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, 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         rcu_assign_pointer(dev->br_port, NULL);
 151
 152         br_multicast_del_port(p);
 153
 154         kobject_uevent(&p->kobj, KOBJ_REMOVE);
 155         kobject_del(&p->kobj);
 156
 157 #ifdef CONFIG_NET_POLL_CONTROLLER
 158         if (br_devices_support_netpoll(br))
 159                 br->dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
 160         if (dev->netdev_ops->ndo_netpoll_cleanup)
 161                 dev->netdev_ops->ndo_netpoll_cleanup(dev);
 162         else
 163                 dev->npinfo = NULL;
 164 #endif
 165         call_rcu(&p->rcu, destroy_nbp_rcu);
 166 }
 167
 168 /* called with RTNL */
 169 static void del_br(struct net_bridge *br, struct list_head *head)
 170 {
 171         struct net_bridge_port *p, *n;
 172
 173         list_for_each_entry_safe(p, n, &br->port_list, list) {
 174                 del_nbp(p);
 175         }
 176
 177         br_netpoll_cleanup(br->dev);
 178
 179         del_timer_sync(&br->gc_timer);
 180
 181         br_sysfs_delbr(br->dev);
 182         unregister_netdevice_queue(br->dev, head);
 183 }
 184
 185 static struct net_device *new_bridge_dev(struct net *net, const char *name)
 186 {
 187         struct net_bridge *br;
 188         struct net_device *dev;
 189
 190         dev = alloc_netdev(sizeof(struct net_bridge), name,
 191                            br_dev_setup);
 192
 193         if (!dev)
 194                 return NULL;
 195         dev_net_set(dev, net);
 196
 197         br = netdev_priv(dev);
 198         br->dev = dev;
 199
 200         br->stats = alloc_percpu(struct br_cpu_netstats);
 201         if (!br->stats) {
 202                 free_netdev(dev);
 203                 return NULL;
 204         }
 205
 206         spin_lock_init(&br->lock);
 207         INIT_LIST_HEAD(&br->port_list);
 208         spin_lock_init(&br->hash_lock);
 209
 210         br->bridge_id.prio[0] = 0x80;
 211         br->bridge_id.prio[1] = 0x00;
 212
 213         memcpy(br->group_addr, br_group_address, ETH_ALEN);
 214
 215         br->feature_mask = dev->features;
 216         br->stp_enabled = BR_NO_STP;
 217         br->designated_root = br->bridge_id;
 218         br->root_path_cost = 0;
 219         br->root_port = 0;
 220         br->bridge_max_age = br->max_age = 20 * HZ;
 221         br->bridge_hello_time = br->hello_time = 2 * HZ;
 222         br->bridge_forward_delay = br->forward_delay = 15 * HZ;
 223         br->topology_change = 0;
 224         br->topology_change_detected = 0;
 225         br->ageing_time = 300 * HZ;
 226
 227         br_netfilter_rtable_init(br);
 228
 229         br_stp_timer_init(br);
 230         br_multicast_init(br);
 231
 232         return dev;
 233 }
 234
 235 /* find an available port number */
 236 static int find_portno(struct net_bridge *br)
 237 {
 238         int index;
 239         struct net_bridge_port *p;
 240         unsigned long *inuse;
 241
 242         inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
 243                         GFP_KERNEL);
 244         if (!inuse)
 245                 return -ENOMEM;
 246
 247         set_bit(0, inuse);      /* zero is reserved */
 248         list_for_each_entry(p, &br->port_list, list) {
 249                 set_bit(p->port_no, inuse);
 250         }
 251         index = find_first_zero_bit(inuse, BR_MAX_PORTS);
 252         kfree(inuse);
 253
 254         return (index >= BR_MAX_PORTS) ? -EXFULL : index;
 255 }
 256
 257 /* called with RTNL but without bridge lock */
 258 static struct net_bridge_port *new_nbp(struct net_bridge *br,
 259                                        struct net_device *dev)
 260 {
 261         int index;
 262         struct net_bridge_port *p;
 263
 264         index = find_portno(br);
 265         if (index < 0)
 266                 return ERR_PTR(index);
 267
 268         p = kzalloc(sizeof(*p), GFP_KERNEL);
 269         if (p == NULL)
 270                 return ERR_PTR(-ENOMEM);
 271
 272         p->br = br;
 273         dev_hold(dev);
 274         p->dev = dev;
 275         p->path_cost = port_cost(dev);
 276         p->priority = 0x8000 >> BR_PORT_BITS;
 277         p->port_no = index;
 278         p->flags = 0;
 279         br_init_port(p);
 280         p->state = BR_STATE_DISABLED;
 281         br_stp_port_timer_init(p);
 282         br_multicast_add_port(p);
 283
 284         return p;
 285 }
 286
 287 static struct device_type br_type = {
 288         .name   = "bridge",
 289 };
 290
 291 int br_add_bridge(struct net *net, const char *name)
 292 {
 293         struct net_device *dev;
 294         int ret;
 295
 296         dev = new_bridge_dev(net, name);
 297         if (!dev)
 298                 return -ENOMEM;
 299
 300         rtnl_lock();
 301         if (strchr(dev->name, '%')) {
 302                 ret = dev_alloc_name(dev, dev->name);
 303                 if (ret < 0)
 304                         goto out_free;
 305         }
 306
 307         SET_NETDEV_DEVTYPE(dev, &br_type);
 308
 309         ret = register_netdevice(dev);
 310         if (ret)
 311                 goto out_free;
 312
 313         ret = br_sysfs_addbr(dev);
 314         if (ret)
 315                 unregister_netdevice(dev);
 316  out:
 317         rtnl_unlock();
 318         return ret;
 319
 320 out_free:
 321         free_netdev(dev);
 322         goto out;
 323 }
 324
 325 int br_del_bridge(struct net *net, const char *name)
 326 {
 327         struct net_device *dev;
 328         int ret = 0;
 329
 330         rtnl_lock();
 331         dev = __dev_get_by_name(net, name);
 332         if (dev == NULL)
 333                 ret =  -ENXIO;  /* Could not find device */
 334
 335         else if (!(dev->priv_flags & IFF_EBRIDGE)) {
 336                 /* Attempt to delete non bridge device! */
 337                 ret = -EPERM;
 338         }
 339
 340         else if (dev->flags & IFF_UP) {
 341                 /* Not shutdown yet. */
 342                 ret = -EBUSY;
 343         }
 344
 345         else
 346                 del_br(netdev_priv(dev), NULL);
 347
 348         rtnl_unlock();
 349         return ret;
 350 }
 351
 352 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
 353 int br_min_mtu(const struct net_bridge *br)
 354 {
 355         const struct net_bridge_port *p;
 356         int mtu = 0;
 357
 358         ASSERT_RTNL();
 359
 360         if (list_empty(&br->port_list))
 361                 mtu = ETH_DATA_LEN;
 362         else {
 363                 list_for_each_entry(p, &br->port_list, list) {
 364                         if (!mtu  || p->dev->mtu < mtu)
 365                                 mtu = p->dev->mtu;
 366                 }
 367         }
 368         return mtu;
 369 }
 370
 371 /*
 372  * Recomputes features using slave's features
 373  */
 374 void br_features_recompute(struct net_bridge *br)
 375 {
 376         struct net_bridge_port *p;
 377         unsigned long features, mask;
 378
 379         features = mask = br->feature_mask;
 380         if (list_empty(&br->port_list))
 381                 goto done;
 382
 383         features &= ~NETIF_F_ONE_FOR_ALL;
 384
 385         list_for_each_entry(p, &br->port_list, list) {
 386                 features = netdev_increment_features(features,
 387                                                      p->dev->features, mask);
 388         }
 389
 390 done:
 391         br->dev->features = netdev_fix_features(features, NULL);
 392 }
 393
 394 /* called with RTNL */
 395 int br_add_if(struct net_bridge *br, struct net_device *dev)
 396 {
 397         struct net_bridge_port *p;
 398         int err = 0;
 399
 400         /* Don't allow bridging non-ethernet like devices */
 401         if ((dev->flags & IFF_LOOPBACK) ||
 402             dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN)
 403                 return -EINVAL;
 404
 405         /* No bridging of bridges */
 406         if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit)
 407                 return -ELOOP;
 408
 409         /* Device is already being bridged */
 410         if (dev->br_port != NULL)
 411                 return -EBUSY;
 412
 413         /* No bridging devices that dislike that (e.g. wireless) */
 414         if (dev->priv_flags & IFF_DONT_BRIDGE)
 415                 return -EOPNOTSUPP;
 416
 417         p = new_nbp(br, dev);
 418         if (IS_ERR(p))
 419                 return PTR_ERR(p);
 420
 421         err = dev_set_promiscuity(dev, 1);
 422         if (err)
 423                 goto put_back;
 424
 425         err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
 426                                    SYSFS_BRIDGE_PORT_ATTR);
 427         if (err)
 428                 goto err0;
 429
 430         err = br_fdb_insert(br, p, dev->dev_addr);
 431         if (err)
 432                 goto err1;
 433
 434         err = br_sysfs_addif(p);
 435         if (err)
 436                 goto err2;
 437
 438         rcu_assign_pointer(dev->br_port, p);
 439         dev_disable_lro(dev);
 440
 441         list_add_rcu(&p->list, &br->port_list);
 442
 443         spin_lock_bh(&br->lock);
 444         br_stp_recalculate_bridge_id(br);
 445         br_features_recompute(br);
 446
 447         if ((dev->flags & IFF_UP) && netif_carrier_ok(dev) &&
 448             (br->dev->flags & IFF_UP))
 449                 br_stp_enable_port(p);
 450         spin_unlock_bh(&br->lock);
 451
 452         br_ifinfo_notify(RTM_NEWLINK, p);
 453
 454         dev_set_mtu(br->dev, br_min_mtu(br));
 455
 456         kobject_uevent(&p->kobj, KOBJ_ADD);
 457
 458 #ifdef CONFIG_NET_POLL_CONTROLLER
 459         if (br_devices_support_netpoll(br)) {
 460                 br->dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
 461                 if (br->dev->npinfo)
 462                         dev->npinfo = br->dev->npinfo;
 463         } else if (!(br->dev->priv_flags & IFF_DISABLE_NETPOLL)) {
 464                 br->dev->priv_flags |= IFF_DISABLE_NETPOLL;
 465                 printk(KERN_INFO "New device %s does not support netpoll\n",
 466                         dev->name);
 467                 printk(KERN_INFO "Disabling netpoll for %s\n",
 468                         br->dev->name);
 469         }
 470 #endif
 471
 472         return 0;
 473 err2:
 474         br_fdb_delete_by_port(br, p, 1);
 475 err1:
 476         kobject_put(&p->kobj);
 477         p = NULL; /* kobject_put frees */
 478 err0:
 479         dev_set_promiscuity(dev, -1);
 480 put_back:
 481         dev_put(dev);
 482         kfree(p);
 483         return err;
 484 }
 485
 486 /* called with RTNL */
 487 int br_del_if(struct net_bridge *br, struct net_device *dev)
 488 {
 489         struct net_bridge_port *p = dev->br_port;
 490
 491         if (!p || p->br != br)
 492                 return -EINVAL;
 493
 494         del_nbp(p);
 495
 496         spin_lock_bh(&br->lock);
 497         br_stp_recalculate_bridge_id(br);
 498         br_features_recompute(br);
 499         spin_unlock_bh(&br->lock);
 500
 501         return 0;
 502 }
 503
 504 void __net_exit br_net_exit(struct net *net)
 505 {
 506         struct net_device *dev;
 507         LIST_HEAD(list);
 508
 509         rtnl_lock();
 510         for_each_netdev(net, dev)
 511                 if (dev->priv_flags & IFF_EBRIDGE)
 512                         del_br(netdev_priv(dev), &list);
 513
 514         unregister_netdevice_many(&list);
 515         rtnl_unlock();
 516
 517 }