net/bridge/br_if.c

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