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)
221 spin_lock_bh(&net->nsid_lock);
222 alloc = atomic_read(&peer->count) == 0 ? false : true;
223 id = __peernet2id_alloc(net, peer, &alloc);
224 spin_unlock_bh(&net->nsid_lock);
225 if (alloc && id >= 0)
226 rtnl_net_notifyid(net, RTM_NEWNSID, id);
230 /* This function returns, if assigned, the id of a peer netns. */
231 int peernet2id(struct net *net, struct net *peer)
235 spin_lock_bh(&net->nsid_lock);
236 id = __peernet2id(net, peer);
237 spin_unlock_bh(&net->nsid_lock);
240 EXPORT_SYMBOL(peernet2id);
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)
258 spin_lock_bh(&net->nsid_lock);
259 peer = idr_find(&net->netns_ids, id);
262 spin_unlock_bh(&net->nsid_lock);
269 * setup_net runs the initializers for the network namespace object.
271 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
273 /* Must be called with net_mutex held */
274 const struct pernet_operations *ops, *saved_ops;
276 LIST_HEAD(net_exit_list);
278 atomic_set(&net->count, 1);
279 atomic_set(&net->passive, 1);
280 net->dev_base_seq = 1;
281 net->user_ns = user_ns;
282 idr_init(&net->netns_ids);
283 spin_lock_init(&net->nsid_lock);
285 list_for_each_entry(ops, &pernet_list, list) {
286 error = ops_init(ops, net);
294 /* Walk through the list backwards calling the exit functions
295 * for the pernet modules whose init functions did not fail.
297 list_add(&net->exit_list, &net_exit_list);
299 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
300 ops_exit_list(ops, &net_exit_list);
303 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
304 ops_free_list(ops, &net_exit_list);
312 static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
314 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
317 static void dec_net_namespaces(struct ucounts *ucounts)
319 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
322 static struct kmem_cache *net_cachep;
323 static struct workqueue_struct *netns_wq;
325 static struct net *net_alloc(void)
327 struct net *net = NULL;
328 struct net_generic *ng;
330 ng = net_alloc_generic();
334 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
338 rcu_assign_pointer(net->gen, ng);
347 static void net_free(struct net *net)
349 kfree(rcu_access_pointer(net->gen));
350 kmem_cache_free(net_cachep, net);
353 void net_drop_ns(void *p)
356 if (ns && atomic_dec_and_test(&ns->passive))
360 struct net *copy_net_ns(unsigned long flags,
361 struct user_namespace *user_ns, struct net *old_net)
363 struct ucounts *ucounts;
367 if (!(flags & CLONE_NEWNET))
368 return get_net(old_net);
370 ucounts = inc_net_namespaces(user_ns);
372 return ERR_PTR(-ENOSPC);
376 dec_net_namespaces(ucounts);
377 return ERR_PTR(-ENOMEM);
380 get_user_ns(user_ns);
382 mutex_lock(&net_mutex);
383 net->ucounts = ucounts;
384 rv = setup_net(net, user_ns);
387 list_add_tail_rcu(&net->list, &net_namespace_list);
390 mutex_unlock(&net_mutex);
392 dec_net_namespaces(ucounts);
393 put_user_ns(user_ns);
400 static DEFINE_SPINLOCK(cleanup_list_lock);
401 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
403 static void cleanup_net(struct work_struct *work)
405 const struct pernet_operations *ops;
406 struct net *net, *tmp;
407 struct list_head net_kill_list;
408 LIST_HEAD(net_exit_list);
410 /* Atomically snapshot the list of namespaces to cleanup */
411 spin_lock_irq(&cleanup_list_lock);
412 list_replace_init(&cleanup_list, &net_kill_list);
413 spin_unlock_irq(&cleanup_list_lock);
415 mutex_lock(&net_mutex);
417 /* Don't let anyone else find us. */
419 list_for_each_entry(net, &net_kill_list, cleanup_list) {
420 list_del_rcu(&net->list);
421 list_add_tail(&net->exit_list, &net_exit_list);
425 spin_lock_bh(&tmp->nsid_lock);
426 id = __peernet2id(tmp, net);
428 idr_remove(&tmp->netns_ids, id);
429 spin_unlock_bh(&tmp->nsid_lock);
431 rtnl_net_notifyid(tmp, RTM_DELNSID, id);
433 spin_lock_bh(&net->nsid_lock);
434 idr_destroy(&net->netns_ids);
435 spin_unlock_bh(&net->nsid_lock);
441 * Another CPU might be rcu-iterating the list, wait for it.
442 * This needs to be before calling the exit() notifiers, so
443 * the rcu_barrier() below isn't sufficient alone.
447 /* Run all of the network namespace exit methods */
448 list_for_each_entry_reverse(ops, &pernet_list, list)
449 ops_exit_list(ops, &net_exit_list);
451 /* Free the net generic variables */
452 list_for_each_entry_reverse(ops, &pernet_list, list)
453 ops_free_list(ops, &net_exit_list);
455 mutex_unlock(&net_mutex);
457 /* Ensure there are no outstanding rcu callbacks using this
462 /* Finally it is safe to free my network namespace structure */
463 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
464 list_del_init(&net->exit_list);
465 dec_net_namespaces(net->ucounts);
466 put_user_ns(net->user_ns);
470 static DECLARE_WORK(net_cleanup_work, cleanup_net);
472 void __put_net(struct net *net)
474 /* Cleanup the network namespace in process context */
477 spin_lock_irqsave(&cleanup_list_lock, flags);
478 list_add(&net->cleanup_list, &cleanup_list);
479 spin_unlock_irqrestore(&cleanup_list_lock, flags);
481 queue_work(netns_wq, &net_cleanup_work);
483 EXPORT_SYMBOL_GPL(__put_net);
485 struct net *get_net_ns_by_fd(int fd)
488 struct ns_common *ns;
491 file = proc_ns_fget(fd);
493 return ERR_CAST(file);
495 ns = get_proc_ns(file_inode(file));
496 if (ns->ops == &netns_operations)
497 net = get_net(container_of(ns, struct net, ns));
499 net = ERR_PTR(-EINVAL);
506 struct net *get_net_ns_by_fd(int fd)
508 return ERR_PTR(-EINVAL);
511 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
513 struct net *get_net_ns_by_pid(pid_t pid)
515 struct task_struct *tsk;
518 /* Lookup the network namespace */
519 net = ERR_PTR(-ESRCH);
521 tsk = find_task_by_vpid(pid);
523 struct nsproxy *nsproxy;
525 nsproxy = tsk->nsproxy;
527 net = get_net(nsproxy->net_ns);
533 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
535 static __net_init int net_ns_net_init(struct net *net)
538 net->ns.ops = &netns_operations;
540 return ns_alloc_inum(&net->ns);
543 static __net_exit void net_ns_net_exit(struct net *net)
545 ns_free_inum(&net->ns);
548 static struct pernet_operations __net_initdata net_ns_ops = {
549 .init = net_ns_net_init,
550 .exit = net_ns_net_exit,
553 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
554 [NETNSA_NONE] = { .type = NLA_UNSPEC },
555 [NETNSA_NSID] = { .type = NLA_S32 },
556 [NETNSA_PID] = { .type = NLA_U32 },
557 [NETNSA_FD] = { .type = NLA_U32 },
560 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh)
562 struct net *net = sock_net(skb->sk);
563 struct nlattr *tb[NETNSA_MAX + 1];
567 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
571 if (!tb[NETNSA_NSID])
573 nsid = nla_get_s32(tb[NETNSA_NSID]);
576 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
577 else if (tb[NETNSA_FD])
578 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
582 return PTR_ERR(peer);
584 spin_lock_bh(&net->nsid_lock);
585 if (__peernet2id(net, peer) >= 0) {
586 spin_unlock_bh(&net->nsid_lock);
591 err = alloc_netid(net, peer, nsid);
592 spin_unlock_bh(&net->nsid_lock);
594 rtnl_net_notifyid(net, RTM_NEWNSID, err);
602 static int rtnl_net_get_size(void)
604 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
605 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
609 static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags,
610 int cmd, struct net *net, int nsid)
612 struct nlmsghdr *nlh;
613 struct rtgenmsg *rth;
615 nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags);
619 rth = nlmsg_data(nlh);
620 rth->rtgen_family = AF_UNSPEC;
622 if (nla_put_s32(skb, NETNSA_NSID, nsid))
623 goto nla_put_failure;
629 nlmsg_cancel(skb, nlh);
633 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh)
635 struct net *net = sock_net(skb->sk);
636 struct nlattr *tb[NETNSA_MAX + 1];
641 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
646 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
647 else if (tb[NETNSA_FD])
648 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
653 return PTR_ERR(peer);
655 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
661 id = peernet2id(net, peer);
662 err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0,
663 RTM_NEWNSID, net, id);
667 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
677 struct rtnl_net_dump_cb {
680 struct netlink_callback *cb;
685 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
687 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
690 if (net_cb->idx < net_cb->s_idx)
693 ret = rtnl_net_fill(net_cb->skb, NETLINK_CB(net_cb->cb->skb).portid,
694 net_cb->cb->nlh->nlmsg_seq, NLM_F_MULTI,
695 RTM_NEWNSID, net_cb->net, id);
704 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
706 struct net *net = sock_net(skb->sk);
707 struct rtnl_net_dump_cb net_cb = {
712 .s_idx = cb->args[0],
715 spin_lock_bh(&net->nsid_lock);
716 idr_for_each(&net->netns_ids, rtnl_net_dumpid_one, &net_cb);
717 spin_unlock_bh(&net->nsid_lock);
719 cb->args[0] = net_cb.idx;
723 static void rtnl_net_notifyid(struct net *net, int cmd, int id)
728 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
732 err = rtnl_net_fill(msg, 0, 0, 0, cmd, net, id);
736 rtnl_notify(msg, net, 0, RTNLGRP_NSID, NULL, 0);
742 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
745 static int __init net_ns_init(void)
747 struct net_generic *ng;
750 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
754 /* Create workqueue for cleanup */
755 netns_wq = create_singlethread_workqueue("netns");
757 panic("Could not create netns workq");
760 ng = net_alloc_generic();
762 panic("Could not allocate generic netns");
764 rcu_assign_pointer(init_net.gen, ng);
766 mutex_lock(&net_mutex);
767 if (setup_net(&init_net, &init_user_ns))
768 panic("Could not setup the initial network namespace");
770 init_net_initialized = true;
773 list_add_tail_rcu(&init_net.list, &net_namespace_list);
776 mutex_unlock(&net_mutex);
778 register_pernet_subsys(&net_ns_ops);
780 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL, NULL);
781 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
787 pure_initcall(net_ns_init);
790 static int __register_pernet_operations(struct list_head *list,
791 struct pernet_operations *ops)
795 LIST_HEAD(net_exit_list);
797 list_add_tail(&ops->list, list);
798 if (ops->init || (ops->id && ops->size)) {
800 error = ops_init(ops, net);
803 list_add_tail(&net->exit_list, &net_exit_list);
809 /* If I have an error cleanup all namespaces I initialized */
810 list_del(&ops->list);
811 ops_exit_list(ops, &net_exit_list);
812 ops_free_list(ops, &net_exit_list);
816 static void __unregister_pernet_operations(struct pernet_operations *ops)
819 LIST_HEAD(net_exit_list);
821 list_del(&ops->list);
823 list_add_tail(&net->exit_list, &net_exit_list);
824 ops_exit_list(ops, &net_exit_list);
825 ops_free_list(ops, &net_exit_list);
830 static int __register_pernet_operations(struct list_head *list,
831 struct pernet_operations *ops)
833 if (!init_net_initialized) {
834 list_add_tail(&ops->list, list);
838 return ops_init(ops, &init_net);
841 static void __unregister_pernet_operations(struct pernet_operations *ops)
843 if (!init_net_initialized) {
844 list_del(&ops->list);
846 LIST_HEAD(net_exit_list);
847 list_add(&init_net.exit_list, &net_exit_list);
848 ops_exit_list(ops, &net_exit_list);
849 ops_free_list(ops, &net_exit_list);
853 #endif /* CONFIG_NET_NS */
855 static DEFINE_IDA(net_generic_ids);
857 static int register_pernet_operations(struct list_head *list,
858 struct pernet_operations *ops)
864 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
866 if (error == -EAGAIN) {
867 ida_pre_get(&net_generic_ids, GFP_KERNEL);
872 max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
874 error = __register_pernet_operations(list, ops);
878 ida_remove(&net_generic_ids, *ops->id);
884 static void unregister_pernet_operations(struct pernet_operations *ops)
887 __unregister_pernet_operations(ops);
890 ida_remove(&net_generic_ids, *ops->id);
894 * register_pernet_subsys - register a network namespace subsystem
895 * @ops: pernet operations structure for the subsystem
897 * Register a subsystem which has init and exit functions
898 * that are called when network namespaces are created and
899 * destroyed respectively.
901 * When registered all network namespace init functions are
902 * called for every existing network namespace. Allowing kernel
903 * modules to have a race free view of the set of network namespaces.
905 * When a new network namespace is created all of the init
906 * methods are called in the order in which they were registered.
908 * When a network namespace is destroyed all of the exit methods
909 * are called in the reverse of the order with which they were
912 int register_pernet_subsys(struct pernet_operations *ops)
915 mutex_lock(&net_mutex);
916 error = register_pernet_operations(first_device, ops);
917 mutex_unlock(&net_mutex);
920 EXPORT_SYMBOL_GPL(register_pernet_subsys);
923 * unregister_pernet_subsys - unregister a network namespace subsystem
924 * @ops: pernet operations structure to manipulate
926 * Remove the pernet operations structure from the list to be
927 * used when network namespaces are created or destroyed. In
928 * addition run the exit method for all existing network
931 void unregister_pernet_subsys(struct pernet_operations *ops)
933 mutex_lock(&net_mutex);
934 unregister_pernet_operations(ops);
935 mutex_unlock(&net_mutex);
937 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
940 * register_pernet_device - register a network namespace device
941 * @ops: pernet operations structure for the subsystem
943 * Register a device which has init and exit functions
944 * that are called when network namespaces are created and
945 * destroyed respectively.
947 * When registered all network namespace init functions are
948 * called for every existing network namespace. Allowing kernel
949 * modules to have a race free view of the set of network namespaces.
951 * When a new network namespace is created all of the init
952 * methods are called in the order in which they were registered.
954 * When a network namespace is destroyed all of the exit methods
955 * are called in the reverse of the order with which they were
958 int register_pernet_device(struct pernet_operations *ops)
961 mutex_lock(&net_mutex);
962 error = register_pernet_operations(&pernet_list, ops);
963 if (!error && (first_device == &pernet_list))
964 first_device = &ops->list;
965 mutex_unlock(&net_mutex);
968 EXPORT_SYMBOL_GPL(register_pernet_device);
971 * unregister_pernet_device - unregister a network namespace netdevice
972 * @ops: pernet operations structure to manipulate
974 * Remove the pernet operations structure from the list to be
975 * used when network namespaces are created or destroyed. In
976 * addition run the exit method for all existing network
979 void unregister_pernet_device(struct pernet_operations *ops)
981 mutex_lock(&net_mutex);
982 if (&ops->list == first_device)
983 first_device = first_device->next;
984 unregister_pernet_operations(ops);
985 mutex_unlock(&net_mutex);
987 EXPORT_SYMBOL_GPL(unregister_pernet_device);
990 static struct ns_common *netns_get(struct task_struct *task)
992 struct net *net = NULL;
993 struct nsproxy *nsproxy;
996 nsproxy = task->nsproxy;
998 net = get_net(nsproxy->net_ns);
1001 return net ? &net->ns : NULL;
1004 static inline struct net *to_net_ns(struct ns_common *ns)
1006 return container_of(ns, struct net, ns);
1009 static void netns_put(struct ns_common *ns)
1011 put_net(to_net_ns(ns));
1014 static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
1016 struct net *net = to_net_ns(ns);
1018 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1019 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1022 put_net(nsproxy->net_ns);
1023 nsproxy->net_ns = get_net(net);
1027 static struct user_namespace *netns_owner(struct ns_common *ns)
1029 return to_net_ns(ns)->user_ns;
1032 const struct proc_ns_operations netns_operations = {
1034 .type = CLONE_NEWNET,
1037 .install = netns_install,
1038 .owner = netns_owner,