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