1 /* 2 * Userspace interface 3 * Linux ethernet bridge 4 * 5 * Authors: 6 * Lennert Buytenhek <buytenh@gnu.org> 7 * 8 * This program is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU General Public License 10 * as published by the Free Software Foundation; either version 11 * 2 of the License, or (at your option) any later version. 12 */ 13 14 #include <linux/kernel.h> 15 #include <linux/netdevice.h> 16 #include <linux/etherdevice.h> 17 #include <linux/netpoll.h> 18 #include <linux/ethtool.h> 19 #include <linux/if_arp.h> 20 #include <linux/module.h> 21 #include <linux/init.h> 22 #include <linux/rtnetlink.h> 23 #include <linux/if_ether.h> 24 #include <linux/slab.h> 25 #include <net/dsa.h> 26 #include <net/sock.h> 27 #include <linux/if_vlan.h> 28 #include <net/switchdev.h> 29 30 #include "br_private.h" 31 32 /* 33 * Determine initial path cost based on speed. 34 * using recommendations from 802.1d standard 35 * 36 * Since driver might sleep need to not be holding any locks. 37 */ 38 static int port_cost(struct net_device *dev) 39 { 40 struct ethtool_link_ksettings ecmd; 41 42 if (!__ethtool_get_link_ksettings(dev, &ecmd)) { 43 switch (ecmd.base.speed) { 44 case SPEED_10000: 45 return 2; 46 case SPEED_1000: 47 return 4; 48 case SPEED_100: 49 return 19; 50 case SPEED_10: 51 return 100; 52 } 53 } 54 55 /* Old silly heuristics based on name */ 56 if (!strncmp(dev->name, "lec", 3)) 57 return 7; 58 59 if (!strncmp(dev->name, "plip", 4)) 60 return 2500; 61 62 return 100; /* assume old 10Mbps */ 63 } 64 65 66 /* Check for port carrier transitions. */ 67 void br_port_carrier_check(struct net_bridge_port *p) 68 { 69 struct net_device *dev = p->dev; 70 struct net_bridge *br = p->br; 71 72 if (!(p->flags & BR_ADMIN_COST) && 73 netif_running(dev) && netif_oper_up(dev)) 74 p->path_cost = port_cost(dev); 75 76 if (!netif_running(br->dev)) 77 return; 78 79 spin_lock_bh(&br->lock); 80 if (netif_running(dev) && netif_oper_up(dev)) { 81 if (p->state == BR_STATE_DISABLED) 82 br_stp_enable_port(p); 83 } else { 84 if (p->state != BR_STATE_DISABLED) 85 br_stp_disable_port(p); 86 } 87 spin_unlock_bh(&br->lock); 88 } 89 90 static void br_port_set_promisc(struct net_bridge_port *p) 91 { 92 int err = 0; 93 94 if (br_promisc_port(p)) 95 return; 96 97 err = dev_set_promiscuity(p->dev, 1); 98 if (err) 99 return; 100 101 br_fdb_unsync_static(p->br, p); 102 p->flags |= BR_PROMISC; 103 } 104 105 static void br_port_clear_promisc(struct net_bridge_port *p) 106 { 107 int err; 108 109 /* Check if the port is already non-promisc or if it doesn't 110 * support UNICAST filtering. Without unicast filtering support 111 * we'll end up re-enabling promisc mode anyway, so just check for 112 * it here. 113 */ 114 if (!br_promisc_port(p) || !(p->dev->priv_flags & IFF_UNICAST_FLT)) 115 return; 116 117 /* Since we'll be clearing the promisc mode, program the port 118 * first so that we don't have interruption in traffic. 119 */ 120 err = br_fdb_sync_static(p->br, p); 121 if (err) 122 return; 123 124 dev_set_promiscuity(p->dev, -1); 125 p->flags &= ~BR_PROMISC; 126 } 127 128 /* When a port is added or removed or when certain port flags 129 * change, this function is called to automatically manage 130 * promiscuity setting of all the bridge ports. We are always called 131 * under RTNL so can skip using rcu primitives. 132 */ 133 void br_manage_promisc(struct net_bridge *br) 134 { 135 struct net_bridge_port *p; 136 bool set_all = false; 137 138 /* If vlan filtering is disabled or bridge interface is placed 139 * into promiscuous mode, place all ports in promiscuous mode. 140 */ 141 if ((br->dev->flags & IFF_PROMISC) || !br_vlan_enabled(br->dev)) 142 set_all = true; 143 144 list_for_each_entry(p, &br->port_list, list) { 145 if (set_all) { 146 br_port_set_promisc(p); 147 } else { 148 /* If the number of auto-ports is <= 1, then all other 149 * ports will have their output configuration 150 * statically specified through fdbs. Since ingress 151 * on the auto-port becomes forwarding/egress to other 152 * ports and egress configuration is statically known, 153 * we can say that ingress configuration of the 154 * auto-port is also statically known. 155 * This lets us disable promiscuous mode and write 156 * this config to hw. 157 */ 158 if (br->auto_cnt == 0 || 159 (br->auto_cnt == 1 && br_auto_port(p))) 160 br_port_clear_promisc(p); 161 else 162 br_port_set_promisc(p); 163 } 164 } 165 } 166 167 static void nbp_update_port_count(struct net_bridge *br) 168 { 169 struct net_bridge_port *p; 170 u32 cnt = 0; 171 172 list_for_each_entry(p, &br->port_list, list) { 173 if (br_auto_port(p)) 174 cnt++; 175 } 176 if (br->auto_cnt != cnt) { 177 br->auto_cnt = cnt; 178 br_manage_promisc(br); 179 } 180 } 181 182 static void nbp_delete_promisc(struct net_bridge_port *p) 183 { 184 /* If port is currently promiscuous, unset promiscuity. 185 * Otherwise, it is a static port so remove all addresses 186 * from it. 187 */ 188 dev_set_allmulti(p->dev, -1); 189 if (br_promisc_port(p)) 190 dev_set_promiscuity(p->dev, -1); 191 else 192 br_fdb_unsync_static(p->br, p); 193 } 194 195 static void release_nbp(struct kobject *kobj) 196 { 197 struct net_bridge_port *p 198 = container_of(kobj, struct net_bridge_port, kobj); 199 kfree(p); 200 } 201 202 static struct kobj_type brport_ktype = { 203 #ifdef CONFIG_SYSFS 204 .sysfs_ops = &brport_sysfs_ops, 205 #endif 206 .release = release_nbp, 207 }; 208 209 static void destroy_nbp(struct net_bridge_port *p) 210 { 211 struct net_device *dev = p->dev; 212 213 p->br = NULL; 214 p->dev = NULL; 215 dev_put(dev); 216 217 kobject_put(&p->kobj); 218 } 219 220 static void destroy_nbp_rcu(struct rcu_head *head) 221 { 222 struct net_bridge_port *p = 223 container_of(head, struct net_bridge_port, rcu); 224 destroy_nbp(p); 225 } 226 227 static unsigned get_max_headroom(struct net_bridge *br) 228 { 229 unsigned max_headroom = 0; 230 struct net_bridge_port *p; 231 232 list_for_each_entry(p, &br->port_list, list) { 233 unsigned dev_headroom = netdev_get_fwd_headroom(p->dev); 234 235 if (dev_headroom > max_headroom) 236 max_headroom = dev_headroom; 237 } 238 239 return max_headroom; 240 } 241 242 static void update_headroom(struct net_bridge *br, int new_hr) 243 { 244 struct net_bridge_port *p; 245 246 list_for_each_entry(p, &br->port_list, list) 247 netdev_set_rx_headroom(p->dev, new_hr); 248 249 br->dev->needed_headroom = new_hr; 250 } 251 252 /* Delete port(interface) from bridge is done in two steps. 253 * via RCU. First step, marks device as down. That deletes 254 * all the timers and stops new packets from flowing through. 255 * 256 * Final cleanup doesn't occur until after all CPU's finished 257 * processing packets. 258 * 259 * Protected from multiple admin operations by RTNL mutex 260 */ 261 static void del_nbp(struct net_bridge_port *p) 262 { 263 struct net_bridge *br = p->br; 264 struct net_device *dev = p->dev; 265 266 sysfs_remove_link(br->ifobj, p->dev->name); 267 268 nbp_delete_promisc(p); 269 270 spin_lock_bh(&br->lock); 271 br_stp_disable_port(p); 272 spin_unlock_bh(&br->lock); 273 274 br_ifinfo_notify(RTM_DELLINK, p); 275 276 list_del_rcu(&p->list); 277 if (netdev_get_fwd_headroom(dev) == br->dev->needed_headroom) 278 update_headroom(br, get_max_headroom(br)); 279 netdev_reset_rx_headroom(dev); 280 281 nbp_vlan_flush(p); 282 br_fdb_delete_by_port(br, p, 0, 1); 283 switchdev_deferred_process(); 284 285 nbp_update_port_count(br); 286 287 netdev_upper_dev_unlink(dev, br->dev); 288 289 dev->priv_flags &= ~IFF_BRIDGE_PORT; 290 291 netdev_rx_handler_unregister(dev); 292 293 br_multicast_del_port(p); 294 295 kobject_uevent(&p->kobj, KOBJ_REMOVE); 296 kobject_del(&p->kobj); 297 298 br_netpoll_disable(p); 299 300 call_rcu(&p->rcu, destroy_nbp_rcu); 301 } 302 303 /* Delete bridge device */ 304 void br_dev_delete(struct net_device *dev, struct list_head *head) 305 { 306 struct net_bridge *br = netdev_priv(dev); 307 struct net_bridge_port *p, *n; 308 309 list_for_each_entry_safe(p, n, &br->port_list, list) { 310 del_nbp(p); 311 } 312 313 br_fdb_delete_by_port(br, NULL, 0, 1); 314 315 cancel_delayed_work_sync(&br->gc_work); 316 317 br_sysfs_delbr(br->dev); 318 unregister_netdevice_queue(br->dev, head); 319 } 320 321 /* find an available port number */ 322 static int find_portno(struct net_bridge *br) 323 { 324 int index; 325 struct net_bridge_port *p; 326 unsigned long *inuse; 327 328 inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long), 329 GFP_KERNEL); 330 if (!inuse) 331 return -ENOMEM; 332 333 set_bit(0, inuse); /* zero is reserved */ 334 list_for_each_entry(p, &br->port_list, list) { 335 set_bit(p->port_no, inuse); 336 } 337 index = find_first_zero_bit(inuse, BR_MAX_PORTS); 338 kfree(inuse); 339 340 return (index >= BR_MAX_PORTS) ? -EXFULL : index; 341 } 342 343 /* called with RTNL but without bridge lock */ 344 static struct net_bridge_port *new_nbp(struct net_bridge *br, 345 struct net_device *dev) 346 { 347 struct net_bridge_port *p; 348 int index, err; 349 350 index = find_portno(br); 351 if (index < 0) 352 return ERR_PTR(index); 353 354 p = kzalloc(sizeof(*p), GFP_KERNEL); 355 if (p == NULL) 356 return ERR_PTR(-ENOMEM); 357 358 p->br = br; 359 dev_hold(dev); 360 p->dev = dev; 361 p->path_cost = port_cost(dev); 362 p->priority = 0x8000 >> BR_PORT_BITS; 363 p->port_no = index; 364 p->flags = BR_LEARNING | BR_FLOOD | BR_MCAST_FLOOD | BR_BCAST_FLOOD; 365 br_init_port(p); 366 br_set_state(p, BR_STATE_DISABLED); 367 br_stp_port_timer_init(p); 368 err = br_multicast_add_port(p); 369 if (err) { 370 dev_put(dev); 371 kfree(p); 372 p = ERR_PTR(err); 373 } 374 375 return p; 376 } 377 378 int br_add_bridge(struct net *net, const char *name) 379 { 380 struct net_device *dev; 381 int res; 382 383 dev = alloc_netdev(sizeof(struct net_bridge), name, NET_NAME_UNKNOWN, 384 br_dev_setup); 385 386 if (!dev) 387 return -ENOMEM; 388 389 dev_net_set(dev, net); 390 dev->rtnl_link_ops = &br_link_ops; 391 392 res = register_netdev(dev); 393 if (res) 394 free_netdev(dev); 395 return res; 396 } 397 398 int br_del_bridge(struct net *net, const char *name) 399 { 400 struct net_device *dev; 401 int ret = 0; 402 403 rtnl_lock(); 404 dev = __dev_get_by_name(net, name); 405 if (dev == NULL) 406 ret = -ENXIO; /* Could not find device */ 407 408 else if (!(dev->priv_flags & IFF_EBRIDGE)) { 409 /* Attempt to delete non bridge device! */ 410 ret = -EPERM; 411 } 412 413 else if (dev->flags & IFF_UP) { 414 /* Not shutdown yet. */ 415 ret = -EBUSY; 416 } 417 418 else 419 br_dev_delete(dev, NULL); 420 421 rtnl_unlock(); 422 return ret; 423 } 424 425 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */ 426 int br_min_mtu(const struct net_bridge *br) 427 { 428 const struct net_bridge_port *p; 429 int mtu = 0; 430 431 ASSERT_RTNL(); 432 433 if (list_empty(&br->port_list)) 434 mtu = ETH_DATA_LEN; 435 else { 436 list_for_each_entry(p, &br->port_list, list) { 437 if (!mtu || p->dev->mtu < mtu) 438 mtu = p->dev->mtu; 439 } 440 } 441 return mtu; 442 } 443 444 static void br_set_gso_limits(struct net_bridge *br) 445 { 446 unsigned int gso_max_size = GSO_MAX_SIZE; 447 u16 gso_max_segs = GSO_MAX_SEGS; 448 const struct net_bridge_port *p; 449 450 list_for_each_entry(p, &br->port_list, list) { 451 gso_max_size = min(gso_max_size, p->dev->gso_max_size); 452 gso_max_segs = min(gso_max_segs, p->dev->gso_max_segs); 453 } 454 br->dev->gso_max_size = gso_max_size; 455 br->dev->gso_max_segs = gso_max_segs; 456 } 457 458 /* 459 * Recomputes features using slave's features 460 */ 461 netdev_features_t br_features_recompute(struct net_bridge *br, 462 netdev_features_t features) 463 { 464 struct net_bridge_port *p; 465 netdev_features_t mask; 466 467 if (list_empty(&br->port_list)) 468 return features; 469 470 mask = features; 471 features &= ~NETIF_F_ONE_FOR_ALL; 472 473 list_for_each_entry(p, &br->port_list, list) { 474 features = netdev_increment_features(features, 475 p->dev->features, mask); 476 } 477 features = netdev_add_tso_features(features, mask); 478 479 return features; 480 } 481 482 /* called with RTNL */ 483 int br_add_if(struct net_bridge *br, struct net_device *dev) 484 { 485 struct net_bridge_port *p; 486 int err = 0; 487 unsigned br_hr, dev_hr; 488 bool changed_addr; 489 490 /* Don't allow bridging non-ethernet like devices, or DSA-enabled 491 * master network devices since the bridge layer rx_handler prevents 492 * the DSA fake ethertype handler to be invoked, so we do not strip off 493 * the DSA switch tag protocol header and the bridge layer just return 494 * RX_HANDLER_CONSUMED, stopping RX processing for these frames. 495 */ 496 if ((dev->flags & IFF_LOOPBACK) || 497 dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN || 498 !is_valid_ether_addr(dev->dev_addr) || 499 netdev_uses_dsa(dev)) 500 return -EINVAL; 501 502 /* No bridging of bridges */ 503 if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit) 504 return -ELOOP; 505 506 /* Device is already being bridged */ 507 if (br_port_exists(dev)) 508 return -EBUSY; 509 510 /* No bridging devices that dislike that (e.g. wireless) */ 511 if (dev->priv_flags & IFF_DONT_BRIDGE) 512 return -EOPNOTSUPP; 513 514 p = new_nbp(br, dev); 515 if (IS_ERR(p)) 516 return PTR_ERR(p); 517 518 call_netdevice_notifiers(NETDEV_JOIN, dev); 519 520 err = dev_set_allmulti(dev, 1); 521 if (err) 522 goto put_back; 523 524 err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj), 525 SYSFS_BRIDGE_PORT_ATTR); 526 if (err) 527 goto err1; 528 529 err = br_sysfs_addif(p); 530 if (err) 531 goto err2; 532 533 err = br_netpoll_enable(p); 534 if (err) 535 goto err3; 536 537 err = netdev_rx_handler_register(dev, br_handle_frame, p); 538 if (err) 539 goto err4; 540 541 dev->priv_flags |= IFF_BRIDGE_PORT; 542 543 err = netdev_master_upper_dev_link(dev, br->dev, NULL, NULL); 544 if (err) 545 goto err5; 546 547 err = nbp_switchdev_mark_set(p); 548 if (err) 549 goto err6; 550 551 dev_disable_lro(dev); 552 553 list_add_rcu(&p->list, &br->port_list); 554 555 nbp_update_port_count(br); 556 557 netdev_update_features(br->dev); 558 559 br_hr = br->dev->needed_headroom; 560 dev_hr = netdev_get_fwd_headroom(dev); 561 if (br_hr < dev_hr) 562 update_headroom(br, dev_hr); 563 else 564 netdev_set_rx_headroom(dev, br_hr); 565 566 if (br_fdb_insert(br, p, dev->dev_addr, 0)) 567 netdev_err(dev, "failed insert local address bridge forwarding table\n"); 568 569 err = nbp_vlan_init(p); 570 if (err) { 571 netdev_err(dev, "failed to initialize vlan filtering on this port\n"); 572 goto err7; 573 } 574 575 spin_lock_bh(&br->lock); 576 changed_addr = br_stp_recalculate_bridge_id(br); 577 578 if (netif_running(dev) && netif_oper_up(dev) && 579 (br->dev->flags & IFF_UP)) 580 br_stp_enable_port(p); 581 spin_unlock_bh(&br->lock); 582 583 br_ifinfo_notify(RTM_NEWLINK, p); 584 585 if (changed_addr) 586 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev); 587 588 dev_set_mtu(br->dev, br_min_mtu(br)); 589 br_set_gso_limits(br); 590 591 kobject_uevent(&p->kobj, KOBJ_ADD); 592 593 return 0; 594 595 err7: 596 list_del_rcu(&p->list); 597 br_fdb_delete_by_port(br, p, 0, 1); 598 nbp_update_port_count(br); 599 err6: 600 netdev_upper_dev_unlink(dev, br->dev); 601 err5: 602 dev->priv_flags &= ~IFF_BRIDGE_PORT; 603 netdev_rx_handler_unregister(dev); 604 err4: 605 br_netpoll_disable(p); 606 err3: 607 sysfs_remove_link(br->ifobj, p->dev->name); 608 err2: 609 kobject_put(&p->kobj); 610 p = NULL; /* kobject_put frees */ 611 err1: 612 dev_set_allmulti(dev, -1); 613 put_back: 614 dev_put(dev); 615 kfree(p); 616 return err; 617 } 618 619 /* called with RTNL */ 620 int br_del_if(struct net_bridge *br, struct net_device *dev) 621 { 622 struct net_bridge_port *p; 623 bool changed_addr; 624 625 p = br_port_get_rtnl(dev); 626 if (!p || p->br != br) 627 return -EINVAL; 628 629 /* Since more than one interface can be attached to a bridge, 630 * there still maybe an alternate path for netconsole to use; 631 * therefore there is no reason for a NETDEV_RELEASE event. 632 */ 633 del_nbp(p); 634 635 dev_set_mtu(br->dev, br_min_mtu(br)); 636 br_set_gso_limits(br); 637 638 spin_lock_bh(&br->lock); 639 changed_addr = br_stp_recalculate_bridge_id(br); 640 spin_unlock_bh(&br->lock); 641 642 if (changed_addr) 643 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev); 644 645 netdev_update_features(br->dev); 646 647 return 0; 648 } 649 650 void br_port_flags_change(struct net_bridge_port *p, unsigned long mask) 651 { 652 struct net_bridge *br = p->br; 653 654 if (mask & BR_AUTO_MASK) 655 nbp_update_port_count(br); 656 } 657