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 old_ng = rcu_dereference_protected(net->gen,
46 lockdep_is_held(&net_mutex));
48 if (old_ng->len >= id)
51 ng = kzalloc(sizeof(struct net_generic) +
52 id * sizeof(void *), GFP_KERNEL);
57 * Some synchronisation notes:
59 * The net_generic explores the net->gen array inside rcu
60 * read section. Besides once set the net->gen->ptr[x]
61 * pointer never changes (see rules in netns/generic.h).
63 * That said, we simply duplicate this array and schedule
64 * the old copy for kfree after a grace period.
68 memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
70 rcu_assign_pointer(net->gen, ng);
71 call_rcu(&old_ng->rcu, net_generic_release);
73 ng->ptr[id - 1] = data;
77 static int ops_init(const struct pernet_operations *ops, struct net *net)
80 if (ops->id && ops->size) {
81 void *data = kzalloc(ops->size, GFP_KERNEL);
85 err = net_assign_generic(net, *ops->id, data);
92 return ops->init(net);
96 static void ops_free(const struct pernet_operations *ops, struct net *net)
98 if (ops->id && ops->size) {
100 kfree(net_generic(net, id));
104 static void ops_exit_list(const struct pernet_operations *ops,
105 struct list_head *net_exit_list)
109 list_for_each_entry(net, net_exit_list, exit_list)
113 ops->exit_batch(net_exit_list);
116 static void ops_free_list(const struct pernet_operations *ops,
117 struct list_head *net_exit_list)
120 if (ops->size && ops->id) {
121 list_for_each_entry(net, net_exit_list, exit_list)
127 * setup_net runs the initializers for the network namespace object.
129 static __net_init int setup_net(struct net *net)
131 /* Must be called with net_mutex held */
132 const struct pernet_operations *ops, *saved_ops;
134 LIST_HEAD(net_exit_list);
136 atomic_set(&net->count, 1);
138 #ifdef NETNS_REFCNT_DEBUG
139 atomic_set(&net->use_count, 0);
142 list_for_each_entry(ops, &pernet_list, list) {
143 error = ops_init(ops, net);
151 /* Walk through the list backwards calling the exit functions
152 * for the pernet modules whose init functions did not fail.
154 list_add(&net->exit_list, &net_exit_list);
156 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
157 ops_exit_list(ops, &net_exit_list);
160 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
161 ops_free_list(ops, &net_exit_list);
167 static struct net_generic *net_alloc_generic(void)
169 struct net_generic *ng;
170 size_t generic_size = sizeof(struct net_generic) +
171 INITIAL_NET_GEN_PTRS * sizeof(void *);
173 ng = kzalloc(generic_size, GFP_KERNEL);
175 ng->len = INITIAL_NET_GEN_PTRS;
181 static struct kmem_cache *net_cachep;
182 static struct workqueue_struct *netns_wq;
184 static struct net *net_alloc(void)
186 struct net *net = NULL;
187 struct net_generic *ng;
189 ng = net_alloc_generic();
193 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
197 rcu_assign_pointer(net->gen, ng);
206 static void net_free(struct net *net)
208 #ifdef NETNS_REFCNT_DEBUG
209 if (unlikely(atomic_read(&net->use_count) != 0)) {
210 printk(KERN_EMERG "network namespace not free! Usage: %d\n",
211 atomic_read(&net->use_count));
216 kmem_cache_free(net_cachep, net);
219 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
224 if (!(flags & CLONE_NEWNET))
225 return get_net(old_net);
229 return ERR_PTR(-ENOMEM);
230 mutex_lock(&net_mutex);
234 list_add_tail_rcu(&net->list, &net_namespace_list);
237 mutex_unlock(&net_mutex);
245 static DEFINE_SPINLOCK(cleanup_list_lock);
246 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
248 static void cleanup_net(struct work_struct *work)
250 const struct pernet_operations *ops;
251 struct net *net, *tmp;
252 LIST_HEAD(net_kill_list);
253 LIST_HEAD(net_exit_list);
255 /* Atomically snapshot the list of namespaces to cleanup */
256 spin_lock_irq(&cleanup_list_lock);
257 list_replace_init(&cleanup_list, &net_kill_list);
258 spin_unlock_irq(&cleanup_list_lock);
260 mutex_lock(&net_mutex);
262 /* Don't let anyone else find us. */
264 list_for_each_entry(net, &net_kill_list, cleanup_list) {
265 list_del_rcu(&net->list);
266 list_add_tail(&net->exit_list, &net_exit_list);
271 * Another CPU might be rcu-iterating the list, wait for it.
272 * This needs to be before calling the exit() notifiers, so
273 * the rcu_barrier() below isn't sufficient alone.
277 /* Run all of the network namespace exit methods */
278 list_for_each_entry_reverse(ops, &pernet_list, list)
279 ops_exit_list(ops, &net_exit_list);
281 /* Free the net generic variables */
282 list_for_each_entry_reverse(ops, &pernet_list, list)
283 ops_free_list(ops, &net_exit_list);
285 mutex_unlock(&net_mutex);
287 /* Ensure there are no outstanding rcu callbacks using this
292 /* Finally it is safe to free my network namespace structure */
293 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
294 list_del_init(&net->exit_list);
298 static DECLARE_WORK(net_cleanup_work, cleanup_net);
300 void __put_net(struct net *net)
302 /* Cleanup the network namespace in process context */
305 spin_lock_irqsave(&cleanup_list_lock, flags);
306 list_add(&net->cleanup_list, &cleanup_list);
307 spin_unlock_irqrestore(&cleanup_list_lock, flags);
309 queue_work(netns_wq, &net_cleanup_work);
311 EXPORT_SYMBOL_GPL(__put_net);
314 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
316 if (flags & CLONE_NEWNET)
317 return ERR_PTR(-EINVAL);
322 struct net *get_net_ns_by_pid(pid_t pid)
324 struct task_struct *tsk;
327 /* Lookup the network namespace */
328 net = ERR_PTR(-ESRCH);
330 tsk = find_task_by_vpid(pid);
332 struct nsproxy *nsproxy;
333 nsproxy = task_nsproxy(tsk);
335 net = get_net(nsproxy->net_ns);
340 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
342 static int __init net_ns_init(void)
344 struct net_generic *ng;
347 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
351 /* Create workqueue for cleanup */
352 netns_wq = create_singlethread_workqueue("netns");
354 panic("Could not create netns workq");
357 ng = net_alloc_generic();
359 panic("Could not allocate generic netns");
361 rcu_assign_pointer(init_net.gen, ng);
363 mutex_lock(&net_mutex);
364 if (setup_net(&init_net))
365 panic("Could not setup the initial network namespace");
368 list_add_tail_rcu(&init_net.list, &net_namespace_list);
371 mutex_unlock(&net_mutex);
376 pure_initcall(net_ns_init);
379 static int __register_pernet_operations(struct list_head *list,
380 struct pernet_operations *ops)
384 LIST_HEAD(net_exit_list);
386 list_add_tail(&ops->list, list);
387 if (ops->init || (ops->id && ops->size)) {
389 error = ops_init(ops, net);
392 list_add_tail(&net->exit_list, &net_exit_list);
398 /* If I have an error cleanup all namespaces I initialized */
399 list_del(&ops->list);
400 ops_exit_list(ops, &net_exit_list);
401 ops_free_list(ops, &net_exit_list);
405 static void __unregister_pernet_operations(struct pernet_operations *ops)
408 LIST_HEAD(net_exit_list);
410 list_del(&ops->list);
412 list_add_tail(&net->exit_list, &net_exit_list);
413 ops_exit_list(ops, &net_exit_list);
414 ops_free_list(ops, &net_exit_list);
419 static int __register_pernet_operations(struct list_head *list,
420 struct pernet_operations *ops)
423 err = ops_init(ops, &init_net);
425 ops_free(ops, &init_net);
430 static void __unregister_pernet_operations(struct pernet_operations *ops)
432 LIST_HEAD(net_exit_list);
433 list_add(&init_net.exit_list, &net_exit_list);
434 ops_exit_list(ops, &net_exit_list);
435 ops_free_list(ops, &net_exit_list);
438 #endif /* CONFIG_NET_NS */
440 static DEFINE_IDA(net_generic_ids);
442 static int register_pernet_operations(struct list_head *list,
443 struct pernet_operations *ops)
449 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
451 if (error == -EAGAIN) {
452 ida_pre_get(&net_generic_ids, GFP_KERNEL);
458 error = __register_pernet_operations(list, ops);
462 ida_remove(&net_generic_ids, *ops->id);
468 static void unregister_pernet_operations(struct pernet_operations *ops)
471 __unregister_pernet_operations(ops);
474 ida_remove(&net_generic_ids, *ops->id);
478 * register_pernet_subsys - register a network namespace subsystem
479 * @ops: pernet operations structure for the subsystem
481 * Register a subsystem which has init and exit functions
482 * that are called when network namespaces are created and
483 * destroyed respectively.
485 * When registered all network namespace init functions are
486 * called for every existing network namespace. Allowing kernel
487 * modules to have a race free view of the set of network namespaces.
489 * When a new network namespace is created all of the init
490 * methods are called in the order in which they were registered.
492 * When a network namespace is destroyed all of the exit methods
493 * are called in the reverse of the order with which they were
496 int register_pernet_subsys(struct pernet_operations *ops)
499 mutex_lock(&net_mutex);
500 error = register_pernet_operations(first_device, ops);
501 mutex_unlock(&net_mutex);
504 EXPORT_SYMBOL_GPL(register_pernet_subsys);
507 * unregister_pernet_subsys - unregister a network namespace subsystem
508 * @ops: pernet operations structure to manipulate
510 * Remove the pernet operations structure from the list to be
511 * used when network namespaces are created or destroyed. In
512 * addition run the exit method for all existing network
515 void unregister_pernet_subsys(struct pernet_operations *ops)
517 mutex_lock(&net_mutex);
518 unregister_pernet_operations(ops);
519 mutex_unlock(&net_mutex);
521 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
524 * register_pernet_device - register a network namespace device
525 * @ops: pernet operations structure for the subsystem
527 * Register a device which has init and exit functions
528 * that are called when network namespaces are created and
529 * destroyed respectively.
531 * When registered all network namespace init functions are
532 * called for every existing network namespace. Allowing kernel
533 * modules to have a race free view of the set of network namespaces.
535 * When a new network namespace is created all of the init
536 * methods are called in the order in which they were registered.
538 * When a network namespace is destroyed all of the exit methods
539 * are called in the reverse of the order with which they were
542 int register_pernet_device(struct pernet_operations *ops)
545 mutex_lock(&net_mutex);
546 error = register_pernet_operations(&pernet_list, ops);
547 if (!error && (first_device == &pernet_list))
548 first_device = &ops->list;
549 mutex_unlock(&net_mutex);
552 EXPORT_SYMBOL_GPL(register_pernet_device);
555 * unregister_pernet_device - unregister a network namespace netdevice
556 * @ops: pernet operations structure to manipulate
558 * Remove the pernet operations structure from the list to be
559 * used when network namespaces are created or destroyed. In
560 * addition run the exit method for all existing network
563 void unregister_pernet_device(struct pernet_operations *ops)
565 mutex_lock(&net_mutex);
566 if (&ops->list == first_device)
567 first_device = first_device->next;
568 unregister_pernet_operations(ops);
569 mutex_unlock(&net_mutex);
571 EXPORT_SYMBOL_GPL(unregister_pernet_device);