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>
39 #include <linux/igmp.h>
40 #include <linux/inetdevice.h>
41 #include <linux/delay.h>
42 #include <linux/completion.h>
48 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
49 static int mcast_debug_level
;
51 module_param(mcast_debug_level
, int, 0644);
52 MODULE_PARM_DESC(mcast_debug_level
,
53 "Enable multicast debug tracing if > 0");
56 static DEFINE_MUTEX(mcast_mutex
);
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 static void ipoib_mcast_free(struct ipoib_mcast
*mcast
)
69 struct net_device
*dev
= mcast
->dev
;
70 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
71 struct ipoib_neigh
*neigh
, *tmp
;
75 ipoib_dbg_mcast(netdev_priv(dev
),
76 "deleting multicast group " IPOIB_GID_FMT
"\n",
77 IPOIB_GID_ARG(mcast
->mcmember
.mgid
));
79 spin_lock_irqsave(&priv
->lock
, flags
);
81 list_for_each_entry_safe(neigh
, tmp
, &mcast
->neigh_list
, list
) {
83 * It's safe to call ipoib_put_ah() inside priv->lock
84 * here, because we know that mcast->ah will always
85 * hold one more reference, so ipoib_put_ah() will
86 * never do more than decrement the ref count.
89 ipoib_put_ah(neigh
->ah
);
90 ipoib_neigh_free(dev
, neigh
);
93 spin_unlock_irqrestore(&priv
->lock
, flags
);
96 ipoib_put_ah(mcast
->ah
);
98 while (!skb_queue_empty(&mcast
->pkt_queue
)) {
100 dev_kfree_skb_any(skb_dequeue(&mcast
->pkt_queue
));
103 spin_lock_irqsave(&priv
->tx_lock
, flags
);
104 dev
->stats
.tx_dropped
+= tx_dropped
;
105 spin_unlock_irqrestore(&priv
->tx_lock
, flags
);
110 static struct ipoib_mcast
*ipoib_mcast_alloc(struct net_device
*dev
,
113 struct ipoib_mcast
*mcast
;
115 mcast
= kzalloc(sizeof *mcast
, can_sleep
? GFP_KERNEL
: GFP_ATOMIC
);
120 mcast
->created
= jiffies
;
123 INIT_LIST_HEAD(&mcast
->list
);
124 INIT_LIST_HEAD(&mcast
->neigh_list
);
125 skb_queue_head_init(&mcast
->pkt_queue
);
130 static struct ipoib_mcast
*__ipoib_mcast_find(struct net_device
*dev
, void *mgid
)
132 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
133 struct rb_node
*n
= priv
->multicast_tree
.rb_node
;
136 struct ipoib_mcast
*mcast
;
139 mcast
= rb_entry(n
, struct ipoib_mcast
, rb_node
);
141 ret
= memcmp(mgid
, mcast
->mcmember
.mgid
.raw
,
142 sizeof (union ib_gid
));
154 static int __ipoib_mcast_add(struct net_device
*dev
, struct ipoib_mcast
*mcast
)
156 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
157 struct rb_node
**n
= &priv
->multicast_tree
.rb_node
, *pn
= NULL
;
160 struct ipoib_mcast
*tmcast
;
164 tmcast
= rb_entry(pn
, struct ipoib_mcast
, rb_node
);
166 ret
= memcmp(mcast
->mcmember
.mgid
.raw
, tmcast
->mcmember
.mgid
.raw
,
167 sizeof (union ib_gid
));
176 rb_link_node(&mcast
->rb_node
, pn
, n
);
177 rb_insert_color(&mcast
->rb_node
, &priv
->multicast_tree
);
182 static int ipoib_mcast_join_finish(struct ipoib_mcast
*mcast
,
183 struct ib_sa_mcmember_rec
*mcmember
)
185 struct net_device
*dev
= mcast
->dev
;
186 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
190 mcast
->mcmember
= *mcmember
;
192 /* Set the cached Q_Key before we attach if it's the broadcast group */
193 if (!memcmp(mcast
->mcmember
.mgid
.raw
, priv
->dev
->broadcast
+ 4,
194 sizeof (union ib_gid
))) {
195 spin_lock_irq(&priv
->lock
);
196 if (!priv
->broadcast
) {
197 spin_unlock_irq(&priv
->lock
);
200 priv
->qkey
= be32_to_cpu(priv
->broadcast
->mcmember
.qkey
);
201 spin_unlock_irq(&priv
->lock
);
202 priv
->tx_wr
.wr
.ud
.remote_qkey
= priv
->qkey
;
205 if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY
, &mcast
->flags
)) {
206 if (test_and_set_bit(IPOIB_MCAST_FLAG_ATTACHED
, &mcast
->flags
)) {
207 ipoib_warn(priv
, "multicast group " IPOIB_GID_FMT
208 " already attached\n",
209 IPOIB_GID_ARG(mcast
->mcmember
.mgid
));
214 ret
= ipoib_mcast_attach(dev
, be16_to_cpu(mcast
->mcmember
.mlid
),
215 &mcast
->mcmember
.mgid
);
217 ipoib_warn(priv
, "couldn't attach QP to multicast group "
219 IPOIB_GID_ARG(mcast
->mcmember
.mgid
));
221 clear_bit(IPOIB_MCAST_FLAG_ATTACHED
, &mcast
->flags
);
227 struct ib_ah_attr av
= {
228 .dlid
= be16_to_cpu(mcast
->mcmember
.mlid
),
229 .port_num
= priv
->port
,
230 .sl
= mcast
->mcmember
.sl
,
231 .ah_flags
= IB_AH_GRH
,
232 .static_rate
= mcast
->mcmember
.rate
,
234 .flow_label
= be32_to_cpu(mcast
->mcmember
.flow_label
),
235 .hop_limit
= mcast
->mcmember
.hop_limit
,
237 .traffic_class
= mcast
->mcmember
.traffic_class
240 av
.grh
.dgid
= mcast
->mcmember
.mgid
;
242 ah
= ipoib_create_ah(dev
, priv
->pd
, &av
);
244 ipoib_warn(priv
, "ib_address_create failed\n");
246 spin_lock_irq(&priv
->lock
);
248 spin_unlock_irq(&priv
->lock
);
250 ipoib_dbg_mcast(priv
, "MGID " IPOIB_GID_FMT
251 " AV %p, LID 0x%04x, SL %d\n",
252 IPOIB_GID_ARG(mcast
->mcmember
.mgid
),
254 be16_to_cpu(mcast
->mcmember
.mlid
),
259 /* actually send any queued packets */
260 spin_lock_irq(&priv
->tx_lock
);
261 while (!skb_queue_empty(&mcast
->pkt_queue
)) {
262 struct sk_buff
*skb
= skb_dequeue(&mcast
->pkt_queue
);
263 spin_unlock_irq(&priv
->tx_lock
);
267 if (!skb
->dst
|| !skb
->dst
->neighbour
) {
268 /* put pseudoheader back on for next time */
269 skb_push(skb
, sizeof (struct ipoib_pseudoheader
));
272 if (dev_queue_xmit(skb
))
273 ipoib_warn(priv
, "dev_queue_xmit failed to requeue packet\n");
274 spin_lock_irq(&priv
->tx_lock
);
276 spin_unlock_irq(&priv
->tx_lock
);
282 ipoib_mcast_sendonly_join_complete(int status
,
283 struct ib_sa_multicast
*multicast
)
285 struct ipoib_mcast
*mcast
= multicast
->context
;
286 struct net_device
*dev
= mcast
->dev
;
287 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
289 /* We trap for port events ourselves. */
290 if (status
== -ENETRESET
)
294 status
= ipoib_mcast_join_finish(mcast
, &multicast
->rec
);
297 if (mcast
->logcount
++ < 20)
298 ipoib_dbg_mcast(netdev_priv(dev
), "multicast join failed for "
299 IPOIB_GID_FMT
", status %d\n",
300 IPOIB_GID_ARG(mcast
->mcmember
.mgid
), status
);
302 /* Flush out any queued packets */
303 spin_lock_irq(&priv
->tx_lock
);
304 while (!skb_queue_empty(&mcast
->pkt_queue
)) {
305 ++dev
->stats
.tx_dropped
;
306 dev_kfree_skb_any(skb_dequeue(&mcast
->pkt_queue
));
308 spin_unlock_irq(&priv
->tx_lock
);
310 /* Clear the busy flag so we try again */
311 status
= test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY
,
317 static int ipoib_mcast_sendonly_join(struct ipoib_mcast
*mcast
)
319 struct net_device
*dev
= mcast
->dev
;
320 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
321 struct ib_sa_mcmember_rec rec
= {
322 #if 0 /* Some SMs don't support send-only yet */
330 if (!test_bit(IPOIB_FLAG_OPER_UP
, &priv
->flags
)) {
331 ipoib_dbg_mcast(priv
, "device shutting down, no multicast joins\n");
335 if (test_and_set_bit(IPOIB_MCAST_FLAG_BUSY
, &mcast
->flags
)) {
336 ipoib_dbg_mcast(priv
, "multicast entry busy, skipping\n");
340 rec
.mgid
= mcast
->mcmember
.mgid
;
341 rec
.port_gid
= priv
->local_gid
;
342 rec
.pkey
= cpu_to_be16(priv
->pkey
);
344 mcast
->mc
= ib_sa_join_multicast(&ipoib_sa_client
, priv
->ca
,
346 IB_SA_MCMEMBER_REC_MGID
|
347 IB_SA_MCMEMBER_REC_PORT_GID
|
348 IB_SA_MCMEMBER_REC_PKEY
|
349 IB_SA_MCMEMBER_REC_JOIN_STATE
,
351 ipoib_mcast_sendonly_join_complete
,
353 if (IS_ERR(mcast
->mc
)) {
354 ret
= PTR_ERR(mcast
->mc
);
355 clear_bit(IPOIB_MCAST_FLAG_BUSY
, &mcast
->flags
);
356 ipoib_warn(priv
, "ib_sa_join_multicast failed (ret = %d)\n",
359 ipoib_dbg_mcast(priv
, "no multicast record for " IPOIB_GID_FMT
361 IPOIB_GID_ARG(mcast
->mcmember
.mgid
));
367 static int ipoib_mcast_join_complete(int status
,
368 struct ib_sa_multicast
*multicast
)
370 struct ipoib_mcast
*mcast
= multicast
->context
;
371 struct net_device
*dev
= mcast
->dev
;
372 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
374 ipoib_dbg_mcast(priv
, "join completion for " IPOIB_GID_FMT
376 IPOIB_GID_ARG(mcast
->mcmember
.mgid
), status
);
378 /* We trap for port events ourselves. */
379 if (status
== -ENETRESET
)
383 status
= ipoib_mcast_join_finish(mcast
, &multicast
->rec
);
387 mutex_lock(&mcast_mutex
);
388 if (test_bit(IPOIB_MCAST_RUN
, &priv
->flags
))
389 queue_delayed_work(ipoib_workqueue
,
390 &priv
->mcast_task
, 0);
391 mutex_unlock(&mcast_mutex
);
393 if (mcast
== priv
->broadcast
)
394 netif_carrier_on(dev
);
399 if (mcast
->logcount
++ < 20) {
400 if (status
== -ETIMEDOUT
) {
401 ipoib_dbg_mcast(priv
, "multicast join failed for " IPOIB_GID_FMT
403 IPOIB_GID_ARG(mcast
->mcmember
.mgid
),
406 ipoib_warn(priv
, "multicast join failed for "
407 IPOIB_GID_FMT
", status %d\n",
408 IPOIB_GID_ARG(mcast
->mcmember
.mgid
),
414 if (mcast
->backoff
> IPOIB_MAX_BACKOFF_SECONDS
)
415 mcast
->backoff
= IPOIB_MAX_BACKOFF_SECONDS
;
417 /* Clear the busy flag so we try again */
418 status
= test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY
, &mcast
->flags
);
420 mutex_lock(&mcast_mutex
);
421 spin_lock_irq(&priv
->lock
);
422 if (test_bit(IPOIB_MCAST_RUN
, &priv
->flags
))
423 queue_delayed_work(ipoib_workqueue
, &priv
->mcast_task
,
424 mcast
->backoff
* HZ
);
425 spin_unlock_irq(&priv
->lock
);
426 mutex_unlock(&mcast_mutex
);
431 static void ipoib_mcast_join(struct net_device
*dev
, struct ipoib_mcast
*mcast
,
434 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
435 struct ib_sa_mcmember_rec rec
= {
438 ib_sa_comp_mask comp_mask
;
441 ipoib_dbg_mcast(priv
, "joining MGID " IPOIB_GID_FMT
"\n",
442 IPOIB_GID_ARG(mcast
->mcmember
.mgid
));
444 rec
.mgid
= mcast
->mcmember
.mgid
;
445 rec
.port_gid
= priv
->local_gid
;
446 rec
.pkey
= cpu_to_be16(priv
->pkey
);
449 IB_SA_MCMEMBER_REC_MGID
|
450 IB_SA_MCMEMBER_REC_PORT_GID
|
451 IB_SA_MCMEMBER_REC_PKEY
|
452 IB_SA_MCMEMBER_REC_JOIN_STATE
;
456 IB_SA_MCMEMBER_REC_QKEY
|
457 IB_SA_MCMEMBER_REC_MTU_SELECTOR
|
458 IB_SA_MCMEMBER_REC_MTU
|
459 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS
|
460 IB_SA_MCMEMBER_REC_RATE_SELECTOR
|
461 IB_SA_MCMEMBER_REC_RATE
|
462 IB_SA_MCMEMBER_REC_SL
|
463 IB_SA_MCMEMBER_REC_FLOW_LABEL
|
464 IB_SA_MCMEMBER_REC_HOP_LIMIT
;
466 rec
.qkey
= priv
->broadcast
->mcmember
.qkey
;
467 rec
.mtu_selector
= IB_SA_EQ
;
468 rec
.mtu
= priv
->broadcast
->mcmember
.mtu
;
469 rec
.traffic_class
= priv
->broadcast
->mcmember
.traffic_class
;
470 rec
.rate_selector
= IB_SA_EQ
;
471 rec
.rate
= priv
->broadcast
->mcmember
.rate
;
472 rec
.sl
= priv
->broadcast
->mcmember
.sl
;
473 rec
.flow_label
= priv
->broadcast
->mcmember
.flow_label
;
474 rec
.hop_limit
= priv
->broadcast
->mcmember
.hop_limit
;
477 set_bit(IPOIB_MCAST_FLAG_BUSY
, &mcast
->flags
);
478 mcast
->mc
= ib_sa_join_multicast(&ipoib_sa_client
, priv
->ca
, priv
->port
,
479 &rec
, comp_mask
, GFP_KERNEL
,
480 ipoib_mcast_join_complete
, mcast
);
481 if (IS_ERR(mcast
->mc
)) {
482 clear_bit(IPOIB_MCAST_FLAG_BUSY
, &mcast
->flags
);
483 ret
= PTR_ERR(mcast
->mc
);
484 ipoib_warn(priv
, "ib_sa_join_multicast failed, status %d\n", ret
);
487 if (mcast
->backoff
> IPOIB_MAX_BACKOFF_SECONDS
)
488 mcast
->backoff
= IPOIB_MAX_BACKOFF_SECONDS
;
490 mutex_lock(&mcast_mutex
);
491 if (test_bit(IPOIB_MCAST_RUN
, &priv
->flags
))
492 queue_delayed_work(ipoib_workqueue
,
494 mcast
->backoff
* HZ
);
495 mutex_unlock(&mcast_mutex
);
499 void ipoib_mcast_join_task(struct work_struct
*work
)
501 struct ipoib_dev_priv
*priv
=
502 container_of(work
, struct ipoib_dev_priv
, mcast_task
.work
);
503 struct net_device
*dev
= priv
->dev
;
505 if (!test_bit(IPOIB_MCAST_RUN
, &priv
->flags
))
508 if (ib_query_gid(priv
->ca
, priv
->port
, 0, &priv
->local_gid
))
509 ipoib_warn(priv
, "ib_query_gid() failed\n");
511 memcpy(priv
->dev
->dev_addr
+ 4, priv
->local_gid
.raw
, sizeof (union ib_gid
));
514 struct ib_port_attr attr
;
516 if (!ib_query_port(priv
->ca
, priv
->port
, &attr
))
517 priv
->local_lid
= attr
.lid
;
519 ipoib_warn(priv
, "ib_query_port failed\n");
522 if (!priv
->broadcast
) {
523 struct ipoib_mcast
*broadcast
;
525 broadcast
= ipoib_mcast_alloc(dev
, 1);
527 ipoib_warn(priv
, "failed to allocate broadcast group\n");
528 mutex_lock(&mcast_mutex
);
529 if (test_bit(IPOIB_MCAST_RUN
, &priv
->flags
))
530 queue_delayed_work(ipoib_workqueue
,
531 &priv
->mcast_task
, HZ
);
532 mutex_unlock(&mcast_mutex
);
536 spin_lock_irq(&priv
->lock
);
537 memcpy(broadcast
->mcmember
.mgid
.raw
, priv
->dev
->broadcast
+ 4,
538 sizeof (union ib_gid
));
539 priv
->broadcast
= broadcast
;
541 __ipoib_mcast_add(dev
, priv
->broadcast
);
542 spin_unlock_irq(&priv
->lock
);
545 if (!test_bit(IPOIB_MCAST_FLAG_ATTACHED
, &priv
->broadcast
->flags
)) {
546 if (!test_bit(IPOIB_MCAST_FLAG_BUSY
, &priv
->broadcast
->flags
))
547 ipoib_mcast_join(dev
, priv
->broadcast
, 0);
552 struct ipoib_mcast
*mcast
= NULL
;
554 spin_lock_irq(&priv
->lock
);
555 list_for_each_entry(mcast
, &priv
->multicast_list
, list
) {
556 if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY
, &mcast
->flags
)
557 && !test_bit(IPOIB_MCAST_FLAG_BUSY
, &mcast
->flags
)
558 && !test_bit(IPOIB_MCAST_FLAG_ATTACHED
, &mcast
->flags
)) {
559 /* Found the next unjoined group */
563 spin_unlock_irq(&priv
->lock
);
565 if (&mcast
->list
== &priv
->multicast_list
) {
570 ipoib_mcast_join(dev
, mcast
, 1);
574 priv
->mcast_mtu
= IPOIB_UD_MTU(ib_mtu_enum_to_int(priv
->broadcast
->mcmember
.mtu
));
576 if (!ipoib_cm_admin_enabled(dev
))
577 dev
->mtu
= min(priv
->mcast_mtu
, priv
->admin_mtu
);
579 ipoib_dbg_mcast(priv
, "successfully joined all multicast groups\n");
581 clear_bit(IPOIB_MCAST_RUN
, &priv
->flags
);
584 int ipoib_mcast_start_thread(struct net_device
*dev
)
586 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
588 ipoib_dbg_mcast(priv
, "starting multicast thread\n");
590 mutex_lock(&mcast_mutex
);
591 if (!test_and_set_bit(IPOIB_MCAST_RUN
, &priv
->flags
))
592 queue_delayed_work(ipoib_workqueue
, &priv
->mcast_task
, 0);
593 mutex_unlock(&mcast_mutex
);
598 int ipoib_mcast_stop_thread(struct net_device
*dev
, int flush
)
600 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
602 ipoib_dbg_mcast(priv
, "stopping multicast thread\n");
604 mutex_lock(&mcast_mutex
);
605 clear_bit(IPOIB_MCAST_RUN
, &priv
->flags
);
606 cancel_delayed_work(&priv
->mcast_task
);
607 mutex_unlock(&mcast_mutex
);
610 flush_workqueue(ipoib_workqueue
);
615 static int ipoib_mcast_leave(struct net_device
*dev
, struct ipoib_mcast
*mcast
)
617 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
620 if (test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY
, &mcast
->flags
))
621 ib_sa_free_multicast(mcast
->mc
);
623 if (test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED
, &mcast
->flags
)) {
624 ipoib_dbg_mcast(priv
, "leaving MGID " IPOIB_GID_FMT
"\n",
625 IPOIB_GID_ARG(mcast
->mcmember
.mgid
));
627 /* Remove ourselves from the multicast group */
628 ret
= ipoib_mcast_detach(dev
, be16_to_cpu(mcast
->mcmember
.mlid
),
629 &mcast
->mcmember
.mgid
);
631 ipoib_warn(priv
, "ipoib_mcast_detach failed (result = %d)\n", ret
);
637 void ipoib_mcast_send(struct net_device
*dev
, void *mgid
, struct sk_buff
*skb
)
639 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
640 struct ipoib_mcast
*mcast
;
643 * We can only be called from ipoib_start_xmit, so we're
644 * inside tx_lock -- no need to save/restore flags.
646 spin_lock(&priv
->lock
);
648 if (!test_bit(IPOIB_FLAG_OPER_UP
, &priv
->flags
) ||
650 !test_bit(IPOIB_MCAST_FLAG_ATTACHED
, &priv
->broadcast
->flags
)) {
651 ++dev
->stats
.tx_dropped
;
652 dev_kfree_skb_any(skb
);
656 mcast
= __ipoib_mcast_find(dev
, mgid
);
658 /* Let's create a new send only group now */
659 ipoib_dbg_mcast(priv
, "setting up send only multicast group for "
660 IPOIB_GID_FMT
"\n", IPOIB_GID_RAW_ARG(mgid
));
662 mcast
= ipoib_mcast_alloc(dev
, 0);
664 ipoib_warn(priv
, "unable to allocate memory for "
665 "multicast structure\n");
666 ++dev
->stats
.tx_dropped
;
667 dev_kfree_skb_any(skb
);
671 set_bit(IPOIB_MCAST_FLAG_SENDONLY
, &mcast
->flags
);
672 memcpy(mcast
->mcmember
.mgid
.raw
, mgid
, sizeof (union ib_gid
));
673 __ipoib_mcast_add(dev
, mcast
);
674 list_add_tail(&mcast
->list
, &priv
->multicast_list
);
678 if (skb_queue_len(&mcast
->pkt_queue
) < IPOIB_MAX_MCAST_QUEUE
)
679 skb_queue_tail(&mcast
->pkt_queue
, skb
);
681 ++dev
->stats
.tx_dropped
;
682 dev_kfree_skb_any(skb
);
685 if (test_bit(IPOIB_MCAST_FLAG_BUSY
, &mcast
->flags
))
686 ipoib_dbg_mcast(priv
, "no address vector, "
687 "but multicast join already started\n");
688 else if (test_bit(IPOIB_MCAST_FLAG_SENDONLY
, &mcast
->flags
))
689 ipoib_mcast_sendonly_join(mcast
);
692 * If lookup completes between here and out:, don't
693 * want to send packet twice.
699 if (mcast
&& mcast
->ah
) {
701 skb
->dst
->neighbour
&&
702 !*to_ipoib_neigh(skb
->dst
->neighbour
)) {
703 struct ipoib_neigh
*neigh
= ipoib_neigh_alloc(skb
->dst
->neighbour
,
707 kref_get(&mcast
->ah
->ref
);
708 neigh
->ah
= mcast
->ah
;
709 list_add_tail(&neigh
->list
, &mcast
->neigh_list
);
713 ipoib_send(dev
, skb
, mcast
->ah
, IB_MULTICAST_QPN
);
717 spin_unlock(&priv
->lock
);
720 void ipoib_mcast_dev_flush(struct net_device
*dev
)
722 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
723 LIST_HEAD(remove_list
);
724 struct ipoib_mcast
*mcast
, *tmcast
;
727 ipoib_dbg_mcast(priv
, "flushing multicast list\n");
729 spin_lock_irqsave(&priv
->lock
, flags
);
731 list_for_each_entry_safe(mcast
, tmcast
, &priv
->multicast_list
, list
) {
732 list_del(&mcast
->list
);
733 rb_erase(&mcast
->rb_node
, &priv
->multicast_tree
);
734 list_add_tail(&mcast
->list
, &remove_list
);
737 if (priv
->broadcast
) {
738 rb_erase(&priv
->broadcast
->rb_node
, &priv
->multicast_tree
);
739 list_add_tail(&priv
->broadcast
->list
, &remove_list
);
740 priv
->broadcast
= NULL
;
743 spin_unlock_irqrestore(&priv
->lock
, flags
);
745 list_for_each_entry_safe(mcast
, tmcast
, &remove_list
, list
) {
746 ipoib_mcast_leave(dev
, mcast
);
747 ipoib_mcast_free(mcast
);
751 void ipoib_mcast_restart_task(struct work_struct
*work
)
753 struct ipoib_dev_priv
*priv
=
754 container_of(work
, struct ipoib_dev_priv
, restart_task
);
755 struct net_device
*dev
= priv
->dev
;
756 struct dev_mc_list
*mclist
;
757 struct ipoib_mcast
*mcast
, *tmcast
;
758 LIST_HEAD(remove_list
);
760 struct ib_sa_mcmember_rec rec
;
762 ipoib_dbg_mcast(priv
, "restarting multicast task\n");
764 ipoib_mcast_stop_thread(dev
, 0);
766 local_irq_save(flags
);
768 spin_lock(&priv
->lock
);
771 * Unfortunately, the networking core only gives us a list of all of
772 * the multicast hardware addresses. We need to figure out which ones
773 * are new and which ones have been removed
776 /* Clear out the found flag */
777 list_for_each_entry(mcast
, &priv
->multicast_list
, list
)
778 clear_bit(IPOIB_MCAST_FLAG_FOUND
, &mcast
->flags
);
780 /* Mark all of the entries that are found or don't exist */
781 for (mclist
= dev
->mc_list
; mclist
; mclist
= mclist
->next
) {
784 memcpy(mgid
.raw
, mclist
->dmi_addr
+ 4, sizeof mgid
);
786 mcast
= __ipoib_mcast_find(dev
, &mgid
);
787 if (!mcast
|| test_bit(IPOIB_MCAST_FLAG_SENDONLY
, &mcast
->flags
)) {
788 struct ipoib_mcast
*nmcast
;
790 /* ignore group which is directly joined by userspace */
791 if (test_bit(IPOIB_FLAG_UMCAST
, &priv
->flags
) &&
792 !ib_sa_get_mcmember_rec(priv
->ca
, priv
->port
, &mgid
, &rec
)) {
793 ipoib_dbg_mcast(priv
, "ignoring multicast entry for mgid "
794 IPOIB_GID_FMT
"\n", IPOIB_GID_ARG(mgid
));
798 /* Not found or send-only group, let's add a new entry */
799 ipoib_dbg_mcast(priv
, "adding multicast entry for mgid "
800 IPOIB_GID_FMT
"\n", IPOIB_GID_ARG(mgid
));
802 nmcast
= ipoib_mcast_alloc(dev
, 0);
804 ipoib_warn(priv
, "unable to allocate memory for multicast structure\n");
808 set_bit(IPOIB_MCAST_FLAG_FOUND
, &nmcast
->flags
);
810 nmcast
->mcmember
.mgid
= mgid
;
813 /* Destroy the send only entry */
814 list_move_tail(&mcast
->list
, &remove_list
);
816 rb_replace_node(&mcast
->rb_node
,
818 &priv
->multicast_tree
);
820 __ipoib_mcast_add(dev
, nmcast
);
822 list_add_tail(&nmcast
->list
, &priv
->multicast_list
);
826 set_bit(IPOIB_MCAST_FLAG_FOUND
, &mcast
->flags
);
829 /* Remove all of the entries don't exist anymore */
830 list_for_each_entry_safe(mcast
, tmcast
, &priv
->multicast_list
, list
) {
831 if (!test_bit(IPOIB_MCAST_FLAG_FOUND
, &mcast
->flags
) &&
832 !test_bit(IPOIB_MCAST_FLAG_SENDONLY
, &mcast
->flags
)) {
833 ipoib_dbg_mcast(priv
, "deleting multicast group " IPOIB_GID_FMT
"\n",
834 IPOIB_GID_ARG(mcast
->mcmember
.mgid
));
836 rb_erase(&mcast
->rb_node
, &priv
->multicast_tree
);
838 /* Move to the remove list */
839 list_move_tail(&mcast
->list
, &remove_list
);
843 spin_unlock(&priv
->lock
);
844 netif_tx_unlock(dev
);
845 local_irq_restore(flags
);
847 /* We have to cancel outside of the spinlock */
848 list_for_each_entry_safe(mcast
, tmcast
, &remove_list
, list
) {
849 ipoib_mcast_leave(mcast
->dev
, mcast
);
850 ipoib_mcast_free(mcast
);
853 if (test_bit(IPOIB_FLAG_ADMIN_UP
, &priv
->flags
))
854 ipoib_mcast_start_thread(dev
);
857 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
859 struct ipoib_mcast_iter
*ipoib_mcast_iter_init(struct net_device
*dev
)
861 struct ipoib_mcast_iter
*iter
;
863 iter
= kmalloc(sizeof *iter
, GFP_KERNEL
);
868 memset(iter
->mgid
.raw
, 0, 16);
870 if (ipoib_mcast_iter_next(iter
)) {
878 int ipoib_mcast_iter_next(struct ipoib_mcast_iter
*iter
)
880 struct ipoib_dev_priv
*priv
= netdev_priv(iter
->dev
);
882 struct ipoib_mcast
*mcast
;
885 spin_lock_irq(&priv
->lock
);
887 n
= rb_first(&priv
->multicast_tree
);
890 mcast
= rb_entry(n
, struct ipoib_mcast
, rb_node
);
892 if (memcmp(iter
->mgid
.raw
, mcast
->mcmember
.mgid
.raw
,
893 sizeof (union ib_gid
)) < 0) {
894 iter
->mgid
= mcast
->mcmember
.mgid
;
895 iter
->created
= mcast
->created
;
896 iter
->queuelen
= skb_queue_len(&mcast
->pkt_queue
);
897 iter
->complete
= !!mcast
->ah
;
898 iter
->send_only
= !!(mcast
->flags
& (1 << IPOIB_MCAST_FLAG_SENDONLY
));
908 spin_unlock_irq(&priv
->lock
);
913 void ipoib_mcast_iter_read(struct ipoib_mcast_iter
*iter
,
915 unsigned long *created
,
916 unsigned int *queuelen
,
917 unsigned int *complete
,
918 unsigned int *send_only
)
921 *created
= iter
->created
;
922 *queuelen
= iter
->queuelen
;
923 *complete
= iter
->complete
;
924 *send_only
= iter
->send_only
;
927 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */