bridge: fix ordering of NEWLINK and NEWNEIGH events
[linux-2.6.git] / 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
27 #include "br_private.h"
28
29 /*
30  * Determine initial path cost based on speed.
31  * using recommendations from 802.1d standard
32  *
33  * Since driver might sleep need to not be holding any locks.
34  */
35 static int port_cost(struct net_device *dev)
36 {
37         struct ethtool_cmd ecmd;
38
39         if (!__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         /* Old silly heuristics based on name */
53         if (!strncmp(dev->name, "lec", 3))
54                 return 7;
55
56         if (!strncmp(dev->name, "plip", 4))
57                 return 2500;
58
59         return 100;     /* assume old 10Mbps */
60 }
61
62
63 /* Check for port carrier transistions. */
64 void br_port_carrier_check(struct net_bridge_port *p)
65 {
66         struct net_device *dev = p->dev;
67         struct net_bridge *br = p->br;
68
69         if (netif_running(dev) && netif_carrier_ok(dev))
70                 p->path_cost = port_cost(dev);
71
72         if (!netif_running(br->dev))
73                 return;
74
75         spin_lock_bh(&br->lock);
76         if (netif_running(dev) && netif_carrier_ok(dev)) {
77                 if (p->state == BR_STATE_DISABLED)
78                         br_stp_enable_port(p);
79         } else {
80                 if (p->state != BR_STATE_DISABLED)
81                         br_stp_disable_port(p);
82         }
83         spin_unlock_bh(&br->lock);
84 }
85
86 static void release_nbp(struct kobject *kobj)
87 {
88         struct net_bridge_port *p
89                 = container_of(kobj, struct net_bridge_port, kobj);
90         kfree(p);
91 }
92
93 static struct kobj_type brport_ktype = {
94 #ifdef CONFIG_SYSFS
95         .sysfs_ops = &brport_sysfs_ops,
96 #endif
97         .release = release_nbp,
98 };
99
100 static void destroy_nbp(struct net_bridge_port *p)
101 {
102         struct net_device *dev = p->dev;
103
104         p->br = NULL;
105         p->dev = NULL;
106         dev_put(dev);
107
108         kobject_put(&p->kobj);
109 }
110
111 static void destroy_nbp_rcu(struct rcu_head *head)
112 {
113         struct net_bridge_port *p =
114                         container_of(head, struct net_bridge_port, rcu);
115         destroy_nbp(p);
116 }
117
118 /* Delete port(interface) from bridge is done in two steps.
119  * via RCU. First step, marks device as down. That deletes
120  * all the timers and stops new packets from flowing through.
121  *
122  * Final cleanup doesn't occur until after all CPU's finished
123  * processing packets.
124  *
125  * Protected from multiple admin operations by RTNL mutex
126  */
127 static void del_nbp(struct net_bridge_port *p)
128 {
129         struct net_bridge *br = p->br;
130         struct net_device *dev = p->dev;
131
132         sysfs_remove_link(br->ifobj, p->dev->name);
133
134         dev_set_promiscuity(dev, -1);
135
136         spin_lock_bh(&br->lock);
137         br_stp_disable_port(p);
138         spin_unlock_bh(&br->lock);
139
140         br_ifinfo_notify(RTM_DELLINK, p);
141
142         br_fdb_delete_by_port(br, p, 1);
143
144         list_del_rcu(&p->list);
145
146         dev->priv_flags &= ~IFF_BRIDGE_PORT;
147
148         netdev_rx_handler_unregister(dev);
149         synchronize_net();
150
151         netdev_set_master(dev, NULL);
152
153         br_multicast_del_port(p);
154
155         kobject_uevent(&p->kobj, KOBJ_REMOVE);
156         kobject_del(&p->kobj);
157
158         br_netpoll_disable(p);
159
160         call_rcu(&p->rcu, destroy_nbp_rcu);
161 }
162
163 /* called with RTNL */
164 static void del_br(struct net_bridge *br, struct list_head *head)
165 {
166         struct net_bridge_port *p, *n;
167
168         list_for_each_entry_safe(p, n, &br->port_list, list) {
169                 del_nbp(p);
170         }
171
172         del_timer_sync(&br->gc_timer);
173
174         br_sysfs_delbr(br->dev);
175         unregister_netdevice_queue(br->dev, head);
176 }
177
178 /* find an available port number */
179 static int find_portno(struct net_bridge *br)
180 {
181         int index;
182         struct net_bridge_port *p;
183         unsigned long *inuse;
184
185         inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
186                         GFP_KERNEL);
187         if (!inuse)
188                 return -ENOMEM;
189
190         set_bit(0, inuse);      /* zero is reserved */
191         list_for_each_entry(p, &br->port_list, list) {
192                 set_bit(p->port_no, inuse);
193         }
194         index = find_first_zero_bit(inuse, BR_MAX_PORTS);
195         kfree(inuse);
196
197         return (index >= BR_MAX_PORTS) ? -EXFULL : index;
198 }
199
200 /* called with RTNL but without bridge lock */
201 static struct net_bridge_port *new_nbp(struct net_bridge *br,
202                                        struct net_device *dev)
203 {
204         int index;
205         struct net_bridge_port *p;
206
207         index = find_portno(br);
208         if (index < 0)
209                 return ERR_PTR(index);
210
211         p = kzalloc(sizeof(*p), GFP_KERNEL);
212         if (p == NULL)
213                 return ERR_PTR(-ENOMEM);
214
215         p->br = br;
216         dev_hold(dev);
217         p->dev = dev;
218         p->path_cost = port_cost(dev);
219         p->priority = 0x8000 >> BR_PORT_BITS;
220         p->port_no = index;
221         p->flags = 0;
222         br_init_port(p);
223         p->state = BR_STATE_DISABLED;
224         br_stp_port_timer_init(p);
225         br_multicast_add_port(p);
226
227         return p;
228 }
229
230 int br_add_bridge(struct net *net, const char *name)
231 {
232         struct net_device *dev;
233         int res;
234
235         dev = alloc_netdev(sizeof(struct net_bridge), name,
236                            br_dev_setup);
237
238         if (!dev)
239                 return -ENOMEM;
240
241         dev_net_set(dev, net);
242
243         res = register_netdev(dev);
244         if (res)
245                 free_netdev(dev);
246         return res;
247 }
248
249 int br_del_bridge(struct net *net, const char *name)
250 {
251         struct net_device *dev;
252         int ret = 0;
253
254         rtnl_lock();
255         dev = __dev_get_by_name(net, name);
256         if (dev == NULL)
257                 ret =  -ENXIO;  /* Could not find device */
258
259         else if (!(dev->priv_flags & IFF_EBRIDGE)) {
260                 /* Attempt to delete non bridge device! */
261                 ret = -EPERM;
262         }
263
264         else if (dev->flags & IFF_UP) {
265                 /* Not shutdown yet. */
266                 ret = -EBUSY;
267         }
268
269         else
270                 del_br(netdev_priv(dev), NULL);
271
272         rtnl_unlock();
273         return ret;
274 }
275
276 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
277 int br_min_mtu(const struct net_bridge *br)
278 {
279         const struct net_bridge_port *p;
280         int mtu = 0;
281
282         ASSERT_RTNL();
283
284         if (list_empty(&br->port_list))
285                 mtu = ETH_DATA_LEN;
286         else {
287                 list_for_each_entry(p, &br->port_list, list) {
288                         if (!mtu  || p->dev->mtu < mtu)
289                                 mtu = p->dev->mtu;
290                 }
291         }
292         return mtu;
293 }
294
295 /*
296  * Recomputes features using slave's features
297  */
298 u32 br_features_recompute(struct net_bridge *br, u32 features)
299 {
300         struct net_bridge_port *p;
301         u32 mask;
302
303         if (list_empty(&br->port_list))
304                 return features;
305
306         mask = features;
307         features &= ~NETIF_F_ONE_FOR_ALL;
308
309         list_for_each_entry(p, &br->port_list, list) {
310                 features = netdev_increment_features(features,
311                                                      p->dev->features, mask);
312         }
313
314         return features;
315 }
316
317 /* called with RTNL */
318 int br_add_if(struct net_bridge *br, struct net_device *dev)
319 {
320         struct net_bridge_port *p;
321         int err = 0;
322         bool changed_addr;
323
324         /* Don't allow bridging non-ethernet like devices */
325         if ((dev->flags & IFF_LOOPBACK) ||
326             dev->type != ARPHRD_ETHER || dev->addr_len != ETH_ALEN ||
327             !is_valid_ether_addr(dev->dev_addr))
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 err1;
356
357         err = br_sysfs_addif(p);
358         if (err)
359                 goto err2;
360
361         if (br_netpoll_info(br) && ((err = br_netpoll_enable(p))))
362                 goto err3;
363
364         err = netdev_set_master(dev, br->dev);
365         if (err)
366                 goto err3;
367
368         err = netdev_rx_handler_register(dev, br_handle_frame, p);
369         if (err)
370                 goto err4;
371
372         dev->priv_flags |= IFF_BRIDGE_PORT;
373
374         dev_disable_lro(dev);
375
376         list_add_rcu(&p->list, &br->port_list);
377
378         netdev_update_features(br->dev);
379
380         spin_lock_bh(&br->lock);
381         changed_addr = br_stp_recalculate_bridge_id(br);
382
383         if ((dev->flags & IFF_UP) && netif_carrier_ok(dev) &&
384             (br->dev->flags & IFF_UP))
385                 br_stp_enable_port(p);
386         spin_unlock_bh(&br->lock);
387
388         br_ifinfo_notify(RTM_NEWLINK, p);
389
390         if (changed_addr)
391                 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
392
393         dev_set_mtu(br->dev, br_min_mtu(br));
394
395         if (br_fdb_insert(br, p, dev->dev_addr))
396                 netdev_err(dev, "failed insert local address bridge forwarding table\n");
397
398         kobject_uevent(&p->kobj, KOBJ_ADD);
399
400         return 0;
401
402 err4:
403         netdev_set_master(dev, NULL);
404 err3:
405         sysfs_remove_link(br->ifobj, p->dev->name);
406 err2:
407         kobject_put(&p->kobj);
408         p = NULL; /* kobject_put frees */
409 err1:
410         dev_set_promiscuity(dev, -1);
411 put_back:
412         dev_put(dev);
413         kfree(p);
414         return err;
415 }
416
417 /* called with RTNL */
418 int br_del_if(struct net_bridge *br, struct net_device *dev)
419 {
420         struct net_bridge_port *p;
421         bool changed_addr;
422
423         p = br_port_get_rtnl(dev);
424         if (!p || p->br != br)
425                 return -EINVAL;
426
427         del_nbp(p);
428
429         spin_lock_bh(&br->lock);
430         changed_addr = br_stp_recalculate_bridge_id(br);
431         spin_unlock_bh(&br->lock);
432
433         if (changed_addr)
434                 call_netdevice_notifiers(NETDEV_CHANGEADDR, br->dev);
435
436         netdev_update_features(br->dev);
437
438         return 0;
439 }
440
441 void __net_exit br_net_exit(struct net *net)
442 {
443         struct net_device *dev;
444         LIST_HEAD(list);
445
446         rtnl_lock();
447         for_each_netdev(net, dev)
448                 if (dev->priv_flags & IFF_EBRIDGE)
449                         del_br(netdev_priv(dev), &list);
450
451         unregister_netdevice_many(&list);
452         rtnl_unlock();
453
454 }