bridge: send correct MTU value in PMTU (revised)
[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/ethtool.h>
17 #include <linux/if_arp.h>
18 #include <linux/module.h>
19 #include <linux/init.h>
20 #include <linux/rtnetlink.h>
21 #include <linux/if_ether.h>
22 #include <net/sock.h>
23
24 #include "br_private.h"
25
26 /*
27  * Determine initial path cost based on speed.
28  * using recommendations from 802.1d standard
29  *
30  * Since driver might sleep need to not be holding any locks.
31  */
32 static int port_cost(struct net_device *dev)
33 {
34         if (dev->ethtool_ops && dev->ethtool_ops->get_settings) {
35                 struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET, };
36
37                 if (!dev->ethtool_ops->get_settings(dev, &ecmd)) {
38                         switch(ecmd.speed) {
39                         case SPEED_10000:
40                                 return 2;
41                         case SPEED_1000:
42                                 return 4;
43                         case SPEED_100:
44                                 return 19;
45                         case SPEED_10:
46                                 return 100;
47                         }
48                 }
49         }
50
51         /* Old silly heuristics based on name */
52         if (!strncmp(dev->name, "lec", 3))
53                 return 7;
54
55         if (!strncmp(dev->name, "plip", 4))
56                 return 2500;
57
58         return 100;     /* assume old 10Mbps */
59 }
60
61
62 /*
63  * Check for port carrier transistions.
64  * Called from work queue to allow for calling functions that
65  * might sleep (such as speed check), and to debounce.
66  */
67 void br_port_carrier_check(struct net_bridge_port *p)
68 {
69         struct net_device *dev = p->dev;
70         struct net_bridge *br = p->br;
71
72         if (netif_carrier_ok(dev))
73                 p->path_cost = port_cost(dev);
74
75         if (netif_running(br->dev)) {
76                 spin_lock_bh(&br->lock);
77                 if (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
88 static void release_nbp(struct kobject *kobj)
89 {
90         struct net_bridge_port *p
91                 = container_of(kobj, struct net_bridge_port, kobj);
92         kfree(p);
93 }
94
95 static struct kobj_type brport_ktype = {
96 #ifdef CONFIG_SYSFS
97         .sysfs_ops = &brport_sysfs_ops,
98 #endif
99         .release = release_nbp,
100 };
101
102 static void destroy_nbp(struct net_bridge_port *p)
103 {
104         struct net_device *dev = p->dev;
105
106         p->br = NULL;
107         p->dev = NULL;
108         dev_put(dev);
109
110         kobject_put(&p->kobj);
111 }
112
113 static void destroy_nbp_rcu(struct rcu_head *head)
114 {
115         struct net_bridge_port *p =
116                         container_of(head, struct net_bridge_port, rcu);
117         destroy_nbp(p);
118 }
119
120 /* Delete port(interface) from bridge is done in two steps.
121  * via RCU. First step, marks device as down. That deletes
122  * all the timers and stops new packets from flowing through.
123  *
124  * Final cleanup doesn't occur until after all CPU's finished
125  * processing packets.
126  *
127  * Protected from multiple admin operations by RTNL mutex
128  */
129 static void del_nbp(struct net_bridge_port *p)
130 {
131         struct net_bridge *br = p->br;
132         struct net_device *dev = p->dev;
133
134         sysfs_remove_link(br->ifobj, dev->name);
135
136         dev_set_promiscuity(dev, -1);
137
138         spin_lock_bh(&br->lock);
139         br_stp_disable_port(p);
140         spin_unlock_bh(&br->lock);
141
142         br_ifinfo_notify(RTM_DELLINK, p);
143
144         br_fdb_delete_by_port(br, p, 1);
145
146         list_del_rcu(&p->list);
147
148         rcu_assign_pointer(dev->br_port, NULL);
149
150         kobject_uevent(&p->kobj, KOBJ_REMOVE);
151         kobject_del(&p->kobj);
152
153         call_rcu(&p->rcu, destroy_nbp_rcu);
154 }
155
156 /* called with RTNL */
157 static void del_br(struct net_bridge *br)
158 {
159         struct net_bridge_port *p, *n;
160
161         list_for_each_entry_safe(p, n, &br->port_list, list) {
162                 del_nbp(p);
163         }
164
165         del_timer_sync(&br->gc_timer);
166
167         br_sysfs_delbr(br->dev);
168         unregister_netdevice(br->dev);
169 }
170
171 static struct net_device *new_bridge_dev(const char *name)
172 {
173         struct net_bridge *br;
174         struct net_device *dev;
175
176         dev = alloc_netdev(sizeof(struct net_bridge), name,
177                            br_dev_setup);
178
179         if (!dev)
180                 return NULL;
181
182         br = netdev_priv(dev);
183         br->dev = dev;
184
185         spin_lock_init(&br->lock);
186         INIT_LIST_HEAD(&br->port_list);
187         spin_lock_init(&br->hash_lock);
188
189         br->bridge_id.prio[0] = 0x80;
190         br->bridge_id.prio[1] = 0x00;
191
192         memcpy(br->group_addr, br_group_address, ETH_ALEN);
193
194         br->feature_mask = dev->features;
195         br->stp_enabled = BR_NO_STP;
196         br->designated_root = br->bridge_id;
197         br->root_path_cost = 0;
198         br->root_port = 0;
199         br->bridge_max_age = br->max_age = 20 * HZ;
200         br->bridge_hello_time = br->hello_time = 2 * HZ;
201         br->bridge_forward_delay = br->forward_delay = 15 * HZ;
202         br->topology_change = 0;
203         br->topology_change_detected = 0;
204         br->ageing_time = 300 * HZ;
205
206         br_netfilter_rtable_init(br);
207
208         INIT_LIST_HEAD(&br->age_list);
209
210         br_stp_timer_init(br);
211
212         return dev;
213 }
214
215 /* find an available port number */
216 static int find_portno(struct net_bridge *br)
217 {
218         int index;
219         struct net_bridge_port *p;
220         unsigned long *inuse;
221
222         inuse = kcalloc(BITS_TO_LONGS(BR_MAX_PORTS), sizeof(unsigned long),
223                         GFP_KERNEL);
224         if (!inuse)
225                 return -ENOMEM;
226
227         set_bit(0, inuse);      /* zero is reserved */
228         list_for_each_entry(p, &br->port_list, list) {
229                 set_bit(p->port_no, inuse);
230         }
231         index = find_first_zero_bit(inuse, BR_MAX_PORTS);
232         kfree(inuse);
233
234         return (index >= BR_MAX_PORTS) ? -EXFULL : index;
235 }
236
237 /* called with RTNL but without bridge lock */
238 static struct net_bridge_port *new_nbp(struct net_bridge *br,
239                                        struct net_device *dev)
240 {
241         int index;
242         struct net_bridge_port *p;
243
244         index = find_portno(br);
245         if (index < 0)
246                 return ERR_PTR(index);
247
248         p = kzalloc(sizeof(*p), GFP_KERNEL);
249         if (p == NULL)
250                 return ERR_PTR(-ENOMEM);
251
252         p->br = br;
253         dev_hold(dev);
254         p->dev = dev;
255         p->path_cost = port_cost(dev);
256         p->priority = 0x8000 >> BR_PORT_BITS;
257         p->port_no = index;
258         br_init_port(p);
259         p->state = BR_STATE_DISABLED;
260         br_stp_port_timer_init(p);
261
262         return p;
263 }
264
265 int br_add_bridge(const char *name)
266 {
267         struct net_device *dev;
268         int ret;
269
270         dev = new_bridge_dev(name);
271         if (!dev)
272                 return -ENOMEM;
273
274         rtnl_lock();
275         if (strchr(dev->name, '%')) {
276                 ret = dev_alloc_name(dev, dev->name);
277                 if (ret < 0)
278                         goto out_free;
279         }
280
281         ret = register_netdevice(dev);
282         if (ret)
283                 goto out_free;
284
285         ret = br_sysfs_addbr(dev);
286         if (ret)
287                 unregister_netdevice(dev);
288  out:
289         rtnl_unlock();
290         return ret;
291
292 out_free:
293         free_netdev(dev);
294         goto out;
295 }
296
297 int br_del_bridge(const char *name)
298 {
299         struct net_device *dev;
300         int ret = 0;
301
302         rtnl_lock();
303         dev = __dev_get_by_name(&init_net, name);
304         if (dev == NULL)
305                 ret =  -ENXIO;  /* Could not find device */
306
307         else if (!(dev->priv_flags & IFF_EBRIDGE)) {
308                 /* Attempt to delete non bridge device! */
309                 ret = -EPERM;
310         }
311
312         else if (dev->flags & IFF_UP) {
313                 /* Not shutdown yet. */
314                 ret = -EBUSY;
315         }
316
317         else
318                 del_br(netdev_priv(dev));
319
320         rtnl_unlock();
321         return ret;
322 }
323
324 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
325 int br_min_mtu(const struct net_bridge *br)
326 {
327         const struct net_bridge_port *p;
328         int mtu = 0;
329
330         ASSERT_RTNL();
331
332         if (list_empty(&br->port_list))
333                 mtu = ETH_DATA_LEN;
334         else {
335                 list_for_each_entry(p, &br->port_list, list) {
336                         if (!mtu  || p->dev->mtu < mtu)
337                                 mtu = p->dev->mtu;
338                 }
339         }
340         return mtu;
341 }
342
343 /*
344  * Recomputes features using slave's features
345  */
346 void br_features_recompute(struct net_bridge *br)
347 {
348         struct net_bridge_port *p;
349         unsigned long features;
350
351         features = br->feature_mask;
352
353         list_for_each_entry(p, &br->port_list, list) {
354                 features = netdev_compute_features(features, p->dev->features);
355         }
356
357         br->dev->features = features;
358 }
359
360 /* called with RTNL */
361 int br_add_if(struct net_bridge *br, struct net_device *dev)
362 {
363         struct net_bridge_port *p;
364         int err = 0;
365
366         if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER)
367                 return -EINVAL;
368
369         if (dev->hard_start_xmit == br_dev_xmit)
370                 return -ELOOP;
371
372         if (dev->br_port != NULL)
373                 return -EBUSY;
374
375         p = new_nbp(br, dev);
376         if (IS_ERR(p))
377                 return PTR_ERR(p);
378
379         err = dev_set_promiscuity(dev, 1);
380         if (err)
381                 goto put_back;
382
383         err = kobject_init_and_add(&p->kobj, &brport_ktype, &(dev->dev.kobj),
384                                    SYSFS_BRIDGE_PORT_ATTR);
385         if (err)
386                 goto err0;
387
388         err = br_fdb_insert(br, p, dev->dev_addr);
389         if (err)
390                 goto err1;
391
392         err = br_sysfs_addif(p);
393         if (err)
394                 goto err2;
395
396         rcu_assign_pointer(dev->br_port, p);
397         dev_disable_lro(dev);
398
399         list_add_rcu(&p->list, &br->port_list);
400
401         spin_lock_bh(&br->lock);
402         br_stp_recalculate_bridge_id(br);
403         br_features_recompute(br);
404
405         if ((dev->flags & IFF_UP) && netif_carrier_ok(dev) &&
406             (br->dev->flags & IFF_UP))
407                 br_stp_enable_port(p);
408         spin_unlock_bh(&br->lock);
409
410         br_ifinfo_notify(RTM_NEWLINK, p);
411
412         dev_set_mtu(br->dev, br_min_mtu(br));
413
414         kobject_uevent(&p->kobj, KOBJ_ADD);
415
416         return 0;
417 err2:
418         br_fdb_delete_by_port(br, p, 1);
419 err1:
420         kobject_del(&p->kobj);
421 err0:
422         kobject_put(&p->kobj);
423         dev_set_promiscuity(dev, -1);
424 put_back:
425         dev_put(dev);
426         kfree(p);
427         return err;
428 }
429
430 /* called with RTNL */
431 int br_del_if(struct net_bridge *br, struct net_device *dev)
432 {
433         struct net_bridge_port *p = dev->br_port;
434
435         if (!p || p->br != br)
436                 return -EINVAL;
437
438         del_nbp(p);
439
440         spin_lock_bh(&br->lock);
441         br_stp_recalculate_bridge_id(br);
442         br_features_recompute(br);
443         spin_unlock_bh(&br->lock);
444
445         return 0;
446 }
447
448 void __exit br_cleanup_bridges(void)
449 {
450         struct net_device *dev;
451
452         rtnl_lock();
453 restart:
454         for_each_netdev(&init_net, dev) {
455                 if (dev->priv_flags & IFF_EBRIDGE) {
456                         del_br(dev->priv);
457                         goto restart;
458                 }
459         }
460         rtnl_unlock();
461
462 }