projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
IB/mad: Handle DR SMPs with a LID routed part
[deliverable/linux.git] / drivers / infiniband / ulp / ipoib / ipoib_multicast.c
CommitLineData
1da177e4
LT		 1/*
2 * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved.
2a1d9b7f
RD		 3 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4 * Copyright (c) 2004 Voltaire, Inc. All rights reserved.
1da177e4
LT		 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 * $Id: ipoib_multicast.c 1362 2004-12-18 15:56:29Z roland $
35 */
36
37#include <linux/skbuff.h>
38#include <linux/rtnetlink.h>
39#include <linux/ip.h>
40#include <linux/in.h>
41#include <linux/igmp.h>
42#include <linux/inetdevice.h>
43#include <linux/delay.h>
44#include <linux/completion.h>
45
14c85021
ACM		 46#include <net/dst.h>
47
1da177e4
LT		 48#include "ipoib.h"
49
50#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
51static int mcast_debug_level;
52
53module_param(mcast_debug_level, int, 0644);
54MODULE_PARM_DESC(mcast_debug_level,
55 "Enable multicast debug tracing if > 0");
56#endif
57
95ed644f 58static DEFINE_MUTEX(mcast_mutex);
1da177e4
LT		 59
60/* Used for all multicast joins (broadcast, IPv4 mcast and IPv6 mcast) */
61struct ipoib_mcast {
62 struct ib_sa_mcmember_rec mcmember;
63 struct ipoib_ah *ah;
64
65 struct rb_node rb_node;
66 struct list_head list;
67 struct completion done;
68
69 int query_id;
70 struct ib_sa_query *query;
71
72 unsigned long created;
73 unsigned long backoff;
74
75 unsigned long flags;
76 unsigned char logcount;
77
78 struct list_head neigh_list;
79
80 struct sk_buff_head pkt_queue;
81
82 struct net_device *dev;
83};
84
85struct ipoib_mcast_iter {
86 struct net_device *dev;
87 union ib_gid mgid;
88 unsigned long created;
89 unsigned int queuelen;
90 unsigned int complete;
91 unsigned int send_only;
92};
93
94static void ipoib_mcast_free(struct ipoib_mcast *mcast)
95{
96 struct net_device *dev = mcast->dev;
97 struct ipoib_dev_priv *priv = netdev_priv(dev);
98 struct ipoib_neigh *neigh, *tmp;
99 unsigned long flags;
b36f170b 100 int tx_dropped = 0;
1da177e4
LT		 101
102 ipoib_dbg_mcast(netdev_priv(dev),
103 "deleting multicast group " IPOIB_GID_FMT "\n",
104 IPOIB_GID_ARG(mcast->mcmember.mgid));
105
106 spin_lock_irqsave(&priv->lock, flags);
107
108 list_for_each_entry_safe(neigh, tmp, &mcast->neigh_list, list) {
97460df3
EC		 109 /*
110 * It's safe to call ipoib_put_ah() inside priv->lock
111 * here, because we know that mcast->ah will always
112 * hold one more reference, so ipoib_put_ah() will
113 * never do more than decrement the ref count.
114 */
1da177e4 115 if (neigh->ah)
97460df3 116 ipoib_put_ah(neigh->ah);
1da177e4
LT		 117 *to_ipoib_neigh(neigh->neighbour) = NULL;
118 neigh->neighbour->ops->destructor = NULL;
119 kfree(neigh);
120 }
121
122 spin_unlock_irqrestore(&priv->lock, flags);
123
1da177e4
LT		 124 if (mcast->ah)
125 ipoib_put_ah(mcast->ah);
126
b36f170b
MT		 127 while (!skb_queue_empty(&mcast->pkt_queue)) {
128 ++tx_dropped;
8c608a32 129 dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
b36f170b
MT		 130 }
131
132 spin_lock_irqsave(&priv->tx_lock, flags);
133 priv->stats.tx_dropped += tx_dropped;
134 spin_unlock_irqrestore(&priv->tx_lock, flags);
1da177e4
LT		 135
136 kfree(mcast);
137}
138
139static struct ipoib_mcast *ipoib_mcast_alloc(struct net_device *dev,
140 int can_sleep)
141{
142 struct ipoib_mcast *mcast;
143
de6eb66b 144 mcast = kzalloc(sizeof *mcast, can_sleep ? GFP_KERNEL : GFP_ATOMIC);
1da177e4
LT		 145 if (!mcast)
146 return NULL;
147
1da177e4
LT		 148 mcast->dev = dev;
149 mcast->created = jiffies;
ce5b65cc 150 mcast->backoff = 1;
1da177e4
LT		 151
152 INIT_LIST_HEAD(&mcast->list);
153 INIT_LIST_HEAD(&mcast->neigh_list);
154 skb_queue_head_init(&mcast->pkt_queue);
155
1da177e4
LT		 156 return mcast;
157}
158
159static struct ipoib_mcast *__ipoib_mcast_find(struct net_device *dev, union ib_gid *mgid)
160{
161 struct ipoib_dev_priv *priv = netdev_priv(dev);
162 struct rb_node *n = priv->multicast_tree.rb_node;
163
164 while (n) {
165 struct ipoib_mcast *mcast;
166 int ret;
167
168 mcast = rb_entry(n, struct ipoib_mcast, rb_node);
169
170 ret = memcmp(mgid->raw, mcast->mcmember.mgid.raw,
171 sizeof (union ib_gid));
172 if (ret < 0)
173 n = n->rb_left;
174 else if (ret > 0)
175 n = n->rb_right;
176 else
177 return mcast;
178 }
179
180 return NULL;
181}
182
183static int __ipoib_mcast_add(struct net_device *dev, struct ipoib_mcast *mcast)
184{
185 struct ipoib_dev_priv *priv = netdev_priv(dev);
186 struct rb_node **n = &priv->multicast_tree.rb_node, *pn = NULL;
187
188 while (*n) {
189 struct ipoib_mcast *tmcast;
190 int ret;
191
192 pn = *n;
193 tmcast = rb_entry(pn, struct ipoib_mcast, rb_node);
194
195 ret = memcmp(mcast->mcmember.mgid.raw, tmcast->mcmember.mgid.raw,
196 sizeof (union ib_gid));
197 if (ret < 0)
198 n = &pn->rb_left;
199 else if (ret > 0)
200 n = &pn->rb_right;
201 else
202 return -EEXIST;
203 }
204
205 rb_link_node(&mcast->rb_node, pn, n);
206 rb_insert_color(&mcast->rb_node, &priv->multicast_tree);
207
208 return 0;
209}
210
211static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
212 struct ib_sa_mcmember_rec *mcmember)
213{
214 struct net_device *dev = mcast->dev;
215 struct ipoib_dev_priv *priv = netdev_priv(dev);
216 int ret;
217
218 mcast->mcmember = *mcmember;
219
220 /* Set the cached Q_Key before we attach if it's the broadcast group */
221 if (!memcmp(mcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
222 sizeof (union ib_gid))) {
223 priv->qkey = be32_to_cpu(priv->broadcast->mcmember.qkey);
224 priv->tx_wr.wr.ud.remote_qkey = priv->qkey;
225 }
226
227 if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
228 if (test_and_set_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
229 ipoib_warn(priv, "multicast group " IPOIB_GID_FMT
230 " already attached\n",
231 IPOIB_GID_ARG(mcast->mcmember.mgid));
232
233 return 0;
234 }
235
236 ret = ipoib_mcast_attach(dev, be16_to_cpu(mcast->mcmember.mlid),
237 &mcast->mcmember.mgid);
238 if (ret < 0) {
239 ipoib_warn(priv, "couldn't attach QP to multicast group "
240 IPOIB_GID_FMT "\n",
241 IPOIB_GID_ARG(mcast->mcmember.mgid));
242
243 clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags);
244 return ret;
245 }
246 }
247
248 {
249 struct ib_ah_attr av = {
250 .dlid = be16_to_cpu(mcast->mcmember.mlid),
251 .port_num = priv->port,
252 .sl = mcast->mcmember.sl,
253 .ah_flags = IB_AH_GRH,
254 .grh = {
255 .flow_label = be32_to_cpu(mcast->mcmember.flow_label),
256 .hop_limit = mcast->mcmember.hop_limit,
257 .sgid_index = 0,
258 .traffic_class = mcast->mcmember.traffic_class
259 }
260 };
e6ded99c 261 int path_rate = ib_sa_rate_enum_to_int(mcast->mcmember.rate);
1da177e4
LT		 262
263 av.grh.dgid = mcast->mcmember.mgid;
264
e6ded99c
RD		 265 if (path_rate > 0 && priv->local_rate > path_rate)
266 av.static_rate = (priv->local_rate - 1) / path_rate;
1da177e4
LT		 267
268 ipoib_dbg_mcast(priv, "static_rate %d for local port %dX, mcmember %dX\n",
269 av.static_rate, priv->local_rate,
270 ib_sa_rate_enum_to_int(mcast->mcmember.rate));
271
272 mcast->ah = ipoib_create_ah(dev, priv->pd, &av);
273 if (!mcast->ah) {
274 ipoib_warn(priv, "ib_address_create failed\n");
275 } else {
276 ipoib_dbg_mcast(priv, "MGID " IPOIB_GID_FMT
277 " AV %p, LID 0x%04x, SL %d\n",
278 IPOIB_GID_ARG(mcast->mcmember.mgid),
279 mcast->ah->ah,
280 be16_to_cpu(mcast->mcmember.mlid),
281 mcast->mcmember.sl);
282 }
283 }
284
285 /* actually send any queued packets */
b36f170b 286 spin_lock_irq(&priv->tx_lock);
1da177e4
LT		 287 while (!skb_queue_empty(&mcast->pkt_queue)) {
288 struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue);
b36f170b 289 spin_unlock_irq(&priv->tx_lock);
1da177e4
LT		 290
291 skb->dev = dev;
292
293 if (!skb->dst || !skb->dst->neighbour) {
294 /* put pseudoheader back on for next time */
295 skb_push(skb, sizeof (struct ipoib_pseudoheader));
296 }
297
298 if (dev_queue_xmit(skb))
299 ipoib_warn(priv, "dev_queue_xmit failed to requeue packet\n");
b36f170b 300 spin_lock_irq(&priv->tx_lock);
1da177e4 301 }
b36f170b 302 spin_unlock_irq(&priv->tx_lock);
1da177e4
LT		 303
304 return 0;
305}
306
307static void
308ipoib_mcast_sendonly_join_complete(int status,
309 struct ib_sa_mcmember_rec *mcmember,
310 void *mcast_ptr)
311{
312 struct ipoib_mcast *mcast = mcast_ptr;
313 struct net_device *dev = mcast->dev;
b36f170b 314 struct ipoib_dev_priv *priv = netdev_priv(dev);
1da177e4
LT		 315
316 if (!status)
317 ipoib_mcast_join_finish(mcast, mcmember);
318 else {
319 if (mcast->logcount++ < 20)
320 ipoib_dbg_mcast(netdev_priv(dev), "multicast join failed for "
321 IPOIB_GID_FMT ", status %d\n",
322 IPOIB_GID_ARG(mcast->mcmember.mgid), status);
323
324 /* Flush out any queued packets */
b36f170b
MT		 325 spin_lock_irq(&priv->tx_lock);
326 while (!skb_queue_empty(&mcast->pkt_queue)) {
327 ++priv->stats.tx_dropped;
8c608a32 328 dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
b36f170b
MT		 329 }
330 spin_unlock_irq(&priv->tx_lock);
1da177e4
LT		 331
332 /* Clear the busy flag so we try again */
333 clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
334 }
335
336 complete(&mcast->done);
337}
338
339static int ipoib_mcast_sendonly_join(struct ipoib_mcast *mcast)
340{
341 struct net_device *dev = mcast->dev;
342 struct ipoib_dev_priv *priv = netdev_priv(dev);
343 struct ib_sa_mcmember_rec rec = {
344#if 0 /* Some SMs don't support send-only yet */
345 .join_state = 4
346#else
347 .join_state = 1
348#endif
349 };
350 int ret = 0;
351
352 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
353 ipoib_dbg_mcast(priv, "device shutting down, no multicast joins\n");
354 return -ENODEV;
355 }
356
357 if (test_and_set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) {
358 ipoib_dbg_mcast(priv, "multicast entry busy, skipping\n");
359 return -EBUSY;
360 }
361
362 rec.mgid = mcast->mcmember.mgid;
363 rec.port_gid = priv->local_gid;
97f52eb4 364 rec.pkey = cpu_to_be16(priv->pkey);
1da177e4 365
de922487
MT		 366 init_completion(&mcast->done);
367
1da177e4
LT		 368 ret = ib_sa_mcmember_rec_set(priv->ca, priv->port, &rec,
369 IB_SA_MCMEMBER_REC_MGID |
370 IB_SA_MCMEMBER_REC_PORT_GID |
371 IB_SA_MCMEMBER_REC_PKEY |
372 IB_SA_MCMEMBER_REC_JOIN_STATE,
373 1000, GFP_ATOMIC,
374 ipoib_mcast_sendonly_join_complete,
375 mcast, &mcast->query);
376 if (ret < 0) {
377 ipoib_warn(priv, "ib_sa_mcmember_rec_set failed (ret = %d)\n",
378 ret);
379 } else {
380 ipoib_dbg_mcast(priv, "no multicast record for " IPOIB_GID_FMT
381 ", starting join\n",
382 IPOIB_GID_ARG(mcast->mcmember.mgid));
383
384 mcast->query_id = ret;
385 }
386
387 return ret;
388}
389
390static void ipoib_mcast_join_complete(int status,
391 struct ib_sa_mcmember_rec *mcmember,
392 void *mcast_ptr)
393{
394 struct ipoib_mcast *mcast = mcast_ptr;
395 struct net_device *dev = mcast->dev;
396 struct ipoib_dev_priv *priv = netdev_priv(dev);
397
398 ipoib_dbg_mcast(priv, "join completion for " IPOIB_GID_FMT
399 " (status %d)\n",
400 IPOIB_GID_ARG(mcast->mcmember.mgid), status);
401
402 if (!status && !ipoib_mcast_join_finish(mcast, mcmember)) {
ce5b65cc 403 mcast->backoff = 1;
95ed644f 404 mutex_lock(&mcast_mutex);
1da177e4
LT		 405 if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
406 queue_work(ipoib_workqueue, &priv->mcast_task);
95ed644f 407 mutex_unlock(&mcast_mutex);
1da177e4
LT		 408 complete(&mcast->done);
409 return;
410 }
411
412 if (status == -EINTR) {
413 complete(&mcast->done);
414 return;
415 }
416
417 if (status && mcast->logcount++ < 20) {
418 if (status == -ETIMEDOUT || status == -EINTR) {
419 ipoib_dbg_mcast(priv, "multicast join failed for " IPOIB_GID_FMT
420 ", status %d\n",
421 IPOIB_GID_ARG(mcast->mcmember.mgid),
422 status);
423 } else {
424 ipoib_warn(priv, "multicast join failed for "
425 IPOIB_GID_FMT ", status %d\n",
426 IPOIB_GID_ARG(mcast->mcmember.mgid),
427 status);
428 }
429 }
430
431 mcast->backoff *= 2;
432 if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
433 mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
434
435 mcast->query = NULL;
436
95ed644f 437 mutex_lock(&mcast_mutex);
1da177e4
LT		 438 if (test_bit(IPOIB_MCAST_RUN, &priv->flags)) {
439 if (status == -ETIMEDOUT)
440 queue_work(ipoib_workqueue, &priv->mcast_task);
441 else
442 queue_delayed_work(ipoib_workqueue, &priv->mcast_task,
443 mcast->backoff * HZ);
444 } else
445 complete(&mcast->done);
95ed644f 446 mutex_unlock(&mcast_mutex);
1da177e4
LT		 447
448 return;
449}
450
451static void ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast,
452 int create)
453{
454 struct ipoib_dev_priv *priv = netdev_priv(dev);
455 struct ib_sa_mcmember_rec rec = {
456 .join_state = 1
457 };
458 ib_sa_comp_mask comp_mask;
459 int ret = 0;
460
461 ipoib_dbg_mcast(priv, "joining MGID " IPOIB_GID_FMT "\n",
462 IPOIB_GID_ARG(mcast->mcmember.mgid));
463
464 rec.mgid = mcast->mcmember.mgid;
465 rec.port_gid = priv->local_gid;
97f52eb4 466 rec.pkey = cpu_to_be16(priv->pkey);
1da177e4
LT		 467
468 comp_mask =
469 IB_SA_MCMEMBER_REC_MGID |
470 IB_SA_MCMEMBER_REC_PORT_GID |
471 IB_SA_MCMEMBER_REC_PKEY |
472 IB_SA_MCMEMBER_REC_JOIN_STATE;
473
474 if (create) {
475 comp_mask |=
476 IB_SA_MCMEMBER_REC_QKEY |
477 IB_SA_MCMEMBER_REC_SL |
478 IB_SA_MCMEMBER_REC_FLOW_LABEL |
479 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
480
481 rec.qkey = priv->broadcast->mcmember.qkey;
482 rec.sl = priv->broadcast->mcmember.sl;
483 rec.flow_label = priv->broadcast->mcmember.flow_label;
484 rec.traffic_class = priv->broadcast->mcmember.traffic_class;
485 }
486
de922487
MT		 487 init_completion(&mcast->done);
488
1da177e4
LT		 489 ret = ib_sa_mcmember_rec_set(priv->ca, priv->port, &rec, comp_mask,
490 mcast->backoff * 1000, GFP_ATOMIC,
491 ipoib_mcast_join_complete,
492 mcast, &mcast->query);
493
494 if (ret < 0) {
495 ipoib_warn(priv, "ib_sa_mcmember_rec_set failed, status %d\n", ret);
496
497 mcast->backoff *= 2;
498 if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
499 mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
500
95ed644f 501 mutex_lock(&mcast_mutex);
1da177e4
LT		 502 if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
503 queue_delayed_work(ipoib_workqueue,
504 &priv->mcast_task,
ce5b65cc 505 mcast->backoff * HZ);
95ed644f 506 mutex_unlock(&mcast_mutex);
1da177e4
LT		 507 } else
508 mcast->query_id = ret;
509}
510
511void ipoib_mcast_join_task(void *dev_ptr)
512{
513 struct net_device *dev = dev_ptr;
514 struct ipoib_dev_priv *priv = netdev_priv(dev);
515
516 if (!test_bit(IPOIB_MCAST_RUN, &priv->flags))
517 return;
518
519 if (ib_query_gid(priv->ca, priv->port, 0, &priv->local_gid))
520 ipoib_warn(priv, "ib_gid_entry_get() failed\n");
521 else
522 memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw, sizeof (union ib_gid));
523
524 {
525 struct ib_port_attr attr;
526
527 if (!ib_query_port(priv->ca, priv->port, &attr)) {
528 priv->local_lid = attr.lid;
529 priv->local_rate = attr.active_speed *
530 ib_width_enum_to_int(attr.active_width);
531 } else
532 ipoib_warn(priv, "ib_query_port failed\n");
533 }
534
535 if (!priv->broadcast) {
536 priv->broadcast = ipoib_mcast_alloc(dev, 1);
537 if (!priv->broadcast) {
538 ipoib_warn(priv, "failed to allocate broadcast group\n");
95ed644f 539 mutex_lock(&mcast_mutex);
1da177e4
LT		 540 if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
541 queue_delayed_work(ipoib_workqueue,
542 &priv->mcast_task, HZ);
95ed644f 543 mutex_unlock(&mcast_mutex);
1da177e4
LT		 544 return;
545 }
546
547 memcpy(priv->broadcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
548 sizeof (union ib_gid));
549
550 spin_lock_irq(&priv->lock);
551 __ipoib_mcast_add(dev, priv->broadcast);
552 spin_unlock_irq(&priv->lock);
553 }
554
555 if (!test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
556 ipoib_mcast_join(dev, priv->broadcast, 0);
557 return;
558 }
559
560 while (1) {
561 struct ipoib_mcast *mcast = NULL;
562
563 spin_lock_irq(&priv->lock);
564 list_for_each_entry(mcast, &priv->multicast_list, list) {
565 if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)
566 && !test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)
567 && !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
568 /* Found the next unjoined group */
569 break;
570 }
571 }
572 spin_unlock_irq(&priv->lock);
573
574 if (&mcast->list == &priv->multicast_list) {
575 /* All done */
576 break;
577 }
578
579 ipoib_mcast_join(dev, mcast, 1);
580 return;
581 }
582
583 priv->mcast_mtu = ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu) -
584 IPOIB_ENCAP_LEN;
585 dev->mtu = min(priv->mcast_mtu, priv->admin_mtu);
586
587 ipoib_dbg_mcast(priv, "successfully joined all multicast groups\n");
588
589 clear_bit(IPOIB_MCAST_RUN, &priv->flags);
590 netif_carrier_on(dev);
591}
592
593int ipoib_mcast_start_thread(struct net_device *dev)
594{
595 struct ipoib_dev_priv *priv = netdev_priv(dev);
596
597 ipoib_dbg_mcast(priv, "starting multicast thread\n");
598
95ed644f 599 mutex_lock(&mcast_mutex);
1da177e4
LT		 600 if (!test_and_set_bit(IPOIB_MCAST_RUN, &priv->flags))
601 queue_work(ipoib_workqueue, &priv->mcast_task);
95ed644f 602 mutex_unlock(&mcast_mutex);
1da177e4
LT		 603
604 return 0;
605}
606
8d2cae06 607int ipoib_mcast_stop_thread(struct net_device *dev, int flush)
1da177e4
LT		 608{
609 struct ipoib_dev_priv *priv = netdev_priv(dev);
610 struct ipoib_mcast *mcast;
611
612 ipoib_dbg_mcast(priv, "stopping multicast thread\n");
613
95ed644f 614 mutex_lock(&mcast_mutex);
1da177e4
LT		 615 clear_bit(IPOIB_MCAST_RUN, &priv->flags);
616 cancel_delayed_work(&priv->mcast_task);
95ed644f 617 mutex_unlock(&mcast_mutex);
1da177e4 618
8d2cae06
RD		 619 if (flush)
620 flush_workqueue(ipoib_workqueue);
1da177e4
LT		 621
622 if (priv->broadcast && priv->broadcast->query) {
623 ib_sa_cancel_query(priv->broadcast->query_id, priv->broadcast->query);
624 priv->broadcast->query = NULL;
625 ipoib_dbg_mcast(priv, "waiting for bcast\n");
626 wait_for_completion(&priv->broadcast->done);
627 }
628
629 list_for_each_entry(mcast, &priv->multicast_list, list) {
630 if (mcast->query) {
631 ib_sa_cancel_query(mcast->query_id, mcast->query);
632 mcast->query = NULL;
633 ipoib_dbg_mcast(priv, "waiting for MGID " IPOIB_GID_FMT "\n",
634 IPOIB_GID_ARG(mcast->mcmember.mgid));
635 wait_for_completion(&mcast->done);
636 }
637 }
638
639 return 0;
640}
641
642static int ipoib_mcast_leave(struct net_device *dev, struct ipoib_mcast *mcast)
643{
644 struct ipoib_dev_priv *priv = netdev_priv(dev);
645 struct ib_sa_mcmember_rec rec = {
646 .join_state = 1
647 };
648 int ret = 0;
649
650 if (!test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags))
651 return 0;
652
653 ipoib_dbg_mcast(priv, "leaving MGID " IPOIB_GID_FMT "\n",
654 IPOIB_GID_ARG(mcast->mcmember.mgid));
655
656 rec.mgid = mcast->mcmember.mgid;
657 rec.port_gid = priv->local_gid;
97f52eb4 658 rec.pkey = cpu_to_be16(priv->pkey);
1da177e4
LT		 659
660 /* Remove ourselves from the multicast group */
661 ret = ipoib_mcast_detach(dev, be16_to_cpu(mcast->mcmember.mlid),
662 &mcast->mcmember.mgid);
663 if (ret)
664 ipoib_warn(priv, "ipoib_mcast_detach failed (result = %d)\n", ret);
665
666 /*
667 * Just make one shot at leaving and don't wait for a reply;
668 * if we fail, too bad.
669 */
670 ret = ib_sa_mcmember_rec_delete(priv->ca, priv->port, &rec,
671 IB_SA_MCMEMBER_REC_MGID |
672 IB_SA_MCMEMBER_REC_PORT_GID |
673 IB_SA_MCMEMBER_REC_PKEY |
674 IB_SA_MCMEMBER_REC_JOIN_STATE,
675 0, GFP_ATOMIC, NULL,
676 mcast, &mcast->query);
677 if (ret < 0)
678 ipoib_warn(priv, "ib_sa_mcmember_rec_delete failed "
679 "for leave (result = %d)\n", ret);
680
681 return 0;
682}
683
684void ipoib_mcast_send(struct net_device *dev, union ib_gid *mgid,
685 struct sk_buff *skb)
686{
687 struct ipoib_dev_priv *priv = netdev_priv(dev);
688 struct ipoib_mcast *mcast;
689
690 /*
691 * We can only be called from ipoib_start_xmit, so we're
692 * inside tx_lock -- no need to save/restore flags.
693 */
694 spin_lock(&priv->lock);
695
696 mcast = __ipoib_mcast_find(dev, mgid);
697 if (!mcast) {
698 /* Let's create a new send only group now */
699 ipoib_dbg_mcast(priv, "setting up send only multicast group for "
700 IPOIB_GID_FMT "\n", IPOIB_GID_ARG(*mgid));
701
702 mcast = ipoib_mcast_alloc(dev, 0);
703 if (!mcast) {
704 ipoib_warn(priv, "unable to allocate memory for "
705 "multicast structure\n");
b36f170b 706 ++priv->stats.tx_dropped;
1da177e4
LT		 707 dev_kfree_skb_any(skb);
708 goto out;
709 }
710
711 set_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags);
712 mcast->mcmember.mgid = *mgid;
713 __ipoib_mcast_add(dev, mcast);
714 list_add_tail(&mcast->list, &priv->multicast_list);
715 }
716
717 if (!mcast->ah) {
718 if (skb_queue_len(&mcast->pkt_queue) < IPOIB_MAX_MCAST_QUEUE)
719 skb_queue_tail(&mcast->pkt_queue, skb);
b36f170b
MT		 720 else {
721 ++priv->stats.tx_dropped;
1da177e4 722 dev_kfree_skb_any(skb);
b36f170b 723 }
1da177e4
LT		 724
725 if (mcast->query)
726 ipoib_dbg_mcast(priv, "no address vector, "
727 "but multicast join already started\n");
728 else if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags))
729 ipoib_mcast_sendonly_join(mcast);
730
731 /*
732 * If lookup completes between here and out:, don't
733 * want to send packet twice.
734 */
735 mcast = NULL;
736 }
737
738out:
739 if (mcast && mcast->ah) {
740 if (skb->dst &&
741 skb->dst->neighbour &&
742 !*to_ipoib_neigh(skb->dst->neighbour)) {
743 struct ipoib_neigh *neigh = kmalloc(sizeof *neigh, GFP_ATOMIC);
744
745 if (neigh) {
746 kref_get(&mcast->ah->ref);
747 neigh->ah = mcast->ah;
748 neigh->neighbour = skb->dst->neighbour;
749 *to_ipoib_neigh(skb->dst->neighbour) = neigh;
750 list_add_tail(&neigh->list, &mcast->neigh_list);
751 }
752 }
753
754 ipoib_send(dev, skb, mcast->ah, IB_MULTICAST_QPN);
755 }
756
757 spin_unlock(&priv->lock);
758}
759
760void ipoib_mcast_dev_flush(struct net_device *dev)
761{
762 struct ipoib_dev_priv *priv = netdev_priv(dev);
763 LIST_HEAD(remove_list);
988bd503 764 struct ipoib_mcast *mcast, *tmcast;
1da177e4
LT		 765 unsigned long flags;
766
767 ipoib_dbg_mcast(priv, "flushing multicast list\n");
768
769 spin_lock_irqsave(&priv->lock, flags);
1da177e4 770
988bd503
EC		 771 list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
772 list_del(&mcast->list);
773 rb_erase(&mcast->rb_node, &priv->multicast_tree);
774 list_add_tail(&mcast->list, &remove_list);
1da177e4
LT		 775 }
776
777 if (priv->broadcast) {
988bd503
EC		 778 rb_erase(&priv->broadcast->rb_node, &priv->multicast_tree);
779 list_add_tail(&priv->broadcast->list, &remove_list);
780 priv->broadcast = NULL;
1da177e4
LT		 781 }
782
783 spin_unlock_irqrestore(&priv->lock, flags);
784
785 list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
786 ipoib_mcast_leave(dev, mcast);
787 ipoib_mcast_free(mcast);
788 }
789}
790
1da177e4
LT		 791void ipoib_mcast_restart_task(void *dev_ptr)
792{
793 struct net_device *dev = dev_ptr;
794 struct ipoib_dev_priv *priv = netdev_priv(dev);
795 struct dev_mc_list *mclist;
796 struct ipoib_mcast *mcast, *tmcast;
797 LIST_HEAD(remove_list);
798 unsigned long flags;
799
800 ipoib_dbg_mcast(priv, "restarting multicast task\n");
801
8d2cae06 802 ipoib_mcast_stop_thread(dev, 0);
1da177e4 803
78bfe0b5
MT		 804 spin_lock_irqsave(&dev->xmit_lock, flags);
805 spin_lock(&priv->lock);
1da177e4
LT		 806
807 /*
808 * Unfortunately, the networking core only gives us a list of all of
809 * the multicast hardware addresses. We need to figure out which ones
810 * are new and which ones have been removed
811 */
812
813 /* Clear out the found flag */
814 list_for_each_entry(mcast, &priv->multicast_list, list)
815 clear_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
816
817 /* Mark all of the entries that are found or don't exist */
818 for (mclist = dev->mc_list; mclist; mclist = mclist->next) {
819 union ib_gid mgid;
820
821 memcpy(mgid.raw, mclist->dmi_addr + 4, sizeof mgid);
822
823 /* Add in the P_Key */
824 mgid.raw[4] = (priv->pkey >> 8) & 0xff;
825 mgid.raw[5] = priv->pkey & 0xff;
826
827 mcast = __ipoib_mcast_find(dev, &mgid);
828 if (!mcast || test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
829 struct ipoib_mcast *nmcast;
830
831 /* Not found or send-only group, let's add a new entry */
832 ipoib_dbg_mcast(priv, "adding multicast entry for mgid "
833 IPOIB_GID_FMT "\n", IPOIB_GID_ARG(mgid));
834
835 nmcast = ipoib_mcast_alloc(dev, 0);
836 if (!nmcast) {
837 ipoib_warn(priv, "unable to allocate memory for multicast structure\n");
838 continue;
839 }
840
841 set_bit(IPOIB_MCAST_FLAG_FOUND, &nmcast->flags);
842
843 nmcast->mcmember.mgid = mgid;
844
845 if (mcast) {
846 /* Destroy the send only entry */
847 list_del(&mcast->list);
848 list_add_tail(&mcast->list, &remove_list);
849
850 rb_replace_node(&mcast->rb_node,
851 &nmcast->rb_node,
852 &priv->multicast_tree);
853 } else
854 __ipoib_mcast_add(dev, nmcast);
855
856 list_add_tail(&nmcast->list, &priv->multicast_list);
857 }
858
859 if (mcast)
860 set_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
861 }
862
863 /* Remove all of the entries don't exist anymore */
864 list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
865 if (!test_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags) &&
866 !test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
867 ipoib_dbg_mcast(priv, "deleting multicast group " IPOIB_GID_FMT "\n",
868 IPOIB_GID_ARG(mcast->mcmember.mgid));
869
870 rb_erase(&mcast->rb_node, &priv->multicast_tree);
871
872 /* Move to the remove list */
873 list_del(&mcast->list);
874 list_add_tail(&mcast->list, &remove_list);
875 }
876 }
78bfe0b5
MT		 877
878 spin_unlock(&priv->lock);
879 spin_unlock_irqrestore(&dev->xmit_lock, flags);
1da177e4
LT		 880
881 /* We have to cancel outside of the spinlock */
882 list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
883 ipoib_mcast_leave(mcast->dev, mcast);
884 ipoib_mcast_free(mcast);
885 }
886
887 if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
888 ipoib_mcast_start_thread(dev);
889}
890
8ae5a8a2
RD		 891#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
892
1da177e4
LT		 893struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev)
894{
895 struct ipoib_mcast_iter *iter;
896
897 iter = kmalloc(sizeof *iter, GFP_KERNEL);
898 if (!iter)
899 return NULL;
900
901 iter->dev = dev;
1732b0ef 902 memset(iter->mgid.raw, 0, 16);
1da177e4
LT		 903
904 if (ipoib_mcast_iter_next(iter)) {
1732b0ef 905 kfree(iter);
1da177e4
LT		 906 return NULL;
907 }
908
909 return iter;
910}
911
1da177e4
LT		 912int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter)
913{
914 struct ipoib_dev_priv *priv = netdev_priv(iter->dev);
915 struct rb_node *n;
916 struct ipoib_mcast *mcast;
917 int ret = 1;
918
919 spin_lock_irq(&priv->lock);
920
921 n = rb_first(&priv->multicast_tree);
922
923 while (n) {
924 mcast = rb_entry(n, struct ipoib_mcast, rb_node);
925
926 if (memcmp(iter->mgid.raw, mcast->mcmember.mgid.raw,
927 sizeof (union ib_gid)) < 0) {
928 iter->mgid = mcast->mcmember.mgid;
929 iter->created = mcast->created;
930 iter->queuelen = skb_queue_len(&mcast->pkt_queue);
931 iter->complete = !!mcast->ah;
932 iter->send_only = !!(mcast->flags & (1 << IPOIB_MCAST_FLAG_SENDONLY));
933
934 ret = 0;
935
936 break;
937 }
938
939 n = rb_next(n);
940 }
941
942 spin_unlock_irq(&priv->lock);
943
944 return ret;
945}
946
947void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
948 union ib_gid *mgid,
949 unsigned long *created,
950 unsigned int *queuelen,
951 unsigned int *complete,
952 unsigned int *send_only)
953{
954 *mgid = iter->mgid;
955 *created = iter->created;
956 *queuelen = iter->queuelen;
957 *complete = iter->complete;
958 *send_only = iter->send_only;
959}
8ae5a8a2
RD		 960
961#endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
Linux kernel - Deliverables
This page took 0.214152 seconds and 5 git commands to generate.