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 static 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 static 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
.wr
.ud
.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);
396 if (mcast
->logcount
++ < 20) {
397 if (status
== -ETIMEDOUT
|| status
== -EAGAIN
) {
398 ipoib_dbg_mcast(priv
, "%smulticast join failed for %pI6, status %d\n",
399 test_bit(IPOIB_MCAST_FLAG_SENDONLY
, &mcast
->flags
) ? "sendonly " : "",
400 mcast
->mcmember
.mgid
.raw
, status
);
402 ipoib_warn(priv
, "%smulticast join failed for %pI6, status %d\n",
403 test_bit(IPOIB_MCAST_FLAG_SENDONLY
, &mcast
->flags
) ? "sendonly " : "",
404 mcast
->mcmember
.mgid
.raw
, status
);
408 if (test_bit(IPOIB_MCAST_FLAG_SENDONLY
, &mcast
->flags
) &&
409 mcast
->backoff
>= 2) {
411 * We only retry sendonly joins once before we drop
412 * the packet and quit trying to deal with the
413 * group. However, we leave the group in the
414 * mcast list as an unjoined group. If we want to
415 * try joining again, we simply queue up a packet
416 * and restart the join thread. The empty queue
417 * is why the join thread ignores this group.
420 netif_tx_lock_bh(dev
);
421 while (!skb_queue_empty(&mcast
->pkt_queue
)) {
422 ++dev
->stats
.tx_dropped
;
423 dev_kfree_skb_any(skb_dequeue(&mcast
->pkt_queue
));
425 netif_tx_unlock_bh(dev
);
427 spin_lock_irq(&priv
->lock
);
428 /* Requeue this join task with a backoff delay */
429 __ipoib_mcast_schedule_join_thread(priv
, mcast
, 1);
434 spin_lock_irq(&priv
->lock
);
437 * Make sure to set mcast->mc before we clear the busy flag to avoid
438 * racing with code that checks for BUSY before checking mcast->mc
443 mcast
->mc
= multicast
;
444 clear_bit(IPOIB_MCAST_FLAG_BUSY
, &mcast
->flags
);
445 spin_unlock_irq(&priv
->lock
);
446 complete(&mcast
->done
);
451 static void ipoib_mcast_join(struct net_device
*dev
, struct ipoib_mcast
*mcast
,
454 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
455 struct ib_sa_multicast
*multicast
;
456 struct ib_sa_mcmember_rec rec
= {
459 ib_sa_comp_mask comp_mask
;
462 ipoib_dbg_mcast(priv
, "joining MGID %pI6\n", mcast
->mcmember
.mgid
.raw
);
464 rec
.mgid
= mcast
->mcmember
.mgid
;
465 rec
.port_gid
= priv
->local_gid
;
466 rec
.pkey
= cpu_to_be16(priv
->pkey
);
469 IB_SA_MCMEMBER_REC_MGID
|
470 IB_SA_MCMEMBER_REC_PORT_GID
|
471 IB_SA_MCMEMBER_REC_PKEY
|
472 IB_SA_MCMEMBER_REC_JOIN_STATE
;
476 IB_SA_MCMEMBER_REC_QKEY
|
477 IB_SA_MCMEMBER_REC_MTU_SELECTOR
|
478 IB_SA_MCMEMBER_REC_MTU
|
479 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS
|
480 IB_SA_MCMEMBER_REC_RATE_SELECTOR
|
481 IB_SA_MCMEMBER_REC_RATE
|
482 IB_SA_MCMEMBER_REC_SL
|
483 IB_SA_MCMEMBER_REC_FLOW_LABEL
|
484 IB_SA_MCMEMBER_REC_HOP_LIMIT
;
486 rec
.qkey
= priv
->broadcast
->mcmember
.qkey
;
487 rec
.mtu_selector
= IB_SA_EQ
;
488 rec
.mtu
= priv
->broadcast
->mcmember
.mtu
;
489 rec
.traffic_class
= priv
->broadcast
->mcmember
.traffic_class
;
490 rec
.rate_selector
= IB_SA_EQ
;
491 rec
.rate
= priv
->broadcast
->mcmember
.rate
;
492 rec
.sl
= priv
->broadcast
->mcmember
.sl
;
493 rec
.flow_label
= priv
->broadcast
->mcmember
.flow_label
;
494 rec
.hop_limit
= priv
->broadcast
->mcmember
.hop_limit
;
497 multicast
= ib_sa_join_multicast(&ipoib_sa_client
, priv
->ca
, priv
->port
,
498 &rec
, comp_mask
, GFP_KERNEL
,
499 ipoib_mcast_join_complete
, mcast
);
500 if (IS_ERR(multicast
)) {
501 ret
= PTR_ERR(multicast
);
502 ipoib_warn(priv
, "ib_sa_join_multicast failed, status %d\n", ret
);
503 spin_lock_irq(&priv
->lock
);
504 /* Requeue this join task with a backoff delay */
505 __ipoib_mcast_schedule_join_thread(priv
, mcast
, 1);
506 clear_bit(IPOIB_MCAST_FLAG_BUSY
, &mcast
->flags
);
507 spin_unlock_irq(&priv
->lock
);
508 complete(&mcast
->done
);
512 void ipoib_mcast_join_task(struct work_struct
*work
)
514 struct ipoib_dev_priv
*priv
=
515 container_of(work
, struct ipoib_dev_priv
, mcast_task
.work
);
516 struct net_device
*dev
= priv
->dev
;
517 struct ib_port_attr port_attr
;
518 unsigned long delay_until
= 0;
519 struct ipoib_mcast
*mcast
= NULL
;
522 if (!test_bit(IPOIB_FLAG_OPER_UP
, &priv
->flags
))
525 if (ib_query_port(priv
->ca
, priv
->port
, &port_attr
) ||
526 port_attr
.state
!= IB_PORT_ACTIVE
) {
527 ipoib_dbg(priv
, "port state is not ACTIVE (state = %d) suspending join task\n",
531 priv
->local_lid
= port_attr
.lid
;
533 if (ib_query_gid(priv
->ca
, priv
->port
, 0, &priv
->local_gid
))
534 ipoib_warn(priv
, "ib_query_gid() failed\n");
536 memcpy(priv
->dev
->dev_addr
+ 4, priv
->local_gid
.raw
, sizeof (union ib_gid
));
538 spin_lock_irq(&priv
->lock
);
539 if (!test_bit(IPOIB_FLAG_OPER_UP
, &priv
->flags
))
542 if (!priv
->broadcast
) {
543 struct ipoib_mcast
*broadcast
;
545 broadcast
= ipoib_mcast_alloc(dev
, 0);
547 ipoib_warn(priv
, "failed to allocate broadcast group\n");
549 * Restart us after a 1 second delay to retry
550 * creating our broadcast group and attaching to
551 * it. Until this succeeds, this ipoib dev is
552 * completely stalled (multicast wise).
554 __ipoib_mcast_schedule_join_thread(priv
, NULL
, 1);
558 memcpy(broadcast
->mcmember
.mgid
.raw
, priv
->dev
->broadcast
+ 4,
559 sizeof (union ib_gid
));
560 priv
->broadcast
= broadcast
;
562 __ipoib_mcast_add(dev
, priv
->broadcast
);
565 if (!test_bit(IPOIB_MCAST_FLAG_ATTACHED
, &priv
->broadcast
->flags
)) {
566 if (IS_ERR_OR_NULL(priv
->broadcast
->mc
) &&
567 !test_bit(IPOIB_MCAST_FLAG_BUSY
, &priv
->broadcast
->flags
)) {
568 mcast
= priv
->broadcast
;
570 if (mcast
->backoff
> 1 &&
571 time_before(jiffies
, mcast
->delay_until
)) {
572 delay_until
= mcast
->delay_until
;
580 * We'll never get here until the broadcast group is both allocated
583 list_for_each_entry(mcast
, &priv
->multicast_list
, list
) {
584 if (IS_ERR_OR_NULL(mcast
->mc
) &&
585 !test_bit(IPOIB_MCAST_FLAG_BUSY
, &mcast
->flags
) &&
586 (!test_bit(IPOIB_MCAST_FLAG_SENDONLY
, &mcast
->flags
) ||
587 !skb_queue_empty(&mcast
->pkt_queue
))) {
588 if (mcast
->backoff
== 1 ||
589 time_after_eq(jiffies
, mcast
->delay_until
)) {
590 /* Found the next unjoined group */
591 init_completion(&mcast
->done
);
592 set_bit(IPOIB_MCAST_FLAG_BUSY
, &mcast
->flags
);
593 if (test_bit(IPOIB_MCAST_FLAG_SENDONLY
, &mcast
->flags
))
597 spin_unlock_irq(&priv
->lock
);
598 ipoib_mcast_join(dev
, mcast
, create
);
599 spin_lock_irq(&priv
->lock
);
600 } else if (!delay_until
||
601 time_before(mcast
->delay_until
, delay_until
))
602 delay_until
= mcast
->delay_until
;
607 ipoib_dbg_mcast(priv
, "successfully started all multicast joins\n");
611 cancel_delayed_work(&priv
->mcast_task
);
612 queue_delayed_work(priv
->wq
, &priv
->mcast_task
,
613 delay_until
- jiffies
);
616 init_completion(&mcast
->done
);
617 set_bit(IPOIB_MCAST_FLAG_BUSY
, &mcast
->flags
);
619 spin_unlock_irq(&priv
->lock
);
621 ipoib_mcast_join(dev
, mcast
, create
);
624 int ipoib_mcast_start_thread(struct net_device
*dev
)
626 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
629 ipoib_dbg_mcast(priv
, "starting multicast thread\n");
631 spin_lock_irqsave(&priv
->lock
, flags
);
632 __ipoib_mcast_schedule_join_thread(priv
, NULL
, 0);
633 spin_unlock_irqrestore(&priv
->lock
, flags
);
638 int ipoib_mcast_stop_thread(struct net_device
*dev
)
640 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
643 ipoib_dbg_mcast(priv
, "stopping multicast thread\n");
645 spin_lock_irqsave(&priv
->lock
, flags
);
646 cancel_delayed_work(&priv
->mcast_task
);
647 spin_unlock_irqrestore(&priv
->lock
, flags
);
649 flush_workqueue(priv
->wq
);
654 static int ipoib_mcast_leave(struct net_device
*dev
, struct ipoib_mcast
*mcast
)
656 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
659 if (test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY
, &mcast
->flags
))
660 ipoib_warn(priv
, "ipoib_mcast_leave on an in-flight join\n");
662 if (!IS_ERR_OR_NULL(mcast
->mc
))
663 ib_sa_free_multicast(mcast
->mc
);
665 if (test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED
, &mcast
->flags
)) {
666 ipoib_dbg_mcast(priv
, "leaving MGID %pI6\n",
667 mcast
->mcmember
.mgid
.raw
);
669 /* Remove ourselves from the multicast group */
670 ret
= ib_detach_mcast(priv
->qp
, &mcast
->mcmember
.mgid
,
671 be16_to_cpu(mcast
->mcmember
.mlid
));
673 ipoib_warn(priv
, "ib_detach_mcast failed (result = %d)\n", ret
);
674 } else if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY
, &mcast
->flags
))
675 ipoib_dbg(priv
, "leaving with no mcmember but not a "
681 void ipoib_mcast_send(struct net_device
*dev
, u8
*daddr
, struct sk_buff
*skb
)
683 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
684 struct ipoib_mcast
*mcast
;
686 void *mgid
= daddr
+ 4;
688 spin_lock_irqsave(&priv
->lock
, flags
);
690 if (!test_bit(IPOIB_FLAG_OPER_UP
, &priv
->flags
) ||
692 !test_bit(IPOIB_MCAST_FLAG_ATTACHED
, &priv
->broadcast
->flags
)) {
693 ++dev
->stats
.tx_dropped
;
694 dev_kfree_skb_any(skb
);
698 mcast
= __ipoib_mcast_find(dev
, mgid
);
699 if (!mcast
|| !mcast
->ah
) {
701 /* Let's create a new send only group now */
702 ipoib_dbg_mcast(priv
, "setting up send only multicast group for %pI6\n",
705 mcast
= ipoib_mcast_alloc(dev
, 0);
707 ipoib_warn(priv
, "unable to allocate memory "
708 "for multicast structure\n");
709 ++dev
->stats
.tx_dropped
;
710 dev_kfree_skb_any(skb
);
714 set_bit(IPOIB_MCAST_FLAG_SENDONLY
, &mcast
->flags
);
715 memcpy(mcast
->mcmember
.mgid
.raw
, mgid
,
716 sizeof (union ib_gid
));
717 __ipoib_mcast_add(dev
, mcast
);
718 list_add_tail(&mcast
->list
, &priv
->multicast_list
);
720 if (skb_queue_len(&mcast
->pkt_queue
) < IPOIB_MAX_MCAST_QUEUE
)
721 skb_queue_tail(&mcast
->pkt_queue
, skb
);
723 ++dev
->stats
.tx_dropped
;
724 dev_kfree_skb_any(skb
);
726 if (!test_bit(IPOIB_MCAST_FLAG_BUSY
, &mcast
->flags
)) {
727 __ipoib_mcast_schedule_join_thread(priv
, NULL
, 0);
730 struct ipoib_neigh
*neigh
;
732 spin_unlock_irqrestore(&priv
->lock
, flags
);
733 neigh
= ipoib_neigh_get(dev
, daddr
);
734 spin_lock_irqsave(&priv
->lock
, flags
);
736 neigh
= ipoib_neigh_alloc(daddr
, dev
);
738 kref_get(&mcast
->ah
->ref
);
739 neigh
->ah
= mcast
->ah
;
740 list_add_tail(&neigh
->list
, &mcast
->neigh_list
);
743 spin_unlock_irqrestore(&priv
->lock
, flags
);
744 ipoib_send(dev
, skb
, mcast
->ah
, IB_MULTICAST_QPN
);
746 ipoib_neigh_put(neigh
);
751 spin_unlock_irqrestore(&priv
->lock
, flags
);
754 void ipoib_mcast_dev_flush(struct net_device
*dev
)
756 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
757 LIST_HEAD(remove_list
);
758 struct ipoib_mcast
*mcast
, *tmcast
;
761 ipoib_dbg_mcast(priv
, "flushing multicast list\n");
763 spin_lock_irqsave(&priv
->lock
, flags
);
765 list_for_each_entry_safe(mcast
, tmcast
, &priv
->multicast_list
, list
) {
766 list_del(&mcast
->list
);
767 rb_erase(&mcast
->rb_node
, &priv
->multicast_tree
);
768 list_add_tail(&mcast
->list
, &remove_list
);
771 if (priv
->broadcast
) {
772 rb_erase(&priv
->broadcast
->rb_node
, &priv
->multicast_tree
);
773 list_add_tail(&priv
->broadcast
->list
, &remove_list
);
774 priv
->broadcast
= NULL
;
777 spin_unlock_irqrestore(&priv
->lock
, flags
);
780 * make sure the in-flight joins have finished before we attempt
783 list_for_each_entry_safe(mcast
, tmcast
, &remove_list
, list
)
784 if (test_bit(IPOIB_MCAST_FLAG_BUSY
, &mcast
->flags
))
785 wait_for_completion(&mcast
->done
);
787 list_for_each_entry_safe(mcast
, tmcast
, &remove_list
, list
) {
788 ipoib_mcast_leave(dev
, mcast
);
789 ipoib_mcast_free(mcast
);
793 static int ipoib_mcast_addr_is_valid(const u8
*addr
, const u8
*broadcast
)
795 /* reserved QPN, prefix, scope */
796 if (memcmp(addr
, broadcast
, 6))
798 /* signature lower, pkey */
799 if (memcmp(addr
+ 7, broadcast
+ 7, 3))
804 void ipoib_mcast_restart_task(struct work_struct
*work
)
806 struct ipoib_dev_priv
*priv
=
807 container_of(work
, struct ipoib_dev_priv
, restart_task
);
808 struct net_device
*dev
= priv
->dev
;
809 struct netdev_hw_addr
*ha
;
810 struct ipoib_mcast
*mcast
, *tmcast
;
811 LIST_HEAD(remove_list
);
813 struct ib_sa_mcmember_rec rec
;
815 if (!test_bit(IPOIB_FLAG_OPER_UP
, &priv
->flags
))
817 * shortcut...on shutdown flush is called next, just
818 * let it do all the work
822 ipoib_dbg_mcast(priv
, "restarting multicast task\n");
824 local_irq_save(flags
);
825 netif_addr_lock(dev
);
826 spin_lock(&priv
->lock
);
829 * Unfortunately, the networking core only gives us a list of all of
830 * the multicast hardware addresses. We need to figure out which ones
831 * are new and which ones have been removed
834 /* Clear out the found flag */
835 list_for_each_entry(mcast
, &priv
->multicast_list
, list
)
836 clear_bit(IPOIB_MCAST_FLAG_FOUND
, &mcast
->flags
);
838 /* Mark all of the entries that are found or don't exist */
839 netdev_for_each_mc_addr(ha
, dev
) {
842 if (!ipoib_mcast_addr_is_valid(ha
->addr
, dev
->broadcast
))
845 memcpy(mgid
.raw
, ha
->addr
+ 4, sizeof mgid
);
847 mcast
= __ipoib_mcast_find(dev
, &mgid
);
848 if (!mcast
|| test_bit(IPOIB_MCAST_FLAG_SENDONLY
, &mcast
->flags
)) {
849 struct ipoib_mcast
*nmcast
;
851 /* ignore group which is directly joined by userspace */
852 if (test_bit(IPOIB_FLAG_UMCAST
, &priv
->flags
) &&
853 !ib_sa_get_mcmember_rec(priv
->ca
, priv
->port
, &mgid
, &rec
)) {
854 ipoib_dbg_mcast(priv
, "ignoring multicast entry for mgid %pI6\n",
859 /* Not found or send-only group, let's add a new entry */
860 ipoib_dbg_mcast(priv
, "adding multicast entry for mgid %pI6\n",
863 nmcast
= ipoib_mcast_alloc(dev
, 0);
865 ipoib_warn(priv
, "unable to allocate memory for multicast structure\n");
869 set_bit(IPOIB_MCAST_FLAG_FOUND
, &nmcast
->flags
);
871 nmcast
->mcmember
.mgid
= mgid
;
874 /* Destroy the send only entry */
875 list_move_tail(&mcast
->list
, &remove_list
);
877 rb_replace_node(&mcast
->rb_node
,
879 &priv
->multicast_tree
);
881 __ipoib_mcast_add(dev
, nmcast
);
883 list_add_tail(&nmcast
->list
, &priv
->multicast_list
);
887 set_bit(IPOIB_MCAST_FLAG_FOUND
, &mcast
->flags
);
890 /* Remove all of the entries don't exist anymore */
891 list_for_each_entry_safe(mcast
, tmcast
, &priv
->multicast_list
, list
) {
892 if (!test_bit(IPOIB_MCAST_FLAG_FOUND
, &mcast
->flags
) &&
893 !test_bit(IPOIB_MCAST_FLAG_SENDONLY
, &mcast
->flags
)) {
894 ipoib_dbg_mcast(priv
, "deleting multicast group %pI6\n",
895 mcast
->mcmember
.mgid
.raw
);
897 rb_erase(&mcast
->rb_node
, &priv
->multicast_tree
);
899 /* Move to the remove list */
900 list_move_tail(&mcast
->list
, &remove_list
);
904 spin_unlock(&priv
->lock
);
905 netif_addr_unlock(dev
);
906 local_irq_restore(flags
);
909 * make sure the in-flight joins have finished before we attempt
912 list_for_each_entry_safe(mcast
, tmcast
, &remove_list
, list
)
913 if (test_bit(IPOIB_MCAST_FLAG_BUSY
, &mcast
->flags
))
914 wait_for_completion(&mcast
->done
);
916 list_for_each_entry_safe(mcast
, tmcast
, &remove_list
, list
) {
917 ipoib_mcast_leave(mcast
->dev
, mcast
);
918 ipoib_mcast_free(mcast
);
922 * Double check that we are still up
924 if (test_bit(IPOIB_FLAG_OPER_UP
, &priv
->flags
)) {
925 spin_lock_irqsave(&priv
->lock
, flags
);
926 __ipoib_mcast_schedule_join_thread(priv
, NULL
, 0);
927 spin_unlock_irqrestore(&priv
->lock
, flags
);
931 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
933 struct ipoib_mcast_iter
*ipoib_mcast_iter_init(struct net_device
*dev
)
935 struct ipoib_mcast_iter
*iter
;
937 iter
= kmalloc(sizeof *iter
, GFP_KERNEL
);
942 memset(iter
->mgid
.raw
, 0, 16);
944 if (ipoib_mcast_iter_next(iter
)) {
952 int ipoib_mcast_iter_next(struct ipoib_mcast_iter
*iter
)
954 struct ipoib_dev_priv
*priv
= netdev_priv(iter
->dev
);
956 struct ipoib_mcast
*mcast
;
959 spin_lock_irq(&priv
->lock
);
961 n
= rb_first(&priv
->multicast_tree
);
964 mcast
= rb_entry(n
, struct ipoib_mcast
, rb_node
);
966 if (memcmp(iter
->mgid
.raw
, mcast
->mcmember
.mgid
.raw
,
967 sizeof (union ib_gid
)) < 0) {
968 iter
->mgid
= mcast
->mcmember
.mgid
;
969 iter
->created
= mcast
->created
;
970 iter
->queuelen
= skb_queue_len(&mcast
->pkt_queue
);
971 iter
->complete
= !!mcast
->ah
;
972 iter
->send_only
= !!(mcast
->flags
& (1 << IPOIB_MCAST_FLAG_SENDONLY
));
982 spin_unlock_irq(&priv
->lock
);
987 void ipoib_mcast_iter_read(struct ipoib_mcast_iter
*iter
,
989 unsigned long *created
,
990 unsigned int *queuelen
,
991 unsigned int *complete
,
992 unsigned int *send_only
)
995 *created
= iter
->created
;
996 *queuelen
= iter
->queuelen
;
997 *complete
= iter
->complete
;
998 *send_only
= iter
->send_only
;
1001 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */