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 <linux/export.h>
14 #include <net/net_namespace.h>
15 #include <net/netns/generic.h>
18 * Our network namespace constructor/destructor lists
21 static LIST_HEAD(pernet_list);
22 static struct list_head *first_device = &pernet_list;
23 static DEFINE_MUTEX(net_mutex);
25 LIST_HEAD(net_namespace_list);
26 EXPORT_SYMBOL_GPL(net_namespace_list);
29 EXPORT_SYMBOL(init_net);
31 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
33 static int net_assign_generic(struct net *net, int id, void *data)
35 struct net_generic *ng, *old_ng;
37 BUG_ON(!mutex_is_locked(&net_mutex));
40 old_ng = rcu_dereference_protected(net->gen,
41 lockdep_is_held(&net_mutex));
43 if (old_ng->len >= id)
46 ng = kzalloc(sizeof(struct net_generic) +
47 id * sizeof(void *), GFP_KERNEL);
52 * Some synchronisation notes:
54 * The net_generic explores the net->gen array inside rcu
55 * read section. Besides once set the net->gen->ptr[x]
56 * pointer never changes (see rules in netns/generic.h).
58 * That said, we simply duplicate this array and schedule
59 * the old copy for kfree after a grace period.
63 memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
65 rcu_assign_pointer(net->gen, ng);
66 kfree_rcu(old_ng, rcu);
68 ng->ptr[id - 1] = data;
72 static int ops_init(const struct pernet_operations *ops, struct net *net)
75 if (ops->id && ops->size) {
76 void *data = kzalloc(ops->size, GFP_KERNEL);
80 err = net_assign_generic(net, *ops->id, data);
87 return ops->init(net);
91 static void ops_free(const struct pernet_operations *ops, struct net *net)
93 if (ops->id && ops->size) {
95 kfree(net_generic(net, id));
99 static void ops_exit_list(const struct pernet_operations *ops,
100 struct list_head *net_exit_list)
104 list_for_each_entry(net, net_exit_list, exit_list)
108 ops->exit_batch(net_exit_list);
111 static void ops_free_list(const struct pernet_operations *ops,
112 struct list_head *net_exit_list)
115 if (ops->size && ops->id) {
116 list_for_each_entry(net, net_exit_list, exit_list)
122 * setup_net runs the initializers for the network namespace object.
124 static __net_init int setup_net(struct net *net)
126 /* Must be called with net_mutex held */
127 const struct pernet_operations *ops, *saved_ops;
129 LIST_HEAD(net_exit_list);
131 atomic_set(&net->count, 1);
132 atomic_set(&net->passive, 1);
133 net->dev_base_seq = 1;
135 #ifdef NETNS_REFCNT_DEBUG
136 atomic_set(&net->use_count, 0);
139 list_for_each_entry(ops, &pernet_list, list) {
140 error = ops_init(ops, net);
148 /* Walk through the list backwards calling the exit functions
149 * for the pernet modules whose init functions did not fail.
151 list_add(&net->exit_list, &net_exit_list);
153 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
154 ops_exit_list(ops, &net_exit_list);
157 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
158 ops_free_list(ops, &net_exit_list);
164 static struct net_generic *net_alloc_generic(void)
166 struct net_generic *ng;
167 size_t generic_size = sizeof(struct net_generic) +
168 INITIAL_NET_GEN_PTRS * sizeof(void *);
170 ng = kzalloc(generic_size, GFP_KERNEL);
172 ng->len = INITIAL_NET_GEN_PTRS;
178 static struct kmem_cache *net_cachep;
179 static struct workqueue_struct *netns_wq;
181 static struct net *net_alloc(void)
183 struct net *net = NULL;
184 struct net_generic *ng;
186 ng = net_alloc_generic();
190 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
194 rcu_assign_pointer(net->gen, ng);
203 static void net_free(struct net *net)
205 #ifdef NETNS_REFCNT_DEBUG
206 if (unlikely(atomic_read(&net->use_count) != 0)) {
207 printk(KERN_EMERG "network namespace not free! Usage: %d\n",
208 atomic_read(&net->use_count));
213 kmem_cache_free(net_cachep, net);
216 void net_drop_ns(void *p)
219 if (ns && atomic_dec_and_test(&ns->passive))
223 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
228 if (!(flags & CLONE_NEWNET))
229 return get_net(old_net);
233 return ERR_PTR(-ENOMEM);
234 mutex_lock(&net_mutex);
238 list_add_tail_rcu(&net->list, &net_namespace_list);
241 mutex_unlock(&net_mutex);
249 static DEFINE_SPINLOCK(cleanup_list_lock);
250 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
252 static void cleanup_net(struct work_struct *work)
254 const struct pernet_operations *ops;
255 struct net *net, *tmp;
256 LIST_HEAD(net_kill_list);
257 LIST_HEAD(net_exit_list);
259 /* Atomically snapshot the list of namespaces to cleanup */
260 spin_lock_irq(&cleanup_list_lock);
261 list_replace_init(&cleanup_list, &net_kill_list);
262 spin_unlock_irq(&cleanup_list_lock);
264 mutex_lock(&net_mutex);
266 /* Don't let anyone else find us. */
268 list_for_each_entry(net, &net_kill_list, cleanup_list) {
269 list_del_rcu(&net->list);
270 list_add_tail(&net->exit_list, &net_exit_list);
275 * Another CPU might be rcu-iterating the list, wait for it.
276 * This needs to be before calling the exit() notifiers, so
277 * the rcu_barrier() below isn't sufficient alone.
281 /* Run all of the network namespace exit methods */
282 list_for_each_entry_reverse(ops, &pernet_list, list)
283 ops_exit_list(ops, &net_exit_list);
285 /* Free the net generic variables */
286 list_for_each_entry_reverse(ops, &pernet_list, list)
287 ops_free_list(ops, &net_exit_list);
289 mutex_unlock(&net_mutex);
291 /* Ensure there are no outstanding rcu callbacks using this
296 /* Finally it is safe to free my network namespace structure */
297 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
298 list_del_init(&net->exit_list);
302 static DECLARE_WORK(net_cleanup_work, cleanup_net);
304 void __put_net(struct net *net)
306 /* Cleanup the network namespace in process context */
309 spin_lock_irqsave(&cleanup_list_lock, flags);
310 list_add(&net->cleanup_list, &cleanup_list);
311 spin_unlock_irqrestore(&cleanup_list_lock, flags);
313 queue_work(netns_wq, &net_cleanup_work);
315 EXPORT_SYMBOL_GPL(__put_net);
317 struct net *get_net_ns_by_fd(int fd)
319 struct proc_inode *ei;
323 file = proc_ns_fget(fd);
325 return ERR_CAST(file);
327 ei = PROC_I(file->f_dentry->d_inode);
328 if (ei->ns_ops == &netns_operations)
329 net = get_net(ei->ns);
331 net = ERR_PTR(-EINVAL);
338 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
340 if (flags & CLONE_NEWNET)
341 return ERR_PTR(-EINVAL);
345 struct net *get_net_ns_by_fd(int fd)
347 return ERR_PTR(-EINVAL);
351 struct net *get_net_ns_by_pid(pid_t pid)
353 struct task_struct *tsk;
356 /* Lookup the network namespace */
357 net = ERR_PTR(-ESRCH);
359 tsk = find_task_by_vpid(pid);
361 struct nsproxy *nsproxy;
362 nsproxy = task_nsproxy(tsk);
364 net = get_net(nsproxy->net_ns);
369 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
371 static int __init net_ns_init(void)
373 struct net_generic *ng;
376 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
380 /* Create workqueue for cleanup */
381 netns_wq = create_singlethread_workqueue("netns");
383 panic("Could not create netns workq");
386 ng = net_alloc_generic();
388 panic("Could not allocate generic netns");
390 rcu_assign_pointer(init_net.gen, ng);
392 mutex_lock(&net_mutex);
393 if (setup_net(&init_net))
394 panic("Could not setup the initial network namespace");
397 list_add_tail_rcu(&init_net.list, &net_namespace_list);
400 mutex_unlock(&net_mutex);
405 pure_initcall(net_ns_init);
408 static int __register_pernet_operations(struct list_head *list,
409 struct pernet_operations *ops)
413 LIST_HEAD(net_exit_list);
415 list_add_tail(&ops->list, list);
416 if (ops->init || (ops->id && ops->size)) {
418 error = ops_init(ops, net);
421 list_add_tail(&net->exit_list, &net_exit_list);
427 /* If I have an error cleanup all namespaces I initialized */
428 list_del(&ops->list);
429 ops_exit_list(ops, &net_exit_list);
430 ops_free_list(ops, &net_exit_list);
434 static void __unregister_pernet_operations(struct pernet_operations *ops)
437 LIST_HEAD(net_exit_list);
439 list_del(&ops->list);
441 list_add_tail(&net->exit_list, &net_exit_list);
442 ops_exit_list(ops, &net_exit_list);
443 ops_free_list(ops, &net_exit_list);
448 static int __register_pernet_operations(struct list_head *list,
449 struct pernet_operations *ops)
452 err = ops_init(ops, &init_net);
454 ops_free(ops, &init_net);
459 static void __unregister_pernet_operations(struct pernet_operations *ops)
461 LIST_HEAD(net_exit_list);
462 list_add(&init_net.exit_list, &net_exit_list);
463 ops_exit_list(ops, &net_exit_list);
464 ops_free_list(ops, &net_exit_list);
467 #endif /* CONFIG_NET_NS */
469 static DEFINE_IDA(net_generic_ids);
471 static int register_pernet_operations(struct list_head *list,
472 struct pernet_operations *ops)
478 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
480 if (error == -EAGAIN) {
481 ida_pre_get(&net_generic_ids, GFP_KERNEL);
487 error = __register_pernet_operations(list, ops);
491 ida_remove(&net_generic_ids, *ops->id);
497 static void unregister_pernet_operations(struct pernet_operations *ops)
500 __unregister_pernet_operations(ops);
503 ida_remove(&net_generic_ids, *ops->id);
507 * register_pernet_subsys - register a network namespace subsystem
508 * @ops: pernet operations structure for the subsystem
510 * Register a subsystem which has init and exit functions
511 * that are called when network namespaces are created and
512 * destroyed respectively.
514 * When registered all network namespace init functions are
515 * called for every existing network namespace. Allowing kernel
516 * modules to have a race free view of the set of network namespaces.
518 * When a new network namespace is created all of the init
519 * methods are called in the order in which they were registered.
521 * When a network namespace is destroyed all of the exit methods
522 * are called in the reverse of the order with which they were
525 int register_pernet_subsys(struct pernet_operations *ops)
528 mutex_lock(&net_mutex);
529 error = register_pernet_operations(first_device, ops);
530 mutex_unlock(&net_mutex);
533 EXPORT_SYMBOL_GPL(register_pernet_subsys);
536 * unregister_pernet_subsys - unregister a network namespace subsystem
537 * @ops: pernet operations structure to manipulate
539 * Remove the pernet operations structure from the list to be
540 * used when network namespaces are created or destroyed. In
541 * addition run the exit method for all existing network
544 void unregister_pernet_subsys(struct pernet_operations *ops)
546 mutex_lock(&net_mutex);
547 unregister_pernet_operations(ops);
548 mutex_unlock(&net_mutex);
550 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
553 * register_pernet_device - register a network namespace device
554 * @ops: pernet operations structure for the subsystem
556 * Register a device which has init and exit functions
557 * that are called when network namespaces are created and
558 * destroyed respectively.
560 * When registered all network namespace init functions are
561 * called for every existing network namespace. Allowing kernel
562 * modules to have a race free view of the set of network namespaces.
564 * When a new network namespace is created all of the init
565 * methods are called in the order in which they were registered.
567 * When a network namespace is destroyed all of the exit methods
568 * are called in the reverse of the order with which they were
571 int register_pernet_device(struct pernet_operations *ops)
574 mutex_lock(&net_mutex);
575 error = register_pernet_operations(&pernet_list, ops);
576 if (!error && (first_device == &pernet_list))
577 first_device = &ops->list;
578 mutex_unlock(&net_mutex);
581 EXPORT_SYMBOL_GPL(register_pernet_device);
584 * unregister_pernet_device - unregister a network namespace netdevice
585 * @ops: pernet operations structure to manipulate
587 * Remove the pernet operations structure from the list to be
588 * used when network namespaces are created or destroyed. In
589 * addition run the exit method for all existing network
592 void unregister_pernet_device(struct pernet_operations *ops)
594 mutex_lock(&net_mutex);
595 if (&ops->list == first_device)
596 first_device = first_device->next;
597 unregister_pernet_operations(ops);
598 mutex_unlock(&net_mutex);
600 EXPORT_SYMBOL_GPL(unregister_pernet_device);
603 static void *netns_get(struct task_struct *task)
605 struct net *net = NULL;
606 struct nsproxy *nsproxy;
609 nsproxy = task_nsproxy(task);
611 net = get_net(nsproxy->net_ns);
617 static void netns_put(void *ns)
622 static int netns_install(struct nsproxy *nsproxy, void *ns)
624 put_net(nsproxy->net_ns);
625 nsproxy->net_ns = get_net(ns);
629 const struct proc_ns_operations netns_operations = {
631 .type = CLONE_NEWNET,
634 .install = netns_install,