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 int net_assign_generic(struct net *net, int id, void *data)
32 struct net_generic *ng, *old_ng;
34 BUG_ON(!mutex_is_locked(&net_mutex));
37 old_ng = rcu_dereference_protected(net->gen,
38 lockdep_is_held(&net_mutex));
40 if (old_ng->len >= id)
43 ng = kzalloc(sizeof(struct net_generic) +
44 id * sizeof(void *), GFP_KERNEL);
49 * Some synchronisation notes:
51 * The net_generic explores the net->gen array inside rcu
52 * read section. Besides once set the net->gen->ptr[x]
53 * pointer never changes (see rules in netns/generic.h).
55 * That said, we simply duplicate this array and schedule
56 * the old copy for kfree after a grace period.
60 memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
62 rcu_assign_pointer(net->gen, ng);
63 kfree_rcu(old_ng, rcu);
65 ng->ptr[id - 1] = data;
69 static int ops_init(const struct pernet_operations *ops, struct net *net)
72 if (ops->id && ops->size) {
73 void *data = kzalloc(ops->size, GFP_KERNEL);
77 err = net_assign_generic(net, *ops->id, data);
84 return ops->init(net);
88 static void ops_free(const struct pernet_operations *ops, struct net *net)
90 if (ops->id && ops->size) {
92 kfree(net_generic(net, id));
96 static void ops_exit_list(const struct pernet_operations *ops,
97 struct list_head *net_exit_list)
101 list_for_each_entry(net, net_exit_list, exit_list)
105 ops->exit_batch(net_exit_list);
108 static void ops_free_list(const struct pernet_operations *ops,
109 struct list_head *net_exit_list)
112 if (ops->size && ops->id) {
113 list_for_each_entry(net, net_exit_list, exit_list)
119 * setup_net runs the initializers for the network namespace object.
121 static __net_init int setup_net(struct net *net)
123 /* Must be called with net_mutex held */
124 const struct pernet_operations *ops, *saved_ops;
126 LIST_HEAD(net_exit_list);
128 atomic_set(&net->count, 1);
130 #ifdef NETNS_REFCNT_DEBUG
131 atomic_set(&net->use_count, 0);
134 list_for_each_entry(ops, &pernet_list, list) {
135 error = ops_init(ops, net);
143 /* Walk through the list backwards calling the exit functions
144 * for the pernet modules whose init functions did not fail.
146 list_add(&net->exit_list, &net_exit_list);
148 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
149 ops_exit_list(ops, &net_exit_list);
152 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
153 ops_free_list(ops, &net_exit_list);
159 static struct net_generic *net_alloc_generic(void)
161 struct net_generic *ng;
162 size_t generic_size = sizeof(struct net_generic) +
163 INITIAL_NET_GEN_PTRS * sizeof(void *);
165 ng = kzalloc(generic_size, GFP_KERNEL);
167 ng->len = INITIAL_NET_GEN_PTRS;
173 static struct kmem_cache *net_cachep;
174 static struct workqueue_struct *netns_wq;
176 static struct net *net_alloc(void)
178 struct net *net = NULL;
179 struct net_generic *ng;
181 ng = net_alloc_generic();
185 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
189 rcu_assign_pointer(net->gen, ng);
198 static void net_free(struct net *net)
200 #ifdef NETNS_REFCNT_DEBUG
201 if (unlikely(atomic_read(&net->use_count) != 0)) {
202 printk(KERN_EMERG "network namespace not free! Usage: %d\n",
203 atomic_read(&net->use_count));
208 kmem_cache_free(net_cachep, net);
211 static struct net *net_create(void)
218 return ERR_PTR(-ENOMEM);
219 mutex_lock(&net_mutex);
223 list_add_tail_rcu(&net->list, &net_namespace_list);
226 mutex_unlock(&net_mutex);
234 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
236 if (!(flags & CLONE_NEWNET))
237 return get_net(old_net);
241 static DEFINE_SPINLOCK(cleanup_list_lock);
242 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
244 static void cleanup_net(struct work_struct *work)
246 const struct pernet_operations *ops;
247 struct net *net, *tmp;
248 LIST_HEAD(net_kill_list);
249 LIST_HEAD(net_exit_list);
251 /* Atomically snapshot the list of namespaces to cleanup */
252 spin_lock_irq(&cleanup_list_lock);
253 list_replace_init(&cleanup_list, &net_kill_list);
254 spin_unlock_irq(&cleanup_list_lock);
256 mutex_lock(&net_mutex);
258 /* Don't let anyone else find us. */
260 list_for_each_entry(net, &net_kill_list, cleanup_list) {
261 list_del_rcu(&net->list);
262 list_add_tail(&net->exit_list, &net_exit_list);
267 * Another CPU might be rcu-iterating the list, wait for it.
268 * This needs to be before calling the exit() notifiers, so
269 * the rcu_barrier() below isn't sufficient alone.
273 /* Run all of the network namespace exit methods */
274 list_for_each_entry_reverse(ops, &pernet_list, list)
275 ops_exit_list(ops, &net_exit_list);
277 /* Free the net generic variables */
278 list_for_each_entry_reverse(ops, &pernet_list, list)
279 ops_free_list(ops, &net_exit_list);
281 mutex_unlock(&net_mutex);
283 /* Ensure there are no outstanding rcu callbacks using this
288 /* Finally it is safe to free my network namespace structure */
289 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
290 list_del_init(&net->exit_list);
294 static DECLARE_WORK(net_cleanup_work, cleanup_net);
296 void __put_net(struct net *net)
298 /* Cleanup the network namespace in process context */
301 spin_lock_irqsave(&cleanup_list_lock, flags);
302 list_add(&net->cleanup_list, &cleanup_list);
303 spin_unlock_irqrestore(&cleanup_list_lock, flags);
305 queue_work(netns_wq, &net_cleanup_work);
307 EXPORT_SYMBOL_GPL(__put_net);
310 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
312 if (flags & CLONE_NEWNET)
313 return ERR_PTR(-EINVAL);
318 struct net *get_net_ns_by_pid(pid_t pid)
320 struct task_struct *tsk;
323 /* Lookup the network namespace */
324 net = ERR_PTR(-ESRCH);
326 tsk = find_task_by_vpid(pid);
328 struct nsproxy *nsproxy;
329 nsproxy = task_nsproxy(tsk);
331 net = get_net(nsproxy->net_ns);
336 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
338 static int __init net_ns_init(void)
340 struct net_generic *ng;
343 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
347 /* Create workqueue for cleanup */
348 netns_wq = create_singlethread_workqueue("netns");
350 panic("Could not create netns workq");
353 ng = net_alloc_generic();
355 panic("Could not allocate generic netns");
357 rcu_assign_pointer(init_net.gen, ng);
359 mutex_lock(&net_mutex);
360 if (setup_net(&init_net))
361 panic("Could not setup the initial network namespace");
364 list_add_tail_rcu(&init_net.list, &net_namespace_list);
367 mutex_unlock(&net_mutex);
372 pure_initcall(net_ns_init);
375 static int __register_pernet_operations(struct list_head *list,
376 struct pernet_operations *ops)
380 LIST_HEAD(net_exit_list);
382 list_add_tail(&ops->list, list);
383 if (ops->init || (ops->id && ops->size)) {
385 error = ops_init(ops, net);
388 list_add_tail(&net->exit_list, &net_exit_list);
394 /* If I have an error cleanup all namespaces I initialized */
395 list_del(&ops->list);
396 ops_exit_list(ops, &net_exit_list);
397 ops_free_list(ops, &net_exit_list);
401 static void __unregister_pernet_operations(struct pernet_operations *ops)
404 LIST_HEAD(net_exit_list);
406 list_del(&ops->list);
408 list_add_tail(&net->exit_list, &net_exit_list);
409 ops_exit_list(ops, &net_exit_list);
410 ops_free_list(ops, &net_exit_list);
415 static int __register_pernet_operations(struct list_head *list,
416 struct pernet_operations *ops)
419 err = ops_init(ops, &init_net);
421 ops_free(ops, &init_net);
426 static void __unregister_pernet_operations(struct pernet_operations *ops)
428 LIST_HEAD(net_exit_list);
429 list_add(&init_net.exit_list, &net_exit_list);
430 ops_exit_list(ops, &net_exit_list);
431 ops_free_list(ops, &net_exit_list);
434 #endif /* CONFIG_NET_NS */
436 static DEFINE_IDA(net_generic_ids);
438 static int register_pernet_operations(struct list_head *list,
439 struct pernet_operations *ops)
445 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
447 if (error == -EAGAIN) {
448 ida_pre_get(&net_generic_ids, GFP_KERNEL);
454 error = __register_pernet_operations(list, ops);
458 ida_remove(&net_generic_ids, *ops->id);
464 static void unregister_pernet_operations(struct pernet_operations *ops)
467 __unregister_pernet_operations(ops);
470 ida_remove(&net_generic_ids, *ops->id);
474 * register_pernet_subsys - register a network namespace subsystem
475 * @ops: pernet operations structure for the subsystem
477 * Register a subsystem which has init and exit functions
478 * that are called when network namespaces are created and
479 * destroyed respectively.
481 * When registered all network namespace init functions are
482 * called for every existing network namespace. Allowing kernel
483 * modules to have a race free view of the set of network namespaces.
485 * When a new network namespace is created all of the init
486 * methods are called in the order in which they were registered.
488 * When a network namespace is destroyed all of the exit methods
489 * are called in the reverse of the order with which they were
492 int register_pernet_subsys(struct pernet_operations *ops)
495 mutex_lock(&net_mutex);
496 error = register_pernet_operations(first_device, ops);
497 mutex_unlock(&net_mutex);
500 EXPORT_SYMBOL_GPL(register_pernet_subsys);
503 * unregister_pernet_subsys - unregister a network namespace subsystem
504 * @ops: pernet operations structure to manipulate
506 * Remove the pernet operations structure from the list to be
507 * used when network namespaces are created or destroyed. In
508 * addition run the exit method for all existing network
511 void unregister_pernet_subsys(struct pernet_operations *ops)
513 mutex_lock(&net_mutex);
514 unregister_pernet_operations(ops);
515 mutex_unlock(&net_mutex);
517 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
520 * register_pernet_device - register a network namespace device
521 * @ops: pernet operations structure for the subsystem
523 * Register a device which has init and exit functions
524 * that are called when network namespaces are created and
525 * destroyed respectively.
527 * When registered all network namespace init functions are
528 * called for every existing network namespace. Allowing kernel
529 * modules to have a race free view of the set of network namespaces.
531 * When a new network namespace is created all of the init
532 * methods are called in the order in which they were registered.
534 * When a network namespace is destroyed all of the exit methods
535 * are called in the reverse of the order with which they were
538 int register_pernet_device(struct pernet_operations *ops)
541 mutex_lock(&net_mutex);
542 error = register_pernet_operations(&pernet_list, ops);
543 if (!error && (first_device == &pernet_list))
544 first_device = &ops->list;
545 mutex_unlock(&net_mutex);
548 EXPORT_SYMBOL_GPL(register_pernet_device);
551 * unregister_pernet_device - unregister a network namespace netdevice
552 * @ops: pernet operations structure to manipulate
554 * Remove the pernet operations structure from the list to be
555 * used when network namespaces are created or destroyed. In
556 * addition run the exit method for all existing network
559 void unregister_pernet_device(struct pernet_operations *ops)
561 mutex_lock(&net_mutex);
562 if (&ops->list == first_device)
563 first_device = first_device->next;
564 unregister_pernet_operations(ops);
565 mutex_unlock(&net_mutex);
567 EXPORT_SYMBOL_GPL(unregister_pernet_device);