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 <linux/proc_fs.h>
12 #include <linux/file.h>
13 #include <net/net_namespace.h>
14 #include <net/netns/generic.h>
17 * Our network namespace constructor/destructor lists
20 static LIST_HEAD(pernet_list);
21 static struct list_head *first_device = &pernet_list;
22 static DEFINE_MUTEX(net_mutex);
24 LIST_HEAD(net_namespace_list);
25 EXPORT_SYMBOL_GPL(net_namespace_list);
28 EXPORT_SYMBOL(init_net);
30 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
32 static int net_assign_generic(struct net *net, int id, void *data)
34 struct net_generic *ng, *old_ng;
36 BUG_ON(!mutex_is_locked(&net_mutex));
39 old_ng = rcu_dereference_protected(net->gen,
40 lockdep_is_held(&net_mutex));
42 if (old_ng->len >= id)
45 ng = kzalloc(sizeof(struct net_generic) +
46 id * sizeof(void *), GFP_KERNEL);
51 * Some synchronisation notes:
53 * The net_generic explores the net->gen array inside rcu
54 * read section. Besides once set the net->gen->ptr[x]
55 * pointer never changes (see rules in netns/generic.h).
57 * That said, we simply duplicate this array and schedule
58 * the old copy for kfree after a grace period.
62 memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
64 rcu_assign_pointer(net->gen, ng);
65 kfree_rcu(old_ng, rcu);
67 ng->ptr[id - 1] = data;
71 static int ops_init(const struct pernet_operations *ops, struct net *net)
74 if (ops->id && ops->size) {
75 void *data = kzalloc(ops->size, GFP_KERNEL);
79 err = net_assign_generic(net, *ops->id, data);
86 return ops->init(net);
90 static void ops_free(const struct pernet_operations *ops, struct net *net)
92 if (ops->id && ops->size) {
94 kfree(net_generic(net, id));
98 static void ops_exit_list(const struct pernet_operations *ops,
99 struct list_head *net_exit_list)
103 list_for_each_entry(net, net_exit_list, exit_list)
107 ops->exit_batch(net_exit_list);
110 static void ops_free_list(const struct pernet_operations *ops,
111 struct list_head *net_exit_list)
114 if (ops->size && ops->id) {
115 list_for_each_entry(net, net_exit_list, exit_list)
121 * setup_net runs the initializers for the network namespace object.
123 static __net_init int setup_net(struct net *net)
125 /* Must be called with net_mutex held */
126 const struct pernet_operations *ops, *saved_ops;
128 LIST_HEAD(net_exit_list);
130 atomic_set(&net->count, 1);
131 atomic_set(&net->passive, 1);
132 net->dev_base_seq = 1;
134 #ifdef NETNS_REFCNT_DEBUG
135 atomic_set(&net->use_count, 0);
138 list_for_each_entry(ops, &pernet_list, list) {
139 error = ops_init(ops, net);
147 /* Walk through the list backwards calling the exit functions
148 * for the pernet modules whose init functions did not fail.
150 list_add(&net->exit_list, &net_exit_list);
152 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
153 ops_exit_list(ops, &net_exit_list);
156 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
157 ops_free_list(ops, &net_exit_list);
163 static struct net_generic *net_alloc_generic(void)
165 struct net_generic *ng;
166 size_t generic_size = sizeof(struct net_generic) +
167 INITIAL_NET_GEN_PTRS * sizeof(void *);
169 ng = kzalloc(generic_size, GFP_KERNEL);
171 ng->len = INITIAL_NET_GEN_PTRS;
177 static struct kmem_cache *net_cachep;
178 static struct workqueue_struct *netns_wq;
180 static struct net *net_alloc(void)
182 struct net *net = NULL;
183 struct net_generic *ng;
185 ng = net_alloc_generic();
189 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
193 rcu_assign_pointer(net->gen, ng);
202 static void net_free(struct net *net)
204 #ifdef NETNS_REFCNT_DEBUG
205 if (unlikely(atomic_read(&net->use_count) != 0)) {
206 printk(KERN_EMERG "network namespace not free! Usage: %d\n",
207 atomic_read(&net->use_count));
212 kmem_cache_free(net_cachep, net);
215 void net_drop_ns(void *p)
218 if (ns && atomic_dec_and_test(&ns->passive))
222 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
227 if (!(flags & CLONE_NEWNET))
228 return get_net(old_net);
232 return ERR_PTR(-ENOMEM);
233 mutex_lock(&net_mutex);
237 list_add_tail_rcu(&net->list, &net_namespace_list);
240 mutex_unlock(&net_mutex);
248 static DEFINE_SPINLOCK(cleanup_list_lock);
249 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
251 static void cleanup_net(struct work_struct *work)
253 const struct pernet_operations *ops;
254 struct net *net, *tmp;
255 LIST_HEAD(net_kill_list);
256 LIST_HEAD(net_exit_list);
258 /* Atomically snapshot the list of namespaces to cleanup */
259 spin_lock_irq(&cleanup_list_lock);
260 list_replace_init(&cleanup_list, &net_kill_list);
261 spin_unlock_irq(&cleanup_list_lock);
263 mutex_lock(&net_mutex);
265 /* Don't let anyone else find us. */
267 list_for_each_entry(net, &net_kill_list, cleanup_list) {
268 list_del_rcu(&net->list);
269 list_add_tail(&net->exit_list, &net_exit_list);
274 * Another CPU might be rcu-iterating the list, wait for it.
275 * This needs to be before calling the exit() notifiers, so
276 * the rcu_barrier() below isn't sufficient alone.
280 /* Run all of the network namespace exit methods */
281 list_for_each_entry_reverse(ops, &pernet_list, list)
282 ops_exit_list(ops, &net_exit_list);
284 /* Free the net generic variables */
285 list_for_each_entry_reverse(ops, &pernet_list, list)
286 ops_free_list(ops, &net_exit_list);
288 mutex_unlock(&net_mutex);
290 /* Ensure there are no outstanding rcu callbacks using this
295 /* Finally it is safe to free my network namespace structure */
296 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
297 list_del_init(&net->exit_list);
301 static DECLARE_WORK(net_cleanup_work, cleanup_net);
303 void __put_net(struct net *net)
305 /* Cleanup the network namespace in process context */
308 spin_lock_irqsave(&cleanup_list_lock, flags);
309 list_add(&net->cleanup_list, &cleanup_list);
310 spin_unlock_irqrestore(&cleanup_list_lock, flags);
312 queue_work(netns_wq, &net_cleanup_work);
314 EXPORT_SYMBOL_GPL(__put_net);
316 struct net *get_net_ns_by_fd(int fd)
318 struct proc_inode *ei;
322 file = proc_ns_fget(fd);
324 return ERR_CAST(file);
326 ei = PROC_I(file->f_dentry->d_inode);
327 if (ei->ns_ops == &netns_operations)
328 net = get_net(ei->ns);
330 net = ERR_PTR(-EINVAL);
337 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
339 if (flags & CLONE_NEWNET)
340 return ERR_PTR(-EINVAL);
344 struct net *get_net_ns_by_fd(int fd)
346 return ERR_PTR(-EINVAL);
350 struct net *get_net_ns_by_pid(pid_t pid)
352 struct task_struct *tsk;
355 /* Lookup the network namespace */
356 net = ERR_PTR(-ESRCH);
358 tsk = find_task_by_vpid(pid);
360 struct nsproxy *nsproxy;
361 nsproxy = task_nsproxy(tsk);
363 net = get_net(nsproxy->net_ns);
368 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
370 static int __init net_ns_init(void)
372 struct net_generic *ng;
375 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
379 /* Create workqueue for cleanup */
380 netns_wq = create_singlethread_workqueue("netns");
382 panic("Could not create netns workq");
385 ng = net_alloc_generic();
387 panic("Could not allocate generic netns");
389 rcu_assign_pointer(init_net.gen, ng);
391 mutex_lock(&net_mutex);
392 if (setup_net(&init_net))
393 panic("Could not setup the initial network namespace");
396 list_add_tail_rcu(&init_net.list, &net_namespace_list);
399 mutex_unlock(&net_mutex);
404 pure_initcall(net_ns_init);
407 static int __register_pernet_operations(struct list_head *list,
408 struct pernet_operations *ops)
412 LIST_HEAD(net_exit_list);
414 list_add_tail(&ops->list, list);
415 if (ops->init || (ops->id && ops->size)) {
417 error = ops_init(ops, net);
420 list_add_tail(&net->exit_list, &net_exit_list);
426 /* If I have an error cleanup all namespaces I initialized */
427 list_del(&ops->list);
428 ops_exit_list(ops, &net_exit_list);
429 ops_free_list(ops, &net_exit_list);
433 static void __unregister_pernet_operations(struct pernet_operations *ops)
436 LIST_HEAD(net_exit_list);
438 list_del(&ops->list);
440 list_add_tail(&net->exit_list, &net_exit_list);
441 ops_exit_list(ops, &net_exit_list);
442 ops_free_list(ops, &net_exit_list);
447 static int __register_pernet_operations(struct list_head *list,
448 struct pernet_operations *ops)
451 err = ops_init(ops, &init_net);
453 ops_free(ops, &init_net);
458 static void __unregister_pernet_operations(struct pernet_operations *ops)
460 LIST_HEAD(net_exit_list);
461 list_add(&init_net.exit_list, &net_exit_list);
462 ops_exit_list(ops, &net_exit_list);
463 ops_free_list(ops, &net_exit_list);
466 #endif /* CONFIG_NET_NS */
468 static DEFINE_IDA(net_generic_ids);
470 static int register_pernet_operations(struct list_head *list,
471 struct pernet_operations *ops)
477 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
479 if (error == -EAGAIN) {
480 ida_pre_get(&net_generic_ids, GFP_KERNEL);
486 error = __register_pernet_operations(list, ops);
490 ida_remove(&net_generic_ids, *ops->id);
496 static void unregister_pernet_operations(struct pernet_operations *ops)
499 __unregister_pernet_operations(ops);
502 ida_remove(&net_generic_ids, *ops->id);
506 * register_pernet_subsys - register a network namespace subsystem
507 * @ops: pernet operations structure for the subsystem
509 * Register a subsystem which has init and exit functions
510 * that are called when network namespaces are created and
511 * destroyed respectively.
513 * When registered all network namespace init functions are
514 * called for every existing network namespace. Allowing kernel
515 * modules to have a race free view of the set of network namespaces.
517 * When a new network namespace is created all of the init
518 * methods are called in the order in which they were registered.
520 * When a network namespace is destroyed all of the exit methods
521 * are called in the reverse of the order with which they were
524 int register_pernet_subsys(struct pernet_operations *ops)
527 mutex_lock(&net_mutex);
528 error = register_pernet_operations(first_device, ops);
529 mutex_unlock(&net_mutex);
532 EXPORT_SYMBOL_GPL(register_pernet_subsys);
535 * unregister_pernet_subsys - unregister a network namespace subsystem
536 * @ops: pernet operations structure to manipulate
538 * Remove the pernet operations structure from the list to be
539 * used when network namespaces are created or destroyed. In
540 * addition run the exit method for all existing network
543 void unregister_pernet_subsys(struct pernet_operations *ops)
545 mutex_lock(&net_mutex);
546 unregister_pernet_operations(ops);
547 mutex_unlock(&net_mutex);
549 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
552 * register_pernet_device - register a network namespace device
553 * @ops: pernet operations structure for the subsystem
555 * Register a device which has init and exit functions
556 * that are called when network namespaces are created and
557 * destroyed respectively.
559 * When registered all network namespace init functions are
560 * called for every existing network namespace. Allowing kernel
561 * modules to have a race free view of the set of network namespaces.
563 * When a new network namespace is created all of the init
564 * methods are called in the order in which they were registered.
566 * When a network namespace is destroyed all of the exit methods
567 * are called in the reverse of the order with which they were
570 int register_pernet_device(struct pernet_operations *ops)
573 mutex_lock(&net_mutex);
574 error = register_pernet_operations(&pernet_list, ops);
575 if (!error && (first_device == &pernet_list))
576 first_device = &ops->list;
577 mutex_unlock(&net_mutex);
580 EXPORT_SYMBOL_GPL(register_pernet_device);
583 * unregister_pernet_device - unregister a network namespace netdevice
584 * @ops: pernet operations structure to manipulate
586 * Remove the pernet operations structure from the list to be
587 * used when network namespaces are created or destroyed. In
588 * addition run the exit method for all existing network
591 void unregister_pernet_device(struct pernet_operations *ops)
593 mutex_lock(&net_mutex);
594 if (&ops->list == first_device)
595 first_device = first_device->next;
596 unregister_pernet_operations(ops);
597 mutex_unlock(&net_mutex);
599 EXPORT_SYMBOL_GPL(unregister_pernet_device);
602 static void *netns_get(struct task_struct *task)
604 struct net *net = NULL;
605 struct nsproxy *nsproxy;
608 nsproxy = task_nsproxy(task);
610 net = get_net(nsproxy->net_ns);
616 static void netns_put(void *ns)
621 static int netns_install(struct nsproxy *nsproxy, void *ns)
623 put_net(nsproxy->net_ns);
624 nsproxy->net_ns = get_net(ns);
628 const struct proc_ns_operations netns_operations = {
630 .type = CLONE_NEWNET,
633 .install = netns_install,