1/* 2 * Copyright (c) 2004 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 <sys/cdefs.h> 36__FBSDID("$FreeBSD$"); 37 38#include "ipoib.h" 39 40static int ipoib_resolvemulti(struct ifnet *, struct sockaddr **, 41 struct sockaddr *); 42 43 44#include <linux/module.h> 45 46#include <linux/slab.h> 47#include <linux/kernel.h> 48#include <linux/vmalloc.h> 49 50#include <linux/if_arp.h> /* For ARPHRD_xxx */ 51#include <linux/if_vlan.h> 52#include <net/ip.h> 53#include <net/ipv6.h> 54 55#include <rdma/ib_cache.h> 56 57MODULE_AUTHOR("Roland Dreier"); 58MODULE_DESCRIPTION("IP-over-InfiniBand net driver"); 59MODULE_LICENSE("Dual BSD/GPL"); 60 61int ipoib_sendq_size = IPOIB_TX_RING_SIZE; 62int ipoib_recvq_size = IPOIB_RX_RING_SIZE; 63 64module_param_named(send_queue_size, ipoib_sendq_size, int, 0444); 65MODULE_PARM_DESC(send_queue_size, "Number of descriptors in send queue"); 66module_param_named(recv_queue_size, ipoib_recvq_size, int, 0444); 67MODULE_PARM_DESC(recv_queue_size, "Number of descriptors in receive queue"); 68 69#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG 70int ipoib_debug_level = 1; 71 72module_param_named(debug_level, ipoib_debug_level, int, 0644); 73MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0"); 74#endif 75 76struct ipoib_path_iter { 77 struct ipoib_dev_priv *priv; 78 struct ipoib_path path; 79}; 80 81static const u8 ipv4_bcast_addr[] = { 82 0x00, 0xff, 0xff, 0xff, 83 0xff, 0x12, 0x40, 0x1b, 0x00, 0x00, 0x00, 0x00, 84 0x00, 0x00, 0x00, 0x00, 0xff, 0xff, 0xff, 0xff 85}; 86 87struct workqueue_struct *ipoib_workqueue; 88 89struct ib_sa_client ipoib_sa_client; 90 91static void ipoib_add_one(struct ib_device *device); 92static void ipoib_remove_one(struct ib_device *device, void *client_data); 93static struct net_device *ipoib_get_net_dev_by_params( 94 struct ib_device *dev, u8 port, u16 pkey, 95 const union ib_gid *gid, const struct sockaddr *addr, 96 void *client_data); 97static void ipoib_start(struct ifnet *dev); 98static int ipoib_output(struct ifnet *ifp, struct mbuf *m, 99 const struct sockaddr *dst, struct route *ro); 100static int ipoib_ioctl(struct ifnet *ifp, u_long command, caddr_t data); 101static void ipoib_input(struct ifnet *ifp, struct mbuf *m); 102 103#define IPOIB_MTAP(_ifp, _m) \ 104do { \ 105 if (bpf_peers_present((_ifp)->if_bpf)) { \ 106 M_ASSERTVALID(_m); \ 107 ipoib_mtap_mb((_ifp), (_m)); \ 108 } \ 109} while (0) 110 111static struct unrhdr *ipoib_unrhdr; 112 113static void 114ipoib_unrhdr_init(void *arg) 115{ 116 117 ipoib_unrhdr = new_unrhdr(0, 65535, NULL); 118} 119SYSINIT(ipoib_unrhdr_init, SI_SUB_KLD - 1, SI_ORDER_ANY, ipoib_unrhdr_init, NULL); 120 121static void 122ipoib_unrhdr_uninit(void *arg) 123{ 124 125 if (ipoib_unrhdr != NULL) { 126 struct unrhdr *hdr; 127 128 hdr = ipoib_unrhdr; 129 ipoib_unrhdr = NULL; 130 131 delete_unrhdr(hdr); 132 } 133} 134SYSUNINIT(ipoib_unrhdr_uninit, SI_SUB_KLD - 1, SI_ORDER_ANY, ipoib_unrhdr_uninit, NULL); 135 136/* 137 * This is for clients that have an ipoib_header in the mbuf. 138 */ 139static void 140ipoib_mtap_mb(struct ifnet *ifp, struct mbuf *mb) 141{ 142 struct ipoib_header *ih; 143 struct ether_header eh; 144 145 ih = mtod(mb, struct ipoib_header *); 146 eh.ether_type = ih->proto; 147 bcopy(ih->hwaddr, &eh.ether_dhost, ETHER_ADDR_LEN); 148 bzero(&eh.ether_shost, ETHER_ADDR_LEN); 149 mb->m_data += sizeof(struct ipoib_header); 150 mb->m_len -= sizeof(struct ipoib_header); 151 bpf_mtap2(ifp->if_bpf, &eh, sizeof(eh), mb); 152 mb->m_data -= sizeof(struct ipoib_header); 153 mb->m_len += sizeof(struct ipoib_header); 154} 155 156void 157ipoib_mtap_proto(struct ifnet *ifp, struct mbuf *mb, uint16_t proto) 158{ 159 struct ether_header eh; 160 161 eh.ether_type = proto; 162 bzero(&eh.ether_shost, ETHER_ADDR_LEN); 163 bzero(&eh.ether_dhost, ETHER_ADDR_LEN); 164 bpf_mtap2(ifp->if_bpf, &eh, sizeof(eh), mb); 165} 166 167static struct ib_client ipoib_client = { 168 .name = "ipoib", 169 .add = ipoib_add_one, 170 .remove = ipoib_remove_one, 171 .get_net_dev_by_params = ipoib_get_net_dev_by_params, 172}; 173 174int 175ipoib_open(struct ipoib_dev_priv *priv) 176{ 177 struct ifnet *dev = priv->dev; 178 179 ipoib_dbg(priv, "bringing up interface\n"); 180 181 set_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags); 182 183 if (ipoib_pkey_dev_delay_open(priv)) 184 return 0; 185 186 if (ipoib_ib_dev_open(priv)) 187 goto err_disable; 188 189 if (ipoib_ib_dev_up(priv)) 190 goto err_stop; 191 192 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) { 193 struct ipoib_dev_priv *cpriv; 194 195 /* Bring up any child interfaces too */ 196 mutex_lock(&priv->vlan_mutex); 197 list_for_each_entry(cpriv, &priv->child_intfs, list) 198 if ((cpriv->dev->if_drv_flags & IFF_DRV_RUNNING) == 0) 199 ipoib_open(cpriv); 200 mutex_unlock(&priv->vlan_mutex); 201 } 202 dev->if_drv_flags |= IFF_DRV_RUNNING; 203 dev->if_drv_flags &= ~IFF_DRV_OACTIVE; 204 205 return 0; 206 207err_stop: 208 ipoib_ib_dev_stop(priv, 1); 209 210err_disable: 211 clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags); 212 213 return -EINVAL; 214} 215 216static void 217ipoib_init(void *arg) 218{ 219 struct ifnet *dev; 220 struct ipoib_dev_priv *priv; 221 222 priv = arg; 223 dev = priv->dev; 224 if ((dev->if_drv_flags & IFF_DRV_RUNNING) == 0) 225 ipoib_open(priv); 226 queue_work(ipoib_workqueue, &priv->flush_light); 227} 228 229 230static int 231ipoib_stop(struct ipoib_dev_priv *priv) 232{ 233 struct ifnet *dev = priv->dev; 234 235 ipoib_dbg(priv, "stopping interface\n"); 236 237 clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags); 238 239 dev->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); 240 241 ipoib_ib_dev_down(priv, 0); 242 ipoib_ib_dev_stop(priv, 0); 243 244 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) { 245 struct ipoib_dev_priv *cpriv; 246 247 /* Bring down any child interfaces too */ 248 mutex_lock(&priv->vlan_mutex); 249 list_for_each_entry(cpriv, &priv->child_intfs, list) 250 if ((cpriv->dev->if_drv_flags & IFF_DRV_RUNNING) != 0) 251 ipoib_stop(cpriv); 252 mutex_unlock(&priv->vlan_mutex); 253 } 254 255 return 0; 256} 257 258static int 259ipoib_propagate_ifnet_mtu(struct ipoib_dev_priv *priv, int new_mtu, 260 bool propagate) 261{ 262 struct ifnet *ifp; 263 struct ifreq ifr; 264 int error; 265 266 ifp = priv->dev; 267 if (ifp->if_mtu == new_mtu) 268 return (0); 269 if (propagate) { 270 strlcpy(ifr.ifr_name, if_name(ifp), IFNAMSIZ); 271 ifr.ifr_mtu = new_mtu; 272 CURVNET_SET(ifp->if_vnet); 273 error = ifhwioctl(SIOCSIFMTU, ifp, (caddr_t)&ifr, curthread); 274 CURVNET_RESTORE(); 275 } else { 276 ifp->if_mtu = new_mtu; 277 error = 0; 278 } 279 return (error); 280} 281 282int 283ipoib_change_mtu(struct ipoib_dev_priv *priv, int new_mtu, bool propagate) 284{ 285 int error, prev_admin_mtu; 286 287 /* dev->if_mtu > 2K ==> connected mode */ 288 if (ipoib_cm_admin_enabled(priv)) { 289 if (new_mtu > IPOIB_CM_MTU(ipoib_cm_max_mtu(priv))) 290 return -EINVAL; 291 292 if (new_mtu > priv->mcast_mtu) 293 ipoib_warn(priv, "mtu > %d will cause multicast packet drops.\n", 294 priv->mcast_mtu); 295 296 return (ipoib_propagate_ifnet_mtu(priv, new_mtu, propagate)); 297 } 298 299 if (new_mtu > IPOIB_UD_MTU(priv->max_ib_mtu)) 300 return -EINVAL; 301 302 prev_admin_mtu = priv->admin_mtu; 303 priv->admin_mtu = new_mtu; 304 error = ipoib_propagate_ifnet_mtu(priv, min(priv->mcast_mtu, 305 priv->admin_mtu), propagate); 306 if (error == 0) { 307 /* check for MTU change to avoid infinite loop */ 308 if (prev_admin_mtu != new_mtu) 309 queue_work(ipoib_workqueue, &priv->flush_light); 310 } else 311 priv->admin_mtu = prev_admin_mtu; 312 return (error); 313} 314 315static int 316ipoib_ioctl(struct ifnet *ifp, u_long command, caddr_t data) 317{ 318 struct ipoib_dev_priv *priv = ifp->if_softc; 319 struct ifaddr *ifa = (struct ifaddr *) data; 320 struct ifreq *ifr = (struct ifreq *) data; 321 int error = 0; 322 323 /* check if detaching */ 324 if (priv == NULL || priv->gone != 0) 325 return (ENXIO); 326 327 switch (command) { 328 case SIOCSIFFLAGS: 329 if (ifp->if_flags & IFF_UP) { 330 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 331 error = -ipoib_open(priv); 332 } else 333 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 334 ipoib_stop(priv); 335 break; 336 case SIOCADDMULTI: 337 case SIOCDELMULTI: 338 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 339 queue_work(ipoib_workqueue, &priv->restart_task); 340 break; 341 case SIOCSIFADDR: 342 ifp->if_flags |= IFF_UP; 343 344 switch (ifa->ifa_addr->sa_family) { 345#ifdef INET 346 case AF_INET: 347 ifp->if_init(ifp->if_softc); /* before arpwhohas */ 348 arp_ifinit(ifp, ifa); 349 break; 350#endif 351 default: 352 ifp->if_init(ifp->if_softc); 353 break; 354 } 355 break; 356 357 case SIOCGIFADDR: 358 bcopy(IF_LLADDR(ifp), &ifr->ifr_addr.sa_data[0], 359 INFINIBAND_ALEN); 360 break; 361 362 case SIOCSIFMTU: 363 /* 364 * Set the interface MTU. 365 */ 366 error = -ipoib_change_mtu(priv, ifr->ifr_mtu, false); 367 break; 368 default: 369 error = EINVAL; 370 break; 371 } 372 return (error); 373} 374 375 376static struct ipoib_path * 377__path_find(struct ipoib_dev_priv *priv, void *gid) 378{ 379 struct rb_node *n = priv->path_tree.rb_node; 380 struct ipoib_path *path; 381 int ret; 382 383 while (n) { 384 path = rb_entry(n, struct ipoib_path, rb_node); 385 386 ret = memcmp(gid, path->pathrec.dgid.raw, 387 sizeof (union ib_gid)); 388 389 if (ret < 0) 390 n = n->rb_left; 391 else if (ret > 0) 392 n = n->rb_right; 393 else 394 return path; 395 } 396 397 return NULL; 398} 399 400static int 401__path_add(struct ipoib_dev_priv *priv, struct ipoib_path *path) 402{ 403 struct rb_node **n = &priv->path_tree.rb_node; 404 struct rb_node *pn = NULL; 405 struct ipoib_path *tpath; 406 int ret; 407 408 while (*n) { 409 pn = *n; 410 tpath = rb_entry(pn, struct ipoib_path, rb_node); 411 412 ret = memcmp(path->pathrec.dgid.raw, tpath->pathrec.dgid.raw, 413 sizeof (union ib_gid)); 414 if (ret < 0) 415 n = &pn->rb_left; 416 else if (ret > 0) 417 n = &pn->rb_right; 418 else 419 return -EEXIST; 420 } 421 422 rb_link_node(&path->rb_node, pn, n); 423 rb_insert_color(&path->rb_node, &priv->path_tree); 424 425 list_add_tail(&path->list, &priv->path_list); 426 427 return 0; 428} 429 430void 431ipoib_path_free(struct ipoib_dev_priv *priv, struct ipoib_path *path) 432{ 433 434 _IF_DRAIN(&path->queue); 435 436 if (path->ah) 437 ipoib_put_ah(path->ah); 438 if (ipoib_cm_get(path)) 439 ipoib_cm_destroy_tx(ipoib_cm_get(path)); 440 441 kfree(path); 442} 443 444#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG 445 446struct ipoib_path_iter * 447ipoib_path_iter_init(struct ipoib_dev_priv *priv) 448{ 449 struct ipoib_path_iter *iter; 450 451 iter = kmalloc(sizeof *iter, GFP_KERNEL); 452 if (!iter) 453 return NULL; 454 455 iter->priv = priv; 456 memset(iter->path.pathrec.dgid.raw, 0, 16); 457 458 if (ipoib_path_iter_next(iter)) { 459 kfree(iter); 460 return NULL; 461 } 462 463 return iter; 464} 465 466int 467ipoib_path_iter_next(struct ipoib_path_iter *iter) 468{ 469 struct ipoib_dev_priv *priv = iter->priv; 470 struct rb_node *n; 471 struct ipoib_path *path; 472 int ret = 1; 473 474 spin_lock_irq(&priv->lock); 475 476 n = rb_first(&priv->path_tree); 477 478 while (n) { 479 path = rb_entry(n, struct ipoib_path, rb_node); 480 481 if (memcmp(iter->path.pathrec.dgid.raw, path->pathrec.dgid.raw, 482 sizeof (union ib_gid)) < 0) { 483 iter->path = *path; 484 ret = 0; 485 break; 486 } 487 488 n = rb_next(n); 489 } 490 491 spin_unlock_irq(&priv->lock); 492 493 return ret; 494} 495 496void 497ipoib_path_iter_read(struct ipoib_path_iter *iter, struct ipoib_path *path) 498{ 499 *path = iter->path; 500} 501 502#endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */ 503 504void 505ipoib_mark_paths_invalid(struct ipoib_dev_priv *priv) 506{ 507 struct ipoib_path *path, *tp; 508 509 spin_lock_irq(&priv->lock); 510 511 list_for_each_entry_safe(path, tp, &priv->path_list, list) { 512 ipoib_dbg(priv, "mark path LID 0x%04x GID %16D invalid\n", 513 be16_to_cpu(path->pathrec.dlid), 514 path->pathrec.dgid.raw, ":"); 515 path->valid = 0; 516 } 517 518 spin_unlock_irq(&priv->lock); 519} 520 521void 522ipoib_flush_paths(struct ipoib_dev_priv *priv) 523{ 524 struct ipoib_path *path, *tp; 525 LIST_HEAD(remove_list); 526 unsigned long flags; 527 528 spin_lock_irqsave(&priv->lock, flags); 529 530 list_splice_init(&priv->path_list, &remove_list); 531 532 list_for_each_entry(path, &remove_list, list) 533 rb_erase(&path->rb_node, &priv->path_tree); 534 535 list_for_each_entry_safe(path, tp, &remove_list, list) { 536 if (path->query) 537 ib_sa_cancel_query(path->query_id, path->query); 538 spin_unlock_irqrestore(&priv->lock, flags); 539 wait_for_completion(&path->done); 540 ipoib_path_free(priv, path); 541 spin_lock_irqsave(&priv->lock, flags); 542 } 543 544 spin_unlock_irqrestore(&priv->lock, flags); 545} 546 547static void 548path_rec_completion(int status, struct ib_sa_path_rec *pathrec, void *path_ptr) 549{ 550 struct ipoib_path *path = path_ptr; 551 struct ipoib_dev_priv *priv = path->priv; 552 struct ifnet *dev = priv->dev; 553 struct ipoib_ah *ah = NULL; 554 struct ipoib_ah *old_ah = NULL; 555 struct ifqueue mbqueue; 556 struct mbuf *mb; 557 unsigned long flags; 558 559 if (!status) 560 ipoib_dbg(priv, "PathRec LID 0x%04x for GID %16D\n", 561 be16_to_cpu(pathrec->dlid), pathrec->dgid.raw, ":"); 562 else 563 ipoib_dbg(priv, "PathRec status %d for GID %16D\n", 564 status, path->pathrec.dgid.raw, ":"); 565 566 bzero(&mbqueue, sizeof(mbqueue)); 567 568 if (!status) { 569 struct ib_ah_attr av; 570 571 if (!ib_init_ah_from_path(priv->ca, priv->port, pathrec, &av)) 572 ah = ipoib_create_ah(priv, priv->pd, &av); 573 } 574 575 spin_lock_irqsave(&priv->lock, flags); 576 577 if (ah) { 578 path->pathrec = *pathrec; 579 580 old_ah = path->ah; 581 path->ah = ah; 582 583 ipoib_dbg(priv, "created address handle %p for LID 0x%04x, SL %d\n", 584 ah, be16_to_cpu(pathrec->dlid), pathrec->sl); 585 586 for (;;) { 587 _IF_DEQUEUE(&path->queue, mb); 588 if (mb == NULL) 589 break; 590 _IF_ENQUEUE(&mbqueue, mb); 591 } 592 593#ifdef CONFIG_INFINIBAND_IPOIB_CM 594 if (ipoib_cm_enabled(priv, path->hwaddr) && !ipoib_cm_get(path)) 595 ipoib_cm_set(path, ipoib_cm_create_tx(priv, path)); 596#endif 597 598 path->valid = 1; 599 } 600 601 path->query = NULL; 602 complete(&path->done); 603 604 spin_unlock_irqrestore(&priv->lock, flags); 605 606 if (old_ah) 607 ipoib_put_ah(old_ah); 608 609 for (;;) { 610 _IF_DEQUEUE(&mbqueue, mb); 611 if (mb == NULL) 612 break; 613 mb->m_pkthdr.rcvif = dev; 614 if (dev->if_transmit(dev, mb)) 615 ipoib_warn(priv, "dev_queue_xmit failed " 616 "to requeue packet\n"); 617 } 618} 619 620static struct ipoib_path * 621path_rec_create(struct ipoib_dev_priv *priv, uint8_t *hwaddr) 622{ 623 struct ipoib_path *path; 624 625 if (!priv->broadcast) 626 return NULL; 627 628 path = kzalloc(sizeof *path, GFP_ATOMIC); 629 if (!path) 630 return NULL; 631 632 path->priv = priv; 633 634 bzero(&path->queue, sizeof(path->queue)); 635 636#ifdef CONFIG_INFINIBAND_IPOIB_CM 637 memcpy(&path->hwaddr, hwaddr, INFINIBAND_ALEN); 638#endif 639 memcpy(path->pathrec.dgid.raw, &hwaddr[4], sizeof (union ib_gid)); 640 path->pathrec.sgid = priv->local_gid; 641 path->pathrec.pkey = cpu_to_be16(priv->pkey); 642 path->pathrec.numb_path = 1; 643 path->pathrec.traffic_class = priv->broadcast->mcmember.traffic_class; 644 645 return path; 646} 647 648static int 649path_rec_start(struct ipoib_dev_priv *priv, struct ipoib_path *path) 650{ 651 struct ifnet *dev = priv->dev; 652 653 ib_sa_comp_mask comp_mask = IB_SA_PATH_REC_MTU_SELECTOR | IB_SA_PATH_REC_MTU; 654 struct ib_sa_path_rec p_rec; 655 656 p_rec = path->pathrec; 657 p_rec.mtu_selector = IB_SA_GT; 658 659 switch (roundup_pow_of_two(dev->if_mtu + IPOIB_ENCAP_LEN)) { 660 case 512: 661 p_rec.mtu = IB_MTU_256; 662 break; 663 case 1024: 664 p_rec.mtu = IB_MTU_512; 665 break; 666 case 2048: 667 p_rec.mtu = IB_MTU_1024; 668 break; 669 case 4096: 670 p_rec.mtu = IB_MTU_2048; 671 break; 672 default: 673 /* Wildcard everything */ 674 comp_mask = 0; 675 p_rec.mtu = 0; 676 p_rec.mtu_selector = 0; 677 } 678 679 ipoib_dbg(priv, "Start path record lookup for %16D MTU > %d\n", 680 p_rec.dgid.raw, ":", 681 comp_mask ? ib_mtu_enum_to_int(p_rec.mtu) : 0); 682 683 init_completion(&path->done); 684 685 path->query_id = 686 ib_sa_path_rec_get(&ipoib_sa_client, priv->ca, priv->port, 687 &p_rec, comp_mask | 688 IB_SA_PATH_REC_DGID | 689 IB_SA_PATH_REC_SGID | 690 IB_SA_PATH_REC_NUMB_PATH | 691 IB_SA_PATH_REC_TRAFFIC_CLASS | 692 IB_SA_PATH_REC_PKEY, 693 1000, GFP_ATOMIC, 694 path_rec_completion, 695 path, &path->query); 696 if (path->query_id < 0) { 697 ipoib_warn(priv, "ib_sa_path_rec_get failed: %d\n", path->query_id); 698 path->query = NULL; 699 complete(&path->done); 700 return path->query_id; 701 } 702 703 return 0; 704} 705 706static void 707ipoib_unicast_send(struct mbuf *mb, struct ipoib_dev_priv *priv, struct ipoib_header *eh) 708{ 709 struct ipoib_path *path; 710 711 path = __path_find(priv, eh->hwaddr + 4); 712 if (!path || !path->valid) { 713 int new_path = 0; 714 715 if (!path) { 716 path = path_rec_create(priv, eh->hwaddr); 717 new_path = 1; 718 } 719 if (path) { 720 if (_IF_QLEN(&path->queue) < IPOIB_MAX_PATH_REC_QUEUE) 721 _IF_ENQUEUE(&path->queue, mb); 722 else { 723 if_inc_counter(priv->dev, IFCOUNTER_OERRORS, 1); 724 m_freem(mb); 725 } 726 727 if (!path->query && path_rec_start(priv, path)) { 728 if (new_path) 729 ipoib_path_free(priv, path); 730 return; 731 } else 732 __path_add(priv, path); 733 } else { 734 if_inc_counter(priv->dev, IFCOUNTER_OERRORS, 1); 735 m_freem(mb); 736 } 737 738 return; 739 } 740 741 if (ipoib_cm_get(path) && ipoib_cm_up(path)) { 742 ipoib_cm_send(priv, mb, ipoib_cm_get(path)); 743 } else if (path->ah) { 744 ipoib_send(priv, mb, path->ah, IPOIB_QPN(eh->hwaddr)); 745 } else if ((path->query || !path_rec_start(priv, path)) && 746 path->queue.ifq_len < IPOIB_MAX_PATH_REC_QUEUE) { 747 _IF_ENQUEUE(&path->queue, mb); 748 } else { 749 if_inc_counter(priv->dev, IFCOUNTER_OERRORS, 1); 750 m_freem(mb); 751 } 752} 753 754static int 755ipoib_send_one(struct ipoib_dev_priv *priv, struct mbuf *mb) 756{ 757 struct ipoib_header *eh; 758 759 eh = mtod(mb, struct ipoib_header *); 760 if (IPOIB_IS_MULTICAST(eh->hwaddr)) { 761 /* Add in the P_Key for multicast*/ 762 eh->hwaddr[8] = (priv->pkey >> 8) & 0xff; 763 eh->hwaddr[9] = priv->pkey & 0xff; 764 765 ipoib_mcast_send(priv, eh->hwaddr + 4, mb); 766 } else 767 ipoib_unicast_send(mb, priv, eh); 768 769 return 0; 770} 771 772void 773ipoib_start_locked(struct ifnet *dev, struct ipoib_dev_priv *priv) 774{ 775 struct mbuf *mb; 776 777 assert_spin_locked(&priv->lock); 778 779 while (!IFQ_DRV_IS_EMPTY(&dev->if_snd) && 780 (dev->if_drv_flags & IFF_DRV_OACTIVE) == 0) { 781 IFQ_DRV_DEQUEUE(&dev->if_snd, mb); 782 if (mb == NULL) 783 break; 784 IPOIB_MTAP(dev, mb); 785 ipoib_send_one(priv, mb); 786 } 787} 788 789static void 790_ipoib_start(struct ifnet *dev, struct ipoib_dev_priv *priv) 791{ 792 793 if ((dev->if_drv_flags & (IFF_DRV_RUNNING|IFF_DRV_OACTIVE)) != 794 IFF_DRV_RUNNING) 795 return; 796 797 spin_lock(&priv->lock); 798 ipoib_start_locked(dev, priv); 799 spin_unlock(&priv->lock); 800} 801 802static void 803ipoib_start(struct ifnet *dev) 804{ 805 _ipoib_start(dev, dev->if_softc); 806} 807 808static void 809ipoib_vlan_start(struct ifnet *dev) 810{ 811 struct ipoib_dev_priv *priv; 812 struct mbuf *mb; 813 814 priv = VLAN_COOKIE(dev); 815 if (priv != NULL) 816 return _ipoib_start(dev, priv); 817 while (!IFQ_DRV_IS_EMPTY(&dev->if_snd)) { 818 IFQ_DRV_DEQUEUE(&dev->if_snd, mb); 819 if (mb == NULL) 820 break; 821 m_freem(mb); 822 if_inc_counter(dev, IFCOUNTER_OERRORS, 1); 823 } 824} 825 826int 827ipoib_dev_init(struct ipoib_dev_priv *priv, struct ib_device *ca, int port) 828{ 829 830 /* Allocate RX/TX "rings" to hold queued mbs */ 831 priv->rx_ring = kzalloc(ipoib_recvq_size * sizeof *priv->rx_ring, 832 GFP_KERNEL); 833 if (!priv->rx_ring) { 834 printk(KERN_WARNING "%s: failed to allocate RX ring (%d entries)\n", 835 ca->name, ipoib_recvq_size); 836 goto out; 837 } 838 839 priv->tx_ring = kzalloc(ipoib_sendq_size * sizeof *priv->tx_ring, GFP_KERNEL); 840 if (!priv->tx_ring) { 841 printk(KERN_WARNING "%s: failed to allocate TX ring (%d entries)\n", 842 ca->name, ipoib_sendq_size); 843 goto out_rx_ring_cleanup; 844 } 845 memset(priv->tx_ring, 0, ipoib_sendq_size * sizeof *priv->tx_ring); 846 847 /* priv->tx_head, tx_tail & tx_outstanding are already 0 */ 848 849 if (ipoib_ib_dev_init(priv, ca, port)) 850 goto out_tx_ring_cleanup; 851 852 return 0; 853 854out_tx_ring_cleanup: 855 kfree(priv->tx_ring); 856 857out_rx_ring_cleanup: 858 kfree(priv->rx_ring); 859 860out: 861 return -ENOMEM; 862} 863 864static void 865ipoib_detach(struct ipoib_dev_priv *priv) 866{ 867 struct ifnet *dev; 868 869 dev = priv->dev; 870 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) { 871 priv->gone = 1; 872 bpfdetach(dev); 873 if_detach(dev); 874 if_free(dev); 875 free_unr(ipoib_unrhdr, priv->unit); 876 } else 877 VLAN_SETCOOKIE(priv->dev, NULL); 878 879 free(priv, M_TEMP); 880} 881 882void 883ipoib_dev_cleanup(struct ipoib_dev_priv *priv) 884{ 885 struct ipoib_dev_priv *cpriv, *tcpriv; 886 887 /* Delete any child interfaces first */ 888 list_for_each_entry_safe(cpriv, tcpriv, &priv->child_intfs, list) { 889 ipoib_dev_cleanup(cpriv); 890 ipoib_detach(cpriv); 891 } 892 893 ipoib_ib_dev_cleanup(priv); 894 895 kfree(priv->rx_ring); 896 kfree(priv->tx_ring); 897 898 priv->rx_ring = NULL; 899 priv->tx_ring = NULL; 900} 901 902static struct ipoib_dev_priv * 903ipoib_priv_alloc(void) 904{ 905 struct ipoib_dev_priv *priv; 906 907 priv = malloc(sizeof(struct ipoib_dev_priv), M_TEMP, M_ZERO|M_WAITOK); 908 spin_lock_init(&priv->lock); 909 spin_lock_init(&priv->drain_lock); 910 mutex_init(&priv->vlan_mutex); 911 INIT_LIST_HEAD(&priv->path_list); 912 INIT_LIST_HEAD(&priv->child_intfs); 913 INIT_LIST_HEAD(&priv->dead_ahs); 914 INIT_LIST_HEAD(&priv->multicast_list); 915 INIT_DELAYED_WORK(&priv->pkey_poll_task, ipoib_pkey_poll); 916 INIT_DELAYED_WORK(&priv->mcast_task, ipoib_mcast_join_task); 917 INIT_WORK(&priv->carrier_on_task, ipoib_mcast_carrier_on_task); 918 INIT_WORK(&priv->flush_light, ipoib_ib_dev_flush_light); 919 INIT_WORK(&priv->flush_normal, ipoib_ib_dev_flush_normal); 920 INIT_WORK(&priv->flush_heavy, ipoib_ib_dev_flush_heavy); 921 INIT_WORK(&priv->restart_task, ipoib_mcast_restart_task); 922 INIT_DELAYED_WORK(&priv->ah_reap_task, ipoib_reap_ah); 923 memcpy(priv->broadcastaddr, ipv4_bcast_addr, INFINIBAND_ALEN); 924 925 return (priv); 926} 927 928struct ipoib_dev_priv * 929ipoib_intf_alloc(const char *name) 930{ 931 struct ipoib_dev_priv *priv; 932 struct sockaddr_dl *sdl; 933 struct ifnet *dev; 934 935 priv = ipoib_priv_alloc(); 936 dev = priv->dev = if_alloc(IFT_INFINIBAND); 937 if (!dev) { 938 free(priv, M_TEMP); 939 return NULL; 940 } 941 dev->if_softc = priv; 942 priv->unit = alloc_unr(ipoib_unrhdr); 943 if (priv->unit == -1) { 944 if_free(dev); 945 free(priv, M_TEMP); 946 return NULL; 947 } 948 if_initname(dev, name, priv->unit); 949 dev->if_flags = IFF_BROADCAST | IFF_MULTICAST; 950 dev->if_addrlen = INFINIBAND_ALEN; 951 dev->if_hdrlen = IPOIB_HEADER_LEN; 952 if_attach(dev); 953 dev->if_init = ipoib_init; 954 dev->if_ioctl = ipoib_ioctl; 955 dev->if_start = ipoib_start; 956 dev->if_output = ipoib_output; 957 dev->if_input = ipoib_input; 958 dev->if_resolvemulti = ipoib_resolvemulti; 959 dev->if_baudrate = IF_Gbps(10); 960 dev->if_broadcastaddr = priv->broadcastaddr; 961 dev->if_snd.ifq_maxlen = ipoib_sendq_size * 2; 962 sdl = (struct sockaddr_dl *)dev->if_addr->ifa_addr; 963 sdl->sdl_type = IFT_INFINIBAND; 964 sdl->sdl_alen = dev->if_addrlen; 965 priv->dev = dev; 966 if_link_state_change(dev, LINK_STATE_DOWN); 967 bpfattach(dev, DLT_EN10MB, ETHER_HDR_LEN); 968 969 return dev->if_softc; 970} 971 972int 973ipoib_set_dev_features(struct ipoib_dev_priv *priv, struct ib_device *hca) 974{ 975 struct ib_device_attr *device_attr = &hca->attrs; 976 977 priv->hca_caps = device_attr->device_cap_flags; 978 979 priv->dev->if_hwassist = 0; 980 priv->dev->if_capabilities = 0; 981 982#ifndef CONFIG_INFINIBAND_IPOIB_CM 983 if (priv->hca_caps & IB_DEVICE_UD_IP_CSUM) { 984 set_bit(IPOIB_FLAG_CSUM, &priv->flags); 985 priv->dev->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP; 986 priv->dev->if_capabilities = IFCAP_HWCSUM | IFCAP_VLAN_HWCSUM; 987 } 988 989#if 0 990 if (priv->dev->features & NETIF_F_SG && priv->hca_caps & IB_DEVICE_UD_TSO) { 991 priv->dev->if_capabilities |= IFCAP_TSO4; 992 priv->dev->if_hwassist |= CSUM_TSO; 993 } 994#endif 995#endif 996 priv->dev->if_capabilities |= 997 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_LINKSTATE; 998 priv->dev->if_capenable = priv->dev->if_capabilities; 999 1000 return 0; 1001} 1002 1003 1004static struct ifnet * 1005ipoib_add_port(const char *format, struct ib_device *hca, u8 port) 1006{ 1007 struct ipoib_dev_priv *priv; 1008 struct ib_port_attr attr; 1009 int result = -ENOMEM; 1010 1011 priv = ipoib_intf_alloc(format); 1012 if (!priv) 1013 goto alloc_mem_failed; 1014 1015 if (!ib_query_port(hca, port, &attr)) 1016 priv->max_ib_mtu = ib_mtu_enum_to_int(attr.max_mtu); 1017 else { 1018 printk(KERN_WARNING "%s: ib_query_port %d failed\n", 1019 hca->name, port); 1020 goto device_init_failed; 1021 } 1022 1023 /* MTU will be reset when mcast join happens */ 1024 priv->dev->if_mtu = IPOIB_UD_MTU(priv->max_ib_mtu); 1025 priv->mcast_mtu = priv->admin_mtu = priv->dev->if_mtu; 1026 1027 result = ib_query_pkey(hca, port, 0, &priv->pkey); 1028 if (result) { 1029 printk(KERN_WARNING "%s: ib_query_pkey port %d failed (ret = %d)\n", 1030 hca->name, port, result); 1031 goto device_init_failed; 1032 } 1033 1034 if (ipoib_set_dev_features(priv, hca)) 1035 goto device_init_failed; 1036 1037 /* 1038 * Set the full membership bit, so that we join the right 1039 * broadcast group, etc. 1040 */ 1041 priv->pkey |= 0x8000; 1042 1043 priv->broadcastaddr[8] = priv->pkey >> 8; 1044 priv->broadcastaddr[9] = priv->pkey & 0xff; 1045 1046 result = ib_query_gid(hca, port, 0, &priv->local_gid, NULL); 1047 if (result) { 1048 printk(KERN_WARNING "%s: ib_query_gid port %d failed (ret = %d)\n", 1049 hca->name, port, result); 1050 goto device_init_failed; 1051 } 1052 memcpy(IF_LLADDR(priv->dev) + 4, priv->local_gid.raw, sizeof (union ib_gid)); 1053 1054 result = ipoib_dev_init(priv, hca, port); 1055 if (result < 0) { 1056 printk(KERN_WARNING "%s: failed to initialize port %d (ret = %d)\n", 1057 hca->name, port, result); 1058 goto device_init_failed; 1059 } 1060 if (ipoib_cm_admin_enabled(priv)) 1061 priv->dev->if_mtu = IPOIB_CM_MTU(ipoib_cm_max_mtu(priv)); 1062 1063 INIT_IB_EVENT_HANDLER(&priv->event_handler, 1064 priv->ca, ipoib_event); 1065 result = ib_register_event_handler(&priv->event_handler); 1066 if (result < 0) { 1067 printk(KERN_WARNING "%s: ib_register_event_handler failed for " 1068 "port %d (ret = %d)\n", 1069 hca->name, port, result); 1070 goto event_failed; 1071 } 1072 if_printf(priv->dev, "Attached to %s port %d\n", hca->name, port); 1073 1074 return priv->dev; 1075 1076event_failed: 1077 ipoib_dev_cleanup(priv); 1078 1079device_init_failed: 1080 ipoib_detach(priv); 1081 1082alloc_mem_failed: 1083 return ERR_PTR(result); 1084} 1085 1086static void 1087ipoib_add_one(struct ib_device *device) 1088{ 1089 struct list_head *dev_list; 1090 struct ifnet *dev; 1091 struct ipoib_dev_priv *priv; 1092 int s, e, p; 1093 1094 if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB) 1095 return; 1096 1097 dev_list = kmalloc(sizeof *dev_list, GFP_KERNEL); 1098 if (!dev_list) 1099 return; 1100 1101 INIT_LIST_HEAD(dev_list); 1102 1103 if (device->node_type == RDMA_NODE_IB_SWITCH) { 1104 s = 0; 1105 e = 0; 1106 } else { 1107 s = 1; 1108 e = device->phys_port_cnt; 1109 } 1110 1111 for (p = s; p <= e; ++p) { 1112 if (rdma_port_get_link_layer(device, p) != IB_LINK_LAYER_INFINIBAND) 1113 continue; 1114 dev = ipoib_add_port("ib", device, p); 1115 if (!IS_ERR(dev)) { 1116 priv = dev->if_softc; 1117 list_add_tail(&priv->list, dev_list); 1118 } 1119 } 1120 1121 ib_set_client_data(device, &ipoib_client, dev_list); 1122} 1123 1124static void 1125ipoib_remove_one(struct ib_device *device, void *client_data) 1126{ 1127 struct ipoib_dev_priv *priv, *tmp; 1128 struct list_head *dev_list = client_data; 1129 1130 if (!dev_list) 1131 return; 1132 1133 if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB) 1134 return; 1135 1136 list_for_each_entry_safe(priv, tmp, dev_list, list) { 1137 if (rdma_port_get_link_layer(device, priv->port) != IB_LINK_LAYER_INFINIBAND) 1138 continue; 1139 1140 ipoib_stop(priv); 1141 1142 ib_unregister_event_handler(&priv->event_handler); 1143 1144 /* dev_change_flags(priv->dev, priv->dev->flags & ~IFF_UP); */ 1145 1146 flush_workqueue(ipoib_workqueue); 1147 1148 ipoib_dev_cleanup(priv); 1149 ipoib_detach(priv); 1150 } 1151 1152 kfree(dev_list); 1153} 1154 1155static int 1156ipoib_match_dev_addr(const struct sockaddr *addr, struct net_device *dev) 1157{ 1158 struct ifaddr *ifa; 1159 int retval = 0; 1160 1161 CURVNET_SET(dev->if_vnet); 1162 IF_ADDR_RLOCK(dev); 1163 TAILQ_FOREACH(ifa, &dev->if_addrhead, ifa_link) { 1164 if (ifa->ifa_addr == NULL || 1165 ifa->ifa_addr->sa_family != addr->sa_family || 1166 ifa->ifa_addr->sa_len != addr->sa_len) { 1167 continue; 1168 } 1169 if (memcmp(ifa->ifa_addr, addr, addr->sa_len) == 0) { 1170 retval = 1; 1171 break; 1172 } 1173 } 1174 IF_ADDR_RUNLOCK(dev); 1175 CURVNET_RESTORE(); 1176 1177 return (retval); 1178} 1179 1180/* 1181 * ipoib_match_gid_pkey_addr - returns the number of IPoIB netdevs on 1182 * top a given ipoib device matching a pkey_index and address, if one 1183 * exists. 1184 * 1185 * @found_net_dev: contains a matching net_device if the return value 1186 * >= 1, with a reference held. 1187 */ 1188static int 1189ipoib_match_gid_pkey_addr(struct ipoib_dev_priv *priv, 1190 const union ib_gid *gid, u16 pkey_index, const struct sockaddr *addr, 1191 struct net_device **found_net_dev) 1192{ 1193 struct ipoib_dev_priv *child_priv; 1194 int matches = 0; 1195 1196 if (priv->pkey_index == pkey_index && 1197 (!gid || !memcmp(gid, &priv->local_gid, sizeof(*gid)))) { 1198 if (addr == NULL || ipoib_match_dev_addr(addr, priv->dev) != 0) { 1199 if (*found_net_dev == NULL) { 1200 struct net_device *net_dev; 1201 1202 if (priv->parent != NULL) 1203 net_dev = priv->parent; 1204 else 1205 net_dev = priv->dev; 1206 *found_net_dev = net_dev; 1207 dev_hold(net_dev); 1208 } 1209 matches++; 1210 } 1211 } 1212 1213 /* Check child interfaces */ 1214 mutex_lock(&priv->vlan_mutex); 1215 list_for_each_entry(child_priv, &priv->child_intfs, list) { 1216 matches += ipoib_match_gid_pkey_addr(child_priv, gid, 1217 pkey_index, addr, found_net_dev); 1218 if (matches > 1) 1219 break; 1220 } 1221 mutex_unlock(&priv->vlan_mutex); 1222 1223 return matches; 1224} 1225 1226/* 1227 * __ipoib_get_net_dev_by_params - returns the number of matching 1228 * net_devs found (between 0 and 2). Also return the matching 1229 * net_device in the @net_dev parameter, holding a reference to the 1230 * net_device, if the number of matches >= 1 1231 */ 1232static int 1233__ipoib_get_net_dev_by_params(struct list_head *dev_list, u8 port, 1234 u16 pkey_index, const union ib_gid *gid, 1235 const struct sockaddr *addr, struct net_device **net_dev) 1236{ 1237 struct ipoib_dev_priv *priv; 1238 int matches = 0; 1239 1240 *net_dev = NULL; 1241 1242 list_for_each_entry(priv, dev_list, list) { 1243 if (priv->port != port) 1244 continue; 1245 1246 matches += ipoib_match_gid_pkey_addr(priv, gid, pkey_index, 1247 addr, net_dev); 1248 1249 if (matches > 1) 1250 break; 1251 } 1252 1253 return matches; 1254} 1255 1256static struct net_device * 1257ipoib_get_net_dev_by_params(struct ib_device *dev, u8 port, u16 pkey, 1258 const union ib_gid *gid, const struct sockaddr *addr, void *client_data) 1259{ 1260 struct net_device *net_dev; 1261 struct list_head *dev_list = client_data; 1262 u16 pkey_index; 1263 int matches; 1264 int ret; 1265 1266 if (!rdma_protocol_ib(dev, port)) 1267 return NULL; 1268 1269 ret = ib_find_cached_pkey(dev, port, pkey, &pkey_index); 1270 if (ret) 1271 return NULL; 1272 1273 if (!dev_list) 1274 return NULL; 1275 1276 /* See if we can find a unique device matching the L2 parameters */ 1277 matches = __ipoib_get_net_dev_by_params(dev_list, port, pkey_index, 1278 gid, NULL, &net_dev); 1279 1280 switch (matches) { 1281 case 0: 1282 return NULL; 1283 case 1: 1284 return net_dev; 1285 } 1286 1287 dev_put(net_dev); 1288 1289 /* Couldn't find a unique device with L2 parameters only. Use L3 1290 * address to uniquely match the net device */ 1291 matches = __ipoib_get_net_dev_by_params(dev_list, port, pkey_index, 1292 gid, addr, &net_dev); 1293 switch (matches) { 1294 case 0: 1295 return NULL; 1296 default: 1297 dev_warn_ratelimited(&dev->dev, 1298 "duplicate IP address detected\n"); 1299 /* Fall through */ 1300 case 1: 1301 return net_dev; 1302 } 1303} 1304 1305static void 1306ipoib_config_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 1307{ 1308 struct ipoib_dev_priv *parent; 1309 struct ipoib_dev_priv *priv; 1310 struct ifnet *dev; 1311 uint16_t pkey; 1312 int error; 1313 1314 if (ifp->if_type != IFT_INFINIBAND) 1315 return; 1316 dev = VLAN_DEVAT(ifp, vtag); 1317 if (dev == NULL) 1318 return; 1319 priv = NULL; 1320 error = 0; 1321 parent = ifp->if_softc; 1322 /* We only support 15 bits of pkey. */ 1323 if (vtag & 0x8000) 1324 return; 1325 pkey = vtag | 0x8000; /* Set full membership bit. */ 1326 if (pkey == parent->pkey) 1327 return; 1328 /* Check for dups */ 1329 mutex_lock(&parent->vlan_mutex); 1330 list_for_each_entry(priv, &parent->child_intfs, list) { 1331 if (priv->pkey == pkey) { 1332 priv = NULL; 1333 error = EBUSY; 1334 goto out; 1335 } 1336 } 1337 priv = ipoib_priv_alloc(); 1338 priv->dev = dev; 1339 priv->max_ib_mtu = parent->max_ib_mtu; 1340 priv->mcast_mtu = priv->admin_mtu = parent->dev->if_mtu; 1341 set_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags); 1342 error = ipoib_set_dev_features(priv, parent->ca); 1343 if (error) 1344 goto out; 1345 priv->pkey = pkey; 1346 priv->broadcastaddr[8] = pkey >> 8; 1347 priv->broadcastaddr[9] = pkey & 0xff; 1348 dev->if_broadcastaddr = priv->broadcastaddr; 1349 error = ipoib_dev_init(priv, parent->ca, parent->port); 1350 if (error) 1351 goto out; 1352 priv->parent = parent->dev; 1353 list_add_tail(&priv->list, &parent->child_intfs); 1354 VLAN_SETCOOKIE(dev, priv); 1355 dev->if_start = ipoib_vlan_start; 1356 dev->if_drv_flags &= ~IFF_DRV_RUNNING; 1357 dev->if_hdrlen = IPOIB_HEADER_LEN; 1358 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 1359 ipoib_open(priv); 1360 mutex_unlock(&parent->vlan_mutex); 1361 return; 1362out: 1363 mutex_unlock(&parent->vlan_mutex); 1364 if (priv) 1365 free(priv, M_TEMP); 1366 if (error) 1367 ipoib_warn(parent, 1368 "failed to initialize subinterface: device %s, port %d vtag 0x%X", 1369 parent->ca->name, parent->port, vtag); 1370 return; 1371} 1372 1373static void 1374ipoib_unconfig_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 1375{ 1376 struct ipoib_dev_priv *parent; 1377 struct ipoib_dev_priv *priv; 1378 struct ifnet *dev; 1379 uint16_t pkey; 1380 1381 if (ifp->if_type != IFT_INFINIBAND) 1382 return; 1383 1384 dev = VLAN_DEVAT(ifp, vtag); 1385 if (dev) 1386 VLAN_SETCOOKIE(dev, NULL); 1387 pkey = vtag | 0x8000; 1388 parent = ifp->if_softc; 1389 mutex_lock(&parent->vlan_mutex); 1390 list_for_each_entry(priv, &parent->child_intfs, list) { 1391 if (priv->pkey == pkey) { 1392 ipoib_dev_cleanup(priv); 1393 list_del(&priv->list); 1394 break; 1395 } 1396 } 1397 mutex_unlock(&parent->vlan_mutex); 1398} 1399 1400eventhandler_tag ipoib_vlan_attach; 1401eventhandler_tag ipoib_vlan_detach; 1402 1403static int __init 1404ipoib_init_module(void) 1405{ 1406 int ret; 1407 1408 ipoib_recvq_size = roundup_pow_of_two(ipoib_recvq_size); 1409 ipoib_recvq_size = min(ipoib_recvq_size, IPOIB_MAX_QUEUE_SIZE); 1410 ipoib_recvq_size = max(ipoib_recvq_size, IPOIB_MIN_QUEUE_SIZE); 1411 1412 ipoib_sendq_size = roundup_pow_of_two(ipoib_sendq_size); 1413 ipoib_sendq_size = min(ipoib_sendq_size, IPOIB_MAX_QUEUE_SIZE); 1414 ipoib_sendq_size = max(ipoib_sendq_size, max(2 * MAX_SEND_CQE, 1415 IPOIB_MIN_QUEUE_SIZE)); 1416#ifdef CONFIG_INFINIBAND_IPOIB_CM 1417 ipoib_max_conn_qp = min(ipoib_max_conn_qp, IPOIB_CM_MAX_CONN_QP); 1418#endif 1419 1420 ipoib_vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 1421 ipoib_config_vlan, NULL, EVENTHANDLER_PRI_FIRST); 1422 ipoib_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 1423 ipoib_unconfig_vlan, NULL, EVENTHANDLER_PRI_FIRST); 1424 1425 /* 1426 * We create our own workqueue mainly because we want to be 1427 * able to flush it when devices are being removed. We can't 1428 * use schedule_work()/flush_scheduled_work() because both 1429 * unregister_netdev() and linkwatch_event take the rtnl lock, 1430 * so flush_scheduled_work() can deadlock during device 1431 * removal. 1432 */ 1433 ipoib_workqueue = create_singlethread_workqueue("ipoib"); 1434 if (!ipoib_workqueue) { 1435 ret = -ENOMEM; 1436 goto err_fs; 1437 } 1438 1439 ib_sa_register_client(&ipoib_sa_client); 1440 1441 ret = ib_register_client(&ipoib_client); 1442 if (ret) 1443 goto err_sa; 1444 1445 return 0; 1446 1447err_sa: 1448 ib_sa_unregister_client(&ipoib_sa_client); 1449 destroy_workqueue(ipoib_workqueue); 1450 1451err_fs: 1452 return ret; 1453} 1454 1455static void __exit 1456ipoib_cleanup_module(void) 1457{ 1458 1459 EVENTHANDLER_DEREGISTER(vlan_config, ipoib_vlan_attach); 1460 EVENTHANDLER_DEREGISTER(vlan_unconfig, ipoib_vlan_detach); 1461 ib_unregister_client(&ipoib_client); 1462 ib_sa_unregister_client(&ipoib_sa_client); 1463 destroy_workqueue(ipoib_workqueue); 1464} 1465 1466/* 1467 * Infiniband output routine. 1468 */ 1469static int 1470ipoib_output(struct ifnet *ifp, struct mbuf *m, 1471 const struct sockaddr *dst, struct route *ro) 1472{ 1473 u_char edst[INFINIBAND_ALEN]; 1474#if defined(INET) || defined(INET6) 1475 struct llentry *lle = NULL; 1476#endif 1477 struct ipoib_header *eh; 1478 int error = 0, is_gw = 0; 1479 short type; 1480 1481 if (ro != NULL) 1482 is_gw = (ro->ro_flags & RT_HAS_GW) != 0; 1483#ifdef MAC 1484 error = mac_ifnet_check_transmit(ifp, m); 1485 if (error) 1486 goto bad; 1487#endif 1488 1489 M_PROFILE(m); 1490 if (ifp->if_flags & IFF_MONITOR) { 1491 error = ENETDOWN; 1492 goto bad; 1493 } 1494 if (!((ifp->if_flags & IFF_UP) && 1495 (ifp->if_drv_flags & IFF_DRV_RUNNING))) { 1496 error = ENETDOWN; 1497 goto bad; 1498 } 1499 1500 switch (dst->sa_family) { 1501#ifdef INET 1502 case AF_INET: 1503 if (lle != NULL && (lle->la_flags & LLE_VALID)) 1504 memcpy(edst, lle->ll_addr, sizeof(edst)); 1505 else if (m->m_flags & M_MCAST) 1506 ip_ib_mc_map(((struct sockaddr_in *)dst)->sin_addr.s_addr, ifp->if_broadcastaddr, edst); 1507 else 1508 error = arpresolve(ifp, is_gw, m, dst, edst, NULL, NULL); 1509 if (error) 1510 return (error == EWOULDBLOCK ? 0 : error); 1511 type = htons(ETHERTYPE_IP); 1512 break; 1513 case AF_ARP: 1514 { 1515 struct arphdr *ah; 1516 ah = mtod(m, struct arphdr *); 1517 ah->ar_hrd = htons(ARPHRD_INFINIBAND); 1518 1519 switch(ntohs(ah->ar_op)) { 1520 case ARPOP_REVREQUEST: 1521 case ARPOP_REVREPLY: 1522 type = htons(ETHERTYPE_REVARP); 1523 break; 1524 case ARPOP_REQUEST: 1525 case ARPOP_REPLY: 1526 default: 1527 type = htons(ETHERTYPE_ARP); 1528 break; 1529 } 1530 1531 if (m->m_flags & M_BCAST) 1532 bcopy(ifp->if_broadcastaddr, edst, INFINIBAND_ALEN); 1533 else 1534 bcopy(ar_tha(ah), edst, INFINIBAND_ALEN); 1535 1536 } 1537 break; 1538#endif 1539#ifdef INET6 1540 case AF_INET6: 1541 if (lle != NULL && (lle->la_flags & LLE_VALID)) 1542 memcpy(edst, lle->ll_addr, sizeof(edst)); 1543 else if (m->m_flags & M_MCAST) 1544 ipv6_ib_mc_map(&((struct sockaddr_in6 *)dst)->sin6_addr, ifp->if_broadcastaddr, edst); 1545 else 1546 error = nd6_resolve(ifp, is_gw, m, dst, edst, NULL, NULL); 1547 if (error) 1548 return error; 1549 type = htons(ETHERTYPE_IPV6); 1550 break; 1551#endif 1552 1553 default: 1554 if_printf(ifp, "can't handle af%d\n", dst->sa_family); 1555 error = EAFNOSUPPORT; 1556 goto bad; 1557 } 1558 1559 /* 1560 * Add local net header. If no space in first mbuf, 1561 * allocate another. 1562 */ 1563 M_PREPEND(m, IPOIB_HEADER_LEN, M_NOWAIT); 1564 if (m == NULL) { 1565 error = ENOBUFS; 1566 goto bad; 1567 } 1568 eh = mtod(m, struct ipoib_header *); 1569 (void)memcpy(&eh->proto, &type, sizeof(eh->proto)); 1570 (void)memcpy(&eh->hwaddr, edst, sizeof (edst)); 1571 1572 /* 1573 * Queue message on interface, update output statistics if 1574 * successful, and start output if interface not yet active. 1575 */ 1576 return ((ifp->if_transmit)(ifp, m)); 1577bad: 1578 if (m != NULL) 1579 m_freem(m); 1580 return (error); 1581} 1582 1583/* 1584 * Upper layer processing for a received Infiniband packet. 1585 */ 1586void 1587ipoib_demux(struct ifnet *ifp, struct mbuf *m, u_short proto) 1588{ 1589 int isr; 1590 1591#ifdef MAC 1592 /* 1593 * Tag the mbuf with an appropriate MAC label before any other 1594 * consumers can get to it. 1595 */ 1596 mac_ifnet_create_mbuf(ifp, m); 1597#endif 1598 /* Allow monitor mode to claim this frame, after stats are updated. */ 1599 if (ifp->if_flags & IFF_MONITOR) { 1600 if_printf(ifp, "discard frame at IFF_MONITOR\n"); 1601 m_freem(m); 1602 return; 1603 } 1604 /* Direct packet to correct FIB based on interface config */ 1605 M_SETFIB(m, ifp->if_fib); 1606 /* 1607 * Dispatch frame to upper layer. 1608 */ 1609 switch (proto) { 1610#ifdef INET 1611 case ETHERTYPE_IP: 1612 isr = NETISR_IP; 1613 break; 1614 1615 case ETHERTYPE_ARP: 1616 if (ifp->if_flags & IFF_NOARP) { 1617 /* Discard packet if ARP is disabled on interface */ 1618 m_freem(m); 1619 return; 1620 } 1621 isr = NETISR_ARP; 1622 break; 1623#endif 1624#ifdef INET6 1625 case ETHERTYPE_IPV6: 1626 isr = NETISR_IPV6; 1627 break; 1628#endif 1629 default: 1630 goto discard; 1631 } 1632 netisr_dispatch(isr, m); 1633 return; 1634 1635discard: 1636 m_freem(m); 1637} 1638 1639/* 1640 * Process a received Infiniband packet. 1641 */ 1642static void 1643ipoib_input(struct ifnet *ifp, struct mbuf *m) 1644{ 1645 struct ipoib_header *eh; 1646 1647 if ((ifp->if_flags & IFF_UP) == 0) { 1648 m_freem(m); 1649 return; 1650 } 1651 CURVNET_SET_QUIET(ifp->if_vnet); 1652 1653 /* Let BPF have it before we strip the header. */ 1654 IPOIB_MTAP(ifp, m); 1655 eh = mtod(m, struct ipoib_header *); 1656 /* 1657 * Reset layer specific mbuf flags to avoid confusing upper layers. 1658 * Strip off Infiniband header. 1659 */ 1660 m->m_flags &= ~M_VLANTAG; 1661 m_clrprotoflags(m); 1662 m_adj(m, IPOIB_HEADER_LEN); 1663 1664 if (IPOIB_IS_MULTICAST(eh->hwaddr)) { 1665 if (memcmp(eh->hwaddr, ifp->if_broadcastaddr, 1666 ifp->if_addrlen) == 0) 1667 m->m_flags |= M_BCAST; 1668 else 1669 m->m_flags |= M_MCAST; 1670 if_inc_counter(ifp, IFCOUNTER_IMCASTS, 1); 1671 } 1672 1673 ipoib_demux(ifp, m, ntohs(eh->proto)); 1674 CURVNET_RESTORE(); 1675} 1676 1677static int 1678ipoib_resolvemulti(struct ifnet *ifp, struct sockaddr **llsa, 1679 struct sockaddr *sa) 1680{ 1681 struct sockaddr_dl *sdl; 1682#ifdef INET 1683 struct sockaddr_in *sin; 1684#endif 1685#ifdef INET6 1686 struct sockaddr_in6 *sin6; 1687#endif 1688 u_char *e_addr; 1689 1690 switch(sa->sa_family) { 1691 case AF_LINK: 1692 /* 1693 * No mapping needed. Just check that it's a valid MC address. 1694 */ 1695 sdl = (struct sockaddr_dl *)sa; 1696 e_addr = LLADDR(sdl); 1697 if (!IPOIB_IS_MULTICAST(e_addr)) 1698 return EADDRNOTAVAIL; 1699 *llsa = NULL; 1700 return 0; 1701 1702#ifdef INET 1703 case AF_INET: 1704 sin = (struct sockaddr_in *)sa; 1705 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) 1706 return EADDRNOTAVAIL; 1707 sdl = link_init_sdl(ifp, *llsa, IFT_INFINIBAND); 1708 sdl->sdl_alen = INFINIBAND_ALEN; 1709 e_addr = LLADDR(sdl); 1710 ip_ib_mc_map(sin->sin_addr.s_addr, ifp->if_broadcastaddr, 1711 e_addr); 1712 *llsa = (struct sockaddr *)sdl; 1713 return 0; 1714#endif 1715#ifdef INET6 1716 case AF_INET6: 1717 sin6 = (struct sockaddr_in6 *)sa; 1718 /* 1719 * An IP6 address of 0 means listen to all 1720 * of the multicast address used for IP6. 1721 * This has no meaning in ipoib. 1722 */ 1723 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 1724 return EADDRNOTAVAIL; 1725 if (!IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 1726 return EADDRNOTAVAIL; 1727 sdl = link_init_sdl(ifp, *llsa, IFT_INFINIBAND); 1728 sdl->sdl_alen = INFINIBAND_ALEN; 1729 e_addr = LLADDR(sdl); 1730 ipv6_ib_mc_map(&sin6->sin6_addr, ifp->if_broadcastaddr, e_addr); 1731 *llsa = (struct sockaddr *)sdl; 1732 return 0; 1733#endif 1734 1735 default: 1736 return EAFNOSUPPORT; 1737 } 1738} 1739 1740module_init_order(ipoib_init_module, SI_ORDER_FIFTH); 1741module_exit_order(ipoib_cleanup_module, SI_ORDER_FIFTH); 1742 1743static int 1744ipoib_evhand(module_t mod, int event, void *arg) 1745{ 1746 return (0); 1747} 1748 1749static moduledata_t ipoib_mod = { 1750 .name = "ipoib", 1751 .evhand = ipoib_evhand, 1752}; 1753 1754DECLARE_MODULE(ipoib, ipoib_mod, SI_SUB_LAST, SI_ORDER_ANY); 1755MODULE_DEPEND(ipoib, ibcore, 1, 1, 1); 1756MODULE_DEPEND(ipoib, linuxkpi, 1, 1, 1); 1757