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