This source file includes following definitions.
- ipoib_netdev_event
- ipoib_open
- ipoib_stop
- ipoib_fix_features
- ipoib_change_mtu
- ipoib_get_stats
- ipoib_is_dev_match_addr_rcu
- ipoib_get_master_net_dev
- ipoib_upper_walk
- ipoib_get_net_dev_match_addr
- ipoib_match_gid_pkey_addr
- __ipoib_get_net_dev_by_params
- ipoib_get_net_dev_by_params
- ipoib_set_mode
- __path_find
- __path_add
- path_free
- ipoib_path_iter_init
- ipoib_path_iter_next
- ipoib_path_iter_read
- ipoib_mark_paths_invalid
- push_pseudo_header
- ipoib_flush_paths
- path_rec_completion
- init_path_rec
- path_rec_create
- path_rec_start
- neigh_refresh_path
- neigh_add_path
- unicast_arp_send
- ipoib_start_xmit
- ipoib_timeout
- ipoib_hard_header
- ipoib_set_mcast_list
- ipoib_get_iflink
- ipoib_addr_hash
- ipoib_neigh_get
- __ipoib_reap_neigh
- ipoib_reap_neigh
- ipoib_neigh_ctor
- ipoib_neigh_alloc
- ipoib_neigh_dtor
- ipoib_neigh_reclaim
- ipoib_neigh_free
- ipoib_neigh_hash_init
- neigh_hash_free_rcu
- ipoib_del_neighs_by_gid
- ipoib_flush_neighs
- ipoib_neigh_hash_uninit
- ipoib_napi_add
- ipoib_napi_del
- ipoib_dev_uninit_default
- ipoib_dev_init_default
- ipoib_ioctl
- ipoib_dev_init
- ipoib_parent_unregister_pre
- ipoib_set_dev_features
- ipoib_parent_init
- ipoib_child_init
- ipoib_ndo_init
- ipoib_ndo_uninit
- ipoib_set_vf_link_state
- ipoib_get_vf_config
- ipoib_set_vf_guid
- ipoib_get_vf_stats
- ipoib_setup_common
- ipoib_build_priv
- ipoib_alloc_netdev
- ipoib_intf_init
- ipoib_intf_alloc
- ipoib_intf_free
- show_pkey
- show_umcast
- ipoib_set_umcast
- set_umcast
- ipoib_add_umcast_attr
- set_base_guid
- ipoib_check_lladdr
- ipoib_set_mac
- create_child
- delete_child
- ipoib_add_pkey_attr
- dev_id_show
- ipoib_intercept_dev_id_attr
- ipoib_add_port
- ipoib_add_one
- ipoib_remove_one
- ipoib_init_module
- ipoib_cleanup_module
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35 #include "ipoib.h"
36
37 #include <linux/module.h>
38
39 #include <linux/init.h>
40 #include <linux/slab.h>
41 #include <linux/kernel.h>
42 #include <linux/vmalloc.h>
43
44 #include <linux/if_arp.h>
45
46 #include <linux/ip.h>
47 #include <linux/in.h>
48
49 #include <linux/jhash.h>
50 #include <net/arp.h>
51 #include <net/addrconf.h>
52 #include <linux/inetdevice.h>
53 #include <rdma/ib_cache.h>
54
55 #define DRV_VERSION "1.0.0"
56
57 const char ipoib_driver_version[] = DRV_VERSION;
58
59 MODULE_AUTHOR("Roland Dreier");
60 MODULE_DESCRIPTION("IP-over-InfiniBand net driver");
61 MODULE_LICENSE("Dual BSD/GPL");
62
63 int ipoib_sendq_size __read_mostly = IPOIB_TX_RING_SIZE;
64 int ipoib_recvq_size __read_mostly = IPOIB_RX_RING_SIZE;
65
66 module_param_named(send_queue_size, ipoib_sendq_size, int, 0444);
67 MODULE_PARM_DESC(send_queue_size, "Number of descriptors in send queue");
68 module_param_named(recv_queue_size, ipoib_recvq_size, int, 0444);
69 MODULE_PARM_DESC(recv_queue_size, "Number of descriptors in receive queue");
70
71 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
72 int ipoib_debug_level;
73
74 module_param_named(debug_level, ipoib_debug_level, int, 0644);
75 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
76 #endif
77
78 struct ipoib_path_iter {
79 struct net_device *dev;
80 struct ipoib_path path;
81 };
82
83 static const u8 ipv4_bcast_addr[] = {
84 0x00, 0xff, 0xff, 0xff,
85 0xff, 0x12, 0x40, 0x1b, 0x00, 0x00, 0x00, 0x00,
86 0x00, 0x00, 0x00, 0x00, 0xff, 0xff, 0xff, 0xff
87 };
88
89 struct workqueue_struct *ipoib_workqueue;
90
91 struct ib_sa_client ipoib_sa_client;
92
93 static void ipoib_add_one(struct ib_device *device);
94 static void ipoib_remove_one(struct ib_device *device, void *client_data);
95 static void ipoib_neigh_reclaim(struct rcu_head *rp);
96 static struct net_device *ipoib_get_net_dev_by_params(
97 struct ib_device *dev, u8 port, u16 pkey,
98 const union ib_gid *gid, const struct sockaddr *addr,
99 void *client_data);
100 static int ipoib_set_mac(struct net_device *dev, void *addr);
101 static int ipoib_ioctl(struct net_device *dev, struct ifreq *ifr,
102 int cmd);
103
104 static struct ib_client ipoib_client = {
105 .name = "ipoib",
106 .add = ipoib_add_one,
107 .remove = ipoib_remove_one,
108 .get_net_dev_by_params = ipoib_get_net_dev_by_params,
109 };
110
111 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
112 static int ipoib_netdev_event(struct notifier_block *this,
113 unsigned long event, void *ptr)
114 {
115 struct netdev_notifier_info *ni = ptr;
116 struct net_device *dev = ni->dev;
117
118 if (dev->netdev_ops->ndo_open != ipoib_open)
119 return NOTIFY_DONE;
120
121 switch (event) {
122 case NETDEV_REGISTER:
123 ipoib_create_debug_files(dev);
124 break;
125 case NETDEV_CHANGENAME:
126 ipoib_delete_debug_files(dev);
127 ipoib_create_debug_files(dev);
128 break;
129 case NETDEV_UNREGISTER:
130 ipoib_delete_debug_files(dev);
131 break;
132 }
133
134 return NOTIFY_DONE;
135 }
136 #endif
137
138 int ipoib_open(struct net_device *dev)
139 {
140 struct ipoib_dev_priv *priv = ipoib_priv(dev);
141
142 ipoib_dbg(priv, "bringing up interface\n");
143
144 netif_carrier_off(dev);
145
146 set_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
147
148 priv->sm_fullmember_sendonly_support = false;
149
150 if (ipoib_ib_dev_open(dev)) {
151 if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags))
152 return 0;
153 goto err_disable;
154 }
155
156 ipoib_ib_dev_up(dev);
157
158 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
159 struct ipoib_dev_priv *cpriv;
160
161
162 down_read(&priv->vlan_rwsem);
163 list_for_each_entry(cpriv, &priv->child_intfs, list) {
164 int flags;
165
166 flags = cpriv->dev->flags;
167 if (flags & IFF_UP)
168 continue;
169
170 dev_change_flags(cpriv->dev, flags | IFF_UP, NULL);
171 }
172 up_read(&priv->vlan_rwsem);
173 }
174
175 netif_start_queue(dev);
176
177 return 0;
178
179 err_disable:
180 clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
181
182 return -EINVAL;
183 }
184
185 static int ipoib_stop(struct net_device *dev)
186 {
187 struct ipoib_dev_priv *priv = ipoib_priv(dev);
188
189 ipoib_dbg(priv, "stopping interface\n");
190
191 clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
192
193 netif_stop_queue(dev);
194
195 ipoib_ib_dev_down(dev);
196 ipoib_ib_dev_stop(dev);
197
198 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
199 struct ipoib_dev_priv *cpriv;
200
201
202 down_read(&priv->vlan_rwsem);
203 list_for_each_entry(cpriv, &priv->child_intfs, list) {
204 int flags;
205
206 flags = cpriv->dev->flags;
207 if (!(flags & IFF_UP))
208 continue;
209
210 dev_change_flags(cpriv->dev, flags & ~IFF_UP, NULL);
211 }
212 up_read(&priv->vlan_rwsem);
213 }
214
215 return 0;
216 }
217
218 static netdev_features_t ipoib_fix_features(struct net_device *dev, netdev_features_t features)
219 {
220 struct ipoib_dev_priv *priv = ipoib_priv(dev);
221
222 if (test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags))
223 features &= ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
224
225 return features;
226 }
227
228 static int ipoib_change_mtu(struct net_device *dev, int new_mtu)
229 {
230 struct ipoib_dev_priv *priv = ipoib_priv(dev);
231 int ret = 0;
232
233
234 if (ipoib_cm_admin_enabled(dev)) {
235 if (new_mtu > ipoib_cm_max_mtu(dev))
236 return -EINVAL;
237
238 if (new_mtu > priv->mcast_mtu)
239 ipoib_warn(priv, "mtu > %d will cause multicast packet drops.\n",
240 priv->mcast_mtu);
241
242 dev->mtu = new_mtu;
243 return 0;
244 }
245
246 if (new_mtu < (ETH_MIN_MTU + IPOIB_ENCAP_LEN) ||
247 new_mtu > IPOIB_UD_MTU(priv->max_ib_mtu))
248 return -EINVAL;
249
250 priv->admin_mtu = new_mtu;
251
252 if (priv->mcast_mtu < priv->admin_mtu)
253 ipoib_dbg(priv, "MTU must be smaller than the underlying "
254 "link layer MTU - 4 (%u)\n", priv->mcast_mtu);
255
256 new_mtu = min(priv->mcast_mtu, priv->admin_mtu);
257
258 if (priv->rn_ops->ndo_change_mtu) {
259 bool carrier_status = netif_carrier_ok(dev);
260
261 netif_carrier_off(dev);
262
263
264 ret = priv->rn_ops->ndo_change_mtu(dev, new_mtu);
265
266 if (carrier_status)
267 netif_carrier_on(dev);
268 } else {
269 dev->mtu = new_mtu;
270 }
271
272 return ret;
273 }
274
275 static void ipoib_get_stats(struct net_device *dev,
276 struct rtnl_link_stats64 *stats)
277 {
278 struct ipoib_dev_priv *priv = ipoib_priv(dev);
279
280 if (priv->rn_ops->ndo_get_stats64)
281 priv->rn_ops->ndo_get_stats64(dev, stats);
282 else
283 netdev_stats_to_stats64(stats, &dev->stats);
284 }
285
286
287 static bool ipoib_is_dev_match_addr_rcu(const struct sockaddr *addr,
288 struct net_device *dev)
289 {
290 struct net *net = dev_net(dev);
291 struct in_device *in_dev;
292 struct sockaddr_in *addr_in = (struct sockaddr_in *)addr;
293 struct sockaddr_in6 *addr_in6 = (struct sockaddr_in6 *)addr;
294 __be32 ret_addr;
295
296 switch (addr->sa_family) {
297 case AF_INET:
298 in_dev = in_dev_get(dev);
299 if (!in_dev)
300 return false;
301
302 ret_addr = inet_confirm_addr(net, in_dev, 0,
303 addr_in->sin_addr.s_addr,
304 RT_SCOPE_HOST);
305 in_dev_put(in_dev);
306 if (ret_addr)
307 return true;
308
309 break;
310 case AF_INET6:
311 if (IS_ENABLED(CONFIG_IPV6) &&
312 ipv6_chk_addr(net, &addr_in6->sin6_addr, dev, 1))
313 return true;
314
315 break;
316 }
317 return false;
318 }
319
320
321
322
323
324
325
326
327 static struct net_device *ipoib_get_master_net_dev(struct net_device *dev)
328 {
329 struct net_device *master;
330
331 rcu_read_lock();
332 master = netdev_master_upper_dev_get_rcu(dev);
333 if (master)
334 dev_hold(master);
335 rcu_read_unlock();
336
337 if (master)
338 return master;
339
340 dev_hold(dev);
341 return dev;
342 }
343
344 struct ipoib_walk_data {
345 const struct sockaddr *addr;
346 struct net_device *result;
347 };
348
349 static int ipoib_upper_walk(struct net_device *upper, void *_data)
350 {
351 struct ipoib_walk_data *data = _data;
352 int ret = 0;
353
354 if (ipoib_is_dev_match_addr_rcu(data->addr, upper)) {
355 dev_hold(upper);
356 data->result = upper;
357 ret = 1;
358 }
359
360 return ret;
361 }
362
363
364
365
366
367
368
369
370
371
372 static struct net_device *ipoib_get_net_dev_match_addr(
373 const struct sockaddr *addr, struct net_device *dev)
374 {
375 struct ipoib_walk_data data = {
376 .addr = addr,
377 };
378
379 rcu_read_lock();
380 if (ipoib_is_dev_match_addr_rcu(addr, dev)) {
381 dev_hold(dev);
382 data.result = dev;
383 goto out;
384 }
385
386 netdev_walk_all_upper_dev_rcu(dev, ipoib_upper_walk, &data);
387 out:
388 rcu_read_unlock();
389 return data.result;
390 }
391
392
393
394
395
396
397 static int ipoib_match_gid_pkey_addr(struct ipoib_dev_priv *priv,
398 const union ib_gid *gid,
399 u16 pkey_index,
400 const struct sockaddr *addr,
401 int nesting,
402 struct net_device **found_net_dev)
403 {
404 struct ipoib_dev_priv *child_priv;
405 struct net_device *net_dev = NULL;
406 int matches = 0;
407
408 if (priv->pkey_index == pkey_index &&
409 (!gid || !memcmp(gid, &priv->local_gid, sizeof(*gid)))) {
410 if (!addr) {
411 net_dev = ipoib_get_master_net_dev(priv->dev);
412 } else {
413
414
415 net_dev = ipoib_get_net_dev_match_addr(addr, priv->dev);
416 }
417 if (net_dev) {
418 if (!*found_net_dev)
419 *found_net_dev = net_dev;
420 else
421 dev_put(net_dev);
422 ++matches;
423 }
424 }
425
426
427 down_read_nested(&priv->vlan_rwsem, nesting);
428 list_for_each_entry(child_priv, &priv->child_intfs, list) {
429 matches += ipoib_match_gid_pkey_addr(child_priv, gid,
430 pkey_index, addr,
431 nesting + 1,
432 found_net_dev);
433 if (matches > 1)
434 break;
435 }
436 up_read(&priv->vlan_rwsem);
437
438 return matches;
439 }
440
441
442
443
444 static int __ipoib_get_net_dev_by_params(struct list_head *dev_list, u8 port,
445 u16 pkey_index,
446 const union ib_gid *gid,
447 const struct sockaddr *addr,
448 struct net_device **net_dev)
449 {
450 struct ipoib_dev_priv *priv;
451 int matches = 0;
452
453 *net_dev = NULL;
454
455 list_for_each_entry(priv, dev_list, list) {
456 if (priv->port != port)
457 continue;
458
459 matches += ipoib_match_gid_pkey_addr(priv, gid, pkey_index,
460 addr, 0, net_dev);
461 if (matches > 1)
462 break;
463 }
464
465 return matches;
466 }
467
468 static struct net_device *ipoib_get_net_dev_by_params(
469 struct ib_device *dev, u8 port, u16 pkey,
470 const union ib_gid *gid, const struct sockaddr *addr,
471 void *client_data)
472 {
473 struct net_device *net_dev;
474 struct list_head *dev_list = client_data;
475 u16 pkey_index;
476 int matches;
477 int ret;
478
479 if (!rdma_protocol_ib(dev, port))
480 return NULL;
481
482 ret = ib_find_cached_pkey(dev, port, pkey, &pkey_index);
483 if (ret)
484 return NULL;
485
486 if (!dev_list)
487 return NULL;
488
489
490 matches = __ipoib_get_net_dev_by_params(dev_list, port, pkey_index,
491 gid, NULL, &net_dev);
492
493 switch (matches) {
494 case 0:
495 return NULL;
496 case 1:
497 return net_dev;
498 }
499
500 dev_put(net_dev);
501
502
503
504 matches = __ipoib_get_net_dev_by_params(dev_list, port, pkey_index,
505 gid, addr, &net_dev);
506 switch (matches) {
507 case 0:
508 return NULL;
509 default:
510 dev_warn_ratelimited(&dev->dev,
511 "duplicate IP address detected\n");
512
513 case 1:
514 return net_dev;
515 }
516 }
517
518 int ipoib_set_mode(struct net_device *dev, const char *buf)
519 {
520 struct ipoib_dev_priv *priv = ipoib_priv(dev);
521
522 if ((test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags) &&
523 !strcmp(buf, "connected\n")) ||
524 (!test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags) &&
525 !strcmp(buf, "datagram\n"))) {
526 return 0;
527 }
528
529
530 if (IPOIB_CM_SUPPORTED(dev->dev_addr) && !strcmp(buf, "connected\n")) {
531 set_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
532 ipoib_warn(priv, "enabling connected mode "
533 "will cause multicast packet drops\n");
534 netdev_update_features(dev);
535 dev_set_mtu(dev, ipoib_cm_max_mtu(dev));
536 rtnl_unlock();
537 priv->tx_wr.wr.send_flags &= ~IB_SEND_IP_CSUM;
538
539 ipoib_flush_paths(dev);
540 return (!rtnl_trylock()) ? -EBUSY : 0;
541 }
542
543 if (!strcmp(buf, "datagram\n")) {
544 clear_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
545 netdev_update_features(dev);
546 dev_set_mtu(dev, min(priv->mcast_mtu, dev->mtu));
547 rtnl_unlock();
548 ipoib_flush_paths(dev);
549 return (!rtnl_trylock()) ? -EBUSY : 0;
550 }
551
552 return -EINVAL;
553 }
554
555 struct ipoib_path *__path_find(struct net_device *dev, void *gid)
556 {
557 struct ipoib_dev_priv *priv = ipoib_priv(dev);
558 struct rb_node *n = priv->path_tree.rb_node;
559 struct ipoib_path *path;
560 int ret;
561
562 while (n) {
563 path = rb_entry(n, struct ipoib_path, rb_node);
564
565 ret = memcmp(gid, path->pathrec.dgid.raw,
566 sizeof (union ib_gid));
567
568 if (ret < 0)
569 n = n->rb_left;
570 else if (ret > 0)
571 n = n->rb_right;
572 else
573 return path;
574 }
575
576 return NULL;
577 }
578
579 static int __path_add(struct net_device *dev, struct ipoib_path *path)
580 {
581 struct ipoib_dev_priv *priv = ipoib_priv(dev);
582 struct rb_node **n = &priv->path_tree.rb_node;
583 struct rb_node *pn = NULL;
584 struct ipoib_path *tpath;
585 int ret;
586
587 while (*n) {
588 pn = *n;
589 tpath = rb_entry(pn, struct ipoib_path, rb_node);
590
591 ret = memcmp(path->pathrec.dgid.raw, tpath->pathrec.dgid.raw,
592 sizeof (union ib_gid));
593 if (ret < 0)
594 n = &pn->rb_left;
595 else if (ret > 0)
596 n = &pn->rb_right;
597 else
598 return -EEXIST;
599 }
600
601 rb_link_node(&path->rb_node, pn, n);
602 rb_insert_color(&path->rb_node, &priv->path_tree);
603
604 list_add_tail(&path->list, &priv->path_list);
605
606 return 0;
607 }
608
609 static void path_free(struct net_device *dev, struct ipoib_path *path)
610 {
611 struct sk_buff *skb;
612
613 while ((skb = __skb_dequeue(&path->queue)))
614 dev_kfree_skb_irq(skb);
615
616 ipoib_dbg(ipoib_priv(dev), "%s\n", __func__);
617
618
619 ipoib_del_neighs_by_gid(dev, path->pathrec.dgid.raw);
620
621 if (path->ah)
622 ipoib_put_ah(path->ah);
623
624 kfree(path);
625 }
626
627 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
628
629 struct ipoib_path_iter *ipoib_path_iter_init(struct net_device *dev)
630 {
631 struct ipoib_path_iter *iter;
632
633 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
634 if (!iter)
635 return NULL;
636
637 iter->dev = dev;
638 memset(iter->path.pathrec.dgid.raw, 0, 16);
639
640 if (ipoib_path_iter_next(iter)) {
641 kfree(iter);
642 return NULL;
643 }
644
645 return iter;
646 }
647
648 int ipoib_path_iter_next(struct ipoib_path_iter *iter)
649 {
650 struct ipoib_dev_priv *priv = ipoib_priv(iter->dev);
651 struct rb_node *n;
652 struct ipoib_path *path;
653 int ret = 1;
654
655 spin_lock_irq(&priv->lock);
656
657 n = rb_first(&priv->path_tree);
658
659 while (n) {
660 path = rb_entry(n, struct ipoib_path, rb_node);
661
662 if (memcmp(iter->path.pathrec.dgid.raw, path->pathrec.dgid.raw,
663 sizeof (union ib_gid)) < 0) {
664 iter->path = *path;
665 ret = 0;
666 break;
667 }
668
669 n = rb_next(n);
670 }
671
672 spin_unlock_irq(&priv->lock);
673
674 return ret;
675 }
676
677 void ipoib_path_iter_read(struct ipoib_path_iter *iter,
678 struct ipoib_path *path)
679 {
680 *path = iter->path;
681 }
682
683 #endif
684
685 void ipoib_mark_paths_invalid(struct net_device *dev)
686 {
687 struct ipoib_dev_priv *priv = ipoib_priv(dev);
688 struct ipoib_path *path, *tp;
689
690 spin_lock_irq(&priv->lock);
691
692 list_for_each_entry_safe(path, tp, &priv->path_list, list) {
693 ipoib_dbg(priv, "mark path LID 0x%08x GID %pI6 invalid\n",
694 be32_to_cpu(sa_path_get_dlid(&path->pathrec)),
695 path->pathrec.dgid.raw);
696 if (path->ah)
697 path->ah->valid = 0;
698 }
699
700 spin_unlock_irq(&priv->lock);
701 }
702
703 static void push_pseudo_header(struct sk_buff *skb, const char *daddr)
704 {
705 struct ipoib_pseudo_header *phdr;
706
707 phdr = skb_push(skb, sizeof(*phdr));
708 memcpy(phdr->hwaddr, daddr, INFINIBAND_ALEN);
709 }
710
711 void ipoib_flush_paths(struct net_device *dev)
712 {
713 struct ipoib_dev_priv *priv = ipoib_priv(dev);
714 struct ipoib_path *path, *tp;
715 LIST_HEAD(remove_list);
716 unsigned long flags;
717
718 netif_tx_lock_bh(dev);
719 spin_lock_irqsave(&priv->lock, flags);
720
721 list_splice_init(&priv->path_list, &remove_list);
722
723 list_for_each_entry(path, &remove_list, list)
724 rb_erase(&path->rb_node, &priv->path_tree);
725
726 list_for_each_entry_safe(path, tp, &remove_list, list) {
727 if (path->query)
728 ib_sa_cancel_query(path->query_id, path->query);
729 spin_unlock_irqrestore(&priv->lock, flags);
730 netif_tx_unlock_bh(dev);
731 wait_for_completion(&path->done);
732 path_free(dev, path);
733 netif_tx_lock_bh(dev);
734 spin_lock_irqsave(&priv->lock, flags);
735 }
736
737 spin_unlock_irqrestore(&priv->lock, flags);
738 netif_tx_unlock_bh(dev);
739 }
740
741 static void path_rec_completion(int status,
742 struct sa_path_rec *pathrec,
743 void *path_ptr)
744 {
745 struct ipoib_path *path = path_ptr;
746 struct net_device *dev = path->dev;
747 struct ipoib_dev_priv *priv = ipoib_priv(dev);
748 struct ipoib_ah *ah = NULL;
749 struct ipoib_ah *old_ah = NULL;
750 struct ipoib_neigh *neigh, *tn;
751 struct sk_buff_head skqueue;
752 struct sk_buff *skb;
753 unsigned long flags;
754
755 if (!status)
756 ipoib_dbg(priv, "PathRec LID 0x%04x for GID %pI6\n",
757 be32_to_cpu(sa_path_get_dlid(pathrec)),
758 pathrec->dgid.raw);
759 else
760 ipoib_dbg(priv, "PathRec status %d for GID %pI6\n",
761 status, path->pathrec.dgid.raw);
762
763 skb_queue_head_init(&skqueue);
764
765 if (!status) {
766 struct rdma_ah_attr av;
767
768 if (!ib_init_ah_attr_from_path(priv->ca, priv->port,
769 pathrec, &av, NULL)) {
770 ah = ipoib_create_ah(dev, priv->pd, &av);
771 rdma_destroy_ah_attr(&av);
772 }
773 }
774
775 spin_lock_irqsave(&priv->lock, flags);
776
777 if (!IS_ERR_OR_NULL(ah)) {
778
779
780
781
782
783 if (memcmp(pathrec->dgid.raw, path->pathrec.dgid.raw,
784 sizeof(union ib_gid))) {
785 ipoib_dbg(
786 priv,
787 "%s got PathRec for gid %pI6 while asked for %pI6\n",
788 dev->name, pathrec->dgid.raw,
789 path->pathrec.dgid.raw);
790 memcpy(pathrec->dgid.raw, path->pathrec.dgid.raw,
791 sizeof(union ib_gid));
792 }
793
794 path->pathrec = *pathrec;
795
796 old_ah = path->ah;
797 path->ah = ah;
798
799 ipoib_dbg(priv, "created address handle %p for LID 0x%04x, SL %d\n",
800 ah, be32_to_cpu(sa_path_get_dlid(pathrec)),
801 pathrec->sl);
802
803 while ((skb = __skb_dequeue(&path->queue)))
804 __skb_queue_tail(&skqueue, skb);
805
806 list_for_each_entry_safe(neigh, tn, &path->neigh_list, list) {
807 if (neigh->ah) {
808 WARN_ON(neigh->ah != old_ah);
809
810
811
812
813
814
815
816 ipoib_put_ah(neigh->ah);
817 }
818 kref_get(&path->ah->ref);
819 neigh->ah = path->ah;
820
821 if (ipoib_cm_enabled(dev, neigh->daddr)) {
822 if (!ipoib_cm_get(neigh))
823 ipoib_cm_set(neigh, ipoib_cm_create_tx(dev,
824 path,
825 neigh));
826 if (!ipoib_cm_get(neigh)) {
827 ipoib_neigh_free(neigh);
828 continue;
829 }
830 }
831
832 while ((skb = __skb_dequeue(&neigh->queue)))
833 __skb_queue_tail(&skqueue, skb);
834 }
835 path->ah->valid = 1;
836 }
837
838 path->query = NULL;
839 complete(&path->done);
840
841 spin_unlock_irqrestore(&priv->lock, flags);
842
843 if (IS_ERR_OR_NULL(ah))
844 ipoib_del_neighs_by_gid(dev, path->pathrec.dgid.raw);
845
846 if (old_ah)
847 ipoib_put_ah(old_ah);
848
849 while ((skb = __skb_dequeue(&skqueue))) {
850 int ret;
851 skb->dev = dev;
852 ret = dev_queue_xmit(skb);
853 if (ret)
854 ipoib_warn(priv, "%s: dev_queue_xmit failed to re-queue packet, ret:%d\n",
855 __func__, ret);
856 }
857 }
858
859 static void init_path_rec(struct ipoib_dev_priv *priv, struct ipoib_path *path,
860 void *gid)
861 {
862 path->dev = priv->dev;
863
864 if (rdma_cap_opa_ah(priv->ca, priv->port))
865 path->pathrec.rec_type = SA_PATH_REC_TYPE_OPA;
866 else
867 path->pathrec.rec_type = SA_PATH_REC_TYPE_IB;
868
869 memcpy(path->pathrec.dgid.raw, gid, sizeof(union ib_gid));
870 path->pathrec.sgid = priv->local_gid;
871 path->pathrec.pkey = cpu_to_be16(priv->pkey);
872 path->pathrec.numb_path = 1;
873 path->pathrec.traffic_class = priv->broadcast->mcmember.traffic_class;
874 }
875
876 static struct ipoib_path *path_rec_create(struct net_device *dev, void *gid)
877 {
878 struct ipoib_dev_priv *priv = ipoib_priv(dev);
879 struct ipoib_path *path;
880
881 if (!priv->broadcast)
882 return NULL;
883
884 path = kzalloc(sizeof(*path), GFP_ATOMIC);
885 if (!path)
886 return NULL;
887
888 skb_queue_head_init(&path->queue);
889
890 INIT_LIST_HEAD(&path->neigh_list);
891
892 init_path_rec(priv, path, gid);
893
894 return path;
895 }
896
897 static int path_rec_start(struct net_device *dev,
898 struct ipoib_path *path)
899 {
900 struct ipoib_dev_priv *priv = ipoib_priv(dev);
901
902 ipoib_dbg(priv, "Start path record lookup for %pI6\n",
903 path->pathrec.dgid.raw);
904
905 init_completion(&path->done);
906
907 path->query_id =
908 ib_sa_path_rec_get(&ipoib_sa_client, priv->ca, priv->port,
909 &path->pathrec,
910 IB_SA_PATH_REC_DGID |
911 IB_SA_PATH_REC_SGID |
912 IB_SA_PATH_REC_NUMB_PATH |
913 IB_SA_PATH_REC_TRAFFIC_CLASS |
914 IB_SA_PATH_REC_PKEY,
915 1000, GFP_ATOMIC,
916 path_rec_completion,
917 path, &path->query);
918 if (path->query_id < 0) {
919 ipoib_warn(priv, "ib_sa_path_rec_get failed: %d\n", path->query_id);
920 path->query = NULL;
921 complete(&path->done);
922 return path->query_id;
923 }
924
925 return 0;
926 }
927
928 static void neigh_refresh_path(struct ipoib_neigh *neigh, u8 *daddr,
929 struct net_device *dev)
930 {
931 struct ipoib_dev_priv *priv = ipoib_priv(dev);
932 struct ipoib_path *path;
933 unsigned long flags;
934
935 spin_lock_irqsave(&priv->lock, flags);
936
937 path = __path_find(dev, daddr + 4);
938 if (!path)
939 goto out;
940 if (!path->query)
941 path_rec_start(dev, path);
942 out:
943 spin_unlock_irqrestore(&priv->lock, flags);
944 }
945
946 static struct ipoib_neigh *neigh_add_path(struct sk_buff *skb, u8 *daddr,
947 struct net_device *dev)
948 {
949 struct ipoib_dev_priv *priv = ipoib_priv(dev);
950 struct rdma_netdev *rn = netdev_priv(dev);
951 struct ipoib_path *path;
952 struct ipoib_neigh *neigh;
953 unsigned long flags;
954
955 spin_lock_irqsave(&priv->lock, flags);
956 neigh = ipoib_neigh_alloc(daddr, dev);
957 if (!neigh) {
958 spin_unlock_irqrestore(&priv->lock, flags);
959 ++dev->stats.tx_dropped;
960 dev_kfree_skb_any(skb);
961 return NULL;
962 }
963
964
965
966
967 if (unlikely(!list_empty(&neigh->list))) {
968 spin_unlock_irqrestore(&priv->lock, flags);
969 return neigh;
970 }
971
972 path = __path_find(dev, daddr + 4);
973 if (!path) {
974 path = path_rec_create(dev, daddr + 4);
975 if (!path)
976 goto err_path;
977
978 __path_add(dev, path);
979 }
980
981 list_add_tail(&neigh->list, &path->neigh_list);
982
983 if (path->ah && path->ah->valid) {
984 kref_get(&path->ah->ref);
985 neigh->ah = path->ah;
986
987 if (ipoib_cm_enabled(dev, neigh->daddr)) {
988 if (!ipoib_cm_get(neigh))
989 ipoib_cm_set(neigh, ipoib_cm_create_tx(dev, path, neigh));
990 if (!ipoib_cm_get(neigh)) {
991 ipoib_neigh_free(neigh);
992 goto err_drop;
993 }
994 if (skb_queue_len(&neigh->queue) <
995 IPOIB_MAX_PATH_REC_QUEUE) {
996 push_pseudo_header(skb, neigh->daddr);
997 __skb_queue_tail(&neigh->queue, skb);
998 } else {
999 ipoib_warn(priv, "queue length limit %d. Packet drop.\n",
1000 skb_queue_len(&neigh->queue));
1001 goto err_drop;
1002 }
1003 } else {
1004 spin_unlock_irqrestore(&priv->lock, flags);
1005 path->ah->last_send = rn->send(dev, skb, path->ah->ah,
1006 IPOIB_QPN(daddr));
1007 ipoib_neigh_put(neigh);
1008 return NULL;
1009 }
1010 } else {
1011 neigh->ah = NULL;
1012
1013 if (!path->query && path_rec_start(dev, path))
1014 goto err_path;
1015 if (skb_queue_len(&neigh->queue) < IPOIB_MAX_PATH_REC_QUEUE) {
1016 push_pseudo_header(skb, neigh->daddr);
1017 __skb_queue_tail(&neigh->queue, skb);
1018 } else {
1019 goto err_drop;
1020 }
1021 }
1022
1023 spin_unlock_irqrestore(&priv->lock, flags);
1024 ipoib_neigh_put(neigh);
1025 return NULL;
1026
1027 err_path:
1028 ipoib_neigh_free(neigh);
1029 err_drop:
1030 ++dev->stats.tx_dropped;
1031 dev_kfree_skb_any(skb);
1032
1033 spin_unlock_irqrestore(&priv->lock, flags);
1034 ipoib_neigh_put(neigh);
1035
1036 return NULL;
1037 }
1038
1039 static void unicast_arp_send(struct sk_buff *skb, struct net_device *dev,
1040 struct ipoib_pseudo_header *phdr)
1041 {
1042 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1043 struct rdma_netdev *rn = netdev_priv(dev);
1044 struct ipoib_path *path;
1045 unsigned long flags;
1046
1047 spin_lock_irqsave(&priv->lock, flags);
1048
1049
1050 if (!priv->broadcast)
1051 goto drop_and_unlock;
1052
1053 path = __path_find(dev, phdr->hwaddr + 4);
1054 if (!path || !path->ah || !path->ah->valid) {
1055 if (!path) {
1056 path = path_rec_create(dev, phdr->hwaddr + 4);
1057 if (!path)
1058 goto drop_and_unlock;
1059 __path_add(dev, path);
1060 } else {
1061
1062
1063
1064
1065 init_path_rec(priv, path, phdr->hwaddr + 4);
1066 }
1067 if (!path->query && path_rec_start(dev, path)) {
1068 goto drop_and_unlock;
1069 }
1070
1071 if (skb_queue_len(&path->queue) < IPOIB_MAX_PATH_REC_QUEUE) {
1072 push_pseudo_header(skb, phdr->hwaddr);
1073 __skb_queue_tail(&path->queue, skb);
1074 goto unlock;
1075 } else {
1076 goto drop_and_unlock;
1077 }
1078 }
1079
1080 spin_unlock_irqrestore(&priv->lock, flags);
1081 ipoib_dbg(priv, "Send unicast ARP to %08x\n",
1082 be32_to_cpu(sa_path_get_dlid(&path->pathrec)));
1083 path->ah->last_send = rn->send(dev, skb, path->ah->ah,
1084 IPOIB_QPN(phdr->hwaddr));
1085 return;
1086
1087 drop_and_unlock:
1088 ++dev->stats.tx_dropped;
1089 dev_kfree_skb_any(skb);
1090 unlock:
1091 spin_unlock_irqrestore(&priv->lock, flags);
1092 }
1093
1094 static netdev_tx_t ipoib_start_xmit(struct sk_buff *skb, struct net_device *dev)
1095 {
1096 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1097 struct rdma_netdev *rn = netdev_priv(dev);
1098 struct ipoib_neigh *neigh;
1099 struct ipoib_pseudo_header *phdr;
1100 struct ipoib_header *header;
1101 unsigned long flags;
1102
1103 phdr = (struct ipoib_pseudo_header *) skb->data;
1104 skb_pull(skb, sizeof(*phdr));
1105 header = (struct ipoib_header *) skb->data;
1106
1107 if (unlikely(phdr->hwaddr[4] == 0xff)) {
1108
1109 if ((header->proto != htons(ETH_P_IP)) &&
1110 (header->proto != htons(ETH_P_IPV6)) &&
1111 (header->proto != htons(ETH_P_ARP)) &&
1112 (header->proto != htons(ETH_P_RARP)) &&
1113 (header->proto != htons(ETH_P_TIPC))) {
1114
1115 ++dev->stats.tx_dropped;
1116 dev_kfree_skb_any(skb);
1117 return NETDEV_TX_OK;
1118 }
1119
1120 phdr->hwaddr[8] = (priv->pkey >> 8) & 0xff;
1121 phdr->hwaddr[9] = priv->pkey & 0xff;
1122
1123 neigh = ipoib_neigh_get(dev, phdr->hwaddr);
1124 if (likely(neigh))
1125 goto send_using_neigh;
1126 ipoib_mcast_send(dev, phdr->hwaddr, skb);
1127 return NETDEV_TX_OK;
1128 }
1129
1130
1131 switch (header->proto) {
1132 case htons(ETH_P_IP):
1133 case htons(ETH_P_IPV6):
1134 case htons(ETH_P_TIPC):
1135 neigh = ipoib_neigh_get(dev, phdr->hwaddr);
1136 if (unlikely(!neigh)) {
1137 neigh = neigh_add_path(skb, phdr->hwaddr, dev);
1138 if (likely(!neigh))
1139 return NETDEV_TX_OK;
1140 }
1141 break;
1142 case htons(ETH_P_ARP):
1143 case htons(ETH_P_RARP):
1144
1145 unicast_arp_send(skb, dev, phdr);
1146 return NETDEV_TX_OK;
1147 default:
1148
1149 ++dev->stats.tx_dropped;
1150 dev_kfree_skb_any(skb);
1151 return NETDEV_TX_OK;
1152 }
1153
1154 send_using_neigh:
1155
1156 if (ipoib_cm_get(neigh)) {
1157 if (ipoib_cm_up(neigh)) {
1158 ipoib_cm_send(dev, skb, ipoib_cm_get(neigh));
1159 goto unref;
1160 }
1161 } else if (neigh->ah && neigh->ah->valid) {
1162 neigh->ah->last_send = rn->send(dev, skb, neigh->ah->ah,
1163 IPOIB_QPN(phdr->hwaddr));
1164 goto unref;
1165 } else if (neigh->ah) {
1166 neigh_refresh_path(neigh, phdr->hwaddr, dev);
1167 }
1168
1169 if (skb_queue_len(&neigh->queue) < IPOIB_MAX_PATH_REC_QUEUE) {
1170 push_pseudo_header(skb, phdr->hwaddr);
1171 spin_lock_irqsave(&priv->lock, flags);
1172 __skb_queue_tail(&neigh->queue, skb);
1173 spin_unlock_irqrestore(&priv->lock, flags);
1174 } else {
1175 ++dev->stats.tx_dropped;
1176 dev_kfree_skb_any(skb);
1177 }
1178
1179 unref:
1180 ipoib_neigh_put(neigh);
1181
1182 return NETDEV_TX_OK;
1183 }
1184
1185 static void ipoib_timeout(struct net_device *dev)
1186 {
1187 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1188
1189 ipoib_warn(priv, "transmit timeout: latency %d msecs\n",
1190 jiffies_to_msecs(jiffies - dev_trans_start(dev)));
1191 ipoib_warn(priv,
1192 "queue stopped %d, tx_head %u, tx_tail %u, global_tx_head %u, global_tx_tail %u\n",
1193 netif_queue_stopped(dev), priv->tx_head, priv->tx_tail,
1194 priv->global_tx_head, priv->global_tx_tail);
1195
1196
1197 }
1198
1199 static int ipoib_hard_header(struct sk_buff *skb,
1200 struct net_device *dev,
1201 unsigned short type,
1202 const void *daddr,
1203 const void *saddr,
1204 unsigned int len)
1205 {
1206 struct ipoib_header *header;
1207
1208 header = skb_push(skb, sizeof(*header));
1209
1210 header->proto = htons(type);
1211 header->reserved = 0;
1212
1213
1214
1215
1216
1217
1218 push_pseudo_header(skb, daddr);
1219
1220 return IPOIB_HARD_LEN;
1221 }
1222
1223 static void ipoib_set_mcast_list(struct net_device *dev)
1224 {
1225 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1226
1227 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
1228 ipoib_dbg(priv, "IPOIB_FLAG_OPER_UP not set");
1229 return;
1230 }
1231
1232 queue_work(priv->wq, &priv->restart_task);
1233 }
1234
1235 static int ipoib_get_iflink(const struct net_device *dev)
1236 {
1237 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1238
1239
1240 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags))
1241 return dev->ifindex;
1242
1243
1244 return priv->parent->ifindex;
1245 }
1246
1247 static u32 ipoib_addr_hash(struct ipoib_neigh_hash *htbl, u8 *daddr)
1248 {
1249
1250
1251
1252
1253
1254
1255
1256 u32 *d32 = (u32 *) daddr;
1257 u32 hv;
1258
1259 hv = jhash_3words(d32[3], d32[4], IPOIB_QPN_MASK & d32[0], 0);
1260 return hv & htbl->mask;
1261 }
1262
1263 struct ipoib_neigh *ipoib_neigh_get(struct net_device *dev, u8 *daddr)
1264 {
1265 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1266 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1267 struct ipoib_neigh_hash *htbl;
1268 struct ipoib_neigh *neigh = NULL;
1269 u32 hash_val;
1270
1271 rcu_read_lock_bh();
1272
1273 htbl = rcu_dereference_bh(ntbl->htbl);
1274
1275 if (!htbl)
1276 goto out_unlock;
1277
1278 hash_val = ipoib_addr_hash(htbl, daddr);
1279 for (neigh = rcu_dereference_bh(htbl->buckets[hash_val]);
1280 neigh != NULL;
1281 neigh = rcu_dereference_bh(neigh->hnext)) {
1282 if (memcmp(daddr, neigh->daddr, INFINIBAND_ALEN) == 0) {
1283
1284 if (!atomic_inc_not_zero(&neigh->refcnt)) {
1285
1286 neigh = NULL;
1287 goto out_unlock;
1288 }
1289
1290 if (likely(skb_queue_len(&neigh->queue) < IPOIB_MAX_PATH_REC_QUEUE))
1291 neigh->alive = jiffies;
1292 goto out_unlock;
1293 }
1294 }
1295
1296 out_unlock:
1297 rcu_read_unlock_bh();
1298 return neigh;
1299 }
1300
1301 static void __ipoib_reap_neigh(struct ipoib_dev_priv *priv)
1302 {
1303 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1304 struct ipoib_neigh_hash *htbl;
1305 unsigned long neigh_obsolete;
1306 unsigned long dt;
1307 unsigned long flags;
1308 int i;
1309 LIST_HEAD(remove_list);
1310
1311 spin_lock_irqsave(&priv->lock, flags);
1312
1313 htbl = rcu_dereference_protected(ntbl->htbl,
1314 lockdep_is_held(&priv->lock));
1315
1316 if (!htbl)
1317 goto out_unlock;
1318
1319
1320 dt = 2 * arp_tbl.gc_interval;
1321 neigh_obsolete = jiffies - dt;
1322
1323 for (i = 0; i < htbl->size; i++) {
1324 struct ipoib_neigh *neigh;
1325 struct ipoib_neigh __rcu **np = &htbl->buckets[i];
1326
1327 while ((neigh = rcu_dereference_protected(*np,
1328 lockdep_is_held(&priv->lock))) != NULL) {
1329
1330 if (time_after(neigh_obsolete, neigh->alive)) {
1331
1332 ipoib_check_and_add_mcast_sendonly(priv, neigh->daddr + 4, &remove_list);
1333
1334 rcu_assign_pointer(*np,
1335 rcu_dereference_protected(neigh->hnext,
1336 lockdep_is_held(&priv->lock)));
1337
1338 list_del_init(&neigh->list);
1339 call_rcu(&neigh->rcu, ipoib_neigh_reclaim);
1340 } else {
1341 np = &neigh->hnext;
1342 }
1343
1344 }
1345 }
1346
1347 out_unlock:
1348 spin_unlock_irqrestore(&priv->lock, flags);
1349 ipoib_mcast_remove_list(&remove_list);
1350 }
1351
1352 static void ipoib_reap_neigh(struct work_struct *work)
1353 {
1354 struct ipoib_dev_priv *priv =
1355 container_of(work, struct ipoib_dev_priv, neigh_reap_task.work);
1356
1357 __ipoib_reap_neigh(priv);
1358
1359 queue_delayed_work(priv->wq, &priv->neigh_reap_task,
1360 arp_tbl.gc_interval);
1361 }
1362
1363
1364 static struct ipoib_neigh *ipoib_neigh_ctor(u8 *daddr,
1365 struct net_device *dev)
1366 {
1367 struct ipoib_neigh *neigh;
1368
1369 neigh = kzalloc(sizeof(*neigh), GFP_ATOMIC);
1370 if (!neigh)
1371 return NULL;
1372
1373 neigh->dev = dev;
1374 memcpy(&neigh->daddr, daddr, sizeof(neigh->daddr));
1375 skb_queue_head_init(&neigh->queue);
1376 INIT_LIST_HEAD(&neigh->list);
1377 ipoib_cm_set(neigh, NULL);
1378
1379 atomic_set(&neigh->refcnt, 1);
1380
1381 return neigh;
1382 }
1383
1384 struct ipoib_neigh *ipoib_neigh_alloc(u8 *daddr,
1385 struct net_device *dev)
1386 {
1387 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1388 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1389 struct ipoib_neigh_hash *htbl;
1390 struct ipoib_neigh *neigh;
1391 u32 hash_val;
1392
1393 htbl = rcu_dereference_protected(ntbl->htbl,
1394 lockdep_is_held(&priv->lock));
1395 if (!htbl) {
1396 neigh = NULL;
1397 goto out_unlock;
1398 }
1399
1400
1401
1402
1403 hash_val = ipoib_addr_hash(htbl, daddr);
1404 for (neigh = rcu_dereference_protected(htbl->buckets[hash_val],
1405 lockdep_is_held(&priv->lock));
1406 neigh != NULL;
1407 neigh = rcu_dereference_protected(neigh->hnext,
1408 lockdep_is_held(&priv->lock))) {
1409 if (memcmp(daddr, neigh->daddr, INFINIBAND_ALEN) == 0) {
1410
1411 if (!atomic_inc_not_zero(&neigh->refcnt)) {
1412
1413 neigh = NULL;
1414 break;
1415 }
1416 neigh->alive = jiffies;
1417 goto out_unlock;
1418 }
1419 }
1420
1421 neigh = ipoib_neigh_ctor(daddr, dev);
1422 if (!neigh)
1423 goto out_unlock;
1424
1425
1426 atomic_inc(&neigh->refcnt);
1427 neigh->alive = jiffies;
1428
1429 rcu_assign_pointer(neigh->hnext,
1430 rcu_dereference_protected(htbl->buckets[hash_val],
1431 lockdep_is_held(&priv->lock)));
1432 rcu_assign_pointer(htbl->buckets[hash_val], neigh);
1433 atomic_inc(&ntbl->entries);
1434
1435 out_unlock:
1436
1437 return neigh;
1438 }
1439
1440 void ipoib_neigh_dtor(struct ipoib_neigh *neigh)
1441 {
1442
1443 struct net_device *dev = neigh->dev;
1444 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1445 struct sk_buff *skb;
1446 if (neigh->ah)
1447 ipoib_put_ah(neigh->ah);
1448 while ((skb = __skb_dequeue(&neigh->queue))) {
1449 ++dev->stats.tx_dropped;
1450 dev_kfree_skb_any(skb);
1451 }
1452 if (ipoib_cm_get(neigh))
1453 ipoib_cm_destroy_tx(ipoib_cm_get(neigh));
1454 ipoib_dbg(ipoib_priv(dev),
1455 "neigh free for %06x %pI6\n",
1456 IPOIB_QPN(neigh->daddr),
1457 neigh->daddr + 4);
1458 kfree(neigh);
1459 if (atomic_dec_and_test(&priv->ntbl.entries)) {
1460 if (test_bit(IPOIB_NEIGH_TBL_FLUSH, &priv->flags))
1461 complete(&priv->ntbl.flushed);
1462 }
1463 }
1464
1465 static void ipoib_neigh_reclaim(struct rcu_head *rp)
1466 {
1467
1468 struct ipoib_neigh *neigh = container_of(rp, struct ipoib_neigh, rcu);
1469
1470 ipoib_neigh_put(neigh);
1471 }
1472
1473 void ipoib_neigh_free(struct ipoib_neigh *neigh)
1474 {
1475 struct net_device *dev = neigh->dev;
1476 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1477 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1478 struct ipoib_neigh_hash *htbl;
1479 struct ipoib_neigh __rcu **np;
1480 struct ipoib_neigh *n;
1481 u32 hash_val;
1482
1483 htbl = rcu_dereference_protected(ntbl->htbl,
1484 lockdep_is_held(&priv->lock));
1485 if (!htbl)
1486 return;
1487
1488 hash_val = ipoib_addr_hash(htbl, neigh->daddr);
1489 np = &htbl->buckets[hash_val];
1490 for (n = rcu_dereference_protected(*np,
1491 lockdep_is_held(&priv->lock));
1492 n != NULL;
1493 n = rcu_dereference_protected(*np,
1494 lockdep_is_held(&priv->lock))) {
1495 if (n == neigh) {
1496
1497 rcu_assign_pointer(*np,
1498 rcu_dereference_protected(neigh->hnext,
1499 lockdep_is_held(&priv->lock)));
1500
1501 list_del_init(&neigh->list);
1502 call_rcu(&neigh->rcu, ipoib_neigh_reclaim);
1503 return;
1504 } else {
1505 np = &n->hnext;
1506 }
1507 }
1508 }
1509
1510 static int ipoib_neigh_hash_init(struct ipoib_dev_priv *priv)
1511 {
1512 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1513 struct ipoib_neigh_hash *htbl;
1514 struct ipoib_neigh __rcu **buckets;
1515 u32 size;
1516
1517 clear_bit(IPOIB_NEIGH_TBL_FLUSH, &priv->flags);
1518 ntbl->htbl = NULL;
1519 htbl = kzalloc(sizeof(*htbl), GFP_KERNEL);
1520 if (!htbl)
1521 return -ENOMEM;
1522 size = roundup_pow_of_two(arp_tbl.gc_thresh3);
1523 buckets = kvcalloc(size, sizeof(*buckets), GFP_KERNEL);
1524 if (!buckets) {
1525 kfree(htbl);
1526 return -ENOMEM;
1527 }
1528 htbl->size = size;
1529 htbl->mask = (size - 1);
1530 htbl->buckets = buckets;
1531 RCU_INIT_POINTER(ntbl->htbl, htbl);
1532 htbl->ntbl = ntbl;
1533 atomic_set(&ntbl->entries, 0);
1534
1535
1536 queue_delayed_work(priv->wq, &priv->neigh_reap_task,
1537 arp_tbl.gc_interval);
1538
1539 return 0;
1540 }
1541
1542 static void neigh_hash_free_rcu(struct rcu_head *head)
1543 {
1544 struct ipoib_neigh_hash *htbl = container_of(head,
1545 struct ipoib_neigh_hash,
1546 rcu);
1547 struct ipoib_neigh __rcu **buckets = htbl->buckets;
1548 struct ipoib_neigh_table *ntbl = htbl->ntbl;
1549
1550 kvfree(buckets);
1551 kfree(htbl);
1552 complete(&ntbl->deleted);
1553 }
1554
1555 void ipoib_del_neighs_by_gid(struct net_device *dev, u8 *gid)
1556 {
1557 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1558 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1559 struct ipoib_neigh_hash *htbl;
1560 unsigned long flags;
1561 int i;
1562
1563
1564 spin_lock_irqsave(&priv->lock, flags);
1565
1566 htbl = rcu_dereference_protected(ntbl->htbl,
1567 lockdep_is_held(&priv->lock));
1568
1569 if (!htbl)
1570 goto out_unlock;
1571
1572 for (i = 0; i < htbl->size; i++) {
1573 struct ipoib_neigh *neigh;
1574 struct ipoib_neigh __rcu **np = &htbl->buckets[i];
1575
1576 while ((neigh = rcu_dereference_protected(*np,
1577 lockdep_is_held(&priv->lock))) != NULL) {
1578
1579 if (!memcmp(gid, neigh->daddr + 4, sizeof (union ib_gid))) {
1580 rcu_assign_pointer(*np,
1581 rcu_dereference_protected(neigh->hnext,
1582 lockdep_is_held(&priv->lock)));
1583
1584 list_del_init(&neigh->list);
1585 call_rcu(&neigh->rcu, ipoib_neigh_reclaim);
1586 } else {
1587 np = &neigh->hnext;
1588 }
1589
1590 }
1591 }
1592 out_unlock:
1593 spin_unlock_irqrestore(&priv->lock, flags);
1594 }
1595
1596 static void ipoib_flush_neighs(struct ipoib_dev_priv *priv)
1597 {
1598 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1599 struct ipoib_neigh_hash *htbl;
1600 unsigned long flags;
1601 int i, wait_flushed = 0;
1602
1603 init_completion(&priv->ntbl.flushed);
1604 set_bit(IPOIB_NEIGH_TBL_FLUSH, &priv->flags);
1605
1606 spin_lock_irqsave(&priv->lock, flags);
1607
1608 htbl = rcu_dereference_protected(ntbl->htbl,
1609 lockdep_is_held(&priv->lock));
1610 if (!htbl)
1611 goto out_unlock;
1612
1613 wait_flushed = atomic_read(&priv->ntbl.entries);
1614 if (!wait_flushed)
1615 goto free_htbl;
1616
1617 for (i = 0; i < htbl->size; i++) {
1618 struct ipoib_neigh *neigh;
1619 struct ipoib_neigh __rcu **np = &htbl->buckets[i];
1620
1621 while ((neigh = rcu_dereference_protected(*np,
1622 lockdep_is_held(&priv->lock))) != NULL) {
1623 rcu_assign_pointer(*np,
1624 rcu_dereference_protected(neigh->hnext,
1625 lockdep_is_held(&priv->lock)));
1626
1627 list_del_init(&neigh->list);
1628 call_rcu(&neigh->rcu, ipoib_neigh_reclaim);
1629 }
1630 }
1631
1632 free_htbl:
1633 rcu_assign_pointer(ntbl->htbl, NULL);
1634 call_rcu(&htbl->rcu, neigh_hash_free_rcu);
1635
1636 out_unlock:
1637 spin_unlock_irqrestore(&priv->lock, flags);
1638 if (wait_flushed)
1639 wait_for_completion(&priv->ntbl.flushed);
1640 }
1641
1642 static void ipoib_neigh_hash_uninit(struct net_device *dev)
1643 {
1644 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1645
1646 ipoib_dbg(priv, "%s\n", __func__);
1647 init_completion(&priv->ntbl.deleted);
1648
1649 cancel_delayed_work_sync(&priv->neigh_reap_task);
1650
1651 ipoib_flush_neighs(priv);
1652
1653 wait_for_completion(&priv->ntbl.deleted);
1654 }
1655
1656 static void ipoib_napi_add(struct net_device *dev)
1657 {
1658 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1659
1660 netif_napi_add(dev, &priv->recv_napi, ipoib_rx_poll, IPOIB_NUM_WC);
1661 netif_napi_add(dev, &priv->send_napi, ipoib_tx_poll, MAX_SEND_CQE);
1662 }
1663
1664 static void ipoib_napi_del(struct net_device *dev)
1665 {
1666 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1667
1668 netif_napi_del(&priv->recv_napi);
1669 netif_napi_del(&priv->send_napi);
1670 }
1671
1672 static void ipoib_dev_uninit_default(struct net_device *dev)
1673 {
1674 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1675
1676 ipoib_transport_dev_cleanup(dev);
1677
1678 ipoib_napi_del(dev);
1679
1680 ipoib_cm_dev_cleanup(dev);
1681
1682 kfree(priv->rx_ring);
1683 vfree(priv->tx_ring);
1684
1685 priv->rx_ring = NULL;
1686 priv->tx_ring = NULL;
1687 }
1688
1689 static int ipoib_dev_init_default(struct net_device *dev)
1690 {
1691 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1692
1693 ipoib_napi_add(dev);
1694
1695
1696 priv->rx_ring = kcalloc(ipoib_recvq_size,
1697 sizeof(*priv->rx_ring),
1698 GFP_KERNEL);
1699 if (!priv->rx_ring)
1700 goto out;
1701
1702 priv->tx_ring = vzalloc(array_size(ipoib_sendq_size,
1703 sizeof(*priv->tx_ring)));
1704 if (!priv->tx_ring) {
1705 pr_warn("%s: failed to allocate TX ring (%d entries)\n",
1706 priv->ca->name, ipoib_sendq_size);
1707 goto out_rx_ring_cleanup;
1708 }
1709
1710
1711
1712 if (ipoib_transport_dev_init(dev, priv->ca)) {
1713 pr_warn("%s: ipoib_transport_dev_init failed\n",
1714 priv->ca->name);
1715 goto out_tx_ring_cleanup;
1716 }
1717
1718
1719 priv->dev->dev_addr[1] = (priv->qp->qp_num >> 16) & 0xff;
1720 priv->dev->dev_addr[2] = (priv->qp->qp_num >> 8) & 0xff;
1721 priv->dev->dev_addr[3] = (priv->qp->qp_num) & 0xff;
1722
1723 return 0;
1724
1725 out_tx_ring_cleanup:
1726 vfree(priv->tx_ring);
1727
1728 out_rx_ring_cleanup:
1729 kfree(priv->rx_ring);
1730
1731 out:
1732 ipoib_napi_del(dev);
1733 return -ENOMEM;
1734 }
1735
1736 static int ipoib_ioctl(struct net_device *dev, struct ifreq *ifr,
1737 int cmd)
1738 {
1739 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1740
1741 if (!priv->rn_ops->ndo_do_ioctl)
1742 return -EOPNOTSUPP;
1743
1744 return priv->rn_ops->ndo_do_ioctl(dev, ifr, cmd);
1745 }
1746
1747 static int ipoib_dev_init(struct net_device *dev)
1748 {
1749 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1750 int ret = -ENOMEM;
1751
1752 priv->qp = NULL;
1753
1754
1755
1756
1757
1758 priv->wq = alloc_ordered_workqueue("ipoib_wq", WQ_MEM_RECLAIM);
1759 if (!priv->wq) {
1760 pr_warn("%s: failed to allocate device WQ\n", dev->name);
1761 goto out;
1762 }
1763
1764
1765 priv->pd = ib_alloc_pd(priv->ca, 0);
1766 if (IS_ERR(priv->pd)) {
1767 pr_warn("%s: failed to allocate PD\n", priv->ca->name);
1768 goto clean_wq;
1769 }
1770
1771 ret = priv->rn_ops->ndo_init(dev);
1772 if (ret) {
1773 pr_warn("%s failed to init HW resource\n", dev->name);
1774 goto out_free_pd;
1775 }
1776
1777 ret = ipoib_neigh_hash_init(priv);
1778 if (ret) {
1779 pr_warn("%s failed to init neigh hash\n", dev->name);
1780 goto out_dev_uninit;
1781 }
1782
1783 if (dev->flags & IFF_UP) {
1784 if (ipoib_ib_dev_open(dev)) {
1785 pr_warn("%s failed to open device\n", dev->name);
1786 ret = -ENODEV;
1787 goto out_hash_uninit;
1788 }
1789 }
1790
1791 return 0;
1792
1793 out_hash_uninit:
1794 ipoib_neigh_hash_uninit(dev);
1795
1796 out_dev_uninit:
1797 ipoib_ib_dev_cleanup(dev);
1798
1799 out_free_pd:
1800 if (priv->pd) {
1801 ib_dealloc_pd(priv->pd);
1802 priv->pd = NULL;
1803 }
1804
1805 clean_wq:
1806 if (priv->wq) {
1807 destroy_workqueue(priv->wq);
1808 priv->wq = NULL;
1809 }
1810
1811 out:
1812 return ret;
1813 }
1814
1815
1816
1817
1818
1819 static void ipoib_parent_unregister_pre(struct net_device *ndev)
1820 {
1821 struct ipoib_dev_priv *priv = ipoib_priv(ndev);
1822
1823
1824
1825
1826
1827 rtnl_lock();
1828 dev_change_flags(priv->dev, priv->dev->flags & ~IFF_UP, NULL);
1829 rtnl_unlock();
1830
1831
1832 ib_unregister_event_handler(&priv->event_handler);
1833
1834
1835
1836
1837
1838 flush_workqueue(ipoib_workqueue);
1839 }
1840
1841 static void ipoib_set_dev_features(struct ipoib_dev_priv *priv)
1842 {
1843 priv->hca_caps = priv->ca->attrs.device_cap_flags;
1844
1845 if (priv->hca_caps & IB_DEVICE_UD_IP_CSUM) {
1846 priv->dev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_RXCSUM;
1847
1848 if (priv->hca_caps & IB_DEVICE_UD_TSO)
1849 priv->dev->hw_features |= NETIF_F_TSO;
1850
1851 priv->dev->features |= priv->dev->hw_features;
1852 }
1853 }
1854
1855 static int ipoib_parent_init(struct net_device *ndev)
1856 {
1857 struct ipoib_dev_priv *priv = ipoib_priv(ndev);
1858 struct ib_port_attr attr;
1859 int result;
1860
1861 result = ib_query_port(priv->ca, priv->port, &attr);
1862 if (result) {
1863 pr_warn("%s: ib_query_port %d failed\n", priv->ca->name,
1864 priv->port);
1865 return result;
1866 }
1867 priv->max_ib_mtu = ib_mtu_enum_to_int(attr.max_mtu);
1868
1869 result = ib_query_pkey(priv->ca, priv->port, 0, &priv->pkey);
1870 if (result) {
1871 pr_warn("%s: ib_query_pkey port %d failed (ret = %d)\n",
1872 priv->ca->name, priv->port, result);
1873 return result;
1874 }
1875
1876 result = rdma_query_gid(priv->ca, priv->port, 0, &priv->local_gid);
1877 if (result) {
1878 pr_warn("%s: rdma_query_gid port %d failed (ret = %d)\n",
1879 priv->ca->name, priv->port, result);
1880 return result;
1881 }
1882 memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw,
1883 sizeof(union ib_gid));
1884
1885 SET_NETDEV_DEV(priv->dev, priv->ca->dev.parent);
1886 priv->dev->dev_port = priv->port - 1;
1887
1888 priv->dev->dev_id = priv->port - 1;
1889
1890 return 0;
1891 }
1892
1893 static void ipoib_child_init(struct net_device *ndev)
1894 {
1895 struct ipoib_dev_priv *priv = ipoib_priv(ndev);
1896 struct ipoib_dev_priv *ppriv = ipoib_priv(priv->parent);
1897
1898 priv->max_ib_mtu = ppriv->max_ib_mtu;
1899 set_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags);
1900 memcpy(priv->dev->dev_addr, ppriv->dev->dev_addr, INFINIBAND_ALEN);
1901 memcpy(&priv->local_gid, &ppriv->local_gid, sizeof(priv->local_gid));
1902 }
1903
1904 static int ipoib_ndo_init(struct net_device *ndev)
1905 {
1906 struct ipoib_dev_priv *priv = ipoib_priv(ndev);
1907 int rc;
1908
1909 if (priv->parent) {
1910 ipoib_child_init(ndev);
1911 } else {
1912 rc = ipoib_parent_init(ndev);
1913 if (rc)
1914 return rc;
1915 }
1916
1917
1918 ndev->mtu = IPOIB_UD_MTU(priv->max_ib_mtu);
1919 priv->mcast_mtu = priv->admin_mtu = ndev->mtu;
1920 ndev->max_mtu = IPOIB_CM_MTU;
1921
1922 ndev->neigh_priv_len = sizeof(struct ipoib_neigh);
1923
1924
1925
1926
1927
1928 priv->pkey |= 0x8000;
1929
1930 ndev->broadcast[8] = priv->pkey >> 8;
1931 ndev->broadcast[9] = priv->pkey & 0xff;
1932 set_bit(IPOIB_FLAG_DEV_ADDR_SET, &priv->flags);
1933
1934 ipoib_set_dev_features(priv);
1935
1936 rc = ipoib_dev_init(ndev);
1937 if (rc) {
1938 pr_warn("%s: failed to initialize device: %s port %d (ret = %d)\n",
1939 priv->ca->name, priv->dev->name, priv->port, rc);
1940 return rc;
1941 }
1942
1943 if (priv->parent) {
1944 struct ipoib_dev_priv *ppriv = ipoib_priv(priv->parent);
1945
1946 dev_hold(priv->parent);
1947
1948 down_write(&ppriv->vlan_rwsem);
1949 list_add_tail(&priv->list, &ppriv->child_intfs);
1950 up_write(&ppriv->vlan_rwsem);
1951 }
1952
1953 return 0;
1954 }
1955
1956 static void ipoib_ndo_uninit(struct net_device *dev)
1957 {
1958 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1959
1960 ASSERT_RTNL();
1961
1962
1963
1964
1965
1966 WARN_ON(!list_empty(&priv->child_intfs));
1967
1968 if (priv->parent) {
1969 struct ipoib_dev_priv *ppriv = ipoib_priv(priv->parent);
1970
1971 down_write(&ppriv->vlan_rwsem);
1972 list_del(&priv->list);
1973 up_write(&ppriv->vlan_rwsem);
1974 }
1975
1976 ipoib_neigh_hash_uninit(dev);
1977
1978 ipoib_ib_dev_cleanup(dev);
1979
1980
1981 if (priv->wq) {
1982 flush_workqueue(priv->wq);
1983 destroy_workqueue(priv->wq);
1984 priv->wq = NULL;
1985 }
1986
1987 if (priv->parent)
1988 dev_put(priv->parent);
1989 }
1990
1991 static int ipoib_set_vf_link_state(struct net_device *dev, int vf, int link_state)
1992 {
1993 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1994
1995 return ib_set_vf_link_state(priv->ca, vf, priv->port, link_state);
1996 }
1997
1998 static int ipoib_get_vf_config(struct net_device *dev, int vf,
1999 struct ifla_vf_info *ivf)
2000 {
2001 struct ipoib_dev_priv *priv = ipoib_priv(dev);
2002 int err;
2003
2004 err = ib_get_vf_config(priv->ca, vf, priv->port, ivf);
2005 if (err)
2006 return err;
2007
2008 ivf->vf = vf;
2009 memcpy(ivf->mac, dev->dev_addr, dev->addr_len);
2010
2011 return 0;
2012 }
2013
2014 static int ipoib_set_vf_guid(struct net_device *dev, int vf, u64 guid, int type)
2015 {
2016 struct ipoib_dev_priv *priv = ipoib_priv(dev);
2017
2018 if (type != IFLA_VF_IB_NODE_GUID && type != IFLA_VF_IB_PORT_GUID)
2019 return -EINVAL;
2020
2021 return ib_set_vf_guid(priv->ca, vf, priv->port, guid, type);
2022 }
2023
2024 static int ipoib_get_vf_stats(struct net_device *dev, int vf,
2025 struct ifla_vf_stats *vf_stats)
2026 {
2027 struct ipoib_dev_priv *priv = ipoib_priv(dev);
2028
2029 return ib_get_vf_stats(priv->ca, vf, priv->port, vf_stats);
2030 }
2031
2032 static const struct header_ops ipoib_header_ops = {
2033 .create = ipoib_hard_header,
2034 };
2035
2036 static const struct net_device_ops ipoib_netdev_ops_pf = {
2037 .ndo_init = ipoib_ndo_init,
2038 .ndo_uninit = ipoib_ndo_uninit,
2039 .ndo_open = ipoib_open,
2040 .ndo_stop = ipoib_stop,
2041 .ndo_change_mtu = ipoib_change_mtu,
2042 .ndo_fix_features = ipoib_fix_features,
2043 .ndo_start_xmit = ipoib_start_xmit,
2044 .ndo_tx_timeout = ipoib_timeout,
2045 .ndo_set_rx_mode = ipoib_set_mcast_list,
2046 .ndo_get_iflink = ipoib_get_iflink,
2047 .ndo_set_vf_link_state = ipoib_set_vf_link_state,
2048 .ndo_get_vf_config = ipoib_get_vf_config,
2049 .ndo_get_vf_stats = ipoib_get_vf_stats,
2050 .ndo_set_vf_guid = ipoib_set_vf_guid,
2051 .ndo_set_mac_address = ipoib_set_mac,
2052 .ndo_get_stats64 = ipoib_get_stats,
2053 .ndo_do_ioctl = ipoib_ioctl,
2054 };
2055
2056 static const struct net_device_ops ipoib_netdev_ops_vf = {
2057 .ndo_init = ipoib_ndo_init,
2058 .ndo_uninit = ipoib_ndo_uninit,
2059 .ndo_open = ipoib_open,
2060 .ndo_stop = ipoib_stop,
2061 .ndo_change_mtu = ipoib_change_mtu,
2062 .ndo_fix_features = ipoib_fix_features,
2063 .ndo_start_xmit = ipoib_start_xmit,
2064 .ndo_tx_timeout = ipoib_timeout,
2065 .ndo_set_rx_mode = ipoib_set_mcast_list,
2066 .ndo_get_iflink = ipoib_get_iflink,
2067 .ndo_get_stats64 = ipoib_get_stats,
2068 .ndo_do_ioctl = ipoib_ioctl,
2069 };
2070
2071 void ipoib_setup_common(struct net_device *dev)
2072 {
2073 dev->header_ops = &ipoib_header_ops;
2074
2075 ipoib_set_ethtool_ops(dev);
2076
2077 dev->watchdog_timeo = HZ;
2078
2079 dev->flags |= IFF_BROADCAST | IFF_MULTICAST;
2080
2081 dev->hard_header_len = IPOIB_HARD_LEN;
2082 dev->addr_len = INFINIBAND_ALEN;
2083 dev->type = ARPHRD_INFINIBAND;
2084 dev->tx_queue_len = ipoib_sendq_size * 2;
2085 dev->features = (NETIF_F_VLAN_CHALLENGED |
2086 NETIF_F_HIGHDMA);
2087 netif_keep_dst(dev);
2088
2089 memcpy(dev->broadcast, ipv4_bcast_addr, INFINIBAND_ALEN);
2090
2091
2092
2093
2094
2095
2096 dev->needs_free_netdev = true;
2097 }
2098
2099 static void ipoib_build_priv(struct net_device *dev)
2100 {
2101 struct ipoib_dev_priv *priv = ipoib_priv(dev);
2102
2103 priv->dev = dev;
2104 spin_lock_init(&priv->lock);
2105 init_rwsem(&priv->vlan_rwsem);
2106 mutex_init(&priv->mcast_mutex);
2107
2108 INIT_LIST_HEAD(&priv->path_list);
2109 INIT_LIST_HEAD(&priv->child_intfs);
2110 INIT_LIST_HEAD(&priv->dead_ahs);
2111 INIT_LIST_HEAD(&priv->multicast_list);
2112
2113 INIT_DELAYED_WORK(&priv->mcast_task, ipoib_mcast_join_task);
2114 INIT_WORK(&priv->carrier_on_task, ipoib_mcast_carrier_on_task);
2115 INIT_WORK(&priv->flush_light, ipoib_ib_dev_flush_light);
2116 INIT_WORK(&priv->flush_normal, ipoib_ib_dev_flush_normal);
2117 INIT_WORK(&priv->flush_heavy, ipoib_ib_dev_flush_heavy);
2118 INIT_WORK(&priv->restart_task, ipoib_mcast_restart_task);
2119 INIT_DELAYED_WORK(&priv->ah_reap_task, ipoib_reap_ah);
2120 INIT_DELAYED_WORK(&priv->neigh_reap_task, ipoib_reap_neigh);
2121 }
2122
2123 static const struct net_device_ops ipoib_netdev_default_pf = {
2124 .ndo_init = ipoib_dev_init_default,
2125 .ndo_uninit = ipoib_dev_uninit_default,
2126 .ndo_open = ipoib_ib_dev_open_default,
2127 .ndo_stop = ipoib_ib_dev_stop_default,
2128 };
2129
2130 static struct net_device *ipoib_alloc_netdev(struct ib_device *hca, u8 port,
2131 const char *name)
2132 {
2133 struct net_device *dev;
2134
2135 dev = rdma_alloc_netdev(hca, port, RDMA_NETDEV_IPOIB, name,
2136 NET_NAME_UNKNOWN, ipoib_setup_common);
2137 if (!IS_ERR(dev) || PTR_ERR(dev) != -EOPNOTSUPP)
2138 return dev;
2139
2140 dev = alloc_netdev(sizeof(struct rdma_netdev), name, NET_NAME_UNKNOWN,
2141 ipoib_setup_common);
2142 if (!dev)
2143 return ERR_PTR(-ENOMEM);
2144 return dev;
2145 }
2146
2147 int ipoib_intf_init(struct ib_device *hca, u8 port, const char *name,
2148 struct net_device *dev)
2149 {
2150 struct rdma_netdev *rn = netdev_priv(dev);
2151 struct ipoib_dev_priv *priv;
2152 int rc;
2153
2154 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
2155 if (!priv)
2156 return -ENOMEM;
2157
2158 priv->ca = hca;
2159 priv->port = port;
2160
2161 rc = rdma_init_netdev(hca, port, RDMA_NETDEV_IPOIB, name,
2162 NET_NAME_UNKNOWN, ipoib_setup_common, dev);
2163 if (rc) {
2164 if (rc != -EOPNOTSUPP)
2165 goto out;
2166
2167 dev->netdev_ops = &ipoib_netdev_default_pf;
2168 rn->send = ipoib_send;
2169 rn->attach_mcast = ipoib_mcast_attach;
2170 rn->detach_mcast = ipoib_mcast_detach;
2171 rn->hca = hca;
2172 }
2173
2174 priv->rn_ops = dev->netdev_ops;
2175
2176 if (hca->attrs.device_cap_flags & IB_DEVICE_VIRTUAL_FUNCTION)
2177 dev->netdev_ops = &ipoib_netdev_ops_vf;
2178 else
2179 dev->netdev_ops = &ipoib_netdev_ops_pf;
2180
2181 rn->clnt_priv = priv;
2182
2183
2184
2185
2186
2187 priv->next_priv_destructor = dev->priv_destructor;
2188 dev->priv_destructor = NULL;
2189
2190 ipoib_build_priv(dev);
2191
2192 return 0;
2193
2194 out:
2195 kfree(priv);
2196 return rc;
2197 }
2198
2199 struct net_device *ipoib_intf_alloc(struct ib_device *hca, u8 port,
2200 const char *name)
2201 {
2202 struct net_device *dev;
2203 int rc;
2204
2205 dev = ipoib_alloc_netdev(hca, port, name);
2206 if (IS_ERR(dev))
2207 return dev;
2208
2209 rc = ipoib_intf_init(hca, port, name, dev);
2210 if (rc) {
2211 free_netdev(dev);
2212 return ERR_PTR(rc);
2213 }
2214
2215
2216
2217
2218
2219
2220 return dev;
2221 }
2222
2223 void ipoib_intf_free(struct net_device *dev)
2224 {
2225 struct ipoib_dev_priv *priv = ipoib_priv(dev);
2226 struct rdma_netdev *rn = netdev_priv(dev);
2227
2228 dev->priv_destructor = priv->next_priv_destructor;
2229 if (dev->priv_destructor)
2230 dev->priv_destructor(dev);
2231
2232
2233
2234
2235
2236 dev->priv_destructor = NULL;
2237
2238
2239 rn->clnt_priv = NULL;
2240
2241 kfree(priv);
2242 }
2243
2244 static ssize_t show_pkey(struct device *dev,
2245 struct device_attribute *attr, char *buf)
2246 {
2247 struct net_device *ndev = to_net_dev(dev);
2248 struct ipoib_dev_priv *priv = ipoib_priv(ndev);
2249
2250 return sprintf(buf, "0x%04x\n", priv->pkey);
2251 }
2252 static DEVICE_ATTR(pkey, S_IRUGO, show_pkey, NULL);
2253
2254 static ssize_t show_umcast(struct device *dev,
2255 struct device_attribute *attr, char *buf)
2256 {
2257 struct net_device *ndev = to_net_dev(dev);
2258 struct ipoib_dev_priv *priv = ipoib_priv(ndev);
2259
2260 return sprintf(buf, "%d\n", test_bit(IPOIB_FLAG_UMCAST, &priv->flags));
2261 }
2262
2263 void ipoib_set_umcast(struct net_device *ndev, int umcast_val)
2264 {
2265 struct ipoib_dev_priv *priv = ipoib_priv(ndev);
2266
2267 if (umcast_val > 0) {
2268 set_bit(IPOIB_FLAG_UMCAST, &priv->flags);
2269 ipoib_warn(priv, "ignoring multicast groups joined directly "
2270 "by userspace\n");
2271 } else
2272 clear_bit(IPOIB_FLAG_UMCAST, &priv->flags);
2273 }
2274
2275 static ssize_t set_umcast(struct device *dev,
2276 struct device_attribute *attr,
2277 const char *buf, size_t count)
2278 {
2279 unsigned long umcast_val = simple_strtoul(buf, NULL, 0);
2280
2281 ipoib_set_umcast(to_net_dev(dev), umcast_val);
2282
2283 return count;
2284 }
2285 static DEVICE_ATTR(umcast, S_IWUSR | S_IRUGO, show_umcast, set_umcast);
2286
2287 int ipoib_add_umcast_attr(struct net_device *dev)
2288 {
2289 return device_create_file(&dev->dev, &dev_attr_umcast);
2290 }
2291
2292 static void set_base_guid(struct ipoib_dev_priv *priv, union ib_gid *gid)
2293 {
2294 struct ipoib_dev_priv *child_priv;
2295 struct net_device *netdev = priv->dev;
2296
2297 netif_addr_lock_bh(netdev);
2298
2299 memcpy(&priv->local_gid.global.interface_id,
2300 &gid->global.interface_id,
2301 sizeof(gid->global.interface_id));
2302 memcpy(netdev->dev_addr + 4, &priv->local_gid, sizeof(priv->local_gid));
2303 clear_bit(IPOIB_FLAG_DEV_ADDR_SET, &priv->flags);
2304
2305 netif_addr_unlock_bh(netdev);
2306
2307 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
2308 down_read(&priv->vlan_rwsem);
2309 list_for_each_entry(child_priv, &priv->child_intfs, list)
2310 set_base_guid(child_priv, gid);
2311 up_read(&priv->vlan_rwsem);
2312 }
2313 }
2314
2315 static int ipoib_check_lladdr(struct net_device *dev,
2316 struct sockaddr_storage *ss)
2317 {
2318 union ib_gid *gid = (union ib_gid *)(ss->__data + 4);
2319 int ret = 0;
2320
2321 netif_addr_lock_bh(dev);
2322
2323
2324
2325
2326 if (memcmp(dev->dev_addr, ss->__data,
2327 4 + sizeof(gid->global.subnet_prefix)) ||
2328 gid->global.interface_id == 0)
2329 ret = -EINVAL;
2330
2331 netif_addr_unlock_bh(dev);
2332
2333 return ret;
2334 }
2335
2336 static int ipoib_set_mac(struct net_device *dev, void *addr)
2337 {
2338 struct ipoib_dev_priv *priv = ipoib_priv(dev);
2339 struct sockaddr_storage *ss = addr;
2340 int ret;
2341
2342 if (!(dev->priv_flags & IFF_LIVE_ADDR_CHANGE) && netif_running(dev))
2343 return -EBUSY;
2344
2345 ret = ipoib_check_lladdr(dev, ss);
2346 if (ret)
2347 return ret;
2348
2349 set_base_guid(priv, (union ib_gid *)(ss->__data + 4));
2350
2351 queue_work(ipoib_workqueue, &priv->flush_light);
2352
2353 return 0;
2354 }
2355
2356 static ssize_t create_child(struct device *dev,
2357 struct device_attribute *attr,
2358 const char *buf, size_t count)
2359 {
2360 int pkey;
2361 int ret;
2362
2363 if (sscanf(buf, "%i", &pkey) != 1)
2364 return -EINVAL;
2365
2366 if (pkey <= 0 || pkey > 0xffff || pkey == 0x8000)
2367 return -EINVAL;
2368
2369 ret = ipoib_vlan_add(to_net_dev(dev), pkey);
2370
2371 return ret ? ret : count;
2372 }
2373 static DEVICE_ATTR(create_child, S_IWUSR, NULL, create_child);
2374
2375 static ssize_t delete_child(struct device *dev,
2376 struct device_attribute *attr,
2377 const char *buf, size_t count)
2378 {
2379 int pkey;
2380 int ret;
2381
2382 if (sscanf(buf, "%i", &pkey) != 1)
2383 return -EINVAL;
2384
2385 if (pkey < 0 || pkey > 0xffff)
2386 return -EINVAL;
2387
2388 ret = ipoib_vlan_delete(to_net_dev(dev), pkey);
2389
2390 return ret ? ret : count;
2391
2392 }
2393 static DEVICE_ATTR(delete_child, S_IWUSR, NULL, delete_child);
2394
2395 int ipoib_add_pkey_attr(struct net_device *dev)
2396 {
2397 return device_create_file(&dev->dev, &dev_attr_pkey);
2398 }
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409 static ssize_t dev_id_show(struct device *dev,
2410 struct device_attribute *attr, char *buf)
2411 {
2412 struct net_device *ndev = to_net_dev(dev);
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425 if (ndev->dev_port && ndev->dev_id == ndev->dev_port)
2426 netdev_info_once(ndev,
2427 "\"%s\" wants to know my dev_id. Should it look at dev_port instead? See Documentation/ABI/testing/sysfs-class-net for more info.\n",
2428 current->comm);
2429
2430 return sprintf(buf, "%#x\n", ndev->dev_id);
2431 }
2432 static DEVICE_ATTR_RO(dev_id);
2433
2434 static int ipoib_intercept_dev_id_attr(struct net_device *dev)
2435 {
2436 device_remove_file(&dev->dev, &dev_attr_dev_id);
2437 return device_create_file(&dev->dev, &dev_attr_dev_id);
2438 }
2439
2440 static struct net_device *ipoib_add_port(const char *format,
2441 struct ib_device *hca, u8 port)
2442 {
2443 struct rtnl_link_ops *ops = ipoib_get_link_ops();
2444 struct rdma_netdev_alloc_params params;
2445 struct ipoib_dev_priv *priv;
2446 struct net_device *ndev;
2447 int result;
2448
2449 ndev = ipoib_intf_alloc(hca, port, format);
2450 if (IS_ERR(ndev)) {
2451 pr_warn("%s, %d: ipoib_intf_alloc failed %ld\n", hca->name, port,
2452 PTR_ERR(ndev));
2453 return ndev;
2454 }
2455 priv = ipoib_priv(ndev);
2456
2457 INIT_IB_EVENT_HANDLER(&priv->event_handler,
2458 priv->ca, ipoib_event);
2459 ib_register_event_handler(&priv->event_handler);
2460
2461
2462 queue_work(ipoib_workqueue, &priv->flush_heavy);
2463
2464 result = register_netdev(ndev);
2465 if (result) {
2466 pr_warn("%s: couldn't register ipoib port %d; error %d\n",
2467 hca->name, port, result);
2468
2469 ipoib_parent_unregister_pre(ndev);
2470 ipoib_intf_free(ndev);
2471 free_netdev(ndev);
2472
2473 return ERR_PTR(result);
2474 }
2475
2476 if (hca->ops.rdma_netdev_get_params) {
2477 int rc = hca->ops.rdma_netdev_get_params(hca, port,
2478 RDMA_NETDEV_IPOIB,
2479 ¶ms);
2480
2481 if (!rc && ops->priv_size < params.sizeof_priv)
2482 ops->priv_size = params.sizeof_priv;
2483 }
2484
2485
2486
2487
2488
2489
2490 ndev->priv_destructor = ipoib_intf_free;
2491
2492 if (ipoib_intercept_dev_id_attr(ndev))
2493 goto sysfs_failed;
2494 if (ipoib_cm_add_mode_attr(ndev))
2495 goto sysfs_failed;
2496 if (ipoib_add_pkey_attr(ndev))
2497 goto sysfs_failed;
2498 if (ipoib_add_umcast_attr(ndev))
2499 goto sysfs_failed;
2500 if (device_create_file(&ndev->dev, &dev_attr_create_child))
2501 goto sysfs_failed;
2502 if (device_create_file(&ndev->dev, &dev_attr_delete_child))
2503 goto sysfs_failed;
2504
2505 return ndev;
2506
2507 sysfs_failed:
2508 ipoib_parent_unregister_pre(ndev);
2509 unregister_netdev(ndev);
2510 return ERR_PTR(-ENOMEM);
2511 }
2512
2513 static void ipoib_add_one(struct ib_device *device)
2514 {
2515 struct list_head *dev_list;
2516 struct net_device *dev;
2517 struct ipoib_dev_priv *priv;
2518 unsigned int p;
2519 int count = 0;
2520
2521 dev_list = kmalloc(sizeof(*dev_list), GFP_KERNEL);
2522 if (!dev_list)
2523 return;
2524
2525 INIT_LIST_HEAD(dev_list);
2526
2527 rdma_for_each_port (device, p) {
2528 if (!rdma_protocol_ib(device, p))
2529 continue;
2530 dev = ipoib_add_port("ib%d", device, p);
2531 if (!IS_ERR(dev)) {
2532 priv = ipoib_priv(dev);
2533 list_add_tail(&priv->list, dev_list);
2534 count++;
2535 }
2536 }
2537
2538 if (!count) {
2539 kfree(dev_list);
2540 return;
2541 }
2542
2543 ib_set_client_data(device, &ipoib_client, dev_list);
2544 }
2545
2546 static void ipoib_remove_one(struct ib_device *device, void *client_data)
2547 {
2548 struct ipoib_dev_priv *priv, *tmp, *cpriv, *tcpriv;
2549 struct list_head *dev_list = client_data;
2550
2551 if (!dev_list)
2552 return;
2553
2554 list_for_each_entry_safe(priv, tmp, dev_list, list) {
2555 LIST_HEAD(head);
2556 ipoib_parent_unregister_pre(priv->dev);
2557
2558 rtnl_lock();
2559
2560 list_for_each_entry_safe(cpriv, tcpriv, &priv->child_intfs,
2561 list)
2562 unregister_netdevice_queue(cpriv->dev, &head);
2563 unregister_netdevice_queue(priv->dev, &head);
2564 unregister_netdevice_many(&head);
2565
2566 rtnl_unlock();
2567 }
2568
2569 kfree(dev_list);
2570 }
2571
2572 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
2573 static struct notifier_block ipoib_netdev_notifier = {
2574 .notifier_call = ipoib_netdev_event,
2575 };
2576 #endif
2577
2578 static int __init ipoib_init_module(void)
2579 {
2580 int ret;
2581
2582 ipoib_recvq_size = roundup_pow_of_two(ipoib_recvq_size);
2583 ipoib_recvq_size = min(ipoib_recvq_size, IPOIB_MAX_QUEUE_SIZE);
2584 ipoib_recvq_size = max(ipoib_recvq_size, IPOIB_MIN_QUEUE_SIZE);
2585
2586 ipoib_sendq_size = roundup_pow_of_two(ipoib_sendq_size);
2587 ipoib_sendq_size = min(ipoib_sendq_size, IPOIB_MAX_QUEUE_SIZE);
2588 ipoib_sendq_size = max3(ipoib_sendq_size, 2 * MAX_SEND_CQE, IPOIB_MIN_QUEUE_SIZE);
2589 #ifdef CONFIG_INFINIBAND_IPOIB_CM
2590 ipoib_max_conn_qp = min(ipoib_max_conn_qp, IPOIB_CM_MAX_CONN_QP);
2591 ipoib_max_conn_qp = max(ipoib_max_conn_qp, 0);
2592 #endif
2593
2594
2595
2596
2597
2598 BUILD_BUG_ON(IPOIB_CM_COPYBREAK > IPOIB_CM_HEAD_SIZE);
2599
2600 ipoib_register_debugfs();
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612 ipoib_workqueue = alloc_ordered_workqueue("ipoib_flush", 0);
2613 if (!ipoib_workqueue) {
2614 ret = -ENOMEM;
2615 goto err_fs;
2616 }
2617
2618 ib_sa_register_client(&ipoib_sa_client);
2619
2620 ret = ib_register_client(&ipoib_client);
2621 if (ret)
2622 goto err_sa;
2623
2624 ret = ipoib_netlink_init();
2625 if (ret)
2626 goto err_client;
2627
2628 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
2629 register_netdevice_notifier(&ipoib_netdev_notifier);
2630 #endif
2631 return 0;
2632
2633 err_client:
2634 ib_unregister_client(&ipoib_client);
2635
2636 err_sa:
2637 ib_sa_unregister_client(&ipoib_sa_client);
2638 destroy_workqueue(ipoib_workqueue);
2639
2640 err_fs:
2641 ipoib_unregister_debugfs();
2642
2643 return ret;
2644 }
2645
2646 static void __exit ipoib_cleanup_module(void)
2647 {
2648 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
2649 unregister_netdevice_notifier(&ipoib_netdev_notifier);
2650 #endif
2651 ipoib_netlink_fini();
2652 ib_unregister_client(&ipoib_client);
2653 ib_sa_unregister_client(&ipoib_sa_client);
2654 ipoib_unregister_debugfs();
2655 destroy_workqueue(ipoib_workqueue);
2656 }
2657
2658 module_init(ipoib_init_module);
2659 module_exit(ipoib_cleanup_module);