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 (netif_running(dev) && netif_oper_up(dev))
  71                 p->path_cost = port_cost(dev);
  72
  73         if (!netif_running(br->dev))
  74                 return;
  75
  76         spin_lock_bh(&br->lock);
  77         if (netif_running(dev) && netif_oper_up(dev)) {
  78                 if (p->state == BR_STATE_DISABLED)
  79                         br_stp_enable_port(p);
  80         } else {
  81                 if (p->state != BR_STATE_DISABLED)
  82                         br_stp_disable_port(p);
  83         }
  84         spin_unlock_bh(&br->lock);
  85 }
  86
  87 static void release_nbp(struct kobject *kobj)
  88 {
  89         struct net_bridge_port *p
  90                 = container_of(kobj, struct net_bridge_port, kobj);
  91         kfree(p);
  92 }
  93
  94 static struct kobj_type brport_ktype = {
  95 #ifdef CONFIG_SYSFS
  96         .sysfs_ops = &brport_sysfs_ops,
  97 #endif
  98         .release = release_nbp,
  99 };
 100
 101 static void destroy_nbp(struct net_bridge_port *p)
 102 {
 103         struct net_device *dev = p->dev;
 104
 105         p->br = NULL;
 106         p->dev = NULL;
 107         dev_put(dev);
 108
 109         kobject_put(&p->kobj);
 110 }
 111
 112 static void destroy_nbp_rcu(struct rcu_head *head)
 113 {
 114         struct net_bridge_port *p =
 115                         container_of(head, struct net_bridge_port, rcu);
 116         destroy_nbp(p);
 117 }
 118
 119 /* Delete port(interface) from bridge is done in two steps.
 120  * via RCU. First step, marks device as down. That deletes
 121  * all the timers and stops new packets from flowing through.
 122  *
 123  * Final cleanup doesn't occur until after all CPU's finished
 124  * processing packets.
 125  *
 126  * Protected from multiple admin operations by RTNL mutex
 127  */
 128 static void del_nbp(struct net_bridge_port *p)
 129 {
 130         struct net_bridge *br = p->br;
 131         struct net_device *dev = p->dev;
 132
 133         sysfs_remove_link(br->ifobj, p->dev->name);
 134
 135         dev_set_promiscuity(dev, -1);
 136
 137         spin_lock_bh(&br->lock);
 138         br_stp_disable_port(p);
 139         spin_unlock_bh(&br->lock);
 140
 141         br_ifinfo_notify(RTM_DELLINK, p);
 142
 143         nbp_vlan_flush(p);
 144         br_fdb_delete_by_port(br, p, 1);
 145
 146         list_del_rcu(&p->list);
 147
 148         dev->priv_flags &= ~IFF_BRIDGE_PORT;
 149
 150         netdev_rx_handler_unregister(dev);
 151         synchronize_net();
 152
 153         netdev_upper_dev_unlink(dev, br->dev);
 154
 155         br_multicast_del_port(p);
 156
 157         kobject_uevent(&p->kobj, KOBJ_REMOVE);
 158         kobject_del(&p->kobj);
 159
 160         br_netpoll_disable(p);
 161
 162         call_rcu(&p->rcu, destroy_nbp_rcu);
 163 }
 164
 165 /* Delete bridge device */
 166 void br_dev_delete(struct net_device *dev, struct list_head *head)
 167 {
 168         struct net_bridge *br = netdev_priv(dev);
 169         struct net_bridge_port *p, *n;
 170
 171         list_for_each_entry_safe(p, n, &br->port_list, list) {
 172                 del_nbp(p);
 173         }
 174
 175         del_timer_sync(&br->gc_timer);
 176
 177         br_sysfs_delbr(br->dev);
 178         unregister_netdevice_queue(br->dev, head);
 179 }
 180
 181 /* find an available port number */
 182 static int find_portno(struct net_bridge *br)
 183 {
 184         int index;
 185         struct net_bridge_port *p;
 186         unsigned long *inuse;
 187
 188         inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
 189                         GFP_KERNEL);
 190         if (!inuse)
 191                 return -ENOMEM;
 192
 193         set_bit(0, inuse);      /* zero is reserved */
 194         list_for_each_entry(p, &br->port_list, list) {
 195                 set_bit(p->port_no, inuse);
 196         }
 197         index = find_first_zero_bit(inuse, BR_MAX_PORTS);
 198         kfree(inuse);
 199
 200         return (index >= BR_MAX_PORTS) ? -EXFULL : index;
 201 }
 202
 203 /* called with RTNL but without bridge lock */
 204 static struct net_bridge_port *new_nbp(struct net_bridge *br,
 205                                        struct net_device *dev)
 206 {
 207         int index;
 208         struct net_bridge_port *p;
 209
 210         index = find_portno(br);
 211         if (index < 0)
 212                 return ERR_PTR(index);
 213
 214         p = kzalloc(sizeof(*p), GFP_KERNEL);
 215         if (p == NULL)
 216                 return ERR_PTR(-ENOMEM);
 217
 218         p->br = br;
 219         dev_hold(dev);
 220         p->dev = dev;
 221         p->path_cost = port_cost(dev);
 222         p->priority = 0x8000 >> BR_PORT_BITS;
 223         p->port_no = index;
 224         p->flags = 0;
 225         br_init_port(p);
 226         p->state = BR_STATE_DISABLED;
 227         br_stp_port_timer_init(p);
 228         br_multicast_add_port(p);
 229
 230         return p;
 231 }
 232
 233 int br_add_bridge(struct net *net, const char *name)
 234 {
 235         struct net_device *dev;
 236         int res;
 237
 238         dev = alloc_netdev(sizeof(struct net_bridge), name,
 239                            br_dev_setup);
 240
 241         if (!dev)
 242                 return -ENOMEM;
 243
 244         dev_net_set(dev, net);
 245         dev->rtnl_link_ops = &br_link_ops;
 246
 247         res = register_netdev(dev);
 248         if (res)
 249                 free_netdev(dev);
 250         return res;
 251 }
 252
 253 int br_del_bridge(struct net *net, const char *name)
 254 {
 255         struct net_device *dev;
 256         int ret = 0;
 257
 258         rtnl_lock();
 259         dev = __dev_get_by_name(net, name);
 260         if (dev == NULL)
 261                 ret =  -ENXIO;  /* Could not find device */
 262
 263         else if (!(dev->priv_flags & IFF_EBRIDGE)) {
 264                 /* Attempt to delete non bridge device! */
 265                 ret = -EPERM;
 266         }
 267
 268         else if (dev->flags & IFF_UP) {
 269                 /* Not shutdown yet. */
 270                 ret = -EBUSY;
 271         }
 272
 273         else
 274                 br_dev_delete(dev, NULL);
 275
 276         rtnl_unlock();
 277         return ret;
 278 }
 279
 280 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
 281 int br_min_mtu(const struct net_bridge *br)
 282 {
 283         const struct net_bridge_port *p;
 284         int mtu = 0;
 285
 286         ASSERT_RTNL();
 287
 288         if (list_empty(&br->port_list))
 289                 mtu = ETH_DATA_LEN;
 290         else {
 291                 list_for_each_entry(p, &br->port_list, list) {
 292                         if (!mtu  || p->dev->mtu < mtu)
 293                                 mtu = p->dev->mtu;
 294                 }
 295         }
 296         return mtu;
 297 }
 298
 299 /*
 300  * Recomputes features using slave's features
 301  */
 302 netdev_features_t br_features_recompute(struct net_bridge *br,
 303         netdev_features_t features)
 304 {
 305         struct net_bridge_port *p;
 306         netdev_features_t mask;
 307
 308         if (list_empty(&br->port_list))
 309                 return features;
 310
 311         mask = features;
 312         features &= ~NETIF_F_ONE_FOR_ALL;
 313
 314         list_for_each_entry(p, &br->port_list, list) {
 315                 features = netdev_increment_features(features,
 316                                                      p->dev->features, mask);
 317         }
 318
 319         return features;
 320 }
 321
 322 /* called with RTNL */
 323 int br_add_if(struct net_bridge *br, struct net_device *dev)
 324 {
 325         struct net_bridge_port *p;
 326         int err = 0;
 327         bool changed_addr;
 328
 329         /* Don't allow bridging non-ethernet like devices */
 330         if ((dev->flags & IFF_LOOPBACK) ||
 331             dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN ||
 332             !is_valid_ether_addr(dev->dev_addr))
 333                 return -EINVAL;
 334
 335         /* No bridging of bridges */
 336         if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit)
 337                 return -ELOOP;
 338
 339         /* Device is already being bridged */
 340         if (br_port_exists(dev))
 341                 return -EBUSY;
 342
 343         /* No bridging devices that dislike that (e.g. wireless) */
 344         if (dev->priv_flags & IFF_DONT_BRIDGE)
 345                 return -EOPNOTSUPP;
 346
 347         p = new_nbp(br, dev);
 348         if (IS_ERR(p))
 349                 return PTR_ERR(p);
 350
 351         call_netdevice_notifiers(NETDEV_JOIN, dev);
 352
 353         err = dev_set_promiscuity(dev, 1);
 354         if (err)
 355                 goto put_back;
 356
 357         err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
 358                                    SYSFS_BRIDGE_PORT_ATTR);
 359         if (err)
 360                 goto err1;
 361
 362         err = br_sysfs_addif(p);
 363         if (err)
 364                 goto err2;
 365
 366         if (br_netpoll_info(br) && ((err = br_netpoll_enable(p, GFP_KERNEL))))
 367                 goto err3;
 368
 369         err = netdev_master_upper_dev_link(dev, br->dev);
 370         if (err)
 371                 goto err4;
 372
 373         err = netdev_rx_handler_register(dev, br_handle_frame, p);
 374         if (err)
 375                 goto err5;
 376
 377         dev->priv_flags |= IFF_BRIDGE_PORT;
 378
 379         dev_disable_lro(dev);
 380
 381         list_add_rcu(&p->list, &br->port_list);
 382
 383         netdev_update_features(br->dev);
 384
 385         spin_lock_bh(&br->lock);
 386         changed_addr = br_stp_recalculate_bridge_id(br);
 387
 388         if (netif_running(dev) && netif_oper_up(dev) &&
 389             (br->dev->flags & IFF_UP))
 390                 br_stp_enable_port(p);
 391         spin_unlock_bh(&br->lock);
 392
 393         br_ifinfo_notify(RTM_NEWLINK, p);
 394
 395         if (changed_addr)
 396                 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
 397
 398         dev_set_mtu(br->dev, br_min_mtu(br));
 399
 400         if (br_fdb_insert(br, p, dev->dev_addr, 0))
 401                 netdev_err(dev, "failed insert local address bridge forwarding table\n");
 402
 403         kobject_uevent(&p->kobj, KOBJ_ADD);
 404
 405         return 0;
 406
 407 err5:
 408         netdev_upper_dev_unlink(dev, br->dev);
 409 err4:
 410         br_netpoll_disable(p);
 411 err3:
 412         sysfs_remove_link(br->ifobj, p->dev->name);
 413 err2:
 414         kobject_put(&p->kobj);
 415         p = NULL; /* kobject_put frees */
 416 err1:
 417         dev_set_promiscuity(dev, -1);
 418 put_back:
 419         dev_put(dev);
 420         kfree(p);
 421         return err;
 422 }
 423
 424 /* called with RTNL */
 425 int br_del_if(struct net_bridge *br, struct net_device *dev)
 426 {
 427         struct net_bridge_port *p;
 428         bool changed_addr;
 429
 430         p = br_port_get_rtnl(dev);
 431         if (!p || p->br != br)
 432                 return -EINVAL;
 433
 434         /* Since more than one interface can be attached to a bridge,
 435          * there still maybe an alternate path for netconsole to use;
 436          * therefore there is no reason for a NETDEV_RELEASE event.
 437          */
 438         del_nbp(p);
 439
 440         spin_lock_bh(&br->lock);
 441         changed_addr = br_stp_recalculate_bridge_id(br);
 442         spin_unlock_bh(&br->lock);
 443
 444         if (changed_addr)
 445                 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
 446
 447         netdev_update_features(br->dev);
 448
 449         return 0;
 450 }
 451
 452 void __net_exit br_net_exit(struct net *net)
 453 {
 454         struct net_device *dev;
 455         LIST_HEAD(list);
 456
 457         rtnl_lock();
 458         for_each_netdev(net, dev)
 459                 if (dev->priv_flags & IFF_EBRIDGE)
 460                         br_dev_delete(dev, &list);
 461
 462         unregister_netdevice_many(&list);
 463         rtnl_unlock();
 464
 465 }