1/* 2 * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved. 3 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved. 4 * Copyright (c) 2004 Voltaire, Inc. All rights reserved. 5 * 6 * This software is available to you under a choice of one of two 7 * licenses. You may choose to be licensed under the terms of the GNU 8 * General Public License (GPL) Version 2, available from the file 9 * COPYING in the main directory of this source tree, or the 10 * OpenIB.org BSD license below: 11 * 12 * Redistribution and use in source and binary forms, with or 13 * without modification, are permitted provided that the following 14 * conditions are met: 15 * 16 * - Redistributions of source code must retain the above 17 * copyright notice, this list of conditions and the following 18 * disclaimer. 19 * 20 * - Redistributions in binary form must reproduce the above 21 * copyright notice, this list of conditions and the following 22 * disclaimer in the documentation and/or other materials 23 * provided with the distribution. 24 * 25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 32 * SOFTWARE. 33 * 34 * $Id: ipoib_multicast.c,v 1.1.1.1 2007/08/03 18:52:32 Exp $ 35 */ 36 37#include <linux/skbuff.h> 38#include <linux/rtnetlink.h> 39#include <linux/ip.h> 40#include <linux/in.h> 41#include <linux/igmp.h> 42#include <linux/inetdevice.h> 43#include <linux/delay.h> 44#include <linux/completion.h> 45 46#include <net/dst.h> 47 48#include "ipoib.h" 49 50#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG 51static int mcast_debug_level; 52 53module_param(mcast_debug_level, int, 0644); 54MODULE_PARM_DESC(mcast_debug_level, 55 "Enable multicast debug tracing if > 0"); 56#endif 57 58static DEFINE_MUTEX(mcast_mutex); 59 60/* Used for all multicast joins (broadcast, IPv4 mcast and IPv6 mcast) */ 61struct ipoib_mcast { 62 struct ib_sa_mcmember_rec mcmember; 63 struct ib_sa_multicast *mc; 64 struct ipoib_ah *ah; 65 66 struct rb_node rb_node; 67 struct list_head list; 68 69 unsigned long created; 70 unsigned long backoff; 71 72 unsigned long flags; 73 unsigned char logcount; 74 75 struct list_head neigh_list; 76 77 struct sk_buff_head pkt_queue; 78 79 struct net_device *dev; 80}; 81 82struct ipoib_mcast_iter { 83 struct net_device *dev; 84 union ib_gid mgid; 85 unsigned long created; 86 unsigned int queuelen; 87 unsigned int complete; 88 unsigned int send_only; 89}; 90 91static void ipoib_mcast_free(struct ipoib_mcast *mcast) 92{ 93 struct net_device *dev = mcast->dev; 94 struct ipoib_dev_priv *priv = netdev_priv(dev); 95 struct ipoib_neigh *neigh, *tmp; 96 unsigned long flags; 97 int tx_dropped = 0; 98 99 ipoib_dbg_mcast(netdev_priv(dev), 100 "deleting multicast group " IPOIB_GID_FMT "\n", 101 IPOIB_GID_ARG(mcast->mcmember.mgid)); 102 103 spin_lock_irqsave(&priv->lock, flags); 104 105 list_for_each_entry_safe(neigh, tmp, &mcast->neigh_list, list) { 106 /* 107 * It's safe to call ipoib_put_ah() inside priv->lock 108 * here, because we know that mcast->ah will always 109 * hold one more reference, so ipoib_put_ah() will 110 * never do more than decrement the ref count. 111 */ 112 if (neigh->ah) 113 ipoib_put_ah(neigh->ah); 114 ipoib_neigh_free(dev, neigh); 115 } 116 117 spin_unlock_irqrestore(&priv->lock, flags); 118 119 if (mcast->ah) 120 ipoib_put_ah(mcast->ah); 121 122 while (!skb_queue_empty(&mcast->pkt_queue)) { 123 ++tx_dropped; 124 dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue)); 125 } 126 127 spin_lock_irqsave(&priv->tx_lock, flags); 128 priv->stats.tx_dropped += tx_dropped; 129 spin_unlock_irqrestore(&priv->tx_lock, flags); 130 131 kfree(mcast); 132} 133 134static struct ipoib_mcast *ipoib_mcast_alloc(struct net_device *dev, 135 int can_sleep) 136{ 137 struct ipoib_mcast *mcast; 138 139 mcast = kzalloc(sizeof *mcast, can_sleep ? GFP_KERNEL : GFP_ATOMIC); 140 if (!mcast) 141 return NULL; 142 143 mcast->dev = dev; 144 mcast->created = jiffies; 145 mcast->backoff = 1; 146 147 INIT_LIST_HEAD(&mcast->list); 148 INIT_LIST_HEAD(&mcast->neigh_list); 149 skb_queue_head_init(&mcast->pkt_queue); 150 151 return mcast; 152} 153 154static struct ipoib_mcast *__ipoib_mcast_find(struct net_device *dev, void *mgid) 155{ 156 struct ipoib_dev_priv *priv = netdev_priv(dev); 157 struct rb_node *n = priv->multicast_tree.rb_node; 158 159 while (n) { 160 struct ipoib_mcast *mcast; 161 int ret; 162 163 mcast = rb_entry(n, struct ipoib_mcast, rb_node); 164 165 ret = memcmp(mgid, mcast->mcmember.mgid.raw, 166 sizeof (union ib_gid)); 167 if (ret < 0) 168 n = n->rb_left; 169 else if (ret > 0) 170 n = n->rb_right; 171 else 172 return mcast; 173 } 174 175 return NULL; 176} 177 178static int __ipoib_mcast_add(struct net_device *dev, struct ipoib_mcast *mcast) 179{ 180 struct ipoib_dev_priv *priv = netdev_priv(dev); 181 struct rb_node **n = &priv->multicast_tree.rb_node, *pn = NULL; 182 183 while (*n) { 184 struct ipoib_mcast *tmcast; 185 int ret; 186 187 pn = *n; 188 tmcast = rb_entry(pn, struct ipoib_mcast, rb_node); 189 190 ret = memcmp(mcast->mcmember.mgid.raw, tmcast->mcmember.mgid.raw, 191 sizeof (union ib_gid)); 192 if (ret < 0) 193 n = &pn->rb_left; 194 else if (ret > 0) 195 n = &pn->rb_right; 196 else 197 return -EEXIST; 198 } 199 200 rb_link_node(&mcast->rb_node, pn, n); 201 rb_insert_color(&mcast->rb_node, &priv->multicast_tree); 202 203 return 0; 204} 205 206static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast, 207 struct ib_sa_mcmember_rec *mcmember) 208{ 209 struct net_device *dev = mcast->dev; 210 struct ipoib_dev_priv *priv = netdev_priv(dev); 211 struct ipoib_ah *ah; 212 int ret; 213 214 mcast->mcmember = *mcmember; 215 216 /* Set the cached Q_Key before we attach if it's the broadcast group */ 217 if (!memcmp(mcast->mcmember.mgid.raw, priv->dev->broadcast + 4, 218 sizeof (union ib_gid))) { 219 priv->qkey = be32_to_cpu(priv->broadcast->mcmember.qkey); 220 priv->tx_wr.wr.ud.remote_qkey = priv->qkey; 221 } 222 223 if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) { 224 if (test_and_set_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) { 225 ipoib_warn(priv, "multicast group " IPOIB_GID_FMT 226 " already attached\n", 227 IPOIB_GID_ARG(mcast->mcmember.mgid)); 228 229 return 0; 230 } 231 232 ret = ipoib_mcast_attach(dev, be16_to_cpu(mcast->mcmember.mlid), 233 &mcast->mcmember.mgid); 234 if (ret < 0) { 235 ipoib_warn(priv, "couldn't attach QP to multicast group " 236 IPOIB_GID_FMT "\n", 237 IPOIB_GID_ARG(mcast->mcmember.mgid)); 238 239 clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags); 240 return ret; 241 } 242 } 243 244 { 245 struct ib_ah_attr av = { 246 .dlid = be16_to_cpu(mcast->mcmember.mlid), 247 .port_num = priv->port, 248 .sl = mcast->mcmember.sl, 249 .ah_flags = IB_AH_GRH, 250 .static_rate = mcast->mcmember.rate, 251 .grh = { 252 .flow_label = be32_to_cpu(mcast->mcmember.flow_label), 253 .hop_limit = mcast->mcmember.hop_limit, 254 .sgid_index = 0, 255 .traffic_class = mcast->mcmember.traffic_class 256 } 257 }; 258 av.grh.dgid = mcast->mcmember.mgid; 259 260 ah = ipoib_create_ah(dev, priv->pd, &av); 261 if (!ah) { 262 ipoib_warn(priv, "ib_address_create failed\n"); 263 } else { 264 spin_lock_irq(&priv->lock); 265 mcast->ah = ah; 266 spin_unlock_irq(&priv->lock); 267 268 ipoib_dbg_mcast(priv, "MGID " IPOIB_GID_FMT 269 " AV %p, LID 0x%04x, SL %d\n", 270 IPOIB_GID_ARG(mcast->mcmember.mgid), 271 mcast->ah->ah, 272 be16_to_cpu(mcast->mcmember.mlid), 273 mcast->mcmember.sl); 274 } 275 } 276 277 /* actually send any queued packets */ 278 spin_lock_irq(&priv->tx_lock); 279 while (!skb_queue_empty(&mcast->pkt_queue)) { 280 struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue); 281 spin_unlock_irq(&priv->tx_lock); 282 283 skb->dev = dev; 284 285 if (!skb->dst || !skb->dst->neighbour) { 286 /* put pseudoheader back on for next time */ 287 skb_push(skb, sizeof (struct ipoib_pseudoheader)); 288 } 289 290 if (dev_queue_xmit(skb)) 291 ipoib_warn(priv, "dev_queue_xmit failed to requeue packet\n"); 292 spin_lock_irq(&priv->tx_lock); 293 } 294 spin_unlock_irq(&priv->tx_lock); 295 296 return 0; 297} 298 299static int 300ipoib_mcast_sendonly_join_complete(int status, 301 struct ib_sa_multicast *multicast) 302{ 303 struct ipoib_mcast *mcast = multicast->context; 304 struct net_device *dev = mcast->dev; 305 struct ipoib_dev_priv *priv = netdev_priv(dev); 306 307 /* We trap for port events ourselves. */ 308 if (status == -ENETRESET) 309 return 0; 310 311 if (!status) 312 status = ipoib_mcast_join_finish(mcast, &multicast->rec); 313 314 if (status) { 315 if (mcast->logcount++ < 20) 316 ipoib_dbg_mcast(netdev_priv(dev), "multicast join failed for " 317 IPOIB_GID_FMT ", status %d\n", 318 IPOIB_GID_ARG(mcast->mcmember.mgid), status); 319 320 /* Flush out any queued packets */ 321 spin_lock_irq(&priv->tx_lock); 322 while (!skb_queue_empty(&mcast->pkt_queue)) { 323 ++priv->stats.tx_dropped; 324 dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue)); 325 } 326 spin_unlock_irq(&priv->tx_lock); 327 328 /* Clear the busy flag so we try again */ 329 status = test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, 330 &mcast->flags); 331 } 332 return status; 333} 334 335static int ipoib_mcast_sendonly_join(struct ipoib_mcast *mcast) 336{ 337 struct net_device *dev = mcast->dev; 338 struct ipoib_dev_priv *priv = netdev_priv(dev); 339 struct ib_sa_mcmember_rec rec = { 340 .join_state = 1 341 }; 342 int ret = 0; 343 344 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) { 345 ipoib_dbg_mcast(priv, "device shutting down, no multicast joins\n"); 346 return -ENODEV; 347 } 348 349 if (test_and_set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) { 350 ipoib_dbg_mcast(priv, "multicast entry busy, skipping\n"); 351 return -EBUSY; 352 } 353 354 rec.mgid = mcast->mcmember.mgid; 355 rec.port_gid = priv->local_gid; 356 rec.pkey = cpu_to_be16(priv->pkey); 357 358 mcast->mc = ib_sa_join_multicast(&ipoib_sa_client, priv->ca, 359 priv->port, &rec, 360 IB_SA_MCMEMBER_REC_MGID | 361 IB_SA_MCMEMBER_REC_PORT_GID | 362 IB_SA_MCMEMBER_REC_PKEY | 363 IB_SA_MCMEMBER_REC_JOIN_STATE, 364 GFP_ATOMIC, 365 ipoib_mcast_sendonly_join_complete, 366 mcast); 367 if (IS_ERR(mcast->mc)) { 368 ret = PTR_ERR(mcast->mc); 369 clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags); 370 ipoib_warn(priv, "ib_sa_join_multicast failed (ret = %d)\n", 371 ret); 372 } else { 373 ipoib_dbg_mcast(priv, "no multicast record for " IPOIB_GID_FMT 374 ", starting join\n", 375 IPOIB_GID_ARG(mcast->mcmember.mgid)); 376 } 377 378 return ret; 379} 380 381static int ipoib_mcast_join_complete(int status, 382 struct ib_sa_multicast *multicast) 383{ 384 struct ipoib_mcast *mcast = multicast->context; 385 struct net_device *dev = mcast->dev; 386 struct ipoib_dev_priv *priv = netdev_priv(dev); 387 388 ipoib_dbg_mcast(priv, "join completion for " IPOIB_GID_FMT 389 " (status %d)\n", 390 IPOIB_GID_ARG(mcast->mcmember.mgid), status); 391 392 /* We trap for port events ourselves. */ 393 if (status == -ENETRESET) 394 return 0; 395 396 if (!status) 397 status = ipoib_mcast_join_finish(mcast, &multicast->rec); 398 399 if (!status) { 400 mcast->backoff = 1; 401 mutex_lock(&mcast_mutex); 402 if (test_bit(IPOIB_MCAST_RUN, &priv->flags)) 403 queue_delayed_work(ipoib_workqueue, 404 &priv->mcast_task, 0); 405 mutex_unlock(&mcast_mutex); 406 407 if (mcast == priv->broadcast) 408 netif_carrier_on(dev); 409 410 return 0; 411 } 412 413 if (mcast->logcount++ < 20) { 414 if (status == -ETIMEDOUT) { 415 ipoib_dbg_mcast(priv, "multicast join failed for " IPOIB_GID_FMT 416 ", status %d\n", 417 IPOIB_GID_ARG(mcast->mcmember.mgid), 418 status); 419 } else { 420 ipoib_warn(priv, "multicast join failed for " 421 IPOIB_GID_FMT ", status %d\n", 422 IPOIB_GID_ARG(mcast->mcmember.mgid), 423 status); 424 } 425 } 426 427 mcast->backoff *= 2; 428 if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS) 429 mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS; 430 431 /* Clear the busy flag so we try again */ 432 status = test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags); 433 434 mutex_lock(&mcast_mutex); 435 spin_lock_irq(&priv->lock); 436 if (test_bit(IPOIB_MCAST_RUN, &priv->flags)) 437 queue_delayed_work(ipoib_workqueue, &priv->mcast_task, 438 mcast->backoff * HZ); 439 spin_unlock_irq(&priv->lock); 440 mutex_unlock(&mcast_mutex); 441 442 return status; 443} 444 445static void ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast, 446 int create) 447{ 448 struct ipoib_dev_priv *priv = netdev_priv(dev); 449 struct ib_sa_mcmember_rec rec = { 450 .join_state = 1 451 }; 452 ib_sa_comp_mask comp_mask; 453 int ret = 0; 454 455 ipoib_dbg_mcast(priv, "joining MGID " IPOIB_GID_FMT "\n", 456 IPOIB_GID_ARG(mcast->mcmember.mgid)); 457 458 rec.mgid = mcast->mcmember.mgid; 459 rec.port_gid = priv->local_gid; 460 rec.pkey = cpu_to_be16(priv->pkey); 461 462 comp_mask = 463 IB_SA_MCMEMBER_REC_MGID | 464 IB_SA_MCMEMBER_REC_PORT_GID | 465 IB_SA_MCMEMBER_REC_PKEY | 466 IB_SA_MCMEMBER_REC_JOIN_STATE; 467 468 if (create) { 469 comp_mask |= 470 IB_SA_MCMEMBER_REC_QKEY | 471 IB_SA_MCMEMBER_REC_MTU_SELECTOR | 472 IB_SA_MCMEMBER_REC_MTU | 473 IB_SA_MCMEMBER_REC_TRAFFIC_CLASS | 474 IB_SA_MCMEMBER_REC_RATE_SELECTOR | 475 IB_SA_MCMEMBER_REC_RATE | 476 IB_SA_MCMEMBER_REC_SL | 477 IB_SA_MCMEMBER_REC_FLOW_LABEL | 478 IB_SA_MCMEMBER_REC_HOP_LIMIT; 479 480 rec.qkey = priv->broadcast->mcmember.qkey; 481 rec.mtu_selector = IB_SA_EQ; 482 rec.mtu = priv->broadcast->mcmember.mtu; 483 rec.traffic_class = priv->broadcast->mcmember.traffic_class; 484 rec.rate_selector = IB_SA_EQ; 485 rec.rate = priv->broadcast->mcmember.rate; 486 rec.sl = priv->broadcast->mcmember.sl; 487 rec.flow_label = priv->broadcast->mcmember.flow_label; 488 rec.hop_limit = priv->broadcast->mcmember.hop_limit; 489 } 490 491 set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags); 492 mcast->mc = ib_sa_join_multicast(&ipoib_sa_client, priv->ca, priv->port, 493 &rec, comp_mask, GFP_KERNEL, 494 ipoib_mcast_join_complete, mcast); 495 if (IS_ERR(mcast->mc)) { 496 clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags); 497 ret = PTR_ERR(mcast->mc); 498 ipoib_warn(priv, "ib_sa_join_multicast failed, status %d\n", ret); 499 500 mcast->backoff *= 2; 501 if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS) 502 mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS; 503 504 mutex_lock(&mcast_mutex); 505 if (test_bit(IPOIB_MCAST_RUN, &priv->flags)) 506 queue_delayed_work(ipoib_workqueue, 507 &priv->mcast_task, 508 mcast->backoff * HZ); 509 mutex_unlock(&mcast_mutex); 510 } 511} 512 513void ipoib_mcast_join_task(struct work_struct *work) 514{ 515 struct ipoib_dev_priv *priv = 516 container_of(work, struct ipoib_dev_priv, mcast_task.work); 517 struct net_device *dev = priv->dev; 518 519 if (!test_bit(IPOIB_MCAST_RUN, &priv->flags)) 520 return; 521 522 if (ib_query_gid(priv->ca, priv->port, 0, &priv->local_gid)) 523 ipoib_warn(priv, "ib_query_gid() failed\n"); 524 else 525 memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw, sizeof (union ib_gid)); 526 527 { 528 struct ib_port_attr attr; 529 530 if (!ib_query_port(priv->ca, priv->port, &attr)) 531 priv->local_lid = attr.lid; 532 else 533 ipoib_warn(priv, "ib_query_port failed\n"); 534 } 535 536 if (!priv->broadcast) { 537 struct ipoib_mcast *broadcast; 538 539 broadcast = ipoib_mcast_alloc(dev, 1); 540 if (!broadcast) { 541 ipoib_warn(priv, "failed to allocate broadcast group\n"); 542 mutex_lock(&mcast_mutex); 543 if (test_bit(IPOIB_MCAST_RUN, &priv->flags)) 544 queue_delayed_work(ipoib_workqueue, 545 &priv->mcast_task, HZ); 546 mutex_unlock(&mcast_mutex); 547 return; 548 } 549 550 spin_lock_irq(&priv->lock); 551 memcpy(broadcast->mcmember.mgid.raw, priv->dev->broadcast + 4, 552 sizeof (union ib_gid)); 553 priv->broadcast = broadcast; 554 555 __ipoib_mcast_add(dev, priv->broadcast); 556 spin_unlock_irq(&priv->lock); 557 } 558 559 if (!test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) { 560 if (!test_bit(IPOIB_MCAST_FLAG_BUSY, &priv->broadcast->flags)) 561 ipoib_mcast_join(dev, priv->broadcast, 0); 562 return; 563 } 564 565 while (1) { 566 struct ipoib_mcast *mcast = NULL; 567 568 spin_lock_irq(&priv->lock); 569 list_for_each_entry(mcast, &priv->multicast_list, list) { 570 if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) 571 && !test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags) 572 && !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) { 573 /* Found the next unjoined group */ 574 break; 575 } 576 } 577 spin_unlock_irq(&priv->lock); 578 579 if (&mcast->list == &priv->multicast_list) { 580 /* All done */ 581 break; 582 } 583 584 ipoib_mcast_join(dev, mcast, 1); 585 return; 586 } 587 588 priv->mcast_mtu = ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu) - 589 IPOIB_ENCAP_LEN; 590 591 if (!ipoib_cm_admin_enabled(dev)) 592 dev->mtu = min(priv->mcast_mtu, priv->admin_mtu); 593 594 ipoib_dbg_mcast(priv, "successfully joined all multicast groups\n"); 595 596 clear_bit(IPOIB_MCAST_RUN, &priv->flags); 597} 598 599int ipoib_mcast_start_thread(struct net_device *dev) 600{ 601 struct ipoib_dev_priv *priv = netdev_priv(dev); 602 603 ipoib_dbg_mcast(priv, "starting multicast thread\n"); 604 605 mutex_lock(&mcast_mutex); 606 if (!test_and_set_bit(IPOIB_MCAST_RUN, &priv->flags)) 607 queue_delayed_work(ipoib_workqueue, &priv->mcast_task, 0); 608 mutex_unlock(&mcast_mutex); 609 610 spin_lock_irq(&priv->lock); 611 set_bit(IPOIB_MCAST_STARTED, &priv->flags); 612 spin_unlock_irq(&priv->lock); 613 614 return 0; 615} 616 617int ipoib_mcast_stop_thread(struct net_device *dev, int flush) 618{ 619 struct ipoib_dev_priv *priv = netdev_priv(dev); 620 621 ipoib_dbg_mcast(priv, "stopping multicast thread\n"); 622 623 spin_lock_irq(&priv->lock); 624 clear_bit(IPOIB_MCAST_STARTED, &priv->flags); 625 spin_unlock_irq(&priv->lock); 626 627 mutex_lock(&mcast_mutex); 628 clear_bit(IPOIB_MCAST_RUN, &priv->flags); 629 cancel_delayed_work(&priv->mcast_task); 630 mutex_unlock(&mcast_mutex); 631 632 if (flush) 633 flush_workqueue(ipoib_workqueue); 634 635 return 0; 636} 637 638static int ipoib_mcast_leave(struct net_device *dev, struct ipoib_mcast *mcast) 639{ 640 struct ipoib_dev_priv *priv = netdev_priv(dev); 641 int ret = 0; 642 643 if (test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) 644 ib_sa_free_multicast(mcast->mc); 645 646 if (test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) { 647 ipoib_dbg_mcast(priv, "leaving MGID " IPOIB_GID_FMT "\n", 648 IPOIB_GID_ARG(mcast->mcmember.mgid)); 649 650 /* Remove ourselves from the multicast group */ 651 ret = ipoib_mcast_detach(dev, be16_to_cpu(mcast->mcmember.mlid), 652 &mcast->mcmember.mgid); 653 if (ret) 654 ipoib_warn(priv, "ipoib_mcast_detach failed (result = %d)\n", ret); 655 } 656 657 return 0; 658} 659 660void ipoib_mcast_send(struct net_device *dev, void *mgid, struct sk_buff *skb) 661{ 662 struct ipoib_dev_priv *priv = netdev_priv(dev); 663 struct ipoib_mcast *mcast; 664 665 /* 666 * We can only be called from ipoib_start_xmit, so we're 667 * inside tx_lock -- no need to save/restore flags. 668 */ 669 spin_lock(&priv->lock); 670 671 if (!test_bit(IPOIB_MCAST_STARTED, &priv->flags) || 672 !priv->broadcast || 673 !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) { 674 ++priv->stats.tx_dropped; 675 dev_kfree_skb_any(skb); 676 goto unlock; 677 } 678 679 mcast = __ipoib_mcast_find(dev, mgid); 680 if (!mcast) { 681 /* Let's create a new send only group now */ 682 ipoib_dbg_mcast(priv, "setting up send only multicast group for " 683 IPOIB_GID_FMT "\n", IPOIB_GID_RAW_ARG(mgid)); 684 685 mcast = ipoib_mcast_alloc(dev, 0); 686 if (!mcast) { 687 ipoib_warn(priv, "unable to allocate memory for " 688 "multicast structure\n"); 689 ++priv->stats.tx_dropped; 690 dev_kfree_skb_any(skb); 691 goto out; 692 } 693 694 set_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags); 695 memcpy(mcast->mcmember.mgid.raw, mgid, sizeof (union ib_gid)); 696 __ipoib_mcast_add(dev, mcast); 697 list_add_tail(&mcast->list, &priv->multicast_list); 698 } 699 700 if (!mcast->ah) { 701 if (skb_queue_len(&mcast->pkt_queue) < IPOIB_MAX_MCAST_QUEUE) 702 skb_queue_tail(&mcast->pkt_queue, skb); 703 else { 704 ++priv->stats.tx_dropped; 705 dev_kfree_skb_any(skb); 706 } 707 708 if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) 709 ipoib_dbg_mcast(priv, "no address vector, " 710 "but multicast join already started\n"); 711 else if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) 712 ipoib_mcast_sendonly_join(mcast); 713 714 /* 715 * If lookup completes between here and out:, don't 716 * want to send packet twice. 717 */ 718 mcast = NULL; 719 } 720 721out: 722 if (mcast && mcast->ah) { 723 if (skb->dst && 724 skb->dst->neighbour && 725 !*to_ipoib_neigh(skb->dst->neighbour)) { 726 struct ipoib_neigh *neigh = ipoib_neigh_alloc(skb->dst->neighbour); 727 728 if (neigh) { 729 kref_get(&mcast->ah->ref); 730 neigh->ah = mcast->ah; 731 list_add_tail(&neigh->list, &mcast->neigh_list); 732 } 733 } 734 735 ipoib_send(dev, skb, mcast->ah, IB_MULTICAST_QPN); 736 } 737 738unlock: 739 spin_unlock(&priv->lock); 740} 741 742void ipoib_mcast_dev_flush(struct net_device *dev) 743{ 744 struct ipoib_dev_priv *priv = netdev_priv(dev); 745 LIST_HEAD(remove_list); 746 struct ipoib_mcast *mcast, *tmcast; 747 unsigned long flags; 748 749 ipoib_dbg_mcast(priv, "flushing multicast list\n"); 750 751 spin_lock_irqsave(&priv->lock, flags); 752 753 list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) { 754 list_del(&mcast->list); 755 rb_erase(&mcast->rb_node, &priv->multicast_tree); 756 list_add_tail(&mcast->list, &remove_list); 757 } 758 759 if (priv->broadcast) { 760 rb_erase(&priv->broadcast->rb_node, &priv->multicast_tree); 761 list_add_tail(&priv->broadcast->list, &remove_list); 762 priv->broadcast = NULL; 763 } 764 765 spin_unlock_irqrestore(&priv->lock, flags); 766 767 list_for_each_entry_safe(mcast, tmcast, &remove_list, list) { 768 ipoib_mcast_leave(dev, mcast); 769 ipoib_mcast_free(mcast); 770 } 771} 772 773void ipoib_mcast_restart_task(struct work_struct *work) 774{ 775 struct ipoib_dev_priv *priv = 776 container_of(work, struct ipoib_dev_priv, restart_task); 777 struct net_device *dev = priv->dev; 778 struct dev_mc_list *mclist; 779 struct ipoib_mcast *mcast, *tmcast; 780 LIST_HEAD(remove_list); 781 unsigned long flags; 782 783 ipoib_dbg_mcast(priv, "restarting multicast task\n"); 784 785 ipoib_mcast_stop_thread(dev, 0); 786 787 local_irq_save(flags); 788 netif_tx_lock(dev); 789 spin_lock(&priv->lock); 790 791 /* 792 * Unfortunately, the networking core only gives us a list of all of 793 * the multicast hardware addresses. We need to figure out which ones 794 * are new and which ones have been removed 795 */ 796 797 /* Clear out the found flag */ 798 list_for_each_entry(mcast, &priv->multicast_list, list) 799 clear_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags); 800 801 /* Mark all of the entries that are found or don't exist */ 802 for (mclist = dev->mc_list; mclist; mclist = mclist->next) { 803 union ib_gid mgid; 804 805 memcpy(mgid.raw, mclist->dmi_addr + 4, sizeof mgid); 806 807 /* Add in the P_Key */ 808 mgid.raw[4] = (priv->pkey >> 8) & 0xff; 809 mgid.raw[5] = priv->pkey & 0xff; 810 811 mcast = __ipoib_mcast_find(dev, &mgid); 812 if (!mcast || test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) { 813 struct ipoib_mcast *nmcast; 814 815 /* Not found or send-only group, let's add a new entry */ 816 ipoib_dbg_mcast(priv, "adding multicast entry for mgid " 817 IPOIB_GID_FMT "\n", IPOIB_GID_ARG(mgid)); 818 819 nmcast = ipoib_mcast_alloc(dev, 0); 820 if (!nmcast) { 821 ipoib_warn(priv, "unable to allocate memory for multicast structure\n"); 822 continue; 823 } 824 825 set_bit(IPOIB_MCAST_FLAG_FOUND, &nmcast->flags); 826 827 nmcast->mcmember.mgid = mgid; 828 829 if (mcast) { 830 /* Destroy the send only entry */ 831 list_move_tail(&mcast->list, &remove_list); 832 833 rb_replace_node(&mcast->rb_node, 834 &nmcast->rb_node, 835 &priv->multicast_tree); 836 } else 837 __ipoib_mcast_add(dev, nmcast); 838 839 list_add_tail(&nmcast->list, &priv->multicast_list); 840 } 841 842 if (mcast) 843 set_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags); 844 } 845 846 /* Remove all of the entries don't exist anymore */ 847 list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) { 848 if (!test_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags) && 849 !test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) { 850 ipoib_dbg_mcast(priv, "deleting multicast group " IPOIB_GID_FMT "\n", 851 IPOIB_GID_ARG(mcast->mcmember.mgid)); 852 853 rb_erase(&mcast->rb_node, &priv->multicast_tree); 854 855 /* Move to the remove list */ 856 list_move_tail(&mcast->list, &remove_list); 857 } 858 } 859 860 spin_unlock(&priv->lock); 861 netif_tx_unlock(dev); 862 local_irq_restore(flags); 863 864 /* We have to cancel outside of the spinlock */ 865 list_for_each_entry_safe(mcast, tmcast, &remove_list, list) { 866 ipoib_mcast_leave(mcast->dev, mcast); 867 ipoib_mcast_free(mcast); 868 } 869 870 if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) 871 ipoib_mcast_start_thread(dev); 872} 873 874#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG 875 876struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev) 877{ 878 struct ipoib_mcast_iter *iter; 879 880 iter = kmalloc(sizeof *iter, GFP_KERNEL); 881 if (!iter) 882 return NULL; 883 884 iter->dev = dev; 885 memset(iter->mgid.raw, 0, 16); 886 887 if (ipoib_mcast_iter_next(iter)) { 888 kfree(iter); 889 return NULL; 890 } 891 892 return iter; 893} 894 895int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter) 896{ 897 struct ipoib_dev_priv *priv = netdev_priv(iter->dev); 898 struct rb_node *n; 899 struct ipoib_mcast *mcast; 900 int ret = 1; 901 902 spin_lock_irq(&priv->lock); 903 904 n = rb_first(&priv->multicast_tree); 905 906 while (n) { 907 mcast = rb_entry(n, struct ipoib_mcast, rb_node); 908 909 if (memcmp(iter->mgid.raw, mcast->mcmember.mgid.raw, 910 sizeof (union ib_gid)) < 0) { 911 iter->mgid = mcast->mcmember.mgid; 912 iter->created = mcast->created; 913 iter->queuelen = skb_queue_len(&mcast->pkt_queue); 914 iter->complete = !!mcast->ah; 915 iter->send_only = !!(mcast->flags & (1 << IPOIB_MCAST_FLAG_SENDONLY)); 916 917 ret = 0; 918 919 break; 920 } 921 922 n = rb_next(n); 923 } 924 925 spin_unlock_irq(&priv->lock); 926 927 return ret; 928} 929 930void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter, 931 union ib_gid *mgid, 932 unsigned long *created, 933 unsigned int *queuelen, 934 unsigned int *complete, 935 unsigned int *send_only) 936{ 937 *mgid = iter->mgid; 938 *created = iter->created; 939 *queuelen = iter->queuelen; 940 *complete = iter->complete; 941 *send_only = iter->send_only; 942} 943 944#endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */ 945