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>
13 #include <linux/proc_fs.h>
14 #include <linux/file.h>
15 #include <linux/export.h>
16 #include <linux/user_namespace.h>
17 #include <net/net_namespace.h>
18 #include <net/netns/generic.h>
21 * Our network namespace constructor/destructor lists
24 static LIST_HEAD(pernet_list);
25 static struct list_head *first_device = &pernet_list;
26 static DEFINE_MUTEX(net_mutex);
28 LIST_HEAD(net_namespace_list);
29 EXPORT_SYMBOL_GPL(net_namespace_list);
31 struct net init_net = {
32 .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head),
34 EXPORT_SYMBOL(init_net);
36 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
38 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
40 static struct net_generic *net_alloc_generic(void)
42 struct net_generic *ng;
43 size_t generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
45 ng = kzalloc(generic_size, GFP_KERNEL);
47 ng->len = max_gen_ptrs;
52 static int net_assign_generic(struct net *net, int id, void *data)
54 struct net_generic *ng, *old_ng;
56 BUG_ON(!mutex_is_locked(&net_mutex));
59 old_ng = rcu_dereference_protected(net->gen,
60 lockdep_is_held(&net_mutex));
62 if (old_ng->len >= id)
65 ng = net_alloc_generic();
70 * Some synchronisation notes:
72 * The net_generic explores the net->gen array inside rcu
73 * read section. Besides once set the net->gen->ptr[x]
74 * pointer never changes (see rules in netns/generic.h).
76 * That said, we simply duplicate this array and schedule
77 * the old copy for kfree after a grace period.
80 memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
82 rcu_assign_pointer(net->gen, ng);
83 kfree_rcu(old_ng, rcu);
85 ng->ptr[id - 1] = data;
89 static int ops_init(const struct pernet_operations *ops, struct net *net)
94 if (ops->id && ops->size) {
95 data = kzalloc(ops->size, GFP_KERNEL);
99 err = net_assign_generic(net, *ops->id, data);
105 err = ops->init(net);
116 static void ops_free(const struct pernet_operations *ops, struct net *net)
118 if (ops->id && ops->size) {
120 kfree(net_generic(net, id));
124 static void ops_exit_list(const struct pernet_operations *ops,
125 struct list_head *net_exit_list)
129 list_for_each_entry(net, net_exit_list, exit_list)
133 ops->exit_batch(net_exit_list);
136 static void ops_free_list(const struct pernet_operations *ops,
137 struct list_head *net_exit_list)
140 if (ops->size && ops->id) {
141 list_for_each_entry(net, net_exit_list, exit_list)
147 * setup_net runs the initializers for the network namespace object.
149 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
151 /* Must be called with net_mutex held */
152 const struct pernet_operations *ops, *saved_ops;
154 LIST_HEAD(net_exit_list);
156 atomic_set(&net->count, 1);
157 atomic_set(&net->passive, 1);
158 net->dev_base_seq = 1;
159 net->user_ns = user_ns;
161 #ifdef NETNS_REFCNT_DEBUG
162 atomic_set(&net->use_count, 0);
165 list_for_each_entry(ops, &pernet_list, list) {
166 error = ops_init(ops, net);
174 /* Walk through the list backwards calling the exit functions
175 * for the pernet modules whose init functions did not fail.
177 list_add(&net->exit_list, &net_exit_list);
179 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
180 ops_exit_list(ops, &net_exit_list);
183 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
184 ops_free_list(ops, &net_exit_list);
192 static struct kmem_cache *net_cachep;
193 static struct workqueue_struct *netns_wq;
195 static struct net *net_alloc(void)
197 struct net *net = NULL;
198 struct net_generic *ng;
200 ng = net_alloc_generic();
204 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
208 rcu_assign_pointer(net->gen, ng);
217 static void net_free(struct net *net)
219 #ifdef NETNS_REFCNT_DEBUG
220 if (unlikely(atomic_read(&net->use_count) != 0)) {
221 pr_emerg("network namespace not free! Usage: %d\n",
222 atomic_read(&net->use_count));
227 kmem_cache_free(net_cachep, net);
230 void net_drop_ns(void *p)
233 if (ns && atomic_dec_and_test(&ns->passive))
237 struct net *copy_net_ns(unsigned long flags,
238 struct user_namespace *user_ns, struct net *old_net)
243 if (!(flags & CLONE_NEWNET))
244 return get_net(old_net);
248 return ERR_PTR(-ENOMEM);
250 get_user_ns(user_ns);
252 mutex_lock(&net_mutex);
253 rv = setup_net(net, user_ns);
256 list_add_tail_rcu(&net->list, &net_namespace_list);
259 mutex_unlock(&net_mutex);
261 put_user_ns(user_ns);
268 static DEFINE_SPINLOCK(cleanup_list_lock);
269 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
271 static void cleanup_net(struct work_struct *work)
273 const struct pernet_operations *ops;
274 struct net *net, *tmp;
275 LIST_HEAD(net_kill_list);
276 LIST_HEAD(net_exit_list);
278 /* Atomically snapshot the list of namespaces to cleanup */
279 spin_lock_irq(&cleanup_list_lock);
280 list_replace_init(&cleanup_list, &net_kill_list);
281 spin_unlock_irq(&cleanup_list_lock);
283 mutex_lock(&net_mutex);
285 /* Don't let anyone else find us. */
287 list_for_each_entry(net, &net_kill_list, cleanup_list) {
288 list_del_rcu(&net->list);
289 list_add_tail(&net->exit_list, &net_exit_list);
294 * Another CPU might be rcu-iterating the list, wait for it.
295 * This needs to be before calling the exit() notifiers, so
296 * the rcu_barrier() below isn't sufficient alone.
300 /* Run all of the network namespace exit methods */
301 list_for_each_entry_reverse(ops, &pernet_list, list)
302 ops_exit_list(ops, &net_exit_list);
304 /* Free the net generic variables */
305 list_for_each_entry_reverse(ops, &pernet_list, list)
306 ops_free_list(ops, &net_exit_list);
308 mutex_unlock(&net_mutex);
310 /* Ensure there are no outstanding rcu callbacks using this
315 /* Finally it is safe to free my network namespace structure */
316 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
317 list_del_init(&net->exit_list);
318 put_user_ns(net->user_ns);
322 static DECLARE_WORK(net_cleanup_work, cleanup_net);
324 void __put_net(struct net *net)
326 /* Cleanup the network namespace in process context */
329 spin_lock_irqsave(&cleanup_list_lock, flags);
330 list_add(&net->cleanup_list, &cleanup_list);
331 spin_unlock_irqrestore(&cleanup_list_lock, flags);
333 queue_work(netns_wq, &net_cleanup_work);
335 EXPORT_SYMBOL_GPL(__put_net);
337 struct net *get_net_ns_by_fd(int fd)
339 struct proc_inode *ei;
343 file = proc_ns_fget(fd);
345 return ERR_CAST(file);
347 ei = PROC_I(file->f_dentry->d_inode);
348 if (ei->ns_ops == &netns_operations)
349 net = get_net(ei->ns);
351 net = ERR_PTR(-EINVAL);
358 struct net *get_net_ns_by_fd(int fd)
360 return ERR_PTR(-EINVAL);
364 struct net *get_net_ns_by_pid(pid_t pid)
366 struct task_struct *tsk;
369 /* Lookup the network namespace */
370 net = ERR_PTR(-ESRCH);
372 tsk = find_task_by_vpid(pid);
374 struct nsproxy *nsproxy;
375 nsproxy = task_nsproxy(tsk);
377 net = get_net(nsproxy->net_ns);
382 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
384 static __net_init int net_ns_net_init(struct net *net)
386 return proc_alloc_inum(&net->proc_inum);
389 static __net_exit void net_ns_net_exit(struct net *net)
391 proc_free_inum(net->proc_inum);
394 static struct pernet_operations __net_initdata net_ns_ops = {
395 .init = net_ns_net_init,
396 .exit = net_ns_net_exit,
399 static int __init net_ns_init(void)
401 struct net_generic *ng;
404 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
408 /* Create workqueue for cleanup */
409 netns_wq = create_singlethread_workqueue("netns");
411 panic("Could not create netns workq");
414 ng = net_alloc_generic();
416 panic("Could not allocate generic netns");
418 rcu_assign_pointer(init_net.gen, ng);
420 mutex_lock(&net_mutex);
421 if (setup_net(&init_net, &init_user_ns))
422 panic("Could not setup the initial network namespace");
425 list_add_tail_rcu(&init_net.list, &net_namespace_list);
428 mutex_unlock(&net_mutex);
430 register_pernet_subsys(&net_ns_ops);
435 pure_initcall(net_ns_init);
438 static int __register_pernet_operations(struct list_head *list,
439 struct pernet_operations *ops)
443 LIST_HEAD(net_exit_list);
445 list_add_tail(&ops->list, list);
446 if (ops->init || (ops->id && ops->size)) {
448 error = ops_init(ops, net);
451 list_add_tail(&net->exit_list, &net_exit_list);
457 /* If I have an error cleanup all namespaces I initialized */
458 list_del(&ops->list);
459 ops_exit_list(ops, &net_exit_list);
460 ops_free_list(ops, &net_exit_list);
464 static void __unregister_pernet_operations(struct pernet_operations *ops)
467 LIST_HEAD(net_exit_list);
469 list_del(&ops->list);
471 list_add_tail(&net->exit_list, &net_exit_list);
472 ops_exit_list(ops, &net_exit_list);
473 ops_free_list(ops, &net_exit_list);
478 static int __register_pernet_operations(struct list_head *list,
479 struct pernet_operations *ops)
481 return ops_init(ops, &init_net);
484 static void __unregister_pernet_operations(struct pernet_operations *ops)
486 LIST_HEAD(net_exit_list);
487 list_add(&init_net.exit_list, &net_exit_list);
488 ops_exit_list(ops, &net_exit_list);
489 ops_free_list(ops, &net_exit_list);
492 #endif /* CONFIG_NET_NS */
494 static DEFINE_IDA(net_generic_ids);
496 static int register_pernet_operations(struct list_head *list,
497 struct pernet_operations *ops)
503 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
505 if (error == -EAGAIN) {
506 ida_pre_get(&net_generic_ids, GFP_KERNEL);
511 max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
513 error = __register_pernet_operations(list, ops);
517 ida_remove(&net_generic_ids, *ops->id);
523 static void unregister_pernet_operations(struct pernet_operations *ops)
526 __unregister_pernet_operations(ops);
529 ida_remove(&net_generic_ids, *ops->id);
533 * register_pernet_subsys - register a network namespace subsystem
534 * @ops: pernet operations structure for the subsystem
536 * Register a subsystem which has init and exit functions
537 * that are called when network namespaces are created and
538 * destroyed respectively.
540 * When registered all network namespace init functions are
541 * called for every existing network namespace. Allowing kernel
542 * modules to have a race free view of the set of network namespaces.
544 * When a new network namespace is created all of the init
545 * methods are called in the order in which they were registered.
547 * When a network namespace is destroyed all of the exit methods
548 * are called in the reverse of the order with which they were
551 int register_pernet_subsys(struct pernet_operations *ops)
554 mutex_lock(&net_mutex);
555 error = register_pernet_operations(first_device, ops);
556 mutex_unlock(&net_mutex);
559 EXPORT_SYMBOL_GPL(register_pernet_subsys);
562 * unregister_pernet_subsys - unregister a network namespace subsystem
563 * @ops: pernet operations structure to manipulate
565 * Remove the pernet operations structure from the list to be
566 * used when network namespaces are created or destroyed. In
567 * addition run the exit method for all existing network
570 void unregister_pernet_subsys(struct pernet_operations *ops)
572 mutex_lock(&net_mutex);
573 unregister_pernet_operations(ops);
574 mutex_unlock(&net_mutex);
576 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
579 * register_pernet_device - register a network namespace device
580 * @ops: pernet operations structure for the subsystem
582 * Register a device which has init and exit functions
583 * that are called when network namespaces are created and
584 * destroyed respectively.
586 * When registered all network namespace init functions are
587 * called for every existing network namespace. Allowing kernel
588 * modules to have a race free view of the set of network namespaces.
590 * When a new network namespace is created all of the init
591 * methods are called in the order in which they were registered.
593 * When a network namespace is destroyed all of the exit methods
594 * are called in the reverse of the order with which they were
597 int register_pernet_device(struct pernet_operations *ops)
600 mutex_lock(&net_mutex);
601 error = register_pernet_operations(&pernet_list, ops);
602 if (!error && (first_device == &pernet_list))
603 first_device = &ops->list;
604 mutex_unlock(&net_mutex);
607 EXPORT_SYMBOL_GPL(register_pernet_device);
610 * unregister_pernet_device - unregister a network namespace netdevice
611 * @ops: pernet operations structure to manipulate
613 * Remove the pernet operations structure from the list to be
614 * used when network namespaces are created or destroyed. In
615 * addition run the exit method for all existing network
618 void unregister_pernet_device(struct pernet_operations *ops)
620 mutex_lock(&net_mutex);
621 if (&ops->list == first_device)
622 first_device = first_device->next;
623 unregister_pernet_operations(ops);
624 mutex_unlock(&net_mutex);
626 EXPORT_SYMBOL_GPL(unregister_pernet_device);
629 static void *netns_get(struct task_struct *task)
631 struct net *net = NULL;
632 struct nsproxy *nsproxy;
635 nsproxy = task_nsproxy(task);
637 net = get_net(nsproxy->net_ns);
643 static void netns_put(void *ns)
648 static int netns_install(struct nsproxy *nsproxy, void *ns)
650 struct net *net = ns;
652 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
653 !nsown_capable(CAP_SYS_ADMIN))
656 put_net(nsproxy->net_ns);
657 nsproxy->net_ns = get_net(net);
661 static unsigned int netns_inum(void *ns)
663 struct net *net = ns;
664 return net->proc_inum;
667 const struct proc_ns_operations netns_operations = {
669 .type = CLONE_NEWNET,
672 .install = netns_install,