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 #include <linux/slab.h>
44 
45 #include <net/dst.h>
46 
47 #include "ipoib.h"
48 
49 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
50 static int mcast_debug_level;
51 
52 module_param(mcast_debug_level, int, 0644);
53 MODULE_PARM_DESC(mcast_debug_level,
54 		 "Enable multicast debug tracing if > 0");
55 #endif
56 
57 static DEFINE_MUTEX(mcast_mutex);
58 
59 struct ipoib_mcast_iter {
60 	struct net_device *dev;
61 	union ib_gid       mgid;
62 	unsigned long      created;
63 	unsigned int       queuelen;
64 	unsigned int       complete;
65 	unsigned int       send_only;
66 };
67 
68 static void ipoib_mcast_free(struct ipoib_mcast *mcast)
69 {
70 	struct net_device *dev = mcast->dev;
71 	struct ipoib_dev_priv *priv = netdev_priv(dev);
72 	struct ipoib_neigh *neigh, *tmp;
73 	int tx_dropped = 0;
74 
75 	ipoib_dbg_mcast(netdev_priv(dev), "deleting multicast group %pI6\n",
76 			mcast->mcmember.mgid.raw);
77 
78 	spin_lock_irq(&priv->lock);
79 
80 	list_for_each_entry_safe(neigh, tmp, &mcast->neigh_list, list) {
81 		/*
82 		 * It's safe to call ipoib_put_ah() inside priv->lock
83 		 * here, because we know that mcast->ah will always
84 		 * hold one more reference, so ipoib_put_ah() will
85 		 * never do more than decrement the ref count.
86 		 */
87 		if (neigh->ah)
88 			ipoib_put_ah(neigh->ah);
89 		ipoib_neigh_free(dev, neigh);
90 	}
91 
92 	spin_unlock_irq(&priv->lock);
93 
94 	if (mcast->ah)
95 		ipoib_put_ah(mcast->ah);
96 
97 	while (!skb_queue_empty(&mcast->pkt_queue)) {
98 		++tx_dropped;
99 		dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
100 	}
101 
102 	netif_tx_lock_bh(dev);
103 	dev->stats.tx_dropped += tx_dropped;
104 	netif_tx_unlock_bh(dev);
105 
106 	kfree(mcast);
107 }
108 
109 static struct ipoib_mcast *ipoib_mcast_alloc(struct net_device *dev,
110 					     int can_sleep)
111 {
112 	struct ipoib_mcast *mcast;
113 
114 	mcast = kzalloc(sizeof *mcast, can_sleep ? GFP_KERNEL : GFP_ATOMIC);
115 	if (!mcast)
116 		return NULL;
117 
118 	mcast->dev = dev;
119 	mcast->created = jiffies;
120 	mcast->backoff = 1;
121 
122 	INIT_LIST_HEAD(&mcast->list);
123 	INIT_LIST_HEAD(&mcast->neigh_list);
124 	skb_queue_head_init(&mcast->pkt_queue);
125 
126 	return mcast;
127 }
128 
129 static struct ipoib_mcast *__ipoib_mcast_find(struct net_device *dev, void *mgid)
130 {
131 	struct ipoib_dev_priv *priv = netdev_priv(dev);
132 	struct rb_node *n = priv->multicast_tree.rb_node;
133 
134 	while (n) {
135 		struct ipoib_mcast *mcast;
136 		int ret;
137 
138 		mcast = rb_entry(n, struct ipoib_mcast, rb_node);
139 
140 		ret = memcmp(mgid, mcast->mcmember.mgid.raw,
141 			     sizeof (union ib_gid));
142 		if (ret < 0)
143 			n = n->rb_left;
144 		else if (ret > 0)
145 			n = n->rb_right;
146 		else
147 			return mcast;
148 	}
149 
150 	return NULL;
151 }
152 
153 static int __ipoib_mcast_add(struct net_device *dev, struct ipoib_mcast *mcast)
154 {
155 	struct ipoib_dev_priv *priv = netdev_priv(dev);
156 	struct rb_node **n = &priv->multicast_tree.rb_node, *pn = NULL;
157 
158 	while (*n) {
159 		struct ipoib_mcast *tmcast;
160 		int ret;
161 
162 		pn = *n;
163 		tmcast = rb_entry(pn, struct ipoib_mcast, rb_node);
164 
165 		ret = memcmp(mcast->mcmember.mgid.raw, tmcast->mcmember.mgid.raw,
166 			     sizeof (union ib_gid));
167 		if (ret < 0)
168 			n = &pn->rb_left;
169 		else if (ret > 0)
170 			n = &pn->rb_right;
171 		else
172 			return -EEXIST;
173 	}
174 
175 	rb_link_node(&mcast->rb_node, pn, n);
176 	rb_insert_color(&mcast->rb_node, &priv->multicast_tree);
177 
178 	return 0;
179 }
180 
181 static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
182 				   struct ib_sa_mcmember_rec *mcmember)
183 {
184 	struct net_device *dev = mcast->dev;
185 	struct ipoib_dev_priv *priv = netdev_priv(dev);
186 	struct ipoib_ah *ah;
187 	int ret;
188 	int set_qkey = 0;
189 
190 	mcast->mcmember = *mcmember;
191 
192 	/* Set the cached Q_Key before we attach if it's the broadcast group */
193 	if (!memcmp(mcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
194 		    sizeof (union ib_gid))) {
195 		spin_lock_irq(&priv->lock);
196 		if (!priv->broadcast) {
197 			spin_unlock_irq(&priv->lock);
198 			return -EAGAIN;
199 		}
200 		priv->qkey = be32_to_cpu(priv->broadcast->mcmember.qkey);
201 		spin_unlock_irq(&priv->lock);
202 		priv->tx_wr.wr.ud.remote_qkey = priv->qkey;
203 		set_qkey = 1;
204 	}
205 
206 	if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
207 		if (test_and_set_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
208 			ipoib_warn(priv, "multicast group %pI6 already attached\n",
209 				   mcast->mcmember.mgid.raw);
210 
211 			return 0;
212 		}
213 
214 		ret = ipoib_mcast_attach(dev, be16_to_cpu(mcast->mcmember.mlid),
215 					 &mcast->mcmember.mgid, set_qkey);
216 		if (ret < 0) {
217 			ipoib_warn(priv, "couldn't attach QP to multicast group %pI6\n",
218 				   mcast->mcmember.mgid.raw);
219 
220 			clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags);
221 			return ret;
222 		}
223 	}
224 
225 	{
226 		struct ib_ah_attr av = {
227 			.dlid	       = be16_to_cpu(mcast->mcmember.mlid),
228 			.port_num      = priv->port,
229 			.sl	       = mcast->mcmember.sl,
230 			.ah_flags      = IB_AH_GRH,
231 			.static_rate   = mcast->mcmember.rate,
232 			.grh	       = {
233 				.flow_label    = be32_to_cpu(mcast->mcmember.flow_label),
234 				.hop_limit     = mcast->mcmember.hop_limit,
235 				.sgid_index    = 0,
236 				.traffic_class = mcast->mcmember.traffic_class
237 			}
238 		};
239 		av.grh.dgid = mcast->mcmember.mgid;
240 
241 		ah = ipoib_create_ah(dev, priv->pd, &av);
242 		if (!ah) {
243 			ipoib_warn(priv, "ib_address_create failed\n");
244 		} else {
245 			spin_lock_irq(&priv->lock);
246 			mcast->ah = ah;
247 			spin_unlock_irq(&priv->lock);
248 
249 			ipoib_dbg_mcast(priv, "MGID %pI6 AV %p, LID 0x%04x, SL %d\n",
250 					mcast->mcmember.mgid.raw,
251 					mcast->ah->ah,
252 					be16_to_cpu(mcast->mcmember.mlid),
253 					mcast->mcmember.sl);
254 		}
255 	}
256 
257 	/* actually send any queued packets */
258 	netif_tx_lock_bh(dev);
259 	while (!skb_queue_empty(&mcast->pkt_queue)) {
260 		struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue);
261 		struct dst_entry *dst = skb_dst(skb);
262 		struct neighbour *n = NULL;
263 
264 		netif_tx_unlock_bh(dev);
265 
266 		skb->dev = dev;
267 		if (dst)
268 			n = dst_get_neighbour(dst);
269 		if (!dst || !n) {
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 		netif_tx_lock_bh(dev);
277 	}
278 	netif_tx_unlock_bh(dev);
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 
290 	/* We trap for port events ourselves. */
291 	if (status == -ENETRESET)
292 		return 0;
293 
294 	if (!status)
295 		status = ipoib_mcast_join_finish(mcast, &multicast->rec);
296 
297 	if (status) {
298 		if (mcast->logcount++ < 20)
299 			ipoib_dbg_mcast(netdev_priv(dev), "multicast join failed for %pI6, status %d\n",
300 					mcast->mcmember.mgid.raw, status);
301 
302 		/* Flush out any queued packets */
303 		netif_tx_lock_bh(dev);
304 		while (!skb_queue_empty(&mcast->pkt_queue)) {
305 			++dev->stats.tx_dropped;
306 			dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
307 		}
308 		netif_tx_unlock_bh(dev);
309 
310 		/* Clear the busy flag so we try again */
311 		status = test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY,
312 					    &mcast->flags);
313 	}
314 	return status;
315 }
316 
317 static int ipoib_mcast_sendonly_join(struct ipoib_mcast *mcast)
318 {
319 	struct net_device *dev = mcast->dev;
320 	struct ipoib_dev_priv *priv = netdev_priv(dev);
321 	struct ib_sa_mcmember_rec rec = {
322 #if 0				/* Some SMs don't support send-only yet */
323 		.join_state = 4
324 #else
325 		.join_state = 1
326 #endif
327 	};
328 	int ret = 0;
329 
330 	if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
331 		ipoib_dbg_mcast(priv, "device shutting down, no multicast joins\n");
332 		return -ENODEV;
333 	}
334 
335 	if (test_and_set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) {
336 		ipoib_dbg_mcast(priv, "multicast entry busy, skipping\n");
337 		return -EBUSY;
338 	}
339 
340 	rec.mgid     = mcast->mcmember.mgid;
341 	rec.port_gid = priv->local_gid;
342 	rec.pkey     = cpu_to_be16(priv->pkey);
343 
344 	mcast->mc = ib_sa_join_multicast(&ipoib_sa_client, priv->ca,
345 					 priv->port, &rec,
346 					 IB_SA_MCMEMBER_REC_MGID	|
347 					 IB_SA_MCMEMBER_REC_PORT_GID	|
348 					 IB_SA_MCMEMBER_REC_PKEY	|
349 					 IB_SA_MCMEMBER_REC_JOIN_STATE,
350 					 GFP_ATOMIC,
351 					 ipoib_mcast_sendonly_join_complete,
352 					 mcast);
353 	if (IS_ERR(mcast->mc)) {
354 		ret = PTR_ERR(mcast->mc);
355 		clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
356 		ipoib_warn(priv, "ib_sa_join_multicast failed (ret = %d)\n",
357 			   ret);
358 	} else {
359 		ipoib_dbg_mcast(priv, "no multicast record for %pI6, starting join\n",
360 				mcast->mcmember.mgid.raw);
361 	}
362 
363 	return ret;
364 }
365 
366 void ipoib_mcast_carrier_on_task(struct work_struct *work)
367 {
368 	struct ipoib_dev_priv *priv = container_of(work, struct ipoib_dev_priv,
369 						   carrier_on_task);
370 	struct ib_port_attr attr;
371 
372 	/*
373 	 * Take rtnl_lock to avoid racing with ipoib_stop() and
374 	 * turning the carrier back on while a device is being
375 	 * removed.
376 	 */
377 	if (ib_query_port(priv->ca, priv->port, &attr) ||
378 	    attr.state != IB_PORT_ACTIVE) {
379 		ipoib_dbg(priv, "Keeping carrier off until IB port is active\n");
380 		return;
381 	}
382 
383 	rtnl_lock();
384 	netif_carrier_on(priv->dev);
385 	rtnl_unlock();
386 }
387 
388 static int ipoib_mcast_join_complete(int status,
389 				     struct ib_sa_multicast *multicast)
390 {
391 	struct ipoib_mcast *mcast = multicast->context;
392 	struct net_device *dev = mcast->dev;
393 	struct ipoib_dev_priv *priv = netdev_priv(dev);
394 
395 	ipoib_dbg_mcast(priv, "join completion for %pI6 (status %d)\n",
396 			mcast->mcmember.mgid.raw, status);
397 
398 	/* We trap for port events ourselves. */
399 	if (status == -ENETRESET)
400 		return 0;
401 
402 	if (!status)
403 		status = ipoib_mcast_join_finish(mcast, &multicast->rec);
404 
405 	if (!status) {
406 		mcast->backoff = 1;
407 		mutex_lock(&mcast_mutex);
408 		if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
409 			queue_delayed_work(ipoib_workqueue,
410 					   &priv->mcast_task, 0);
411 		mutex_unlock(&mcast_mutex);
412 
413 		/*
414 		 * Defer carrier on work to ipoib_workqueue to avoid a
415 		 * deadlock on rtnl_lock here.
416 		 */
417 		if (mcast == priv->broadcast)
418 			queue_work(ipoib_workqueue, &priv->carrier_on_task);
419 
420 		return 0;
421 	}
422 
423 	if (mcast->logcount++ < 20) {
424 		if (status == -ETIMEDOUT || status == -EAGAIN) {
425 			ipoib_dbg_mcast(priv, "multicast join failed for %pI6, status %d\n",
426 					mcast->mcmember.mgid.raw, status);
427 		} else {
428 			ipoib_warn(priv, "multicast join failed for %pI6, status %d\n",
429 				   mcast->mcmember.mgid.raw, status);
430 		}
431 	}
432 
433 	mcast->backoff *= 2;
434 	if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
435 		mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
436 
437 	/* Clear the busy flag so we try again */
438 	status = test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
439 
440 	mutex_lock(&mcast_mutex);
441 	spin_lock_irq(&priv->lock);
442 	if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
443 		queue_delayed_work(ipoib_workqueue, &priv->mcast_task,
444 				   mcast->backoff * HZ);
445 	spin_unlock_irq(&priv->lock);
446 	mutex_unlock(&mcast_mutex);
447 
448 	return status;
449 }
450 
451 static void ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast,
452 			     int create)
453 {
454 	struct ipoib_dev_priv *priv = netdev_priv(dev);
455 	struct ib_sa_mcmember_rec rec = {
456 		.join_state = 1
457 	};
458 	ib_sa_comp_mask comp_mask;
459 	int ret = 0;
460 
461 	ipoib_dbg_mcast(priv, "joining MGID %pI6\n", mcast->mcmember.mgid.raw);
462 
463 	rec.mgid     = mcast->mcmember.mgid;
464 	rec.port_gid = priv->local_gid;
465 	rec.pkey     = cpu_to_be16(priv->pkey);
466 
467 	comp_mask =
468 		IB_SA_MCMEMBER_REC_MGID		|
469 		IB_SA_MCMEMBER_REC_PORT_GID	|
470 		IB_SA_MCMEMBER_REC_PKEY		|
471 		IB_SA_MCMEMBER_REC_JOIN_STATE;
472 
473 	if (create) {
474 		comp_mask |=
475 			IB_SA_MCMEMBER_REC_QKEY			|
476 			IB_SA_MCMEMBER_REC_MTU_SELECTOR		|
477 			IB_SA_MCMEMBER_REC_MTU			|
478 			IB_SA_MCMEMBER_REC_TRAFFIC_CLASS	|
479 			IB_SA_MCMEMBER_REC_RATE_SELECTOR	|
480 			IB_SA_MCMEMBER_REC_RATE			|
481 			IB_SA_MCMEMBER_REC_SL			|
482 			IB_SA_MCMEMBER_REC_FLOW_LABEL		|
483 			IB_SA_MCMEMBER_REC_HOP_LIMIT;
484 
485 		rec.qkey	  = priv->broadcast->mcmember.qkey;
486 		rec.mtu_selector  = IB_SA_EQ;
487 		rec.mtu		  = priv->broadcast->mcmember.mtu;
488 		rec.traffic_class = priv->broadcast->mcmember.traffic_class;
489 		rec.rate_selector = IB_SA_EQ;
490 		rec.rate	  = priv->broadcast->mcmember.rate;
491 		rec.sl		  = priv->broadcast->mcmember.sl;
492 		rec.flow_label	  = priv->broadcast->mcmember.flow_label;
493 		rec.hop_limit	  = priv->broadcast->mcmember.hop_limit;
494 	}
495 
496 	set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
497 	mcast->mc = ib_sa_join_multicast(&ipoib_sa_client, priv->ca, priv->port,
498 					 &rec, comp_mask, GFP_KERNEL,
499 					 ipoib_mcast_join_complete, mcast);
500 	if (IS_ERR(mcast->mc)) {
501 		clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
502 		ret = PTR_ERR(mcast->mc);
503 		ipoib_warn(priv, "ib_sa_join_multicast failed, status %d\n", ret);
504 
505 		mcast->backoff *= 2;
506 		if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
507 			mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
508 
509 		mutex_lock(&mcast_mutex);
510 		if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
511 			queue_delayed_work(ipoib_workqueue,
512 					   &priv->mcast_task,
513 					   mcast->backoff * HZ);
514 		mutex_unlock(&mcast_mutex);
515 	}
516 }
517 
518 void ipoib_mcast_join_task(struct work_struct *work)
519 {
520 	struct ipoib_dev_priv *priv =
521 		container_of(work, struct ipoib_dev_priv, mcast_task.work);
522 	struct net_device *dev = priv->dev;
523 
524 	if (!test_bit(IPOIB_MCAST_RUN, &priv->flags))
525 		return;
526 
527 	if (ib_query_gid(priv->ca, priv->port, 0, &priv->local_gid))
528 		ipoib_warn(priv, "ib_query_gid() failed\n");
529 	else
530 		memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw, sizeof (union ib_gid));
531 
532 	{
533 		struct ib_port_attr attr;
534 
535 		if (!ib_query_port(priv->ca, priv->port, &attr))
536 			priv->local_lid = attr.lid;
537 		else
538 			ipoib_warn(priv, "ib_query_port failed\n");
539 	}
540 
541 	if (!priv->broadcast) {
542 		struct ipoib_mcast *broadcast;
543 
544 		if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
545 			return;
546 
547 		broadcast = ipoib_mcast_alloc(dev, 1);
548 		if (!broadcast) {
549 			ipoib_warn(priv, "failed to allocate broadcast group\n");
550 			mutex_lock(&mcast_mutex);
551 			if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
552 				queue_delayed_work(ipoib_workqueue,
553 						   &priv->mcast_task, HZ);
554 			mutex_unlock(&mcast_mutex);
555 			return;
556 		}
557 
558 		spin_lock_irq(&priv->lock);
559 		memcpy(broadcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
560 		       sizeof (union ib_gid));
561 		priv->broadcast = broadcast;
562 
563 		__ipoib_mcast_add(dev, priv->broadcast);
564 		spin_unlock_irq(&priv->lock);
565 	}
566 
567 	if (!test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
568 		if (!test_bit(IPOIB_MCAST_FLAG_BUSY, &priv->broadcast->flags))
569 			ipoib_mcast_join(dev, priv->broadcast, 0);
570 		return;
571 	}
572 
573 	while (1) {
574 		struct ipoib_mcast *mcast = NULL;
575 
576 		spin_lock_irq(&priv->lock);
577 		list_for_each_entry(mcast, &priv->multicast_list, list) {
578 			if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)
579 			    && !test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)
580 			    && !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
581 				/* Found the next unjoined group */
582 				break;
583 			}
584 		}
585 		spin_unlock_irq(&priv->lock);
586 
587 		if (&mcast->list == &priv->multicast_list) {
588 			/* All done */
589 			break;
590 		}
591 
592 		ipoib_mcast_join(dev, mcast, 1);
593 		return;
594 	}
595 
596 	priv->mcast_mtu = IPOIB_UD_MTU(ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu));
597 
598 	if (!ipoib_cm_admin_enabled(dev)) {
599 		rtnl_lock();
600 		dev_set_mtu(dev, min(priv->mcast_mtu, priv->admin_mtu));
601 		rtnl_unlock();
602 	}
603 
604 	ipoib_dbg_mcast(priv, "successfully joined all multicast groups\n");
605 
606 	clear_bit(IPOIB_MCAST_RUN, &priv->flags);
607 }
608 
609 int ipoib_mcast_start_thread(struct net_device *dev)
610 {
611 	struct ipoib_dev_priv *priv = netdev_priv(dev);
612 
613 	ipoib_dbg_mcast(priv, "starting multicast thread\n");
614 
615 	mutex_lock(&mcast_mutex);
616 	if (!test_and_set_bit(IPOIB_MCAST_RUN, &priv->flags))
617 		queue_delayed_work(ipoib_workqueue, &priv->mcast_task, 0);
618 	mutex_unlock(&mcast_mutex);
619 
620 	return 0;
621 }
622 
623 int ipoib_mcast_stop_thread(struct net_device *dev, int flush)
624 {
625 	struct ipoib_dev_priv *priv = netdev_priv(dev);
626 
627 	ipoib_dbg_mcast(priv, "stopping multicast thread\n");
628 
629 	mutex_lock(&mcast_mutex);
630 	clear_bit(IPOIB_MCAST_RUN, &priv->flags);
631 	cancel_delayed_work(&priv->mcast_task);
632 	mutex_unlock(&mcast_mutex);
633 
634 	if (flush)
635 		flush_workqueue(ipoib_workqueue);
636 
637 	return 0;
638 }
639 
640 static int ipoib_mcast_leave(struct net_device *dev, struct ipoib_mcast *mcast)
641 {
642 	struct ipoib_dev_priv *priv = netdev_priv(dev);
643 	int ret = 0;
644 
645 	if (test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
646 		ib_sa_free_multicast(mcast->mc);
647 
648 	if (test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
649 		ipoib_dbg_mcast(priv, "leaving MGID %pI6\n",
650 				mcast->mcmember.mgid.raw);
651 
652 		/* Remove ourselves from the multicast group */
653 		ret = ib_detach_mcast(priv->qp, &mcast->mcmember.mgid,
654 				      be16_to_cpu(mcast->mcmember.mlid));
655 		if (ret)
656 			ipoib_warn(priv, "ib_detach_mcast failed (result = %d)\n", ret);
657 	}
658 
659 	return 0;
660 }
661 
662 void ipoib_mcast_send(struct net_device *dev, void *mgid, struct sk_buff *skb)
663 {
664 	struct ipoib_dev_priv *priv = netdev_priv(dev);
665 	struct ipoib_mcast *mcast;
666 	unsigned long flags;
667 
668 	spin_lock_irqsave(&priv->lock, flags);
669 
670 	if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)		||
671 	    !priv->broadcast					||
672 	    !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
673 		++dev->stats.tx_dropped;
674 		dev_kfree_skb_any(skb);
675 		goto unlock;
676 	}
677 
678 	mcast = __ipoib_mcast_find(dev, mgid);
679 	if (!mcast) {
680 		/* Let's create a new send only group now */
681 		ipoib_dbg_mcast(priv, "setting up send only multicast group for %pI6\n",
682 				mgid);
683 
684 		mcast = ipoib_mcast_alloc(dev, 0);
685 		if (!mcast) {
686 			ipoib_warn(priv, "unable to allocate memory for "
687 				   "multicast structure\n");
688 			++dev->stats.tx_dropped;
689 			dev_kfree_skb_any(skb);
690 			goto out;
691 		}
692 
693 		set_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags);
694 		memcpy(mcast->mcmember.mgid.raw, mgid, sizeof (union ib_gid));
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->stats.tx_dropped;
704 			dev_kfree_skb_any(skb);
705 		}
706 
707 		if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
708 			ipoib_dbg_mcast(priv, "no address vector, "
709 					"but multicast join already started\n");
710 		else if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags))
711 			ipoib_mcast_sendonly_join(mcast);
712 
713 		/*
714 		 * If lookup completes between here and out:, don't
715 		 * want to send packet twice.
716 		 */
717 		mcast = NULL;
718 	}
719 
720 out:
721 	if (mcast && mcast->ah) {
722 		struct dst_entry *dst = skb_dst(skb);
723 		struct neighbour *n = NULL;
724 		if (dst)
725 			n = dst_get_neighbour(dst);
726 		if (n && !*to_ipoib_neigh(n)) {
727 			struct ipoib_neigh *neigh = ipoib_neigh_alloc(n,
728 								      skb->dev);
729 
730 			if (neigh) {
731 				kref_get(&mcast->ah->ref);
732 				neigh->ah	= mcast->ah;
733 				list_add_tail(&neigh->list, &mcast->neigh_list);
734 			}
735 		}
736 
737 		spin_unlock_irqrestore(&priv->lock, flags);
738 		ipoib_send(dev, skb, mcast->ah, IB_MULTICAST_QPN);
739 		return;
740 	}
741 
742 unlock:
743 	spin_unlock_irqrestore(&priv->lock, flags);
744 }
745 
746 void ipoib_mcast_dev_flush(struct net_device *dev)
747 {
748 	struct ipoib_dev_priv *priv = netdev_priv(dev);
749 	LIST_HEAD(remove_list);
750 	struct ipoib_mcast *mcast, *tmcast;
751 	unsigned long flags;
752 
753 	ipoib_dbg_mcast(priv, "flushing multicast list\n");
754 
755 	spin_lock_irqsave(&priv->lock, flags);
756 
757 	list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
758 		list_del(&mcast->list);
759 		rb_erase(&mcast->rb_node, &priv->multicast_tree);
760 		list_add_tail(&mcast->list, &remove_list);
761 	}
762 
763 	if (priv->broadcast) {
764 		rb_erase(&priv->broadcast->rb_node, &priv->multicast_tree);
765 		list_add_tail(&priv->broadcast->list, &remove_list);
766 		priv->broadcast = NULL;
767 	}
768 
769 	spin_unlock_irqrestore(&priv->lock, flags);
770 
771 	list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
772 		ipoib_mcast_leave(dev, mcast);
773 		ipoib_mcast_free(mcast);
774 	}
775 }
776 
777 static int ipoib_mcast_addr_is_valid(const u8 *addr, const u8 *broadcast)
778 {
779 	/* reserved QPN, prefix, scope */
780 	if (memcmp(addr, broadcast, 6))
781 		return 0;
782 	/* signature lower, pkey */
783 	if (memcmp(addr + 7, broadcast + 7, 3))
784 		return 0;
785 	return 1;
786 }
787 
788 void ipoib_mcast_restart_task(struct work_struct *work)
789 {
790 	struct ipoib_dev_priv *priv =
791 		container_of(work, struct ipoib_dev_priv, restart_task);
792 	struct net_device *dev = priv->dev;
793 	struct netdev_hw_addr *ha;
794 	struct ipoib_mcast *mcast, *tmcast;
795 	LIST_HEAD(remove_list);
796 	unsigned long flags;
797 	struct ib_sa_mcmember_rec rec;
798 
799 	ipoib_dbg_mcast(priv, "restarting multicast task\n");
800 
801 	ipoib_mcast_stop_thread(dev, 0);
802 
803 	local_irq_save(flags);
804 	netif_addr_lock(dev);
805 	spin_lock(&priv->lock);
806 
807 	/*
808 	 * Unfortunately, the networking core only gives us a list of all of
809 	 * the multicast hardware addresses. We need to figure out which ones
810 	 * are new and which ones have been removed
811 	 */
812 
813 	/* Clear out the found flag */
814 	list_for_each_entry(mcast, &priv->multicast_list, list)
815 		clear_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
816 
817 	/* Mark all of the entries that are found or don't exist */
818 	netdev_for_each_mc_addr(ha, dev) {
819 		union ib_gid mgid;
820 
821 		if (!ipoib_mcast_addr_is_valid(ha->addr, dev->broadcast))
822 			continue;
823 
824 		memcpy(mgid.raw, ha->addr + 4, sizeof mgid);
825 
826 		mcast = __ipoib_mcast_find(dev, &mgid);
827 		if (!mcast || test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
828 			struct ipoib_mcast *nmcast;
829 
830 			/* ignore group which is directly joined by userspace */
831 			if (test_bit(IPOIB_FLAG_UMCAST, &priv->flags) &&
832 			    !ib_sa_get_mcmember_rec(priv->ca, priv->port, &mgid, &rec)) {
833 				ipoib_dbg_mcast(priv, "ignoring multicast entry for mgid %pI6\n",
834 						mgid.raw);
835 				continue;
836 			}
837 
838 			/* Not found or send-only group, let's add a new entry */
839 			ipoib_dbg_mcast(priv, "adding multicast entry for mgid %pI6\n",
840 					mgid.raw);
841 
842 			nmcast = ipoib_mcast_alloc(dev, 0);
843 			if (!nmcast) {
844 				ipoib_warn(priv, "unable to allocate memory for multicast structure\n");
845 				continue;
846 			}
847 
848 			set_bit(IPOIB_MCAST_FLAG_FOUND, &nmcast->flags);
849 
850 			nmcast->mcmember.mgid = mgid;
851 
852 			if (mcast) {
853 				/* Destroy the send only entry */
854 				list_move_tail(&mcast->list, &remove_list);
855 
856 				rb_replace_node(&mcast->rb_node,
857 						&nmcast->rb_node,
858 						&priv->multicast_tree);
859 			} else
860 				__ipoib_mcast_add(dev, nmcast);
861 
862 			list_add_tail(&nmcast->list, &priv->multicast_list);
863 		}
864 
865 		if (mcast)
866 			set_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
867 	}
868 
869 	/* Remove all of the entries don't exist anymore */
870 	list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
871 		if (!test_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags) &&
872 		    !test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
873 			ipoib_dbg_mcast(priv, "deleting multicast group %pI6\n",
874 					mcast->mcmember.mgid.raw);
875 
876 			rb_erase(&mcast->rb_node, &priv->multicast_tree);
877 
878 			/* Move to the remove list */
879 			list_move_tail(&mcast->list, &remove_list);
880 		}
881 	}
882 
883 	spin_unlock(&priv->lock);
884 	netif_addr_unlock(dev);
885 	local_irq_restore(flags);
886 
887 	/* We have to cancel outside of the spinlock */
888 	list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
889 		ipoib_mcast_leave(mcast->dev, mcast);
890 		ipoib_mcast_free(mcast);
891 	}
892 
893 	if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
894 		ipoib_mcast_start_thread(dev);
895 }
896 
897 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
898 
899 struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev)
900 {
901 	struct ipoib_mcast_iter *iter;
902 
903 	iter = kmalloc(sizeof *iter, GFP_KERNEL);
904 	if (!iter)
905 		return NULL;
906 
907 	iter->dev = dev;
908 	memset(iter->mgid.raw, 0, 16);
909 
910 	if (ipoib_mcast_iter_next(iter)) {
911 		kfree(iter);
912 		return NULL;
913 	}
914 
915 	return iter;
916 }
917 
918 int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter)
919 {
920 	struct ipoib_dev_priv *priv = netdev_priv(iter->dev);
921 	struct rb_node *n;
922 	struct ipoib_mcast *mcast;
923 	int ret = 1;
924 
925 	spin_lock_irq(&priv->lock);
926 
927 	n = rb_first(&priv->multicast_tree);
928 
929 	while (n) {
930 		mcast = rb_entry(n, struct ipoib_mcast, rb_node);
931 
932 		if (memcmp(iter->mgid.raw, mcast->mcmember.mgid.raw,
933 			   sizeof (union ib_gid)) < 0) {
934 			iter->mgid      = mcast->mcmember.mgid;
935 			iter->created   = mcast->created;
936 			iter->queuelen  = skb_queue_len(&mcast->pkt_queue);
937 			iter->complete  = !!mcast->ah;
938 			iter->send_only = !!(mcast->flags & (1 << IPOIB_MCAST_FLAG_SENDONLY));
939 
940 			ret = 0;
941 
942 			break;
943 		}
944 
945 		n = rb_next(n);
946 	}
947 
948 	spin_unlock_irq(&priv->lock);
949 
950 	return ret;
951 }
952 
953 void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
954 			   union ib_gid *mgid,
955 			   unsigned long *created,
956 			   unsigned int *queuelen,
957 			   unsigned int *complete,
958 			   unsigned int *send_only)
959 {
960 	*mgid      = iter->mgid;
961 	*created   = iter->created;
962 	*queuelen  = iter->queuelen;
963 	*complete  = iter->complete;
964 	*send_only = iter->send_only;
965 }
966 
967 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
968