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(void *arg)
  81 {
  82         struct net_device *dev = arg;
  83         struct net_bridge_port *p;
  84
  85         rtnl_lock();
  86         p = dev->br_port;
  87         if (!p)
  88                 goto done;
  89
  90         if (netif_carrier_ok(p->dev)) {
  91                 u32 cost = port_cost(p->dev);
  92
  93                 spin_lock_bh(&p->br->lock);
  94                 if (p->state == BR_STATE_DISABLED) {
  95                         p->path_cost = cost;
  96                         br_stp_enable_port(p);
  97                 }
  98                 spin_unlock_bh(&p->br->lock);
  99         } else {
 100                 spin_lock_bh(&p->br->lock);
 101                 if (p->state != BR_STATE_DISABLED)
 102                         br_stp_disable_port(p);
 103                 spin_unlock_bh(&p->br->lock);
 104         }
 105 done:
 106         rtnl_unlock();
 107 }
 108
 109 static void release_nbp(struct kobject *kobj)
 110 {
 111         struct net_bridge_port *p
 112                 = container_of(kobj, struct net_bridge_port, kobj);
 113         kfree(p);
 114 }
 115
 116 static struct kobj_type brport_ktype = {
 117 #ifdef CONFIG_SYSFS
 118         .sysfs_ops = &brport_sysfs_ops,
 119 #endif
 120         .release = release_nbp,
 121 };
 122
 123 static void destroy_nbp(struct net_bridge_port *p)
 124 {
 125         struct net_device *dev = p->dev;
 126
 127         p->br = NULL;
 128         p->dev = NULL;
 129         dev_put(dev);
 130
 131         kobject_put(&p->kobj);
 132 }
 133
 134 static void destroy_nbp_rcu(struct rcu_head *head)
 135 {
 136         struct net_bridge_port *p =
 137                         container_of(head, struct net_bridge_port, rcu);
 138         destroy_nbp(p);
 139 }
 140
 141 /* Delete port(interface) from bridge is done in two steps.
 142  * via RCU. First step, marks device as down. That deletes
 143  * all the timers and stops new packets from flowing through.
 144  *
 145  * Final cleanup doesn't occur until after all CPU's finished
 146  * processing packets.
 147  *
 148  * Protected from multiple admin operations by RTNL mutex
 149  */
 150 static void del_nbp(struct net_bridge_port *p)
 151 {
 152         struct net_bridge *br = p->br;
 153         struct net_device *dev = p->dev;
 154
 155         sysfs_remove_link(&br->ifobj, dev->name);
 156
 157         dev_set_promiscuity(dev, -1);
 158
 159         cancel_delayed_work(&p->carrier_check);
 160
 161         spin_lock_bh(&br->lock);
 162         br_stp_disable_port(p);
 163         spin_unlock_bh(&br->lock);
 164
 165         br_fdb_delete_by_port(br, p);
 166
 167         list_del_rcu(&p->list);
 168
 169         rcu_assign_pointer(dev->br_port, NULL);
 170
 171         kobject_del(&p->kobj);
 172
 173         call_rcu(&p->rcu, destroy_nbp_rcu);
 174 }
 175
 176 /* called with RTNL */
 177 static void del_br(struct net_bridge *br)
 178 {
 179         struct net_bridge_port *p, *n;
 180
 181         list_for_each_entry_safe(p, n, &br->port_list, list) {
 182                 del_nbp(p);
 183         }
 184
 185         del_timer_sync(&br->gc_timer);
 186
 187         br_sysfs_delbr(br->dev);
 188         unregister_netdevice(br->dev);
 189 }
 190
 191 static struct net_device *new_bridge_dev(const char *name)
 192 {
 193         struct net_bridge *br;
 194         struct net_device *dev;
 195
 196         dev = alloc_netdev(sizeof(struct net_bridge), name,
 197                            br_dev_setup);
 198
 199         if (!dev)
 200                 return NULL;
 201
 202         br = netdev_priv(dev);
 203         br->dev = dev;
 204
 205         spin_lock_init(&br->lock);
 206         INIT_LIST_HEAD(&br->port_list);
 207         spin_lock_init(&br->hash_lock);
 208
 209         br->bridge_id.prio[0] = 0x80;
 210         br->bridge_id.prio[1] = 0x00;
 211         memset(br->bridge_id.addr, 0, ETH_ALEN);
 212
 213         br->feature_mask = dev->features;
 214         br->stp_enabled = 0;
 215         br->designated_root = br->bridge_id;
 216         br->root_path_cost = 0;
 217         br->root_port = 0;
 218         br->bridge_max_age = br->max_age = 20 * HZ;
 219         br->bridge_hello_time = br->hello_time = 2 * HZ;
 220         br->bridge_forward_delay = br->forward_delay = 15 * HZ;
 221         br->topology_change = 0;
 222         br->topology_change_detected = 0;
 223         br->ageing_time = 300 * HZ;
 224         INIT_LIST_HEAD(&br->age_list);
 225
 226         br_stp_timer_init(br);
 227
 228         return dev;
 229 }
 230
 231 /* find an available port number */
 232 static int find_portno(struct net_bridge *br)
 233 {
 234         int index;
 235         struct net_bridge_port *p;
 236         unsigned long *inuse;
 237
 238         inuse = kmalloc(BITS_TO_LONGS(BR_MAX_PORTS)*sizeof(unsigned long),
 239                         GFP_KERNEL);
 240         if (!inuse)
 241                 return -ENOMEM;
 242
 243         memset(inuse, 0, BITS_TO_LONGS(BR_MAX_PORTS)*sizeof(unsigned long));
 244         set_bit(0, inuse);      /* zero is reserved */
 245         list_for_each_entry(p, &br->port_list, list) {
 246                 set_bit(p->port_no, inuse);
 247         }
 248         index = find_first_zero_bit(inuse, BR_MAX_PORTS);
 249         kfree(inuse);
 250
 251         return (index >= BR_MAX_PORTS) ? -EXFULL : index;
 252 }
 253
 254 /* called with RTNL but without bridge lock */
 255 static struct net_bridge_port *new_nbp(struct net_bridge *br,
 256                                        struct net_device *dev)
 257 {
 258         int index;
 259         struct net_bridge_port *p;
 260
 261         index = find_portno(br);
 262         if (index < 0)
 263                 return ERR_PTR(index);
 264
 265         p = kmalloc(sizeof(*p), GFP_KERNEL);
 266         if (p == NULL)
 267                 return ERR_PTR(-ENOMEM);
 268
 269         memset(p, 0, sizeof(*p));
 270         p->br = br;
 271         dev_hold(dev);
 272         p->dev = dev;
 273         p->path_cost = port_cost(dev);
 274         p->priority = 0x8000 >> BR_PORT_BITS;
 275         p->port_no = index;
 276         br_init_port(p);
 277         p->state = BR_STATE_DISABLED;
 278         INIT_WORK(&p->carrier_check, port_carrier_check, dev);
 279         kobject_init(&p->kobj);
 280
 281         kobject_set_name(&p->kobj, SYSFS_BRIDGE_PORT_ATTR);
 282         p->kobj.ktype = &brport_ktype;
 283         p->kobj.parent = &(dev->class_dev.kobj);
 284         p->kobj.kset = NULL;
 285
 286         return p;
 287 }
 288
 289 int br_add_bridge(const char *name)
 290 {
 291         struct net_device *dev;
 292         int ret;
 293
 294         dev = new_bridge_dev(name);
 295         if (!dev)
 296                 return -ENOMEM;
 297
 298         rtnl_lock();
 299         if (strchr(dev->name, '%')) {
 300                 ret = dev_alloc_name(dev, dev->name);
 301                 if (ret < 0)
 302                         goto err1;
 303         }
 304
 305         ret = register_netdevice(dev);
 306         if (ret)
 307                 goto err2;
 308
 309         /* network device kobject is not setup until
 310          * after rtnl_unlock does it's hotplug magic.
 311          * so hold reference to avoid race.
 312          */
 313         dev_hold(dev);
 314         rtnl_unlock();
 315
 316         ret = br_sysfs_addbr(dev);
 317         dev_put(dev);
 318
 319         if (ret)
 320                 unregister_netdev(dev);
 321  out:
 322         return ret;
 323
 324  err2:
 325         free_netdev(dev);
 326  err1:
 327         rtnl_unlock();
 328         goto out;
 329 }
 330
 331 int br_del_bridge(const char *name)
 332 {
 333         struct net_device *dev;
 334         int ret = 0;
 335
 336         rtnl_lock();
 337         dev = __dev_get_by_name(name);
 338         if (dev == NULL)
 339                 ret =  -ENXIO;  /* Could not find device */
 340
 341         else if (!(dev->priv_flags & IFF_EBRIDGE)) {
 342                 /* Attempt to delete non bridge device! */
 343                 ret = -EPERM;
 344         }
 345
 346         else if (dev->flags & IFF_UP) {
 347                 /* Not shutdown yet. */
 348                 ret = -EBUSY;
 349         }
 350
 351         else
 352                 del_br(netdev_priv(dev));
 353
 354         rtnl_unlock();
 355         return ret;
 356 }
 357
 358 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
 359 int br_min_mtu(const struct net_bridge *br)
 360 {
 361         const struct net_bridge_port *p;
 362         int mtu = 0;
 363
 364         ASSERT_RTNL();
 365
 366         if (list_empty(&br->port_list))
 367                 mtu = ETH_DATA_LEN;
 368         else {
 369                 list_for_each_entry(p, &br->port_list, list) {
 370                         if (!mtu  || p->dev->mtu < mtu)
 371                                 mtu = p->dev->mtu;
 372                 }
 373         }
 374         return mtu;
 375 }
 376
 377 /*
 378  * Recomputes features using slave's features
 379  */
 380 void br_features_recompute(struct net_bridge *br)
 381 {
 382         struct net_bridge_port *p;
 383         unsigned long features, checksum;
 384
 385         features = br->feature_mask &~ NETIF_F_IP_CSUM;
 386         checksum = br->feature_mask & NETIF_F_IP_CSUM;
 387
 388         list_for_each_entry(p, &br->port_list, list) {
 389                 if (!(p->dev->features
 390                       & (NETIF_F_IP_CSUM|NETIF_F_NO_CSUM|NETIF_F_HW_CSUM)))
 391                         checksum = 0;
 392                 features &= p->dev->features;
 393         }
 394
 395         br->dev->features = features | checksum | NETIF_F_LLTX;
 396 }
 397
 398 /* called with RTNL */
 399 int br_add_if(struct net_bridge *br, struct net_device *dev)
 400 {
 401         struct net_bridge_port *p;
 402         int err = 0;
 403
 404         if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER)
 405                 return -EINVAL;
 406
 407         if (dev->hard_start_xmit == br_dev_xmit)
 408                 return -ELOOP;
 409
 410         if (dev->br_port != NULL)
 411                 return -EBUSY;
 412
 413         p = new_nbp(br, dev);
 414         if (IS_ERR(p))
 415                 return PTR_ERR(p);
 416
 417         err = kobject_add(&p->kobj);
 418         if (err)
 419                 goto err0;
 420
 421         err = br_fdb_insert(br, p, dev->dev_addr);
 422         if (err)
 423                 goto err1;
 424
 425         err = br_sysfs_addif(p);
 426         if (err)
 427                 goto err2;
 428
 429         rcu_assign_pointer(dev->br_port, p);
 430         dev_set_promiscuity(dev, 1);
 431
 432         list_add_rcu(&p->list, &br->port_list);
 433
 434         spin_lock_bh(&br->lock);
 435         br_stp_recalculate_bridge_id(br);
 436         br_features_recompute(br);
 437         schedule_delayed_work(&p->carrier_check, BR_PORT_DEBOUNCE);
 438         spin_unlock_bh(&br->lock);
 439
 440         dev_set_mtu(br->dev, br_min_mtu(br));
 441         kobject_uevent(&p->kobj, KOBJ_ADD);
 442
 443         return 0;
 444 err2:
 445         br_fdb_delete_by_port(br, p);
 446 err1:
 447         kobject_del(&p->kobj);
 448 err0:
 449         kobject_put(&p->kobj);
 450         return err;
 451 }
 452
 453 /* called with RTNL */
 454 int br_del_if(struct net_bridge *br, struct net_device *dev)
 455 {
 456         struct net_bridge_port *p = dev->br_port;
 457
 458         if (!p || p->br != br)
 459                 return -EINVAL;
 460
 461         del_nbp(p);
 462
 463         spin_lock_bh(&br->lock);
 464         br_stp_recalculate_bridge_id(br);
 465         br_features_recompute(br);
 466         spin_unlock_bh(&br->lock);
 467
 468         return 0;
 469 }
 470
 471 void __exit br_cleanup_bridges(void)
 472 {
 473         struct net_device *dev, *nxt;
 474
 475         rtnl_lock();
 476         for (dev = dev_base; dev; dev = nxt) {
 477                 nxt = dev->next;
 478                 if (dev->priv_flags & IFF_EBRIDGE)
 479                         del_br(dev->priv);
 480         }
 481         rtnl_unlock();
 482
 483 }