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>
19 #include <linux/rtnetlink.h>
21 #include <net/netlink.h>
22 #include <net/net_namespace.h>
23 #include <net/netns/generic.h>
26 * Our network namespace constructor/destructor lists
29 static LIST_HEAD(pernet_list);
30 static struct list_head *first_device = &pernet_list;
31 DEFINE_MUTEX(net_mutex);
33 LIST_HEAD(net_namespace_list);
34 EXPORT_SYMBOL_GPL(net_namespace_list);
36 struct net init_net = {
37 .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head),
39 EXPORT_SYMBOL(init_net);
41 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
43 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
45 static struct net_generic *net_alloc_generic(void)
47 struct net_generic *ng;
48 size_t generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
50 ng = kzalloc(generic_size, GFP_KERNEL);
52 ng->len = max_gen_ptrs;
57 static int net_assign_generic(struct net *net, int id, void *data)
59 struct net_generic *ng, *old_ng;
61 BUG_ON(!mutex_is_locked(&net_mutex));
64 old_ng = rcu_dereference_protected(net->gen,
65 lockdep_is_held(&net_mutex));
67 if (old_ng->len >= id)
70 ng = net_alloc_generic();
75 * Some synchronisation notes:
77 * The net_generic explores the net->gen array inside rcu
78 * read section. Besides once set the net->gen->ptr[x]
79 * pointer never changes (see rules in netns/generic.h).
81 * That said, we simply duplicate this array and schedule
82 * the old copy for kfree after a grace period.
85 memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
87 rcu_assign_pointer(net->gen, ng);
88 kfree_rcu(old_ng, rcu);
90 ng->ptr[id - 1] = data;
94 static int ops_init(const struct pernet_operations *ops, struct net *net)
99 if (ops->id && ops->size) {
100 data = kzalloc(ops->size, GFP_KERNEL);
104 err = net_assign_generic(net, *ops->id, data);
110 err = ops->init(net);
121 static void ops_free(const struct pernet_operations *ops, struct net *net)
123 if (ops->id && ops->size) {
125 kfree(net_generic(net, id));
129 static void ops_exit_list(const struct pernet_operations *ops,
130 struct list_head *net_exit_list)
134 list_for_each_entry(net, net_exit_list, exit_list)
138 ops->exit_batch(net_exit_list);
141 static void ops_free_list(const struct pernet_operations *ops,
142 struct list_head *net_exit_list)
145 if (ops->size && ops->id) {
146 list_for_each_entry(net, net_exit_list, exit_list)
151 static int alloc_netid(struct net *net, struct net *peer, int reqid)
153 int min = 0, max = 0;
162 return idr_alloc(&net->netns_ids, peer, min, max, GFP_KERNEL);
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 static int __peernet2id(struct net *net, struct net *peer, bool alloc)
180 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
184 /* Magic value for id 0. */
185 if (id == NET_ID_ZERO)
191 return alloc_netid(net, peer, -1);
196 /* This function returns the id of a peer netns. If no id is assigned, one will
197 * be allocated and returned.
199 int peernet2id(struct net *net, struct net *peer)
201 int id = __peernet2id(net, peer, true);
203 return id >= 0 ? id : NETNSA_NSID_NOT_ASSIGNED;
206 struct net *get_net_ns_by_id(struct net *net, int id)
214 peer = idr_find(&net->netns_ids, id);
223 * setup_net runs the initializers for the network namespace object.
225 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
227 /* Must be called with net_mutex held */
228 const struct pernet_operations *ops, *saved_ops;
230 LIST_HEAD(net_exit_list);
232 atomic_set(&net->count, 1);
233 atomic_set(&net->passive, 1);
234 net->dev_base_seq = 1;
235 net->user_ns = user_ns;
236 idr_init(&net->netns_ids);
238 #ifdef NETNS_REFCNT_DEBUG
239 atomic_set(&net->use_count, 0);
242 list_for_each_entry(ops, &pernet_list, list) {
243 error = ops_init(ops, net);
251 /* Walk through the list backwards calling the exit functions
252 * for the pernet modules whose init functions did not fail.
254 list_add(&net->exit_list, &net_exit_list);
256 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
257 ops_exit_list(ops, &net_exit_list);
260 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
261 ops_free_list(ops, &net_exit_list);
269 static struct kmem_cache *net_cachep;
270 static struct workqueue_struct *netns_wq;
272 static struct net *net_alloc(void)
274 struct net *net = NULL;
275 struct net_generic *ng;
277 ng = net_alloc_generic();
281 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
285 rcu_assign_pointer(net->gen, ng);
294 static void net_free(struct net *net)
296 #ifdef NETNS_REFCNT_DEBUG
297 if (unlikely(atomic_read(&net->use_count) != 0)) {
298 pr_emerg("network namespace not free! Usage: %d\n",
299 atomic_read(&net->use_count));
303 kfree(rcu_access_pointer(net->gen));
304 kmem_cache_free(net_cachep, net);
307 void net_drop_ns(void *p)
310 if (ns && atomic_dec_and_test(&ns->passive))
314 struct net *copy_net_ns(unsigned long flags,
315 struct user_namespace *user_ns, struct net *old_net)
320 if (!(flags & CLONE_NEWNET))
321 return get_net(old_net);
325 return ERR_PTR(-ENOMEM);
327 get_user_ns(user_ns);
329 mutex_lock(&net_mutex);
330 rv = setup_net(net, user_ns);
333 list_add_tail_rcu(&net->list, &net_namespace_list);
336 mutex_unlock(&net_mutex);
338 put_user_ns(user_ns);
345 static DEFINE_SPINLOCK(cleanup_list_lock);
346 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
348 static void cleanup_net(struct work_struct *work)
350 const struct pernet_operations *ops;
351 struct net *net, *tmp;
352 struct list_head net_kill_list;
353 LIST_HEAD(net_exit_list);
355 /* Atomically snapshot the list of namespaces to cleanup */
356 spin_lock_irq(&cleanup_list_lock);
357 list_replace_init(&cleanup_list, &net_kill_list);
358 spin_unlock_irq(&cleanup_list_lock);
360 mutex_lock(&net_mutex);
362 /* Don't let anyone else find us. */
364 list_for_each_entry(net, &net_kill_list, cleanup_list) {
365 list_del_rcu(&net->list);
366 list_add_tail(&net->exit_list, &net_exit_list);
368 int id = __peernet2id(tmp, net, false);
371 idr_remove(&tmp->netns_ids, id);
373 idr_destroy(&net->netns_ids);
379 * Another CPU might be rcu-iterating the list, wait for it.
380 * This needs to be before calling the exit() notifiers, so
381 * the rcu_barrier() below isn't sufficient alone.
385 /* Run all of the network namespace exit methods */
386 list_for_each_entry_reverse(ops, &pernet_list, list)
387 ops_exit_list(ops, &net_exit_list);
389 /* Free the net generic variables */
390 list_for_each_entry_reverse(ops, &pernet_list, list)
391 ops_free_list(ops, &net_exit_list);
393 mutex_unlock(&net_mutex);
395 /* Ensure there are no outstanding rcu callbacks using this
400 /* Finally it is safe to free my network namespace structure */
401 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
402 list_del_init(&net->exit_list);
403 put_user_ns(net->user_ns);
407 static DECLARE_WORK(net_cleanup_work, cleanup_net);
409 void __put_net(struct net *net)
411 /* Cleanup the network namespace in process context */
414 spin_lock_irqsave(&cleanup_list_lock, flags);
415 list_add(&net->cleanup_list, &cleanup_list);
416 spin_unlock_irqrestore(&cleanup_list_lock, flags);
418 queue_work(netns_wq, &net_cleanup_work);
420 EXPORT_SYMBOL_GPL(__put_net);
422 struct net *get_net_ns_by_fd(int fd)
425 struct ns_common *ns;
428 file = proc_ns_fget(fd);
430 return ERR_CAST(file);
432 ns = get_proc_ns(file_inode(file));
433 if (ns->ops == &netns_operations)
434 net = get_net(container_of(ns, struct net, ns));
436 net = ERR_PTR(-EINVAL);
443 struct net *get_net_ns_by_fd(int fd)
445 return ERR_PTR(-EINVAL);
449 struct net *get_net_ns_by_pid(pid_t pid)
451 struct task_struct *tsk;
454 /* Lookup the network namespace */
455 net = ERR_PTR(-ESRCH);
457 tsk = find_task_by_vpid(pid);
459 struct nsproxy *nsproxy;
461 nsproxy = tsk->nsproxy;
463 net = get_net(nsproxy->net_ns);
469 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
471 static __net_init int net_ns_net_init(struct net *net)
474 net->ns.ops = &netns_operations;
476 return ns_alloc_inum(&net->ns);
479 static __net_exit void net_ns_net_exit(struct net *net)
481 ns_free_inum(&net->ns);
484 static struct pernet_operations __net_initdata net_ns_ops = {
485 .init = net_ns_net_init,
486 .exit = net_ns_net_exit,
489 static struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
490 [NETNSA_NONE] = { .type = NLA_UNSPEC },
491 [NETNSA_NSID] = { .type = NLA_S32 },
492 [NETNSA_PID] = { .type = NLA_U32 },
493 [NETNSA_FD] = { .type = NLA_U32 },
496 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh)
498 struct net *net = sock_net(skb->sk);
499 struct nlattr *tb[NETNSA_MAX + 1];
503 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
507 if (!tb[NETNSA_NSID])
509 nsid = nla_get_s32(tb[NETNSA_NSID]);
512 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
513 else if (tb[NETNSA_FD])
514 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
518 return PTR_ERR(peer);
520 if (__peernet2id(net, peer, false) >= 0) {
525 err = alloc_netid(net, peer, nsid);
533 static int rtnl_net_get_size(void)
535 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
536 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
540 static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags,
541 int cmd, struct net *net, struct net *peer)
543 struct nlmsghdr *nlh;
544 struct rtgenmsg *rth;
549 nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags);
553 rth = nlmsg_data(nlh);
554 rth->rtgen_family = AF_UNSPEC;
556 id = __peernet2id(net, peer, false);
558 id = NETNSA_NSID_NOT_ASSIGNED;
559 if (nla_put_s32(skb, NETNSA_NSID, id))
560 goto nla_put_failure;
566 nlmsg_cancel(skb, nlh);
570 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh)
572 struct net *net = sock_net(skb->sk);
573 struct nlattr *tb[NETNSA_MAX + 1];
578 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
583 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
584 else if (tb[NETNSA_FD])
585 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
590 return PTR_ERR(peer);
592 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
598 err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0,
599 RTM_GETNSID, net, peer);
603 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
613 static int __init net_ns_init(void)
615 struct net_generic *ng;
618 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
622 /* Create workqueue for cleanup */
623 netns_wq = create_singlethread_workqueue("netns");
625 panic("Could not create netns workq");
628 ng = net_alloc_generic();
630 panic("Could not allocate generic netns");
632 rcu_assign_pointer(init_net.gen, ng);
634 mutex_lock(&net_mutex);
635 if (setup_net(&init_net, &init_user_ns))
636 panic("Could not setup the initial network namespace");
639 list_add_tail_rcu(&init_net.list, &net_namespace_list);
642 mutex_unlock(&net_mutex);
644 register_pernet_subsys(&net_ns_ops);
646 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL, NULL);
647 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, NULL, NULL);
652 pure_initcall(net_ns_init);
655 static int __register_pernet_operations(struct list_head *list,
656 struct pernet_operations *ops)
660 LIST_HEAD(net_exit_list);
662 list_add_tail(&ops->list, list);
663 if (ops->init || (ops->id && ops->size)) {
665 error = ops_init(ops, net);
668 list_add_tail(&net->exit_list, &net_exit_list);
674 /* If I have an error cleanup all namespaces I initialized */
675 list_del(&ops->list);
676 ops_exit_list(ops, &net_exit_list);
677 ops_free_list(ops, &net_exit_list);
681 static void __unregister_pernet_operations(struct pernet_operations *ops)
684 LIST_HEAD(net_exit_list);
686 list_del(&ops->list);
688 list_add_tail(&net->exit_list, &net_exit_list);
689 ops_exit_list(ops, &net_exit_list);
690 ops_free_list(ops, &net_exit_list);
695 static int __register_pernet_operations(struct list_head *list,
696 struct pernet_operations *ops)
698 return ops_init(ops, &init_net);
701 static void __unregister_pernet_operations(struct pernet_operations *ops)
703 LIST_HEAD(net_exit_list);
704 list_add(&init_net.exit_list, &net_exit_list);
705 ops_exit_list(ops, &net_exit_list);
706 ops_free_list(ops, &net_exit_list);
709 #endif /* CONFIG_NET_NS */
711 static DEFINE_IDA(net_generic_ids);
713 static int register_pernet_operations(struct list_head *list,
714 struct pernet_operations *ops)
720 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
722 if (error == -EAGAIN) {
723 ida_pre_get(&net_generic_ids, GFP_KERNEL);
728 max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
730 error = __register_pernet_operations(list, ops);
734 ida_remove(&net_generic_ids, *ops->id);
740 static void unregister_pernet_operations(struct pernet_operations *ops)
743 __unregister_pernet_operations(ops);
746 ida_remove(&net_generic_ids, *ops->id);
750 * register_pernet_subsys - register a network namespace subsystem
751 * @ops: pernet operations structure for the subsystem
753 * Register a subsystem which has init and exit functions
754 * that are called when network namespaces are created and
755 * destroyed respectively.
757 * When registered all network namespace init functions are
758 * called for every existing network namespace. Allowing kernel
759 * modules to have a race free view of the set of network namespaces.
761 * When a new network namespace is created all of the init
762 * methods are called in the order in which they were registered.
764 * When a network namespace is destroyed all of the exit methods
765 * are called in the reverse of the order with which they were
768 int register_pernet_subsys(struct pernet_operations *ops)
771 mutex_lock(&net_mutex);
772 error = register_pernet_operations(first_device, ops);
773 mutex_unlock(&net_mutex);
776 EXPORT_SYMBOL_GPL(register_pernet_subsys);
779 * unregister_pernet_subsys - unregister a network namespace subsystem
780 * @ops: pernet operations structure to manipulate
782 * Remove the pernet operations structure from the list to be
783 * used when network namespaces are created or destroyed. In
784 * addition run the exit method for all existing network
787 void unregister_pernet_subsys(struct pernet_operations *ops)
789 mutex_lock(&net_mutex);
790 unregister_pernet_operations(ops);
791 mutex_unlock(&net_mutex);
793 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
796 * register_pernet_device - register a network namespace device
797 * @ops: pernet operations structure for the subsystem
799 * Register a device which has init and exit functions
800 * that are called when network namespaces are created and
801 * destroyed respectively.
803 * When registered all network namespace init functions are
804 * called for every existing network namespace. Allowing kernel
805 * modules to have a race free view of the set of network namespaces.
807 * When a new network namespace is created all of the init
808 * methods are called in the order in which they were registered.
810 * When a network namespace is destroyed all of the exit methods
811 * are called in the reverse of the order with which they were
814 int register_pernet_device(struct pernet_operations *ops)
817 mutex_lock(&net_mutex);
818 error = register_pernet_operations(&pernet_list, ops);
819 if (!error && (first_device == &pernet_list))
820 first_device = &ops->list;
821 mutex_unlock(&net_mutex);
824 EXPORT_SYMBOL_GPL(register_pernet_device);
827 * unregister_pernet_device - unregister a network namespace netdevice
828 * @ops: pernet operations structure to manipulate
830 * Remove the pernet operations structure from the list to be
831 * used when network namespaces are created or destroyed. In
832 * addition run the exit method for all existing network
835 void unregister_pernet_device(struct pernet_operations *ops)
837 mutex_lock(&net_mutex);
838 if (&ops->list == first_device)
839 first_device = first_device->next;
840 unregister_pernet_operations(ops);
841 mutex_unlock(&net_mutex);
843 EXPORT_SYMBOL_GPL(unregister_pernet_device);
846 static struct ns_common *netns_get(struct task_struct *task)
848 struct net *net = NULL;
849 struct nsproxy *nsproxy;
852 nsproxy = task->nsproxy;
854 net = get_net(nsproxy->net_ns);
857 return net ? &net->ns : NULL;
860 static inline struct net *to_net_ns(struct ns_common *ns)
862 return container_of(ns, struct net, ns);
865 static void netns_put(struct ns_common *ns)
867 put_net(to_net_ns(ns));
870 static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
872 struct net *net = to_net_ns(ns);
874 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
875 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
878 put_net(nsproxy->net_ns);
879 nsproxy->net_ns = get_net(net);
883 const struct proc_ns_operations netns_operations = {
885 .type = CLONE_NEWNET,
888 .install = netns_install,