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);
269 static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
271 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
274 static void dec_net_namespaces(struct ucounts *ucounts)
276 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
280 * setup_net runs the initializers for the network namespace object.
282 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
284 /* Must be called with net_mutex held */
285 const struct pernet_operations *ops, *saved_ops;
287 LIST_HEAD(net_exit_list);
289 atomic_set(&net->count, 1);
290 atomic_set(&net->passive, 1);
291 net->dev_base_seq = 1;
292 net->user_ns = user_ns;
293 idr_init(&net->netns_ids);
294 spin_lock_init(&net->nsid_lock);
296 list_for_each_entry(ops, &pernet_list, list) {
297 error = ops_init(ops, net);
305 /* Walk through the list backwards calling the exit functions
306 * for the pernet modules whose init functions did not fail.
308 list_add(&net->exit_list, &net_exit_list);
310 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
311 ops_exit_list(ops, &net_exit_list);
314 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
315 ops_free_list(ops, &net_exit_list);
323 static struct kmem_cache *net_cachep;
324 static struct workqueue_struct *netns_wq;
326 static struct net *net_alloc(void)
328 struct net *net = NULL;
329 struct net_generic *ng;
331 ng = net_alloc_generic();
335 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
339 rcu_assign_pointer(net->gen, ng);
348 static void net_free(struct net *net)
350 kfree(rcu_access_pointer(net->gen));
351 kmem_cache_free(net_cachep, net);
354 void net_drop_ns(void *p)
357 if (ns && atomic_dec_and_test(&ns->passive))
361 struct net *copy_net_ns(unsigned long flags,
362 struct user_namespace *user_ns, struct net *old_net)
364 struct ucounts *ucounts;
368 if (!(flags & CLONE_NEWNET))
369 return get_net(old_net);
371 ucounts = inc_net_namespaces(user_ns);
373 return ERR_PTR(-ENOSPC);
377 dec_net_namespaces(ucounts);
378 return ERR_PTR(-ENOMEM);
381 get_user_ns(user_ns);
383 mutex_lock(&net_mutex);
384 net->ucounts = ucounts;
385 rv = setup_net(net, user_ns);
388 list_add_tail_rcu(&net->list, &net_namespace_list);
391 mutex_unlock(&net_mutex);
393 dec_net_namespaces(ucounts);
394 put_user_ns(user_ns);
401 static DEFINE_SPINLOCK(cleanup_list_lock);
402 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
404 static void cleanup_net(struct work_struct *work)
406 const struct pernet_operations *ops;
407 struct net *net, *tmp;
408 struct list_head net_kill_list;
409 LIST_HEAD(net_exit_list);
411 /* Atomically snapshot the list of namespaces to cleanup */
412 spin_lock_irq(&cleanup_list_lock);
413 list_replace_init(&cleanup_list, &net_kill_list);
414 spin_unlock_irq(&cleanup_list_lock);
416 mutex_lock(&net_mutex);
418 /* Don't let anyone else find us. */
420 list_for_each_entry(net, &net_kill_list, cleanup_list) {
421 list_del_rcu(&net->list);
422 list_add_tail(&net->exit_list, &net_exit_list);
426 spin_lock_irq(&tmp->nsid_lock);
427 id = __peernet2id(tmp, net);
429 idr_remove(&tmp->netns_ids, id);
430 spin_unlock_irq(&tmp->nsid_lock);
432 rtnl_net_notifyid(tmp, RTM_DELNSID, id);
434 spin_lock_irq(&net->nsid_lock);
435 idr_destroy(&net->netns_ids);
436 spin_unlock_irq(&net->nsid_lock);
442 * Another CPU might be rcu-iterating the list, wait for it.
443 * This needs to be before calling the exit() notifiers, so
444 * the rcu_barrier() below isn't sufficient alone.
448 /* Run all of the network namespace exit methods */
449 list_for_each_entry_reverse(ops, &pernet_list, list)
450 ops_exit_list(ops, &net_exit_list);
452 /* Free the net generic variables */
453 list_for_each_entry_reverse(ops, &pernet_list, list)
454 ops_free_list(ops, &net_exit_list);
456 mutex_unlock(&net_mutex);
458 /* Ensure there are no outstanding rcu callbacks using this
463 /* Finally it is safe to free my network namespace structure */
464 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
465 list_del_init(&net->exit_list);
466 dec_net_namespaces(net->ucounts);
467 put_user_ns(net->user_ns);
471 static DECLARE_WORK(net_cleanup_work, cleanup_net);
473 void __put_net(struct net *net)
475 /* Cleanup the network namespace in process context */
478 spin_lock_irqsave(&cleanup_list_lock, flags);
479 list_add(&net->cleanup_list, &cleanup_list);
480 spin_unlock_irqrestore(&cleanup_list_lock, flags);
482 queue_work(netns_wq, &net_cleanup_work);
484 EXPORT_SYMBOL_GPL(__put_net);
486 struct net *get_net_ns_by_fd(int fd)
489 struct ns_common *ns;
492 file = proc_ns_fget(fd);
494 return ERR_CAST(file);
496 ns = get_proc_ns(file_inode(file));
497 if (ns->ops == &netns_operations)
498 net = get_net(container_of(ns, struct net, ns));
500 net = ERR_PTR(-EINVAL);
507 struct net *get_net_ns_by_fd(int fd)
509 return ERR_PTR(-EINVAL);
512 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
514 struct net *get_net_ns_by_pid(pid_t pid)
516 struct task_struct *tsk;
519 /* Lookup the network namespace */
520 net = ERR_PTR(-ESRCH);
522 tsk = find_task_by_vpid(pid);
524 struct nsproxy *nsproxy;
526 nsproxy = tsk->nsproxy;
528 net = get_net(nsproxy->net_ns);
534 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
536 static __net_init int net_ns_net_init(struct net *net)
539 net->ns.ops = &netns_operations;
541 return ns_alloc_inum(&net->ns);
544 static __net_exit void net_ns_net_exit(struct net *net)
546 ns_free_inum(&net->ns);
549 static struct pernet_operations __net_initdata net_ns_ops = {
550 .init = net_ns_net_init,
551 .exit = net_ns_net_exit,
554 static struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
555 [NETNSA_NONE] = { .type = NLA_UNSPEC },
556 [NETNSA_NSID] = { .type = NLA_S32 },
557 [NETNSA_PID] = { .type = NLA_U32 },
558 [NETNSA_FD] = { .type = NLA_U32 },
561 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh)
563 struct net *net = sock_net(skb->sk);
564 struct nlattr *tb[NETNSA_MAX + 1];
569 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
573 if (!tb[NETNSA_NSID])
575 nsid = nla_get_s32(tb[NETNSA_NSID]);
578 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
579 else if (tb[NETNSA_FD])
580 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
584 return PTR_ERR(peer);
586 spin_lock_irqsave(&net->nsid_lock, flags);
587 if (__peernet2id(net, peer) >= 0) {
588 spin_unlock_irqrestore(&net->nsid_lock, flags);
593 err = alloc_netid(net, peer, nsid);
594 spin_unlock_irqrestore(&net->nsid_lock, flags);
596 rtnl_net_notifyid(net, RTM_NEWNSID, err);
604 static int rtnl_net_get_size(void)
606 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
607 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
611 static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags,
612 int cmd, struct net *net, int nsid)
614 struct nlmsghdr *nlh;
615 struct rtgenmsg *rth;
617 nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags);
621 rth = nlmsg_data(nlh);
622 rth->rtgen_family = AF_UNSPEC;
624 if (nla_put_s32(skb, NETNSA_NSID, nsid))
625 goto nla_put_failure;
631 nlmsg_cancel(skb, nlh);
635 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh)
637 struct net *net = sock_net(skb->sk);
638 struct nlattr *tb[NETNSA_MAX + 1];
643 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
648 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
649 else if (tb[NETNSA_FD])
650 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
655 return PTR_ERR(peer);
657 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
663 id = peernet2id(net, peer);
664 err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0,
665 RTM_NEWNSID, net, id);
669 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
679 struct rtnl_net_dump_cb {
682 struct netlink_callback *cb;
687 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
689 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
692 if (net_cb->idx < net_cb->s_idx)
695 ret = rtnl_net_fill(net_cb->skb, NETLINK_CB(net_cb->cb->skb).portid,
696 net_cb->cb->nlh->nlmsg_seq, NLM_F_MULTI,
697 RTM_NEWNSID, net_cb->net, id);
706 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
708 struct net *net = sock_net(skb->sk);
709 struct rtnl_net_dump_cb net_cb = {
714 .s_idx = cb->args[0],
718 spin_lock_irqsave(&net->nsid_lock, flags);
719 idr_for_each(&net->netns_ids, rtnl_net_dumpid_one, &net_cb);
720 spin_unlock_irqrestore(&net->nsid_lock, flags);
722 cb->args[0] = net_cb.idx;
726 static void rtnl_net_notifyid(struct net *net, int cmd, int id)
731 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
735 err = rtnl_net_fill(msg, 0, 0, 0, cmd, net, id);
739 rtnl_notify(msg, net, 0, RTNLGRP_NSID, NULL, 0);
745 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
748 static int __init net_ns_init(void)
750 struct net_generic *ng;
753 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
757 /* Create workqueue for cleanup */
758 netns_wq = create_singlethread_workqueue("netns");
760 panic("Could not create netns workq");
763 ng = net_alloc_generic();
765 panic("Could not allocate generic netns");
767 rcu_assign_pointer(init_net.gen, ng);
769 mutex_lock(&net_mutex);
770 if (setup_net(&init_net, &init_user_ns))
771 panic("Could not setup the initial network namespace");
774 list_add_tail_rcu(&init_net.list, &net_namespace_list);
777 mutex_unlock(&net_mutex);
779 register_pernet_subsys(&net_ns_ops);
781 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL, NULL);
782 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
788 pure_initcall(net_ns_init);
791 static int __register_pernet_operations(struct list_head *list,
792 struct pernet_operations *ops)
796 LIST_HEAD(net_exit_list);
798 list_add_tail(&ops->list, list);
799 if (ops->init || (ops->id && ops->size)) {
801 error = ops_init(ops, net);
804 list_add_tail(&net->exit_list, &net_exit_list);
810 /* If I have an error cleanup all namespaces I initialized */
811 list_del(&ops->list);
812 ops_exit_list(ops, &net_exit_list);
813 ops_free_list(ops, &net_exit_list);
817 static void __unregister_pernet_operations(struct pernet_operations *ops)
820 LIST_HEAD(net_exit_list);
822 list_del(&ops->list);
824 list_add_tail(&net->exit_list, &net_exit_list);
825 ops_exit_list(ops, &net_exit_list);
826 ops_free_list(ops, &net_exit_list);
831 static int __register_pernet_operations(struct list_head *list,
832 struct pernet_operations *ops)
834 return ops_init(ops, &init_net);
837 static void __unregister_pernet_operations(struct pernet_operations *ops)
839 LIST_HEAD(net_exit_list);
840 list_add(&init_net.exit_list, &net_exit_list);
841 ops_exit_list(ops, &net_exit_list);
842 ops_free_list(ops, &net_exit_list);
845 #endif /* CONFIG_NET_NS */
847 static DEFINE_IDA(net_generic_ids);
849 static int register_pernet_operations(struct list_head *list,
850 struct pernet_operations *ops)
856 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
858 if (error == -EAGAIN) {
859 ida_pre_get(&net_generic_ids, GFP_KERNEL);
864 max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
866 error = __register_pernet_operations(list, ops);
870 ida_remove(&net_generic_ids, *ops->id);
876 static void unregister_pernet_operations(struct pernet_operations *ops)
879 __unregister_pernet_operations(ops);
882 ida_remove(&net_generic_ids, *ops->id);
886 * register_pernet_subsys - register a network namespace subsystem
887 * @ops: pernet operations structure for the subsystem
889 * Register a subsystem which has init and exit functions
890 * that are called when network namespaces are created and
891 * destroyed respectively.
893 * When registered all network namespace init functions are
894 * called for every existing network namespace. Allowing kernel
895 * modules to have a race free view of the set of network namespaces.
897 * When a new network namespace is created all of the init
898 * methods are called in the order in which they were registered.
900 * When a network namespace is destroyed all of the exit methods
901 * are called in the reverse of the order with which they were
904 int register_pernet_subsys(struct pernet_operations *ops)
907 mutex_lock(&net_mutex);
908 error = register_pernet_operations(first_device, ops);
909 mutex_unlock(&net_mutex);
912 EXPORT_SYMBOL_GPL(register_pernet_subsys);
915 * unregister_pernet_subsys - unregister a network namespace subsystem
916 * @ops: pernet operations structure to manipulate
918 * Remove the pernet operations structure from the list to be
919 * used when network namespaces are created or destroyed. In
920 * addition run the exit method for all existing network
923 void unregister_pernet_subsys(struct pernet_operations *ops)
925 mutex_lock(&net_mutex);
926 unregister_pernet_operations(ops);
927 mutex_unlock(&net_mutex);
929 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
932 * register_pernet_device - register a network namespace device
933 * @ops: pernet operations structure for the subsystem
935 * Register a device which has init and exit functions
936 * that are called when network namespaces are created and
937 * destroyed respectively.
939 * When registered all network namespace init functions are
940 * called for every existing network namespace. Allowing kernel
941 * modules to have a race free view of the set of network namespaces.
943 * When a new network namespace is created all of the init
944 * methods are called in the order in which they were registered.
946 * When a network namespace is destroyed all of the exit methods
947 * are called in the reverse of the order with which they were
950 int register_pernet_device(struct pernet_operations *ops)
953 mutex_lock(&net_mutex);
954 error = register_pernet_operations(&pernet_list, ops);
955 if (!error && (first_device == &pernet_list))
956 first_device = &ops->list;
957 mutex_unlock(&net_mutex);
960 EXPORT_SYMBOL_GPL(register_pernet_device);
963 * unregister_pernet_device - unregister a network namespace netdevice
964 * @ops: pernet operations structure to manipulate
966 * Remove the pernet operations structure from the list to be
967 * used when network namespaces are created or destroyed. In
968 * addition run the exit method for all existing network
971 void unregister_pernet_device(struct pernet_operations *ops)
973 mutex_lock(&net_mutex);
974 if (&ops->list == first_device)
975 first_device = first_device->next;
976 unregister_pernet_operations(ops);
977 mutex_unlock(&net_mutex);
979 EXPORT_SYMBOL_GPL(unregister_pernet_device);
982 static struct ns_common *netns_get(struct task_struct *task)
984 struct net *net = NULL;
985 struct nsproxy *nsproxy;
988 nsproxy = task->nsproxy;
990 net = get_net(nsproxy->net_ns);
993 return net ? &net->ns : NULL;
996 static inline struct net *to_net_ns(struct ns_common *ns)
998 return container_of(ns, struct net, ns);
1001 static void netns_put(struct ns_common *ns)
1003 put_net(to_net_ns(ns));
1006 static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
1008 struct net *net = to_net_ns(ns);
1010 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1011 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1014 put_net(nsproxy->net_ns);
1015 nsproxy->net_ns = get_net(net);
1019 static struct user_namespace *netns_owner(struct ns_common *ns)
1021 return to_net_ns(ns)->user_ns;
1024 const struct proc_ns_operations netns_operations = {
1026 .type = CLONE_NEWNET,
1029 .install = netns_install,
1030 .owner = netns_owner,