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/moduleparam.h>
38 #include <linux/ip.h>
39 #include <linux/in.h>
40 #include <linux/igmp.h>
41 #include <linux/inetdevice.h>
42 #include <linux/delay.h>
43 #include <linux/completion.h>
44 #include <linux/slab.h>
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 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 /*
68  * This should be called with the priv->lock held
69  */
70 static void __ipoib_mcast_schedule_join_thread(struct ipoib_dev_priv *priv,
71 					       struct ipoib_mcast *mcast,
72 					       bool delay)
73 {
74 	if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
75 		return;
76 
77 	/*
78 	 * We will be scheduling *something*, so cancel whatever is
79 	 * currently scheduled first
80 	 */
81 	cancel_delayed_work(&priv->mcast_task);
82 	if (mcast && delay) {
83 		/*
84 		 * We had a failure and want to schedule a retry later
85 		 */
86 		mcast->backoff *= 2;
87 		if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
88 			mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
89 		mcast->delay_until = jiffies + (mcast->backoff * HZ);
90 		/*
91 		 * Mark this mcast for its delay, but restart the
92 		 * task immediately.  The join task will make sure to
93 		 * clear out all entries without delays, and then
94 		 * schedule itself to run again when the earliest
95 		 * delay expires
96 		 */
97 		queue_delayed_work(priv->wq, &priv->mcast_task, 0);
98 	} else if (delay) {
99 		/*
100 		 * Special case of retrying after a failure to
101 		 * allocate the broadcast multicast group, wait
102 		 * 1 second and try again
103 		 */
104 		queue_delayed_work(priv->wq, &priv->mcast_task, HZ);
105 	} else
106 		queue_delayed_work(priv->wq, &priv->mcast_task, 0);
107 }
108 
109 static void ipoib_mcast_free(struct ipoib_mcast *mcast)
110 {
111 	struct net_device *dev = mcast->dev;
112 	int tx_dropped = 0;
113 
114 	ipoib_dbg_mcast(netdev_priv(dev), "deleting multicast group %pI6\n",
115 			mcast->mcmember.mgid.raw);
116 
117 	/* remove all neigh connected to this mcast */
118 	ipoib_del_neighs_by_gid(dev, mcast->mcmember.mgid.raw);
119 
120 	if (mcast->ah)
121 		ipoib_put_ah(mcast->ah);
122 
123 	while (!skb_queue_empty(&mcast->pkt_queue)) {
124 		++tx_dropped;
125 		dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
126 	}
127 
128 	netif_tx_lock_bh(dev);
129 	dev->stats.tx_dropped += tx_dropped;
130 	netif_tx_unlock_bh(dev);
131 
132 	kfree(mcast);
133 }
134 
135 static struct ipoib_mcast *ipoib_mcast_alloc(struct net_device *dev,
136 					     int can_sleep)
137 {
138 	struct ipoib_mcast *mcast;
139 
140 	mcast = kzalloc(sizeof *mcast, can_sleep ? GFP_KERNEL : GFP_ATOMIC);
141 	if (!mcast)
142 		return NULL;
143 
144 	mcast->dev = dev;
145 	mcast->created = jiffies;
146 	mcast->delay_until = jiffies;
147 	mcast->backoff = 1;
148 
149 	INIT_LIST_HEAD(&mcast->list);
150 	INIT_LIST_HEAD(&mcast->neigh_list);
151 	skb_queue_head_init(&mcast->pkt_queue);
152 
153 	return mcast;
154 }
155 
156 static struct ipoib_mcast *__ipoib_mcast_find(struct net_device *dev, void *mgid)
157 {
158 	struct ipoib_dev_priv *priv = netdev_priv(dev);
159 	struct rb_node *n = priv->multicast_tree.rb_node;
160 
161 	while (n) {
162 		struct ipoib_mcast *mcast;
163 		int ret;
164 
165 		mcast = rb_entry(n, struct ipoib_mcast, rb_node);
166 
167 		ret = memcmp(mgid, mcast->mcmember.mgid.raw,
168 			     sizeof (union ib_gid));
169 		if (ret < 0)
170 			n = n->rb_left;
171 		else if (ret > 0)
172 			n = n->rb_right;
173 		else
174 			return mcast;
175 	}
176 
177 	return NULL;
178 }
179 
180 static int __ipoib_mcast_add(struct net_device *dev, struct ipoib_mcast *mcast)
181 {
182 	struct ipoib_dev_priv *priv = netdev_priv(dev);
183 	struct rb_node **n = &priv->multicast_tree.rb_node, *pn = NULL;
184 
185 	while (*n) {
186 		struct ipoib_mcast *tmcast;
187 		int ret;
188 
189 		pn = *n;
190 		tmcast = rb_entry(pn, struct ipoib_mcast, rb_node);
191 
192 		ret = memcmp(mcast->mcmember.mgid.raw, tmcast->mcmember.mgid.raw,
193 			     sizeof (union ib_gid));
194 		if (ret < 0)
195 			n = &pn->rb_left;
196 		else if (ret > 0)
197 			n = &pn->rb_right;
198 		else
199 			return -EEXIST;
200 	}
201 
202 	rb_link_node(&mcast->rb_node, pn, n);
203 	rb_insert_color(&mcast->rb_node, &priv->multicast_tree);
204 
205 	return 0;
206 }
207 
208 static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
209 				   struct ib_sa_mcmember_rec *mcmember)
210 {
211 	struct net_device *dev = mcast->dev;
212 	struct ipoib_dev_priv *priv = netdev_priv(dev);
213 	struct ipoib_ah *ah;
214 	int ret;
215 	int set_qkey = 0;
216 
217 	mcast->mcmember = *mcmember;
218 
219 	/* Set the multicast MTU and cached Q_Key before we attach if it's
220 	 * the broadcast group.
221 	 */
222 	if (!memcmp(mcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
223 		    sizeof (union ib_gid))) {
224 		spin_lock_irq(&priv->lock);
225 		if (!priv->broadcast) {
226 			spin_unlock_irq(&priv->lock);
227 			return -EAGAIN;
228 		}
229 		/*update priv member according to the new mcast*/
230 		priv->broadcast->mcmember.qkey = mcmember->qkey;
231 		priv->broadcast->mcmember.mtu = mcmember->mtu;
232 		priv->broadcast->mcmember.traffic_class = mcmember->traffic_class;
233 		priv->broadcast->mcmember.rate = mcmember->rate;
234 		priv->broadcast->mcmember.sl = mcmember->sl;
235 		priv->broadcast->mcmember.flow_label = mcmember->flow_label;
236 		priv->broadcast->mcmember.hop_limit = mcmember->hop_limit;
237 		/* assume if the admin and the mcast are the same both can be changed */
238 		if (priv->mcast_mtu == priv->admin_mtu)
239 			priv->admin_mtu =
240 			priv->mcast_mtu =
241 			IPOIB_UD_MTU(ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu));
242 		else
243 			priv->mcast_mtu =
244 			IPOIB_UD_MTU(ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu));
245 
246 		priv->qkey = be32_to_cpu(priv->broadcast->mcmember.qkey);
247 		spin_unlock_irq(&priv->lock);
248 		priv->tx_wr.wr.ud.remote_qkey = priv->qkey;
249 		set_qkey = 1;
250 	}
251 
252 	if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
253 		if (test_and_set_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
254 			ipoib_warn(priv, "multicast group %pI6 already attached\n",
255 				   mcast->mcmember.mgid.raw);
256 
257 			return 0;
258 		}
259 
260 		ret = ipoib_mcast_attach(dev, be16_to_cpu(mcast->mcmember.mlid),
261 					 &mcast->mcmember.mgid, set_qkey);
262 		if (ret < 0) {
263 			ipoib_warn(priv, "couldn't attach QP to multicast group %pI6\n",
264 				   mcast->mcmember.mgid.raw);
265 
266 			clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags);
267 			return ret;
268 		}
269 	}
270 
271 	{
272 		struct ib_ah_attr av = {
273 			.dlid	       = be16_to_cpu(mcast->mcmember.mlid),
274 			.port_num      = priv->port,
275 			.sl	       = mcast->mcmember.sl,
276 			.ah_flags      = IB_AH_GRH,
277 			.static_rate   = mcast->mcmember.rate,
278 			.grh	       = {
279 				.flow_label    = be32_to_cpu(mcast->mcmember.flow_label),
280 				.hop_limit     = mcast->mcmember.hop_limit,
281 				.sgid_index    = 0,
282 				.traffic_class = mcast->mcmember.traffic_class
283 			}
284 		};
285 		av.grh.dgid = mcast->mcmember.mgid;
286 
287 		ah = ipoib_create_ah(dev, priv->pd, &av);
288 		if (IS_ERR(ah)) {
289 			ipoib_warn(priv, "ib_address_create failed %ld\n",
290 				-PTR_ERR(ah));
291 			/* use original error */
292 			return PTR_ERR(ah);
293 		} else {
294 			spin_lock_irq(&priv->lock);
295 			mcast->ah = ah;
296 			spin_unlock_irq(&priv->lock);
297 
298 			ipoib_dbg_mcast(priv, "MGID %pI6 AV %p, LID 0x%04x, SL %d\n",
299 					mcast->mcmember.mgid.raw,
300 					mcast->ah->ah,
301 					be16_to_cpu(mcast->mcmember.mlid),
302 					mcast->mcmember.sl);
303 		}
304 	}
305 
306 	/* actually send any queued packets */
307 	netif_tx_lock_bh(dev);
308 	while (!skb_queue_empty(&mcast->pkt_queue)) {
309 		struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue);
310 
311 		netif_tx_unlock_bh(dev);
312 
313 		skb->dev = dev;
314 		if (dev_queue_xmit(skb))
315 			ipoib_warn(priv, "dev_queue_xmit failed to requeue packet\n");
316 
317 		netif_tx_lock_bh(dev);
318 	}
319 	netif_tx_unlock_bh(dev);
320 
321 	return 0;
322 }
323 
324 void ipoib_mcast_carrier_on_task(struct work_struct *work)
325 {
326 	struct ipoib_dev_priv *priv = container_of(work, struct ipoib_dev_priv,
327 						   carrier_on_task);
328 	struct ib_port_attr attr;
329 
330 	if (ib_query_port(priv->ca, priv->port, &attr) ||
331 	    attr.state != IB_PORT_ACTIVE) {
332 		ipoib_dbg(priv, "Keeping carrier off until IB port is active\n");
333 		return;
334 	}
335 
336 	/*
337 	 * Take rtnl_lock to avoid racing with ipoib_stop() and
338 	 * turning the carrier back on while a device is being
339 	 * removed.  However, ipoib_stop() will attempt to flush
340 	 * the workqueue while holding the rtnl lock, so loop
341 	 * on trylock until either we get the lock or we see
342 	 * FLAG_OPER_UP go away as that signals that we are bailing
343 	 * and can safely ignore the carrier on work.
344 	 */
345 	while (!rtnl_trylock()) {
346 		if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
347 			return;
348 		else
349 			msleep(20);
350 	}
351 	if (!ipoib_cm_admin_enabled(priv->dev))
352 		dev_set_mtu(priv->dev, min(priv->mcast_mtu, priv->admin_mtu));
353 	netif_carrier_on(priv->dev);
354 	rtnl_unlock();
355 }
356 
357 static int ipoib_mcast_join_complete(int status,
358 				     struct ib_sa_multicast *multicast)
359 {
360 	struct ipoib_mcast *mcast = multicast->context;
361 	struct net_device *dev = mcast->dev;
362 	struct ipoib_dev_priv *priv = netdev_priv(dev);
363 
364 	ipoib_dbg_mcast(priv, "%sjoin completion for %pI6 (status %d)\n",
365 			test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) ?
366 			"sendonly " : "",
367 			mcast->mcmember.mgid.raw, status);
368 
369 	/* We trap for port events ourselves. */
370 	if (status == -ENETRESET) {
371 		status = 0;
372 		goto out;
373 	}
374 
375 	if (!status)
376 		status = ipoib_mcast_join_finish(mcast, &multicast->rec);
377 
378 	if (!status) {
379 		mcast->backoff = 1;
380 		mcast->delay_until = jiffies;
381 
382 		/*
383 		 * Defer carrier on work to priv->wq to avoid a
384 		 * deadlock on rtnl_lock here.  Requeue our multicast
385 		 * work too, which will end up happening right after
386 		 * our carrier on task work and will allow us to
387 		 * send out all of the non-broadcast joins
388 		 */
389 		if (mcast == priv->broadcast) {
390 			spin_lock_irq(&priv->lock);
391 			queue_work(priv->wq, &priv->carrier_on_task);
392 			__ipoib_mcast_schedule_join_thread(priv, NULL, 0);
393 			goto out_locked;
394 		}
395 	} else {
396 		if (mcast->logcount++ < 20) {
397 			if (status == -ETIMEDOUT || status == -EAGAIN) {
398 				ipoib_dbg_mcast(priv, "%smulticast join failed for %pI6, status %d\n",
399 						test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) ? "sendonly " : "",
400 						mcast->mcmember.mgid.raw, status);
401 			} else {
402 				ipoib_warn(priv, "%smulticast join failed for %pI6, status %d\n",
403 						test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) ? "sendonly " : "",
404 					   mcast->mcmember.mgid.raw, status);
405 			}
406 		}
407 
408 		if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) &&
409 		    mcast->backoff >= 2) {
410 			/*
411 			 * We only retry sendonly joins once before we drop
412 			 * the packet and quit trying to deal with the
413 			 * group.  However, we leave the group in the
414 			 * mcast list as an unjoined group.  If we want to
415 			 * try joining again, we simply queue up a packet
416 			 * and restart the join thread.  The empty queue
417 			 * is why the join thread ignores this group.
418 			 */
419 			mcast->backoff = 1;
420 			netif_tx_lock_bh(dev);
421 			while (!skb_queue_empty(&mcast->pkt_queue)) {
422 				++dev->stats.tx_dropped;
423 				dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue));
424 			}
425 			netif_tx_unlock_bh(dev);
426 		} else {
427 			spin_lock_irq(&priv->lock);
428 			/* Requeue this join task with a backoff delay */
429 			__ipoib_mcast_schedule_join_thread(priv, mcast, 1);
430 			goto out_locked;
431 		}
432 	}
433 out:
434 	spin_lock_irq(&priv->lock);
435 out_locked:
436 	/*
437 	 * Make sure to set mcast->mc before we clear the busy flag to avoid
438 	 * racing with code that checks for BUSY before checking mcast->mc
439 	 */
440 	if (status)
441 		mcast->mc = NULL;
442 	else
443 		mcast->mc = multicast;
444 	clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
445 	spin_unlock_irq(&priv->lock);
446 	complete(&mcast->done);
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_multicast *multicast;
456 	struct ib_sa_mcmember_rec rec = {
457 		.join_state = 1
458 	};
459 	ib_sa_comp_mask comp_mask;
460 	int ret = 0;
461 
462 	ipoib_dbg_mcast(priv, "joining MGID %pI6\n", mcast->mcmember.mgid.raw);
463 
464 	rec.mgid     = mcast->mcmember.mgid;
465 	rec.port_gid = priv->local_gid;
466 	rec.pkey     = cpu_to_be16(priv->pkey);
467 
468 	comp_mask =
469 		IB_SA_MCMEMBER_REC_MGID		|
470 		IB_SA_MCMEMBER_REC_PORT_GID	|
471 		IB_SA_MCMEMBER_REC_PKEY		|
472 		IB_SA_MCMEMBER_REC_JOIN_STATE;
473 
474 	if (create) {
475 		comp_mask |=
476 			IB_SA_MCMEMBER_REC_QKEY			|
477 			IB_SA_MCMEMBER_REC_MTU_SELECTOR		|
478 			IB_SA_MCMEMBER_REC_MTU			|
479 			IB_SA_MCMEMBER_REC_TRAFFIC_CLASS	|
480 			IB_SA_MCMEMBER_REC_RATE_SELECTOR	|
481 			IB_SA_MCMEMBER_REC_RATE			|
482 			IB_SA_MCMEMBER_REC_SL			|
483 			IB_SA_MCMEMBER_REC_FLOW_LABEL		|
484 			IB_SA_MCMEMBER_REC_HOP_LIMIT;
485 
486 		rec.qkey	  = priv->broadcast->mcmember.qkey;
487 		rec.mtu_selector  = IB_SA_EQ;
488 		rec.mtu		  = priv->broadcast->mcmember.mtu;
489 		rec.traffic_class = priv->broadcast->mcmember.traffic_class;
490 		rec.rate_selector = IB_SA_EQ;
491 		rec.rate	  = priv->broadcast->mcmember.rate;
492 		rec.sl		  = priv->broadcast->mcmember.sl;
493 		rec.flow_label	  = priv->broadcast->mcmember.flow_label;
494 		rec.hop_limit	  = priv->broadcast->mcmember.hop_limit;
495 	}
496 
497 	multicast = 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(multicast)) {
501 		ret = PTR_ERR(multicast);
502 		ipoib_warn(priv, "ib_sa_join_multicast failed, status %d\n", ret);
503 		spin_lock_irq(&priv->lock);
504 		/* Requeue this join task with a backoff delay */
505 		__ipoib_mcast_schedule_join_thread(priv, mcast, 1);
506 		clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
507 		spin_unlock_irq(&priv->lock);
508 		complete(&mcast->done);
509 	}
510 }
511 
512 void ipoib_mcast_join_task(struct work_struct *work)
513 {
514 	struct ipoib_dev_priv *priv =
515 		container_of(work, struct ipoib_dev_priv, mcast_task.work);
516 	struct net_device *dev = priv->dev;
517 	struct ib_port_attr port_attr;
518 	unsigned long delay_until = 0;
519 	struct ipoib_mcast *mcast = NULL;
520 	int create = 1;
521 
522 	if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
523 		return;
524 
525 	if (ib_query_port(priv->ca, priv->port, &port_attr) ||
526 	    port_attr.state != IB_PORT_ACTIVE) {
527 		ipoib_dbg(priv, "port state is not ACTIVE (state = %d) suspending join task\n",
528 			  port_attr.state);
529 		return;
530 	}
531 	priv->local_lid = port_attr.lid;
532 
533 	if (ib_query_gid(priv->ca, priv->port, 0, &priv->local_gid))
534 		ipoib_warn(priv, "ib_query_gid() failed\n");
535 	else
536 		memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw, sizeof (union ib_gid));
537 
538 	spin_lock_irq(&priv->lock);
539 	if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
540 		goto out;
541 
542 	if (!priv->broadcast) {
543 		struct ipoib_mcast *broadcast;
544 
545 		broadcast = ipoib_mcast_alloc(dev, 0);
546 		if (!broadcast) {
547 			ipoib_warn(priv, "failed to allocate broadcast group\n");
548 			/*
549 			 * Restart us after a 1 second delay to retry
550 			 * creating our broadcast group and attaching to
551 			 * it.  Until this succeeds, this ipoib dev is
552 			 * completely stalled (multicast wise).
553 			 */
554 			__ipoib_mcast_schedule_join_thread(priv, NULL, 1);
555 			goto out;
556 		}
557 
558 		memcpy(broadcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
559 		       sizeof (union ib_gid));
560 		priv->broadcast = broadcast;
561 
562 		__ipoib_mcast_add(dev, priv->broadcast);
563 	}
564 
565 	if (!test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
566 		if (IS_ERR_OR_NULL(priv->broadcast->mc) &&
567 		    !test_bit(IPOIB_MCAST_FLAG_BUSY, &priv->broadcast->flags)) {
568 			mcast = priv->broadcast;
569 			create = 0;
570 			if (mcast->backoff > 1 &&
571 			    time_before(jiffies, mcast->delay_until)) {
572 				delay_until = mcast->delay_until;
573 				mcast = NULL;
574 			}
575 		}
576 		goto out;
577 	}
578 
579 	/*
580 	 * We'll never get here until the broadcast group is both allocated
581 	 * and attached
582 	 */
583 	list_for_each_entry(mcast, &priv->multicast_list, list) {
584 		if (IS_ERR_OR_NULL(mcast->mc) &&
585 		    !test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags) &&
586 		    (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) ||
587 		     !skb_queue_empty(&mcast->pkt_queue))) {
588 			if (mcast->backoff == 1 ||
589 			    time_after_eq(jiffies, mcast->delay_until)) {
590 				/* Found the next unjoined group */
591 				init_completion(&mcast->done);
592 				set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
593 				if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags))
594 					create = 0;
595 				else
596 					create = 1;
597 				spin_unlock_irq(&priv->lock);
598 				ipoib_mcast_join(dev, mcast, create);
599 				spin_lock_irq(&priv->lock);
600 			} else if (!delay_until ||
601 				 time_before(mcast->delay_until, delay_until))
602 				delay_until = mcast->delay_until;
603 		}
604 	}
605 
606 	mcast = NULL;
607 	ipoib_dbg_mcast(priv, "successfully started all multicast joins\n");
608 
609 out:
610 	if (delay_until) {
611 		cancel_delayed_work(&priv->mcast_task);
612 		queue_delayed_work(priv->wq, &priv->mcast_task,
613 				   delay_until - jiffies);
614 	}
615 	if (mcast) {
616 		init_completion(&mcast->done);
617 		set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
618 	}
619 	spin_unlock_irq(&priv->lock);
620 	if (mcast)
621 		ipoib_mcast_join(dev, mcast, create);
622 }
623 
624 int ipoib_mcast_start_thread(struct net_device *dev)
625 {
626 	struct ipoib_dev_priv *priv = netdev_priv(dev);
627 	unsigned long flags;
628 
629 	ipoib_dbg_mcast(priv, "starting multicast thread\n");
630 
631 	spin_lock_irqsave(&priv->lock, flags);
632 	__ipoib_mcast_schedule_join_thread(priv, NULL, 0);
633 	spin_unlock_irqrestore(&priv->lock, flags);
634 
635 	return 0;
636 }
637 
638 int ipoib_mcast_stop_thread(struct net_device *dev)
639 {
640 	struct ipoib_dev_priv *priv = netdev_priv(dev);
641 	unsigned long flags;
642 
643 	ipoib_dbg_mcast(priv, "stopping multicast thread\n");
644 
645 	spin_lock_irqsave(&priv->lock, flags);
646 	cancel_delayed_work(&priv->mcast_task);
647 	spin_unlock_irqrestore(&priv->lock, flags);
648 
649 	flush_workqueue(priv->wq);
650 
651 	return 0;
652 }
653 
654 static int ipoib_mcast_leave(struct net_device *dev, struct ipoib_mcast *mcast)
655 {
656 	struct ipoib_dev_priv *priv = netdev_priv(dev);
657 	int ret = 0;
658 
659 	if (test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
660 		ipoib_warn(priv, "ipoib_mcast_leave on an in-flight join\n");
661 
662 	if (!IS_ERR_OR_NULL(mcast->mc))
663 		ib_sa_free_multicast(mcast->mc);
664 
665 	if (test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
666 		ipoib_dbg_mcast(priv, "leaving MGID %pI6\n",
667 				mcast->mcmember.mgid.raw);
668 
669 		/* Remove ourselves from the multicast group */
670 		ret = ib_detach_mcast(priv->qp, &mcast->mcmember.mgid,
671 				      be16_to_cpu(mcast->mcmember.mlid));
672 		if (ret)
673 			ipoib_warn(priv, "ib_detach_mcast failed (result = %d)\n", ret);
674 	} else if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags))
675 		ipoib_dbg(priv, "leaving with no mcmember but not a "
676 			  "SENDONLY join\n");
677 
678 	return 0;
679 }
680 
681 void ipoib_mcast_send(struct net_device *dev, u8 *daddr, struct sk_buff *skb)
682 {
683 	struct ipoib_dev_priv *priv = netdev_priv(dev);
684 	struct ipoib_mcast *mcast;
685 	unsigned long flags;
686 	void *mgid = daddr + 4;
687 
688 	spin_lock_irqsave(&priv->lock, flags);
689 
690 	if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)		||
691 	    !priv->broadcast					||
692 	    !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
693 		++dev->stats.tx_dropped;
694 		dev_kfree_skb_any(skb);
695 		goto unlock;
696 	}
697 
698 	mcast = __ipoib_mcast_find(dev, mgid);
699 	if (!mcast || !mcast->ah) {
700 		if (!mcast) {
701 			/* Let's create a new send only group now */
702 			ipoib_dbg_mcast(priv, "setting up send only multicast group for %pI6\n",
703 					mgid);
704 
705 			mcast = ipoib_mcast_alloc(dev, 0);
706 			if (!mcast) {
707 				ipoib_warn(priv, "unable to allocate memory "
708 					   "for multicast structure\n");
709 				++dev->stats.tx_dropped;
710 				dev_kfree_skb_any(skb);
711 				goto unlock;
712 			}
713 
714 			set_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags);
715 			memcpy(mcast->mcmember.mgid.raw, mgid,
716 			       sizeof (union ib_gid));
717 			__ipoib_mcast_add(dev, mcast);
718 			list_add_tail(&mcast->list, &priv->multicast_list);
719 		}
720 		if (skb_queue_len(&mcast->pkt_queue) < IPOIB_MAX_MCAST_QUEUE)
721 			skb_queue_tail(&mcast->pkt_queue, skb);
722 		else {
723 			++dev->stats.tx_dropped;
724 			dev_kfree_skb_any(skb);
725 		}
726 		if (!test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) {
727 			__ipoib_mcast_schedule_join_thread(priv, NULL, 0);
728 		}
729 	} else {
730 		struct ipoib_neigh *neigh;
731 
732 		spin_unlock_irqrestore(&priv->lock, flags);
733 		neigh = ipoib_neigh_get(dev, daddr);
734 		spin_lock_irqsave(&priv->lock, flags);
735 		if (!neigh) {
736 			neigh = ipoib_neigh_alloc(daddr, dev);
737 			if (neigh) {
738 				kref_get(&mcast->ah->ref);
739 				neigh->ah	= mcast->ah;
740 				list_add_tail(&neigh->list, &mcast->neigh_list);
741 			}
742 		}
743 		spin_unlock_irqrestore(&priv->lock, flags);
744 		ipoib_send(dev, skb, mcast->ah, IB_MULTICAST_QPN);
745 		if (neigh)
746 			ipoib_neigh_put(neigh);
747 		return;
748 	}
749 
750 unlock:
751 	spin_unlock_irqrestore(&priv->lock, flags);
752 }
753 
754 void ipoib_mcast_dev_flush(struct net_device *dev)
755 {
756 	struct ipoib_dev_priv *priv = netdev_priv(dev);
757 	LIST_HEAD(remove_list);
758 	struct ipoib_mcast *mcast, *tmcast;
759 	unsigned long flags;
760 
761 	ipoib_dbg_mcast(priv, "flushing multicast list\n");
762 
763 	spin_lock_irqsave(&priv->lock, flags);
764 
765 	list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
766 		list_del(&mcast->list);
767 		rb_erase(&mcast->rb_node, &priv->multicast_tree);
768 		list_add_tail(&mcast->list, &remove_list);
769 	}
770 
771 	if (priv->broadcast) {
772 		rb_erase(&priv->broadcast->rb_node, &priv->multicast_tree);
773 		list_add_tail(&priv->broadcast->list, &remove_list);
774 		priv->broadcast = NULL;
775 	}
776 
777 	spin_unlock_irqrestore(&priv->lock, flags);
778 
779 	/*
780 	 * make sure the in-flight joins have finished before we attempt
781 	 * to leave
782 	 */
783 	list_for_each_entry_safe(mcast, tmcast, &remove_list, list)
784 		if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
785 			wait_for_completion(&mcast->done);
786 
787 	list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
788 		ipoib_mcast_leave(dev, mcast);
789 		ipoib_mcast_free(mcast);
790 	}
791 }
792 
793 static int ipoib_mcast_addr_is_valid(const u8 *addr, const u8 *broadcast)
794 {
795 	/* reserved QPN, prefix, scope */
796 	if (memcmp(addr, broadcast, 6))
797 		return 0;
798 	/* signature lower, pkey */
799 	if (memcmp(addr + 7, broadcast + 7, 3))
800 		return 0;
801 	return 1;
802 }
803 
804 void ipoib_mcast_restart_task(struct work_struct *work)
805 {
806 	struct ipoib_dev_priv *priv =
807 		container_of(work, struct ipoib_dev_priv, restart_task);
808 	struct net_device *dev = priv->dev;
809 	struct netdev_hw_addr *ha;
810 	struct ipoib_mcast *mcast, *tmcast;
811 	LIST_HEAD(remove_list);
812 	unsigned long flags;
813 	struct ib_sa_mcmember_rec rec;
814 
815 	if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags))
816 		/*
817 		 * shortcut...on shutdown flush is called next, just
818 		 * let it do all the work
819 		 */
820 		return;
821 
822 	ipoib_dbg_mcast(priv, "restarting multicast task\n");
823 
824 	local_irq_save(flags);
825 	netif_addr_lock(dev);
826 	spin_lock(&priv->lock);
827 
828 	/*
829 	 * Unfortunately, the networking core only gives us a list of all of
830 	 * the multicast hardware addresses. We need to figure out which ones
831 	 * are new and which ones have been removed
832 	 */
833 
834 	/* Clear out the found flag */
835 	list_for_each_entry(mcast, &priv->multicast_list, list)
836 		clear_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
837 
838 	/* Mark all of the entries that are found or don't exist */
839 	netdev_for_each_mc_addr(ha, dev) {
840 		union ib_gid mgid;
841 
842 		if (!ipoib_mcast_addr_is_valid(ha->addr, dev->broadcast))
843 			continue;
844 
845 		memcpy(mgid.raw, ha->addr + 4, sizeof mgid);
846 
847 		mcast = __ipoib_mcast_find(dev, &mgid);
848 		if (!mcast || test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
849 			struct ipoib_mcast *nmcast;
850 
851 			/* ignore group which is directly joined by userspace */
852 			if (test_bit(IPOIB_FLAG_UMCAST, &priv->flags) &&
853 			    !ib_sa_get_mcmember_rec(priv->ca, priv->port, &mgid, &rec)) {
854 				ipoib_dbg_mcast(priv, "ignoring multicast entry for mgid %pI6\n",
855 						mgid.raw);
856 				continue;
857 			}
858 
859 			/* Not found or send-only group, let's add a new entry */
860 			ipoib_dbg_mcast(priv, "adding multicast entry for mgid %pI6\n",
861 					mgid.raw);
862 
863 			nmcast = ipoib_mcast_alloc(dev, 0);
864 			if (!nmcast) {
865 				ipoib_warn(priv, "unable to allocate memory for multicast structure\n");
866 				continue;
867 			}
868 
869 			set_bit(IPOIB_MCAST_FLAG_FOUND, &nmcast->flags);
870 
871 			nmcast->mcmember.mgid = mgid;
872 
873 			if (mcast) {
874 				/* Destroy the send only entry */
875 				list_move_tail(&mcast->list, &remove_list);
876 
877 				rb_replace_node(&mcast->rb_node,
878 						&nmcast->rb_node,
879 						&priv->multicast_tree);
880 			} else
881 				__ipoib_mcast_add(dev, nmcast);
882 
883 			list_add_tail(&nmcast->list, &priv->multicast_list);
884 		}
885 
886 		if (mcast)
887 			set_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
888 	}
889 
890 	/* Remove all of the entries don't exist anymore */
891 	list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
892 		if (!test_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags) &&
893 		    !test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
894 			ipoib_dbg_mcast(priv, "deleting multicast group %pI6\n",
895 					mcast->mcmember.mgid.raw);
896 
897 			rb_erase(&mcast->rb_node, &priv->multicast_tree);
898 
899 			/* Move to the remove list */
900 			list_move_tail(&mcast->list, &remove_list);
901 		}
902 	}
903 
904 	spin_unlock(&priv->lock);
905 	netif_addr_unlock(dev);
906 	local_irq_restore(flags);
907 
908 	/*
909 	 * make sure the in-flight joins have finished before we attempt
910 	 * to leave
911 	 */
912 	list_for_each_entry_safe(mcast, tmcast, &remove_list, list)
913 		if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
914 			wait_for_completion(&mcast->done);
915 
916 	list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
917 		ipoib_mcast_leave(mcast->dev, mcast);
918 		ipoib_mcast_free(mcast);
919 	}
920 
921 	/*
922 	 * Double check that we are still up
923 	 */
924 	if (test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
925 		spin_lock_irqsave(&priv->lock, flags);
926 		__ipoib_mcast_schedule_join_thread(priv, NULL, 0);
927 		spin_unlock_irqrestore(&priv->lock, flags);
928 	}
929 }
930 
931 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
932 
933 struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev)
934 {
935 	struct ipoib_mcast_iter *iter;
936 
937 	iter = kmalloc(sizeof *iter, GFP_KERNEL);
938 	if (!iter)
939 		return NULL;
940 
941 	iter->dev = dev;
942 	memset(iter->mgid.raw, 0, 16);
943 
944 	if (ipoib_mcast_iter_next(iter)) {
945 		kfree(iter);
946 		return NULL;
947 	}
948 
949 	return iter;
950 }
951 
952 int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter)
953 {
954 	struct ipoib_dev_priv *priv = netdev_priv(iter->dev);
955 	struct rb_node *n;
956 	struct ipoib_mcast *mcast;
957 	int ret = 1;
958 
959 	spin_lock_irq(&priv->lock);
960 
961 	n = rb_first(&priv->multicast_tree);
962 
963 	while (n) {
964 		mcast = rb_entry(n, struct ipoib_mcast, rb_node);
965 
966 		if (memcmp(iter->mgid.raw, mcast->mcmember.mgid.raw,
967 			   sizeof (union ib_gid)) < 0) {
968 			iter->mgid      = mcast->mcmember.mgid;
969 			iter->created   = mcast->created;
970 			iter->queuelen  = skb_queue_len(&mcast->pkt_queue);
971 			iter->complete  = !!mcast->ah;
972 			iter->send_only = !!(mcast->flags & (1 << IPOIB_MCAST_FLAG_SENDONLY));
973 
974 			ret = 0;
975 
976 			break;
977 		}
978 
979 		n = rb_next(n);
980 	}
981 
982 	spin_unlock_irq(&priv->lock);
983 
984 	return ret;
985 }
986 
987 void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
988 			   union ib_gid *mgid,
989 			   unsigned long *created,
990 			   unsigned int *queuelen,
991 			   unsigned int *complete,
992 			   unsigned int *send_only)
993 {
994 	*mgid      = iter->mgid;
995 	*created   = iter->created;
996 	*queuelen  = iter->queuelen;
997 	*complete  = iter->complete;
998 	*send_only = iter->send_only;
999 }
1000 
1001 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
1002