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/netpoll.h>
  17 #include <linux/ethtool.h>
  18 #include <linux/if_arp.h>
  19 #include <linux/module.h>
  20 #include <linux/init.h>
  21 #include <linux/rtnetlink.h>
  22 #include <linux/if_ether.h>
  23 #include <linux/slab.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  * Since driver might sleep need to not be holding any locks.
  33  */
  34 static int port_cost(struct net_device *dev)
  35 {
  36         if (dev->ethtool_ops && dev->ethtool_ops->get_settings) {
  37                 struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET, };
  38
  39                 if (!dev_ethtool_get_settings(dev, &ecmd)) {
  40                         switch (ethtool_cmd_speed(&ecmd)) {
  41                         case SPEED_10000:
  42                                 return 2;
  43                         case SPEED_1000:
  44                                 return 4;
  45                         case SPEED_100:
  46                                 return 19;
  47                         case SPEED_10:
  48                                 return 100;
  49                         }
  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 transistions. */
  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 (netif_running(dev) && netif_carrier_ok(dev))
  71                 p->path_cost = port_cost(dev);
  72
  73         if (!netif_running(br->dev))
  74                 return;
  75
  76         spin_lock_bh(&br->lock);
  77         if (netif_running(dev) && netif_carrier_ok(dev)) {
  78                 if (p->state == BR_STATE_DISABLED)
  79                         br_stp_enable_port(p);
  80         } else {
  81                 if (p->state != BR_STATE_DISABLED)
  82                         br_stp_disable_port(p);
  83         }
  84         spin_unlock_bh(&br->lock);
  85 }
  86
  87 static void release_nbp(struct kobject *kobj)
  88 {
  89         struct net_bridge_port *p
  90                 = container_of(kobj, struct net_bridge_port, kobj);
  91         kfree(p);
  92 }
  93
  94 static struct kobj_type brport_ktype = {
  95 #ifdef CONFIG_SYSFS
  96         .sysfs_ops = &brport_sysfs_ops,
  97 #endif
  98         .release = release_nbp,
  99 };
 100
 101 static void destroy_nbp(struct net_bridge_port *p)
 102 {
 103         struct net_device *dev = p->dev;
 104
 105         p->br = NULL;
 106         p->dev = NULL;
 107         dev_put(dev);
 108
 109         kobject_put(&p->kobj);
 110 }
 111
 112 static void destroy_nbp_rcu(struct rcu_head *head)
 113 {
 114         struct net_bridge_port *p =
 115                         container_of(head, struct net_bridge_port, rcu);
 116         destroy_nbp(p);
 117 }
 118
 119 /* Delete port(interface) from bridge is done in two steps.
 120  * via RCU. First step, marks device as down. That deletes
 121  * all the timers and stops new packets from flowing through.
 122  *
 123  * Final cleanup doesn't occur until after all CPU's finished
 124  * processing packets.
 125  *
 126  * Protected from multiple admin operations by RTNL mutex
 127  */
 128 static void del_nbp(struct net_bridge_port *p)
 129 {
 130         struct net_bridge *br = p->br;
 131         struct net_device *dev = p->dev;
 132
 133         sysfs_remove_link(br->ifobj, p->dev->name);
 134
 135         dev_set_promiscuity(dev, -1);
 136
 137         spin_lock_bh(&br->lock);
 138         br_stp_disable_port(p);
 139         spin_unlock_bh(&br->lock);
 140
 141         br_ifinfo_notify(RTM_DELLINK, p);
 142
 143         br_fdb_delete_by_port(br, p, 1);
 144
 145         list_del_rcu(&p->list);
 146
 147         dev->priv_flags &= ~IFF_BRIDGE_PORT;
 148
 149         netdev_rx_handler_unregister(dev);
 150         synchronize_net();
 151
 152         netdev_set_master(dev, NULL);
 153
 154         br_multicast_del_port(p);
 155
 156         kobject_uevent(&p->kobj, KOBJ_REMOVE);
 157         kobject_del(&p->kobj);
 158
 159         br_netpoll_disable(p);
 160
 161         call_rcu(&p->rcu, destroy_nbp_rcu);
 162 }
 163
 164 /* called with RTNL */
 165 static void del_br(struct net_bridge *br, struct list_head *head)
 166 {
 167         struct net_bridge_port *p, *n;
 168
 169         list_for_each_entry_safe(p, n, &br->port_list, list) {
 170                 del_nbp(p);
 171         }
 172
 173         del_timer_sync(&br->gc_timer);
 174
 175         br_sysfs_delbr(br->dev);
 176         unregister_netdevice_queue(br->dev, head);
 177 }
 178
 179 /* find an available port number */
 180 static int find_portno(struct net_bridge *br)
 181 {
 182         int index;
 183         struct net_bridge_port *p;
 184         unsigned long *inuse;
 185
 186         inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
 187                         GFP_KERNEL);
 188         if (!inuse)
 189                 return -ENOMEM;
 190
 191         set_bit(0, inuse);      /* zero is reserved */
 192         list_for_each_entry(p, &br->port_list, list) {
 193                 set_bit(p->port_no, inuse);
 194         }
 195         index = find_first_zero_bit(inuse, BR_MAX_PORTS);
 196         kfree(inuse);
 197
 198         return (index >= BR_MAX_PORTS) ? -EXFULL : index;
 199 }
 200
 201 /* called with RTNL but without bridge lock */
 202 static struct net_bridge_port *new_nbp(struct net_bridge *br,
 203                                        struct net_device *dev)
 204 {
 205         int index;
 206         struct net_bridge_port *p;
 207
 208         index = find_portno(br);
 209         if (index < 0)
 210                 return ERR_PTR(index);
 211
 212         p = kzalloc(sizeof(*p), GFP_KERNEL);
 213         if (p == NULL)
 214                 return ERR_PTR(-ENOMEM);
 215
 216         p->br = br;
 217         dev_hold(dev);
 218         p->dev = dev;
 219         p->path_cost = port_cost(dev);
 220         p->priority = 0x8000 >> BR_PORT_BITS;
 221         p->port_no = index;
 222         p->flags = 0;
 223         br_init_port(p);
 224         p->state = BR_STATE_DISABLED;
 225         br_stp_port_timer_init(p);
 226         br_multicast_add_port(p);
 227
 228         return p;
 229 }
 230
 231 int br_add_bridge(struct net *net, const char *name)
 232 {
 233         struct net_device *dev;
 234         int res;
 235
 236         dev = alloc_netdev(sizeof(struct net_bridge), name,
 237                            br_dev_setup);
 238
 239         if (!dev)
 240                 return -ENOMEM;
 241
 242         dev_net_set(dev, net);
 243
 244         res = register_netdev(dev);
 245         if (res)
 246                 free_netdev(dev);
 247         return res;
 248 }
 249
 250 int br_del_bridge(struct net *net, const char *name)
 251 {
 252         struct net_device *dev;
 253         int ret = 0;
 254
 255         rtnl_lock();
 256         dev = __dev_get_by_name(net, name);
 257         if (dev == NULL)
 258                 ret =  -ENXIO;  /* Could not find device */
 259
 260         else if (!(dev->priv_flags & IFF_EBRIDGE)) {
 261                 /* Attempt to delete non bridge device! */
 262                 ret = -EPERM;
 263         }
 264
 265         else if (dev->flags & IFF_UP) {
 266                 /* Not shutdown yet. */
 267                 ret = -EBUSY;
 268         }
 269
 270         else
 271                 del_br(netdev_priv(dev), NULL);
 272
 273         rtnl_unlock();
 274         return ret;
 275 }
 276
 277 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
 278 int br_min_mtu(const struct net_bridge *br)
 279 {
 280         const struct net_bridge_port *p;
 281         int mtu = 0;
 282
 283         ASSERT_RTNL();
 284
 285         if (list_empty(&br->port_list))
 286                 mtu = ETH_DATA_LEN;
 287         else {
 288                 list_for_each_entry(p, &br->port_list, list) {
 289                         if (!mtu  || p->dev->mtu < mtu)
 290                                 mtu = p->dev->mtu;
 291                 }
 292         }
 293         return mtu;
 294 }
 295
 296 /*
 297  * Recomputes features using slave's features
 298  */
 299 u32 br_features_recompute(struct net_bridge *br, u32 features)
 300 {
 301         struct net_bridge_port *p;
 302         u32 mask;
 303
 304         if (list_empty(&br->port_list))
 305                 return features;
 306
 307         mask = features;
 308         features &= ~NETIF_F_ONE_FOR_ALL;
 309
 310         list_for_each_entry(p, &br->port_list, list) {
 311                 features = netdev_increment_features(features,
 312                                                      p->dev->features, mask);
 313         }
 314
 315         return features;
 316 }
 317
 318 /* called with RTNL */
 319 int br_add_if(struct net_bridge *br, struct net_device *dev)
 320 {
 321         struct net_bridge_port *p;
 322         int err = 0;
 323         bool changed_addr;
 324
 325         /* Don't allow bridging non-ethernet like devices */
 326         if ((dev->flags & IFF_LOOPBACK) ||
 327             dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN)
 328                 return -EINVAL;
 329
 330         /* No bridging of bridges */
 331         if (dev->netdev_ops->ndo_start_xmit == br_dev_xmit)
 332                 return -ELOOP;
 333
 334         /* Device is already being bridged */
 335         if (br_port_exists(dev))
 336                 return -EBUSY;
 337
 338         /* No bridging devices that dislike that (e.g. wireless) */
 339         if (dev->priv_flags & IFF_DONT_BRIDGE)
 340                 return -EOPNOTSUPP;
 341
 342         p = new_nbp(br, dev);
 343         if (IS_ERR(p))
 344                 return PTR_ERR(p);
 345
 346         call_netdevice_notifiers(NETDEV_JOIN, dev);
 347
 348         err = dev_set_promiscuity(dev, 1);
 349         if (err)
 350                 goto put_back;
 351
 352         err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
 353                                    SYSFS_BRIDGE_PORT_ATTR);
 354         if (err)
 355                 goto err0;
 356
 357         err = br_fdb_insert(br, p, dev->dev_addr);
 358         if (err)
 359                 goto err1;
 360
 361         err = br_sysfs_addif(p);
 362         if (err)
 363                 goto err2;
 364
 365         if (br_netpoll_info(br) && ((err = br_netpoll_enable(p))))
 366                 goto err3;
 367
 368         err = netdev_set_master(dev, br->dev);
 369         if (err)
 370                 goto err3;
 371
 372         err = netdev_rx_handler_register(dev, br_handle_frame, p);
 373         if (err)
 374                 goto err4;
 375
 376         dev->priv_flags |= IFF_BRIDGE_PORT;
 377
 378         dev_disable_lro(dev);
 379
 380         list_add_rcu(&p->list, &br->port_list);
 381
 382         netdev_update_features(br->dev);
 383
 384         spin_lock_bh(&br->lock);
 385         changed_addr = br_stp_recalculate_bridge_id(br);
 386
 387         if ((dev->flags & IFF_UP) && netif_carrier_ok(dev) &&
 388             (br->dev->flags & IFF_UP))
 389                 br_stp_enable_port(p);
 390         spin_unlock_bh(&br->lock);
 391
 392         br_ifinfo_notify(RTM_NEWLINK, p);
 393
 394         if (changed_addr)
 395                 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
 396
 397         dev_set_mtu(br->dev, br_min_mtu(br));
 398
 399         kobject_uevent(&p->kobj, KOBJ_ADD);
 400
 401         return 0;
 402
 403 err4:
 404         netdev_set_master(dev, NULL);
 405 err3:
 406         sysfs_remove_link(br->ifobj, p->dev->name);
 407 err2:
 408         br_fdb_delete_by_port(br, p, 1);
 409 err1:
 410         kobject_put(&p->kobj);
 411         p = NULL; /* kobject_put frees */
 412 err0:
 413         dev_set_promiscuity(dev, -1);
 414 put_back:
 415         dev_put(dev);
 416         kfree(p);
 417         return err;
 418 }
 419
 420 /* called with RTNL */
 421 int br_del_if(struct net_bridge *br, struct net_device *dev)
 422 {
 423         struct net_bridge_port *p;
 424         bool changed_addr;
 425
 426         p = br_port_get_rtnl(dev);
 427         if (!p || p->br != br)
 428                 return -EINVAL;
 429
 430         del_nbp(p);
 431
 432         spin_lock_bh(&br->lock);
 433         changed_addr = br_stp_recalculate_bridge_id(br);
 434         spin_unlock_bh(&br->lock);
 435
 436         if (changed_addr)
 437                 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
 438
 439         netdev_update_features(br->dev);
 440
 441         return 0;
 442 }
 443
 444 void __net_exit br_net_exit(struct net *net)
 445 {
 446         struct net_device *dev;
 447         LIST_HEAD(list);
 448
 449         rtnl_lock();
 450         for_each_netdev(net, dev)
 451                 if (dev->priv_flags & IFF_EBRIDGE)
 452                         del_br(netdev_priv(dev), &list);
 453
 454         unregister_netdevice_many(&list);
 455         rtnl_unlock();
 456
 457 }