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 transistions. */
  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 release_nbp(struct kobject *kobj)
  89 {
  90         struct net_bridge_port *p
  91                 = container_of(kobj, struct net_bridge_port, kobj);
  92         kfree(p);
  93 }
  94
  95 static struct kobj_type brport_ktype = {
  96 #ifdef CONFIG_SYSFS
  97         .sysfs_ops = &brport_sysfs_ops,
  98 #endif
  99         .release = release_nbp,
 100 };
 101
 102 static void destroy_nbp(struct net_bridge_port *p)
 103 {
 104         struct net_device *dev = p->dev;
 105
 106         p->br = NULL;
 107         p->dev = NULL;
 108         dev_put(dev);
 109
 110         kobject_put(&p->kobj);
 111 }
 112
 113 static void destroy_nbp_rcu(struct rcu_head *head)
 114 {
 115         struct net_bridge_port *p =
 116                         container_of(head, struct net_bridge_port, rcu);
 117         destroy_nbp(p);
 118 }
 119
 120 /* Delete port(interface) from bridge is done in two steps.
 121  * via RCU. First step, marks device as down. That deletes
 122  * all the timers and stops new packets from flowing through.
 123  *
 124  * Final cleanup doesn't occur until after all CPU's finished
 125  * processing packets.
 126  *
 127  * Protected from multiple admin operations by RTNL mutex
 128  */
 129 static void del_nbp(struct net_bridge_port *p)
 130 {
 131         struct net_bridge *br = p->br;
 132         struct net_device *dev = p->dev;
 133
 134         sysfs_remove_link(br->ifobj, p->dev->name);
 135
 136         dev_set_promiscuity(dev, -1);
 137
 138         spin_lock_bh(&br->lock);
 139         br_stp_disable_port(p);
 140         spin_unlock_bh(&br->lock);
 141
 142         br_ifinfo_notify(RTM_DELLINK, p);
 143
 144         nbp_vlan_flush(p);
 145         br_fdb_delete_by_port(br, p, 1);
 146
 147         list_del_rcu(&p->list);
 148
 149         dev->priv_flags &= ~IFF_BRIDGE_PORT;
 150
 151         netdev_rx_handler_unregister(dev);
 152         synchronize_net();
 153
 154         netdev_upper_dev_unlink(dev, br->dev);
 155
 156         br_multicast_del_port(p);
 157
 158         kobject_uevent(&p->kobj, KOBJ_REMOVE);
 159         kobject_del(&p->kobj);
 160
 161         br_netpoll_disable(p);
 162
 163         call_rcu(&p->rcu, destroy_nbp_rcu);
 164 }
 165
 166 /* Delete bridge device */
 167 void br_dev_delete(struct net_device *dev, struct list_head *head)
 168 {
 169         struct net_bridge *br = netdev_priv(dev);
 170         struct net_bridge_port *p, *n;
 171
 172         list_for_each_entry_safe(p, n, &br->port_list, list) {
 173                 del_nbp(p);
 174         }
 175
 176         del_timer_sync(&br->gc_timer);
 177
 178         br_sysfs_delbr(br->dev);
 179         unregister_netdevice_queue(br->dev, head);
 180 }
 181
 182 /* find an available port number */
 183 static int find_portno(struct net_bridge *br)
 184 {
 185         int index;
 186         struct net_bridge_port *p;
 187         unsigned long *inuse;
 188
 189         inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
 190                         GFP_KERNEL);
 191         if (!inuse)
 192                 return -ENOMEM;
 193
 194         set_bit(0, inuse);      /* zero is reserved */
 195         list_for_each_entry(p, &br->port_list, list) {
 196                 set_bit(p->port_no, inuse);
 197         }
 198         index = find_first_zero_bit(inuse, BR_MAX_PORTS);
 199         kfree(inuse);
 200
 201         return (index >= BR_MAX_PORTS) ? -EXFULL : index;
 202 }
 203
 204 /* called with RTNL but without bridge lock */
 205 static struct net_bridge_port *new_nbp(struct net_bridge *br,
 206                                        struct net_device *dev)
 207 {
 208         int index;
 209         struct net_bridge_port *p;
 210
 211         index = find_portno(br);
 212         if (index < 0)
 213                 return ERR_PTR(index);
 214
 215         p = kzalloc(sizeof(*p), GFP_KERNEL);
 216         if (p == NULL)
 217                 return ERR_PTR(-ENOMEM);
 218
 219         p->br = br;
 220         dev_hold(dev);
 221         p->dev = dev;
 222         p->path_cost = port_cost(dev);
 223         p->priority = 0x8000 >> BR_PORT_BITS;
 224         p->port_no = index;
 225         p->flags = 0;
 226         br_init_port(p);
 227         p->state = BR_STATE_DISABLED;
 228         br_stp_port_timer_init(p);
 229         br_multicast_add_port(p);
 230
 231         return p;
 232 }
 233
 234 int br_add_bridge(struct net *net, const char *name)
 235 {
 236         struct net_device *dev;
 237         int res;
 238
 239         dev = alloc_netdev(sizeof(struct net_bridge), name,
 240                            br_dev_setup);
 241
 242         if (!dev)
 243                 return -ENOMEM;
 244
 245         dev_net_set(dev, net);
 246         dev->rtnl_link_ops = &br_link_ops;
 247
 248         res = register_netdev(dev);
 249         if (res)
 250                 free_netdev(dev);
 251         return res;
 252 }
 253
 254 int br_del_bridge(struct net *net, const char *name)
 255 {
 256         struct net_device *dev;
 257         int ret = 0;
 258
 259         rtnl_lock();
 260         dev = __dev_get_by_name(net, name);
 261         if (dev == NULL)
 262                 ret =  -ENXIO;  /* Could not find device */
 263
 264         else if (!(dev->priv_flags & IFF_EBRIDGE)) {
 265                 /* Attempt to delete non bridge device! */
 266                 ret = -EPERM;
 267         }
 268
 269         else if (dev->flags & IFF_UP) {
 270                 /* Not shutdown yet. */
 271                 ret = -EBUSY;
 272         }
 273
 274         else
 275                 br_dev_delete(dev, NULL);
 276
 277         rtnl_unlock();
 278         return ret;
 279 }
 280
 281 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
 282 int br_min_mtu(const struct net_bridge *br)
 283 {
 284         const struct net_bridge_port *p;
 285         int mtu = 0;
 286
 287         ASSERT_RTNL();
 288
 289         if (list_empty(&br->port_list))
 290                 mtu = ETH_DATA_LEN;
 291         else {
 292                 list_for_each_entry(p, &br->port_list, list) {
 293                         if (!mtu  || p->dev->mtu < mtu)
 294                                 mtu = p->dev->mtu;
 295                 }
 296         }
 297         return mtu;
 298 }
 299
 300 /*
 301  * Recomputes features using slave's features
 302  */
 303 netdev_features_t br_features_recompute(struct net_bridge *br,
 304         netdev_features_t features)
 305 {
 306         struct net_bridge_port *p;
 307         netdev_features_t mask;
 308
 309         if (list_empty(&br->port_list))
 310                 return features;
 311
 312         mask = features;
 313         features &= ~NETIF_F_ONE_FOR_ALL;
 314
 315         list_for_each_entry(p, &br->port_list, list) {
 316                 features = netdev_increment_features(features,
 317                                                      p->dev->features, mask);
 318         }
 319
 320         return features;
 321 }
 322
 323 /* called with RTNL */
 324 int br_add_if(struct net_bridge *br, struct net_device *dev)
 325 {
 326         struct net_bridge_port *p;
 327         int err = 0;
 328         bool changed_addr;
 329
 330         /* Don't allow bridging non-ethernet like devices */
 331         if ((dev->flags & IFF_LOOPBACK) ||
 332             dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN ||
 333             !is_valid_ether_addr(dev->dev_addr))
 334                 return -EINVAL;
 335
 336         /* No bridging of bridges */
 337         if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit)
 338                 return -ELOOP;
 339
 340         /* Device is already being bridged */
 341         if (br_port_exists(dev))
 342                 return -EBUSY;
 343
 344         /* No bridging devices that dislike that (e.g. wireless) */
 345         if (dev->priv_flags & IFF_DONT_BRIDGE)
 346                 return -EOPNOTSUPP;
 347
 348         p = new_nbp(br, dev);
 349         if (IS_ERR(p))
 350                 return PTR_ERR(p);
 351
 352         call_netdevice_notifiers(NETDEV_JOIN, dev);
 353
 354         err = dev_set_promiscuity(dev, 1);
 355         if (err)
 356                 goto put_back;
 357
 358         err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
 359                                    SYSFS_BRIDGE_PORT_ATTR);
 360         if (err)
 361                 goto err1;
 362
 363         err = br_sysfs_addif(p);
 364         if (err)
 365                 goto err2;
 366
 367         if (br_netpoll_info(br) && ((err = br_netpoll_enable(p, GFP_KERNEL))))
 368                 goto err3;
 369
 370         err = netdev_master_upper_dev_link(dev, br->dev);
 371         if (err)
 372                 goto err4;
 373
 374         err = netdev_rx_handler_register(dev, br_handle_frame, p);
 375         if (err)
 376                 goto err5;
 377
 378         dev->priv_flags |= IFF_BRIDGE_PORT;
 379
 380         dev_disable_lro(dev);
 381
 382         list_add_rcu(&p->list, &br->port_list);
 383
 384         netdev_update_features(br->dev);
 385
 386         spin_lock_bh(&br->lock);
 387         changed_addr = br_stp_recalculate_bridge_id(br);
 388
 389         if (netif_running(dev) && netif_oper_up(dev) &&
 390             (br->dev->flags & IFF_UP))
 391                 br_stp_enable_port(p);
 392         spin_unlock_bh(&br->lock);
 393
 394         br_ifinfo_notify(RTM_NEWLINK, p);
 395
 396         if (changed_addr)
 397                 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
 398
 399         dev_set_mtu(br->dev, br_min_mtu(br));
 400
 401         if (br_fdb_insert(br, p, dev->dev_addr, 0))
 402                 netdev_err(dev, "failed insert local address bridge forwarding table\n");
 403
 404         kobject_uevent(&p->kobj, KOBJ_ADD);
 405
 406         return 0;
 407
 408 err5:
 409         netdev_upper_dev_unlink(dev, br->dev);
 410 err4:
 411         br_netpoll_disable(p);
 412 err3:
 413         sysfs_remove_link(br->ifobj, p->dev->name);
 414 err2:
 415         kobject_put(&p->kobj);
 416         p = NULL; /* kobject_put frees */
 417 err1:
 418         dev_set_promiscuity(dev, -1);
 419 put_back:
 420         dev_put(dev);
 421         kfree(p);
 422         return err;
 423 }
 424
 425 /* called with RTNL */
 426 int br_del_if(struct net_bridge *br, struct net_device *dev)
 427 {
 428         struct net_bridge_port *p;
 429         bool changed_addr;
 430
 431         p = br_port_get_rtnl(dev);
 432         if (!p || p->br != br)
 433                 return -EINVAL;
 434
 435         /* Since more than one interface can be attached to a bridge,
 436          * there still maybe an alternate path for netconsole to use;
 437          * therefore there is no reason for a NETDEV_RELEASE event.
 438          */
 439         del_nbp(p);
 440
 441         spin_lock_bh(&br->lock);
 442         changed_addr = br_stp_recalculate_bridge_id(br);
 443         spin_unlock_bh(&br->lock);
 444
 445         if (changed_addr)
 446                 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
 447
 448         netdev_update_features(br->dev);
 449
 450         return 0;
 451 }
 452
 453 void __net_exit br_net_exit(struct net *net)
 454 {
 455         struct net_device *dev;
 456         LIST_HEAD(list);
 457
 458         rtnl_lock();
 459         for_each_netdev(net, dev)
 460                 if (dev->priv_flags & IFF_EBRIDGE)
 461                         br_dev_delete(dev, &list);
 462
 463         unregister_netdevice_many(&list);
 464         rtnl_unlock();
 465
 466 }