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 struct ipoib_mcast_iter {
59 struct net_device *dev;
61 unsigned long created;
62 unsigned int queuelen;
63 unsigned int complete;
64 unsigned int send_only;
68 * This should be called with the priv->lock held
70 static void __ipoib_mcast_schedule_join_thread(struct ipoib_dev_priv *priv,
71 struct ipoib_mcast *mcast,
74 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
78 * We will be scheduling *something*, so cancel whatever is
79 * currently scheduled first
81 cancel_delayed_work(&priv->mcast_task);
84 * We had a failure and want to schedule a retry later
87 if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
88 mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
89 mcast->delay_until = jiffies + (mcast->backoff * HZ);
91 * Mark this mcast for its delay, but restart the
92 * task immediately. The join task will make sure to
93 * clear out all entries without delays, and then
94 * schedule itself to run again when the earliest
97 queue_delayed_work(priv->wq, &priv->mcast_task, 0);
100 * Special case of retrying after a failure to
101 * allocate the broadcast multicast group, wait
102 * 1 second and try again
104 queue_delayed_work(priv->wq, &priv->mcast_task, HZ);
106 queue_delayed_work(priv->wq, &priv->mcast_task, 0);
109 void ipoib_mcast_free(struct ipoib_mcast *mcast)
111 struct net_device *dev = mcast->dev;
114 ipoib_dbg_mcast(netdev_priv(dev), "deleting multicast group %pI6\n",
115 mcast->mcmember.mgid.raw);
117 /* remove all neigh connected to this mcast */
118 ipoib_del_neighs_by_gid(dev, mcast->mcmember.mgid.raw);
121 ipoib_put_ah(mcast->ah);
123 while (!skb_queue_empty(&mcast->pkt_queue)) {
125 dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
128 netif_tx_lock_bh(dev);
129 dev->stats.tx_dropped += tx_dropped;
130 netif_tx_unlock_bh(dev);
135 static struct ipoib_mcast *ipoib_mcast_alloc(struct net_device *dev,
138 struct ipoib_mcast *mcast;
140 mcast = kzalloc(sizeof *mcast, can_sleep ? GFP_KERNEL : GFP_ATOMIC);
145 mcast->created = jiffies;
146 mcast->delay_until = jiffies;
149 INIT_LIST_HEAD(&mcast->list);
150 INIT_LIST_HEAD(&mcast->neigh_list);
151 skb_queue_head_init(&mcast->pkt_queue);
156 struct ipoib_mcast *__ipoib_mcast_find(struct net_device *dev, void *mgid)
158 struct ipoib_dev_priv *priv = netdev_priv(dev);
159 struct rb_node *n = priv->multicast_tree.rb_node;
162 struct ipoib_mcast *mcast;
165 mcast = rb_entry(n, struct ipoib_mcast, rb_node);
167 ret = memcmp(mgid, mcast->mcmember.mgid.raw,
168 sizeof (union ib_gid));
180 static int __ipoib_mcast_add(struct net_device *dev, struct ipoib_mcast *mcast)
182 struct ipoib_dev_priv *priv = netdev_priv(dev);
183 struct rb_node **n = &priv->multicast_tree.rb_node, *pn = NULL;
186 struct ipoib_mcast *tmcast;
190 tmcast = rb_entry(pn, struct ipoib_mcast, rb_node);
192 ret = memcmp(mcast->mcmember.mgid.raw, tmcast->mcmember.mgid.raw,
193 sizeof (union ib_gid));
202 rb_link_node(&mcast->rb_node, pn, n);
203 rb_insert_color(&mcast->rb_node, &priv->multicast_tree);
208 static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
209 struct ib_sa_mcmember_rec *mcmember)
211 struct net_device *dev = mcast->dev;
212 struct ipoib_dev_priv *priv = netdev_priv(dev);
217 mcast->mcmember = *mcmember;
219 /* Set the multicast MTU and cached Q_Key before we attach if it's
220 * the broadcast group.
222 if (!memcmp(mcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
223 sizeof (union ib_gid))) {
224 spin_lock_irq(&priv->lock);
225 if (!priv->broadcast) {
226 spin_unlock_irq(&priv->lock);
229 /*update priv member according to the new mcast*/
230 priv->broadcast->mcmember.qkey = mcmember->qkey;
231 priv->broadcast->mcmember.mtu = mcmember->mtu;
232 priv->broadcast->mcmember.traffic_class = mcmember->traffic_class;
233 priv->broadcast->mcmember.rate = mcmember->rate;
234 priv->broadcast->mcmember.sl = mcmember->sl;
235 priv->broadcast->mcmember.flow_label = mcmember->flow_label;
236 priv->broadcast->mcmember.hop_limit = mcmember->hop_limit;
237 /* assume if the admin and the mcast are the same both can be changed */
238 if (priv->mcast_mtu == priv->admin_mtu)
241 IPOIB_UD_MTU(ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu));
244 IPOIB_UD_MTU(ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu));
246 priv->qkey = be32_to_cpu(priv->broadcast->mcmember.qkey);
247 spin_unlock_irq(&priv->lock);
248 priv->tx_wr.remote_qkey = priv->qkey;
252 if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
253 if (test_and_set_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
254 ipoib_warn(priv, "multicast group %pI6 already attached\n",
255 mcast->mcmember.mgid.raw);
260 ret = ipoib_mcast_attach(dev, be16_to_cpu(mcast->mcmember.mlid),
261 &mcast->mcmember.mgid, set_qkey);
263 ipoib_warn(priv, "couldn't attach QP to multicast group %pI6\n",
264 mcast->mcmember.mgid.raw);
266 clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags);
272 struct ib_ah_attr av = {
273 .dlid = be16_to_cpu(mcast->mcmember.mlid),
274 .port_num = priv->port,
275 .sl = mcast->mcmember.sl,
276 .ah_flags = IB_AH_GRH,
277 .static_rate = mcast->mcmember.rate,
279 .flow_label = be32_to_cpu(mcast->mcmember.flow_label),
280 .hop_limit = mcast->mcmember.hop_limit,
282 .traffic_class = mcast->mcmember.traffic_class
285 av.grh.dgid = mcast->mcmember.mgid;
287 ah = ipoib_create_ah(dev, priv->pd, &av);
289 ipoib_warn(priv, "ib_address_create failed %ld\n",
291 /* use original error */
294 spin_lock_irq(&priv->lock);
296 spin_unlock_irq(&priv->lock);
298 ipoib_dbg_mcast(priv, "MGID %pI6 AV %p, LID 0x%04x, SL %d\n",
299 mcast->mcmember.mgid.raw,
301 be16_to_cpu(mcast->mcmember.mlid),
306 /* actually send any queued packets */
307 netif_tx_lock_bh(dev);
308 while (!skb_queue_empty(&mcast->pkt_queue)) {
309 struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue);
311 netif_tx_unlock_bh(dev);
314 if (dev_queue_xmit(skb))
315 ipoib_warn(priv, "dev_queue_xmit failed to requeue packet\n");
317 netif_tx_lock_bh(dev);
319 netif_tx_unlock_bh(dev);
324 void ipoib_mcast_carrier_on_task(struct work_struct *work)
326 struct ipoib_dev_priv *priv = container_of(work, struct ipoib_dev_priv,
328 struct ib_port_attr attr;
330 if (ib_query_port(priv->ca, priv->port, &attr) ||
331 attr.state != IB_PORT_ACTIVE) {
332 ipoib_dbg(priv, "Keeping carrier off until IB port is active\n");
337 * Take rtnl_lock to avoid racing with ipoib_stop() and
338 * turning the carrier back on while a device is being
339 * removed. However, ipoib_stop() will attempt to flush
340 * the workqueue while holding the rtnl lock, so loop
341 * on trylock until either we get the lock or we see
342 * FLAG_OPER_UP go away as that signals that we are bailing
343 * and can safely ignore the carrier on work.
345 while (!rtnl_trylock()) {
346 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
351 if (!ipoib_cm_admin_enabled(priv->dev))
352 dev_set_mtu(priv->dev, min(priv->mcast_mtu, priv->admin_mtu));
353 netif_carrier_on(priv->dev);
357 static int ipoib_mcast_join_complete(int status,
358 struct ib_sa_multicast *multicast)
360 struct ipoib_mcast *mcast = multicast->context;
361 struct net_device *dev = mcast->dev;
362 struct ipoib_dev_priv *priv = netdev_priv(dev);
364 ipoib_dbg_mcast(priv, "%sjoin completion for %pI6 (status %d)\n",
365 test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) ?
367 mcast->mcmember.mgid.raw, status);
369 /* We trap for port events ourselves. */
370 if (status == -ENETRESET) {
376 status = ipoib_mcast_join_finish(mcast, &multicast->rec);
380 mcast->delay_until = jiffies;
383 * Defer carrier on work to priv->wq to avoid a
384 * deadlock on rtnl_lock here. Requeue our multicast
385 * work too, which will end up happening right after
386 * our carrier on task work and will allow us to
387 * send out all of the non-broadcast joins
389 if (mcast == priv->broadcast) {
390 spin_lock_irq(&priv->lock);
391 queue_work(priv->wq, &priv->carrier_on_task);
392 __ipoib_mcast_schedule_join_thread(priv, NULL, 0);
397 test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) &&
400 if (mcast->logcount < 20) {
401 if (status == -ETIMEDOUT || status == -EAGAIN ||
403 ipoib_dbg_mcast(priv, "%smulticast join failed for %pI6, status %d\n",
404 test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) ? "sendonly " : "",
405 mcast->mcmember.mgid.raw, status);
407 ipoib_warn(priv, "%smulticast join failed for %pI6, status %d\n",
408 test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) ? "sendonly " : "",
409 mcast->mcmember.mgid.raw, status);
416 if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) &&
417 mcast->backoff >= 2) {
419 * We only retry sendonly joins once before we drop
420 * the packet and quit trying to deal with the
421 * group. However, we leave the group in the
422 * mcast list as an unjoined group. If we want to
423 * try joining again, we simply queue up a packet
424 * and restart the join thread. The empty queue
425 * is why the join thread ignores this group.
428 netif_tx_lock_bh(dev);
429 while (!skb_queue_empty(&mcast->pkt_queue)) {
430 ++dev->stats.tx_dropped;
431 dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
433 netif_tx_unlock_bh(dev);
435 spin_lock_irq(&priv->lock);
436 /* Requeue this join task with a backoff delay */
437 __ipoib_mcast_schedule_join_thread(priv, mcast, 1);
442 spin_lock_irq(&priv->lock);
445 * Make sure to set mcast->mc before we clear the busy flag to avoid
446 * racing with code that checks for BUSY before checking mcast->mc
451 mcast->mc = multicast;
452 clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
453 spin_unlock_irq(&priv->lock);
454 complete(&mcast->done);
460 * Caller must hold 'priv->lock'
462 static int ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast)
464 struct ipoib_dev_priv *priv = netdev_priv(dev);
465 struct ib_sa_multicast *multicast;
466 struct ib_sa_mcmember_rec rec = {
469 ib_sa_comp_mask comp_mask;
472 if (!priv->broadcast ||
473 !test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
476 init_completion(&mcast->done);
477 set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
479 ipoib_dbg_mcast(priv, "joining MGID %pI6\n", mcast->mcmember.mgid.raw);
481 rec.mgid = mcast->mcmember.mgid;
482 rec.port_gid = priv->local_gid;
483 rec.pkey = cpu_to_be16(priv->pkey);
486 IB_SA_MCMEMBER_REC_MGID |
487 IB_SA_MCMEMBER_REC_PORT_GID |
488 IB_SA_MCMEMBER_REC_PKEY |
489 IB_SA_MCMEMBER_REC_JOIN_STATE;
491 if (mcast != priv->broadcast) {
494 * The MGID MUST use the same P_Key, Q_Key, SL, MTU,
495 * and HopLimit as those used in the broadcast-GID. The rest
496 * of attributes SHOULD follow the values used in the
497 * broadcast-GID as well.
500 IB_SA_MCMEMBER_REC_QKEY |
501 IB_SA_MCMEMBER_REC_MTU_SELECTOR |
502 IB_SA_MCMEMBER_REC_MTU |
503 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS |
504 IB_SA_MCMEMBER_REC_RATE_SELECTOR |
505 IB_SA_MCMEMBER_REC_RATE |
506 IB_SA_MCMEMBER_REC_SL |
507 IB_SA_MCMEMBER_REC_FLOW_LABEL |
508 IB_SA_MCMEMBER_REC_HOP_LIMIT;
510 rec.qkey = priv->broadcast->mcmember.qkey;
511 rec.mtu_selector = IB_SA_EQ;
512 rec.mtu = priv->broadcast->mcmember.mtu;
513 rec.traffic_class = priv->broadcast->mcmember.traffic_class;
514 rec.rate_selector = IB_SA_EQ;
515 rec.rate = priv->broadcast->mcmember.rate;
516 rec.sl = priv->broadcast->mcmember.sl;
517 rec.flow_label = priv->broadcast->mcmember.flow_label;
518 rec.hop_limit = priv->broadcast->mcmember.hop_limit;
521 * Send-only IB Multicast joins do not work at the core
522 * IB layer yet, so we can't use them here. However,
523 * we are emulating an Ethernet multicast send, which
524 * does not require a multicast subscription and will
525 * still send properly. The most appropriate thing to
526 * do is to create the group if it doesn't exist as that
527 * most closely emulates the behavior, from a user space
528 * application perspecitive, of Ethernet multicast
529 * operation. For now, we do a full join, maybe later
530 * when the core IB layers support send only joins we
534 if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags))
538 spin_unlock_irq(&priv->lock);
540 multicast = ib_sa_join_multicast(&ipoib_sa_client, priv->ca, priv->port,
541 &rec, comp_mask, GFP_KERNEL,
542 ipoib_mcast_join_complete, mcast);
543 spin_lock_irq(&priv->lock);
544 if (IS_ERR(multicast)) {
545 ret = PTR_ERR(multicast);
546 ipoib_warn(priv, "ib_sa_join_multicast failed, status %d\n", ret);
547 /* Requeue this join task with a backoff delay */
548 __ipoib_mcast_schedule_join_thread(priv, mcast, 1);
549 clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
550 spin_unlock_irq(&priv->lock);
551 complete(&mcast->done);
552 spin_lock_irq(&priv->lock);
557 void ipoib_mcast_join_task(struct work_struct *work)
559 struct ipoib_dev_priv *priv =
560 container_of(work, struct ipoib_dev_priv, mcast_task.work);
561 struct net_device *dev = priv->dev;
562 struct ib_port_attr port_attr;
563 unsigned long delay_until = 0;
564 struct ipoib_mcast *mcast = NULL;
566 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
569 if (ib_query_port(priv->ca, priv->port, &port_attr)) {
570 ipoib_dbg(priv, "ib_query_port() failed\n");
573 if (port_attr.state != IB_PORT_ACTIVE) {
574 ipoib_dbg(priv, "port state is not ACTIVE (state = %d) suspending join task\n",
578 priv->local_lid = port_attr.lid;
580 if (ib_query_gid(priv->ca, priv->port, 0, &priv->local_gid, NULL))
581 ipoib_warn(priv, "ib_query_gid() failed\n");
583 memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw, sizeof (union ib_gid));
585 spin_lock_irq(&priv->lock);
586 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
589 if (!priv->broadcast) {
590 struct ipoib_mcast *broadcast;
592 broadcast = ipoib_mcast_alloc(dev, 0);
594 ipoib_warn(priv, "failed to allocate broadcast group\n");
596 * Restart us after a 1 second delay to retry
597 * creating our broadcast group and attaching to
598 * it. Until this succeeds, this ipoib dev is
599 * completely stalled (multicast wise).
601 __ipoib_mcast_schedule_join_thread(priv, NULL, 1);
605 memcpy(broadcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
606 sizeof (union ib_gid));
607 priv->broadcast = broadcast;
609 __ipoib_mcast_add(dev, priv->broadcast);
612 if (!test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
613 if (IS_ERR_OR_NULL(priv->broadcast->mc) &&
614 !test_bit(IPOIB_MCAST_FLAG_BUSY, &priv->broadcast->flags)) {
615 mcast = priv->broadcast;
616 if (mcast->backoff > 1 &&
617 time_before(jiffies, mcast->delay_until)) {
618 delay_until = mcast->delay_until;
626 * We'll never get here until the broadcast group is both allocated
629 list_for_each_entry(mcast, &priv->multicast_list, list) {
630 if (IS_ERR_OR_NULL(mcast->mc) &&
631 !test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags) &&
632 (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) ||
633 !skb_queue_empty(&mcast->pkt_queue))) {
634 if (mcast->backoff == 1 ||
635 time_after_eq(jiffies, mcast->delay_until)) {
636 /* Found the next unjoined group */
637 if (ipoib_mcast_join(dev, mcast)) {
638 spin_unlock_irq(&priv->lock);
641 } else if (!delay_until ||
642 time_before(mcast->delay_until, delay_until))
643 delay_until = mcast->delay_until;
648 ipoib_dbg_mcast(priv, "successfully started all multicast joins\n");
652 cancel_delayed_work(&priv->mcast_task);
653 queue_delayed_work(priv->wq, &priv->mcast_task,
654 delay_until - jiffies);
657 ipoib_mcast_join(dev, mcast);
659 spin_unlock_irq(&priv->lock);
662 int ipoib_mcast_start_thread(struct net_device *dev)
664 struct ipoib_dev_priv *priv = netdev_priv(dev);
667 ipoib_dbg_mcast(priv, "starting multicast thread\n");
669 spin_lock_irqsave(&priv->lock, flags);
670 __ipoib_mcast_schedule_join_thread(priv, NULL, 0);
671 spin_unlock_irqrestore(&priv->lock, flags);
676 int ipoib_mcast_stop_thread(struct net_device *dev)
678 struct ipoib_dev_priv *priv = netdev_priv(dev);
681 ipoib_dbg_mcast(priv, "stopping multicast thread\n");
683 spin_lock_irqsave(&priv->lock, flags);
684 cancel_delayed_work(&priv->mcast_task);
685 spin_unlock_irqrestore(&priv->lock, flags);
687 flush_workqueue(priv->wq);
692 int ipoib_mcast_leave(struct net_device *dev, struct ipoib_mcast *mcast)
694 struct ipoib_dev_priv *priv = netdev_priv(dev);
697 if (test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
698 ipoib_warn(priv, "ipoib_mcast_leave on an in-flight join\n");
700 if (!IS_ERR_OR_NULL(mcast->mc))
701 ib_sa_free_multicast(mcast->mc);
703 if (test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
704 ipoib_dbg_mcast(priv, "leaving MGID %pI6\n",
705 mcast->mcmember.mgid.raw);
707 /* Remove ourselves from the multicast group */
708 ret = ib_detach_mcast(priv->qp, &mcast->mcmember.mgid,
709 be16_to_cpu(mcast->mcmember.mlid));
711 ipoib_warn(priv, "ib_detach_mcast failed (result = %d)\n", ret);
712 } else if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags))
713 ipoib_dbg(priv, "leaving with no mcmember but not a "
719 void ipoib_mcast_send(struct net_device *dev, u8 *daddr, struct sk_buff *skb)
721 struct ipoib_dev_priv *priv = netdev_priv(dev);
722 struct ipoib_mcast *mcast;
724 void *mgid = daddr + 4;
726 spin_lock_irqsave(&priv->lock, flags);
728 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags) ||
730 !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
731 ++dev->stats.tx_dropped;
732 dev_kfree_skb_any(skb);
736 mcast = __ipoib_mcast_find(dev, mgid);
737 if (!mcast || !mcast->ah) {
739 /* Let's create a new send only group now */
740 ipoib_dbg_mcast(priv, "setting up send only multicast group for %pI6\n",
743 mcast = ipoib_mcast_alloc(dev, 0);
745 ipoib_warn(priv, "unable to allocate memory "
746 "for multicast structure\n");
747 ++dev->stats.tx_dropped;
748 dev_kfree_skb_any(skb);
752 set_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags);
753 memcpy(mcast->mcmember.mgid.raw, mgid,
754 sizeof (union ib_gid));
755 __ipoib_mcast_add(dev, mcast);
756 list_add_tail(&mcast->list, &priv->multicast_list);
758 if (skb_queue_len(&mcast->pkt_queue) < IPOIB_MAX_MCAST_QUEUE) {
759 /* put pseudoheader back on for next time */
760 skb_push(skb, sizeof(struct ipoib_pseudo_header));
761 skb_queue_tail(&mcast->pkt_queue, skb);
763 ++dev->stats.tx_dropped;
764 dev_kfree_skb_any(skb);
766 if (!test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) {
767 __ipoib_mcast_schedule_join_thread(priv, NULL, 0);
770 struct ipoib_neigh *neigh;
772 spin_unlock_irqrestore(&priv->lock, flags);
773 neigh = ipoib_neigh_get(dev, daddr);
774 spin_lock_irqsave(&priv->lock, flags);
776 neigh = ipoib_neigh_alloc(daddr, dev);
777 /* Make sure that the neigh will be added only
778 * once to mcast list.
780 if (neigh && list_empty(&neigh->list)) {
781 kref_get(&mcast->ah->ref);
782 neigh->ah = mcast->ah;
783 list_add_tail(&neigh->list, &mcast->neigh_list);
786 spin_unlock_irqrestore(&priv->lock, flags);
787 ipoib_send(dev, skb, mcast->ah, IB_MULTICAST_QPN);
789 ipoib_neigh_put(neigh);
794 spin_unlock_irqrestore(&priv->lock, flags);
797 void ipoib_mcast_dev_flush(struct net_device *dev)
799 struct ipoib_dev_priv *priv = netdev_priv(dev);
800 LIST_HEAD(remove_list);
801 struct ipoib_mcast *mcast, *tmcast;
804 ipoib_dbg_mcast(priv, "flushing multicast list\n");
806 spin_lock_irqsave(&priv->lock, flags);
808 list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
809 list_del(&mcast->list);
810 rb_erase(&mcast->rb_node, &priv->multicast_tree);
811 list_add_tail(&mcast->list, &remove_list);
814 if (priv->broadcast) {
815 rb_erase(&priv->broadcast->rb_node, &priv->multicast_tree);
816 list_add_tail(&priv->broadcast->list, &remove_list);
817 priv->broadcast = NULL;
820 spin_unlock_irqrestore(&priv->lock, flags);
823 * make sure the in-flight joins have finished before we attempt
826 list_for_each_entry_safe(mcast, tmcast, &remove_list, list)
827 if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
828 wait_for_completion(&mcast->done);
830 list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
831 ipoib_mcast_leave(dev, mcast);
832 ipoib_mcast_free(mcast);
836 static int ipoib_mcast_addr_is_valid(const u8 *addr, const u8 *broadcast)
838 /* reserved QPN, prefix, scope */
839 if (memcmp(addr, broadcast, 6))
841 /* signature lower, pkey */
842 if (memcmp(addr + 7, broadcast + 7, 3))
847 void ipoib_mcast_restart_task(struct work_struct *work)
849 struct ipoib_dev_priv *priv =
850 container_of(work, struct ipoib_dev_priv, restart_task);
851 struct net_device *dev = priv->dev;
852 struct netdev_hw_addr *ha;
853 struct ipoib_mcast *mcast, *tmcast;
854 LIST_HEAD(remove_list);
856 struct ib_sa_mcmember_rec rec;
858 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
860 * shortcut...on shutdown flush is called next, just
861 * let it do all the work
865 ipoib_dbg_mcast(priv, "restarting multicast task\n");
867 local_irq_save(flags);
868 netif_addr_lock(dev);
869 spin_lock(&priv->lock);
872 * Unfortunately, the networking core only gives us a list of all of
873 * the multicast hardware addresses. We need to figure out which ones
874 * are new and which ones have been removed
877 /* Clear out the found flag */
878 list_for_each_entry(mcast, &priv->multicast_list, list)
879 clear_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
881 /* Mark all of the entries that are found or don't exist */
882 netdev_for_each_mc_addr(ha, dev) {
885 if (!ipoib_mcast_addr_is_valid(ha->addr, dev->broadcast))
888 memcpy(mgid.raw, ha->addr + 4, sizeof mgid);
890 mcast = __ipoib_mcast_find(dev, &mgid);
891 if (!mcast || test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
892 struct ipoib_mcast *nmcast;
894 /* ignore group which is directly joined by userspace */
895 if (test_bit(IPOIB_FLAG_UMCAST, &priv->flags) &&
896 !ib_sa_get_mcmember_rec(priv->ca, priv->port, &mgid, &rec)) {
897 ipoib_dbg_mcast(priv, "ignoring multicast entry for mgid %pI6\n",
902 /* Not found or send-only group, let's add a new entry */
903 ipoib_dbg_mcast(priv, "adding multicast entry for mgid %pI6\n",
906 nmcast = ipoib_mcast_alloc(dev, 0);
908 ipoib_warn(priv, "unable to allocate memory for multicast structure\n");
912 set_bit(IPOIB_MCAST_FLAG_FOUND, &nmcast->flags);
914 nmcast->mcmember.mgid = mgid;
917 /* Destroy the send only entry */
918 list_move_tail(&mcast->list, &remove_list);
920 rb_replace_node(&mcast->rb_node,
922 &priv->multicast_tree);
924 __ipoib_mcast_add(dev, nmcast);
926 list_add_tail(&nmcast->list, &priv->multicast_list);
930 set_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
933 /* Remove all of the entries don't exist anymore */
934 list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
935 if (!test_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags) &&
936 !test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
937 ipoib_dbg_mcast(priv, "deleting multicast group %pI6\n",
938 mcast->mcmember.mgid.raw);
940 rb_erase(&mcast->rb_node, &priv->multicast_tree);
942 /* Move to the remove list */
943 list_move_tail(&mcast->list, &remove_list);
947 spin_unlock(&priv->lock);
948 netif_addr_unlock(dev);
949 local_irq_restore(flags);
952 * make sure the in-flight joins have finished before we attempt
955 list_for_each_entry_safe(mcast, tmcast, &remove_list, list)
956 if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
957 wait_for_completion(&mcast->done);
959 list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
960 ipoib_mcast_leave(mcast->dev, mcast);
961 ipoib_mcast_free(mcast);
965 * Double check that we are still up
967 if (test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
968 spin_lock_irqsave(&priv->lock, flags);
969 __ipoib_mcast_schedule_join_thread(priv, NULL, 0);
970 spin_unlock_irqrestore(&priv->lock, flags);
974 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
976 struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev)
978 struct ipoib_mcast_iter *iter;
980 iter = kmalloc(sizeof *iter, GFP_KERNEL);
985 memset(iter->mgid.raw, 0, 16);
987 if (ipoib_mcast_iter_next(iter)) {
995 int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter)
997 struct ipoib_dev_priv *priv = netdev_priv(iter->dev);
999 struct ipoib_mcast *mcast;
1002 spin_lock_irq(&priv->lock);
1004 n = rb_first(&priv->multicast_tree);
1007 mcast = rb_entry(n, struct ipoib_mcast, rb_node);
1009 if (memcmp(iter->mgid.raw, mcast->mcmember.mgid.raw,
1010 sizeof (union ib_gid)) < 0) {
1011 iter->mgid = mcast->mcmember.mgid;
1012 iter->created = mcast->created;
1013 iter->queuelen = skb_queue_len(&mcast->pkt_queue);
1014 iter->complete = !!mcast->ah;
1015 iter->send_only = !!(mcast->flags & (1 << IPOIB_MCAST_FLAG_SENDONLY));
1025 spin_unlock_irq(&priv->lock);
1030 void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
1032 unsigned long *created,
1033 unsigned int *queuelen,
1034 unsigned int *complete,
1035 unsigned int *send_only)
1038 *created = iter->created;
1039 *queuelen = iter->queuelen;
1040 *complete = iter->complete;
1041 *send_only = iter->send_only;
1044 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */