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;
67 /* join state that allows creating mcg with sendonly member request */
68 #define SENDONLY_FULLMEMBER_JOIN 8
71 * This should be called with the priv->lock held
73 static void __ipoib_mcast_schedule_join_thread(struct ipoib_dev_priv *priv,
74 struct ipoib_mcast *mcast,
77 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
81 * We will be scheduling *something*, so cancel whatever is
82 * currently scheduled first
84 cancel_delayed_work(&priv->mcast_task);
87 * We had a failure and want to schedule a retry later
90 if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
91 mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
92 mcast->delay_until = jiffies + (mcast->backoff * HZ);
94 * Mark this mcast for its delay, but restart the
95 * task immediately. The join task will make sure to
96 * clear out all entries without delays, and then
97 * schedule itself to run again when the earliest
100 queue_delayed_work(priv->wq, &priv->mcast_task, 0);
103 * Special case of retrying after a failure to
104 * allocate the broadcast multicast group, wait
105 * 1 second and try again
107 queue_delayed_work(priv->wq, &priv->mcast_task, HZ);
109 queue_delayed_work(priv->wq, &priv->mcast_task, 0);
112 static void ipoib_mcast_free(struct ipoib_mcast *mcast)
114 struct net_device *dev = mcast->dev;
117 ipoib_dbg_mcast(netdev_priv(dev), "deleting multicast group %pI6\n",
118 mcast->mcmember.mgid.raw);
120 /* remove all neigh connected to this mcast */
121 ipoib_del_neighs_by_gid(dev, mcast->mcmember.mgid.raw);
124 ipoib_put_ah(mcast->ah);
126 while (!skb_queue_empty(&mcast->pkt_queue)) {
128 dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
131 netif_tx_lock_bh(dev);
132 dev->stats.tx_dropped += tx_dropped;
133 netif_tx_unlock_bh(dev);
138 static struct ipoib_mcast *ipoib_mcast_alloc(struct net_device *dev,
141 struct ipoib_mcast *mcast;
143 mcast = kzalloc(sizeof *mcast, can_sleep ? GFP_KERNEL : GFP_ATOMIC);
148 mcast->created = jiffies;
149 mcast->delay_until = jiffies;
152 INIT_LIST_HEAD(&mcast->list);
153 INIT_LIST_HEAD(&mcast->neigh_list);
154 skb_queue_head_init(&mcast->pkt_queue);
159 static struct ipoib_mcast *__ipoib_mcast_find(struct net_device *dev, void *mgid)
161 struct ipoib_dev_priv *priv = netdev_priv(dev);
162 struct rb_node *n = priv->multicast_tree.rb_node;
165 struct ipoib_mcast *mcast;
168 mcast = rb_entry(n, struct ipoib_mcast, rb_node);
170 ret = memcmp(mgid, mcast->mcmember.mgid.raw,
171 sizeof (union ib_gid));
183 static int __ipoib_mcast_add(struct net_device *dev, struct ipoib_mcast *mcast)
185 struct ipoib_dev_priv *priv = netdev_priv(dev);
186 struct rb_node **n = &priv->multicast_tree.rb_node, *pn = NULL;
189 struct ipoib_mcast *tmcast;
193 tmcast = rb_entry(pn, struct ipoib_mcast, rb_node);
195 ret = memcmp(mcast->mcmember.mgid.raw, tmcast->mcmember.mgid.raw,
196 sizeof (union ib_gid));
205 rb_link_node(&mcast->rb_node, pn, n);
206 rb_insert_color(&mcast->rb_node, &priv->multicast_tree);
211 static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
212 struct ib_sa_mcmember_rec *mcmember)
214 struct net_device *dev = mcast->dev;
215 struct ipoib_dev_priv *priv = netdev_priv(dev);
220 mcast->mcmember = *mcmember;
222 /* Set the multicast MTU and cached Q_Key before we attach if it's
223 * the broadcast group.
225 if (!memcmp(mcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
226 sizeof (union ib_gid))) {
227 spin_lock_irq(&priv->lock);
228 if (!priv->broadcast) {
229 spin_unlock_irq(&priv->lock);
232 /*update priv member according to the new mcast*/
233 priv->broadcast->mcmember.qkey = mcmember->qkey;
234 priv->broadcast->mcmember.mtu = mcmember->mtu;
235 priv->broadcast->mcmember.traffic_class = mcmember->traffic_class;
236 priv->broadcast->mcmember.rate = mcmember->rate;
237 priv->broadcast->mcmember.sl = mcmember->sl;
238 priv->broadcast->mcmember.flow_label = mcmember->flow_label;
239 priv->broadcast->mcmember.hop_limit = mcmember->hop_limit;
240 /* assume if the admin and the mcast are the same both can be changed */
241 if (priv->mcast_mtu == priv->admin_mtu)
244 IPOIB_UD_MTU(ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu));
247 IPOIB_UD_MTU(ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu));
249 priv->qkey = be32_to_cpu(priv->broadcast->mcmember.qkey);
250 spin_unlock_irq(&priv->lock);
251 priv->tx_wr.remote_qkey = priv->qkey;
255 if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
256 if (test_and_set_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
257 ipoib_warn(priv, "multicast group %pI6 already attached\n",
258 mcast->mcmember.mgid.raw);
263 ret = ipoib_mcast_attach(dev, be16_to_cpu(mcast->mcmember.mlid),
264 &mcast->mcmember.mgid, set_qkey);
266 ipoib_warn(priv, "couldn't attach QP to multicast group %pI6\n",
267 mcast->mcmember.mgid.raw);
269 clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags);
275 struct ib_ah_attr av = {
276 .dlid = be16_to_cpu(mcast->mcmember.mlid),
277 .port_num = priv->port,
278 .sl = mcast->mcmember.sl,
279 .ah_flags = IB_AH_GRH,
280 .static_rate = mcast->mcmember.rate,
282 .flow_label = be32_to_cpu(mcast->mcmember.flow_label),
283 .hop_limit = mcast->mcmember.hop_limit,
285 .traffic_class = mcast->mcmember.traffic_class
288 av.grh.dgid = mcast->mcmember.mgid;
290 ah = ipoib_create_ah(dev, priv->pd, &av);
292 ipoib_warn(priv, "ib_address_create failed %ld\n",
294 /* use original error */
297 spin_lock_irq(&priv->lock);
299 spin_unlock_irq(&priv->lock);
301 ipoib_dbg_mcast(priv, "MGID %pI6 AV %p, LID 0x%04x, SL %d\n",
302 mcast->mcmember.mgid.raw,
304 be16_to_cpu(mcast->mcmember.mlid),
309 /* actually send any queued packets */
310 netif_tx_lock_bh(dev);
311 while (!skb_queue_empty(&mcast->pkt_queue)) {
312 struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue);
314 netif_tx_unlock_bh(dev);
317 if (dev_queue_xmit(skb))
318 ipoib_warn(priv, "dev_queue_xmit failed to requeue packet\n");
320 netif_tx_lock_bh(dev);
322 netif_tx_unlock_bh(dev);
327 void ipoib_mcast_carrier_on_task(struct work_struct *work)
329 struct ipoib_dev_priv *priv = container_of(work, struct ipoib_dev_priv,
331 struct ib_port_attr attr;
334 if (ib_query_port(priv->ca, priv->port, &attr) ||
335 attr.state != IB_PORT_ACTIVE) {
336 ipoib_dbg(priv, "Keeping carrier off until IB port is active\n");
340 * Check if can send sendonly MCG's with sendonly-fullmember join state.
341 * It done here after the successfully join to the broadcast group,
342 * because the broadcast group must always be joined first and is always
343 * re-joined if the SM changes substantially.
345 ret = ipoib_check_sm_sendonly_fullmember_support(priv);
347 pr_debug("%s failed query sm support for sendonly-fullmember (ret: %d)\n",
348 priv->dev->name, ret);
351 * Take rtnl_lock to avoid racing with ipoib_stop() and
352 * turning the carrier back on while a device is being
353 * removed. However, ipoib_stop() will attempt to flush
354 * the workqueue while holding the rtnl lock, so loop
355 * on trylock until either we get the lock or we see
356 * FLAG_OPER_UP go away as that signals that we are bailing
357 * and can safely ignore the carrier on work.
359 while (!rtnl_trylock()) {
360 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
365 if (!ipoib_cm_admin_enabled(priv->dev))
366 dev_set_mtu(priv->dev, min(priv->mcast_mtu, priv->admin_mtu));
367 netif_carrier_on(priv->dev);
371 static int ipoib_mcast_join_complete(int status,
372 struct ib_sa_multicast *multicast)
374 struct ipoib_mcast *mcast = multicast->context;
375 struct net_device *dev = mcast->dev;
376 struct ipoib_dev_priv *priv = netdev_priv(dev);
378 ipoib_dbg_mcast(priv, "%sjoin completion for %pI6 (status %d)\n",
379 test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) ?
381 mcast->mcmember.mgid.raw, status);
383 /* We trap for port events ourselves. */
384 if (status == -ENETRESET) {
390 status = ipoib_mcast_join_finish(mcast, &multicast->rec);
394 mcast->delay_until = jiffies;
397 * Defer carrier on work to priv->wq to avoid a
398 * deadlock on rtnl_lock here. Requeue our multicast
399 * work too, which will end up happening right after
400 * our carrier on task work and will allow us to
401 * send out all of the non-broadcast joins
403 if (mcast == priv->broadcast) {
404 spin_lock_irq(&priv->lock);
405 queue_work(priv->wq, &priv->carrier_on_task);
406 __ipoib_mcast_schedule_join_thread(priv, NULL, 0);
411 test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) &&
414 if (mcast->logcount < 20) {
415 if (status == -ETIMEDOUT || status == -EAGAIN ||
417 ipoib_dbg_mcast(priv, "%smulticast join failed for %pI6, status %d\n",
418 test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) ? "sendonly " : "",
419 mcast->mcmember.mgid.raw, status);
421 ipoib_warn(priv, "%smulticast join failed for %pI6, status %d\n",
422 test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) ? "sendonly " : "",
423 mcast->mcmember.mgid.raw, status);
430 if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) &&
431 mcast->backoff >= 2) {
433 * We only retry sendonly joins once before we drop
434 * the packet and quit trying to deal with the
435 * group. However, we leave the group in the
436 * mcast list as an unjoined group. If we want to
437 * try joining again, we simply queue up a packet
438 * and restart the join thread. The empty queue
439 * is why the join thread ignores this group.
442 netif_tx_lock_bh(dev);
443 while (!skb_queue_empty(&mcast->pkt_queue)) {
444 ++dev->stats.tx_dropped;
445 dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
447 netif_tx_unlock_bh(dev);
449 spin_lock_irq(&priv->lock);
450 /* Requeue this join task with a backoff delay */
451 __ipoib_mcast_schedule_join_thread(priv, mcast, 1);
456 spin_lock_irq(&priv->lock);
459 * Make sure to set mcast->mc before we clear the busy flag to avoid
460 * racing with code that checks for BUSY before checking mcast->mc
465 mcast->mc = multicast;
466 clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
467 spin_unlock_irq(&priv->lock);
468 complete(&mcast->done);
474 * Caller must hold 'priv->lock'
476 static int ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast)
478 struct ipoib_dev_priv *priv = netdev_priv(dev);
479 struct ib_sa_multicast *multicast;
480 struct ib_sa_mcmember_rec rec = {
483 ib_sa_comp_mask comp_mask;
486 if (!priv->broadcast ||
487 !test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
490 init_completion(&mcast->done);
491 set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
493 ipoib_dbg_mcast(priv, "joining MGID %pI6\n", mcast->mcmember.mgid.raw);
495 rec.mgid = mcast->mcmember.mgid;
496 rec.port_gid = priv->local_gid;
497 rec.pkey = cpu_to_be16(priv->pkey);
500 IB_SA_MCMEMBER_REC_MGID |
501 IB_SA_MCMEMBER_REC_PORT_GID |
502 IB_SA_MCMEMBER_REC_PKEY |
503 IB_SA_MCMEMBER_REC_JOIN_STATE;
505 if (mcast != priv->broadcast) {
508 * The MGID MUST use the same P_Key, Q_Key, SL, MTU,
509 * and HopLimit as those used in the broadcast-GID. The rest
510 * of attributes SHOULD follow the values used in the
511 * broadcast-GID as well.
514 IB_SA_MCMEMBER_REC_QKEY |
515 IB_SA_MCMEMBER_REC_MTU_SELECTOR |
516 IB_SA_MCMEMBER_REC_MTU |
517 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS |
518 IB_SA_MCMEMBER_REC_RATE_SELECTOR |
519 IB_SA_MCMEMBER_REC_RATE |
520 IB_SA_MCMEMBER_REC_SL |
521 IB_SA_MCMEMBER_REC_FLOW_LABEL |
522 IB_SA_MCMEMBER_REC_HOP_LIMIT;
524 rec.qkey = priv->broadcast->mcmember.qkey;
525 rec.mtu_selector = IB_SA_EQ;
526 rec.mtu = priv->broadcast->mcmember.mtu;
527 rec.traffic_class = priv->broadcast->mcmember.traffic_class;
528 rec.rate_selector = IB_SA_EQ;
529 rec.rate = priv->broadcast->mcmember.rate;
530 rec.sl = priv->broadcast->mcmember.sl;
531 rec.flow_label = priv->broadcast->mcmember.flow_label;
532 rec.hop_limit = priv->broadcast->mcmember.hop_limit;
535 * Send-only IB Multicast joins work at the core IB layer but
536 * require specific SM support.
537 * We can use such joins here only if the current SM supports that feature.
538 * However, if not, we emulate an Ethernet multicast send,
539 * which does not require a multicast subscription and will
540 * still send properly. The most appropriate thing to
541 * do is to create the group if it doesn't exist as that
542 * most closely emulates the behavior, from a user space
543 * application perspective, of Ethernet multicast operation.
545 if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) &&
546 priv->sm_fullmember_sendonly_support)
547 /* SM supports sendonly-fullmember, otherwise fallback to full-member */
548 rec.join_state = SENDONLY_FULLMEMBER_JOIN;
550 spin_unlock_irq(&priv->lock);
552 multicast = ib_sa_join_multicast(&ipoib_sa_client, priv->ca, priv->port,
553 &rec, comp_mask, GFP_KERNEL,
554 ipoib_mcast_join_complete, mcast);
555 spin_lock_irq(&priv->lock);
556 if (IS_ERR(multicast)) {
557 ret = PTR_ERR(multicast);
558 ipoib_warn(priv, "ib_sa_join_multicast failed, status %d\n", ret);
559 /* Requeue this join task with a backoff delay */
560 __ipoib_mcast_schedule_join_thread(priv, mcast, 1);
561 clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
562 spin_unlock_irq(&priv->lock);
563 complete(&mcast->done);
564 spin_lock_irq(&priv->lock);
569 void ipoib_mcast_join_task(struct work_struct *work)
571 struct ipoib_dev_priv *priv =
572 container_of(work, struct ipoib_dev_priv, mcast_task.work);
573 struct net_device *dev = priv->dev;
574 struct ib_port_attr port_attr;
575 unsigned long delay_until = 0;
576 struct ipoib_mcast *mcast = NULL;
578 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
581 if (ib_query_port(priv->ca, priv->port, &port_attr)) {
582 ipoib_dbg(priv, "ib_query_port() failed\n");
585 if (port_attr.state != IB_PORT_ACTIVE) {
586 ipoib_dbg(priv, "port state is not ACTIVE (state = %d) suspending join task\n",
590 priv->local_lid = port_attr.lid;
591 netif_addr_lock_bh(dev);
593 if (!test_bit(IPOIB_FLAG_DEV_ADDR_SET, &priv->flags)) {
594 netif_addr_unlock_bh(dev);
597 netif_addr_unlock_bh(dev);
599 spin_lock_irq(&priv->lock);
600 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
603 if (!priv->broadcast) {
604 struct ipoib_mcast *broadcast;
606 broadcast = ipoib_mcast_alloc(dev, 0);
608 ipoib_warn(priv, "failed to allocate broadcast group\n");
610 * Restart us after a 1 second delay to retry
611 * creating our broadcast group and attaching to
612 * it. Until this succeeds, this ipoib dev is
613 * completely stalled (multicast wise).
615 __ipoib_mcast_schedule_join_thread(priv, NULL, 1);
619 memcpy(broadcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
620 sizeof (union ib_gid));
621 priv->broadcast = broadcast;
623 __ipoib_mcast_add(dev, priv->broadcast);
626 if (!test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
627 if (IS_ERR_OR_NULL(priv->broadcast->mc) &&
628 !test_bit(IPOIB_MCAST_FLAG_BUSY, &priv->broadcast->flags)) {
629 mcast = priv->broadcast;
630 if (mcast->backoff > 1 &&
631 time_before(jiffies, mcast->delay_until)) {
632 delay_until = mcast->delay_until;
640 * We'll never get here until the broadcast group is both allocated
643 list_for_each_entry(mcast, &priv->multicast_list, list) {
644 if (IS_ERR_OR_NULL(mcast->mc) &&
645 !test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags) &&
646 (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) ||
647 !skb_queue_empty(&mcast->pkt_queue))) {
648 if (mcast->backoff == 1 ||
649 time_after_eq(jiffies, mcast->delay_until)) {
650 /* Found the next unjoined group */
651 if (ipoib_mcast_join(dev, mcast)) {
652 spin_unlock_irq(&priv->lock);
655 } else if (!delay_until ||
656 time_before(mcast->delay_until, delay_until))
657 delay_until = mcast->delay_until;
662 ipoib_dbg_mcast(priv, "successfully started all multicast joins\n");
666 cancel_delayed_work(&priv->mcast_task);
667 queue_delayed_work(priv->wq, &priv->mcast_task,
668 delay_until - jiffies);
671 ipoib_mcast_join(dev, mcast);
673 spin_unlock_irq(&priv->lock);
676 int ipoib_mcast_start_thread(struct net_device *dev)
678 struct ipoib_dev_priv *priv = netdev_priv(dev);
681 ipoib_dbg_mcast(priv, "starting multicast thread\n");
683 spin_lock_irqsave(&priv->lock, flags);
684 __ipoib_mcast_schedule_join_thread(priv, NULL, 0);
685 spin_unlock_irqrestore(&priv->lock, flags);
690 int ipoib_mcast_stop_thread(struct net_device *dev)
692 struct ipoib_dev_priv *priv = netdev_priv(dev);
695 ipoib_dbg_mcast(priv, "stopping multicast thread\n");
697 spin_lock_irqsave(&priv->lock, flags);
698 cancel_delayed_work(&priv->mcast_task);
699 spin_unlock_irqrestore(&priv->lock, flags);
701 flush_workqueue(priv->wq);
706 static int ipoib_mcast_leave(struct net_device *dev, struct ipoib_mcast *mcast)
708 struct ipoib_dev_priv *priv = netdev_priv(dev);
711 if (test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
712 ipoib_warn(priv, "ipoib_mcast_leave on an in-flight join\n");
714 if (!IS_ERR_OR_NULL(mcast->mc))
715 ib_sa_free_multicast(mcast->mc);
717 if (test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
718 ipoib_dbg_mcast(priv, "leaving MGID %pI6\n",
719 mcast->mcmember.mgid.raw);
721 /* Remove ourselves from the multicast group */
722 ret = ib_detach_mcast(priv->qp, &mcast->mcmember.mgid,
723 be16_to_cpu(mcast->mcmember.mlid));
725 ipoib_warn(priv, "ib_detach_mcast failed (result = %d)\n", ret);
726 } else if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags))
727 ipoib_dbg(priv, "leaving with no mcmember but not a "
734 * Check if the multicast group is sendonly. If so remove it from the maps
735 * and add to the remove list
737 void ipoib_check_and_add_mcast_sendonly(struct ipoib_dev_priv *priv, u8 *mgid,
738 struct list_head *remove_list)
740 /* Is this multicast ? */
742 struct ipoib_mcast *mcast = __ipoib_mcast_find(priv->dev, mgid);
744 if (mcast && test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
745 list_del(&mcast->list);
746 rb_erase(&mcast->rb_node, &priv->multicast_tree);
747 list_add_tail(&mcast->list, remove_list);
752 void ipoib_mcast_remove_list(struct list_head *remove_list)
754 struct ipoib_mcast *mcast, *tmcast;
756 list_for_each_entry_safe(mcast, tmcast, remove_list, list) {
757 ipoib_mcast_leave(mcast->dev, mcast);
758 ipoib_mcast_free(mcast);
762 void ipoib_mcast_send(struct net_device *dev, u8 *daddr, struct sk_buff *skb)
764 struct ipoib_dev_priv *priv = netdev_priv(dev);
765 struct ipoib_mcast *mcast;
767 void *mgid = daddr + 4;
769 spin_lock_irqsave(&priv->lock, flags);
771 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags) ||
773 !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
774 ++dev->stats.tx_dropped;
775 dev_kfree_skb_any(skb);
779 mcast = __ipoib_mcast_find(dev, mgid);
780 if (!mcast || !mcast->ah) {
782 /* Let's create a new send only group now */
783 ipoib_dbg_mcast(priv, "setting up send only multicast group for %pI6\n",
786 mcast = ipoib_mcast_alloc(dev, 0);
788 ipoib_warn(priv, "unable to allocate memory "
789 "for multicast structure\n");
790 ++dev->stats.tx_dropped;
791 dev_kfree_skb_any(skb);
795 set_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags);
796 memcpy(mcast->mcmember.mgid.raw, mgid,
797 sizeof (union ib_gid));
798 __ipoib_mcast_add(dev, mcast);
799 list_add_tail(&mcast->list, &priv->multicast_list);
801 if (skb_queue_len(&mcast->pkt_queue) < IPOIB_MAX_MCAST_QUEUE) {
802 /* put pseudoheader back on for next time */
803 skb_push(skb, sizeof(struct ipoib_pseudo_header));
804 skb_queue_tail(&mcast->pkt_queue, skb);
806 ++dev->stats.tx_dropped;
807 dev_kfree_skb_any(skb);
809 if (!test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) {
810 __ipoib_mcast_schedule_join_thread(priv, NULL, 0);
813 struct ipoib_neigh *neigh;
815 spin_unlock_irqrestore(&priv->lock, flags);
816 neigh = ipoib_neigh_get(dev, daddr);
817 spin_lock_irqsave(&priv->lock, flags);
819 neigh = ipoib_neigh_alloc(daddr, dev);
820 /* Make sure that the neigh will be added only
821 * once to mcast list.
823 if (neigh && list_empty(&neigh->list)) {
824 kref_get(&mcast->ah->ref);
825 neigh->ah = mcast->ah;
826 list_add_tail(&neigh->list, &mcast->neigh_list);
829 spin_unlock_irqrestore(&priv->lock, flags);
830 ipoib_send(dev, skb, mcast->ah, IB_MULTICAST_QPN);
832 ipoib_neigh_put(neigh);
837 spin_unlock_irqrestore(&priv->lock, flags);
840 void ipoib_mcast_dev_flush(struct net_device *dev)
842 struct ipoib_dev_priv *priv = netdev_priv(dev);
843 LIST_HEAD(remove_list);
844 struct ipoib_mcast *mcast, *tmcast;
847 ipoib_dbg_mcast(priv, "flushing multicast list\n");
849 spin_lock_irqsave(&priv->lock, flags);
851 list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
852 list_del(&mcast->list);
853 rb_erase(&mcast->rb_node, &priv->multicast_tree);
854 list_add_tail(&mcast->list, &remove_list);
857 if (priv->broadcast) {
858 rb_erase(&priv->broadcast->rb_node, &priv->multicast_tree);
859 list_add_tail(&priv->broadcast->list, &remove_list);
860 priv->broadcast = NULL;
863 spin_unlock_irqrestore(&priv->lock, flags);
866 * make sure the in-flight joins have finished before we attempt
869 list_for_each_entry_safe(mcast, tmcast, &remove_list, list)
870 if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
871 wait_for_completion(&mcast->done);
873 ipoib_mcast_remove_list(&remove_list);
876 static int ipoib_mcast_addr_is_valid(const u8 *addr, const u8 *broadcast)
878 /* reserved QPN, prefix, scope */
879 if (memcmp(addr, broadcast, 6))
881 /* signature lower, pkey */
882 if (memcmp(addr + 7, broadcast + 7, 3))
887 void ipoib_mcast_restart_task(struct work_struct *work)
889 struct ipoib_dev_priv *priv =
890 container_of(work, struct ipoib_dev_priv, restart_task);
891 struct net_device *dev = priv->dev;
892 struct netdev_hw_addr *ha;
893 struct ipoib_mcast *mcast, *tmcast;
894 LIST_HEAD(remove_list);
896 struct ib_sa_mcmember_rec rec;
898 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
900 * shortcut...on shutdown flush is called next, just
901 * let it do all the work
905 ipoib_dbg_mcast(priv, "restarting multicast task\n");
907 local_irq_save(flags);
908 netif_addr_lock(dev);
909 spin_lock(&priv->lock);
912 * Unfortunately, the networking core only gives us a list of all of
913 * the multicast hardware addresses. We need to figure out which ones
914 * are new and which ones have been removed
917 /* Clear out the found flag */
918 list_for_each_entry(mcast, &priv->multicast_list, list)
919 clear_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
921 /* Mark all of the entries that are found or don't exist */
922 netdev_for_each_mc_addr(ha, dev) {
925 if (!ipoib_mcast_addr_is_valid(ha->addr, dev->broadcast))
928 memcpy(mgid.raw, ha->addr + 4, sizeof mgid);
930 mcast = __ipoib_mcast_find(dev, &mgid);
931 if (!mcast || test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
932 struct ipoib_mcast *nmcast;
934 /* ignore group which is directly joined by userspace */
935 if (test_bit(IPOIB_FLAG_UMCAST, &priv->flags) &&
936 !ib_sa_get_mcmember_rec(priv->ca, priv->port, &mgid, &rec)) {
937 ipoib_dbg_mcast(priv, "ignoring multicast entry for mgid %pI6\n",
942 /* Not found or send-only group, let's add a new entry */
943 ipoib_dbg_mcast(priv, "adding multicast entry for mgid %pI6\n",
946 nmcast = ipoib_mcast_alloc(dev, 0);
948 ipoib_warn(priv, "unable to allocate memory for multicast structure\n");
952 set_bit(IPOIB_MCAST_FLAG_FOUND, &nmcast->flags);
954 nmcast->mcmember.mgid = mgid;
957 /* Destroy the send only entry */
958 list_move_tail(&mcast->list, &remove_list);
960 rb_replace_node(&mcast->rb_node,
962 &priv->multicast_tree);
964 __ipoib_mcast_add(dev, nmcast);
966 list_add_tail(&nmcast->list, &priv->multicast_list);
970 set_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
973 /* Remove all of the entries don't exist anymore */
974 list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
975 if (!test_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags) &&
976 !test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
977 ipoib_dbg_mcast(priv, "deleting multicast group %pI6\n",
978 mcast->mcmember.mgid.raw);
980 rb_erase(&mcast->rb_node, &priv->multicast_tree);
982 /* Move to the remove list */
983 list_move_tail(&mcast->list, &remove_list);
987 spin_unlock(&priv->lock);
988 netif_addr_unlock(dev);
989 local_irq_restore(flags);
992 * make sure the in-flight joins have finished before we attempt
995 list_for_each_entry_safe(mcast, tmcast, &remove_list, list)
996 if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
997 wait_for_completion(&mcast->done);
999 ipoib_mcast_remove_list(&remove_list);
1002 * Double check that we are still up
1004 if (test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
1005 spin_lock_irqsave(&priv->lock, flags);
1006 __ipoib_mcast_schedule_join_thread(priv, NULL, 0);
1007 spin_unlock_irqrestore(&priv->lock, flags);
1011 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
1013 struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev)
1015 struct ipoib_mcast_iter *iter;
1017 iter = kmalloc(sizeof *iter, GFP_KERNEL);
1022 memset(iter->mgid.raw, 0, 16);
1024 if (ipoib_mcast_iter_next(iter)) {
1032 int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter)
1034 struct ipoib_dev_priv *priv = netdev_priv(iter->dev);
1036 struct ipoib_mcast *mcast;
1039 spin_lock_irq(&priv->lock);
1041 n = rb_first(&priv->multicast_tree);
1044 mcast = rb_entry(n, struct ipoib_mcast, rb_node);
1046 if (memcmp(iter->mgid.raw, mcast->mcmember.mgid.raw,
1047 sizeof (union ib_gid)) < 0) {
1048 iter->mgid = mcast->mcmember.mgid;
1049 iter->created = mcast->created;
1050 iter->queuelen = skb_queue_len(&mcast->pkt_queue);
1051 iter->complete = !!mcast->ah;
1052 iter->send_only = !!(mcast->flags & (1 << IPOIB_MCAST_FLAG_SENDONLY));
1062 spin_unlock_irq(&priv->lock);
1067 void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
1069 unsigned long *created,
1070 unsigned int *queuelen,
1071 unsigned int *complete,
1072 unsigned int *send_only)
1075 *created = iter->created;
1076 *queuelen = iter->queuelen;
1077 *complete = iter->complete;
1078 *send_only = iter->send_only;
1081 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */