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>
20 #include <net/netlink.h>
21 #include <net/net_namespace.h>
22 #include <net/netns/generic.h>
25 * Our network namespace constructor/destructor lists
28 static LIST_HEAD(pernet_list);
29 static struct list_head *first_device = &pernet_list;
30 DEFINE_MUTEX(net_mutex);
32 LIST_HEAD(net_namespace_list);
33 EXPORT_SYMBOL_GPL(net_namespace_list);
35 struct net init_net = {
36 .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head),
38 EXPORT_SYMBOL(init_net);
40 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
42 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
44 static struct net_generic *net_alloc_generic(void)
46 struct net_generic *ng;
47 size_t generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
49 ng = kzalloc(generic_size, GFP_KERNEL);
51 ng->len = max_gen_ptrs;
56 static int net_assign_generic(struct net *net, int id, void *data)
58 struct net_generic *ng, *old_ng;
60 BUG_ON(!mutex_is_locked(&net_mutex));
63 old_ng = rcu_dereference_protected(net->gen,
64 lockdep_is_held(&net_mutex));
66 if (old_ng->len >= id)
69 ng = net_alloc_generic();
74 * Some synchronisation notes:
76 * The net_generic explores the net->gen array inside rcu
77 * read section. Besides once set the net->gen->ptr[x]
78 * pointer never changes (see rules in netns/generic.h).
80 * That said, we simply duplicate this array and schedule
81 * the old copy for kfree after a grace period.
84 memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
86 rcu_assign_pointer(net->gen, ng);
87 kfree_rcu(old_ng, rcu);
89 ng->ptr[id - 1] = data;
93 static int ops_init(const struct pernet_operations *ops, struct net *net)
98 if (ops->id && ops->size) {
99 data = kzalloc(ops->size, GFP_KERNEL);
103 err = net_assign_generic(net, *ops->id, data);
109 err = ops->init(net);
120 static void ops_free(const struct pernet_operations *ops, struct net *net)
122 if (ops->id && ops->size) {
124 kfree(net_generic(net, id));
128 static void ops_exit_list(const struct pernet_operations *ops,
129 struct list_head *net_exit_list)
133 list_for_each_entry(net, net_exit_list, exit_list)
137 ops->exit_batch(net_exit_list);
140 static void ops_free_list(const struct pernet_operations *ops,
141 struct list_head *net_exit_list)
144 if (ops->size && ops->id) {
145 list_for_each_entry(net, net_exit_list, exit_list)
150 /* should be called with nsid_lock held */
151 static int alloc_netid(struct net *net, struct net *peer, int reqid)
153 int min = 0, max = 0;
160 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
163 /* This function is used by idr_for_each(). If net is equal to peer, the
164 * function returns the id so that idr_for_each() stops. Because we cannot
165 * returns the id 0 (idr_for_each() will not stop), we return the magic value
166 * NET_ID_ZERO (-1) for it.
168 #define NET_ID_ZERO -1
169 static int net_eq_idr(int id, void *net, void *peer)
171 if (net_eq(net, peer))
172 return id ? : NET_ID_ZERO;
176 /* Should be called with nsid_lock held. If a new id is assigned, the bool alloc
177 * is set to true, thus the caller knows that the new id must be notified via
180 static int __peernet2id_alloc(struct net *net, struct net *peer, bool *alloc)
182 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
183 bool alloc_it = *alloc;
187 /* Magic value for id 0. */
188 if (id == NET_ID_ZERO)
194 id = alloc_netid(net, peer, -1);
196 return id >= 0 ? id : NETNSA_NSID_NOT_ASSIGNED;
199 return NETNSA_NSID_NOT_ASSIGNED;
202 /* should be called with nsid_lock held */
203 static int __peernet2id(struct net *net, struct net *peer)
207 return __peernet2id_alloc(net, peer, &no);
210 static void rtnl_net_notifyid(struct net *net, int cmd, int id);
211 /* This function returns the id of a peer netns. If no id is assigned, one will
212 * be allocated and returned.
214 int peernet2id_alloc(struct net *net, struct net *peer)
220 spin_lock_irqsave(&net->nsid_lock, flags);
221 alloc = atomic_read(&peer->count) == 0 ? false : true;
222 id = __peernet2id_alloc(net, peer, &alloc);
223 spin_unlock_irqrestore(&net->nsid_lock, flags);
224 if (alloc && id >= 0)
225 rtnl_net_notifyid(net, RTM_NEWNSID, id);
228 EXPORT_SYMBOL(peernet2id_alloc);
230 /* This function returns, if assigned, the id of a peer netns. */
231 int peernet2id(struct net *net, struct net *peer)
236 spin_lock_irqsave(&net->nsid_lock, flags);
237 id = __peernet2id(net, peer);
238 spin_unlock_irqrestore(&net->nsid_lock, flags);
242 /* This function returns true is the peer netns has an id assigned into the
245 bool peernet_has_id(struct net *net, struct net *peer)
247 return peernet2id(net, peer) >= 0;
250 struct net *get_net_ns_by_id(struct net *net, int id)
259 spin_lock_irqsave(&net->nsid_lock, flags);
260 peer = idr_find(&net->netns_ids, id);
263 spin_unlock_irqrestore(&net->nsid_lock, flags);
270 * setup_net runs the initializers for the network namespace object.
272 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
274 /* Must be called with net_mutex held */
275 const struct pernet_operations *ops, *saved_ops;
277 LIST_HEAD(net_exit_list);
279 atomic_set(&net->count, 1);
280 atomic_set(&net->passive, 1);
281 net->dev_base_seq = 1;
282 net->user_ns = user_ns;
283 idr_init(&net->netns_ids);
284 spin_lock_init(&net->nsid_lock);
286 list_for_each_entry(ops, &pernet_list, list) {
287 error = ops_init(ops, net);
295 /* Walk through the list backwards calling the exit functions
296 * for the pernet modules whose init functions did not fail.
298 list_add(&net->exit_list, &net_exit_list);
300 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
301 ops_exit_list(ops, &net_exit_list);
304 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
305 ops_free_list(ops, &net_exit_list);
313 static struct kmem_cache *net_cachep;
314 static struct workqueue_struct *netns_wq;
316 static struct net *net_alloc(void)
318 struct net *net = NULL;
319 struct net_generic *ng;
321 ng = net_alloc_generic();
325 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
329 rcu_assign_pointer(net->gen, ng);
338 static void net_free(struct net *net)
340 kfree(rcu_access_pointer(net->gen));
341 kmem_cache_free(net_cachep, net);
344 void net_drop_ns(void *p)
347 if (ns && atomic_dec_and_test(&ns->passive))
351 struct net *copy_net_ns(unsigned long flags,
352 struct user_namespace *user_ns, struct net *old_net)
357 if (!(flags & CLONE_NEWNET))
358 return get_net(old_net);
362 return ERR_PTR(-ENOMEM);
364 get_user_ns(user_ns);
366 mutex_lock(&net_mutex);
367 rv = setup_net(net, user_ns);
370 list_add_tail_rcu(&net->list, &net_namespace_list);
373 mutex_unlock(&net_mutex);
375 put_user_ns(user_ns);
382 static DEFINE_SPINLOCK(cleanup_list_lock);
383 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
385 static void cleanup_net(struct work_struct *work)
387 const struct pernet_operations *ops;
388 struct net *net, *tmp;
389 struct list_head net_kill_list;
390 LIST_HEAD(net_exit_list);
392 /* Atomically snapshot the list of namespaces to cleanup */
393 spin_lock_irq(&cleanup_list_lock);
394 list_replace_init(&cleanup_list, &net_kill_list);
395 spin_unlock_irq(&cleanup_list_lock);
397 mutex_lock(&net_mutex);
399 /* Don't let anyone else find us. */
401 list_for_each_entry(net, &net_kill_list, cleanup_list) {
402 list_del_rcu(&net->list);
403 list_add_tail(&net->exit_list, &net_exit_list);
407 spin_lock_irq(&tmp->nsid_lock);
408 id = __peernet2id(tmp, net);
410 idr_remove(&tmp->netns_ids, id);
411 spin_unlock_irq(&tmp->nsid_lock);
413 rtnl_net_notifyid(tmp, RTM_DELNSID, id);
415 spin_lock_irq(&net->nsid_lock);
416 idr_destroy(&net->netns_ids);
417 spin_unlock_irq(&net->nsid_lock);
423 * Another CPU might be rcu-iterating the list, wait for it.
424 * This needs to be before calling the exit() notifiers, so
425 * the rcu_barrier() below isn't sufficient alone.
429 /* Run all of the network namespace exit methods */
430 list_for_each_entry_reverse(ops, &pernet_list, list)
431 ops_exit_list(ops, &net_exit_list);
433 /* Free the net generic variables */
434 list_for_each_entry_reverse(ops, &pernet_list, list)
435 ops_free_list(ops, &net_exit_list);
437 mutex_unlock(&net_mutex);
439 /* Ensure there are no outstanding rcu callbacks using this
444 /* Finally it is safe to free my network namespace structure */
445 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
446 list_del_init(&net->exit_list);
447 put_user_ns(net->user_ns);
451 static DECLARE_WORK(net_cleanup_work, cleanup_net);
453 void __put_net(struct net *net)
455 /* Cleanup the network namespace in process context */
458 spin_lock_irqsave(&cleanup_list_lock, flags);
459 list_add(&net->cleanup_list, &cleanup_list);
460 spin_unlock_irqrestore(&cleanup_list_lock, flags);
462 queue_work(netns_wq, &net_cleanup_work);
464 EXPORT_SYMBOL_GPL(__put_net);
466 struct net *get_net_ns_by_fd(int fd)
469 struct ns_common *ns;
472 file = proc_ns_fget(fd);
474 return ERR_CAST(file);
476 ns = get_proc_ns(file_inode(file));
477 if (ns->ops == &netns_operations)
478 net = get_net(container_of(ns, struct net, ns));
480 net = ERR_PTR(-EINVAL);
487 struct net *get_net_ns_by_fd(int fd)
489 return ERR_PTR(-EINVAL);
492 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
494 struct net *get_net_ns_by_pid(pid_t pid)
496 struct task_struct *tsk;
499 /* Lookup the network namespace */
500 net = ERR_PTR(-ESRCH);
502 tsk = find_task_by_vpid(pid);
504 struct nsproxy *nsproxy;
506 nsproxy = tsk->nsproxy;
508 net = get_net(nsproxy->net_ns);
514 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
516 static __net_init int net_ns_net_init(struct net *net)
519 net->ns.ops = &netns_operations;
521 return ns_alloc_inum(&net->ns);
524 static __net_exit void net_ns_net_exit(struct net *net)
526 ns_free_inum(&net->ns);
529 static struct pernet_operations __net_initdata net_ns_ops = {
530 .init = net_ns_net_init,
531 .exit = net_ns_net_exit,
534 static struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
535 [NETNSA_NONE] = { .type = NLA_UNSPEC },
536 [NETNSA_NSID] = { .type = NLA_S32 },
537 [NETNSA_PID] = { .type = NLA_U32 },
538 [NETNSA_FD] = { .type = NLA_U32 },
541 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh)
543 struct net *net = sock_net(skb->sk);
544 struct nlattr *tb[NETNSA_MAX + 1];
549 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
553 if (!tb[NETNSA_NSID])
555 nsid = nla_get_s32(tb[NETNSA_NSID]);
558 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
559 else if (tb[NETNSA_FD])
560 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
564 return PTR_ERR(peer);
566 spin_lock_irqsave(&net->nsid_lock, flags);
567 if (__peernet2id(net, peer) >= 0) {
568 spin_unlock_irqrestore(&net->nsid_lock, flags);
573 err = alloc_netid(net, peer, nsid);
574 spin_unlock_irqrestore(&net->nsid_lock, flags);
576 rtnl_net_notifyid(net, RTM_NEWNSID, err);
584 static int rtnl_net_get_size(void)
586 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
587 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
591 static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags,
592 int cmd, struct net *net, int nsid)
594 struct nlmsghdr *nlh;
595 struct rtgenmsg *rth;
597 nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags);
601 rth = nlmsg_data(nlh);
602 rth->rtgen_family = AF_UNSPEC;
604 if (nla_put_s32(skb, NETNSA_NSID, nsid))
605 goto nla_put_failure;
611 nlmsg_cancel(skb, nlh);
615 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh)
617 struct net *net = sock_net(skb->sk);
618 struct nlattr *tb[NETNSA_MAX + 1];
623 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
628 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
629 else if (tb[NETNSA_FD])
630 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
635 return PTR_ERR(peer);
637 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
643 id = peernet2id(net, peer);
644 err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0,
645 RTM_NEWNSID, net, id);
649 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
659 struct rtnl_net_dump_cb {
662 struct netlink_callback *cb;
667 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
669 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
672 if (net_cb->idx < net_cb->s_idx)
675 ret = rtnl_net_fill(net_cb->skb, NETLINK_CB(net_cb->cb->skb).portid,
676 net_cb->cb->nlh->nlmsg_seq, NLM_F_MULTI,
677 RTM_NEWNSID, net_cb->net, id);
686 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
688 struct net *net = sock_net(skb->sk);
689 struct rtnl_net_dump_cb net_cb = {
694 .s_idx = cb->args[0],
698 spin_lock_irqsave(&net->nsid_lock, flags);
699 idr_for_each(&net->netns_ids, rtnl_net_dumpid_one, &net_cb);
700 spin_unlock_irqrestore(&net->nsid_lock, flags);
702 cb->args[0] = net_cb.idx;
706 static void rtnl_net_notifyid(struct net *net, int cmd, int id)
711 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
715 err = rtnl_net_fill(msg, 0, 0, 0, cmd, net, id);
719 rtnl_notify(msg, net, 0, RTNLGRP_NSID, NULL, 0);
725 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
728 static int __init net_ns_init(void)
730 struct net_generic *ng;
733 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
737 /* Create workqueue for cleanup */
738 netns_wq = create_singlethread_workqueue("netns");
740 panic("Could not create netns workq");
743 ng = net_alloc_generic();
745 panic("Could not allocate generic netns");
747 rcu_assign_pointer(init_net.gen, ng);
749 mutex_lock(&net_mutex);
750 if (setup_net(&init_net, &init_user_ns))
751 panic("Could not setup the initial network namespace");
754 list_add_tail_rcu(&init_net.list, &net_namespace_list);
757 mutex_unlock(&net_mutex);
759 register_pernet_subsys(&net_ns_ops);
761 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL, NULL);
762 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
768 pure_initcall(net_ns_init);
771 static int __register_pernet_operations(struct list_head *list,
772 struct pernet_operations *ops)
776 LIST_HEAD(net_exit_list);
778 list_add_tail(&ops->list, list);
779 if (ops->init || (ops->id && ops->size)) {
781 error = ops_init(ops, net);
784 list_add_tail(&net->exit_list, &net_exit_list);
790 /* If I have an error cleanup all namespaces I initialized */
791 list_del(&ops->list);
792 ops_exit_list(ops, &net_exit_list);
793 ops_free_list(ops, &net_exit_list);
797 static void __unregister_pernet_operations(struct pernet_operations *ops)
800 LIST_HEAD(net_exit_list);
802 list_del(&ops->list);
804 list_add_tail(&net->exit_list, &net_exit_list);
805 ops_exit_list(ops, &net_exit_list);
806 ops_free_list(ops, &net_exit_list);
811 static int __register_pernet_operations(struct list_head *list,
812 struct pernet_operations *ops)
814 return ops_init(ops, &init_net);
817 static void __unregister_pernet_operations(struct pernet_operations *ops)
819 LIST_HEAD(net_exit_list);
820 list_add(&init_net.exit_list, &net_exit_list);
821 ops_exit_list(ops, &net_exit_list);
822 ops_free_list(ops, &net_exit_list);
825 #endif /* CONFIG_NET_NS */
827 static DEFINE_IDA(net_generic_ids);
829 static int register_pernet_operations(struct list_head *list,
830 struct pernet_operations *ops)
836 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
838 if (error == -EAGAIN) {
839 ida_pre_get(&net_generic_ids, GFP_KERNEL);
844 max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
846 error = __register_pernet_operations(list, ops);
850 ida_remove(&net_generic_ids, *ops->id);
856 static void unregister_pernet_operations(struct pernet_operations *ops)
859 __unregister_pernet_operations(ops);
862 ida_remove(&net_generic_ids, *ops->id);
866 * register_pernet_subsys - register a network namespace subsystem
867 * @ops: pernet operations structure for the subsystem
869 * Register a subsystem which has init and exit functions
870 * that are called when network namespaces are created and
871 * destroyed respectively.
873 * When registered all network namespace init functions are
874 * called for every existing network namespace. Allowing kernel
875 * modules to have a race free view of the set of network namespaces.
877 * When a new network namespace is created all of the init
878 * methods are called in the order in which they were registered.
880 * When a network namespace is destroyed all of the exit methods
881 * are called in the reverse of the order with which they were
884 int register_pernet_subsys(struct pernet_operations *ops)
887 mutex_lock(&net_mutex);
888 error = register_pernet_operations(first_device, ops);
889 mutex_unlock(&net_mutex);
892 EXPORT_SYMBOL_GPL(register_pernet_subsys);
895 * unregister_pernet_subsys - unregister a network namespace subsystem
896 * @ops: pernet operations structure to manipulate
898 * Remove the pernet operations structure from the list to be
899 * used when network namespaces are created or destroyed. In
900 * addition run the exit method for all existing network
903 void unregister_pernet_subsys(struct pernet_operations *ops)
905 mutex_lock(&net_mutex);
906 unregister_pernet_operations(ops);
907 mutex_unlock(&net_mutex);
909 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
912 * register_pernet_device - register a network namespace device
913 * @ops: pernet operations structure for the subsystem
915 * Register a device which has init and exit functions
916 * that are called when network namespaces are created and
917 * destroyed respectively.
919 * When registered all network namespace init functions are
920 * called for every existing network namespace. Allowing kernel
921 * modules to have a race free view of the set of network namespaces.
923 * When a new network namespace is created all of the init
924 * methods are called in the order in which they were registered.
926 * When a network namespace is destroyed all of the exit methods
927 * are called in the reverse of the order with which they were
930 int register_pernet_device(struct pernet_operations *ops)
933 mutex_lock(&net_mutex);
934 error = register_pernet_operations(&pernet_list, ops);
935 if (!error && (first_device == &pernet_list))
936 first_device = &ops->list;
937 mutex_unlock(&net_mutex);
940 EXPORT_SYMBOL_GPL(register_pernet_device);
943 * unregister_pernet_device - unregister a network namespace netdevice
944 * @ops: pernet operations structure to manipulate
946 * Remove the pernet operations structure from the list to be
947 * used when network namespaces are created or destroyed. In
948 * addition run the exit method for all existing network
951 void unregister_pernet_device(struct pernet_operations *ops)
953 mutex_lock(&net_mutex);
954 if (&ops->list == first_device)
955 first_device = first_device->next;
956 unregister_pernet_operations(ops);
957 mutex_unlock(&net_mutex);
959 EXPORT_SYMBOL_GPL(unregister_pernet_device);
962 static struct ns_common *netns_get(struct task_struct *task)
964 struct net *net = NULL;
965 struct nsproxy *nsproxy;
968 nsproxy = task->nsproxy;
970 net = get_net(nsproxy->net_ns);
973 return net ? &net->ns : NULL;
976 static inline struct net *to_net_ns(struct ns_common *ns)
978 return container_of(ns, struct net, ns);
981 static void netns_put(struct ns_common *ns)
983 put_net(to_net_ns(ns));
986 static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
988 struct net *net = to_net_ns(ns);
990 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
991 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
994 put_net(nsproxy->net_ns);
995 nsproxy->net_ns = get_net(net);
999 const struct proc_ns_operations netns_operations = {
1001 .type = CLONE_NEWNET,
1004 .install = netns_install,