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