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 static bool init_net_initialized;
42 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
44 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
46 static struct net_generic *net_alloc_generic(void)
48 struct net_generic *ng;
49 size_t generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
51 ng = kzalloc(generic_size, GFP_KERNEL);
53 ng->len = max_gen_ptrs;
58 static int net_assign_generic(struct net *net, int id, void *data)
60 struct net_generic *ng, *old_ng;
62 BUG_ON(!mutex_is_locked(&net_mutex));
65 old_ng = rcu_dereference_protected(net->gen,
66 lockdep_is_held(&net_mutex));
68 if (old_ng->len >= id)
71 ng = net_alloc_generic();
76 * Some synchronisation notes:
78 * The net_generic explores the net->gen array inside rcu
79 * read section. Besides once set the net->gen->ptr[x]
80 * pointer never changes (see rules in netns/generic.h).
82 * That said, we simply duplicate this array and schedule
83 * the old copy for kfree after a grace period.
86 memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
88 rcu_assign_pointer(net->gen, ng);
89 kfree_rcu(old_ng, rcu);
91 ng->ptr[id - 1] = data;
95 static int ops_init(const struct pernet_operations *ops, struct net *net)
100 if (ops->id && ops->size) {
101 data = kzalloc(ops->size, GFP_KERNEL);
105 err = net_assign_generic(net, *ops->id, data);
111 err = ops->init(net);
122 static void ops_free(const struct pernet_operations *ops, struct net *net)
124 if (ops->id && ops->size) {
126 kfree(net_generic(net, id));
130 static void ops_exit_list(const struct pernet_operations *ops,
131 struct list_head *net_exit_list)
135 list_for_each_entry(net, net_exit_list, exit_list)
139 ops->exit_batch(net_exit_list);
142 static void ops_free_list(const struct pernet_operations *ops,
143 struct list_head *net_exit_list)
146 if (ops->size && ops->id) {
147 list_for_each_entry(net, net_exit_list, exit_list)
152 /* should be called with nsid_lock held */
153 static int alloc_netid(struct net *net, struct net *peer, int reqid)
155 int min = 0, max = 0;
162 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
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 /* Should be called with nsid_lock held. If a new id is assigned, the bool alloc
179 * is set to true, thus the caller knows that the new id must be notified via
182 static int __peernet2id_alloc(struct net *net, struct net *peer, bool *alloc)
184 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
185 bool alloc_it = *alloc;
189 /* Magic value for id 0. */
190 if (id == NET_ID_ZERO)
196 id = alloc_netid(net, peer, -1);
198 return id >= 0 ? id : NETNSA_NSID_NOT_ASSIGNED;
201 return NETNSA_NSID_NOT_ASSIGNED;
204 /* should be called with nsid_lock held */
205 static int __peernet2id(struct net *net, struct net *peer)
209 return __peernet2id_alloc(net, peer, &no);
212 static void rtnl_net_notifyid(struct net *net, int cmd, int id);
213 /* This function returns the id of a peer netns. If no id is assigned, one will
214 * be allocated and returned.
216 int peernet2id_alloc(struct net *net, struct net *peer)
222 spin_lock_irqsave(&net->nsid_lock, flags);
223 alloc = atomic_read(&peer->count) == 0 ? false : true;
224 id = __peernet2id_alloc(net, peer, &alloc);
225 spin_unlock_irqrestore(&net->nsid_lock, flags);
226 if (alloc && id >= 0)
227 rtnl_net_notifyid(net, RTM_NEWNSID, id);
230 EXPORT_SYMBOL(peernet2id_alloc);
232 /* This function returns, if assigned, the id of a peer netns. */
233 int peernet2id(struct net *net, struct net *peer)
238 spin_lock_irqsave(&net->nsid_lock, flags);
239 id = __peernet2id(net, peer);
240 spin_unlock_irqrestore(&net->nsid_lock, flags);
244 /* This function returns true is the peer netns has an id assigned into the
247 bool peernet_has_id(struct net *net, struct net *peer)
249 return peernet2id(net, peer) >= 0;
252 struct net *get_net_ns_by_id(struct net *net, int id)
261 spin_lock_irqsave(&net->nsid_lock, flags);
262 peer = idr_find(&net->netns_ids, id);
265 spin_unlock_irqrestore(&net->nsid_lock, flags);
272 * setup_net runs the initializers for the network namespace object.
274 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
276 /* Must be called with net_mutex held */
277 const struct pernet_operations *ops, *saved_ops;
279 LIST_HEAD(net_exit_list);
281 atomic_set(&net->count, 1);
282 atomic_set(&net->passive, 1);
283 net->dev_base_seq = 1;
284 net->user_ns = user_ns;
285 idr_init(&net->netns_ids);
286 spin_lock_init(&net->nsid_lock);
288 list_for_each_entry(ops, &pernet_list, list) {
289 error = ops_init(ops, net);
297 /* Walk through the list backwards calling the exit functions
298 * for the pernet modules whose init functions did not fail.
300 list_add(&net->exit_list, &net_exit_list);
302 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
303 ops_exit_list(ops, &net_exit_list);
306 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
307 ops_free_list(ops, &net_exit_list);
315 static struct kmem_cache *net_cachep;
316 static struct workqueue_struct *netns_wq;
318 static struct net *net_alloc(void)
320 struct net *net = NULL;
321 struct net_generic *ng;
323 ng = net_alloc_generic();
327 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
331 rcu_assign_pointer(net->gen, ng);
340 static void net_free(struct net *net)
342 kfree(rcu_access_pointer(net->gen));
343 kmem_cache_free(net_cachep, net);
346 void net_drop_ns(void *p)
349 if (ns && atomic_dec_and_test(&ns->passive))
353 struct net *copy_net_ns(unsigned long flags,
354 struct user_namespace *user_ns, struct net *old_net)
359 if (!(flags & CLONE_NEWNET))
360 return get_net(old_net);
364 return ERR_PTR(-ENOMEM);
366 get_user_ns(user_ns);
368 mutex_lock(&net_mutex);
369 rv = setup_net(net, user_ns);
372 list_add_tail_rcu(&net->list, &net_namespace_list);
375 mutex_unlock(&net_mutex);
377 put_user_ns(user_ns);
384 static DEFINE_SPINLOCK(cleanup_list_lock);
385 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
387 static void cleanup_net(struct work_struct *work)
389 const struct pernet_operations *ops;
390 struct net *net, *tmp;
391 struct list_head net_kill_list;
392 LIST_HEAD(net_exit_list);
394 /* Atomically snapshot the list of namespaces to cleanup */
395 spin_lock_irq(&cleanup_list_lock);
396 list_replace_init(&cleanup_list, &net_kill_list);
397 spin_unlock_irq(&cleanup_list_lock);
399 mutex_lock(&net_mutex);
401 /* Don't let anyone else find us. */
403 list_for_each_entry(net, &net_kill_list, cleanup_list) {
404 list_del_rcu(&net->list);
405 list_add_tail(&net->exit_list, &net_exit_list);
409 spin_lock_irq(&tmp->nsid_lock);
410 id = __peernet2id(tmp, net);
412 idr_remove(&tmp->netns_ids, id);
413 spin_unlock_irq(&tmp->nsid_lock);
415 rtnl_net_notifyid(tmp, RTM_DELNSID, id);
417 spin_lock_irq(&net->nsid_lock);
418 idr_destroy(&net->netns_ids);
419 spin_unlock_irq(&net->nsid_lock);
425 * Another CPU might be rcu-iterating the list, wait for it.
426 * This needs to be before calling the exit() notifiers, so
427 * the rcu_barrier() below isn't sufficient alone.
431 /* Run all of the network namespace exit methods */
432 list_for_each_entry_reverse(ops, &pernet_list, list)
433 ops_exit_list(ops, &net_exit_list);
435 /* Free the net generic variables */
436 list_for_each_entry_reverse(ops, &pernet_list, list)
437 ops_free_list(ops, &net_exit_list);
439 mutex_unlock(&net_mutex);
441 /* Ensure there are no outstanding rcu callbacks using this
446 /* Finally it is safe to free my network namespace structure */
447 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
448 list_del_init(&net->exit_list);
449 put_user_ns(net->user_ns);
453 static DECLARE_WORK(net_cleanup_work, cleanup_net);
455 void __put_net(struct net *net)
457 /* Cleanup the network namespace in process context */
460 spin_lock_irqsave(&cleanup_list_lock, flags);
461 list_add(&net->cleanup_list, &cleanup_list);
462 spin_unlock_irqrestore(&cleanup_list_lock, flags);
464 queue_work(netns_wq, &net_cleanup_work);
466 EXPORT_SYMBOL_GPL(__put_net);
468 struct net *get_net_ns_by_fd(int fd)
471 struct ns_common *ns;
474 file = proc_ns_fget(fd);
476 return ERR_CAST(file);
478 ns = get_proc_ns(file_inode(file));
479 if (ns->ops == &netns_operations)
480 net = get_net(container_of(ns, struct net, ns));
482 net = ERR_PTR(-EINVAL);
489 struct net *get_net_ns_by_fd(int fd)
491 return ERR_PTR(-EINVAL);
494 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
496 struct net *get_net_ns_by_pid(pid_t pid)
498 struct task_struct *tsk;
501 /* Lookup the network namespace */
502 net = ERR_PTR(-ESRCH);
504 tsk = find_task_by_vpid(pid);
506 struct nsproxy *nsproxy;
508 nsproxy = tsk->nsproxy;
510 net = get_net(nsproxy->net_ns);
516 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
518 static __net_init int net_ns_net_init(struct net *net)
521 net->ns.ops = &netns_operations;
523 return ns_alloc_inum(&net->ns);
526 static __net_exit void net_ns_net_exit(struct net *net)
528 ns_free_inum(&net->ns);
531 static struct pernet_operations __net_initdata net_ns_ops = {
532 .init = net_ns_net_init,
533 .exit = net_ns_net_exit,
536 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
537 [NETNSA_NONE] = { .type = NLA_UNSPEC },
538 [NETNSA_NSID] = { .type = NLA_S32 },
539 [NETNSA_PID] = { .type = NLA_U32 },
540 [NETNSA_FD] = { .type = NLA_U32 },
543 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh)
545 struct net *net = sock_net(skb->sk);
546 struct nlattr *tb[NETNSA_MAX + 1];
551 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
555 if (!tb[NETNSA_NSID])
557 nsid = nla_get_s32(tb[NETNSA_NSID]);
560 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
561 else if (tb[NETNSA_FD])
562 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
566 return PTR_ERR(peer);
568 spin_lock_irqsave(&net->nsid_lock, flags);
569 if (__peernet2id(net, peer) >= 0) {
570 spin_unlock_irqrestore(&net->nsid_lock, flags);
575 err = alloc_netid(net, peer, nsid);
576 spin_unlock_irqrestore(&net->nsid_lock, flags);
578 rtnl_net_notifyid(net, RTM_NEWNSID, err);
586 static int rtnl_net_get_size(void)
588 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
589 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
593 static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags,
594 int cmd, struct net *net, int nsid)
596 struct nlmsghdr *nlh;
597 struct rtgenmsg *rth;
599 nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags);
603 rth = nlmsg_data(nlh);
604 rth->rtgen_family = AF_UNSPEC;
606 if (nla_put_s32(skb, NETNSA_NSID, nsid))
607 goto nla_put_failure;
613 nlmsg_cancel(skb, nlh);
617 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh)
619 struct net *net = sock_net(skb->sk);
620 struct nlattr *tb[NETNSA_MAX + 1];
625 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
630 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
631 else if (tb[NETNSA_FD])
632 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
637 return PTR_ERR(peer);
639 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
645 id = peernet2id(net, peer);
646 err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0,
647 RTM_NEWNSID, net, id);
651 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
661 struct rtnl_net_dump_cb {
664 struct netlink_callback *cb;
669 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
671 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
674 if (net_cb->idx < net_cb->s_idx)
677 ret = rtnl_net_fill(net_cb->skb, NETLINK_CB(net_cb->cb->skb).portid,
678 net_cb->cb->nlh->nlmsg_seq, NLM_F_MULTI,
679 RTM_NEWNSID, net_cb->net, id);
688 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
690 struct net *net = sock_net(skb->sk);
691 struct rtnl_net_dump_cb net_cb = {
696 .s_idx = cb->args[0],
700 spin_lock_irqsave(&net->nsid_lock, flags);
701 idr_for_each(&net->netns_ids, rtnl_net_dumpid_one, &net_cb);
702 spin_unlock_irqrestore(&net->nsid_lock, flags);
704 cb->args[0] = net_cb.idx;
708 static void rtnl_net_notifyid(struct net *net, int cmd, int id)
713 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
717 err = rtnl_net_fill(msg, 0, 0, 0, cmd, net, id);
721 rtnl_notify(msg, net, 0, RTNLGRP_NSID, NULL, 0);
727 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
730 static int __init net_ns_init(void)
732 struct net_generic *ng;
735 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
739 /* Create workqueue for cleanup */
740 netns_wq = create_singlethread_workqueue("netns");
742 panic("Could not create netns workq");
745 ng = net_alloc_generic();
747 panic("Could not allocate generic netns");
749 rcu_assign_pointer(init_net.gen, ng);
751 mutex_lock(&net_mutex);
752 if (setup_net(&init_net, &init_user_ns))
753 panic("Could not setup the initial network namespace");
755 init_net_initialized = true;
758 list_add_tail_rcu(&init_net.list, &net_namespace_list);
761 mutex_unlock(&net_mutex);
763 register_pernet_subsys(&net_ns_ops);
765 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL, NULL);
766 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
772 pure_initcall(net_ns_init);
775 static int __register_pernet_operations(struct list_head *list,
776 struct pernet_operations *ops)
780 LIST_HEAD(net_exit_list);
782 list_add_tail(&ops->list, list);
783 if (ops->init || (ops->id && ops->size)) {
785 error = ops_init(ops, net);
788 list_add_tail(&net->exit_list, &net_exit_list);
794 /* If I have an error cleanup all namespaces I initialized */
795 list_del(&ops->list);
796 ops_exit_list(ops, &net_exit_list);
797 ops_free_list(ops, &net_exit_list);
801 static void __unregister_pernet_operations(struct pernet_operations *ops)
804 LIST_HEAD(net_exit_list);
806 list_del(&ops->list);
808 list_add_tail(&net->exit_list, &net_exit_list);
809 ops_exit_list(ops, &net_exit_list);
810 ops_free_list(ops, &net_exit_list);
815 static int __register_pernet_operations(struct list_head *list,
816 struct pernet_operations *ops)
818 if (!init_net_initialized) {
819 list_add_tail(&ops->list, list);
823 return ops_init(ops, &init_net);
826 static void __unregister_pernet_operations(struct pernet_operations *ops)
828 if (!init_net_initialized) {
829 list_del(&ops->list);
831 LIST_HEAD(net_exit_list);
832 list_add(&init_net.exit_list, &net_exit_list);
833 ops_exit_list(ops, &net_exit_list);
834 ops_free_list(ops, &net_exit_list);
838 #endif /* CONFIG_NET_NS */
840 static DEFINE_IDA(net_generic_ids);
842 static int register_pernet_operations(struct list_head *list,
843 struct pernet_operations *ops)
849 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
851 if (error == -EAGAIN) {
852 ida_pre_get(&net_generic_ids, GFP_KERNEL);
857 max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
859 error = __register_pernet_operations(list, ops);
863 ida_remove(&net_generic_ids, *ops->id);
869 static void unregister_pernet_operations(struct pernet_operations *ops)
872 __unregister_pernet_operations(ops);
875 ida_remove(&net_generic_ids, *ops->id);
879 * register_pernet_subsys - register a network namespace subsystem
880 * @ops: pernet operations structure for the subsystem
882 * Register a subsystem which has init and exit functions
883 * that are called when network namespaces are created and
884 * destroyed respectively.
886 * When registered all network namespace init functions are
887 * called for every existing network namespace. Allowing kernel
888 * modules to have a race free view of the set of network namespaces.
890 * When a new network namespace is created all of the init
891 * methods are called in the order in which they were registered.
893 * When a network namespace is destroyed all of the exit methods
894 * are called in the reverse of the order with which they were
897 int register_pernet_subsys(struct pernet_operations *ops)
900 mutex_lock(&net_mutex);
901 error = register_pernet_operations(first_device, ops);
902 mutex_unlock(&net_mutex);
905 EXPORT_SYMBOL_GPL(register_pernet_subsys);
908 * unregister_pernet_subsys - unregister a network namespace subsystem
909 * @ops: pernet operations structure to manipulate
911 * Remove the pernet operations structure from the list to be
912 * used when network namespaces are created or destroyed. In
913 * addition run the exit method for all existing network
916 void unregister_pernet_subsys(struct pernet_operations *ops)
918 mutex_lock(&net_mutex);
919 unregister_pernet_operations(ops);
920 mutex_unlock(&net_mutex);
922 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
925 * register_pernet_device - register a network namespace device
926 * @ops: pernet operations structure for the subsystem
928 * Register a device which has init and exit functions
929 * that are called when network namespaces are created and
930 * destroyed respectively.
932 * When registered all network namespace init functions are
933 * called for every existing network namespace. Allowing kernel
934 * modules to have a race free view of the set of network namespaces.
936 * When a new network namespace is created all of the init
937 * methods are called in the order in which they were registered.
939 * When a network namespace is destroyed all of the exit methods
940 * are called in the reverse of the order with which they were
943 int register_pernet_device(struct pernet_operations *ops)
946 mutex_lock(&net_mutex);
947 error = register_pernet_operations(&pernet_list, ops);
948 if (!error && (first_device == &pernet_list))
949 first_device = &ops->list;
950 mutex_unlock(&net_mutex);
953 EXPORT_SYMBOL_GPL(register_pernet_device);
956 * unregister_pernet_device - unregister a network namespace netdevice
957 * @ops: pernet operations structure to manipulate
959 * Remove the pernet operations structure from the list to be
960 * used when network namespaces are created or destroyed. In
961 * addition run the exit method for all existing network
964 void unregister_pernet_device(struct pernet_operations *ops)
966 mutex_lock(&net_mutex);
967 if (&ops->list == first_device)
968 first_device = first_device->next;
969 unregister_pernet_operations(ops);
970 mutex_unlock(&net_mutex);
972 EXPORT_SYMBOL_GPL(unregister_pernet_device);
975 static struct ns_common *netns_get(struct task_struct *task)
977 struct net *net = NULL;
978 struct nsproxy *nsproxy;
981 nsproxy = task->nsproxy;
983 net = get_net(nsproxy->net_ns);
986 return net ? &net->ns : NULL;
989 static inline struct net *to_net_ns(struct ns_common *ns)
991 return container_of(ns, struct net, ns);
994 static void netns_put(struct ns_common *ns)
996 put_net(to_net_ns(ns));
999 static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
1001 struct net *net = to_net_ns(ns);
1003 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1004 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1007 put_net(nsproxy->net_ns);
1008 nsproxy->net_ns = get_net(net);
1012 const struct proc_ns_operations netns_operations = {
1014 .type = CLONE_NEWNET,
1017 .install = netns_install,