1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Userspace interface 4 * Linux ethernet bridge 5 * 6 * Authors: 7 * Lennert Buytenhek <buytenh@gnu.org> 8 */ 9 10 #include <linux/kernel.h> 11 #include <linux/netdevice.h> 12 #include <linux/etherdevice.h> 13 #include <linux/netpoll.h> 14 #include <linux/ethtool.h> 15 #include <linux/if_arp.h> 16 #include <linux/module.h> 17 #include <linux/init.h> 18 #include <linux/rtnetlink.h> 19 #include <linux/if_ether.h> 20 #include <linux/slab.h> 21 #include <net/dsa.h> 22 #include <net/sock.h> 23 #include <linux/if_vlan.h> 24 #include <net/switchdev.h> 25 #include <net/net_namespace.h> 26 27 #include "br_private.h" 28 29 /* 30 * Determine initial path cost based on speed. 31 * using recommendations from 802.1d standard 32 * 33 * Since driver might sleep need to not be holding any locks. 34 */ 35 static int port_cost(struct net_device *dev) 36 { 37 struct ethtool_link_ksettings ecmd; 38 39 if (!__ethtool_get_link_ksettings(dev, &ecmd)) { 40 switch (ecmd.base.speed) { 41 case SPEED_10000: 42 return 2; 43 case SPEED_5000: 44 return 3; 45 case SPEED_2500: 46 return 4; 47 case SPEED_1000: 48 return 5; 49 case SPEED_100: 50 return 19; 51 case SPEED_10: 52 return 100; 53 case SPEED_UNKNOWN: 54 return 100; 55 default: 56 if (ecmd.base.speed > SPEED_10000) 57 return 1; 58 } 59 } 60 61 /* Old silly heuristics based on name */ 62 if (!strncmp(dev->name, "lec", 3)) 63 return 7; 64 65 if (!strncmp(dev->name, "plip", 4)) 66 return 2500; 67 68 return 100; /* assume old 10Mbps */ 69 } 70 71 72 /* Check for port carrier transitions. */ 73 void br_port_carrier_check(struct net_bridge_port *p, bool *notified) 74 { 75 struct net_device *dev = p->dev; 76 struct net_bridge *br = p->br; 77 78 if (!(p->flags & BR_ADMIN_COST) && 79 netif_running(dev) && netif_oper_up(dev)) 80 p->path_cost = port_cost(dev); 81 82 *notified = false; 83 if (!netif_running(br->dev)) 84 return; 85 86 spin_lock_bh(&br->lock); 87 if (netif_running(dev) && netif_oper_up(dev)) { 88 if (p->state == BR_STATE_DISABLED) { 89 br_stp_enable_port(p); 90 *notified = true; 91 } 92 } else { 93 if (p->state != BR_STATE_DISABLED) { 94 br_stp_disable_port(p); 95 *notified = true; 96 } 97 } 98 spin_unlock_bh(&br->lock); 99 } 100 101 static void br_port_set_promisc(struct net_bridge_port *p) 102 { 103 int err = 0; 104 105 if (br_promisc_port(p)) 106 return; 107 108 err = dev_set_promiscuity(p->dev, 1); 109 if (err) 110 return; 111 112 br_fdb_unsync_static(p->br, p); 113 p->flags |= BR_PROMISC; 114 } 115 116 static void br_port_clear_promisc(struct net_bridge_port *p) 117 { 118 int err; 119 120 /* Check if the port is already non-promisc or if it doesn't 121 * support UNICAST filtering. Without unicast filtering support 122 * we'll end up re-enabling promisc mode anyway, so just check for 123 * it here. 124 */ 125 if (!br_promisc_port(p) || !(p->dev->priv_flags & IFF_UNICAST_FLT)) 126 return; 127 128 /* Since we'll be clearing the promisc mode, program the port 129 * first so that we don't have interruption in traffic. 130 */ 131 err = br_fdb_sync_static(p->br, p); 132 if (err) 133 return; 134 135 dev_set_promiscuity(p->dev, -1); 136 p->flags &= ~BR_PROMISC; 137 } 138 139 /* When a port is added or removed or when certain port flags 140 * change, this function is called to automatically manage 141 * promiscuity setting of all the bridge ports. We are always called 142 * under RTNL so can skip using rcu primitives. 143 */ 144 void br_manage_promisc(struct net_bridge *br) 145 { 146 struct net_bridge_port *p; 147 bool set_all = false; 148 149 /* If vlan filtering is disabled or bridge interface is placed 150 * into promiscuous mode, place all ports in promiscuous mode. 151 */ 152 if ((br->dev->flags & IFF_PROMISC) || !br_vlan_enabled(br->dev)) 153 set_all = true; 154 155 list_for_each_entry(p, &br->port_list, list) { 156 if (set_all) { 157 br_port_set_promisc(p); 158 } else { 159 /* If the number of auto-ports is <= 1, then all other 160 * ports will have their output configuration 161 * statically specified through fdbs. Since ingress 162 * on the auto-port becomes forwarding/egress to other 163 * ports and egress configuration is statically known, 164 * we can say that ingress configuration of the 165 * auto-port is also statically known. 166 * This lets us disable promiscuous mode and write 167 * this config to hw. 168 */ 169 if ((p->dev->priv_flags & IFF_UNICAST_FLT) && 170 (br->auto_cnt == 0 || 171 (br->auto_cnt == 1 && br_auto_port(p)))) 172 br_port_clear_promisc(p); 173 else 174 br_port_set_promisc(p); 175 } 176 } 177 } 178 179 int nbp_backup_change(struct net_bridge_port *p, 180 struct net_device *backup_dev) 181 { 182 struct net_bridge_port *old_backup = rtnl_dereference(p->backup_port); 183 struct net_bridge_port *backup_p = NULL; 184 185 ASSERT_RTNL(); 186 187 if (backup_dev) { 188 if (!netif_is_bridge_port(backup_dev)) 189 return -ENOENT; 190 191 backup_p = br_port_get_rtnl(backup_dev); 192 if (backup_p->br != p->br) 193 return -EINVAL; 194 } 195 196 if (p == backup_p) 197 return -EINVAL; 198 199 if (old_backup == backup_p) 200 return 0; 201 202 /* if the backup link is already set, clear it */ 203 if (old_backup) 204 old_backup->backup_redirected_cnt--; 205 206 if (backup_p) 207 backup_p->backup_redirected_cnt++; 208 rcu_assign_pointer(p->backup_port, backup_p); 209 210 return 0; 211 } 212 213 static void nbp_backup_clear(struct net_bridge_port *p) 214 { 215 nbp_backup_change(p, NULL); 216 if (p->backup_redirected_cnt) { 217 struct net_bridge_port *cur_p; 218 219 list_for_each_entry(cur_p, &p->br->port_list, list) { 220 struct net_bridge_port *backup_p; 221 222 backup_p = rtnl_dereference(cur_p->backup_port); 223 if (backup_p == p) 224 nbp_backup_change(cur_p, NULL); 225 } 226 } 227 228 WARN_ON(rcu_access_pointer(p->backup_port) || p->backup_redirected_cnt); 229 } 230 231 static void nbp_update_port_count(struct net_bridge *br) 232 { 233 struct net_bridge_port *p; 234 u32 cnt = 0; 235 236 list_for_each_entry(p, &br->port_list, list) { 237 if (br_auto_port(p)) 238 cnt++; 239 } 240 if (br->auto_cnt != cnt) { 241 br->auto_cnt = cnt; 242 br_manage_promisc(br); 243 } 244 } 245 246 static void nbp_delete_promisc(struct net_bridge_port *p) 247 { 248 /* If port is currently promiscuous, unset promiscuity. 249 * Otherwise, it is a static port so remove all addresses 250 * from it. 251 */ 252 dev_set_allmulti(p->dev, -1); 253 if (br_promisc_port(p)) 254 dev_set_promiscuity(p->dev, -1); 255 else 256 br_fdb_unsync_static(p->br, p); 257 } 258 259 static void release_nbp(struct kobject *kobj) 260 { 261 struct net_bridge_port *p 262 = container_of(kobj, struct net_bridge_port, kobj); 263 kfree(p); 264 } 265 266 static void brport_get_ownership(const struct kobject *kobj, kuid_t *uid, kgid_t *gid) 267 { 268 struct net_bridge_port *p = kobj_to_brport(kobj); 269 270 net_ns_get_ownership(dev_net(p->dev), uid, gid); 271 } 272 273 static const struct kobj_type brport_ktype = { 274 #ifdef CONFIG_SYSFS 275 .sysfs_ops = &brport_sysfs_ops, 276 #endif 277 .release = release_nbp, 278 .get_ownership = brport_get_ownership, 279 }; 280 281 static void destroy_nbp(struct net_bridge_port *p) 282 { 283 struct net_device *dev = p->dev; 284 285 p->br = NULL; 286 p->dev = NULL; 287 netdev_put(dev, &p->dev_tracker); 288 289 kobject_put(&p->kobj); 290 } 291 292 static void destroy_nbp_rcu(struct rcu_head *head) 293 { 294 struct net_bridge_port *p = 295 container_of(head, struct net_bridge_port, rcu); 296 destroy_nbp(p); 297 } 298 299 static unsigned get_max_headroom(struct net_bridge *br) 300 { 301 unsigned max_headroom = 0; 302 struct net_bridge_port *p; 303 304 list_for_each_entry(p, &br->port_list, list) { 305 unsigned dev_headroom = netdev_get_fwd_headroom(p->dev); 306 307 if (dev_headroom > max_headroom) 308 max_headroom = dev_headroom; 309 } 310 311 return max_headroom; 312 } 313 314 static void update_headroom(struct net_bridge *br, int new_hr) 315 { 316 struct net_bridge_port *p; 317 318 list_for_each_entry(p, &br->port_list, list) 319 netdev_set_rx_headroom(p->dev, new_hr); 320 321 br->dev->needed_headroom = new_hr; 322 } 323 324 /* Delete port(interface) from bridge is done in two steps. 325 * via RCU. First step, marks device as down. That deletes 326 * all the timers and stops new packets from flowing through. 327 * 328 * Final cleanup doesn't occur until after all CPU's finished 329 * processing packets. 330 * 331 * Protected from multiple admin operations by RTNL mutex 332 */ 333 static void del_nbp(struct net_bridge_port *p) 334 { 335 struct net_bridge *br = p->br; 336 struct net_device *dev = p->dev; 337 338 sysfs_remove_link(br->ifobj, p->dev->name); 339 340 nbp_delete_promisc(p); 341 342 spin_lock_bh(&br->lock); 343 br_stp_disable_port(p); 344 spin_unlock_bh(&br->lock); 345 346 br_mrp_port_del(br, p); 347 br_cfm_port_del(br, p); 348 349 br_ifinfo_notify(RTM_DELLINK, NULL, p); 350 351 list_del_rcu(&p->list); 352 if (netdev_get_fwd_headroom(dev) == br->dev->needed_headroom) 353 update_headroom(br, get_max_headroom(br)); 354 netdev_reset_rx_headroom(dev); 355 356 nbp_vlan_flush(p); 357 br_fdb_delete_by_port(br, p, 0, 1); 358 switchdev_deferred_process(); 359 nbp_backup_clear(p); 360 361 nbp_update_port_count(br); 362 363 netdev_upper_dev_unlink(dev, br->dev); 364 365 dev->priv_flags &= ~IFF_BRIDGE_PORT; 366 367 netdev_rx_handler_unregister(dev); 368 369 br_multicast_del_port(p); 370 371 kobject_uevent(&p->kobj, KOBJ_REMOVE); 372 kobject_del(&p->kobj); 373 374 br_netpoll_disable(p); 375 376 call_rcu(&p->rcu, destroy_nbp_rcu); 377 } 378 379 /* Delete bridge device */ 380 void br_dev_delete(struct net_device *dev, struct list_head *head) 381 { 382 struct net_bridge *br = netdev_priv(dev); 383 struct net_bridge_port *p, *n; 384 385 list_for_each_entry_safe(p, n, &br->port_list, list) { 386 del_nbp(p); 387 } 388 389 br_mst_uninit(br); 390 br_recalculate_neigh_suppress_enabled(br); 391 392 br_fdb_delete_by_port(br, NULL, 0, 1); 393 394 cancel_delayed_work_sync(&br->gc_work); 395 396 br_sysfs_delbr(br->dev); 397 unregister_netdevice_queue(br->dev, head); 398 } 399 400 /* find an available port number */ 401 static int find_portno(struct net_bridge *br) 402 { 403 int index; 404 struct net_bridge_port *p; 405 unsigned long *inuse; 406 407 inuse = bitmap_zalloc(BR_MAX_PORTS, GFP_KERNEL); 408 if (!inuse) 409 return -ENOMEM; 410 411 __set_bit(0, inuse); /* zero is reserved */ 412 list_for_each_entry(p, &br->port_list, list) 413 __set_bit(p->port_no, inuse); 414 415 index = find_first_zero_bit(inuse, BR_MAX_PORTS); 416 bitmap_free(inuse); 417 418 return (index >= BR_MAX_PORTS) ? -EXFULL : index; 419 } 420 421 /* called with RTNL but without bridge lock */ 422 static struct net_bridge_port *new_nbp(struct net_bridge *br, 423 struct net_device *dev) 424 { 425 struct net_bridge_port *p; 426 int index, err; 427 428 index = find_portno(br); 429 if (index < 0) 430 return ERR_PTR(index); 431 432 p = kzalloc(sizeof(*p), GFP_KERNEL); 433 if (p == NULL) 434 return ERR_PTR(-ENOMEM); 435 436 p->br = br; 437 netdev_hold(dev, &p->dev_tracker, GFP_KERNEL); 438 p->dev = dev; 439 p->path_cost = port_cost(dev); 440 p->priority = 0x8000 >> BR_PORT_BITS; 441 p->port_no = index; 442 p->flags = BR_LEARNING | BR_FLOOD | BR_MCAST_FLOOD | BR_BCAST_FLOOD; 443 br_init_port(p); 444 br_set_state(p, BR_STATE_DISABLED); 445 br_stp_port_timer_init(p); 446 err = br_multicast_add_port(p); 447 if (err) { 448 netdev_put(dev, &p->dev_tracker); 449 kfree(p); 450 p = ERR_PTR(err); 451 } 452 453 return p; 454 } 455 456 int br_add_bridge(struct net *net, const char *name) 457 { 458 struct net_device *dev; 459 int res; 460 461 dev = alloc_netdev(sizeof(struct net_bridge), name, NET_NAME_UNKNOWN, 462 br_dev_setup); 463 464 if (!dev) 465 return -ENOMEM; 466 467 dev_net_set(dev, net); 468 dev->rtnl_link_ops = &br_link_ops; 469 470 res = register_netdevice(dev); 471 if (res) 472 free_netdev(dev); 473 return res; 474 } 475 476 int br_del_bridge(struct net *net, const char *name) 477 { 478 struct net_device *dev; 479 int ret = 0; 480 481 dev = __dev_get_by_name(net, name); 482 if (dev == NULL) 483 ret = -ENXIO; /* Could not find device */ 484 485 else if (!netif_is_bridge_master(dev)) { 486 /* Attempt to delete non bridge device! */ 487 ret = -EPERM; 488 } 489 490 else if (dev->flags & IFF_UP) { 491 /* Not shutdown yet. */ 492 ret = -EBUSY; 493 } 494 495 else 496 br_dev_delete(dev, NULL); 497 498 return ret; 499 } 500 501 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */ 502 static int br_mtu_min(const struct net_bridge *br) 503 { 504 const struct net_bridge_port *p; 505 int ret_mtu = 0; 506 507 list_for_each_entry(p, &br->port_list, list) 508 if (!ret_mtu || ret_mtu > p->dev->mtu) 509 ret_mtu = p->dev->mtu; 510 511 return ret_mtu ? ret_mtu : ETH_DATA_LEN; 512 } 513 514 void br_mtu_auto_adjust(struct net_bridge *br) 515 { 516 ASSERT_RTNL(); 517 518 /* if the bridge MTU was manually configured don't mess with it */ 519 if (br_opt_get(br, BROPT_MTU_SET_BY_USER)) 520 return; 521 522 /* change to the minimum MTU and clear the flag which was set by 523 * the bridge ndo_change_mtu callback 524 */ 525 dev_set_mtu(br->dev, br_mtu_min(br)); 526 br_opt_toggle(br, BROPT_MTU_SET_BY_USER, false); 527 } 528 529 /* 530 * Recomputes features using slave's features 531 */ 532 netdev_features_t br_features_recompute(struct net_bridge *br, 533 netdev_features_t features) 534 { 535 struct net_bridge_port *p; 536 netdev_features_t mask; 537 538 if (list_empty(&br->port_list)) 539 return features; 540 541 mask = features; 542 features &= ~NETIF_F_ONE_FOR_ALL; 543 544 list_for_each_entry(p, &br->port_list, list) { 545 features = netdev_increment_features(features, 546 p->dev->features, mask); 547 } 548 features = netdev_add_tso_features(features, mask); 549 550 return features; 551 } 552 553 /* called with RTNL */ 554 int br_add_if(struct net_bridge *br, struct net_device *dev, 555 struct netlink_ext_ack *extack) 556 { 557 struct net_bridge_port *p; 558 int err = 0; 559 unsigned br_hr, dev_hr; 560 bool changed_addr, fdb_synced = false; 561 562 /* Don't allow bridging non-ethernet like devices. */ 563 if ((dev->flags & IFF_LOOPBACK) || 564 dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN || 565 !is_valid_ether_addr(dev->dev_addr)) 566 return -EINVAL; 567 568 /* No bridging of bridges */ 569 if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit) { 570 NL_SET_ERR_MSG(extack, 571 "Can not enslave a bridge to a bridge"); 572 return -ELOOP; 573 } 574 575 /* Device has master upper dev */ 576 if (netdev_master_upper_dev_get(dev)) 577 return -EBUSY; 578 579 /* No bridging devices that dislike that (e.g. wireless) */ 580 if (dev->priv_flags & IFF_DONT_BRIDGE) { 581 NL_SET_ERR_MSG(extack, 582 "Device does not allow enslaving to a bridge"); 583 return -EOPNOTSUPP; 584 } 585 586 p = new_nbp(br, dev); 587 if (IS_ERR(p)) 588 return PTR_ERR(p); 589 590 call_netdevice_notifiers(NETDEV_JOIN, dev); 591 592 err = dev_set_allmulti(dev, 1); 593 if (err) { 594 br_multicast_del_port(p); 595 netdev_put(dev, &p->dev_tracker); 596 kfree(p); /* kobject not yet init'd, manually free */ 597 goto err1; 598 } 599 600 err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj), 601 SYSFS_BRIDGE_PORT_ATTR); 602 if (err) 603 goto err2; 604 605 err = br_sysfs_addif(p); 606 if (err) 607 goto err2; 608 609 err = br_netpoll_enable(p); 610 if (err) 611 goto err3; 612 613 err = netdev_rx_handler_register(dev, br_get_rx_handler(dev), p); 614 if (err) 615 goto err4; 616 617 dev->priv_flags |= IFF_BRIDGE_PORT; 618 619 err = netdev_master_upper_dev_link(dev, br->dev, NULL, NULL, extack); 620 if (err) 621 goto err5; 622 623 dev_disable_lro(dev); 624 625 list_add_rcu(&p->list, &br->port_list); 626 627 nbp_update_port_count(br); 628 if (!br_promisc_port(p) && (p->dev->priv_flags & IFF_UNICAST_FLT)) { 629 /* When updating the port count we also update all ports' 630 * promiscuous mode. 631 * A port leaving promiscuous mode normally gets the bridge's 632 * fdb synced to the unicast filter (if supported), however, 633 * `br_port_clear_promisc` does not distinguish between 634 * non-promiscuous ports and *new* ports, so we need to 635 * sync explicitly here. 636 */ 637 fdb_synced = br_fdb_sync_static(br, p) == 0; 638 if (!fdb_synced) 639 netdev_err(dev, "failed to sync bridge static fdb addresses to this port\n"); 640 } 641 642 br_hr = br->dev->needed_headroom; 643 dev_hr = netdev_get_fwd_headroom(dev); 644 if (br_hr < dev_hr) 645 update_headroom(br, dev_hr); 646 else 647 netdev_set_rx_headroom(dev, br_hr); 648 649 if (br_fdb_add_local(br, p, dev->dev_addr, 0)) 650 netdev_err(dev, "failed insert local address bridge forwarding table\n"); 651 652 if (br->dev->addr_assign_type != NET_ADDR_SET) { 653 /* Ask for permission to use this MAC address now, even if we 654 * don't end up choosing it below. 655 */ 656 err = netif_pre_changeaddr_notify(br->dev, dev->dev_addr, 657 extack); 658 if (err) 659 goto err6; 660 } 661 662 err = nbp_vlan_init(p, extack); 663 if (err) { 664 netdev_err(dev, "failed to initialize vlan filtering on this port\n"); 665 goto err6; 666 } 667 668 spin_lock_bh(&br->lock); 669 changed_addr = br_stp_recalculate_bridge_id(br); 670 671 if (netif_running(dev) && netif_oper_up(dev) && 672 (br->dev->flags & IFF_UP)) 673 br_stp_enable_port(p); 674 spin_unlock_bh(&br->lock); 675 676 br_ifinfo_notify(RTM_NEWLINK, NULL, p); 677 678 if (changed_addr) 679 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev); 680 681 br_mtu_auto_adjust(br); 682 683 netdev_compute_master_upper_features(br->dev, false); 684 685 kobject_uevent(&p->kobj, KOBJ_ADD); 686 687 return 0; 688 689 err6: 690 if (fdb_synced) 691 br_fdb_unsync_static(br, p); 692 list_del_rcu(&p->list); 693 br_fdb_delete_by_port(br, p, 0, 1); 694 nbp_update_port_count(br); 695 netdev_upper_dev_unlink(dev, br->dev); 696 err5: 697 dev->priv_flags &= ~IFF_BRIDGE_PORT; 698 netdev_rx_handler_unregister(dev); 699 err4: 700 br_netpoll_disable(p); 701 err3: 702 sysfs_remove_link(br->ifobj, p->dev->name); 703 err2: 704 br_multicast_del_port(p); 705 netdev_put(dev, &p->dev_tracker); 706 kobject_put(&p->kobj); 707 dev_set_allmulti(dev, -1); 708 err1: 709 return err; 710 } 711 712 /* called with RTNL */ 713 int br_del_if(struct net_bridge *br, struct net_device *dev) 714 { 715 struct net_bridge_port *p; 716 bool changed_addr; 717 718 p = br_port_get_rtnl(dev); 719 if (!p || p->br != br) 720 return -EINVAL; 721 722 /* Since more than one interface can be attached to a bridge, 723 * there still maybe an alternate path for netconsole to use; 724 * therefore there is no reason for a NETDEV_RELEASE event. 725 */ 726 del_nbp(p); 727 728 br_mtu_auto_adjust(br); 729 730 spin_lock_bh(&br->lock); 731 changed_addr = br_stp_recalculate_bridge_id(br); 732 spin_unlock_bh(&br->lock); 733 734 if (changed_addr) 735 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev); 736 737 netdev_compute_master_upper_features(br->dev, false); 738 739 return 0; 740 } 741 742 void br_port_flags_change(struct net_bridge_port *p, unsigned long mask) 743 { 744 struct net_bridge *br = p->br; 745 746 if (mask & BR_AUTO_MASK) 747 nbp_update_port_count(br); 748 749 if (mask & (BR_NEIGH_SUPPRESS | BR_NEIGH_VLAN_SUPPRESS)) 750 br_recalculate_neigh_suppress_enabled(br); 751 } 752 753 bool br_port_flag_is_set(const struct net_device *dev, unsigned long flag) 754 { 755 struct net_bridge_port *p; 756 757 p = br_port_get_rtnl_rcu(dev); 758 if (!p) 759 return false; 760 761 return p->flags & flag; 762 } 763 EXPORT_SYMBOL_GPL(br_port_flag_is_set); 764