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 static struct net *net_create(void)
226 return ERR_PTR(-ENOMEM);
227 mutex_lock(&net_mutex);
231 list_add_tail_rcu(&net->list, &net_namespace_list);
234 mutex_unlock(&net_mutex);
242 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
244 if (!(flags & CLONE_NEWNET))
245 return get_net(old_net);
249 static DEFINE_SPINLOCK(cleanup_list_lock);
250 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
252 static void cleanup_net(struct work_struct *work)
254 const struct pernet_operations *ops;
255 struct net *net, *tmp;
256 LIST_HEAD(net_kill_list);
257 LIST_HEAD(net_exit_list);
259 /* Atomically snapshot the list of namespaces to cleanup */
260 spin_lock_irq(&cleanup_list_lock);
261 list_replace_init(&cleanup_list, &net_kill_list);
262 spin_unlock_irq(&cleanup_list_lock);
264 mutex_lock(&net_mutex);
266 /* Don't let anyone else find us. */
268 list_for_each_entry(net, &net_kill_list, cleanup_list) {
269 list_del_rcu(&net->list);
270 list_add_tail(&net->exit_list, &net_exit_list);
275 * Another CPU might be rcu-iterating the list, wait for it.
276 * This needs to be before calling the exit() notifiers, so
277 * the rcu_barrier() below isn't sufficient alone.
281 /* Run all of the network namespace exit methods */
282 list_for_each_entry_reverse(ops, &pernet_list, list)
283 ops_exit_list(ops, &net_exit_list);
285 /* Free the net generic variables */
286 list_for_each_entry_reverse(ops, &pernet_list, list)
287 ops_free_list(ops, &net_exit_list);
289 mutex_unlock(&net_mutex);
291 /* Ensure there are no outstanding rcu callbacks using this
296 /* Finally it is safe to free my network namespace structure */
297 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
298 list_del_init(&net->exit_list);
302 static DECLARE_WORK(net_cleanup_work, cleanup_net);
304 void __put_net(struct net *net)
306 /* Cleanup the network namespace in process context */
309 spin_lock_irqsave(&cleanup_list_lock, flags);
310 list_add(&net->cleanup_list, &cleanup_list);
311 spin_unlock_irqrestore(&cleanup_list_lock, flags);
313 queue_work(netns_wq, &net_cleanup_work);
315 EXPORT_SYMBOL_GPL(__put_net);
318 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
320 if (flags & CLONE_NEWNET)
321 return ERR_PTR(-EINVAL);
326 struct net *get_net_ns_by_pid(pid_t pid)
328 struct task_struct *tsk;
331 /* Lookup the network namespace */
332 net = ERR_PTR(-ESRCH);
334 tsk = find_task_by_vpid(pid);
336 struct nsproxy *nsproxy;
337 nsproxy = task_nsproxy(tsk);
339 net = get_net(nsproxy->net_ns);
344 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
346 static int __init net_ns_init(void)
348 struct net_generic *ng;
351 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
355 /* Create workqueue for cleanup */
356 netns_wq = create_singlethread_workqueue("netns");
358 panic("Could not create netns workq");
361 ng = net_alloc_generic();
363 panic("Could not allocate generic netns");
365 rcu_assign_pointer(init_net.gen, ng);
367 mutex_lock(&net_mutex);
368 if (setup_net(&init_net))
369 panic("Could not setup the initial network namespace");
372 list_add_tail_rcu(&init_net.list, &net_namespace_list);
375 mutex_unlock(&net_mutex);
380 pure_initcall(net_ns_init);
383 static int __register_pernet_operations(struct list_head *list,
384 struct pernet_operations *ops)
388 LIST_HEAD(net_exit_list);
390 list_add_tail(&ops->list, list);
391 if (ops->init || (ops->id && ops->size)) {
393 error = ops_init(ops, net);
396 list_add_tail(&net->exit_list, &net_exit_list);
402 /* If I have an error cleanup all namespaces I initialized */
403 list_del(&ops->list);
404 ops_exit_list(ops, &net_exit_list);
405 ops_free_list(ops, &net_exit_list);
409 static void __unregister_pernet_operations(struct pernet_operations *ops)
412 LIST_HEAD(net_exit_list);
414 list_del(&ops->list);
416 list_add_tail(&net->exit_list, &net_exit_list);
417 ops_exit_list(ops, &net_exit_list);
418 ops_free_list(ops, &net_exit_list);
423 static int __register_pernet_operations(struct list_head *list,
424 struct pernet_operations *ops)
427 err = ops_init(ops, &init_net);
429 ops_free(ops, &init_net);
434 static void __unregister_pernet_operations(struct pernet_operations *ops)
436 LIST_HEAD(net_exit_list);
437 list_add(&init_net.exit_list, &net_exit_list);
438 ops_exit_list(ops, &net_exit_list);
439 ops_free_list(ops, &net_exit_list);
442 #endif /* CONFIG_NET_NS */
444 static DEFINE_IDA(net_generic_ids);
446 static int register_pernet_operations(struct list_head *list,
447 struct pernet_operations *ops)
453 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
455 if (error == -EAGAIN) {
456 ida_pre_get(&net_generic_ids, GFP_KERNEL);
462 error = __register_pernet_operations(list, ops);
466 ida_remove(&net_generic_ids, *ops->id);
472 static void unregister_pernet_operations(struct pernet_operations *ops)
475 __unregister_pernet_operations(ops);
478 ida_remove(&net_generic_ids, *ops->id);
482 * register_pernet_subsys - register a network namespace subsystem
483 * @ops: pernet operations structure for the subsystem
485 * Register a subsystem which has init and exit functions
486 * that are called when network namespaces are created and
487 * destroyed respectively.
489 * When registered all network namespace init functions are
490 * called for every existing network namespace. Allowing kernel
491 * modules to have a race free view of the set of network namespaces.
493 * When a new network namespace is created all of the init
494 * methods are called in the order in which they were registered.
496 * When a network namespace is destroyed all of the exit methods
497 * are called in the reverse of the order with which they were
500 int register_pernet_subsys(struct pernet_operations *ops)
503 mutex_lock(&net_mutex);
504 error = register_pernet_operations(first_device, ops);
505 mutex_unlock(&net_mutex);
508 EXPORT_SYMBOL_GPL(register_pernet_subsys);
511 * unregister_pernet_subsys - unregister a network namespace subsystem
512 * @ops: pernet operations structure to manipulate
514 * Remove the pernet operations structure from the list to be
515 * used when network namespaces are created or destroyed. In
516 * addition run the exit method for all existing network
519 void unregister_pernet_subsys(struct pernet_operations *ops)
521 mutex_lock(&net_mutex);
522 unregister_pernet_operations(ops);
523 mutex_unlock(&net_mutex);
525 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
528 * register_pernet_device - register a network namespace device
529 * @ops: pernet operations structure for the subsystem
531 * Register a device which has init and exit functions
532 * that are called when network namespaces are created and
533 * destroyed respectively.
535 * When registered all network namespace init functions are
536 * called for every existing network namespace. Allowing kernel
537 * modules to have a race free view of the set of network namespaces.
539 * When a new network namespace is created all of the init
540 * methods are called in the order in which they were registered.
542 * When a network namespace is destroyed all of the exit methods
543 * are called in the reverse of the order with which they were
546 int register_pernet_device(struct pernet_operations *ops)
549 mutex_lock(&net_mutex);
550 error = register_pernet_operations(&pernet_list, ops);
551 if (!error && (first_device == &pernet_list))
552 first_device = &ops->list;
553 mutex_unlock(&net_mutex);
556 EXPORT_SYMBOL_GPL(register_pernet_device);
559 * unregister_pernet_device - unregister a network namespace netdevice
560 * @ops: pernet operations structure to manipulate
562 * Remove the pernet operations structure from the list to be
563 * used when network namespaces are created or destroyed. In
564 * addition run the exit method for all existing network
567 void unregister_pernet_device(struct pernet_operations *ops)
569 mutex_lock(&net_mutex);
570 if (&ops->list == first_device)
571 first_device = first_device->next;
572 unregister_pernet_operations(ops);
573 mutex_unlock(&net_mutex);
575 EXPORT_SYMBOL_GPL(unregister_pernet_device);