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/ethtool.h>
  17 #include <linux/if_arp.h>
  18 #include <linux/module.h>
  19 #include <linux/init.h>
  20 #include <linux/rtnetlink.h>
  21 #include <linux/if_ether.h>
  22 #include <linux/slab.h>
  23 #include <net/sock.h>
  24
  25 #include "br_private.h"
  26
  27 /*
  28  * Determine initial path cost based on speed.
  29  * using recommendations from 802.1d standard
  30  *
  31  * Since driver might sleep need to not be holding any locks.
  32  */
  33 static int port_cost(struct net_device *dev)
  34 {
  35         if (dev->ethtool_ops && dev->ethtool_ops->get_settings) {
  36                 struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET, };
  37
  38                 if (!dev->ethtool_ops->get_settings(dev, &ecmd)) {
  39                         switch(ecmd.speed) {
  40                         case SPEED_10000:
  41                                 return 2;
  42                         case SPEED_1000:
  43                                 return 4;
  44                         case SPEED_100:
  45                                 return 19;
  46                         case SPEED_10:
  47                                 return 100;
  48                         }
  49                 }
  50         }
  51
  52         /* Old silly heuristics based on name */
  53         if (!strncmp(dev->name, "lec", 3))
  54                 return 7;
  55
  56         if (!strncmp(dev->name, "plip", 4))
  57                 return 2500;
  58
  59         return 100;     /* assume old 10Mbps */
  60 }
  61
  62
  63 /*
  64  * Check for port carrier transistions.
  65  * Called from work queue to allow for calling functions that
  66  * might sleep (such as speed check), and to debounce.
  67  */
  68 void br_port_carrier_check(struct net_bridge_port *p)
  69 {
  70         struct net_device *dev = p->dev;
  71         struct net_bridge *br = p->br;
  72
  73         if (netif_carrier_ok(dev))
  74                 p->path_cost = port_cost(dev);
  75
  76         if (netif_running(br->dev)) {
  77                 spin_lock_bh(&br->lock);
  78                 if (netif_carrier_ok(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
  89 static void release_nbp(struct kobject *kobj)
  90 {
  91         struct net_bridge_port *p
  92                 = container_of(kobj, struct net_bridge_port, kobj);
  93         kfree(p);
  94 }
  95
  96 static struct kobj_type brport_ktype = {
  97 #ifdef CONFIG_SYSFS
  98         .sysfs_ops = &brport_sysfs_ops,
  99 #endif
 100         .release = release_nbp,
 101 };
 102
 103 static void destroy_nbp(struct net_bridge_port *p)
 104 {
 105         struct net_device *dev = p->dev;
 106
 107         p->br = NULL;
 108         p->dev = NULL;
 109         dev_put(dev);
 110
 111         kobject_put(&p->kobj);
 112 }
 113
 114 static void destroy_nbp_rcu(struct rcu_head *head)
 115 {
 116         struct net_bridge_port *p =
 117                         container_of(head, struct net_bridge_port, rcu);
 118         destroy_nbp(p);
 119 }
 120
 121 /* Delete port(interface) from bridge is done in two steps.
 122  * via RCU. First step, marks device as down. That deletes
 123  * all the timers and stops new packets from flowing through.
 124  *
 125  * Final cleanup doesn't occur until after all CPU's finished
 126  * processing packets.
 127  *
 128  * Protected from multiple admin operations by RTNL mutex
 129  */
 130 static void del_nbp(struct net_bridge_port *p)
 131 {
 132         struct net_bridge *br = p->br;
 133         struct net_device *dev = p->dev;
 134
 135         sysfs_remove_link(br->ifobj, dev->name);
 136
 137         dev_set_promiscuity(dev, -1);
 138
 139         spin_lock_bh(&br->lock);
 140         br_stp_disable_port(p);
 141         spin_unlock_bh(&br->lock);
 142
 143         br_ifinfo_notify(RTM_DELLINK, p);
 144
 145         br_fdb_delete_by_port(br, p, 1);
 146
 147         list_del_rcu(&p->list);
 148
 149         rcu_assign_pointer(dev->br_port, NULL);
 150
 151         br_multicast_del_port(p);
 152
 153         kobject_uevent(&p->kobj, KOBJ_REMOVE);
 154         kobject_del(&p->kobj);
 155
 156         call_rcu(&p->rcu, destroy_nbp_rcu);
 157 }
 158
 159 /* called with RTNL */
 160 static void del_br(struct net_bridge *br, struct list_head *head)
 161 {
 162         struct net_bridge_port *p, *n;
 163
 164         list_for_each_entry_safe(p, n, &br->port_list, list) {
 165                 del_nbp(p);
 166         }
 167
 168         del_timer_sync(&br->gc_timer);
 169
 170         br_sysfs_delbr(br->dev);
 171         unregister_netdevice_queue(br->dev, head);
 172 }
 173
 174 static struct net_device *new_bridge_dev(struct net *net, const char *name)
 175 {
 176         struct net_bridge *br;
 177         struct net_device *dev;
 178
 179         dev = alloc_netdev(sizeof(struct net_bridge), name,
 180                            br_dev_setup);
 181
 182         if (!dev)
 183                 return NULL;
 184         dev_net_set(dev, net);
 185
 186         br = netdev_priv(dev);
 187         br->dev = dev;
 188
 189         br->stats = alloc_percpu(struct br_cpu_netstats);
 190         if (!br->stats) {
 191                 free_netdev(dev);
 192                 return NULL;
 193         }
 194
 195         spin_lock_init(&br->lock);
 196         INIT_LIST_HEAD(&br->port_list);
 197         spin_lock_init(&br->hash_lock);
 198
 199         br->bridge_id.prio[0] = 0x80;
 200         br->bridge_id.prio[1] = 0x00;
 201
 202         memcpy(br->group_addr, br_group_address, ETH_ALEN);
 203
 204         br->feature_mask = dev->features;
 205         br->stp_enabled = BR_NO_STP;
 206         br->designated_root = br->bridge_id;
 207         br->root_path_cost = 0;
 208         br->root_port = 0;
 209         br->bridge_max_age = br->max_age = 20 * HZ;
 210         br->bridge_hello_time = br->hello_time = 2 * HZ;
 211         br->bridge_forward_delay = br->forward_delay = 15 * HZ;
 212         br->topology_change = 0;
 213         br->topology_change_detected = 0;
 214         br->ageing_time = 300 * HZ;
 215
 216         br_netfilter_rtable_init(br);
 217
 218         br_stp_timer_init(br);
 219         br_multicast_init(br);
 220
 221         return dev;
 222 }
 223
 224 /* find an available port number */
 225 static int find_portno(struct net_bridge *br)
 226 {
 227         int index;
 228         struct net_bridge_port *p;
 229         unsigned long *inuse;
 230
 231         inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
 232                         GFP_KERNEL);
 233         if (!inuse)
 234                 return -ENOMEM;
 235
 236         set_bit(0, inuse);      /* zero is reserved */
 237         list_for_each_entry(p, &br->port_list, list) {
 238                 set_bit(p->port_no, inuse);
 239         }
 240         index = find_first_zero_bit(inuse, BR_MAX_PORTS);
 241         kfree(inuse);
 242
 243         return (index >= BR_MAX_PORTS) ? -EXFULL : index;
 244 }
 245
 246 /* called with RTNL but without bridge lock */
 247 static struct net_bridge_port *new_nbp(struct net_bridge *br,
 248                                        struct net_device *dev)
 249 {
 250         int index;
 251         struct net_bridge_port *p;
 252
 253         index = find_portno(br);
 254         if (index < 0)
 255                 return ERR_PTR(index);
 256
 257         p = kzalloc(sizeof(*p), GFP_KERNEL);
 258         if (p == NULL)
 259                 return ERR_PTR(-ENOMEM);
 260
 261         p->br = br;
 262         dev_hold(dev);
 263         p->dev = dev;
 264         p->path_cost = port_cost(dev);
 265         p->priority = 0x8000 >> BR_PORT_BITS;
 266         p->port_no = index;
 267         p->flags = 0;
 268         br_init_port(p);
 269         p->state = BR_STATE_DISABLED;
 270         br_stp_port_timer_init(p);
 271         br_multicast_add_port(p);
 272
 273         return p;
 274 }
 275
 276 static struct device_type br_type = {
 277         .name   = "bridge",
 278 };
 279
 280 int br_add_bridge(struct net *net, const char *name)
 281 {
 282         struct net_device *dev;
 283         int ret;
 284
 285         dev = new_bridge_dev(net, name);
 286         if (!dev)
 287                 return -ENOMEM;
 288
 289         rtnl_lock();
 290         if (strchr(dev->name, '%')) {
 291                 ret = dev_alloc_name(dev, dev->name);
 292                 if (ret < 0)
 293                         goto out_free;
 294         }
 295
 296         SET_NETDEV_DEVTYPE(dev, &br_type);
 297
 298         ret = register_netdevice(dev);
 299         if (ret)
 300                 goto out_free;
 301
 302         ret = br_sysfs_addbr(dev);
 303         if (ret)
 304                 unregister_netdevice(dev);
 305  out:
 306         rtnl_unlock();
 307         return ret;
 308
 309 out_free:
 310         free_netdev(dev);
 311         goto out;
 312 }
 313
 314 int br_del_bridge(struct net *net, const char *name)
 315 {
 316         struct net_device *dev;
 317         int ret = 0;
 318
 319         rtnl_lock();
 320         dev = __dev_get_by_name(net, name);
 321         if (dev == NULL)
 322                 ret =  -ENXIO;  /* Could not find device */
 323
 324         else if (!(dev->priv_flags & IFF_EBRIDGE)) {
 325                 /* Attempt to delete non bridge device! */
 326                 ret = -EPERM;
 327         }
 328
 329         else if (dev->flags & IFF_UP) {
 330                 /* Not shutdown yet. */
 331                 ret = -EBUSY;
 332         }
 333
 334         else
 335                 del_br(netdev_priv(dev), NULL);
 336
 337         rtnl_unlock();
 338         return ret;
 339 }
 340
 341 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
 342 int br_min_mtu(const struct net_bridge *br)
 343 {
 344         const struct net_bridge_port *p;
 345         int mtu = 0;
 346
 347         ASSERT_RTNL();
 348
 349         if (list_empty(&br->port_list))
 350                 mtu = ETH_DATA_LEN;
 351         else {
 352                 list_for_each_entry(p, &br->port_list, list) {
 353                         if (!mtu  || p->dev->mtu < mtu)
 354                                 mtu = p->dev->mtu;
 355                 }
 356         }
 357         return mtu;
 358 }
 359
 360 /*
 361  * Recomputes features using slave's features
 362  */
 363 void br_features_recompute(struct net_bridge *br)
 364 {
 365         struct net_bridge_port *p;
 366         unsigned long features, mask;
 367
 368         features = mask = br->feature_mask;
 369         if (list_empty(&br->port_list))
 370                 goto done;
 371
 372         features &= ~NETIF_F_ONE_FOR_ALL;
 373
 374         list_for_each_entry(p, &br->port_list, list) {
 375                 features = netdev_increment_features(features,
 376                                                      p->dev->features, mask);
 377         }
 378
 379 done:
 380         br->dev->features = netdev_fix_features(features, NULL);
 381 }
 382
 383 /* called with RTNL */
 384 int br_add_if(struct net_bridge *br, struct net_device *dev)
 385 {
 386         struct net_bridge_port *p;
 387         int err = 0;
 388
 389         /* Don't allow bridging non-ethernet like devices */
 390         if ((dev->flags & IFF_LOOPBACK) ||
 391             dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN)
 392                 return -EINVAL;
 393
 394         /* No bridging of bridges */
 395         if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit)
 396                 return -ELOOP;
 397
 398         /* Device is already being bridged */
 399         if (dev->br_port != NULL)
 400                 return -EBUSY;
 401
 402         /* No bridging devices that dislike that (e.g. wireless) */
 403         if (dev->priv_flags & IFF_DONT_BRIDGE)
 404                 return -EOPNOTSUPP;
 405
 406         p = new_nbp(br, dev);
 407         if (IS_ERR(p))
 408                 return PTR_ERR(p);
 409
 410         err = dev_set_promiscuity(dev, 1);
 411         if (err)
 412                 goto put_back;
 413
 414         err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
 415                                    SYSFS_BRIDGE_PORT_ATTR);
 416         if (err)
 417                 goto err0;
 418
 419         err = br_fdb_insert(br, p, dev->dev_addr);
 420         if (err)
 421                 goto err1;
 422
 423         err = br_sysfs_addif(p);
 424         if (err)
 425                 goto err2;
 426
 427         rcu_assign_pointer(dev->br_port, p);
 428         dev_disable_lro(dev);
 429
 430         list_add_rcu(&p->list, &br->port_list);
 431
 432         spin_lock_bh(&br->lock);
 433         br_stp_recalculate_bridge_id(br);
 434         br_features_recompute(br);
 435
 436         if ((dev->flags & IFF_UP) && netif_carrier_ok(dev) &&
 437             (br->dev->flags & IFF_UP))
 438                 br_stp_enable_port(p);
 439         spin_unlock_bh(&br->lock);
 440
 441         br_ifinfo_notify(RTM_NEWLINK, p);
 442
 443         dev_set_mtu(br->dev, br_min_mtu(br));
 444
 445         kobject_uevent(&p->kobj, KOBJ_ADD);
 446
 447         return 0;
 448 err2:
 449         br_fdb_delete_by_port(br, p, 1);
 450 err1:
 451         kobject_put(&p->kobj);
 452         p = NULL; /* kobject_put frees */
 453 err0:
 454         dev_set_promiscuity(dev, -1);
 455 put_back:
 456         dev_put(dev);
 457         kfree(p);
 458         return err;
 459 }
 460
 461 /* called with RTNL */
 462 int br_del_if(struct net_bridge *br, struct net_device *dev)
 463 {
 464         struct net_bridge_port *p = dev->br_port;
 465
 466         if (!p || p->br != br)
 467                 return -EINVAL;
 468
 469         del_nbp(p);
 470
 471         spin_lock_bh(&br->lock);
 472         br_stp_recalculate_bridge_id(br);
 473         br_features_recompute(br);
 474         spin_unlock_bh(&br->lock);
 475
 476         return 0;
 477 }
 478
 479 void __net_exit br_net_exit(struct net *net)
 480 {
 481         struct net_device *dev;
 482         LIST_HEAD(list);
 483
 484         rtnl_lock();
 485         for_each_netdev(net, dev)
 486                 if (dev->priv_flags & IFF_EBRIDGE)
 487                         del_br(netdev_priv(dev), &list);
 488
 489         unregister_netdevice_many(&list);
 490         rtnl_unlock();
 491
 492 }