1 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
3 #include <linux/workqueue.h>
4 #include <linux/rtnetlink.h>
5 #include <linux/cache.h>
6 #include <linux/slab.h>
7 #include <linux/list.h>
8 #include <linux/delay.h>
9 #include <linux/sched.h>
10 #include <linux/idr.h>
11 #include <linux/rculist.h>
12 #include <linux/nsproxy.h>
14 #include <linux/proc_ns.h>
15 #include <linux/file.h>
16 #include <linux/export.h>
17 #include <linux/user_namespace.h>
18 #include <net/net_namespace.h>
19 #include <net/netns/generic.h>
22 * Our network namespace constructor/destructor lists
25 static LIST_HEAD(pernet_list);
26 static struct list_head *first_device = &pernet_list;
27 DEFINE_MUTEX(net_mutex);
29 LIST_HEAD(net_namespace_list);
30 EXPORT_SYMBOL_GPL(net_namespace_list);
32 struct net init_net = {
33 .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head),
35 EXPORT_SYMBOL(init_net);
37 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
39 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
41 static struct net_generic *net_alloc_generic(void)
43 struct net_generic *ng;
44 size_t generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
46 ng = kzalloc(generic_size, GFP_KERNEL);
48 ng->len = max_gen_ptrs;
53 static int net_assign_generic(struct net *net, int id, void *data)
55 struct net_generic *ng, *old_ng;
57 BUG_ON(!mutex_is_locked(&net_mutex));
60 old_ng = rcu_dereference_protected(net->gen,
61 lockdep_is_held(&net_mutex));
63 if (old_ng->len >= id)
66 ng = net_alloc_generic();
71 * Some synchronisation notes:
73 * The net_generic explores the net->gen array inside rcu
74 * read section. Besides once set the net->gen->ptr[x]
75 * pointer never changes (see rules in netns/generic.h).
77 * That said, we simply duplicate this array and schedule
78 * the old copy for kfree after a grace period.
81 memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
83 rcu_assign_pointer(net->gen, ng);
84 kfree_rcu(old_ng, rcu);
86 ng->ptr[id - 1] = data;
90 static int ops_init(const struct pernet_operations *ops, struct net *net)
95 if (ops->id && ops->size) {
96 data = kzalloc(ops->size, GFP_KERNEL);
100 err = net_assign_generic(net, *ops->id, data);
106 err = ops->init(net);
117 static void ops_free(const struct pernet_operations *ops, struct net *net)
119 if (ops->id && ops->size) {
121 kfree(net_generic(net, id));
125 static void ops_exit_list(const struct pernet_operations *ops,
126 struct list_head *net_exit_list)
130 list_for_each_entry(net, net_exit_list, exit_list)
134 ops->exit_batch(net_exit_list);
137 static void ops_free_list(const struct pernet_operations *ops,
138 struct list_head *net_exit_list)
141 if (ops->size && ops->id) {
142 list_for_each_entry(net, net_exit_list, exit_list)
148 * setup_net runs the initializers for the network namespace object.
150 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
152 /* Must be called with net_mutex held */
153 const struct pernet_operations *ops, *saved_ops;
155 LIST_HEAD(net_exit_list);
157 atomic_set(&net->count, 1);
158 atomic_set(&net->passive, 1);
159 net->dev_base_seq = 1;
160 net->user_ns = user_ns;
162 #ifdef NETNS_REFCNT_DEBUG
163 atomic_set(&net->use_count, 0);
166 list_for_each_entry(ops, &pernet_list, list) {
167 error = ops_init(ops, net);
175 /* Walk through the list backwards calling the exit functions
176 * for the pernet modules whose init functions did not fail.
178 list_add(&net->exit_list, &net_exit_list);
180 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
181 ops_exit_list(ops, &net_exit_list);
184 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
185 ops_free_list(ops, &net_exit_list);
193 static struct kmem_cache *net_cachep;
194 static struct workqueue_struct *netns_wq;
196 static struct net *net_alloc(void)
198 struct net *net = NULL;
199 struct net_generic *ng;
201 ng = net_alloc_generic();
205 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
209 rcu_assign_pointer(net->gen, ng);
218 static void net_free(struct net *net)
220 #ifdef NETNS_REFCNT_DEBUG
221 if (unlikely(atomic_read(&net->use_count) != 0)) {
222 pr_emerg("network namespace not free! Usage: %d\n",
223 atomic_read(&net->use_count));
227 kfree(rcu_access_pointer(net->gen));
228 kmem_cache_free(net_cachep, net);
231 void net_drop_ns(void *p)
234 if (ns && atomic_dec_and_test(&ns->passive))
238 struct net *copy_net_ns(unsigned long flags,
239 struct user_namespace *user_ns, struct net *old_net)
244 if (!(flags & CLONE_NEWNET))
245 return get_net(old_net);
249 return ERR_PTR(-ENOMEM);
251 get_user_ns(user_ns);
253 mutex_lock(&net_mutex);
254 rv = setup_net(net, user_ns);
257 list_add_tail_rcu(&net->list, &net_namespace_list);
260 mutex_unlock(&net_mutex);
262 put_user_ns(user_ns);
269 static DEFINE_SPINLOCK(cleanup_list_lock);
270 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
272 static void cleanup_net(struct work_struct *work)
274 const struct pernet_operations *ops;
275 struct net *net, *tmp;
276 struct list_head net_kill_list;
277 LIST_HEAD(net_exit_list);
279 /* Atomically snapshot the list of namespaces to cleanup */
280 spin_lock_irq(&cleanup_list_lock);
281 list_replace_init(&cleanup_list, &net_kill_list);
282 spin_unlock_irq(&cleanup_list_lock);
284 mutex_lock(&net_mutex);
286 /* Don't let anyone else find us. */
288 list_for_each_entry(net, &net_kill_list, cleanup_list) {
289 list_del_rcu(&net->list);
290 list_add_tail(&net->exit_list, &net_exit_list);
295 * Another CPU might be rcu-iterating the list, wait for it.
296 * This needs to be before calling the exit() notifiers, so
297 * the rcu_barrier() below isn't sufficient alone.
301 /* Run all of the network namespace exit methods */
302 list_for_each_entry_reverse(ops, &pernet_list, list)
303 ops_exit_list(ops, &net_exit_list);
305 /* Free the net generic variables */
306 list_for_each_entry_reverse(ops, &pernet_list, list)
307 ops_free_list(ops, &net_exit_list);
309 mutex_unlock(&net_mutex);
311 /* Ensure there are no outstanding rcu callbacks using this
316 /* Finally it is safe to free my network namespace structure */
317 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
318 list_del_init(&net->exit_list);
319 put_user_ns(net->user_ns);
323 static DECLARE_WORK(net_cleanup_work, cleanup_net);
325 void __put_net(struct net *net)
327 /* Cleanup the network namespace in process context */
330 spin_lock_irqsave(&cleanup_list_lock, flags);
331 list_add(&net->cleanup_list, &cleanup_list);
332 spin_unlock_irqrestore(&cleanup_list_lock, flags);
334 queue_work(netns_wq, &net_cleanup_work);
336 EXPORT_SYMBOL_GPL(__put_net);
338 struct net *get_net_ns_by_fd(int fd)
344 file = proc_ns_fget(fd);
346 return ERR_CAST(file);
348 ei = get_proc_ns(file_inode(file));
349 if (ei->ns_ops == &netns_operations)
350 net = get_net(ei->ns);
352 net = ERR_PTR(-EINVAL);
359 struct net *get_net_ns_by_fd(int fd)
361 return ERR_PTR(-EINVAL);
365 struct net *get_net_ns_by_pid(pid_t pid)
367 struct task_struct *tsk;
370 /* Lookup the network namespace */
371 net = ERR_PTR(-ESRCH);
373 tsk = find_task_by_vpid(pid);
375 struct nsproxy *nsproxy;
377 nsproxy = tsk->nsproxy;
379 net = get_net(nsproxy->net_ns);
385 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
387 static __net_init int net_ns_net_init(struct net *net)
389 return proc_alloc_inum(&net->proc_inum);
392 static __net_exit void net_ns_net_exit(struct net *net)
394 proc_free_inum(net->proc_inum);
397 static struct pernet_operations __net_initdata net_ns_ops = {
398 .init = net_ns_net_init,
399 .exit = net_ns_net_exit,
402 static int __init net_ns_init(void)
404 struct net_generic *ng;
407 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
411 /* Create workqueue for cleanup */
412 netns_wq = create_singlethread_workqueue("netns");
414 panic("Could not create netns workq");
417 ng = net_alloc_generic();
419 panic("Could not allocate generic netns");
421 rcu_assign_pointer(init_net.gen, ng);
423 mutex_lock(&net_mutex);
424 if (setup_net(&init_net, &init_user_ns))
425 panic("Could not setup the initial network namespace");
428 list_add_tail_rcu(&init_net.list, &net_namespace_list);
431 mutex_unlock(&net_mutex);
433 register_pernet_subsys(&net_ns_ops);
438 pure_initcall(net_ns_init);
441 static int __register_pernet_operations(struct list_head *list,
442 struct pernet_operations *ops)
446 LIST_HEAD(net_exit_list);
448 list_add_tail(&ops->list, list);
449 if (ops->init || (ops->id && ops->size)) {
451 error = ops_init(ops, net);
454 list_add_tail(&net->exit_list, &net_exit_list);
460 /* If I have an error cleanup all namespaces I initialized */
461 list_del(&ops->list);
462 ops_exit_list(ops, &net_exit_list);
463 ops_free_list(ops, &net_exit_list);
467 static void __unregister_pernet_operations(struct pernet_operations *ops)
470 LIST_HEAD(net_exit_list);
472 list_del(&ops->list);
474 list_add_tail(&net->exit_list, &net_exit_list);
475 ops_exit_list(ops, &net_exit_list);
476 ops_free_list(ops, &net_exit_list);
481 static int __register_pernet_operations(struct list_head *list,
482 struct pernet_operations *ops)
484 return ops_init(ops, &init_net);
487 static void __unregister_pernet_operations(struct pernet_operations *ops)
489 LIST_HEAD(net_exit_list);
490 list_add(&init_net.exit_list, &net_exit_list);
491 ops_exit_list(ops, &net_exit_list);
492 ops_free_list(ops, &net_exit_list);
495 #endif /* CONFIG_NET_NS */
497 static DEFINE_IDA(net_generic_ids);
499 static int register_pernet_operations(struct list_head *list,
500 struct pernet_operations *ops)
506 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
508 if (error == -EAGAIN) {
509 ida_pre_get(&net_generic_ids, GFP_KERNEL);
514 max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
516 error = __register_pernet_operations(list, ops);
520 ida_remove(&net_generic_ids, *ops->id);
526 static void unregister_pernet_operations(struct pernet_operations *ops)
529 __unregister_pernet_operations(ops);
532 ida_remove(&net_generic_ids, *ops->id);
536 * register_pernet_subsys - register a network namespace subsystem
537 * @ops: pernet operations structure for the subsystem
539 * Register a subsystem which has init and exit functions
540 * that are called when network namespaces are created and
541 * destroyed respectively.
543 * When registered all network namespace init functions are
544 * called for every existing network namespace. Allowing kernel
545 * modules to have a race free view of the set of network namespaces.
547 * When a new network namespace is created all of the init
548 * methods are called in the order in which they were registered.
550 * When a network namespace is destroyed all of the exit methods
551 * are called in the reverse of the order with which they were
554 int register_pernet_subsys(struct pernet_operations *ops)
557 mutex_lock(&net_mutex);
558 error = register_pernet_operations(first_device, ops);
559 mutex_unlock(&net_mutex);
562 EXPORT_SYMBOL_GPL(register_pernet_subsys);
565 * unregister_pernet_subsys - unregister a network namespace subsystem
566 * @ops: pernet operations structure to manipulate
568 * Remove the pernet operations structure from the list to be
569 * used when network namespaces are created or destroyed. In
570 * addition run the exit method for all existing network
573 void unregister_pernet_subsys(struct pernet_operations *ops)
575 mutex_lock(&net_mutex);
576 unregister_pernet_operations(ops);
577 mutex_unlock(&net_mutex);
579 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
582 * register_pernet_device - register a network namespace device
583 * @ops: pernet operations structure for the subsystem
585 * Register a device which has init and exit functions
586 * that are called when network namespaces are created and
587 * destroyed respectively.
589 * When registered all network namespace init functions are
590 * called for every existing network namespace. Allowing kernel
591 * modules to have a race free view of the set of network namespaces.
593 * When a new network namespace is created all of the init
594 * methods are called in the order in which they were registered.
596 * When a network namespace is destroyed all of the exit methods
597 * are called in the reverse of the order with which they were
600 int register_pernet_device(struct pernet_operations *ops)
603 mutex_lock(&net_mutex);
604 error = register_pernet_operations(&pernet_list, ops);
605 if (!error && (first_device == &pernet_list))
606 first_device = &ops->list;
607 mutex_unlock(&net_mutex);
610 EXPORT_SYMBOL_GPL(register_pernet_device);
613 * unregister_pernet_device - unregister a network namespace netdevice
614 * @ops: pernet operations structure to manipulate
616 * Remove the pernet operations structure from the list to be
617 * used when network namespaces are created or destroyed. In
618 * addition run the exit method for all existing network
621 void unregister_pernet_device(struct pernet_operations *ops)
623 mutex_lock(&net_mutex);
624 if (&ops->list == first_device)
625 first_device = first_device->next;
626 unregister_pernet_operations(ops);
627 mutex_unlock(&net_mutex);
629 EXPORT_SYMBOL_GPL(unregister_pernet_device);
632 static void *netns_get(struct task_struct *task)
634 struct net *net = NULL;
635 struct nsproxy *nsproxy;
638 nsproxy = task->nsproxy;
640 net = get_net(nsproxy->net_ns);
646 static void netns_put(void *ns)
651 static int netns_install(struct nsproxy *nsproxy, void *ns)
653 struct net *net = ns;
655 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
656 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
659 put_net(nsproxy->net_ns);
660 nsproxy->net_ns = get_net(net);
664 static unsigned int netns_inum(void *ns)
666 struct net *net = ns;
667 return net->proc_inum;
670 const struct proc_ns_operations netns_operations = {
672 .type = CLONE_NEWNET,
675 .install = netns_install,