2 * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4 * Copyright (c) 2004 Voltaire, Inc. All rights reserved.
6 * This software is available to you under a choice of one of two
7 * licenses. You may choose to be licensed under the terms of the GNU
8 * General Public License (GPL) Version 2, available from the file
9 * COPYING in the main directory of this source tree, or the
10 * OpenIB.org BSD license below:
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
20 * - Redistributions in binary form must reproduce the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer in the documentation and/or other materials
23 * provided with the distribution.
25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
35 #include <linux/skbuff.h>
36 #include <linux/rtnetlink.h>
37 #include <linux/moduleparam.h>
40 #include <linux/igmp.h>
41 #include <linux/inetdevice.h>
42 #include <linux/delay.h>
43 #include <linux/completion.h>
44 #include <linux/slab.h>
50 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
51 static int mcast_debug_level;
53 module_param(mcast_debug_level, int, 0644);
54 MODULE_PARM_DESC(mcast_debug_level,
55 "Enable multicast debug tracing if > 0");
58 static DEFINE_MUTEX(mcast_mutex);
60 struct ipoib_mcast_iter {
61 struct net_device *dev;
63 unsigned long created;
64 unsigned int queuelen;
65 unsigned int complete;
66 unsigned int send_only;
69 static void ipoib_mcast_free(struct ipoib_mcast *mcast)
71 struct net_device *dev = mcast->dev;
74 ipoib_dbg_mcast(netdev_priv(dev), "deleting multicast group %pI6\n",
75 mcast->mcmember.mgid.raw);
77 /* remove all neigh connected to this mcast */
78 ipoib_del_neighs_by_gid(dev, mcast->mcmember.mgid.raw);
81 ipoib_put_ah(mcast->ah);
83 while (!skb_queue_empty(&mcast->pkt_queue)) {
85 dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
88 netif_tx_lock_bh(dev);
89 dev->stats.tx_dropped += tx_dropped;
90 netif_tx_unlock_bh(dev);
95 static struct ipoib_mcast *ipoib_mcast_alloc(struct net_device *dev,
98 struct ipoib_mcast *mcast;
100 mcast = kzalloc(sizeof *mcast, can_sleep ? GFP_KERNEL : GFP_ATOMIC);
105 mcast->created = jiffies;
108 INIT_LIST_HEAD(&mcast->list);
109 INIT_LIST_HEAD(&mcast->neigh_list);
110 skb_queue_head_init(&mcast->pkt_queue);
115 static struct ipoib_mcast *__ipoib_mcast_find(struct net_device *dev, void *mgid)
117 struct ipoib_dev_priv *priv = netdev_priv(dev);
118 struct rb_node *n = priv->multicast_tree.rb_node;
121 struct ipoib_mcast *mcast;
124 mcast = rb_entry(n, struct ipoib_mcast, rb_node);
126 ret = memcmp(mgid, mcast->mcmember.mgid.raw,
127 sizeof (union ib_gid));
139 static int __ipoib_mcast_add(struct net_device *dev, struct ipoib_mcast *mcast)
141 struct ipoib_dev_priv *priv = netdev_priv(dev);
142 struct rb_node **n = &priv->multicast_tree.rb_node, *pn = NULL;
145 struct ipoib_mcast *tmcast;
149 tmcast = rb_entry(pn, struct ipoib_mcast, rb_node);
151 ret = memcmp(mcast->mcmember.mgid.raw, tmcast->mcmember.mgid.raw,
152 sizeof (union ib_gid));
161 rb_link_node(&mcast->rb_node, pn, n);
162 rb_insert_color(&mcast->rb_node, &priv->multicast_tree);
167 static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
168 struct ib_sa_mcmember_rec *mcmember)
170 struct net_device *dev = mcast->dev;
171 struct ipoib_dev_priv *priv = netdev_priv(dev);
176 mcast->mcmember = *mcmember;
178 /* Set the multicast MTU and cached Q_Key before we attach if it's
179 * the broadcast group.
181 if (!memcmp(mcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
182 sizeof (union ib_gid))) {
183 spin_lock_irq(&priv->lock);
184 if (!priv->broadcast) {
185 spin_unlock_irq(&priv->lock);
188 priv->mcast_mtu = IPOIB_UD_MTU(ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu));
189 priv->qkey = be32_to_cpu(priv->broadcast->mcmember.qkey);
190 spin_unlock_irq(&priv->lock);
191 priv->tx_wr.wr.ud.remote_qkey = priv->qkey;
195 if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
196 if (test_and_set_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
197 ipoib_warn(priv, "multicast group %pI6 already attached\n",
198 mcast->mcmember.mgid.raw);
203 ret = ipoib_mcast_attach(dev, be16_to_cpu(mcast->mcmember.mlid),
204 &mcast->mcmember.mgid, set_qkey);
206 ipoib_warn(priv, "couldn't attach QP to multicast group %pI6\n",
207 mcast->mcmember.mgid.raw);
209 clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags);
215 struct ib_ah_attr av = {
216 .dlid = be16_to_cpu(mcast->mcmember.mlid),
217 .port_num = priv->port,
218 .sl = mcast->mcmember.sl,
219 .ah_flags = IB_AH_GRH,
220 .static_rate = mcast->mcmember.rate,
222 .flow_label = be32_to_cpu(mcast->mcmember.flow_label),
223 .hop_limit = mcast->mcmember.hop_limit,
225 .traffic_class = mcast->mcmember.traffic_class
228 av.grh.dgid = mcast->mcmember.mgid;
230 ah = ipoib_create_ah(dev, priv->pd, &av);
232 ipoib_warn(priv, "ib_address_create failed %ld\n",
234 /* use original error */
237 spin_lock_irq(&priv->lock);
239 spin_unlock_irq(&priv->lock);
241 ipoib_dbg_mcast(priv, "MGID %pI6 AV %p, LID 0x%04x, SL %d\n",
242 mcast->mcmember.mgid.raw,
244 be16_to_cpu(mcast->mcmember.mlid),
249 /* actually send any queued packets */
250 netif_tx_lock_bh(dev);
251 while (!skb_queue_empty(&mcast->pkt_queue)) {
252 struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue);
254 netif_tx_unlock_bh(dev);
257 if (dev_queue_xmit(skb))
258 ipoib_warn(priv, "dev_queue_xmit failed to requeue packet\n");
260 netif_tx_lock_bh(dev);
262 netif_tx_unlock_bh(dev);
268 ipoib_mcast_sendonly_join_complete(int status,
269 struct ib_sa_multicast *multicast)
271 struct ipoib_mcast *mcast = multicast->context;
272 struct net_device *dev = mcast->dev;
274 /* We trap for port events ourselves. */
275 if (status == -ENETRESET)
279 status = ipoib_mcast_join_finish(mcast, &multicast->rec);
282 if (mcast->logcount++ < 20)
283 ipoib_dbg_mcast(netdev_priv(dev), "multicast join failed for %pI6, status %d\n",
284 mcast->mcmember.mgid.raw, status);
286 /* Flush out any queued packets */
287 netif_tx_lock_bh(dev);
288 while (!skb_queue_empty(&mcast->pkt_queue)) {
289 ++dev->stats.tx_dropped;
290 dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
292 netif_tx_unlock_bh(dev);
294 /* Clear the busy flag so we try again */
295 status = test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY,
301 static int ipoib_mcast_sendonly_join(struct ipoib_mcast *mcast)
303 struct net_device *dev = mcast->dev;
304 struct ipoib_dev_priv *priv = netdev_priv(dev);
305 struct ib_sa_mcmember_rec rec = {
306 #if 0 /* Some SMs don't support send-only yet */
314 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
315 ipoib_dbg_mcast(priv, "device shutting down, no multicast joins\n");
319 if (test_and_set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) {
320 ipoib_dbg_mcast(priv, "multicast entry busy, skipping\n");
324 rec.mgid = mcast->mcmember.mgid;
325 rec.port_gid = priv->local_gid;
326 rec.pkey = cpu_to_be16(priv->pkey);
328 mcast->mc = ib_sa_join_multicast(&ipoib_sa_client, priv->ca,
330 IB_SA_MCMEMBER_REC_MGID |
331 IB_SA_MCMEMBER_REC_PORT_GID |
332 IB_SA_MCMEMBER_REC_PKEY |
333 IB_SA_MCMEMBER_REC_JOIN_STATE,
335 ipoib_mcast_sendonly_join_complete,
337 if (IS_ERR(mcast->mc)) {
338 ret = PTR_ERR(mcast->mc);
339 clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
340 ipoib_warn(priv, "ib_sa_join_multicast failed (ret = %d)\n",
343 ipoib_dbg_mcast(priv, "no multicast record for %pI6, starting join\n",
344 mcast->mcmember.mgid.raw);
350 void ipoib_mcast_carrier_on_task(struct work_struct *work)
352 struct ipoib_dev_priv *priv = container_of(work, struct ipoib_dev_priv,
354 struct ib_port_attr attr;
356 if (ib_query_port(priv->ca, priv->port, &attr) ||
357 attr.state != IB_PORT_ACTIVE) {
358 ipoib_dbg(priv, "Keeping carrier off until IB port is active\n");
363 * Take rtnl_lock to avoid racing with ipoib_stop() and
364 * turning the carrier back on while a device is being
365 * removed. However, ipoib_stop() will attempt to flush
366 * the workqueue while holding the rtnl lock, so loop
367 * on trylock until either we get the lock or we see
368 * FLAG_OPER_UP go away as that signals that we are bailing
369 * and can safely ignore the carrier on work.
371 while (!rtnl_trylock()) {
372 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
377 if (!ipoib_cm_admin_enabled(priv->dev))
378 dev_set_mtu(priv->dev, min(priv->mcast_mtu, priv->admin_mtu));
379 netif_carrier_on(priv->dev);
383 static int ipoib_mcast_join_complete(int status,
384 struct ib_sa_multicast *multicast)
386 struct ipoib_mcast *mcast = multicast->context;
387 struct net_device *dev = mcast->dev;
388 struct ipoib_dev_priv *priv = netdev_priv(dev);
390 ipoib_dbg_mcast(priv, "join completion for %pI6 (status %d)\n",
391 mcast->mcmember.mgid.raw, status);
393 /* We trap for port events ourselves. */
394 if (status == -ENETRESET) {
400 status = ipoib_mcast_join_finish(mcast, &multicast->rec);
404 mutex_lock(&mcast_mutex);
405 if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
406 queue_delayed_work(priv->wq, &priv->mcast_task, 0);
407 mutex_unlock(&mcast_mutex);
410 * Defer carrier on work to priv->wq to avoid a
411 * deadlock on rtnl_lock here.
413 if (mcast == priv->broadcast)
414 queue_work(priv->wq, &priv->carrier_on_task);
420 if (mcast->logcount++ < 20) {
421 if (status == -ETIMEDOUT || status == -EAGAIN) {
422 ipoib_dbg_mcast(priv, "multicast join failed for %pI6, status %d\n",
423 mcast->mcmember.mgid.raw, status);
425 ipoib_warn(priv, "multicast join failed for %pI6, status %d\n",
426 mcast->mcmember.mgid.raw, status);
431 if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
432 mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
434 /* Clear the busy flag so we try again */
435 status = test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
437 mutex_lock(&mcast_mutex);
438 spin_lock_irq(&priv->lock);
439 if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
440 queue_delayed_work(priv->wq, &priv->mcast_task,
441 mcast->backoff * HZ);
442 spin_unlock_irq(&priv->lock);
443 mutex_unlock(&mcast_mutex);
445 complete(&mcast->done);
449 static void ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast,
452 struct ipoib_dev_priv *priv = netdev_priv(dev);
453 struct ib_sa_mcmember_rec rec = {
456 ib_sa_comp_mask comp_mask;
459 ipoib_dbg_mcast(priv, "joining MGID %pI6\n", mcast->mcmember.mgid.raw);
461 rec.mgid = mcast->mcmember.mgid;
462 rec.port_gid = priv->local_gid;
463 rec.pkey = cpu_to_be16(priv->pkey);
466 IB_SA_MCMEMBER_REC_MGID |
467 IB_SA_MCMEMBER_REC_PORT_GID |
468 IB_SA_MCMEMBER_REC_PKEY |
469 IB_SA_MCMEMBER_REC_JOIN_STATE;
473 IB_SA_MCMEMBER_REC_QKEY |
474 IB_SA_MCMEMBER_REC_MTU_SELECTOR |
475 IB_SA_MCMEMBER_REC_MTU |
476 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS |
477 IB_SA_MCMEMBER_REC_RATE_SELECTOR |
478 IB_SA_MCMEMBER_REC_RATE |
479 IB_SA_MCMEMBER_REC_SL |
480 IB_SA_MCMEMBER_REC_FLOW_LABEL |
481 IB_SA_MCMEMBER_REC_HOP_LIMIT;
483 rec.qkey = priv->broadcast->mcmember.qkey;
484 rec.mtu_selector = IB_SA_EQ;
485 rec.mtu = priv->broadcast->mcmember.mtu;
486 rec.traffic_class = priv->broadcast->mcmember.traffic_class;
487 rec.rate_selector = IB_SA_EQ;
488 rec.rate = priv->broadcast->mcmember.rate;
489 rec.sl = priv->broadcast->mcmember.sl;
490 rec.flow_label = priv->broadcast->mcmember.flow_label;
491 rec.hop_limit = priv->broadcast->mcmember.hop_limit;
494 set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
495 init_completion(&mcast->done);
496 set_bit(IPOIB_MCAST_JOIN_STARTED, &mcast->flags);
498 mcast->mc = ib_sa_join_multicast(&ipoib_sa_client, priv->ca, priv->port,
499 &rec, comp_mask, GFP_KERNEL,
500 ipoib_mcast_join_complete, mcast);
501 if (IS_ERR(mcast->mc)) {
502 clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
503 complete(&mcast->done);
504 ret = PTR_ERR(mcast->mc);
505 ipoib_warn(priv, "ib_sa_join_multicast failed, status %d\n", ret);
508 if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
509 mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
511 mutex_lock(&mcast_mutex);
512 if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
513 queue_delayed_work(priv->wq, &priv->mcast_task,
514 mcast->backoff * HZ);
515 mutex_unlock(&mcast_mutex);
519 void ipoib_mcast_join_task(struct work_struct *work)
521 struct ipoib_dev_priv *priv =
522 container_of(work, struct ipoib_dev_priv, mcast_task.work);
523 struct net_device *dev = priv->dev;
524 struct ib_port_attr port_attr;
526 if (!test_bit(IPOIB_MCAST_RUN, &priv->flags))
529 if (ib_query_port(priv->ca, priv->port, &port_attr) ||
530 port_attr.state != IB_PORT_ACTIVE) {
531 ipoib_dbg(priv, "port state is not ACTIVE (state = %d) suspending join task\n",
535 priv->local_lid = port_attr.lid;
537 if (ib_query_gid(priv->ca, priv->port, 0, &priv->local_gid))
538 ipoib_warn(priv, "ib_query_gid() failed\n");
540 memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw, sizeof (union ib_gid));
542 if (!priv->broadcast) {
543 struct ipoib_mcast *broadcast;
545 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
548 broadcast = ipoib_mcast_alloc(dev, 1);
550 ipoib_warn(priv, "failed to allocate broadcast group\n");
551 mutex_lock(&mcast_mutex);
552 if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
553 queue_delayed_work(priv->wq, &priv->mcast_task,
555 mutex_unlock(&mcast_mutex);
559 spin_lock_irq(&priv->lock);
560 memcpy(broadcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
561 sizeof (union ib_gid));
562 priv->broadcast = broadcast;
564 __ipoib_mcast_add(dev, priv->broadcast);
565 spin_unlock_irq(&priv->lock);
568 if (!test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
569 if (!test_bit(IPOIB_MCAST_FLAG_BUSY, &priv->broadcast->flags))
570 ipoib_mcast_join(dev, priv->broadcast, 0);
575 struct ipoib_mcast *mcast = NULL;
577 spin_lock_irq(&priv->lock);
578 list_for_each_entry(mcast, &priv->multicast_list, list) {
579 if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)
580 && !test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)
581 && !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
582 /* Found the next unjoined group */
586 spin_unlock_irq(&priv->lock);
588 if (&mcast->list == &priv->multicast_list) {
593 ipoib_mcast_join(dev, mcast, 1);
597 ipoib_dbg_mcast(priv, "successfully joined all multicast groups\n");
599 clear_bit(IPOIB_MCAST_RUN, &priv->flags);
602 int ipoib_mcast_start_thread(struct net_device *dev)
604 struct ipoib_dev_priv *priv = netdev_priv(dev);
606 ipoib_dbg_mcast(priv, "starting multicast thread\n");
608 mutex_lock(&mcast_mutex);
609 if (!test_and_set_bit(IPOIB_MCAST_RUN, &priv->flags))
610 queue_delayed_work(priv->wq, &priv->mcast_task, 0);
611 mutex_unlock(&mcast_mutex);
616 int ipoib_mcast_stop_thread(struct net_device *dev, int flush)
618 struct ipoib_dev_priv *priv = netdev_priv(dev);
620 ipoib_dbg_mcast(priv, "stopping multicast thread\n");
622 mutex_lock(&mcast_mutex);
623 clear_bit(IPOIB_MCAST_RUN, &priv->flags);
624 cancel_delayed_work(&priv->mcast_task);
625 mutex_unlock(&mcast_mutex);
628 flush_workqueue(priv->wq);
633 static int ipoib_mcast_leave(struct net_device *dev, struct ipoib_mcast *mcast)
635 struct ipoib_dev_priv *priv = netdev_priv(dev);
638 if (test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
639 ib_sa_free_multicast(mcast->mc);
641 if (test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
642 ipoib_dbg_mcast(priv, "leaving MGID %pI6\n",
643 mcast->mcmember.mgid.raw);
645 /* Remove ourselves from the multicast group */
646 ret = ib_detach_mcast(priv->qp, &mcast->mcmember.mgid,
647 be16_to_cpu(mcast->mcmember.mlid));
649 ipoib_warn(priv, "ib_detach_mcast failed (result = %d)\n", ret);
655 void ipoib_mcast_send(struct net_device *dev, u8 *daddr, struct sk_buff *skb)
657 struct ipoib_dev_priv *priv = netdev_priv(dev);
658 struct ipoib_mcast *mcast;
660 void *mgid = daddr + 4;
662 spin_lock_irqsave(&priv->lock, flags);
664 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags) ||
666 !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
667 ++dev->stats.tx_dropped;
668 dev_kfree_skb_any(skb);
672 mcast = __ipoib_mcast_find(dev, mgid);
674 /* Let's create a new send only group now */
675 ipoib_dbg_mcast(priv, "setting up send only multicast group for %pI6\n",
678 mcast = ipoib_mcast_alloc(dev, 0);
680 ipoib_warn(priv, "unable to allocate memory for "
681 "multicast structure\n");
682 ++dev->stats.tx_dropped;
683 dev_kfree_skb_any(skb);
687 set_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags);
688 memcpy(mcast->mcmember.mgid.raw, mgid, sizeof (union ib_gid));
689 __ipoib_mcast_add(dev, mcast);
690 list_add_tail(&mcast->list, &priv->multicast_list);
694 if (skb_queue_len(&mcast->pkt_queue) < IPOIB_MAX_MCAST_QUEUE)
695 skb_queue_tail(&mcast->pkt_queue, skb);
697 ++dev->stats.tx_dropped;
698 dev_kfree_skb_any(skb);
701 if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
702 ipoib_dbg_mcast(priv, "no address vector, "
703 "but multicast join already started\n");
704 else if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags))
705 ipoib_mcast_sendonly_join(mcast);
708 * If lookup completes between here and out:, don't
709 * want to send packet twice.
715 if (mcast && mcast->ah) {
716 struct ipoib_neigh *neigh;
718 spin_unlock_irqrestore(&priv->lock, flags);
719 neigh = ipoib_neigh_get(dev, daddr);
720 spin_lock_irqsave(&priv->lock, flags);
722 neigh = ipoib_neigh_alloc(daddr, dev);
724 kref_get(&mcast->ah->ref);
725 neigh->ah = mcast->ah;
726 list_add_tail(&neigh->list, &mcast->neigh_list);
729 spin_unlock_irqrestore(&priv->lock, flags);
730 ipoib_send(dev, skb, mcast->ah, IB_MULTICAST_QPN);
732 ipoib_neigh_put(neigh);
737 spin_unlock_irqrestore(&priv->lock, flags);
740 void ipoib_mcast_dev_flush(struct net_device *dev)
742 struct ipoib_dev_priv *priv = netdev_priv(dev);
743 LIST_HEAD(remove_list);
744 struct ipoib_mcast *mcast, *tmcast;
747 ipoib_dbg_mcast(priv, "flushing multicast list\n");
749 spin_lock_irqsave(&priv->lock, flags);
751 list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
752 list_del(&mcast->list);
753 rb_erase(&mcast->rb_node, &priv->multicast_tree);
754 list_add_tail(&mcast->list, &remove_list);
757 if (priv->broadcast) {
758 rb_erase(&priv->broadcast->rb_node, &priv->multicast_tree);
759 list_add_tail(&priv->broadcast->list, &remove_list);
760 priv->broadcast = NULL;
763 spin_unlock_irqrestore(&priv->lock, flags);
765 /* seperate between the wait to the leave*/
766 list_for_each_entry_safe(mcast, tmcast, &remove_list, list)
767 if (test_bit(IPOIB_MCAST_JOIN_STARTED, &mcast->flags))
768 wait_for_completion(&mcast->done);
770 list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
771 ipoib_mcast_leave(dev, mcast);
772 ipoib_mcast_free(mcast);
776 static int ipoib_mcast_addr_is_valid(const u8 *addr, const u8 *broadcast)
778 /* reserved QPN, prefix, scope */
779 if (memcmp(addr, broadcast, 6))
781 /* signature lower, pkey */
782 if (memcmp(addr + 7, broadcast + 7, 3))
787 void ipoib_mcast_restart_task(struct work_struct *work)
789 struct ipoib_dev_priv *priv =
790 container_of(work, struct ipoib_dev_priv, restart_task);
791 struct net_device *dev = priv->dev;
792 struct netdev_hw_addr *ha;
793 struct ipoib_mcast *mcast, *tmcast;
794 LIST_HEAD(remove_list);
796 struct ib_sa_mcmember_rec rec;
798 ipoib_dbg_mcast(priv, "restarting multicast task\n");
800 ipoib_mcast_stop_thread(dev, 0);
802 local_irq_save(flags);
803 netif_addr_lock(dev);
804 spin_lock(&priv->lock);
807 * Unfortunately, the networking core only gives us a list of all of
808 * the multicast hardware addresses. We need to figure out which ones
809 * are new and which ones have been removed
812 /* Clear out the found flag */
813 list_for_each_entry(mcast, &priv->multicast_list, list)
814 clear_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
816 /* Mark all of the entries that are found or don't exist */
817 netdev_for_each_mc_addr(ha, dev) {
820 if (!ipoib_mcast_addr_is_valid(ha->addr, dev->broadcast))
823 memcpy(mgid.raw, ha->addr + 4, sizeof mgid);
825 mcast = __ipoib_mcast_find(dev, &mgid);
826 if (!mcast || test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
827 struct ipoib_mcast *nmcast;
829 /* ignore group which is directly joined by userspace */
830 if (test_bit(IPOIB_FLAG_UMCAST, &priv->flags) &&
831 !ib_sa_get_mcmember_rec(priv->ca, priv->port, &mgid, &rec)) {
832 ipoib_dbg_mcast(priv, "ignoring multicast entry for mgid %pI6\n",
837 /* Not found or send-only group, let's add a new entry */
838 ipoib_dbg_mcast(priv, "adding multicast entry for mgid %pI6\n",
841 nmcast = ipoib_mcast_alloc(dev, 0);
843 ipoib_warn(priv, "unable to allocate memory for multicast structure\n");
847 set_bit(IPOIB_MCAST_FLAG_FOUND, &nmcast->flags);
849 nmcast->mcmember.mgid = mgid;
852 /* Destroy the send only entry */
853 list_move_tail(&mcast->list, &remove_list);
855 rb_replace_node(&mcast->rb_node,
857 &priv->multicast_tree);
859 __ipoib_mcast_add(dev, nmcast);
861 list_add_tail(&nmcast->list, &priv->multicast_list);
865 set_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
868 /* Remove all of the entries don't exist anymore */
869 list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
870 if (!test_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags) &&
871 !test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
872 ipoib_dbg_mcast(priv, "deleting multicast group %pI6\n",
873 mcast->mcmember.mgid.raw);
875 rb_erase(&mcast->rb_node, &priv->multicast_tree);
877 /* Move to the remove list */
878 list_move_tail(&mcast->list, &remove_list);
882 spin_unlock(&priv->lock);
883 netif_addr_unlock(dev);
884 local_irq_restore(flags);
886 /* We have to cancel outside of the spinlock */
887 list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
888 ipoib_mcast_leave(mcast->dev, mcast);
889 ipoib_mcast_free(mcast);
892 if (test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
893 ipoib_mcast_start_thread(dev);
896 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
898 struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev)
900 struct ipoib_mcast_iter *iter;
902 iter = kmalloc(sizeof *iter, GFP_KERNEL);
907 memset(iter->mgid.raw, 0, 16);
909 if (ipoib_mcast_iter_next(iter)) {
917 int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter)
919 struct ipoib_dev_priv *priv = netdev_priv(iter->dev);
921 struct ipoib_mcast *mcast;
924 spin_lock_irq(&priv->lock);
926 n = rb_first(&priv->multicast_tree);
929 mcast = rb_entry(n, struct ipoib_mcast, rb_node);
931 if (memcmp(iter->mgid.raw, mcast->mcmember.mgid.raw,
932 sizeof (union ib_gid)) < 0) {
933 iter->mgid = mcast->mcmember.mgid;
934 iter->created = mcast->created;
935 iter->queuelen = skb_queue_len(&mcast->pkt_queue);
936 iter->complete = !!mcast->ah;
937 iter->send_only = !!(mcast->flags & (1 << IPOIB_MCAST_FLAG_SENDONLY));
947 spin_unlock_irq(&priv->lock);
952 void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
954 unsigned long *created,
955 unsigned int *queuelen,
956 unsigned int *complete,
957 unsigned int *send_only)
960 *created = iter->created;
961 *queuelen = iter->queuelen;
962 *complete = iter->complete;
963 *send_only = iter->send_only;
966 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */