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);
132 #ifdef NETNS_REFCNT_DEBUG
133 atomic_set(&net->use_count, 0);
136 list_for_each_entry(ops, &pernet_list, list) {
137 error = ops_init(ops, net);
145 /* Walk through the list backwards calling the exit functions
146 * for the pernet modules whose init functions did not fail.
148 list_add(&net->exit_list, &net_exit_list);
150 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
151 ops_exit_list(ops, &net_exit_list);
154 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
155 ops_free_list(ops, &net_exit_list);
161 static struct net_generic *net_alloc_generic(void)
163 struct net_generic *ng;
164 size_t generic_size = sizeof(struct net_generic) +
165 INITIAL_NET_GEN_PTRS * sizeof(void *);
167 ng = kzalloc(generic_size, GFP_KERNEL);
169 ng->len = INITIAL_NET_GEN_PTRS;
175 static struct kmem_cache *net_cachep;
176 static struct workqueue_struct *netns_wq;
178 static struct net *net_alloc(void)
180 struct net *net = NULL;
181 struct net_generic *ng;
183 ng = net_alloc_generic();
187 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
191 rcu_assign_pointer(net->gen, ng);
200 static void net_free(struct net *net)
202 #ifdef NETNS_REFCNT_DEBUG
203 if (unlikely(atomic_read(&net->use_count) != 0)) {
204 printk(KERN_EMERG "network namespace not free! Usage: %d\n",
205 atomic_read(&net->use_count));
210 kmem_cache_free(net_cachep, net);
213 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
218 if (!(flags & CLONE_NEWNET))
219 return get_net(old_net);
223 return ERR_PTR(-ENOMEM);
224 mutex_lock(&net_mutex);
228 list_add_tail_rcu(&net->list, &net_namespace_list);
231 mutex_unlock(&net_mutex);
239 static DEFINE_SPINLOCK(cleanup_list_lock);
240 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
242 static void cleanup_net(struct work_struct *work)
244 const struct pernet_operations *ops;
245 struct net *net, *tmp;
246 LIST_HEAD(net_kill_list);
247 LIST_HEAD(net_exit_list);
249 /* Atomically snapshot the list of namespaces to cleanup */
250 spin_lock_irq(&cleanup_list_lock);
251 list_replace_init(&cleanup_list, &net_kill_list);
252 spin_unlock_irq(&cleanup_list_lock);
254 mutex_lock(&net_mutex);
256 /* Don't let anyone else find us. */
258 list_for_each_entry(net, &net_kill_list, cleanup_list) {
259 list_del_rcu(&net->list);
260 list_add_tail(&net->exit_list, &net_exit_list);
265 * Another CPU might be rcu-iterating the list, wait for it.
266 * This needs to be before calling the exit() notifiers, so
267 * the rcu_barrier() below isn't sufficient alone.
271 /* Run all of the network namespace exit methods */
272 list_for_each_entry_reverse(ops, &pernet_list, list)
273 ops_exit_list(ops, &net_exit_list);
275 /* Free the net generic variables */
276 list_for_each_entry_reverse(ops, &pernet_list, list)
277 ops_free_list(ops, &net_exit_list);
279 mutex_unlock(&net_mutex);
281 /* Ensure there are no outstanding rcu callbacks using this
286 /* Finally it is safe to free my network namespace structure */
287 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
288 list_del_init(&net->exit_list);
292 static DECLARE_WORK(net_cleanup_work, cleanup_net);
294 void __put_net(struct net *net)
296 /* Cleanup the network namespace in process context */
299 spin_lock_irqsave(&cleanup_list_lock, flags);
300 list_add(&net->cleanup_list, &cleanup_list);
301 spin_unlock_irqrestore(&cleanup_list_lock, flags);
303 queue_work(netns_wq, &net_cleanup_work);
305 EXPORT_SYMBOL_GPL(__put_net);
307 struct net *get_net_ns_by_fd(int fd)
309 struct proc_inode *ei;
313 net = ERR_PTR(-EINVAL);
314 file = proc_ns_fget(fd);
318 ei = PROC_I(file->f_dentry->d_inode);
319 if (ei->ns_ops != &netns_operations)
322 net = get_net(ei->ns);
330 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
332 if (flags & CLONE_NEWNET)
333 return ERR_PTR(-EINVAL);
337 struct net *get_net_ns_by_fd(int fd)
339 return ERR_PTR(-EINVAL);
343 struct net *get_net_ns_by_pid(pid_t pid)
345 struct task_struct *tsk;
348 /* Lookup the network namespace */
349 net = ERR_PTR(-ESRCH);
351 tsk = find_task_by_vpid(pid);
353 struct nsproxy *nsproxy;
354 nsproxy = task_nsproxy(tsk);
356 net = get_net(nsproxy->net_ns);
361 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
363 static int __init net_ns_init(void)
365 struct net_generic *ng;
368 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
372 /* Create workqueue for cleanup */
373 netns_wq = create_singlethread_workqueue("netns");
375 panic("Could not create netns workq");
378 ng = net_alloc_generic();
380 panic("Could not allocate generic netns");
382 rcu_assign_pointer(init_net.gen, ng);
384 mutex_lock(&net_mutex);
385 if (setup_net(&init_net))
386 panic("Could not setup the initial network namespace");
389 list_add_tail_rcu(&init_net.list, &net_namespace_list);
392 mutex_unlock(&net_mutex);
397 pure_initcall(net_ns_init);
400 static int __register_pernet_operations(struct list_head *list,
401 struct pernet_operations *ops)
405 LIST_HEAD(net_exit_list);
407 list_add_tail(&ops->list, list);
408 if (ops->init || (ops->id && ops->size)) {
410 error = ops_init(ops, net);
413 list_add_tail(&net->exit_list, &net_exit_list);
419 /* If I have an error cleanup all namespaces I initialized */
420 list_del(&ops->list);
421 ops_exit_list(ops, &net_exit_list);
422 ops_free_list(ops, &net_exit_list);
426 static void __unregister_pernet_operations(struct pernet_operations *ops)
429 LIST_HEAD(net_exit_list);
431 list_del(&ops->list);
433 list_add_tail(&net->exit_list, &net_exit_list);
434 ops_exit_list(ops, &net_exit_list);
435 ops_free_list(ops, &net_exit_list);
440 static int __register_pernet_operations(struct list_head *list,
441 struct pernet_operations *ops)
444 err = ops_init(ops, &init_net);
446 ops_free(ops, &init_net);
451 static void __unregister_pernet_operations(struct pernet_operations *ops)
453 LIST_HEAD(net_exit_list);
454 list_add(&init_net.exit_list, &net_exit_list);
455 ops_exit_list(ops, &net_exit_list);
456 ops_free_list(ops, &net_exit_list);
459 #endif /* CONFIG_NET_NS */
461 static DEFINE_IDA(net_generic_ids);
463 static int register_pernet_operations(struct list_head *list,
464 struct pernet_operations *ops)
470 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
472 if (error == -EAGAIN) {
473 ida_pre_get(&net_generic_ids, GFP_KERNEL);
479 error = __register_pernet_operations(list, ops);
483 ida_remove(&net_generic_ids, *ops->id);
489 static void unregister_pernet_operations(struct pernet_operations *ops)
492 __unregister_pernet_operations(ops);
495 ida_remove(&net_generic_ids, *ops->id);
499 * register_pernet_subsys - register a network namespace subsystem
500 * @ops: pernet operations structure for the subsystem
502 * Register a subsystem which has init and exit functions
503 * that are called when network namespaces are created and
504 * destroyed respectively.
506 * When registered all network namespace init functions are
507 * called for every existing network namespace. Allowing kernel
508 * modules to have a race free view of the set of network namespaces.
510 * When a new network namespace is created all of the init
511 * methods are called in the order in which they were registered.
513 * When a network namespace is destroyed all of the exit methods
514 * are called in the reverse of the order with which they were
517 int register_pernet_subsys(struct pernet_operations *ops)
520 mutex_lock(&net_mutex);
521 error = register_pernet_operations(first_device, ops);
522 mutex_unlock(&net_mutex);
525 EXPORT_SYMBOL_GPL(register_pernet_subsys);
528 * unregister_pernet_subsys - unregister a network namespace subsystem
529 * @ops: pernet operations structure to manipulate
531 * Remove the pernet operations structure from the list to be
532 * used when network namespaces are created or destroyed. In
533 * addition run the exit method for all existing network
536 void unregister_pernet_subsys(struct pernet_operations *ops)
538 mutex_lock(&net_mutex);
539 unregister_pernet_operations(ops);
540 mutex_unlock(&net_mutex);
542 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
545 * register_pernet_device - register a network namespace device
546 * @ops: pernet operations structure for the subsystem
548 * Register a device which has init and exit functions
549 * that are called when network namespaces are created and
550 * destroyed respectively.
552 * When registered all network namespace init functions are
553 * called for every existing network namespace. Allowing kernel
554 * modules to have a race free view of the set of network namespaces.
556 * When a new network namespace is created all of the init
557 * methods are called in the order in which they were registered.
559 * When a network namespace is destroyed all of the exit methods
560 * are called in the reverse of the order with which they were
563 int register_pernet_device(struct pernet_operations *ops)
566 mutex_lock(&net_mutex);
567 error = register_pernet_operations(&pernet_list, ops);
568 if (!error && (first_device == &pernet_list))
569 first_device = &ops->list;
570 mutex_unlock(&net_mutex);
573 EXPORT_SYMBOL_GPL(register_pernet_device);
576 * unregister_pernet_device - unregister a network namespace netdevice
577 * @ops: pernet operations structure to manipulate
579 * Remove the pernet operations structure from the list to be
580 * used when network namespaces are created or destroyed. In
581 * addition run the exit method for all existing network
584 void unregister_pernet_device(struct pernet_operations *ops)
586 mutex_lock(&net_mutex);
587 if (&ops->list == first_device)
588 first_device = first_device->next;
589 unregister_pernet_operations(ops);
590 mutex_unlock(&net_mutex);
592 EXPORT_SYMBOL_GPL(unregister_pernet_device);
595 static void *netns_get(struct task_struct *task)
597 struct net *net = NULL;
598 struct nsproxy *nsproxy;
601 nsproxy = task_nsproxy(task);
603 net = get_net(nsproxy->net_ns);
609 static void netns_put(void *ns)
614 static int netns_install(struct nsproxy *nsproxy, void *ns)
616 put_net(nsproxy->net_ns);
617 nsproxy->net_ns = get_net(ns);
621 const struct proc_ns_operations netns_operations = {
623 .type = CLONE_NEWNET,
626 .install = netns_install,