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 bool alloc = atomic_read(&peer->count) == 0 ? false : true;
204 id = __peernet2id(net, peer, alloc);
205 return id >= 0 ? id : NETNSA_NSID_NOT_ASSIGNED;
207 EXPORT_SYMBOL(peernet2id);
209 struct net *get_net_ns_by_id(struct net *net, int id)
217 peer = idr_find(&net->netns_ids, id);
226 * setup_net runs the initializers for the network namespace object.
228 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
230 /* Must be called with net_mutex held */
231 const struct pernet_operations *ops, *saved_ops;
233 LIST_HEAD(net_exit_list);
235 atomic_set(&net->count, 1);
236 atomic_set(&net->passive, 1);
237 net->dev_base_seq = 1;
238 net->user_ns = user_ns;
239 idr_init(&net->netns_ids);
241 #ifdef NETNS_REFCNT_DEBUG
242 atomic_set(&net->use_count, 0);
245 list_for_each_entry(ops, &pernet_list, list) {
246 error = ops_init(ops, net);
254 /* Walk through the list backwards calling the exit functions
255 * for the pernet modules whose init functions did not fail.
257 list_add(&net->exit_list, &net_exit_list);
259 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
260 ops_exit_list(ops, &net_exit_list);
263 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
264 ops_free_list(ops, &net_exit_list);
272 static struct kmem_cache *net_cachep;
273 static struct workqueue_struct *netns_wq;
275 static struct net *net_alloc(void)
277 struct net *net = NULL;
278 struct net_generic *ng;
280 ng = net_alloc_generic();
284 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
288 rcu_assign_pointer(net->gen, ng);
297 static void net_free(struct net *net)
299 #ifdef NETNS_REFCNT_DEBUG
300 if (unlikely(atomic_read(&net->use_count) != 0)) {
301 pr_emerg("network namespace not free! Usage: %d\n",
302 atomic_read(&net->use_count));
306 kfree(rcu_access_pointer(net->gen));
307 kmem_cache_free(net_cachep, net);
310 void net_drop_ns(void *p)
313 if (ns && atomic_dec_and_test(&ns->passive))
317 struct net *copy_net_ns(unsigned long flags,
318 struct user_namespace *user_ns, struct net *old_net)
323 if (!(flags & CLONE_NEWNET))
324 return get_net(old_net);
328 return ERR_PTR(-ENOMEM);
330 get_user_ns(user_ns);
332 mutex_lock(&net_mutex);
333 rv = setup_net(net, user_ns);
336 list_add_tail_rcu(&net->list, &net_namespace_list);
339 mutex_unlock(&net_mutex);
341 put_user_ns(user_ns);
348 static DEFINE_SPINLOCK(cleanup_list_lock);
349 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
351 static void cleanup_net(struct work_struct *work)
353 const struct pernet_operations *ops;
354 struct net *net, *tmp;
355 struct list_head net_kill_list;
356 LIST_HEAD(net_exit_list);
358 /* Atomically snapshot the list of namespaces to cleanup */
359 spin_lock_irq(&cleanup_list_lock);
360 list_replace_init(&cleanup_list, &net_kill_list);
361 spin_unlock_irq(&cleanup_list_lock);
363 mutex_lock(&net_mutex);
365 /* Don't let anyone else find us. */
367 list_for_each_entry(net, &net_kill_list, cleanup_list) {
368 list_del_rcu(&net->list);
369 list_add_tail(&net->exit_list, &net_exit_list);
371 int id = __peernet2id(tmp, net, false);
374 idr_remove(&tmp->netns_ids, id);
376 idr_destroy(&net->netns_ids);
382 * Another CPU might be rcu-iterating the list, wait for it.
383 * This needs to be before calling the exit() notifiers, so
384 * the rcu_barrier() below isn't sufficient alone.
388 /* Run all of the network namespace exit methods */
389 list_for_each_entry_reverse(ops, &pernet_list, list)
390 ops_exit_list(ops, &net_exit_list);
392 /* Free the net generic variables */
393 list_for_each_entry_reverse(ops, &pernet_list, list)
394 ops_free_list(ops, &net_exit_list);
396 mutex_unlock(&net_mutex);
398 /* Ensure there are no outstanding rcu callbacks using this
403 /* Finally it is safe to free my network namespace structure */
404 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
405 list_del_init(&net->exit_list);
406 put_user_ns(net->user_ns);
410 static DECLARE_WORK(net_cleanup_work, cleanup_net);
412 void __put_net(struct net *net)
414 /* Cleanup the network namespace in process context */
417 spin_lock_irqsave(&cleanup_list_lock, flags);
418 list_add(&net->cleanup_list, &cleanup_list);
419 spin_unlock_irqrestore(&cleanup_list_lock, flags);
421 queue_work(netns_wq, &net_cleanup_work);
423 EXPORT_SYMBOL_GPL(__put_net);
425 struct net *get_net_ns_by_fd(int fd)
428 struct ns_common *ns;
431 file = proc_ns_fget(fd);
433 return ERR_CAST(file);
435 ns = get_proc_ns(file_inode(file));
436 if (ns->ops == &netns_operations)
437 net = get_net(container_of(ns, struct net, ns));
439 net = ERR_PTR(-EINVAL);
446 struct net *get_net_ns_by_fd(int fd)
448 return ERR_PTR(-EINVAL);
451 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
453 struct net *get_net_ns_by_pid(pid_t pid)
455 struct task_struct *tsk;
458 /* Lookup the network namespace */
459 net = ERR_PTR(-ESRCH);
461 tsk = find_task_by_vpid(pid);
463 struct nsproxy *nsproxy;
465 nsproxy = tsk->nsproxy;
467 net = get_net(nsproxy->net_ns);
473 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
475 static __net_init int net_ns_net_init(struct net *net)
478 net->ns.ops = &netns_operations;
480 return ns_alloc_inum(&net->ns);
483 static __net_exit void net_ns_net_exit(struct net *net)
485 ns_free_inum(&net->ns);
488 static struct pernet_operations __net_initdata net_ns_ops = {
489 .init = net_ns_net_init,
490 .exit = net_ns_net_exit,
493 static struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
494 [NETNSA_NONE] = { .type = NLA_UNSPEC },
495 [NETNSA_NSID] = { .type = NLA_S32 },
496 [NETNSA_PID] = { .type = NLA_U32 },
497 [NETNSA_FD] = { .type = NLA_U32 },
500 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh)
502 struct net *net = sock_net(skb->sk);
503 struct nlattr *tb[NETNSA_MAX + 1];
507 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
511 if (!tb[NETNSA_NSID])
513 nsid = nla_get_s32(tb[NETNSA_NSID]);
516 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
517 else if (tb[NETNSA_FD])
518 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
522 return PTR_ERR(peer);
524 if (__peernet2id(net, peer, false) >= 0) {
529 err = alloc_netid(net, peer, nsid);
537 static int rtnl_net_get_size(void)
539 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
540 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
544 static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags,
545 int cmd, struct net *net, struct net *peer)
547 struct nlmsghdr *nlh;
548 struct rtgenmsg *rth;
553 nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags);
557 rth = nlmsg_data(nlh);
558 rth->rtgen_family = AF_UNSPEC;
560 id = __peernet2id(net, peer, false);
562 id = NETNSA_NSID_NOT_ASSIGNED;
563 if (nla_put_s32(skb, NETNSA_NSID, id))
564 goto nla_put_failure;
570 nlmsg_cancel(skb, nlh);
574 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh)
576 struct net *net = sock_net(skb->sk);
577 struct nlattr *tb[NETNSA_MAX + 1];
582 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
587 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
588 else if (tb[NETNSA_FD])
589 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
594 return PTR_ERR(peer);
596 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
602 err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0,
603 RTM_GETNSID, net, peer);
607 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
617 static int __init net_ns_init(void)
619 struct net_generic *ng;
622 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
626 /* Create workqueue for cleanup */
627 netns_wq = create_singlethread_workqueue("netns");
629 panic("Could not create netns workq");
632 ng = net_alloc_generic();
634 panic("Could not allocate generic netns");
636 rcu_assign_pointer(init_net.gen, ng);
638 mutex_lock(&net_mutex);
639 if (setup_net(&init_net, &init_user_ns))
640 panic("Could not setup the initial network namespace");
643 list_add_tail_rcu(&init_net.list, &net_namespace_list);
646 mutex_unlock(&net_mutex);
648 register_pernet_subsys(&net_ns_ops);
650 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL, NULL);
651 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, NULL, NULL);
656 pure_initcall(net_ns_init);
659 static int __register_pernet_operations(struct list_head *list,
660 struct pernet_operations *ops)
664 LIST_HEAD(net_exit_list);
666 list_add_tail(&ops->list, list);
667 if (ops->init || (ops->id && ops->size)) {
669 error = ops_init(ops, net);
672 list_add_tail(&net->exit_list, &net_exit_list);
678 /* If I have an error cleanup all namespaces I initialized */
679 list_del(&ops->list);
680 ops_exit_list(ops, &net_exit_list);
681 ops_free_list(ops, &net_exit_list);
685 static void __unregister_pernet_operations(struct pernet_operations *ops)
688 LIST_HEAD(net_exit_list);
690 list_del(&ops->list);
692 list_add_tail(&net->exit_list, &net_exit_list);
693 ops_exit_list(ops, &net_exit_list);
694 ops_free_list(ops, &net_exit_list);
699 static int __register_pernet_operations(struct list_head *list,
700 struct pernet_operations *ops)
702 return ops_init(ops, &init_net);
705 static void __unregister_pernet_operations(struct pernet_operations *ops)
707 LIST_HEAD(net_exit_list);
708 list_add(&init_net.exit_list, &net_exit_list);
709 ops_exit_list(ops, &net_exit_list);
710 ops_free_list(ops, &net_exit_list);
713 #endif /* CONFIG_NET_NS */
715 static DEFINE_IDA(net_generic_ids);
717 static int register_pernet_operations(struct list_head *list,
718 struct pernet_operations *ops)
724 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
726 if (error == -EAGAIN) {
727 ida_pre_get(&net_generic_ids, GFP_KERNEL);
732 max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
734 error = __register_pernet_operations(list, ops);
738 ida_remove(&net_generic_ids, *ops->id);
744 static void unregister_pernet_operations(struct pernet_operations *ops)
747 __unregister_pernet_operations(ops);
750 ida_remove(&net_generic_ids, *ops->id);
754 * register_pernet_subsys - register a network namespace subsystem
755 * @ops: pernet operations structure for the subsystem
757 * Register a subsystem which has init and exit functions
758 * that are called when network namespaces are created and
759 * destroyed respectively.
761 * When registered all network namespace init functions are
762 * called for every existing network namespace. Allowing kernel
763 * modules to have a race free view of the set of network namespaces.
765 * When a new network namespace is created all of the init
766 * methods are called in the order in which they were registered.
768 * When a network namespace is destroyed all of the exit methods
769 * are called in the reverse of the order with which they were
772 int register_pernet_subsys(struct pernet_operations *ops)
775 mutex_lock(&net_mutex);
776 error = register_pernet_operations(first_device, ops);
777 mutex_unlock(&net_mutex);
780 EXPORT_SYMBOL_GPL(register_pernet_subsys);
783 * unregister_pernet_subsys - unregister a network namespace subsystem
784 * @ops: pernet operations structure to manipulate
786 * Remove the pernet operations structure from the list to be
787 * used when network namespaces are created or destroyed. In
788 * addition run the exit method for all existing network
791 void unregister_pernet_subsys(struct pernet_operations *ops)
793 mutex_lock(&net_mutex);
794 unregister_pernet_operations(ops);
795 mutex_unlock(&net_mutex);
797 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
800 * register_pernet_device - register a network namespace device
801 * @ops: pernet operations structure for the subsystem
803 * Register a device which has init and exit functions
804 * that are called when network namespaces are created and
805 * destroyed respectively.
807 * When registered all network namespace init functions are
808 * called for every existing network namespace. Allowing kernel
809 * modules to have a race free view of the set of network namespaces.
811 * When a new network namespace is created all of the init
812 * methods are called in the order in which they were registered.
814 * When a network namespace is destroyed all of the exit methods
815 * are called in the reverse of the order with which they were
818 int register_pernet_device(struct pernet_operations *ops)
821 mutex_lock(&net_mutex);
822 error = register_pernet_operations(&pernet_list, ops);
823 if (!error && (first_device == &pernet_list))
824 first_device = &ops->list;
825 mutex_unlock(&net_mutex);
828 EXPORT_SYMBOL_GPL(register_pernet_device);
831 * unregister_pernet_device - unregister a network namespace netdevice
832 * @ops: pernet operations structure to manipulate
834 * Remove the pernet operations structure from the list to be
835 * used when network namespaces are created or destroyed. In
836 * addition run the exit method for all existing network
839 void unregister_pernet_device(struct pernet_operations *ops)
841 mutex_lock(&net_mutex);
842 if (&ops->list == first_device)
843 first_device = first_device->next;
844 unregister_pernet_operations(ops);
845 mutex_unlock(&net_mutex);
847 EXPORT_SYMBOL_GPL(unregister_pernet_device);
850 static struct ns_common *netns_get(struct task_struct *task)
852 struct net *net = NULL;
853 struct nsproxy *nsproxy;
856 nsproxy = task->nsproxy;
858 net = get_net(nsproxy->net_ns);
861 return net ? &net->ns : NULL;
864 static inline struct net *to_net_ns(struct ns_common *ns)
866 return container_of(ns, struct net, ns);
869 static void netns_put(struct ns_common *ns)
871 put_net(to_net_ns(ns));
874 static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
876 struct net *net = to_net_ns(ns);
878 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
879 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
882 put_net(nsproxy->net_ns);
883 nsproxy->net_ns = get_net(net);
887 const struct proc_ns_operations netns_operations = {
889 .type = CLONE_NEWNET,
892 .install = netns_install,