Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved. | |
2a1d9b7f RD |
3 | * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved. |
4 | * Copyright (c) 2004 Voltaire, Inc. All rights reserved. | |
1da177e4 LT |
5 | * |
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: | |
11 | * | |
12 | * Redistribution and use in source and binary forms, with or | |
13 | * without modification, are permitted provided that the following | |
14 | * conditions are met: | |
15 | * | |
16 | * - Redistributions of source code must retain the above | |
17 | * copyright notice, this list of conditions and the following | |
18 | * disclaimer. | |
19 | * | |
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. | |
24 | * | |
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 | |
32 | * SOFTWARE. | |
33 | * | |
34 | * $Id: ipoib_multicast.c 1362 2004-12-18 15:56:29Z roland $ | |
35 | */ | |
36 | ||
37 | #include <linux/skbuff.h> | |
38 | #include <linux/rtnetlink.h> | |
39 | #include <linux/ip.h> | |
40 | #include <linux/in.h> | |
41 | #include <linux/igmp.h> | |
42 | #include <linux/inetdevice.h> | |
43 | #include <linux/delay.h> | |
44 | #include <linux/completion.h> | |
45 | ||
46 | #include "ipoib.h" | |
47 | ||
48 | #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG | |
49 | static int mcast_debug_level; | |
50 | ||
51 | module_param(mcast_debug_level, int, 0644); | |
52 | MODULE_PARM_DESC(mcast_debug_level, | |
53 | "Enable multicast debug tracing if > 0"); | |
54 | #endif | |
55 | ||
56 | static DECLARE_MUTEX(mcast_mutex); | |
57 | ||
58 | /* Used for all multicast joins (broadcast, IPv4 mcast and IPv6 mcast) */ | |
59 | struct ipoib_mcast { | |
60 | struct ib_sa_mcmember_rec mcmember; | |
61 | struct ipoib_ah *ah; | |
62 | ||
63 | struct rb_node rb_node; | |
64 | struct list_head list; | |
65 | struct completion done; | |
66 | ||
67 | int query_id; | |
68 | struct ib_sa_query *query; | |
69 | ||
70 | unsigned long created; | |
71 | unsigned long backoff; | |
72 | ||
73 | unsigned long flags; | |
74 | unsigned char logcount; | |
75 | ||
76 | struct list_head neigh_list; | |
77 | ||
78 | struct sk_buff_head pkt_queue; | |
79 | ||
80 | struct net_device *dev; | |
81 | }; | |
82 | ||
83 | struct ipoib_mcast_iter { | |
84 | struct net_device *dev; | |
85 | union ib_gid mgid; | |
86 | unsigned long created; | |
87 | unsigned int queuelen; | |
88 | unsigned int complete; | |
89 | unsigned int send_only; | |
90 | }; | |
91 | ||
92 | static void ipoib_mcast_free(struct ipoib_mcast *mcast) | |
93 | { | |
94 | struct net_device *dev = mcast->dev; | |
95 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
96 | struct ipoib_neigh *neigh, *tmp; | |
97 | unsigned long flags; | |
98 | LIST_HEAD(ah_list); | |
99 | struct ipoib_ah *ah, *tah; | |
100 | ||
101 | ipoib_dbg_mcast(netdev_priv(dev), | |
102 | "deleting multicast group " IPOIB_GID_FMT "\n", | |
103 | IPOIB_GID_ARG(mcast->mcmember.mgid)); | |
104 | ||
105 | spin_lock_irqsave(&priv->lock, flags); | |
106 | ||
107 | list_for_each_entry_safe(neigh, tmp, &mcast->neigh_list, list) { | |
108 | if (neigh->ah) | |
109 | list_add_tail(&neigh->ah->list, &ah_list); | |
110 | *to_ipoib_neigh(neigh->neighbour) = NULL; | |
111 | neigh->neighbour->ops->destructor = NULL; | |
112 | kfree(neigh); | |
113 | } | |
114 | ||
115 | spin_unlock_irqrestore(&priv->lock, flags); | |
116 | ||
117 | list_for_each_entry_safe(ah, tah, &ah_list, list) | |
118 | ipoib_put_ah(ah); | |
119 | ||
120 | if (mcast->ah) | |
121 | ipoib_put_ah(mcast->ah); | |
122 | ||
8c608a32 RD |
123 | while (!skb_queue_empty(&mcast->pkt_queue)) |
124 | dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue)); | |
1da177e4 LT |
125 | |
126 | kfree(mcast); | |
127 | } | |
128 | ||
129 | static struct ipoib_mcast *ipoib_mcast_alloc(struct net_device *dev, | |
130 | int can_sleep) | |
131 | { | |
132 | struct ipoib_mcast *mcast; | |
133 | ||
de6eb66b | 134 | mcast = kzalloc(sizeof *mcast, can_sleep ? GFP_KERNEL : GFP_ATOMIC); |
1da177e4 LT |
135 | if (!mcast) |
136 | return NULL; | |
137 | ||
1da177e4 LT |
138 | init_completion(&mcast->done); |
139 | ||
140 | mcast->dev = dev; | |
141 | mcast->created = jiffies; | |
ce5b65cc | 142 | mcast->backoff = 1; |
1da177e4 LT |
143 | mcast->logcount = 0; |
144 | ||
145 | INIT_LIST_HEAD(&mcast->list); | |
146 | INIT_LIST_HEAD(&mcast->neigh_list); | |
147 | skb_queue_head_init(&mcast->pkt_queue); | |
148 | ||
149 | mcast->ah = NULL; | |
150 | mcast->query = NULL; | |
151 | ||
152 | return mcast; | |
153 | } | |
154 | ||
155 | static struct ipoib_mcast *__ipoib_mcast_find(struct net_device *dev, union ib_gid *mgid) | |
156 | { | |
157 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
158 | struct rb_node *n = priv->multicast_tree.rb_node; | |
159 | ||
160 | while (n) { | |
161 | struct ipoib_mcast *mcast; | |
162 | int ret; | |
163 | ||
164 | mcast = rb_entry(n, struct ipoib_mcast, rb_node); | |
165 | ||
166 | ret = memcmp(mgid->raw, mcast->mcmember.mgid.raw, | |
167 | sizeof (union ib_gid)); | |
168 | if (ret < 0) | |
169 | n = n->rb_left; | |
170 | else if (ret > 0) | |
171 | n = n->rb_right; | |
172 | else | |
173 | return mcast; | |
174 | } | |
175 | ||
176 | return NULL; | |
177 | } | |
178 | ||
179 | static int __ipoib_mcast_add(struct net_device *dev, struct ipoib_mcast *mcast) | |
180 | { | |
181 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
182 | struct rb_node **n = &priv->multicast_tree.rb_node, *pn = NULL; | |
183 | ||
184 | while (*n) { | |
185 | struct ipoib_mcast *tmcast; | |
186 | int ret; | |
187 | ||
188 | pn = *n; | |
189 | tmcast = rb_entry(pn, struct ipoib_mcast, rb_node); | |
190 | ||
191 | ret = memcmp(mcast->mcmember.mgid.raw, tmcast->mcmember.mgid.raw, | |
192 | sizeof (union ib_gid)); | |
193 | if (ret < 0) | |
194 | n = &pn->rb_left; | |
195 | else if (ret > 0) | |
196 | n = &pn->rb_right; | |
197 | else | |
198 | return -EEXIST; | |
199 | } | |
200 | ||
201 | rb_link_node(&mcast->rb_node, pn, n); | |
202 | rb_insert_color(&mcast->rb_node, &priv->multicast_tree); | |
203 | ||
204 | return 0; | |
205 | } | |
206 | ||
207 | static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast, | |
208 | struct ib_sa_mcmember_rec *mcmember) | |
209 | { | |
210 | struct net_device *dev = mcast->dev; | |
211 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
212 | int ret; | |
213 | ||
214 | mcast->mcmember = *mcmember; | |
215 | ||
216 | /* Set the cached Q_Key before we attach if it's the broadcast group */ | |
217 | if (!memcmp(mcast->mcmember.mgid.raw, priv->dev->broadcast + 4, | |
218 | sizeof (union ib_gid))) { | |
219 | priv->qkey = be32_to_cpu(priv->broadcast->mcmember.qkey); | |
220 | priv->tx_wr.wr.ud.remote_qkey = priv->qkey; | |
221 | } | |
222 | ||
223 | if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) { | |
224 | if (test_and_set_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) { | |
225 | ipoib_warn(priv, "multicast group " IPOIB_GID_FMT | |
226 | " already attached\n", | |
227 | IPOIB_GID_ARG(mcast->mcmember.mgid)); | |
228 | ||
229 | return 0; | |
230 | } | |
231 | ||
232 | ret = ipoib_mcast_attach(dev, be16_to_cpu(mcast->mcmember.mlid), | |
233 | &mcast->mcmember.mgid); | |
234 | if (ret < 0) { | |
235 | ipoib_warn(priv, "couldn't attach QP to multicast group " | |
236 | IPOIB_GID_FMT "\n", | |
237 | IPOIB_GID_ARG(mcast->mcmember.mgid)); | |
238 | ||
239 | clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags); | |
240 | return ret; | |
241 | } | |
242 | } | |
243 | ||
244 | { | |
245 | struct ib_ah_attr av = { | |
246 | .dlid = be16_to_cpu(mcast->mcmember.mlid), | |
247 | .port_num = priv->port, | |
248 | .sl = mcast->mcmember.sl, | |
249 | .ah_flags = IB_AH_GRH, | |
250 | .grh = { | |
251 | .flow_label = be32_to_cpu(mcast->mcmember.flow_label), | |
252 | .hop_limit = mcast->mcmember.hop_limit, | |
253 | .sgid_index = 0, | |
254 | .traffic_class = mcast->mcmember.traffic_class | |
255 | } | |
256 | }; | |
e6ded99c | 257 | int path_rate = ib_sa_rate_enum_to_int(mcast->mcmember.rate); |
1da177e4 LT |
258 | |
259 | av.grh.dgid = mcast->mcmember.mgid; | |
260 | ||
e6ded99c RD |
261 | if (path_rate > 0 && priv->local_rate > path_rate) |
262 | av.static_rate = (priv->local_rate - 1) / path_rate; | |
1da177e4 LT |
263 | |
264 | ipoib_dbg_mcast(priv, "static_rate %d for local port %dX, mcmember %dX\n", | |
265 | av.static_rate, priv->local_rate, | |
266 | ib_sa_rate_enum_to_int(mcast->mcmember.rate)); | |
267 | ||
268 | mcast->ah = ipoib_create_ah(dev, priv->pd, &av); | |
269 | if (!mcast->ah) { | |
270 | ipoib_warn(priv, "ib_address_create failed\n"); | |
271 | } else { | |
272 | ipoib_dbg_mcast(priv, "MGID " IPOIB_GID_FMT | |
273 | " AV %p, LID 0x%04x, SL %d\n", | |
274 | IPOIB_GID_ARG(mcast->mcmember.mgid), | |
275 | mcast->ah->ah, | |
276 | be16_to_cpu(mcast->mcmember.mlid), | |
277 | mcast->mcmember.sl); | |
278 | } | |
279 | } | |
280 | ||
281 | /* actually send any queued packets */ | |
282 | while (!skb_queue_empty(&mcast->pkt_queue)) { | |
283 | struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue); | |
284 | ||
285 | skb->dev = dev; | |
286 | ||
287 | if (!skb->dst || !skb->dst->neighbour) { | |
288 | /* put pseudoheader back on for next time */ | |
289 | skb_push(skb, sizeof (struct ipoib_pseudoheader)); | |
290 | } | |
291 | ||
292 | if (dev_queue_xmit(skb)) | |
293 | ipoib_warn(priv, "dev_queue_xmit failed to requeue packet\n"); | |
294 | } | |
295 | ||
296 | return 0; | |
297 | } | |
298 | ||
299 | static void | |
300 | ipoib_mcast_sendonly_join_complete(int status, | |
301 | struct ib_sa_mcmember_rec *mcmember, | |
302 | void *mcast_ptr) | |
303 | { | |
304 | struct ipoib_mcast *mcast = mcast_ptr; | |
305 | struct net_device *dev = mcast->dev; | |
306 | ||
307 | if (!status) | |
308 | ipoib_mcast_join_finish(mcast, mcmember); | |
309 | else { | |
310 | if (mcast->logcount++ < 20) | |
311 | ipoib_dbg_mcast(netdev_priv(dev), "multicast join failed for " | |
312 | IPOIB_GID_FMT ", status %d\n", | |
313 | IPOIB_GID_ARG(mcast->mcmember.mgid), status); | |
314 | ||
315 | /* Flush out any queued packets */ | |
8c608a32 RD |
316 | while (!skb_queue_empty(&mcast->pkt_queue)) |
317 | dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue)); | |
1da177e4 LT |
318 | |
319 | /* Clear the busy flag so we try again */ | |
320 | clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags); | |
321 | } | |
322 | ||
323 | complete(&mcast->done); | |
324 | } | |
325 | ||
326 | static int ipoib_mcast_sendonly_join(struct ipoib_mcast *mcast) | |
327 | { | |
328 | struct net_device *dev = mcast->dev; | |
329 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
330 | struct ib_sa_mcmember_rec rec = { | |
331 | #if 0 /* Some SMs don't support send-only yet */ | |
332 | .join_state = 4 | |
333 | #else | |
334 | .join_state = 1 | |
335 | #endif | |
336 | }; | |
337 | int ret = 0; | |
338 | ||
339 | if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) { | |
340 | ipoib_dbg_mcast(priv, "device shutting down, no multicast joins\n"); | |
341 | return -ENODEV; | |
342 | } | |
343 | ||
344 | if (test_and_set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) { | |
345 | ipoib_dbg_mcast(priv, "multicast entry busy, skipping\n"); | |
346 | return -EBUSY; | |
347 | } | |
348 | ||
349 | rec.mgid = mcast->mcmember.mgid; | |
350 | rec.port_gid = priv->local_gid; | |
97f52eb4 | 351 | rec.pkey = cpu_to_be16(priv->pkey); |
1da177e4 LT |
352 | |
353 | ret = ib_sa_mcmember_rec_set(priv->ca, priv->port, &rec, | |
354 | IB_SA_MCMEMBER_REC_MGID | | |
355 | IB_SA_MCMEMBER_REC_PORT_GID | | |
356 | IB_SA_MCMEMBER_REC_PKEY | | |
357 | IB_SA_MCMEMBER_REC_JOIN_STATE, | |
358 | 1000, GFP_ATOMIC, | |
359 | ipoib_mcast_sendonly_join_complete, | |
360 | mcast, &mcast->query); | |
361 | if (ret < 0) { | |
362 | ipoib_warn(priv, "ib_sa_mcmember_rec_set failed (ret = %d)\n", | |
363 | ret); | |
364 | } else { | |
365 | ipoib_dbg_mcast(priv, "no multicast record for " IPOIB_GID_FMT | |
366 | ", starting join\n", | |
367 | IPOIB_GID_ARG(mcast->mcmember.mgid)); | |
368 | ||
369 | mcast->query_id = ret; | |
370 | } | |
371 | ||
372 | return ret; | |
373 | } | |
374 | ||
375 | static void ipoib_mcast_join_complete(int status, | |
376 | struct ib_sa_mcmember_rec *mcmember, | |
377 | void *mcast_ptr) | |
378 | { | |
379 | struct ipoib_mcast *mcast = mcast_ptr; | |
380 | struct net_device *dev = mcast->dev; | |
381 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
382 | ||
383 | ipoib_dbg_mcast(priv, "join completion for " IPOIB_GID_FMT | |
384 | " (status %d)\n", | |
385 | IPOIB_GID_ARG(mcast->mcmember.mgid), status); | |
386 | ||
387 | if (!status && !ipoib_mcast_join_finish(mcast, mcmember)) { | |
ce5b65cc | 388 | mcast->backoff = 1; |
1da177e4 LT |
389 | down(&mcast_mutex); |
390 | if (test_bit(IPOIB_MCAST_RUN, &priv->flags)) | |
391 | queue_work(ipoib_workqueue, &priv->mcast_task); | |
392 | up(&mcast_mutex); | |
393 | complete(&mcast->done); | |
394 | return; | |
395 | } | |
396 | ||
397 | if (status == -EINTR) { | |
398 | complete(&mcast->done); | |
399 | return; | |
400 | } | |
401 | ||
402 | if (status && mcast->logcount++ < 20) { | |
403 | if (status == -ETIMEDOUT || status == -EINTR) { | |
404 | ipoib_dbg_mcast(priv, "multicast join failed for " IPOIB_GID_FMT | |
405 | ", status %d\n", | |
406 | IPOIB_GID_ARG(mcast->mcmember.mgid), | |
407 | status); | |
408 | } else { | |
409 | ipoib_warn(priv, "multicast join failed for " | |
410 | IPOIB_GID_FMT ", status %d\n", | |
411 | IPOIB_GID_ARG(mcast->mcmember.mgid), | |
412 | status); | |
413 | } | |
414 | } | |
415 | ||
416 | mcast->backoff *= 2; | |
417 | if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS) | |
418 | mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS; | |
419 | ||
420 | mcast->query = NULL; | |
421 | ||
422 | down(&mcast_mutex); | |
423 | if (test_bit(IPOIB_MCAST_RUN, &priv->flags)) { | |
424 | if (status == -ETIMEDOUT) | |
425 | queue_work(ipoib_workqueue, &priv->mcast_task); | |
426 | else | |
427 | queue_delayed_work(ipoib_workqueue, &priv->mcast_task, | |
428 | mcast->backoff * HZ); | |
429 | } else | |
430 | complete(&mcast->done); | |
431 | up(&mcast_mutex); | |
432 | ||
433 | return; | |
434 | } | |
435 | ||
436 | static void ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast, | |
437 | int create) | |
438 | { | |
439 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
440 | struct ib_sa_mcmember_rec rec = { | |
441 | .join_state = 1 | |
442 | }; | |
443 | ib_sa_comp_mask comp_mask; | |
444 | int ret = 0; | |
445 | ||
446 | ipoib_dbg_mcast(priv, "joining MGID " IPOIB_GID_FMT "\n", | |
447 | IPOIB_GID_ARG(mcast->mcmember.mgid)); | |
448 | ||
449 | rec.mgid = mcast->mcmember.mgid; | |
450 | rec.port_gid = priv->local_gid; | |
97f52eb4 | 451 | rec.pkey = cpu_to_be16(priv->pkey); |
1da177e4 LT |
452 | |
453 | comp_mask = | |
454 | IB_SA_MCMEMBER_REC_MGID | | |
455 | IB_SA_MCMEMBER_REC_PORT_GID | | |
456 | IB_SA_MCMEMBER_REC_PKEY | | |
457 | IB_SA_MCMEMBER_REC_JOIN_STATE; | |
458 | ||
459 | if (create) { | |
460 | comp_mask |= | |
461 | IB_SA_MCMEMBER_REC_QKEY | | |
462 | IB_SA_MCMEMBER_REC_SL | | |
463 | IB_SA_MCMEMBER_REC_FLOW_LABEL | | |
464 | IB_SA_MCMEMBER_REC_TRAFFIC_CLASS; | |
465 | ||
466 | rec.qkey = priv->broadcast->mcmember.qkey; | |
467 | rec.sl = priv->broadcast->mcmember.sl; | |
468 | rec.flow_label = priv->broadcast->mcmember.flow_label; | |
469 | rec.traffic_class = priv->broadcast->mcmember.traffic_class; | |
470 | } | |
471 | ||
472 | ret = ib_sa_mcmember_rec_set(priv->ca, priv->port, &rec, comp_mask, | |
473 | mcast->backoff * 1000, GFP_ATOMIC, | |
474 | ipoib_mcast_join_complete, | |
475 | mcast, &mcast->query); | |
476 | ||
477 | if (ret < 0) { | |
478 | ipoib_warn(priv, "ib_sa_mcmember_rec_set failed, status %d\n", ret); | |
479 | ||
480 | mcast->backoff *= 2; | |
481 | if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS) | |
482 | mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS; | |
483 | ||
484 | down(&mcast_mutex); | |
485 | if (test_bit(IPOIB_MCAST_RUN, &priv->flags)) | |
486 | queue_delayed_work(ipoib_workqueue, | |
487 | &priv->mcast_task, | |
ce5b65cc | 488 | mcast->backoff * HZ); |
1da177e4 LT |
489 | up(&mcast_mutex); |
490 | } else | |
491 | mcast->query_id = ret; | |
492 | } | |
493 | ||
494 | void ipoib_mcast_join_task(void *dev_ptr) | |
495 | { | |
496 | struct net_device *dev = dev_ptr; | |
497 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
498 | ||
499 | if (!test_bit(IPOIB_MCAST_RUN, &priv->flags)) | |
500 | return; | |
501 | ||
502 | if (ib_query_gid(priv->ca, priv->port, 0, &priv->local_gid)) | |
503 | ipoib_warn(priv, "ib_gid_entry_get() failed\n"); | |
504 | else | |
505 | memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw, sizeof (union ib_gid)); | |
506 | ||
507 | { | |
508 | struct ib_port_attr attr; | |
509 | ||
510 | if (!ib_query_port(priv->ca, priv->port, &attr)) { | |
511 | priv->local_lid = attr.lid; | |
512 | priv->local_rate = attr.active_speed * | |
513 | ib_width_enum_to_int(attr.active_width); | |
514 | } else | |
515 | ipoib_warn(priv, "ib_query_port failed\n"); | |
516 | } | |
517 | ||
518 | if (!priv->broadcast) { | |
519 | priv->broadcast = ipoib_mcast_alloc(dev, 1); | |
520 | if (!priv->broadcast) { | |
521 | ipoib_warn(priv, "failed to allocate broadcast group\n"); | |
522 | down(&mcast_mutex); | |
523 | if (test_bit(IPOIB_MCAST_RUN, &priv->flags)) | |
524 | queue_delayed_work(ipoib_workqueue, | |
525 | &priv->mcast_task, HZ); | |
526 | up(&mcast_mutex); | |
527 | return; | |
528 | } | |
529 | ||
530 | memcpy(priv->broadcast->mcmember.mgid.raw, priv->dev->broadcast + 4, | |
531 | sizeof (union ib_gid)); | |
532 | ||
533 | spin_lock_irq(&priv->lock); | |
534 | __ipoib_mcast_add(dev, priv->broadcast); | |
535 | spin_unlock_irq(&priv->lock); | |
536 | } | |
537 | ||
538 | if (!test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) { | |
539 | ipoib_mcast_join(dev, priv->broadcast, 0); | |
540 | return; | |
541 | } | |
542 | ||
543 | while (1) { | |
544 | struct ipoib_mcast *mcast = NULL; | |
545 | ||
546 | spin_lock_irq(&priv->lock); | |
547 | list_for_each_entry(mcast, &priv->multicast_list, list) { | |
548 | if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) | |
549 | && !test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags) | |
550 | && !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) { | |
551 | /* Found the next unjoined group */ | |
552 | break; | |
553 | } | |
554 | } | |
555 | spin_unlock_irq(&priv->lock); | |
556 | ||
557 | if (&mcast->list == &priv->multicast_list) { | |
558 | /* All done */ | |
559 | break; | |
560 | } | |
561 | ||
562 | ipoib_mcast_join(dev, mcast, 1); | |
563 | return; | |
564 | } | |
565 | ||
566 | priv->mcast_mtu = ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu) - | |
567 | IPOIB_ENCAP_LEN; | |
568 | dev->mtu = min(priv->mcast_mtu, priv->admin_mtu); | |
569 | ||
570 | ipoib_dbg_mcast(priv, "successfully joined all multicast groups\n"); | |
571 | ||
572 | clear_bit(IPOIB_MCAST_RUN, &priv->flags); | |
573 | netif_carrier_on(dev); | |
574 | } | |
575 | ||
576 | int ipoib_mcast_start_thread(struct net_device *dev) | |
577 | { | |
578 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
579 | ||
580 | ipoib_dbg_mcast(priv, "starting multicast thread\n"); | |
581 | ||
582 | down(&mcast_mutex); | |
583 | if (!test_and_set_bit(IPOIB_MCAST_RUN, &priv->flags)) | |
584 | queue_work(ipoib_workqueue, &priv->mcast_task); | |
585 | up(&mcast_mutex); | |
586 | ||
587 | return 0; | |
588 | } | |
589 | ||
8d2cae06 | 590 | int ipoib_mcast_stop_thread(struct net_device *dev, int flush) |
1da177e4 LT |
591 | { |
592 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
593 | struct ipoib_mcast *mcast; | |
594 | ||
595 | ipoib_dbg_mcast(priv, "stopping multicast thread\n"); | |
596 | ||
597 | down(&mcast_mutex); | |
598 | clear_bit(IPOIB_MCAST_RUN, &priv->flags); | |
599 | cancel_delayed_work(&priv->mcast_task); | |
600 | up(&mcast_mutex); | |
601 | ||
8d2cae06 RD |
602 | if (flush) |
603 | flush_workqueue(ipoib_workqueue); | |
1da177e4 LT |
604 | |
605 | if (priv->broadcast && priv->broadcast->query) { | |
606 | ib_sa_cancel_query(priv->broadcast->query_id, priv->broadcast->query); | |
607 | priv->broadcast->query = NULL; | |
608 | ipoib_dbg_mcast(priv, "waiting for bcast\n"); | |
609 | wait_for_completion(&priv->broadcast->done); | |
610 | } | |
611 | ||
612 | list_for_each_entry(mcast, &priv->multicast_list, list) { | |
613 | if (mcast->query) { | |
614 | ib_sa_cancel_query(mcast->query_id, mcast->query); | |
615 | mcast->query = NULL; | |
616 | ipoib_dbg_mcast(priv, "waiting for MGID " IPOIB_GID_FMT "\n", | |
617 | IPOIB_GID_ARG(mcast->mcmember.mgid)); | |
618 | wait_for_completion(&mcast->done); | |
619 | } | |
620 | } | |
621 | ||
622 | return 0; | |
623 | } | |
624 | ||
625 | static int ipoib_mcast_leave(struct net_device *dev, struct ipoib_mcast *mcast) | |
626 | { | |
627 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
628 | struct ib_sa_mcmember_rec rec = { | |
629 | .join_state = 1 | |
630 | }; | |
631 | int ret = 0; | |
632 | ||
633 | if (!test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) | |
634 | return 0; | |
635 | ||
636 | ipoib_dbg_mcast(priv, "leaving MGID " IPOIB_GID_FMT "\n", | |
637 | IPOIB_GID_ARG(mcast->mcmember.mgid)); | |
638 | ||
639 | rec.mgid = mcast->mcmember.mgid; | |
640 | rec.port_gid = priv->local_gid; | |
97f52eb4 | 641 | rec.pkey = cpu_to_be16(priv->pkey); |
1da177e4 LT |
642 | |
643 | /* Remove ourselves from the multicast group */ | |
644 | ret = ipoib_mcast_detach(dev, be16_to_cpu(mcast->mcmember.mlid), | |
645 | &mcast->mcmember.mgid); | |
646 | if (ret) | |
647 | ipoib_warn(priv, "ipoib_mcast_detach failed (result = %d)\n", ret); | |
648 | ||
649 | /* | |
650 | * Just make one shot at leaving and don't wait for a reply; | |
651 | * if we fail, too bad. | |
652 | */ | |
653 | ret = ib_sa_mcmember_rec_delete(priv->ca, priv->port, &rec, | |
654 | IB_SA_MCMEMBER_REC_MGID | | |
655 | IB_SA_MCMEMBER_REC_PORT_GID | | |
656 | IB_SA_MCMEMBER_REC_PKEY | | |
657 | IB_SA_MCMEMBER_REC_JOIN_STATE, | |
658 | 0, GFP_ATOMIC, NULL, | |
659 | mcast, &mcast->query); | |
660 | if (ret < 0) | |
661 | ipoib_warn(priv, "ib_sa_mcmember_rec_delete failed " | |
662 | "for leave (result = %d)\n", ret); | |
663 | ||
664 | return 0; | |
665 | } | |
666 | ||
667 | void ipoib_mcast_send(struct net_device *dev, union ib_gid *mgid, | |
668 | struct sk_buff *skb) | |
669 | { | |
670 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
671 | struct ipoib_mcast *mcast; | |
672 | ||
673 | /* | |
674 | * We can only be called from ipoib_start_xmit, so we're | |
675 | * inside tx_lock -- no need to save/restore flags. | |
676 | */ | |
677 | spin_lock(&priv->lock); | |
678 | ||
679 | mcast = __ipoib_mcast_find(dev, mgid); | |
680 | if (!mcast) { | |
681 | /* Let's create a new send only group now */ | |
682 | ipoib_dbg_mcast(priv, "setting up send only multicast group for " | |
683 | IPOIB_GID_FMT "\n", IPOIB_GID_ARG(*mgid)); | |
684 | ||
685 | mcast = ipoib_mcast_alloc(dev, 0); | |
686 | if (!mcast) { | |
687 | ipoib_warn(priv, "unable to allocate memory for " | |
688 | "multicast structure\n"); | |
689 | dev_kfree_skb_any(skb); | |
690 | goto out; | |
691 | } | |
692 | ||
693 | set_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags); | |
694 | mcast->mcmember.mgid = *mgid; | |
695 | __ipoib_mcast_add(dev, mcast); | |
696 | list_add_tail(&mcast->list, &priv->multicast_list); | |
697 | } | |
698 | ||
699 | if (!mcast->ah) { | |
700 | if (skb_queue_len(&mcast->pkt_queue) < IPOIB_MAX_MCAST_QUEUE) | |
701 | skb_queue_tail(&mcast->pkt_queue, skb); | |
702 | else | |
703 | dev_kfree_skb_any(skb); | |
704 | ||
705 | if (mcast->query) | |
706 | ipoib_dbg_mcast(priv, "no address vector, " | |
707 | "but multicast join already started\n"); | |
708 | else if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) | |
709 | ipoib_mcast_sendonly_join(mcast); | |
710 | ||
711 | /* | |
712 | * If lookup completes between here and out:, don't | |
713 | * want to send packet twice. | |
714 | */ | |
715 | mcast = NULL; | |
716 | } | |
717 | ||
718 | out: | |
719 | if (mcast && mcast->ah) { | |
720 | if (skb->dst && | |
721 | skb->dst->neighbour && | |
722 | !*to_ipoib_neigh(skb->dst->neighbour)) { | |
723 | struct ipoib_neigh *neigh = kmalloc(sizeof *neigh, GFP_ATOMIC); | |
724 | ||
725 | if (neigh) { | |
726 | kref_get(&mcast->ah->ref); | |
727 | neigh->ah = mcast->ah; | |
728 | neigh->neighbour = skb->dst->neighbour; | |
729 | *to_ipoib_neigh(skb->dst->neighbour) = neigh; | |
730 | list_add_tail(&neigh->list, &mcast->neigh_list); | |
731 | } | |
732 | } | |
733 | ||
734 | ipoib_send(dev, skb, mcast->ah, IB_MULTICAST_QPN); | |
735 | } | |
736 | ||
737 | spin_unlock(&priv->lock); | |
738 | } | |
739 | ||
740 | void ipoib_mcast_dev_flush(struct net_device *dev) | |
741 | { | |
742 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
743 | LIST_HEAD(remove_list); | |
744 | struct ipoib_mcast *mcast, *tmcast, *nmcast; | |
745 | unsigned long flags; | |
746 | ||
747 | ipoib_dbg_mcast(priv, "flushing multicast list\n"); | |
748 | ||
749 | spin_lock_irqsave(&priv->lock, flags); | |
750 | list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) { | |
751 | nmcast = ipoib_mcast_alloc(dev, 0); | |
752 | if (nmcast) { | |
753 | nmcast->flags = | |
754 | mcast->flags & (1 << IPOIB_MCAST_FLAG_SENDONLY); | |
755 | ||
756 | nmcast->mcmember.mgid = mcast->mcmember.mgid; | |
757 | ||
758 | /* Add the new group in before the to-be-destroyed group */ | |
759 | list_add_tail(&nmcast->list, &mcast->list); | |
760 | list_del_init(&mcast->list); | |
761 | ||
762 | rb_replace_node(&mcast->rb_node, &nmcast->rb_node, | |
763 | &priv->multicast_tree); | |
764 | ||
765 | list_add_tail(&mcast->list, &remove_list); | |
766 | } else { | |
767 | ipoib_warn(priv, "could not reallocate multicast group " | |
768 | IPOIB_GID_FMT "\n", | |
769 | IPOIB_GID_ARG(mcast->mcmember.mgid)); | |
770 | } | |
771 | } | |
772 | ||
773 | if (priv->broadcast) { | |
774 | nmcast = ipoib_mcast_alloc(dev, 0); | |
775 | if (nmcast) { | |
776 | nmcast->mcmember.mgid = priv->broadcast->mcmember.mgid; | |
777 | ||
778 | rb_replace_node(&priv->broadcast->rb_node, | |
779 | &nmcast->rb_node, | |
780 | &priv->multicast_tree); | |
781 | ||
782 | list_add_tail(&priv->broadcast->list, &remove_list); | |
783 | } | |
784 | ||
785 | priv->broadcast = nmcast; | |
786 | } | |
787 | ||
788 | spin_unlock_irqrestore(&priv->lock, flags); | |
789 | ||
790 | list_for_each_entry_safe(mcast, tmcast, &remove_list, list) { | |
791 | ipoib_mcast_leave(dev, mcast); | |
792 | ipoib_mcast_free(mcast); | |
793 | } | |
794 | } | |
795 | ||
796 | void ipoib_mcast_dev_down(struct net_device *dev) | |
797 | { | |
798 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
799 | unsigned long flags; | |
800 | ||
801 | /* Delete broadcast since it will be recreated */ | |
802 | if (priv->broadcast) { | |
803 | ipoib_dbg_mcast(priv, "deleting broadcast group\n"); | |
804 | ||
805 | spin_lock_irqsave(&priv->lock, flags); | |
806 | rb_erase(&priv->broadcast->rb_node, &priv->multicast_tree); | |
807 | spin_unlock_irqrestore(&priv->lock, flags); | |
808 | ipoib_mcast_leave(dev, priv->broadcast); | |
809 | ipoib_mcast_free(priv->broadcast); | |
810 | priv->broadcast = NULL; | |
811 | } | |
812 | } | |
813 | ||
814 | void ipoib_mcast_restart_task(void *dev_ptr) | |
815 | { | |
816 | struct net_device *dev = dev_ptr; | |
817 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
818 | struct dev_mc_list *mclist; | |
819 | struct ipoib_mcast *mcast, *tmcast; | |
820 | LIST_HEAD(remove_list); | |
821 | unsigned long flags; | |
822 | ||
823 | ipoib_dbg_mcast(priv, "restarting multicast task\n"); | |
824 | ||
8d2cae06 | 825 | ipoib_mcast_stop_thread(dev, 0); |
1da177e4 LT |
826 | |
827 | spin_lock_irqsave(&priv->lock, flags); | |
828 | ||
829 | /* | |
830 | * Unfortunately, the networking core only gives us a list of all of | |
831 | * the multicast hardware addresses. We need to figure out which ones | |
832 | * are new and which ones have been removed | |
833 | */ | |
834 | ||
835 | /* Clear out the found flag */ | |
836 | list_for_each_entry(mcast, &priv->multicast_list, list) | |
837 | clear_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags); | |
838 | ||
839 | /* Mark all of the entries that are found or don't exist */ | |
840 | for (mclist = dev->mc_list; mclist; mclist = mclist->next) { | |
841 | union ib_gid mgid; | |
842 | ||
843 | memcpy(mgid.raw, mclist->dmi_addr + 4, sizeof mgid); | |
844 | ||
845 | /* Add in the P_Key */ | |
846 | mgid.raw[4] = (priv->pkey >> 8) & 0xff; | |
847 | mgid.raw[5] = priv->pkey & 0xff; | |
848 | ||
849 | mcast = __ipoib_mcast_find(dev, &mgid); | |
850 | if (!mcast || test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) { | |
851 | struct ipoib_mcast *nmcast; | |
852 | ||
853 | /* Not found or send-only group, let's add a new entry */ | |
854 | ipoib_dbg_mcast(priv, "adding multicast entry for mgid " | |
855 | IPOIB_GID_FMT "\n", IPOIB_GID_ARG(mgid)); | |
856 | ||
857 | nmcast = ipoib_mcast_alloc(dev, 0); | |
858 | if (!nmcast) { | |
859 | ipoib_warn(priv, "unable to allocate memory for multicast structure\n"); | |
860 | continue; | |
861 | } | |
862 | ||
863 | set_bit(IPOIB_MCAST_FLAG_FOUND, &nmcast->flags); | |
864 | ||
865 | nmcast->mcmember.mgid = mgid; | |
866 | ||
867 | if (mcast) { | |
868 | /* Destroy the send only entry */ | |
869 | list_del(&mcast->list); | |
870 | list_add_tail(&mcast->list, &remove_list); | |
871 | ||
872 | rb_replace_node(&mcast->rb_node, | |
873 | &nmcast->rb_node, | |
874 | &priv->multicast_tree); | |
875 | } else | |
876 | __ipoib_mcast_add(dev, nmcast); | |
877 | ||
878 | list_add_tail(&nmcast->list, &priv->multicast_list); | |
879 | } | |
880 | ||
881 | if (mcast) | |
882 | set_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags); | |
883 | } | |
884 | ||
885 | /* Remove all of the entries don't exist anymore */ | |
886 | list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) { | |
887 | if (!test_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags) && | |
888 | !test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) { | |
889 | ipoib_dbg_mcast(priv, "deleting multicast group " IPOIB_GID_FMT "\n", | |
890 | IPOIB_GID_ARG(mcast->mcmember.mgid)); | |
891 | ||
892 | rb_erase(&mcast->rb_node, &priv->multicast_tree); | |
893 | ||
894 | /* Move to the remove list */ | |
895 | list_del(&mcast->list); | |
896 | list_add_tail(&mcast->list, &remove_list); | |
897 | } | |
898 | } | |
899 | spin_unlock_irqrestore(&priv->lock, flags); | |
900 | ||
901 | /* We have to cancel outside of the spinlock */ | |
902 | list_for_each_entry_safe(mcast, tmcast, &remove_list, list) { | |
903 | ipoib_mcast_leave(mcast->dev, mcast); | |
904 | ipoib_mcast_free(mcast); | |
905 | } | |
906 | ||
907 | if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) | |
908 | ipoib_mcast_start_thread(dev); | |
909 | } | |
910 | ||
8ae5a8a2 RD |
911 | #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG |
912 | ||
1da177e4 LT |
913 | struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev) |
914 | { | |
915 | struct ipoib_mcast_iter *iter; | |
916 | ||
917 | iter = kmalloc(sizeof *iter, GFP_KERNEL); | |
918 | if (!iter) | |
919 | return NULL; | |
920 | ||
921 | iter->dev = dev; | |
1732b0ef | 922 | memset(iter->mgid.raw, 0, 16); |
1da177e4 LT |
923 | |
924 | if (ipoib_mcast_iter_next(iter)) { | |
1732b0ef | 925 | kfree(iter); |
1da177e4 LT |
926 | return NULL; |
927 | } | |
928 | ||
929 | return iter; | |
930 | } | |
931 | ||
1da177e4 LT |
932 | int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter) |
933 | { | |
934 | struct ipoib_dev_priv *priv = netdev_priv(iter->dev); | |
935 | struct rb_node *n; | |
936 | struct ipoib_mcast *mcast; | |
937 | int ret = 1; | |
938 | ||
939 | spin_lock_irq(&priv->lock); | |
940 | ||
941 | n = rb_first(&priv->multicast_tree); | |
942 | ||
943 | while (n) { | |
944 | mcast = rb_entry(n, struct ipoib_mcast, rb_node); | |
945 | ||
946 | if (memcmp(iter->mgid.raw, mcast->mcmember.mgid.raw, | |
947 | sizeof (union ib_gid)) < 0) { | |
948 | iter->mgid = mcast->mcmember.mgid; | |
949 | iter->created = mcast->created; | |
950 | iter->queuelen = skb_queue_len(&mcast->pkt_queue); | |
951 | iter->complete = !!mcast->ah; | |
952 | iter->send_only = !!(mcast->flags & (1 << IPOIB_MCAST_FLAG_SENDONLY)); | |
953 | ||
954 | ret = 0; | |
955 | ||
956 | break; | |
957 | } | |
958 | ||
959 | n = rb_next(n); | |
960 | } | |
961 | ||
962 | spin_unlock_irq(&priv->lock); | |
963 | ||
964 | return ret; | |
965 | } | |
966 | ||
967 | void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter, | |
968 | union ib_gid *mgid, | |
969 | unsigned long *created, | |
970 | unsigned int *queuelen, | |
971 | unsigned int *complete, | |
972 | unsigned int *send_only) | |
973 | { | |
974 | *mgid = iter->mgid; | |
975 | *created = iter->created; | |
976 | *queuelen = iter->queuelen; | |
977 | *complete = iter->complete; | |
978 | *send_only = iter->send_only; | |
979 | } | |
8ae5a8a2 RD |
980 | |
981 | #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */ |