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);
308 * setup_net runs the initializers for the network namespace object.
310 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
312 /* Must be called with pernet_ops_rwsem held */
313 const struct pernet_operations *ops, *saved_ops;
315 LIST_HEAD(net_exit_list);
317 refcount_set(&net->ns.count, 1);
318 ref_tracker_dir_init(&net->refcnt_tracker, 128);
320 refcount_set(&net->passive, 1);
321 get_random_bytes(&net->hash_mix, sizeof(u32));
323 net->net_cookie = gen_cookie_next(&net_cookie);
325 net->dev_base_seq = 1;
326 net->user_ns = user_ns;
327 idr_init(&net->netns_ids);
328 spin_lock_init(&net->nsid_lock);
329 mutex_init(&net->ipv4.ra_mutex);
331 list_for_each_entry(ops, &pernet_list, list) {
332 error = ops_init(ops, net);
336 down_write(&net_rwsem);
337 list_add_tail_rcu(&net->list, &net_namespace_list);
338 up_write(&net_rwsem);
343 /* Walk through the list backwards calling the exit functions
344 * for the pernet modules whose init functions did not fail.
346 list_add(&net->exit_list, &net_exit_list);
348 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
349 ops_pre_exit_list(ops, &net_exit_list);
354 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
355 ops_exit_list(ops, &net_exit_list);
358 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
359 ops_free_list(ops, &net_exit_list);
365 static int __net_init net_defaults_init_net(struct net *net)
367 net->core.sysctl_somaxconn = SOMAXCONN;
368 net->core.sysctl_txrehash = SOCK_TXREHASH_ENABLED;
373 static struct pernet_operations net_defaults_ops = {
374 .init = net_defaults_init_net,
377 static __init int net_defaults_init(void)
379 if (register_pernet_subsys(&net_defaults_ops))
380 panic("Cannot initialize net default settings");
385 core_initcall(net_defaults_init);
388 static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
390 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
393 static void dec_net_namespaces(struct ucounts *ucounts)
395 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
398 static struct kmem_cache *net_cachep __ro_after_init;
399 static struct workqueue_struct *netns_wq;
401 static struct net *net_alloc(void)
403 struct net *net = NULL;
404 struct net_generic *ng;
406 ng = net_alloc_generic();
410 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
415 net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
416 if (!net->key_domain)
418 refcount_set(&net->key_domain->usage, 1);
421 rcu_assign_pointer(net->gen, ng);
427 kmem_cache_free(net_cachep, net);
435 static void net_free(struct net *net)
437 if (refcount_dec_and_test(&net->passive)) {
438 kfree(rcu_access_pointer(net->gen));
439 kmem_cache_free(net_cachep, net);
443 void net_drop_ns(void *p)
445 struct net *net = (struct net *)p;
451 struct net *copy_net_ns(unsigned long flags,
452 struct user_namespace *user_ns, struct net *old_net)
454 struct ucounts *ucounts;
458 if (!(flags & CLONE_NEWNET))
459 return get_net(old_net);
461 ucounts = inc_net_namespaces(user_ns);
463 return ERR_PTR(-ENOSPC);
470 refcount_set(&net->passive, 1);
471 net->ucounts = ucounts;
472 get_user_ns(user_ns);
474 rv = down_read_killable(&pernet_ops_rwsem);
478 rv = setup_net(net, user_ns);
480 up_read(&pernet_ops_rwsem);
485 key_remove_domain(net->key_domain);
487 put_user_ns(user_ns);
490 dec_net_namespaces(ucounts);
497 * net_ns_get_ownership - get sysfs ownership data for @net
498 * @net: network namespace in question (can be NULL)
499 * @uid: kernel user ID for sysfs objects
500 * @gid: kernel group ID for sysfs objects
502 * Returns the uid/gid pair of root in the user namespace associated with the
503 * given network namespace.
505 void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
508 kuid_t ns_root_uid = make_kuid(net->user_ns, 0);
509 kgid_t ns_root_gid = make_kgid(net->user_ns, 0);
511 if (uid_valid(ns_root_uid))
514 if (gid_valid(ns_root_gid))
517 *uid = GLOBAL_ROOT_UID;
518 *gid = GLOBAL_ROOT_GID;
521 EXPORT_SYMBOL_GPL(net_ns_get_ownership);
523 static void unhash_nsid(struct net *net, struct net *last)
526 /* This function is only called from cleanup_net() work,
527 * and this work is the only process, that may delete
528 * a net from net_namespace_list. So, when the below
529 * is executing, the list may only grow. Thus, we do not
530 * use for_each_net_rcu() or net_rwsem.
535 spin_lock_bh(&tmp->nsid_lock);
536 id = __peernet2id(tmp, net);
538 idr_remove(&tmp->netns_ids, id);
539 spin_unlock_bh(&tmp->nsid_lock);
541 rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL,
546 spin_lock_bh(&net->nsid_lock);
547 idr_destroy(&net->netns_ids);
548 spin_unlock_bh(&net->nsid_lock);
551 static LLIST_HEAD(cleanup_list);
553 static void cleanup_net(struct work_struct *work)
555 const struct pernet_operations *ops;
556 struct net *net, *tmp, *last;
557 struct llist_node *net_kill_list;
558 LIST_HEAD(net_exit_list);
560 /* Atomically snapshot the list of namespaces to cleanup */
561 net_kill_list = llist_del_all(&cleanup_list);
563 down_read(&pernet_ops_rwsem);
565 /* Don't let anyone else find us. */
566 down_write(&net_rwsem);
567 llist_for_each_entry(net, net_kill_list, cleanup_list)
568 list_del_rcu(&net->list);
569 /* Cache last net. After we unlock rtnl, no one new net
570 * added to net_namespace_list can assign nsid pointer
571 * to a net from net_kill_list (see peernet2id_alloc()).
572 * So, we skip them in unhash_nsid().
574 * Note, that unhash_nsid() does not delete nsid links
575 * between net_kill_list's nets, as they've already
576 * deleted from net_namespace_list. But, this would be
577 * useless anyway, as netns_ids are destroyed there.
579 last = list_last_entry(&net_namespace_list, struct net, list);
580 up_write(&net_rwsem);
582 llist_for_each_entry(net, net_kill_list, cleanup_list) {
583 unhash_nsid(net, last);
584 list_add_tail(&net->exit_list, &net_exit_list);
587 /* Run all of the network namespace pre_exit methods */
588 list_for_each_entry_reverse(ops, &pernet_list, list)
589 ops_pre_exit_list(ops, &net_exit_list);
592 * Another CPU might be rcu-iterating the list, wait for it.
593 * This needs to be before calling the exit() notifiers, so
594 * the rcu_barrier() below isn't sufficient alone.
595 * Also the pre_exit() and exit() methods need this barrier.
599 /* Run all of the network namespace exit methods */
600 list_for_each_entry_reverse(ops, &pernet_list, list)
601 ops_exit_list(ops, &net_exit_list);
603 /* Free the net generic variables */
604 list_for_each_entry_reverse(ops, &pernet_list, list)
605 ops_free_list(ops, &net_exit_list);
607 up_read(&pernet_ops_rwsem);
609 /* Ensure there are no outstanding rcu callbacks using this
614 /* Finally it is safe to free my network namespace structure */
615 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
616 list_del_init(&net->exit_list);
617 dec_net_namespaces(net->ucounts);
619 key_remove_domain(net->key_domain);
621 put_user_ns(net->user_ns);
627 * net_ns_barrier - wait until concurrent net_cleanup_work is done
629 * cleanup_net runs from work queue and will first remove namespaces
630 * from the global list, then run net exit functions.
632 * Call this in module exit path to make sure that all netns
633 * ->exit ops have been invoked before the function is removed.
635 void net_ns_barrier(void)
637 down_write(&pernet_ops_rwsem);
638 up_write(&pernet_ops_rwsem);
640 EXPORT_SYMBOL(net_ns_barrier);
642 static DECLARE_WORK(net_cleanup_work, cleanup_net);
644 void __put_net(struct net *net)
646 ref_tracker_dir_exit(&net->refcnt_tracker);
647 /* Cleanup the network namespace in process context */
648 if (llist_add(&net->cleanup_list, &cleanup_list))
649 queue_work(netns_wq, &net_cleanup_work);
651 EXPORT_SYMBOL_GPL(__put_net);
654 * get_net_ns - increment the refcount of the network namespace
655 * @ns: common namespace (net)
657 * Returns the net's common namespace.
659 struct ns_common *get_net_ns(struct ns_common *ns)
661 return &get_net(container_of(ns, struct net, ns))->ns;
663 EXPORT_SYMBOL_GPL(get_net_ns);
665 struct net *get_net_ns_by_fd(int fd)
668 struct ns_common *ns;
671 file = proc_ns_fget(fd);
673 return ERR_CAST(file);
675 ns = get_proc_ns(file_inode(file));
676 if (ns->ops == &netns_operations)
677 net = get_net(container_of(ns, struct net, ns));
679 net = ERR_PTR(-EINVAL);
684 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
687 struct net *get_net_ns_by_pid(pid_t pid)
689 struct task_struct *tsk;
692 /* Lookup the network namespace */
693 net = ERR_PTR(-ESRCH);
695 tsk = find_task_by_vpid(pid);
697 struct nsproxy *nsproxy;
699 nsproxy = tsk->nsproxy;
701 net = get_net(nsproxy->net_ns);
707 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
709 static __net_init int net_ns_net_init(struct net *net)
712 net->ns.ops = &netns_operations;
714 return ns_alloc_inum(&net->ns);
717 static __net_exit void net_ns_net_exit(struct net *net)
719 ns_free_inum(&net->ns);
722 static struct pernet_operations __net_initdata net_ns_ops = {
723 .init = net_ns_net_init,
724 .exit = net_ns_net_exit,
727 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
728 [NETNSA_NONE] = { .type = NLA_UNSPEC },
729 [NETNSA_NSID] = { .type = NLA_S32 },
730 [NETNSA_PID] = { .type = NLA_U32 },
731 [NETNSA_FD] = { .type = NLA_U32 },
732 [NETNSA_TARGET_NSID] = { .type = NLA_S32 },
735 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
736 struct netlink_ext_ack *extack)
738 struct net *net = sock_net(skb->sk);
739 struct nlattr *tb[NETNSA_MAX + 1];
744 err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
745 NETNSA_MAX, rtnl_net_policy, extack);
748 if (!tb[NETNSA_NSID]) {
749 NL_SET_ERR_MSG(extack, "nsid is missing");
752 nsid = nla_get_s32(tb[NETNSA_NSID]);
754 if (tb[NETNSA_PID]) {
755 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
756 nla = tb[NETNSA_PID];
757 } else if (tb[NETNSA_FD]) {
758 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
761 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
765 NL_SET_BAD_ATTR(extack, nla);
766 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
767 return PTR_ERR(peer);
770 spin_lock_bh(&net->nsid_lock);
771 if (__peernet2id(net, peer) >= 0) {
772 spin_unlock_bh(&net->nsid_lock);
774 NL_SET_BAD_ATTR(extack, nla);
775 NL_SET_ERR_MSG(extack,
776 "Peer netns already has a nsid assigned");
780 err = alloc_netid(net, peer, nsid);
781 spin_unlock_bh(&net->nsid_lock);
783 rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid,
786 } else if (err == -ENOSPC && nsid >= 0) {
788 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
789 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
796 static int rtnl_net_get_size(void)
798 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
799 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
800 + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
804 struct net_fill_args {
814 static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
816 struct nlmsghdr *nlh;
817 struct rtgenmsg *rth;
819 nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
824 rth = nlmsg_data(nlh);
825 rth->rtgen_family = AF_UNSPEC;
827 if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
828 goto nla_put_failure;
831 nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
832 goto nla_put_failure;
838 nlmsg_cancel(skb, nlh);
842 static int rtnl_net_valid_getid_req(struct sk_buff *skb,
843 const struct nlmsghdr *nlh,
845 struct netlink_ext_ack *extack)
849 if (!netlink_strict_get_check(skb))
850 return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
851 tb, NETNSA_MAX, rtnl_net_policy,
854 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
855 NETNSA_MAX, rtnl_net_policy,
860 for (i = 0; i <= NETNSA_MAX; i++) {
868 case NETNSA_TARGET_NSID:
871 NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
879 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
880 struct netlink_ext_ack *extack)
882 struct net *net = sock_net(skb->sk);
883 struct nlattr *tb[NETNSA_MAX + 1];
884 struct net_fill_args fillargs = {
885 .portid = NETLINK_CB(skb).portid,
886 .seq = nlh->nlmsg_seq,
889 struct net *peer, *target = net;
894 err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
897 if (tb[NETNSA_PID]) {
898 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
899 nla = tb[NETNSA_PID];
900 } else if (tb[NETNSA_FD]) {
901 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
903 } else if (tb[NETNSA_NSID]) {
904 peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
906 peer = ERR_PTR(-ENOENT);
907 nla = tb[NETNSA_NSID];
909 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
914 NL_SET_BAD_ATTR(extack, nla);
915 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
916 return PTR_ERR(peer);
919 if (tb[NETNSA_TARGET_NSID]) {
920 int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
922 target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
923 if (IS_ERR(target)) {
924 NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
925 NL_SET_ERR_MSG(extack,
926 "Target netns reference is invalid");
927 err = PTR_ERR(target);
930 fillargs.add_ref = true;
931 fillargs.ref_nsid = peernet2id(net, peer);
934 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
940 fillargs.nsid = peernet2id(target, peer);
941 err = rtnl_net_fill(msg, &fillargs);
945 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
951 if (fillargs.add_ref)
957 struct rtnl_net_dump_cb {
961 struct net_fill_args fillargs;
966 /* Runs in RCU-critical section. */
967 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
969 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
972 if (net_cb->idx < net_cb->s_idx)
975 net_cb->fillargs.nsid = id;
976 if (net_cb->fillargs.add_ref)
977 net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
978 ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
987 static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
988 struct rtnl_net_dump_cb *net_cb,
989 struct netlink_callback *cb)
991 struct netlink_ext_ack *extack = cb->extack;
992 struct nlattr *tb[NETNSA_MAX + 1];
995 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
996 NETNSA_MAX, rtnl_net_policy,
1001 for (i = 0; i <= NETNSA_MAX; i++) {
1005 if (i == NETNSA_TARGET_NSID) {
1008 net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
1010 NL_SET_BAD_ATTR(extack, tb[i]);
1011 NL_SET_ERR_MSG(extack,
1012 "Invalid target network namespace id");
1013 return PTR_ERR(net);
1015 net_cb->fillargs.add_ref = true;
1016 net_cb->ref_net = net_cb->tgt_net;
1017 net_cb->tgt_net = net;
1019 NL_SET_BAD_ATTR(extack, tb[i]);
1020 NL_SET_ERR_MSG(extack,
1021 "Unsupported attribute in dump request");
1029 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1031 struct rtnl_net_dump_cb net_cb = {
1032 .tgt_net = sock_net(skb->sk),
1035 .portid = NETLINK_CB(cb->skb).portid,
1036 .seq = cb->nlh->nlmsg_seq,
1037 .flags = NLM_F_MULTI,
1041 .s_idx = cb->args[0],
1045 if (cb->strict_check) {
1046 err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1052 idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1055 cb->args[0] = net_cb.idx;
1057 if (net_cb.fillargs.add_ref)
1058 put_net(net_cb.tgt_net);
1059 return err < 0 ? err : skb->len;
1062 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1063 struct nlmsghdr *nlh, gfp_t gfp)
1065 struct net_fill_args fillargs = {
1067 .seq = nlh ? nlh->nlmsg_seq : 0,
1071 struct sk_buff *msg;
1074 msg = nlmsg_new(rtnl_net_get_size(), gfp);
1078 err = rtnl_net_fill(msg, &fillargs);
1082 rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, gfp);
1088 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1091 void __init net_ns_init(void)
1093 struct net_generic *ng;
1095 #ifdef CONFIG_NET_NS
1096 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1098 SLAB_PANIC|SLAB_ACCOUNT, NULL);
1100 /* Create workqueue for cleanup */
1101 netns_wq = create_singlethread_workqueue("netns");
1103 panic("Could not create netns workq");
1106 ng = net_alloc_generic();
1108 panic("Could not allocate generic netns");
1110 rcu_assign_pointer(init_net.gen, ng);
1113 init_net.key_domain = &init_net_key_domain;
1115 down_write(&pernet_ops_rwsem);
1116 if (setup_net(&init_net, &init_user_ns))
1117 panic("Could not setup the initial network namespace");
1119 init_net_initialized = true;
1120 up_write(&pernet_ops_rwsem);
1122 if (register_pernet_subsys(&net_ns_ops))
1123 panic("Could not register network namespace subsystems");
1125 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
1126 RTNL_FLAG_DOIT_UNLOCKED);
1127 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
1128 RTNL_FLAG_DOIT_UNLOCKED);
1131 static void free_exit_list(struct pernet_operations *ops, struct list_head *net_exit_list)
1133 ops_pre_exit_list(ops, net_exit_list);
1135 ops_exit_list(ops, net_exit_list);
1136 ops_free_list(ops, net_exit_list);
1139 #ifdef CONFIG_NET_NS
1140 static int __register_pernet_operations(struct list_head *list,
1141 struct pernet_operations *ops)
1145 LIST_HEAD(net_exit_list);
1147 list_add_tail(&ops->list, list);
1148 if (ops->init || (ops->id && ops->size)) {
1149 /* We held write locked pernet_ops_rwsem, and parallel
1150 * setup_net() and cleanup_net() are not possible.
1153 error = ops_init(ops, net);
1156 list_add_tail(&net->exit_list, &net_exit_list);
1162 /* If I have an error cleanup all namespaces I initialized */
1163 list_del(&ops->list);
1164 free_exit_list(ops, &net_exit_list);
1168 static void __unregister_pernet_operations(struct pernet_operations *ops)
1171 LIST_HEAD(net_exit_list);
1173 list_del(&ops->list);
1174 /* See comment in __register_pernet_operations() */
1176 list_add_tail(&net->exit_list, &net_exit_list);
1178 free_exit_list(ops, &net_exit_list);
1183 static int __register_pernet_operations(struct list_head *list,
1184 struct pernet_operations *ops)
1186 if (!init_net_initialized) {
1187 list_add_tail(&ops->list, list);
1191 return ops_init(ops, &init_net);
1194 static void __unregister_pernet_operations(struct pernet_operations *ops)
1196 if (!init_net_initialized) {
1197 list_del(&ops->list);
1199 LIST_HEAD(net_exit_list);
1200 list_add(&init_net.exit_list, &net_exit_list);
1201 free_exit_list(ops, &net_exit_list);
1205 #endif /* CONFIG_NET_NS */
1207 static DEFINE_IDA(net_generic_ids);
1209 static int register_pernet_operations(struct list_head *list,
1210 struct pernet_operations *ops)
1215 error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID,
1220 max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
1222 error = __register_pernet_operations(list, ops);
1226 ida_free(&net_generic_ids, *ops->id);
1232 static void unregister_pernet_operations(struct pernet_operations *ops)
1234 __unregister_pernet_operations(ops);
1237 ida_free(&net_generic_ids, *ops->id);
1241 * register_pernet_subsys - register a network namespace subsystem
1242 * @ops: pernet operations structure for the subsystem
1244 * Register a subsystem which has init and exit functions
1245 * that are called when network namespaces are created and
1246 * destroyed respectively.
1248 * When registered all network namespace init functions are
1249 * called for every existing network namespace. Allowing kernel
1250 * modules to have a race free view of the set of network namespaces.
1252 * When a new network namespace is created all of the init
1253 * methods are called in the order in which they were registered.
1255 * When a network namespace is destroyed all of the exit methods
1256 * are called in the reverse of the order with which they were
1259 int register_pernet_subsys(struct pernet_operations *ops)
1262 down_write(&pernet_ops_rwsem);
1263 error = register_pernet_operations(first_device, ops);
1264 up_write(&pernet_ops_rwsem);
1267 EXPORT_SYMBOL_GPL(register_pernet_subsys);
1270 * unregister_pernet_subsys - unregister a network namespace subsystem
1271 * @ops: pernet operations structure to manipulate
1273 * Remove the pernet operations structure from the list to be
1274 * used when network namespaces are created or destroyed. In
1275 * addition run the exit method for all existing network
1278 void unregister_pernet_subsys(struct pernet_operations *ops)
1280 down_write(&pernet_ops_rwsem);
1281 unregister_pernet_operations(ops);
1282 up_write(&pernet_ops_rwsem);
1284 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1287 * register_pernet_device - register a network namespace device
1288 * @ops: pernet operations structure for the subsystem
1290 * Register a device which has init and exit functions
1291 * that are called when network namespaces are created and
1292 * destroyed respectively.
1294 * When registered all network namespace init functions are
1295 * called for every existing network namespace. Allowing kernel
1296 * modules to have a race free view of the set of network namespaces.
1298 * When a new network namespace is created all of the init
1299 * methods are called in the order in which they were registered.
1301 * When a network namespace is destroyed all of the exit methods
1302 * are called in the reverse of the order with which they were
1305 int register_pernet_device(struct pernet_operations *ops)
1308 down_write(&pernet_ops_rwsem);
1309 error = register_pernet_operations(&pernet_list, ops);
1310 if (!error && (first_device == &pernet_list))
1311 first_device = &ops->list;
1312 up_write(&pernet_ops_rwsem);
1315 EXPORT_SYMBOL_GPL(register_pernet_device);
1318 * unregister_pernet_device - unregister a network namespace netdevice
1319 * @ops: pernet operations structure to manipulate
1321 * Remove the pernet operations structure from the list to be
1322 * used when network namespaces are created or destroyed. In
1323 * addition run the exit method for all existing network
1326 void unregister_pernet_device(struct pernet_operations *ops)
1328 down_write(&pernet_ops_rwsem);
1329 if (&ops->list == first_device)
1330 first_device = first_device->next;
1331 unregister_pernet_operations(ops);
1332 up_write(&pernet_ops_rwsem);
1334 EXPORT_SYMBOL_GPL(unregister_pernet_device);
1336 #ifdef CONFIG_NET_NS
1337 static struct ns_common *netns_get(struct task_struct *task)
1339 struct net *net = NULL;
1340 struct nsproxy *nsproxy;
1343 nsproxy = task->nsproxy;
1345 net = get_net(nsproxy->net_ns);
1348 return net ? &net->ns : NULL;
1351 static inline struct net *to_net_ns(struct ns_common *ns)
1353 return container_of(ns, struct net, ns);
1356 static void netns_put(struct ns_common *ns)
1358 put_net(to_net_ns(ns));
1361 static int netns_install(struct nsset *nsset, struct ns_common *ns)
1363 struct nsproxy *nsproxy = nsset->nsproxy;
1364 struct net *net = to_net_ns(ns);
1366 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1367 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
1370 put_net(nsproxy->net_ns);
1371 nsproxy->net_ns = get_net(net);
1375 static struct user_namespace *netns_owner(struct ns_common *ns)
1377 return to_net_ns(ns)->user_ns;
1380 const struct proc_ns_operations netns_operations = {
1382 .type = CLONE_NEWNET,
1385 .install = netns_install,
1386 .owner = netns_owner,