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);
318 ret = dev_queue_xmit(skb);
320 ipoib_warn(priv, "%s:dev_queue_xmit failed to re-queue packet, ret:%d\n",
322 netif_tx_lock_bh(dev);
324 netif_tx_unlock_bh(dev);
329 void ipoib_mcast_carrier_on_task(struct work_struct *work)
331 struct ipoib_dev_priv *priv = container_of(work, struct ipoib_dev_priv,
333 struct ib_port_attr attr;
336 if (ib_query_port(priv->ca, priv->port, &attr) ||
337 attr.state != IB_PORT_ACTIVE) {
338 ipoib_dbg(priv, "Keeping carrier off until IB port is active\n");
342 * Check if can send sendonly MCG's with sendonly-fullmember join state.
343 * It done here after the successfully join to the broadcast group,
344 * because the broadcast group must always be joined first and is always
345 * re-joined if the SM changes substantially.
347 ret = ipoib_check_sm_sendonly_fullmember_support(priv);
349 pr_debug("%s failed query sm support for sendonly-fullmember (ret: %d)\n",
350 priv->dev->name, ret);
353 * Take rtnl_lock to avoid racing with ipoib_stop() and
354 * turning the carrier back on while a device is being
355 * removed. However, ipoib_stop() will attempt to flush
356 * the workqueue while holding the rtnl lock, so loop
357 * on trylock until either we get the lock or we see
358 * FLAG_OPER_UP go away as that signals that we are bailing
359 * and can safely ignore the carrier on work.
361 while (!rtnl_trylock()) {
362 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
367 if (!ipoib_cm_admin_enabled(priv->dev))
368 dev_set_mtu(priv->dev, min(priv->mcast_mtu, priv->admin_mtu));
369 netif_carrier_on(priv->dev);
373 static int ipoib_mcast_join_complete(int status,
374 struct ib_sa_multicast *multicast)
376 struct ipoib_mcast *mcast = multicast->context;
377 struct net_device *dev = mcast->dev;
378 struct ipoib_dev_priv *priv = netdev_priv(dev);
380 ipoib_dbg_mcast(priv, "%sjoin completion for %pI6 (status %d)\n",
381 test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) ?
383 mcast->mcmember.mgid.raw, status);
385 /* We trap for port events ourselves. */
386 if (status == -ENETRESET) {
392 status = ipoib_mcast_join_finish(mcast, &multicast->rec);
396 mcast->delay_until = jiffies;
399 * Defer carrier on work to priv->wq to avoid a
400 * deadlock on rtnl_lock here. Requeue our multicast
401 * work too, which will end up happening right after
402 * our carrier on task work and will allow us to
403 * send out all of the non-broadcast joins
405 if (mcast == priv->broadcast) {
406 spin_lock_irq(&priv->lock);
407 queue_work(priv->wq, &priv->carrier_on_task);
408 __ipoib_mcast_schedule_join_thread(priv, NULL, 0);
413 test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) &&
416 if (mcast->logcount < 20) {
417 if (status == -ETIMEDOUT || status == -EAGAIN ||
419 ipoib_dbg_mcast(priv, "%smulticast join failed for %pI6, status %d\n",
420 test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) ? "sendonly " : "",
421 mcast->mcmember.mgid.raw, status);
423 ipoib_warn(priv, "%smulticast join failed for %pI6, status %d\n",
424 test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) ? "sendonly " : "",
425 mcast->mcmember.mgid.raw, status);
432 if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) &&
433 mcast->backoff >= 2) {
435 * We only retry sendonly joins once before we drop
436 * the packet and quit trying to deal with the
437 * group. However, we leave the group in the
438 * mcast list as an unjoined group. If we want to
439 * try joining again, we simply queue up a packet
440 * and restart the join thread. The empty queue
441 * is why the join thread ignores this group.
444 netif_tx_lock_bh(dev);
445 while (!skb_queue_empty(&mcast->pkt_queue)) {
446 ++dev->stats.tx_dropped;
447 dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
449 netif_tx_unlock_bh(dev);
451 spin_lock_irq(&priv->lock);
452 /* Requeue this join task with a backoff delay */
453 __ipoib_mcast_schedule_join_thread(priv, mcast, 1);
458 spin_lock_irq(&priv->lock);
461 * Make sure to set mcast->mc before we clear the busy flag to avoid
462 * racing with code that checks for BUSY before checking mcast->mc
467 mcast->mc = multicast;
468 clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
469 spin_unlock_irq(&priv->lock);
470 complete(&mcast->done);
476 * Caller must hold 'priv->lock'
478 static int ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast)
480 struct ipoib_dev_priv *priv = netdev_priv(dev);
481 struct ib_sa_multicast *multicast;
482 struct ib_sa_mcmember_rec rec = {
485 ib_sa_comp_mask comp_mask;
488 if (!priv->broadcast ||
489 !test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
492 ipoib_dbg_mcast(priv, "joining MGID %pI6\n", mcast->mcmember.mgid.raw);
494 rec.mgid = mcast->mcmember.mgid;
495 rec.port_gid = priv->local_gid;
496 rec.pkey = cpu_to_be16(priv->pkey);
499 IB_SA_MCMEMBER_REC_MGID |
500 IB_SA_MCMEMBER_REC_PORT_GID |
501 IB_SA_MCMEMBER_REC_PKEY |
502 IB_SA_MCMEMBER_REC_JOIN_STATE;
504 if (mcast != priv->broadcast) {
507 * The MGID MUST use the same P_Key, Q_Key, SL, MTU,
508 * and HopLimit as those used in the broadcast-GID. The rest
509 * of attributes SHOULD follow the values used in the
510 * broadcast-GID as well.
513 IB_SA_MCMEMBER_REC_QKEY |
514 IB_SA_MCMEMBER_REC_MTU_SELECTOR |
515 IB_SA_MCMEMBER_REC_MTU |
516 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS |
517 IB_SA_MCMEMBER_REC_RATE_SELECTOR |
518 IB_SA_MCMEMBER_REC_RATE |
519 IB_SA_MCMEMBER_REC_SL |
520 IB_SA_MCMEMBER_REC_FLOW_LABEL |
521 IB_SA_MCMEMBER_REC_HOP_LIMIT;
523 rec.qkey = priv->broadcast->mcmember.qkey;
524 rec.mtu_selector = IB_SA_EQ;
525 rec.mtu = priv->broadcast->mcmember.mtu;
526 rec.traffic_class = priv->broadcast->mcmember.traffic_class;
527 rec.rate_selector = IB_SA_EQ;
528 rec.rate = priv->broadcast->mcmember.rate;
529 rec.sl = priv->broadcast->mcmember.sl;
530 rec.flow_label = priv->broadcast->mcmember.flow_label;
531 rec.hop_limit = priv->broadcast->mcmember.hop_limit;
534 * Send-only IB Multicast joins work at the core IB layer but
535 * require specific SM support.
536 * We can use such joins here only if the current SM supports that feature.
537 * However, if not, we emulate an Ethernet multicast send,
538 * which does not require a multicast subscription and will
539 * still send properly. The most appropriate thing to
540 * do is to create the group if it doesn't exist as that
541 * most closely emulates the behavior, from a user space
542 * application perspective, of Ethernet multicast operation.
544 if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) &&
545 priv->sm_fullmember_sendonly_support)
546 /* SM supports sendonly-fullmember, otherwise fallback to full-member */
547 rec.join_state = SENDONLY_FULLMEMBER_JOIN;
549 spin_unlock_irq(&priv->lock);
551 multicast = ib_sa_join_multicast(&ipoib_sa_client, priv->ca, priv->port,
552 &rec, comp_mask, GFP_KERNEL,
553 ipoib_mcast_join_complete, mcast);
554 spin_lock_irq(&priv->lock);
555 if (IS_ERR(multicast)) {
556 ret = PTR_ERR(multicast);
557 ipoib_warn(priv, "ib_sa_join_multicast failed, status %d\n", ret);
558 /* Requeue this join task with a backoff delay */
559 __ipoib_mcast_schedule_join_thread(priv, mcast, 1);
560 clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
561 spin_unlock_irq(&priv->lock);
562 complete(&mcast->done);
563 spin_lock_irq(&priv->lock);
568 void ipoib_mcast_join_task(struct work_struct *work)
570 struct ipoib_dev_priv *priv =
571 container_of(work, struct ipoib_dev_priv, mcast_task.work);
572 struct net_device *dev = priv->dev;
573 struct ib_port_attr port_attr;
574 unsigned long delay_until = 0;
575 struct ipoib_mcast *mcast = NULL;
577 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
580 if (ib_query_port(priv->ca, priv->port, &port_attr)) {
581 ipoib_dbg(priv, "ib_query_port() failed\n");
584 if (port_attr.state != IB_PORT_ACTIVE) {
585 ipoib_dbg(priv, "port state is not ACTIVE (state = %d) suspending join task\n",
589 priv->local_lid = port_attr.lid;
590 netif_addr_lock_bh(dev);
592 if (!test_bit(IPOIB_FLAG_DEV_ADDR_SET, &priv->flags)) {
593 netif_addr_unlock_bh(dev);
596 netif_addr_unlock_bh(dev);
598 spin_lock_irq(&priv->lock);
599 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
602 if (!priv->broadcast) {
603 struct ipoib_mcast *broadcast;
605 broadcast = ipoib_mcast_alloc(dev, 0);
607 ipoib_warn(priv, "failed to allocate broadcast group\n");
609 * Restart us after a 1 second delay to retry
610 * creating our broadcast group and attaching to
611 * it. Until this succeeds, this ipoib dev is
612 * completely stalled (multicast wise).
614 __ipoib_mcast_schedule_join_thread(priv, NULL, 1);
618 memcpy(broadcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
619 sizeof (union ib_gid));
620 priv->broadcast = broadcast;
622 __ipoib_mcast_add(dev, priv->broadcast);
625 if (!test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
626 if (IS_ERR_OR_NULL(priv->broadcast->mc) &&
627 !test_bit(IPOIB_MCAST_FLAG_BUSY, &priv->broadcast->flags)) {
628 mcast = priv->broadcast;
629 if (mcast->backoff > 1 &&
630 time_before(jiffies, mcast->delay_until)) {
631 delay_until = mcast->delay_until;
639 * We'll never get here until the broadcast group is both allocated
642 list_for_each_entry(mcast, &priv->multicast_list, list) {
643 if (IS_ERR_OR_NULL(mcast->mc) &&
644 !test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags) &&
645 (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) ||
646 !skb_queue_empty(&mcast->pkt_queue))) {
647 if (mcast->backoff == 1 ||
648 time_after_eq(jiffies, mcast->delay_until)) {
649 /* Found the next unjoined group */
650 init_completion(&mcast->done);
651 set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
652 if (ipoib_mcast_join(dev, mcast)) {
653 spin_unlock_irq(&priv->lock);
656 } else if (!delay_until ||
657 time_before(mcast->delay_until, delay_until))
658 delay_until = mcast->delay_until;
663 ipoib_dbg_mcast(priv, "successfully started all multicast joins\n");
667 cancel_delayed_work(&priv->mcast_task);
668 queue_delayed_work(priv->wq, &priv->mcast_task,
669 delay_until - jiffies);
672 init_completion(&mcast->done);
673 set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
674 ipoib_mcast_join(dev, mcast);
676 spin_unlock_irq(&priv->lock);
679 void ipoib_mcast_start_thread(struct net_device *dev)
681 struct ipoib_dev_priv *priv = netdev_priv(dev);
684 ipoib_dbg_mcast(priv, "starting multicast thread\n");
686 spin_lock_irqsave(&priv->lock, flags);
687 __ipoib_mcast_schedule_join_thread(priv, NULL, 0);
688 spin_unlock_irqrestore(&priv->lock, flags);
691 int ipoib_mcast_stop_thread(struct net_device *dev)
693 struct ipoib_dev_priv *priv = netdev_priv(dev);
696 ipoib_dbg_mcast(priv, "stopping multicast thread\n");
698 spin_lock_irqsave(&priv->lock, flags);
699 cancel_delayed_work(&priv->mcast_task);
700 spin_unlock_irqrestore(&priv->lock, flags);
702 flush_workqueue(priv->wq);
707 static int ipoib_mcast_leave(struct net_device *dev, struct ipoib_mcast *mcast)
709 struct ipoib_dev_priv *priv = netdev_priv(dev);
712 if (test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
713 ipoib_warn(priv, "ipoib_mcast_leave on an in-flight join\n");
715 if (!IS_ERR_OR_NULL(mcast->mc))
716 ib_sa_free_multicast(mcast->mc);
718 if (test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
719 ipoib_dbg_mcast(priv, "leaving MGID %pI6\n",
720 mcast->mcmember.mgid.raw);
722 /* Remove ourselves from the multicast group */
723 ret = ib_detach_mcast(priv->qp, &mcast->mcmember.mgid,
724 be16_to_cpu(mcast->mcmember.mlid));
726 ipoib_warn(priv, "ib_detach_mcast failed (result = %d)\n", ret);
727 } else if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags))
728 ipoib_dbg(priv, "leaving with no mcmember but not a "
735 * Check if the multicast group is sendonly. If so remove it from the maps
736 * and add to the remove list
738 void ipoib_check_and_add_mcast_sendonly(struct ipoib_dev_priv *priv, u8 *mgid,
739 struct list_head *remove_list)
741 /* Is this multicast ? */
743 struct ipoib_mcast *mcast = __ipoib_mcast_find(priv->dev, mgid);
745 if (mcast && test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
746 list_del(&mcast->list);
747 rb_erase(&mcast->rb_node, &priv->multicast_tree);
748 list_add_tail(&mcast->list, remove_list);
753 void ipoib_mcast_remove_list(struct list_head *remove_list)
755 struct ipoib_mcast *mcast, *tmcast;
757 list_for_each_entry_safe(mcast, tmcast, remove_list, list) {
758 ipoib_mcast_leave(mcast->dev, mcast);
759 ipoib_mcast_free(mcast);
763 void ipoib_mcast_send(struct net_device *dev, u8 *daddr, struct sk_buff *skb)
765 struct ipoib_dev_priv *priv = netdev_priv(dev);
766 struct ipoib_mcast *mcast;
768 void *mgid = daddr + 4;
770 spin_lock_irqsave(&priv->lock, flags);
772 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags) ||
774 !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
775 ++dev->stats.tx_dropped;
776 dev_kfree_skb_any(skb);
780 mcast = __ipoib_mcast_find(dev, mgid);
781 if (!mcast || !mcast->ah) {
783 /* Let's create a new send only group now */
784 ipoib_dbg_mcast(priv, "setting up send only multicast group for %pI6\n",
787 mcast = ipoib_mcast_alloc(dev, 0);
789 ipoib_warn(priv, "unable to allocate memory "
790 "for multicast structure\n");
791 ++dev->stats.tx_dropped;
792 dev_kfree_skb_any(skb);
796 set_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags);
797 memcpy(mcast->mcmember.mgid.raw, mgid,
798 sizeof (union ib_gid));
799 __ipoib_mcast_add(dev, mcast);
800 list_add_tail(&mcast->list, &priv->multicast_list);
802 if (skb_queue_len(&mcast->pkt_queue) < IPOIB_MAX_MCAST_QUEUE) {
803 /* put pseudoheader back on for next time */
804 skb_push(skb, sizeof(struct ipoib_pseudo_header));
805 skb_queue_tail(&mcast->pkt_queue, skb);
807 ++dev->stats.tx_dropped;
808 dev_kfree_skb_any(skb);
810 if (!test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) {
811 __ipoib_mcast_schedule_join_thread(priv, NULL, 0);
814 struct ipoib_neigh *neigh;
816 spin_unlock_irqrestore(&priv->lock, flags);
817 neigh = ipoib_neigh_get(dev, daddr);
818 spin_lock_irqsave(&priv->lock, flags);
820 neigh = ipoib_neigh_alloc(daddr, dev);
822 kref_get(&mcast->ah->ref);
823 neigh->ah = mcast->ah;
824 list_add_tail(&neigh->list, &mcast->neigh_list);
827 spin_unlock_irqrestore(&priv->lock, flags);
828 ipoib_send(dev, skb, mcast->ah, IB_MULTICAST_QPN);
830 ipoib_neigh_put(neigh);
835 spin_unlock_irqrestore(&priv->lock, flags);
838 void ipoib_mcast_dev_flush(struct net_device *dev)
840 struct ipoib_dev_priv *priv = netdev_priv(dev);
841 LIST_HEAD(remove_list);
842 struct ipoib_mcast *mcast, *tmcast;
845 ipoib_dbg_mcast(priv, "flushing multicast list\n");
847 spin_lock_irqsave(&priv->lock, flags);
849 list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
850 list_del(&mcast->list);
851 rb_erase(&mcast->rb_node, &priv->multicast_tree);
852 list_add_tail(&mcast->list, &remove_list);
855 if (priv->broadcast) {
856 rb_erase(&priv->broadcast->rb_node, &priv->multicast_tree);
857 list_add_tail(&priv->broadcast->list, &remove_list);
858 priv->broadcast = NULL;
861 spin_unlock_irqrestore(&priv->lock, flags);
864 * make sure the in-flight joins have finished before we attempt
867 list_for_each_entry_safe(mcast, tmcast, &remove_list, list)
868 if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
869 wait_for_completion(&mcast->done);
871 ipoib_mcast_remove_list(&remove_list);
874 static int ipoib_mcast_addr_is_valid(const u8 *addr, const u8 *broadcast)
876 /* reserved QPN, prefix, scope */
877 if (memcmp(addr, broadcast, 6))
879 /* signature lower, pkey */
880 if (memcmp(addr + 7, broadcast + 7, 3))
885 void ipoib_mcast_restart_task(struct work_struct *work)
887 struct ipoib_dev_priv *priv =
888 container_of(work, struct ipoib_dev_priv, restart_task);
889 struct net_device *dev = priv->dev;
890 struct netdev_hw_addr *ha;
891 struct ipoib_mcast *mcast, *tmcast;
892 LIST_HEAD(remove_list);
894 struct ib_sa_mcmember_rec rec;
896 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
898 * shortcut...on shutdown flush is called next, just
899 * let it do all the work
903 ipoib_dbg_mcast(priv, "restarting multicast task\n");
905 local_irq_save(flags);
906 netif_addr_lock(dev);
907 spin_lock(&priv->lock);
910 * Unfortunately, the networking core only gives us a list of all of
911 * the multicast hardware addresses. We need to figure out which ones
912 * are new and which ones have been removed
915 /* Clear out the found flag */
916 list_for_each_entry(mcast, &priv->multicast_list, list)
917 clear_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
919 /* Mark all of the entries that are found or don't exist */
920 netdev_for_each_mc_addr(ha, dev) {
923 if (!ipoib_mcast_addr_is_valid(ha->addr, dev->broadcast))
926 memcpy(mgid.raw, ha->addr + 4, sizeof mgid);
928 mcast = __ipoib_mcast_find(dev, &mgid);
929 if (!mcast || test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
930 struct ipoib_mcast *nmcast;
932 /* ignore group which is directly joined by userspace */
933 if (test_bit(IPOIB_FLAG_UMCAST, &priv->flags) &&
934 !ib_sa_get_mcmember_rec(priv->ca, priv->port, &mgid, &rec)) {
935 ipoib_dbg_mcast(priv, "ignoring multicast entry for mgid %pI6\n",
940 /* Not found or send-only group, let's add a new entry */
941 ipoib_dbg_mcast(priv, "adding multicast entry for mgid %pI6\n",
944 nmcast = ipoib_mcast_alloc(dev, 0);
946 ipoib_warn(priv, "unable to allocate memory for multicast structure\n");
950 set_bit(IPOIB_MCAST_FLAG_FOUND, &nmcast->flags);
952 nmcast->mcmember.mgid = mgid;
955 /* Destroy the send only entry */
956 list_move_tail(&mcast->list, &remove_list);
958 rb_replace_node(&mcast->rb_node,
960 &priv->multicast_tree);
962 __ipoib_mcast_add(dev, nmcast);
964 list_add_tail(&nmcast->list, &priv->multicast_list);
968 set_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
971 /* Remove all of the entries don't exist anymore */
972 list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
973 if (!test_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags) &&
974 !test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
975 ipoib_dbg_mcast(priv, "deleting multicast group %pI6\n",
976 mcast->mcmember.mgid.raw);
978 rb_erase(&mcast->rb_node, &priv->multicast_tree);
980 /* Move to the remove list */
981 list_move_tail(&mcast->list, &remove_list);
985 spin_unlock(&priv->lock);
986 netif_addr_unlock(dev);
987 local_irq_restore(flags);
990 * make sure the in-flight joins have finished before we attempt
993 list_for_each_entry_safe(mcast, tmcast, &remove_list, list)
994 if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
995 wait_for_completion(&mcast->done);
997 ipoib_mcast_remove_list(&remove_list);
1000 * Double check that we are still up
1002 if (test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
1003 spin_lock_irqsave(&priv->lock, flags);
1004 __ipoib_mcast_schedule_join_thread(priv, NULL, 0);
1005 spin_unlock_irqrestore(&priv->lock, flags);
1009 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
1011 struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev)
1013 struct ipoib_mcast_iter *iter;
1015 iter = kmalloc(sizeof *iter, GFP_KERNEL);
1020 memset(iter->mgid.raw, 0, 16);
1022 if (ipoib_mcast_iter_next(iter)) {
1030 int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter)
1032 struct ipoib_dev_priv *priv = netdev_priv(iter->dev);
1034 struct ipoib_mcast *mcast;
1037 spin_lock_irq(&priv->lock);
1039 n = rb_first(&priv->multicast_tree);
1042 mcast = rb_entry(n, struct ipoib_mcast, rb_node);
1044 if (memcmp(iter->mgid.raw, mcast->mcmember.mgid.raw,
1045 sizeof (union ib_gid)) < 0) {
1046 iter->mgid = mcast->mcmember.mgid;
1047 iter->created = mcast->created;
1048 iter->queuelen = skb_queue_len(&mcast->pkt_queue);
1049 iter->complete = !!mcast->ah;
1050 iter->send_only = !!(mcast->flags & (1 << IPOIB_MCAST_FLAG_SENDONLY));
1060 spin_unlock_irq(&priv->lock);
1065 void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
1067 unsigned long *created,
1068 unsigned int *queuelen,
1069 unsigned int *complete,
1070 unsigned int *send_only)
1073 *created = iter->created;
1074 *queuelen = iter->queuelen;
1075 *complete = iter->complete;
1076 *send_only = iter->send_only;
1079 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */