]>
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> |
11a28d37 | 9 | #include <linux/rculist.h> |
5f256bec | 10 | #include <net/net_namespace.h> |
dec827d1 | 11 | #include <net/netns/generic.h> |
5f256bec EB |
12 | |
13 | /* | |
14 | * Our network namespace constructor/destructor lists | |
15 | */ | |
16 | ||
17 | static LIST_HEAD(pernet_list); | |
18 | static struct list_head *first_device = &pernet_list; | |
19 | static DEFINE_MUTEX(net_mutex); | |
20 | ||
5f256bec | 21 | LIST_HEAD(net_namespace_list); |
b76a461f | 22 | EXPORT_SYMBOL_GPL(net_namespace_list); |
5f256bec | 23 | |
5f256bec | 24 | struct net init_net; |
ff4b9502 | 25 | EXPORT_SYMBOL(init_net); |
5f256bec | 26 | |
dec827d1 PE |
27 | #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */ |
28 | ||
5f256bec EB |
29 | /* |
30 | * setup_net runs the initializers for the network namespace object. | |
31 | */ | |
1a2ee93d | 32 | static __net_init int setup_net(struct net *net) |
5f256bec EB |
33 | { |
34 | /* Must be called with net_mutex held */ | |
35 | struct pernet_operations *ops; | |
486a87f1 | 36 | int error = 0; |
5f256bec | 37 | |
5f256bec | 38 | atomic_set(&net->count, 1); |
486a87f1 | 39 | |
5d1e4468 | 40 | #ifdef NETNS_REFCNT_DEBUG |
5f256bec | 41 | atomic_set(&net->use_count, 0); |
5d1e4468 | 42 | #endif |
5f256bec | 43 | |
768f3591 | 44 | list_for_each_entry(ops, &pernet_list, list) { |
5f256bec EB |
45 | if (ops->init) { |
46 | error = ops->init(net); | |
47 | if (error < 0) | |
48 | goto out_undo; | |
49 | } | |
50 | } | |
51 | out: | |
52 | return error; | |
768f3591 | 53 | |
5f256bec EB |
54 | out_undo: |
55 | /* Walk through the list backwards calling the exit functions | |
56 | * for the pernet modules whose init functions did not fail. | |
57 | */ | |
768f3591 | 58 | list_for_each_entry_continue_reverse(ops, &pernet_list, list) { |
5f256bec EB |
59 | if (ops->exit) |
60 | ops->exit(net); | |
61 | } | |
310928d9 DL |
62 | |
63 | rcu_barrier(); | |
5f256bec EB |
64 | goto out; |
65 | } | |
66 | ||
486a87f1 | 67 | static struct net_generic *net_alloc_generic(void) |
6a1a3b9f | 68 | { |
486a87f1 DL |
69 | struct net_generic *ng; |
70 | size_t generic_size = sizeof(struct net_generic) + | |
71 | INITIAL_NET_GEN_PTRS * sizeof(void *); | |
72 | ||
73 | ng = kzalloc(generic_size, GFP_KERNEL); | |
74 | if (ng) | |
75 | ng->len = INITIAL_NET_GEN_PTRS; | |
76 | ||
77 | return ng; | |
6a1a3b9f PE |
78 | } |
79 | ||
ebe47d47 CN |
80 | #ifdef CONFIG_NET_NS |
81 | static struct kmem_cache *net_cachep; | |
82 | static struct workqueue_struct *netns_wq; | |
83 | ||
486a87f1 | 84 | static struct net *net_alloc(void) |
45a19b0a | 85 | { |
486a87f1 DL |
86 | struct net *net = NULL; |
87 | struct net_generic *ng; | |
88 | ||
89 | ng = net_alloc_generic(); | |
90 | if (!ng) | |
91 | goto out; | |
92 | ||
93 | net = kmem_cache_zalloc(net_cachep, GFP_KERNEL); | |
45a19b0a | 94 | if (!net) |
486a87f1 | 95 | goto out_free; |
45a19b0a | 96 | |
486a87f1 DL |
97 | rcu_assign_pointer(net->gen, ng); |
98 | out: | |
99 | return net; | |
100 | ||
101 | out_free: | |
102 | kfree(ng); | |
103 | goto out; | |
104 | } | |
105 | ||
106 | static void net_free(struct net *net) | |
107 | { | |
5d1e4468 | 108 | #ifdef NETNS_REFCNT_DEBUG |
45a19b0a JFS |
109 | if (unlikely(atomic_read(&net->use_count) != 0)) { |
110 | printk(KERN_EMERG "network namespace not free! Usage: %d\n", | |
111 | atomic_read(&net->use_count)); | |
112 | return; | |
113 | } | |
5d1e4468 | 114 | #endif |
4ef079cc | 115 | kfree(net->gen); |
45a19b0a JFS |
116 | kmem_cache_free(net_cachep, net); |
117 | } | |
118 | ||
088eb2d9 | 119 | static struct net *net_create(void) |
9dd776b6 | 120 | { |
088eb2d9 AD |
121 | struct net *net; |
122 | int rv; | |
9dd776b6 | 123 | |
088eb2d9 AD |
124 | net = net_alloc(); |
125 | if (!net) | |
126 | return ERR_PTR(-ENOMEM); | |
9dd776b6 | 127 | mutex_lock(&net_mutex); |
088eb2d9 AD |
128 | rv = setup_net(net); |
129 | if (rv == 0) { | |
486a87f1 | 130 | rtnl_lock(); |
11a28d37 | 131 | list_add_tail_rcu(&net->list, &net_namespace_list); |
486a87f1 DL |
132 | rtnl_unlock(); |
133 | } | |
9dd776b6 | 134 | mutex_unlock(&net_mutex); |
088eb2d9 AD |
135 | if (rv < 0) { |
136 | net_free(net); | |
137 | return ERR_PTR(rv); | |
138 | } | |
139 | return net; | |
140 | } | |
486a87f1 | 141 | |
088eb2d9 AD |
142 | struct net *copy_net_ns(unsigned long flags, struct net *old_net) |
143 | { | |
144 | if (!(flags & CLONE_NEWNET)) | |
145 | return get_net(old_net); | |
146 | return net_create(); | |
9dd776b6 EB |
147 | } |
148 | ||
6a1a3b9f PE |
149 | static void cleanup_net(struct work_struct *work) |
150 | { | |
151 | struct pernet_operations *ops; | |
152 | struct net *net; | |
153 | ||
154 | net = container_of(work, struct net, work); | |
155 | ||
156 | mutex_lock(&net_mutex); | |
157 | ||
158 | /* Don't let anyone else find us. */ | |
159 | rtnl_lock(); | |
11a28d37 | 160 | list_del_rcu(&net->list); |
6a1a3b9f PE |
161 | rtnl_unlock(); |
162 | ||
11a28d37 JB |
163 | /* |
164 | * Another CPU might be rcu-iterating the list, wait for it. | |
165 | * This needs to be before calling the exit() notifiers, so | |
166 | * the rcu_barrier() below isn't sufficient alone. | |
167 | */ | |
168 | synchronize_rcu(); | |
169 | ||
6a1a3b9f PE |
170 | /* Run all of the network namespace exit methods */ |
171 | list_for_each_entry_reverse(ops, &pernet_list, list) { | |
172 | if (ops->exit) | |
173 | ops->exit(net); | |
174 | } | |
175 | ||
176 | mutex_unlock(&net_mutex); | |
177 | ||
178 | /* Ensure there are no outstanding rcu callbacks using this | |
179 | * network namespace. | |
180 | */ | |
181 | rcu_barrier(); | |
182 | ||
183 | /* Finally it is safe to free my network namespace structure */ | |
184 | net_free(net); | |
185 | } | |
186 | ||
187 | void __put_net(struct net *net) | |
188 | { | |
189 | /* Cleanup the network namespace in process context */ | |
190 | INIT_WORK(&net->work, cleanup_net); | |
3ef1355d | 191 | queue_work(netns_wq, &net->work); |
6a1a3b9f PE |
192 | } |
193 | EXPORT_SYMBOL_GPL(__put_net); | |
194 | ||
195 | #else | |
196 | struct net *copy_net_ns(unsigned long flags, struct net *old_net) | |
197 | { | |
198 | if (flags & CLONE_NEWNET) | |
199 | return ERR_PTR(-EINVAL); | |
200 | return old_net; | |
201 | } | |
202 | #endif | |
203 | ||
5f256bec EB |
204 | static int __init net_ns_init(void) |
205 | { | |
486a87f1 | 206 | struct net_generic *ng; |
5f256bec | 207 | |
d57a9212 | 208 | #ifdef CONFIG_NET_NS |
5f256bec EB |
209 | net_cachep = kmem_cache_create("net_namespace", sizeof(struct net), |
210 | SMP_CACHE_BYTES, | |
211 | SLAB_PANIC, NULL); | |
3ef1355d BT |
212 | |
213 | /* Create workqueue for cleanup */ | |
214 | netns_wq = create_singlethread_workqueue("netns"); | |
215 | if (!netns_wq) | |
216 | panic("Could not create netns workq"); | |
d57a9212 | 217 | #endif |
3ef1355d | 218 | |
486a87f1 DL |
219 | ng = net_alloc_generic(); |
220 | if (!ng) | |
221 | panic("Could not allocate generic netns"); | |
222 | ||
223 | rcu_assign_pointer(init_net.gen, ng); | |
224 | ||
5f256bec | 225 | mutex_lock(&net_mutex); |
ca0f3112 SH |
226 | if (setup_net(&init_net)) |
227 | panic("Could not setup the initial network namespace"); | |
5f256bec | 228 | |
f4618d39 | 229 | rtnl_lock(); |
11a28d37 | 230 | list_add_tail_rcu(&init_net.list, &net_namespace_list); |
f4618d39 | 231 | rtnl_unlock(); |
5f256bec EB |
232 | |
233 | mutex_unlock(&net_mutex); | |
5f256bec EB |
234 | |
235 | return 0; | |
236 | } | |
237 | ||
238 | pure_initcall(net_ns_init); | |
239 | ||
ed160e83 | 240 | #ifdef CONFIG_NET_NS |
5f256bec EB |
241 | static int register_pernet_operations(struct list_head *list, |
242 | struct pernet_operations *ops) | |
243 | { | |
244 | struct net *net, *undo_net; | |
245 | int error; | |
246 | ||
5f256bec | 247 | list_add_tail(&ops->list, list); |
1dba323b PE |
248 | if (ops->init) { |
249 | for_each_net(net) { | |
5f256bec EB |
250 | error = ops->init(net); |
251 | if (error) | |
252 | goto out_undo; | |
253 | } | |
254 | } | |
1dba323b | 255 | return 0; |
5f256bec EB |
256 | |
257 | out_undo: | |
258 | /* If I have an error cleanup all namespaces I initialized */ | |
259 | list_del(&ops->list); | |
1dba323b PE |
260 | if (ops->exit) { |
261 | for_each_net(undo_net) { | |
262 | if (undo_net == net) | |
263 | goto undone; | |
5f256bec | 264 | ops->exit(undo_net); |
1dba323b | 265 | } |
5f256bec EB |
266 | } |
267 | undone: | |
1dba323b | 268 | return error; |
5f256bec EB |
269 | } |
270 | ||
271 | static void unregister_pernet_operations(struct pernet_operations *ops) | |
272 | { | |
273 | struct net *net; | |
274 | ||
275 | list_del(&ops->list); | |
1dba323b PE |
276 | if (ops->exit) |
277 | for_each_net(net) | |
5f256bec EB |
278 | ops->exit(net); |
279 | } | |
280 | ||
ed160e83 DL |
281 | #else |
282 | ||
283 | static int register_pernet_operations(struct list_head *list, | |
284 | struct pernet_operations *ops) | |
285 | { | |
286 | if (ops->init == NULL) | |
287 | return 0; | |
288 | return ops->init(&init_net); | |
289 | } | |
290 | ||
291 | static void unregister_pernet_operations(struct pernet_operations *ops) | |
292 | { | |
293 | if (ops->exit) | |
294 | ops->exit(&init_net); | |
295 | } | |
296 | #endif | |
297 | ||
c93cf61f PE |
298 | static DEFINE_IDA(net_generic_ids); |
299 | ||
5f256bec EB |
300 | /** |
301 | * register_pernet_subsys - register a network namespace subsystem | |
302 | * @ops: pernet operations structure for the subsystem | |
303 | * | |
304 | * Register a subsystem which has init and exit functions | |
305 | * that are called when network namespaces are created and | |
306 | * destroyed respectively. | |
307 | * | |
308 | * When registered all network namespace init functions are | |
309 | * called for every existing network namespace. Allowing kernel | |
310 | * modules to have a race free view of the set of network namespaces. | |
311 | * | |
312 | * When a new network namespace is created all of the init | |
313 | * methods are called in the order in which they were registered. | |
314 | * | |
315 | * When a network namespace is destroyed all of the exit methods | |
316 | * are called in the reverse of the order with which they were | |
317 | * registered. | |
318 | */ | |
319 | int register_pernet_subsys(struct pernet_operations *ops) | |
320 | { | |
321 | int error; | |
322 | mutex_lock(&net_mutex); | |
323 | error = register_pernet_operations(first_device, ops); | |
324 | mutex_unlock(&net_mutex); | |
325 | return error; | |
326 | } | |
327 | EXPORT_SYMBOL_GPL(register_pernet_subsys); | |
328 | ||
329 | /** | |
330 | * unregister_pernet_subsys - unregister a network namespace subsystem | |
331 | * @ops: pernet operations structure to manipulate | |
332 | * | |
333 | * Remove the pernet operations structure from the list to be | |
53379e57 | 334 | * used when network namespaces are created or destroyed. In |
5f256bec EB |
335 | * addition run the exit method for all existing network |
336 | * namespaces. | |
337 | */ | |
338 | void unregister_pernet_subsys(struct pernet_operations *module) | |
339 | { | |
340 | mutex_lock(&net_mutex); | |
341 | unregister_pernet_operations(module); | |
342 | mutex_unlock(&net_mutex); | |
343 | } | |
344 | EXPORT_SYMBOL_GPL(unregister_pernet_subsys); | |
345 | ||
485ac57b AD |
346 | int register_pernet_gen_subsys(int *id, struct pernet_operations *ops) |
347 | { | |
348 | int rv; | |
349 | ||
350 | mutex_lock(&net_mutex); | |
351 | again: | |
352 | rv = ida_get_new_above(&net_generic_ids, 1, id); | |
353 | if (rv < 0) { | |
354 | if (rv == -EAGAIN) { | |
355 | ida_pre_get(&net_generic_ids, GFP_KERNEL); | |
356 | goto again; | |
357 | } | |
358 | goto out; | |
359 | } | |
360 | rv = register_pernet_operations(first_device, ops); | |
361 | if (rv < 0) | |
362 | ida_remove(&net_generic_ids, *id); | |
485ac57b | 363 | out: |
357f5b0b | 364 | mutex_unlock(&net_mutex); |
485ac57b AD |
365 | return rv; |
366 | } | |
367 | EXPORT_SYMBOL_GPL(register_pernet_gen_subsys); | |
368 | ||
369 | void unregister_pernet_gen_subsys(int id, struct pernet_operations *ops) | |
370 | { | |
371 | mutex_lock(&net_mutex); | |
372 | unregister_pernet_operations(ops); | |
373 | ida_remove(&net_generic_ids, id); | |
374 | mutex_unlock(&net_mutex); | |
375 | } | |
376 | EXPORT_SYMBOL_GPL(unregister_pernet_gen_subsys); | |
377 | ||
5f256bec EB |
378 | /** |
379 | * register_pernet_device - register a network namespace device | |
380 | * @ops: pernet operations structure for the subsystem | |
381 | * | |
382 | * Register a device which has init and exit functions | |
383 | * that are called when network namespaces are created and | |
384 | * destroyed respectively. | |
385 | * | |
386 | * When registered all network namespace init functions are | |
387 | * called for every existing network namespace. Allowing kernel | |
388 | * modules to have a race free view of the set of network namespaces. | |
389 | * | |
390 | * When a new network namespace is created all of the init | |
391 | * methods are called in the order in which they were registered. | |
392 | * | |
393 | * When a network namespace is destroyed all of the exit methods | |
394 | * are called in the reverse of the order with which they were | |
395 | * registered. | |
396 | */ | |
397 | int register_pernet_device(struct pernet_operations *ops) | |
398 | { | |
399 | int error; | |
400 | mutex_lock(&net_mutex); | |
401 | error = register_pernet_operations(&pernet_list, ops); | |
402 | if (!error && (first_device == &pernet_list)) | |
403 | first_device = &ops->list; | |
404 | mutex_unlock(&net_mutex); | |
405 | return error; | |
406 | } | |
407 | EXPORT_SYMBOL_GPL(register_pernet_device); | |
408 | ||
c93cf61f PE |
409 | int register_pernet_gen_device(int *id, struct pernet_operations *ops) |
410 | { | |
411 | int error; | |
412 | mutex_lock(&net_mutex); | |
413 | again: | |
414 | error = ida_get_new_above(&net_generic_ids, 1, id); | |
415 | if (error) { | |
416 | if (error == -EAGAIN) { | |
417 | ida_pre_get(&net_generic_ids, GFP_KERNEL); | |
418 | goto again; | |
419 | } | |
420 | goto out; | |
421 | } | |
422 | error = register_pernet_operations(&pernet_list, ops); | |
423 | if (error) | |
424 | ida_remove(&net_generic_ids, *id); | |
425 | else if (first_device == &pernet_list) | |
426 | first_device = &ops->list; | |
427 | out: | |
428 | mutex_unlock(&net_mutex); | |
429 | return error; | |
430 | } | |
431 | EXPORT_SYMBOL_GPL(register_pernet_gen_device); | |
432 | ||
5f256bec EB |
433 | /** |
434 | * unregister_pernet_device - unregister a network namespace netdevice | |
435 | * @ops: pernet operations structure to manipulate | |
436 | * | |
437 | * Remove the pernet operations structure from the list to be | |
53379e57 | 438 | * used when network namespaces are created or destroyed. In |
5f256bec EB |
439 | * addition run the exit method for all existing network |
440 | * namespaces. | |
441 | */ | |
442 | void unregister_pernet_device(struct pernet_operations *ops) | |
443 | { | |
444 | mutex_lock(&net_mutex); | |
445 | if (&ops->list == first_device) | |
446 | first_device = first_device->next; | |
447 | unregister_pernet_operations(ops); | |
448 | mutex_unlock(&net_mutex); | |
449 | } | |
450 | EXPORT_SYMBOL_GPL(unregister_pernet_device); | |
c93cf61f PE |
451 | |
452 | void unregister_pernet_gen_device(int id, struct pernet_operations *ops) | |
453 | { | |
454 | mutex_lock(&net_mutex); | |
455 | if (&ops->list == first_device) | |
456 | first_device = first_device->next; | |
457 | unregister_pernet_operations(ops); | |
458 | ida_remove(&net_generic_ids, id); | |
459 | mutex_unlock(&net_mutex); | |
460 | } | |
461 | EXPORT_SYMBOL_GPL(unregister_pernet_gen_device); | |
dec827d1 PE |
462 | |
463 | static void net_generic_release(struct rcu_head *rcu) | |
464 | { | |
465 | struct net_generic *ng; | |
466 | ||
467 | ng = container_of(rcu, struct net_generic, rcu); | |
468 | kfree(ng); | |
469 | } | |
470 | ||
471 | int net_assign_generic(struct net *net, int id, void *data) | |
472 | { | |
473 | struct net_generic *ng, *old_ng; | |
474 | ||
475 | BUG_ON(!mutex_is_locked(&net_mutex)); | |
476 | BUG_ON(id == 0); | |
477 | ||
478 | ng = old_ng = net->gen; | |
479 | if (old_ng->len >= id) | |
480 | goto assign; | |
481 | ||
482 | ng = kzalloc(sizeof(struct net_generic) + | |
483 | id * sizeof(void *), GFP_KERNEL); | |
484 | if (ng == NULL) | |
485 | return -ENOMEM; | |
486 | ||
487 | /* | |
488 | * Some synchronisation notes: | |
489 | * | |
490 | * The net_generic explores the net->gen array inside rcu | |
491 | * read section. Besides once set the net->gen->ptr[x] | |
492 | * pointer never changes (see rules in netns/generic.h). | |
493 | * | |
494 | * That said, we simply duplicate this array and schedule | |
495 | * the old copy for kfree after a grace period. | |
496 | */ | |
497 | ||
498 | ng->len = id; | |
dec827d1 PE |
499 | memcpy(&ng->ptr, &old_ng->ptr, old_ng->len); |
500 | ||
501 | rcu_assign_pointer(net->gen, ng); | |
502 | call_rcu(&old_ng->rcu, net_generic_release); | |
503 | assign: | |
504 | ng->ptr[id - 1] = data; | |
505 | return 0; | |
506 | } | |
507 | EXPORT_SYMBOL_GPL(net_assign_generic); |