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