projects / deliverable / linux.git / blob
? search:


0b7d129161e1d009b2738dbb4f79956793b041dd
[deliverable/linux.git] / drivers / infiniband / ulp / ipoib / ipoib_multicast.c
1 /*
2 * Copyright (c) 2004, 2005 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 <linux/skbuff.h>
36 #include <linux/rtnetlink.h>
37 #include <linux/ip.h>
38 #include <linux/in.h>
39 #include <linux/igmp.h>
40 #include <linux/inetdevice.h>
41 #include <linux/delay.h>
42 #include <linux/completion.h>
43
44 #include <net/dst.h>
45
46 #include "ipoib.h"
47
48 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
49 static int mcast_debug_level;
50
51 module_param(mcast_debug_level, int, 0644);
52 MODULE_PARM_DESC(mcast_debug_level,
53 "Enable multicast debug tracing if > 0");
54 #endif
55
56 static DEFINE_MUTEX(mcast_mutex);
57
58 struct ipoib_mcast_iter {
59 struct net_device *dev;
60 union ib_gid mgid;
61 unsigned long created;
62 unsigned int queuelen;
63 unsigned int complete;
64 unsigned int send_only;
65 };
66
67 static void ipoib_mcast_free(struct ipoib_mcast *mcast)
68 {
69 struct net_device *dev = mcast->dev;
70 struct ipoib_dev_priv *priv = netdev_priv(dev);
71 struct ipoib_neigh *neigh, *tmp;
72 unsigned long flags;
73 int tx_dropped = 0;
74
75 ipoib_dbg_mcast(netdev_priv(dev),
76 "deleting multicast group " IPOIB_GID_FMT "\n",
77 IPOIB_GID_ARG(mcast->mcmember.mgid));
78
79 spin_lock_irqsave(&priv->lock, flags);
80
81 list_for_each_entry_safe(neigh, tmp, &mcast->neigh_list, list) {
82 /*
83 * It's safe to call ipoib_put_ah() inside priv->lock
84 * here, because we know that mcast->ah will always
85 * hold one more reference, so ipoib_put_ah() will
86 * never do more than decrement the ref count.
87 */
88 if (neigh->ah)
89 ipoib_put_ah(neigh->ah);
90 ipoib_neigh_free(dev, neigh);
91 }
92
93 spin_unlock_irqrestore(&priv->lock, flags);
94
95 if (mcast->ah)
96 ipoib_put_ah(mcast->ah);
97
98 while (!skb_queue_empty(&mcast->pkt_queue)) {
99 ++tx_dropped;
100 dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
101 }
102
103 spin_lock_irqsave(&priv->tx_lock, flags);
104 dev->stats.tx_dropped += tx_dropped;
105 spin_unlock_irqrestore(&priv->tx_lock, flags);
106
107 kfree(mcast);
108 }
109
110 static struct ipoib_mcast *ipoib_mcast_alloc(struct net_device *dev,
111 int can_sleep)
112 {
113 struct ipoib_mcast *mcast;
114
115 mcast = kzalloc(sizeof *mcast, can_sleep ? GFP_KERNEL : GFP_ATOMIC);
116 if (!mcast)
117 return NULL;
118
119 mcast->dev = dev;
120 mcast->created = jiffies;
121 mcast->backoff = 1;
122
123 INIT_LIST_HEAD(&mcast->list);
124 INIT_LIST_HEAD(&mcast->neigh_list);
125 skb_queue_head_init(&mcast->pkt_queue);
126
127 return mcast;
128 }
129
130 static struct ipoib_mcast *__ipoib_mcast_find(struct net_device *dev, void *mgid)
131 {
132 struct ipoib_dev_priv *priv = netdev_priv(dev);
133 struct rb_node *n = priv->multicast_tree.rb_node;
134
135 while (n) {
136 struct ipoib_mcast *mcast;
137 int ret;
138
139 mcast = rb_entry(n, struct ipoib_mcast, rb_node);
140
141 ret = memcmp(mgid, mcast->mcmember.mgid.raw,
142 sizeof (union ib_gid));
143 if (ret < 0)
144 n = n->rb_left;
145 else if (ret > 0)
146 n = n->rb_right;
147 else
148 return mcast;
149 }
150
151 return NULL;
152 }
153
154 static int __ipoib_mcast_add(struct net_device *dev, struct ipoib_mcast *mcast)
155 {
156 struct ipoib_dev_priv *priv = netdev_priv(dev);
157 struct rb_node **n = &priv->multicast_tree.rb_node, *pn = NULL;
158
159 while (*n) {
160 struct ipoib_mcast *tmcast;
161 int ret;
162
163 pn = *n;
164 tmcast = rb_entry(pn, struct ipoib_mcast, rb_node);
165
166 ret = memcmp(mcast->mcmember.mgid.raw, tmcast->mcmember.mgid.raw,
167 sizeof (union ib_gid));
168 if (ret < 0)
169 n = &pn->rb_left;
170 else if (ret > 0)
171 n = &pn->rb_right;
172 else
173 return -EEXIST;
174 }
175
176 rb_link_node(&mcast->rb_node, pn, n);
177 rb_insert_color(&mcast->rb_node, &priv->multicast_tree);
178
179 return 0;
180 }
181
182 static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
183 struct ib_sa_mcmember_rec *mcmember)
184 {
185 struct net_device *dev = mcast->dev;
186 struct ipoib_dev_priv *priv = netdev_priv(dev);
187 struct ipoib_ah *ah;
188 int ret;
189
190 mcast->mcmember = *mcmember;
191
192 /* Set the cached Q_Key before we attach if it's the broadcast group */
193 if (!memcmp(mcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
194 sizeof (union ib_gid))) {
195 spin_lock_irq(&priv->lock);
196 if (!priv->broadcast) {
197 spin_unlock_irq(&priv->lock);
198 return -EAGAIN;
199 }
200 priv->qkey = be32_to_cpu(priv->broadcast->mcmember.qkey);
201 spin_unlock_irq(&priv->lock);
202 priv->tx_wr.wr.ud.remote_qkey = priv->qkey;
203 }
204
205 if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
206 if (test_and_set_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
207 ipoib_warn(priv, "multicast group " IPOIB_GID_FMT
208 " already attached\n",
209 IPOIB_GID_ARG(mcast->mcmember.mgid));
210
211 return 0;
212 }
213
214 ret = ipoib_mcast_attach(dev, be16_to_cpu(mcast->mcmember.mlid),
215 &mcast->mcmember.mgid);
216 if (ret < 0) {
217 ipoib_warn(priv, "couldn't attach QP to multicast group "
218 IPOIB_GID_FMT "\n",
219 IPOIB_GID_ARG(mcast->mcmember.mgid));
220
221 clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags);
222 return ret;
223 }
224 }
225
226 {
227 struct ib_ah_attr av = {
228 .dlid = be16_to_cpu(mcast->mcmember.mlid),
229 .port_num = priv->port,
230 .sl = mcast->mcmember.sl,
231 .ah_flags = IB_AH_GRH,
232 .static_rate = mcast->mcmember.rate,
233 .grh = {
234 .flow_label = be32_to_cpu(mcast->mcmember.flow_label),
235 .hop_limit = mcast->mcmember.hop_limit,
236 .sgid_index = 0,
237 .traffic_class = mcast->mcmember.traffic_class
238 }
239 };
240 av.grh.dgid = mcast->mcmember.mgid;
241
242 ah = ipoib_create_ah(dev, priv->pd, &av);
243 if (!ah) {
244 ipoib_warn(priv, "ib_address_create failed\n");
245 } else {
246 spin_lock_irq(&priv->lock);
247 mcast->ah = ah;
248 spin_unlock_irq(&priv->lock);
249
250 ipoib_dbg_mcast(priv, "MGID " IPOIB_GID_FMT
251 " AV %p, LID 0x%04x, SL %d\n",
252 IPOIB_GID_ARG(mcast->mcmember.mgid),
253 mcast->ah->ah,
254 be16_to_cpu(mcast->mcmember.mlid),
255 mcast->mcmember.sl);
256 }
257 }
258
259 /* actually send any queued packets */
260 spin_lock_irq(&priv->tx_lock);
261 while (!skb_queue_empty(&mcast->pkt_queue)) {
262 struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue);
263 spin_unlock_irq(&priv->tx_lock);
264
265 skb->dev = dev;
266
267 if (!skb->dst || !skb->dst->neighbour) {
268 /* put pseudoheader back on for next time */
269 skb_push(skb, sizeof (struct ipoib_pseudoheader));
270 }
271
272 if (dev_queue_xmit(skb))
273 ipoib_warn(priv, "dev_queue_xmit failed to requeue packet\n");
274 spin_lock_irq(&priv->tx_lock);
275 }
276 spin_unlock_irq(&priv->tx_lock);
277
278 return 0;
279 }
280
281 static int
282 ipoib_mcast_sendonly_join_complete(int status,
283 struct ib_sa_multicast *multicast)
284 {
285 struct ipoib_mcast *mcast = multicast->context;
286 struct net_device *dev = mcast->dev;
287 struct ipoib_dev_priv *priv = netdev_priv(dev);
288
289 /* We trap for port events ourselves. */
290 if (status == -ENETRESET)
291 return 0;
292
293 if (!status)
294 status = ipoib_mcast_join_finish(mcast, &multicast->rec);
295
296 if (status) {
297 if (mcast->logcount++ < 20)
298 ipoib_dbg_mcast(netdev_priv(dev), "multicast join failed for "
299 IPOIB_GID_FMT ", status %d\n",
300 IPOIB_GID_ARG(mcast->mcmember.mgid), status);
301
302 /* Flush out any queued packets */
303 spin_lock_irq(&priv->tx_lock);
304 while (!skb_queue_empty(&mcast->pkt_queue)) {
305 ++dev->stats.tx_dropped;
306 dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
307 }
308 spin_unlock_irq(&priv->tx_lock);
309
310 /* Clear the busy flag so we try again */
311 status = test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY,
312 &mcast->flags);
313 }
314 return status;
315 }
316
317 static int ipoib_mcast_sendonly_join(struct ipoib_mcast *mcast)
318 {
319 struct net_device *dev = mcast->dev;
320 struct ipoib_dev_priv *priv = netdev_priv(dev);
321 struct ib_sa_mcmember_rec rec = {
322 #if 0 /* Some SMs don't support send-only yet */
323 .join_state = 4
324 #else
325 .join_state = 1
326 #endif
327 };
328 int ret = 0;
329
330 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
331 ipoib_dbg_mcast(priv, "device shutting down, no multicast joins\n");
332 return -ENODEV;
333 }
334
335 if (test_and_set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) {
336 ipoib_dbg_mcast(priv, "multicast entry busy, skipping\n");
337 return -EBUSY;
338 }
339
340 rec.mgid = mcast->mcmember.mgid;
341 rec.port_gid = priv->local_gid;
342 rec.pkey = cpu_to_be16(priv->pkey);
343
344 mcast->mc = ib_sa_join_multicast(&ipoib_sa_client, priv->ca,
345 priv->port, &rec,
346 IB_SA_MCMEMBER_REC_MGID |
347 IB_SA_MCMEMBER_REC_PORT_GID |
348 IB_SA_MCMEMBER_REC_PKEY |
349 IB_SA_MCMEMBER_REC_JOIN_STATE,
350 GFP_ATOMIC,
351 ipoib_mcast_sendonly_join_complete,
352 mcast);
353 if (IS_ERR(mcast->mc)) {
354 ret = PTR_ERR(mcast->mc);
355 clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
356 ipoib_warn(priv, "ib_sa_join_multicast failed (ret = %d)\n",
357 ret);
358 } else {
359 ipoib_dbg_mcast(priv, "no multicast record for " IPOIB_GID_FMT
360 ", starting join\n",
361 IPOIB_GID_ARG(mcast->mcmember.mgid));
362 }
363
364 return ret;
365 }
366
367 static int ipoib_mcast_join_complete(int status,
368 struct ib_sa_multicast *multicast)
369 {
370 struct ipoib_mcast *mcast = multicast->context;
371 struct net_device *dev = mcast->dev;
372 struct ipoib_dev_priv *priv = netdev_priv(dev);
373
374 ipoib_dbg_mcast(priv, "join completion for " IPOIB_GID_FMT
375 " (status %d)\n",
376 IPOIB_GID_ARG(mcast->mcmember.mgid), status);
377
378 /* We trap for port events ourselves. */
379 if (status == -ENETRESET)
380 return 0;
381
382 if (!status)
383 status = ipoib_mcast_join_finish(mcast, &multicast->rec);
384
385 if (!status) {
386 mcast->backoff = 1;
387 mutex_lock(&mcast_mutex);
388 if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
389 queue_delayed_work(ipoib_workqueue,
390 &priv->mcast_task, 0);
391 mutex_unlock(&mcast_mutex);
392
393 if (mcast == priv->broadcast)
394 netif_carrier_on(dev);
395
396 return 0;
397 }
398
399 if (mcast->logcount++ < 20) {
400 if (status == -ETIMEDOUT) {
401 ipoib_dbg_mcast(priv, "multicast join failed for " IPOIB_GID_FMT
402 ", status %d\n",
403 IPOIB_GID_ARG(mcast->mcmember.mgid),
404 status);
405 } else {
406 ipoib_warn(priv, "multicast join failed for "
407 IPOIB_GID_FMT ", status %d\n",
408 IPOIB_GID_ARG(mcast->mcmember.mgid),
409 status);
410 }
411 }
412
413 mcast->backoff *= 2;
414 if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
415 mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
416
417 /* Clear the busy flag so we try again */
418 status = test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
419
420 mutex_lock(&mcast_mutex);
421 spin_lock_irq(&priv->lock);
422 if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
423 queue_delayed_work(ipoib_workqueue, &priv->mcast_task,
424 mcast->backoff * HZ);
425 spin_unlock_irq(&priv->lock);
426 mutex_unlock(&mcast_mutex);
427
428 return status;
429 }
430
431 static void ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast,
432 int create)
433 {
434 struct ipoib_dev_priv *priv = netdev_priv(dev);
435 struct ib_sa_mcmember_rec rec = {
436 .join_state = 1
437 };
438 ib_sa_comp_mask comp_mask;
439 int ret = 0;
440
441 ipoib_dbg_mcast(priv, "joining MGID " IPOIB_GID_FMT "\n",
442 IPOIB_GID_ARG(mcast->mcmember.mgid));
443
444 rec.mgid = mcast->mcmember.mgid;
445 rec.port_gid = priv->local_gid;
446 rec.pkey = cpu_to_be16(priv->pkey);
447
448 comp_mask =
449 IB_SA_MCMEMBER_REC_MGID |
450 IB_SA_MCMEMBER_REC_PORT_GID |
451 IB_SA_MCMEMBER_REC_PKEY |
452 IB_SA_MCMEMBER_REC_JOIN_STATE;
453
454 if (create) {
455 comp_mask |=
456 IB_SA_MCMEMBER_REC_QKEY |
457 IB_SA_MCMEMBER_REC_MTU_SELECTOR |
458 IB_SA_MCMEMBER_REC_MTU |
459 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS |
460 IB_SA_MCMEMBER_REC_RATE_SELECTOR |
461 IB_SA_MCMEMBER_REC_RATE |
462 IB_SA_MCMEMBER_REC_SL |
463 IB_SA_MCMEMBER_REC_FLOW_LABEL |
464 IB_SA_MCMEMBER_REC_HOP_LIMIT;
465
466 rec.qkey = priv->broadcast->mcmember.qkey;
467 rec.mtu_selector = IB_SA_EQ;
468 rec.mtu = priv->broadcast->mcmember.mtu;
469 rec.traffic_class = priv->broadcast->mcmember.traffic_class;
470 rec.rate_selector = IB_SA_EQ;
471 rec.rate = priv->broadcast->mcmember.rate;
472 rec.sl = priv->broadcast->mcmember.sl;
473 rec.flow_label = priv->broadcast->mcmember.flow_label;
474 rec.hop_limit = priv->broadcast->mcmember.hop_limit;
475 }
476
477 set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
478 mcast->mc = ib_sa_join_multicast(&ipoib_sa_client, priv->ca, priv->port,
479 &rec, comp_mask, GFP_KERNEL,
480 ipoib_mcast_join_complete, mcast);
481 if (IS_ERR(mcast->mc)) {
482 clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
483 ret = PTR_ERR(mcast->mc);
484 ipoib_warn(priv, "ib_sa_join_multicast failed, status %d\n", ret);
485
486 mcast->backoff *= 2;
487 if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
488 mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
489
490 mutex_lock(&mcast_mutex);
491 if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
492 queue_delayed_work(ipoib_workqueue,
493 &priv->mcast_task,
494 mcast->backoff * HZ);
495 mutex_unlock(&mcast_mutex);
496 }
497 }
498
499 void ipoib_mcast_join_task(struct work_struct *work)
500 {
501 struct ipoib_dev_priv *priv =
502 container_of(work, struct ipoib_dev_priv, mcast_task.work);
503 struct net_device *dev = priv->dev;
504
505 if (!test_bit(IPOIB_MCAST_RUN, &priv->flags))
506 return;
507
508 if (ib_query_gid(priv->ca, priv->port, 0, &priv->local_gid))
509 ipoib_warn(priv, "ib_query_gid() failed\n");
510 else
511 memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw, sizeof (union ib_gid));
512
513 {
514 struct ib_port_attr attr;
515
516 if (!ib_query_port(priv->ca, priv->port, &attr))
517 priv->local_lid = attr.lid;
518 else
519 ipoib_warn(priv, "ib_query_port failed\n");
520 }
521
522 if (!priv->broadcast) {
523 struct ipoib_mcast *broadcast;
524
525 broadcast = ipoib_mcast_alloc(dev, 1);
526 if (!broadcast) {
527 ipoib_warn(priv, "failed to allocate broadcast group\n");
528 mutex_lock(&mcast_mutex);
529 if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
530 queue_delayed_work(ipoib_workqueue,
531 &priv->mcast_task, HZ);
532 mutex_unlock(&mcast_mutex);
533 return;
534 }
535
536 spin_lock_irq(&priv->lock);
537 memcpy(broadcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
538 sizeof (union ib_gid));
539 priv->broadcast = broadcast;
540
541 __ipoib_mcast_add(dev, priv->broadcast);
542 spin_unlock_irq(&priv->lock);
543 }
544
545 if (!test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
546 if (!test_bit(IPOIB_MCAST_FLAG_BUSY, &priv->broadcast->flags))
547 ipoib_mcast_join(dev, priv->broadcast, 0);
548 return;
549 }
550
551 while (1) {
552 struct ipoib_mcast *mcast = NULL;
553
554 spin_lock_irq(&priv->lock);
555 list_for_each_entry(mcast, &priv->multicast_list, list) {
556 if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)
557 && !test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)
558 && !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
559 /* Found the next unjoined group */
560 break;
561 }
562 }
563 spin_unlock_irq(&priv->lock);
564
565 if (&mcast->list == &priv->multicast_list) {
566 /* All done */
567 break;
568 }
569
570 ipoib_mcast_join(dev, mcast, 1);
571 return;
572 }
573
574 priv->mcast_mtu = IPOIB_UD_MTU(ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu));
575
576 if (!ipoib_cm_admin_enabled(dev))
577 dev->mtu = min(priv->mcast_mtu, priv->admin_mtu);
578
579 ipoib_dbg_mcast(priv, "successfully joined all multicast groups\n");
580
581 clear_bit(IPOIB_MCAST_RUN, &priv->flags);
582 }
583
584 int ipoib_mcast_start_thread(struct net_device *dev)
585 {
586 struct ipoib_dev_priv *priv = netdev_priv(dev);
587
588 ipoib_dbg_mcast(priv, "starting multicast thread\n");
589
590 mutex_lock(&mcast_mutex);
591 if (!test_and_set_bit(IPOIB_MCAST_RUN, &priv->flags))
592 queue_delayed_work(ipoib_workqueue, &priv->mcast_task, 0);
593 mutex_unlock(&mcast_mutex);
594
595 return 0;
596 }
597
598 int ipoib_mcast_stop_thread(struct net_device *dev, int flush)
599 {
600 struct ipoib_dev_priv *priv = netdev_priv(dev);
601
602 ipoib_dbg_mcast(priv, "stopping multicast thread\n");
603
604 mutex_lock(&mcast_mutex);
605 clear_bit(IPOIB_MCAST_RUN, &priv->flags);
606 cancel_delayed_work(&priv->mcast_task);
607 mutex_unlock(&mcast_mutex);
608
609 if (flush)
610 flush_workqueue(ipoib_workqueue);
611
612 return 0;
613 }
614
615 static int ipoib_mcast_leave(struct net_device *dev, struct ipoib_mcast *mcast)
616 {
617 struct ipoib_dev_priv *priv = netdev_priv(dev);
618 int ret = 0;
619
620 if (test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
621 ib_sa_free_multicast(mcast->mc);
622
623 if (test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
624 ipoib_dbg_mcast(priv, "leaving MGID " IPOIB_GID_FMT "\n",
625 IPOIB_GID_ARG(mcast->mcmember.mgid));
626
627 /* Remove ourselves from the multicast group */
628 ret = ipoib_mcast_detach(dev, be16_to_cpu(mcast->mcmember.mlid),
629 &mcast->mcmember.mgid);
630 if (ret)
631 ipoib_warn(priv, "ipoib_mcast_detach failed (result = %d)\n", ret);
632 }
633
634 return 0;
635 }
636
637 void ipoib_mcast_send(struct net_device *dev, void *mgid, struct sk_buff *skb)
638 {
639 struct ipoib_dev_priv *priv = netdev_priv(dev);
640 struct ipoib_mcast *mcast;
641
642 /*
643 * We can only be called from ipoib_start_xmit, so we're
644 * inside tx_lock -- no need to save/restore flags.
645 */
646 spin_lock(&priv->lock);
647
648 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags) ||
649 !priv->broadcast ||
650 !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
651 ++dev->stats.tx_dropped;
652 dev_kfree_skb_any(skb);
653 goto unlock;
654 }
655
656 mcast = __ipoib_mcast_find(dev, mgid);
657 if (!mcast) {
658 /* Let's create a new send only group now */
659 ipoib_dbg_mcast(priv, "setting up send only multicast group for "
660 IPOIB_GID_FMT "\n", IPOIB_GID_RAW_ARG(mgid));
661
662 mcast = ipoib_mcast_alloc(dev, 0);
663 if (!mcast) {
664 ipoib_warn(priv, "unable to allocate memory for "
665 "multicast structure\n");
666 ++dev->stats.tx_dropped;
667 dev_kfree_skb_any(skb);
668 goto out;
669 }
670
671 set_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags);
672 memcpy(mcast->mcmember.mgid.raw, mgid, sizeof (union ib_gid));
673 __ipoib_mcast_add(dev, mcast);
674 list_add_tail(&mcast->list, &priv->multicast_list);
675 }
676
677 if (!mcast->ah) {
678 if (skb_queue_len(&mcast->pkt_queue) < IPOIB_MAX_MCAST_QUEUE)
679 skb_queue_tail(&mcast->pkt_queue, skb);
680 else {
681 ++dev->stats.tx_dropped;
682 dev_kfree_skb_any(skb);
683 }
684
685 if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
686 ipoib_dbg_mcast(priv, "no address vector, "
687 "but multicast join already started\n");
688 else if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags))
689 ipoib_mcast_sendonly_join(mcast);
690
691 /*
692 * If lookup completes between here and out:, don't
693 * want to send packet twice.
694 */
695 mcast = NULL;
696 }
697
698 out:
699 if (mcast && mcast->ah) {
700 if (skb->dst &&
701 skb->dst->neighbour &&
702 !*to_ipoib_neigh(skb->dst->neighbour)) {
703 struct ipoib_neigh *neigh = ipoib_neigh_alloc(skb->dst->neighbour,
704 skb->dev);
705
706 if (neigh) {
707 kref_get(&mcast->ah->ref);
708 neigh->ah = mcast->ah;
709 list_add_tail(&neigh->list, &mcast->neigh_list);
710 }
711 }
712
713 ipoib_send(dev, skb, mcast->ah, IB_MULTICAST_QPN);
714 }
715
716 unlock:
717 spin_unlock(&priv->lock);
718 }
719
720 void ipoib_mcast_dev_flush(struct net_device *dev)
721 {
722 struct ipoib_dev_priv *priv = netdev_priv(dev);
723 LIST_HEAD(remove_list);
724 struct ipoib_mcast *mcast, *tmcast;
725 unsigned long flags;
726
727 ipoib_dbg_mcast(priv, "flushing multicast list\n");
728
729 spin_lock_irqsave(&priv->lock, flags);
730
731 list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
732 list_del(&mcast->list);
733 rb_erase(&mcast->rb_node, &priv->multicast_tree);
734 list_add_tail(&mcast->list, &remove_list);
735 }
736
737 if (priv->broadcast) {
738 rb_erase(&priv->broadcast->rb_node, &priv->multicast_tree);
739 list_add_tail(&priv->broadcast->list, &remove_list);
740 priv->broadcast = NULL;
741 }
742
743 spin_unlock_irqrestore(&priv->lock, flags);
744
745 list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
746 ipoib_mcast_leave(dev, mcast);
747 ipoib_mcast_free(mcast);
748 }
749 }
750
751 void ipoib_mcast_restart_task(struct work_struct *work)
752 {
753 struct ipoib_dev_priv *priv =
754 container_of(work, struct ipoib_dev_priv, restart_task);
755 struct net_device *dev = priv->dev;
756 struct dev_mc_list *mclist;
757 struct ipoib_mcast *mcast, *tmcast;
758 LIST_HEAD(remove_list);
759 unsigned long flags;
760 struct ib_sa_mcmember_rec rec;
761
762 ipoib_dbg_mcast(priv, "restarting multicast task\n");
763
764 ipoib_mcast_stop_thread(dev, 0);
765
766 local_irq_save(flags);
767 netif_tx_lock(dev);
768 spin_lock(&priv->lock);
769
770 /*
771 * Unfortunately, the networking core only gives us a list of all of
772 * the multicast hardware addresses. We need to figure out which ones
773 * are new and which ones have been removed
774 */
775
776 /* Clear out the found flag */
777 list_for_each_entry(mcast, &priv->multicast_list, list)
778 clear_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
779
780 /* Mark all of the entries that are found or don't exist */
781 for (mclist = dev->mc_list; mclist; mclist = mclist->next) {
782 union ib_gid mgid;
783
784 memcpy(mgid.raw, mclist->dmi_addr + 4, sizeof mgid);
785
786 mcast = __ipoib_mcast_find(dev, &mgid);
787 if (!mcast || test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
788 struct ipoib_mcast *nmcast;
789
790 /* ignore group which is directly joined by userspace */
791 if (test_bit(IPOIB_FLAG_UMCAST, &priv->flags) &&
792 !ib_sa_get_mcmember_rec(priv->ca, priv->port, &mgid, &rec)) {
793 ipoib_dbg_mcast(priv, "ignoring multicast entry for mgid "
794 IPOIB_GID_FMT "\n", IPOIB_GID_ARG(mgid));
795 continue;
796 }
797
798 /* Not found or send-only group, let's add a new entry */
799 ipoib_dbg_mcast(priv, "adding multicast entry for mgid "
800 IPOIB_GID_FMT "\n", IPOIB_GID_ARG(mgid));
801
802 nmcast = ipoib_mcast_alloc(dev, 0);
803 if (!nmcast) {
804 ipoib_warn(priv, "unable to allocate memory for multicast structure\n");
805 continue;
806 }
807
808 set_bit(IPOIB_MCAST_FLAG_FOUND, &nmcast->flags);
809
810 nmcast->mcmember.mgid = mgid;
811
812 if (mcast) {
813 /* Destroy the send only entry */
814 list_move_tail(&mcast->list, &remove_list);
815
816 rb_replace_node(&mcast->rb_node,
817 &nmcast->rb_node,
818 &priv->multicast_tree);
819 } else
820 __ipoib_mcast_add(dev, nmcast);
821
822 list_add_tail(&nmcast->list, &priv->multicast_list);
823 }
824
825 if (mcast)
826 set_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
827 }
828
829 /* Remove all of the entries don't exist anymore */
830 list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
831 if (!test_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags) &&
832 !test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
833 ipoib_dbg_mcast(priv, "deleting multicast group " IPOIB_GID_FMT "\n",
834 IPOIB_GID_ARG(mcast->mcmember.mgid));
835
836 rb_erase(&mcast->rb_node, &priv->multicast_tree);
837
838 /* Move to the remove list */
839 list_move_tail(&mcast->list, &remove_list);
840 }
841 }
842
843 spin_unlock(&priv->lock);
844 netif_tx_unlock(dev);
845 local_irq_restore(flags);
846
847 /* We have to cancel outside of the spinlock */
848 list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
849 ipoib_mcast_leave(mcast->dev, mcast);
850 ipoib_mcast_free(mcast);
851 }
852
853 if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
854 ipoib_mcast_start_thread(dev);
855 }
856
857 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
858
859 struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev)
860 {
861 struct ipoib_mcast_iter *iter;
862
863 iter = kmalloc(sizeof *iter, GFP_KERNEL);
864 if (!iter)
865 return NULL;
866
867 iter->dev = dev;
868 memset(iter->mgid.raw, 0, 16);
869
870 if (ipoib_mcast_iter_next(iter)) {
871 kfree(iter);
872 return NULL;
873 }
874
875 return iter;
876 }
877
878 int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter)
879 {
880 struct ipoib_dev_priv *priv = netdev_priv(iter->dev);
881 struct rb_node *n;
882 struct ipoib_mcast *mcast;
883 int ret = 1;
884
885 spin_lock_irq(&priv->lock);
886
887 n = rb_first(&priv->multicast_tree);
888
889 while (n) {
890 mcast = rb_entry(n, struct ipoib_mcast, rb_node);
891
892 if (memcmp(iter->mgid.raw, mcast->mcmember.mgid.raw,
893 sizeof (union ib_gid)) < 0) {
894 iter->mgid = mcast->mcmember.mgid;
895 iter->created = mcast->created;
896 iter->queuelen = skb_queue_len(&mcast->pkt_queue);
897 iter->complete = !!mcast->ah;
898 iter->send_only = !!(mcast->flags & (1 << IPOIB_MCAST_FLAG_SENDONLY));
899
900 ret = 0;
901
902 break;
903 }
904
905 n = rb_next(n);
906 }
907
908 spin_unlock_irq(&priv->lock);
909
910 return ret;
911 }
912
913 void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
914 union ib_gid *mgid,
915 unsigned long *created,
916 unsigned int *queuelen,
917 unsigned int *complete,
918 unsigned int *send_only)
919 {
920 *mgid = iter->mgid;
921 *created = iter->created;
922 *queuelen = iter->queuelen;
923 *complete = iter->complete;
924 *send_only = iter->send_only;
925 }
926
927 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
Linux kernel - Deliverables
This page took 0.049975 seconds and 4 git commands to generate.