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
34 * $Id: ipoib_multicast.c 1362 2004-12-18 15:56:29Z roland $
37 #include <linux/skbuff.h>
38 #include <linux/rtnetlink.h>
41 #include <linux/igmp.h>
42 #include <linux/inetdevice.h>
43 #include <linux/delay.h>
44 #include <linux/completion.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 /* Used for all multicast joins (broadcast, IPv4 mcast and IPv6 mcast) */
62 struct ib_sa_mcmember_rec mcmember;
65 struct rb_node rb_node;
66 struct list_head list;
67 struct completion done;
70 struct ib_sa_query *query;
72 unsigned long created;
73 unsigned long backoff;
76 unsigned char logcount;
78 struct list_head neigh_list;
80 struct sk_buff_head pkt_queue;
82 struct net_device *dev;
85 struct ipoib_mcast_iter {
86 struct net_device *dev;
88 unsigned long created;
89 unsigned int queuelen;
90 unsigned int complete;
91 unsigned int send_only;
94 static void ipoib_mcast_free(struct ipoib_mcast *mcast)
96 struct net_device *dev = mcast->dev;
97 struct ipoib_dev_priv *priv = netdev_priv(dev);
98 struct ipoib_neigh *neigh, *tmp;
102 ipoib_dbg_mcast(netdev_priv(dev),
103 "deleting multicast group " IPOIB_GID_FMT "\n",
104 IPOIB_GID_ARG(mcast->mcmember.mgid));
106 spin_lock_irqsave(&priv->lock, flags);
108 list_for_each_entry_safe(neigh, tmp, &mcast->neigh_list, list) {
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.
116 ipoib_put_ah(neigh->ah);
117 ipoib_neigh_free(neigh);
120 spin_unlock_irqrestore(&priv->lock, flags);
123 ipoib_put_ah(mcast->ah);
125 while (!skb_queue_empty(&mcast->pkt_queue)) {
127 dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
130 spin_lock_irqsave(&priv->tx_lock, flags);
131 priv->stats.tx_dropped += tx_dropped;
132 spin_unlock_irqrestore(&priv->tx_lock, flags);
137 static struct ipoib_mcast *ipoib_mcast_alloc(struct net_device *dev,
140 struct ipoib_mcast *mcast;
142 mcast = kzalloc(sizeof *mcast, can_sleep ? GFP_KERNEL : GFP_ATOMIC);
147 mcast->created = jiffies;
150 INIT_LIST_HEAD(&mcast->list);
151 INIT_LIST_HEAD(&mcast->neigh_list);
152 skb_queue_head_init(&mcast->pkt_queue);
157 static struct ipoib_mcast *__ipoib_mcast_find(struct net_device *dev, void *mgid)
159 struct ipoib_dev_priv *priv = netdev_priv(dev);
160 struct rb_node *n = priv->multicast_tree.rb_node;
163 struct ipoib_mcast *mcast;
166 mcast = rb_entry(n, struct ipoib_mcast, rb_node);
168 ret = memcmp(mgid, mcast->mcmember.mgid.raw,
169 sizeof (union ib_gid));
181 static int __ipoib_mcast_add(struct net_device *dev, struct ipoib_mcast *mcast)
183 struct ipoib_dev_priv *priv = netdev_priv(dev);
184 struct rb_node **n = &priv->multicast_tree.rb_node, *pn = NULL;
187 struct ipoib_mcast *tmcast;
191 tmcast = rb_entry(pn, struct ipoib_mcast, rb_node);
193 ret = memcmp(mcast->mcmember.mgid.raw, tmcast->mcmember.mgid.raw,
194 sizeof (union ib_gid));
203 rb_link_node(&mcast->rb_node, pn, n);
204 rb_insert_color(&mcast->rb_node, &priv->multicast_tree);
209 static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
210 struct ib_sa_mcmember_rec *mcmember)
212 struct net_device *dev = mcast->dev;
213 struct ipoib_dev_priv *priv = netdev_priv(dev);
217 mcast->mcmember = *mcmember;
219 /* Set the cached Q_Key before we attach if it's the broadcast group */
220 if (!memcmp(mcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
221 sizeof (union ib_gid))) {
222 priv->qkey = be32_to_cpu(priv->broadcast->mcmember.qkey);
223 priv->tx_wr.wr.ud.remote_qkey = priv->qkey;
226 if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
227 if (test_and_set_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
228 ipoib_warn(priv, "multicast group " IPOIB_GID_FMT
229 " already attached\n",
230 IPOIB_GID_ARG(mcast->mcmember.mgid));
235 ret = ipoib_mcast_attach(dev, be16_to_cpu(mcast->mcmember.mlid),
236 &mcast->mcmember.mgid);
238 ipoib_warn(priv, "couldn't attach QP to multicast group "
240 IPOIB_GID_ARG(mcast->mcmember.mgid));
242 clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags);
248 struct ib_ah_attr av = {
249 .dlid = be16_to_cpu(mcast->mcmember.mlid),
250 .port_num = priv->port,
251 .sl = mcast->mcmember.sl,
252 .ah_flags = IB_AH_GRH,
253 .static_rate = mcast->mcmember.rate,
255 .flow_label = be32_to_cpu(mcast->mcmember.flow_label),
256 .hop_limit = mcast->mcmember.hop_limit,
258 .traffic_class = mcast->mcmember.traffic_class
261 av.grh.dgid = mcast->mcmember.mgid;
263 ah = ipoib_create_ah(dev, priv->pd, &av);
265 ipoib_warn(priv, "ib_address_create failed\n");
267 spin_lock_irq(&priv->lock);
269 spin_unlock_irq(&priv->lock);
271 ipoib_dbg_mcast(priv, "MGID " IPOIB_GID_FMT
272 " AV %p, LID 0x%04x, SL %d\n",
273 IPOIB_GID_ARG(mcast->mcmember.mgid),
275 be16_to_cpu(mcast->mcmember.mlid),
280 /* actually send any queued packets */
281 spin_lock_irq(&priv->tx_lock);
282 while (!skb_queue_empty(&mcast->pkt_queue)) {
283 struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue);
284 spin_unlock_irq(&priv->tx_lock);
288 if (!skb->dst || !skb->dst->neighbour) {
289 /* put pseudoheader back on for next time */
290 skb_push(skb, sizeof (struct ipoib_pseudoheader));
293 if (dev_queue_xmit(skb))
294 ipoib_warn(priv, "dev_queue_xmit failed to requeue packet\n");
295 spin_lock_irq(&priv->tx_lock);
297 spin_unlock_irq(&priv->tx_lock);
303 ipoib_mcast_sendonly_join_complete(int status,
304 struct ib_sa_mcmember_rec *mcmember,
307 struct ipoib_mcast *mcast = mcast_ptr;
308 struct net_device *dev = mcast->dev;
309 struct ipoib_dev_priv *priv = netdev_priv(dev);
312 ipoib_mcast_join_finish(mcast, mcmember);
314 if (mcast->logcount++ < 20)
315 ipoib_dbg_mcast(netdev_priv(dev), "multicast join failed for "
316 IPOIB_GID_FMT ", status %d\n",
317 IPOIB_GID_ARG(mcast->mcmember.mgid), status);
319 /* Flush out any queued packets */
320 spin_lock_irq(&priv->tx_lock);
321 while (!skb_queue_empty(&mcast->pkt_queue)) {
322 ++priv->stats.tx_dropped;
323 dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
325 spin_unlock_irq(&priv->tx_lock);
327 /* Clear the busy flag so we try again */
328 clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
332 complete(&mcast->done);
335 static int ipoib_mcast_sendonly_join(struct ipoib_mcast *mcast)
337 struct net_device *dev = mcast->dev;
338 struct ipoib_dev_priv *priv = netdev_priv(dev);
339 struct ib_sa_mcmember_rec rec = {
340 #if 0 /* Some SMs don't support send-only yet */
348 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
349 ipoib_dbg_mcast(priv, "device shutting down, no multicast joins\n");
353 if (test_and_set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) {
354 ipoib_dbg_mcast(priv, "multicast entry busy, skipping\n");
358 rec.mgid = mcast->mcmember.mgid;
359 rec.port_gid = priv->local_gid;
360 rec.pkey = cpu_to_be16(priv->pkey);
362 init_completion(&mcast->done);
364 ret = ib_sa_mcmember_rec_set(priv->ca, priv->port, &rec,
365 IB_SA_MCMEMBER_REC_MGID |
366 IB_SA_MCMEMBER_REC_PORT_GID |
367 IB_SA_MCMEMBER_REC_PKEY |
368 IB_SA_MCMEMBER_REC_JOIN_STATE,
370 ipoib_mcast_sendonly_join_complete,
371 mcast, &mcast->query);
373 ipoib_warn(priv, "ib_sa_mcmember_rec_set failed (ret = %d)\n",
376 ipoib_dbg_mcast(priv, "no multicast record for " IPOIB_GID_FMT
378 IPOIB_GID_ARG(mcast->mcmember.mgid));
380 mcast->query_id = ret;
386 static void ipoib_mcast_join_complete(int status,
387 struct ib_sa_mcmember_rec *mcmember,
390 struct ipoib_mcast *mcast = mcast_ptr;
391 struct net_device *dev = mcast->dev;
392 struct ipoib_dev_priv *priv = netdev_priv(dev);
394 ipoib_dbg_mcast(priv, "join completion for " IPOIB_GID_FMT
396 IPOIB_GID_ARG(mcast->mcmember.mgid), status);
398 if (!status && !ipoib_mcast_join_finish(mcast, mcmember)) {
400 mutex_lock(&mcast_mutex);
401 if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
402 queue_work(ipoib_workqueue, &priv->mcast_task);
403 mutex_unlock(&mcast_mutex);
404 complete(&mcast->done);
408 if (status == -EINTR) {
409 complete(&mcast->done);
413 if (status && mcast->logcount++ < 20) {
414 if (status == -ETIMEDOUT || status == -EINTR) {
415 ipoib_dbg_mcast(priv, "multicast join failed for " IPOIB_GID_FMT
417 IPOIB_GID_ARG(mcast->mcmember.mgid),
420 ipoib_warn(priv, "multicast join failed for "
421 IPOIB_GID_FMT ", status %d\n",
422 IPOIB_GID_ARG(mcast->mcmember.mgid),
428 if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
429 mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
431 mutex_lock(&mcast_mutex);
433 spin_lock_irq(&priv->lock);
436 if (test_bit(IPOIB_MCAST_RUN, &priv->flags)) {
437 if (status == -ETIMEDOUT)
438 queue_work(ipoib_workqueue, &priv->mcast_task);
440 queue_delayed_work(ipoib_workqueue, &priv->mcast_task,
441 mcast->backoff * HZ);
443 complete(&mcast->done);
444 spin_unlock_irq(&priv->lock);
445 mutex_unlock(&mcast_mutex);
450 static void ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast,
453 struct ipoib_dev_priv *priv = netdev_priv(dev);
454 struct ib_sa_mcmember_rec rec = {
457 ib_sa_comp_mask comp_mask;
460 ipoib_dbg_mcast(priv, "joining MGID " IPOIB_GID_FMT "\n",
461 IPOIB_GID_ARG(mcast->mcmember.mgid));
463 rec.mgid = mcast->mcmember.mgid;
464 rec.port_gid = priv->local_gid;
465 rec.pkey = cpu_to_be16(priv->pkey);
468 IB_SA_MCMEMBER_REC_MGID |
469 IB_SA_MCMEMBER_REC_PORT_GID |
470 IB_SA_MCMEMBER_REC_PKEY |
471 IB_SA_MCMEMBER_REC_JOIN_STATE;
475 IB_SA_MCMEMBER_REC_QKEY |
476 IB_SA_MCMEMBER_REC_SL |
477 IB_SA_MCMEMBER_REC_FLOW_LABEL |
478 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
480 rec.qkey = priv->broadcast->mcmember.qkey;
481 rec.sl = priv->broadcast->mcmember.sl;
482 rec.flow_label = priv->broadcast->mcmember.flow_label;
483 rec.traffic_class = priv->broadcast->mcmember.traffic_class;
486 init_completion(&mcast->done);
488 ret = ib_sa_mcmember_rec_set(priv->ca, priv->port, &rec, comp_mask,
489 mcast->backoff * 1000, GFP_ATOMIC,
490 ipoib_mcast_join_complete,
491 mcast, &mcast->query);
494 ipoib_warn(priv, "ib_sa_mcmember_rec_set failed, status %d\n", ret);
497 if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
498 mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
500 mutex_lock(&mcast_mutex);
501 if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
502 queue_delayed_work(ipoib_workqueue,
504 mcast->backoff * HZ);
505 mutex_unlock(&mcast_mutex);
507 mcast->query_id = ret;
510 void ipoib_mcast_join_task(void *dev_ptr)
512 struct net_device *dev = dev_ptr;
513 struct ipoib_dev_priv *priv = netdev_priv(dev);
515 if (!test_bit(IPOIB_MCAST_RUN, &priv->flags))
518 if (ib_query_gid(priv->ca, priv->port, 0, &priv->local_gid))
519 ipoib_warn(priv, "ib_gid_entry_get() failed\n");
521 memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw, sizeof (union ib_gid));
524 struct ib_port_attr attr;
526 if (!ib_query_port(priv->ca, priv->port, &attr)) {
527 priv->local_lid = attr.lid;
528 priv->local_rate = attr.active_speed *
529 ib_width_enum_to_int(attr.active_width);
531 ipoib_warn(priv, "ib_query_port failed\n");
534 if (!priv->broadcast) {
535 struct ipoib_mcast *broadcast;
537 broadcast = ipoib_mcast_alloc(dev, 1);
539 ipoib_warn(priv, "failed to allocate broadcast group\n");
540 mutex_lock(&mcast_mutex);
541 if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
542 queue_delayed_work(ipoib_workqueue,
543 &priv->mcast_task, HZ);
544 mutex_unlock(&mcast_mutex);
548 spin_lock_irq(&priv->lock);
549 memcpy(broadcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
550 sizeof (union ib_gid));
551 priv->broadcast = broadcast;
553 __ipoib_mcast_add(dev, priv->broadcast);
554 spin_unlock_irq(&priv->lock);
557 if (!test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
558 ipoib_mcast_join(dev, priv->broadcast, 0);
563 struct ipoib_mcast *mcast = NULL;
565 spin_lock_irq(&priv->lock);
566 list_for_each_entry(mcast, &priv->multicast_list, list) {
567 if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)
568 && !test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)
569 && !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
570 /* Found the next unjoined group */
574 spin_unlock_irq(&priv->lock);
576 if (&mcast->list == &priv->multicast_list) {
581 ipoib_mcast_join(dev, mcast, 1);
585 priv->mcast_mtu = ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu) -
587 dev->mtu = min(priv->mcast_mtu, priv->admin_mtu);
589 ipoib_dbg_mcast(priv, "successfully joined all multicast groups\n");
591 clear_bit(IPOIB_MCAST_RUN, &priv->flags);
592 netif_carrier_on(dev);
595 int ipoib_mcast_start_thread(struct net_device *dev)
597 struct ipoib_dev_priv *priv = netdev_priv(dev);
599 ipoib_dbg_mcast(priv, "starting multicast thread\n");
601 mutex_lock(&mcast_mutex);
602 if (!test_and_set_bit(IPOIB_MCAST_RUN, &priv->flags))
603 queue_work(ipoib_workqueue, &priv->mcast_task);
604 mutex_unlock(&mcast_mutex);
606 spin_lock_irq(&priv->lock);
607 set_bit(IPOIB_MCAST_STARTED, &priv->flags);
608 spin_unlock_irq(&priv->lock);
613 static void wait_for_mcast_join(struct ipoib_dev_priv *priv,
614 struct ipoib_mcast *mcast)
616 spin_lock_irq(&priv->lock);
617 if (mcast && mcast->query) {
618 ib_sa_cancel_query(mcast->query_id, mcast->query);
620 spin_unlock_irq(&priv->lock);
621 ipoib_dbg_mcast(priv, "waiting for MGID " IPOIB_GID_FMT "\n",
622 IPOIB_GID_ARG(mcast->mcmember.mgid));
623 wait_for_completion(&mcast->done);
626 spin_unlock_irq(&priv->lock);
629 int ipoib_mcast_stop_thread(struct net_device *dev, int flush)
631 struct ipoib_dev_priv *priv = netdev_priv(dev);
632 struct ipoib_mcast *mcast;
634 ipoib_dbg_mcast(priv, "stopping multicast thread\n");
636 spin_lock_irq(&priv->lock);
637 clear_bit(IPOIB_MCAST_STARTED, &priv->flags);
638 spin_unlock_irq(&priv->lock);
640 mutex_lock(&mcast_mutex);
641 clear_bit(IPOIB_MCAST_RUN, &priv->flags);
642 cancel_delayed_work(&priv->mcast_task);
643 mutex_unlock(&mcast_mutex);
646 flush_workqueue(ipoib_workqueue);
648 wait_for_mcast_join(priv, priv->broadcast);
650 list_for_each_entry(mcast, &priv->multicast_list, list)
651 wait_for_mcast_join(priv, mcast);
656 static int ipoib_mcast_leave(struct net_device *dev, struct ipoib_mcast *mcast)
658 struct ipoib_dev_priv *priv = netdev_priv(dev);
659 struct ib_sa_mcmember_rec rec = {
664 if (!test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags))
667 ipoib_dbg_mcast(priv, "leaving MGID " IPOIB_GID_FMT "\n",
668 IPOIB_GID_ARG(mcast->mcmember.mgid));
670 rec.mgid = mcast->mcmember.mgid;
671 rec.port_gid = priv->local_gid;
672 rec.pkey = cpu_to_be16(priv->pkey);
674 /* Remove ourselves from the multicast group */
675 ret = ipoib_mcast_detach(dev, be16_to_cpu(mcast->mcmember.mlid),
676 &mcast->mcmember.mgid);
678 ipoib_warn(priv, "ipoib_mcast_detach failed (result = %d)\n", ret);
681 * Just make one shot at leaving and don't wait for a reply;
682 * if we fail, too bad.
684 ret = ib_sa_mcmember_rec_delete(priv->ca, priv->port, &rec,
685 IB_SA_MCMEMBER_REC_MGID |
686 IB_SA_MCMEMBER_REC_PORT_GID |
687 IB_SA_MCMEMBER_REC_PKEY |
688 IB_SA_MCMEMBER_REC_JOIN_STATE,
690 mcast, &mcast->query);
692 ipoib_warn(priv, "ib_sa_mcmember_rec_delete failed "
693 "for leave (result = %d)\n", ret);
698 void ipoib_mcast_send(struct net_device *dev, void *mgid, struct sk_buff *skb)
700 struct ipoib_dev_priv *priv = netdev_priv(dev);
701 struct ipoib_mcast *mcast;
704 * We can only be called from ipoib_start_xmit, so we're
705 * inside tx_lock -- no need to save/restore flags.
707 spin_lock(&priv->lock);
709 if (!test_bit(IPOIB_MCAST_STARTED, &priv->flags) ||
711 !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
712 ++priv->stats.tx_dropped;
713 dev_kfree_skb_any(skb);
717 mcast = __ipoib_mcast_find(dev, mgid);
719 /* Let's create a new send only group now */
720 ipoib_dbg_mcast(priv, "setting up send only multicast group for "
721 IPOIB_GID_FMT "\n", IPOIB_GID_RAW_ARG(mgid));
723 mcast = ipoib_mcast_alloc(dev, 0);
725 ipoib_warn(priv, "unable to allocate memory for "
726 "multicast structure\n");
727 ++priv->stats.tx_dropped;
728 dev_kfree_skb_any(skb);
732 set_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags);
733 memcpy(mcast->mcmember.mgid.raw, mgid, sizeof (union ib_gid));
734 __ipoib_mcast_add(dev, mcast);
735 list_add_tail(&mcast->list, &priv->multicast_list);
739 if (skb_queue_len(&mcast->pkt_queue) < IPOIB_MAX_MCAST_QUEUE)
740 skb_queue_tail(&mcast->pkt_queue, skb);
742 ++priv->stats.tx_dropped;
743 dev_kfree_skb_any(skb);
747 ipoib_dbg_mcast(priv, "no address vector, "
748 "but multicast join already started\n");
749 else if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags))
750 ipoib_mcast_sendonly_join(mcast);
753 * If lookup completes between here and out:, don't
754 * want to send packet twice.
760 if (mcast && mcast->ah) {
762 skb->dst->neighbour &&
763 !*to_ipoib_neigh(skb->dst->neighbour)) {
764 struct ipoib_neigh *neigh = ipoib_neigh_alloc(skb->dst->neighbour);
767 kref_get(&mcast->ah->ref);
768 neigh->ah = mcast->ah;
769 list_add_tail(&neigh->list, &mcast->neigh_list);
773 ipoib_send(dev, skb, mcast->ah, IB_MULTICAST_QPN);
777 spin_unlock(&priv->lock);
780 void ipoib_mcast_dev_flush(struct net_device *dev)
782 struct ipoib_dev_priv *priv = netdev_priv(dev);
783 LIST_HEAD(remove_list);
784 struct ipoib_mcast *mcast, *tmcast;
787 ipoib_dbg_mcast(priv, "flushing multicast list\n");
789 spin_lock_irqsave(&priv->lock, flags);
791 list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
792 list_del(&mcast->list);
793 rb_erase(&mcast->rb_node, &priv->multicast_tree);
794 list_add_tail(&mcast->list, &remove_list);
797 if (priv->broadcast) {
798 rb_erase(&priv->broadcast->rb_node, &priv->multicast_tree);
799 list_add_tail(&priv->broadcast->list, &remove_list);
800 priv->broadcast = NULL;
803 spin_unlock_irqrestore(&priv->lock, flags);
805 list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
806 ipoib_mcast_leave(dev, mcast);
807 ipoib_mcast_free(mcast);
811 void ipoib_mcast_restart_task(void *dev_ptr)
813 struct net_device *dev = dev_ptr;
814 struct ipoib_dev_priv *priv = netdev_priv(dev);
815 struct dev_mc_list *mclist;
816 struct ipoib_mcast *mcast, *tmcast;
817 LIST_HEAD(remove_list);
820 ipoib_dbg_mcast(priv, "restarting multicast task\n");
822 ipoib_mcast_stop_thread(dev, 0);
824 local_irq_save(flags);
826 spin_lock(&priv->lock);
829 * Unfortunately, the networking core only gives us a list of all of
830 * the multicast hardware addresses. We need to figure out which ones
831 * are new and which ones have been removed
834 /* Clear out the found flag */
835 list_for_each_entry(mcast, &priv->multicast_list, list)
836 clear_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
838 /* Mark all of the entries that are found or don't exist */
839 for (mclist = dev->mc_list; mclist; mclist = mclist->next) {
842 memcpy(mgid.raw, mclist->dmi_addr + 4, sizeof mgid);
844 /* Add in the P_Key */
845 mgid.raw[4] = (priv->pkey >> 8) & 0xff;
846 mgid.raw[5] = priv->pkey & 0xff;
848 mcast = __ipoib_mcast_find(dev, &mgid);
849 if (!mcast || test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
850 struct ipoib_mcast *nmcast;
852 /* Not found or send-only group, let's add a new entry */
853 ipoib_dbg_mcast(priv, "adding multicast entry for mgid "
854 IPOIB_GID_FMT "\n", IPOIB_GID_ARG(mgid));
856 nmcast = ipoib_mcast_alloc(dev, 0);
858 ipoib_warn(priv, "unable to allocate memory for multicast structure\n");
862 set_bit(IPOIB_MCAST_FLAG_FOUND, &nmcast->flags);
864 nmcast->mcmember.mgid = mgid;
867 /* Destroy the send only entry */
868 list_move_tail(&mcast->list, &remove_list);
870 rb_replace_node(&mcast->rb_node,
872 &priv->multicast_tree);
874 __ipoib_mcast_add(dev, nmcast);
876 list_add_tail(&nmcast->list, &priv->multicast_list);
880 set_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
883 /* Remove all of the entries don't exist anymore */
884 list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
885 if (!test_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags) &&
886 !test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
887 ipoib_dbg_mcast(priv, "deleting multicast group " IPOIB_GID_FMT "\n",
888 IPOIB_GID_ARG(mcast->mcmember.mgid));
890 rb_erase(&mcast->rb_node, &priv->multicast_tree);
892 /* Move to the remove list */
893 list_move_tail(&mcast->list, &remove_list);
897 spin_unlock(&priv->lock);
898 netif_tx_unlock(dev);
899 local_irq_restore(flags);
901 /* We have to cancel outside of the spinlock */
902 list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
903 wait_for_mcast_join(priv, mcast);
904 ipoib_mcast_leave(mcast->dev, mcast);
905 ipoib_mcast_free(mcast);
908 if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
909 ipoib_mcast_start_thread(dev);
912 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
914 struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev)
916 struct ipoib_mcast_iter *iter;
918 iter = kmalloc(sizeof *iter, GFP_KERNEL);
923 memset(iter->mgid.raw, 0, 16);
925 if (ipoib_mcast_iter_next(iter)) {
933 int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter)
935 struct ipoib_dev_priv *priv = netdev_priv(iter->dev);
937 struct ipoib_mcast *mcast;
940 spin_lock_irq(&priv->lock);
942 n = rb_first(&priv->multicast_tree);
945 mcast = rb_entry(n, struct ipoib_mcast, rb_node);
947 if (memcmp(iter->mgid.raw, mcast->mcmember.mgid.raw,
948 sizeof (union ib_gid)) < 0) {
949 iter->mgid = mcast->mcmember.mgid;
950 iter->created = mcast->created;
951 iter->queuelen = skb_queue_len(&mcast->pkt_queue);
952 iter->complete = !!mcast->ah;
953 iter->send_only = !!(mcast->flags & (1 << IPOIB_MCAST_FLAG_SENDONLY));
963 spin_unlock_irq(&priv->lock);
968 void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
970 unsigned long *created,
971 unsigned int *queuelen,
972 unsigned int *complete,
973 unsigned int *send_only)
976 *created = iter->created;
977 *queuelen = iter->queuelen;
978 *complete = iter->complete;
979 *send_only = iter->send_only;
982 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */