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>
22 #include <linux/cookie.h>
25 #include <net/netlink.h>
26 #include <net/net_namespace.h>
27 #include <net/netns/generic.h>
30 * Our network namespace constructor/destructor lists
33 static LIST_HEAD(pernet_list);
34 static struct list_head *first_device = &pernet_list;
36 LIST_HEAD(net_namespace_list);
37 EXPORT_SYMBOL_GPL(net_namespace_list);
39 /* Protects net_namespace_list. Nests iside rtnl_lock() */
40 DECLARE_RWSEM(net_rwsem);
41 EXPORT_SYMBOL_GPL(net_rwsem);
44 static struct key_tag init_net_key_domain = { .usage = REFCOUNT_INIT(1) };
48 EXPORT_SYMBOL(init_net);
50 static bool init_net_initialized;
52 * pernet_ops_rwsem: protects: pernet_list, net_generic_ids,
53 * init_net_initialized and first_device pointer.
54 * This is internal net namespace object. Please, don't use it
57 DECLARE_RWSEM(pernet_ops_rwsem);
58 EXPORT_SYMBOL_GPL(pernet_ops_rwsem);
60 #define MIN_PERNET_OPS_ID \
61 ((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *))
63 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
65 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
67 DEFINE_COOKIE(net_cookie);
69 static struct net_generic *net_alloc_generic(void)
71 struct net_generic *ng;
72 unsigned int generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
74 ng = kzalloc(generic_size, GFP_KERNEL);
76 ng->s.len = max_gen_ptrs;
81 static int net_assign_generic(struct net *net, unsigned int id, void *data)
83 struct net_generic *ng, *old_ng;
85 BUG_ON(id < MIN_PERNET_OPS_ID);
87 old_ng = rcu_dereference_protected(net->gen,
88 lockdep_is_held(&pernet_ops_rwsem));
89 if (old_ng->s.len > id) {
90 old_ng->ptr[id] = data;
94 ng = net_alloc_generic();
99 * Some synchronisation notes:
101 * The net_generic explores the net->gen array inside rcu
102 * read section. Besides once set the net->gen->ptr[x]
103 * pointer never changes (see rules in netns/generic.h).
105 * That said, we simply duplicate this array and schedule
106 * the old copy for kfree after a grace period.
109 memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
110 (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
113 rcu_assign_pointer(net->gen, ng);
114 kfree_rcu(old_ng, s.rcu);
118 static int ops_init(const struct pernet_operations *ops, struct net *net)
120 struct net_generic *ng;
124 if (ops->id && ops->size) {
125 data = kzalloc(ops->size, GFP_KERNEL);
129 err = net_assign_generic(net, *ops->id, data);
135 err = ops->init(net);
139 if (ops->id && ops->size) {
140 ng = rcu_dereference_protected(net->gen,
141 lockdep_is_held(&pernet_ops_rwsem));
142 ng->ptr[*ops->id] = NULL;
152 static void ops_pre_exit_list(const struct pernet_operations *ops,
153 struct list_head *net_exit_list)
158 list_for_each_entry(net, net_exit_list, exit_list)
163 static void ops_exit_list(const struct pernet_operations *ops,
164 struct list_head *net_exit_list)
168 list_for_each_entry(net, net_exit_list, exit_list) {
174 ops->exit_batch(net_exit_list);
177 static void ops_free_list(const struct pernet_operations *ops,
178 struct list_head *net_exit_list)
181 if (ops->size && ops->id) {
182 list_for_each_entry(net, net_exit_list, exit_list)
183 kfree(net_generic(net, *ops->id));
187 /* should be called with nsid_lock held */
188 static int alloc_netid(struct net *net, struct net *peer, int reqid)
190 int min = 0, max = 0;
197 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
200 /* This function is used by idr_for_each(). If net is equal to peer, the
201 * function returns the id so that idr_for_each() stops. Because we cannot
202 * returns the id 0 (idr_for_each() will not stop), we return the magic value
203 * NET_ID_ZERO (-1) for it.
205 #define NET_ID_ZERO -1
206 static int net_eq_idr(int id, void *net, void *peer)
208 if (net_eq(net, peer))
209 return id ? : NET_ID_ZERO;
213 /* Must be called from RCU-critical section or with nsid_lock held */
214 static int __peernet2id(const struct net *net, struct net *peer)
216 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
218 /* Magic value for id 0. */
219 if (id == NET_ID_ZERO)
224 return NETNSA_NSID_NOT_ASSIGNED;
227 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
228 struct nlmsghdr *nlh, gfp_t gfp);
229 /* This function returns the id of a peer netns. If no id is assigned, one will
230 * be allocated and returned.
232 int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp)
236 if (refcount_read(&net->ns.count) == 0)
237 return NETNSA_NSID_NOT_ASSIGNED;
239 spin_lock_bh(&net->nsid_lock);
240 id = __peernet2id(net, peer);
242 spin_unlock_bh(&net->nsid_lock);
246 /* When peer is obtained from RCU lists, we may race with
247 * its cleanup. Check whether it's alive, and this guarantees
248 * we never hash a peer back to net->netns_ids, after it has
249 * just been idr_remove()'d from there in cleanup_net().
251 if (!maybe_get_net(peer)) {
252 spin_unlock_bh(&net->nsid_lock);
253 return NETNSA_NSID_NOT_ASSIGNED;
256 id = alloc_netid(net, peer, -1);
257 spin_unlock_bh(&net->nsid_lock);
261 return NETNSA_NSID_NOT_ASSIGNED;
263 rtnl_net_notifyid(net, RTM_NEWNSID, id, 0, NULL, gfp);
267 EXPORT_SYMBOL_GPL(peernet2id_alloc);
269 /* This function returns, if assigned, the id of a peer netns. */
270 int peernet2id(const struct net *net, struct net *peer)
275 id = __peernet2id(net, peer);
280 EXPORT_SYMBOL(peernet2id);
282 /* This function returns true is the peer netns has an id assigned into the
285 bool peernet_has_id(const struct net *net, struct net *peer)
287 return peernet2id(net, peer) >= 0;
290 struct net *get_net_ns_by_id(const struct net *net, int id)
298 peer = idr_find(&net->netns_ids, id);
300 peer = maybe_get_net(peer);
305 EXPORT_SYMBOL_GPL(get_net_ns_by_id);
307 /* init code that must occur even if setup_net() is not called. */
308 static __net_init void preinit_net(struct net *net)
310 ref_tracker_dir_init(&net->notrefcnt_tracker, 128);
314 * setup_net runs the initializers for the network namespace object.
316 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
318 /* Must be called with pernet_ops_rwsem held */
319 const struct pernet_operations *ops, *saved_ops;
321 LIST_HEAD(net_exit_list);
323 refcount_set(&net->ns.count, 1);
324 ref_tracker_dir_init(&net->refcnt_tracker, 128);
326 refcount_set(&net->passive, 1);
327 get_random_bytes(&net->hash_mix, sizeof(u32));
329 net->net_cookie = gen_cookie_next(&net_cookie);
331 net->dev_base_seq = 1;
332 net->user_ns = user_ns;
333 idr_init(&net->netns_ids);
334 spin_lock_init(&net->nsid_lock);
335 mutex_init(&net->ipv4.ra_mutex);
337 list_for_each_entry(ops, &pernet_list, list) {
338 error = ops_init(ops, net);
342 down_write(&net_rwsem);
343 list_add_tail_rcu(&net->list, &net_namespace_list);
344 up_write(&net_rwsem);
349 /* Walk through the list backwards calling the exit functions
350 * for the pernet modules whose init functions did not fail.
352 list_add(&net->exit_list, &net_exit_list);
354 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
355 ops_pre_exit_list(ops, &net_exit_list);
360 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
361 ops_exit_list(ops, &net_exit_list);
364 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
365 ops_free_list(ops, &net_exit_list);
371 static int __net_init net_defaults_init_net(struct net *net)
373 net->core.sysctl_somaxconn = SOMAXCONN;
374 net->core.sysctl_txrehash = SOCK_TXREHASH_ENABLED;
379 static struct pernet_operations net_defaults_ops = {
380 .init = net_defaults_init_net,
383 static __init int net_defaults_init(void)
385 if (register_pernet_subsys(&net_defaults_ops))
386 panic("Cannot initialize net default settings");
391 core_initcall(net_defaults_init);
394 static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
396 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
399 static void dec_net_namespaces(struct ucounts *ucounts)
401 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
404 static struct kmem_cache *net_cachep __ro_after_init;
405 static struct workqueue_struct *netns_wq;
407 static struct net *net_alloc(void)
409 struct net *net = NULL;
410 struct net_generic *ng;
412 ng = net_alloc_generic();
416 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
421 net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
422 if (!net->key_domain)
424 refcount_set(&net->key_domain->usage, 1);
427 rcu_assign_pointer(net->gen, ng);
433 kmem_cache_free(net_cachep, net);
441 static void net_free(struct net *net)
443 if (refcount_dec_and_test(&net->passive)) {
444 kfree(rcu_access_pointer(net->gen));
446 /* There should not be any trackers left there. */
447 ref_tracker_dir_exit(&net->notrefcnt_tracker);
449 kmem_cache_free(net_cachep, net);
453 void net_drop_ns(void *p)
455 struct net *net = (struct net *)p;
461 struct net *copy_net_ns(unsigned long flags,
462 struct user_namespace *user_ns, struct net *old_net)
464 struct ucounts *ucounts;
468 if (!(flags & CLONE_NEWNET))
469 return get_net(old_net);
471 ucounts = inc_net_namespaces(user_ns);
473 return ERR_PTR(-ENOSPC);
482 refcount_set(&net->passive, 1);
483 net->ucounts = ucounts;
484 get_user_ns(user_ns);
486 rv = down_read_killable(&pernet_ops_rwsem);
490 rv = setup_net(net, user_ns);
492 up_read(&pernet_ops_rwsem);
497 key_remove_domain(net->key_domain);
499 put_user_ns(user_ns);
502 dec_net_namespaces(ucounts);
509 * net_ns_get_ownership - get sysfs ownership data for @net
510 * @net: network namespace in question (can be NULL)
511 * @uid: kernel user ID for sysfs objects
512 * @gid: kernel group ID for sysfs objects
514 * Returns the uid/gid pair of root in the user namespace associated with the
515 * given network namespace.
517 void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
520 kuid_t ns_root_uid = make_kuid(net->user_ns, 0);
521 kgid_t ns_root_gid = make_kgid(net->user_ns, 0);
523 if (uid_valid(ns_root_uid))
526 if (gid_valid(ns_root_gid))
529 *uid = GLOBAL_ROOT_UID;
530 *gid = GLOBAL_ROOT_GID;
533 EXPORT_SYMBOL_GPL(net_ns_get_ownership);
535 static void unhash_nsid(struct net *net, struct net *last)
538 /* This function is only called from cleanup_net() work,
539 * and this work is the only process, that may delete
540 * a net from net_namespace_list. So, when the below
541 * is executing, the list may only grow. Thus, we do not
542 * use for_each_net_rcu() or net_rwsem.
547 spin_lock_bh(&tmp->nsid_lock);
548 id = __peernet2id(tmp, net);
550 idr_remove(&tmp->netns_ids, id);
551 spin_unlock_bh(&tmp->nsid_lock);
553 rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL,
558 spin_lock_bh(&net->nsid_lock);
559 idr_destroy(&net->netns_ids);
560 spin_unlock_bh(&net->nsid_lock);
563 static LLIST_HEAD(cleanup_list);
565 static void cleanup_net(struct work_struct *work)
567 const struct pernet_operations *ops;
568 struct net *net, *tmp, *last;
569 struct llist_node *net_kill_list;
570 LIST_HEAD(net_exit_list);
572 /* Atomically snapshot the list of namespaces to cleanup */
573 net_kill_list = llist_del_all(&cleanup_list);
575 down_read(&pernet_ops_rwsem);
577 /* Don't let anyone else find us. */
578 down_write(&net_rwsem);
579 llist_for_each_entry(net, net_kill_list, cleanup_list)
580 list_del_rcu(&net->list);
581 /* Cache last net. After we unlock rtnl, no one new net
582 * added to net_namespace_list can assign nsid pointer
583 * to a net from net_kill_list (see peernet2id_alloc()).
584 * So, we skip them in unhash_nsid().
586 * Note, that unhash_nsid() does not delete nsid links
587 * between net_kill_list's nets, as they've already
588 * deleted from net_namespace_list. But, this would be
589 * useless anyway, as netns_ids are destroyed there.
591 last = list_last_entry(&net_namespace_list, struct net, list);
592 up_write(&net_rwsem);
594 llist_for_each_entry(net, net_kill_list, cleanup_list) {
595 unhash_nsid(net, last);
596 list_add_tail(&net->exit_list, &net_exit_list);
599 /* Run all of the network namespace pre_exit methods */
600 list_for_each_entry_reverse(ops, &pernet_list, list)
601 ops_pre_exit_list(ops, &net_exit_list);
604 * Another CPU might be rcu-iterating the list, wait for it.
605 * This needs to be before calling the exit() notifiers, so
606 * the rcu_barrier() below isn't sufficient alone.
607 * Also the pre_exit() and exit() methods need this barrier.
611 /* Run all of the network namespace exit methods */
612 list_for_each_entry_reverse(ops, &pernet_list, list)
613 ops_exit_list(ops, &net_exit_list);
615 /* Free the net generic variables */
616 list_for_each_entry_reverse(ops, &pernet_list, list)
617 ops_free_list(ops, &net_exit_list);
619 up_read(&pernet_ops_rwsem);
621 /* Ensure there are no outstanding rcu callbacks using this
626 /* Finally it is safe to free my network namespace structure */
627 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
628 list_del_init(&net->exit_list);
629 dec_net_namespaces(net->ucounts);
631 key_remove_domain(net->key_domain);
633 put_user_ns(net->user_ns);
639 * net_ns_barrier - wait until concurrent net_cleanup_work is done
641 * cleanup_net runs from work queue and will first remove namespaces
642 * from the global list, then run net exit functions.
644 * Call this in module exit path to make sure that all netns
645 * ->exit ops have been invoked before the function is removed.
647 void net_ns_barrier(void)
649 down_write(&pernet_ops_rwsem);
650 up_write(&pernet_ops_rwsem);
652 EXPORT_SYMBOL(net_ns_barrier);
654 static DECLARE_WORK(net_cleanup_work, cleanup_net);
656 void __put_net(struct net *net)
658 ref_tracker_dir_exit(&net->refcnt_tracker);
659 /* Cleanup the network namespace in process context */
660 if (llist_add(&net->cleanup_list, &cleanup_list))
661 queue_work(netns_wq, &net_cleanup_work);
663 EXPORT_SYMBOL_GPL(__put_net);
666 * get_net_ns - increment the refcount of the network namespace
667 * @ns: common namespace (net)
669 * Returns the net's common namespace.
671 struct ns_common *get_net_ns(struct ns_common *ns)
673 return &get_net(container_of(ns, struct net, ns))->ns;
675 EXPORT_SYMBOL_GPL(get_net_ns);
677 struct net *get_net_ns_by_fd(int fd)
680 struct ns_common *ns;
683 file = proc_ns_fget(fd);
685 return ERR_CAST(file);
687 ns = get_proc_ns(file_inode(file));
688 if (ns->ops == &netns_operations)
689 net = get_net(container_of(ns, struct net, ns));
691 net = ERR_PTR(-EINVAL);
696 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
699 struct net *get_net_ns_by_pid(pid_t pid)
701 struct task_struct *tsk;
704 /* Lookup the network namespace */
705 net = ERR_PTR(-ESRCH);
707 tsk = find_task_by_vpid(pid);
709 struct nsproxy *nsproxy;
711 nsproxy = tsk->nsproxy;
713 net = get_net(nsproxy->net_ns);
719 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
721 static __net_init int net_ns_net_init(struct net *net)
724 net->ns.ops = &netns_operations;
726 return ns_alloc_inum(&net->ns);
729 static __net_exit void net_ns_net_exit(struct net *net)
731 ns_free_inum(&net->ns);
734 static struct pernet_operations __net_initdata net_ns_ops = {
735 .init = net_ns_net_init,
736 .exit = net_ns_net_exit,
739 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
740 [NETNSA_NONE] = { .type = NLA_UNSPEC },
741 [NETNSA_NSID] = { .type = NLA_S32 },
742 [NETNSA_PID] = { .type = NLA_U32 },
743 [NETNSA_FD] = { .type = NLA_U32 },
744 [NETNSA_TARGET_NSID] = { .type = NLA_S32 },
747 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
748 struct netlink_ext_ack *extack)
750 struct net *net = sock_net(skb->sk);
751 struct nlattr *tb[NETNSA_MAX + 1];
756 err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
757 NETNSA_MAX, rtnl_net_policy, extack);
760 if (!tb[NETNSA_NSID]) {
761 NL_SET_ERR_MSG(extack, "nsid is missing");
764 nsid = nla_get_s32(tb[NETNSA_NSID]);
766 if (tb[NETNSA_PID]) {
767 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
768 nla = tb[NETNSA_PID];
769 } else if (tb[NETNSA_FD]) {
770 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
773 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
777 NL_SET_BAD_ATTR(extack, nla);
778 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
779 return PTR_ERR(peer);
782 spin_lock_bh(&net->nsid_lock);
783 if (__peernet2id(net, peer) >= 0) {
784 spin_unlock_bh(&net->nsid_lock);
786 NL_SET_BAD_ATTR(extack, nla);
787 NL_SET_ERR_MSG(extack,
788 "Peer netns already has a nsid assigned");
792 err = alloc_netid(net, peer, nsid);
793 spin_unlock_bh(&net->nsid_lock);
795 rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid,
798 } else if (err == -ENOSPC && nsid >= 0) {
800 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
801 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
808 static int rtnl_net_get_size(void)
810 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
811 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
812 + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
816 struct net_fill_args {
826 static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
828 struct nlmsghdr *nlh;
829 struct rtgenmsg *rth;
831 nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
836 rth = nlmsg_data(nlh);
837 rth->rtgen_family = AF_UNSPEC;
839 if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
840 goto nla_put_failure;
843 nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
844 goto nla_put_failure;
850 nlmsg_cancel(skb, nlh);
854 static int rtnl_net_valid_getid_req(struct sk_buff *skb,
855 const struct nlmsghdr *nlh,
857 struct netlink_ext_ack *extack)
861 if (!netlink_strict_get_check(skb))
862 return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
863 tb, NETNSA_MAX, rtnl_net_policy,
866 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
867 NETNSA_MAX, rtnl_net_policy,
872 for (i = 0; i <= NETNSA_MAX; i++) {
880 case NETNSA_TARGET_NSID:
883 NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
891 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
892 struct netlink_ext_ack *extack)
894 struct net *net = sock_net(skb->sk);
895 struct nlattr *tb[NETNSA_MAX + 1];
896 struct net_fill_args fillargs = {
897 .portid = NETLINK_CB(skb).portid,
898 .seq = nlh->nlmsg_seq,
901 struct net *peer, *target = net;
906 err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
909 if (tb[NETNSA_PID]) {
910 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
911 nla = tb[NETNSA_PID];
912 } else if (tb[NETNSA_FD]) {
913 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
915 } else if (tb[NETNSA_NSID]) {
916 peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
918 peer = ERR_PTR(-ENOENT);
919 nla = tb[NETNSA_NSID];
921 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
926 NL_SET_BAD_ATTR(extack, nla);
927 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
928 return PTR_ERR(peer);
931 if (tb[NETNSA_TARGET_NSID]) {
932 int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
934 target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
935 if (IS_ERR(target)) {
936 NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
937 NL_SET_ERR_MSG(extack,
938 "Target netns reference is invalid");
939 err = PTR_ERR(target);
942 fillargs.add_ref = true;
943 fillargs.ref_nsid = peernet2id(net, peer);
946 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
952 fillargs.nsid = peernet2id(target, peer);
953 err = rtnl_net_fill(msg, &fillargs);
957 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
963 if (fillargs.add_ref)
969 struct rtnl_net_dump_cb {
973 struct net_fill_args fillargs;
978 /* Runs in RCU-critical section. */
979 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
981 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
984 if (net_cb->idx < net_cb->s_idx)
987 net_cb->fillargs.nsid = id;
988 if (net_cb->fillargs.add_ref)
989 net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
990 ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
999 static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
1000 struct rtnl_net_dump_cb *net_cb,
1001 struct netlink_callback *cb)
1003 struct netlink_ext_ack *extack = cb->extack;
1004 struct nlattr *tb[NETNSA_MAX + 1];
1007 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
1008 NETNSA_MAX, rtnl_net_policy,
1013 for (i = 0; i <= NETNSA_MAX; i++) {
1017 if (i == NETNSA_TARGET_NSID) {
1020 net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
1022 NL_SET_BAD_ATTR(extack, tb[i]);
1023 NL_SET_ERR_MSG(extack,
1024 "Invalid target network namespace id");
1025 return PTR_ERR(net);
1027 net_cb->fillargs.add_ref = true;
1028 net_cb->ref_net = net_cb->tgt_net;
1029 net_cb->tgt_net = net;
1031 NL_SET_BAD_ATTR(extack, tb[i]);
1032 NL_SET_ERR_MSG(extack,
1033 "Unsupported attribute in dump request");
1041 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1043 struct rtnl_net_dump_cb net_cb = {
1044 .tgt_net = sock_net(skb->sk),
1047 .portid = NETLINK_CB(cb->skb).portid,
1048 .seq = cb->nlh->nlmsg_seq,
1049 .flags = NLM_F_MULTI,
1053 .s_idx = cb->args[0],
1057 if (cb->strict_check) {
1058 err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1064 idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1067 cb->args[0] = net_cb.idx;
1069 if (net_cb.fillargs.add_ref)
1070 put_net(net_cb.tgt_net);
1071 return err < 0 ? err : skb->len;
1074 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1075 struct nlmsghdr *nlh, gfp_t gfp)
1077 struct net_fill_args fillargs = {
1079 .seq = nlh ? nlh->nlmsg_seq : 0,
1083 struct sk_buff *msg;
1086 msg = nlmsg_new(rtnl_net_get_size(), gfp);
1090 err = rtnl_net_fill(msg, &fillargs);
1094 rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, gfp);
1100 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1103 void __init net_ns_init(void)
1105 struct net_generic *ng;
1107 #ifdef CONFIG_NET_NS
1108 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1110 SLAB_PANIC|SLAB_ACCOUNT, NULL);
1112 /* Create workqueue for cleanup */
1113 netns_wq = create_singlethread_workqueue("netns");
1115 panic("Could not create netns workq");
1118 ng = net_alloc_generic();
1120 panic("Could not allocate generic netns");
1122 rcu_assign_pointer(init_net.gen, ng);
1125 init_net.key_domain = &init_net_key_domain;
1127 down_write(&pernet_ops_rwsem);
1128 preinit_net(&init_net);
1129 if (setup_net(&init_net, &init_user_ns))
1130 panic("Could not setup the initial network namespace");
1132 init_net_initialized = true;
1133 up_write(&pernet_ops_rwsem);
1135 if (register_pernet_subsys(&net_ns_ops))
1136 panic("Could not register network namespace subsystems");
1138 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
1139 RTNL_FLAG_DOIT_UNLOCKED);
1140 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
1141 RTNL_FLAG_DOIT_UNLOCKED);
1144 static void free_exit_list(struct pernet_operations *ops, struct list_head *net_exit_list)
1146 ops_pre_exit_list(ops, net_exit_list);
1148 ops_exit_list(ops, net_exit_list);
1149 ops_free_list(ops, net_exit_list);
1152 #ifdef CONFIG_NET_NS
1153 static int __register_pernet_operations(struct list_head *list,
1154 struct pernet_operations *ops)
1158 LIST_HEAD(net_exit_list);
1160 list_add_tail(&ops->list, list);
1161 if (ops->init || (ops->id && ops->size)) {
1162 /* We held write locked pernet_ops_rwsem, and parallel
1163 * setup_net() and cleanup_net() are not possible.
1166 error = ops_init(ops, net);
1169 list_add_tail(&net->exit_list, &net_exit_list);
1175 /* If I have an error cleanup all namespaces I initialized */
1176 list_del(&ops->list);
1177 free_exit_list(ops, &net_exit_list);
1181 static void __unregister_pernet_operations(struct pernet_operations *ops)
1184 LIST_HEAD(net_exit_list);
1186 list_del(&ops->list);
1187 /* See comment in __register_pernet_operations() */
1189 list_add_tail(&net->exit_list, &net_exit_list);
1191 free_exit_list(ops, &net_exit_list);
1196 static int __register_pernet_operations(struct list_head *list,
1197 struct pernet_operations *ops)
1199 if (!init_net_initialized) {
1200 list_add_tail(&ops->list, list);
1204 return ops_init(ops, &init_net);
1207 static void __unregister_pernet_operations(struct pernet_operations *ops)
1209 if (!init_net_initialized) {
1210 list_del(&ops->list);
1212 LIST_HEAD(net_exit_list);
1213 list_add(&init_net.exit_list, &net_exit_list);
1214 free_exit_list(ops, &net_exit_list);
1218 #endif /* CONFIG_NET_NS */
1220 static DEFINE_IDA(net_generic_ids);
1222 static int register_pernet_operations(struct list_head *list,
1223 struct pernet_operations *ops)
1228 error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID,
1233 max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
1235 error = __register_pernet_operations(list, ops);
1239 ida_free(&net_generic_ids, *ops->id);
1245 static void unregister_pernet_operations(struct pernet_operations *ops)
1247 __unregister_pernet_operations(ops);
1250 ida_free(&net_generic_ids, *ops->id);
1254 * register_pernet_subsys - register a network namespace subsystem
1255 * @ops: pernet operations structure for the subsystem
1257 * Register a subsystem which has init and exit functions
1258 * that are called when network namespaces are created and
1259 * destroyed respectively.
1261 * When registered all network namespace init functions are
1262 * called for every existing network namespace. Allowing kernel
1263 * modules to have a race free view of the set of network namespaces.
1265 * When a new network namespace is created all of the init
1266 * methods are called in the order in which they were registered.
1268 * When a network namespace is destroyed all of the exit methods
1269 * are called in the reverse of the order with which they were
1272 int register_pernet_subsys(struct pernet_operations *ops)
1275 down_write(&pernet_ops_rwsem);
1276 error = register_pernet_operations(first_device, ops);
1277 up_write(&pernet_ops_rwsem);
1280 EXPORT_SYMBOL_GPL(register_pernet_subsys);
1283 * unregister_pernet_subsys - unregister a network namespace subsystem
1284 * @ops: pernet operations structure to manipulate
1286 * Remove the pernet operations structure from the list to be
1287 * used when network namespaces are created or destroyed. In
1288 * addition run the exit method for all existing network
1291 void unregister_pernet_subsys(struct pernet_operations *ops)
1293 down_write(&pernet_ops_rwsem);
1294 unregister_pernet_operations(ops);
1295 up_write(&pernet_ops_rwsem);
1297 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1300 * register_pernet_device - register a network namespace device
1301 * @ops: pernet operations structure for the subsystem
1303 * Register a device which has init and exit functions
1304 * that are called when network namespaces are created and
1305 * destroyed respectively.
1307 * When registered all network namespace init functions are
1308 * called for every existing network namespace. Allowing kernel
1309 * modules to have a race free view of the set of network namespaces.
1311 * When a new network namespace is created all of the init
1312 * methods are called in the order in which they were registered.
1314 * When a network namespace is destroyed all of the exit methods
1315 * are called in the reverse of the order with which they were
1318 int register_pernet_device(struct pernet_operations *ops)
1321 down_write(&pernet_ops_rwsem);
1322 error = register_pernet_operations(&pernet_list, ops);
1323 if (!error && (first_device == &pernet_list))
1324 first_device = &ops->list;
1325 up_write(&pernet_ops_rwsem);
1328 EXPORT_SYMBOL_GPL(register_pernet_device);
1331 * unregister_pernet_device - unregister a network namespace netdevice
1332 * @ops: pernet operations structure to manipulate
1334 * Remove the pernet operations structure from the list to be
1335 * used when network namespaces are created or destroyed. In
1336 * addition run the exit method for all existing network
1339 void unregister_pernet_device(struct pernet_operations *ops)
1341 down_write(&pernet_ops_rwsem);
1342 if (&ops->list == first_device)
1343 first_device = first_device->next;
1344 unregister_pernet_operations(ops);
1345 up_write(&pernet_ops_rwsem);
1347 EXPORT_SYMBOL_GPL(unregister_pernet_device);
1349 #ifdef CONFIG_NET_NS
1350 static struct ns_common *netns_get(struct task_struct *task)
1352 struct net *net = NULL;
1353 struct nsproxy *nsproxy;
1356 nsproxy = task->nsproxy;
1358 net = get_net(nsproxy->net_ns);
1361 return net ? &net->ns : NULL;
1364 static inline struct net *to_net_ns(struct ns_common *ns)
1366 return container_of(ns, struct net, ns);
1369 static void netns_put(struct ns_common *ns)
1371 put_net(to_net_ns(ns));
1374 static int netns_install(struct nsset *nsset, struct ns_common *ns)
1376 struct nsproxy *nsproxy = nsset->nsproxy;
1377 struct net *net = to_net_ns(ns);
1379 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1380 !ns_capable(nsset->cred->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,