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  * $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 <net/dst.h>
47 
48 #include "ipoib.h"
49 
50 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
51 static int mcast_debug_level;
52 
53 module_param(mcast_debug_level, int, 0644);
54 MODULE_PARM_DESC(mcast_debug_level,
55 		 "Enable multicast debug tracing if > 0");
56 #endif
57 
58 static DEFINE_MUTEX(mcast_mutex);
59 
60 struct ipoib_mcast_iter {
61 	struct net_device *dev;
62 	union ib_gid       mgid;
63 	unsigned long      created;
64 	unsigned int       queuelen;
65 	unsigned int       complete;
66 	unsigned int       send_only;
67 };
68 
69 static void ipoib_mcast_free(struct ipoib_mcast *mcast)
70 {
71 	struct net_device *dev = mcast->dev;
72 	struct ipoib_dev_priv *priv = netdev_priv(dev);
73 	struct ipoib_neigh *neigh, *tmp;
74 	unsigned long flags;
75 	int tx_dropped = 0;
76 
77 	ipoib_dbg_mcast(netdev_priv(dev),
78 			"deleting multicast group " IPOIB_GID_FMT "\n",
79 			IPOIB_GID_ARG(mcast->mcmember.mgid));
80 
81 	spin_lock_irqsave(&priv->lock, flags);
82 
83 	list_for_each_entry_safe(neigh, tmp, &mcast->neigh_list, list) {
84 		/*
85 		 * It's safe to call ipoib_put_ah() inside priv->lock
86 		 * here, because we know that mcast->ah will always
87 		 * hold one more reference, so ipoib_put_ah() will
88 		 * never do more than decrement the ref count.
89 		 */
90 		if (neigh->ah)
91 			ipoib_put_ah(neigh->ah);
92 		ipoib_neigh_free(dev, neigh);
93 	}
94 
95 	spin_unlock_irqrestore(&priv->lock, flags);
96 
97 	if (mcast->ah)
98 		ipoib_put_ah(mcast->ah);
99 
100 	while (!skb_queue_empty(&mcast->pkt_queue)) {
101 		++tx_dropped;
102 		dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
103 	}
104 
105 	spin_lock_irqsave(&priv->tx_lock, flags);
106 	dev->stats.tx_dropped += tx_dropped;
107 	spin_unlock_irqrestore(&priv->tx_lock, flags);
108 
109 	kfree(mcast);
110 }
111 
112 static struct ipoib_mcast *ipoib_mcast_alloc(struct net_device *dev,
113 					     int can_sleep)
114 {
115 	struct ipoib_mcast *mcast;
116 
117 	mcast = kzalloc(sizeof *mcast, can_sleep ? GFP_KERNEL : GFP_ATOMIC);
118 	if (!mcast)
119 		return NULL;
120 
121 	mcast->dev = dev;
122 	mcast->created = jiffies;
123 	mcast->backoff = 1;
124 
125 	INIT_LIST_HEAD(&mcast->list);
126 	INIT_LIST_HEAD(&mcast->neigh_list);
127 	skb_queue_head_init(&mcast->pkt_queue);
128 
129 	return mcast;
130 }
131 
132 static struct ipoib_mcast *__ipoib_mcast_find(struct net_device *dev, void *mgid)
133 {
134 	struct ipoib_dev_priv *priv = netdev_priv(dev);
135 	struct rb_node *n = priv->multicast_tree.rb_node;
136 
137 	while (n) {
138 		struct ipoib_mcast *mcast;
139 		int ret;
140 
141 		mcast = rb_entry(n, struct ipoib_mcast, rb_node);
142 
143 		ret = memcmp(mgid, mcast->mcmember.mgid.raw,
144 			     sizeof (union ib_gid));
145 		if (ret < 0)
146 			n = n->rb_left;
147 		else if (ret > 0)
148 			n = n->rb_right;
149 		else
150 			return mcast;
151 	}
152 
153 	return NULL;
154 }
155 
156 static int __ipoib_mcast_add(struct net_device *dev, struct ipoib_mcast *mcast)
157 {
158 	struct ipoib_dev_priv *priv = netdev_priv(dev);
159 	struct rb_node **n = &priv->multicast_tree.rb_node, *pn = NULL;
160 
161 	while (*n) {
162 		struct ipoib_mcast *tmcast;
163 		int ret;
164 
165 		pn = *n;
166 		tmcast = rb_entry(pn, struct ipoib_mcast, rb_node);
167 
168 		ret = memcmp(mcast->mcmember.mgid.raw, tmcast->mcmember.mgid.raw,
169 			     sizeof (union ib_gid));
170 		if (ret < 0)
171 			n = &pn->rb_left;
172 		else if (ret > 0)
173 			n = &pn->rb_right;
174 		else
175 			return -EEXIST;
176 	}
177 
178 	rb_link_node(&mcast->rb_node, pn, n);
179 	rb_insert_color(&mcast->rb_node, &priv->multicast_tree);
180 
181 	return 0;
182 }
183 
184 static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
185 				   struct ib_sa_mcmember_rec *mcmember)
186 {
187 	struct net_device *dev = mcast->dev;
188 	struct ipoib_dev_priv *priv = netdev_priv(dev);
189 	struct ipoib_ah *ah;
190 	int ret;
191 
192 	mcast->mcmember = *mcmember;
193 
194 	/* Set the cached Q_Key before we attach if it's the broadcast group */
195 	if (!memcmp(mcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
196 		    sizeof (union ib_gid))) {
197 		priv->qkey = be32_to_cpu(priv->broadcast->mcmember.qkey);
198 		priv->tx_wr.wr.ud.remote_qkey = priv->qkey;
199 	}
200 
201 	if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
202 		if (test_and_set_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
203 			ipoib_warn(priv, "multicast group " IPOIB_GID_FMT
204 				   " already attached\n",
205 				   IPOIB_GID_ARG(mcast->mcmember.mgid));
206 
207 			return 0;
208 		}
209 
210 		ret = ipoib_mcast_attach(dev, be16_to_cpu(mcast->mcmember.mlid),
211 					 &mcast->mcmember.mgid);
212 		if (ret < 0) {
213 			ipoib_warn(priv, "couldn't attach QP to multicast group "
214 				   IPOIB_GID_FMT "\n",
215 				   IPOIB_GID_ARG(mcast->mcmember.mgid));
216 
217 			clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags);
218 			return ret;
219 		}
220 	}
221 
222 	{
223 		struct ib_ah_attr av = {
224 			.dlid	       = be16_to_cpu(mcast->mcmember.mlid),
225 			.port_num      = priv->port,
226 			.sl	       = mcast->mcmember.sl,
227 			.ah_flags      = IB_AH_GRH,
228 			.static_rate   = mcast->mcmember.rate,
229 			.grh	       = {
230 				.flow_label    = be32_to_cpu(mcast->mcmember.flow_label),
231 				.hop_limit     = mcast->mcmember.hop_limit,
232 				.sgid_index    = 0,
233 				.traffic_class = mcast->mcmember.traffic_class
234 			}
235 		};
236 		av.grh.dgid = mcast->mcmember.mgid;
237 
238 		ah = ipoib_create_ah(dev, priv->pd, &av);
239 		if (!ah) {
240 			ipoib_warn(priv, "ib_address_create failed\n");
241 		} else {
242 			spin_lock_irq(&priv->lock);
243 			mcast->ah = ah;
244 			spin_unlock_irq(&priv->lock);
245 
246 			ipoib_dbg_mcast(priv, "MGID " IPOIB_GID_FMT
247 					" AV %p, LID 0x%04x, SL %d\n",
248 					IPOIB_GID_ARG(mcast->mcmember.mgid),
249 					mcast->ah->ah,
250 					be16_to_cpu(mcast->mcmember.mlid),
251 					mcast->mcmember.sl);
252 		}
253 	}
254 
255 	/* actually send any queued packets */
256 	spin_lock_irq(&priv->tx_lock);
257 	while (!skb_queue_empty(&mcast->pkt_queue)) {
258 		struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue);
259 		spin_unlock_irq(&priv->tx_lock);
260 
261 		skb->dev = dev;
262 
263 		if (!skb->dst || !skb->dst->neighbour) {
264 			/* put pseudoheader back on for next time */
265 			skb_push(skb, sizeof (struct ipoib_pseudoheader));
266 		}
267 
268 		if (dev_queue_xmit(skb))
269 			ipoib_warn(priv, "dev_queue_xmit failed to requeue packet\n");
270 		spin_lock_irq(&priv->tx_lock);
271 	}
272 	spin_unlock_irq(&priv->tx_lock);
273 
274 	return 0;
275 }
276 
277 static int
278 ipoib_mcast_sendonly_join_complete(int status,
279 				   struct ib_sa_multicast *multicast)
280 {
281 	struct ipoib_mcast *mcast = multicast->context;
282 	struct net_device *dev = mcast->dev;
283 	struct ipoib_dev_priv *priv = netdev_priv(dev);
284 
285 	/* We trap for port events ourselves. */
286 	if (status == -ENETRESET)
287 		return 0;
288 
289 	if (!status)
290 		status = ipoib_mcast_join_finish(mcast, &multicast->rec);
291 
292 	if (status) {
293 		if (mcast->logcount++ < 20)
294 			ipoib_dbg_mcast(netdev_priv(dev), "multicast join failed for "
295 					IPOIB_GID_FMT ", status %d\n",
296 					IPOIB_GID_ARG(mcast->mcmember.mgid), status);
297 
298 		/* Flush out any queued packets */
299 		spin_lock_irq(&priv->tx_lock);
300 		while (!skb_queue_empty(&mcast->pkt_queue)) {
301 			++dev->stats.tx_dropped;
302 			dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
303 		}
304 		spin_unlock_irq(&priv->tx_lock);
305 
306 		/* Clear the busy flag so we try again */
307 		status = test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY,
308 					    &mcast->flags);
309 	}
310 	return status;
311 }
312 
313 static int ipoib_mcast_sendonly_join(struct ipoib_mcast *mcast)
314 {
315 	struct net_device *dev = mcast->dev;
316 	struct ipoib_dev_priv *priv = netdev_priv(dev);
317 	struct ib_sa_mcmember_rec rec = {
318 #if 0				/* Some SMs don't support send-only yet */
319 		.join_state = 4
320 #else
321 		.join_state = 1
322 #endif
323 	};
324 	int ret = 0;
325 
326 	if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
327 		ipoib_dbg_mcast(priv, "device shutting down, no multicast joins\n");
328 		return -ENODEV;
329 	}
330 
331 	if (test_and_set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) {
332 		ipoib_dbg_mcast(priv, "multicast entry busy, skipping\n");
333 		return -EBUSY;
334 	}
335 
336 	rec.mgid     = mcast->mcmember.mgid;
337 	rec.port_gid = priv->local_gid;
338 	rec.pkey     = cpu_to_be16(priv->pkey);
339 
340 	mcast->mc = ib_sa_join_multicast(&ipoib_sa_client, priv->ca,
341 					 priv->port, &rec,
342 					 IB_SA_MCMEMBER_REC_MGID	|
343 					 IB_SA_MCMEMBER_REC_PORT_GID	|
344 					 IB_SA_MCMEMBER_REC_PKEY	|
345 					 IB_SA_MCMEMBER_REC_JOIN_STATE,
346 					 GFP_ATOMIC,
347 					 ipoib_mcast_sendonly_join_complete,
348 					 mcast);
349 	if (IS_ERR(mcast->mc)) {
350 		ret = PTR_ERR(mcast->mc);
351 		clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
352 		ipoib_warn(priv, "ib_sa_join_multicast failed (ret = %d)\n",
353 			   ret);
354 	} else {
355 		ipoib_dbg_mcast(priv, "no multicast record for " IPOIB_GID_FMT
356 				", starting join\n",
357 				IPOIB_GID_ARG(mcast->mcmember.mgid));
358 	}
359 
360 	return ret;
361 }
362 
363 static int ipoib_mcast_join_complete(int status,
364 				     struct ib_sa_multicast *multicast)
365 {
366 	struct ipoib_mcast *mcast = multicast->context;
367 	struct net_device *dev = mcast->dev;
368 	struct ipoib_dev_priv *priv = netdev_priv(dev);
369 
370 	ipoib_dbg_mcast(priv, "join completion for " IPOIB_GID_FMT
371 			" (status %d)\n",
372 			IPOIB_GID_ARG(mcast->mcmember.mgid), status);
373 
374 	/* We trap for port events ourselves. */
375 	if (status == -ENETRESET)
376 		return 0;
377 
378 	if (!status)
379 		status = ipoib_mcast_join_finish(mcast, &multicast->rec);
380 
381 	if (!status) {
382 		mcast->backoff = 1;
383 		mutex_lock(&mcast_mutex);
384 		if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
385 			queue_delayed_work(ipoib_workqueue,
386 					   &priv->mcast_task, 0);
387 		mutex_unlock(&mcast_mutex);
388 
389 		if (mcast == priv->broadcast)
390 			netif_carrier_on(dev);
391 
392 		return 0;
393 	}
394 
395 	if (mcast->logcount++ < 20) {
396 		if (status == -ETIMEDOUT) {
397 			ipoib_dbg_mcast(priv, "multicast join failed for " IPOIB_GID_FMT
398 					", status %d\n",
399 					IPOIB_GID_ARG(mcast->mcmember.mgid),
400 					status);
401 		} else {
402 			ipoib_warn(priv, "multicast join failed for "
403 				   IPOIB_GID_FMT ", status %d\n",
404 				   IPOIB_GID_ARG(mcast->mcmember.mgid),
405 				   status);
406 		}
407 	}
408 
409 	mcast->backoff *= 2;
410 	if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
411 		mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
412 
413 	/* Clear the busy flag so we try again */
414 	status = test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
415 
416 	mutex_lock(&mcast_mutex);
417 	spin_lock_irq(&priv->lock);
418 	if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
419 		queue_delayed_work(ipoib_workqueue, &priv->mcast_task,
420 				   mcast->backoff * HZ);
421 	spin_unlock_irq(&priv->lock);
422 	mutex_unlock(&mcast_mutex);
423 
424 	return status;
425 }
426 
427 static void ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast,
428 			     int create)
429 {
430 	struct ipoib_dev_priv *priv = netdev_priv(dev);
431 	struct ib_sa_mcmember_rec rec = {
432 		.join_state = 1
433 	};
434 	ib_sa_comp_mask comp_mask;
435 	int ret = 0;
436 
437 	ipoib_dbg_mcast(priv, "joining MGID " IPOIB_GID_FMT "\n",
438 			IPOIB_GID_ARG(mcast->mcmember.mgid));
439 
440 	rec.mgid     = mcast->mcmember.mgid;
441 	rec.port_gid = priv->local_gid;
442 	rec.pkey     = cpu_to_be16(priv->pkey);
443 
444 	comp_mask =
445 		IB_SA_MCMEMBER_REC_MGID		|
446 		IB_SA_MCMEMBER_REC_PORT_GID	|
447 		IB_SA_MCMEMBER_REC_PKEY		|
448 		IB_SA_MCMEMBER_REC_JOIN_STATE;
449 
450 	if (create) {
451 		comp_mask |=
452 			IB_SA_MCMEMBER_REC_QKEY			|
453 			IB_SA_MCMEMBER_REC_MTU_SELECTOR		|
454 			IB_SA_MCMEMBER_REC_MTU			|
455 			IB_SA_MCMEMBER_REC_TRAFFIC_CLASS	|
456 			IB_SA_MCMEMBER_REC_RATE_SELECTOR	|
457 			IB_SA_MCMEMBER_REC_RATE			|
458 			IB_SA_MCMEMBER_REC_SL			|
459 			IB_SA_MCMEMBER_REC_FLOW_LABEL		|
460 			IB_SA_MCMEMBER_REC_HOP_LIMIT;
461 
462 		rec.qkey	  = priv->broadcast->mcmember.qkey;
463 		rec.mtu_selector  = IB_SA_EQ;
464 		rec.mtu		  = priv->broadcast->mcmember.mtu;
465 		rec.traffic_class = priv->broadcast->mcmember.traffic_class;
466 		rec.rate_selector = IB_SA_EQ;
467 		rec.rate	  = priv->broadcast->mcmember.rate;
468 		rec.sl		  = priv->broadcast->mcmember.sl;
469 		rec.flow_label	  = priv->broadcast->mcmember.flow_label;
470 		rec.hop_limit	  = priv->broadcast->mcmember.hop_limit;
471 	}
472 
473 	set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
474 	mcast->mc = ib_sa_join_multicast(&ipoib_sa_client, priv->ca, priv->port,
475 					 &rec, comp_mask, GFP_KERNEL,
476 					 ipoib_mcast_join_complete, mcast);
477 	if (IS_ERR(mcast->mc)) {
478 		clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
479 		ret = PTR_ERR(mcast->mc);
480 		ipoib_warn(priv, "ib_sa_join_multicast failed, status %d\n", ret);
481 
482 		mcast->backoff *= 2;
483 		if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
484 			mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
485 
486 		mutex_lock(&mcast_mutex);
487 		if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
488 			queue_delayed_work(ipoib_workqueue,
489 					   &priv->mcast_task,
490 					   mcast->backoff * HZ);
491 		mutex_unlock(&mcast_mutex);
492 	}
493 }
494 
495 void ipoib_mcast_join_task(struct work_struct *work)
496 {
497 	struct ipoib_dev_priv *priv =
498 		container_of(work, struct ipoib_dev_priv, mcast_task.work);
499 	struct net_device *dev = priv->dev;
500 
501 	if (!test_bit(IPOIB_MCAST_RUN, &priv->flags))
502 		return;
503 
504 	if (ib_query_gid(priv->ca, priv->port, 0, &priv->local_gid))
505 		ipoib_warn(priv, "ib_query_gid() failed\n");
506 	else
507 		memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw, sizeof (union ib_gid));
508 
509 	{
510 		struct ib_port_attr attr;
511 
512 		if (!ib_query_port(priv->ca, priv->port, &attr))
513 			priv->local_lid = attr.lid;
514 		else
515 			ipoib_warn(priv, "ib_query_port failed\n");
516 	}
517 
518 	if (!priv->broadcast) {
519 		struct ipoib_mcast *broadcast;
520 
521 		broadcast = ipoib_mcast_alloc(dev, 1);
522 		if (!broadcast) {
523 			ipoib_warn(priv, "failed to allocate broadcast group\n");
524 			mutex_lock(&mcast_mutex);
525 			if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
526 				queue_delayed_work(ipoib_workqueue,
527 						   &priv->mcast_task, HZ);
528 			mutex_unlock(&mcast_mutex);
529 			return;
530 		}
531 
532 		spin_lock_irq(&priv->lock);
533 		memcpy(broadcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
534 		       sizeof (union ib_gid));
535 		priv->broadcast = broadcast;
536 
537 		__ipoib_mcast_add(dev, priv->broadcast);
538 		spin_unlock_irq(&priv->lock);
539 	}
540 
541 	if (!test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
542 		if (!test_bit(IPOIB_MCAST_FLAG_BUSY, &priv->broadcast->flags))
543 			ipoib_mcast_join(dev, priv->broadcast, 0);
544 		return;
545 	}
546 
547 	while (1) {
548 		struct ipoib_mcast *mcast = NULL;
549 
550 		spin_lock_irq(&priv->lock);
551 		list_for_each_entry(mcast, &priv->multicast_list, list) {
552 			if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)
553 			    && !test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)
554 			    && !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
555 				/* Found the next unjoined group */
556 				break;
557 			}
558 		}
559 		spin_unlock_irq(&priv->lock);
560 
561 		if (&mcast->list == &priv->multicast_list) {
562 			/* All done */
563 			break;
564 		}
565 
566 		ipoib_mcast_join(dev, mcast, 1);
567 		return;
568 	}
569 
570 	priv->mcast_mtu = ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu) -
571 		IPOIB_ENCAP_LEN;
572 
573 	if (!ipoib_cm_admin_enabled(dev))
574 		dev->mtu = min(priv->mcast_mtu, priv->admin_mtu);
575 
576 	ipoib_dbg_mcast(priv, "successfully joined all multicast groups\n");
577 
578 	clear_bit(IPOIB_MCAST_RUN, &priv->flags);
579 }
580 
581 int ipoib_mcast_start_thread(struct net_device *dev)
582 {
583 	struct ipoib_dev_priv *priv = netdev_priv(dev);
584 
585 	ipoib_dbg_mcast(priv, "starting multicast thread\n");
586 
587 	mutex_lock(&mcast_mutex);
588 	if (!test_and_set_bit(IPOIB_MCAST_RUN, &priv->flags))
589 		queue_delayed_work(ipoib_workqueue, &priv->mcast_task, 0);
590 	mutex_unlock(&mcast_mutex);
591 
592 	spin_lock_irq(&priv->lock);
593 	set_bit(IPOIB_MCAST_STARTED, &priv->flags);
594 	spin_unlock_irq(&priv->lock);
595 
596 	return 0;
597 }
598 
599 int ipoib_mcast_stop_thread(struct net_device *dev, int flush)
600 {
601 	struct ipoib_dev_priv *priv = netdev_priv(dev);
602 
603 	ipoib_dbg_mcast(priv, "stopping multicast thread\n");
604 
605 	spin_lock_irq(&priv->lock);
606 	clear_bit(IPOIB_MCAST_STARTED, &priv->flags);
607 	spin_unlock_irq(&priv->lock);
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 = ipoib_mcast_detach(dev, be16_to_cpu(mcast->mcmember.mlid),
634 					 &mcast->mcmember.mgid);
635 		if (ret)
636 			ipoib_warn(priv, "ipoib_mcast_detach 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_tx_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_tx_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