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(br, dev);
 159         call_rcu(&p->rcu, destroy_nbp_rcu);
 160 }
 161
 162 /* called with RTNL */
 163 static void del_br(struct net_bridge *br, struct list_head *head)
 164 {
 165         struct net_bridge_port *p, *n;
 166
 167         list_for_each_entry_safe(p, n, &br->port_list, list) {
 168                 del_nbp(p);
 169         }
 170
 171         br_netpoll_cleanup(br->dev);
 172
 173         del_timer_sync(&br->gc_timer);
 174
 175         br_sysfs_delbr(br->dev);
 176         unregister_netdevice_queue(br->dev, head);
 177 }
 178
 179 static struct net_device *new_bridge_dev(struct net *net, const char *name)
 180 {
 181         struct net_bridge *br;
 182         struct net_device *dev;
 183
 184         dev = alloc_netdev(sizeof(struct net_bridge), name,
 185                            br_dev_setup);
 186
 187         if (!dev)
 188                 return NULL;
 189         dev_net_set(dev, net);
 190
 191         br = netdev_priv(dev);
 192         br->dev = dev;
 193
 194         br->stats = alloc_percpu(struct br_cpu_netstats);
 195         if (!br->stats) {
 196                 free_netdev(dev);
 197                 return NULL;
 198         }
 199
 200         spin_lock_init(&br->lock);
 201         INIT_LIST_HEAD(&br->port_list);
 202         spin_lock_init(&br->hash_lock);
 203
 204         br->bridge_id.prio[0] = 0x80;
 205         br->bridge_id.prio[1] = 0x00;
 206
 207         memcpy(br->group_addr, br_group_address, ETH_ALEN);
 208
 209         br->feature_mask = dev->features;
 210         br->stp_enabled = BR_NO_STP;
 211         br->designated_root = br->bridge_id;
 212         br->root_path_cost = 0;
 213         br->root_port = 0;
 214         br->bridge_max_age = br->max_age = 20 * HZ;
 215         br->bridge_hello_time = br->hello_time = 2 * HZ;
 216         br->bridge_forward_delay = br->forward_delay = 15 * HZ;
 217         br->topology_change = 0;
 218         br->topology_change_detected = 0;
 219         br->ageing_time = 300 * HZ;
 220
 221         br_netfilter_rtable_init(br);
 222
 223         br_stp_timer_init(br);
 224         br_multicast_init(br);
 225
 226         return dev;
 227 }
 228
 229 /* find an available port number */
 230 static int find_portno(struct net_bridge *br)
 231 {
 232         int index;
 233         struct net_bridge_port *p;
 234         unsigned long *inuse;
 235
 236         inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
 237                         GFP_KERNEL);
 238         if (!inuse)
 239                 return -ENOMEM;
 240
 241         set_bit(0, inuse);      /* zero is reserved */
 242         list_for_each_entry(p, &br->port_list, list) {
 243                 set_bit(p->port_no, inuse);
 244         }
 245         index = find_first_zero_bit(inuse, BR_MAX_PORTS);
 246         kfree(inuse);
 247
 248         return (index >= BR_MAX_PORTS) ? -EXFULL : index;
 249 }
 250
 251 /* called with RTNL but without bridge lock */
 252 static struct net_bridge_port *new_nbp(struct net_bridge *br,
 253                                        struct net_device *dev)
 254 {
 255         int index;
 256         struct net_bridge_port *p;
 257
 258         index = find_portno(br);
 259         if (index < 0)
 260                 return ERR_PTR(index);
 261
 262         p = kzalloc(sizeof(*p), GFP_KERNEL);
 263         if (p == NULL)
 264                 return ERR_PTR(-ENOMEM);
 265
 266         p->br = br;
 267         dev_hold(dev);
 268         p->dev = dev;
 269         p->path_cost = port_cost(dev);
 270         p->priority = 0x8000 >> BR_PORT_BITS;
 271         p->port_no = index;
 272         p->flags = 0;
 273         br_init_port(p);
 274         p->state = BR_STATE_DISABLED;
 275         br_stp_port_timer_init(p);
 276         br_multicast_add_port(p);
 277
 278         return p;
 279 }
 280
 281 static struct device_type br_type = {
 282         .name   = "bridge",
 283 };
 284
 285 int br_add_bridge(struct net *net, const char *name)
 286 {
 287         struct net_device *dev;
 288         int ret;
 289
 290         dev = new_bridge_dev(net, name);
 291         if (!dev)
 292                 return -ENOMEM;
 293
 294         rtnl_lock();
 295         if (strchr(dev->name, '%')) {
 296                 ret = dev_alloc_name(dev, dev->name);
 297                 if (ret < 0)
 298                         goto out_free;
 299         }
 300
 301         SET_NETDEV_DEVTYPE(dev, &br_type);
 302
 303         ret = register_netdevice(dev);
 304         if (ret)
 305                 goto out_free;
 306
 307         ret = br_sysfs_addbr(dev);
 308         if (ret)
 309                 unregister_netdevice(dev);
 310  out:
 311         rtnl_unlock();
 312         return ret;
 313
 314 out_free:
 315         free_netdev(dev);
 316         goto out;
 317 }
 318
 319 int br_del_bridge(struct net *net, const char *name)
 320 {
 321         struct net_device *dev;
 322         int ret = 0;
 323
 324         rtnl_lock();
 325         dev = __dev_get_by_name(net, name);
 326         if (dev == NULL)
 327                 ret =  -ENXIO;  /* Could not find device */
 328
 329         else if (!(dev->priv_flags & IFF_EBRIDGE)) {
 330                 /* Attempt to delete non bridge device! */
 331                 ret = -EPERM;
 332         }
 333
 334         else if (dev->flags & IFF_UP) {
 335                 /* Not shutdown yet. */
 336                 ret = -EBUSY;
 337         }
 338
 339         else
 340                 del_br(netdev_priv(dev), NULL);
 341
 342         rtnl_unlock();
 343         return ret;
 344 }
 345
 346 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
 347 int br_min_mtu(const struct net_bridge *br)
 348 {
 349         const struct net_bridge_port *p;
 350         int mtu = 0;
 351
 352         ASSERT_RTNL();
 353
 354         if (list_empty(&br->port_list))
 355                 mtu = ETH_DATA_LEN;
 356         else {
 357                 list_for_each_entry(p, &br->port_list, list) {
 358                         if (!mtu  || p->dev->mtu < mtu)
 359                                 mtu = p->dev->mtu;
 360                 }
 361         }
 362         return mtu;
 363 }
 364
 365 /*
 366  * Recomputes features using slave's features
 367  */
 368 void br_features_recompute(struct net_bridge *br)
 369 {
 370         struct net_bridge_port *p;
 371         unsigned long features, mask;
 372
 373         features = mask = br->feature_mask;
 374         if (list_empty(&br->port_list))
 375                 goto done;
 376
 377         features &= ~NETIF_F_ONE_FOR_ALL;
 378
 379         list_for_each_entry(p, &br->port_list, list) {
 380                 features = netdev_increment_features(features,
 381                                                      p->dev->features, mask);
 382         }
 383
 384 done:
 385         br->dev->features = netdev_fix_features(features, NULL);
 386 }
 387
 388 /* called with RTNL */
 389 int br_add_if(struct net_bridge *br, struct net_device *dev)
 390 {
 391         struct net_bridge_port *p;
 392         int err = 0;
 393
 394         /* Don't allow bridging non-ethernet like devices */
 395         if ((dev->flags & IFF_LOOPBACK) ||
 396             dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN)
 397                 return -EINVAL;
 398
 399         /* No bridging of bridges */
 400         if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit)
 401                 return -ELOOP;
 402
 403         /* Device is already being bridged */
 404         if (dev->br_port != NULL)
 405                 return -EBUSY;
 406
 407         /* No bridging devices that dislike that (e.g. wireless) */
 408         if (dev->priv_flags & IFF_DONT_BRIDGE)
 409                 return -EOPNOTSUPP;
 410
 411         p = new_nbp(br, dev);
 412         if (IS_ERR(p))
 413                 return PTR_ERR(p);
 414
 415         err = dev_set_promiscuity(dev, 1);
 416         if (err)
 417                 goto put_back;
 418
 419         err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
 420                                    SYSFS_BRIDGE_PORT_ATTR);
 421         if (err)
 422                 goto err0;
 423
 424         err = br_fdb_insert(br, p, dev->dev_addr);
 425         if (err)
 426                 goto err1;
 427
 428         err = br_sysfs_addif(p);
 429         if (err)
 430                 goto err2;
 431
 432         rcu_assign_pointer(dev->br_port, p);
 433
 434         err = netdev_rx_handler_register(dev, br_handle_frame);
 435         if (err)
 436                 goto err3;
 437
 438         dev_disable_lro(dev);
 439
 440         list_add_rcu(&p->list, &br->port_list);
 441
 442         spin_lock_bh(&br->lock);
 443         br_stp_recalculate_bridge_id(br);
 444         br_features_recompute(br);
 445
 446         if ((dev->flags & IFF_UP) && netif_carrier_ok(dev) &&
 447             (br->dev->flags & IFF_UP))
 448                 br_stp_enable_port(p);
 449         spin_unlock_bh(&br->lock);
 450
 451         br_ifinfo_notify(RTM_NEWLINK, p);
 452
 453         dev_set_mtu(br->dev, br_min_mtu(br));
 454
 455         kobject_uevent(&p->kobj, KOBJ_ADD);
 456
 457         br_netpoll_enable(br, dev);
 458
 459         return 0;
 460 err3:
 461         rcu_assign_pointer(dev->br_port, NULL);
 462 err2:
 463         br_fdb_delete_by_port(br, p, 1);
 464 err1:
 465         kobject_put(&p->kobj);
 466         p = NULL; /* kobject_put frees */
 467 err0:
 468         dev_set_promiscuity(dev, -1);
 469 put_back:
 470         dev_put(dev);
 471         kfree(p);
 472         return err;
 473 }
 474
 475 /* called with RTNL */
 476 int br_del_if(struct net_bridge *br, struct net_device *dev)
 477 {
 478         struct net_bridge_port *p = dev->br_port;
 479
 480         if (!p || p->br != br)
 481                 return -EINVAL;
 482
 483         del_nbp(p);
 484
 485         spin_lock_bh(&br->lock);
 486         br_stp_recalculate_bridge_id(br);
 487         br_features_recompute(br);
 488         spin_unlock_bh(&br->lock);
 489
 490         return 0;
 491 }
 492
 493 void __net_exit br_net_exit(struct net *net)
 494 {
 495         struct net_device *dev;
 496         LIST_HEAD(list);
 497
 498         rtnl_lock();
 499         for_each_netdev(net, dev)
 500                 if (dev->priv_flags & IFF_EBRIDGE)
 501                         del_br(netdev_priv(dev), &list);
 502
 503         unregister_netdevice_many(&list);
 504         rtnl_unlock();
 505
 506 }