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  * Need to simulate user ioctl because not all device's that support
  33  * ethtool, use ethtool_ops.  Also, since driver might sleep need to
  34  * not be holding any locks.
  35  */
  36 static int port_cost(struct net_device *dev)
  37 {
  38         struct ethtool_cmd ecmd = { ETHTOOL_GSET };
  39         struct ifreq ifr;
  40         mm_segment_t old_fs;
  41         int err;
  42
  43         strncpy(ifr.ifr_name, dev->name, IFNAMSIZ);
  44         ifr.ifr_data = (void __user *) &ecmd;
  45
  46         old_fs = get_fs();
  47         set_fs(KERNEL_DS);
  48         err = dev_ethtool(&ifr);
  49         set_fs(old_fs);
  50
  51         if (!err) {
  52                 switch(ecmd.speed) {
  53                 case SPEED_100:
  54                         return 19;
  55                 case SPEED_1000:
  56                         return 4;
  57                 case SPEED_10000:
  58                         return 2;
  59                 case SPEED_10:
  60                         return 100;
  61                 }
  62         }
  63
  64         /* Old silly heuristics based on name */
  65         if (!strncmp(dev->name, "lec", 3))
  66                 return 7;
  67
  68         if (!strncmp(dev->name, "plip", 4))
  69                 return 2500;
  70
  71         return 100;     /* assume old 10Mbps */
  72 }
  73
  74
  75 /*
  76  * Check for port carrier transistions.
  77  * Called from work queue to allow for calling functions that
  78  * might sleep (such as speed check), and to debounce.
  79  */
  80 static void port_carrier_check(struct work_struct *work)
  81 {
  82         struct net_bridge_port *p;
  83         struct net_device *dev;
  84         struct net_bridge *br;
  85
  86         dev = container_of(work, struct net_bridge_port,
  87                            carrier_check.work)->dev;
  88         work_release(work);
  89
  90         rtnl_lock();
  91         p = dev->br_port;
  92         if (!p)
  93                 goto done;
  94         br = p->br;
  95
  96         if (netif_carrier_ok(dev))
  97                 p->path_cost = port_cost(dev);
  98
  99         if (br->dev->flags & IFF_UP) {
 100                 spin_lock_bh(&br->lock);
 101                 if (netif_carrier_ok(dev)) {
 102                         if (p->state == BR_STATE_DISABLED)
 103                                 br_stp_enable_port(p);
 104                 } else {
 105                         if (p->state != BR_STATE_DISABLED)
 106                                 br_stp_disable_port(p);
 107                 }
 108                 spin_unlock_bh(&br->lock);
 109         }
 110 done:
 111         dev_put(dev);
 112         rtnl_unlock();
 113 }
 114
 115 static void release_nbp(struct kobject *kobj)
 116 {
 117         struct net_bridge_port *p
 118                 = container_of(kobj, struct net_bridge_port, kobj);
 119         kfree(p);
 120 }
 121
 122 static struct kobj_type brport_ktype = {
 123 #ifdef CONFIG_SYSFS
 124         .sysfs_ops = &brport_sysfs_ops,
 125 #endif
 126         .release = release_nbp,
 127 };
 128
 129 static void destroy_nbp(struct net_bridge_port *p)
 130 {
 131         struct net_device *dev = p->dev;
 132
 133         p->br = NULL;
 134         p->dev = NULL;
 135         dev_put(dev);
 136
 137         kobject_put(&p->kobj);
 138 }
 139
 140 static void destroy_nbp_rcu(struct rcu_head *head)
 141 {
 142         struct net_bridge_port *p =
 143                         container_of(head, struct net_bridge_port, rcu);
 144         destroy_nbp(p);
 145 }
 146
 147 /* Delete port(interface) from bridge is done in two steps.
 148  * via RCU. First step, marks device as down. That deletes
 149  * all the timers and stops new packets from flowing through.
 150  *
 151  * Final cleanup doesn't occur until after all CPU's finished
 152  * processing packets.
 153  *
 154  * Protected from multiple admin operations by RTNL mutex
 155  */
 156 static void del_nbp(struct net_bridge_port *p)
 157 {
 158         struct net_bridge *br = p->br;
 159         struct net_device *dev = p->dev;
 160
 161         sysfs_remove_link(&br->ifobj, dev->name);
 162
 163         dev_set_promiscuity(dev, -1);
 164
 165         if (cancel_delayed_work(&p->carrier_check))
 166                 dev_put(dev);
 167
 168         spin_lock_bh(&br->lock);
 169         br_stp_disable_port(p);
 170         spin_unlock_bh(&br->lock);
 171
 172         br_fdb_delete_by_port(br, p, 1);
 173
 174         list_del_rcu(&p->list);
 175
 176         rcu_assign_pointer(dev->br_port, NULL);
 177
 178         kobject_uevent(&p->kobj, KOBJ_REMOVE);
 179         kobject_del(&p->kobj);
 180
 181         call_rcu(&p->rcu, destroy_nbp_rcu);
 182 }
 183
 184 /* called with RTNL */
 185 static void del_br(struct net_bridge *br)
 186 {
 187         struct net_bridge_port *p, *n;
 188
 189         list_for_each_entry_safe(p, n, &br->port_list, list) {
 190                 del_nbp(p);
 191         }
 192
 193         del_timer_sync(&br->gc_timer);
 194
 195         br_sysfs_delbr(br->dev);
 196         unregister_netdevice(br->dev);
 197 }
 198
 199 static struct net_device *new_bridge_dev(const char *name)
 200 {
 201         struct net_bridge *br;
 202         struct net_device *dev;
 203
 204         dev = alloc_netdev(sizeof(struct net_bridge), name,
 205                            br_dev_setup);
 206
 207         if (!dev)
 208                 return NULL;
 209
 210         br = netdev_priv(dev);
 211         br->dev = dev;
 212
 213         spin_lock_init(&br->lock);
 214         INIT_LIST_HEAD(&br->port_list);
 215         spin_lock_init(&br->hash_lock);
 216
 217         br->bridge_id.prio[0] = 0x80;
 218         br->bridge_id.prio[1] = 0x00;
 219
 220         memcpy(br->group_addr, br_group_address, ETH_ALEN);
 221
 222         br->feature_mask = dev->features;
 223         br->stp_enabled = 0;
 224         br->designated_root = br->bridge_id;
 225         br->root_path_cost = 0;
 226         br->root_port = 0;
 227         br->bridge_max_age = br->max_age = 20 * HZ;
 228         br->bridge_hello_time = br->hello_time = 2 * HZ;
 229         br->bridge_forward_delay = br->forward_delay = 15 * HZ;
 230         br->topology_change = 0;
 231         br->topology_change_detected = 0;
 232         br->ageing_time = 300 * HZ;
 233         INIT_LIST_HEAD(&br->age_list);
 234
 235         br_stp_timer_init(br);
 236
 237         return dev;
 238 }
 239
 240 /* find an available port number */
 241 static int find_portno(struct net_bridge *br)
 242 {
 243         int index;
 244         struct net_bridge_port *p;
 245         unsigned long *inuse;
 246
 247         inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
 248                         GFP_KERNEL);
 249         if (!inuse)
 250                 return -ENOMEM;
 251
 252         set_bit(0, inuse);      /* zero is reserved */
 253         list_for_each_entry(p, &br->port_list, list) {
 254                 set_bit(p->port_no, inuse);
 255         }
 256         index = find_first_zero_bit(inuse, BR_MAX_PORTS);
 257         kfree(inuse);
 258
 259         return (index >= BR_MAX_PORTS) ? -EXFULL : index;
 260 }
 261
 262 /* called with RTNL but without bridge lock */
 263 static struct net_bridge_port *new_nbp(struct net_bridge *br,
 264                                        struct net_device *dev)
 265 {
 266         int index;
 267         struct net_bridge_port *p;
 268
 269         index = find_portno(br);
 270         if (index < 0)
 271                 return ERR_PTR(index);
 272
 273         p = kzalloc(sizeof(*p), GFP_KERNEL);
 274         if (p == NULL)
 275                 return ERR_PTR(-ENOMEM);
 276
 277         p->br = br;
 278         dev_hold(dev);
 279         p->dev = dev;
 280         p->path_cost = port_cost(dev);
 281         p->priority = 0x8000 >> BR_PORT_BITS;
 282         p->port_no = index;
 283         br_init_port(p);
 284         p->state = BR_STATE_DISABLED;
 285         INIT_DELAYED_WORK_NAR(&p->carrier_check, port_carrier_check);
 286         br_stp_port_timer_init(p);
 287
 288         kobject_init(&p->kobj);
 289         kobject_set_name(&p->kobj, SYSFS_BRIDGE_PORT_ATTR);
 290         p->kobj.ktype = &brport_ktype;
 291         p->kobj.parent = &(dev->dev.kobj);
 292         p->kobj.kset = NULL;
 293
 294         return p;
 295 }
 296
 297 int br_add_bridge(const char *name)
 298 {
 299         struct net_device *dev;
 300         int ret;
 301
 302         dev = new_bridge_dev(name);
 303         if (!dev)
 304                 return -ENOMEM;
 305
 306         rtnl_lock();
 307         if (strchr(dev->name, '%')) {
 308                 ret = dev_alloc_name(dev, dev->name);
 309                 if (ret < 0) {
 310                         free_netdev(dev);
 311                         goto out;
 312                 }
 313         }
 314
 315         ret = register_netdevice(dev);
 316         if (ret)
 317                 goto out;
 318
 319         ret = br_sysfs_addbr(dev);
 320         if (ret)
 321                 unregister_netdevice(dev);
 322  out:
 323         rtnl_unlock();
 324         return ret;
 325 }
 326
 327 int br_del_bridge(const char *name)
 328 {
 329         struct net_device *dev;
 330         int ret = 0;
 331
 332         rtnl_lock();
 333         dev = __dev_get_by_name(name);
 334         if (dev == NULL)
 335                 ret =  -ENXIO;  /* Could not find device */
 336
 337         else if (!(dev->priv_flags & IFF_EBRIDGE)) {
 338                 /* Attempt to delete non bridge device! */
 339                 ret = -EPERM;
 340         }
 341
 342         else if (dev->flags & IFF_UP) {
 343                 /* Not shutdown yet. */
 344                 ret = -EBUSY;
 345         }
 346
 347         else
 348                 del_br(netdev_priv(dev));
 349
 350         rtnl_unlock();
 351         return ret;
 352 }
 353
 354 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
 355 int br_min_mtu(const struct net_bridge *br)
 356 {
 357         const struct net_bridge_port *p;
 358         int mtu = 0;
 359
 360         ASSERT_RTNL();
 361
 362         if (list_empty(&br->port_list))
 363                 mtu = ETH_DATA_LEN;
 364         else {
 365                 list_for_each_entry(p, &br->port_list, list) {
 366                         if (!mtu  || p->dev->mtu < mtu)
 367                                 mtu = p->dev->mtu;
 368                 }
 369         }
 370         return mtu;
 371 }
 372
 373 /*
 374  * Recomputes features using slave's features
 375  */
 376 void br_features_recompute(struct net_bridge *br)
 377 {
 378         struct net_bridge_port *p;
 379         unsigned long features, checksum;
 380
 381         checksum = br->feature_mask & NETIF_F_ALL_CSUM ? NETIF_F_NO_CSUM : 0;
 382         features = br->feature_mask & ~NETIF_F_ALL_CSUM;
 383
 384         list_for_each_entry(p, &br->port_list, list) {
 385                 unsigned long feature = p->dev->features;
 386
 387                 if (checksum & NETIF_F_NO_CSUM && !(feature & NETIF_F_NO_CSUM))
 388                         checksum ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM;
 389                 if (checksum & NETIF_F_HW_CSUM && !(feature & NETIF_F_HW_CSUM))
 390                         checksum ^= NETIF_F_HW_CSUM | NETIF_F_IP_CSUM;
 391                 if (!(feature & NETIF_F_IP_CSUM))
 392                         checksum = 0;
 393
 394                 if (feature & NETIF_F_GSO)
 395                         feature |= NETIF_F_GSO_SOFTWARE;
 396                 feature |= NETIF_F_GSO;
 397
 398                 features &= feature;
 399         }
 400
 401         if (!(checksum & NETIF_F_ALL_CSUM))
 402                 features &= ~NETIF_F_SG;
 403         if (!(features & NETIF_F_SG))
 404                 features &= ~NETIF_F_GSO_MASK;
 405
 406         br->dev->features = features | checksum | NETIF_F_LLTX |
 407                             NETIF_F_GSO_ROBUST;
 408 }
 409
 410 /* called with RTNL */
 411 int br_add_if(struct net_bridge *br, struct net_device *dev)
 412 {
 413         struct net_bridge_port *p;
 414         int err = 0;
 415
 416         if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER)
 417                 return -EINVAL;
 418
 419         if (dev->hard_start_xmit == br_dev_xmit)
 420                 return -ELOOP;
 421
 422         if (dev->br_port != NULL)
 423                 return -EBUSY;
 424
 425         p = new_nbp(br, dev);
 426         if (IS_ERR(p))
 427                 return PTR_ERR(p);
 428
 429         err = kobject_add(&p->kobj);
 430         if (err)
 431                 goto err0;
 432
 433         err = br_fdb_insert(br, p, dev->dev_addr);
 434         if (err)
 435                 goto err1;
 436
 437         err = br_sysfs_addif(p);
 438         if (err)
 439                 goto err2;
 440
 441         rcu_assign_pointer(dev->br_port, p);
 442         dev_set_promiscuity(dev, 1);
 443
 444         list_add_rcu(&p->list, &br->port_list);
 445
 446         spin_lock_bh(&br->lock);
 447         br_stp_recalculate_bridge_id(br);
 448         br_features_recompute(br);
 449         if (schedule_delayed_work(&p->carrier_check, BR_PORT_DEBOUNCE))
 450                 dev_hold(dev);
 451
 452         spin_unlock_bh(&br->lock);
 453
 454         dev_set_mtu(br->dev, br_min_mtu(br));
 455         kobject_uevent(&p->kobj, KOBJ_ADD);
 456
 457         return 0;
 458 err2:
 459         br_fdb_delete_by_port(br, p, 1);
 460 err1:
 461         kobject_del(&p->kobj);
 462 err0:
 463         kobject_put(&p->kobj);
 464         return err;
 465 }
 466
 467 /* called with RTNL */
 468 int br_del_if(struct net_bridge *br, struct net_device *dev)
 469 {
 470         struct net_bridge_port *p = dev->br_port;
 471
 472         if (!p || p->br != br)
 473                 return -EINVAL;
 474
 475         del_nbp(p);
 476
 477         spin_lock_bh(&br->lock);
 478         br_stp_recalculate_bridge_id(br);
 479         br_features_recompute(br);
 480         spin_unlock_bh(&br->lock);
 481
 482         return 0;
 483 }
 484
 485 void __exit br_cleanup_bridges(void)
 486 {
 487         struct net_device *dev, *nxt;
 488
 489         rtnl_lock();
 490         for (dev = dev_base; dev; dev = nxt) {
 491                 nxt = dev->next;
 492                 if (dev->priv_flags & IFF_EBRIDGE)
 493                         del_br(dev->priv);
 494         }
 495         rtnl_unlock();
 496
 497 }