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
  27 #include "br_private.h"
  28
  29 /*
  30  * Determine initial path cost based on speed.
  31  * using recommendations from 802.1d standard
  32  *
  33  * Since driver might sleep need to not be holding any locks.
  34  */
  35 static int port_cost(struct net_device *dev)
  36 {
  37         struct ethtool_cmd ecmd;
  38
  39         if (!__ethtool_get_settings(dev, &ecmd)) {
  40                 switch (ethtool_cmd_speed(&ecmd)) {
  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         /* 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 /* Check for port carrier transistions. */
  64 void br_port_carrier_check(struct net_bridge_port *p)
  65 {
  66         struct net_device *dev = p->dev;
  67         struct net_bridge *br = p->br;
  68
  69         if (netif_running(dev) && netif_carrier_ok(dev))
  70                 p->path_cost = port_cost(dev);
  71
  72         if (!netif_running(br->dev))
  73                 return;
  74
  75         spin_lock_bh(&br->lock);
  76         if (netif_running(dev) && netif_carrier_ok(dev)) {
  77                 if (p->state == BR_STATE_DISABLED)
  78                         br_stp_enable_port(p);
  79         } else {
  80                 if (p->state != BR_STATE_DISABLED)
  81                         br_stp_disable_port(p);
  82         }
  83         spin_unlock_bh(&br->lock);
  84 }
  85
  86 static void release_nbp(struct kobject *kobj)
  87 {
  88         struct net_bridge_port *p
  89                 = container_of(kobj, struct net_bridge_port, kobj);
  90         kfree(p);
  91 }
  92
  93 static struct kobj_type brport_ktype = {
  94 #ifdef CONFIG_SYSFS
  95         .sysfs_ops = &brport_sysfs_ops,
  96 #endif
  97         .release = release_nbp,
  98 };
  99
 100 static void destroy_nbp(struct net_bridge_port *p)
 101 {
 102         struct net_device *dev = p->dev;
 103
 104         p->br = NULL;
 105         p->dev = NULL;
 106         dev_put(dev);
 107
 108         kobject_put(&p->kobj);
 109 }
 110
 111 static void destroy_nbp_rcu(struct rcu_head *head)
 112 {
 113         struct net_bridge_port *p =
 114                         container_of(head, struct net_bridge_port, rcu);
 115         destroy_nbp(p);
 116 }
 117
 118 /* Delete port(interface) from bridge is done in two steps.
 119  * via RCU. First step, marks device as down. That deletes
 120  * all the timers and stops new packets from flowing through.
 121  *
 122  * Final cleanup doesn't occur until after all CPU's finished
 123  * processing packets.
 124  *
 125  * Protected from multiple admin operations by RTNL mutex
 126  */
 127 static void del_nbp(struct net_bridge_port *p)
 128 {
 129         struct net_bridge *br = p->br;
 130         struct net_device *dev = p->dev;
 131
 132         sysfs_remove_link(br->ifobj, p->dev->name);
 133
 134         dev_set_promiscuity(dev, -1);
 135
 136         spin_lock_bh(&br->lock);
 137         br_stp_disable_port(p);
 138         spin_unlock_bh(&br->lock);
 139
 140         br_ifinfo_notify(RTM_DELLINK, p);
 141
 142         br_fdb_delete_by_port(br, p, 1);
 143
 144         list_del_rcu(&p->list);
 145
 146         dev->priv_flags &= ~IFF_BRIDGE_PORT;
 147
 148         netdev_rx_handler_unregister(dev);
 149         synchronize_net();
 150
 151         netdev_set_master(dev, 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(p);
 159
 160         call_rcu(&p->rcu, destroy_nbp_rcu);
 161 }
 162
 163 /* Delete bridge device */
 164 void br_dev_delete(struct net_device *dev, struct list_head *head)
 165 {
 166         struct net_bridge *br = netdev_priv(dev);
 167         struct net_bridge_port *p, *n;
 168
 169         list_for_each_entry_safe(p, n, &br->port_list, list) {
 170                 del_nbp(p);
 171         }
 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 /* find an available port number */
 180 static int find_portno(struct net_bridge *br)
 181 {
 182         int index;
 183         struct net_bridge_port *p;
 184         unsigned long *inuse;
 185
 186         inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
 187                         GFP_KERNEL);
 188         if (!inuse)
 189                 return -ENOMEM;
 190
 191         set_bit(0, inuse);      /* zero is reserved */
 192         list_for_each_entry(p, &br->port_list, list) {
 193                 set_bit(p->port_no, inuse);
 194         }
 195         index = find_first_zero_bit(inuse, BR_MAX_PORTS);
 196         kfree(inuse);
 197
 198         return (index >= BR_MAX_PORTS) ? -EXFULL : index;
 199 }
 200
 201 /* called with RTNL but without bridge lock */
 202 static struct net_bridge_port *new_nbp(struct net_bridge *br,
 203                                        struct net_device *dev)
 204 {
 205         int index;
 206         struct net_bridge_port *p;
 207
 208         index = find_portno(br);
 209         if (index < 0)
 210                 return ERR_PTR(index);
 211
 212         p = kzalloc(sizeof(*p), GFP_KERNEL);
 213         if (p == NULL)
 214                 return ERR_PTR(-ENOMEM);
 215
 216         p->br = br;
 217         dev_hold(dev);
 218         p->dev = dev;
 219         p->path_cost = port_cost(dev);
 220         p->priority = 0x8000 >> BR_PORT_BITS;
 221         p->port_no = index;
 222         p->flags = 0;
 223         br_init_port(p);
 224         p->state = BR_STATE_DISABLED;
 225         br_stp_port_timer_init(p);
 226         br_multicast_add_port(p);
 227
 228         return p;
 229 }
 230
 231 int br_add_bridge(struct net *net, const char *name)
 232 {
 233         struct net_device *dev;
 234         int res;
 235
 236         dev = alloc_netdev(sizeof(struct net_bridge), name,
 237                            br_dev_setup);
 238
 239         if (!dev)
 240                 return -ENOMEM;
 241
 242         dev_net_set(dev, net);
 243         dev->rtnl_link_ops = &br_link_ops;
 244
 245         res = register_netdev(dev);
 246         if (res)
 247                 free_netdev(dev);
 248         return res;
 249 }
 250
 251 int br_del_bridge(struct net *net, const char *name)
 252 {
 253         struct net_device *dev;
 254         int ret = 0;
 255
 256         rtnl_lock();
 257         dev = __dev_get_by_name(net, name);
 258         if (dev == NULL)
 259                 ret =  -ENXIO;  /* Could not find device */
 260
 261         else if (!(dev->priv_flags & IFF_EBRIDGE)) {
 262                 /* Attempt to delete non bridge device! */
 263                 ret = -EPERM;
 264         }
 265
 266         else if (dev->flags & IFF_UP) {
 267                 /* Not shutdown yet. */
 268                 ret = -EBUSY;
 269         }
 270
 271         else
 272                 br_dev_delete(dev, NULL);
 273
 274         rtnl_unlock();
 275         return ret;
 276 }
 277
 278 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
 279 int br_min_mtu(const struct net_bridge *br)
 280 {
 281         const struct net_bridge_port *p;
 282         int mtu = 0;
 283
 284         ASSERT_RTNL();
 285
 286         if (list_empty(&br->port_list))
 287                 mtu = ETH_DATA_LEN;
 288         else {
 289                 list_for_each_entry(p, &br->port_list, list) {
 290                         if (!mtu  || p->dev->mtu < mtu)
 291                                 mtu = p->dev->mtu;
 292                 }
 293         }
 294         return mtu;
 295 }
 296
 297 /*
 298  * Recomputes features using slave's features
 299  */
 300 netdev_features_t br_features_recompute(struct net_bridge *br,
 301         netdev_features_t features)
 302 {
 303         struct net_bridge_port *p;
 304         netdev_features_t mask;
 305
 306         if (list_empty(&br->port_list))
 307                 return features;
 308
 309         mask = features;
 310         features &= ~NETIF_F_ONE_FOR_ALL;
 311
 312         list_for_each_entry(p, &br->port_list, list) {
 313                 features = netdev_increment_features(features,
 314                                                      p->dev->features, mask);
 315         }
 316
 317         return features;
 318 }
 319
 320 /* called with RTNL */
 321 int br_add_if(struct net_bridge *br, struct net_device *dev)
 322 {
 323         struct net_bridge_port *p;
 324         int err = 0;
 325         bool changed_addr;
 326
 327         /* Don't allow bridging non-ethernet like devices */
 328         if ((dev->flags & IFF_LOOPBACK) ||
 329             dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN ||
 330             !is_valid_ether_addr(dev->dev_addr))
 331                 return -EINVAL;
 332
 333         /* No bridging of bridges */
 334         if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit)
 335                 return -ELOOP;
 336
 337         /* Device is already being bridged */
 338         if (br_port_exists(dev))
 339                 return -EBUSY;
 340
 341         /* No bridging devices that dislike that (e.g. wireless) */
 342         if (dev->priv_flags & IFF_DONT_BRIDGE)
 343                 return -EOPNOTSUPP;
 344
 345         p = new_nbp(br, dev);
 346         if (IS_ERR(p))
 347                 return PTR_ERR(p);
 348
 349         call_netdevice_notifiers(NETDEV_JOIN, dev);
 350
 351         err = dev_set_promiscuity(dev, 1);
 352         if (err)
 353                 goto put_back;
 354
 355         err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
 356                                    SYSFS_BRIDGE_PORT_ATTR);
 357         if (err)
 358                 goto err1;
 359
 360         err = br_sysfs_addif(p);
 361         if (err)
 362                 goto err2;
 363
 364         if (br_netpoll_info(br) && ((err = br_netpoll_enable(p, GFP_KERNEL))))
 365                 goto err3;
 366
 367         err = netdev_set_master(dev, br->dev);
 368         if (err)
 369                 goto err4;
 370
 371         err = netdev_rx_handler_register(dev, br_handle_frame, p);
 372         if (err)
 373                 goto err5;
 374
 375         dev->priv_flags |= IFF_BRIDGE_PORT;
 376
 377         dev_disable_lro(dev);
 378
 379         list_add_rcu(&p->list, &br->port_list);
 380
 381         netdev_update_features(br->dev);
 382
 383         spin_lock_bh(&br->lock);
 384         changed_addr = br_stp_recalculate_bridge_id(br);
 385
 386         if ((dev->flags & IFF_UP) && netif_carrier_ok(dev) &&
 387             (br->dev->flags & IFF_UP))
 388                 br_stp_enable_port(p);
 389         spin_unlock_bh(&br->lock);
 390
 391         br_ifinfo_notify(RTM_NEWLINK, p);
 392
 393         if (changed_addr)
 394                 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
 395
 396         dev_set_mtu(br->dev, br_min_mtu(br));
 397
 398         if (br_fdb_insert(br, p, dev->dev_addr))
 399                 netdev_err(dev, "failed insert local address bridge forwarding table\n");
 400
 401         kobject_uevent(&p->kobj, KOBJ_ADD);
 402
 403         return 0;
 404
 405 err5:
 406         netdev_set_master(dev, NULL);
 407 err4:
 408         br_netpoll_disable(p);
 409 err3:
 410         sysfs_remove_link(br->ifobj, p->dev->name);
 411 err2:
 412         kobject_put(&p->kobj);
 413         p = NULL; /* kobject_put frees */
 414 err1:
 415         dev_set_promiscuity(dev, -1);
 416 put_back:
 417         dev_put(dev);
 418         kfree(p);
 419         return err;
 420 }
 421
 422 /* called with RTNL */
 423 int br_del_if(struct net_bridge *br, struct net_device *dev)
 424 {
 425         struct net_bridge_port *p;
 426         bool changed_addr;
 427
 428         p = br_port_get_rtnl(dev);
 429         if (!p || p->br != br)
 430                 return -EINVAL;
 431
 432         /* Since more than one interface can be attached to a bridge,
 433          * there still maybe an alternate path for netconsole to use;
 434          * therefore there is no reason for a NETDEV_RELEASE event.
 435          */
 436         del_nbp(p);
 437
 438         spin_lock_bh(&br->lock);
 439         changed_addr = br_stp_recalculate_bridge_id(br);
 440         spin_unlock_bh(&br->lock);
 441
 442         if (changed_addr)
 443                 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
 444
 445         netdev_update_features(br->dev);
 446
 447         return 0;
 448 }
 449
 450 void __net_exit br_net_exit(struct net *net)
 451 {
 452         struct net_device *dev;
 453         LIST_HEAD(list);
 454
 455         rtnl_lock();
 456         for_each_netdev(net, dev)
 457                 if (dev->priv_flags & IFF_EBRIDGE)
 458                         br_dev_delete(dev, &list);
 459
 460         unregister_netdevice_many(&list);
 461         rtnl_unlock();
 462
 463 }