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);
30 struct net init_net = {
31 .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head),
33 EXPORT_SYMBOL(init_net);
35 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
37 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
39 static struct net_generic *net_alloc_generic(void)
41 struct net_generic *ng;
42 size_t generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
44 ng = kzalloc(generic_size, GFP_KERNEL);
46 ng->len = max_gen_ptrs;
51 static int net_assign_generic(struct net *net, int id, void *data)
53 struct net_generic *ng, *old_ng;
55 BUG_ON(!mutex_is_locked(&net_mutex));
58 old_ng = rcu_dereference_protected(net->gen,
59 lockdep_is_held(&net_mutex));
61 if (old_ng->len >= id)
64 ng = net_alloc_generic();
69 * Some synchronisation notes:
71 * The net_generic explores the net->gen array inside rcu
72 * read section. Besides once set the net->gen->ptr[x]
73 * pointer never changes (see rules in netns/generic.h).
75 * That said, we simply duplicate this array and schedule
76 * the old copy for kfree after a grace period.
79 memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
81 rcu_assign_pointer(net->gen, ng);
82 kfree_rcu(old_ng, rcu);
84 ng->ptr[id - 1] = data;
88 static int ops_init(const struct pernet_operations *ops, struct net *net)
93 if (ops->id && ops->size) {
94 data = kzalloc(ops->size, GFP_KERNEL);
98 err = net_assign_generic(net, *ops->id, data);
104 err = ops->init(net);
115 static void ops_free(const struct pernet_operations *ops, struct net *net)
117 if (ops->id && ops->size) {
119 kfree(net_generic(net, id));
123 static void ops_exit_list(const struct pernet_operations *ops,
124 struct list_head *net_exit_list)
128 list_for_each_entry(net, net_exit_list, exit_list)
132 ops->exit_batch(net_exit_list);
135 static void ops_free_list(const struct pernet_operations *ops,
136 struct list_head *net_exit_list)
139 if (ops->size && ops->id) {
140 list_for_each_entry(net, net_exit_list, exit_list)
146 * setup_net runs the initializers for the network namespace object.
148 static __net_init int setup_net(struct net *net)
150 /* Must be called with net_mutex held */
151 const struct pernet_operations *ops, *saved_ops;
153 LIST_HEAD(net_exit_list);
155 atomic_set(&net->count, 1);
156 atomic_set(&net->passive, 1);
157 net->dev_base_seq = 1;
159 #ifdef NETNS_REFCNT_DEBUG
160 atomic_set(&net->use_count, 0);
163 list_for_each_entry(ops, &pernet_list, list) {
164 error = ops_init(ops, net);
172 /* Walk through the list backwards calling the exit functions
173 * for the pernet modules whose init functions did not fail.
175 list_add(&net->exit_list, &net_exit_list);
177 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
178 ops_exit_list(ops, &net_exit_list);
181 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
182 ops_free_list(ops, &net_exit_list);
190 static struct kmem_cache *net_cachep;
191 static struct workqueue_struct *netns_wq;
193 static struct net *net_alloc(void)
195 struct net *net = NULL;
196 struct net_generic *ng;
198 ng = net_alloc_generic();
202 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
206 rcu_assign_pointer(net->gen, ng);
215 static void net_free(struct net *net)
217 #ifdef NETNS_REFCNT_DEBUG
218 if (unlikely(atomic_read(&net->use_count) != 0)) {
219 pr_emerg("network namespace not free! Usage: %d\n",
220 atomic_read(&net->use_count));
225 kmem_cache_free(net_cachep, net);
228 void net_drop_ns(void *p)
231 if (ns && atomic_dec_and_test(&ns->passive))
235 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
240 if (!(flags & CLONE_NEWNET))
241 return get_net(old_net);
245 return ERR_PTR(-ENOMEM);
246 mutex_lock(&net_mutex);
250 list_add_tail_rcu(&net->list, &net_namespace_list);
253 mutex_unlock(&net_mutex);
261 static DEFINE_SPINLOCK(cleanup_list_lock);
262 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
264 static void cleanup_net(struct work_struct *work)
266 const struct pernet_operations *ops;
267 struct net *net, *tmp;
268 LIST_HEAD(net_kill_list);
269 LIST_HEAD(net_exit_list);
271 /* Atomically snapshot the list of namespaces to cleanup */
272 spin_lock_irq(&cleanup_list_lock);
273 list_replace_init(&cleanup_list, &net_kill_list);
274 spin_unlock_irq(&cleanup_list_lock);
276 mutex_lock(&net_mutex);
278 /* Don't let anyone else find us. */
280 list_for_each_entry(net, &net_kill_list, cleanup_list) {
281 list_del_rcu(&net->list);
282 list_add_tail(&net->exit_list, &net_exit_list);
287 * Another CPU might be rcu-iterating the list, wait for it.
288 * This needs to be before calling the exit() notifiers, so
289 * the rcu_barrier() below isn't sufficient alone.
293 /* Run all of the network namespace exit methods */
294 list_for_each_entry_reverse(ops, &pernet_list, list)
295 ops_exit_list(ops, &net_exit_list);
297 /* Free the net generic variables */
298 list_for_each_entry_reverse(ops, &pernet_list, list)
299 ops_free_list(ops, &net_exit_list);
301 mutex_unlock(&net_mutex);
303 /* Ensure there are no outstanding rcu callbacks using this
308 /* Finally it is safe to free my network namespace structure */
309 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
310 list_del_init(&net->exit_list);
314 static DECLARE_WORK(net_cleanup_work, cleanup_net);
316 void __put_net(struct net *net)
318 /* Cleanup the network namespace in process context */
321 spin_lock_irqsave(&cleanup_list_lock, flags);
322 list_add(&net->cleanup_list, &cleanup_list);
323 spin_unlock_irqrestore(&cleanup_list_lock, flags);
325 queue_work(netns_wq, &net_cleanup_work);
327 EXPORT_SYMBOL_GPL(__put_net);
329 struct net *get_net_ns_by_fd(int fd)
331 struct proc_inode *ei;
335 file = proc_ns_fget(fd);
337 return ERR_CAST(file);
339 ei = PROC_I(file->f_dentry->d_inode);
340 if (ei->ns_ops == &netns_operations)
341 net = get_net(ei->ns);
343 net = ERR_PTR(-EINVAL);
350 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
352 if (flags & CLONE_NEWNET)
353 return ERR_PTR(-EINVAL);
357 struct net *get_net_ns_by_fd(int fd)
359 return ERR_PTR(-EINVAL);
363 struct net *get_net_ns_by_pid(pid_t pid)
365 struct task_struct *tsk;
368 /* Lookup the network namespace */
369 net = ERR_PTR(-ESRCH);
371 tsk = find_task_by_vpid(pid);
373 struct nsproxy *nsproxy;
374 nsproxy = task_nsproxy(tsk);
376 net = get_net(nsproxy->net_ns);
381 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
383 static int __init net_ns_init(void)
385 struct net_generic *ng;
388 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
392 /* Create workqueue for cleanup */
393 netns_wq = create_singlethread_workqueue("netns");
395 panic("Could not create netns workq");
398 ng = net_alloc_generic();
400 panic("Could not allocate generic netns");
402 rcu_assign_pointer(init_net.gen, ng);
404 mutex_lock(&net_mutex);
405 if (setup_net(&init_net))
406 panic("Could not setup the initial network namespace");
409 list_add_tail_rcu(&init_net.list, &net_namespace_list);
412 mutex_unlock(&net_mutex);
417 pure_initcall(net_ns_init);
420 static int __register_pernet_operations(struct list_head *list,
421 struct pernet_operations *ops)
425 LIST_HEAD(net_exit_list);
427 list_add_tail(&ops->list, list);
428 if (ops->init || (ops->id && ops->size)) {
430 error = ops_init(ops, net);
433 list_add_tail(&net->exit_list, &net_exit_list);
439 /* If I have an error cleanup all namespaces I initialized */
440 list_del(&ops->list);
441 ops_exit_list(ops, &net_exit_list);
442 ops_free_list(ops, &net_exit_list);
446 static void __unregister_pernet_operations(struct pernet_operations *ops)
449 LIST_HEAD(net_exit_list);
451 list_del(&ops->list);
453 list_add_tail(&net->exit_list, &net_exit_list);
454 ops_exit_list(ops, &net_exit_list);
455 ops_free_list(ops, &net_exit_list);
460 static int __register_pernet_operations(struct list_head *list,
461 struct pernet_operations *ops)
463 return ops_init(ops, &init_net);
466 static void __unregister_pernet_operations(struct pernet_operations *ops)
468 LIST_HEAD(net_exit_list);
469 list_add(&init_net.exit_list, &net_exit_list);
470 ops_exit_list(ops, &net_exit_list);
471 ops_free_list(ops, &net_exit_list);
474 #endif /* CONFIG_NET_NS */
476 static DEFINE_IDA(net_generic_ids);
478 static int register_pernet_operations(struct list_head *list,
479 struct pernet_operations *ops)
485 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
487 if (error == -EAGAIN) {
488 ida_pre_get(&net_generic_ids, GFP_KERNEL);
493 max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
495 error = __register_pernet_operations(list, ops);
499 ida_remove(&net_generic_ids, *ops->id);
505 static void unregister_pernet_operations(struct pernet_operations *ops)
508 __unregister_pernet_operations(ops);
511 ida_remove(&net_generic_ids, *ops->id);
515 * register_pernet_subsys - register a network namespace subsystem
516 * @ops: pernet operations structure for the subsystem
518 * Register a subsystem which has init and exit functions
519 * that are called when network namespaces are created and
520 * destroyed respectively.
522 * When registered all network namespace init functions are
523 * called for every existing network namespace. Allowing kernel
524 * modules to have a race free view of the set of network namespaces.
526 * When a new network namespace is created all of the init
527 * methods are called in the order in which they were registered.
529 * When a network namespace is destroyed all of the exit methods
530 * are called in the reverse of the order with which they were
533 int register_pernet_subsys(struct pernet_operations *ops)
536 mutex_lock(&net_mutex);
537 error = register_pernet_operations(first_device, ops);
538 mutex_unlock(&net_mutex);
541 EXPORT_SYMBOL_GPL(register_pernet_subsys);
544 * unregister_pernet_subsys - unregister a network namespace subsystem
545 * @ops: pernet operations structure to manipulate
547 * Remove the pernet operations structure from the list to be
548 * used when network namespaces are created or destroyed. In
549 * addition run the exit method for all existing network
552 void unregister_pernet_subsys(struct pernet_operations *ops)
554 mutex_lock(&net_mutex);
555 unregister_pernet_operations(ops);
556 mutex_unlock(&net_mutex);
558 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
561 * register_pernet_device - register a network namespace device
562 * @ops: pernet operations structure for the subsystem
564 * Register a device which has init and exit functions
565 * that are called when network namespaces are created and
566 * destroyed respectively.
568 * When registered all network namespace init functions are
569 * called for every existing network namespace. Allowing kernel
570 * modules to have a race free view of the set of network namespaces.
572 * When a new network namespace is created all of the init
573 * methods are called in the order in which they were registered.
575 * When a network namespace is destroyed all of the exit methods
576 * are called in the reverse of the order with which they were
579 int register_pernet_device(struct pernet_operations *ops)
582 mutex_lock(&net_mutex);
583 error = register_pernet_operations(&pernet_list, ops);
584 if (!error && (first_device == &pernet_list))
585 first_device = &ops->list;
586 mutex_unlock(&net_mutex);
589 EXPORT_SYMBOL_GPL(register_pernet_device);
592 * unregister_pernet_device - unregister a network namespace netdevice
593 * @ops: pernet operations structure to manipulate
595 * Remove the pernet operations structure from the list to be
596 * used when network namespaces are created or destroyed. In
597 * addition run the exit method for all existing network
600 void unregister_pernet_device(struct pernet_operations *ops)
602 mutex_lock(&net_mutex);
603 if (&ops->list == first_device)
604 first_device = first_device->next;
605 unregister_pernet_operations(ops);
606 mutex_unlock(&net_mutex);
608 EXPORT_SYMBOL_GPL(unregister_pernet_device);
611 static void *netns_get(struct task_struct *task)
613 struct net *net = NULL;
614 struct nsproxy *nsproxy;
617 nsproxy = task_nsproxy(task);
619 net = get_net(nsproxy->net_ns);
625 static void netns_put(void *ns)
630 static int netns_install(struct nsproxy *nsproxy, void *ns)
632 put_net(nsproxy->net_ns);
633 nsproxy->net_ns = get_net(ns);
637 const struct proc_ns_operations netns_operations = {
639 .type = CLONE_NEWNET,
642 .install = netns_install,