1 // SPDX-License-Identifier: GPL-2.0-only
2 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
4 #include <linux/workqueue.h>
5 #include <linux/rtnetlink.h>
6 #include <linux/cache.h>
7 #include <linux/slab.h>
8 #include <linux/list.h>
9 #include <linux/delay.h>
10 #include <linux/sched.h>
11 #include <linux/idr.h>
12 #include <linux/rculist.h>
13 #include <linux/nsproxy.h>
15 #include <linux/proc_ns.h>
16 #include <linux/file.h>
17 #include <linux/export.h>
18 #include <linux/user_namespace.h>
19 #include <linux/net_namespace.h>
20 #include <linux/sched/task.h>
21 #include <linux/uidgid.h>
24 #include <net/netlink.h>
25 #include <net/net_namespace.h>
26 #include <net/netns/generic.h>
29 * Our network namespace constructor/destructor lists
32 static LIST_HEAD(pernet_list);
33 static struct list_head *first_device = &pernet_list;
35 LIST_HEAD(net_namespace_list);
36 EXPORT_SYMBOL_GPL(net_namespace_list);
38 /* Protects net_namespace_list. Nests iside rtnl_lock() */
39 DECLARE_RWSEM(net_rwsem);
40 EXPORT_SYMBOL_GPL(net_rwsem);
43 static struct key_tag init_net_key_domain = { .usage = REFCOUNT_INIT(1) };
46 struct net init_net = {
47 .count = REFCOUNT_INIT(1),
48 .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head),
50 .key_domain = &init_net_key_domain,
53 EXPORT_SYMBOL(init_net);
55 static bool init_net_initialized;
57 * pernet_ops_rwsem: protects: pernet_list, net_generic_ids,
58 * init_net_initialized and first_device pointer.
59 * This is internal net namespace object. Please, don't use it
62 DECLARE_RWSEM(pernet_ops_rwsem);
63 EXPORT_SYMBOL_GPL(pernet_ops_rwsem);
65 #define MIN_PERNET_OPS_ID \
66 ((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *))
68 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
70 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
72 static struct net_generic *net_alloc_generic(void)
74 struct net_generic *ng;
75 unsigned int generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
77 ng = kzalloc(generic_size, GFP_KERNEL);
79 ng->s.len = max_gen_ptrs;
84 static int net_assign_generic(struct net *net, unsigned int id, void *data)
86 struct net_generic *ng, *old_ng;
88 BUG_ON(id < MIN_PERNET_OPS_ID);
90 old_ng = rcu_dereference_protected(net->gen,
91 lockdep_is_held(&pernet_ops_rwsem));
92 if (old_ng->s.len > id) {
93 old_ng->ptr[id] = data;
97 ng = net_alloc_generic();
102 * Some synchronisation notes:
104 * The net_generic explores the net->gen array inside rcu
105 * read section. Besides once set the net->gen->ptr[x]
106 * pointer never changes (see rules in netns/generic.h).
108 * That said, we simply duplicate this array and schedule
109 * the old copy for kfree after a grace period.
112 memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
113 (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
116 rcu_assign_pointer(net->gen, ng);
117 kfree_rcu(old_ng, s.rcu);
121 static int ops_init(const struct pernet_operations *ops, struct net *net)
123 struct net_generic *ng;
127 if (ops->id && ops->size) {
128 data = kzalloc(ops->size, GFP_KERNEL);
132 err = net_assign_generic(net, *ops->id, data);
138 err = ops->init(net);
142 if (ops->id && ops->size) {
143 ng = rcu_dereference_protected(net->gen,
144 lockdep_is_held(&pernet_ops_rwsem));
145 ng->ptr[*ops->id] = NULL;
155 static void ops_free(const struct pernet_operations *ops, struct net *net)
157 if (ops->id && ops->size) {
158 kfree(net_generic(net, *ops->id));
162 static void ops_pre_exit_list(const struct pernet_operations *ops,
163 struct list_head *net_exit_list)
168 list_for_each_entry(net, net_exit_list, exit_list)
173 static void ops_exit_list(const struct pernet_operations *ops,
174 struct list_head *net_exit_list)
178 list_for_each_entry(net, net_exit_list, exit_list) {
184 ops->exit_batch(net_exit_list);
187 static void ops_free_list(const struct pernet_operations *ops,
188 struct list_head *net_exit_list)
191 if (ops->size && ops->id) {
192 list_for_each_entry(net, net_exit_list, exit_list)
197 /* should be called with nsid_lock held */
198 static int alloc_netid(struct net *net, struct net *peer, int reqid)
200 int min = 0, max = 0;
207 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
210 /* This function is used by idr_for_each(). If net is equal to peer, the
211 * function returns the id so that idr_for_each() stops. Because we cannot
212 * returns the id 0 (idr_for_each() will not stop), we return the magic value
213 * NET_ID_ZERO (-1) for it.
215 #define NET_ID_ZERO -1
216 static int net_eq_idr(int id, void *net, void *peer)
218 if (net_eq(net, peer))
219 return id ? : NET_ID_ZERO;
223 /* Must be called from RCU-critical section or with nsid_lock held. If
224 * a new id is assigned, the bool alloc is set to true, thus the
225 * caller knows that the new id must be notified via rtnl.
227 static int __peernet2id_alloc(struct net *net, struct net *peer, bool *alloc)
229 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
230 bool alloc_it = *alloc;
234 /* Magic value for id 0. */
235 if (id == NET_ID_ZERO)
241 id = alloc_netid(net, peer, -1);
243 return id >= 0 ? id : NETNSA_NSID_NOT_ASSIGNED;
246 return NETNSA_NSID_NOT_ASSIGNED;
249 /* Must be called from RCU-critical section or with nsid_lock held */
250 static int __peernet2id(struct net *net, struct net *peer)
254 return __peernet2id_alloc(net, peer, &no);
257 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
258 struct nlmsghdr *nlh, gfp_t gfp);
259 /* This function returns the id of a peer netns. If no id is assigned, one will
260 * be allocated and returned.
262 int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp)
264 bool alloc = false, alive = false;
267 if (refcount_read(&net->count) == 0)
268 return NETNSA_NSID_NOT_ASSIGNED;
269 spin_lock_bh(&net->nsid_lock);
271 * When peer is obtained from RCU lists, we may race with
272 * its cleanup. Check whether it's alive, and this guarantees
273 * we never hash a peer back to net->netns_ids, after it has
274 * just been idr_remove()'d from there in cleanup_net().
276 if (maybe_get_net(peer))
277 alive = alloc = true;
278 id = __peernet2id_alloc(net, peer, &alloc);
279 spin_unlock_bh(&net->nsid_lock);
280 if (alloc && id >= 0)
281 rtnl_net_notifyid(net, RTM_NEWNSID, id, 0, NULL, gfp);
286 EXPORT_SYMBOL_GPL(peernet2id_alloc);
288 /* This function returns, if assigned, the id of a peer netns. */
289 int peernet2id(struct net *net, struct net *peer)
294 id = __peernet2id(net, peer);
299 EXPORT_SYMBOL(peernet2id);
301 /* This function returns true is the peer netns has an id assigned into the
304 bool peernet_has_id(struct net *net, struct net *peer)
306 return peernet2id(net, peer) >= 0;
309 struct net *get_net_ns_by_id(struct net *net, int id)
317 peer = idr_find(&net->netns_ids, id);
319 peer = maybe_get_net(peer);
326 * setup_net runs the initializers for the network namespace object.
328 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
330 /* Must be called with pernet_ops_rwsem held */
331 const struct pernet_operations *ops, *saved_ops;
333 LIST_HEAD(net_exit_list);
335 refcount_set(&net->count, 1);
336 refcount_set(&net->passive, 1);
337 get_random_bytes(&net->hash_mix, sizeof(u32));
338 net->dev_base_seq = 1;
339 net->user_ns = user_ns;
340 idr_init(&net->netns_ids);
341 spin_lock_init(&net->nsid_lock);
342 mutex_init(&net->ipv4.ra_mutex);
344 list_for_each_entry(ops, &pernet_list, list) {
345 error = ops_init(ops, net);
349 down_write(&net_rwsem);
350 list_add_tail_rcu(&net->list, &net_namespace_list);
351 up_write(&net_rwsem);
356 /* Walk through the list backwards calling the exit functions
357 * for the pernet modules whose init functions did not fail.
359 list_add(&net->exit_list, &net_exit_list);
361 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
362 ops_pre_exit_list(ops, &net_exit_list);
367 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
368 ops_exit_list(ops, &net_exit_list);
371 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
372 ops_free_list(ops, &net_exit_list);
378 static int __net_init net_defaults_init_net(struct net *net)
380 net->core.sysctl_somaxconn = SOMAXCONN;
384 static struct pernet_operations net_defaults_ops = {
385 .init = net_defaults_init_net,
388 static __init int net_defaults_init(void)
390 if (register_pernet_subsys(&net_defaults_ops))
391 panic("Cannot initialize net default settings");
396 core_initcall(net_defaults_init);
399 static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
401 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
404 static void dec_net_namespaces(struct ucounts *ucounts)
406 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
409 static struct kmem_cache *net_cachep __ro_after_init;
410 static struct workqueue_struct *netns_wq;
412 static struct net *net_alloc(void)
414 struct net *net = NULL;
415 struct net_generic *ng;
417 ng = net_alloc_generic();
421 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
426 net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
427 if (!net->key_domain)
429 refcount_set(&net->key_domain->usage, 1);
432 rcu_assign_pointer(net->gen, ng);
438 kmem_cache_free(net_cachep, net);
446 static void net_free(struct net *net)
448 kfree(rcu_access_pointer(net->gen));
449 kmem_cache_free(net_cachep, net);
452 void net_drop_ns(void *p)
455 if (ns && refcount_dec_and_test(&ns->passive))
459 struct net *copy_net_ns(unsigned long flags,
460 struct user_namespace *user_ns, struct net *old_net)
462 struct ucounts *ucounts;
466 if (!(flags & CLONE_NEWNET))
467 return get_net(old_net);
469 ucounts = inc_net_namespaces(user_ns);
471 return ERR_PTR(-ENOSPC);
478 refcount_set(&net->passive, 1);
479 net->ucounts = ucounts;
480 get_user_ns(user_ns);
482 rv = down_read_killable(&pernet_ops_rwsem);
486 rv = setup_net(net, user_ns);
488 up_read(&pernet_ops_rwsem);
493 key_remove_domain(net->key_domain);
495 put_user_ns(user_ns);
498 dec_net_namespaces(ucounts);
505 * net_ns_get_ownership - get sysfs ownership data for @net
506 * @net: network namespace in question (can be NULL)
507 * @uid: kernel user ID for sysfs objects
508 * @gid: kernel group ID for sysfs objects
510 * Returns the uid/gid pair of root in the user namespace associated with the
511 * given network namespace.
513 void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
516 kuid_t ns_root_uid = make_kuid(net->user_ns, 0);
517 kgid_t ns_root_gid = make_kgid(net->user_ns, 0);
519 if (uid_valid(ns_root_uid))
522 if (gid_valid(ns_root_gid))
525 *uid = GLOBAL_ROOT_UID;
526 *gid = GLOBAL_ROOT_GID;
529 EXPORT_SYMBOL_GPL(net_ns_get_ownership);
531 static void unhash_nsid(struct net *net, struct net *last)
534 /* This function is only called from cleanup_net() work,
535 * and this work is the only process, that may delete
536 * a net from net_namespace_list. So, when the below
537 * is executing, the list may only grow. Thus, we do not
538 * use for_each_net_rcu() or net_rwsem.
543 spin_lock_bh(&tmp->nsid_lock);
544 id = __peernet2id(tmp, net);
546 idr_remove(&tmp->netns_ids, id);
547 spin_unlock_bh(&tmp->nsid_lock);
549 rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL,
554 spin_lock_bh(&net->nsid_lock);
555 idr_destroy(&net->netns_ids);
556 spin_unlock_bh(&net->nsid_lock);
559 static LLIST_HEAD(cleanup_list);
561 static void cleanup_net(struct work_struct *work)
563 const struct pernet_operations *ops;
564 struct net *net, *tmp, *last;
565 struct llist_node *net_kill_list;
566 LIST_HEAD(net_exit_list);
568 /* Atomically snapshot the list of namespaces to cleanup */
569 net_kill_list = llist_del_all(&cleanup_list);
571 down_read(&pernet_ops_rwsem);
573 /* Don't let anyone else find us. */
574 down_write(&net_rwsem);
575 llist_for_each_entry(net, net_kill_list, cleanup_list)
576 list_del_rcu(&net->list);
577 /* Cache last net. After we unlock rtnl, no one new net
578 * added to net_namespace_list can assign nsid pointer
579 * to a net from net_kill_list (see peernet2id_alloc()).
580 * So, we skip them in unhash_nsid().
582 * Note, that unhash_nsid() does not delete nsid links
583 * between net_kill_list's nets, as they've already
584 * deleted from net_namespace_list. But, this would be
585 * useless anyway, as netns_ids are destroyed there.
587 last = list_last_entry(&net_namespace_list, struct net, list);
588 up_write(&net_rwsem);
590 llist_for_each_entry(net, net_kill_list, cleanup_list) {
591 unhash_nsid(net, last);
592 list_add_tail(&net->exit_list, &net_exit_list);
595 /* Run all of the network namespace pre_exit methods */
596 list_for_each_entry_reverse(ops, &pernet_list, list)
597 ops_pre_exit_list(ops, &net_exit_list);
600 * Another CPU might be rcu-iterating the list, wait for it.
601 * This needs to be before calling the exit() notifiers, so
602 * the rcu_barrier() below isn't sufficient alone.
603 * Also the pre_exit() and exit() methods need this barrier.
607 /* Run all of the network namespace exit methods */
608 list_for_each_entry_reverse(ops, &pernet_list, list)
609 ops_exit_list(ops, &net_exit_list);
611 /* Free the net generic variables */
612 list_for_each_entry_reverse(ops, &pernet_list, list)
613 ops_free_list(ops, &net_exit_list);
615 up_read(&pernet_ops_rwsem);
617 /* Ensure there are no outstanding rcu callbacks using this
622 /* Finally it is safe to free my network namespace structure */
623 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
624 list_del_init(&net->exit_list);
625 dec_net_namespaces(net->ucounts);
627 key_remove_domain(net->key_domain);
629 put_user_ns(net->user_ns);
635 * net_ns_barrier - wait until concurrent net_cleanup_work is done
637 * cleanup_net runs from work queue and will first remove namespaces
638 * from the global list, then run net exit functions.
640 * Call this in module exit path to make sure that all netns
641 * ->exit ops have been invoked before the function is removed.
643 void net_ns_barrier(void)
645 down_write(&pernet_ops_rwsem);
646 up_write(&pernet_ops_rwsem);
648 EXPORT_SYMBOL(net_ns_barrier);
650 static DECLARE_WORK(net_cleanup_work, cleanup_net);
652 void __put_net(struct net *net)
654 /* Cleanup the network namespace in process context */
655 if (llist_add(&net->cleanup_list, &cleanup_list))
656 queue_work(netns_wq, &net_cleanup_work);
658 EXPORT_SYMBOL_GPL(__put_net);
661 * get_net_ns - increment the refcount of the network namespace
662 * @ns: common namespace (net)
664 * Returns the net's common namespace.
666 struct ns_common *get_net_ns(struct ns_common *ns)
668 return &get_net(container_of(ns, struct net, ns))->ns;
670 EXPORT_SYMBOL_GPL(get_net_ns);
672 struct net *get_net_ns_by_fd(int fd)
675 struct ns_common *ns;
678 file = proc_ns_fget(fd);
680 return ERR_CAST(file);
682 ns = get_proc_ns(file_inode(file));
683 if (ns->ops == &netns_operations)
684 net = get_net(container_of(ns, struct net, ns));
686 net = ERR_PTR(-EINVAL);
693 struct net *get_net_ns_by_fd(int fd)
695 return ERR_PTR(-EINVAL);
698 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
700 struct net *get_net_ns_by_pid(pid_t pid)
702 struct task_struct *tsk;
705 /* Lookup the network namespace */
706 net = ERR_PTR(-ESRCH);
708 tsk = find_task_by_vpid(pid);
710 struct nsproxy *nsproxy;
712 nsproxy = tsk->nsproxy;
714 net = get_net(nsproxy->net_ns);
720 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
722 static __net_init int net_ns_net_init(struct net *net)
725 net->ns.ops = &netns_operations;
727 return ns_alloc_inum(&net->ns);
730 static __net_exit void net_ns_net_exit(struct net *net)
732 ns_free_inum(&net->ns);
735 static struct pernet_operations __net_initdata net_ns_ops = {
736 .init = net_ns_net_init,
737 .exit = net_ns_net_exit,
740 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
741 [NETNSA_NONE] = { .type = NLA_UNSPEC },
742 [NETNSA_NSID] = { .type = NLA_S32 },
743 [NETNSA_PID] = { .type = NLA_U32 },
744 [NETNSA_FD] = { .type = NLA_U32 },
745 [NETNSA_TARGET_NSID] = { .type = NLA_S32 },
748 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
749 struct netlink_ext_ack *extack)
751 struct net *net = sock_net(skb->sk);
752 struct nlattr *tb[NETNSA_MAX + 1];
757 err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
758 NETNSA_MAX, rtnl_net_policy, extack);
761 if (!tb[NETNSA_NSID]) {
762 NL_SET_ERR_MSG(extack, "nsid is missing");
765 nsid = nla_get_s32(tb[NETNSA_NSID]);
767 if (tb[NETNSA_PID]) {
768 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
769 nla = tb[NETNSA_PID];
770 } else if (tb[NETNSA_FD]) {
771 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
774 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
778 NL_SET_BAD_ATTR(extack, nla);
779 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
780 return PTR_ERR(peer);
783 spin_lock_bh(&net->nsid_lock);
784 if (__peernet2id(net, peer) >= 0) {
785 spin_unlock_bh(&net->nsid_lock);
787 NL_SET_BAD_ATTR(extack, nla);
788 NL_SET_ERR_MSG(extack,
789 "Peer netns already has a nsid assigned");
793 err = alloc_netid(net, peer, nsid);
794 spin_unlock_bh(&net->nsid_lock);
796 rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid,
799 } else if (err == -ENOSPC && nsid >= 0) {
801 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
802 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
809 static int rtnl_net_get_size(void)
811 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
812 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
813 + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
817 struct net_fill_args {
827 static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
829 struct nlmsghdr *nlh;
830 struct rtgenmsg *rth;
832 nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
837 rth = nlmsg_data(nlh);
838 rth->rtgen_family = AF_UNSPEC;
840 if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
841 goto nla_put_failure;
844 nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
845 goto nla_put_failure;
851 nlmsg_cancel(skb, nlh);
855 static int rtnl_net_valid_getid_req(struct sk_buff *skb,
856 const struct nlmsghdr *nlh,
858 struct netlink_ext_ack *extack)
862 if (!netlink_strict_get_check(skb))
863 return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
864 tb, NETNSA_MAX, rtnl_net_policy,
867 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
868 NETNSA_MAX, rtnl_net_policy,
873 for (i = 0; i <= NETNSA_MAX; i++) {
881 case NETNSA_TARGET_NSID:
884 NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
892 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
893 struct netlink_ext_ack *extack)
895 struct net *net = sock_net(skb->sk);
896 struct nlattr *tb[NETNSA_MAX + 1];
897 struct net_fill_args fillargs = {
898 .portid = NETLINK_CB(skb).portid,
899 .seq = nlh->nlmsg_seq,
902 struct net *peer, *target = net;
907 err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
910 if (tb[NETNSA_PID]) {
911 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
912 nla = tb[NETNSA_PID];
913 } else if (tb[NETNSA_FD]) {
914 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
916 } else if (tb[NETNSA_NSID]) {
917 peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
919 peer = ERR_PTR(-ENOENT);
920 nla = tb[NETNSA_NSID];
922 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
927 NL_SET_BAD_ATTR(extack, nla);
928 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
929 return PTR_ERR(peer);
932 if (tb[NETNSA_TARGET_NSID]) {
933 int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
935 target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
936 if (IS_ERR(target)) {
937 NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
938 NL_SET_ERR_MSG(extack,
939 "Target netns reference is invalid");
940 err = PTR_ERR(target);
943 fillargs.add_ref = true;
944 fillargs.ref_nsid = peernet2id(net, peer);
947 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
953 fillargs.nsid = peernet2id(target, peer);
954 err = rtnl_net_fill(msg, &fillargs);
958 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
964 if (fillargs.add_ref)
970 struct rtnl_net_dump_cb {
974 struct net_fill_args fillargs;
979 /* Runs in RCU-critical section. */
980 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
982 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
985 if (net_cb->idx < net_cb->s_idx)
988 net_cb->fillargs.nsid = id;
989 if (net_cb->fillargs.add_ref)
990 net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
991 ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
1000 static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
1001 struct rtnl_net_dump_cb *net_cb,
1002 struct netlink_callback *cb)
1004 struct netlink_ext_ack *extack = cb->extack;
1005 struct nlattr *tb[NETNSA_MAX + 1];
1008 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
1009 NETNSA_MAX, rtnl_net_policy,
1014 for (i = 0; i <= NETNSA_MAX; i++) {
1018 if (i == NETNSA_TARGET_NSID) {
1021 net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
1023 NL_SET_BAD_ATTR(extack, tb[i]);
1024 NL_SET_ERR_MSG(extack,
1025 "Invalid target network namespace id");
1026 return PTR_ERR(net);
1028 net_cb->fillargs.add_ref = true;
1029 net_cb->ref_net = net_cb->tgt_net;
1030 net_cb->tgt_net = net;
1032 NL_SET_BAD_ATTR(extack, tb[i]);
1033 NL_SET_ERR_MSG(extack,
1034 "Unsupported attribute in dump request");
1042 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1044 struct rtnl_net_dump_cb net_cb = {
1045 .tgt_net = sock_net(skb->sk),
1048 .portid = NETLINK_CB(cb->skb).portid,
1049 .seq = cb->nlh->nlmsg_seq,
1050 .flags = NLM_F_MULTI,
1054 .s_idx = cb->args[0],
1058 if (cb->strict_check) {
1059 err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1065 idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1068 cb->args[0] = net_cb.idx;
1070 if (net_cb.fillargs.add_ref)
1071 put_net(net_cb.tgt_net);
1072 return err < 0 ? err : skb->len;
1075 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1076 struct nlmsghdr *nlh, gfp_t gfp)
1078 struct net_fill_args fillargs = {
1080 .seq = nlh ? nlh->nlmsg_seq : 0,
1084 struct sk_buff *msg;
1087 msg = nlmsg_new(rtnl_net_get_size(), gfp);
1091 err = rtnl_net_fill(msg, &fillargs);
1095 rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, gfp);
1101 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1104 static int __init net_ns_init(void)
1106 struct net_generic *ng;
1108 #ifdef CONFIG_NET_NS
1109 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1111 SLAB_PANIC|SLAB_ACCOUNT, NULL);
1113 /* Create workqueue for cleanup */
1114 netns_wq = create_singlethread_workqueue("netns");
1116 panic("Could not create netns workq");
1119 ng = net_alloc_generic();
1121 panic("Could not allocate generic netns");
1123 rcu_assign_pointer(init_net.gen, ng);
1125 down_write(&pernet_ops_rwsem);
1126 if (setup_net(&init_net, &init_user_ns))
1127 panic("Could not setup the initial network namespace");
1129 init_net_initialized = true;
1130 up_write(&pernet_ops_rwsem);
1132 if (register_pernet_subsys(&net_ns_ops))
1133 panic("Could not register network namespace subsystems");
1135 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
1136 RTNL_FLAG_DOIT_UNLOCKED);
1137 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
1138 RTNL_FLAG_DOIT_UNLOCKED);
1143 pure_initcall(net_ns_init);
1145 #ifdef CONFIG_NET_NS
1146 static int __register_pernet_operations(struct list_head *list,
1147 struct pernet_operations *ops)
1151 LIST_HEAD(net_exit_list);
1153 list_add_tail(&ops->list, list);
1154 if (ops->init || (ops->id && ops->size)) {
1155 /* We held write locked pernet_ops_rwsem, and parallel
1156 * setup_net() and cleanup_net() are not possible.
1159 error = ops_init(ops, net);
1162 list_add_tail(&net->exit_list, &net_exit_list);
1168 /* If I have an error cleanup all namespaces I initialized */
1169 list_del(&ops->list);
1170 ops_pre_exit_list(ops, &net_exit_list);
1172 ops_exit_list(ops, &net_exit_list);
1173 ops_free_list(ops, &net_exit_list);
1177 static void __unregister_pernet_operations(struct pernet_operations *ops)
1180 LIST_HEAD(net_exit_list);
1182 list_del(&ops->list);
1183 /* See comment in __register_pernet_operations() */
1185 list_add_tail(&net->exit_list, &net_exit_list);
1186 ops_pre_exit_list(ops, &net_exit_list);
1188 ops_exit_list(ops, &net_exit_list);
1189 ops_free_list(ops, &net_exit_list);
1194 static int __register_pernet_operations(struct list_head *list,
1195 struct pernet_operations *ops)
1197 if (!init_net_initialized) {
1198 list_add_tail(&ops->list, list);
1202 return ops_init(ops, &init_net);
1205 static void __unregister_pernet_operations(struct pernet_operations *ops)
1207 if (!init_net_initialized) {
1208 list_del(&ops->list);
1210 LIST_HEAD(net_exit_list);
1211 list_add(&init_net.exit_list, &net_exit_list);
1212 ops_pre_exit_list(ops, &net_exit_list);
1214 ops_exit_list(ops, &net_exit_list);
1215 ops_free_list(ops, &net_exit_list);
1219 #endif /* CONFIG_NET_NS */
1221 static DEFINE_IDA(net_generic_ids);
1223 static int register_pernet_operations(struct list_head *list,
1224 struct pernet_operations *ops)
1229 error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID,
1234 max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
1236 error = __register_pernet_operations(list, ops);
1240 ida_free(&net_generic_ids, *ops->id);
1246 static void unregister_pernet_operations(struct pernet_operations *ops)
1248 __unregister_pernet_operations(ops);
1251 ida_free(&net_generic_ids, *ops->id);
1255 * register_pernet_subsys - register a network namespace subsystem
1256 * @ops: pernet operations structure for the subsystem
1258 * Register a subsystem which has init and exit functions
1259 * that are called when network namespaces are created and
1260 * destroyed respectively.
1262 * When registered all network namespace init functions are
1263 * called for every existing network namespace. Allowing kernel
1264 * modules to have a race free view of the set of network namespaces.
1266 * When a new network namespace is created all of the init
1267 * methods are called in the order in which they were registered.
1269 * When a network namespace is destroyed all of the exit methods
1270 * are called in the reverse of the order with which they were
1273 int register_pernet_subsys(struct pernet_operations *ops)
1276 down_write(&pernet_ops_rwsem);
1277 error = register_pernet_operations(first_device, ops);
1278 up_write(&pernet_ops_rwsem);
1281 EXPORT_SYMBOL_GPL(register_pernet_subsys);
1284 * unregister_pernet_subsys - unregister a network namespace subsystem
1285 * @ops: pernet operations structure to manipulate
1287 * Remove the pernet operations structure from the list to be
1288 * used when network namespaces are created or destroyed. In
1289 * addition run the exit method for all existing network
1292 void unregister_pernet_subsys(struct pernet_operations *ops)
1294 down_write(&pernet_ops_rwsem);
1295 unregister_pernet_operations(ops);
1296 up_write(&pernet_ops_rwsem);
1298 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1301 * register_pernet_device - register a network namespace device
1302 * @ops: pernet operations structure for the subsystem
1304 * Register a device which has init and exit functions
1305 * that are called when network namespaces are created and
1306 * destroyed respectively.
1308 * When registered all network namespace init functions are
1309 * called for every existing network namespace. Allowing kernel
1310 * modules to have a race free view of the set of network namespaces.
1312 * When a new network namespace is created all of the init
1313 * methods are called in the order in which they were registered.
1315 * When a network namespace is destroyed all of the exit methods
1316 * are called in the reverse of the order with which they were
1319 int register_pernet_device(struct pernet_operations *ops)
1322 down_write(&pernet_ops_rwsem);
1323 error = register_pernet_operations(&pernet_list, ops);
1324 if (!error && (first_device == &pernet_list))
1325 first_device = &ops->list;
1326 up_write(&pernet_ops_rwsem);
1329 EXPORT_SYMBOL_GPL(register_pernet_device);
1332 * unregister_pernet_device - unregister a network namespace netdevice
1333 * @ops: pernet operations structure to manipulate
1335 * Remove the pernet operations structure from the list to be
1336 * used when network namespaces are created or destroyed. In
1337 * addition run the exit method for all existing network
1340 void unregister_pernet_device(struct pernet_operations *ops)
1342 down_write(&pernet_ops_rwsem);
1343 if (&ops->list == first_device)
1344 first_device = first_device->next;
1345 unregister_pernet_operations(ops);
1346 up_write(&pernet_ops_rwsem);
1348 EXPORT_SYMBOL_GPL(unregister_pernet_device);
1350 #ifdef CONFIG_NET_NS
1351 static struct ns_common *netns_get(struct task_struct *task)
1353 struct net *net = NULL;
1354 struct nsproxy *nsproxy;
1357 nsproxy = task->nsproxy;
1359 net = get_net(nsproxy->net_ns);
1362 return net ? &net->ns : NULL;
1365 static inline struct net *to_net_ns(struct ns_common *ns)
1367 return container_of(ns, struct net, ns);
1370 static void netns_put(struct ns_common *ns)
1372 put_net(to_net_ns(ns));
1375 static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
1377 struct net *net = to_net_ns(ns);
1379 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1380 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1383 put_net(nsproxy->net_ns);
1384 nsproxy->net_ns = get_net(net);
1388 static struct user_namespace *netns_owner(struct ns_common *ns)
1390 return to_net_ns(ns)->user_ns;
1393 const struct proc_ns_operations netns_operations = {
1395 .type = CLONE_NEWNET,
1398 .install = netns_install,
1399 .owner = netns_owner,