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