/drivers/infiniband/ulp/ipoib/ipoib_multicast.c

https://bitbucket.org/evzijst/gittest · C · 991 lines · 726 code · 195 blank · 70 comment · 99 complexity · 112c47de3ed3d3668b9be27279386f59 MD5 · raw file 

    

  1. /*
    2. 

  2.  * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
    3. 

  3.  *
    4. 

  4.  * This software is available to you under a choice of one of two
    5. 

  5.  * licenses.  You may choose to be licensed under the terms of the GNU
    6. 

  6.  * General Public License (GPL) Version 2, available from the file
    7. 

  7.  * COPYING in the main directory of this source tree, or the
    8. 

  8.  * OpenIB.org BSD license below:
    9. 

  9.  *
    10. 

  10.  *     Redistribution and use in source and binary forms, with or
    11. 

  11.  *     without modification, are permitted provided that the following
    12. 

  12.  *     conditions are met:
    13. 

  13.  *
    14. 

  14.  *      - Redistributions of source code must retain the above
    15. 

  15.  *        copyright notice, this list of conditions and the following
    16. 

  16.  *        disclaimer.
    17. 

  17.  *
    18. 

  18.  *      - Redistributions in binary form must reproduce the above
    19. 

  19.  *        copyright notice, this list of conditions and the following
    20. 

  20.  *        disclaimer in the documentation and/or other materials
    21. 

  21.  *        provided with the distribution.
    22. 

  22.  *
    23. 

  23.  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
    24. 

  24.  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
    25. 

  25.  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
    26. 

  26.  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
    27. 

  27.  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
    28. 

  28.  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
    29. 

  29.  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
    30. 

  30.  * SOFTWARE.
    31. 

  31.  *
    32. 

  32.  * $Id: ipoib_multicast.c 1362 2004-12-18 15:56:29Z roland $
    33. 

  33.  */
    34. 

  34. 

  35. #include <linux/skbuff.h>
    36. 

  36. #include <linux/rtnetlink.h>
    37. 

  37. #include <linux/ip.h>
    38. 

  38. #include <linux/in.h>
    39. 

  39. #include <linux/igmp.h>
    40. 

  40. #include <linux/inetdevice.h>
    41. 

  41. #include <linux/delay.h>
    42. 

  42. #include <linux/completion.h>
    43. 

  43. 

  44. #include "ipoib.h"
    45. 

  45. 

  46. #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
    47. 

  47. static int mcast_debug_level;
    48. 

  48. 

  49. module_param(mcast_debug_level, int, 0644);
    50. 

  50. MODULE_PARM_DESC(mcast_debug_level,
    51. 

  51. 		 "Enable multicast debug tracing if > 0");
    52. 

  52. #endif
    53. 

  53. 

  54. static DECLARE_MUTEX(mcast_mutex);
    55. 

  55. 

  56. /* Used for all multicast joins (broadcast, IPv4 mcast and IPv6 mcast) */
    57. 

  57. struct ipoib_mcast {
    58. 

  58. 	struct ib_sa_mcmember_rec mcmember;
    59. 

  59. 	struct ipoib_ah          *ah;
    60. 

  60. 

  61. 	struct rb_node    rb_node;
    62. 

  62. 	struct list_head  list;
    63. 

  63. 	struct completion done;
    64. 

  64. 

  65. 	int                 query_id;
    66. 

  66. 	struct ib_sa_query *query;
    67. 

  67. 

  68. 	unsigned long created;
    69. 

  69. 	unsigned long backoff;
    70. 

  70. 

  71. 	unsigned long flags;
    72. 

  72. 	unsigned char logcount;
    73. 

  73. 

  74. 	struct list_head  neigh_list;
    75. 

  75. 

  76. 	struct sk_buff_head pkt_queue;
    77. 

  77. 

  78. 	struct net_device *dev;
    79. 

  79. };
    80. 

  80. 

  81. struct ipoib_mcast_iter {
    82. 

  82. 	struct net_device *dev;
    83. 

  83. 	union ib_gid       mgid;
    84. 

  84. 	unsigned long      created;
    85. 

  85. 	unsigned int       queuelen;
    86. 

  86. 	unsigned int       complete;
    87. 

  87. 	unsigned int       send_only;
    88. 

  88. };
    89. 

  89. 

  90. static void ipoib_mcast_free(struct ipoib_mcast *mcast)
    91. 

  91. {
    92. 

  92. 	struct net_device *dev = mcast->dev;
    93. 

  93. 	struct ipoib_dev_priv *priv = netdev_priv(dev);
    94. 

  94. 	struct ipoib_neigh *neigh, *tmp;
    95. 

  95. 	unsigned long flags;
    96. 

  96. 	LIST_HEAD(ah_list);
    97. 

  97. 	struct ipoib_ah *ah, *tah;
    98. 

  98. 

  99. 	ipoib_dbg_mcast(netdev_priv(dev),
    100. 

  100. 			"deleting multicast group " IPOIB_GID_FMT "\n",
    101. 

  101. 			IPOIB_GID_ARG(mcast->mcmember.mgid));
    102. 

  102. 

  103. 	spin_lock_irqsave(&priv->lock, flags);
    104. 

  104. 

  105. 	list_for_each_entry_safe(neigh, tmp, &mcast->neigh_list, list) {
    106. 

  106. 		if (neigh->ah)
    107. 

  107. 			list_add_tail(&neigh->ah->list, &ah_list);
    108. 

  108. 		*to_ipoib_neigh(neigh->neighbour) = NULL;
    109. 

  109. 		neigh->neighbour->ops->destructor = NULL;
    110. 

  110. 		kfree(neigh);
    111. 

  111. 	}
    112. 

  112. 

  113. 	spin_unlock_irqrestore(&priv->lock, flags);
    114. 

  114. 

  115. 	list_for_each_entry_safe(ah, tah, &ah_list, list)
    116. 

  116. 		ipoib_put_ah(ah);
    117. 

  117. 

  118. 	if (mcast->ah)
    119. 

  119. 		ipoib_put_ah(mcast->ah);
    120. 

  120. 

  121. 	while (!skb_queue_empty(&mcast->pkt_queue)) {
    122. 

  122. 		struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue);
    123. 

  123. 

  124. 		skb->dev = dev;
    125. 

  125. 		dev_kfree_skb_any(skb);
    126. 

  126. 	}
    127. 

  127. 

  128. 	kfree(mcast);
    129. 

  129. }
    130. 

  130. 

  131. static struct ipoib_mcast *ipoib_mcast_alloc(struct net_device *dev,
    132. 

  132. 					     int can_sleep)
    133. 

  133. {
    134. 

  134. 	struct ipoib_mcast *mcast;
    135. 

  135. 

  136. 	mcast = kmalloc(sizeof (*mcast), can_sleep ? GFP_KERNEL : GFP_ATOMIC);
    137. 

  137. 	if (!mcast)
    138. 

  138. 		return NULL;
    139. 

  139. 

  140. 	memset(mcast, 0, sizeof (*mcast));
    141. 

  141. 

  142. 	init_completion(&mcast->done);
    143. 

  143. 

  144. 	mcast->dev = dev;
    145. 

  145. 	mcast->created = jiffies;
    146. 

  146. 	mcast->backoff = HZ;
    147. 

  147. 	mcast->logcount = 0;
    148. 

  148. 

  149. 	INIT_LIST_HEAD(&mcast->list);
    150. 

  150. 	INIT_LIST_HEAD(&mcast->neigh_list);
    151. 

  151. 	skb_queue_head_init(&mcast->pkt_queue);
    152. 

  152. 

  153. 	mcast->ah    = NULL;
    154. 

  154. 	mcast->query = NULL;
    155. 

  155. 

  156. 	return mcast;
    157. 

  157. }
    158. 

  158. 

  159. static struct ipoib_mcast *__ipoib_mcast_find(struct net_device *dev, union ib_gid *mgid)
    160. 

  160. {
    161. 

  161. 	struct ipoib_dev_priv *priv = netdev_priv(dev);
    162. 

  162. 	struct rb_node *n = priv->multicast_tree.rb_node;
    163. 

  163. 

  164. 	while (n) {
    165. 

  165. 		struct ipoib_mcast *mcast;
    166. 

  166. 		int ret;
    167. 

  167. 

  168. 		mcast = rb_entry(n, struct ipoib_mcast, rb_node);
    169. 

  169. 

  170. 		ret = memcmp(mgid->raw, mcast->mcmember.mgid.raw,
    171. 

  171. 			     sizeof (union ib_gid));
    172. 

  172. 		if (ret < 0)
    173. 

  173. 			n = n->rb_left;
    174. 

  174. 		else if (ret > 0)
    175. 

  175. 			n = n->rb_right;
    176. 

  176. 		else
    177. 

  177. 			return mcast;
    178. 

  178. 	}
    179. 

  179. 

  180. 	return NULL;
    181. 

  181. }
    182. 

  182. 

  183. static int __ipoib_mcast_add(struct net_device *dev, struct ipoib_mcast *mcast)
    184. 

  184. {
    185. 

  185. 	struct ipoib_dev_priv *priv = netdev_priv(dev);
    186. 

  186. 	struct rb_node **n = &priv->multicast_tree.rb_node, *pn = NULL;
    187. 

  187. 

  188. 	while (*n) {
    189. 

  189. 		struct ipoib_mcast *tmcast;
    190. 

  190. 		int ret;
    191. 

  191. 

  192. 		pn = *n;
    193. 

  193. 		tmcast = rb_entry(pn, struct ipoib_mcast, rb_node);
    194. 

  194. 

  195. 		ret = memcmp(mcast->mcmember.mgid.raw, tmcast->mcmember.mgid.raw,
    196. 

  196. 			     sizeof (union ib_gid));
    197. 

  197. 		if (ret < 0)
    198. 

  198. 			n = &pn->rb_left;
    199. 

  199. 		else if (ret > 0)
    200. 

  200. 			n = &pn->rb_right;
    201. 

  201. 		else
    202. 

  202. 			return -EEXIST;
    203. 

  203. 	}
    204. 

  204. 

  205. 	rb_link_node(&mcast->rb_node, pn, n);
    206. 

  206. 	rb_insert_color(&mcast->rb_node, &priv->multicast_tree);
    207. 

  207. 

  208. 	return 0;
    209. 

  209. }
    210. 

  210. 

  211. static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
    212. 

  212. 				   struct ib_sa_mcmember_rec *mcmember)
    213. 

  213. {
    214. 

  214. 	struct net_device *dev = mcast->dev;
    215. 

  215. 	struct ipoib_dev_priv *priv = netdev_priv(dev);
    216. 

  216. 	int ret;
    217. 

  217. 

  218. 	mcast->mcmember = *mcmember;
    219. 

  219. 

  220. 	/* Set the cached Q_Key before we attach if it's the broadcast group */
    221. 

  221. 	if (!memcmp(mcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
    222. 

  222. 		    sizeof (union ib_gid))) {
    223. 

  223. 		priv->qkey = be32_to_cpu(priv->broadcast->mcmember.qkey);
    224. 

  224. 		priv->tx_wr.wr.ud.remote_qkey = priv->qkey;
    225. 

  225. 	}
    226. 

  226. 

  227. 	if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
    228. 

  228. 		if (test_and_set_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
    229. 

  229. 			ipoib_warn(priv, "multicast group " IPOIB_GID_FMT
    230. 

  230. 				   " already attached\n",
    231. 

  231. 				   IPOIB_GID_ARG(mcast->mcmember.mgid));
    232. 

  232. 

  233. 			return 0;
    234. 

  234. 		}
    235. 

  235. 

  236. 		ret = ipoib_mcast_attach(dev, be16_to_cpu(mcast->mcmember.mlid),
    237. 

  237. 					 &mcast->mcmember.mgid);
    238. 

  238. 		if (ret < 0) {
    239. 

  239. 			ipoib_warn(priv, "couldn't attach QP to multicast group "
    240. 

  240. 				   IPOIB_GID_FMT "\n",
    241. 

  241. 				   IPOIB_GID_ARG(mcast->mcmember.mgid));
    242. 

  242. 

  243. 			clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags);
    244. 

  244. 			return ret;
    245. 

  245. 		}
    246. 

  246. 	}
    247. 

  247. 

  248. 	{
    249. 

  249. 		struct ib_ah_attr av = {
    250. 

  250. 			.dlid	       = be16_to_cpu(mcast->mcmember.mlid),
    251. 

  251. 			.port_num      = priv->port,
    252. 

  252. 			.sl	       = mcast->mcmember.sl,
    253. 

  253. 			.ah_flags      = IB_AH_GRH,
    254. 

  254. 			.grh	       = {
    255. 

  255. 				.flow_label    = be32_to_cpu(mcast->mcmember.flow_label),
    256. 

  256. 				.hop_limit     = mcast->mcmember.hop_limit,
    257. 

  257. 				.sgid_index    = 0,
    258. 

  258. 				.traffic_class = mcast->mcmember.traffic_class
    259. 

  259. 			}
    260. 

  260. 		};
    261. 

  261. 

  262. 		av.grh.dgid = mcast->mcmember.mgid;
    263. 

  263. 

  264. 		if (ib_sa_rate_enum_to_int(mcast->mcmember.rate) > 0)
    265. 

  265. 			av.static_rate = (2 * priv->local_rate -
    266. 

  266. 					  ib_sa_rate_enum_to_int(mcast->mcmember.rate) - 1) /
    267. 

  267. 				(priv->local_rate ? priv->local_rate : 1);
    268. 

  268. 

  269. 		ipoib_dbg_mcast(priv, "static_rate %d for local port %dX, mcmember %dX\n",
    270. 

  270. 				av.static_rate, priv->local_rate,
    271. 

  271. 				ib_sa_rate_enum_to_int(mcast->mcmember.rate));
    272. 

  272. 

  273. 		mcast->ah = ipoib_create_ah(dev, priv->pd, &av);
    274. 

  274. 		if (!mcast->ah) {
    275. 

  275. 			ipoib_warn(priv, "ib_address_create failed\n");
    276. 

  276. 		} else {
    277. 

  277. 			ipoib_dbg_mcast(priv, "MGID " IPOIB_GID_FMT
    278. 

  278. 					" AV %p, LID 0x%04x, SL %d\n",
    279. 

  279. 					IPOIB_GID_ARG(mcast->mcmember.mgid),
    280. 

  280. 					mcast->ah->ah,
    281. 

  281. 					be16_to_cpu(mcast->mcmember.mlid),
    282. 

  282. 					mcast->mcmember.sl);
    283. 

  283. 		}
    284. 

  284. 	}
    285. 

  285. 

  286. 	/* actually send any queued packets */
    287. 

  287. 	while (!skb_queue_empty(&mcast->pkt_queue)) {
    288. 

  288. 		struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue);
    289. 

  289. 

  290. 		skb->dev = dev;
    291. 

  291. 

  292. 		if (!skb->dst || !skb->dst->neighbour) {
    293. 

  293. 			/* put pseudoheader back on for next time */
    294. 

  294. 			skb_push(skb, sizeof (struct ipoib_pseudoheader));
    295. 

  295. 		}
    296. 

  296. 

  297. 		if (dev_queue_xmit(skb))
    298. 

  298. 			ipoib_warn(priv, "dev_queue_xmit failed to requeue packet\n");
    299. 

  299. 	}
    300. 

  300. 

  301. 	return 0;
    302. 

  302. }
    303. 

  303. 

  304. static void
    305. 

  305. ipoib_mcast_sendonly_join_complete(int status,
    306. 

  306. 				   struct ib_sa_mcmember_rec *mcmember,
    307. 

  307. 				   void *mcast_ptr)
    308. 

  308. {
    309. 

  309. 	struct ipoib_mcast *mcast = mcast_ptr;
    310. 

  310. 	struct net_device *dev = mcast->dev;
    311. 

  311. 

  312. 	if (!status)
    313. 

  313. 		ipoib_mcast_join_finish(mcast, mcmember);
    314. 

  314. 	else {
    315. 

  315. 		if (mcast->logcount++ < 20)
    316. 

  316. 			ipoib_dbg_mcast(netdev_priv(dev), "multicast join failed for "
    317. 

  317. 					IPOIB_GID_FMT ", status %d\n",
    318. 

  318. 					IPOIB_GID_ARG(mcast->mcmember.mgid), status);
    319. 

  319. 

  320. 		/* Flush out any queued packets */
    321. 

  321. 		while (!skb_queue_empty(&mcast->pkt_queue)) {
    322. 

  322. 			struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue);
    323. 

  323. 

  324. 			skb->dev = dev;
    325. 

  325. 

  326. 			dev_kfree_skb_any(skb);
    327. 

  327. 		}
    328. 

  328. 

  329. 		/* Clear the busy flag so we try again */
    330. 

  330. 		clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
    331. 

  331. 	}
    332. 

  332. 

  333. 	complete(&mcast->done);
    334. 

  334. }
    335. 

  335. 

  336. static int ipoib_mcast_sendonly_join(struct ipoib_mcast *mcast)
    337. 

  337. {
    338. 

  338. 	struct net_device *dev = mcast->dev;
    339. 

  339. 	struct ipoib_dev_priv *priv = netdev_priv(dev);
    340. 

  340. 	struct ib_sa_mcmember_rec rec = {
    341. 

  341. #if 0				/* Some SMs don't support send-only yet */
    342. 

  342. 		.join_state = 4
    343. 

  343. #else
    344. 

  344. 		.join_state = 1
    345. 

  345. #endif
    346. 

  346. 	};
    347. 

  347. 	int ret = 0;
    348. 

  348. 

  349. 	if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
    350. 

  350. 		ipoib_dbg_mcast(priv, "device shutting down, no multicast joins\n");
    351. 

  351. 		return -ENODEV;
    352. 

  352. 	}
    353. 

  353. 

  354. 	if (test_and_set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) {
    355. 

  355. 		ipoib_dbg_mcast(priv, "multicast entry busy, skipping\n");
    356. 

  356. 		return -EBUSY;
    357. 

  357. 	}
    358. 

  358. 

  359. 	rec.mgid     = mcast->mcmember.mgid;
    360. 

  360. 	rec.port_gid = priv->local_gid;
    361. 

  361. 	rec.pkey     = be16_to_cpu(priv->pkey);
    362. 

  362. 

  363. 	ret = ib_sa_mcmember_rec_set(priv->ca, priv->port, &rec,
    364. 

  364. 				     IB_SA_MCMEMBER_REC_MGID		|
    365. 

  365. 				     IB_SA_MCMEMBER_REC_PORT_GID	|
    366. 

  366. 				     IB_SA_MCMEMBER_REC_PKEY		|
    367. 

  367. 				     IB_SA_MCMEMBER_REC_JOIN_STATE,
    368. 

  368. 				     1000, GFP_ATOMIC,
    369. 

  369. 				     ipoib_mcast_sendonly_join_complete,
    370. 

  370. 				     mcast, &mcast->query);
    371. 

  371. 	if (ret < 0) {
    372. 

  372. 		ipoib_warn(priv, "ib_sa_mcmember_rec_set failed (ret = %d)\n",
    373. 

  373. 			   ret);
    374. 

  374. 	} else {
    375. 

  375. 		ipoib_dbg_mcast(priv, "no multicast record for " IPOIB_GID_FMT
    376. 

  376. 				", starting join\n",
    377. 

  377. 				IPOIB_GID_ARG(mcast->mcmember.mgid));
    378. 

  378. 

  379. 		mcast->query_id = ret;
    380. 

  380. 	}
    381. 

  381. 

  382. 	return ret;
    383. 

  383. }
    384. 

  384. 

  385. static void ipoib_mcast_join_complete(int status,
    386. 

  386. 				      struct ib_sa_mcmember_rec *mcmember,
    387. 

  387. 				      void *mcast_ptr)
    388. 

  388. {
    389. 

  389. 	struct ipoib_mcast *mcast = mcast_ptr;
    390. 

  390. 	struct net_device *dev = mcast->dev;
    391. 

  391. 	struct ipoib_dev_priv *priv = netdev_priv(dev);
    392. 

  392. 

  393. 	ipoib_dbg_mcast(priv, "join completion for " IPOIB_GID_FMT
    394. 

  394. 			" (status %d)\n",
    395. 

  395. 			IPOIB_GID_ARG(mcast->mcmember.mgid), status);
    396. 

  396. 

  397. 	if (!status && !ipoib_mcast_join_finish(mcast, mcmember)) {
    398. 

  398. 		mcast->backoff = HZ;
    399. 

  399. 		down(&mcast_mutex);
    400. 

  400. 		if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
    401. 

  401. 			queue_work(ipoib_workqueue, &priv->mcast_task);
    402. 

  402. 		up(&mcast_mutex);
    403. 

  403. 		complete(&mcast->done);
    404. 

  404. 		return;
    405. 

  405. 	}
    406. 

  406. 

  407. 	if (status == -EINTR) {
    408. 

  408. 		complete(&mcast->done);
    409. 

  409. 		return;
    410. 

  410. 	}
    411. 

  411. 

  412. 	if (status && mcast->logcount++ < 20) {
    413. 

  413. 		if (status == -ETIMEDOUT || status == -EINTR) {
    414. 

  414. 			ipoib_dbg_mcast(priv, "multicast join failed for " IPOIB_GID_FMT
    415. 

  415. 					", status %d\n",
    416. 

  416. 					IPOIB_GID_ARG(mcast->mcmember.mgid),
    417. 

  417. 					status);
    418. 

  418. 		} else {
    419. 

  419. 			ipoib_warn(priv, "multicast join failed for "
    420. 

  420. 				   IPOIB_GID_FMT ", status %d\n",
    421. 

  421. 				   IPOIB_GID_ARG(mcast->mcmember.mgid),
    422. 

  422. 				   status);
    423. 

  423. 		}
    424. 

  424. 	}
    425. 

  425. 

  426. 	mcast->backoff *= 2;
    427. 

  427. 	if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
    428. 

  428. 		mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
    429. 

  429. 

  430. 	mcast->query = NULL;
    431. 

  431. 

  432. 	down(&mcast_mutex);
    433. 

  433. 	if (test_bit(IPOIB_MCAST_RUN, &priv->flags)) {
    434. 

  434. 		if (status == -ETIMEDOUT)
    435. 

  435. 			queue_work(ipoib_workqueue, &priv->mcast_task);
    436. 

  436. 		else
    437. 

  437. 			queue_delayed_work(ipoib_workqueue, &priv->mcast_task,
    438. 

  438. 					   mcast->backoff * HZ);
    439. 

  439. 	} else
    440. 

  440. 		complete(&mcast->done);
    441. 

  441. 	up(&mcast_mutex);
    442. 

  442. 

  443. 	return;
    444. 

  444. }
    445. 

  445. 

  446. static void ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast,
    447. 

  447. 			     int create)
    448. 

  448. {
    449. 

  449. 	struct ipoib_dev_priv *priv = netdev_priv(dev);
    450. 

  450. 	struct ib_sa_mcmember_rec rec = {
    451. 

  451. 		.join_state = 1
    452. 

  452. 	};
    453. 

  453. 	ib_sa_comp_mask comp_mask;
    454. 

  454. 	int ret = 0;
    455. 

  455. 

  456. 	ipoib_dbg_mcast(priv, "joining MGID " IPOIB_GID_FMT "\n",
    457. 

  457. 			IPOIB_GID_ARG(mcast->mcmember.mgid));
    458. 

  458. 

  459. 	rec.mgid     = mcast->mcmember.mgid;
    460. 

  460. 	rec.port_gid = priv->local_gid;
    461. 

  461. 	rec.pkey     = be16_to_cpu(priv->pkey);
    462. 

  462. 

  463. 	comp_mask =
    464. 

  464. 		IB_SA_MCMEMBER_REC_MGID		|
    465. 

  465. 		IB_SA_MCMEMBER_REC_PORT_GID	|
    466. 

  466. 		IB_SA_MCMEMBER_REC_PKEY		|
    467. 

  467. 		IB_SA_MCMEMBER_REC_JOIN_STATE;
    468. 

  468. 

  469. 	if (create) {
    470. 

  470. 		comp_mask |=
    471. 

  471. 			IB_SA_MCMEMBER_REC_QKEY		|
    472. 

  472. 			IB_SA_MCMEMBER_REC_SL		|
    473. 

  473. 			IB_SA_MCMEMBER_REC_FLOW_LABEL	|
    474. 

  474. 			IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
    475. 

  475. 

  476. 		rec.qkey	  = priv->broadcast->mcmember.qkey;
    477. 

  477. 		rec.sl		  = priv->broadcast->mcmember.sl;
    478. 

  478. 		rec.flow_label	  = priv->broadcast->mcmember.flow_label;
    479. 

  479. 		rec.traffic_class = priv->broadcast->mcmember.traffic_class;
    480. 

  480. 	}
    481. 

  481. 

  482. 	ret = ib_sa_mcmember_rec_set(priv->ca, priv->port, &rec, comp_mask,
    483. 

  483. 				     mcast->backoff * 1000, GFP_ATOMIC,
    484. 

  484. 				     ipoib_mcast_join_complete,
    485. 

  485. 				     mcast, &mcast->query);
    486. 

  486. 

  487. 	if (ret < 0) {
    488. 

  488. 		ipoib_warn(priv, "ib_sa_mcmember_rec_set failed, status %d\n", ret);
    489. 

  489. 

  490. 		mcast->backoff *= 2;
    491. 

  491. 		if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
    492. 

  492. 			mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
    493. 

  493. 

  494. 		down(&mcast_mutex);
    495. 

  495. 		if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
    496. 

  496. 			queue_delayed_work(ipoib_workqueue,
    497. 

  497. 					   &priv->mcast_task,
    498. 

  498. 					   mcast->backoff);
    499. 

  499. 		up(&mcast_mutex);
    500. 

  500. 	} else
    501. 

  501. 		mcast->query_id = ret;
    502. 

  502. }
    503. 

  503. 

  504. void ipoib_mcast_join_task(void *dev_ptr)
    505. 

  505. {
    506. 

  506. 	struct net_device *dev = dev_ptr;
    507. 

  507. 	struct ipoib_dev_priv *priv = netdev_priv(dev);
    508. 

  508. 

  509. 	if (!test_bit(IPOIB_MCAST_RUN, &priv->flags))
    510. 

  510. 		return;
    511. 

  511. 

  512. 	if (ib_query_gid(priv->ca, priv->port, 0, &priv->local_gid))
    513. 

  513. 		ipoib_warn(priv, "ib_gid_entry_get() failed\n");
    514. 

  514. 	else
    515. 

  515. 		memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw, sizeof (union ib_gid));
    516. 

  516. 

  517. 	{
    518. 

  518. 		struct ib_port_attr attr;
    519. 

  519. 

  520. 		if (!ib_query_port(priv->ca, priv->port, &attr)) {
    521. 

  521. 			priv->local_lid  = attr.lid;
    522. 

  522. 			priv->local_rate = attr.active_speed *
    523. 

  523. 				ib_width_enum_to_int(attr.active_width);
    524. 

  524. 		} else
    525. 

  525. 			ipoib_warn(priv, "ib_query_port failed\n");
    526. 

  526. 	}
    527. 

  527. 

  528. 	if (!priv->broadcast) {
    529. 

  529. 		priv->broadcast = ipoib_mcast_alloc(dev, 1);
    530. 

  530. 		if (!priv->broadcast) {
    531. 

  531. 			ipoib_warn(priv, "failed to allocate broadcast group\n");
    532. 

  532. 			down(&mcast_mutex);
    533. 

  533. 			if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
    534. 

  534. 				queue_delayed_work(ipoib_workqueue,
    535. 

  535. 						   &priv->mcast_task, HZ);
    536. 

  536. 			up(&mcast_mutex);
    537. 

  537. 			return;
    538. 

  538. 		}
    539. 

  539. 

  540. 		memcpy(priv->broadcast->mcmember.mgid.raw, priv->dev->broadcast + 4,
    541. 

  541. 		       sizeof (union ib_gid));
    542. 

  542. 

  543. 		spin_lock_irq(&priv->lock);
    544. 

  544. 		__ipoib_mcast_add(dev, priv->broadcast);
    545. 

  545. 		spin_unlock_irq(&priv->lock);
    546. 

  546. 	}
    547. 

  547. 

  548. 	if (!test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
    549. 

  549. 		ipoib_mcast_join(dev, priv->broadcast, 0);
    550. 

  550. 		return;
    551. 

  551. 	}
    552. 

  552. 

  553. 	while (1) {
    554. 

  554. 		struct ipoib_mcast *mcast = NULL;
    555. 

  555. 

  556. 		spin_lock_irq(&priv->lock);
    557. 

  557. 		list_for_each_entry(mcast, &priv->multicast_list, list) {
    558. 

  558. 			if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)
    559. 

  559. 			    && !test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)
    560. 

  560. 			    && !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
    561. 

  561. 				/* Found the next unjoined group */
    562. 

  562. 				break;
    563. 

  563. 			}
    564. 

  564. 		}
    565. 

  565. 		spin_unlock_irq(&priv->lock);
    566. 

  566. 

  567. 		if (&mcast->list == &priv->multicast_list) {
    568. 

  568. 			/* All done */
    569. 

  569. 			break;
    570. 

  570. 		}
    571. 

  571. 

  572. 		ipoib_mcast_join(dev, mcast, 1);
    573. 

  573. 		return;
    574. 

  574. 	}
    575. 

  575. 

  576. 	priv->mcast_mtu = ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu) -
    577. 

  577. 		IPOIB_ENCAP_LEN;
    578. 

  578. 	dev->mtu = min(priv->mcast_mtu, priv->admin_mtu);
    579. 

  579. 

  580. 	ipoib_dbg_mcast(priv, "successfully joined all multicast groups\n");
    581. 

  581. 

  582. 	clear_bit(IPOIB_MCAST_RUN, &priv->flags);
    583. 

  583. 	netif_carrier_on(dev);
    584. 

  584. }
    585. 

  585. 

  586. int ipoib_mcast_start_thread(struct net_device *dev)
    587. 

  587. {
    588. 

  588. 	struct ipoib_dev_priv *priv = netdev_priv(dev);
    589. 

  589. 

  590. 	ipoib_dbg_mcast(priv, "starting multicast thread\n");
    591. 

  591. 

  592. 	down(&mcast_mutex);
    593. 

  593. 	if (!test_and_set_bit(IPOIB_MCAST_RUN, &priv->flags))
    594. 

  594. 		queue_work(ipoib_workqueue, &priv->mcast_task);
    595. 

  595. 	up(&mcast_mutex);
    596. 

  596. 

  597. 	return 0;
    598. 

  598. }
    599. 

  599. 

  600. int ipoib_mcast_stop_thread(struct net_device *dev)
    601. 

  601. {
    602. 

  602. 	struct ipoib_dev_priv *priv = netdev_priv(dev);
    603. 

  603. 	struct ipoib_mcast *mcast;
    604. 

  604. 

  605. 	ipoib_dbg_mcast(priv, "stopping multicast thread\n");
    606. 

  606. 

  607. 	down(&mcast_mutex);
    608. 

  608. 	clear_bit(IPOIB_MCAST_RUN, &priv->flags);
    609. 

  609. 	cancel_delayed_work(&priv->mcast_task);
    610. 

  610. 	up(&mcast_mutex);
    611. 

  611. 

  612. 	flush_workqueue(ipoib_workqueue);
    613. 

  613. 

  614. 	if (priv->broadcast && priv->broadcast->query) {
    615. 

  615. 		ib_sa_cancel_query(priv->broadcast->query_id, priv->broadcast->query);
    616. 

  616. 		priv->broadcast->query = NULL;
    617. 

  617. 		ipoib_dbg_mcast(priv, "waiting for bcast\n");
    618. 

  618. 		wait_for_completion(&priv->broadcast->done);
    619. 

  619. 	}
    620. 

  620. 

  621. 	list_for_each_entry(mcast, &priv->multicast_list, list) {
    622. 

  622. 		if (mcast->query) {
    623. 

  623. 			ib_sa_cancel_query(mcast->query_id, mcast->query);
    624. 

  624. 			mcast->query = NULL;
    625. 

  625. 			ipoib_dbg_mcast(priv, "waiting for MGID " IPOIB_GID_FMT "\n",
    626. 

  626. 					IPOIB_GID_ARG(mcast->mcmember.mgid));
    627. 

  627. 			wait_for_completion(&mcast->done);
    628. 

  628. 		}
    629. 

  629. 	}
    630. 

  630. 

  631. 	return 0;
    632. 

  632. }
    633. 

  633. 

  634. static int ipoib_mcast_leave(struct net_device *dev, struct ipoib_mcast *mcast)
    635. 

  635. {
    636. 

  636. 	struct ipoib_dev_priv *priv = netdev_priv(dev);
    637. 

  637. 	struct ib_sa_mcmember_rec rec = {
    638. 

  638. 		.join_state = 1
    639. 

  639. 	};
    640. 

  640. 	int ret = 0;
    641. 

  641. 

  642. 	if (!test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags))
    643. 

  643. 		return 0;
    644. 

  644. 

  645. 	ipoib_dbg_mcast(priv, "leaving MGID " IPOIB_GID_FMT "\n",
    646. 

  646. 			IPOIB_GID_ARG(mcast->mcmember.mgid));
    647. 

  647. 

  648. 	rec.mgid     = mcast->mcmember.mgid;
    649. 

  649. 	rec.port_gid = priv->local_gid;
    650. 

  650. 	rec.pkey     = be16_to_cpu(priv->pkey);
    651. 

  651. 

  652. 	/* Remove ourselves from the multicast group */
    653. 

  653. 	ret = ipoib_mcast_detach(dev, be16_to_cpu(mcast->mcmember.mlid),
    654. 

  654. 				 &mcast->mcmember.mgid);
    655. 

  655. 	if (ret)
    656. 

  656. 		ipoib_warn(priv, "ipoib_mcast_detach failed (result = %d)\n", ret);
    657. 

  657. 

  658. 	/*
    659. 

  659. 	 * Just make one shot at leaving and don't wait for a reply;
    660. 

  660. 	 * if we fail, too bad.
    661. 

  661. 	 */
    662. 

  662. 	ret = ib_sa_mcmember_rec_delete(priv->ca, priv->port, &rec,
    663. 

  663. 					IB_SA_MCMEMBER_REC_MGID		|
    664. 

  664. 					IB_SA_MCMEMBER_REC_PORT_GID	|
    665. 

  665. 					IB_SA_MCMEMBER_REC_PKEY		|
    666. 

  666. 					IB_SA_MCMEMBER_REC_JOIN_STATE,
    667. 

  667. 					0, GFP_ATOMIC, NULL,
    668. 

  668. 					mcast, &mcast->query);
    669. 

  669. 	if (ret < 0)
    670. 

  670. 		ipoib_warn(priv, "ib_sa_mcmember_rec_delete failed "
    671. 

  671. 			   "for leave (result = %d)\n", ret);
    672. 

  672. 

  673. 	return 0;
    674. 

  674. }
    675. 

  675. 

  676. void ipoib_mcast_send(struct net_device *dev, union ib_gid *mgid,
    677. 

  677. 		      struct sk_buff *skb)
    678. 

  678. {
    679. 

  679. 	struct ipoib_dev_priv *priv = netdev_priv(dev);
    680. 

  680. 	struct ipoib_mcast *mcast;
    681. 

  681. 

  682. 	/*
    683. 

  683. 	 * We can only be called from ipoib_start_xmit, so we're
    684. 

  684. 	 * inside tx_lock -- no need to save/restore flags.
    685. 

  685. 	 */
    686. 

  686. 	spin_lock(&priv->lock);
    687. 

  687. 

  688. 	mcast = __ipoib_mcast_find(dev, mgid);
    689. 

  689. 	if (!mcast) {
    690. 

  690. 		/* Let's create a new send only group now */
    691. 

  691. 		ipoib_dbg_mcast(priv, "setting up send only multicast group for "
    692. 

  692. 				IPOIB_GID_FMT "\n", IPOIB_GID_ARG(*mgid));
    693. 

  693. 

  694. 		mcast = ipoib_mcast_alloc(dev, 0);
    695. 

  695. 		if (!mcast) {
    696. 

  696. 			ipoib_warn(priv, "unable to allocate memory for "
    697. 

  697. 				   "multicast structure\n");
    698. 

  698. 			dev_kfree_skb_any(skb);
    699. 

  699. 			goto out;
    700. 

  700. 		}
    701. 

  701. 

  702. 		set_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags);
    703. 

  703. 		mcast->mcmember.mgid = *mgid;
    704. 

  704. 		__ipoib_mcast_add(dev, mcast);
    705. 

  705. 		list_add_tail(&mcast->list, &priv->multicast_list);
    706. 

  706. 	}
    707. 

  707. 

  708. 	if (!mcast->ah) {
    709. 

  709. 		if (skb_queue_len(&mcast->pkt_queue) < IPOIB_MAX_MCAST_QUEUE)
    710. 

  710. 			skb_queue_tail(&mcast->pkt_queue, skb);
    711. 

  711. 		else
    712. 

  712. 			dev_kfree_skb_any(skb);
    713. 

  713. 

  714. 		if (mcast->query)
    715. 

  715. 			ipoib_dbg_mcast(priv, "no address vector, "
    716. 

  716. 					"but multicast join already started\n");
    717. 

  717. 		else if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags))
    718. 

  718. 			ipoib_mcast_sendonly_join(mcast);
    719. 

  719. 

  720. 		/*
    721. 

  721. 		 * If lookup completes between here and out:, don't
    722. 

  722. 		 * want to send packet twice.
    723. 

  723. 		 */
    724. 

  724. 		mcast = NULL;
    725. 

  725. 	}
    726. 

  726. 

  727. out:
    728. 

  728. 	if (mcast && mcast->ah) {
    729. 

  729. 		if (skb->dst            &&
    730. 

  730. 		    skb->dst->neighbour &&
    731. 

  731. 		    !*to_ipoib_neigh(skb->dst->neighbour)) {
    732. 

  732. 			struct ipoib_neigh *neigh = kmalloc(sizeof *neigh, GFP_ATOMIC);
    733. 

  733. 

  734. 			if (neigh) {
    735. 

  735. 				kref_get(&mcast->ah->ref);
    736. 

  736. 				neigh->ah  	= mcast->ah;
    737. 

  737. 				neigh->neighbour = skb->dst->neighbour;
    738. 

  738. 				*to_ipoib_neigh(skb->dst->neighbour) = neigh;
    739. 

  739. 				list_add_tail(&neigh->list, &mcast->neigh_list);
    740. 

  740. 			}
    741. 

  741. 		}
    742. 

  742. 

  743. 		ipoib_send(dev, skb, mcast->ah, IB_MULTICAST_QPN);
    744. 

  744. 	}
    745. 

  745. 

  746. 	spin_unlock(&priv->lock);
    747. 

  747. }
    748. 

  748. 

  749. void ipoib_mcast_dev_flush(struct net_device *dev)
    750. 

  750. {
    751. 

  751. 	struct ipoib_dev_priv *priv = netdev_priv(dev);
    752. 

  752. 	LIST_HEAD(remove_list);
    753. 

  753. 	struct ipoib_mcast *mcast, *tmcast, *nmcast;
    754. 

  754. 	unsigned long flags;
    755. 

  755. 

  756. 	ipoib_dbg_mcast(priv, "flushing multicast list\n");
    757. 

  757. 

  758. 	spin_lock_irqsave(&priv->lock, flags);
    759. 

  759. 	list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
    760. 

  760. 		nmcast = ipoib_mcast_alloc(dev, 0);
    761. 

  761. 		if (nmcast) {
    762. 

  762. 			nmcast->flags =
    763. 

  763. 				mcast->flags & (1 << IPOIB_MCAST_FLAG_SENDONLY);
    764. 

  764. 

  765. 			nmcast->mcmember.mgid = mcast->mcmember.mgid;
    766. 

  766. 

  767. 			/* Add the new group in before the to-be-destroyed group */
    768. 

  768. 			list_add_tail(&nmcast->list, &mcast->list);
    769. 

  769. 			list_del_init(&mcast->list);
    770. 

  770. 

  771. 			rb_replace_node(&mcast->rb_node, &nmcast->rb_node,
    772. 

  772. 					&priv->multicast_tree);
    773. 

  773. 

  774. 			list_add_tail(&mcast->list, &remove_list);
    775. 

  775. 		} else {
    776. 

  776. 			ipoib_warn(priv, "could not reallocate multicast group "
    777. 

  777. 				   IPOIB_GID_FMT "\n",
    778. 

  778. 				   IPOIB_GID_ARG(mcast->mcmember.mgid));
    779. 

  779. 		}
    780. 

  780. 	}
    781. 

  781. 

  782. 	if (priv->broadcast) {
    783. 

  783. 		nmcast = ipoib_mcast_alloc(dev, 0);
    784. 

  784. 		if (nmcast) {
    785. 

  785. 			nmcast->mcmember.mgid = priv->broadcast->mcmember.mgid;
    786. 

  786. 

  787. 			rb_replace_node(&priv->broadcast->rb_node,
    788. 

  788. 					&nmcast->rb_node,
    789. 

  789. 					&priv->multicast_tree);
    790. 

  790. 

  791. 			list_add_tail(&priv->broadcast->list, &remove_list);
    792. 

  792. 		}
    793. 

  793. 

  794. 		priv->broadcast = nmcast;
    795. 

  795. 	}
    796. 

  796. 

  797. 	spin_unlock_irqrestore(&priv->lock, flags);
    798. 

  798. 

  799. 	list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
    800. 

  800. 		ipoib_mcast_leave(dev, mcast);
    801. 

  801. 		ipoib_mcast_free(mcast);
    802. 

  802. 	}
    803. 

  803. }
    804. 

  804. 

  805. void ipoib_mcast_dev_down(struct net_device *dev)
    806. 

  806. {
    807. 

  807. 	struct ipoib_dev_priv *priv = netdev_priv(dev);
    808. 

  808. 	unsigned long flags;
    809. 

  809. 

  810. 	/* Delete broadcast since it will be recreated */
    811. 

  811. 	if (priv->broadcast) {
    812. 

  812. 		ipoib_dbg_mcast(priv, "deleting broadcast group\n");
    813. 

  813. 

  814. 		spin_lock_irqsave(&priv->lock, flags);
    815. 

  815. 		rb_erase(&priv->broadcast->rb_node, &priv->multicast_tree);
    816. 

  816. 		spin_unlock_irqrestore(&priv->lock, flags);
    817. 

  817. 		ipoib_mcast_leave(dev, priv->broadcast);
    818. 

  818. 		ipoib_mcast_free(priv->broadcast);
    819. 

  819. 		priv->broadcast = NULL;
    820. 

  820. 	}
    821. 

  821. }
    822. 

  822. 

  823. void ipoib_mcast_restart_task(void *dev_ptr)
    824. 

  824. {
    825. 

  825. 	struct net_device *dev = dev_ptr;
    826. 

  826. 	struct ipoib_dev_priv *priv = netdev_priv(dev);
    827. 

  827. 	struct dev_mc_list *mclist;
    828. 

  828. 	struct ipoib_mcast *mcast, *tmcast;
    829. 

  829. 	LIST_HEAD(remove_list);
    830. 

  830. 	unsigned long flags;
    831. 

  831. 

  832. 	ipoib_dbg_mcast(priv, "restarting multicast task\n");
    833. 

  833. 

  834. 	ipoib_mcast_stop_thread(dev);
    835. 

  835. 

  836. 	spin_lock_irqsave(&priv->lock, flags);
    837. 

  837. 

  838. 	/*
    839. 

  839. 	 * Unfortunately, the networking core only gives us a list of all of
    840. 

  840. 	 * the multicast hardware addresses. We need to figure out which ones
    841. 

  841. 	 * are new and which ones have been removed
    842. 

  842. 	 */
    843. 

  843. 

  844. 	/* Clear out the found flag */
    845. 

  845. 	list_for_each_entry(mcast, &priv->multicast_list, list)
    846. 

  846. 		clear_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
    847. 

  847. 

  848. 	/* Mark all of the entries that are found or don't exist */
    849. 

  849. 	for (mclist = dev->mc_list; mclist; mclist = mclist->next) {
    850. 

  850. 		union ib_gid mgid;
    851. 

  851. 

  852. 		memcpy(mgid.raw, mclist->dmi_addr + 4, sizeof mgid);
    853. 

  853. 

  854. 		/* Add in the P_Key */
    855. 

  855. 		mgid.raw[4] = (priv->pkey >> 8) & 0xff;
    856. 

  856. 		mgid.raw[5] = priv->pkey & 0xff;
    857. 

  857. 

  858. 		mcast = __ipoib_mcast_find(dev, &mgid);
    859. 

  859. 		if (!mcast || test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
    860. 

  860. 			struct ipoib_mcast *nmcast;
    861. 

  861. 

  862. 			/* Not found or send-only group, let's add a new entry */
    863. 

  863. 			ipoib_dbg_mcast(priv, "adding multicast entry for mgid "
    864. 

  864. 					IPOIB_GID_FMT "\n", IPOIB_GID_ARG(mgid));
    865. 

  865. 

  866. 			nmcast = ipoib_mcast_alloc(dev, 0);
    867. 

  867. 			if (!nmcast) {
    868. 

  868. 				ipoib_warn(priv, "unable to allocate memory for multicast structure\n");
    869. 

  869. 				continue;
    870. 

  870. 			}
    871. 

  871. 

  872. 			set_bit(IPOIB_MCAST_FLAG_FOUND, &nmcast->flags);
    873. 

  873. 

  874. 			nmcast->mcmember.mgid = mgid;
    875. 

  875. 

  876. 			if (mcast) {
    877. 

  877. 				/* Destroy the send only entry */
    878. 

  878. 				list_del(&mcast->list);
    879. 

  879. 				list_add_tail(&mcast->list, &remove_list);
    880. 

  880. 

  881. 				rb_replace_node(&mcast->rb_node,
    882. 

  882. 						&nmcast->rb_node,
    883. 

  883. 						&priv->multicast_tree);
    884. 

  884. 			} else
    885. 

  885. 				__ipoib_mcast_add(dev, nmcast);
    886. 

  886. 

  887. 			list_add_tail(&nmcast->list, &priv->multicast_list);
    888. 

  888. 		}
    889. 

  889. 

  890. 		if (mcast)
    891. 

  891. 			set_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
    892. 

  892. 	}
    893. 

  893. 

  894. 	/* Remove all of the entries don't exist anymore */
    895. 

  895. 	list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
    896. 

  896. 		if (!test_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags) &&
    897. 

  897. 		    !test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
    898. 

  898. 			ipoib_dbg_mcast(priv, "deleting multicast group " IPOIB_GID_FMT "\n",
    899. 

  899. 					IPOIB_GID_ARG(mcast->mcmember.mgid));
    900. 

  900. 

  901. 			rb_erase(&mcast->rb_node, &priv->multicast_tree);
    902. 

  902. 

  903. 			/* Move to the remove list */
    904. 

  904. 			list_del(&mcast->list);
    905. 

  905. 			list_add_tail(&mcast->list, &remove_list);
    906. 

  906. 		}
    907. 

  907. 	}
    908. 

  908. 	spin_unlock_irqrestore(&priv->lock, flags);
    909. 

  909. 

  910. 	/* We have to cancel outside of the spinlock */
    911. 

  911. 	list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
    912. 

  912. 		ipoib_mcast_leave(mcast->dev, mcast);
    913. 

  913. 		ipoib_mcast_free(mcast);
    914. 

  914. 	}
    915. 

  915. 

  916. 	if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
    917. 

  917. 		ipoib_mcast_start_thread(dev);
    918. 

  918. }
    919. 

  919. 

  920. struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev)
    921. 

  921. {
    922. 

  922. 	struct ipoib_mcast_iter *iter;
    923. 

  923. 

  924. 	iter = kmalloc(sizeof *iter, GFP_KERNEL);
    925. 

  925. 	if (!iter)
    926. 

  926. 		return NULL;
    927. 

  927. 

  928. 	iter->dev = dev;
    929. 

  929. 	memset(iter->mgid.raw, 0, sizeof iter->mgid);
    930. 

  930. 

  931. 	if (ipoib_mcast_iter_next(iter)) {
    932. 

  932. 		ipoib_mcast_iter_free(iter);
    933. 

  933. 		return NULL;
    934. 

  934. 	}
    935. 

  935. 

  936. 	return iter;
    937. 

  937. }
    938. 

  938. 

  939. void ipoib_mcast_iter_free(struct ipoib_mcast_iter *iter)
    940. 

  940. {
    941. 

  941. 	kfree(iter);
    942. 

  942. }
    943. 

  943. 

  944. int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter)
    945. 

  945. {
    946. 

  946. 	struct ipoib_dev_priv *priv = netdev_priv(iter->dev);
    947. 

  947. 	struct rb_node *n;
    948. 

  948. 	struct ipoib_mcast *mcast;
    949. 

  949. 	int ret = 1;
    950. 

  950. 

  951. 	spin_lock_irq(&priv->lock);
    952. 

  952. 

  953. 	n = rb_first(&priv->multicast_tree);
    954. 

  954. 

  955. 	while (n) {
    956. 

  956. 		mcast = rb_entry(n, struct ipoib_mcast, rb_node);
    957. 

  957. 

  958. 		if (memcmp(iter->mgid.raw, mcast->mcmember.mgid.raw,
    959. 

  959. 			   sizeof (union ib_gid)) < 0) {
    960. 

  960. 			iter->mgid      = mcast->mcmember.mgid;
    961. 

  961. 			iter->created   = mcast->created;
    962. 

  962. 			iter->queuelen  = skb_queue_len(&mcast->pkt_queue);
    963. 

  963. 			iter->complete  = !!mcast->ah;
    964. 

  964. 			iter->send_only = !!(mcast->flags & (1 << IPOIB_MCAST_FLAG_SENDONLY));
    965. 

  965. 

  966. 			ret = 0;
    967. 

  967. 

  968. 			break;
    969. 

  969. 		}
    970. 

  970. 

  971. 		n = rb_next(n);
    972. 

  972. 	}
    973. 

  973. 

  974. 	spin_unlock_irq(&priv->lock);
    975. 

  975. 

  976. 	return ret;
    977. 

  977. }
    978. 

  978. 

  979. void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
    980. 

  980. 			   union ib_gid *mgid,
    981. 

  981. 			   unsigned long *created,
    982. 

  982. 			   unsigned int *queuelen,
    983. 

  983. 			   unsigned int *complete,
    984. 

  984. 			   unsigned int *send_only)
    985. 

  985. {
    986. 

  986. 	*mgid      = iter->mgid;
    987. 

  987. 	*created   = iter->created;
    988. 

  988. 	*queuelen  = iter->queuelen;
    989. 

  989. 	*complete  = iter->complete;
    990. 

  990. 	*send_only = iter->send_only;
    991. 

  991. }
    992.