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/netpoll.h>
  17 #include <linux/ethtool.h>
  18 #include <linux/if_arp.h>
  19 #include <linux/module.h>
  20 #include <linux/init.h>
  21 #include <linux/rtnetlink.h>
  22 #include <linux/if_ether.h>
  23 #include <linux/slab.h>
  24 #include <net/sock.h>
  25
  26 #include "br_private.h"
  27
  28 /*
  29  * Determine initial path cost based on speed.
  30  * using recommendations from 802.1d standard
  31  *
  32  * Since driver might sleep need to not be holding any locks.
  33  */
  34 static int port_cost(struct net_device *dev)
  35 {
  36         if (dev->ethtool_ops && dev->ethtool_ops->get_settings) {
  37                 struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET, };
  38
  39                 if (!dev->ethtool_ops->get_settings(dev, &ecmd)) {
  40                         switch(ecmd.speed) {
  41                         case SPEED_10000:
  42                                 return 2;
  43                         case SPEED_1000:
  44                                 return 4;
  45                         case SPEED_100:
  46                                 return 19;
  47                         case SPEED_10:
  48                                 return 100;
  49                         }
  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 /*
  65  * Check for port carrier transistions.
  66  * Called from work queue to allow for calling functions that
  67  * might sleep (such as speed check), and to debounce.
  68  */
  69 void br_port_carrier_check(struct net_bridge_port *p)
  70 {
  71         struct net_device *dev = p->dev;
  72         struct net_bridge *br = p->br;
  73
  74         if (netif_carrier_ok(dev))
  75                 p->path_cost = port_cost(dev);
  76
  77         if (netif_running(br->dev)) {
  78                 spin_lock_bh(&br->lock);
  79                 if (netif_carrier_ok(dev)) {
  80                         if (p->state == BR_STATE_DISABLED)
  81                                 br_stp_enable_port(p);
  82                 } else {
  83                         if (p->state != BR_STATE_DISABLED)
  84                                 br_stp_disable_port(p);
  85                 }
  86                 spin_unlock_bh(&br->lock);
  87         }
  88 }
  89
  90 static void release_nbp(struct kobject *kobj)
  91 {
  92         struct net_bridge_port *p
  93                 = container_of(kobj, struct net_bridge_port, kobj);
  94         kfree(p);
  95 }
  96
  97 static struct kobj_type brport_ktype = {
  98 #ifdef CONFIG_SYSFS
  99         .sysfs_ops = &brport_sysfs_ops,
 100 #endif
 101         .release = release_nbp,
 102 };
 103
 104 static void destroy_nbp(struct net_bridge_port *p)
 105 {
 106         struct net_device *dev = p->dev;
 107
 108         p->br = NULL;
 109         p->dev = NULL;
 110         dev_put(dev);
 111
 112         kobject_put(&p->kobj);
 113 }
 114
 115 static void destroy_nbp_rcu(struct rcu_head *head)
 116 {
 117         struct net_bridge_port *p =
 118                         container_of(head, struct net_bridge_port, rcu);
 119         destroy_nbp(p);
 120 }
 121
 122 /* Delete port(interface) from bridge is done in two steps.
 123  * via RCU. First step, marks device as down. That deletes
 124  * all the timers and stops new packets from flowing through.
 125  *
 126  * Final cleanup doesn't occur until after all CPU's finished
 127  * processing packets.
 128  *
 129  * Protected from multiple admin operations by RTNL mutex
 130  */
 131 static void del_nbp(struct net_bridge_port *p)
 132 {
 133         struct net_bridge *br = p->br;
 134         struct net_device *dev = p->dev;
 135
 136         sysfs_remove_link(br->ifobj, p->dev->name);
 137
 138         dev_set_promiscuity(dev, -1);
 139
 140         spin_lock_bh(&br->lock);
 141         br_stp_disable_port(p);
 142         spin_unlock_bh(&br->lock);
 143
 144         br_ifinfo_notify(RTM_DELLINK, p);
 145
 146         br_fdb_delete_by_port(br, p, 1);
 147
 148         list_del_rcu(&p->list);
 149
 150         netdev_rx_handler_unregister(dev);
 151         rcu_assign_pointer(dev->br_port, NULL);
 152
 153         br_multicast_del_port(p);
 154
 155         kobject_uevent(&p->kobj, KOBJ_REMOVE);
 156         kobject_del(&p->kobj);
 157
 158         br_netpoll_disable(p);
 159
 160         call_rcu(&p->rcu, destroy_nbp_rcu);
 161 }
 162
 163 /* called with RTNL */
 164 static void del_br(struct net_bridge *br, struct list_head *head)
 165 {
 166         struct net_bridge_port *p, *n;
 167
 168         list_for_each_entry_safe(p, n, &br->port_list, list) {
 169                 del_nbp(p);
 170         }
 171
 172         del_timer_sync(&br->gc_timer);
 173
 174         br_sysfs_delbr(br->dev);
 175         unregister_netdevice_queue(br->dev, head);
 176 }
 177
 178 static struct net_device *new_bridge_dev(struct net *net, const char *name)
 179 {
 180         struct net_bridge *br;
 181         struct net_device *dev;
 182
 183         dev = alloc_netdev(sizeof(struct net_bridge), name,
 184                            br_dev_setup);
 185
 186         if (!dev)
 187                 return NULL;
 188         dev_net_set(dev, net);
 189
 190         br = netdev_priv(dev);
 191         br->dev = dev;
 192
 193         br->stats = alloc_percpu(struct br_cpu_netstats);
 194         if (!br->stats) {
 195                 free_netdev(dev);
 196                 return NULL;
 197         }
 198
 199         spin_lock_init(&br->lock);
 200         INIT_LIST_HEAD(&br->port_list);
 201         spin_lock_init(&br->hash_lock);
 202
 203         br->bridge_id.prio[0] = 0x80;
 204         br->bridge_id.prio[1] = 0x00;
 205
 206         memcpy(br->group_addr, br_group_address, ETH_ALEN);
 207
 208         br->feature_mask = dev->features;
 209         br->stp_enabled = BR_NO_STP;
 210         br->designated_root = br->bridge_id;
 211         br->root_path_cost = 0;
 212         br->root_port = 0;
 213         br->bridge_max_age = br->max_age = 20 * HZ;
 214         br->bridge_hello_time = br->hello_time = 2 * HZ;
 215         br->bridge_forward_delay = br->forward_delay = 15 * HZ;
 216         br->topology_change = 0;
 217         br->topology_change_detected = 0;
 218         br->ageing_time = 300 * HZ;
 219
 220         br_netfilter_rtable_init(br);
 221
 222         br_stp_timer_init(br);
 223         br_multicast_init(br);
 224
 225         return dev;
 226 }
 227
 228 /* find an available port number */
 229 static int find_portno(struct net_bridge *br)
 230 {
 231         int index;
 232         struct net_bridge_port *p;
 233         unsigned long *inuse;
 234
 235         inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
 236                         GFP_KERNEL);
 237         if (!inuse)
 238                 return -ENOMEM;
 239
 240         set_bit(0, inuse);      /* zero is reserved */
 241         list_for_each_entry(p, &br->port_list, list) {
 242                 set_bit(p->port_no, inuse);
 243         }
 244         index = find_first_zero_bit(inuse, BR_MAX_PORTS);
 245         kfree(inuse);
 246
 247         return (index >= BR_MAX_PORTS) ? -EXFULL : index;
 248 }
 249
 250 /* called with RTNL but without bridge lock */
 251 static struct net_bridge_port *new_nbp(struct net_bridge *br,
 252                                        struct net_device *dev)
 253 {
 254         int index;
 255         struct net_bridge_port *p;
 256
 257         index = find_portno(br);
 258         if (index < 0)
 259                 return ERR_PTR(index);
 260
 261         p = kzalloc(sizeof(*p), GFP_KERNEL);
 262         if (p == NULL)
 263                 return ERR_PTR(-ENOMEM);
 264
 265         p->br = br;
 266         dev_hold(dev);
 267         p->dev = dev;
 268         p->path_cost = port_cost(dev);
 269         p->priority = 0x8000 >> BR_PORT_BITS;
 270         p->port_no = index;
 271         p->flags = 0;
 272         br_init_port(p);
 273         p->state = BR_STATE_DISABLED;
 274         br_stp_port_timer_init(p);
 275         br_multicast_add_port(p);
 276
 277         return p;
 278 }
 279
 280 static struct device_type br_type = {
 281         .name   = "bridge",
 282 };
 283
 284 int br_add_bridge(struct net *net, const char *name)
 285 {
 286         struct net_device *dev;
 287         int ret;
 288
 289         dev = new_bridge_dev(net, name);
 290         if (!dev)
 291                 return -ENOMEM;
 292
 293         rtnl_lock();
 294         if (strchr(dev->name, '%')) {
 295                 ret = dev_alloc_name(dev, dev->name);
 296                 if (ret < 0)
 297                         goto out_free;
 298         }
 299
 300         SET_NETDEV_DEVTYPE(dev, &br_type);
 301
 302         ret = register_netdevice(dev);
 303         if (ret)
 304                 goto out_free;
 305
 306         ret = br_sysfs_addbr(dev);
 307         if (ret)
 308                 unregister_netdevice(dev);
 309  out:
 310         rtnl_unlock();
 311         return ret;
 312
 313 out_free:
 314         free_netdev(dev);
 315         goto out;
 316 }
 317
 318 int br_del_bridge(struct net *net, const char *name)
 319 {
 320         struct net_device *dev;
 321         int ret = 0;
 322
 323         rtnl_lock();
 324         dev = __dev_get_by_name(net, name);
 325         if (dev == NULL)
 326                 ret =  -ENXIO;  /* Could not find device */
 327
 328         else if (!(dev->priv_flags & IFF_EBRIDGE)) {
 329                 /* Attempt to delete non bridge device! */
 330                 ret = -EPERM;
 331         }
 332
 333         else if (dev->flags & IFF_UP) {
 334                 /* Not shutdown yet. */
 335                 ret = -EBUSY;
 336         }
 337
 338         else
 339                 del_br(netdev_priv(dev), NULL);
 340
 341         rtnl_unlock();
 342         return ret;
 343 }
 344
 345 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
 346 int br_min_mtu(const struct net_bridge *br)
 347 {
 348         const struct net_bridge_port *p;
 349         int mtu = 0;
 350
 351         ASSERT_RTNL();
 352
 353         if (list_empty(&br->port_list))
 354                 mtu = ETH_DATA_LEN;
 355         else {
 356                 list_for_each_entry(p, &br->port_list, list) {
 357                         if (!mtu  || p->dev->mtu < mtu)
 358                                 mtu = p->dev->mtu;
 359                 }
 360         }
 361         return mtu;
 362 }
 363
 364 /*
 365  * Recomputes features using slave's features
 366  */
 367 void br_features_recompute(struct net_bridge *br)
 368 {
 369         struct net_bridge_port *p;
 370         unsigned long features, mask;
 371
 372         features = mask = br->feature_mask;
 373         if (list_empty(&br->port_list))
 374                 goto done;
 375
 376         features &= ~NETIF_F_ONE_FOR_ALL;
 377
 378         list_for_each_entry(p, &br->port_list, list) {
 379                 features = netdev_increment_features(features,
 380                                                      p->dev->features, mask);
 381         }
 382
 383 done:
 384         br->dev->features = netdev_fix_features(features, NULL);
 385 }
 386
 387 /* called with RTNL */
 388 int br_add_if(struct net_bridge *br, struct net_device *dev)
 389 {
 390         struct net_bridge_port *p;
 391         int err = 0;
 392
 393         /* Don't allow bridging non-ethernet like devices */
 394         if ((dev->flags & IFF_LOOPBACK) ||
 395             dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN)
 396                 return -EINVAL;
 397
 398         /* No bridging of bridges */
 399         if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit)
 400                 return -ELOOP;
 401
 402         /* Device is already being bridged */
 403         if (dev->br_port != NULL)
 404                 return -EBUSY;
 405
 406         /* No bridging devices that dislike that (e.g. wireless) */
 407         if (dev->priv_flags & IFF_DONT_BRIDGE)
 408                 return -EOPNOTSUPP;
 409
 410         p = new_nbp(br, dev);
 411         if (IS_ERR(p))
 412                 return PTR_ERR(p);
 413
 414         err = dev_set_promiscuity(dev, 1);
 415         if (err)
 416                 goto put_back;
 417
 418         err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
 419                                    SYSFS_BRIDGE_PORT_ATTR);
 420         if (err)
 421                 goto err0;
 422
 423         err = br_fdb_insert(br, p, dev->dev_addr);
 424         if (err)
 425                 goto err1;
 426
 427         err = br_sysfs_addif(p);
 428         if (err)
 429                 goto err2;
 430
 431         if (br_netpoll_info(br) && ((err = br_netpoll_enable(p))))
 432                 goto err3;
 433
 434         rcu_assign_pointer(dev->br_port, p);
 435
 436         err = netdev_rx_handler_register(dev, br_handle_frame, NULL);
 437         if (err)
 438                 goto err4;
 439
 440         dev_disable_lro(dev);
 441
 442         list_add_rcu(&p->list, &br->port_list);
 443
 444         spin_lock_bh(&br->lock);
 445         br_stp_recalculate_bridge_id(br);
 446         br_features_recompute(br);
 447
 448         if ((dev->flags & IFF_UP) && netif_carrier_ok(dev) &&
 449             (br->dev->flags & IFF_UP))
 450                 br_stp_enable_port(p);
 451         spin_unlock_bh(&br->lock);
 452
 453         br_ifinfo_notify(RTM_NEWLINK, p);
 454
 455         dev_set_mtu(br->dev, br_min_mtu(br));
 456
 457         kobject_uevent(&p->kobj, KOBJ_ADD);
 458
 459         return 0;
 460 err4:
 461         rcu_assign_pointer(dev->br_port, NULL);
 462 err3:
 463         sysfs_remove_link(br->ifobj, p->dev->name);
 464 err2:
 465         br_fdb_delete_by_port(br, p, 1);
 466 err1:
 467         kobject_put(&p->kobj);
 468         p = NULL; /* kobject_put frees */
 469 err0:
 470         dev_set_promiscuity(dev, -1);
 471 put_back:
 472         dev_put(dev);
 473         kfree(p);
 474         return err;
 475 }
 476
 477 /* called with RTNL */
 478 int br_del_if(struct net_bridge *br, struct net_device *dev)
 479 {
 480         struct net_bridge_port *p = dev->br_port;
 481
 482         if (!p || p->br != br)
 483                 return -EINVAL;
 484
 485         del_nbp(p);
 486
 487         spin_lock_bh(&br->lock);
 488         br_stp_recalculate_bridge_id(br);
 489         br_features_recompute(br);
 490         spin_unlock_bh(&br->lock);
 491
 492         return 0;
 493 }
 494
 495 void __net_exit br_net_exit(struct net *net)
 496 {
 497         struct net_device *dev;
 498         LIST_HEAD(list);
 499
 500         rtnl_lock();
 501         for_each_netdev(net, dev)
 502                 if (dev->priv_flags & IFF_EBRIDGE)
 503                         del_br(netdev_priv(dev), &list);
 504
 505         unregister_netdevice_many(&list);
 506         rtnl_unlock();
 507
 508 }