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_uevent(&p->kobj, KOBJ_REMOVE);
 173         kobject_del(&p->kobj);
 174
 175         call_rcu(&p->rcu, destroy_nbp_rcu);
 176 }
 177
 178 /* called with RTNL */
 179 static void del_br(struct net_bridge *br)
 180 {
 181         struct net_bridge_port *p, *n;
 182
 183         list_for_each_entry_safe(p, n, &br->port_list, list) {
 184                 del_nbp(p);
 185         }
 186
 187         del_timer_sync(&br->gc_timer);
 188
 189         br_sysfs_delbr(br->dev);
 190         unregister_netdevice(br->dev);
 191 }
 192
 193 static struct net_device *new_bridge_dev(const char *name)
 194 {
 195         struct net_bridge *br;
 196         struct net_device *dev;
 197
 198         dev = alloc_netdev(sizeof(struct net_bridge), name,
 199                            br_dev_setup);
 200
 201         if (!dev)
 202                 return NULL;
 203
 204         br = netdev_priv(dev);
 205         br->dev = dev;
 206
 207         spin_lock_init(&br->lock);
 208         INIT_LIST_HEAD(&br->port_list);
 209         spin_lock_init(&br->hash_lock);
 210
 211         br->bridge_id.prio[0] = 0x80;
 212         br->bridge_id.prio[1] = 0x00;
 213
 214         memcpy(br->group_addr, br_group_address, ETH_ALEN);
 215
 216         br->feature_mask = dev->features;
 217         br->stp_enabled = 0;
 218         br->designated_root = br->bridge_id;
 219         br->root_path_cost = 0;
 220         br->root_port = 0;
 221         br->bridge_max_age = br->max_age = 20 * HZ;
 222         br->bridge_hello_time = br->hello_time = 2 * HZ;
 223         br->bridge_forward_delay = br->forward_delay = 15 * HZ;
 224         br->topology_change = 0;
 225         br->topology_change_detected = 0;
 226         br->ageing_time = 300 * HZ;
 227         INIT_LIST_HEAD(&br->age_list);
 228
 229         br_stp_timer_init(br);
 230
 231         return dev;
 232 }
 233
 234 /* find an available port number */
 235 static int find_portno(struct net_bridge *br)
 236 {
 237         int index;
 238         struct net_bridge_port *p;
 239         unsigned long *inuse;
 240
 241         inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
 242                         GFP_KERNEL);
 243         if (!inuse)
 244                 return -ENOMEM;
 245
 246         set_bit(0, inuse);      /* zero is reserved */
 247         list_for_each_entry(p, &br->port_list, list) {
 248                 set_bit(p->port_no, inuse);
 249         }
 250         index = find_first_zero_bit(inuse, BR_MAX_PORTS);
 251         kfree(inuse);
 252
 253         return (index >= BR_MAX_PORTS) ? -EXFULL : index;
 254 }
 255
 256 /* called with RTNL but without bridge lock */
 257 static struct net_bridge_port *new_nbp(struct net_bridge *br,
 258                                        struct net_device *dev)
 259 {
 260         int index;
 261         struct net_bridge_port *p;
 262
 263         index = find_portno(br);
 264         if (index < 0)
 265                 return ERR_PTR(index);
 266
 267         p = kzalloc(sizeof(*p), GFP_KERNEL);
 268         if (p == NULL)
 269                 return ERR_PTR(-ENOMEM);
 270
 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         br_stp_port_timer_init(p);
 281
 282         kobject_init(&p->kobj);
 283         kobject_set_name(&p->kobj, SYSFS_BRIDGE_PORT_ATTR);
 284         p->kobj.ktype = &brport_ktype;
 285         p->kobj.parent = &(dev->class_dev.kobj);
 286         p->kobj.kset = NULL;
 287
 288         return p;
 289 }
 290
 291 int br_add_bridge(const char *name)
 292 {
 293         struct net_device *dev;
 294         int ret;
 295
 296         dev = new_bridge_dev(name);
 297         if (!dev)
 298                 return -ENOMEM;
 299
 300         rtnl_lock();
 301         if (strchr(dev->name, '%')) {
 302                 ret = dev_alloc_name(dev, dev->name);
 303                 if (ret < 0) {
 304                         free_netdev(dev);
 305                         goto out;
 306                 }
 307         }
 308
 309         ret = register_netdevice(dev);
 310         if (ret)
 311                 goto out;
 312
 313         ret = br_sysfs_addbr(dev);
 314         if (ret)
 315                 unregister_netdevice(dev);
 316  out:
 317         rtnl_unlock();
 318         return ret;
 319 }
 320
 321 int br_del_bridge(const char *name)
 322 {
 323         struct net_device *dev;
 324         int ret = 0;
 325
 326         rtnl_lock();
 327         dev = __dev_get_by_name(name);
 328         if (dev == NULL)
 329                 ret =  -ENXIO;  /* Could not find device */
 330
 331         else if (!(dev->priv_flags & IFF_EBRIDGE)) {
 332                 /* Attempt to delete non bridge device! */
 333                 ret = -EPERM;
 334         }
 335
 336         else if (dev->flags & IFF_UP) {
 337                 /* Not shutdown yet. */
 338                 ret = -EBUSY;
 339         }
 340
 341         else
 342                 del_br(netdev_priv(dev));
 343
 344         rtnl_unlock();
 345         return ret;
 346 }
 347
 348 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
 349 int br_min_mtu(const struct net_bridge *br)
 350 {
 351         const struct net_bridge_port *p;
 352         int mtu = 0;
 353
 354         ASSERT_RTNL();
 355
 356         if (list_empty(&br->port_list))
 357                 mtu = ETH_DATA_LEN;
 358         else {
 359                 list_for_each_entry(p, &br->port_list, list) {
 360                         if (!mtu  || p->dev->mtu < mtu)
 361                                 mtu = p->dev->mtu;
 362                 }
 363         }
 364         return mtu;
 365 }
 366
 367 /*
 368  * Recomputes features using slave's features
 369  */
 370 void br_features_recompute(struct net_bridge *br)
 371 {
 372         struct net_bridge_port *p;
 373         unsigned long features, checksum;
 374
 375         checksum = br->feature_mask & NETIF_F_ALL_CSUM ? NETIF_F_NO_CSUM : 0;
 376         features = br->feature_mask & ~NETIF_F_ALL_CSUM;
 377
 378         list_for_each_entry(p, &br->port_list, list) {
 379                 unsigned long feature = p->dev->features;
 380
 381                 if (checksum & NETIF_F_NO_CSUM && !(feature & NETIF_F_NO_CSUM))
 382                         checksum ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM;
 383                 if (checksum & NETIF_F_HW_CSUM && !(feature & NETIF_F_HW_CSUM))
 384                         checksum ^= NETIF_F_HW_CSUM | NETIF_F_IP_CSUM;
 385                 if (!(feature & NETIF_F_IP_CSUM))
 386                         checksum = 0;
 387
 388                 if (feature & NETIF_F_GSO)
 389                         feature |= NETIF_F_TSO;
 390                 feature |= NETIF_F_GSO;
 391
 392                 features &= feature;
 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 }