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->get_settings) {
  37                 struct ethtool_cmd ecmd = { ETHTOOL_GSET };
  38                 int err = dev->ethtool_ops->get_settings(dev, &ecmd);
  39                 if (!err) {
  40                         switch(ecmd.speed) {
  41                         case SPEED_100:
  42                                 return 19;
  43                         case SPEED_1000:
  44                                 return 4;
  45                         case SPEED_10000:
  46                                 return 2;
  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         kobject_init(&p->kobj);
 262         kobject_set_name(&p->kobj, SYSFS_BRIDGE_PORT_ATTR);
 263         p->kobj.ktype = &brport_ktype;
 264         p->kobj.parent = &(dev->dev.kobj);
 265         p->kobj.kset = NULL;
 266
 267         return p;
 268 }
 269
 270 int br_add_bridge(const char *name)
 271 {
 272         struct net_device *dev;
 273         int ret;
 274
 275         dev = new_bridge_dev(name);
 276         if (!dev)
 277                 return -ENOMEM;
 278
 279         rtnl_lock();
 280         if (strchr(dev->name, '%')) {
 281                 ret = dev_alloc_name(dev, dev->name);
 282                 if (ret < 0) {
 283                         free_netdev(dev);
 284                         goto out;
 285                 }
 286         }
 287
 288         ret = register_netdevice(dev);
 289         if (ret)
 290                 goto out;
 291
 292         ret = br_sysfs_addbr(dev);
 293         if (ret)
 294                 unregister_netdevice(dev);
 295  out:
 296         rtnl_unlock();
 297         return ret;
 298 }
 299
 300 int br_del_bridge(const char *name)
 301 {
 302         struct net_device *dev;
 303         int ret = 0;
 304
 305         rtnl_lock();
 306         dev = __dev_get_by_name(name);
 307         if (dev == NULL)
 308                 ret =  -ENXIO;  /* Could not find device */
 309
 310         else if (!(dev->priv_flags & IFF_EBRIDGE)) {
 311                 /* Attempt to delete non bridge device! */
 312                 ret = -EPERM;
 313         }
 314
 315         else if (dev->flags & IFF_UP) {
 316                 /* Not shutdown yet. */
 317                 ret = -EBUSY;
 318         }
 319
 320         else
 321                 del_br(netdev_priv(dev));
 322
 323         rtnl_unlock();
 324         return ret;
 325 }
 326
 327 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
 328 int br_min_mtu(const struct net_bridge *br)
 329 {
 330         const struct net_bridge_port *p;
 331         int mtu = 0;
 332
 333         ASSERT_RTNL();
 334
 335         if (list_empty(&br->port_list))
 336                 mtu = ETH_DATA_LEN;
 337         else {
 338                 list_for_each_entry(p, &br->port_list, list) {
 339                         if (!mtu  || p->dev->mtu < mtu)
 340                                 mtu = p->dev->mtu;
 341                 }
 342         }
 343         return mtu;
 344 }
 345
 346 /*
 347  * Recomputes features using slave's features
 348  */
 349 void br_features_recompute(struct net_bridge *br)
 350 {
 351         struct net_bridge_port *p;
 352         unsigned long features, checksum;
 353
 354         checksum = br->feature_mask & NETIF_F_ALL_CSUM ? NETIF_F_NO_CSUM : 0;
 355         features = br->feature_mask & ~NETIF_F_ALL_CSUM;
 356
 357         list_for_each_entry(p, &br->port_list, list) {
 358                 unsigned long feature = p->dev->features;
 359
 360                 /* if device needs checksumming, downgrade to hw checksumming */
 361                 if (checksum & NETIF_F_NO_CSUM && !(feature & NETIF_F_NO_CSUM))
 362                         checksum ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM;
 363
 364                 /* if device can't do all checksum, downgrade to ipv4/ipv6 */
 365                 if (checksum & NETIF_F_HW_CSUM && !(feature & NETIF_F_HW_CSUM))
 366                         checksum ^= NETIF_F_HW_CSUM
 367                                 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
 368
 369                 if (checksum & NETIF_F_IPV6_CSUM && !(feature & NETIF_F_IPV6_CSUM))
 370                         checksum &= ~NETIF_F_IPV6_CSUM;
 371
 372                 if (!(feature & NETIF_F_IP_CSUM))
 373                         checksum = 0;
 374
 375                 if (feature & NETIF_F_GSO)
 376                         feature |= NETIF_F_GSO_SOFTWARE;
 377                 feature |= NETIF_F_GSO;
 378
 379                 features &= feature;
 380         }
 381
 382         if (!(checksum & NETIF_F_ALL_CSUM))
 383                 features &= ~NETIF_F_SG;
 384         if (!(features & NETIF_F_SG))
 385                 features &= ~NETIF_F_GSO_MASK;
 386
 387         br->dev->features = features | checksum | NETIF_F_LLTX |
 388                             NETIF_F_GSO_ROBUST;
 389 }
 390
 391 /* called with RTNL */
 392 int br_add_if(struct net_bridge *br, struct net_device *dev)
 393 {
 394         struct net_bridge_port *p;
 395         int err = 0;
 396
 397         if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER)
 398                 return -EINVAL;
 399
 400         if (dev->hard_start_xmit == br_dev_xmit)
 401                 return -ELOOP;
 402
 403         if (dev->br_port != NULL)
 404                 return -EBUSY;
 405
 406         p = new_nbp(br, dev);
 407         if (IS_ERR(p))
 408                 return PTR_ERR(p);
 409
 410         err = kobject_add(&p->kobj);
 411         if (err)
 412                 goto err0;
 413
 414         err = br_fdb_insert(br, p, dev->dev_addr);
 415         if (err)
 416                 goto err1;
 417
 418         err = br_sysfs_addif(p);
 419         if (err)
 420                 goto err2;
 421
 422         rcu_assign_pointer(dev->br_port, p);
 423         dev_set_promiscuity(dev, 1);
 424
 425         list_add_rcu(&p->list, &br->port_list);
 426
 427         spin_lock_bh(&br->lock);
 428         br_stp_recalculate_bridge_id(br);
 429         br_features_recompute(br);
 430
 431         if ((dev->flags & IFF_UP) && netif_carrier_ok(dev) &&
 432             (br->dev->flags & IFF_UP))
 433                 br_stp_enable_port(p);
 434         spin_unlock_bh(&br->lock);
 435
 436         br_ifinfo_notify(RTM_NEWLINK, p);
 437
 438         dev_set_mtu(br->dev, br_min_mtu(br));
 439
 440         kobject_uevent(&p->kobj, KOBJ_ADD);
 441
 442         return 0;
 443 err2:
 444         br_fdb_delete_by_port(br, p, 1);
 445 err1:
 446         kobject_del(&p->kobj);
 447 err0:
 448         kobject_put(&p->kobj);
 449         return err;
 450 }
 451
 452 /* called with RTNL */
 453 int br_del_if(struct net_bridge *br, struct net_device *dev)
 454 {
 455         struct net_bridge_port *p = dev->br_port;
 456
 457         if (!p || p->br != br)
 458                 return -EINVAL;
 459
 460         del_nbp(p);
 461
 462         spin_lock_bh(&br->lock);
 463         br_stp_recalculate_bridge_id(br);
 464         br_features_recompute(br);
 465         spin_unlock_bh(&br->lock);
 466
 467         return 0;
 468 }
 469
 470 void __exit br_cleanup_bridges(void)
 471 {
 472         struct net_device *dev, *nxt;
 473
 474         rtnl_lock();
 475         for_each_netdev_safe(dev, nxt)
 476                 if (dev->priv_flags & IFF_EBRIDGE)
 477                         del_br(dev->priv);
 478         rtnl_unlock();
 479
 480 }