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/netpoll.h>
  17 #include <linux/ethtool.h>
  18 #include <linux/if_arp.h>
  19 #include <linux/module.h>
  20 #include <linux/init.h>
  21 #include <linux/rtnetlink.h>
  22 #include <linux/if_ether.h>
  23 #include <linux/slab.h>
  24 #include <net/sock.h>
  25
  26 #include "br_private.h"
  27
  28 /*
  29  * Determine initial path cost based on speed.
  30  * using recommendations from 802.1d standard
  31  *
  32  * Since driver might sleep need to not be holding any locks.
  33  */
  34 static int port_cost(struct net_device *dev)
  35 {
  36         if (dev->ethtool_ops && dev->ethtool_ops->get_settings) {
  37                 struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET, };
  38
  39                 if (!dev_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
  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_carrier_ok(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_carrier_ok(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         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
 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 /* called with RTNL */
 164 static void del_br(struct net_bridge *br, struct list_head *head)
 165 {
 166         struct net_bridge_port *p, *n;
 167
 168         list_for_each_entry_safe(p, n, &br->port_list, list) {
 169                 del_nbp(p);
 170         }
 171
 172         del_timer_sync(&br->gc_timer);
 173
 174         br_sysfs_delbr(br->dev);
 175         unregister_netdevice_queue(br->dev, head);
 176 }
 177
 178 /* find an available port number */
 179 static int find_portno(struct net_bridge *br)
 180 {
 181         int index;
 182         struct net_bridge_port *p;
 183         unsigned long *inuse;
 184
 185         inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
 186                         GFP_KERNEL);
 187         if (!inuse)
 188                 return -ENOMEM;
 189
 190         set_bit(0, inuse);      /* zero is reserved */
 191         list_for_each_entry(p, &br->port_list, list) {
 192                 set_bit(p->port_no, inuse);
 193         }
 194         index = find_first_zero_bit(inuse, BR_MAX_PORTS);
 195         kfree(inuse);
 196
 197         return (index >= BR_MAX_PORTS) ? -EXFULL : index;
 198 }
 199
 200 /* called with RTNL but without bridge lock */
 201 static struct net_bridge_port *new_nbp(struct net_bridge *br,
 202                                        struct net_device *dev)
 203 {
 204         int index;
 205         struct net_bridge_port *p;
 206
 207         index = find_portno(br);
 208         if (index < 0)
 209                 return ERR_PTR(index);
 210
 211         p = kzalloc(sizeof(*p), GFP_KERNEL);
 212         if (p == NULL)
 213                 return ERR_PTR(-ENOMEM);
 214
 215         p->br = br;
 216         dev_hold(dev);
 217         p->dev = dev;
 218         p->path_cost = port_cost(dev);
 219         p->priority = 0x8000 >> BR_PORT_BITS;
 220         p->port_no = index;
 221         p->flags = 0;
 222         br_init_port(p);
 223         p->state = BR_STATE_DISABLED;
 224         br_stp_port_timer_init(p);
 225         br_multicast_add_port(p);
 226
 227         return p;
 228 }
 229
 230 int br_add_bridge(struct net *net, const char *name)
 231 {
 232         struct net_device *dev;
 233
 234         dev = alloc_netdev(sizeof(struct net_bridge), name,
 235                            br_dev_setup);
 236
 237         if (!dev)
 238                 return -ENOMEM;
 239
 240         dev_net_set(dev, net);
 241
 242         return register_netdev(dev);
 243 }
 244
 245 int br_del_bridge(struct net *net, const char *name)
 246 {
 247         struct net_device *dev;
 248         int ret = 0;
 249
 250         rtnl_lock();
 251         dev = __dev_get_by_name(net, name);
 252         if (dev == NULL)
 253                 ret =  -ENXIO;  /* Could not find device */
 254
 255         else if (!(dev->priv_flags & IFF_EBRIDGE)) {
 256                 /* Attempt to delete non bridge device! */
 257                 ret = -EPERM;
 258         }
 259
 260         else if (dev->flags & IFF_UP) {
 261                 /* Not shutdown yet. */
 262                 ret = -EBUSY;
 263         }
 264
 265         else
 266                 del_br(netdev_priv(dev), NULL);
 267
 268         rtnl_unlock();
 269         return ret;
 270 }
 271
 272 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
 273 int br_min_mtu(const struct net_bridge *br)
 274 {
 275         const struct net_bridge_port *p;
 276         int mtu = 0;
 277
 278         ASSERT_RTNL();
 279
 280         if (list_empty(&br->port_list))
 281                 mtu = ETH_DATA_LEN;
 282         else {
 283                 list_for_each_entry(p, &br->port_list, list) {
 284                         if (!mtu  || p->dev->mtu < mtu)
 285                                 mtu = p->dev->mtu;
 286                 }
 287         }
 288         return mtu;
 289 }
 290
 291 /*
 292  * Recomputes features using slave's features
 293  */
 294 u32 br_features_recompute(struct net_bridge *br, u32 features)
 295 {
 296         struct net_bridge_port *p;
 297         u32 mask;
 298
 299         if (list_empty(&br->port_list))
 300                 return features;
 301
 302         mask = features;
 303         features &= ~NETIF_F_ONE_FOR_ALL;
 304
 305         list_for_each_entry(p, &br->port_list, list) {
 306                 features = netdev_increment_features(features,
 307                                                      p->dev->features, mask);
 308         }
 309
 310         return features;
 311 }
 312
 313 /* called with RTNL */
 314 int br_add_if(struct net_bridge *br, struct net_device *dev)
 315 {
 316         struct net_bridge_port *p;
 317         int err = 0;
 318         bool changed_addr;
 319
 320         /* Don't allow bridging non-ethernet like devices */
 321         if ((dev->flags & IFF_LOOPBACK) ||
 322             dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN)
 323                 return -EINVAL;
 324
 325         /* No bridging of bridges */
 326         if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit)
 327                 return -ELOOP;
 328
 329         /* Device is already being bridged */
 330         if (br_port_exists(dev))
 331                 return -EBUSY;
 332
 333         /* No bridging devices that dislike that (e.g. wireless) */
 334         if (dev->priv_flags & IFF_DONT_BRIDGE)
 335                 return -EOPNOTSUPP;
 336
 337         p = new_nbp(br, dev);
 338         if (IS_ERR(p))
 339                 return PTR_ERR(p);
 340
 341         err = dev_set_promiscuity(dev, 1);
 342         if (err)
 343                 goto put_back;
 344
 345         err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
 346                                    SYSFS_BRIDGE_PORT_ATTR);
 347         if (err)
 348                 goto err0;
 349
 350         err = br_fdb_insert(br, p, dev->dev_addr);
 351         if (err)
 352                 goto err1;
 353
 354         err = br_sysfs_addif(p);
 355         if (err)
 356                 goto err2;
 357
 358         if (br_netpoll_info(br) && ((err = br_netpoll_enable(p))))
 359                 goto err3;
 360
 361         err = netdev_set_master(dev, br->dev);
 362         if (err)
 363                 goto err3;
 364
 365         err = netdev_rx_handler_register(dev, br_handle_frame, p);
 366         if (err)
 367                 goto err4;
 368
 369         dev->priv_flags |= IFF_BRIDGE_PORT;
 370
 371         dev_disable_lro(dev);
 372
 373         list_add_rcu(&p->list, &br->port_list);
 374
 375         netdev_update_features(br->dev);
 376
 377         spin_lock_bh(&br->lock);
 378         changed_addr = br_stp_recalculate_bridge_id(br);
 379
 380         if ((dev->flags & IFF_UP) && netif_carrier_ok(dev) &&
 381             (br->dev->flags & IFF_UP))
 382                 br_stp_enable_port(p);
 383         spin_unlock_bh(&br->lock);
 384
 385         br_ifinfo_notify(RTM_NEWLINK, p);
 386
 387         if (changed_addr)
 388                 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
 389
 390         dev_set_mtu(br->dev, br_min_mtu(br));
 391
 392         kobject_uevent(&p->kobj, KOBJ_ADD);
 393
 394         return 0;
 395
 396 err4:
 397         netdev_set_master(dev, NULL);
 398 err3:
 399         sysfs_remove_link(br->ifobj, p->dev->name);
 400 err2:
 401         br_fdb_delete_by_port(br, p, 1);
 402 err1:
 403         kobject_put(&p->kobj);
 404         p = NULL; /* kobject_put frees */
 405 err0:
 406         dev_set_promiscuity(dev, -1);
 407 put_back:
 408         dev_put(dev);
 409         kfree(p);
 410         return err;
 411 }
 412
 413 /* called with RTNL */
 414 int br_del_if(struct net_bridge *br, struct net_device *dev)
 415 {
 416         struct net_bridge_port *p;
 417
 418         p = br_port_get_rtnl(dev);
 419         if (!p || p->br != br)
 420                 return -EINVAL;
 421
 422         del_nbp(p);
 423
 424         spin_lock_bh(&br->lock);
 425         br_stp_recalculate_bridge_id(br);
 426         spin_unlock_bh(&br->lock);
 427
 428         netdev_update_features(br->dev);
 429
 430         return 0;
 431 }
 432
 433 void __net_exit br_net_exit(struct net *net)
 434 {
 435         struct net_device *dev;
 436         LIST_HEAD(list);
 437
 438         rtnl_lock();
 439         for_each_netdev(net, dev)
 440                 if (dev->priv_flags & IFF_EBRIDGE)
 441                         del_br(netdev_priv(dev), &list);
 442
 443         unregister_netdevice_many(&list);
 444         rtnl_unlock();
 445
 446 }