]>
Commit | Line | Data |
---|---|---|
5f256bec EB |
1 | #include <linux/workqueue.h> |
2 | #include <linux/rtnetlink.h> | |
3 | #include <linux/cache.h> | |
4 | #include <linux/slab.h> | |
5 | #include <linux/list.h> | |
6 | #include <linux/delay.h> | |
9dd776b6 | 7 | #include <linux/sched.h> |
c93cf61f | 8 | #include <linux/idr.h> |
5f256bec | 9 | #include <net/net_namespace.h> |
dec827d1 | 10 | #include <net/netns/generic.h> |
5f256bec EB |
11 | |
12 | /* | |
13 | * Our network namespace constructor/destructor lists | |
14 | */ | |
15 | ||
16 | static LIST_HEAD(pernet_list); | |
17 | static struct list_head *first_device = &pernet_list; | |
18 | static DEFINE_MUTEX(net_mutex); | |
19 | ||
5f256bec | 20 | LIST_HEAD(net_namespace_list); |
b76a461f | 21 | EXPORT_SYMBOL_GPL(net_namespace_list); |
5f256bec | 22 | |
5f256bec | 23 | struct net init_net; |
ff4b9502 | 24 | EXPORT_SYMBOL(init_net); |
5f256bec | 25 | |
dec827d1 PE |
26 | #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */ |
27 | ||
5f256bec EB |
28 | /* |
29 | * setup_net runs the initializers for the network namespace object. | |
30 | */ | |
1a2ee93d | 31 | static __net_init int setup_net(struct net *net) |
5f256bec EB |
32 | { |
33 | /* Must be called with net_mutex held */ | |
34 | struct pernet_operations *ops; | |
5f256bec | 35 | int error; |
dec827d1 | 36 | struct net_generic *ng; |
5f256bec | 37 | |
5f256bec | 38 | atomic_set(&net->count, 1); |
5d1e4468 | 39 | #ifdef NETNS_REFCNT_DEBUG |
5f256bec | 40 | atomic_set(&net->use_count, 0); |
5d1e4468 | 41 | #endif |
5f256bec | 42 | |
dec827d1 PE |
43 | error = -ENOMEM; |
44 | ng = kzalloc(sizeof(struct net_generic) + | |
45 | INITIAL_NET_GEN_PTRS * sizeof(void *), GFP_KERNEL); | |
46 | if (ng == NULL) | |
47 | goto out; | |
48 | ||
49 | ng->len = INITIAL_NET_GEN_PTRS; | |
50 | INIT_RCU_HEAD(&ng->rcu); | |
51 | rcu_assign_pointer(net->gen, ng); | |
52 | ||
5f256bec | 53 | error = 0; |
768f3591 | 54 | list_for_each_entry(ops, &pernet_list, list) { |
5f256bec EB |
55 | if (ops->init) { |
56 | error = ops->init(net); | |
57 | if (error < 0) | |
58 | goto out_undo; | |
59 | } | |
60 | } | |
61 | out: | |
62 | return error; | |
768f3591 | 63 | |
5f256bec EB |
64 | out_undo: |
65 | /* Walk through the list backwards calling the exit functions | |
66 | * for the pernet modules whose init functions did not fail. | |
67 | */ | |
768f3591 | 68 | list_for_each_entry_continue_reverse(ops, &pernet_list, list) { |
5f256bec EB |
69 | if (ops->exit) |
70 | ops->exit(net); | |
71 | } | |
310928d9 DL |
72 | |
73 | rcu_barrier(); | |
dec827d1 | 74 | kfree(ng); |
5f256bec EB |
75 | goto out; |
76 | } | |
77 | ||
6a1a3b9f | 78 | #ifdef CONFIG_NET_NS |
d57a9212 | 79 | static struct kmem_cache *net_cachep; |
3ef1355d | 80 | static struct workqueue_struct *netns_wq; |
d57a9212 | 81 | |
6a1a3b9f PE |
82 | static struct net *net_alloc(void) |
83 | { | |
84 | return kmem_cache_zalloc(net_cachep, GFP_KERNEL); | |
85 | } | |
86 | ||
45a19b0a JFS |
87 | static void net_free(struct net *net) |
88 | { | |
89 | if (!net) | |
90 | return; | |
91 | ||
5d1e4468 | 92 | #ifdef NETNS_REFCNT_DEBUG |
45a19b0a JFS |
93 | if (unlikely(atomic_read(&net->use_count) != 0)) { |
94 | printk(KERN_EMERG "network namespace not free! Usage: %d\n", | |
95 | atomic_read(&net->use_count)); | |
96 | return; | |
97 | } | |
5d1e4468 | 98 | #endif |
45a19b0a JFS |
99 | |
100 | kmem_cache_free(net_cachep, net); | |
101 | } | |
102 | ||
9dd776b6 EB |
103 | struct net *copy_net_ns(unsigned long flags, struct net *old_net) |
104 | { | |
105 | struct net *new_net = NULL; | |
106 | int err; | |
107 | ||
108 | get_net(old_net); | |
109 | ||
110 | if (!(flags & CLONE_NEWNET)) | |
111 | return old_net; | |
112 | ||
9dd776b6 EB |
113 | err = -ENOMEM; |
114 | new_net = net_alloc(); | |
115 | if (!new_net) | |
116 | goto out; | |
117 | ||
118 | mutex_lock(&net_mutex); | |
119 | err = setup_net(new_net); | |
120 | if (err) | |
121 | goto out_unlock; | |
122 | ||
f4618d39 | 123 | rtnl_lock(); |
9dd776b6 | 124 | list_add_tail(&new_net->list, &net_namespace_list); |
f4618d39 | 125 | rtnl_unlock(); |
9dd776b6 EB |
126 | |
127 | ||
128 | out_unlock: | |
129 | mutex_unlock(&net_mutex); | |
130 | out: | |
131 | put_net(old_net); | |
132 | if (err) { | |
133 | net_free(new_net); | |
134 | new_net = ERR_PTR(err); | |
135 | } | |
136 | return new_net; | |
137 | } | |
138 | ||
6a1a3b9f PE |
139 | static void cleanup_net(struct work_struct *work) |
140 | { | |
141 | struct pernet_operations *ops; | |
142 | struct net *net; | |
143 | ||
b9f75f45 EB |
144 | /* Be very certain incoming network packets will not find us */ |
145 | rcu_barrier(); | |
146 | ||
6a1a3b9f PE |
147 | net = container_of(work, struct net, work); |
148 | ||
149 | mutex_lock(&net_mutex); | |
150 | ||
151 | /* Don't let anyone else find us. */ | |
152 | rtnl_lock(); | |
153 | list_del(&net->list); | |
154 | rtnl_unlock(); | |
155 | ||
156 | /* Run all of the network namespace exit methods */ | |
157 | list_for_each_entry_reverse(ops, &pernet_list, list) { | |
158 | if (ops->exit) | |
159 | ops->exit(net); | |
160 | } | |
161 | ||
162 | mutex_unlock(&net_mutex); | |
163 | ||
164 | /* Ensure there are no outstanding rcu callbacks using this | |
165 | * network namespace. | |
166 | */ | |
167 | rcu_barrier(); | |
168 | ||
169 | /* Finally it is safe to free my network namespace structure */ | |
170 | net_free(net); | |
171 | } | |
172 | ||
173 | void __put_net(struct net *net) | |
174 | { | |
175 | /* Cleanup the network namespace in process context */ | |
176 | INIT_WORK(&net->work, cleanup_net); | |
3ef1355d | 177 | queue_work(netns_wq, &net->work); |
6a1a3b9f PE |
178 | } |
179 | EXPORT_SYMBOL_GPL(__put_net); | |
180 | ||
181 | #else | |
182 | struct net *copy_net_ns(unsigned long flags, struct net *old_net) | |
183 | { | |
184 | if (flags & CLONE_NEWNET) | |
185 | return ERR_PTR(-EINVAL); | |
186 | return old_net; | |
187 | } | |
188 | #endif | |
189 | ||
5f256bec EB |
190 | static int __init net_ns_init(void) |
191 | { | |
192 | int err; | |
193 | ||
194 | printk(KERN_INFO "net_namespace: %zd bytes\n", sizeof(struct net)); | |
d57a9212 | 195 | #ifdef CONFIG_NET_NS |
5f256bec EB |
196 | net_cachep = kmem_cache_create("net_namespace", sizeof(struct net), |
197 | SMP_CACHE_BYTES, | |
198 | SLAB_PANIC, NULL); | |
3ef1355d BT |
199 | |
200 | /* Create workqueue for cleanup */ | |
201 | netns_wq = create_singlethread_workqueue("netns"); | |
202 | if (!netns_wq) | |
203 | panic("Could not create netns workq"); | |
d57a9212 | 204 | #endif |
3ef1355d | 205 | |
5f256bec EB |
206 | mutex_lock(&net_mutex); |
207 | err = setup_net(&init_net); | |
208 | ||
f4618d39 | 209 | rtnl_lock(); |
5f256bec | 210 | list_add_tail(&init_net.list, &net_namespace_list); |
f4618d39 | 211 | rtnl_unlock(); |
5f256bec EB |
212 | |
213 | mutex_unlock(&net_mutex); | |
214 | if (err) | |
215 | panic("Could not setup the initial network namespace"); | |
216 | ||
217 | return 0; | |
218 | } | |
219 | ||
220 | pure_initcall(net_ns_init); | |
221 | ||
ed160e83 | 222 | #ifdef CONFIG_NET_NS |
5f256bec EB |
223 | static int register_pernet_operations(struct list_head *list, |
224 | struct pernet_operations *ops) | |
225 | { | |
226 | struct net *net, *undo_net; | |
227 | int error; | |
228 | ||
5f256bec | 229 | list_add_tail(&ops->list, list); |
1dba323b PE |
230 | if (ops->init) { |
231 | for_each_net(net) { | |
5f256bec EB |
232 | error = ops->init(net); |
233 | if (error) | |
234 | goto out_undo; | |
235 | } | |
236 | } | |
1dba323b | 237 | return 0; |
5f256bec EB |
238 | |
239 | out_undo: | |
240 | /* If I have an error cleanup all namespaces I initialized */ | |
241 | list_del(&ops->list); | |
1dba323b PE |
242 | if (ops->exit) { |
243 | for_each_net(undo_net) { | |
244 | if (undo_net == net) | |
245 | goto undone; | |
5f256bec | 246 | ops->exit(undo_net); |
1dba323b | 247 | } |
5f256bec EB |
248 | } |
249 | undone: | |
1dba323b | 250 | return error; |
5f256bec EB |
251 | } |
252 | ||
253 | static void unregister_pernet_operations(struct pernet_operations *ops) | |
254 | { | |
255 | struct net *net; | |
256 | ||
257 | list_del(&ops->list); | |
1dba323b PE |
258 | if (ops->exit) |
259 | for_each_net(net) | |
5f256bec EB |
260 | ops->exit(net); |
261 | } | |
262 | ||
ed160e83 DL |
263 | #else |
264 | ||
265 | static int register_pernet_operations(struct list_head *list, | |
266 | struct pernet_operations *ops) | |
267 | { | |
268 | if (ops->init == NULL) | |
269 | return 0; | |
270 | return ops->init(&init_net); | |
271 | } | |
272 | ||
273 | static void unregister_pernet_operations(struct pernet_operations *ops) | |
274 | { | |
275 | if (ops->exit) | |
276 | ops->exit(&init_net); | |
277 | } | |
278 | #endif | |
279 | ||
c93cf61f PE |
280 | static DEFINE_IDA(net_generic_ids); |
281 | ||
5f256bec EB |
282 | /** |
283 | * register_pernet_subsys - register a network namespace subsystem | |
284 | * @ops: pernet operations structure for the subsystem | |
285 | * | |
286 | * Register a subsystem which has init and exit functions | |
287 | * that are called when network namespaces are created and | |
288 | * destroyed respectively. | |
289 | * | |
290 | * When registered all network namespace init functions are | |
291 | * called for every existing network namespace. Allowing kernel | |
292 | * modules to have a race free view of the set of network namespaces. | |
293 | * | |
294 | * When a new network namespace is created all of the init | |
295 | * methods are called in the order in which they were registered. | |
296 | * | |
297 | * When a network namespace is destroyed all of the exit methods | |
298 | * are called in the reverse of the order with which they were | |
299 | * registered. | |
300 | */ | |
301 | int register_pernet_subsys(struct pernet_operations *ops) | |
302 | { | |
303 | int error; | |
304 | mutex_lock(&net_mutex); | |
305 | error = register_pernet_operations(first_device, ops); | |
306 | mutex_unlock(&net_mutex); | |
307 | return error; | |
308 | } | |
309 | EXPORT_SYMBOL_GPL(register_pernet_subsys); | |
310 | ||
311 | /** | |
312 | * unregister_pernet_subsys - unregister a network namespace subsystem | |
313 | * @ops: pernet operations structure to manipulate | |
314 | * | |
315 | * Remove the pernet operations structure from the list to be | |
53379e57 | 316 | * used when network namespaces are created or destroyed. In |
5f256bec EB |
317 | * addition run the exit method for all existing network |
318 | * namespaces. | |
319 | */ | |
320 | void unregister_pernet_subsys(struct pernet_operations *module) | |
321 | { | |
322 | mutex_lock(&net_mutex); | |
323 | unregister_pernet_operations(module); | |
324 | mutex_unlock(&net_mutex); | |
325 | } | |
326 | EXPORT_SYMBOL_GPL(unregister_pernet_subsys); | |
327 | ||
328 | /** | |
329 | * register_pernet_device - register a network namespace device | |
330 | * @ops: pernet operations structure for the subsystem | |
331 | * | |
332 | * Register a device which has init and exit functions | |
333 | * that are called when network namespaces are created and | |
334 | * destroyed respectively. | |
335 | * | |
336 | * When registered all network namespace init functions are | |
337 | * called for every existing network namespace. Allowing kernel | |
338 | * modules to have a race free view of the set of network namespaces. | |
339 | * | |
340 | * When a new network namespace is created all of the init | |
341 | * methods are called in the order in which they were registered. | |
342 | * | |
343 | * When a network namespace is destroyed all of the exit methods | |
344 | * are called in the reverse of the order with which they were | |
345 | * registered. | |
346 | */ | |
347 | int register_pernet_device(struct pernet_operations *ops) | |
348 | { | |
349 | int error; | |
350 | mutex_lock(&net_mutex); | |
351 | error = register_pernet_operations(&pernet_list, ops); | |
352 | if (!error && (first_device == &pernet_list)) | |
353 | first_device = &ops->list; | |
354 | mutex_unlock(&net_mutex); | |
355 | return error; | |
356 | } | |
357 | EXPORT_SYMBOL_GPL(register_pernet_device); | |
358 | ||
c93cf61f PE |
359 | int register_pernet_gen_device(int *id, struct pernet_operations *ops) |
360 | { | |
361 | int error; | |
362 | mutex_lock(&net_mutex); | |
363 | again: | |
364 | error = ida_get_new_above(&net_generic_ids, 1, id); | |
365 | if (error) { | |
366 | if (error == -EAGAIN) { | |
367 | ida_pre_get(&net_generic_ids, GFP_KERNEL); | |
368 | goto again; | |
369 | } | |
370 | goto out; | |
371 | } | |
372 | error = register_pernet_operations(&pernet_list, ops); | |
373 | if (error) | |
374 | ida_remove(&net_generic_ids, *id); | |
375 | else if (first_device == &pernet_list) | |
376 | first_device = &ops->list; | |
377 | out: | |
378 | mutex_unlock(&net_mutex); | |
379 | return error; | |
380 | } | |
381 | EXPORT_SYMBOL_GPL(register_pernet_gen_device); | |
382 | ||
5f256bec EB |
383 | /** |
384 | * unregister_pernet_device - unregister a network namespace netdevice | |
385 | * @ops: pernet operations structure to manipulate | |
386 | * | |
387 | * Remove the pernet operations structure from the list to be | |
53379e57 | 388 | * used when network namespaces are created or destroyed. In |
5f256bec EB |
389 | * addition run the exit method for all existing network |
390 | * namespaces. | |
391 | */ | |
392 | void unregister_pernet_device(struct pernet_operations *ops) | |
393 | { | |
394 | mutex_lock(&net_mutex); | |
395 | if (&ops->list == first_device) | |
396 | first_device = first_device->next; | |
397 | unregister_pernet_operations(ops); | |
398 | mutex_unlock(&net_mutex); | |
399 | } | |
400 | EXPORT_SYMBOL_GPL(unregister_pernet_device); | |
c93cf61f PE |
401 | |
402 | void unregister_pernet_gen_device(int id, struct pernet_operations *ops) | |
403 | { | |
404 | mutex_lock(&net_mutex); | |
405 | if (&ops->list == first_device) | |
406 | first_device = first_device->next; | |
407 | unregister_pernet_operations(ops); | |
408 | ida_remove(&net_generic_ids, id); | |
409 | mutex_unlock(&net_mutex); | |
410 | } | |
411 | EXPORT_SYMBOL_GPL(unregister_pernet_gen_device); | |
dec827d1 PE |
412 | |
413 | static void net_generic_release(struct rcu_head *rcu) | |
414 | { | |
415 | struct net_generic *ng; | |
416 | ||
417 | ng = container_of(rcu, struct net_generic, rcu); | |
418 | kfree(ng); | |
419 | } | |
420 | ||
421 | int net_assign_generic(struct net *net, int id, void *data) | |
422 | { | |
423 | struct net_generic *ng, *old_ng; | |
424 | ||
425 | BUG_ON(!mutex_is_locked(&net_mutex)); | |
426 | BUG_ON(id == 0); | |
427 | ||
428 | ng = old_ng = net->gen; | |
429 | if (old_ng->len >= id) | |
430 | goto assign; | |
431 | ||
432 | ng = kzalloc(sizeof(struct net_generic) + | |
433 | id * sizeof(void *), GFP_KERNEL); | |
434 | if (ng == NULL) | |
435 | return -ENOMEM; | |
436 | ||
437 | /* | |
438 | * Some synchronisation notes: | |
439 | * | |
440 | * The net_generic explores the net->gen array inside rcu | |
441 | * read section. Besides once set the net->gen->ptr[x] | |
442 | * pointer never changes (see rules in netns/generic.h). | |
443 | * | |
444 | * That said, we simply duplicate this array and schedule | |
445 | * the old copy for kfree after a grace period. | |
446 | */ | |
447 | ||
448 | ng->len = id; | |
449 | INIT_RCU_HEAD(&ng->rcu); | |
450 | memcpy(&ng->ptr, &old_ng->ptr, old_ng->len); | |
451 | ||
452 | rcu_assign_pointer(net->gen, ng); | |
453 | call_rcu(&old_ng->rcu, net_generic_release); | |
454 | assign: | |
455 | ng->ptr[id - 1] = data; | |
456 | return 0; | |
457 | } | |
458 | EXPORT_SYMBOL_GPL(net_assign_generic); |