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 void br_port_carrier_check(struct net_bridge_port *p)
  81 {
  82         struct net_device *dev = p->dev;
  83         struct net_bridge *br = p->br;
  84
  85         if (netif_carrier_ok(dev))
  86                 p->path_cost = port_cost(dev);
  87
  88         if (netif_running(br->dev)) {
  89                 spin_lock_bh(&br->lock);
  90                 if (netif_carrier_ok(dev)) {
  91                         if (p->state == BR_STATE_DISABLED)
  92                                 br_stp_enable_port(p);
  93                 } else {
  94                         if (p->state != BR_STATE_DISABLED)
  95                                 br_stp_disable_port(p);
  96                 }
  97                 spin_unlock_bh(&br->lock);
  98         }
  99 }
 100
 101 static void release_nbp(struct kobject *kobj)
 102 {
 103         struct net_bridge_port *p
 104                 = container_of(kobj, struct net_bridge_port, kobj);
 105         kfree(p);
 106 }
 107
 108 static struct kobj_type brport_ktype = {
 109 #ifdef CONFIG_SYSFS
 110         .sysfs_ops = &brport_sysfs_ops,
 111 #endif
 112         .release = release_nbp,
 113 };
 114
 115 static void destroy_nbp(struct net_bridge_port *p)
 116 {
 117         struct net_device *dev = p->dev;
 118
 119         p->br = NULL;
 120         p->dev = NULL;
 121         dev_put(dev);
 122
 123         kobject_put(&p->kobj);
 124 }
 125
 126 static void destroy_nbp_rcu(struct rcu_head *head)
 127 {
 128         struct net_bridge_port *p =
 129                         container_of(head, struct net_bridge_port, rcu);
 130         destroy_nbp(p);
 131 }
 132
 133 /* Delete port(interface) from bridge is done in two steps.
 134  * via RCU. First step, marks device as down. That deletes
 135  * all the timers and stops new packets from flowing through.
 136  *
 137  * Final cleanup doesn't occur until after all CPU's finished
 138  * processing packets.
 139  *
 140  * Protected from multiple admin operations by RTNL mutex
 141  */
 142 static void del_nbp(struct net_bridge_port *p)
 143 {
 144         struct net_bridge *br = p->br;
 145         struct net_device *dev = p->dev;
 146
 147         sysfs_remove_link(&br->ifobj, dev->name);
 148
 149         dev_set_promiscuity(dev, -1);
 150
 151         spin_lock_bh(&br->lock);
 152         br_stp_disable_port(p);
 153         spin_unlock_bh(&br->lock);
 154
 155         br_ifinfo_notify(RTM_DELLINK, p);
 156
 157         br_fdb_delete_by_port(br, p, 1);
 158
 159         list_del_rcu(&p->list);
 160
 161         rcu_assign_pointer(dev->br_port, NULL);
 162
 163         kobject_uevent(&p->kobj, KOBJ_REMOVE);
 164         kobject_del(&p->kobj);
 165
 166         call_rcu(&p->rcu, destroy_nbp_rcu);
 167 }
 168
 169 /* called with RTNL */
 170 static void del_br(struct net_bridge *br)
 171 {
 172         struct net_bridge_port *p, *n;
 173
 174         list_for_each_entry_safe(p, n, &br->port_list, list) {
 175                 del_nbp(p);
 176         }
 177
 178         del_timer_sync(&br->gc_timer);
 179
 180         br_sysfs_delbr(br->dev);
 181         unregister_netdevice(br->dev);
 182 }
 183
 184 static struct net_device *new_bridge_dev(const char *name)
 185 {
 186         struct net_bridge *br;
 187         struct net_device *dev;
 188
 189         dev = alloc_netdev(sizeof(struct net_bridge), name,
 190                            br_dev_setup);
 191
 192         if (!dev)
 193                 return NULL;
 194
 195         br = netdev_priv(dev);
 196         br->dev = dev;
 197
 198         spin_lock_init(&br->lock);
 199         INIT_LIST_HEAD(&br->port_list);
 200         spin_lock_init(&br->hash_lock);
 201
 202         br->bridge_id.prio[0] = 0x80;
 203         br->bridge_id.prio[1] = 0x00;
 204
 205         memcpy(br->group_addr, br_group_address, ETH_ALEN);
 206
 207         br->feature_mask = dev->features;
 208         br->stp_enabled = BR_NO_STP;
 209         br->designated_root = br->bridge_id;
 210         br->root_path_cost = 0;
 211         br->root_port = 0;
 212         br->bridge_max_age = br->max_age = 20 * HZ;
 213         br->bridge_hello_time = br->hello_time = 2 * HZ;
 214         br->bridge_forward_delay = br->forward_delay = 15 * HZ;
 215         br->topology_change = 0;
 216         br->topology_change_detected = 0;
 217         br->ageing_time = 300 * HZ;
 218         INIT_LIST_HEAD(&br->age_list);
 219
 220         br_stp_timer_init(br);
 221
 222         return dev;
 223 }
 224
 225 /* find an available port number */
 226 static int find_portno(struct net_bridge *br)
 227 {
 228         int index;
 229         struct net_bridge_port *p;
 230         unsigned long *inuse;
 231
 232         inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
 233                         GFP_KERNEL);
 234         if (!inuse)
 235                 return -ENOMEM;
 236
 237         set_bit(0, inuse);      /* zero is reserved */
 238         list_for_each_entry(p, &br->port_list, list) {
 239                 set_bit(p->port_no, inuse);
 240         }
 241         index = find_first_zero_bit(inuse, BR_MAX_PORTS);
 242         kfree(inuse);
 243
 244         return (index >= BR_MAX_PORTS) ? -EXFULL : index;
 245 }
 246
 247 /* called with RTNL but without bridge lock */
 248 static struct net_bridge_port *new_nbp(struct net_bridge *br,
 249                                        struct net_device *dev)
 250 {
 251         int index;
 252         struct net_bridge_port *p;
 253
 254         index = find_portno(br);
 255         if (index < 0)
 256                 return ERR_PTR(index);
 257
 258         p = kzalloc(sizeof(*p), GFP_KERNEL);
 259         if (p == NULL)
 260                 return ERR_PTR(-ENOMEM);
 261
 262         p->br = br;
 263         dev_hold(dev);
 264         p->dev = dev;
 265         p->path_cost = port_cost(dev);
 266         p->priority = 0x8000 >> BR_PORT_BITS;
 267         p->port_no = index;
 268         br_init_port(p);
 269         p->state = BR_STATE_DISABLED;
 270         br_stp_port_timer_init(p);
 271
 272         kobject_init(&p->kobj);
 273         kobject_set_name(&p->kobj, SYSFS_BRIDGE_PORT_ATTR);
 274         p->kobj.ktype = &brport_ktype;
 275         p->kobj.parent = &(dev->dev.kobj);
 276         p->kobj.kset = NULL;
 277
 278         return p;
 279 }
 280
 281 int br_add_bridge(const char *name)
 282 {
 283         struct net_device *dev;
 284         int ret;
 285
 286         dev = new_bridge_dev(name);
 287         if (!dev)
 288                 return -ENOMEM;
 289
 290         rtnl_lock();
 291         if (strchr(dev->name, '%')) {
 292                 ret = dev_alloc_name(dev, dev->name);
 293                 if (ret < 0) {
 294                         free_netdev(dev);
 295                         goto out;
 296                 }
 297         }
 298
 299         ret = register_netdevice(dev);
 300         if (ret)
 301                 goto out;
 302
 303         ret = br_sysfs_addbr(dev);
 304         if (ret)
 305                 unregister_netdevice(dev);
 306  out:
 307         rtnl_unlock();
 308         return ret;
 309 }
 310
 311 int br_del_bridge(const char *name)
 312 {
 313         struct net_device *dev;
 314         int ret = 0;
 315
 316         rtnl_lock();
 317         dev = __dev_get_by_name(name);
 318         if (dev == NULL)
 319                 ret =  -ENXIO;  /* Could not find device */
 320
 321         else if (!(dev->priv_flags & IFF_EBRIDGE)) {
 322                 /* Attempt to delete non bridge device! */
 323                 ret = -EPERM;
 324         }
 325
 326         else if (dev->flags & IFF_UP) {
 327                 /* Not shutdown yet. */
 328                 ret = -EBUSY;
 329         }
 330
 331         else
 332                 del_br(netdev_priv(dev));
 333
 334         rtnl_unlock();
 335         return ret;
 336 }
 337
 338 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
 339 int br_min_mtu(const struct net_bridge *br)
 340 {
 341         const struct net_bridge_port *p;
 342         int mtu = 0;
 343
 344         ASSERT_RTNL();
 345
 346         if (list_empty(&br->port_list))
 347                 mtu = ETH_DATA_LEN;
 348         else {
 349                 list_for_each_entry(p, &br->port_list, list) {
 350                         if (!mtu  || p->dev->mtu < mtu)
 351                                 mtu = p->dev->mtu;
 352                 }
 353         }
 354         return mtu;
 355 }
 356
 357 /*
 358  * Recomputes features using slave's features
 359  */
 360 void br_features_recompute(struct net_bridge *br)
 361 {
 362         struct net_bridge_port *p;
 363         unsigned long features, checksum;
 364
 365         checksum = br->feature_mask & NETIF_F_ALL_CSUM ? NETIF_F_NO_CSUM : 0;
 366         features = br->feature_mask & ~NETIF_F_ALL_CSUM;
 367
 368         list_for_each_entry(p, &br->port_list, list) {
 369                 unsigned long feature = p->dev->features;
 370
 371                 /* if device needs checksumming, downgrade to hw checksumming */
 372                 if (checksum & NETIF_F_NO_CSUM && !(feature & NETIF_F_NO_CSUM))
 373                         checksum ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM;
 374
 375                 /* if device can't do all checksum, downgrade to ipv4/ipv6 */
 376                 if (checksum & NETIF_F_HW_CSUM && !(feature & NETIF_F_HW_CSUM))
 377                         checksum ^= NETIF_F_HW_CSUM
 378                                 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
 379
 380                 if (checksum & NETIF_F_IPV6_CSUM && !(feature & NETIF_F_IPV6_CSUM))
 381                         checksum &= ~NETIF_F_IPV6_CSUM;
 382
 383                 if (!(feature & NETIF_F_IP_CSUM))
 384                         checksum = 0;
 385
 386                 if (feature & NETIF_F_GSO)
 387                         feature |= NETIF_F_GSO_SOFTWARE;
 388                 feature |= NETIF_F_GSO;
 389
 390                 features &= feature;
 391         }
 392
 393         if (!(checksum & NETIF_F_ALL_CSUM))
 394                 features &= ~NETIF_F_SG;
 395         if (!(features & NETIF_F_SG))
 396                 features &= ~NETIF_F_GSO_MASK;
 397
 398         br->dev->features = features | checksum | NETIF_F_LLTX |
 399                             NETIF_F_GSO_ROBUST;
 400 }
 401
 402 /* called with RTNL */
 403 int br_add_if(struct net_bridge *br, struct net_device *dev)
 404 {
 405         struct net_bridge_port *p;
 406         int err = 0;
 407
 408         if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER)
 409                 return -EINVAL;
 410
 411         if (dev->hard_start_xmit == br_dev_xmit)
 412                 return -ELOOP;
 413
 414         if (dev->br_port != NULL)
 415                 return -EBUSY;
 416
 417         p = new_nbp(br, dev);
 418         if (IS_ERR(p))
 419                 return PTR_ERR(p);
 420
 421         err = kobject_add(&p->kobj);
 422         if (err)
 423                 goto err0;
 424
 425         err = br_fdb_insert(br, p, dev->dev_addr);
 426         if (err)
 427                 goto err1;
 428
 429         err = br_sysfs_addif(p);
 430         if (err)
 431                 goto err2;
 432
 433         rcu_assign_pointer(dev->br_port, p);
 434         dev_set_promiscuity(dev, 1);
 435
 436         list_add_rcu(&p->list, &br->port_list);
 437
 438         spin_lock_bh(&br->lock);
 439         br_stp_recalculate_bridge_id(br);
 440         br_features_recompute(br);
 441
 442         if ((dev->flags & IFF_UP) && netif_carrier_ok(dev) &&
 443             (br->dev->flags & IFF_UP))
 444                 br_stp_enable_port(p);
 445         spin_unlock_bh(&br->lock);
 446
 447         br_ifinfo_notify(RTM_NEWLINK, p);
 448
 449         dev_set_mtu(br->dev, br_min_mtu(br));
 450
 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_each_netdev_safe(dev, nxt)
 487                 if (dev->priv_flags & IFF_EBRIDGE)
 488                         del_br(dev->priv);
 489         rtnl_unlock();
 490
 491 }