net/bridge/br_if.c

   1 /*
   2  *      Userspace interface
   3  *      Linux ethernet bridge
   4  *
   5  *      Authors:
   6  *      Lennert Buytenhek               <buytenh@gnu.org>
   7  *
   8  *      This program is free software; you can redistribute it and/or
   9  *      modify it under the terms of the GNU General Public License
  10  *      as published by the Free Software Foundation; either version
  11  *      2 of the License, or (at your option) any later version.
  12  */
  13
  14 #include <linux/kernel.h>
  15 #include <linux/netdevice.h>
  16 #include <linux/etherdevice.h>
  17 #include <linux/netpoll.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 <linux/slab.h>
  25 #include <net/sock.h>
  26 #include <linux/if_vlan.h>
  27
  28 #include "br_private.h"
  29
  30 /*
  31  * Determine initial path cost based on speed.
  32  * using recommendations from 802.1d standard
  33  *
  34  * Since driver might sleep need to not be holding any locks.
  35  */
  36 static int port_cost(struct net_device *dev)
  37 {
  38         struct ethtool_cmd ecmd;
  39
  40         if (!__ethtool_get_settings(dev, &ecmd)) {
  41                 switch (ethtool_cmd_speed(&ecmd)) {
  42                 case SPEED_10000:
  43                         return 2;
  44                 case SPEED_1000:
  45                         return 4;
  46                 case SPEED_100:
  47                         return 19;
  48                 case SPEED_10:
  49                         return 100;
  50                 }
  51         }
  52
  53         /* Old silly heuristics based on name */
  54         if (!strncmp(dev->name, "lec", 3))
  55                 return 7;
  56
  57         if (!strncmp(dev->name, "plip", 4))
  58                 return 2500;
  59
  60         return 100;     /* assume old 10Mbps */
  61 }
  62
  63
  64 /* Check for port carrier transitions. */
  65 void br_port_carrier_check(struct net_bridge_port *p)
  66 {
  67         struct net_device *dev = p->dev;
  68         struct net_bridge *br = p->br;
  69
  70         if (!(p->flags & BR_ADMIN_COST) &&
  71             netif_running(dev) && netif_oper_up(dev))
  72                 p->path_cost = port_cost(dev);
  73
  74         if (!netif_running(br->dev))
  75                 return;
  76
  77         spin_lock_bh(&br->lock);
  78         if (netif_running(dev) && netif_oper_up(dev)) {
  79                 if (p->state == BR_STATE_DISABLED)
  80                         br_stp_enable_port(p);
  81         } else {
  82                 if (p->state != BR_STATE_DISABLED)
  83                         br_stp_disable_port(p);
  84         }
  85         spin_unlock_bh(&br->lock);
  86 }
  87
  88 static void nbp_update_port_count(struct net_bridge *br)
  89 {
  90         struct net_bridge_port *p;
  91         u32 cnt = 0;
  92
  93         list_for_each_entry(p, &br->port_list, list) {
  94                 if (br_auto_port(p))
  95                         cnt++;
  96         }
  97         br->auto_cnt = cnt;
  98 }
  99
 100 static void release_nbp(struct kobject *kobj)
 101 {
 102         struct net_bridge_port *p
 103                 = container_of(kobj, struct net_bridge_port, kobj);
 104         kfree(p);
 105 }
 106
 107 static struct kobj_type brport_ktype = {
 108 #ifdef CONFIG_SYSFS
 109         .sysfs_ops = &brport_sysfs_ops,
 110 #endif
 111         .release = release_nbp,
 112 };
 113
 114 static void destroy_nbp(struct net_bridge_port *p)
 115 {
 116         struct net_device *dev = p->dev;
 117
 118         p->br = NULL;
 119         p->dev = NULL;
 120         dev_put(dev);
 121
 122         kobject_put(&p->kobj);
 123 }
 124
 125 static void destroy_nbp_rcu(struct rcu_head *head)
 126 {
 127         struct net_bridge_port *p =
 128                         container_of(head, struct net_bridge_port, rcu);
 129         destroy_nbp(p);
 130 }
 131
 132 /* Delete port(interface) from bridge is done in two steps.
 133  * via RCU. First step, marks device as down. That deletes
 134  * all the timers and stops new packets from flowing through.
 135  *
 136  * Final cleanup doesn't occur until after all CPU's finished
 137  * processing packets.
 138  *
 139  * Protected from multiple admin operations by RTNL mutex
 140  */
 141 static void del_nbp(struct net_bridge_port *p)
 142 {
 143         struct net_bridge *br = p->br;
 144         struct net_device *dev = p->dev;
 145
 146         sysfs_remove_link(br->ifobj, p->dev->name);
 147
 148         dev_set_promiscuity(dev, -1);
 149
 150         spin_lock_bh(&br->lock);
 151         br_stp_disable_port(p);
 152         spin_unlock_bh(&br->lock);
 153
 154         br_ifinfo_notify(RTM_DELLINK, p);
 155
 156         nbp_vlan_flush(p);
 157         br_fdb_delete_by_port(br, p, 1);
 158
 159         list_del_rcu(&p->list);
 160
 161         nbp_update_port_count(br);
 162
 163         dev->priv_flags &= ~IFF_BRIDGE_PORT;
 164
 165         netdev_rx_handler_unregister(dev);
 166
 167         netdev_upper_dev_unlink(dev, br->dev);
 168
 169         br_multicast_del_port(p);
 170
 171         kobject_uevent(&p->kobj, KOBJ_REMOVE);
 172         kobject_del(&p->kobj);
 173
 174         br_netpoll_disable(p);
 175
 176         call_rcu(&p->rcu, destroy_nbp_rcu);
 177 }
 178
 179 /* Delete bridge device */
 180 void br_dev_delete(struct net_device *dev, struct list_head *head)
 181 {
 182         struct net_bridge *br = netdev_priv(dev);
 183         struct net_bridge_port *p, *n;
 184
 185         list_for_each_entry_safe(p, n, &br->port_list, list) {
 186                 del_nbp(p);
 187         }
 188
 189         br_fdb_delete_by_port(br, NULL, 1);
 190
 191         br_vlan_flush(br);
 192         del_timer_sync(&br->gc_timer);
 193
 194         br_sysfs_delbr(br->dev);
 195         unregister_netdevice_queue(br->dev, head);
 196 }
 197
 198 /* find an available port number */
 199 static int find_portno(struct net_bridge *br)
 200 {
 201         int index;
 202         struct net_bridge_port *p;
 203         unsigned long *inuse;
 204
 205         inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
 206                         GFP_KERNEL);
 207         if (!inuse)
 208                 return -ENOMEM;
 209
 210         set_bit(0, inuse);      /* zero is reserved */
 211         list_for_each_entry(p, &br->port_list, list) {
 212                 set_bit(p->port_no, inuse);
 213         }
 214         index = find_first_zero_bit(inuse, BR_MAX_PORTS);
 215         kfree(inuse);
 216
 217         return (index >= BR_MAX_PORTS) ? -EXFULL : index;
 218 }
 219
 220 /* called with RTNL but without bridge lock */
 221 static struct net_bridge_port *new_nbp(struct net_bridge *br,
 222                                        struct net_device *dev)
 223 {
 224         int index;
 225         struct net_bridge_port *p;
 226
 227         index = find_portno(br);
 228         if (index < 0)
 229                 return ERR_PTR(index);
 230
 231         p = kzalloc(sizeof(*p), GFP_KERNEL);
 232         if (p == NULL)
 233                 return ERR_PTR(-ENOMEM);
 234
 235         p->br = br;
 236         dev_hold(dev);
 237         p->dev = dev;
 238         p->path_cost = port_cost(dev);
 239         p->priority = 0x8000 >> BR_PORT_BITS;
 240         p->port_no = index;
 241         p->flags = BR_LEARNING | BR_FLOOD | BR_PROMISC;
 242         br_init_port(p);
 243         p->state = BR_STATE_DISABLED;
 244         br_stp_port_timer_init(p);
 245         br_multicast_add_port(p);
 246
 247         return p;
 248 }
 249
 250 int br_add_bridge(struct net *net, const char *name)
 251 {
 252         struct net_device *dev;
 253         int res;
 254
 255         dev = alloc_netdev(sizeof(struct net_bridge), name,
 256                            br_dev_setup);
 257
 258         if (!dev)
 259                 return -ENOMEM;
 260
 261         dev_net_set(dev, net);
 262         dev->rtnl_link_ops = &br_link_ops;
 263
 264         res = register_netdev(dev);
 265         if (res)
 266                 free_netdev(dev);
 267         return res;
 268 }
 269
 270 int br_del_bridge(struct net *net, const char *name)
 271 {
 272         struct net_device *dev;
 273         int ret = 0;
 274
 275         rtnl_lock();
 276         dev = __dev_get_by_name(net, name);
 277         if (dev == NULL)
 278                 ret =  -ENXIO;  /* Could not find device */
 279
 280         else if (!(dev->priv_flags & IFF_EBRIDGE)) {
 281                 /* Attempt to delete non bridge device! */
 282                 ret = -EPERM;
 283         }
 284
 285         else if (dev->flags & IFF_UP) {
 286                 /* Not shutdown yet. */
 287                 ret = -EBUSY;
 288         }
 289
 290         else
 291                 br_dev_delete(dev, NULL);
 292
 293         rtnl_unlock();
 294         return ret;
 295 }
 296
 297 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
 298 int br_min_mtu(const struct net_bridge *br)
 299 {
 300         const struct net_bridge_port *p;
 301         int mtu = 0;
 302
 303         ASSERT_RTNL();
 304
 305         if (list_empty(&br->port_list))
 306                 mtu = ETH_DATA_LEN;
 307         else {
 308                 list_for_each_entry(p, &br->port_list, list) {
 309                         if (!mtu  || p->dev->mtu < mtu)
 310                                 mtu = p->dev->mtu;
 311                 }
 312         }
 313         return mtu;
 314 }
 315
 316 /*
 317  * Recomputes features using slave's features
 318  */
 319 netdev_features_t br_features_recompute(struct net_bridge *br,
 320         netdev_features_t features)
 321 {
 322         struct net_bridge_port *p;
 323         netdev_features_t mask;
 324
 325         if (list_empty(&br->port_list))
 326                 return features;
 327
 328         mask = features;
 329         features &= ~NETIF_F_ONE_FOR_ALL;
 330
 331         list_for_each_entry(p, &br->port_list, list) {
 332                 features = netdev_increment_features(features,
 333                                                      p->dev->features, mask);
 334         }
 335
 336         return features;
 337 }
 338
 339 /* called with RTNL */
 340 int br_add_if(struct net_bridge *br, struct net_device *dev)
 341 {
 342         struct net_bridge_port *p;
 343         int err = 0;
 344         bool changed_addr;
 345
 346         /* Don't allow bridging non-ethernet like devices */
 347         if ((dev->flags & IFF_LOOPBACK) ||
 348             dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN ||
 349             !is_valid_ether_addr(dev->dev_addr))
 350                 return -EINVAL;
 351
 352         /* No bridging of bridges */
 353         if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit)
 354                 return -ELOOP;
 355
 356         /* Device is already being bridged */
 357         if (br_port_exists(dev))
 358                 return -EBUSY;
 359
 360         /* No bridging devices that dislike that (e.g. wireless) */
 361         if (dev->priv_flags & IFF_DONT_BRIDGE)
 362                 return -EOPNOTSUPP;
 363
 364         p = new_nbp(br, dev);
 365         if (IS_ERR(p))
 366                 return PTR_ERR(p);
 367
 368         call_netdevice_notifiers(NETDEV_JOIN, dev);
 369
 370         err = dev_set_promiscuity(dev, 1);
 371         if (err)
 372                 goto put_back;
 373
 374         err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
 375                                    SYSFS_BRIDGE_PORT_ATTR);
 376         if (err)
 377                 goto err1;
 378
 379         err = br_sysfs_addif(p);
 380         if (err)
 381                 goto err2;
 382
 383         err = br_netpoll_enable(p);
 384         if (err)
 385                 goto err3;
 386
 387         err = netdev_master_upper_dev_link(dev, br->dev);
 388         if (err)
 389                 goto err4;
 390
 391         err = netdev_rx_handler_register(dev, br_handle_frame, p);
 392         if (err)
 393                 goto err5;
 394
 395         dev->priv_flags |= IFF_BRIDGE_PORT;
 396
 397         dev_disable_lro(dev);
 398
 399         list_add_rcu(&p->list, &br->port_list);
 400
 401         nbp_update_port_count(br);
 402
 403         netdev_update_features(br->dev);
 404
 405         if (br->dev->needed_headroom < dev->needed_headroom)
 406                 br->dev->needed_headroom = dev->needed_headroom;
 407
 408         if (br_fdb_insert(br, p, dev->dev_addr, 0))
 409                 netdev_err(dev, "failed insert local address bridge forwarding table\n");
 410
 411         spin_lock_bh(&br->lock);
 412         changed_addr = br_stp_recalculate_bridge_id(br);
 413
 414         if (netif_running(dev) && netif_oper_up(dev) &&
 415             (br->dev->flags & IFF_UP))
 416                 br_stp_enable_port(p);
 417         spin_unlock_bh(&br->lock);
 418
 419         br_ifinfo_notify(RTM_NEWLINK, p);
 420
 421         if (changed_addr)
 422                 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
 423
 424         dev_set_mtu(br->dev, br_min_mtu(br));
 425
 426         kobject_uevent(&p->kobj, KOBJ_ADD);
 427
 428         return 0;
 429
 430 err5:
 431         netdev_upper_dev_unlink(dev, br->dev);
 432 err4:
 433         br_netpoll_disable(p);
 434 err3:
 435         sysfs_remove_link(br->ifobj, p->dev->name);
 436 err2:
 437         kobject_put(&p->kobj);
 438         p = NULL; /* kobject_put frees */
 439 err1:
 440         dev_set_promiscuity(dev, -1);
 441 put_back:
 442         dev_put(dev);
 443         kfree(p);
 444         return err;
 445 }
 446
 447 /* called with RTNL */
 448 int br_del_if(struct net_bridge *br, struct net_device *dev)
 449 {
 450         struct net_bridge_port *p;
 451         bool changed_addr;
 452
 453         p = br_port_get_rtnl(dev);
 454         if (!p || p->br != br)
 455                 return -EINVAL;
 456
 457         /* Since more than one interface can be attached to a bridge,
 458          * there still maybe an alternate path for netconsole to use;
 459          * therefore there is no reason for a NETDEV_RELEASE event.
 460          */
 461         del_nbp(p);
 462
 463         spin_lock_bh(&br->lock);
 464         changed_addr = br_stp_recalculate_bridge_id(br);
 465         spin_unlock_bh(&br->lock);
 466
 467         if (changed_addr)
 468                 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
 469
 470         netdev_update_features(br->dev);
 471
 472         return 0;
 473 }
 474
 475 void br_port_flags_change(struct net_bridge_port *p, unsigned long mask)
 476 {
 477         struct net_bridge *br = p->br;
 478
 479         if (mask & BR_AUTO_MASK)
 480                 nbp_update_port_count(br);
 481 }