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 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
ipoib_mtap_mb(struct ifnet * ifp,struct mbuf * mb)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
ipoib_mtap_proto(struct ifnet * ifp,struct mbuf * mb,uint16_t proto)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
ipoib_open(struct ipoib_dev_priv * priv)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
ipoib_init(void * arg)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
ipoib_stop(struct ipoib_dev_priv * priv)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
ipoib_change_mtu(struct ipoib_dev_priv * priv,int new_mtu)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
ipoib_ioctl(struct ifnet * ifp,u_long command,caddr_t data)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 /* check if detaching */
262 if (priv == NULL || priv->gone != 0)
263 return (ENXIO);
264
265 switch (command) {
266 case SIOCSIFFLAGS:
267 if (ifp->if_flags & IFF_UP) {
268 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
269 error = -ipoib_open(priv);
270 } else
271 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
272 ipoib_stop(priv);
273 break;
274 case SIOCADDMULTI:
275 case SIOCDELMULTI:
276 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
277 queue_work(ipoib_workqueue, &priv->restart_task);
278 break;
279 case SIOCSIFADDR:
280 ifp->if_flags |= IFF_UP;
281
282 switch (ifa->ifa_addr->sa_family) {
283 #ifdef INET
284 case AF_INET:
285 ifp->if_init(ifp->if_softc); /* before arpwhohas */
286 arp_ifinit(ifp, ifa);
287 break;
288 #endif
289 default:
290 ifp->if_init(ifp->if_softc);
291 break;
292 }
293 break;
294
295 case SIOCGIFADDR:
296 {
297 struct sockaddr *sa;
298
299 sa = (struct sockaddr *) & ifr->ifr_data;
300 bcopy(IF_LLADDR(ifp),
301 (caddr_t) sa->sa_data, INFINIBAND_ALEN);
302 }
303 break;
304
305 case SIOCSIFMTU:
306 /*
307 * Set the interface MTU.
308 */
309 error = -ipoib_change_mtu(priv, ifr->ifr_mtu);
310 break;
311 default:
312 error = EINVAL;
313 break;
314 }
315 return (error);
316 }
317
318
319 static struct ipoib_path *
__path_find(struct ipoib_dev_priv * priv,void * gid)320 __path_find(struct ipoib_dev_priv *priv, void *gid)
321 {
322 struct rb_node *n = priv->path_tree.rb_node;
323 struct ipoib_path *path;
324 int ret;
325
326 while (n) {
327 path = rb_entry(n, struct ipoib_path, rb_node);
328
329 ret = memcmp(gid, path->pathrec.dgid.raw,
330 sizeof (union ib_gid));
331
332 if (ret < 0)
333 n = n->rb_left;
334 else if (ret > 0)
335 n = n->rb_right;
336 else
337 return path;
338 }
339
340 return NULL;
341 }
342
343 static int
__path_add(struct ipoib_dev_priv * priv,struct ipoib_path * path)344 __path_add(struct ipoib_dev_priv *priv, struct ipoib_path *path)
345 {
346 struct rb_node **n = &priv->path_tree.rb_node;
347 struct rb_node *pn = NULL;
348 struct ipoib_path *tpath;
349 int ret;
350
351 while (*n) {
352 pn = *n;
353 tpath = rb_entry(pn, struct ipoib_path, rb_node);
354
355 ret = memcmp(path->pathrec.dgid.raw, tpath->pathrec.dgid.raw,
356 sizeof (union ib_gid));
357 if (ret < 0)
358 n = &pn->rb_left;
359 else if (ret > 0)
360 n = &pn->rb_right;
361 else
362 return -EEXIST;
363 }
364
365 rb_link_node(&path->rb_node, pn, n);
366 rb_insert_color(&path->rb_node, &priv->path_tree);
367
368 list_add_tail(&path->list, &priv->path_list);
369
370 return 0;
371 }
372
373 void
ipoib_path_free(struct ipoib_dev_priv * priv,struct ipoib_path * path)374 ipoib_path_free(struct ipoib_dev_priv *priv, struct ipoib_path *path)
375 {
376
377 _IF_DRAIN(&path->queue);
378
379 if (path->ah)
380 ipoib_put_ah(path->ah);
381 if (ipoib_cm_get(path))
382 ipoib_cm_destroy_tx(ipoib_cm_get(path));
383
384 kfree(path);
385 }
386
387 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
388
389 struct ipoib_path_iter *
ipoib_path_iter_init(struct ipoib_dev_priv * priv)390 ipoib_path_iter_init(struct ipoib_dev_priv *priv)
391 {
392 struct ipoib_path_iter *iter;
393
394 iter = kmalloc(sizeof *iter, GFP_KERNEL);
395 if (!iter)
396 return NULL;
397
398 iter->priv = priv;
399 memset(iter->path.pathrec.dgid.raw, 0, 16);
400
401 if (ipoib_path_iter_next(iter)) {
402 kfree(iter);
403 return NULL;
404 }
405
406 return iter;
407 }
408
409 int
ipoib_path_iter_next(struct ipoib_path_iter * iter)410 ipoib_path_iter_next(struct ipoib_path_iter *iter)
411 {
412 struct ipoib_dev_priv *priv = iter->priv;
413 struct rb_node *n;
414 struct ipoib_path *path;
415 int ret = 1;
416
417 spin_lock_irq(&priv->lock);
418
419 n = rb_first(&priv->path_tree);
420
421 while (n) {
422 path = rb_entry(n, struct ipoib_path, rb_node);
423
424 if (memcmp(iter->path.pathrec.dgid.raw, path->pathrec.dgid.raw,
425 sizeof (union ib_gid)) < 0) {
426 iter->path = *path;
427 ret = 0;
428 break;
429 }
430
431 n = rb_next(n);
432 }
433
434 spin_unlock_irq(&priv->lock);
435
436 return ret;
437 }
438
439 void
ipoib_path_iter_read(struct ipoib_path_iter * iter,struct ipoib_path * path)440 ipoib_path_iter_read(struct ipoib_path_iter *iter, struct ipoib_path *path)
441 {
442 *path = iter->path;
443 }
444
445 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
446
447 void
ipoib_mark_paths_invalid(struct ipoib_dev_priv * priv)448 ipoib_mark_paths_invalid(struct ipoib_dev_priv *priv)
449 {
450 struct ipoib_path *path, *tp;
451
452 spin_lock_irq(&priv->lock);
453
454 list_for_each_entry_safe(path, tp, &priv->path_list, list) {
455 ipoib_dbg(priv, "mark path LID 0x%04x GID %16D invalid\n",
456 be16_to_cpu(path->pathrec.dlid),
457 path->pathrec.dgid.raw, ":");
458 path->valid = 0;
459 }
460
461 spin_unlock_irq(&priv->lock);
462 }
463
464 void
ipoib_flush_paths(struct ipoib_dev_priv * priv)465 ipoib_flush_paths(struct ipoib_dev_priv *priv)
466 {
467 struct ipoib_path *path, *tp;
468 LIST_HEAD(remove_list);
469 unsigned long flags;
470
471 spin_lock_irqsave(&priv->lock, flags);
472
473 list_splice_init(&priv->path_list, &remove_list);
474
475 list_for_each_entry(path, &remove_list, list)
476 rb_erase(&path->rb_node, &priv->path_tree);
477
478 list_for_each_entry_safe(path, tp, &remove_list, list) {
479 if (path->query)
480 ib_sa_cancel_query(path->query_id, path->query);
481 spin_unlock_irqrestore(&priv->lock, flags);
482 wait_for_completion(&path->done);
483 ipoib_path_free(priv, path);
484 spin_lock_irqsave(&priv->lock, flags);
485 }
486
487 spin_unlock_irqrestore(&priv->lock, flags);
488 }
489
490 static void
path_rec_completion(int status,struct ib_sa_path_rec * pathrec,void * path_ptr)491 path_rec_completion(int status, struct ib_sa_path_rec *pathrec, void *path_ptr)
492 {
493 struct ipoib_path *path = path_ptr;
494 struct ipoib_dev_priv *priv = path->priv;
495 struct ifnet *dev = priv->dev;
496 struct ipoib_ah *ah = NULL;
497 struct ipoib_ah *old_ah = NULL;
498 struct ifqueue mbqueue;
499 struct mbuf *mb;
500 unsigned long flags;
501
502 if (!status)
503 ipoib_dbg(priv, "PathRec LID 0x%04x for GID %16D\n",
504 be16_to_cpu(pathrec->dlid), pathrec->dgid.raw, ":");
505 else
506 ipoib_dbg(priv, "PathRec status %d for GID %16D\n",
507 status, path->pathrec.dgid.raw, ":");
508
509 bzero(&mbqueue, sizeof(mbqueue));
510
511 if (!status) {
512 struct ib_ah_attr av;
513
514 if (!ib_init_ah_from_path(priv->ca, priv->port, pathrec, &av))
515 ah = ipoib_create_ah(priv, priv->pd, &av);
516 }
517
518 spin_lock_irqsave(&priv->lock, flags);
519
520 if (ah) {
521 path->pathrec = *pathrec;
522
523 old_ah = path->ah;
524 path->ah = ah;
525
526 ipoib_dbg(priv, "created address handle %p for LID 0x%04x, SL %d\n",
527 ah, be16_to_cpu(pathrec->dlid), pathrec->sl);
528
529 for (;;) {
530 _IF_DEQUEUE(&path->queue, mb);
531 if (mb == NULL)
532 break;
533 _IF_ENQUEUE(&mbqueue, mb);
534 }
535
536 #ifdef CONFIG_INFINIBAND_IPOIB_CM
537 if (ipoib_cm_enabled(priv, path->hwaddr) && !ipoib_cm_get(path))
538 ipoib_cm_set(path, ipoib_cm_create_tx(priv, path));
539 #endif
540
541 path->valid = 1;
542 }
543
544 path->query = NULL;
545 complete(&path->done);
546
547 spin_unlock_irqrestore(&priv->lock, flags);
548
549 if (old_ah)
550 ipoib_put_ah(old_ah);
551
552 for (;;) {
553 _IF_DEQUEUE(&mbqueue, mb);
554 if (mb == NULL)
555 break;
556 mb->m_pkthdr.rcvif = dev;
557 if (dev->if_transmit(dev, mb))
558 ipoib_warn(priv, "dev_queue_xmit failed "
559 "to requeue packet\n");
560 }
561 }
562
563 static struct ipoib_path *
path_rec_create(struct ipoib_dev_priv * priv,uint8_t * hwaddr)564 path_rec_create(struct ipoib_dev_priv *priv, uint8_t *hwaddr)
565 {
566 struct ipoib_path *path;
567
568 if (!priv->broadcast)
569 return NULL;
570
571 path = kzalloc(sizeof *path, GFP_ATOMIC);
572 if (!path)
573 return NULL;
574
575 path->priv = priv;
576
577 bzero(&path->queue, sizeof(path->queue));
578
579 #ifdef CONFIG_INFINIBAND_IPOIB_CM
580 memcpy(&path->hwaddr, hwaddr, INFINIBAND_ALEN);
581 #endif
582 memcpy(path->pathrec.dgid.raw, &hwaddr[4], sizeof (union ib_gid));
583 path->pathrec.sgid = priv->local_gid;
584 path->pathrec.pkey = cpu_to_be16(priv->pkey);
585 path->pathrec.numb_path = 1;
586 path->pathrec.traffic_class = priv->broadcast->mcmember.traffic_class;
587
588 return path;
589 }
590
591 static int
path_rec_start(struct ipoib_dev_priv * priv,struct ipoib_path * path)592 path_rec_start(struct ipoib_dev_priv *priv, struct ipoib_path *path)
593 {
594 struct ifnet *dev = priv->dev;
595
596 ib_sa_comp_mask comp_mask = IB_SA_PATH_REC_MTU_SELECTOR | IB_SA_PATH_REC_MTU;
597 struct ib_sa_path_rec p_rec;
598
599 p_rec = path->pathrec;
600 p_rec.mtu_selector = IB_SA_GT;
601
602 switch (roundup_pow_of_two(dev->if_mtu + IPOIB_ENCAP_LEN)) {
603 case 512:
604 p_rec.mtu = IB_MTU_256;
605 break;
606 case 1024:
607 p_rec.mtu = IB_MTU_512;
608 break;
609 case 2048:
610 p_rec.mtu = IB_MTU_1024;
611 break;
612 case 4096:
613 p_rec.mtu = IB_MTU_2048;
614 break;
615 default:
616 /* Wildcard everything */
617 comp_mask = 0;
618 p_rec.mtu = 0;
619 p_rec.mtu_selector = 0;
620 }
621
622 ipoib_dbg(priv, "Start path record lookup for %16D MTU > %d\n",
623 p_rec.dgid.raw, ":",
624 comp_mask ? ib_mtu_enum_to_int(p_rec.mtu) : 0);
625
626 init_completion(&path->done);
627
628 path->query_id =
629 ib_sa_path_rec_get(&ipoib_sa_client, priv->ca, priv->port,
630 &p_rec, comp_mask |
631 IB_SA_PATH_REC_DGID |
632 IB_SA_PATH_REC_SGID |
633 IB_SA_PATH_REC_NUMB_PATH |
634 IB_SA_PATH_REC_TRAFFIC_CLASS |
635 IB_SA_PATH_REC_PKEY,
636 1000, GFP_ATOMIC,
637 path_rec_completion,
638 path, &path->query);
639 if (path->query_id < 0) {
640 ipoib_warn(priv, "ib_sa_path_rec_get failed: %d\n", path->query_id);
641 path->query = NULL;
642 complete(&path->done);
643 return path->query_id;
644 }
645
646 return 0;
647 }
648
649 static void
ipoib_unicast_send(struct mbuf * mb,struct ipoib_dev_priv * priv,struct ipoib_header * eh)650 ipoib_unicast_send(struct mbuf *mb, struct ipoib_dev_priv *priv, struct ipoib_header *eh)
651 {
652 struct ipoib_path *path;
653
654 path = __path_find(priv, eh->hwaddr + 4);
655 if (!path || !path->valid) {
656 int new_path = 0;
657
658 if (!path) {
659 path = path_rec_create(priv, eh->hwaddr);
660 new_path = 1;
661 }
662 if (path) {
663 _IF_ENQUEUE(&path->queue, mb);
664 if (!path->query && path_rec_start(priv, path)) {
665 if (new_path)
666 ipoib_path_free(priv, path);
667 return;
668 } else
669 __path_add(priv, path);
670 } else {
671 ++priv->dev->if_oerrors;
672 m_freem(mb);
673 }
674
675 return;
676 }
677
678 if (ipoib_cm_get(path) && ipoib_cm_up(path)) {
679 ipoib_cm_send(priv, mb, ipoib_cm_get(path));
680 } else if (path->ah) {
681 ipoib_send(priv, mb, path->ah, IPOIB_QPN(eh->hwaddr));
682 } else if ((path->query || !path_rec_start(priv, path)) &&
683 path->queue.ifq_len < IPOIB_MAX_PATH_REC_QUEUE) {
684 _IF_ENQUEUE(&path->queue, mb);
685 } else {
686 ++priv->dev->if_oerrors;
687 m_freem(mb);
688 }
689 }
690
691 static int
ipoib_send_one(struct ipoib_dev_priv * priv,struct mbuf * mb)692 ipoib_send_one(struct ipoib_dev_priv *priv, struct mbuf *mb)
693 {
694 struct ipoib_header *eh;
695
696 eh = mtod(mb, struct ipoib_header *);
697 if (IPOIB_IS_MULTICAST(eh->hwaddr)) {
698 /* Add in the P_Key for multicast*/
699 eh->hwaddr[8] = (priv->pkey >> 8) & 0xff;
700 eh->hwaddr[9] = priv->pkey & 0xff;
701
702 ipoib_mcast_send(priv, eh->hwaddr + 4, mb);
703 } else
704 ipoib_unicast_send(mb, priv, eh);
705
706 return 0;
707 }
708
709
710 static void
_ipoib_start(struct ifnet * dev,struct ipoib_dev_priv * priv)711 _ipoib_start(struct ifnet *dev, struct ipoib_dev_priv *priv)
712 {
713 struct mbuf *mb;
714
715 if ((dev->if_drv_flags & (IFF_DRV_RUNNING|IFF_DRV_OACTIVE)) !=
716 IFF_DRV_RUNNING)
717 return;
718
719 spin_lock(&priv->lock);
720 while (!IFQ_DRV_IS_EMPTY(&dev->if_snd) &&
721 (dev->if_drv_flags & IFF_DRV_OACTIVE) == 0) {
722 IFQ_DRV_DEQUEUE(&dev->if_snd, mb);
723 if (mb == NULL)
724 break;
725 IPOIB_MTAP(dev, mb);
726 ipoib_send_one(priv, mb);
727 }
728 spin_unlock(&priv->lock);
729 }
730
731 static void
ipoib_start(struct ifnet * dev)732 ipoib_start(struct ifnet *dev)
733 {
734 _ipoib_start(dev, dev->if_softc);
735 }
736
737 static void
ipoib_vlan_start(struct ifnet * dev)738 ipoib_vlan_start(struct ifnet *dev)
739 {
740 struct ipoib_dev_priv *priv;
741 struct mbuf *mb;
742
743 priv = VLAN_COOKIE(dev);
744 if (priv != NULL)
745 return _ipoib_start(dev, priv);
746 while (!IFQ_DRV_IS_EMPTY(&dev->if_snd)) {
747 IFQ_DRV_DEQUEUE(&dev->if_snd, mb);
748 if (mb == NULL)
749 break;
750 m_freem(mb);
751 dev->if_oerrors++;
752 }
753 }
754
755 int
ipoib_dev_init(struct ipoib_dev_priv * priv,struct ib_device * ca,int port)756 ipoib_dev_init(struct ipoib_dev_priv *priv, struct ib_device *ca, int port)
757 {
758
759 /* Allocate RX/TX "rings" to hold queued mbs */
760 priv->rx_ring = kzalloc(ipoib_recvq_size * sizeof *priv->rx_ring,
761 GFP_KERNEL);
762 if (!priv->rx_ring) {
763 printk(KERN_WARNING "%s: failed to allocate RX ring (%d entries)\n",
764 ca->name, ipoib_recvq_size);
765 goto out;
766 }
767
768 priv->tx_ring = kzalloc(ipoib_sendq_size * sizeof *priv->tx_ring, GFP_KERNEL);
769 if (!priv->tx_ring) {
770 printk(KERN_WARNING "%s: failed to allocate TX ring (%d entries)\n",
771 ca->name, ipoib_sendq_size);
772 goto out_rx_ring_cleanup;
773 }
774 memset(priv->tx_ring, 0, ipoib_sendq_size * sizeof *priv->tx_ring);
775
776 /* priv->tx_head, tx_tail & tx_outstanding are already 0 */
777
778 if (ipoib_ib_dev_init(priv, ca, port))
779 goto out_tx_ring_cleanup;
780
781 return 0;
782
783 out_tx_ring_cleanup:
784 kfree(priv->tx_ring);
785
786 out_rx_ring_cleanup:
787 kfree(priv->rx_ring);
788
789 out:
790 return -ENOMEM;
791 }
792
793 static void
ipoib_detach(struct ipoib_dev_priv * priv)794 ipoib_detach(struct ipoib_dev_priv *priv)
795 {
796 struct ifnet *dev;
797
798 dev = priv->dev;
799 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
800 priv->gone = 1;
801 bpfdetach(dev);
802 if_detach(dev);
803 if_free(dev);
804 } else
805 VLAN_SETCOOKIE(priv->dev, NULL);
806
807 free(priv, M_TEMP);
808 }
809
810 void
ipoib_dev_cleanup(struct ipoib_dev_priv * priv)811 ipoib_dev_cleanup(struct ipoib_dev_priv *priv)
812 {
813 struct ipoib_dev_priv *cpriv, *tcpriv;
814
815 /* Delete any child interfaces first */
816 list_for_each_entry_safe(cpriv, tcpriv, &priv->child_intfs, list) {
817 ipoib_dev_cleanup(cpriv);
818 ipoib_detach(cpriv);
819 }
820
821 ipoib_ib_dev_cleanup(priv);
822
823 kfree(priv->rx_ring);
824 kfree(priv->tx_ring);
825
826 priv->rx_ring = NULL;
827 priv->tx_ring = NULL;
828 }
829
830 static volatile int ipoib_unit;
831
832 static struct ipoib_dev_priv *
ipoib_priv_alloc(void)833 ipoib_priv_alloc(void)
834 {
835 struct ipoib_dev_priv *priv;
836
837 priv = malloc(sizeof(struct ipoib_dev_priv), M_TEMP, M_ZERO|M_WAITOK);
838 spin_lock_init(&priv->lock);
839 mutex_init(&priv->vlan_mutex);
840 INIT_LIST_HEAD(&priv->path_list);
841 INIT_LIST_HEAD(&priv->child_intfs);
842 INIT_LIST_HEAD(&priv->dead_ahs);
843 INIT_LIST_HEAD(&priv->multicast_list);
844 INIT_DELAYED_WORK(&priv->pkey_poll_task, ipoib_pkey_poll);
845 INIT_DELAYED_WORK(&priv->mcast_task, ipoib_mcast_join_task);
846 INIT_WORK(&priv->carrier_on_task, ipoib_mcast_carrier_on_task);
847 INIT_WORK(&priv->flush_light, ipoib_ib_dev_flush_light);
848 INIT_WORK(&priv->flush_normal, ipoib_ib_dev_flush_normal);
849 INIT_WORK(&priv->flush_heavy, ipoib_ib_dev_flush_heavy);
850 INIT_WORK(&priv->restart_task, ipoib_mcast_restart_task);
851 INIT_DELAYED_WORK(&priv->ah_reap_task, ipoib_reap_ah);
852 memcpy(priv->broadcastaddr, ipv4_bcast_addr, INFINIBAND_ALEN);
853
854 return (priv);
855 }
856
857 struct ipoib_dev_priv *
ipoib_intf_alloc(const char * name)858 ipoib_intf_alloc(const char *name)
859 {
860 struct ipoib_dev_priv *priv;
861 struct sockaddr_dl *sdl;
862 struct ifnet *dev;
863
864 priv = ipoib_priv_alloc();
865 dev = priv->dev = if_alloc(IFT_INFINIBAND);
866 if (!dev) {
867 free(priv, M_TEMP);
868 return NULL;
869 }
870 dev->if_softc = priv;
871 if_initname(dev, name, atomic_fetchadd_int(&ipoib_unit, 1));
872 dev->if_flags = IFF_BROADCAST | IFF_MULTICAST;
873 dev->if_addrlen = INFINIBAND_ALEN;
874 dev->if_hdrlen = IPOIB_HEADER_LEN;
875 if_attach(dev);
876 dev->if_init = ipoib_init;
877 dev->if_ioctl = ipoib_ioctl;
878 dev->if_start = ipoib_start;
879 dev->if_output = ipoib_output;
880 dev->if_input = ipoib_input;
881 dev->if_resolvemulti = ipoib_resolvemulti;
882 dev->if_baudrate = IF_Gbps(10UL);
883 dev->if_broadcastaddr = priv->broadcastaddr;
884 dev->if_snd.ifq_maxlen = ipoib_sendq_size * 2;
885 sdl = (struct sockaddr_dl *)dev->if_addr->ifa_addr;
886 sdl->sdl_type = IFT_INFINIBAND;
887 sdl->sdl_alen = dev->if_addrlen;
888 priv->dev = dev;
889 if_link_state_change(dev, LINK_STATE_DOWN);
890 bpfattach(dev, DLT_EN10MB, ETHER_HDR_LEN);
891
892 return dev->if_softc;
893 }
894
895 int
ipoib_set_dev_features(struct ipoib_dev_priv * priv,struct ib_device * hca)896 ipoib_set_dev_features(struct ipoib_dev_priv *priv, struct ib_device *hca)
897 {
898 struct ib_device_attr *device_attr;
899 int result = -ENOMEM;
900
901 device_attr = kmalloc(sizeof *device_attr, GFP_KERNEL);
902 if (!device_attr) {
903 printk(KERN_WARNING "%s: allocation of %zu bytes failed\n",
904 hca->name, sizeof *device_attr);
905 return result;
906 }
907
908 result = ib_query_device(hca, device_attr);
909 if (result) {
910 printk(KERN_WARNING "%s: ib_query_device failed (ret = %d)\n",
911 hca->name, result);
912 kfree(device_attr);
913 return result;
914 }
915 priv->hca_caps = device_attr->device_cap_flags;
916
917 kfree(device_attr);
918
919 priv->dev->if_hwassist = 0;
920 priv->dev->if_capabilities = 0;
921
922 #ifndef CONFIG_INFINIBAND_IPOIB_CM
923 if (priv->hca_caps & IB_DEVICE_UD_IP_CSUM) {
924 set_bit(IPOIB_FLAG_CSUM, &priv->flags);
925 priv->dev->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP;
926 priv->dev->if_capabilities = IFCAP_HWCSUM | IFCAP_VLAN_HWCSUM;
927 }
928
929 #if 0
930 if (priv->dev->features & NETIF_F_SG && priv->hca_caps & IB_DEVICE_UD_TSO) {
931 priv->dev->if_capabilities |= IFCAP_TSO4;
932 priv->dev->if_hwassist |= CSUM_TSO;
933 }
934 #endif
935 #endif
936 priv->dev->if_capabilities |=
937 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_LINKSTATE;
938 priv->dev->if_capenable = priv->dev->if_capabilities;
939
940 return 0;
941 }
942
943
944 static struct ifnet *
ipoib_add_port(const char * format,struct ib_device * hca,u8 port)945 ipoib_add_port(const char *format, struct ib_device *hca, u8 port)
946 {
947 struct ipoib_dev_priv *priv;
948 struct ib_port_attr attr;
949 int result = -ENOMEM;
950
951 priv = ipoib_intf_alloc(format);
952 if (!priv)
953 goto alloc_mem_failed;
954
955 if (!ib_query_port(hca, port, &attr))
956 priv->max_ib_mtu = ib_mtu_enum_to_int(attr.max_mtu);
957 else {
958 printk(KERN_WARNING "%s: ib_query_port %d failed\n",
959 hca->name, port);
960 goto device_init_failed;
961 }
962
963 /* MTU will be reset when mcast join happens */
964 priv->dev->if_mtu = IPOIB_UD_MTU(priv->max_ib_mtu);
965 priv->mcast_mtu = priv->admin_mtu = priv->dev->if_mtu;
966
967 result = ib_query_pkey(hca, port, 0, &priv->pkey);
968 if (result) {
969 printk(KERN_WARNING "%s: ib_query_pkey port %d failed (ret = %d)\n",
970 hca->name, port, result);
971 goto device_init_failed;
972 }
973
974 if (ipoib_set_dev_features(priv, hca))
975 goto device_init_failed;
976
977 /*
978 * Set the full membership bit, so that we join the right
979 * broadcast group, etc.
980 */
981 priv->pkey |= 0x8000;
982
983 priv->broadcastaddr[8] = priv->pkey >> 8;
984 priv->broadcastaddr[9] = priv->pkey & 0xff;
985
986 result = ib_query_gid(hca, port, 0, &priv->local_gid);
987 if (result) {
988 printk(KERN_WARNING "%s: ib_query_gid port %d failed (ret = %d)\n",
989 hca->name, port, result);
990 goto device_init_failed;
991 }
992 memcpy(IF_LLADDR(priv->dev) + 4, priv->local_gid.raw, sizeof (union ib_gid));
993
994 result = ipoib_dev_init(priv, hca, port);
995 if (result < 0) {
996 printk(KERN_WARNING "%s: failed to initialize port %d (ret = %d)\n",
997 hca->name, port, result);
998 goto device_init_failed;
999 }
1000 if (ipoib_cm_admin_enabled(priv))
1001 priv->dev->if_mtu = IPOIB_CM_MTU(ipoib_cm_max_mtu(priv));
1002
1003 INIT_IB_EVENT_HANDLER(&priv->event_handler,
1004 priv->ca, ipoib_event);
1005 result = ib_register_event_handler(&priv->event_handler);
1006 if (result < 0) {
1007 printk(KERN_WARNING "%s: ib_register_event_handler failed for "
1008 "port %d (ret = %d)\n",
1009 hca->name, port, result);
1010 goto event_failed;
1011 }
1012 if_printf(priv->dev, "Attached to %s port %d\n", hca->name, port);
1013
1014 return priv->dev;
1015
1016 event_failed:
1017 ipoib_dev_cleanup(priv);
1018
1019 device_init_failed:
1020 ipoib_detach(priv);
1021
1022 alloc_mem_failed:
1023 return ERR_PTR(result);
1024 }
1025
1026 static void
ipoib_add_one(struct ib_device * device)1027 ipoib_add_one(struct ib_device *device)
1028 {
1029 struct list_head *dev_list;
1030 struct ifnet *dev;
1031 struct ipoib_dev_priv *priv;
1032 int s, e, p;
1033
1034 if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
1035 return;
1036
1037 dev_list = kmalloc(sizeof *dev_list, GFP_KERNEL);
1038 if (!dev_list)
1039 return;
1040
1041 INIT_LIST_HEAD(dev_list);
1042
1043 if (device->node_type == RDMA_NODE_IB_SWITCH) {
1044 s = 0;
1045 e = 0;
1046 } else {
1047 s = 1;
1048 e = device->phys_port_cnt;
1049 }
1050
1051 for (p = s; p <= e; ++p) {
1052 if (rdma_port_get_link_layer(device, p) != IB_LINK_LAYER_INFINIBAND)
1053 continue;
1054 dev = ipoib_add_port("ib", device, p);
1055 if (!IS_ERR(dev)) {
1056 priv = dev->if_softc;
1057 list_add_tail(&priv->list, dev_list);
1058 }
1059 }
1060
1061 ib_set_client_data(device, &ipoib_client, dev_list);
1062 }
1063
1064 static void
ipoib_remove_one(struct ib_device * device)1065 ipoib_remove_one(struct ib_device *device)
1066 {
1067 struct ipoib_dev_priv *priv, *tmp;
1068 struct list_head *dev_list;
1069
1070 if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
1071 return;
1072
1073 dev_list = ib_get_client_data(device, &ipoib_client);
1074
1075 list_for_each_entry_safe(priv, tmp, dev_list, list) {
1076 if (rdma_port_get_link_layer(device, priv->port) != IB_LINK_LAYER_INFINIBAND)
1077 continue;
1078
1079 ipoib_stop(priv);
1080
1081 ib_unregister_event_handler(&priv->event_handler);
1082
1083 /* dev_change_flags(priv->dev, priv->dev->flags & ~IFF_UP); */
1084
1085 flush_workqueue(ipoib_workqueue);
1086
1087 ipoib_dev_cleanup(priv);
1088 ipoib_detach(priv);
1089 }
1090
1091 kfree(dev_list);
1092 }
1093
1094 static void
ipoib_config_vlan(void * arg,struct ifnet * ifp,u_int16_t vtag)1095 ipoib_config_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
1096 {
1097 struct ipoib_dev_priv *parent;
1098 struct ipoib_dev_priv *priv;
1099 struct ifnet *dev;
1100 uint16_t pkey;
1101 int error;
1102
1103 if (ifp->if_type != IFT_INFINIBAND)
1104 return;
1105 dev = VLAN_DEVAT(ifp, vtag);
1106 if (dev == NULL)
1107 return;
1108 priv = NULL;
1109 error = 0;
1110 parent = ifp->if_softc;
1111 /* We only support 15 bits of pkey. */
1112 if (vtag & 0x8000)
1113 return;
1114 pkey = vtag | 0x8000; /* Set full membership bit. */
1115 if (pkey == parent->pkey)
1116 return;
1117 /* Check for dups */
1118 mutex_lock(&parent->vlan_mutex);
1119 list_for_each_entry(priv, &parent->child_intfs, list) {
1120 if (priv->pkey == pkey) {
1121 priv = NULL;
1122 error = EBUSY;
1123 goto out;
1124 }
1125 }
1126 priv = ipoib_priv_alloc();
1127 priv->dev = dev;
1128 priv->max_ib_mtu = parent->max_ib_mtu;
1129 priv->mcast_mtu = priv->admin_mtu = parent->dev->if_mtu;
1130 set_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags);
1131 error = ipoib_set_dev_features(priv, parent->ca);
1132 if (error)
1133 goto out;
1134 priv->pkey = pkey;
1135 priv->broadcastaddr[8] = pkey >> 8;
1136 priv->broadcastaddr[9] = pkey & 0xff;
1137 dev->if_broadcastaddr = priv->broadcastaddr;
1138 error = ipoib_dev_init(priv, parent->ca, parent->port);
1139 if (error)
1140 goto out;
1141 priv->parent = parent->dev;
1142 list_add_tail(&priv->list, &parent->child_intfs);
1143 VLAN_SETCOOKIE(dev, priv);
1144 dev->if_start = ipoib_vlan_start;
1145 dev->if_drv_flags &= ~IFF_DRV_RUNNING;
1146 dev->if_hdrlen = IPOIB_HEADER_LEN;
1147 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1148 ipoib_open(priv);
1149 mutex_unlock(&parent->vlan_mutex);
1150 return;
1151 out:
1152 mutex_unlock(&parent->vlan_mutex);
1153 if (priv)
1154 free(priv, M_TEMP);
1155 if (error)
1156 ipoib_warn(parent,
1157 "failed to initialize subinterface: device %s, port %d vtag 0x%X",
1158 parent->ca->name, parent->port, vtag);
1159 return;
1160 }
1161
1162 static void
ipoib_unconfig_vlan(void * arg,struct ifnet * ifp,u_int16_t vtag)1163 ipoib_unconfig_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
1164 {
1165 struct ipoib_dev_priv *parent;
1166 struct ipoib_dev_priv *priv;
1167 struct ifnet *dev;
1168 uint16_t pkey;
1169
1170 if (ifp->if_type != IFT_INFINIBAND)
1171 return;
1172
1173 dev = VLAN_DEVAT(ifp, vtag);
1174 if (dev)
1175 VLAN_SETCOOKIE(dev, NULL);
1176 pkey = vtag | 0x8000;
1177 parent = ifp->if_softc;
1178 mutex_lock(&parent->vlan_mutex);
1179 list_for_each_entry(priv, &parent->child_intfs, list) {
1180 if (priv->pkey == pkey) {
1181 ipoib_dev_cleanup(priv);
1182 list_del(&priv->list);
1183 break;
1184 }
1185 }
1186 mutex_unlock(&parent->vlan_mutex);
1187 }
1188
1189 eventhandler_tag ipoib_vlan_attach;
1190 eventhandler_tag ipoib_vlan_detach;
1191
1192 static int __init
ipoib_init_module(void)1193 ipoib_init_module(void)
1194 {
1195 int ret;
1196
1197 ipoib_recvq_size = roundup_pow_of_two(ipoib_recvq_size);
1198 ipoib_recvq_size = min(ipoib_recvq_size, IPOIB_MAX_QUEUE_SIZE);
1199 ipoib_recvq_size = max(ipoib_recvq_size, IPOIB_MIN_QUEUE_SIZE);
1200
1201 ipoib_sendq_size = roundup_pow_of_two(ipoib_sendq_size);
1202 ipoib_sendq_size = min(ipoib_sendq_size, IPOIB_MAX_QUEUE_SIZE);
1203 ipoib_sendq_size = max(ipoib_sendq_size, max(2 * MAX_SEND_CQE,
1204 IPOIB_MIN_QUEUE_SIZE));
1205 #ifdef CONFIG_INFINIBAND_IPOIB_CM
1206 ipoib_max_conn_qp = min(ipoib_max_conn_qp, IPOIB_CM_MAX_CONN_QP);
1207 #endif
1208
1209 ipoib_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
1210 ipoib_config_vlan, NULL, EVENTHANDLER_PRI_FIRST);
1211 ipoib_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1212 ipoib_unconfig_vlan, NULL, EVENTHANDLER_PRI_FIRST);
1213
1214 /*
1215 * We create our own workqueue mainly because we want to be
1216 * able to flush it when devices are being removed. We can't
1217 * use schedule_work()/flush_scheduled_work() because both
1218 * unregister_netdev() and linkwatch_event take the rtnl lock,
1219 * so flush_scheduled_work() can deadlock during device
1220 * removal.
1221 */
1222 ipoib_workqueue = create_singlethread_workqueue("ipoib");
1223 if (!ipoib_workqueue) {
1224 ret = -ENOMEM;
1225 goto err_fs;
1226 }
1227
1228 ib_sa_register_client(&ipoib_sa_client);
1229
1230 ret = ib_register_client(&ipoib_client);
1231 if (ret)
1232 goto err_sa;
1233
1234 return 0;
1235
1236 err_sa:
1237 ib_sa_unregister_client(&ipoib_sa_client);
1238 destroy_workqueue(ipoib_workqueue);
1239
1240 err_fs:
1241 return ret;
1242 }
1243
1244 static void __exit
ipoib_cleanup_module(void)1245 ipoib_cleanup_module(void)
1246 {
1247
1248 EVENTHANDLER_DEREGISTER(vlan_config, ipoib_vlan_attach);
1249 EVENTHANDLER_DEREGISTER(vlan_unconfig, ipoib_vlan_detach);
1250 ib_unregister_client(&ipoib_client);
1251 ib_sa_unregister_client(&ipoib_sa_client);
1252 destroy_workqueue(ipoib_workqueue);
1253 }
1254
1255 /*
1256 * Infiniband output routine.
1257 */
1258 static int
ipoib_output(struct ifnet * ifp,struct mbuf * m,struct sockaddr * dst,struct route * ro)1259 ipoib_output(struct ifnet *ifp, struct mbuf *m,
1260 struct sockaddr *dst, struct route *ro)
1261 {
1262 u_char edst[INFINIBAND_ALEN];
1263 struct llentry *lle = NULL;
1264 struct rtentry *rt0 = NULL;
1265 struct ipoib_header *eh;
1266 int error = 0;
1267 short type;
1268
1269 if (ro != NULL) {
1270 if (!(m->m_flags & (M_BCAST | M_MCAST)))
1271 lle = ro->ro_lle;
1272 rt0 = ro->ro_rt;
1273 }
1274 #ifdef MAC
1275 error = mac_ifnet_check_transmit(ifp, m);
1276 if (error)
1277 goto bad;
1278 #endif
1279
1280 M_PROFILE(m);
1281 if (ifp->if_flags & IFF_MONITOR) {
1282 error = ENETDOWN;
1283 goto bad;
1284 }
1285 if (!((ifp->if_flags & IFF_UP) &&
1286 (ifp->if_drv_flags & IFF_DRV_RUNNING))) {
1287 error = ENETDOWN;
1288 goto bad;
1289 }
1290
1291 switch (dst->sa_family) {
1292 #ifdef INET
1293 case AF_INET:
1294 if (lle != NULL && (lle->la_flags & LLE_VALID))
1295 memcpy(edst, &lle->ll_addr.mac8, sizeof(edst));
1296 else if (m->m_flags & M_MCAST)
1297 ip_ib_mc_map(((struct sockaddr_in *)dst)->sin_addr.s_addr, ifp->if_broadcastaddr, edst);
1298 else
1299 error = arpresolve(ifp, rt0, m, dst, edst, &lle);
1300 if (error)
1301 return (error == EWOULDBLOCK ? 0 : error);
1302 type = htons(ETHERTYPE_IP);
1303 break;
1304 case AF_ARP:
1305 {
1306 struct arphdr *ah;
1307 ah = mtod(m, struct arphdr *);
1308 ah->ar_hrd = htons(ARPHRD_INFINIBAND);
1309
1310 switch(ntohs(ah->ar_op)) {
1311 case ARPOP_REVREQUEST:
1312 case ARPOP_REVREPLY:
1313 type = htons(ETHERTYPE_REVARP);
1314 break;
1315 case ARPOP_REQUEST:
1316 case ARPOP_REPLY:
1317 default:
1318 type = htons(ETHERTYPE_ARP);
1319 break;
1320 }
1321
1322 if (m->m_flags & M_BCAST)
1323 bcopy(ifp->if_broadcastaddr, edst, INFINIBAND_ALEN);
1324 else
1325 bcopy(ar_tha(ah), edst, INFINIBAND_ALEN);
1326
1327 }
1328 break;
1329 #endif
1330 #ifdef INET6
1331 case AF_INET6:
1332 if (lle != NULL && (lle->la_flags & LLE_VALID))
1333 memcpy(edst, &lle->ll_addr.mac8, sizeof(edst));
1334 else if (m->m_flags & M_MCAST)
1335 ipv6_ib_mc_map(&((struct sockaddr_in6 *)dst)->sin6_addr, ifp->if_broadcastaddr, edst);
1336 else
1337 error = nd6_storelladdr(ifp, m, dst, (u_char *)edst, &lle);
1338 if (error)
1339 return error;
1340 type = htons(ETHERTYPE_IPV6);
1341 break;
1342 #endif
1343
1344 default:
1345 if_printf(ifp, "can't handle af%d\n", dst->sa_family);
1346 error = EAFNOSUPPORT;
1347 goto bad;
1348 }
1349
1350 /*
1351 * Add local net header. If no space in first mbuf,
1352 * allocate another.
1353 */
1354 M_PREPEND(m, IPOIB_HEADER_LEN, M_DONTWAIT);
1355 if (m == NULL) {
1356 error = ENOBUFS;
1357 goto bad;
1358 }
1359 eh = mtod(m, struct ipoib_header *);
1360 (void)memcpy(&eh->proto, &type, sizeof(eh->proto));
1361 (void)memcpy(&eh->hwaddr, edst, sizeof (edst));
1362
1363 /*
1364 * Queue message on interface, update output statistics if
1365 * successful, and start output if interface not yet active.
1366 */
1367 return ((ifp->if_transmit)(ifp, m));
1368 bad:
1369 if (m != NULL)
1370 m_freem(m);
1371 return (error);
1372 }
1373
1374 /*
1375 * Upper layer processing for a received Infiniband packet.
1376 */
1377 void
ipoib_demux(struct ifnet * ifp,struct mbuf * m,u_short proto)1378 ipoib_demux(struct ifnet *ifp, struct mbuf *m, u_short proto)
1379 {
1380 int isr;
1381
1382 #ifdef MAC
1383 /*
1384 * Tag the mbuf with an appropriate MAC label before any other
1385 * consumers can get to it.
1386 */
1387 mac_ifnet_create_mbuf(ifp, m);
1388 #endif
1389 /* Allow monitor mode to claim this frame, after stats are updated. */
1390 if (ifp->if_flags & IFF_MONITOR) {
1391 if_printf(ifp, "discard frame at IFF_MONITOR\n");
1392 m_freem(m);
1393 return;
1394 }
1395 /*
1396 * Dispatch frame to upper layer.
1397 */
1398 switch (proto) {
1399 #ifdef INET
1400 case ETHERTYPE_IP:
1401 isr = NETISR_IP;
1402 break;
1403
1404 case ETHERTYPE_ARP:
1405 if (ifp->if_flags & IFF_NOARP) {
1406 /* Discard packet if ARP is disabled on interface */
1407 m_freem(m);
1408 return;
1409 }
1410 isr = NETISR_ARP;
1411 break;
1412 #endif
1413 #ifdef INET6
1414 case ETHERTYPE_IPV6:
1415 isr = NETISR_IPV6;
1416 break;
1417 #endif
1418 default:
1419 goto discard;
1420 }
1421 netisr_dispatch(isr, m);
1422 return;
1423
1424 discard:
1425 m_freem(m);
1426 }
1427
1428 /*
1429 * Process a received Infiniband packet.
1430 */
1431 static void
ipoib_input(struct ifnet * ifp,struct mbuf * m)1432 ipoib_input(struct ifnet *ifp, struct mbuf *m)
1433 {
1434 struct ipoib_header *eh;
1435
1436 if ((ifp->if_flags & IFF_UP) == 0) {
1437 m_freem(m);
1438 return;
1439 }
1440 CURVNET_SET_QUIET(ifp->if_vnet);
1441
1442 /* Let BPF have it before we strip the header. */
1443 IPOIB_MTAP(ifp, m);
1444 eh = mtod(m, struct ipoib_header *);
1445 /*
1446 * Reset layer specific mbuf flags to avoid confusing upper layers.
1447 * Strip off Infiniband header.
1448 */
1449 m->m_flags &= ~M_VLANTAG;
1450 m->m_flags &= ~(M_PROTOFLAGS);
1451 m_adj(m, IPOIB_HEADER_LEN);
1452
1453 if (IPOIB_IS_MULTICAST(eh->hwaddr)) {
1454 if (memcmp(eh->hwaddr, ifp->if_broadcastaddr,
1455 ifp->if_addrlen) == 0)
1456 m->m_flags |= M_BCAST;
1457 else
1458 m->m_flags |= M_MCAST;
1459 ifp->if_imcasts++;
1460 }
1461
1462 ipoib_demux(ifp, m, ntohs(eh->proto));
1463 CURVNET_RESTORE();
1464 }
1465
1466 static int
ipoib_resolvemulti(struct ifnet * ifp,struct sockaddr ** llsa,struct sockaddr * sa)1467 ipoib_resolvemulti(struct ifnet *ifp, struct sockaddr **llsa,
1468 struct sockaddr *sa)
1469 {
1470 struct sockaddr_dl *sdl;
1471 #ifdef INET
1472 struct sockaddr_in *sin;
1473 #endif
1474 #ifdef INET6
1475 struct sockaddr_in6 *sin6;
1476 #endif
1477 u_char *e_addr;
1478
1479 switch(sa->sa_family) {
1480 case AF_LINK:
1481 /*
1482 * No mapping needed. Just check that it's a valid MC address.
1483 */
1484 sdl = (struct sockaddr_dl *)sa;
1485 e_addr = LLADDR(sdl);
1486 if (!IPOIB_IS_MULTICAST(e_addr))
1487 return EADDRNOTAVAIL;
1488 *llsa = 0;
1489 return 0;
1490
1491 #ifdef INET
1492 case AF_INET:
1493 sin = (struct sockaddr_in *)sa;
1494 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
1495 return EADDRNOTAVAIL;
1496 sdl = malloc(sizeof *sdl, M_IFMADDR,
1497 M_NOWAIT|M_ZERO);
1498 if (sdl == NULL)
1499 return ENOMEM;
1500 sdl->sdl_len = sizeof *sdl;
1501 sdl->sdl_family = AF_LINK;
1502 sdl->sdl_index = ifp->if_index;
1503 sdl->sdl_type = IFT_INFINIBAND;
1504 sdl->sdl_alen = INFINIBAND_ALEN;
1505 e_addr = LLADDR(sdl);
1506 ip_ib_mc_map(sin->sin_addr.s_addr, ifp->if_broadcastaddr,
1507 e_addr);
1508 *llsa = (struct sockaddr *)sdl;
1509 return 0;
1510 #endif
1511 #ifdef INET6
1512 case AF_INET6:
1513 sin6 = (struct sockaddr_in6 *)sa;
1514 /*
1515 * An IP6 address of 0 means listen to all
1516 * of the multicast address used for IP6.
1517 * This has no meaning in ipoib.
1518 */
1519 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
1520 return EADDRNOTAVAIL;
1521 if (!IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
1522 return EADDRNOTAVAIL;
1523 sdl = malloc(sizeof *sdl, M_IFMADDR,
1524 M_NOWAIT|M_ZERO);
1525 if (sdl == NULL)
1526 return (ENOMEM);
1527 sdl->sdl_len = sizeof *sdl;
1528 sdl->sdl_family = AF_LINK;
1529 sdl->sdl_index = ifp->if_index;
1530 sdl->sdl_type = IFT_INFINIBAND;
1531 sdl->sdl_alen = INFINIBAND_ALEN;
1532 e_addr = LLADDR(sdl);
1533 ipv6_ib_mc_map(&sin6->sin6_addr, ifp->if_broadcastaddr, e_addr);
1534 *llsa = (struct sockaddr *)sdl;
1535 return 0;
1536 #endif
1537
1538 default:
1539 return EAFNOSUPPORT;
1540 }
1541 }
1542
1543 module_init(ipoib_init_module);
1544 module_exit(ipoib_cleanup_module);
1545
1546 #undef MODULE_VERSION
1547 #include <sys/module.h>
1548 static int
ipoib_evhand(module_t mod,int event,void * arg)1549 ipoib_evhand(module_t mod, int event, void *arg)
1550 {
1551 return (0);
1552 }
1553
1554 static moduledata_t ipoib_mod = {
1555 .name = "ipoib",
1556 .evhand = ipoib_evhand,
1557 };
1558
1559 DECLARE_MODULE(ipoib, ipoib_mod, SI_SUB_SMP, SI_ORDER_ANY);
1560 MODULE_DEPEND(ipoib, ibcore, 1, 1, 1);
1561
1562