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>
7 #include <linux/sched.h>
9 #include <linux/rculist.h>
10 #include <linux/nsproxy.h>
11 #include <net/net_namespace.h>
12 #include <net/netns/generic.h>
15 * Our network namespace constructor/destructor lists
18 static LIST_HEAD(pernet_list);
19 static struct list_head *first_device = &pernet_list;
20 static DEFINE_MUTEX(net_mutex);
22 LIST_HEAD(net_namespace_list);
23 EXPORT_SYMBOL_GPL(net_namespace_list);
26 EXPORT_SYMBOL(init_net);
28 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
30 static void net_generic_release(struct rcu_head *rcu)
32 struct net_generic *ng;
34 ng = container_of(rcu, struct net_generic, rcu);
38 static int net_assign_generic(struct net *net, int id, void *data)
40 struct net_generic *ng, *old_ng;
42 BUG_ON(!mutex_is_locked(&net_mutex));
45 ng = old_ng = net->gen;
46 if (old_ng->len >= id)
49 ng = kzalloc(sizeof(struct net_generic) +
50 id * sizeof(void *), GFP_KERNEL);
55 * Some synchronisation notes:
57 * The net_generic explores the net->gen array inside rcu
58 * read section. Besides once set the net->gen->ptr[x]
59 * pointer never changes (see rules in netns/generic.h).
61 * That said, we simply duplicate this array and schedule
62 * the old copy for kfree after a grace period.
66 memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
68 rcu_assign_pointer(net->gen, ng);
69 call_rcu(&old_ng->rcu, net_generic_release);
71 ng->ptr[id - 1] = data;
75 static int ops_init(const struct pernet_operations *ops, struct net *net)
78 if (ops->id && ops->size) {
79 void *data = kzalloc(ops->size, GFP_KERNEL);
83 err = net_assign_generic(net, *ops->id, data);
90 return ops->init(net);
94 static void ops_free(const struct pernet_operations *ops, struct net *net)
96 if (ops->id && ops->size) {
98 kfree(net_generic(net, id));
102 static void ops_exit_list(const struct pernet_operations *ops,
103 struct list_head *net_exit_list)
107 list_for_each_entry(net, net_exit_list, exit_list)
111 ops->exit_batch(net_exit_list);
114 static void ops_free_list(const struct pernet_operations *ops,
115 struct list_head *net_exit_list)
118 if (ops->size && ops->id) {
119 list_for_each_entry(net, net_exit_list, exit_list)
125 * setup_net runs the initializers for the network namespace object.
127 static __net_init int setup_net(struct net *net)
129 /* Must be called with net_mutex held */
130 const struct pernet_operations *ops, *saved_ops;
132 LIST_HEAD(net_exit_list);
134 atomic_set(&net->count, 1);
136 #ifdef NETNS_REFCNT_DEBUG
137 atomic_set(&net->use_count, 0);
140 list_for_each_entry(ops, &pernet_list, list) {
141 error = ops_init(ops, net);
149 /* Walk through the list backwards calling the exit functions
150 * for the pernet modules whose init functions did not fail.
152 list_add(&net->exit_list, &net_exit_list);
154 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
155 ops_exit_list(ops, &net_exit_list);
158 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
159 ops_free_list(ops, &net_exit_list);
165 static struct net_generic *net_alloc_generic(void)
167 struct net_generic *ng;
168 size_t generic_size = sizeof(struct net_generic) +
169 INITIAL_NET_GEN_PTRS * sizeof(void *);
171 ng = kzalloc(generic_size, GFP_KERNEL);
173 ng->len = INITIAL_NET_GEN_PTRS;
179 static struct kmem_cache *net_cachep;
180 static struct workqueue_struct *netns_wq;
182 static struct net *net_alloc(void)
184 struct net *net = NULL;
185 struct net_generic *ng;
187 ng = net_alloc_generic();
191 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
195 rcu_assign_pointer(net->gen, ng);
204 static void net_free(struct net *net)
206 #ifdef NETNS_REFCNT_DEBUG
207 if (unlikely(atomic_read(&net->use_count) != 0)) {
208 printk(KERN_EMERG "network namespace not free! Usage: %d\n",
209 atomic_read(&net->use_count));
214 kmem_cache_free(net_cachep, net);
217 static struct net *net_create(void)
224 return ERR_PTR(-ENOMEM);
225 mutex_lock(&net_mutex);
229 list_add_tail_rcu(&net->list, &net_namespace_list);
232 mutex_unlock(&net_mutex);
240 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
242 if (!(flags & CLONE_NEWNET))
243 return get_net(old_net);
247 static DEFINE_SPINLOCK(cleanup_list_lock);
248 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
250 static void cleanup_net(struct work_struct *work)
252 const struct pernet_operations *ops;
253 struct net *net, *tmp;
254 LIST_HEAD(net_kill_list);
255 LIST_HEAD(net_exit_list);
257 /* Atomically snapshot the list of namespaces to cleanup */
258 spin_lock_irq(&cleanup_list_lock);
259 list_replace_init(&cleanup_list, &net_kill_list);
260 spin_unlock_irq(&cleanup_list_lock);
262 mutex_lock(&net_mutex);
264 /* Don't let anyone else find us. */
266 list_for_each_entry(net, &net_kill_list, cleanup_list) {
267 list_del_rcu(&net->list);
268 list_add_tail(&net->exit_list, &net_exit_list);
273 * Another CPU might be rcu-iterating the list, wait for it.
274 * This needs to be before calling the exit() notifiers, so
275 * the rcu_barrier() below isn't sufficient alone.
279 /* Run all of the network namespace exit methods */
280 list_for_each_entry_reverse(ops, &pernet_list, list)
281 ops_exit_list(ops, &net_exit_list);
283 /* Free the net generic variables */
284 list_for_each_entry_reverse(ops, &pernet_list, list)
285 ops_free_list(ops, &net_exit_list);
287 mutex_unlock(&net_mutex);
289 /* Ensure there are no outstanding rcu callbacks using this
294 /* Finally it is safe to free my network namespace structure */
295 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
296 list_del_init(&net->exit_list);
300 static DECLARE_WORK(net_cleanup_work, cleanup_net);
302 void __put_net(struct net *net)
304 /* Cleanup the network namespace in process context */
307 spin_lock_irqsave(&cleanup_list_lock, flags);
308 list_add(&net->cleanup_list, &cleanup_list);
309 spin_unlock_irqrestore(&cleanup_list_lock, flags);
311 queue_work(netns_wq, &net_cleanup_work);
313 EXPORT_SYMBOL_GPL(__put_net);
316 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
318 if (flags & CLONE_NEWNET)
319 return ERR_PTR(-EINVAL);
324 struct net *get_net_ns_by_pid(pid_t pid)
326 struct task_struct *tsk;
329 /* Lookup the network namespace */
330 net = ERR_PTR(-ESRCH);
332 tsk = find_task_by_vpid(pid);
334 struct nsproxy *nsproxy;
335 nsproxy = task_nsproxy(tsk);
337 net = get_net(nsproxy->net_ns);
342 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
344 static int __init net_ns_init(void)
346 struct net_generic *ng;
349 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
353 /* Create workqueue for cleanup */
354 netns_wq = create_singlethread_workqueue("netns");
356 panic("Could not create netns workq");
359 ng = net_alloc_generic();
361 panic("Could not allocate generic netns");
363 rcu_assign_pointer(init_net.gen, ng);
365 mutex_lock(&net_mutex);
366 if (setup_net(&init_net))
367 panic("Could not setup the initial network namespace");
370 list_add_tail_rcu(&init_net.list, &net_namespace_list);
373 mutex_unlock(&net_mutex);
378 pure_initcall(net_ns_init);
381 static int __register_pernet_operations(struct list_head *list,
382 struct pernet_operations *ops)
386 LIST_HEAD(net_exit_list);
388 list_add_tail(&ops->list, list);
389 if (ops->init || (ops->id && ops->size)) {
391 error = ops_init(ops, net);
394 list_add_tail(&net->exit_list, &net_exit_list);
400 /* If I have an error cleanup all namespaces I initialized */
401 list_del(&ops->list);
402 ops_exit_list(ops, &net_exit_list);
403 ops_free_list(ops, &net_exit_list);
407 static void __unregister_pernet_operations(struct pernet_operations *ops)
410 LIST_HEAD(net_exit_list);
412 list_del(&ops->list);
414 list_add_tail(&net->exit_list, &net_exit_list);
415 ops_exit_list(ops, &net_exit_list);
416 ops_free_list(ops, &net_exit_list);
421 static int __register_pernet_operations(struct list_head *list,
422 struct pernet_operations *ops)
425 err = ops_init(ops, &init_net);
427 ops_free(ops, &init_net);
432 static void __unregister_pernet_operations(struct pernet_operations *ops)
434 LIST_HEAD(net_exit_list);
435 list_add(&init_net.exit_list, &net_exit_list);
436 ops_exit_list(ops, &net_exit_list);
437 ops_free_list(ops, &net_exit_list);
440 #endif /* CONFIG_NET_NS */
442 static DEFINE_IDA(net_generic_ids);
444 static int register_pernet_operations(struct list_head *list,
445 struct pernet_operations *ops)
451 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
453 if (error == -EAGAIN) {
454 ida_pre_get(&net_generic_ids, GFP_KERNEL);
460 error = __register_pernet_operations(list, ops);
464 ida_remove(&net_generic_ids, *ops->id);
470 static void unregister_pernet_operations(struct pernet_operations *ops)
473 __unregister_pernet_operations(ops);
476 ida_remove(&net_generic_ids, *ops->id);
480 * register_pernet_subsys - register a network namespace subsystem
481 * @ops: pernet operations structure for the subsystem
483 * Register a subsystem which has init and exit functions
484 * that are called when network namespaces are created and
485 * destroyed respectively.
487 * When registered all network namespace init functions are
488 * called for every existing network namespace. Allowing kernel
489 * modules to have a race free view of the set of network namespaces.
491 * When a new network namespace is created all of the init
492 * methods are called in the order in which they were registered.
494 * When a network namespace is destroyed all of the exit methods
495 * are called in the reverse of the order with which they were
498 int register_pernet_subsys(struct pernet_operations *ops)
501 mutex_lock(&net_mutex);
502 error = register_pernet_operations(first_device, ops);
503 mutex_unlock(&net_mutex);
506 EXPORT_SYMBOL_GPL(register_pernet_subsys);
509 * unregister_pernet_subsys - unregister a network namespace subsystem
510 * @ops: pernet operations structure to manipulate
512 * Remove the pernet operations structure from the list to be
513 * used when network namespaces are created or destroyed. In
514 * addition run the exit method for all existing network
517 void unregister_pernet_subsys(struct pernet_operations *ops)
519 mutex_lock(&net_mutex);
520 unregister_pernet_operations(ops);
521 mutex_unlock(&net_mutex);
523 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
526 * register_pernet_device - register a network namespace device
527 * @ops: pernet operations structure for the subsystem
529 * Register a device which has init and exit functions
530 * that are called when network namespaces are created and
531 * destroyed respectively.
533 * When registered all network namespace init functions are
534 * called for every existing network namespace. Allowing kernel
535 * modules to have a race free view of the set of network namespaces.
537 * When a new network namespace is created all of the init
538 * methods are called in the order in which they were registered.
540 * When a network namespace is destroyed all of the exit methods
541 * are called in the reverse of the order with which they were
544 int register_pernet_device(struct pernet_operations *ops)
547 mutex_lock(&net_mutex);
548 error = register_pernet_operations(&pernet_list, ops);
549 if (!error && (first_device == &pernet_list))
550 first_device = &ops->list;
551 mutex_unlock(&net_mutex);
554 EXPORT_SYMBOL_GPL(register_pernet_device);
557 * unregister_pernet_device - unregister a network namespace netdevice
558 * @ops: pernet operations structure to manipulate
560 * Remove the pernet operations structure from the list to be
561 * used when network namespaces are created or destroyed. In
562 * addition run the exit method for all existing network
565 void unregister_pernet_device(struct pernet_operations *ops)
567 mutex_lock(&net_mutex);
568 if (&ops->list == first_device)
569 first_device = first_device->next;
570 unregister_pernet_operations(ops);
571 mutex_unlock(&net_mutex);
573 EXPORT_SYMBOL_GPL(unregister_pernet_device);