net/bridge/br_if.c

   1 /*
   2  *      Userspace interface
   3  *      Linux ethernet bridge
   4  *
   5  *      Authors:
   6  *      Lennert Buytenhek               <buytenh@gnu.org>
   7  *
   8  *      $Id: br_if.c,v 1.7 2001/12/24 00:59:55 davem Exp $
   9  *
  10  *      This program is free software; you can redistribute it and/or
  11  *      modify it under the terms of the GNU General Public License
  12  *      as published by the Free Software Foundation; either version
  13  *      2 of the License, or (at your option) any later version.
  14  */
  15
  16 #include <linux/kernel.h>
  17 #include <linux/netdevice.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 <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_ops->get_settings(dev, &ecmd)) {
  40                         switch(ecmd.speed) {
  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 /*
  65  * Check for port carrier transistions.
  66  * Called from work queue to allow for calling functions that
  67  * might sleep (such as speed check), and to debounce.
  68  */
  69 void br_port_carrier_check(struct net_bridge_port *p)
  70 {
  71         struct net_device *dev = p->dev;
  72         struct net_bridge *br = p->br;
  73
  74         if (netif_carrier_ok(dev))
  75                 p->path_cost = port_cost(dev);
  76
  77         if (netif_running(br->dev)) {
  78                 spin_lock_bh(&br->lock);
  79                 if (netif_carrier_ok(dev)) {
  80                         if (p->state == BR_STATE_DISABLED)
  81                                 br_stp_enable_port(p);
  82                 } else {
  83                         if (p->state != BR_STATE_DISABLED)
  84                                 br_stp_disable_port(p);
  85                 }
  86                 spin_unlock_bh(&br->lock);
  87         }
  88 }
  89
  90 static void release_nbp(struct kobject *kobj)
  91 {
  92         struct net_bridge_port *p
  93                 = container_of(kobj, struct net_bridge_port, kobj);
  94         kfree(p);
  95 }
  96
  97 static struct kobj_type brport_ktype = {
  98 #ifdef CONFIG_SYSFS
  99         .sysfs_ops = &brport_sysfs_ops,
 100 #endif
 101         .release = release_nbp,
 102 };
 103
 104 static void destroy_nbp(struct net_bridge_port *p)
 105 {
 106         struct net_device *dev = p->dev;
 107
 108         p->br = NULL;
 109         p->dev = NULL;
 110         dev_put(dev);
 111
 112         kobject_put(&p->kobj);
 113 }
 114
 115 static void destroy_nbp_rcu(struct rcu_head *head)
 116 {
 117         struct net_bridge_port *p =
 118                         container_of(head, struct net_bridge_port, rcu);
 119         destroy_nbp(p);
 120 }
 121
 122 /* Delete port(interface) from bridge is done in two steps.
 123  * via RCU. First step, marks device as down. That deletes
 124  * all the timers and stops new packets from flowing through.
 125  *
 126  * Final cleanup doesn't occur until after all CPU's finished
 127  * processing packets.
 128  *
 129  * Protected from multiple admin operations by RTNL mutex
 130  */
 131 static void del_nbp(struct net_bridge_port *p)
 132 {
 133         struct net_bridge *br = p->br;
 134         struct net_device *dev = p->dev;
 135
 136         sysfs_remove_link(br->ifobj, dev->name);
 137
 138         dev_set_promiscuity(dev, -1);
 139
 140         spin_lock_bh(&br->lock);
 141         br_stp_disable_port(p);
 142         spin_unlock_bh(&br->lock);
 143
 144         br_ifinfo_notify(RTM_DELLINK, p);
 145
 146         br_fdb_delete_by_port(br, p, 1);
 147
 148         list_del_rcu(&p->list);
 149
 150         rcu_assign_pointer(dev->br_port, NULL);
 151
 152         kobject_uevent(&p->kobj, KOBJ_REMOVE);
 153         kobject_del(&p->kobj);
 154
 155         call_rcu(&p->rcu, destroy_nbp_rcu);
 156 }
 157
 158 /* called with RTNL */
 159 static void del_br(struct net_bridge *br)
 160 {
 161         struct net_bridge_port *p, *n;
 162
 163         list_for_each_entry_safe(p, n, &br->port_list, list) {
 164                 del_nbp(p);
 165         }
 166
 167         del_timer_sync(&br->gc_timer);
 168
 169         br_sysfs_delbr(br->dev);
 170         unregister_netdevice(br->dev);
 171 }
 172
 173 static struct net_device *new_bridge_dev(const char *name)
 174 {
 175         struct net_bridge *br;
 176         struct net_device *dev;
 177
 178         dev = alloc_netdev(sizeof(struct net_bridge), name,
 179                            br_dev_setup);
 180
 181         if (!dev)
 182                 return NULL;
 183
 184         br = netdev_priv(dev);
 185         br->dev = dev;
 186
 187         spin_lock_init(&br->lock);
 188         INIT_LIST_HEAD(&br->port_list);
 189         spin_lock_init(&br->hash_lock);
 190
 191         br->bridge_id.prio[0] = 0x80;
 192         br->bridge_id.prio[1] = 0x00;
 193
 194         memcpy(br->group_addr, br_group_address, ETH_ALEN);
 195
 196         br->feature_mask = dev->features;
 197         br->stp_enabled = BR_NO_STP;
 198         br->designated_root = br->bridge_id;
 199         br->root_path_cost = 0;
 200         br->root_port = 0;
 201         br->bridge_max_age = br->max_age = 20 * HZ;
 202         br->bridge_hello_time = br->hello_time = 2 * HZ;
 203         br->bridge_forward_delay = br->forward_delay = 15 * HZ;
 204         br->topology_change = 0;
 205         br->topology_change_detected = 0;
 206         br->ageing_time = 300 * HZ;
 207         INIT_LIST_HEAD(&br->age_list);
 208
 209         br_stp_timer_init(br);
 210
 211         return dev;
 212 }
 213
 214 /* find an available port number */
 215 static int find_portno(struct net_bridge *br)
 216 {
 217         int index;
 218         struct net_bridge_port *p;
 219         unsigned long *inuse;
 220
 221         inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
 222                         GFP_KERNEL);
 223         if (!inuse)
 224                 return -ENOMEM;
 225
 226         set_bit(0, inuse);      /* zero is reserved */
 227         list_for_each_entry(p, &br->port_list, list) {
 228                 set_bit(p->port_no, inuse);
 229         }
 230         index = find_first_zero_bit(inuse, BR_MAX_PORTS);
 231         kfree(inuse);
 232
 233         return (index >= BR_MAX_PORTS) ? -EXFULL : index;
 234 }
 235
 236 /* called with RTNL but without bridge lock */
 237 static struct net_bridge_port *new_nbp(struct net_bridge *br,
 238                                        struct net_device *dev)
 239 {
 240         int index;
 241         struct net_bridge_port *p;
 242
 243         index = find_portno(br);
 244         if (index < 0)
 245                 return ERR_PTR(index);
 246
 247         p = kzalloc(sizeof(*p), GFP_KERNEL);
 248         if (p == NULL)
 249                 return ERR_PTR(-ENOMEM);
 250
 251         p->br = br;
 252         dev_hold(dev);
 253         p->dev = dev;
 254         p->path_cost = port_cost(dev);
 255         p->priority = 0x8000 >> BR_PORT_BITS;
 256         p->port_no = index;
 257         br_init_port(p);
 258         p->state = BR_STATE_DISABLED;
 259         br_stp_port_timer_init(p);
 260
 261         return p;
 262 }
 263
 264 int br_add_bridge(const char *name)
 265 {
 266         struct net_device *dev;
 267         int ret;
 268
 269         dev = new_bridge_dev(name);
 270         if (!dev)
 271                 return -ENOMEM;
 272
 273         rtnl_lock();
 274         if (strchr(dev->name, '%')) {
 275                 ret = dev_alloc_name(dev, dev->name);
 276                 if (ret < 0)
 277                         goto out_free;
 278         }
 279
 280         ret = register_netdevice(dev);
 281         if (ret)
 282                 goto out_free;
 283
 284         ret = br_sysfs_addbr(dev);
 285         if (ret)
 286                 unregister_netdevice(dev);
 287  out:
 288         rtnl_unlock();
 289         return ret;
 290
 291 out_free:
 292         free_netdev(dev);
 293         goto out;
 294 }
 295
 296 int br_del_bridge(const char *name)
 297 {
 298         struct net_device *dev;
 299         int ret = 0;
 300
 301         rtnl_lock();
 302         dev = __dev_get_by_name(&init_net, name);
 303         if (dev == NULL)
 304                 ret =  -ENXIO;  /* Could not find device */
 305
 306         else if (!(dev->priv_flags & IFF_EBRIDGE)) {
 307                 /* Attempt to delete non bridge device! */
 308                 ret = -EPERM;
 309         }
 310
 311         else if (dev->flags & IFF_UP) {
 312                 /* Not shutdown yet. */
 313                 ret = -EBUSY;
 314         }
 315
 316         else
 317                 del_br(netdev_priv(dev));
 318
 319         rtnl_unlock();
 320         return ret;
 321 }
 322
 323 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
 324 int br_min_mtu(const struct net_bridge *br)
 325 {
 326         const struct net_bridge_port *p;
 327         int mtu = 0;
 328
 329         ASSERT_RTNL();
 330
 331         if (list_empty(&br->port_list))
 332                 mtu = ETH_DATA_LEN;
 333         else {
 334                 list_for_each_entry(p, &br->port_list, list) {
 335                         if (!mtu  || p->dev->mtu < mtu)
 336                                 mtu = p->dev->mtu;
 337                 }
 338         }
 339         return mtu;
 340 }
 341
 342 /*
 343  * Recomputes features using slave's features
 344  */
 345 void br_features_recompute(struct net_bridge *br)
 346 {
 347         struct net_bridge_port *p;
 348         unsigned long features;
 349
 350         features = br->feature_mask;
 351
 352         list_for_each_entry(p, &br->port_list, list) {
 353                 features = netdev_compute_features(features, p->dev->features);
 354         }
 355
 356         br->dev->features = features;
 357 }
 358
 359 /* called with RTNL */
 360 int br_add_if(struct net_bridge *br, struct net_device *dev)
 361 {
 362         struct net_bridge_port *p;
 363         int err = 0;
 364
 365         if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER)
 366                 return -EINVAL;
 367
 368         if (dev->hard_start_xmit == br_dev_xmit)
 369                 return -ELOOP;
 370
 371         if (dev->br_port != NULL)
 372                 return -EBUSY;
 373
 374         p = new_nbp(br, dev);
 375         if (IS_ERR(p))
 376                 return PTR_ERR(p);
 377
 378         err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
 379                                    SYSFS_BRIDGE_PORT_ATTR);
 380         if (err)
 381                 goto err0;
 382
 383         err = br_fdb_insert(br, p, dev->dev_addr);
 384         if (err)
 385                 goto err1;
 386
 387         err = br_sysfs_addif(p);
 388         if (err)
 389                 goto err2;
 390
 391         rcu_assign_pointer(dev->br_port, p);
 392         dev_set_promiscuity(dev, 1);
 393
 394         list_add_rcu(&p->list, &br->port_list);
 395
 396         spin_lock_bh(&br->lock);
 397         br_stp_recalculate_bridge_id(br);
 398         br_features_recompute(br);
 399
 400         if ((dev->flags & IFF_UP) && netif_carrier_ok(dev) &&
 401             (br->dev->flags & IFF_UP))
 402                 br_stp_enable_port(p);
 403         spin_unlock_bh(&br->lock);
 404
 405         br_ifinfo_notify(RTM_NEWLINK, p);
 406
 407         dev_set_mtu(br->dev, br_min_mtu(br));
 408
 409         kobject_uevent(&p->kobj, KOBJ_ADD);
 410
 411         return 0;
 412 err2:
 413         br_fdb_delete_by_port(br, p, 1);
 414 err1:
 415         kobject_del(&p->kobj);
 416         goto put_back;
 417 err0:
 418         kobject_put(&p->kobj);
 419
 420 put_back:
 421         dev_put(dev);
 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 = dev->br_port;
 429
 430         if (!p || p->br != br)
 431                 return -EINVAL;
 432
 433         del_nbp(p);
 434
 435         spin_lock_bh(&br->lock);
 436         br_stp_recalculate_bridge_id(br);
 437         br_features_recompute(br);
 438         spin_unlock_bh(&br->lock);
 439
 440         return 0;
 441 }
 442
 443 void __exit br_cleanup_bridges(void)
 444 {
 445         struct net_device *dev, *nxt;
 446
 447         rtnl_lock();
 448         for_each_netdev_safe(&init_net, dev, nxt)
 449                 if (dev->priv_flags & IFF_EBRIDGE)
 450                         del_br(dev->priv);
 451         rtnl_unlock();
 452
 453 }