[BRIDGE] netfilter: vlan + hw checksum = bug?
[safe/jmp/linux-2.6] / 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_bridge_port *p = arg;
83
84         rtnl_lock();
85         if (netif_carrier_ok(p->dev)) {
86                 u32 cost = port_cost(p->dev);
87
88                 spin_lock_bh(&p->br->lock);
89                 if (p->state == BR_STATE_DISABLED) {
90                         p->path_cost = cost;
91                         br_stp_enable_port(p);
92                 }
93                 spin_unlock_bh(&p->br->lock);
94         } else {
95                 spin_lock_bh(&p->br->lock);
96                 if (p->state != BR_STATE_DISABLED)
97                         br_stp_disable_port(p);
98                 spin_unlock_bh(&p->br->lock);
99         }
100         rtnl_unlock();
101 }
102
103 static void destroy_nbp(struct net_bridge_port *p)
104 {
105         struct net_device *dev = p->dev;
106
107         p->br = NULL;
108         p->dev = NULL;
109         dev_put(dev);
110
111         br_sysfs_freeif(p);
112 }
113
114 static void destroy_nbp_rcu(struct rcu_head *head)
115 {
116         struct net_bridge_port *p =
117                         container_of(head, struct net_bridge_port, rcu);
118         destroy_nbp(p);
119 }
120
121 /* called with RTNL */
122 static void del_nbp(struct net_bridge_port *p)
123 {
124         struct net_bridge *br = p->br;
125         struct net_device *dev = p->dev;
126
127         dev->br_port = NULL;
128         dev_set_promiscuity(dev, -1);
129
130         cancel_delayed_work(&p->carrier_check);
131         flush_scheduled_work();
132
133         spin_lock_bh(&br->lock);
134         br_stp_disable_port(p);
135         spin_unlock_bh(&br->lock);
136
137         br_fdb_delete_by_port(br, p);
138
139         list_del_rcu(&p->list);
140
141         del_timer_sync(&p->message_age_timer);
142         del_timer_sync(&p->forward_delay_timer);
143         del_timer_sync(&p->hold_timer);
144         
145         call_rcu(&p->rcu, destroy_nbp_rcu);
146 }
147
148 /* called with RTNL */
149 static void del_br(struct net_bridge *br)
150 {
151         struct net_bridge_port *p, *n;
152
153         list_for_each_entry_safe(p, n, &br->port_list, list) {
154                 br_sysfs_removeif(p);
155                 del_nbp(p);
156         }
157
158         del_timer_sync(&br->gc_timer);
159
160         br_sysfs_delbr(br->dev);
161         unregister_netdevice(br->dev);
162 }
163
164 static struct net_device *new_bridge_dev(const char *name)
165 {
166         struct net_bridge *br;
167         struct net_device *dev;
168
169         dev = alloc_netdev(sizeof(struct net_bridge), name,
170                            br_dev_setup);
171         
172         if (!dev)
173                 return NULL;
174
175         br = netdev_priv(dev);
176         br->dev = dev;
177
178         spin_lock_init(&br->lock);
179         INIT_LIST_HEAD(&br->port_list);
180         spin_lock_init(&br->hash_lock);
181
182         br->bridge_id.prio[0] = 0x80;
183         br->bridge_id.prio[1] = 0x00;
184         memset(br->bridge_id.addr, 0, ETH_ALEN);
185
186         br->feature_mask = dev->features;
187         br->stp_enabled = 0;
188         br->designated_root = br->bridge_id;
189         br->root_path_cost = 0;
190         br->root_port = 0;
191         br->bridge_max_age = br->max_age = 20 * HZ;
192         br->bridge_hello_time = br->hello_time = 2 * HZ;
193         br->bridge_forward_delay = br->forward_delay = 15 * HZ;
194         br->topology_change = 0;
195         br->topology_change_detected = 0;
196         br->ageing_time = 300 * HZ;
197         INIT_LIST_HEAD(&br->age_list);
198
199         br_stp_timer_init(br);
200
201         return dev;
202 }
203
204 /* find an available port number */
205 static int find_portno(struct net_bridge *br)
206 {
207         int index;
208         struct net_bridge_port *p;
209         unsigned long *inuse;
210
211         inuse = kmalloc(BITS_TO_LONGS(BR_MAX_PORTS)*sizeof(unsigned long),
212                         GFP_KERNEL);
213         if (!inuse)
214                 return -ENOMEM;
215
216         memset(inuse, 0, BITS_TO_LONGS(BR_MAX_PORTS)*sizeof(unsigned long));
217         set_bit(0, inuse);      /* zero is reserved */
218         list_for_each_entry(p, &br->port_list, list) {
219                 set_bit(p->port_no, inuse);
220         }
221         index = find_first_zero_bit(inuse, BR_MAX_PORTS);
222         kfree(inuse);
223
224         return (index >= BR_MAX_PORTS) ? -EXFULL : index;
225 }
226
227 /* called with RTNL but without bridge lock */
228 static struct net_bridge_port *new_nbp(struct net_bridge *br, 
229                                        struct net_device *dev)
230 {
231         int index;
232         struct net_bridge_port *p;
233         
234         index = find_portno(br);
235         if (index < 0)
236                 return ERR_PTR(index);
237
238         p = kmalloc(sizeof(*p), GFP_KERNEL);
239         if (p == NULL)
240                 return ERR_PTR(-ENOMEM);
241
242         memset(p, 0, sizeof(*p));
243         p->br = br;
244         dev_hold(dev);
245         p->dev = dev;
246         p->path_cost = port_cost(dev);
247         p->priority = 0x8000 >> BR_PORT_BITS;
248         dev->br_port = p;
249         p->port_no = index;
250         br_init_port(p);
251         p->state = BR_STATE_DISABLED;
252         INIT_WORK(&p->carrier_check, port_carrier_check, p);
253         kobject_init(&p->kobj);
254
255         return p;
256 }
257
258 int br_add_bridge(const char *name)
259 {
260         struct net_device *dev;
261         int ret;
262
263         dev = new_bridge_dev(name);
264         if (!dev) 
265                 return -ENOMEM;
266
267         rtnl_lock();
268         if (strchr(dev->name, '%')) {
269                 ret = dev_alloc_name(dev, dev->name);
270                 if (ret < 0)
271                         goto err1;
272         }
273
274         ret = register_netdevice(dev);
275         if (ret)
276                 goto err2;
277
278         /* network device kobject is not setup until
279          * after rtnl_unlock does it's hotplug magic.
280          * so hold reference to avoid race.
281          */
282         dev_hold(dev);
283         rtnl_unlock();
284
285         ret = br_sysfs_addbr(dev);
286         dev_put(dev);
287
288         if (ret) 
289                 unregister_netdev(dev);
290  out:
291         return ret;
292
293  err2:
294         free_netdev(dev);
295  err1:
296         rtnl_unlock();
297         goto out;
298 }
299
300 int br_del_bridge(const char *name)
301 {
302         struct net_device *dev;
303         int ret = 0;
304
305         rtnl_lock();
306         dev = __dev_get_by_name(name);
307         if (dev == NULL) 
308                 ret =  -ENXIO;  /* Could not find device */
309
310         else if (!(dev->priv_flags & IFF_EBRIDGE)) {
311                 /* Attempt to delete non bridge device! */
312                 ret = -EPERM;
313         }
314
315         else if (dev->flags & IFF_UP) {
316                 /* Not shutdown yet. */
317                 ret = -EBUSY;
318         } 
319
320         else 
321                 del_br(netdev_priv(dev));
322
323         rtnl_unlock();
324         return ret;
325 }
326
327 /* MTU of the bridge pseudo-device: ETH_DATA_LEN or the minimum of the ports */
328 int br_min_mtu(const struct net_bridge *br)
329 {
330         const struct net_bridge_port *p;
331         int mtu = 0;
332
333         ASSERT_RTNL();
334
335         if (list_empty(&br->port_list))
336                 mtu = ETH_DATA_LEN;
337         else {
338                 list_for_each_entry(p, &br->port_list, list) {
339                         if (!mtu  || p->dev->mtu < mtu)
340                                 mtu = p->dev->mtu;
341                 }
342         }
343         return mtu;
344 }
345
346 /*
347  * Recomputes features using slave's features
348  */
349 void br_features_recompute(struct net_bridge *br)
350 {
351         struct net_bridge_port *p;
352         unsigned long features, checksum;
353
354         features = br->feature_mask &~ NETIF_F_IP_CSUM;
355         checksum = br->feature_mask & NETIF_F_IP_CSUM;
356
357         list_for_each_entry(p, &br->port_list, list) {
358                 if (!(p->dev->features 
359                       & (NETIF_F_IP_CSUM|NETIF_F_NO_CSUM|NETIF_F_HW_CSUM)))
360                         checksum = 0;
361                 features &= p->dev->features;
362         }
363
364         br->dev->features = features | checksum | NETIF_F_LLTX;
365 }
366
367 /* called with RTNL */
368 int br_add_if(struct net_bridge *br, struct net_device *dev)
369 {
370         struct net_bridge_port *p;
371         int err = 0;
372
373         if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER)
374                 return -EINVAL;
375
376         if (dev->hard_start_xmit == br_dev_xmit)
377                 return -ELOOP;
378
379         if (dev->br_port != NULL)
380                 return -EBUSY;
381
382         if (IS_ERR(p = new_nbp(br, dev)))
383                 return PTR_ERR(p);
384
385         if ((err = br_fdb_insert(br, p, dev->dev_addr)))
386                 destroy_nbp(p);
387  
388         else if ((err = br_sysfs_addif(p)))
389                 del_nbp(p);
390         else {
391                 dev_set_promiscuity(dev, 1);
392
393                 list_add_rcu(&p->list, &br->port_list);
394
395                 spin_lock_bh(&br->lock);
396                 br_stp_recalculate_bridge_id(br);
397                 br_features_recompute(br);
398                 if ((br->dev->flags & IFF_UP) 
399                     && (dev->flags & IFF_UP) && netif_carrier_ok(dev))
400                         br_stp_enable_port(p);
401                 spin_unlock_bh(&br->lock);
402
403                 dev_set_mtu(br->dev, br_min_mtu(br));
404         }
405
406         return err;
407 }
408
409 /* called with RTNL */
410 int br_del_if(struct net_bridge *br, struct net_device *dev)
411 {
412         struct net_bridge_port *p = dev->br_port;
413         
414         if (!p || p->br != br) 
415                 return -EINVAL;
416
417         br_sysfs_removeif(p);
418         del_nbp(p);
419
420         spin_lock_bh(&br->lock);
421         br_stp_recalculate_bridge_id(br);
422         br_features_recompute(br);
423         spin_unlock_bh(&br->lock);
424
425         return 0;
426 }
427
428 void __exit br_cleanup_bridges(void)
429 {
430         struct net_device *dev, *nxt;
431
432         rtnl_lock();
433         for (dev = dev_base; dev; dev = nxt) {
434                 nxt = dev->next;
435                 if (dev->priv_flags & IFF_EBRIDGE)
436                         del_br(dev->priv);
437         }
438         rtnl_unlock();
439
440 }