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) {
141 ng = rcu_dereference_protected(net->gen,
142 lockdep_is_held(&pernet_ops_rwsem));
143 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);
319 ref_tracker_dir_init(&net->notrefcnt_tracker, 128);
321 refcount_set(&net->passive, 1);
322 get_random_bytes(&net->hash_mix, sizeof(u32));
324 net->net_cookie = gen_cookie_next(&net_cookie);
326 net->dev_base_seq = 1;
327 net->user_ns = user_ns;
328 idr_init(&net->netns_ids);
329 spin_lock_init(&net->nsid_lock);
330 mutex_init(&net->ipv4.ra_mutex);
332 list_for_each_entry(ops, &pernet_list, list) {
333 error = ops_init(ops, net);
337 down_write(&net_rwsem);
338 list_add_tail_rcu(&net->list, &net_namespace_list);
339 up_write(&net_rwsem);
344 /* Walk through the list backwards calling the exit functions
345 * for the pernet modules whose init functions did not fail.
347 list_add(&net->exit_list, &net_exit_list);
349 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
350 ops_pre_exit_list(ops, &net_exit_list);
355 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
356 ops_exit_list(ops, &net_exit_list);
359 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
360 ops_free_list(ops, &net_exit_list);
366 static int __net_init net_defaults_init_net(struct net *net)
368 net->core.sysctl_somaxconn = SOMAXCONN;
369 net->core.sysctl_txrehash = SOCK_TXREHASH_ENABLED;
374 static struct pernet_operations net_defaults_ops = {
375 .init = net_defaults_init_net,
378 static __init int net_defaults_init(void)
380 if (register_pernet_subsys(&net_defaults_ops))
381 panic("Cannot initialize net default settings");
386 core_initcall(net_defaults_init);
389 static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
391 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
394 static void dec_net_namespaces(struct ucounts *ucounts)
396 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
399 static struct kmem_cache *net_cachep __ro_after_init;
400 static struct workqueue_struct *netns_wq;
402 static struct net *net_alloc(void)
404 struct net *net = NULL;
405 struct net_generic *ng;
407 ng = net_alloc_generic();
411 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
416 net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
417 if (!net->key_domain)
419 refcount_set(&net->key_domain->usage, 1);
422 rcu_assign_pointer(net->gen, ng);
428 kmem_cache_free(net_cachep, net);
436 static void net_free(struct net *net)
438 if (refcount_dec_and_test(&net->passive)) {
439 kfree(rcu_access_pointer(net->gen));
441 /* There should not be any trackers left there. */
442 ref_tracker_dir_exit(&net->notrefcnt_tracker);
444 kmem_cache_free(net_cachep, net);
448 void net_drop_ns(void *p)
450 struct net *net = (struct net *)p;
456 struct net *copy_net_ns(unsigned long flags,
457 struct user_namespace *user_ns, struct net *old_net)
459 struct ucounts *ucounts;
463 if (!(flags & CLONE_NEWNET))
464 return get_net(old_net);
466 ucounts = inc_net_namespaces(user_ns);
468 return ERR_PTR(-ENOSPC);
475 refcount_set(&net->passive, 1);
476 net->ucounts = ucounts;
477 get_user_ns(user_ns);
479 rv = down_read_killable(&pernet_ops_rwsem);
483 rv = setup_net(net, user_ns);
485 up_read(&pernet_ops_rwsem);
490 key_remove_domain(net->key_domain);
492 put_user_ns(user_ns);
495 dec_net_namespaces(ucounts);
502 * net_ns_get_ownership - get sysfs ownership data for @net
503 * @net: network namespace in question (can be NULL)
504 * @uid: kernel user ID for sysfs objects
505 * @gid: kernel group ID for sysfs objects
507 * Returns the uid/gid pair of root in the user namespace associated with the
508 * given network namespace.
510 void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
513 kuid_t ns_root_uid = make_kuid(net->user_ns, 0);
514 kgid_t ns_root_gid = make_kgid(net->user_ns, 0);
516 if (uid_valid(ns_root_uid))
519 if (gid_valid(ns_root_gid))
522 *uid = GLOBAL_ROOT_UID;
523 *gid = GLOBAL_ROOT_GID;
526 EXPORT_SYMBOL_GPL(net_ns_get_ownership);
528 static void unhash_nsid(struct net *net, struct net *last)
531 /* This function is only called from cleanup_net() work,
532 * and this work is the only process, that may delete
533 * a net from net_namespace_list. So, when the below
534 * is executing, the list may only grow. Thus, we do not
535 * use for_each_net_rcu() or net_rwsem.
540 spin_lock_bh(&tmp->nsid_lock);
541 id = __peernet2id(tmp, net);
543 idr_remove(&tmp->netns_ids, id);
544 spin_unlock_bh(&tmp->nsid_lock);
546 rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL,
551 spin_lock_bh(&net->nsid_lock);
552 idr_destroy(&net->netns_ids);
553 spin_unlock_bh(&net->nsid_lock);
556 static LLIST_HEAD(cleanup_list);
558 static void cleanup_net(struct work_struct *work)
560 const struct pernet_operations *ops;
561 struct net *net, *tmp, *last;
562 struct llist_node *net_kill_list;
563 LIST_HEAD(net_exit_list);
565 /* Atomically snapshot the list of namespaces to cleanup */
566 net_kill_list = llist_del_all(&cleanup_list);
568 down_read(&pernet_ops_rwsem);
570 /* Don't let anyone else find us. */
571 down_write(&net_rwsem);
572 llist_for_each_entry(net, net_kill_list, cleanup_list)
573 list_del_rcu(&net->list);
574 /* Cache last net. After we unlock rtnl, no one new net
575 * added to net_namespace_list can assign nsid pointer
576 * to a net from net_kill_list (see peernet2id_alloc()).
577 * So, we skip them in unhash_nsid().
579 * Note, that unhash_nsid() does not delete nsid links
580 * between net_kill_list's nets, as they've already
581 * deleted from net_namespace_list. But, this would be
582 * useless anyway, as netns_ids are destroyed there.
584 last = list_last_entry(&net_namespace_list, struct net, list);
585 up_write(&net_rwsem);
587 llist_for_each_entry(net, net_kill_list, cleanup_list) {
588 unhash_nsid(net, last);
589 list_add_tail(&net->exit_list, &net_exit_list);
592 /* Run all of the network namespace pre_exit methods */
593 list_for_each_entry_reverse(ops, &pernet_list, list)
594 ops_pre_exit_list(ops, &net_exit_list);
597 * Another CPU might be rcu-iterating the list, wait for it.
598 * This needs to be before calling the exit() notifiers, so
599 * the rcu_barrier() below isn't sufficient alone.
600 * Also the pre_exit() and exit() methods need this barrier.
604 /* Run all of the network namespace exit methods */
605 list_for_each_entry_reverse(ops, &pernet_list, list)
606 ops_exit_list(ops, &net_exit_list);
608 /* Free the net generic variables */
609 list_for_each_entry_reverse(ops, &pernet_list, list)
610 ops_free_list(ops, &net_exit_list);
612 up_read(&pernet_ops_rwsem);
614 /* Ensure there are no outstanding rcu callbacks using this
619 /* Finally it is safe to free my network namespace structure */
620 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
621 list_del_init(&net->exit_list);
622 dec_net_namespaces(net->ucounts);
624 key_remove_domain(net->key_domain);
626 put_user_ns(net->user_ns);
632 * net_ns_barrier - wait until concurrent net_cleanup_work is done
634 * cleanup_net runs from work queue and will first remove namespaces
635 * from the global list, then run net exit functions.
637 * Call this in module exit path to make sure that all netns
638 * ->exit ops have been invoked before the function is removed.
640 void net_ns_barrier(void)
642 down_write(&pernet_ops_rwsem);
643 up_write(&pernet_ops_rwsem);
645 EXPORT_SYMBOL(net_ns_barrier);
647 static DECLARE_WORK(net_cleanup_work, cleanup_net);
649 void __put_net(struct net *net)
651 ref_tracker_dir_exit(&net->refcnt_tracker);
652 /* Cleanup the network namespace in process context */
653 if (llist_add(&net->cleanup_list, &cleanup_list))
654 queue_work(netns_wq, &net_cleanup_work);
656 EXPORT_SYMBOL_GPL(__put_net);
659 * get_net_ns - increment the refcount of the network namespace
660 * @ns: common namespace (net)
662 * Returns the net's common namespace.
664 struct ns_common *get_net_ns(struct ns_common *ns)
666 return &get_net(container_of(ns, struct net, ns))->ns;
668 EXPORT_SYMBOL_GPL(get_net_ns);
670 struct net *get_net_ns_by_fd(int fd)
673 struct ns_common *ns;
676 file = proc_ns_fget(fd);
678 return ERR_CAST(file);
680 ns = get_proc_ns(file_inode(file));
681 if (ns->ops == &netns_operations)
682 net = get_net(container_of(ns, struct net, ns));
684 net = ERR_PTR(-EINVAL);
689 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
692 struct net *get_net_ns_by_pid(pid_t pid)
694 struct task_struct *tsk;
697 /* Lookup the network namespace */
698 net = ERR_PTR(-ESRCH);
700 tsk = find_task_by_vpid(pid);
702 struct nsproxy *nsproxy;
704 nsproxy = tsk->nsproxy;
706 net = get_net(nsproxy->net_ns);
712 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
714 static __net_init int net_ns_net_init(struct net *net)
717 net->ns.ops = &netns_operations;
719 return ns_alloc_inum(&net->ns);
722 static __net_exit void net_ns_net_exit(struct net *net)
724 ns_free_inum(&net->ns);
727 static struct pernet_operations __net_initdata net_ns_ops = {
728 .init = net_ns_net_init,
729 .exit = net_ns_net_exit,
732 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
733 [NETNSA_NONE] = { .type = NLA_UNSPEC },
734 [NETNSA_NSID] = { .type = NLA_S32 },
735 [NETNSA_PID] = { .type = NLA_U32 },
736 [NETNSA_FD] = { .type = NLA_U32 },
737 [NETNSA_TARGET_NSID] = { .type = NLA_S32 },
740 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
741 struct netlink_ext_ack *extack)
743 struct net *net = sock_net(skb->sk);
744 struct nlattr *tb[NETNSA_MAX + 1];
749 err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
750 NETNSA_MAX, rtnl_net_policy, extack);
753 if (!tb[NETNSA_NSID]) {
754 NL_SET_ERR_MSG(extack, "nsid is missing");
757 nsid = nla_get_s32(tb[NETNSA_NSID]);
759 if (tb[NETNSA_PID]) {
760 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
761 nla = tb[NETNSA_PID];
762 } else if (tb[NETNSA_FD]) {
763 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
766 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
770 NL_SET_BAD_ATTR(extack, nla);
771 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
772 return PTR_ERR(peer);
775 spin_lock_bh(&net->nsid_lock);
776 if (__peernet2id(net, peer) >= 0) {
777 spin_unlock_bh(&net->nsid_lock);
779 NL_SET_BAD_ATTR(extack, nla);
780 NL_SET_ERR_MSG(extack,
781 "Peer netns already has a nsid assigned");
785 err = alloc_netid(net, peer, nsid);
786 spin_unlock_bh(&net->nsid_lock);
788 rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid,
791 } else if (err == -ENOSPC && nsid >= 0) {
793 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
794 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
801 static int rtnl_net_get_size(void)
803 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
804 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
805 + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
809 struct net_fill_args {
819 static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
821 struct nlmsghdr *nlh;
822 struct rtgenmsg *rth;
824 nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
829 rth = nlmsg_data(nlh);
830 rth->rtgen_family = AF_UNSPEC;
832 if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
833 goto nla_put_failure;
836 nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
837 goto nla_put_failure;
843 nlmsg_cancel(skb, nlh);
847 static int rtnl_net_valid_getid_req(struct sk_buff *skb,
848 const struct nlmsghdr *nlh,
850 struct netlink_ext_ack *extack)
854 if (!netlink_strict_get_check(skb))
855 return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
856 tb, NETNSA_MAX, rtnl_net_policy,
859 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
860 NETNSA_MAX, rtnl_net_policy,
865 for (i = 0; i <= NETNSA_MAX; i++) {
873 case NETNSA_TARGET_NSID:
876 NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
884 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
885 struct netlink_ext_ack *extack)
887 struct net *net = sock_net(skb->sk);
888 struct nlattr *tb[NETNSA_MAX + 1];
889 struct net_fill_args fillargs = {
890 .portid = NETLINK_CB(skb).portid,
891 .seq = nlh->nlmsg_seq,
894 struct net *peer, *target = net;
899 err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
902 if (tb[NETNSA_PID]) {
903 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
904 nla = tb[NETNSA_PID];
905 } else if (tb[NETNSA_FD]) {
906 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
908 } else if (tb[NETNSA_NSID]) {
909 peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
911 peer = ERR_PTR(-ENOENT);
912 nla = tb[NETNSA_NSID];
914 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
919 NL_SET_BAD_ATTR(extack, nla);
920 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
921 return PTR_ERR(peer);
924 if (tb[NETNSA_TARGET_NSID]) {
925 int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
927 target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
928 if (IS_ERR(target)) {
929 NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
930 NL_SET_ERR_MSG(extack,
931 "Target netns reference is invalid");
932 err = PTR_ERR(target);
935 fillargs.add_ref = true;
936 fillargs.ref_nsid = peernet2id(net, peer);
939 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
945 fillargs.nsid = peernet2id(target, peer);
946 err = rtnl_net_fill(msg, &fillargs);
950 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
956 if (fillargs.add_ref)
962 struct rtnl_net_dump_cb {
966 struct net_fill_args fillargs;
971 /* Runs in RCU-critical section. */
972 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
974 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
977 if (net_cb->idx < net_cb->s_idx)
980 net_cb->fillargs.nsid = id;
981 if (net_cb->fillargs.add_ref)
982 net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
983 ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
992 static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
993 struct rtnl_net_dump_cb *net_cb,
994 struct netlink_callback *cb)
996 struct netlink_ext_ack *extack = cb->extack;
997 struct nlattr *tb[NETNSA_MAX + 1];
1000 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
1001 NETNSA_MAX, rtnl_net_policy,
1006 for (i = 0; i <= NETNSA_MAX; i++) {
1010 if (i == NETNSA_TARGET_NSID) {
1013 net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
1015 NL_SET_BAD_ATTR(extack, tb[i]);
1016 NL_SET_ERR_MSG(extack,
1017 "Invalid target network namespace id");
1018 return PTR_ERR(net);
1020 net_cb->fillargs.add_ref = true;
1021 net_cb->ref_net = net_cb->tgt_net;
1022 net_cb->tgt_net = net;
1024 NL_SET_BAD_ATTR(extack, tb[i]);
1025 NL_SET_ERR_MSG(extack,
1026 "Unsupported attribute in dump request");
1034 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1036 struct rtnl_net_dump_cb net_cb = {
1037 .tgt_net = sock_net(skb->sk),
1040 .portid = NETLINK_CB(cb->skb).portid,
1041 .seq = cb->nlh->nlmsg_seq,
1042 .flags = NLM_F_MULTI,
1046 .s_idx = cb->args[0],
1050 if (cb->strict_check) {
1051 err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1057 idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1060 cb->args[0] = net_cb.idx;
1062 if (net_cb.fillargs.add_ref)
1063 put_net(net_cb.tgt_net);
1064 return err < 0 ? err : skb->len;
1067 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1068 struct nlmsghdr *nlh, gfp_t gfp)
1070 struct net_fill_args fillargs = {
1072 .seq = nlh ? nlh->nlmsg_seq : 0,
1076 struct sk_buff *msg;
1079 msg = nlmsg_new(rtnl_net_get_size(), gfp);
1083 err = rtnl_net_fill(msg, &fillargs);
1087 rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, gfp);
1093 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1096 void __init net_ns_init(void)
1098 struct net_generic *ng;
1100 #ifdef CONFIG_NET_NS
1101 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1103 SLAB_PANIC|SLAB_ACCOUNT, NULL);
1105 /* Create workqueue for cleanup */
1106 netns_wq = create_singlethread_workqueue("netns");
1108 panic("Could not create netns workq");
1111 ng = net_alloc_generic();
1113 panic("Could not allocate generic netns");
1115 rcu_assign_pointer(init_net.gen, ng);
1118 init_net.key_domain = &init_net_key_domain;
1120 down_write(&pernet_ops_rwsem);
1121 if (setup_net(&init_net, &init_user_ns))
1122 panic("Could not setup the initial network namespace");
1124 init_net_initialized = true;
1125 up_write(&pernet_ops_rwsem);
1127 if (register_pernet_subsys(&net_ns_ops))
1128 panic("Could not register network namespace subsystems");
1130 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
1131 RTNL_FLAG_DOIT_UNLOCKED);
1132 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
1133 RTNL_FLAG_DOIT_UNLOCKED);
1136 static void free_exit_list(struct pernet_operations *ops, struct list_head *net_exit_list)
1138 ops_pre_exit_list(ops, net_exit_list);
1140 ops_exit_list(ops, net_exit_list);
1141 ops_free_list(ops, net_exit_list);
1144 #ifdef CONFIG_NET_NS
1145 static int __register_pernet_operations(struct list_head *list,
1146 struct pernet_operations *ops)
1150 LIST_HEAD(net_exit_list);
1152 list_add_tail(&ops->list, list);
1153 if (ops->init || (ops->id && ops->size)) {
1154 /* We held write locked pernet_ops_rwsem, and parallel
1155 * setup_net() and cleanup_net() are not possible.
1158 error = ops_init(ops, net);
1161 list_add_tail(&net->exit_list, &net_exit_list);
1167 /* If I have an error cleanup all namespaces I initialized */
1168 list_del(&ops->list);
1169 free_exit_list(ops, &net_exit_list);
1173 static void __unregister_pernet_operations(struct pernet_operations *ops)
1176 LIST_HEAD(net_exit_list);
1178 list_del(&ops->list);
1179 /* See comment in __register_pernet_operations() */
1181 list_add_tail(&net->exit_list, &net_exit_list);
1183 free_exit_list(ops, &net_exit_list);
1188 static int __register_pernet_operations(struct list_head *list,
1189 struct pernet_operations *ops)
1191 if (!init_net_initialized) {
1192 list_add_tail(&ops->list, list);
1196 return ops_init(ops, &init_net);
1199 static void __unregister_pernet_operations(struct pernet_operations *ops)
1201 if (!init_net_initialized) {
1202 list_del(&ops->list);
1204 LIST_HEAD(net_exit_list);
1205 list_add(&init_net.exit_list, &net_exit_list);
1206 free_exit_list(ops, &net_exit_list);
1210 #endif /* CONFIG_NET_NS */
1212 static DEFINE_IDA(net_generic_ids);
1214 static int register_pernet_operations(struct list_head *list,
1215 struct pernet_operations *ops)
1220 error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID,
1225 max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
1227 error = __register_pernet_operations(list, ops);
1231 ida_free(&net_generic_ids, *ops->id);
1237 static void unregister_pernet_operations(struct pernet_operations *ops)
1239 __unregister_pernet_operations(ops);
1242 ida_free(&net_generic_ids, *ops->id);
1246 * register_pernet_subsys - register a network namespace subsystem
1247 * @ops: pernet operations structure for the subsystem
1249 * Register a subsystem which has init and exit functions
1250 * that are called when network namespaces are created and
1251 * destroyed respectively.
1253 * When registered all network namespace init functions are
1254 * called for every existing network namespace. Allowing kernel
1255 * modules to have a race free view of the set of network namespaces.
1257 * When a new network namespace is created all of the init
1258 * methods are called in the order in which they were registered.
1260 * When a network namespace is destroyed all of the exit methods
1261 * are called in the reverse of the order with which they were
1264 int register_pernet_subsys(struct pernet_operations *ops)
1267 down_write(&pernet_ops_rwsem);
1268 error = register_pernet_operations(first_device, ops);
1269 up_write(&pernet_ops_rwsem);
1272 EXPORT_SYMBOL_GPL(register_pernet_subsys);
1275 * unregister_pernet_subsys - unregister a network namespace subsystem
1276 * @ops: pernet operations structure to manipulate
1278 * Remove the pernet operations structure from the list to be
1279 * used when network namespaces are created or destroyed. In
1280 * addition run the exit method for all existing network
1283 void unregister_pernet_subsys(struct pernet_operations *ops)
1285 down_write(&pernet_ops_rwsem);
1286 unregister_pernet_operations(ops);
1287 up_write(&pernet_ops_rwsem);
1289 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1292 * register_pernet_device - register a network namespace device
1293 * @ops: pernet operations structure for the subsystem
1295 * Register a device which has init and exit functions
1296 * that are called when network namespaces are created and
1297 * destroyed respectively.
1299 * When registered all network namespace init functions are
1300 * called for every existing network namespace. Allowing kernel
1301 * modules to have a race free view of the set of network namespaces.
1303 * When a new network namespace is created all of the init
1304 * methods are called in the order in which they were registered.
1306 * When a network namespace is destroyed all of the exit methods
1307 * are called in the reverse of the order with which they were
1310 int register_pernet_device(struct pernet_operations *ops)
1313 down_write(&pernet_ops_rwsem);
1314 error = register_pernet_operations(&pernet_list, ops);
1315 if (!error && (first_device == &pernet_list))
1316 first_device = &ops->list;
1317 up_write(&pernet_ops_rwsem);
1320 EXPORT_SYMBOL_GPL(register_pernet_device);
1323 * unregister_pernet_device - unregister a network namespace netdevice
1324 * @ops: pernet operations structure to manipulate
1326 * Remove the pernet operations structure from the list to be
1327 * used when network namespaces are created or destroyed. In
1328 * addition run the exit method for all existing network
1331 void unregister_pernet_device(struct pernet_operations *ops)
1333 down_write(&pernet_ops_rwsem);
1334 if (&ops->list == first_device)
1335 first_device = first_device->next;
1336 unregister_pernet_operations(ops);
1337 up_write(&pernet_ops_rwsem);
1339 EXPORT_SYMBOL_GPL(unregister_pernet_device);
1341 #ifdef CONFIG_NET_NS
1342 static struct ns_common *netns_get(struct task_struct *task)
1344 struct net *net = NULL;
1345 struct nsproxy *nsproxy;
1348 nsproxy = task->nsproxy;
1350 net = get_net(nsproxy->net_ns);
1353 return net ? &net->ns : NULL;
1356 static inline struct net *to_net_ns(struct ns_common *ns)
1358 return container_of(ns, struct net, ns);
1361 static void netns_put(struct ns_common *ns)
1363 put_net(to_net_ns(ns));
1366 static int netns_install(struct nsset *nsset, struct ns_common *ns)
1368 struct nsproxy *nsproxy = nsset->nsproxy;
1369 struct net *net = to_net_ns(ns);
1371 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1372 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
1375 put_net(nsproxy->net_ns);
1376 nsproxy->net_ns = get_net(net);
1380 static struct user_namespace *netns_owner(struct ns_common *ns)
1382 return to_net_ns(ns)->user_ns;
1385 const struct proc_ns_operations netns_operations = {
1387 .type = CLONE_NEWNET,
1390 .install = netns_install,
1391 .owner = netns_owner,