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 <net/net_namespace.h>
17 #include <net/netns/generic.h>
20 * Our network namespace constructor/destructor lists
23 static LIST_HEAD(pernet_list);
24 static struct list_head *first_device = &pernet_list;
25 static DEFINE_MUTEX(net_mutex);
27 LIST_HEAD(net_namespace_list);
28 EXPORT_SYMBOL_GPL(net_namespace_list);
31 EXPORT_SYMBOL(init_net);
33 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
35 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
37 static struct net_generic *net_alloc_generic(void)
39 struct net_generic *ng;
40 size_t generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
42 ng = kzalloc(generic_size, GFP_KERNEL);
44 ng->len = max_gen_ptrs;
49 static int net_assign_generic(struct net *net, int id, void *data)
51 struct net_generic *ng, *old_ng;
53 BUG_ON(!mutex_is_locked(&net_mutex));
56 old_ng = rcu_dereference_protected(net->gen,
57 lockdep_is_held(&net_mutex));
59 if (old_ng->len >= id)
62 ng = net_alloc_generic();
67 * Some synchronisation notes:
69 * The net_generic explores the net->gen array inside rcu
70 * read section. Besides once set the net->gen->ptr[x]
71 * pointer never changes (see rules in netns/generic.h).
73 * That said, we simply duplicate this array and schedule
74 * the old copy for kfree after a grace period.
77 memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
79 rcu_assign_pointer(net->gen, ng);
80 kfree_rcu(old_ng, rcu);
82 ng->ptr[id - 1] = data;
86 static int ops_init(const struct pernet_operations *ops, struct net *net)
91 if (ops->id && ops->size) {
92 data = kzalloc(ops->size, GFP_KERNEL);
96 err = net_assign_generic(net, *ops->id, data);
102 err = ops->init(net);
113 static void ops_free(const struct pernet_operations *ops, struct net *net)
115 if (ops->id && ops->size) {
117 kfree(net_generic(net, id));
121 static void ops_exit_list(const struct pernet_operations *ops,
122 struct list_head *net_exit_list)
126 list_for_each_entry(net, net_exit_list, exit_list)
130 ops->exit_batch(net_exit_list);
133 static void ops_free_list(const struct pernet_operations *ops,
134 struct list_head *net_exit_list)
137 if (ops->size && ops->id) {
138 list_for_each_entry(net, net_exit_list, exit_list)
144 * setup_net runs the initializers for the network namespace object.
146 static __net_init int setup_net(struct net *net)
148 /* Must be called with net_mutex held */
149 const struct pernet_operations *ops, *saved_ops;
151 LIST_HEAD(net_exit_list);
153 atomic_set(&net->count, 1);
154 atomic_set(&net->passive, 1);
155 net->dev_base_seq = 1;
157 #ifdef NETNS_REFCNT_DEBUG
158 atomic_set(&net->use_count, 0);
161 list_for_each_entry(ops, &pernet_list, list) {
162 error = ops_init(ops, net);
170 /* Walk through the list backwards calling the exit functions
171 * for the pernet modules whose init functions did not fail.
173 list_add(&net->exit_list, &net_exit_list);
175 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
176 ops_exit_list(ops, &net_exit_list);
179 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
180 ops_free_list(ops, &net_exit_list);
188 static struct kmem_cache *net_cachep;
189 static struct workqueue_struct *netns_wq;
191 static struct net *net_alloc(void)
193 struct net *net = NULL;
194 struct net_generic *ng;
196 ng = net_alloc_generic();
200 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
204 rcu_assign_pointer(net->gen, ng);
213 static void net_free(struct net *net)
215 #ifdef NETNS_REFCNT_DEBUG
216 if (unlikely(atomic_read(&net->use_count) != 0)) {
217 pr_emerg("network namespace not free! Usage: %d\n",
218 atomic_read(&net->use_count));
223 kmem_cache_free(net_cachep, net);
226 void net_drop_ns(void *p)
229 if (ns && atomic_dec_and_test(&ns->passive))
233 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
238 if (!(flags & CLONE_NEWNET))
239 return get_net(old_net);
243 return ERR_PTR(-ENOMEM);
244 mutex_lock(&net_mutex);
248 list_add_tail_rcu(&net->list, &net_namespace_list);
251 mutex_unlock(&net_mutex);
259 static DEFINE_SPINLOCK(cleanup_list_lock);
260 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
262 static void cleanup_net(struct work_struct *work)
264 const struct pernet_operations *ops;
265 struct net *net, *tmp;
266 LIST_HEAD(net_kill_list);
267 LIST_HEAD(net_exit_list);
269 /* Atomically snapshot the list of namespaces to cleanup */
270 spin_lock_irq(&cleanup_list_lock);
271 list_replace_init(&cleanup_list, &net_kill_list);
272 spin_unlock_irq(&cleanup_list_lock);
274 mutex_lock(&net_mutex);
276 /* Don't let anyone else find us. */
278 list_for_each_entry(net, &net_kill_list, cleanup_list) {
279 list_del_rcu(&net->list);
280 list_add_tail(&net->exit_list, &net_exit_list);
285 * Another CPU might be rcu-iterating the list, wait for it.
286 * This needs to be before calling the exit() notifiers, so
287 * the rcu_barrier() below isn't sufficient alone.
291 /* Run all of the network namespace exit methods */
292 list_for_each_entry_reverse(ops, &pernet_list, list)
293 ops_exit_list(ops, &net_exit_list);
295 /* Free the net generic variables */
296 list_for_each_entry_reverse(ops, &pernet_list, list)
297 ops_free_list(ops, &net_exit_list);
299 mutex_unlock(&net_mutex);
301 /* Ensure there are no outstanding rcu callbacks using this
306 /* Finally it is safe to free my network namespace structure */
307 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
308 list_del_init(&net->exit_list);
312 static DECLARE_WORK(net_cleanup_work, cleanup_net);
314 void __put_net(struct net *net)
316 /* Cleanup the network namespace in process context */
319 spin_lock_irqsave(&cleanup_list_lock, flags);
320 list_add(&net->cleanup_list, &cleanup_list);
321 spin_unlock_irqrestore(&cleanup_list_lock, flags);
323 queue_work(netns_wq, &net_cleanup_work);
325 EXPORT_SYMBOL_GPL(__put_net);
327 struct net *get_net_ns_by_fd(int fd)
329 struct proc_inode *ei;
333 file = proc_ns_fget(fd);
335 return ERR_CAST(file);
337 ei = PROC_I(file->f_dentry->d_inode);
338 if (ei->ns_ops == &netns_operations)
339 net = get_net(ei->ns);
341 net = ERR_PTR(-EINVAL);
348 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
350 if (flags & CLONE_NEWNET)
351 return ERR_PTR(-EINVAL);
355 struct net *get_net_ns_by_fd(int fd)
357 return ERR_PTR(-EINVAL);
361 struct net *get_net_ns_by_pid(pid_t pid)
363 struct task_struct *tsk;
366 /* Lookup the network namespace */
367 net = ERR_PTR(-ESRCH);
369 tsk = find_task_by_vpid(pid);
371 struct nsproxy *nsproxy;
372 nsproxy = task_nsproxy(tsk);
374 net = get_net(nsproxy->net_ns);
379 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
381 static int __init net_ns_init(void)
383 struct net_generic *ng;
386 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
390 /* Create workqueue for cleanup */
391 netns_wq = create_singlethread_workqueue("netns");
393 panic("Could not create netns workq");
396 ng = net_alloc_generic();
398 panic("Could not allocate generic netns");
400 rcu_assign_pointer(init_net.gen, ng);
402 mutex_lock(&net_mutex);
403 if (setup_net(&init_net))
404 panic("Could not setup the initial network namespace");
407 list_add_tail_rcu(&init_net.list, &net_namespace_list);
410 mutex_unlock(&net_mutex);
415 pure_initcall(net_ns_init);
418 static int __register_pernet_operations(struct list_head *list,
419 struct pernet_operations *ops)
423 LIST_HEAD(net_exit_list);
425 list_add_tail(&ops->list, list);
426 if (ops->init || (ops->id && ops->size)) {
428 error = ops_init(ops, net);
431 list_add_tail(&net->exit_list, &net_exit_list);
437 /* If I have an error cleanup all namespaces I initialized */
438 list_del(&ops->list);
439 ops_exit_list(ops, &net_exit_list);
440 ops_free_list(ops, &net_exit_list);
444 static void __unregister_pernet_operations(struct pernet_operations *ops)
447 LIST_HEAD(net_exit_list);
449 list_del(&ops->list);
451 list_add_tail(&net->exit_list, &net_exit_list);
452 ops_exit_list(ops, &net_exit_list);
453 ops_free_list(ops, &net_exit_list);
458 static int __register_pernet_operations(struct list_head *list,
459 struct pernet_operations *ops)
461 return ops_init(ops, &init_net);
464 static void __unregister_pernet_operations(struct pernet_operations *ops)
466 LIST_HEAD(net_exit_list);
467 list_add(&init_net.exit_list, &net_exit_list);
468 ops_exit_list(ops, &net_exit_list);
469 ops_free_list(ops, &net_exit_list);
472 #endif /* CONFIG_NET_NS */
474 static DEFINE_IDA(net_generic_ids);
476 static int register_pernet_operations(struct list_head *list,
477 struct pernet_operations *ops)
483 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
485 if (error == -EAGAIN) {
486 ida_pre_get(&net_generic_ids, GFP_KERNEL);
491 max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
493 error = __register_pernet_operations(list, ops);
497 ida_remove(&net_generic_ids, *ops->id);
503 static void unregister_pernet_operations(struct pernet_operations *ops)
506 __unregister_pernet_operations(ops);
509 ida_remove(&net_generic_ids, *ops->id);
513 * register_pernet_subsys - register a network namespace subsystem
514 * @ops: pernet operations structure for the subsystem
516 * Register a subsystem which has init and exit functions
517 * that are called when network namespaces are created and
518 * destroyed respectively.
520 * When registered all network namespace init functions are
521 * called for every existing network namespace. Allowing kernel
522 * modules to have a race free view of the set of network namespaces.
524 * When a new network namespace is created all of the init
525 * methods are called in the order in which they were registered.
527 * When a network namespace is destroyed all of the exit methods
528 * are called in the reverse of the order with which they were
531 int register_pernet_subsys(struct pernet_operations *ops)
534 mutex_lock(&net_mutex);
535 error = register_pernet_operations(first_device, ops);
536 mutex_unlock(&net_mutex);
539 EXPORT_SYMBOL_GPL(register_pernet_subsys);
542 * unregister_pernet_subsys - unregister a network namespace subsystem
543 * @ops: pernet operations structure to manipulate
545 * Remove the pernet operations structure from the list to be
546 * used when network namespaces are created or destroyed. In
547 * addition run the exit method for all existing network
550 void unregister_pernet_subsys(struct pernet_operations *ops)
552 mutex_lock(&net_mutex);
553 unregister_pernet_operations(ops);
554 mutex_unlock(&net_mutex);
556 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
559 * register_pernet_device - register a network namespace device
560 * @ops: pernet operations structure for the subsystem
562 * Register a device which has init and exit functions
563 * that are called when network namespaces are created and
564 * destroyed respectively.
566 * When registered all network namespace init functions are
567 * called for every existing network namespace. Allowing kernel
568 * modules to have a race free view of the set of network namespaces.
570 * When a new network namespace is created all of the init
571 * methods are called in the order in which they were registered.
573 * When a network namespace is destroyed all of the exit methods
574 * are called in the reverse of the order with which they were
577 int register_pernet_device(struct pernet_operations *ops)
580 mutex_lock(&net_mutex);
581 error = register_pernet_operations(&pernet_list, ops);
582 if (!error && (first_device == &pernet_list))
583 first_device = &ops->list;
584 mutex_unlock(&net_mutex);
587 EXPORT_SYMBOL_GPL(register_pernet_device);
590 * unregister_pernet_device - unregister a network namespace netdevice
591 * @ops: pernet operations structure to manipulate
593 * Remove the pernet operations structure from the list to be
594 * used when network namespaces are created or destroyed. In
595 * addition run the exit method for all existing network
598 void unregister_pernet_device(struct pernet_operations *ops)
600 mutex_lock(&net_mutex);
601 if (&ops->list == first_device)
602 first_device = first_device->next;
603 unregister_pernet_operations(ops);
604 mutex_unlock(&net_mutex);
606 EXPORT_SYMBOL_GPL(unregister_pernet_device);
609 static void *netns_get(struct task_struct *task)
611 struct net *net = NULL;
612 struct nsproxy *nsproxy;
615 nsproxy = task_nsproxy(task);
617 net = get_net(nsproxy->net_ns);
623 static void netns_put(void *ns)
628 static int netns_install(struct nsproxy *nsproxy, void *ns)
630 put_net(nsproxy->net_ns);
631 nsproxy->net_ns = get_net(ns);
635 const struct proc_ns_operations netns_operations = {
637 .type = CLONE_NEWNET,
640 .install = netns_install,