1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause OR GPL-2.0 3 * 4 * Copyright (c) 2004 Topspin Communications. All rights reserved. 5 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved. 6 * Copyright (c) 2004 Voltaire, Inc. All rights reserved. 7 * 8 * This software is available to you under a choice of one of two 9 * licenses. You may choose to be licensed under the terms of the GNU 10 * General Public License (GPL) Version 2, available from the file 11 * COPYING in the main directory of this source tree, or the 12 * OpenIB.org BSD license below: 13 * 14 * Redistribution and use in source and binary forms, with or 15 * without modification, are permitted provided that the following 16 * conditions are met: 17 * 18 * - Redistributions of source code must retain the above 19 * copyright notice, this list of conditions and the following 20 * disclaimer. 21 * 22 * - Redistributions in binary form must reproduce the above 23 * copyright notice, this list of conditions and the following 24 * disclaimer in the documentation and/or other materials 25 * provided with the distribution. 26 * 27 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 28 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 29 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 30 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 31 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 32 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 33 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 34 * SOFTWARE. 35 */ 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 #include "ipoib.h" 41 #include <sys/eventhandler.h> 42 43 static int ipoib_resolvemulti(struct ifnet *, struct sockaddr **, 44 struct sockaddr *); 45 46 47 #include <linux/module.h> 48 49 #include <linux/slab.h> 50 #include <linux/kernel.h> 51 #include <linux/vmalloc.h> 52 53 #include <linux/if_arp.h> /* For ARPHRD_xxx */ 54 #include <linux/if_vlan.h> 55 #include <net/ip.h> 56 #include <net/ipv6.h> 57 58 #include <rdma/ib_cache.h> 59 60 MODULE_AUTHOR("Roland Dreier"); 61 MODULE_DESCRIPTION("IP-over-InfiniBand net driver"); 62 MODULE_LICENSE("Dual BSD/GPL"); 63 64 int ipoib_sendq_size = IPOIB_TX_RING_SIZE; 65 int ipoib_recvq_size = IPOIB_RX_RING_SIZE; 66 67 module_param_named(send_queue_size, ipoib_sendq_size, int, 0444); 68 MODULE_PARM_DESC(send_queue_size, "Number of descriptors in send queue"); 69 module_param_named(recv_queue_size, ipoib_recvq_size, int, 0444); 70 MODULE_PARM_DESC(recv_queue_size, "Number of descriptors in receive queue"); 71 72 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG 73 int ipoib_debug_level = 1; 74 75 module_param_named(debug_level, ipoib_debug_level, int, 0644); 76 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0"); 77 #endif 78 79 struct ipoib_path_iter { 80 struct ipoib_dev_priv *priv; 81 struct ipoib_path path; 82 }; 83 84 static const u8 ipv4_bcast_addr[] = { 85 0x00, 0xff, 0xff, 0xff, 86 0xff, 0x12, 0x40, 0x1b, 0x00, 0x00, 0x00, 0x00, 87 0x00, 0x00, 0x00, 0x00, 0xff, 0xff, 0xff, 0xff 88 }; 89 90 struct workqueue_struct *ipoib_workqueue; 91 92 struct ib_sa_client ipoib_sa_client; 93 94 static void ipoib_add_one(struct ib_device *device); 95 static void ipoib_remove_one(struct ib_device *device, void *client_data); 96 static struct net_device *ipoib_get_net_dev_by_params( 97 struct ib_device *dev, u8 port, u16 pkey, 98 const union ib_gid *gid, const struct sockaddr *addr, 99 void *client_data); 100 static void ipoib_start(struct ifnet *dev); 101 static int ipoib_output(struct ifnet *ifp, struct mbuf *m, 102 const struct sockaddr *dst, struct route *ro); 103 static int ipoib_ioctl(struct ifnet *ifp, u_long command, caddr_t data); 104 static void ipoib_input(struct ifnet *ifp, struct mbuf *m); 105 106 #define IPOIB_MTAP(_ifp, _m) \ 107 do { \ 108 if (bpf_peers_present((_ifp)->if_bpf)) { \ 109 M_ASSERTVALID(_m); \ 110 ipoib_mtap_mb((_ifp), (_m)); \ 111 } \ 112 } while (0) 113 114 static struct unrhdr *ipoib_unrhdr; 115 116 static void 117 ipoib_unrhdr_init(void *arg) 118 { 119 120 ipoib_unrhdr = new_unrhdr(0, 65535, NULL); 121 } 122 SYSINIT(ipoib_unrhdr_init, SI_SUB_KLD - 1, SI_ORDER_ANY, ipoib_unrhdr_init, NULL); 123 124 static void 125 ipoib_unrhdr_uninit(void *arg) 126 { 127 128 if (ipoib_unrhdr != NULL) { 129 struct unrhdr *hdr; 130 131 hdr = ipoib_unrhdr; 132 ipoib_unrhdr = NULL; 133 134 delete_unrhdr(hdr); 135 } 136 } 137 SYSUNINIT(ipoib_unrhdr_uninit, SI_SUB_KLD - 1, SI_ORDER_ANY, ipoib_unrhdr_uninit, NULL); 138 139 /* 140 * This is for clients that have an ipoib_header in the mbuf. 141 */ 142 static void 143 ipoib_mtap_mb(struct ifnet *ifp, struct mbuf *mb) 144 { 145 struct ipoib_header *ih; 146 struct ether_header eh; 147 148 ih = mtod(mb, struct ipoib_header *); 149 eh.ether_type = ih->proto; 150 bcopy(ih->hwaddr, &eh.ether_dhost, ETHER_ADDR_LEN); 151 bzero(&eh.ether_shost, ETHER_ADDR_LEN); 152 mb->m_data += sizeof(struct ipoib_header); 153 mb->m_len -= sizeof(struct ipoib_header); 154 bpf_mtap2(ifp->if_bpf, &eh, sizeof(eh), mb); 155 mb->m_data -= sizeof(struct ipoib_header); 156 mb->m_len += sizeof(struct ipoib_header); 157 } 158 159 void 160 ipoib_mtap_proto(struct ifnet *ifp, struct mbuf *mb, uint16_t proto) 161 { 162 struct ether_header eh; 163 164 eh.ether_type = proto; 165 bzero(&eh.ether_shost, ETHER_ADDR_LEN); 166 bzero(&eh.ether_dhost, ETHER_ADDR_LEN); 167 bpf_mtap2(ifp->if_bpf, &eh, sizeof(eh), mb); 168 } 169 170 static struct ib_client ipoib_client = { 171 .name = "ipoib", 172 .add = ipoib_add_one, 173 .remove = ipoib_remove_one, 174 .get_net_dev_by_params = ipoib_get_net_dev_by_params, 175 }; 176 177 int 178 ipoib_open(struct ipoib_dev_priv *priv) 179 { 180 struct ifnet *dev = priv->dev; 181 182 ipoib_dbg(priv, "bringing up interface\n"); 183 184 set_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags); 185 186 if (ipoib_pkey_dev_delay_open(priv)) 187 return 0; 188 189 if (ipoib_ib_dev_open(priv)) 190 goto err_disable; 191 192 if (ipoib_ib_dev_up(priv)) 193 goto err_stop; 194 195 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) { 196 struct ipoib_dev_priv *cpriv; 197 198 /* Bring up any child interfaces too */ 199 mutex_lock(&priv->vlan_mutex); 200 list_for_each_entry(cpriv, &priv->child_intfs, list) 201 if ((cpriv->dev->if_drv_flags & IFF_DRV_RUNNING) == 0) 202 ipoib_open(cpriv); 203 mutex_unlock(&priv->vlan_mutex); 204 } 205 dev->if_drv_flags |= IFF_DRV_RUNNING; 206 dev->if_drv_flags &= ~IFF_DRV_OACTIVE; 207 208 return 0; 209 210 err_stop: 211 ipoib_ib_dev_stop(priv, 1); 212 213 err_disable: 214 clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags); 215 216 return -EINVAL; 217 } 218 219 static void 220 ipoib_init(void *arg) 221 { 222 struct ifnet *dev; 223 struct ipoib_dev_priv *priv; 224 225 priv = arg; 226 dev = priv->dev; 227 if ((dev->if_drv_flags & IFF_DRV_RUNNING) == 0) 228 ipoib_open(priv); 229 queue_work(ipoib_workqueue, &priv->flush_light); 230 } 231 232 233 static int 234 ipoib_stop(struct ipoib_dev_priv *priv) 235 { 236 struct ifnet *dev = priv->dev; 237 238 ipoib_dbg(priv, "stopping interface\n"); 239 240 clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags); 241 242 dev->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); 243 244 ipoib_ib_dev_down(priv, 0); 245 ipoib_ib_dev_stop(priv, 0); 246 247 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) { 248 struct ipoib_dev_priv *cpriv; 249 250 /* Bring down any child interfaces too */ 251 mutex_lock(&priv->vlan_mutex); 252 list_for_each_entry(cpriv, &priv->child_intfs, list) 253 if ((cpriv->dev->if_drv_flags & IFF_DRV_RUNNING) != 0) 254 ipoib_stop(cpriv); 255 mutex_unlock(&priv->vlan_mutex); 256 } 257 258 return 0; 259 } 260 261 static int 262 ipoib_propagate_ifnet_mtu(struct ipoib_dev_priv *priv, int new_mtu, 263 bool propagate) 264 { 265 struct ifnet *ifp; 266 struct ifreq ifr; 267 int error; 268 269 ifp = priv->dev; 270 if (ifp->if_mtu == new_mtu) 271 return (0); 272 if (propagate) { 273 strlcpy(ifr.ifr_name, if_name(ifp), IFNAMSIZ); 274 ifr.ifr_mtu = new_mtu; 275 CURVNET_SET(ifp->if_vnet); 276 error = ifhwioctl(SIOCSIFMTU, ifp, (caddr_t)&ifr, curthread); 277 CURVNET_RESTORE(); 278 } else { 279 ifp->if_mtu = new_mtu; 280 error = 0; 281 } 282 return (error); 283 } 284 285 int 286 ipoib_change_mtu(struct ipoib_dev_priv *priv, int new_mtu, bool propagate) 287 { 288 int error, prev_admin_mtu; 289 290 /* dev->if_mtu > 2K ==> connected mode */ 291 if (ipoib_cm_admin_enabled(priv)) { 292 if (new_mtu > IPOIB_CM_MTU(ipoib_cm_max_mtu(priv))) 293 return -EINVAL; 294 295 if (new_mtu > priv->mcast_mtu) 296 ipoib_warn(priv, "mtu > %d will cause multicast packet drops.\n", 297 priv->mcast_mtu); 298 299 return (ipoib_propagate_ifnet_mtu(priv, new_mtu, propagate)); 300 } 301 302 if (new_mtu > IPOIB_UD_MTU(priv->max_ib_mtu)) 303 return -EINVAL; 304 305 prev_admin_mtu = priv->admin_mtu; 306 priv->admin_mtu = new_mtu; 307 error = ipoib_propagate_ifnet_mtu(priv, min(priv->mcast_mtu, 308 priv->admin_mtu), propagate); 309 if (error == 0) { 310 /* check for MTU change to avoid infinite loop */ 311 if (prev_admin_mtu != new_mtu) 312 queue_work(ipoib_workqueue, &priv->flush_light); 313 } else 314 priv->admin_mtu = prev_admin_mtu; 315 return (error); 316 } 317 318 static int 319 ipoib_ioctl(struct ifnet *ifp, u_long command, caddr_t data) 320 { 321 struct ipoib_dev_priv *priv = ifp->if_softc; 322 struct ifaddr *ifa = (struct ifaddr *) data; 323 struct ifreq *ifr = (struct ifreq *) data; 324 int error = 0; 325 326 /* check if detaching */ 327 if (priv == NULL || priv->gone != 0) 328 return (ENXIO); 329 330 switch (command) { 331 case SIOCSIFFLAGS: 332 if (ifp->if_flags & IFF_UP) { 333 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 334 error = -ipoib_open(priv); 335 } else 336 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 337 ipoib_stop(priv); 338 break; 339 case SIOCADDMULTI: 340 case SIOCDELMULTI: 341 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 342 queue_work(ipoib_workqueue, &priv->restart_task); 343 break; 344 case SIOCSIFADDR: 345 ifp->if_flags |= IFF_UP; 346 347 switch (ifa->ifa_addr->sa_family) { 348 #ifdef INET 349 case AF_INET: 350 ifp->if_init(ifp->if_softc); /* before arpwhohas */ 351 arp_ifinit(ifp, ifa); 352 break; 353 #endif 354 default: 355 ifp->if_init(ifp->if_softc); 356 break; 357 } 358 break; 359 360 case SIOCGIFADDR: 361 bcopy(IF_LLADDR(ifp), &ifr->ifr_addr.sa_data[0], 362 INFINIBAND_ALEN); 363 break; 364 365 case SIOCSIFMTU: 366 /* 367 * Set the interface MTU. 368 */ 369 error = -ipoib_change_mtu(priv, ifr->ifr_mtu, false); 370 break; 371 default: 372 error = EINVAL; 373 break; 374 } 375 return (error); 376 } 377 378 379 static struct ipoib_path * 380 __path_find(struct ipoib_dev_priv *priv, void *gid) 381 { 382 struct rb_node *n = priv->path_tree.rb_node; 383 struct ipoib_path *path; 384 int ret; 385 386 while (n) { 387 path = rb_entry(n, struct ipoib_path, rb_node); 388 389 ret = memcmp(gid, path->pathrec.dgid.raw, 390 sizeof (union ib_gid)); 391 392 if (ret < 0) 393 n = n->rb_left; 394 else if (ret > 0) 395 n = n->rb_right; 396 else 397 return path; 398 } 399 400 return NULL; 401 } 402 403 static int 404 __path_add(struct ipoib_dev_priv *priv, struct ipoib_path *path) 405 { 406 struct rb_node **n = &priv->path_tree.rb_node; 407 struct rb_node *pn = NULL; 408 struct ipoib_path *tpath; 409 int ret; 410 411 while (*n) { 412 pn = *n; 413 tpath = rb_entry(pn, struct ipoib_path, rb_node); 414 415 ret = memcmp(path->pathrec.dgid.raw, tpath->pathrec.dgid.raw, 416 sizeof (union ib_gid)); 417 if (ret < 0) 418 n = &pn->rb_left; 419 else if (ret > 0) 420 n = &pn->rb_right; 421 else 422 return -EEXIST; 423 } 424 425 rb_link_node(&path->rb_node, pn, n); 426 rb_insert_color(&path->rb_node, &priv->path_tree); 427 428 list_add_tail(&path->list, &priv->path_list); 429 430 return 0; 431 } 432 433 void 434 ipoib_path_free(struct ipoib_dev_priv *priv, struct ipoib_path *path) 435 { 436 437 _IF_DRAIN(&path->queue); 438 439 if (path->ah) 440 ipoib_put_ah(path->ah); 441 if (ipoib_cm_get(path)) 442 ipoib_cm_destroy_tx(ipoib_cm_get(path)); 443 444 kfree(path); 445 } 446 447 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG 448 449 struct ipoib_path_iter * 450 ipoib_path_iter_init(struct ipoib_dev_priv *priv) 451 { 452 struct ipoib_path_iter *iter; 453 454 iter = kmalloc(sizeof *iter, GFP_KERNEL); 455 if (!iter) 456 return NULL; 457 458 iter->priv = priv; 459 memset(iter->path.pathrec.dgid.raw, 0, 16); 460 461 if (ipoib_path_iter_next(iter)) { 462 kfree(iter); 463 return NULL; 464 } 465 466 return iter; 467 } 468 469 int 470 ipoib_path_iter_next(struct ipoib_path_iter *iter) 471 { 472 struct ipoib_dev_priv *priv = iter->priv; 473 struct rb_node *n; 474 struct ipoib_path *path; 475 int ret = 1; 476 477 spin_lock_irq(&priv->lock); 478 479 n = rb_first(&priv->path_tree); 480 481 while (n) { 482 path = rb_entry(n, struct ipoib_path, rb_node); 483 484 if (memcmp(iter->path.pathrec.dgid.raw, path->pathrec.dgid.raw, 485 sizeof (union ib_gid)) < 0) { 486 iter->path = *path; 487 ret = 0; 488 break; 489 } 490 491 n = rb_next(n); 492 } 493 494 spin_unlock_irq(&priv->lock); 495 496 return ret; 497 } 498 499 void 500 ipoib_path_iter_read(struct ipoib_path_iter *iter, struct ipoib_path *path) 501 { 502 *path = iter->path; 503 } 504 505 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */ 506 507 void 508 ipoib_mark_paths_invalid(struct ipoib_dev_priv *priv) 509 { 510 struct ipoib_path *path, *tp; 511 512 spin_lock_irq(&priv->lock); 513 514 list_for_each_entry_safe(path, tp, &priv->path_list, list) { 515 ipoib_dbg(priv, "mark path LID 0x%04x GID %16D invalid\n", 516 be16_to_cpu(path->pathrec.dlid), 517 path->pathrec.dgid.raw, ":"); 518 path->valid = 0; 519 } 520 521 spin_unlock_irq(&priv->lock); 522 } 523 524 void 525 ipoib_flush_paths(struct ipoib_dev_priv *priv) 526 { 527 struct ipoib_path *path, *tp; 528 LIST_HEAD(remove_list); 529 unsigned long flags; 530 531 spin_lock_irqsave(&priv->lock, flags); 532 533 list_splice_init(&priv->path_list, &remove_list); 534 535 list_for_each_entry(path, &remove_list, list) 536 rb_erase(&path->rb_node, &priv->path_tree); 537 538 list_for_each_entry_safe(path, tp, &remove_list, list) { 539 if (path->query) 540 ib_sa_cancel_query(path->query_id, path->query); 541 spin_unlock_irqrestore(&priv->lock, flags); 542 wait_for_completion(&path->done); 543 ipoib_path_free(priv, path); 544 spin_lock_irqsave(&priv->lock, flags); 545 } 546 547 spin_unlock_irqrestore(&priv->lock, flags); 548 } 549 550 static void 551 path_rec_completion(int status, struct ib_sa_path_rec *pathrec, void *path_ptr) 552 { 553 struct ipoib_path *path = path_ptr; 554 struct ipoib_dev_priv *priv = path->priv; 555 struct ifnet *dev = priv->dev; 556 struct ipoib_ah *ah = NULL; 557 struct ipoib_ah *old_ah = NULL; 558 struct epoch_tracker et; 559 struct ifqueue mbqueue; 560 struct mbuf *mb; 561 unsigned long flags; 562 563 if (!status) 564 ipoib_dbg(priv, "PathRec LID 0x%04x for GID %16D\n", 565 be16_to_cpu(pathrec->dlid), pathrec->dgid.raw, ":"); 566 else 567 ipoib_dbg(priv, "PathRec status %d for GID %16D\n", 568 status, path->pathrec.dgid.raw, ":"); 569 570 bzero(&mbqueue, sizeof(mbqueue)); 571 572 if (!status) { 573 struct ib_ah_attr av; 574 575 if (!ib_init_ah_from_path(priv->ca, priv->port, pathrec, &av)) 576 ah = ipoib_create_ah(priv, priv->pd, &av); 577 } 578 579 spin_lock_irqsave(&priv->lock, flags); 580 581 if (ah) { 582 path->pathrec = *pathrec; 583 584 old_ah = path->ah; 585 path->ah = ah; 586 587 ipoib_dbg(priv, "created address handle %p for LID 0x%04x, SL %d\n", 588 ah, be16_to_cpu(pathrec->dlid), pathrec->sl); 589 590 for (;;) { 591 _IF_DEQUEUE(&path->queue, mb); 592 if (mb == NULL) 593 break; 594 _IF_ENQUEUE(&mbqueue, mb); 595 } 596 597 #ifdef CONFIG_INFINIBAND_IPOIB_CM 598 if (ipoib_cm_enabled(priv, path->hwaddr) && !ipoib_cm_get(path)) 599 ipoib_cm_set(path, ipoib_cm_create_tx(priv, path)); 600 #endif 601 602 path->valid = 1; 603 } 604 605 path->query = NULL; 606 complete(&path->done); 607 608 spin_unlock_irqrestore(&priv->lock, flags); 609 610 if (old_ah) 611 ipoib_put_ah(old_ah); 612 613 NET_EPOCH_ENTER(et); 614 for (;;) { 615 _IF_DEQUEUE(&mbqueue, mb); 616 if (mb == NULL) 617 break; 618 mb->m_pkthdr.rcvif = dev; 619 if (dev->if_transmit(dev, mb)) 620 ipoib_warn(priv, "dev_queue_xmit failed " 621 "to requeue packet\n"); 622 } 623 NET_EPOCH_EXIT(et); 624 } 625 626 static struct ipoib_path * 627 path_rec_create(struct ipoib_dev_priv *priv, uint8_t *hwaddr) 628 { 629 struct ipoib_path *path; 630 631 if (!priv->broadcast) 632 return NULL; 633 634 path = kzalloc(sizeof *path, GFP_ATOMIC); 635 if (!path) 636 return NULL; 637 638 path->priv = priv; 639 640 bzero(&path->queue, sizeof(path->queue)); 641 642 #ifdef CONFIG_INFINIBAND_IPOIB_CM 643 memcpy(&path->hwaddr, hwaddr, INFINIBAND_ALEN); 644 #endif 645 memcpy(path->pathrec.dgid.raw, &hwaddr[4], sizeof (union ib_gid)); 646 path->pathrec.sgid = priv->local_gid; 647 path->pathrec.pkey = cpu_to_be16(priv->pkey); 648 path->pathrec.numb_path = 1; 649 path->pathrec.traffic_class = priv->broadcast->mcmember.traffic_class; 650 651 return path; 652 } 653 654 static int 655 path_rec_start(struct ipoib_dev_priv *priv, struct ipoib_path *path) 656 { 657 struct ifnet *dev = priv->dev; 658 659 ib_sa_comp_mask comp_mask = IB_SA_PATH_REC_MTU_SELECTOR | IB_SA_PATH_REC_MTU; 660 struct ib_sa_path_rec p_rec; 661 662 p_rec = path->pathrec; 663 p_rec.mtu_selector = IB_SA_GT; 664 665 switch (roundup_pow_of_two(dev->if_mtu + IPOIB_ENCAP_LEN)) { 666 case 512: 667 p_rec.mtu = IB_MTU_256; 668 break; 669 case 1024: 670 p_rec.mtu = IB_MTU_512; 671 break; 672 case 2048: 673 p_rec.mtu = IB_MTU_1024; 674 break; 675 case 4096: 676 p_rec.mtu = IB_MTU_2048; 677 break; 678 default: 679 /* Wildcard everything */ 680 comp_mask = 0; 681 p_rec.mtu = 0; 682 p_rec.mtu_selector = 0; 683 } 684 685 ipoib_dbg(priv, "Start path record lookup for %16D MTU > %d\n", 686 p_rec.dgid.raw, ":", 687 comp_mask ? ib_mtu_enum_to_int(p_rec.mtu) : 0); 688 689 init_completion(&path->done); 690 691 path->query_id = 692 ib_sa_path_rec_get(&ipoib_sa_client, priv->ca, priv->port, 693 &p_rec, comp_mask | 694 IB_SA_PATH_REC_DGID | 695 IB_SA_PATH_REC_SGID | 696 IB_SA_PATH_REC_NUMB_PATH | 697 IB_SA_PATH_REC_TRAFFIC_CLASS | 698 IB_SA_PATH_REC_PKEY, 699 1000, GFP_ATOMIC, 700 path_rec_completion, 701 path, &path->query); 702 if (path->query_id < 0) { 703 ipoib_warn(priv, "ib_sa_path_rec_get failed: %d\n", path->query_id); 704 path->query = NULL; 705 complete(&path->done); 706 return path->query_id; 707 } 708 709 return 0; 710 } 711 712 static void 713 ipoib_unicast_send(struct mbuf *mb, struct ipoib_dev_priv *priv, struct ipoib_header *eh) 714 { 715 struct ipoib_path *path; 716 717 path = __path_find(priv, eh->hwaddr + 4); 718 if (!path || !path->valid) { 719 int new_path = 0; 720 721 if (!path) { 722 path = path_rec_create(priv, eh->hwaddr); 723 new_path = 1; 724 } 725 if (path) { 726 if (_IF_QLEN(&path->queue) < IPOIB_MAX_PATH_REC_QUEUE) 727 _IF_ENQUEUE(&path->queue, mb); 728 else { 729 if_inc_counter(priv->dev, IFCOUNTER_OERRORS, 1); 730 m_freem(mb); 731 } 732 733 if (!path->query && path_rec_start(priv, path)) { 734 if (new_path) 735 ipoib_path_free(priv, path); 736 return; 737 } else 738 __path_add(priv, path); 739 } else { 740 if_inc_counter(priv->dev, IFCOUNTER_OERRORS, 1); 741 m_freem(mb); 742 } 743 744 return; 745 } 746 747 if (ipoib_cm_get(path) && ipoib_cm_up(path)) { 748 ipoib_cm_send(priv, mb, ipoib_cm_get(path)); 749 } else if (path->ah) { 750 ipoib_send(priv, mb, path->ah, IPOIB_QPN(eh->hwaddr)); 751 } else if ((path->query || !path_rec_start(priv, path)) && 752 path->queue.ifq_len < IPOIB_MAX_PATH_REC_QUEUE) { 753 _IF_ENQUEUE(&path->queue, mb); 754 } else { 755 if_inc_counter(priv->dev, IFCOUNTER_OERRORS, 1); 756 m_freem(mb); 757 } 758 } 759 760 static int 761 ipoib_send_one(struct ipoib_dev_priv *priv, struct mbuf *mb) 762 { 763 struct ipoib_header *eh; 764 765 eh = mtod(mb, struct ipoib_header *); 766 if (IPOIB_IS_MULTICAST(eh->hwaddr)) { 767 /* Add in the P_Key for multicast*/ 768 eh->hwaddr[8] = (priv->pkey >> 8) & 0xff; 769 eh->hwaddr[9] = priv->pkey & 0xff; 770 771 ipoib_mcast_send(priv, eh->hwaddr + 4, mb); 772 } else 773 ipoib_unicast_send(mb, priv, eh); 774 775 return 0; 776 } 777 778 void 779 ipoib_start_locked(struct ifnet *dev, struct ipoib_dev_priv *priv) 780 { 781 struct mbuf *mb; 782 783 assert_spin_locked(&priv->lock); 784 785 while (!IFQ_DRV_IS_EMPTY(&dev->if_snd) && 786 (dev->if_drv_flags & IFF_DRV_OACTIVE) == 0) { 787 IFQ_DRV_DEQUEUE(&dev->if_snd, mb); 788 if (mb == NULL) 789 break; 790 IPOIB_MTAP(dev, mb); 791 ipoib_send_one(priv, mb); 792 } 793 } 794 795 static void 796 _ipoib_start(struct ifnet *dev, struct ipoib_dev_priv *priv) 797 { 798 799 if ((dev->if_drv_flags & (IFF_DRV_RUNNING|IFF_DRV_OACTIVE)) != 800 IFF_DRV_RUNNING) 801 return; 802 803 spin_lock(&priv->lock); 804 ipoib_start_locked(dev, priv); 805 spin_unlock(&priv->lock); 806 } 807 808 static void 809 ipoib_start(struct ifnet *dev) 810 { 811 _ipoib_start(dev, dev->if_softc); 812 } 813 814 static void 815 ipoib_vlan_start(struct ifnet *dev) 816 { 817 struct ipoib_dev_priv *priv; 818 struct mbuf *mb; 819 820 priv = VLAN_COOKIE(dev); 821 if (priv != NULL) 822 return _ipoib_start(dev, priv); 823 while (!IFQ_DRV_IS_EMPTY(&dev->if_snd)) { 824 IFQ_DRV_DEQUEUE(&dev->if_snd, mb); 825 if (mb == NULL) 826 break; 827 m_freem(mb); 828 if_inc_counter(dev, IFCOUNTER_OERRORS, 1); 829 } 830 } 831 832 int 833 ipoib_dev_init(struct ipoib_dev_priv *priv, struct ib_device *ca, int port) 834 { 835 836 /* Allocate RX/TX "rings" to hold queued mbs */ 837 priv->rx_ring = kzalloc(ipoib_recvq_size * sizeof *priv->rx_ring, 838 GFP_KERNEL); 839 if (!priv->rx_ring) { 840 printk(KERN_WARNING "%s: failed to allocate RX ring (%d entries)\n", 841 ca->name, ipoib_recvq_size); 842 goto out; 843 } 844 845 priv->tx_ring = kzalloc(ipoib_sendq_size * sizeof *priv->tx_ring, GFP_KERNEL); 846 if (!priv->tx_ring) { 847 printk(KERN_WARNING "%s: failed to allocate TX ring (%d entries)\n", 848 ca->name, ipoib_sendq_size); 849 goto out_rx_ring_cleanup; 850 } 851 memset(priv->tx_ring, 0, ipoib_sendq_size * sizeof *priv->tx_ring); 852 853 /* priv->tx_head, tx_tail & tx_outstanding are already 0 */ 854 855 if (ipoib_ib_dev_init(priv, ca, port)) 856 goto out_tx_ring_cleanup; 857 858 return 0; 859 860 out_tx_ring_cleanup: 861 kfree(priv->tx_ring); 862 863 out_rx_ring_cleanup: 864 kfree(priv->rx_ring); 865 866 out: 867 return -ENOMEM; 868 } 869 870 static void 871 ipoib_detach(struct ipoib_dev_priv *priv) 872 { 873 struct ifnet *dev; 874 875 dev = priv->dev; 876 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) { 877 priv->gone = 1; 878 bpfdetach(dev); 879 if_detach(dev); 880 if_free(dev); 881 free_unr(ipoib_unrhdr, priv->unit); 882 } else 883 VLAN_SETCOOKIE(priv->dev, NULL); 884 885 free(priv, M_TEMP); 886 } 887 888 void 889 ipoib_dev_cleanup(struct ipoib_dev_priv *priv) 890 { 891 struct ipoib_dev_priv *cpriv, *tcpriv; 892 893 /* Delete any child interfaces first */ 894 list_for_each_entry_safe(cpriv, tcpriv, &priv->child_intfs, list) { 895 ipoib_dev_cleanup(cpriv); 896 ipoib_detach(cpriv); 897 } 898 899 ipoib_ib_dev_cleanup(priv); 900 901 kfree(priv->rx_ring); 902 kfree(priv->tx_ring); 903 904 priv->rx_ring = NULL; 905 priv->tx_ring = NULL; 906 } 907 908 static struct ipoib_dev_priv * 909 ipoib_priv_alloc(void) 910 { 911 struct ipoib_dev_priv *priv; 912 913 priv = malloc(sizeof(struct ipoib_dev_priv), M_TEMP, M_ZERO|M_WAITOK); 914 spin_lock_init(&priv->lock); 915 spin_lock_init(&priv->drain_lock); 916 mutex_init(&priv->vlan_mutex); 917 INIT_LIST_HEAD(&priv->path_list); 918 INIT_LIST_HEAD(&priv->child_intfs); 919 INIT_LIST_HEAD(&priv->dead_ahs); 920 INIT_LIST_HEAD(&priv->multicast_list); 921 INIT_DELAYED_WORK(&priv->pkey_poll_task, ipoib_pkey_poll); 922 INIT_DELAYED_WORK(&priv->mcast_task, ipoib_mcast_join_task); 923 INIT_WORK(&priv->carrier_on_task, ipoib_mcast_carrier_on_task); 924 INIT_WORK(&priv->flush_light, ipoib_ib_dev_flush_light); 925 INIT_WORK(&priv->flush_normal, ipoib_ib_dev_flush_normal); 926 INIT_WORK(&priv->flush_heavy, ipoib_ib_dev_flush_heavy); 927 INIT_WORK(&priv->restart_task, ipoib_mcast_restart_task); 928 INIT_DELAYED_WORK(&priv->ah_reap_task, ipoib_reap_ah); 929 memcpy(priv->broadcastaddr, ipv4_bcast_addr, INFINIBAND_ALEN); 930 931 return (priv); 932 } 933 934 struct ipoib_dev_priv * 935 ipoib_intf_alloc(const char *name) 936 { 937 struct ipoib_dev_priv *priv; 938 struct sockaddr_dl *sdl; 939 struct ifnet *dev; 940 941 priv = ipoib_priv_alloc(); 942 dev = priv->dev = if_alloc(IFT_INFINIBAND); 943 if (!dev) { 944 free(priv, M_TEMP); 945 return NULL; 946 } 947 dev->if_softc = priv; 948 priv->unit = alloc_unr(ipoib_unrhdr); 949 if (priv->unit == -1) { 950 if_free(dev); 951 free(priv, M_TEMP); 952 return NULL; 953 } 954 if_initname(dev, name, priv->unit); 955 dev->if_flags = IFF_BROADCAST | IFF_MULTICAST; 956 dev->if_addrlen = INFINIBAND_ALEN; 957 dev->if_hdrlen = IPOIB_HEADER_LEN; 958 if_attach(dev); 959 dev->if_init = ipoib_init; 960 dev->if_ioctl = ipoib_ioctl; 961 dev->if_start = ipoib_start; 962 dev->if_output = ipoib_output; 963 dev->if_input = ipoib_input; 964 dev->if_resolvemulti = ipoib_resolvemulti; 965 dev->if_baudrate = IF_Gbps(10); 966 dev->if_broadcastaddr = priv->broadcastaddr; 967 dev->if_snd.ifq_maxlen = ipoib_sendq_size * 2; 968 sdl = (struct sockaddr_dl *)dev->if_addr->ifa_addr; 969 sdl->sdl_type = IFT_INFINIBAND; 970 sdl->sdl_alen = dev->if_addrlen; 971 priv->dev = dev; 972 if_link_state_change(dev, LINK_STATE_DOWN); 973 bpfattach(dev, DLT_EN10MB, ETHER_HDR_LEN); 974 975 return dev->if_softc; 976 } 977 978 int 979 ipoib_set_dev_features(struct ipoib_dev_priv *priv, struct ib_device *hca) 980 { 981 struct ib_device_attr *device_attr = &hca->attrs; 982 983 priv->hca_caps = device_attr->device_cap_flags; 984 985 priv->dev->if_hwassist = 0; 986 priv->dev->if_capabilities = 0; 987 988 #ifndef CONFIG_INFINIBAND_IPOIB_CM 989 if (priv->hca_caps & IB_DEVICE_UD_IP_CSUM) { 990 set_bit(IPOIB_FLAG_CSUM, &priv->flags); 991 priv->dev->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP; 992 priv->dev->if_capabilities = IFCAP_HWCSUM | IFCAP_VLAN_HWCSUM; 993 } 994 995 #if 0 996 if (priv->dev->features & NETIF_F_SG && priv->hca_caps & IB_DEVICE_UD_TSO) { 997 priv->dev->if_capabilities |= IFCAP_TSO4; 998 priv->dev->if_hwassist |= CSUM_TSO; 999 } 1000 #endif 1001 #endif 1002 priv->dev->if_capabilities |= 1003 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_LINKSTATE; 1004 priv->dev->if_capenable = priv->dev->if_capabilities; 1005 1006 return 0; 1007 } 1008 1009 1010 static struct ifnet * 1011 ipoib_add_port(const char *format, struct ib_device *hca, u8 port) 1012 { 1013 struct ipoib_dev_priv *priv; 1014 struct ib_port_attr attr; 1015 int result = -ENOMEM; 1016 1017 priv = ipoib_intf_alloc(format); 1018 if (!priv) 1019 goto alloc_mem_failed; 1020 1021 if (!ib_query_port(hca, port, &attr)) 1022 priv->max_ib_mtu = ib_mtu_enum_to_int(attr.max_mtu); 1023 else { 1024 printk(KERN_WARNING "%s: ib_query_port %d failed\n", 1025 hca->name, port); 1026 goto device_init_failed; 1027 } 1028 1029 /* MTU will be reset when mcast join happens */ 1030 priv->dev->if_mtu = IPOIB_UD_MTU(priv->max_ib_mtu); 1031 priv->mcast_mtu = priv->admin_mtu = priv->dev->if_mtu; 1032 1033 result = ib_query_pkey(hca, port, 0, &priv->pkey); 1034 if (result) { 1035 printk(KERN_WARNING "%s: ib_query_pkey port %d failed (ret = %d)\n", 1036 hca->name, port, result); 1037 goto device_init_failed; 1038 } 1039 1040 if (ipoib_set_dev_features(priv, hca)) 1041 goto device_init_failed; 1042 1043 /* 1044 * Set the full membership bit, so that we join the right 1045 * broadcast group, etc. 1046 */ 1047 priv->pkey |= 0x8000; 1048 1049 priv->broadcastaddr[8] = priv->pkey >> 8; 1050 priv->broadcastaddr[9] = priv->pkey & 0xff; 1051 1052 result = ib_query_gid(hca, port, 0, &priv->local_gid, NULL); 1053 if (result) { 1054 printk(KERN_WARNING "%s: ib_query_gid port %d failed (ret = %d)\n", 1055 hca->name, port, result); 1056 goto device_init_failed; 1057 } 1058 memcpy(IF_LLADDR(priv->dev) + 4, priv->local_gid.raw, sizeof (union ib_gid)); 1059 1060 result = ipoib_dev_init(priv, hca, port); 1061 if (result < 0) { 1062 printk(KERN_WARNING "%s: failed to initialize port %d (ret = %d)\n", 1063 hca->name, port, result); 1064 goto device_init_failed; 1065 } 1066 if (ipoib_cm_admin_enabled(priv)) 1067 priv->dev->if_mtu = IPOIB_CM_MTU(ipoib_cm_max_mtu(priv)); 1068 1069 INIT_IB_EVENT_HANDLER(&priv->event_handler, 1070 priv->ca, ipoib_event); 1071 result = ib_register_event_handler(&priv->event_handler); 1072 if (result < 0) { 1073 printk(KERN_WARNING "%s: ib_register_event_handler failed for " 1074 "port %d (ret = %d)\n", 1075 hca->name, port, result); 1076 goto event_failed; 1077 } 1078 if_printf(priv->dev, "Attached to %s port %d\n", hca->name, port); 1079 1080 return priv->dev; 1081 1082 event_failed: 1083 ipoib_dev_cleanup(priv); 1084 1085 device_init_failed: 1086 ipoib_detach(priv); 1087 1088 alloc_mem_failed: 1089 return ERR_PTR(result); 1090 } 1091 1092 static void 1093 ipoib_add_one(struct ib_device *device) 1094 { 1095 struct list_head *dev_list; 1096 struct ifnet *dev; 1097 struct ipoib_dev_priv *priv; 1098 int s, e, p; 1099 1100 if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB) 1101 return; 1102 1103 dev_list = kmalloc(sizeof *dev_list, GFP_KERNEL); 1104 if (!dev_list) 1105 return; 1106 1107 INIT_LIST_HEAD(dev_list); 1108 1109 if (device->node_type == RDMA_NODE_IB_SWITCH) { 1110 s = 0; 1111 e = 0; 1112 } else { 1113 s = 1; 1114 e = device->phys_port_cnt; 1115 } 1116 1117 for (p = s; p <= e; ++p) { 1118 if (rdma_port_get_link_layer(device, p) != IB_LINK_LAYER_INFINIBAND) 1119 continue; 1120 dev = ipoib_add_port("ib", device, p); 1121 if (!IS_ERR(dev)) { 1122 priv = dev->if_softc; 1123 list_add_tail(&priv->list, dev_list); 1124 } 1125 } 1126 1127 ib_set_client_data(device, &ipoib_client, dev_list); 1128 } 1129 1130 static void 1131 ipoib_remove_one(struct ib_device *device, void *client_data) 1132 { 1133 struct ipoib_dev_priv *priv, *tmp; 1134 struct list_head *dev_list = client_data; 1135 1136 if (!dev_list) 1137 return; 1138 1139 if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB) 1140 return; 1141 1142 list_for_each_entry_safe(priv, tmp, dev_list, list) { 1143 if (rdma_port_get_link_layer(device, priv->port) != IB_LINK_LAYER_INFINIBAND) 1144 continue; 1145 1146 ipoib_stop(priv); 1147 1148 ib_unregister_event_handler(&priv->event_handler); 1149 1150 /* dev_change_flags(priv->dev, priv->dev->flags & ~IFF_UP); */ 1151 1152 flush_workqueue(ipoib_workqueue); 1153 1154 ipoib_dev_cleanup(priv); 1155 ipoib_detach(priv); 1156 } 1157 1158 kfree(dev_list); 1159 } 1160 1161 static int 1162 ipoib_match_dev_addr(const struct sockaddr *addr, struct net_device *dev) 1163 { 1164 struct epoch_tracker et; 1165 struct ifaddr *ifa; 1166 int retval = 0; 1167 1168 CURVNET_SET(dev->if_vnet); 1169 NET_EPOCH_ENTER(et); 1170 CK_STAILQ_FOREACH(ifa, &dev->if_addrhead, ifa_link) { 1171 if (ifa->ifa_addr == NULL || 1172 ifa->ifa_addr->sa_family != addr->sa_family || 1173 ifa->ifa_addr->sa_len != addr->sa_len) { 1174 continue; 1175 } 1176 if (memcmp(ifa->ifa_addr, addr, addr->sa_len) == 0) { 1177 retval = 1; 1178 break; 1179 } 1180 } 1181 NET_EPOCH_EXIT(et); 1182 CURVNET_RESTORE(); 1183 1184 return (retval); 1185 } 1186 1187 /* 1188 * ipoib_match_gid_pkey_addr - returns the number of IPoIB netdevs on 1189 * top a given ipoib device matching a pkey_index and address, if one 1190 * exists. 1191 * 1192 * @found_net_dev: contains a matching net_device if the return value 1193 * >= 1, with a reference held. 1194 */ 1195 static int 1196 ipoib_match_gid_pkey_addr(struct ipoib_dev_priv *priv, 1197 const union ib_gid *gid, u16 pkey_index, const struct sockaddr *addr, 1198 struct net_device **found_net_dev) 1199 { 1200 struct ipoib_dev_priv *child_priv; 1201 int matches = 0; 1202 1203 if (priv->pkey_index == pkey_index && 1204 (!gid || !memcmp(gid, &priv->local_gid, sizeof(*gid)))) { 1205 if (addr == NULL || ipoib_match_dev_addr(addr, priv->dev) != 0) { 1206 if (*found_net_dev == NULL) { 1207 struct net_device *net_dev; 1208 1209 if (priv->parent != NULL) 1210 net_dev = priv->parent; 1211 else 1212 net_dev = priv->dev; 1213 *found_net_dev = net_dev; 1214 dev_hold(net_dev); 1215 } 1216 matches++; 1217 } 1218 } 1219 1220 /* Check child interfaces */ 1221 mutex_lock(&priv->vlan_mutex); 1222 list_for_each_entry(child_priv, &priv->child_intfs, list) { 1223 matches += ipoib_match_gid_pkey_addr(child_priv, gid, 1224 pkey_index, addr, found_net_dev); 1225 if (matches > 1) 1226 break; 1227 } 1228 mutex_unlock(&priv->vlan_mutex); 1229 1230 return matches; 1231 } 1232 1233 /* 1234 * __ipoib_get_net_dev_by_params - returns the number of matching 1235 * net_devs found (between 0 and 2). Also return the matching 1236 * net_device in the @net_dev parameter, holding a reference to the 1237 * net_device, if the number of matches >= 1 1238 */ 1239 static int 1240 __ipoib_get_net_dev_by_params(struct list_head *dev_list, u8 port, 1241 u16 pkey_index, const union ib_gid *gid, 1242 const struct sockaddr *addr, struct net_device **net_dev) 1243 { 1244 struct ipoib_dev_priv *priv; 1245 int matches = 0; 1246 1247 *net_dev = NULL; 1248 1249 list_for_each_entry(priv, dev_list, list) { 1250 if (priv->port != port) 1251 continue; 1252 1253 matches += ipoib_match_gid_pkey_addr(priv, gid, pkey_index, 1254 addr, net_dev); 1255 1256 if (matches > 1) 1257 break; 1258 } 1259 1260 return matches; 1261 } 1262 1263 static struct net_device * 1264 ipoib_get_net_dev_by_params(struct ib_device *dev, u8 port, u16 pkey, 1265 const union ib_gid *gid, const struct sockaddr *addr, void *client_data) 1266 { 1267 struct net_device *net_dev; 1268 struct list_head *dev_list = client_data; 1269 u16 pkey_index; 1270 int matches; 1271 int ret; 1272 1273 if (!rdma_protocol_ib(dev, port)) 1274 return NULL; 1275 1276 ret = ib_find_cached_pkey(dev, port, pkey, &pkey_index); 1277 if (ret) 1278 return NULL; 1279 1280 if (!dev_list) 1281 return NULL; 1282 1283 /* See if we can find a unique device matching the L2 parameters */ 1284 matches = __ipoib_get_net_dev_by_params(dev_list, port, pkey_index, 1285 gid, NULL, &net_dev); 1286 1287 switch (matches) { 1288 case 0: 1289 return NULL; 1290 case 1: 1291 return net_dev; 1292 } 1293 1294 dev_put(net_dev); 1295 1296 /* Couldn't find a unique device with L2 parameters only. Use L3 1297 * address to uniquely match the net device */ 1298 matches = __ipoib_get_net_dev_by_params(dev_list, port, pkey_index, 1299 gid, addr, &net_dev); 1300 switch (matches) { 1301 case 0: 1302 return NULL; 1303 default: 1304 dev_warn_ratelimited(&dev->dev, 1305 "duplicate IP address detected\n"); 1306 /* Fall through */ 1307 case 1: 1308 return net_dev; 1309 } 1310 } 1311 1312 static void 1313 ipoib_config_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 1314 { 1315 struct ipoib_dev_priv *parent; 1316 struct ipoib_dev_priv *priv; 1317 struct epoch_tracker et; 1318 struct ifnet *dev; 1319 uint16_t pkey; 1320 int error; 1321 1322 if (ifp->if_type != IFT_INFINIBAND) 1323 return; 1324 NET_EPOCH_ENTER(et); 1325 dev = VLAN_DEVAT(ifp, vtag); 1326 NET_EPOCH_EXIT(et); 1327 if (dev == NULL) 1328 return; 1329 priv = NULL; 1330 error = 0; 1331 parent = ifp->if_softc; 1332 /* We only support 15 bits of pkey. */ 1333 if (vtag & 0x8000) 1334 return; 1335 pkey = vtag | 0x8000; /* Set full membership bit. */ 1336 if (pkey == parent->pkey) 1337 return; 1338 /* Check for dups */ 1339 mutex_lock(&parent->vlan_mutex); 1340 list_for_each_entry(priv, &parent->child_intfs, list) { 1341 if (priv->pkey == pkey) { 1342 priv = NULL; 1343 error = EBUSY; 1344 goto out; 1345 } 1346 } 1347 priv = ipoib_priv_alloc(); 1348 priv->dev = dev; 1349 priv->max_ib_mtu = parent->max_ib_mtu; 1350 priv->mcast_mtu = priv->admin_mtu = parent->dev->if_mtu; 1351 set_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags); 1352 error = ipoib_set_dev_features(priv, parent->ca); 1353 if (error) 1354 goto out; 1355 priv->pkey = pkey; 1356 priv->broadcastaddr[8] = pkey >> 8; 1357 priv->broadcastaddr[9] = pkey & 0xff; 1358 dev->if_broadcastaddr = priv->broadcastaddr; 1359 error = ipoib_dev_init(priv, parent->ca, parent->port); 1360 if (error) 1361 goto out; 1362 priv->parent = parent->dev; 1363 list_add_tail(&priv->list, &parent->child_intfs); 1364 VLAN_SETCOOKIE(dev, priv); 1365 dev->if_start = ipoib_vlan_start; 1366 dev->if_drv_flags &= ~IFF_DRV_RUNNING; 1367 dev->if_hdrlen = IPOIB_HEADER_LEN; 1368 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 1369 ipoib_open(priv); 1370 mutex_unlock(&parent->vlan_mutex); 1371 return; 1372 out: 1373 mutex_unlock(&parent->vlan_mutex); 1374 if (priv) 1375 free(priv, M_TEMP); 1376 if (error) 1377 ipoib_warn(parent, 1378 "failed to initialize subinterface: device %s, port %d vtag 0x%X", 1379 parent->ca->name, parent->port, vtag); 1380 return; 1381 } 1382 1383 static void 1384 ipoib_unconfig_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 1385 { 1386 struct ipoib_dev_priv *parent; 1387 struct ipoib_dev_priv *priv; 1388 struct epoch_tracker et; 1389 struct ifnet *dev; 1390 uint16_t pkey; 1391 1392 if (ifp->if_type != IFT_INFINIBAND) 1393 return; 1394 1395 NET_EPOCH_ENTER(et); 1396 dev = VLAN_DEVAT(ifp, vtag); 1397 NET_EPOCH_EXIT(et); 1398 if (dev) 1399 VLAN_SETCOOKIE(dev, NULL); 1400 pkey = vtag | 0x8000; 1401 parent = ifp->if_softc; 1402 mutex_lock(&parent->vlan_mutex); 1403 list_for_each_entry(priv, &parent->child_intfs, list) { 1404 if (priv->pkey == pkey) { 1405 ipoib_dev_cleanup(priv); 1406 list_del(&priv->list); 1407 break; 1408 } 1409 } 1410 mutex_unlock(&parent->vlan_mutex); 1411 } 1412 1413 eventhandler_tag ipoib_vlan_attach; 1414 eventhandler_tag ipoib_vlan_detach; 1415 1416 static int __init 1417 ipoib_init_module(void) 1418 { 1419 int ret; 1420 1421 ipoib_recvq_size = roundup_pow_of_two(ipoib_recvq_size); 1422 ipoib_recvq_size = min(ipoib_recvq_size, IPOIB_MAX_QUEUE_SIZE); 1423 ipoib_recvq_size = max(ipoib_recvq_size, IPOIB_MIN_QUEUE_SIZE); 1424 1425 ipoib_sendq_size = roundup_pow_of_two(ipoib_sendq_size); 1426 ipoib_sendq_size = min(ipoib_sendq_size, IPOIB_MAX_QUEUE_SIZE); 1427 ipoib_sendq_size = max(ipoib_sendq_size, max(2 * MAX_SEND_CQE, 1428 IPOIB_MIN_QUEUE_SIZE)); 1429 #ifdef CONFIG_INFINIBAND_IPOIB_CM 1430 ipoib_max_conn_qp = min(ipoib_max_conn_qp, IPOIB_CM_MAX_CONN_QP); 1431 #endif 1432 1433 ipoib_vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 1434 ipoib_config_vlan, NULL, EVENTHANDLER_PRI_FIRST); 1435 ipoib_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 1436 ipoib_unconfig_vlan, NULL, EVENTHANDLER_PRI_FIRST); 1437 1438 /* 1439 * We create our own workqueue mainly because we want to be 1440 * able to flush it when devices are being removed. We can't 1441 * use schedule_work()/flush_scheduled_work() because both 1442 * unregister_netdev() and linkwatch_event take the rtnl lock, 1443 * so flush_scheduled_work() can deadlock during device 1444 * removal. 1445 */ 1446 ipoib_workqueue = create_singlethread_workqueue("ipoib"); 1447 if (!ipoib_workqueue) { 1448 ret = -ENOMEM; 1449 goto err_fs; 1450 } 1451 1452 ib_sa_register_client(&ipoib_sa_client); 1453 1454 ret = ib_register_client(&ipoib_client); 1455 if (ret) 1456 goto err_sa; 1457 1458 return 0; 1459 1460 err_sa: 1461 ib_sa_unregister_client(&ipoib_sa_client); 1462 destroy_workqueue(ipoib_workqueue); 1463 1464 err_fs: 1465 return ret; 1466 } 1467 1468 static void __exit 1469 ipoib_cleanup_module(void) 1470 { 1471 1472 EVENTHANDLER_DEREGISTER(vlan_config, ipoib_vlan_attach); 1473 EVENTHANDLER_DEREGISTER(vlan_unconfig, ipoib_vlan_detach); 1474 ib_unregister_client(&ipoib_client); 1475 ib_sa_unregister_client(&ipoib_sa_client); 1476 destroy_workqueue(ipoib_workqueue); 1477 } 1478 1479 /* 1480 * Infiniband output routine. 1481 */ 1482 static int 1483 ipoib_output(struct ifnet *ifp, struct mbuf *m, 1484 const struct sockaddr *dst, struct route *ro) 1485 { 1486 u_char edst[INFINIBAND_ALEN]; 1487 #if defined(INET) || defined(INET6) 1488 struct llentry *lle = NULL; 1489 #endif 1490 struct ipoib_header *eh; 1491 int error = 0, is_gw = 0; 1492 short type; 1493 1494 NET_EPOCH_ASSERT(); 1495 1496 if (ro != NULL) 1497 is_gw = (ro->ro_flags & RT_HAS_GW) != 0; 1498 #ifdef MAC 1499 error = mac_ifnet_check_transmit(ifp, m); 1500 if (error) 1501 goto bad; 1502 #endif 1503 1504 M_PROFILE(m); 1505 if (ifp->if_flags & IFF_MONITOR) { 1506 error = ENETDOWN; 1507 goto bad; 1508 } 1509 if (!((ifp->if_flags & IFF_UP) && 1510 (ifp->if_drv_flags & IFF_DRV_RUNNING))) { 1511 error = ENETDOWN; 1512 goto bad; 1513 } 1514 1515 switch (dst->sa_family) { 1516 #ifdef INET 1517 case AF_INET: 1518 if (lle != NULL && (lle->la_flags & LLE_VALID)) 1519 memcpy(edst, lle->ll_addr, sizeof(edst)); 1520 else if (m->m_flags & M_MCAST) 1521 ip_ib_mc_map(((struct sockaddr_in *)dst)->sin_addr.s_addr, ifp->if_broadcastaddr, edst); 1522 else 1523 error = arpresolve(ifp, is_gw, m, dst, edst, NULL, NULL); 1524 if (error) 1525 return (error == EWOULDBLOCK ? 0 : error); 1526 type = htons(ETHERTYPE_IP); 1527 break; 1528 case AF_ARP: 1529 { 1530 struct arphdr *ah; 1531 ah = mtod(m, struct arphdr *); 1532 ah->ar_hrd = htons(ARPHRD_INFINIBAND); 1533 1534 switch(ntohs(ah->ar_op)) { 1535 case ARPOP_REVREQUEST: 1536 case ARPOP_REVREPLY: 1537 type = htons(ETHERTYPE_REVARP); 1538 break; 1539 case ARPOP_REQUEST: 1540 case ARPOP_REPLY: 1541 default: 1542 type = htons(ETHERTYPE_ARP); 1543 break; 1544 } 1545 1546 if (m->m_flags & M_BCAST) 1547 bcopy(ifp->if_broadcastaddr, edst, INFINIBAND_ALEN); 1548 else 1549 bcopy(ar_tha(ah), edst, INFINIBAND_ALEN); 1550 1551 } 1552 break; 1553 #endif 1554 #ifdef INET6 1555 case AF_INET6: 1556 if (lle != NULL && (lle->la_flags & LLE_VALID)) 1557 memcpy(edst, lle->ll_addr, sizeof(edst)); 1558 else if (m->m_flags & M_MCAST) 1559 ipv6_ib_mc_map(&((struct sockaddr_in6 *)dst)->sin6_addr, ifp->if_broadcastaddr, edst); 1560 else 1561 error = nd6_resolve(ifp, is_gw, m, dst, edst, NULL, NULL); 1562 if (error) 1563 return error; 1564 type = htons(ETHERTYPE_IPV6); 1565 break; 1566 #endif 1567 1568 default: 1569 if_printf(ifp, "can't handle af%d\n", dst->sa_family); 1570 error = EAFNOSUPPORT; 1571 goto bad; 1572 } 1573 1574 /* 1575 * Add local net header. If no space in first mbuf, 1576 * allocate another. 1577 */ 1578 M_PREPEND(m, IPOIB_HEADER_LEN, M_NOWAIT); 1579 if (m == NULL) { 1580 error = ENOBUFS; 1581 goto bad; 1582 } 1583 eh = mtod(m, struct ipoib_header *); 1584 (void)memcpy(&eh->proto, &type, sizeof(eh->proto)); 1585 (void)memcpy(&eh->hwaddr, edst, sizeof (edst)); 1586 1587 /* 1588 * Queue message on interface, update output statistics if 1589 * successful, and start output if interface not yet active. 1590 */ 1591 return ((ifp->if_transmit)(ifp, m)); 1592 bad: 1593 if (m != NULL) 1594 m_freem(m); 1595 return (error); 1596 } 1597 1598 /* 1599 * Upper layer processing for a received Infiniband packet. 1600 */ 1601 void 1602 ipoib_demux(struct ifnet *ifp, struct mbuf *m, u_short proto) 1603 { 1604 struct epoch_tracker et; 1605 int isr; 1606 1607 #ifdef MAC 1608 /* 1609 * Tag the mbuf with an appropriate MAC label before any other 1610 * consumers can get to it. 1611 */ 1612 mac_ifnet_create_mbuf(ifp, m); 1613 #endif 1614 /* Allow monitor mode to claim this frame, after stats are updated. */ 1615 if (ifp->if_flags & IFF_MONITOR) { 1616 if_printf(ifp, "discard frame at IFF_MONITOR\n"); 1617 m_freem(m); 1618 return; 1619 } 1620 /* 1621 * Dispatch frame to upper layer. 1622 */ 1623 switch (proto) { 1624 #ifdef INET 1625 case ETHERTYPE_IP: 1626 isr = NETISR_IP; 1627 break; 1628 1629 case ETHERTYPE_ARP: 1630 if (ifp->if_flags & IFF_NOARP) { 1631 /* Discard packet if ARP is disabled on interface */ 1632 m_freem(m); 1633 return; 1634 } 1635 isr = NETISR_ARP; 1636 break; 1637 #endif 1638 #ifdef INET6 1639 case ETHERTYPE_IPV6: 1640 isr = NETISR_IPV6; 1641 break; 1642 #endif 1643 default: 1644 goto discard; 1645 } 1646 NET_EPOCH_ENTER(et); 1647 netisr_dispatch(isr, m); 1648 NET_EPOCH_EXIT(et); 1649 return; 1650 1651 discard: 1652 m_freem(m); 1653 } 1654 1655 /* 1656 * Process a received Infiniband packet. 1657 */ 1658 static void 1659 ipoib_input(struct ifnet *ifp, struct mbuf *m) 1660 { 1661 struct ipoib_header *eh; 1662 1663 if ((ifp->if_flags & IFF_UP) == 0) { 1664 m_freem(m); 1665 return; 1666 } 1667 CURVNET_SET_QUIET(ifp->if_vnet); 1668 1669 /* Let BPF have it before we strip the header. */ 1670 IPOIB_MTAP(ifp, m); 1671 eh = mtod(m, struct ipoib_header *); 1672 /* 1673 * Reset layer specific mbuf flags to avoid confusing upper layers. 1674 * Strip off Infiniband header. 1675 */ 1676 m->m_flags &= ~M_VLANTAG; 1677 m_clrprotoflags(m); 1678 m_adj(m, IPOIB_HEADER_LEN); 1679 1680 if (IPOIB_IS_MULTICAST(eh->hwaddr)) { 1681 if (memcmp(eh->hwaddr, ifp->if_broadcastaddr, 1682 ifp->if_addrlen) == 0) 1683 m->m_flags |= M_BCAST; 1684 else 1685 m->m_flags |= M_MCAST; 1686 if_inc_counter(ifp, IFCOUNTER_IMCASTS, 1); 1687 } 1688 1689 ipoib_demux(ifp, m, ntohs(eh->proto)); 1690 CURVNET_RESTORE(); 1691 } 1692 1693 static int 1694 ipoib_resolvemulti(struct ifnet *ifp, struct sockaddr **llsa, 1695 struct sockaddr *sa) 1696 { 1697 struct sockaddr_dl *sdl; 1698 #ifdef INET 1699 struct sockaddr_in *sin; 1700 #endif 1701 #ifdef INET6 1702 struct sockaddr_in6 *sin6; 1703 #endif 1704 u_char *e_addr; 1705 1706 switch(sa->sa_family) { 1707 case AF_LINK: 1708 /* 1709 * No mapping needed. Just check that it's a valid MC address. 1710 */ 1711 sdl = (struct sockaddr_dl *)sa; 1712 e_addr = LLADDR(sdl); 1713 if (!IPOIB_IS_MULTICAST(e_addr)) 1714 return EADDRNOTAVAIL; 1715 *llsa = NULL; 1716 return 0; 1717 1718 #ifdef INET 1719 case AF_INET: 1720 sin = (struct sockaddr_in *)sa; 1721 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) 1722 return EADDRNOTAVAIL; 1723 sdl = link_init_sdl(ifp, *llsa, IFT_INFINIBAND); 1724 sdl->sdl_alen = INFINIBAND_ALEN; 1725 e_addr = LLADDR(sdl); 1726 ip_ib_mc_map(sin->sin_addr.s_addr, ifp->if_broadcastaddr, 1727 e_addr); 1728 *llsa = (struct sockaddr *)sdl; 1729 return 0; 1730 #endif 1731 #ifdef INET6 1732 case AF_INET6: 1733 sin6 = (struct sockaddr_in6 *)sa; 1734 /* 1735 * An IP6 address of 0 means listen to all 1736 * of the multicast address used for IP6. 1737 * This has no meaning in ipoib. 1738 */ 1739 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 1740 return EADDRNOTAVAIL; 1741 if (!IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 1742 return EADDRNOTAVAIL; 1743 sdl = link_init_sdl(ifp, *llsa, IFT_INFINIBAND); 1744 sdl->sdl_alen = INFINIBAND_ALEN; 1745 e_addr = LLADDR(sdl); 1746 ipv6_ib_mc_map(&sin6->sin6_addr, ifp->if_broadcastaddr, e_addr); 1747 *llsa = (struct sockaddr *)sdl; 1748 return 0; 1749 #endif 1750 1751 default: 1752 return EAFNOSUPPORT; 1753 } 1754 } 1755 1756 module_init_order(ipoib_init_module, SI_ORDER_FIFTH); 1757 module_exit_order(ipoib_cleanup_module, SI_ORDER_FIFTH); 1758 1759 static int 1760 ipoib_evhand(module_t mod, int event, void *arg) 1761 { 1762 return (0); 1763 } 1764 1765 static moduledata_t ipoib_mod = { 1766 .name = "ipoib", 1767 .evhand = ipoib_evhand, 1768 }; 1769 1770 DECLARE_MODULE(ipoib, ipoib_mod, SI_SUB_LAST, SI_ORDER_ANY); 1771 MODULE_DEPEND(ipoib, ibcore, 1, 1, 1); 1772 MODULE_DEPEND(ipoib, linuxkpi, 1, 1, 1); 1773