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>
14 #include <linux/proc_ns.h>
15 #include <linux/file.h>
16 #include <linux/export.h>
17 #include <linux/user_namespace.h>
18 #include <linux/net_namespace.h>
19 #include <linux/rtnetlink.h>
21 #include <net/netlink.h>
22 #include <net/net_namespace.h>
23 #include <net/netns/generic.h>
26 * Our network namespace constructor/destructor lists
29 static LIST_HEAD(pernet_list);
30 static struct list_head *first_device = &pernet_list;
31 DEFINE_MUTEX(net_mutex);
33 LIST_HEAD(net_namespace_list);
34 EXPORT_SYMBOL_GPL(net_namespace_list);
36 struct net init_net = {
37 .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head),
39 EXPORT_SYMBOL(init_net);
41 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
43 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
45 static struct net_generic *net_alloc_generic(void)
47 struct net_generic *ng;
48 size_t generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
50 ng = kzalloc(generic_size, GFP_KERNEL);
52 ng->len = max_gen_ptrs;
57 static int net_assign_generic(struct net *net, int id, void *data)
59 struct net_generic *ng, *old_ng;
61 BUG_ON(!mutex_is_locked(&net_mutex));
64 old_ng = rcu_dereference_protected(net->gen,
65 lockdep_is_held(&net_mutex));
67 if (old_ng->len >= id)
70 ng = net_alloc_generic();
75 * Some synchronisation notes:
77 * The net_generic explores the net->gen array inside rcu
78 * read section. Besides once set the net->gen->ptr[x]
79 * pointer never changes (see rules in netns/generic.h).
81 * That said, we simply duplicate this array and schedule
82 * the old copy for kfree after a grace period.
85 memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
87 rcu_assign_pointer(net->gen, ng);
88 kfree_rcu(old_ng, rcu);
90 ng->ptr[id - 1] = data;
94 static int ops_init(const struct pernet_operations *ops, struct net *net)
99 if (ops->id && ops->size) {
100 data = kzalloc(ops->size, GFP_KERNEL);
104 err = net_assign_generic(net, *ops->id, data);
110 err = ops->init(net);
121 static void ops_free(const struct pernet_operations *ops, struct net *net)
123 if (ops->id && ops->size) {
125 kfree(net_generic(net, id));
129 static void ops_exit_list(const struct pernet_operations *ops,
130 struct list_head *net_exit_list)
134 list_for_each_entry(net, net_exit_list, exit_list)
138 ops->exit_batch(net_exit_list);
141 static void ops_free_list(const struct pernet_operations *ops,
142 struct list_head *net_exit_list)
145 if (ops->size && ops->id) {
146 list_for_each_entry(net, net_exit_list, exit_list)
151 static int alloc_netid(struct net *net, struct net *peer, int reqid)
153 int min = 0, max = 0;
162 return idr_alloc(&net->netns_ids, peer, min, max, GFP_KERNEL);
165 /* This function is used by idr_for_each(). If net is equal to peer, the
166 * function returns the id so that idr_for_each() stops. Because we cannot
167 * returns the id 0 (idr_for_each() will not stop), we return the magic value
168 * NET_ID_ZERO (-1) for it.
170 #define NET_ID_ZERO -1
171 static int net_eq_idr(int id, void *net, void *peer)
173 if (net_eq(net, peer))
174 return id ? : NET_ID_ZERO;
178 static int __peernet2id(struct net *net, struct net *peer, bool alloc)
180 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
184 /* Magic value for id 0. */
185 if (id == NET_ID_ZERO)
191 return alloc_netid(net, peer, -1);
196 /* This function returns the id of a peer netns. If no id is assigned, one will
197 * be allocated and returned.
199 int peernet2id(struct net *net, struct net *peer)
201 int id = __peernet2id(net, peer, true);
203 return id >= 0 ? id : NETNSA_NSID_NOT_ASSIGNED;
205 EXPORT_SYMBOL(peernet2id);
207 struct net *get_net_ns_by_id(struct net *net, int id)
215 peer = idr_find(&net->netns_ids, id);
224 * setup_net runs the initializers for the network namespace object.
226 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
228 /* Must be called with net_mutex held */
229 const struct pernet_operations *ops, *saved_ops;
231 LIST_HEAD(net_exit_list);
233 atomic_set(&net->count, 1);
234 atomic_set(&net->passive, 1);
235 net->dev_base_seq = 1;
236 net->user_ns = user_ns;
237 idr_init(&net->netns_ids);
239 list_for_each_entry(ops, &pernet_list, list) {
240 error = ops_init(ops, net);
248 /* Walk through the list backwards calling the exit functions
249 * for the pernet modules whose init functions did not fail.
251 list_add(&net->exit_list, &net_exit_list);
253 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
254 ops_exit_list(ops, &net_exit_list);
257 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
258 ops_free_list(ops, &net_exit_list);
266 static struct kmem_cache *net_cachep;
267 static struct workqueue_struct *netns_wq;
269 static struct net *net_alloc(void)
271 struct net *net = NULL;
272 struct net_generic *ng;
274 ng = net_alloc_generic();
278 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
282 rcu_assign_pointer(net->gen, ng);
291 static void net_free(struct net *net)
293 kfree(rcu_access_pointer(net->gen));
294 kmem_cache_free(net_cachep, net);
297 void net_drop_ns(void *p)
300 if (ns && atomic_dec_and_test(&ns->passive))
304 struct net *copy_net_ns(unsigned long flags,
305 struct user_namespace *user_ns, struct net *old_net)
310 if (!(flags & CLONE_NEWNET))
311 return get_net(old_net);
315 return ERR_PTR(-ENOMEM);
317 get_user_ns(user_ns);
319 mutex_lock(&net_mutex);
320 rv = setup_net(net, user_ns);
323 list_add_tail_rcu(&net->list, &net_namespace_list);
326 mutex_unlock(&net_mutex);
328 put_user_ns(user_ns);
335 static DEFINE_SPINLOCK(cleanup_list_lock);
336 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
338 static void cleanup_net(struct work_struct *work)
340 const struct pernet_operations *ops;
341 struct net *net, *tmp;
342 struct list_head net_kill_list;
343 LIST_HEAD(net_exit_list);
345 /* Atomically snapshot the list of namespaces to cleanup */
346 spin_lock_irq(&cleanup_list_lock);
347 list_replace_init(&cleanup_list, &net_kill_list);
348 spin_unlock_irq(&cleanup_list_lock);
350 mutex_lock(&net_mutex);
352 /* Don't let anyone else find us. */
354 list_for_each_entry(net, &net_kill_list, cleanup_list) {
355 list_del_rcu(&net->list);
356 list_add_tail(&net->exit_list, &net_exit_list);
358 int id = __peernet2id(tmp, net, false);
361 idr_remove(&tmp->netns_ids, id);
363 idr_destroy(&net->netns_ids);
369 * Another CPU might be rcu-iterating the list, wait for it.
370 * This needs to be before calling the exit() notifiers, so
371 * the rcu_barrier() below isn't sufficient alone.
375 /* Run all of the network namespace exit methods */
376 list_for_each_entry_reverse(ops, &pernet_list, list)
377 ops_exit_list(ops, &net_exit_list);
379 /* Free the net generic variables */
380 list_for_each_entry_reverse(ops, &pernet_list, list)
381 ops_free_list(ops, &net_exit_list);
383 mutex_unlock(&net_mutex);
385 /* Ensure there are no outstanding rcu callbacks using this
390 /* Finally it is safe to free my network namespace structure */
391 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
392 list_del_init(&net->exit_list);
393 put_user_ns(net->user_ns);
397 static DECLARE_WORK(net_cleanup_work, cleanup_net);
399 void __put_net(struct net *net)
401 /* Cleanup the network namespace in process context */
404 spin_lock_irqsave(&cleanup_list_lock, flags);
405 list_add(&net->cleanup_list, &cleanup_list);
406 spin_unlock_irqrestore(&cleanup_list_lock, flags);
408 queue_work(netns_wq, &net_cleanup_work);
410 EXPORT_SYMBOL_GPL(__put_net);
412 struct net *get_net_ns_by_fd(int fd)
415 struct ns_common *ns;
418 file = proc_ns_fget(fd);
420 return ERR_CAST(file);
422 ns = get_proc_ns(file_inode(file));
423 if (ns->ops == &netns_operations)
424 net = get_net(container_of(ns, struct net, ns));
426 net = ERR_PTR(-EINVAL);
433 struct net *get_net_ns_by_fd(int fd)
435 return ERR_PTR(-EINVAL);
438 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
440 struct net *get_net_ns_by_pid(pid_t pid)
442 struct task_struct *tsk;
445 /* Lookup the network namespace */
446 net = ERR_PTR(-ESRCH);
448 tsk = find_task_by_vpid(pid);
450 struct nsproxy *nsproxy;
452 nsproxy = tsk->nsproxy;
454 net = get_net(nsproxy->net_ns);
460 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
462 static __net_init int net_ns_net_init(struct net *net)
465 net->ns.ops = &netns_operations;
467 return ns_alloc_inum(&net->ns);
470 static __net_exit void net_ns_net_exit(struct net *net)
472 ns_free_inum(&net->ns);
475 static struct pernet_operations __net_initdata net_ns_ops = {
476 .init = net_ns_net_init,
477 .exit = net_ns_net_exit,
480 static struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
481 [NETNSA_NONE] = { .type = NLA_UNSPEC },
482 [NETNSA_NSID] = { .type = NLA_S32 },
483 [NETNSA_PID] = { .type = NLA_U32 },
484 [NETNSA_FD] = { .type = NLA_U32 },
487 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh)
489 struct net *net = sock_net(skb->sk);
490 struct nlattr *tb[NETNSA_MAX + 1];
494 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
498 if (!tb[NETNSA_NSID])
500 nsid = nla_get_s32(tb[NETNSA_NSID]);
503 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
504 else if (tb[NETNSA_FD])
505 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
509 return PTR_ERR(peer);
511 if (__peernet2id(net, peer, false) >= 0) {
516 err = alloc_netid(net, peer, nsid);
524 static int rtnl_net_get_size(void)
526 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
527 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
531 static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags,
532 int cmd, struct net *net, struct net *peer)
534 struct nlmsghdr *nlh;
535 struct rtgenmsg *rth;
540 nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags);
544 rth = nlmsg_data(nlh);
545 rth->rtgen_family = AF_UNSPEC;
547 id = __peernet2id(net, peer, false);
549 id = NETNSA_NSID_NOT_ASSIGNED;
550 if (nla_put_s32(skb, NETNSA_NSID, id))
551 goto nla_put_failure;
557 nlmsg_cancel(skb, nlh);
561 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh)
563 struct net *net = sock_net(skb->sk);
564 struct nlattr *tb[NETNSA_MAX + 1];
569 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
574 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
575 else if (tb[NETNSA_FD])
576 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
581 return PTR_ERR(peer);
583 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
589 err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0,
590 RTM_GETNSID, net, peer);
594 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
604 static int __init net_ns_init(void)
606 struct net_generic *ng;
609 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
613 /* Create workqueue for cleanup */
614 netns_wq = create_singlethread_workqueue("netns");
616 panic("Could not create netns workq");
619 ng = net_alloc_generic();
621 panic("Could not allocate generic netns");
623 rcu_assign_pointer(init_net.gen, ng);
625 mutex_lock(&net_mutex);
626 if (setup_net(&init_net, &init_user_ns))
627 panic("Could not setup the initial network namespace");
630 list_add_tail_rcu(&init_net.list, &net_namespace_list);
633 mutex_unlock(&net_mutex);
635 register_pernet_subsys(&net_ns_ops);
637 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL, NULL);
638 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, NULL, NULL);
643 pure_initcall(net_ns_init);
646 static int __register_pernet_operations(struct list_head *list,
647 struct pernet_operations *ops)
651 LIST_HEAD(net_exit_list);
653 list_add_tail(&ops->list, list);
654 if (ops->init || (ops->id && ops->size)) {
656 error = ops_init(ops, net);
659 list_add_tail(&net->exit_list, &net_exit_list);
665 /* If I have an error cleanup all namespaces I initialized */
666 list_del(&ops->list);
667 ops_exit_list(ops, &net_exit_list);
668 ops_free_list(ops, &net_exit_list);
672 static void __unregister_pernet_operations(struct pernet_operations *ops)
675 LIST_HEAD(net_exit_list);
677 list_del(&ops->list);
679 list_add_tail(&net->exit_list, &net_exit_list);
680 ops_exit_list(ops, &net_exit_list);
681 ops_free_list(ops, &net_exit_list);
686 static int __register_pernet_operations(struct list_head *list,
687 struct pernet_operations *ops)
689 return ops_init(ops, &init_net);
692 static void __unregister_pernet_operations(struct pernet_operations *ops)
694 LIST_HEAD(net_exit_list);
695 list_add(&init_net.exit_list, &net_exit_list);
696 ops_exit_list(ops, &net_exit_list);
697 ops_free_list(ops, &net_exit_list);
700 #endif /* CONFIG_NET_NS */
702 static DEFINE_IDA(net_generic_ids);
704 static int register_pernet_operations(struct list_head *list,
705 struct pernet_operations *ops)
711 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
713 if (error == -EAGAIN) {
714 ida_pre_get(&net_generic_ids, GFP_KERNEL);
719 max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
721 error = __register_pernet_operations(list, ops);
725 ida_remove(&net_generic_ids, *ops->id);
731 static void unregister_pernet_operations(struct pernet_operations *ops)
734 __unregister_pernet_operations(ops);
737 ida_remove(&net_generic_ids, *ops->id);
741 * register_pernet_subsys - register a network namespace subsystem
742 * @ops: pernet operations structure for the subsystem
744 * Register a subsystem which has init and exit functions
745 * that are called when network namespaces are created and
746 * destroyed respectively.
748 * When registered all network namespace init functions are
749 * called for every existing network namespace. Allowing kernel
750 * modules to have a race free view of the set of network namespaces.
752 * When a new network namespace is created all of the init
753 * methods are called in the order in which they were registered.
755 * When a network namespace is destroyed all of the exit methods
756 * are called in the reverse of the order with which they were
759 int register_pernet_subsys(struct pernet_operations *ops)
762 mutex_lock(&net_mutex);
763 error = register_pernet_operations(first_device, ops);
764 mutex_unlock(&net_mutex);
767 EXPORT_SYMBOL_GPL(register_pernet_subsys);
770 * unregister_pernet_subsys - unregister a network namespace subsystem
771 * @ops: pernet operations structure to manipulate
773 * Remove the pernet operations structure from the list to be
774 * used when network namespaces are created or destroyed. In
775 * addition run the exit method for all existing network
778 void unregister_pernet_subsys(struct pernet_operations *ops)
780 mutex_lock(&net_mutex);
781 unregister_pernet_operations(ops);
782 mutex_unlock(&net_mutex);
784 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
787 * register_pernet_device - register a network namespace device
788 * @ops: pernet operations structure for the subsystem
790 * Register a device which has init and exit functions
791 * that are called when network namespaces are created and
792 * destroyed respectively.
794 * When registered all network namespace init functions are
795 * called for every existing network namespace. Allowing kernel
796 * modules to have a race free view of the set of network namespaces.
798 * When a new network namespace is created all of the init
799 * methods are called in the order in which they were registered.
801 * When a network namespace is destroyed all of the exit methods
802 * are called in the reverse of the order with which they were
805 int register_pernet_device(struct pernet_operations *ops)
808 mutex_lock(&net_mutex);
809 error = register_pernet_operations(&pernet_list, ops);
810 if (!error && (first_device == &pernet_list))
811 first_device = &ops->list;
812 mutex_unlock(&net_mutex);
815 EXPORT_SYMBOL_GPL(register_pernet_device);
818 * unregister_pernet_device - unregister a network namespace netdevice
819 * @ops: pernet operations structure to manipulate
821 * Remove the pernet operations structure from the list to be
822 * used when network namespaces are created or destroyed. In
823 * addition run the exit method for all existing network
826 void unregister_pernet_device(struct pernet_operations *ops)
828 mutex_lock(&net_mutex);
829 if (&ops->list == first_device)
830 first_device = first_device->next;
831 unregister_pernet_operations(ops);
832 mutex_unlock(&net_mutex);
834 EXPORT_SYMBOL_GPL(unregister_pernet_device);
837 static struct ns_common *netns_get(struct task_struct *task)
839 struct net *net = NULL;
840 struct nsproxy *nsproxy;
843 nsproxy = task->nsproxy;
845 net = get_net(nsproxy->net_ns);
848 return net ? &net->ns : NULL;
851 static inline struct net *to_net_ns(struct ns_common *ns)
853 return container_of(ns, struct net, ns);
856 static void netns_put(struct ns_common *ns)
858 put_net(to_net_ns(ns));
861 static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
863 struct net *net = to_net_ns(ns);
865 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
866 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
869 put_net(nsproxy->net_ns);
870 nsproxy->net_ns = get_net(net);
874 const struct proc_ns_operations netns_operations = {
876 .type = CLONE_NEWNET,
879 .install = netns_install,