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.
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:
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
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.
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
35 #include <linux/skbuff.h>
36 #include <linux/rtnetlink.h>
37 #include <linux/moduleparam.h>
40 #include <linux/igmp.h>
41 #include <linux/inetdevice.h>
42 #include <linux/delay.h>
43 #include <linux/completion.h>
44 #include <linux/slab.h>
50 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
51 static int mcast_debug_level;
53 module_param(mcast_debug_level, int, 0644);
54 MODULE_PARM_DESC(mcast_debug_level,
55 "Enable multicast debug tracing if > 0");
58 static DEFINE_MUTEX(mcast_mutex);
60 struct ipoib_mcast_iter {
61 struct net_device *dev;
63 unsigned long created;
64 unsigned int queuelen;
65 unsigned int complete;
66 unsigned int send_only;
69 static void ipoib_mcast_free(struct ipoib_mcast *mcast)
71 struct net_device *dev = mcast->dev;
74 ipoib_dbg_mcast(netdev_priv(dev), "deleting multicast group %pI6\n",
75 mcast->mcmember.mgid.raw);
77 /* remove all neigh connected to this mcast */
78 ipoib_del_neighs_by_gid(dev, mcast->mcmember.mgid.raw);
81 ipoib_put_ah(mcast->ah);
83 while (!skb_queue_empty(&mcast->pkt_queue)) {
85 dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
88 netif_tx_lock_bh(dev);
89 dev->stats.tx_dropped += tx_dropped;
90 netif_tx_unlock_bh(dev);
95 static struct ipoib_mcast *ipoib_mcast_alloc(struct net_device *dev,
98 struct ipoib_mcast *mcast;
100 mcast = kzalloc(sizeof *mcast, can_sleep ? GFP_KERNEL : GFP_ATOMIC);
105 mcast->created = jiffies;
108 INIT_LIST_HEAD(&mcast->list);
109 INIT_LIST_HEAD(&mcast->neigh_list);
110 skb_queue_head_init(&mcast->pkt_queue);
115 static struct ipoib_mcast *__ipoib_mcast_find(struct net_device *dev, void *mgid)
117 struct ipoib_dev_priv *priv = netdev_priv(dev);
118 struct rb_node *n = priv->multicast_tree.rb_node;
121 struct ipoib_mcast *mcast;
124 mcast = rb_entry(n, struct ipoib_mcast, rb_node);
126 ret = memcmp(mgid, mcast->mcmember.mgid.raw,
127 sizeof (union ib_gid));
139 static int __ipoib_mcast_add(struct net_device *dev, struct ipoib_mcast *mcast)
141 struct ipoib_dev_priv *priv = netdev_priv(dev);
142 struct rb_node **n = &priv->multicast_tree.rb_node, *pn = NULL;
145 struct ipoib_mcast *tmcast;
149 tmcast = rb_entry(pn, struct ipoib_mcast, rb_node);
151 ret = memcmp(mcast->mcmember.mgid.raw, tmcast->mcmember.mgid.raw,
152 sizeof (union ib_gid));
161 rb_link_node(&mcast->rb_node, pn, n);
162 rb_insert_color(&mcast->rb_node, &priv->multicast_tree);
167 static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
168 struct ib_sa_mcmember_rec *mcmember)
170 struct net_device *dev = mcast->dev;
171 struct ipoib_dev_priv *priv = netdev_priv(dev);
176 mcast->mcmember = *mcmember;
178 /* Set the multicast MTU and cached Q_Key before we attach if it's
179 * the broadcast group.
181 if (!memcmp(mcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
182 sizeof (union ib_gid))) {
183 spin_lock_irq(&priv->lock);
184 if (!priv->broadcast) {
185 spin_unlock_irq(&priv->lock);
188 priv->mcast_mtu = IPOIB_UD_MTU(ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu));
189 priv->qkey = be32_to_cpu(priv->broadcast->mcmember.qkey);
190 spin_unlock_irq(&priv->lock);
191 priv->tx_wr.wr.ud.remote_qkey = priv->qkey;
195 if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
196 if (test_and_set_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
197 ipoib_warn(priv, "multicast group %pI6 already attached\n",
198 mcast->mcmember.mgid.raw);
203 ret = ipoib_mcast_attach(dev, be16_to_cpu(mcast->mcmember.mlid),
204 &mcast->mcmember.mgid, set_qkey);
206 ipoib_warn(priv, "couldn't attach QP to multicast group %pI6\n",
207 mcast->mcmember.mgid.raw);
209 clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags);
215 struct ib_ah_attr av = {
216 .dlid = be16_to_cpu(mcast->mcmember.mlid),
217 .port_num = priv->port,
218 .sl = mcast->mcmember.sl,
219 .ah_flags = IB_AH_GRH,
220 .static_rate = mcast->mcmember.rate,
222 .flow_label = be32_to_cpu(mcast->mcmember.flow_label),
223 .hop_limit = mcast->mcmember.hop_limit,
225 .traffic_class = mcast->mcmember.traffic_class
228 av.grh.dgid = mcast->mcmember.mgid;
230 ah = ipoib_create_ah(dev, priv->pd, &av);
232 ipoib_warn(priv, "ib_address_create failed %ld\n",
234 /* use original error */
237 spin_lock_irq(&priv->lock);
239 spin_unlock_irq(&priv->lock);
241 ipoib_dbg_mcast(priv, "MGID %pI6 AV %p, LID 0x%04x, SL %d\n",
242 mcast->mcmember.mgid.raw,
244 be16_to_cpu(mcast->mcmember.mlid),
249 /* actually send any queued packets */
250 netif_tx_lock_bh(dev);
251 while (!skb_queue_empty(&mcast->pkt_queue)) {
252 struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue);
254 netif_tx_unlock_bh(dev);
257 if (dev_queue_xmit(skb))
258 ipoib_warn(priv, "dev_queue_xmit failed to requeue packet\n");
260 netif_tx_lock_bh(dev);
262 netif_tx_unlock_bh(dev);
268 ipoib_mcast_sendonly_join_complete(int status,
269 struct ib_sa_multicast *multicast)
271 struct ipoib_mcast *mcast = multicast->context;
272 struct net_device *dev = mcast->dev;
275 * We have to take the mutex to force mcast_sendonly_join to
276 * return from ib_sa_multicast_join and set mcast->mc to a
277 * valid value. Otherwise we were racing with ourselves in
278 * that we might fail here, but get a valid return from
279 * ib_sa_multicast_join after we had cleared mcast->mc here,
280 * resulting in mis-matched joins and leaves and a deadlock
282 mutex_lock(&mcast_mutex);
284 /* We trap for port events ourselves. */
285 if (status == -ENETRESET)
289 status = ipoib_mcast_join_finish(mcast, &multicast->rec);
292 if (mcast->logcount++ < 20)
293 ipoib_dbg_mcast(netdev_priv(dev), "sendonly multicast "
294 "join failed for %pI6, status %d\n",
295 mcast->mcmember.mgid.raw, status);
297 /* Flush out any queued packets */
298 netif_tx_lock_bh(dev);
299 while (!skb_queue_empty(&mcast->pkt_queue)) {
300 ++dev->stats.tx_dropped;
301 dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
303 netif_tx_unlock_bh(dev);
306 clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
309 complete(&mcast->done);
310 if (status == -ENETRESET)
312 mutex_unlock(&mcast_mutex);
316 static int ipoib_mcast_sendonly_join(struct ipoib_mcast *mcast)
318 struct net_device *dev = mcast->dev;
319 struct ipoib_dev_priv *priv = netdev_priv(dev);
320 struct ib_sa_mcmember_rec rec = {
321 #if 0 /* Some SMs don't support send-only yet */
329 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
330 ipoib_dbg_mcast(priv, "device shutting down, no sendonly "
331 "multicast joins\n");
335 if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) {
336 ipoib_dbg_mcast(priv, "multicast entry busy, skipping "
341 rec.mgid = mcast->mcmember.mgid;
342 rec.port_gid = priv->local_gid;
343 rec.pkey = cpu_to_be16(priv->pkey);
345 mutex_lock(&mcast_mutex);
346 init_completion(&mcast->done);
347 set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
348 mcast->mc = ib_sa_join_multicast(&ipoib_sa_client, priv->ca,
350 IB_SA_MCMEMBER_REC_MGID |
351 IB_SA_MCMEMBER_REC_PORT_GID |
352 IB_SA_MCMEMBER_REC_PKEY |
353 IB_SA_MCMEMBER_REC_JOIN_STATE,
355 ipoib_mcast_sendonly_join_complete,
357 if (IS_ERR(mcast->mc)) {
358 ret = PTR_ERR(mcast->mc);
359 clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
360 complete(&mcast->done);
361 ipoib_warn(priv, "ib_sa_join_multicast for sendonly join "
362 "failed (ret = %d)\n", ret);
364 ipoib_dbg_mcast(priv, "no multicast record for %pI6, starting "
365 "sendonly join\n", mcast->mcmember.mgid.raw);
367 mutex_unlock(&mcast_mutex);
372 void ipoib_mcast_carrier_on_task(struct work_struct *work)
374 struct ipoib_dev_priv *priv = container_of(work, struct ipoib_dev_priv,
376 struct ib_port_attr attr;
378 if (ib_query_port(priv->ca, priv->port, &attr) ||
379 attr.state != IB_PORT_ACTIVE) {
380 ipoib_dbg(priv, "Keeping carrier off until IB port is active\n");
385 * Take rtnl_lock to avoid racing with ipoib_stop() and
386 * turning the carrier back on while a device is being
387 * removed. However, ipoib_stop() will attempt to flush
388 * the workqueue while holding the rtnl lock, so loop
389 * on trylock until either we get the lock or we see
390 * FLAG_ADMIN_UP go away as that signals that we are bailing
391 * and can safely ignore the carrier on work.
393 while (!rtnl_trylock()) {
394 if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
399 if (!ipoib_cm_admin_enabled(priv->dev))
400 dev_set_mtu(priv->dev, min(priv->mcast_mtu, priv->admin_mtu));
401 netif_carrier_on(priv->dev);
405 static int ipoib_mcast_join_complete(int status,
406 struct ib_sa_multicast *multicast)
408 struct ipoib_mcast *mcast = multicast->context;
409 struct net_device *dev = mcast->dev;
410 struct ipoib_dev_priv *priv = netdev_priv(dev);
412 ipoib_dbg_mcast(priv, "join completion for %pI6 (status %d)\n",
413 mcast->mcmember.mgid.raw, status);
416 * We have to take the mutex to force mcast_join to
417 * return from ib_sa_multicast_join and set mcast->mc to a
418 * valid value. Otherwise we were racing with ourselves in
419 * that we might fail here, but get a valid return from
420 * ib_sa_multicast_join after we had cleared mcast->mc here,
421 * resulting in mis-matched joins and leaves and a deadlock
423 mutex_lock(&mcast_mutex);
425 /* We trap for port events ourselves. */
426 if (status == -ENETRESET)
430 status = ipoib_mcast_join_finish(mcast, &multicast->rec);
434 if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
435 queue_delayed_work(priv->wq, &priv->mcast_task, 0);
438 * Defer carrier on work to priv->wq to avoid a
439 * deadlock on rtnl_lock here.
441 if (mcast == priv->broadcast)
442 queue_work(priv->wq, &priv->carrier_on_task);
444 if (mcast->logcount++ < 20) {
445 if (status == -ETIMEDOUT || status == -EAGAIN) {
446 ipoib_dbg_mcast(priv, "multicast join failed for %pI6, status %d\n",
447 mcast->mcmember.mgid.raw, status);
449 ipoib_warn(priv, "multicast join failed for %pI6, status %d\n",
450 mcast->mcmember.mgid.raw, status);
455 if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
456 mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
459 spin_lock_irq(&priv->lock);
460 clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
463 complete(&mcast->done);
464 if (status == -ENETRESET)
466 if (status && test_bit(IPOIB_MCAST_RUN, &priv->flags))
467 queue_delayed_work(priv->wq, &priv->mcast_task,
468 mcast->backoff * HZ);
469 spin_unlock_irq(&priv->lock);
470 mutex_unlock(&mcast_mutex);
475 static void ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast,
478 struct ipoib_dev_priv *priv = netdev_priv(dev);
479 struct ib_sa_mcmember_rec rec = {
482 ib_sa_comp_mask comp_mask;
485 ipoib_dbg_mcast(priv, "joining MGID %pI6\n", mcast->mcmember.mgid.raw);
487 rec.mgid = mcast->mcmember.mgid;
488 rec.port_gid = priv->local_gid;
489 rec.pkey = cpu_to_be16(priv->pkey);
492 IB_SA_MCMEMBER_REC_MGID |
493 IB_SA_MCMEMBER_REC_PORT_GID |
494 IB_SA_MCMEMBER_REC_PKEY |
495 IB_SA_MCMEMBER_REC_JOIN_STATE;
499 IB_SA_MCMEMBER_REC_QKEY |
500 IB_SA_MCMEMBER_REC_MTU_SELECTOR |
501 IB_SA_MCMEMBER_REC_MTU |
502 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS |
503 IB_SA_MCMEMBER_REC_RATE_SELECTOR |
504 IB_SA_MCMEMBER_REC_RATE |
505 IB_SA_MCMEMBER_REC_SL |
506 IB_SA_MCMEMBER_REC_FLOW_LABEL |
507 IB_SA_MCMEMBER_REC_HOP_LIMIT;
509 rec.qkey = priv->broadcast->mcmember.qkey;
510 rec.mtu_selector = IB_SA_EQ;
511 rec.mtu = priv->broadcast->mcmember.mtu;
512 rec.traffic_class = priv->broadcast->mcmember.traffic_class;
513 rec.rate_selector = IB_SA_EQ;
514 rec.rate = priv->broadcast->mcmember.rate;
515 rec.sl = priv->broadcast->mcmember.sl;
516 rec.flow_label = priv->broadcast->mcmember.flow_label;
517 rec.hop_limit = priv->broadcast->mcmember.hop_limit;
520 mutex_lock(&mcast_mutex);
521 init_completion(&mcast->done);
522 set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
523 mcast->mc = ib_sa_join_multicast(&ipoib_sa_client, priv->ca, priv->port,
524 &rec, comp_mask, GFP_KERNEL,
525 ipoib_mcast_join_complete, mcast);
526 if (IS_ERR(mcast->mc)) {
527 clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
528 complete(&mcast->done);
529 ret = PTR_ERR(mcast->mc);
530 ipoib_warn(priv, "ib_sa_join_multicast failed, status %d\n", ret);
533 if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
534 mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
536 if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
537 queue_delayed_work(priv->wq, &priv->mcast_task,
538 mcast->backoff * HZ);
540 mutex_unlock(&mcast_mutex);
543 void ipoib_mcast_join_task(struct work_struct *work)
545 struct ipoib_dev_priv *priv =
546 container_of(work, struct ipoib_dev_priv, mcast_task.work);
547 struct net_device *dev = priv->dev;
548 struct ib_port_attr port_attr;
550 if (!test_bit(IPOIB_MCAST_RUN, &priv->flags))
553 if (ib_query_port(priv->ca, priv->port, &port_attr) ||
554 port_attr.state != IB_PORT_ACTIVE) {
555 ipoib_dbg(priv, "port state is not ACTIVE (state = %d) suspending join task\n",
559 priv->local_lid = port_attr.lid;
561 if (ib_query_gid(priv->ca, priv->port, 0, &priv->local_gid))
562 ipoib_warn(priv, "ib_query_gid() failed\n");
564 memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw, sizeof (union ib_gid));
566 if (!priv->broadcast) {
567 struct ipoib_mcast *broadcast;
569 if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
572 broadcast = ipoib_mcast_alloc(dev, 1);
574 ipoib_warn(priv, "failed to allocate broadcast group\n");
575 mutex_lock(&mcast_mutex);
576 if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
577 queue_delayed_work(priv->wq, &priv->mcast_task,
579 mutex_unlock(&mcast_mutex);
583 spin_lock_irq(&priv->lock);
584 memcpy(broadcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
585 sizeof (union ib_gid));
586 priv->broadcast = broadcast;
588 __ipoib_mcast_add(dev, priv->broadcast);
589 spin_unlock_irq(&priv->lock);
592 if (!test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
593 if (IS_ERR_OR_NULL(priv->broadcast->mc) &&
594 !test_bit(IPOIB_MCAST_FLAG_BUSY, &priv->broadcast->flags))
595 ipoib_mcast_join(dev, priv->broadcast, 0);
600 struct ipoib_mcast *mcast = NULL;
603 * Need the mutex so our flags are consistent, need the
604 * priv->lock so we don't race with list removals in either
605 * mcast_dev_flush or mcast_restart_task
607 mutex_lock(&mcast_mutex);
608 spin_lock_irq(&priv->lock);
609 list_for_each_entry(mcast, &priv->multicast_list, list) {
610 if (IS_ERR_OR_NULL(mcast->mc) &&
611 !test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags) &&
612 !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
613 /* Found the next unjoined group */
617 spin_unlock_irq(&priv->lock);
618 mutex_unlock(&mcast_mutex);
620 if (&mcast->list == &priv->multicast_list) {
625 if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags))
626 ipoib_mcast_sendonly_join(mcast);
628 ipoib_mcast_join(dev, mcast, 1);
632 ipoib_dbg_mcast(priv, "successfully joined all multicast groups\n");
634 clear_bit(IPOIB_MCAST_RUN, &priv->flags);
637 int ipoib_mcast_start_thread(struct net_device *dev)
639 struct ipoib_dev_priv *priv = netdev_priv(dev);
641 ipoib_dbg_mcast(priv, "starting multicast thread\n");
643 mutex_lock(&mcast_mutex);
644 if (!test_and_set_bit(IPOIB_MCAST_RUN, &priv->flags))
645 queue_delayed_work(priv->wq, &priv->mcast_task, 0);
646 mutex_unlock(&mcast_mutex);
651 int ipoib_mcast_stop_thread(struct net_device *dev, int flush)
653 struct ipoib_dev_priv *priv = netdev_priv(dev);
655 ipoib_dbg_mcast(priv, "stopping multicast thread\n");
657 mutex_lock(&mcast_mutex);
658 clear_bit(IPOIB_MCAST_RUN, &priv->flags);
659 cancel_delayed_work(&priv->mcast_task);
660 mutex_unlock(&mcast_mutex);
663 flush_workqueue(priv->wq);
668 static int ipoib_mcast_leave(struct net_device *dev, struct ipoib_mcast *mcast)
670 struct ipoib_dev_priv *priv = netdev_priv(dev);
673 if (test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
674 ipoib_warn(priv, "ipoib_mcast_leave on an in-flight join\n");
676 if (!IS_ERR_OR_NULL(mcast->mc))
677 ib_sa_free_multicast(mcast->mc);
679 if (test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
680 ipoib_dbg_mcast(priv, "leaving MGID %pI6\n",
681 mcast->mcmember.mgid.raw);
683 /* Remove ourselves from the multicast group */
684 ret = ib_detach_mcast(priv->qp, &mcast->mcmember.mgid,
685 be16_to_cpu(mcast->mcmember.mlid));
687 ipoib_warn(priv, "ib_detach_mcast failed (result = %d)\n", ret);
693 void ipoib_mcast_send(struct net_device *dev, u8 *daddr, struct sk_buff *skb)
695 struct ipoib_dev_priv *priv = netdev_priv(dev);
696 struct ipoib_mcast *mcast;
698 void *mgid = daddr + 4;
700 spin_lock_irqsave(&priv->lock, flags);
702 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags) ||
704 !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
705 ++dev->stats.tx_dropped;
706 dev_kfree_skb_any(skb);
710 mcast = __ipoib_mcast_find(dev, mgid);
712 /* Let's create a new send only group now */
713 ipoib_dbg_mcast(priv, "setting up send only multicast group for %pI6\n",
716 mcast = ipoib_mcast_alloc(dev, 0);
718 ipoib_warn(priv, "unable to allocate memory for "
719 "multicast structure\n");
720 ++dev->stats.tx_dropped;
721 dev_kfree_skb_any(skb);
725 set_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags);
726 memcpy(mcast->mcmember.mgid.raw, mgid, sizeof (union ib_gid));
727 __ipoib_mcast_add(dev, mcast);
728 list_add_tail(&mcast->list, &priv->multicast_list);
729 if (!test_and_set_bit(IPOIB_MCAST_RUN, &priv->flags))
730 queue_delayed_work(priv->wq, &priv->mcast_task, 0);
734 if (skb_queue_len(&mcast->pkt_queue) < IPOIB_MAX_MCAST_QUEUE)
735 skb_queue_tail(&mcast->pkt_queue, skb);
737 ++dev->stats.tx_dropped;
738 dev_kfree_skb_any(skb);
741 if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
742 ipoib_dbg_mcast(priv, "no address vector, "
743 "but multicast join already started\n");
746 * If lookup completes between here and out:, don't
747 * want to send packet twice.
753 if (mcast && mcast->ah) {
754 struct ipoib_neigh *neigh;
756 spin_unlock_irqrestore(&priv->lock, flags);
757 neigh = ipoib_neigh_get(dev, daddr);
758 spin_lock_irqsave(&priv->lock, flags);
760 neigh = ipoib_neigh_alloc(daddr, dev);
762 kref_get(&mcast->ah->ref);
763 neigh->ah = mcast->ah;
764 list_add_tail(&neigh->list, &mcast->neigh_list);
767 spin_unlock_irqrestore(&priv->lock, flags);
768 ipoib_send(dev, skb, mcast->ah, IB_MULTICAST_QPN);
770 ipoib_neigh_put(neigh);
775 spin_unlock_irqrestore(&priv->lock, flags);
778 void ipoib_mcast_dev_flush(struct net_device *dev)
780 struct ipoib_dev_priv *priv = netdev_priv(dev);
781 LIST_HEAD(remove_list);
782 struct ipoib_mcast *mcast, *tmcast;
785 ipoib_dbg_mcast(priv, "flushing multicast list\n");
787 spin_lock_irqsave(&priv->lock, flags);
789 list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
790 list_del(&mcast->list);
791 rb_erase(&mcast->rb_node, &priv->multicast_tree);
792 list_add_tail(&mcast->list, &remove_list);
795 if (priv->broadcast) {
796 rb_erase(&priv->broadcast->rb_node, &priv->multicast_tree);
797 list_add_tail(&priv->broadcast->list, &remove_list);
798 priv->broadcast = NULL;
801 spin_unlock_irqrestore(&priv->lock, flags);
804 * make sure the in-flight joins have finished before we attempt
807 list_for_each_entry_safe(mcast, tmcast, &remove_list, list)
808 if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
809 wait_for_completion(&mcast->done);
811 list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
812 ipoib_mcast_leave(dev, mcast);
813 ipoib_mcast_free(mcast);
817 static int ipoib_mcast_addr_is_valid(const u8 *addr, const u8 *broadcast)
819 /* reserved QPN, prefix, scope */
820 if (memcmp(addr, broadcast, 6))
822 /* signature lower, pkey */
823 if (memcmp(addr + 7, broadcast + 7, 3))
828 void ipoib_mcast_restart_task(struct work_struct *work)
830 struct ipoib_dev_priv *priv =
831 container_of(work, struct ipoib_dev_priv, restart_task);
832 struct net_device *dev = priv->dev;
833 struct netdev_hw_addr *ha;
834 struct ipoib_mcast *mcast, *tmcast;
835 LIST_HEAD(remove_list);
837 struct ib_sa_mcmember_rec rec;
839 ipoib_dbg_mcast(priv, "restarting multicast task\n");
841 ipoib_mcast_stop_thread(dev, 0);
843 local_irq_save(flags);
844 netif_addr_lock(dev);
845 spin_lock(&priv->lock);
848 * Unfortunately, the networking core only gives us a list of all of
849 * the multicast hardware addresses. We need to figure out which ones
850 * are new and which ones have been removed
853 /* Clear out the found flag */
854 list_for_each_entry(mcast, &priv->multicast_list, list)
855 clear_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
857 /* Mark all of the entries that are found or don't exist */
858 netdev_for_each_mc_addr(ha, dev) {
861 if (!ipoib_mcast_addr_is_valid(ha->addr, dev->broadcast))
864 memcpy(mgid.raw, ha->addr + 4, sizeof mgid);
866 mcast = __ipoib_mcast_find(dev, &mgid);
867 if (!mcast || test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
868 struct ipoib_mcast *nmcast;
870 /* ignore group which is directly joined by userspace */
871 if (test_bit(IPOIB_FLAG_UMCAST, &priv->flags) &&
872 !ib_sa_get_mcmember_rec(priv->ca, priv->port, &mgid, &rec)) {
873 ipoib_dbg_mcast(priv, "ignoring multicast entry for mgid %pI6\n",
878 /* Not found or send-only group, let's add a new entry */
879 ipoib_dbg_mcast(priv, "adding multicast entry for mgid %pI6\n",
882 nmcast = ipoib_mcast_alloc(dev, 0);
884 ipoib_warn(priv, "unable to allocate memory for multicast structure\n");
888 set_bit(IPOIB_MCAST_FLAG_FOUND, &nmcast->flags);
890 nmcast->mcmember.mgid = mgid;
893 /* Destroy the send only entry */
894 list_move_tail(&mcast->list, &remove_list);
896 rb_replace_node(&mcast->rb_node,
898 &priv->multicast_tree);
900 __ipoib_mcast_add(dev, nmcast);
902 list_add_tail(&nmcast->list, &priv->multicast_list);
906 set_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
909 /* Remove all of the entries don't exist anymore */
910 list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
911 if (!test_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags) &&
912 !test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
913 ipoib_dbg_mcast(priv, "deleting multicast group %pI6\n",
914 mcast->mcmember.mgid.raw);
916 rb_erase(&mcast->rb_node, &priv->multicast_tree);
918 /* Move to the remove list */
919 list_move_tail(&mcast->list, &remove_list);
923 spin_unlock(&priv->lock);
924 netif_addr_unlock(dev);
925 local_irq_restore(flags);
928 * make sure the in-flight joins have finished before we attempt
931 list_for_each_entry_safe(mcast, tmcast, &remove_list, list)
932 if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
933 wait_for_completion(&mcast->done);
936 * We have to cancel outside of the spinlock, but we have to
937 * take the rtnl lock or else we race with the removal of
938 * entries from the remove list in mcast_dev_flush as part
939 * of ipoib_stop() which will call mcast_stop_thread with
940 * flush == 1 while holding the rtnl lock, and the
941 * flush_workqueue won't complete until this restart_mcast_task
942 * completes. So do like the carrier on task and attempt to
943 * take the rtnl lock, but if we can't before the ADMIN_UP flag
944 * goes away, then just return and know that the remove list will
945 * get flushed later by mcast_stop_thread.
947 while (!rtnl_trylock()) {
948 if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
953 list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
954 ipoib_mcast_leave(mcast->dev, mcast);
955 ipoib_mcast_free(mcast);
957 ipoib_mcast_start_thread(dev);
961 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
963 struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev)
965 struct ipoib_mcast_iter *iter;
967 iter = kmalloc(sizeof *iter, GFP_KERNEL);
972 memset(iter->mgid.raw, 0, 16);
974 if (ipoib_mcast_iter_next(iter)) {
982 int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter)
984 struct ipoib_dev_priv *priv = netdev_priv(iter->dev);
986 struct ipoib_mcast *mcast;
989 spin_lock_irq(&priv->lock);
991 n = rb_first(&priv->multicast_tree);
994 mcast = rb_entry(n, struct ipoib_mcast, rb_node);
996 if (memcmp(iter->mgid.raw, mcast->mcmember.mgid.raw,
997 sizeof (union ib_gid)) < 0) {
998 iter->mgid = mcast->mcmember.mgid;
999 iter->created = mcast->created;
1000 iter->queuelen = skb_queue_len(&mcast->pkt_queue);
1001 iter->complete = !!mcast->ah;
1002 iter->send_only = !!(mcast->flags & (1 << IPOIB_MCAST_FLAG_SENDONLY));
1012 spin_unlock_irq(&priv->lock);
1017 void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
1019 unsigned long *created,
1020 unsigned int *queuelen,
1021 unsigned int *complete,
1022 unsigned int *send_only)
1025 *created = iter->created;
1026 *queuelen = iter->queuelen;
1027 *complete = iter->complete;
1028 *send_only = iter->send_only;
1031 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */