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 "ipoib.h"
36
37 #include <linux/delay.h>
38 #include <linux/completion.h>
39
40 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
41 static int mcast_debug_level = 1;
42
43 module_param(mcast_debug_level, int, 0644);
44 MODULE_PARM_DESC(mcast_debug_level,
45 "Enable multicast debug tracing if > 0");
46 #endif
47
48 static DEFINE_MUTEX(mcast_mutex);
49
50 struct ipoib_mcast_iter {
51 struct ipoib_dev_priv *priv;
52 union ib_gid mgid;
53 unsigned long created;
54 unsigned int queuelen;
55 unsigned int complete;
56 unsigned int send_only;
57 };
58
ipoib_mcast_free(struct ipoib_mcast * mcast)59 static void ipoib_mcast_free(struct ipoib_mcast *mcast)
60 {
61 struct ifnet *dev = mcast->priv->dev;
62 int tx_dropped = 0;
63
64 ipoib_dbg_mcast(mcast->priv, "deleting multicast group %16D\n",
65 mcast->mcmember.mgid.raw, ":");
66
67 if (mcast->ah)
68 ipoib_put_ah(mcast->ah);
69
70 tx_dropped = mcast->pkt_queue.ifq_len;
71 _IF_DRAIN(&mcast->pkt_queue); /* XXX Locking. */
72
73 if_inc_counter(dev, IFCOUNTER_OERRORS, tx_dropped);
74
75 kfree(mcast);
76 }
77
ipoib_mcast_alloc(struct ipoib_dev_priv * priv,int can_sleep)78 static struct ipoib_mcast *ipoib_mcast_alloc(struct ipoib_dev_priv *priv,
79 int can_sleep)
80 {
81 struct ipoib_mcast *mcast;
82
83 mcast = kzalloc(sizeof *mcast, can_sleep ? GFP_KERNEL : GFP_ATOMIC);
84 if (!mcast)
85 return NULL;
86
87 mcast->priv = priv;
88 mcast->created = jiffies;
89 mcast->backoff = 1;
90
91 INIT_LIST_HEAD(&mcast->list);
92 bzero(&mcast->pkt_queue, sizeof(mcast->pkt_queue));
93
94 return mcast;
95 }
96
__ipoib_mcast_find(struct ipoib_dev_priv * priv,void * mgid)97 static struct ipoib_mcast *__ipoib_mcast_find(struct ipoib_dev_priv *priv,
98 void *mgid)
99 {
100 struct rb_node *n = priv->multicast_tree.rb_node;
101
102 while (n) {
103 struct ipoib_mcast *mcast;
104 int ret;
105
106 mcast = rb_entry(n, struct ipoib_mcast, rb_node);
107
108 ret = memcmp(mgid, mcast->mcmember.mgid.raw,
109 sizeof (union ib_gid));
110 if (ret < 0)
111 n = n->rb_left;
112 else if (ret > 0)
113 n = n->rb_right;
114 else
115 return mcast;
116 }
117
118 return NULL;
119 }
120
__ipoib_mcast_add(struct ipoib_dev_priv * priv,struct ipoib_mcast * mcast)121 static int __ipoib_mcast_add(struct ipoib_dev_priv *priv,
122 struct ipoib_mcast *mcast)
123 {
124 struct rb_node **n = &priv->multicast_tree.rb_node, *pn = NULL;
125
126 while (*n) {
127 struct ipoib_mcast *tmcast;
128 int ret;
129
130 pn = *n;
131 tmcast = rb_entry(pn, struct ipoib_mcast, rb_node);
132
133 ret = memcmp(mcast->mcmember.mgid.raw, tmcast->mcmember.mgid.raw,
134 sizeof (union ib_gid));
135 if (ret < 0)
136 n = &pn->rb_left;
137 else if (ret > 0)
138 n = &pn->rb_right;
139 else
140 return -EEXIST;
141 }
142
143 rb_link_node(&mcast->rb_node, pn, n);
144 rb_insert_color(&mcast->rb_node, &priv->multicast_tree);
145
146 return 0;
147 }
148
ipoib_mcast_join_finish(struct ipoib_mcast * mcast,struct ib_sa_mcmember_rec * mcmember)149 static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
150 struct ib_sa_mcmember_rec *mcmember)
151 {
152 struct ipoib_dev_priv *priv = mcast->priv;
153 struct ifnet *dev = priv->dev;
154 struct ipoib_ah *ah;
155 int ret;
156 int set_qkey = 0;
157
158 mcast->mcmember = *mcmember;
159
160 /* Set the cached Q_Key before we attach if it's the broadcast group */
161 if (!memcmp(mcast->mcmember.mgid.raw, dev->if_broadcastaddr + 4,
162 sizeof (union ib_gid))) {
163 spin_lock_irq(&priv->lock);
164 if (!priv->broadcast) {
165 spin_unlock_irq(&priv->lock);
166 return -EAGAIN;
167 }
168 priv->qkey = be32_to_cpu(priv->broadcast->mcmember.qkey);
169 spin_unlock_irq(&priv->lock);
170 priv->tx_wr.wr.ud.remote_qkey = priv->qkey;
171 set_qkey = 1;
172 }
173
174 if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
175 if (test_and_set_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
176 ipoib_warn(priv, "multicast group %16D already attached\n",
177 mcast->mcmember.mgid.raw, ":");
178
179 return 0;
180 }
181
182 ret = ipoib_mcast_attach(priv, be16_to_cpu(mcast->mcmember.mlid),
183 &mcast->mcmember.mgid, set_qkey);
184 if (ret < 0) {
185 ipoib_warn(priv, "couldn't attach QP to multicast group %16D\n",
186 mcast->mcmember.mgid.raw, ":");
187
188 clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags);
189 return ret;
190 }
191 }
192
193 {
194 struct ib_ah_attr av = {
195 .dlid = be16_to_cpu(mcast->mcmember.mlid),
196 .port_num = priv->port,
197 .sl = mcast->mcmember.sl,
198 .ah_flags = IB_AH_GRH,
199 .static_rate = mcast->mcmember.rate,
200 .grh = {
201 .flow_label = be32_to_cpu(mcast->mcmember.flow_label),
202 .hop_limit = mcast->mcmember.hop_limit,
203 .sgid_index = 0,
204 .traffic_class = mcast->mcmember.traffic_class
205 }
206 };
207 av.grh.dgid = mcast->mcmember.mgid;
208
209 ah = ipoib_create_ah(priv, priv->pd, &av);
210 if (!ah) {
211 ipoib_warn(priv, "ib_address_create failed\n");
212 } else {
213 spin_lock_irq(&priv->lock);
214 mcast->ah = ah;
215 spin_unlock_irq(&priv->lock);
216
217 ipoib_dbg_mcast(priv, "MGID %16D AV %p, LID 0x%04x, SL %d\n",
218 mcast->mcmember.mgid.raw, ":",
219 mcast->ah->ah,
220 be16_to_cpu(mcast->mcmember.mlid),
221 mcast->mcmember.sl);
222 }
223 }
224
225 /* actually send any queued packets */
226 while (mcast->pkt_queue.ifq_len) {
227 struct mbuf *mb;
228 _IF_DEQUEUE(&mcast->pkt_queue, mb);
229 mb->m_pkthdr.rcvif = dev;
230
231 if (dev->if_transmit(dev, mb))
232 ipoib_warn(priv, "dev_queue_xmit failed to requeue packet\n");
233 }
234
235 return 0;
236 }
237
238 static int
ipoib_mcast_sendonly_join_complete(int status,struct ib_sa_multicast * multicast)239 ipoib_mcast_sendonly_join_complete(int status,
240 struct ib_sa_multicast *multicast)
241 {
242 struct ipoib_mcast *mcast = multicast->context;
243 struct ipoib_dev_priv *priv = mcast->priv;
244
245 /* We trap for port events ourselves. */
246 if (status == -ENETRESET)
247 return 0;
248
249 if (!status)
250 status = ipoib_mcast_join_finish(mcast, &multicast->rec);
251
252 if (status) {
253 if (mcast->logcount++ < 20)
254 ipoib_dbg_mcast(priv, "multicast join failed for %16D, status %d\n",
255 mcast->mcmember.mgid.raw, ":", status);
256
257 /* Flush out any queued packets */
258 if_inc_counter(priv->dev, IFCOUNTER_OERRORS, mcast->pkt_queue.ifq_len);
259 _IF_DRAIN(&mcast->pkt_queue);
260
261 /* Clear the busy flag so we try again */
262 status = test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY,
263 &mcast->flags);
264 }
265 return status;
266 }
267
ipoib_mcast_sendonly_join(struct ipoib_mcast * mcast)268 static int ipoib_mcast_sendonly_join(struct ipoib_mcast *mcast)
269 {
270 struct ipoib_dev_priv *priv = mcast->priv;
271 struct ib_sa_mcmember_rec rec = {
272 #if 0 /* Some SMs don't support send-only yet */
273 .join_state = 4
274 #else
275 .join_state = 1
276 #endif
277 };
278 int ret = 0;
279
280 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
281 ipoib_dbg_mcast(priv, "device shutting down, no multicast joins\n");
282 return -ENODEV;
283 }
284
285 if (test_and_set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) {
286 ipoib_dbg_mcast(priv, "multicast entry busy, skipping\n");
287 return -EBUSY;
288 }
289
290 rec.mgid = mcast->mcmember.mgid;
291 rec.port_gid = priv->local_gid;
292 rec.pkey = cpu_to_be16(priv->pkey);
293
294 mcast->mc = ib_sa_join_multicast(&ipoib_sa_client, priv->ca,
295 priv->port, &rec,
296 IB_SA_MCMEMBER_REC_MGID |
297 IB_SA_MCMEMBER_REC_PORT_GID |
298 IB_SA_MCMEMBER_REC_PKEY |
299 IB_SA_MCMEMBER_REC_JOIN_STATE,
300 GFP_ATOMIC,
301 ipoib_mcast_sendonly_join_complete,
302 mcast);
303 if (IS_ERR(mcast->mc)) {
304 ret = PTR_ERR(mcast->mc);
305 clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
306 ipoib_warn(priv, "ib_sa_join_multicast failed (ret = %d)\n",
307 ret);
308 } else {
309 ipoib_dbg_mcast(priv, "no multicast record for %16D, starting join\n",
310 mcast->mcmember.mgid.raw, ":");
311 }
312
313 return ret;
314 }
315
ipoib_mcast_carrier_on_task(struct work_struct * work)316 void ipoib_mcast_carrier_on_task(struct work_struct *work)
317 {
318 struct ipoib_dev_priv *priv = container_of(work, struct ipoib_dev_priv,
319 carrier_on_task);
320 struct ib_port_attr attr;
321
322 /*
323 * Take rtnl_lock to avoid racing with ipoib_stop() and
324 * turning the carrier back on while a device is being
325 * removed.
326 */
327 if (ib_query_port(priv->ca, priv->port, &attr) ||
328 attr.state != IB_PORT_ACTIVE) {
329 ipoib_dbg(priv, "Keeping carrier off until IB port is active\n");
330 return;
331 }
332 if_link_state_change(priv->dev, LINK_STATE_UP);
333 }
334
ipoib_mcast_join_complete(int status,struct ib_sa_multicast * multicast)335 static int ipoib_mcast_join_complete(int status,
336 struct ib_sa_multicast *multicast)
337 {
338 struct ipoib_mcast *mcast = multicast->context;
339 struct ipoib_dev_priv *priv = mcast->priv;
340
341 ipoib_dbg_mcast(priv, "join completion for %16D (status %d)\n",
342 mcast->mcmember.mgid.raw, ":", status);
343
344 /* We trap for port events ourselves. */
345 if (status == -ENETRESET)
346 return 0;
347
348 if (!status)
349 status = ipoib_mcast_join_finish(mcast, &multicast->rec);
350
351 if (!status) {
352 mcast->backoff = 1;
353 mutex_lock(&mcast_mutex);
354 if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
355 queue_delayed_work(ipoib_workqueue,
356 &priv->mcast_task, 0);
357 mutex_unlock(&mcast_mutex);
358
359 /*
360 * Defer carrier on work to ipoib_workqueue to avoid a
361 * deadlock on rtnl_lock here.
362 */
363 if (mcast == priv->broadcast)
364 queue_work(ipoib_workqueue, &priv->carrier_on_task);
365
366 return 0;
367 }
368
369 if (mcast->logcount++ < 20) {
370 if (status == -ETIMEDOUT || status == -EAGAIN) {
371 ipoib_dbg_mcast(priv, "multicast join failed for %16D, status %d\n",
372 mcast->mcmember.mgid.raw, ":", status);
373 } else {
374 ipoib_warn(priv, "multicast join failed for %16D, status %d\n",
375 mcast->mcmember.mgid.raw, ":", status);
376 }
377 }
378
379 mcast->backoff *= 2;
380 if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
381 mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
382
383 /* Clear the busy flag so we try again */
384 status = test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
385
386 mutex_lock(&mcast_mutex);
387 spin_lock_irq(&priv->lock);
388 if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
389 queue_delayed_work(ipoib_workqueue, &priv->mcast_task,
390 mcast->backoff * HZ);
391 spin_unlock_irq(&priv->lock);
392 mutex_unlock(&mcast_mutex);
393
394 return status;
395 }
396
ipoib_mcast_join(struct ipoib_dev_priv * priv,struct ipoib_mcast * mcast,int create)397 static void ipoib_mcast_join(struct ipoib_dev_priv *priv,
398 struct ipoib_mcast *mcast, int create)
399 {
400 struct ib_sa_mcmember_rec rec = {
401 .join_state = 1
402 };
403 ib_sa_comp_mask comp_mask;
404 int ret = 0;
405
406 ipoib_dbg_mcast(priv, "joining MGID %16D\n",
407 mcast->mcmember.mgid.raw, ":");
408
409 rec.mgid = mcast->mcmember.mgid;
410 rec.port_gid = priv->local_gid;
411 rec.pkey = cpu_to_be16(priv->pkey);
412
413 comp_mask =
414 IB_SA_MCMEMBER_REC_MGID |
415 IB_SA_MCMEMBER_REC_PORT_GID |
416 IB_SA_MCMEMBER_REC_PKEY |
417 IB_SA_MCMEMBER_REC_JOIN_STATE;
418
419 if (create) {
420 comp_mask |=
421 IB_SA_MCMEMBER_REC_QKEY |
422 IB_SA_MCMEMBER_REC_MTU_SELECTOR |
423 IB_SA_MCMEMBER_REC_MTU |
424 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS |
425 IB_SA_MCMEMBER_REC_RATE_SELECTOR |
426 IB_SA_MCMEMBER_REC_RATE |
427 IB_SA_MCMEMBER_REC_SL |
428 IB_SA_MCMEMBER_REC_FLOW_LABEL |
429 IB_SA_MCMEMBER_REC_HOP_LIMIT;
430
431 rec.qkey = priv->broadcast->mcmember.qkey;
432 rec.mtu_selector = IB_SA_EQ;
433 rec.mtu = priv->broadcast->mcmember.mtu;
434 rec.traffic_class = priv->broadcast->mcmember.traffic_class;
435 rec.rate_selector = IB_SA_EQ;
436 rec.rate = priv->broadcast->mcmember.rate;
437 rec.sl = priv->broadcast->mcmember.sl;
438 rec.flow_label = priv->broadcast->mcmember.flow_label;
439 rec.hop_limit = priv->broadcast->mcmember.hop_limit;
440 }
441
442 set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
443 mcast->mc = ib_sa_join_multicast(&ipoib_sa_client, priv->ca, priv->port,
444 &rec, comp_mask, GFP_KERNEL,
445 ipoib_mcast_join_complete, mcast);
446 if (IS_ERR(mcast->mc)) {
447 clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
448 ret = PTR_ERR(mcast->mc);
449 ipoib_warn(priv, "ib_sa_join_multicast failed, status %d\n", ret);
450
451 mcast->backoff *= 2;
452 if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
453 mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
454
455 mutex_lock(&mcast_mutex);
456 if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
457 queue_delayed_work(ipoib_workqueue,
458 &priv->mcast_task,
459 mcast->backoff * HZ);
460 mutex_unlock(&mcast_mutex);
461 }
462 }
463
ipoib_mcast_join_task(struct work_struct * work)464 void ipoib_mcast_join_task(struct work_struct *work)
465 {
466 struct ipoib_dev_priv *priv =
467 container_of(work, struct ipoib_dev_priv, mcast_task.work);
468 struct ifnet *dev = priv->dev;
469 struct ib_port_attr attr;
470
471 ipoib_dbg_mcast(priv, "Running join task. flags 0x%lX\n", priv->flags);
472
473 if (!test_bit(IPOIB_MCAST_RUN, &priv->flags))
474 return;
475
476 if (ib_query_port(priv->ca, priv->port, &attr) ||
477 attr.state != IB_PORT_ACTIVE) {
478 ipoib_dbg(priv, "%s: port state is not ACTIVE (state = %d) suspend task.\n",
479 __func__, attr.state);
480 return;
481 }
482
483 if (ib_query_gid(priv->ca, priv->port, 0, &priv->local_gid))
484 ipoib_warn(priv, "ib_query_gid() failed\n");
485 else
486 memcpy(IF_LLADDR(dev) + 4, priv->local_gid.raw, sizeof (union ib_gid));
487
488 {
489 struct ib_port_attr attr;
490
491 if (!ib_query_port(priv->ca, priv->port, &attr))
492 priv->local_lid = attr.lid;
493 else
494 ipoib_warn(priv, "ib_query_port failed\n");
495 }
496
497 if (!priv->broadcast) {
498 struct ipoib_mcast *broadcast;
499
500 if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
501 return;
502
503 broadcast = ipoib_mcast_alloc(priv, 1);
504 if (!broadcast) {
505 ipoib_warn(priv, "failed to allocate broadcast group\n");
506 mutex_lock(&mcast_mutex);
507 if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
508 queue_delayed_work(ipoib_workqueue,
509 &priv->mcast_task, HZ);
510 mutex_unlock(&mcast_mutex);
511 return;
512 }
513
514 spin_lock_irq(&priv->lock);
515 memcpy(broadcast->mcmember.mgid.raw, dev->if_broadcastaddr + 4,
516 sizeof (union ib_gid));
517 priv->broadcast = broadcast;
518
519 __ipoib_mcast_add(priv, priv->broadcast);
520 spin_unlock_irq(&priv->lock);
521 }
522
523 if (priv->broadcast &&
524 !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
525 if (priv->broadcast &&
526 !test_bit(IPOIB_MCAST_FLAG_BUSY, &priv->broadcast->flags))
527 ipoib_mcast_join(priv, priv->broadcast, 0);
528 return;
529 }
530
531 while (1) {
532 struct ipoib_mcast *mcast = NULL;
533
534 spin_lock_irq(&priv->lock);
535 list_for_each_entry(mcast, &priv->multicast_list, list) {
536 if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)
537 && !test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)
538 && !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
539 /* Found the next unjoined group */
540 break;
541 }
542 }
543 spin_unlock_irq(&priv->lock);
544
545 if (&mcast->list == &priv->multicast_list) {
546 /* All done */
547 break;
548 }
549
550 ipoib_mcast_join(priv, mcast, 1);
551 return;
552 }
553
554 spin_lock_irq(&priv->lock);
555 if (priv->broadcast)
556 priv->mcast_mtu = IPOIB_UD_MTU(ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu));
557 else
558 priv->mcast_mtu = priv->admin_mtu;
559 spin_unlock_irq(&priv->lock);
560
561 if (!ipoib_cm_admin_enabled(priv))
562 ipoib_change_mtu(priv, min(priv->mcast_mtu, priv->admin_mtu));
563
564 ipoib_dbg_mcast(priv, "successfully joined all multicast groups\n");
565
566 clear_bit(IPOIB_MCAST_RUN, &priv->flags);
567 }
568
ipoib_mcast_start_thread(struct ipoib_dev_priv * priv)569 int ipoib_mcast_start_thread(struct ipoib_dev_priv *priv)
570 {
571 ipoib_dbg_mcast(priv, "starting multicast thread flags 0x%lX\n",
572 priv->flags);
573
574 mutex_lock(&mcast_mutex);
575 if (!test_and_set_bit(IPOIB_MCAST_RUN, &priv->flags))
576 queue_delayed_work(ipoib_workqueue, &priv->mcast_task, 0);
577 mutex_unlock(&mcast_mutex);
578
579 return 0;
580 }
581
ipoib_mcast_stop_thread(struct ipoib_dev_priv * priv,int flush)582 int ipoib_mcast_stop_thread(struct ipoib_dev_priv *priv, int flush)
583 {
584
585 ipoib_dbg_mcast(priv, "stopping multicast thread\n");
586
587 mutex_lock(&mcast_mutex);
588 clear_bit(IPOIB_MCAST_RUN, &priv->flags);
589 cancel_delayed_work(&priv->mcast_task);
590 mutex_unlock(&mcast_mutex);
591
592 if (flush)
593 flush_workqueue(ipoib_workqueue);
594
595 return 0;
596 }
597
ipoib_mcast_leave(struct ipoib_dev_priv * priv,struct ipoib_mcast * mcast)598 static int ipoib_mcast_leave(struct ipoib_dev_priv *priv, struct ipoib_mcast *mcast)
599 {
600 int ret = 0;
601
602 if (test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
603 ib_sa_free_multicast(mcast->mc);
604
605 if (test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
606 ipoib_dbg_mcast(priv, "leaving MGID %16D\n",
607 mcast->mcmember.mgid.raw, ":");
608
609 /* Remove ourselves from the multicast group */
610 ret = ib_detach_mcast(priv->qp, &mcast->mcmember.mgid,
611 be16_to_cpu(mcast->mcmember.mlid));
612 if (ret)
613 ipoib_warn(priv, "ib_detach_mcast failed (result = %d)\n", ret);
614 }
615
616 return 0;
617 }
618
619 void
ipoib_mcast_send(struct ipoib_dev_priv * priv,void * mgid,struct mbuf * mb)620 ipoib_mcast_send(struct ipoib_dev_priv *priv, void *mgid, struct mbuf *mb)
621 {
622 struct ifnet *dev = priv->dev;
623 struct ipoib_mcast *mcast;
624
625 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags) ||
626 !priv->broadcast ||
627 !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
628 if_inc_counter(dev, IFCOUNTER_OERRORS, 1);
629 m_freem(mb);
630 return;
631 }
632
633 mcast = __ipoib_mcast_find(priv, mgid);
634 if (!mcast) {
635 /* Let's create a new send only group now */
636 ipoib_dbg_mcast(priv, "setting up send only multicast group for %16D\n",
637 mgid, ":");
638
639 mcast = ipoib_mcast_alloc(priv, 0);
640 if (!mcast) {
641 ipoib_warn(priv, "unable to allocate memory for "
642 "multicast structure\n");
643 if_inc_counter(dev, IFCOUNTER_OERRORS, 1);
644 m_freem(mb);
645 goto out;
646 }
647
648 set_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags);
649 memcpy(mcast->mcmember.mgid.raw, mgid, sizeof (union ib_gid));
650 __ipoib_mcast_add(priv, mcast);
651 list_add_tail(&mcast->list, &priv->multicast_list);
652 }
653
654 if (!mcast->ah) {
655 if (mcast->pkt_queue.ifq_len < IPOIB_MAX_MCAST_QUEUE) {
656 _IF_ENQUEUE(&mcast->pkt_queue, mb);
657 } else {
658 if_inc_counter(dev, IFCOUNTER_OERRORS, 1);
659 m_freem(mb);
660 }
661
662 if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
663 ipoib_dbg_mcast(priv, "no address vector, "
664 "but multicast join already started\n");
665 else if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags))
666 ipoib_mcast_sendonly_join(mcast);
667
668 /*
669 * If lookup completes between here and out:, don't
670 * want to send packet twice.
671 */
672 mcast = NULL;
673 }
674
675 out:
676 if (mcast && mcast->ah)
677 ipoib_send(priv, mb, mcast->ah, IB_MULTICAST_QPN);
678 }
679
ipoib_mcast_dev_flush(struct ipoib_dev_priv * priv)680 void ipoib_mcast_dev_flush(struct ipoib_dev_priv *priv)
681 {
682 LIST_HEAD(remove_list);
683 struct ipoib_mcast *mcast, *tmcast;
684 unsigned long flags;
685
686 ipoib_dbg_mcast(priv, "flushing multicast list\n");
687
688 spin_lock_irqsave(&priv->lock, flags);
689
690 list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
691 list_del(&mcast->list);
692 rb_erase(&mcast->rb_node, &priv->multicast_tree);
693 list_add_tail(&mcast->list, &remove_list);
694 }
695
696 if (priv->broadcast) {
697 rb_erase(&priv->broadcast->rb_node, &priv->multicast_tree);
698 list_add_tail(&priv->broadcast->list, &remove_list);
699 priv->broadcast = NULL;
700 }
701
702 spin_unlock_irqrestore(&priv->lock, flags);
703
704 list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
705 ipoib_mcast_leave(priv, mcast);
706 ipoib_mcast_free(mcast);
707 }
708 }
709
ipoib_mcast_addr_is_valid(const u8 * addr,unsigned int addrlen,const u8 * broadcast)710 static int ipoib_mcast_addr_is_valid(const u8 *addr, unsigned int addrlen,
711 const u8 *broadcast)
712 {
713 if (addrlen != INFINIBAND_ALEN)
714 return 0;
715 /* reserved QPN, prefix, scope */
716 if (memcmp(addr, broadcast, 6))
717 return 0;
718 /* signature lower, pkey */
719 if (memcmp(addr + 7, broadcast + 7, 3))
720 return 0;
721 return 1;
722 }
723
ipoib_mcast_restart_task(struct work_struct * work)724 void ipoib_mcast_restart_task(struct work_struct *work)
725 {
726 struct ipoib_dev_priv *priv =
727 container_of(work, struct ipoib_dev_priv, restart_task);
728 ipoib_mcast_restart(priv);
729 }
730
ipoib_mcast_restart(struct ipoib_dev_priv * priv)731 void ipoib_mcast_restart(struct ipoib_dev_priv *priv)
732 {
733 struct ifnet *dev = priv->dev;
734 struct ifmultiaddr *ifma;
735 struct ipoib_mcast *mcast, *tmcast;
736 LIST_HEAD(remove_list);
737 struct ib_sa_mcmember_rec rec;
738 int addrlen;
739
740 ipoib_dbg_mcast(priv, "restarting multicast task flags 0x%lX\n",
741 priv->flags);
742
743 ipoib_mcast_stop_thread(priv, 0);
744
745 if_maddr_rlock(dev);
746 spin_lock(&priv->lock);
747
748 /*
749 * Unfortunately, the networking core only gives us a list of all of
750 * the multicast hardware addresses. We need to figure out which ones
751 * are new and which ones have been removed
752 */
753
754 /* Clear out the found flag */
755 list_for_each_entry(mcast, &priv->multicast_list, list)
756 clear_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
757
758 /* Mark all of the entries that are found or don't exist */
759
760
761 TAILQ_FOREACH(ifma, &dev->if_multiaddrs, ifma_link) {
762 union ib_gid mgid;
763 uint8_t *addr;
764
765 if (ifma->ifma_addr->sa_family != AF_LINK)
766 continue;
767 addr = LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
768 addrlen = ((struct sockaddr_dl *)ifma->ifma_addr)->sdl_alen;
769 if (!ipoib_mcast_addr_is_valid(addr, addrlen,
770 dev->if_broadcastaddr))
771 continue;
772
773 memcpy(mgid.raw, addr + 4, sizeof mgid);
774
775 mcast = __ipoib_mcast_find(priv, &mgid);
776 if (!mcast || test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
777 struct ipoib_mcast *nmcast;
778
779 /* ignore group which is directly joined by userspace */
780 if (test_bit(IPOIB_FLAG_UMCAST, &priv->flags) &&
781 !ib_sa_get_mcmember_rec(priv->ca, priv->port, &mgid, &rec)) {
782 ipoib_dbg_mcast(priv, "ignoring multicast entry for mgid %16D\n",
783 mgid.raw, ":");
784 continue;
785 }
786
787 /* Not found or send-only group, let's add a new entry */
788 ipoib_dbg_mcast(priv, "adding multicast entry for mgid %16D\n",
789 mgid.raw, ":");
790
791 nmcast = ipoib_mcast_alloc(priv, 0);
792 if (!nmcast) {
793 ipoib_warn(priv, "unable to allocate memory for multicast structure\n");
794 continue;
795 }
796
797 set_bit(IPOIB_MCAST_FLAG_FOUND, &nmcast->flags);
798
799 nmcast->mcmember.mgid = mgid;
800
801 if (mcast) {
802 /* Destroy the send only entry */
803 list_move_tail(&mcast->list, &remove_list);
804
805 rb_replace_node(&mcast->rb_node,
806 &nmcast->rb_node,
807 &priv->multicast_tree);
808 } else
809 __ipoib_mcast_add(priv, nmcast);
810
811 list_add_tail(&nmcast->list, &priv->multicast_list);
812 }
813
814 if (mcast)
815 set_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
816 }
817
818 /* Remove all of the entries don't exist anymore */
819 list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
820 if (!test_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags) &&
821 !test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
822 ipoib_dbg_mcast(priv, "deleting multicast group %16D\n",
823 mcast->mcmember.mgid.raw, ":");
824
825 rb_erase(&mcast->rb_node, &priv->multicast_tree);
826
827 /* Move to the remove list */
828 list_move_tail(&mcast->list, &remove_list);
829 }
830 }
831
832 spin_unlock(&priv->lock);
833 if_maddr_runlock(dev);
834
835 /* We have to cancel outside of the spinlock */
836 list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
837 ipoib_mcast_leave(mcast->priv, mcast);
838 ipoib_mcast_free(mcast);
839 }
840
841 if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
842 ipoib_mcast_start_thread(priv);
843 }
844
845 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
846
ipoib_mcast_iter_init(struct ipoib_dev_priv * priv)847 struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct ipoib_dev_priv *priv)
848 {
849 struct ipoib_mcast_iter *iter;
850
851 iter = kmalloc(sizeof *iter, GFP_KERNEL);
852 if (!iter)
853 return NULL;
854
855 iter->priv = priv;
856 memset(iter->mgid.raw, 0, 16);
857
858 if (ipoib_mcast_iter_next(iter)) {
859 kfree(iter);
860 return NULL;
861 }
862
863 return iter;
864 }
865
ipoib_mcast_iter_next(struct ipoib_mcast_iter * iter)866 int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter)
867 {
868 struct ipoib_dev_priv *priv = iter->priv;
869 struct rb_node *n;
870 struct ipoib_mcast *mcast;
871 int ret = 1;
872
873 spin_lock_irq(&priv->lock);
874
875 n = rb_first(&priv->multicast_tree);
876
877 while (n) {
878 mcast = rb_entry(n, struct ipoib_mcast, rb_node);
879
880 if (memcmp(iter->mgid.raw, mcast->mcmember.mgid.raw,
881 sizeof (union ib_gid)) < 0) {
882 iter->mgid = mcast->mcmember.mgid;
883 iter->created = mcast->created;
884 iter->queuelen = mcast->pkt_queue.ifq_len;
885 iter->complete = !!mcast->ah;
886 iter->send_only = !!(mcast->flags & (1 << IPOIB_MCAST_FLAG_SENDONLY));
887
888 ret = 0;
889
890 break;
891 }
892
893 n = rb_next(n);
894 }
895
896 spin_unlock_irq(&priv->lock);
897
898 return ret;
899 }
900
ipoib_mcast_iter_read(struct ipoib_mcast_iter * iter,union ib_gid * mgid,unsigned long * created,unsigned int * queuelen,unsigned int * complete,unsigned int * send_only)901 void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
902 union ib_gid *mgid,
903 unsigned long *created,
904 unsigned int *queuelen,
905 unsigned int *complete,
906 unsigned int *send_only)
907 {
908 *mgid = iter->mgid;
909 *created = iter->created;
910 *queuelen = iter->queuelen;
911 *complete = iter->complete;
912 *send_only = iter->send_only;
913 }
914
915 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
916