 	dev_list = kmalloc(sizeof *dev_list, GFP_KERNEL);
1035 	if (!dev_list)
1036 		return;
1037 
1038 	INIT_LIST_HEAD(dev_list);
1039 
1040 	if (device->node_type == RDMA_NODE_IB_SWITCH) {
1041 		s = 0;
1042 		e = 0;
1043 	} else {
1044 		s = 1;
1045 		e = device->phys_port_cnt;
1046 	}
1047 
1048 	for (p = s; p <= e; ++p) {
1049 		if (rdma_port_get_link_layer(device, p) != IB_LINK_LAYER_INFINIBAND)
1050 			continue;
1051 		dev = ipoib_add_port("ib", device, p);
1052 		if (!IS_ERR(dev)) {
1053 			priv = dev->if_softc;
1054 			list_add_tail(&priv->list, dev_list);
1055 		}
1056 	}
1057 
1058 	ib_set_client_data(device, &ipoib_client, dev_list);
1059 }
1060 
1061 static void
1062 ipoib_remove_one(struct ib_device *device)
1063 {
1064 	struct ipoib_dev_priv *priv, *tmp;
1065 	struct list_head *dev_list;
1066 
1067 	if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
1068 		return;
1069 
1070 	dev_list = ib_get_client_data(device, &ipoib_client);
1071 
1072 	list_for_each_entry_safe(priv, tmp, dev_list, list) {
1073 		if (rdma_port_get_link_layer(device, priv->port) != IB_LINK_LAYER_INFINIBAND)
1074 			continue;
1075 
1076 		ipoib_stop(priv);
1077 
1078 		ib_unregister_event_handler(&priv->event_handler);
1079 
1080 		/* dev_change_flags(priv->dev, priv->dev->flags & ~IFF_UP); */
1081 
1082 		flush_workqueue(ipoib_workqueue);
1083 
1084 		ipoib_dev_cleanup(priv);
1085 		ipoib_detach(priv);
1086 	}
1087 
1088 	kfree(dev_list);
1089 }
1090 
1091 static void
1092 ipoib_config_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
1093 {
1094 	struct ipoib_dev_priv *parent;
1095 	struct ipoib_dev_priv *priv;
1096 	struct ifnet *dev;
1097 	uint16_t pkey;
1098 	int error;
1099 
1100 	if (ifp->if_type != IFT_INFINIBAND)
1101 		return;
1102 	dev = VLAN_DEVAT(ifp, vtag);
1103 	if (dev == NULL)
1104 		return;
1105 	priv = NULL;
1106 	error = 0;
1107 	parent = ifp->if_softc;
1108 	/* We only support 15 bits of pkey. */
1109 	if (vtag & 0x8000)
1110 		return;
1111 	pkey = vtag | 0x8000;	/* Set full membership bit. */
1112 	if (pkey == parent->pkey)
1113 		return;
1114 	/* Check for dups */
1115 	mutex_lock(&parent->vlan_mutex);
1116 	list_for_each_entry(priv, &parent->child_intfs, list) {
1117 		if (priv->pkey == pkey) {
1118 			priv = NULL;
1119 			error = EBUSY;
1120 			goto out;
1121 		}
1122 	}
1123 	priv = ipoib_priv_alloc();
1124 	priv->dev = dev;
1125 	priv->max_ib_mtu = parent->max_ib_mtu;
1126 	priv->mcast_mtu = priv->admin_mtu = parent->dev->if_mtu;
1127 	set_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags);
1128 	error = ipoib_set_dev_features(priv, parent->ca);
1129 	if (error)
1130 		goto out;
1131 	priv->pkey = pkey;
1132 	priv->broadcastaddr[8] = pkey >> 8;
1133 	priv->broadcastaddr[9] = pkey & 0xff;
1134 	dev->if_broadcastaddr = priv->broadcastaddr;
1135 	error = ipoib_dev_init(priv, parent->ca, parent->port);
1136 	if (error)
1137 		goto out;
1138 	priv->parent = parent->dev;
1139 	list_add_tail(&priv->list, &parent->child_intfs);
1140 	VLAN_SETCOOKIE(dev, priv);
1141 	dev->if_start = ipoib_vlan_start;
1142 	dev->if_drv_flags &= ~IFF_DRV_RUNNING;
1143 	dev->if_hdrlen = IPOIB_HEADER_LEN;
1144 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1145 		ipoib_open(priv);
1146 	mutex_unlock(&parent->vlan_mutex);
1147 	return;
1148 out:
1149 	mutex_unlock(&parent->vlan_mutex);
1150 	if (priv)
1151 		free(priv, M_TEMP);
1152 	if (error)
1153 		ipoib_warn(parent,
1154 		    "failed to initialize subinterface: device %s, port %d vtag 0x%X",
1155 		    parent->ca->name, parent->port, vtag);
1156 	return;
1157 }
1158 
1159 static void
1160 ipoib_unconfig_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
1161 {
1162 	struct ipoib_dev_priv *parent;
1163 	struct ipoib_dev_priv *priv;
1164 	struct ifnet *dev;
1165 	uint16_t pkey;
1166 
1167 	if (ifp->if_type != IFT_INFINIBAND)
1168 		return;
1169 
1170 	dev = VLAN_DEVAT(ifp, vtag);
1171 	if (dev)
1172 		VLAN_SETCOOKIE(dev, NULL);
1173 	pkey = vtag | 0x8000;
1174 	parent = ifp->if_softc;
1175 	mutex_lock(&parent->vlan_mutex);
1176 	list_for_each_entry(priv, &parent->child_intfs, list) {
1177 		if (priv->pkey == pkey) {
1178 			ipoib_dev_cleanup(priv);
1179 			list_del(&priv->list);
1180 			break;
1181 		}
1182 	}
1183 	mutex_unlock(&parent->vlan_mutex);
1184 }
1185 
1186 eventhandler_tag ipoib_vlan_attach;
1187 eventhandler_tag ipoib_vlan_detach;
1188 
1189 static int __init
1190 ipoib_init_module(void)
1191 {
1192 	int ret;
1193 
1194 	ipoib_recvq_size = roundup_pow_of_two(ipoib_recvq_size);
1195 	ipoib_recvq_size = min(ipoib_recvq_size, IPOIB_MAX_QUEUE_SIZE);
1196 	ipoib_recvq_size = max(ipoib_recvq_size, IPOIB_MIN_QUEUE_SIZE);
1197 
1198 	ipoib_sendq_size = roundup_pow_of_two(ipoib_sendq_size);
1199 	ipoib_sendq_size = min(ipoib_sendq_size, IPOIB_MAX_QUEUE_SIZE);
1200 	ipoib_sendq_size = max(ipoib_sendq_size, max(2 * MAX_SEND_CQE,
1201 						     IPOIB_MIN_QUEUE_SIZE));
1202 #ifdef CONFIG_INFINIBAND_IPOIB_CM
1203 	ipoib_max_conn_qp = min(ipoib_max_conn_qp, IPOIB_CM_MAX_CONN_QP);
1204 #endif
1205 
1206 	ipoib_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
1207 		ipoib_config_vlan, NULL, EVENTHANDLER_PRI_FIRST);
1208 	ipoib_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1209 		ipoib_unconfig_vlan, NULL, EVENTHANDLER_PRI_FIRST);
1210 
1211 	/*
1212 	 * We create our own workqueue mainly because we want to be
1213 	 * able to flush it when devices are being removed.  We can't
1214 	 * use schedule_work()/flush_scheduled_work() because both
1215 	 * unregister_netdev() and linkwatch_event take the rtnl lock,
1216 	 * so flush_scheduled_work() can deadlock during device
1217 	 * removal.
1218 	 */
1219 	ipoib_workqueue = create_singlethread_workqueue("ipoib");
1220 	if (!ipoib_workqueue) {
1221 		ret = -ENOMEM;
1222 		goto err_fs;
1223 	}
1224 
1225 	ib_sa_register_client(&ipoib_sa_client);
1226 
1227 	ret = ib_register_client(&ipoib_client);
1228 	if (ret)
1229 		goto err_sa;
1230 
1231 	return 0;
1232 
1233 err_sa:
1234 	ib_sa_unregister_client(&ipoib_sa_client);
1235 	destroy_workqueue(ipoib_workqueue);
1236 
1237 err_fs:
1238 	return ret;
1239 }
1240 
1241 static void __exit
1242 ipoib_cleanup_module(void)
1243 {
1244 
1245 	EVENTHANDLER_DEREGISTER(vlan_config, ipoib_vlan_attach);
1246 	EVENTHANDLER_DEREGISTER(vlan_unconfig, ipoib_vlan_detach);
1247 	ib_unregister_client(&ipoib_client);
1248 	ib_sa_unregister_client(&ipoib_sa_client);
1249 	destroy_workqueue(ipoib_workqueue);
1250 }
1251 
1252 /*
1253  * Infiniband output routine.
1254  */
1255 static int
1256 ipoib_output(struct ifnet *ifp, struct mbuf *m,
1257 	const struct sockaddr *dst, struct route *ro)
1258 {
1259 	u_char edst[INFINIBAND_ALEN];
1260 #if defined(INET) || defined(INET6)
1261 	struct llentry *lle = NULL;
1262 #endif
1263 	struct ipoib_header *eh;
1264 	int error = 0, is_gw = 0;
1265 	short type;
1266 
1267 	if (ro != NULL)
1268 		is_gw = (ro->ro_flags & RT_HAS_GW) != 0;
1269 #ifdef MAC
1270 	error = mac_ifnet_check_transmit(ifp, m);
1271 	if (error)
1272 		goto bad;
1273 #endif
1274 
1275 	M_PROFILE(m);
1276 	if (ifp->if_flags & IFF_MONITOR) {
1277 		error = ENETDOWN;
1278 		goto bad;
1279 	}
1280 	if (!((ifp->if_flags & IFF_UP) &&
1281 	    (ifp->if_drv_flags & IFF_DRV_RUNNING))) {
1282 		error = ENETDOWN;
1283 		goto bad;
1284 	}
1285 
1286 	switch (dst->sa_family) {
1287 #ifdef INET
1288 	case AF_INET:
1289 		if (lle != NULL && (lle->la_flags & LLE_VALID))
1290 			memcpy(edst, lle->ll_addr, sizeof(edst));
1291 		else if (m->m_flags & M_MCAST)
1292 			ip_ib_mc_map(((struct sockaddr_in *)dst)->sin_addr.s_addr, ifp->if_broadcastaddr, edst);
1293 		else
1294 			error = arpresolve(ifp, is_gw, m, dst, edst, NULL);
1295 		if (error)
1296 			return (error == EWOULDBLOCK ? 0 : error);
1297 		type = htons(ETHERTYPE_IP);
1298 		break;
1299 	case AF_ARP:
1300 	{
1301 		struct arphdr *ah;
1302 		ah = mtod(m, struct arphdr *);
1303 		ah->ar_hrd = htons(ARPHRD_INFINIBAND);
1304 
1305 		switch(ntohs(ah->ar_op)) {
1306 		case ARPOP_REVREQUEST:
1307 		case ARPOP_REVREPLY:
1308 			type = htons(ETHERTYPE_REVARP);
1309 			break;
1310 		case ARPOP_REQUEST:
1311 		case ARPOP_REPLY:
1312 		default:
1313 			type = htons(ETHERTYPE_ARP);
1314 			break;
1315 		}
1316 
1317 		if (m->m_flags & M_BCAST)
1318 			bcopy(ifp->if_broadcastaddr, edst, INFINIBAND_ALEN);
1319 		else
1320 			bcopy(ar_tha(ah), edst, INFINIBAND_ALEN);
1321 
1322 	}
1323 	break;
1324 #endif
1325 #ifdef INET6
1326 	case AF_INET6:
1327 		if (lle != NULL && (lle->la_flags & LLE_VALID))
1328 			memcpy(edst, lle->ll_addr, sizeof(edst));
1329 		else if (m->m_flags & M_MCAST)
1330 			ipv6_ib_mc_map(&((struct sockaddr_in6 *)dst)->sin6_addr, ifp->if_broadcastaddr, edst);
1331 		else
1332 			error = nd6_resolve(ifp, is_gw, m, dst, edst, NULL);
1333 		if (error)
1334 			return error;
1335 		type = htons(ETHERTYPE_IPV6);
1336 		break;
1337 #endif
1338 
1339 	default:
1340 		if_printf(ifp, "can't handle af%d\n", dst->sa_family);
1341 		error = EAFNOSUPPORT;
1342 		goto bad;
1343 	}
1344 
1345 	/*
1346 	 * Add local net header.  If no space in first mbuf,
1347 	 * allocate another.
1348 	 */
1349 	M_PREPEND(m, IPOIB_HEADER_LEN, M_NOWAIT);
1350 	if (m == NULL) {
1351 		error = ENOBUFS;
1352 		goto bad;
1353 	}
1354 	eh = mtod(m, struct ipoib_header *);
1355 	(void)memcpy(&eh->proto, &type, sizeof(eh->proto));
1356 	(void)memcpy(&eh->hwaddr, edst, sizeof (edst));
1357 
1358 	/*
1359 	 * Queue message on interface, update output statistics if
1360 	 * successful, and start output if interface not yet active.
1361 	 */
1362 	return ((ifp->if_transmit)(ifp, m));
1363 bad:
1364 	if (m != NULL)
1365 		m_freem(m);
1366 	return (error);
1367 }
1368 
1369 /*
1370  * Upper layer processing for a received Infiniband packet.
1371  */
1372 void
1373 ipoib_demux(struct ifnet *ifp, struct mbuf *m, u_short proto)
1374 {
1375 	int isr;
1376 
1377 #ifdef MAC
1378 	/*
1379 	 * Tag the mbuf with an appropriate MAC label before any other
1380 	 * consumers can get to it.
1381 	 */
1382 	mac_ifnet_create_mbuf(ifp, m);
1383 #endif
1384 	/* Allow monitor mode to claim this frame, after stats are updated. */
1385 	if (ifp->if_flags & IFF_MONITOR) {
1386 		if_printf(ifp, "discard frame at IFF_MONITOR\n");
1387 		m_freem(m);
1388 		return;
1389 	}
1390 	/*
1391 	 * Dispatch frame to upper layer.
1392 	 */
1393 	switch (proto) {
1394 #ifdef INET
1395 	case ETHERTYPE_IP:
1396 		isr = NETISR_IP;
1397 		break;
1398 
1399 	case ETHERTYPE_ARP:
1400 		if (ifp->if_flags & IFF_NOARP) {
1401 			/* Discard packet if ARP is disabled on interface */
1402 			m_freem(m);
1403 			return;
1404 		}
1405 		isr = NETISR_ARP;
1406 		break;
1407 #endif
1408 #ifdef INET6
1409 	case ETHERTYPE_IPV6:
1410 		isr = NETISR_IPV6;
1411 		break;
1412 #endif
1413 	default:
1414 		goto discard;
1415 	}
1416 	netisr_dispatch(isr, m);
1417 	return;
1418 
1419 discard:
1420 	m_freem(m);
1421 }
1422 
1423 /*
1424  * Process a received Infiniband packet.
1425  */
1426 static void
1427 ipoib_input(struct ifnet *ifp, struct mbuf *m)
1428 {
1429 	struct ipoib_header *eh;
1430 
1431 	if ((ifp->if_flags & IFF_UP) == 0) {
1432 		m_freem(m);
1433 		return;
1434 	}
1435 	CURVNET_SET_QUIET(ifp->if_vnet);
1436 
1437 	/* Let BPF have it before we strip the header. */
1438 	IPOIB_MTAP(ifp, m);
1439 	eh = mtod(m, struct ipoib_header *);
1440 	/*
1441 	 * Reset layer specific mbuf flags to avoid confusing upper layers.
1442 	 * Strip off Infiniband header.
1443 	 */
1444 	m->m_flags &= ~M_VLANTAG;
1445 	m_clrprotoflags(m);
1446 	m_adj(m, IPOIB_HEADER_LEN);
1447 
1448 	if (IPOIB_IS_MULTICAST(eh->hwaddr)) {
1449 		if (memcmp(eh->hwaddr, ifp->if_broadcastaddr,
1450 		    ifp->if_addrlen) == 0)
1451 			m->m_flags |= M_BCAST;
1452 		else
1453 			m->m_flags |= M_MCAST;
1454 		if_inc_counter(ifp, IFCOUNTER_IMCASTS, 1);
1455 	}
1456 
1457 	ipoib_demux(ifp, m, ntohs(eh->proto));
1458 	CURVNET_RESTORE();
1459 }
1460 
1461 static int
1462 ipoib_resolvemulti(struct ifnet *ifp, struct sockaddr **llsa,
1463 	struct sockaddr *sa)
1464 {
1465 	struct sockaddr_dl *sdl;
1466 #ifdef INET
1467 	struct sockaddr_in *sin;
1468 #endif
1469 #ifdef INET6
1470 	struct sockaddr_in6 *sin6;
1471 #endif
1472 	u_char *e_addr;
1473 
1474 	switch(sa->sa_family) {
1475 	case AF_LINK:
1476 		/*
1477 		 * No mapping needed. Just check that it's a valid MC address.
1478 		 */
1479 		sdl = (struct sockaddr_dl *)sa;
1480 		e_addr = LLADDR(sdl);
1481 		if (!IPOIB_IS_MULTICAST(e_addr))
1482 			return EADDRNOTAVAIL;
1483 		*llsa = 0;
1484 		return 0;
1485 
1486 #ifdef INET
1487 	case AF_INET:
1488 		sin = (struct sockaddr_in *)sa;
1489 		if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
1490 			return EADDRNOTAVAIL;
1491 		sdl = link_init_sdl(ifp, *llsa, IFT_INFINIBAND);
1492 		sdl->sdl_alen = INFINIBAND_ALEN;
1493 		e_addr = LLADDR(sdl);
1494 		ip_ib_mc_map(sin->sin_addr.s_addr, ifp->if_broadcastaddr,
1495 		    e_addr);
1496 		*llsa = (struct sockaddr *)sdl;
1497 		return 0;
1498 #endif
1499 #ifdef INET6
1500 	case AF_INET6:
1501 		sin6 = (struct sockaddr_in6 *)sa;
1502 		/*
1503 		 * An IP6 address of 0 means listen to all
1504 		 * of the multicast address used for IP6.
1505 		 * This has no meaning in ipoib.
1506 		 */
1507 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
1508 			return EADDRNOTAVAIL;
1509 		if (!IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
1510 			return EADDRNOTAVAIL;
1511 		sdl = link_init_sdl(ifp, *llsa, IFT_INFINIBAND);
1512 		sdl->sdl_alen = INFINIBAND_ALEN;
1513 		e_addr = LLADDR(sdl);
1514 		ipv6_ib_mc_map(&sin6->sin6_addr, ifp->if_broadcastaddr, e_addr);
1515 		*llsa = (struct sockaddr *)sdl;
1516 		return 0;
1517 #endif
1518 
1519 	default:
1520 		return EAFNOSUPPORT;
1521 	}
1522 }
1523 
1524 module_init(ipoib_init_module);
1525 module_exit(ipoib_cleanup_module);
1526 
1527 static int
1528 ipoib_evhand(module_t mod, int event, void *arg)
1529 {
1530 	                return (0);
1531 }
1532 
1533 static moduledata_t ipoib_mod = {
1534 	                .name = "ipoib",
1535 			                .evhand = ipoib_evhand,
1536 };
1537 
1538 DECLARE_MODULE(ipoib, ipoib_mod, SI_SUB_SMP, SI_ORDER_ANY);
1539 MODULE_DEPEND(ipoib, ibcore, 1, 1, 1);
1540 MODULE_DEPEND(ipoib, linuxkpi, 1, 1, 1);
1541