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> |
30ffee84 | 10 | #include <linux/nsproxy.h> |
5f256bec | 11 | #include <net/net_namespace.h> |
dec827d1 | 12 | #include <net/netns/generic.h> |
5f256bec EB |
13 | |
14 | /* | |
15 | * Our network namespace constructor/destructor lists | |
16 | */ | |
17 | ||
18 | static LIST_HEAD(pernet_list); | |
19 | static struct list_head *first_device = &pernet_list; | |
20 | static DEFINE_MUTEX(net_mutex); | |
21 | ||
5f256bec | 22 | LIST_HEAD(net_namespace_list); |
b76a461f | 23 | EXPORT_SYMBOL_GPL(net_namespace_list); |
5f256bec | 24 | |
5f256bec | 25 | struct net init_net; |
ff4b9502 | 26 | EXPORT_SYMBOL(init_net); |
5f256bec | 27 | |
dec827d1 PE |
28 | #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */ |
29 | ||
5f256bec EB |
30 | /* |
31 | * setup_net runs the initializers for the network namespace object. | |
32 | */ | |
1a2ee93d | 33 | static __net_init int setup_net(struct net *net) |
5f256bec EB |
34 | { |
35 | /* Must be called with net_mutex held */ | |
36 | struct pernet_operations *ops; | |
486a87f1 | 37 | int error = 0; |
5f256bec | 38 | |
5f256bec | 39 | atomic_set(&net->count, 1); |
486a87f1 | 40 | |
5d1e4468 | 41 | #ifdef NETNS_REFCNT_DEBUG |
5f256bec | 42 | atomic_set(&net->use_count, 0); |
5d1e4468 | 43 | #endif |
5f256bec | 44 | |
768f3591 | 45 | list_for_each_entry(ops, &pernet_list, list) { |
5f256bec EB |
46 | if (ops->init) { |
47 | error = ops->init(net); | |
48 | if (error < 0) | |
49 | goto out_undo; | |
50 | } | |
51 | } | |
52 | out: | |
53 | return error; | |
768f3591 | 54 | |
5f256bec EB |
55 | out_undo: |
56 | /* Walk through the list backwards calling the exit functions | |
57 | * for the pernet modules whose init functions did not fail. | |
58 | */ | |
768f3591 | 59 | list_for_each_entry_continue_reverse(ops, &pernet_list, list) { |
5f256bec EB |
60 | if (ops->exit) |
61 | ops->exit(net); | |
62 | } | |
310928d9 DL |
63 | |
64 | rcu_barrier(); | |
5f256bec EB |
65 | goto out; |
66 | } | |
67 | ||
486a87f1 | 68 | static struct net_generic *net_alloc_generic(void) |
6a1a3b9f | 69 | { |
486a87f1 DL |
70 | struct net_generic *ng; |
71 | size_t generic_size = sizeof(struct net_generic) + | |
72 | INITIAL_NET_GEN_PTRS * sizeof(void *); | |
73 | ||
74 | ng = kzalloc(generic_size, GFP_KERNEL); | |
75 | if (ng) | |
76 | ng->len = INITIAL_NET_GEN_PTRS; | |
77 | ||
78 | return ng; | |
6a1a3b9f PE |
79 | } |
80 | ||
ebe47d47 CN |
81 | #ifdef CONFIG_NET_NS |
82 | static struct kmem_cache *net_cachep; | |
83 | static struct workqueue_struct *netns_wq; | |
84 | ||
486a87f1 | 85 | static struct net *net_alloc(void) |
45a19b0a | 86 | { |
486a87f1 DL |
87 | struct net *net = NULL; |
88 | struct net_generic *ng; | |
89 | ||
90 | ng = net_alloc_generic(); | |
91 | if (!ng) | |
92 | goto out; | |
93 | ||
94 | net = kmem_cache_zalloc(net_cachep, GFP_KERNEL); | |
45a19b0a | 95 | if (!net) |
486a87f1 | 96 | goto out_free; |
45a19b0a | 97 | |
486a87f1 DL |
98 | rcu_assign_pointer(net->gen, ng); |
99 | out: | |
100 | return net; | |
101 | ||
102 | out_free: | |
103 | kfree(ng); | |
104 | goto out; | |
105 | } | |
106 | ||
107 | static void net_free(struct net *net) | |
108 | { | |
5d1e4468 | 109 | #ifdef NETNS_REFCNT_DEBUG |
45a19b0a JFS |
110 | if (unlikely(atomic_read(&net->use_count) != 0)) { |
111 | printk(KERN_EMERG "network namespace not free! Usage: %d\n", | |
112 | atomic_read(&net->use_count)); | |
113 | return; | |
114 | } | |
5d1e4468 | 115 | #endif |
4ef079cc | 116 | kfree(net->gen); |
45a19b0a JFS |
117 | kmem_cache_free(net_cachep, net); |
118 | } | |
119 | ||
088eb2d9 | 120 | static struct net *net_create(void) |
9dd776b6 | 121 | { |
088eb2d9 AD |
122 | struct net *net; |
123 | int rv; | |
9dd776b6 | 124 | |
088eb2d9 AD |
125 | net = net_alloc(); |
126 | if (!net) | |
127 | return ERR_PTR(-ENOMEM); | |
9dd776b6 | 128 | mutex_lock(&net_mutex); |
088eb2d9 AD |
129 | rv = setup_net(net); |
130 | if (rv == 0) { | |
486a87f1 | 131 | rtnl_lock(); |
11a28d37 | 132 | list_add_tail_rcu(&net->list, &net_namespace_list); |
486a87f1 DL |
133 | rtnl_unlock(); |
134 | } | |
9dd776b6 | 135 | mutex_unlock(&net_mutex); |
088eb2d9 AD |
136 | if (rv < 0) { |
137 | net_free(net); | |
138 | return ERR_PTR(rv); | |
139 | } | |
140 | return net; | |
141 | } | |
486a87f1 | 142 | |
088eb2d9 AD |
143 | struct net *copy_net_ns(unsigned long flags, struct net *old_net) |
144 | { | |
145 | if (!(flags & CLONE_NEWNET)) | |
146 | return get_net(old_net); | |
147 | return net_create(); | |
9dd776b6 EB |
148 | } |
149 | ||
6a1a3b9f PE |
150 | static void cleanup_net(struct work_struct *work) |
151 | { | |
152 | struct pernet_operations *ops; | |
153 | struct net *net; | |
154 | ||
155 | net = container_of(work, struct net, work); | |
156 | ||
157 | mutex_lock(&net_mutex); | |
158 | ||
159 | /* Don't let anyone else find us. */ | |
160 | rtnl_lock(); | |
11a28d37 | 161 | list_del_rcu(&net->list); |
6a1a3b9f PE |
162 | rtnl_unlock(); |
163 | ||
11a28d37 JB |
164 | /* |
165 | * Another CPU might be rcu-iterating the list, wait for it. | |
166 | * This needs to be before calling the exit() notifiers, so | |
167 | * the rcu_barrier() below isn't sufficient alone. | |
168 | */ | |
169 | synchronize_rcu(); | |
170 | ||
6a1a3b9f PE |
171 | /* Run all of the network namespace exit methods */ |
172 | list_for_each_entry_reverse(ops, &pernet_list, list) { | |
173 | if (ops->exit) | |
174 | ops->exit(net); | |
175 | } | |
176 | ||
177 | mutex_unlock(&net_mutex); | |
178 | ||
179 | /* Ensure there are no outstanding rcu callbacks using this | |
180 | * network namespace. | |
181 | */ | |
182 | rcu_barrier(); | |
183 | ||
184 | /* Finally it is safe to free my network namespace structure */ | |
185 | net_free(net); | |
186 | } | |
187 | ||
188 | void __put_net(struct net *net) | |
189 | { | |
190 | /* Cleanup the network namespace in process context */ | |
191 | INIT_WORK(&net->work, cleanup_net); | |
3ef1355d | 192 | queue_work(netns_wq, &net->work); |
6a1a3b9f PE |
193 | } |
194 | EXPORT_SYMBOL_GPL(__put_net); | |
195 | ||
196 | #else | |
197 | struct net *copy_net_ns(unsigned long flags, struct net *old_net) | |
198 | { | |
199 | if (flags & CLONE_NEWNET) | |
200 | return ERR_PTR(-EINVAL); | |
201 | return old_net; | |
202 | } | |
203 | #endif | |
204 | ||
30ffee84 JB |
205 | struct net *get_net_ns_by_pid(pid_t pid) |
206 | { | |
207 | struct task_struct *tsk; | |
208 | struct net *net; | |
209 | ||
210 | /* Lookup the network namespace */ | |
211 | net = ERR_PTR(-ESRCH); | |
212 | rcu_read_lock(); | |
213 | tsk = find_task_by_vpid(pid); | |
214 | if (tsk) { | |
215 | struct nsproxy *nsproxy; | |
216 | nsproxy = task_nsproxy(tsk); | |
217 | if (nsproxy) | |
218 | net = get_net(nsproxy->net_ns); | |
219 | } | |
220 | rcu_read_unlock(); | |
221 | return net; | |
222 | } | |
223 | EXPORT_SYMBOL_GPL(get_net_ns_by_pid); | |
224 | ||
5f256bec EB |
225 | static int __init net_ns_init(void) |
226 | { | |
486a87f1 | 227 | struct net_generic *ng; |
5f256bec | 228 | |
d57a9212 | 229 | #ifdef CONFIG_NET_NS |
5f256bec EB |
230 | net_cachep = kmem_cache_create("net_namespace", sizeof(struct net), |
231 | SMP_CACHE_BYTES, | |
232 | SLAB_PANIC, NULL); | |
3ef1355d BT |
233 | |
234 | /* Create workqueue for cleanup */ | |
235 | netns_wq = create_singlethread_workqueue("netns"); | |
236 | if (!netns_wq) | |
237 | panic("Could not create netns workq"); | |
d57a9212 | 238 | #endif |
3ef1355d | 239 | |
486a87f1 DL |
240 | ng = net_alloc_generic(); |
241 | if (!ng) | |
242 | panic("Could not allocate generic netns"); | |
243 | ||
244 | rcu_assign_pointer(init_net.gen, ng); | |
245 | ||
5f256bec | 246 | mutex_lock(&net_mutex); |
ca0f3112 SH |
247 | if (setup_net(&init_net)) |
248 | panic("Could not setup the initial network namespace"); | |
5f256bec | 249 | |
f4618d39 | 250 | rtnl_lock(); |
11a28d37 | 251 | list_add_tail_rcu(&init_net.list, &net_namespace_list); |
f4618d39 | 252 | rtnl_unlock(); |
5f256bec EB |
253 | |
254 | mutex_unlock(&net_mutex); | |
5f256bec EB |
255 | |
256 | return 0; | |
257 | } | |
258 | ||
259 | pure_initcall(net_ns_init); | |
260 | ||
ed160e83 | 261 | #ifdef CONFIG_NET_NS |
5f256bec EB |
262 | static int register_pernet_operations(struct list_head *list, |
263 | struct pernet_operations *ops) | |
264 | { | |
265 | struct net *net, *undo_net; | |
266 | int error; | |
267 | ||
5f256bec | 268 | list_add_tail(&ops->list, list); |
1dba323b PE |
269 | if (ops->init) { |
270 | for_each_net(net) { | |
5f256bec EB |
271 | error = ops->init(net); |
272 | if (error) | |
273 | goto out_undo; | |
274 | } | |
275 | } | |
1dba323b | 276 | return 0; |
5f256bec EB |
277 | |
278 | out_undo: | |
279 | /* If I have an error cleanup all namespaces I initialized */ | |
280 | list_del(&ops->list); | |
1dba323b PE |
281 | if (ops->exit) { |
282 | for_each_net(undo_net) { | |
283 | if (undo_net == net) | |
284 | goto undone; | |
5f256bec | 285 | ops->exit(undo_net); |
1dba323b | 286 | } |
5f256bec EB |
287 | } |
288 | undone: | |
1dba323b | 289 | return error; |
5f256bec EB |
290 | } |
291 | ||
292 | static void unregister_pernet_operations(struct pernet_operations *ops) | |
293 | { | |
294 | struct net *net; | |
295 | ||
296 | list_del(&ops->list); | |
1dba323b PE |
297 | if (ops->exit) |
298 | for_each_net(net) | |
5f256bec EB |
299 | ops->exit(net); |
300 | } | |
301 | ||
ed160e83 DL |
302 | #else |
303 | ||
304 | static int register_pernet_operations(struct list_head *list, | |
305 | struct pernet_operations *ops) | |
306 | { | |
307 | if (ops->init == NULL) | |
308 | return 0; | |
309 | return ops->init(&init_net); | |
310 | } | |
311 | ||
312 | static void unregister_pernet_operations(struct pernet_operations *ops) | |
313 | { | |
314 | if (ops->exit) | |
315 | ops->exit(&init_net); | |
316 | } | |
317 | #endif | |
318 | ||
c93cf61f PE |
319 | static DEFINE_IDA(net_generic_ids); |
320 | ||
5f256bec EB |
321 | /** |
322 | * register_pernet_subsys - register a network namespace subsystem | |
323 | * @ops: pernet operations structure for the subsystem | |
324 | * | |
325 | * Register a subsystem which has init and exit functions | |
326 | * that are called when network namespaces are created and | |
327 | * destroyed respectively. | |
328 | * | |
329 | * When registered all network namespace init functions are | |
330 | * called for every existing network namespace. Allowing kernel | |
331 | * modules to have a race free view of the set of network namespaces. | |
332 | * | |
333 | * When a new network namespace is created all of the init | |
334 | * methods are called in the order in which they were registered. | |
335 | * | |
336 | * When a network namespace is destroyed all of the exit methods | |
337 | * are called in the reverse of the order with which they were | |
338 | * registered. | |
339 | */ | |
340 | int register_pernet_subsys(struct pernet_operations *ops) | |
341 | { | |
342 | int error; | |
343 | mutex_lock(&net_mutex); | |
344 | error = register_pernet_operations(first_device, ops); | |
345 | mutex_unlock(&net_mutex); | |
346 | return error; | |
347 | } | |
348 | EXPORT_SYMBOL_GPL(register_pernet_subsys); | |
349 | ||
350 | /** | |
351 | * unregister_pernet_subsys - unregister a network namespace subsystem | |
352 | * @ops: pernet operations structure to manipulate | |
353 | * | |
354 | * Remove the pernet operations structure from the list to be | |
53379e57 | 355 | * used when network namespaces are created or destroyed. In |
5f256bec EB |
356 | * addition run the exit method for all existing network |
357 | * namespaces. | |
358 | */ | |
359 | void unregister_pernet_subsys(struct pernet_operations *module) | |
360 | { | |
361 | mutex_lock(&net_mutex); | |
362 | unregister_pernet_operations(module); | |
363 | mutex_unlock(&net_mutex); | |
364 | } | |
365 | EXPORT_SYMBOL_GPL(unregister_pernet_subsys); | |
366 | ||
485ac57b AD |
367 | int register_pernet_gen_subsys(int *id, struct pernet_operations *ops) |
368 | { | |
369 | int rv; | |
370 | ||
371 | mutex_lock(&net_mutex); | |
372 | again: | |
373 | rv = ida_get_new_above(&net_generic_ids, 1, id); | |
374 | if (rv < 0) { | |
375 | if (rv == -EAGAIN) { | |
376 | ida_pre_get(&net_generic_ids, GFP_KERNEL); | |
377 | goto again; | |
378 | } | |
379 | goto out; | |
380 | } | |
381 | rv = register_pernet_operations(first_device, ops); | |
382 | if (rv < 0) | |
383 | ida_remove(&net_generic_ids, *id); | |
485ac57b | 384 | out: |
357f5b0b | 385 | mutex_unlock(&net_mutex); |
485ac57b AD |
386 | return rv; |
387 | } | |
388 | EXPORT_SYMBOL_GPL(register_pernet_gen_subsys); | |
389 | ||
390 | void unregister_pernet_gen_subsys(int id, struct pernet_operations *ops) | |
391 | { | |
392 | mutex_lock(&net_mutex); | |
393 | unregister_pernet_operations(ops); | |
394 | ida_remove(&net_generic_ids, id); | |
395 | mutex_unlock(&net_mutex); | |
396 | } | |
397 | EXPORT_SYMBOL_GPL(unregister_pernet_gen_subsys); | |
398 | ||
5f256bec EB |
399 | /** |
400 | * register_pernet_device - register a network namespace device | |
401 | * @ops: pernet operations structure for the subsystem | |
402 | * | |
403 | * Register a device which has init and exit functions | |
404 | * that are called when network namespaces are created and | |
405 | * destroyed respectively. | |
406 | * | |
407 | * When registered all network namespace init functions are | |
408 | * called for every existing network namespace. Allowing kernel | |
409 | * modules to have a race free view of the set of network namespaces. | |
410 | * | |
411 | * When a new network namespace is created all of the init | |
412 | * methods are called in the order in which they were registered. | |
413 | * | |
414 | * When a network namespace is destroyed all of the exit methods | |
415 | * are called in the reverse of the order with which they were | |
416 | * registered. | |
417 | */ | |
418 | int register_pernet_device(struct pernet_operations *ops) | |
419 | { | |
420 | int error; | |
421 | mutex_lock(&net_mutex); | |
422 | error = register_pernet_operations(&pernet_list, ops); | |
423 | if (!error && (first_device == &pernet_list)) | |
424 | first_device = &ops->list; | |
425 | mutex_unlock(&net_mutex); | |
426 | return error; | |
427 | } | |
428 | EXPORT_SYMBOL_GPL(register_pernet_device); | |
429 | ||
c93cf61f PE |
430 | int register_pernet_gen_device(int *id, struct pernet_operations *ops) |
431 | { | |
432 | int error; | |
433 | mutex_lock(&net_mutex); | |
434 | again: | |
435 | error = ida_get_new_above(&net_generic_ids, 1, id); | |
436 | if (error) { | |
437 | if (error == -EAGAIN) { | |
438 | ida_pre_get(&net_generic_ids, GFP_KERNEL); | |
439 | goto again; | |
440 | } | |
441 | goto out; | |
442 | } | |
443 | error = register_pernet_operations(&pernet_list, ops); | |
444 | if (error) | |
445 | ida_remove(&net_generic_ids, *id); | |
446 | else if (first_device == &pernet_list) | |
447 | first_device = &ops->list; | |
448 | out: | |
449 | mutex_unlock(&net_mutex); | |
450 | return error; | |
451 | } | |
452 | EXPORT_SYMBOL_GPL(register_pernet_gen_device); | |
453 | ||
5f256bec EB |
454 | /** |
455 | * unregister_pernet_device - unregister a network namespace netdevice | |
456 | * @ops: pernet operations structure to manipulate | |
457 | * | |
458 | * Remove the pernet operations structure from the list to be | |
53379e57 | 459 | * used when network namespaces are created or destroyed. In |
5f256bec EB |
460 | * addition run the exit method for all existing network |
461 | * namespaces. | |
462 | */ | |
463 | void unregister_pernet_device(struct pernet_operations *ops) | |
464 | { | |
465 | mutex_lock(&net_mutex); | |
466 | if (&ops->list == first_device) | |
467 | first_device = first_device->next; | |
468 | unregister_pernet_operations(ops); | |
469 | mutex_unlock(&net_mutex); | |
470 | } | |
471 | EXPORT_SYMBOL_GPL(unregister_pernet_device); | |
c93cf61f PE |
472 | |
473 | void unregister_pernet_gen_device(int id, struct pernet_operations *ops) | |
474 | { | |
475 | mutex_lock(&net_mutex); | |
476 | if (&ops->list == first_device) | |
477 | first_device = first_device->next; | |
478 | unregister_pernet_operations(ops); | |
479 | ida_remove(&net_generic_ids, id); | |
480 | mutex_unlock(&net_mutex); | |
481 | } | |
482 | EXPORT_SYMBOL_GPL(unregister_pernet_gen_device); | |
dec827d1 PE |
483 | |
484 | static void net_generic_release(struct rcu_head *rcu) | |
485 | { | |
486 | struct net_generic *ng; | |
487 | ||
488 | ng = container_of(rcu, struct net_generic, rcu); | |
489 | kfree(ng); | |
490 | } | |
491 | ||
492 | int net_assign_generic(struct net *net, int id, void *data) | |
493 | { | |
494 | struct net_generic *ng, *old_ng; | |
495 | ||
496 | BUG_ON(!mutex_is_locked(&net_mutex)); | |
497 | BUG_ON(id == 0); | |
498 | ||
499 | ng = old_ng = net->gen; | |
500 | if (old_ng->len >= id) | |
501 | goto assign; | |
502 | ||
503 | ng = kzalloc(sizeof(struct net_generic) + | |
504 | id * sizeof(void *), GFP_KERNEL); | |
505 | if (ng == NULL) | |
506 | return -ENOMEM; | |
507 | ||
508 | /* | |
509 | * Some synchronisation notes: | |
510 | * | |
511 | * The net_generic explores the net->gen array inside rcu | |
512 | * read section. Besides once set the net->gen->ptr[x] | |
513 | * pointer never changes (see rules in netns/generic.h). | |
514 | * | |
515 | * That said, we simply duplicate this array and schedule | |
516 | * the old copy for kfree after a grace period. | |
517 | */ | |
518 | ||
519 | ng->len = id; | |
14458630 | 520 | memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*)); |
dec827d1 PE |
521 | |
522 | rcu_assign_pointer(net->gen, ng); | |
523 | call_rcu(&old_ng->rcu, net_generic_release); | |
524 | assign: | |
525 | ng->ptr[id - 1] = data; | |
526 | return 0; | |
527 | } | |
528 | EXPORT_SYMBOL_GPL(net_assign_generic); |