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 #ifdef NETNS_REFCNT_DEBUG
240 atomic_set(&net->use_count, 0);
243 list_for_each_entry(ops, &pernet_list, list) {
244 error = ops_init(ops, net);
252 /* Walk through the list backwards calling the exit functions
253 * for the pernet modules whose init functions did not fail.
255 list_add(&net->exit_list, &net_exit_list);
257 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
258 ops_exit_list(ops, &net_exit_list);
261 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
262 ops_free_list(ops, &net_exit_list);
270 static struct kmem_cache *net_cachep;
271 static struct workqueue_struct *netns_wq;
273 static struct net *net_alloc(void)
275 struct net *net = NULL;
276 struct net_generic *ng;
278 ng = net_alloc_generic();
282 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
286 rcu_assign_pointer(net->gen, ng);
295 static void net_free(struct net *net)
297 #ifdef NETNS_REFCNT_DEBUG
298 if (unlikely(atomic_read(&net->use_count) != 0)) {
299 pr_emerg("network namespace not free! Usage: %d\n",
300 atomic_read(&net->use_count));
304 kfree(rcu_access_pointer(net->gen));
305 kmem_cache_free(net_cachep, net);
308 void net_drop_ns(void *p)
311 if (ns && atomic_dec_and_test(&ns->passive))
315 struct net *copy_net_ns(unsigned long flags,
316 struct user_namespace *user_ns, struct net *old_net)
321 if (!(flags & CLONE_NEWNET))
322 return get_net(old_net);
326 return ERR_PTR(-ENOMEM);
328 get_user_ns(user_ns);
330 mutex_lock(&net_mutex);
331 rv = setup_net(net, user_ns);
334 list_add_tail_rcu(&net->list, &net_namespace_list);
337 mutex_unlock(&net_mutex);
339 put_user_ns(user_ns);
346 static DEFINE_SPINLOCK(cleanup_list_lock);
347 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
349 static void cleanup_net(struct work_struct *work)
351 const struct pernet_operations *ops;
352 struct net *net, *tmp, *peer;
353 struct list_head net_kill_list;
354 LIST_HEAD(net_exit_list);
356 /* Atomically snapshot the list of namespaces to cleanup */
357 spin_lock_irq(&cleanup_list_lock);
358 list_replace_init(&cleanup_list, &net_kill_list);
359 spin_unlock_irq(&cleanup_list_lock);
361 mutex_lock(&net_mutex);
363 /* Don't let anyone else find us. */
365 list_for_each_entry(net, &net_kill_list, cleanup_list) {
366 list_del_rcu(&net->list);
367 list_add_tail(&net->exit_list, &net_exit_list);
372 * Another CPU might be rcu-iterating the list, wait for it.
373 * This needs to be before calling the exit() notifiers, so
374 * the rcu_barrier() below isn't sufficient alone.
378 /* Run all of the network namespace exit methods */
379 list_for_each_entry_reverse(ops, &pernet_list, list)
380 ops_exit_list(ops, &net_exit_list);
382 /* Free the net generic variables */
383 list_for_each_entry_reverse(ops, &pernet_list, list)
384 ops_free_list(ops, &net_exit_list);
386 mutex_unlock(&net_mutex);
388 /* Ensure there are no outstanding rcu callbacks using this
394 /* Finally it is safe to free my network namespace structure */
395 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
396 /* Unreference net from all peers (no need to loop over
397 * net_exit_list because idr_destroy() will be called for each
398 * element of this list.
401 int id = __peernet2id(peer, net, false);
404 idr_remove(&peer->netns_ids, id);
406 idr_destroy(&net->netns_ids);
408 list_del_init(&net->exit_list);
409 put_user_ns(net->user_ns);
414 static DECLARE_WORK(net_cleanup_work, cleanup_net);
416 void __put_net(struct net *net)
418 /* Cleanup the network namespace in process context */
421 spin_lock_irqsave(&cleanup_list_lock, flags);
422 list_add(&net->cleanup_list, &cleanup_list);
423 spin_unlock_irqrestore(&cleanup_list_lock, flags);
425 queue_work(netns_wq, &net_cleanup_work);
427 EXPORT_SYMBOL_GPL(__put_net);
429 struct net *get_net_ns_by_fd(int fd)
432 struct ns_common *ns;
435 file = proc_ns_fget(fd);
437 return ERR_CAST(file);
439 ns = get_proc_ns(file_inode(file));
440 if (ns->ops == &netns_operations)
441 net = get_net(container_of(ns, struct net, ns));
443 net = ERR_PTR(-EINVAL);
450 struct net *get_net_ns_by_fd(int fd)
452 return ERR_PTR(-EINVAL);
455 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
457 struct net *get_net_ns_by_pid(pid_t pid)
459 struct task_struct *tsk;
462 /* Lookup the network namespace */
463 net = ERR_PTR(-ESRCH);
465 tsk = find_task_by_vpid(pid);
467 struct nsproxy *nsproxy;
469 nsproxy = tsk->nsproxy;
471 net = get_net(nsproxy->net_ns);
477 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
479 static __net_init int net_ns_net_init(struct net *net)
482 net->ns.ops = &netns_operations;
484 return ns_alloc_inum(&net->ns);
487 static __net_exit void net_ns_net_exit(struct net *net)
489 ns_free_inum(&net->ns);
492 static struct pernet_operations __net_initdata net_ns_ops = {
493 .init = net_ns_net_init,
494 .exit = net_ns_net_exit,
497 static struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
498 [NETNSA_NONE] = { .type = NLA_UNSPEC },
499 [NETNSA_NSID] = { .type = NLA_S32 },
500 [NETNSA_PID] = { .type = NLA_U32 },
501 [NETNSA_FD] = { .type = NLA_U32 },
504 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh)
506 struct net *net = sock_net(skb->sk);
507 struct nlattr *tb[NETNSA_MAX + 1];
511 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
515 if (!tb[NETNSA_NSID])
517 nsid = nla_get_s32(tb[NETNSA_NSID]);
520 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
521 else if (tb[NETNSA_FD])
522 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
526 return PTR_ERR(peer);
528 if (__peernet2id(net, peer, false) >= 0) {
533 err = alloc_netid(net, peer, nsid);
541 static int rtnl_net_get_size(void)
543 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
544 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
548 static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags,
549 int cmd, struct net *net, struct net *peer)
551 struct nlmsghdr *nlh;
552 struct rtgenmsg *rth;
557 nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags);
561 rth = nlmsg_data(nlh);
562 rth->rtgen_family = AF_UNSPEC;
564 id = __peernet2id(net, peer, false);
566 id = NETNSA_NSID_NOT_ASSIGNED;
567 if (nla_put_s32(skb, NETNSA_NSID, id))
568 goto nla_put_failure;
574 nlmsg_cancel(skb, nlh);
578 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh)
580 struct net *net = sock_net(skb->sk);
581 struct nlattr *tb[NETNSA_MAX + 1];
586 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
591 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
592 else if (tb[NETNSA_FD])
593 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
598 return PTR_ERR(peer);
600 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
606 err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0,
607 RTM_GETNSID, net, peer);
611 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
621 static int __init net_ns_init(void)
623 struct net_generic *ng;
626 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
630 /* Create workqueue for cleanup */
631 netns_wq = create_singlethread_workqueue("netns");
633 panic("Could not create netns workq");
636 ng = net_alloc_generic();
638 panic("Could not allocate generic netns");
640 rcu_assign_pointer(init_net.gen, ng);
642 mutex_lock(&net_mutex);
643 if (setup_net(&init_net, &init_user_ns))
644 panic("Could not setup the initial network namespace");
647 list_add_tail_rcu(&init_net.list, &net_namespace_list);
650 mutex_unlock(&net_mutex);
652 register_pernet_subsys(&net_ns_ops);
654 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL, NULL);
655 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, NULL, NULL);
660 pure_initcall(net_ns_init);
663 static int __register_pernet_operations(struct list_head *list,
664 struct pernet_operations *ops)
668 LIST_HEAD(net_exit_list);
670 list_add_tail(&ops->list, list);
671 if (ops->init || (ops->id && ops->size)) {
673 error = ops_init(ops, net);
676 list_add_tail(&net->exit_list, &net_exit_list);
682 /* If I have an error cleanup all namespaces I initialized */
683 list_del(&ops->list);
684 ops_exit_list(ops, &net_exit_list);
685 ops_free_list(ops, &net_exit_list);
689 static void __unregister_pernet_operations(struct pernet_operations *ops)
692 LIST_HEAD(net_exit_list);
694 list_del(&ops->list);
696 list_add_tail(&net->exit_list, &net_exit_list);
697 ops_exit_list(ops, &net_exit_list);
698 ops_free_list(ops, &net_exit_list);
703 static int __register_pernet_operations(struct list_head *list,
704 struct pernet_operations *ops)
706 return ops_init(ops, &init_net);
709 static void __unregister_pernet_operations(struct pernet_operations *ops)
711 LIST_HEAD(net_exit_list);
712 list_add(&init_net.exit_list, &net_exit_list);
713 ops_exit_list(ops, &net_exit_list);
714 ops_free_list(ops, &net_exit_list);
717 #endif /* CONFIG_NET_NS */
719 static DEFINE_IDA(net_generic_ids);
721 static int register_pernet_operations(struct list_head *list,
722 struct pernet_operations *ops)
728 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
730 if (error == -EAGAIN) {
731 ida_pre_get(&net_generic_ids, GFP_KERNEL);
736 max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
738 error = __register_pernet_operations(list, ops);
742 ida_remove(&net_generic_ids, *ops->id);
748 static void unregister_pernet_operations(struct pernet_operations *ops)
751 __unregister_pernet_operations(ops);
754 ida_remove(&net_generic_ids, *ops->id);
758 * register_pernet_subsys - register a network namespace subsystem
759 * @ops: pernet operations structure for the subsystem
761 * Register a subsystem which has init and exit functions
762 * that are called when network namespaces are created and
763 * destroyed respectively.
765 * When registered all network namespace init functions are
766 * called for every existing network namespace. Allowing kernel
767 * modules to have a race free view of the set of network namespaces.
769 * When a new network namespace is created all of the init
770 * methods are called in the order in which they were registered.
772 * When a network namespace is destroyed all of the exit methods
773 * are called in the reverse of the order with which they were
776 int register_pernet_subsys(struct pernet_operations *ops)
779 mutex_lock(&net_mutex);
780 error = register_pernet_operations(first_device, ops);
781 mutex_unlock(&net_mutex);
784 EXPORT_SYMBOL_GPL(register_pernet_subsys);
787 * unregister_pernet_subsys - unregister a network namespace subsystem
788 * @ops: pernet operations structure to manipulate
790 * Remove the pernet operations structure from the list to be
791 * used when network namespaces are created or destroyed. In
792 * addition run the exit method for all existing network
795 void unregister_pernet_subsys(struct pernet_operations *ops)
797 mutex_lock(&net_mutex);
798 unregister_pernet_operations(ops);
799 mutex_unlock(&net_mutex);
801 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
804 * register_pernet_device - register a network namespace device
805 * @ops: pernet operations structure for the subsystem
807 * Register a device which has init and exit functions
808 * that are called when network namespaces are created and
809 * destroyed respectively.
811 * When registered all network namespace init functions are
812 * called for every existing network namespace. Allowing kernel
813 * modules to have a race free view of the set of network namespaces.
815 * When a new network namespace is created all of the init
816 * methods are called in the order in which they were registered.
818 * When a network namespace is destroyed all of the exit methods
819 * are called in the reverse of the order with which they were
822 int register_pernet_device(struct pernet_operations *ops)
825 mutex_lock(&net_mutex);
826 error = register_pernet_operations(&pernet_list, ops);
827 if (!error && (first_device == &pernet_list))
828 first_device = &ops->list;
829 mutex_unlock(&net_mutex);
832 EXPORT_SYMBOL_GPL(register_pernet_device);
835 * unregister_pernet_device - unregister a network namespace netdevice
836 * @ops: pernet operations structure to manipulate
838 * Remove the pernet operations structure from the list to be
839 * used when network namespaces are created or destroyed. In
840 * addition run the exit method for all existing network
843 void unregister_pernet_device(struct pernet_operations *ops)
845 mutex_lock(&net_mutex);
846 if (&ops->list == first_device)
847 first_device = first_device->next;
848 unregister_pernet_operations(ops);
849 mutex_unlock(&net_mutex);
851 EXPORT_SYMBOL_GPL(unregister_pernet_device);
854 static struct ns_common *netns_get(struct task_struct *task)
856 struct net *net = NULL;
857 struct nsproxy *nsproxy;
860 nsproxy = task->nsproxy;
862 net = get_net(nsproxy->net_ns);
865 return net ? &net->ns : NULL;
868 static inline struct net *to_net_ns(struct ns_common *ns)
870 return container_of(ns, struct net, ns);
873 static void netns_put(struct ns_common *ns)
875 put_net(to_net_ns(ns));
878 static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
880 struct net *net = to_net_ns(ns);
882 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
883 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
886 put_net(nsproxy->net_ns);
887 nsproxy->net_ns = get_net(net);
891 const struct proc_ns_operations netns_operations = {
893 .type = CLONE_NEWNET,
896 .install = netns_install,