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
;
72 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
73 struct ipoib_neigh
*neigh
, *tmp
;
76 ipoib_dbg_mcast(netdev_priv(dev
), "deleting multicast group %pI6\n",
77 mcast
->mcmember
.mgid
.raw
);
79 spin_lock_irq(&priv
->lock
);
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_irq(&priv
->lock
);
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 netif_tx_lock_bh(dev
);
104 dev
->stats
.tx_dropped
+= tx_dropped
;
105 netif_tx_unlock_bh(dev
);
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
);
191 mcast
->mcmember
= *mcmember
;
193 /* Set the cached Q_Key before we attach if it's the broadcast group */
194 if (!memcmp(mcast
->mcmember
.mgid
.raw
, priv
->dev
->broadcast
+ 4,
195 sizeof (union ib_gid
))) {
196 spin_lock_irq(&priv
->lock
);
197 if (!priv
->broadcast
) {
198 spin_unlock_irq(&priv
->lock
);
201 priv
->qkey
= be32_to_cpu(priv
->broadcast
->mcmember
.qkey
);
202 spin_unlock_irq(&priv
->lock
);
203 priv
->tx_wr
.wr
.ud
.remote_qkey
= priv
->qkey
;
207 if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY
, &mcast
->flags
)) {
208 if (test_and_set_bit(IPOIB_MCAST_FLAG_ATTACHED
, &mcast
->flags
)) {
209 ipoib_warn(priv
, "multicast group %pI6 already attached\n",
210 mcast
->mcmember
.mgid
.raw
);
215 ret
= ipoib_mcast_attach(dev
, be16_to_cpu(mcast
->mcmember
.mlid
),
216 &mcast
->mcmember
.mgid
, set_qkey
);
218 ipoib_warn(priv
, "couldn't attach QP to multicast group %pI6\n",
219 mcast
->mcmember
.mgid
.raw
);
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 %pI6 AV %p, LID 0x%04x, SL %d\n",
251 mcast
->mcmember
.mgid
.raw
,
253 be16_to_cpu(mcast
->mcmember
.mlid
),
258 /* actually send any queued packets */
259 netif_tx_lock_bh(dev
);
260 while (!skb_queue_empty(&mcast
->pkt_queue
)) {
261 struct sk_buff
*skb
= skb_dequeue(&mcast
->pkt_queue
);
262 struct dst_entry
*dst
= skb_dst(skb
);
263 struct neighbour
*n
= NULL
;
265 netif_tx_unlock_bh(dev
);
269 n
= dst_get_neighbour(dst
);
271 /* put pseudoheader back on for next time */
272 skb_push(skb
, sizeof (struct ipoib_pseudoheader
));
275 if (dev_queue_xmit(skb
))
276 ipoib_warn(priv
, "dev_queue_xmit failed to requeue packet\n");
277 netif_tx_lock_bh(dev
);
279 netif_tx_unlock_bh(dev
);
285 ipoib_mcast_sendonly_join_complete(int status
,
286 struct ib_sa_multicast
*multicast
)
288 struct ipoib_mcast
*mcast
= multicast
->context
;
289 struct net_device
*dev
= mcast
->dev
;
291 /* We trap for port events ourselves. */
292 if (status
== -ENETRESET
)
296 status
= ipoib_mcast_join_finish(mcast
, &multicast
->rec
);
299 if (mcast
->logcount
++ < 20)
300 ipoib_dbg_mcast(netdev_priv(dev
), "multicast join failed for %pI6, status %d\n",
301 mcast
->mcmember
.mgid
.raw
, status
);
303 /* Flush out any queued packets */
304 netif_tx_lock_bh(dev
);
305 while (!skb_queue_empty(&mcast
->pkt_queue
)) {
306 ++dev
->stats
.tx_dropped
;
307 dev_kfree_skb_any(skb_dequeue(&mcast
->pkt_queue
));
309 netif_tx_unlock_bh(dev
);
311 /* Clear the busy flag so we try again */
312 status
= test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY
,
318 static int ipoib_mcast_sendonly_join(struct ipoib_mcast
*mcast
)
320 struct net_device
*dev
= mcast
->dev
;
321 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
322 struct ib_sa_mcmember_rec rec
= {
323 #if 0 /* Some SMs don't support send-only yet */
331 if (!test_bit(IPOIB_FLAG_OPER_UP
, &priv
->flags
)) {
332 ipoib_dbg_mcast(priv
, "device shutting down, no multicast joins\n");
336 if (test_and_set_bit(IPOIB_MCAST_FLAG_BUSY
, &mcast
->flags
)) {
337 ipoib_dbg_mcast(priv
, "multicast entry busy, skipping\n");
341 rec
.mgid
= mcast
->mcmember
.mgid
;
342 rec
.port_gid
= priv
->local_gid
;
343 rec
.pkey
= cpu_to_be16(priv
->pkey
);
345 mcast
->mc
= ib_sa_join_multicast(&ipoib_sa_client
, priv
->ca
,
347 IB_SA_MCMEMBER_REC_MGID
|
348 IB_SA_MCMEMBER_REC_PORT_GID
|
349 IB_SA_MCMEMBER_REC_PKEY
|
350 IB_SA_MCMEMBER_REC_JOIN_STATE
,
352 ipoib_mcast_sendonly_join_complete
,
354 if (IS_ERR(mcast
->mc
)) {
355 ret
= PTR_ERR(mcast
->mc
);
356 clear_bit(IPOIB_MCAST_FLAG_BUSY
, &mcast
->flags
);
357 ipoib_warn(priv
, "ib_sa_join_multicast failed (ret = %d)\n",
360 ipoib_dbg_mcast(priv
, "no multicast record for %pI6, starting join\n",
361 mcast
->mcmember
.mgid
.raw
);
367 void ipoib_mcast_carrier_on_task(struct work_struct
*work
)
369 struct ipoib_dev_priv
*priv
= container_of(work
, struct ipoib_dev_priv
,
371 struct ib_port_attr attr
;
374 * Take rtnl_lock to avoid racing with ipoib_stop() and
375 * turning the carrier back on while a device is being
378 if (ib_query_port(priv
->ca
, priv
->port
, &attr
) ||
379 attr
.state
!= IB_PORT_ACTIVE
) {
380 ipoib_dbg(priv
, "Keeping carrier off until IB port is active\n");
385 netif_carrier_on(priv
->dev
);
389 static int ipoib_mcast_join_complete(int status
,
390 struct ib_sa_multicast
*multicast
)
392 struct ipoib_mcast
*mcast
= multicast
->context
;
393 struct net_device
*dev
= mcast
->dev
;
394 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
396 ipoib_dbg_mcast(priv
, "join completion for %pI6 (status %d)\n",
397 mcast
->mcmember
.mgid
.raw
, status
);
399 /* We trap for port events ourselves. */
400 if (status
== -ENETRESET
)
404 status
= ipoib_mcast_join_finish(mcast
, &multicast
->rec
);
408 mutex_lock(&mcast_mutex
);
409 if (test_bit(IPOIB_MCAST_RUN
, &priv
->flags
))
410 queue_delayed_work(ipoib_workqueue
,
411 &priv
->mcast_task
, 0);
412 mutex_unlock(&mcast_mutex
);
415 * Defer carrier on work to ipoib_workqueue to avoid a
416 * deadlock on rtnl_lock here.
418 if (mcast
== priv
->broadcast
)
419 queue_work(ipoib_workqueue
, &priv
->carrier_on_task
);
424 if (mcast
->logcount
++ < 20) {
425 if (status
== -ETIMEDOUT
|| status
== -EAGAIN
) {
426 ipoib_dbg_mcast(priv
, "multicast join failed for %pI6, status %d\n",
427 mcast
->mcmember
.mgid
.raw
, status
);
429 ipoib_warn(priv
, "multicast join failed for %pI6, status %d\n",
430 mcast
->mcmember
.mgid
.raw
, status
);
435 if (mcast
->backoff
> IPOIB_MAX_BACKOFF_SECONDS
)
436 mcast
->backoff
= IPOIB_MAX_BACKOFF_SECONDS
;
438 /* Clear the busy flag so we try again */
439 status
= test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY
, &mcast
->flags
);
441 mutex_lock(&mcast_mutex
);
442 spin_lock_irq(&priv
->lock
);
443 if (test_bit(IPOIB_MCAST_RUN
, &priv
->flags
))
444 queue_delayed_work(ipoib_workqueue
, &priv
->mcast_task
,
445 mcast
->backoff
* HZ
);
446 spin_unlock_irq(&priv
->lock
);
447 mutex_unlock(&mcast_mutex
);
452 static void ipoib_mcast_join(struct net_device
*dev
, struct ipoib_mcast
*mcast
,
455 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
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 set_bit(IPOIB_MCAST_FLAG_BUSY
, &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 ret
= PTR_ERR(mcast
->mc
);
504 ipoib_warn(priv
, "ib_sa_join_multicast failed, status %d\n", ret
);
507 if (mcast
->backoff
> IPOIB_MAX_BACKOFF_SECONDS
)
508 mcast
->backoff
= IPOIB_MAX_BACKOFF_SECONDS
;
510 mutex_lock(&mcast_mutex
);
511 if (test_bit(IPOIB_MCAST_RUN
, &priv
->flags
))
512 queue_delayed_work(ipoib_workqueue
,
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
;
525 if (!test_bit(IPOIB_MCAST_RUN
, &priv
->flags
))
528 if (ib_query_gid(priv
->ca
, priv
->port
, 0, &priv
->local_gid
))
529 ipoib_warn(priv
, "ib_query_gid() failed\n");
531 memcpy(priv
->dev
->dev_addr
+ 4, priv
->local_gid
.raw
, sizeof (union ib_gid
));
534 struct ib_port_attr attr
;
536 if (!ib_query_port(priv
->ca
, priv
->port
, &attr
))
537 priv
->local_lid
= attr
.lid
;
539 ipoib_warn(priv
, "ib_query_port failed\n");
542 if (!priv
->broadcast
) {
543 struct ipoib_mcast
*broadcast
;
545 if (!test_bit(IPOIB_FLAG_ADMIN_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(ipoib_workqueue
,
554 &priv
->mcast_task
, HZ
);
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 priv
->mcast_mtu
= IPOIB_UD_MTU(ib_mtu_enum_to_int(priv
->broadcast
->mcmember
.mtu
));
599 if (!ipoib_cm_admin_enabled(dev
)) {
601 dev_set_mtu(dev
, min(priv
->mcast_mtu
, priv
->admin_mtu
));
605 ipoib_dbg_mcast(priv
, "successfully joined all multicast groups\n");
607 clear_bit(IPOIB_MCAST_RUN
, &priv
->flags
);
610 int ipoib_mcast_start_thread(struct net_device
*dev
)
612 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
614 ipoib_dbg_mcast(priv
, "starting multicast thread\n");
616 mutex_lock(&mcast_mutex
);
617 if (!test_and_set_bit(IPOIB_MCAST_RUN
, &priv
->flags
))
618 queue_delayed_work(ipoib_workqueue
, &priv
->mcast_task
, 0);
619 mutex_unlock(&mcast_mutex
);
624 int ipoib_mcast_stop_thread(struct net_device
*dev
, int flush
)
626 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
628 ipoib_dbg_mcast(priv
, "stopping multicast thread\n");
630 mutex_lock(&mcast_mutex
);
631 clear_bit(IPOIB_MCAST_RUN
, &priv
->flags
);
632 cancel_delayed_work(&priv
->mcast_task
);
633 mutex_unlock(&mcast_mutex
);
636 flush_workqueue(ipoib_workqueue
);
641 static int ipoib_mcast_leave(struct net_device
*dev
, struct ipoib_mcast
*mcast
)
643 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
646 if (test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY
, &mcast
->flags
))
647 ib_sa_free_multicast(mcast
->mc
);
649 if (test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED
, &mcast
->flags
)) {
650 ipoib_dbg_mcast(priv
, "leaving MGID %pI6\n",
651 mcast
->mcmember
.mgid
.raw
);
653 /* Remove ourselves from the multicast group */
654 ret
= ib_detach_mcast(priv
->qp
, &mcast
->mcmember
.mgid
,
655 be16_to_cpu(mcast
->mcmember
.mlid
));
657 ipoib_warn(priv
, "ib_detach_mcast failed (result = %d)\n", ret
);
663 void ipoib_mcast_send(struct net_device
*dev
, void *mgid
, struct sk_buff
*skb
)
665 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
666 struct ipoib_mcast
*mcast
;
669 spin_lock_irqsave(&priv
->lock
, flags
);
671 if (!test_bit(IPOIB_FLAG_OPER_UP
, &priv
->flags
) ||
673 !test_bit(IPOIB_MCAST_FLAG_ATTACHED
, &priv
->broadcast
->flags
)) {
674 ++dev
->stats
.tx_dropped
;
675 dev_kfree_skb_any(skb
);
679 mcast
= __ipoib_mcast_find(dev
, mgid
);
681 /* Let's create a new send only group now */
682 ipoib_dbg_mcast(priv
, "setting up send only multicast group for %pI6\n",
685 mcast
= ipoib_mcast_alloc(dev
, 0);
687 ipoib_warn(priv
, "unable to allocate memory for "
688 "multicast structure\n");
689 ++dev
->stats
.tx_dropped
;
690 dev_kfree_skb_any(skb
);
694 set_bit(IPOIB_MCAST_FLAG_SENDONLY
, &mcast
->flags
);
695 memcpy(mcast
->mcmember
.mgid
.raw
, mgid
, sizeof (union ib_gid
));
696 __ipoib_mcast_add(dev
, mcast
);
697 list_add_tail(&mcast
->list
, &priv
->multicast_list
);
701 if (skb_queue_len(&mcast
->pkt_queue
) < IPOIB_MAX_MCAST_QUEUE
)
702 skb_queue_tail(&mcast
->pkt_queue
, skb
);
704 ++dev
->stats
.tx_dropped
;
705 dev_kfree_skb_any(skb
);
708 if (test_bit(IPOIB_MCAST_FLAG_BUSY
, &mcast
->flags
))
709 ipoib_dbg_mcast(priv
, "no address vector, "
710 "but multicast join already started\n");
711 else if (test_bit(IPOIB_MCAST_FLAG_SENDONLY
, &mcast
->flags
))
712 ipoib_mcast_sendonly_join(mcast
);
715 * If lookup completes between here and out:, don't
716 * want to send packet twice.
722 if (mcast
&& mcast
->ah
) {
723 struct dst_entry
*dst
= skb_dst(skb
);
724 struct neighbour
*n
= NULL
;
726 n
= dst_get_neighbour(dst
);
727 if (n
&& !*to_ipoib_neigh(n
)) {
728 struct ipoib_neigh
*neigh
= ipoib_neigh_alloc(n
,
732 kref_get(&mcast
->ah
->ref
);
733 neigh
->ah
= mcast
->ah
;
734 list_add_tail(&neigh
->list
, &mcast
->neigh_list
);
738 spin_unlock_irqrestore(&priv
->lock
, flags
);
739 ipoib_send(dev
, skb
, mcast
->ah
, IB_MULTICAST_QPN
);
744 spin_unlock_irqrestore(&priv
->lock
, flags
);
747 void ipoib_mcast_dev_flush(struct net_device
*dev
)
749 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
750 LIST_HEAD(remove_list
);
751 struct ipoib_mcast
*mcast
, *tmcast
;
754 ipoib_dbg_mcast(priv
, "flushing multicast list\n");
756 spin_lock_irqsave(&priv
->lock
, flags
);
758 list_for_each_entry_safe(mcast
, tmcast
, &priv
->multicast_list
, list
) {
759 list_del(&mcast
->list
);
760 rb_erase(&mcast
->rb_node
, &priv
->multicast_tree
);
761 list_add_tail(&mcast
->list
, &remove_list
);
764 if (priv
->broadcast
) {
765 rb_erase(&priv
->broadcast
->rb_node
, &priv
->multicast_tree
);
766 list_add_tail(&priv
->broadcast
->list
, &remove_list
);
767 priv
->broadcast
= NULL
;
770 spin_unlock_irqrestore(&priv
->lock
, flags
);
772 list_for_each_entry_safe(mcast
, tmcast
, &remove_list
, list
) {
773 ipoib_mcast_leave(dev
, mcast
);
774 ipoib_mcast_free(mcast
);
778 static int ipoib_mcast_addr_is_valid(const u8
*addr
, const u8
*broadcast
)
780 /* reserved QPN, prefix, scope */
781 if (memcmp(addr
, broadcast
, 6))
783 /* signature lower, pkey */
784 if (memcmp(addr
+ 7, broadcast
+ 7, 3))
789 void ipoib_mcast_restart_task(struct work_struct
*work
)
791 struct ipoib_dev_priv
*priv
=
792 container_of(work
, struct ipoib_dev_priv
, restart_task
);
793 struct net_device
*dev
= priv
->dev
;
794 struct netdev_hw_addr
*ha
;
795 struct ipoib_mcast
*mcast
, *tmcast
;
796 LIST_HEAD(remove_list
);
798 struct ib_sa_mcmember_rec rec
;
800 ipoib_dbg_mcast(priv
, "restarting multicast task\n");
802 ipoib_mcast_stop_thread(dev
, 0);
804 local_irq_save(flags
);
805 netif_addr_lock(dev
);
806 spin_lock(&priv
->lock
);
809 * Unfortunately, the networking core only gives us a list of all of
810 * the multicast hardware addresses. We need to figure out which ones
811 * are new and which ones have been removed
814 /* Clear out the found flag */
815 list_for_each_entry(mcast
, &priv
->multicast_list
, list
)
816 clear_bit(IPOIB_MCAST_FLAG_FOUND
, &mcast
->flags
);
818 /* Mark all of the entries that are found or don't exist */
819 netdev_for_each_mc_addr(ha
, dev
) {
822 if (!ipoib_mcast_addr_is_valid(ha
->addr
, dev
->broadcast
))
825 memcpy(mgid
.raw
, ha
->addr
+ 4, sizeof mgid
);
827 mcast
= __ipoib_mcast_find(dev
, &mgid
);
828 if (!mcast
|| test_bit(IPOIB_MCAST_FLAG_SENDONLY
, &mcast
->flags
)) {
829 struct ipoib_mcast
*nmcast
;
831 /* ignore group which is directly joined by userspace */
832 if (test_bit(IPOIB_FLAG_UMCAST
, &priv
->flags
) &&
833 !ib_sa_get_mcmember_rec(priv
->ca
, priv
->port
, &mgid
, &rec
)) {
834 ipoib_dbg_mcast(priv
, "ignoring multicast entry for mgid %pI6\n",
839 /* Not found or send-only group, let's add a new entry */
840 ipoib_dbg_mcast(priv
, "adding multicast entry for mgid %pI6\n",
843 nmcast
= ipoib_mcast_alloc(dev
, 0);
845 ipoib_warn(priv
, "unable to allocate memory for multicast structure\n");
849 set_bit(IPOIB_MCAST_FLAG_FOUND
, &nmcast
->flags
);
851 nmcast
->mcmember
.mgid
= mgid
;
854 /* Destroy the send only entry */
855 list_move_tail(&mcast
->list
, &remove_list
);
857 rb_replace_node(&mcast
->rb_node
,
859 &priv
->multicast_tree
);
861 __ipoib_mcast_add(dev
, nmcast
);
863 list_add_tail(&nmcast
->list
, &priv
->multicast_list
);
867 set_bit(IPOIB_MCAST_FLAG_FOUND
, &mcast
->flags
);
870 /* Remove all of the entries don't exist anymore */
871 list_for_each_entry_safe(mcast
, tmcast
, &priv
->multicast_list
, list
) {
872 if (!test_bit(IPOIB_MCAST_FLAG_FOUND
, &mcast
->flags
) &&
873 !test_bit(IPOIB_MCAST_FLAG_SENDONLY
, &mcast
->flags
)) {
874 ipoib_dbg_mcast(priv
, "deleting multicast group %pI6\n",
875 mcast
->mcmember
.mgid
.raw
);
877 rb_erase(&mcast
->rb_node
, &priv
->multicast_tree
);
879 /* Move to the remove list */
880 list_move_tail(&mcast
->list
, &remove_list
);
884 spin_unlock(&priv
->lock
);
885 netif_addr_unlock(dev
);
886 local_irq_restore(flags
);
888 /* We have to cancel outside of the spinlock */
889 list_for_each_entry_safe(mcast
, tmcast
, &remove_list
, list
) {
890 ipoib_mcast_leave(mcast
->dev
, mcast
);
891 ipoib_mcast_free(mcast
);
894 if (test_bit(IPOIB_FLAG_ADMIN_UP
, &priv
->flags
))
895 ipoib_mcast_start_thread(dev
);
898 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
900 struct ipoib_mcast_iter
*ipoib_mcast_iter_init(struct net_device
*dev
)
902 struct ipoib_mcast_iter
*iter
;
904 iter
= kmalloc(sizeof *iter
, GFP_KERNEL
);
909 memset(iter
->mgid
.raw
, 0, 16);
911 if (ipoib_mcast_iter_next(iter
)) {
919 int ipoib_mcast_iter_next(struct ipoib_mcast_iter
*iter
)
921 struct ipoib_dev_priv
*priv
= netdev_priv(iter
->dev
);
923 struct ipoib_mcast
*mcast
;
926 spin_lock_irq(&priv
->lock
);
928 n
= rb_first(&priv
->multicast_tree
);
931 mcast
= rb_entry(n
, struct ipoib_mcast
, rb_node
);
933 if (memcmp(iter
->mgid
.raw
, mcast
->mcmember
.mgid
.raw
,
934 sizeof (union ib_gid
)) < 0) {
935 iter
->mgid
= mcast
->mcmember
.mgid
;
936 iter
->created
= mcast
->created
;
937 iter
->queuelen
= skb_queue_len(&mcast
->pkt_queue
);
938 iter
->complete
= !!mcast
->ah
;
939 iter
->send_only
= !!(mcast
->flags
& (1 << IPOIB_MCAST_FLAG_SENDONLY
));
949 spin_unlock_irq(&priv
->lock
);
954 void ipoib_mcast_iter_read(struct ipoib_mcast_iter
*iter
,
956 unsigned long *created
,
957 unsigned int *queuelen
,
958 unsigned int *complete
,
959 unsigned int *send_only
)
962 *created
= iter
->created
;
963 *queuelen
= iter
->queuelen
;
964 *complete
= iter
->complete
;
965 *send_only
= iter
->send_only
;
968 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */