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 if (atomic_read(&net->count) == 0)
221 return NETNSA_NSID_NOT_ASSIGNED;
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);
264 peer = maybe_get_net(peer);
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 get_random_bytes(&net->hash_mix, sizeof(u32));
284 net->dev_base_seq = 1;
285 net->user_ns = user_ns;
286 idr_init(&net->netns_ids);
287 spin_lock_init(&net->nsid_lock);
289 list_for_each_entry(ops, &pernet_list, list) {
290 error = ops_init(ops, net);
298 /* Walk through the list backwards calling the exit functions
299 * for the pernet modules whose init functions did not fail.
301 list_add(&net->exit_list, &net_exit_list);
303 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
304 ops_exit_list(ops, &net_exit_list);
307 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
308 ops_free_list(ops, &net_exit_list);
314 static int __net_init net_defaults_init_net(struct net *net)
316 net->core.sysctl_somaxconn = SOMAXCONN;
320 static struct pernet_operations net_defaults_ops = {
321 .init = net_defaults_init_net,
324 static __init int net_defaults_init(void)
326 if (register_pernet_subsys(&net_defaults_ops))
327 panic("Cannot initialize net default settings");
332 core_initcall(net_defaults_init);
335 static struct kmem_cache *net_cachep;
336 static struct workqueue_struct *netns_wq;
338 static struct net *net_alloc(void)
340 struct net *net = NULL;
341 struct net_generic *ng;
343 ng = net_alloc_generic();
347 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
351 rcu_assign_pointer(net->gen, ng);
360 static void net_free(struct net *net)
362 kfree(rcu_access_pointer(net->gen));
363 kmem_cache_free(net_cachep, net);
366 void net_drop_ns(void *p)
369 if (ns && atomic_dec_and_test(&ns->passive))
373 struct net *copy_net_ns(unsigned long flags,
374 struct user_namespace *user_ns, struct net *old_net)
379 if (!(flags & CLONE_NEWNET))
380 return get_net(old_net);
384 return ERR_PTR(-ENOMEM);
386 get_user_ns(user_ns);
388 mutex_lock(&net_mutex);
389 rv = setup_net(net, user_ns);
392 list_add_tail_rcu(&net->list, &net_namespace_list);
395 mutex_unlock(&net_mutex);
397 put_user_ns(user_ns);
404 static DEFINE_SPINLOCK(cleanup_list_lock);
405 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
407 static void cleanup_net(struct work_struct *work)
409 const struct pernet_operations *ops;
410 struct net *net, *tmp;
411 struct list_head net_kill_list;
412 LIST_HEAD(net_exit_list);
414 /* Atomically snapshot the list of namespaces to cleanup */
415 spin_lock_irq(&cleanup_list_lock);
416 list_replace_init(&cleanup_list, &net_kill_list);
417 spin_unlock_irq(&cleanup_list_lock);
419 mutex_lock(&net_mutex);
421 /* Don't let anyone else find us. */
423 list_for_each_entry(net, &net_kill_list, cleanup_list) {
424 list_del_rcu(&net->list);
425 list_add_tail(&net->exit_list, &net_exit_list);
429 spin_lock_irq(&tmp->nsid_lock);
430 id = __peernet2id(tmp, net);
432 idr_remove(&tmp->netns_ids, id);
433 spin_unlock_irq(&tmp->nsid_lock);
435 rtnl_net_notifyid(tmp, RTM_DELNSID, id);
437 spin_lock_irq(&net->nsid_lock);
438 idr_destroy(&net->netns_ids);
439 spin_unlock_irq(&net->nsid_lock);
445 * Another CPU might be rcu-iterating the list, wait for it.
446 * This needs to be before calling the exit() notifiers, so
447 * the rcu_barrier() below isn't sufficient alone.
451 /* Run all of the network namespace exit methods */
452 list_for_each_entry_reverse(ops, &pernet_list, list)
453 ops_exit_list(ops, &net_exit_list);
455 /* Free the net generic variables */
456 list_for_each_entry_reverse(ops, &pernet_list, list)
457 ops_free_list(ops, &net_exit_list);
459 mutex_unlock(&net_mutex);
461 /* Ensure there are no outstanding rcu callbacks using this
466 /* Finally it is safe to free my network namespace structure */
467 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
468 list_del_init(&net->exit_list);
469 put_user_ns(net->user_ns);
473 static DECLARE_WORK(net_cleanup_work, cleanup_net);
475 void __put_net(struct net *net)
477 /* Cleanup the network namespace in process context */
480 spin_lock_irqsave(&cleanup_list_lock, flags);
481 list_add(&net->cleanup_list, &cleanup_list);
482 spin_unlock_irqrestore(&cleanup_list_lock, flags);
484 queue_work(netns_wq, &net_cleanup_work);
486 EXPORT_SYMBOL_GPL(__put_net);
488 struct net *get_net_ns_by_fd(int fd)
491 struct ns_common *ns;
494 file = proc_ns_fget(fd);
496 return ERR_CAST(file);
498 ns = get_proc_ns(file_inode(file));
499 if (ns->ops == &netns_operations)
500 net = get_net(container_of(ns, struct net, ns));
502 net = ERR_PTR(-EINVAL);
509 struct net *get_net_ns_by_fd(int fd)
511 return ERR_PTR(-EINVAL);
514 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
516 struct net *get_net_ns_by_pid(pid_t pid)
518 struct task_struct *tsk;
521 /* Lookup the network namespace */
522 net = ERR_PTR(-ESRCH);
524 tsk = find_task_by_vpid(pid);
526 struct nsproxy *nsproxy;
528 nsproxy = tsk->nsproxy;
530 net = get_net(nsproxy->net_ns);
536 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
538 static __net_init int net_ns_net_init(struct net *net)
541 net->ns.ops = &netns_operations;
543 return ns_alloc_inum(&net->ns);
546 static __net_exit void net_ns_net_exit(struct net *net)
548 ns_free_inum(&net->ns);
551 static struct pernet_operations __net_initdata net_ns_ops = {
552 .init = net_ns_net_init,
553 .exit = net_ns_net_exit,
556 static struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
557 [NETNSA_NONE] = { .type = NLA_UNSPEC },
558 [NETNSA_NSID] = { .type = NLA_S32 },
559 [NETNSA_PID] = { .type = NLA_U32 },
560 [NETNSA_FD] = { .type = NLA_U32 },
563 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh)
565 struct net *net = sock_net(skb->sk);
566 struct nlattr *tb[NETNSA_MAX + 1];
571 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
575 if (!tb[NETNSA_NSID])
577 nsid = nla_get_s32(tb[NETNSA_NSID]);
580 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
581 else if (tb[NETNSA_FD])
582 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
586 return PTR_ERR(peer);
588 spin_lock_irqsave(&net->nsid_lock, flags);
589 if (__peernet2id(net, peer) >= 0) {
590 spin_unlock_irqrestore(&net->nsid_lock, flags);
595 err = alloc_netid(net, peer, nsid);
596 spin_unlock_irqrestore(&net->nsid_lock, flags);
598 rtnl_net_notifyid(net, RTM_NEWNSID, err);
606 static int rtnl_net_get_size(void)
608 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
609 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
613 static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags,
614 int cmd, struct net *net, int nsid)
616 struct nlmsghdr *nlh;
617 struct rtgenmsg *rth;
619 nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags);
623 rth = nlmsg_data(nlh);
624 rth->rtgen_family = AF_UNSPEC;
626 if (nla_put_s32(skb, NETNSA_NSID, nsid))
627 goto nla_put_failure;
633 nlmsg_cancel(skb, nlh);
637 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh)
639 struct net *net = sock_net(skb->sk);
640 struct nlattr *tb[NETNSA_MAX + 1];
645 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
650 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
651 else if (tb[NETNSA_FD])
652 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
657 return PTR_ERR(peer);
659 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
665 id = peernet2id(net, peer);
666 err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0,
667 RTM_NEWNSID, net, id);
671 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
681 struct rtnl_net_dump_cb {
684 struct netlink_callback *cb;
689 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
691 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
694 if (net_cb->idx < net_cb->s_idx)
697 ret = rtnl_net_fill(net_cb->skb, NETLINK_CB(net_cb->cb->skb).portid,
698 net_cb->cb->nlh->nlmsg_seq, NLM_F_MULTI,
699 RTM_NEWNSID, net_cb->net, id);
708 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
710 struct net *net = sock_net(skb->sk);
711 struct rtnl_net_dump_cb net_cb = {
716 .s_idx = cb->args[0],
720 spin_lock_irqsave(&net->nsid_lock, flags);
721 idr_for_each(&net->netns_ids, rtnl_net_dumpid_one, &net_cb);
722 spin_unlock_irqrestore(&net->nsid_lock, flags);
724 cb->args[0] = net_cb.idx;
728 static void rtnl_net_notifyid(struct net *net, int cmd, int id)
733 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
737 err = rtnl_net_fill(msg, 0, 0, 0, cmd, net, id);
741 rtnl_notify(msg, net, 0, RTNLGRP_NSID, NULL, 0);
747 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
750 static int __init net_ns_init(void)
752 struct net_generic *ng;
755 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
759 /* Create workqueue for cleanup */
760 netns_wq = create_singlethread_workqueue("netns");
762 panic("Could not create netns workq");
765 ng = net_alloc_generic();
767 panic("Could not allocate generic netns");
769 rcu_assign_pointer(init_net.gen, ng);
771 mutex_lock(&net_mutex);
772 if (setup_net(&init_net, &init_user_ns))
773 panic("Could not setup the initial network namespace");
776 list_add_tail_rcu(&init_net.list, &net_namespace_list);
779 mutex_unlock(&net_mutex);
781 if (register_pernet_subsys(&net_ns_ops))
782 panic("Could not register network namespace subsystems");
784 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL, NULL);
785 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
791 pure_initcall(net_ns_init);
794 static int __register_pernet_operations(struct list_head *list,
795 struct pernet_operations *ops)
799 LIST_HEAD(net_exit_list);
801 list_add_tail(&ops->list, list);
802 if (ops->init || (ops->id && ops->size)) {
804 error = ops_init(ops, net);
807 list_add_tail(&net->exit_list, &net_exit_list);
813 /* If I have an error cleanup all namespaces I initialized */
814 list_del(&ops->list);
815 ops_exit_list(ops, &net_exit_list);
816 ops_free_list(ops, &net_exit_list);
820 static void __unregister_pernet_operations(struct pernet_operations *ops)
823 LIST_HEAD(net_exit_list);
825 list_del(&ops->list);
827 list_add_tail(&net->exit_list, &net_exit_list);
828 ops_exit_list(ops, &net_exit_list);
829 ops_free_list(ops, &net_exit_list);
834 static int __register_pernet_operations(struct list_head *list,
835 struct pernet_operations *ops)
837 return ops_init(ops, &init_net);
840 static void __unregister_pernet_operations(struct pernet_operations *ops)
842 LIST_HEAD(net_exit_list);
843 list_add(&init_net.exit_list, &net_exit_list);
844 ops_exit_list(ops, &net_exit_list);
845 ops_free_list(ops, &net_exit_list);
848 #endif /* CONFIG_NET_NS */
850 static DEFINE_IDA(net_generic_ids);
852 static int register_pernet_operations(struct list_head *list,
853 struct pernet_operations *ops)
859 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
861 if (error == -EAGAIN) {
862 ida_pre_get(&net_generic_ids, GFP_KERNEL);
867 max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
869 error = __register_pernet_operations(list, ops);
873 ida_remove(&net_generic_ids, *ops->id);
879 static void unregister_pernet_operations(struct pernet_operations *ops)
882 __unregister_pernet_operations(ops);
885 ida_remove(&net_generic_ids, *ops->id);
889 * register_pernet_subsys - register a network namespace subsystem
890 * @ops: pernet operations structure for the subsystem
892 * Register a subsystem which has init and exit functions
893 * that are called when network namespaces are created and
894 * destroyed respectively.
896 * When registered all network namespace init functions are
897 * called for every existing network namespace. Allowing kernel
898 * modules to have a race free view of the set of network namespaces.
900 * When a new network namespace is created all of the init
901 * methods are called in the order in which they were registered.
903 * When a network namespace is destroyed all of the exit methods
904 * are called in the reverse of the order with which they were
907 int register_pernet_subsys(struct pernet_operations *ops)
910 mutex_lock(&net_mutex);
911 error = register_pernet_operations(first_device, ops);
912 mutex_unlock(&net_mutex);
915 EXPORT_SYMBOL_GPL(register_pernet_subsys);
918 * unregister_pernet_subsys - unregister a network namespace subsystem
919 * @ops: pernet operations structure to manipulate
921 * Remove the pernet operations structure from the list to be
922 * used when network namespaces are created or destroyed. In
923 * addition run the exit method for all existing network
926 void unregister_pernet_subsys(struct pernet_operations *ops)
928 mutex_lock(&net_mutex);
929 unregister_pernet_operations(ops);
930 mutex_unlock(&net_mutex);
932 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
935 * register_pernet_device - register a network namespace device
936 * @ops: pernet operations structure for the subsystem
938 * Register a device which has init and exit functions
939 * that are called when network namespaces are created and
940 * destroyed respectively.
942 * When registered all network namespace init functions are
943 * called for every existing network namespace. Allowing kernel
944 * modules to have a race free view of the set of network namespaces.
946 * When a new network namespace is created all of the init
947 * methods are called in the order in which they were registered.
949 * When a network namespace is destroyed all of the exit methods
950 * are called in the reverse of the order with which they were
953 int register_pernet_device(struct pernet_operations *ops)
956 mutex_lock(&net_mutex);
957 error = register_pernet_operations(&pernet_list, ops);
958 if (!error && (first_device == &pernet_list))
959 first_device = &ops->list;
960 mutex_unlock(&net_mutex);
963 EXPORT_SYMBOL_GPL(register_pernet_device);
966 * unregister_pernet_device - unregister a network namespace netdevice
967 * @ops: pernet operations structure to manipulate
969 * Remove the pernet operations structure from the list to be
970 * used when network namespaces are created or destroyed. In
971 * addition run the exit method for all existing network
974 void unregister_pernet_device(struct pernet_operations *ops)
976 mutex_lock(&net_mutex);
977 if (&ops->list == first_device)
978 first_device = first_device->next;
979 unregister_pernet_operations(ops);
980 mutex_unlock(&net_mutex);
982 EXPORT_SYMBOL_GPL(unregister_pernet_device);
985 static struct ns_common *netns_get(struct task_struct *task)
987 struct net *net = NULL;
988 struct nsproxy *nsproxy;
991 nsproxy = task->nsproxy;
993 net = get_net(nsproxy->net_ns);
996 return net ? &net->ns : NULL;
999 static inline struct net *to_net_ns(struct ns_common *ns)
1001 return container_of(ns, struct net, ns);
1004 static void netns_put(struct ns_common *ns)
1006 put_net(to_net_ns(ns));
1009 static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
1011 struct net *net = to_net_ns(ns);
1013 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1014 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1017 put_net(nsproxy->net_ns);
1018 nsproxy->net_ns = get_net(net);
1022 const struct proc_ns_operations netns_operations = {
1024 .type = CLONE_NEWNET,
1027 .install = netns_install,