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>
23 #include <linux/proc_fs.h>
26 #include <net/netlink.h>
27 #include <net/net_namespace.h>
28 #include <net/netns/generic.h>
31 * Our network namespace constructor/destructor lists
34 static LIST_HEAD(pernet_list);
35 static struct list_head *first_device = &pernet_list;
37 LIST_HEAD(net_namespace_list);
38 EXPORT_SYMBOL_GPL(net_namespace_list);
40 /* Protects net_namespace_list. Nests iside rtnl_lock() */
41 DECLARE_RWSEM(net_rwsem);
42 EXPORT_SYMBOL_GPL(net_rwsem);
45 static struct key_tag init_net_key_domain = { .usage = REFCOUNT_INIT(1) };
49 EXPORT_SYMBOL(init_net);
51 static bool init_net_initialized;
53 * pernet_ops_rwsem: protects: pernet_list, net_generic_ids,
54 * init_net_initialized and first_device pointer.
55 * This is internal net namespace object. Please, don't use it
58 DECLARE_RWSEM(pernet_ops_rwsem);
59 EXPORT_SYMBOL_GPL(pernet_ops_rwsem);
61 #define MIN_PERNET_OPS_ID \
62 ((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *))
64 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
66 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
68 DEFINE_COOKIE(net_cookie);
70 static struct net_generic *net_alloc_generic(void)
72 unsigned int gen_ptrs = READ_ONCE(max_gen_ptrs);
73 unsigned int generic_size;
74 struct net_generic *ng;
76 generic_size = offsetof(struct net_generic, ptr[gen_ptrs]);
78 ng = kzalloc(generic_size, GFP_KERNEL);
85 static int net_assign_generic(struct net *net, unsigned int id, void *data)
87 struct net_generic *ng, *old_ng;
89 BUG_ON(id < MIN_PERNET_OPS_ID);
91 old_ng = rcu_dereference_protected(net->gen,
92 lockdep_is_held(&pernet_ops_rwsem));
93 if (old_ng->s.len > id) {
94 old_ng->ptr[id] = data;
98 ng = net_alloc_generic();
103 * Some synchronisation notes:
105 * The net_generic explores the net->gen array inside rcu
106 * read section. Besides once set the net->gen->ptr[x]
107 * pointer never changes (see rules in netns/generic.h).
109 * That said, we simply duplicate this array and schedule
110 * the old copy for kfree after a grace period.
113 memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
114 (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
117 rcu_assign_pointer(net->gen, ng);
118 kfree_rcu(old_ng, s.rcu);
122 static int ops_init(const struct pernet_operations *ops, struct net *net)
124 struct net_generic *ng;
128 if (ops->id && ops->size) {
129 data = kzalloc(ops->size, GFP_KERNEL);
133 err = net_assign_generic(net, *ops->id, data);
139 err = ops->init(net);
143 if (ops->id && ops->size) {
144 ng = rcu_dereference_protected(net->gen,
145 lockdep_is_held(&pernet_ops_rwsem));
146 ng->ptr[*ops->id] = NULL;
156 static void ops_pre_exit_list(const struct pernet_operations *ops,
157 struct list_head *net_exit_list)
162 list_for_each_entry(net, net_exit_list, exit_list)
167 static void ops_exit_list(const struct pernet_operations *ops,
168 struct list_head *net_exit_list)
172 list_for_each_entry(net, net_exit_list, exit_list) {
178 ops->exit_batch(net_exit_list);
181 static void ops_free_list(const struct pernet_operations *ops,
182 struct list_head *net_exit_list)
185 if (ops->size && ops->id) {
186 list_for_each_entry(net, net_exit_list, exit_list)
187 kfree(net_generic(net, *ops->id));
191 /* should be called with nsid_lock held */
192 static int alloc_netid(struct net *net, struct net *peer, int reqid)
194 int min = 0, max = 0;
201 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
204 /* This function is used by idr_for_each(). If net is equal to peer, the
205 * function returns the id so that idr_for_each() stops. Because we cannot
206 * returns the id 0 (idr_for_each() will not stop), we return the magic value
207 * NET_ID_ZERO (-1) for it.
209 #define NET_ID_ZERO -1
210 static int net_eq_idr(int id, void *net, void *peer)
212 if (net_eq(net, peer))
213 return id ? : NET_ID_ZERO;
217 /* Must be called from RCU-critical section or with nsid_lock held */
218 static int __peernet2id(const struct net *net, struct net *peer)
220 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
222 /* Magic value for id 0. */
223 if (id == NET_ID_ZERO)
228 return NETNSA_NSID_NOT_ASSIGNED;
231 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
232 struct nlmsghdr *nlh, gfp_t gfp);
233 /* This function returns the id of a peer netns. If no id is assigned, one will
234 * be allocated and returned.
236 int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp)
240 if (refcount_read(&net->ns.count) == 0)
241 return NETNSA_NSID_NOT_ASSIGNED;
243 spin_lock_bh(&net->nsid_lock);
244 id = __peernet2id(net, peer);
246 spin_unlock_bh(&net->nsid_lock);
250 /* When peer is obtained from RCU lists, we may race with
251 * its cleanup. Check whether it's alive, and this guarantees
252 * we never hash a peer back to net->netns_ids, after it has
253 * just been idr_remove()'d from there in cleanup_net().
255 if (!maybe_get_net(peer)) {
256 spin_unlock_bh(&net->nsid_lock);
257 return NETNSA_NSID_NOT_ASSIGNED;
260 id = alloc_netid(net, peer, -1);
261 spin_unlock_bh(&net->nsid_lock);
265 return NETNSA_NSID_NOT_ASSIGNED;
267 rtnl_net_notifyid(net, RTM_NEWNSID, id, 0, NULL, gfp);
271 EXPORT_SYMBOL_GPL(peernet2id_alloc);
273 /* This function returns, if assigned, the id of a peer netns. */
274 int peernet2id(const struct net *net, struct net *peer)
279 id = __peernet2id(net, peer);
284 EXPORT_SYMBOL(peernet2id);
286 /* This function returns true is the peer netns has an id assigned into the
289 bool peernet_has_id(const struct net *net, struct net *peer)
291 return peernet2id(net, peer) >= 0;
294 struct net *get_net_ns_by_id(const struct net *net, int id)
302 peer = idr_find(&net->netns_ids, id);
304 peer = maybe_get_net(peer);
309 EXPORT_SYMBOL_GPL(get_net_ns_by_id);
311 /* init code that must occur even if setup_net() is not called. */
312 static __net_init void preinit_net(struct net *net)
314 ref_tracker_dir_init(&net->notrefcnt_tracker, 128, "net notrefcnt");
318 * setup_net runs the initializers for the network namespace object.
320 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
322 /* Must be called with pernet_ops_rwsem held */
323 const struct pernet_operations *ops, *saved_ops;
324 LIST_HEAD(net_exit_list);
325 LIST_HEAD(dev_kill_list);
328 refcount_set(&net->ns.count, 1);
329 ref_tracker_dir_init(&net->refcnt_tracker, 128, "net refcnt");
331 refcount_set(&net->passive, 1);
332 get_random_bytes(&net->hash_mix, sizeof(u32));
334 net->net_cookie = gen_cookie_next(&net_cookie);
336 net->dev_base_seq = 1;
337 net->user_ns = user_ns;
338 idr_init(&net->netns_ids);
339 spin_lock_init(&net->nsid_lock);
340 mutex_init(&net->ipv4.ra_mutex);
342 list_for_each_entry(ops, &pernet_list, list) {
343 error = ops_init(ops, net);
347 down_write(&net_rwsem);
348 list_add_tail_rcu(&net->list, &net_namespace_list);
349 up_write(&net_rwsem);
354 /* Walk through the list backwards calling the exit functions
355 * for the pernet modules whose init functions did not fail.
357 list_add(&net->exit_list, &net_exit_list);
359 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
360 ops_pre_exit_list(ops, &net_exit_list);
366 list_for_each_entry_continue_reverse(ops, &pernet_list, list) {
367 if (ops->exit_batch_rtnl)
368 ops->exit_batch_rtnl(&net_exit_list, &dev_kill_list);
370 unregister_netdevice_many(&dev_kill_list);
374 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
375 ops_exit_list(ops, &net_exit_list);
378 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
379 ops_free_list(ops, &net_exit_list);
385 static int __net_init net_defaults_init_net(struct net *net)
387 net->core.sysctl_somaxconn = SOMAXCONN;
388 /* Limits per socket sk_omem_alloc usage.
389 * TCP zerocopy regular usage needs 128 KB.
391 net->core.sysctl_optmem_max = 128 * 1024;
392 net->core.sysctl_txrehash = SOCK_TXREHASH_ENABLED;
397 static struct pernet_operations net_defaults_ops = {
398 .init = net_defaults_init_net,
401 static __init int net_defaults_init(void)
403 if (register_pernet_subsys(&net_defaults_ops))
404 panic("Cannot initialize net default settings");
409 core_initcall(net_defaults_init);
412 static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
414 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
417 static void dec_net_namespaces(struct ucounts *ucounts)
419 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
422 static struct kmem_cache *net_cachep __ro_after_init;
423 static struct workqueue_struct *netns_wq;
425 static struct net *net_alloc(void)
427 struct net *net = NULL;
428 struct net_generic *ng;
430 ng = net_alloc_generic();
434 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
439 net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
440 if (!net->key_domain)
442 refcount_set(&net->key_domain->usage, 1);
445 rcu_assign_pointer(net->gen, ng);
451 kmem_cache_free(net_cachep, net);
459 static void net_free(struct net *net)
461 if (refcount_dec_and_test(&net->passive)) {
462 kfree(rcu_access_pointer(net->gen));
464 /* There should not be any trackers left there. */
465 ref_tracker_dir_exit(&net->notrefcnt_tracker);
467 kmem_cache_free(net_cachep, net);
471 void net_drop_ns(void *p)
473 struct net *net = (struct net *)p;
479 struct net *copy_net_ns(unsigned long flags,
480 struct user_namespace *user_ns, struct net *old_net)
482 struct ucounts *ucounts;
486 if (!(flags & CLONE_NEWNET))
487 return get_net(old_net);
489 ucounts = inc_net_namespaces(user_ns);
491 return ERR_PTR(-ENOSPC);
500 refcount_set(&net->passive, 1);
501 net->ucounts = ucounts;
502 get_user_ns(user_ns);
504 rv = down_read_killable(&pernet_ops_rwsem);
508 rv = setup_net(net, user_ns);
510 up_read(&pernet_ops_rwsem);
515 key_remove_domain(net->key_domain);
517 put_user_ns(user_ns);
520 dec_net_namespaces(ucounts);
527 * net_ns_get_ownership - get sysfs ownership data for @net
528 * @net: network namespace in question (can be NULL)
529 * @uid: kernel user ID for sysfs objects
530 * @gid: kernel group ID for sysfs objects
532 * Returns the uid/gid pair of root in the user namespace associated with the
533 * given network namespace.
535 void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
538 kuid_t ns_root_uid = make_kuid(net->user_ns, 0);
539 kgid_t ns_root_gid = make_kgid(net->user_ns, 0);
541 if (uid_valid(ns_root_uid))
544 if (gid_valid(ns_root_gid))
547 *uid = GLOBAL_ROOT_UID;
548 *gid = GLOBAL_ROOT_GID;
551 EXPORT_SYMBOL_GPL(net_ns_get_ownership);
553 static void unhash_nsid(struct net *net, struct net *last)
556 /* This function is only called from cleanup_net() work,
557 * and this work is the only process, that may delete
558 * a net from net_namespace_list. So, when the below
559 * is executing, the list may only grow. Thus, we do not
560 * use for_each_net_rcu() or net_rwsem.
565 spin_lock_bh(&tmp->nsid_lock);
566 id = __peernet2id(tmp, net);
568 idr_remove(&tmp->netns_ids, id);
569 spin_unlock_bh(&tmp->nsid_lock);
571 rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL,
576 spin_lock_bh(&net->nsid_lock);
577 idr_destroy(&net->netns_ids);
578 spin_unlock_bh(&net->nsid_lock);
581 static LLIST_HEAD(cleanup_list);
583 static void cleanup_net(struct work_struct *work)
585 const struct pernet_operations *ops;
586 struct net *net, *tmp, *last;
587 struct llist_node *net_kill_list;
588 LIST_HEAD(net_exit_list);
589 LIST_HEAD(dev_kill_list);
591 /* Atomically snapshot the list of namespaces to cleanup */
592 net_kill_list = llist_del_all(&cleanup_list);
594 down_read(&pernet_ops_rwsem);
596 /* Don't let anyone else find us. */
597 down_write(&net_rwsem);
598 llist_for_each_entry(net, net_kill_list, cleanup_list)
599 list_del_rcu(&net->list);
600 /* Cache last net. After we unlock rtnl, no one new net
601 * added to net_namespace_list can assign nsid pointer
602 * to a net from net_kill_list (see peernet2id_alloc()).
603 * So, we skip them in unhash_nsid().
605 * Note, that unhash_nsid() does not delete nsid links
606 * between net_kill_list's nets, as they've already
607 * deleted from net_namespace_list. But, this would be
608 * useless anyway, as netns_ids are destroyed there.
610 last = list_last_entry(&net_namespace_list, struct net, list);
611 up_write(&net_rwsem);
613 llist_for_each_entry(net, net_kill_list, cleanup_list) {
614 unhash_nsid(net, last);
615 list_add_tail(&net->exit_list, &net_exit_list);
618 /* Run all of the network namespace pre_exit methods */
619 list_for_each_entry_reverse(ops, &pernet_list, list)
620 ops_pre_exit_list(ops, &net_exit_list);
623 * Another CPU might be rcu-iterating the list, wait for it.
624 * This needs to be before calling the exit() notifiers, so
625 * the rcu_barrier() below isn't sufficient alone.
626 * Also the pre_exit() and exit() methods need this barrier.
628 synchronize_rcu_expedited();
631 list_for_each_entry_reverse(ops, &pernet_list, list) {
632 if (ops->exit_batch_rtnl)
633 ops->exit_batch_rtnl(&net_exit_list, &dev_kill_list);
635 unregister_netdevice_many(&dev_kill_list);
638 /* Run all of the network namespace exit methods */
639 list_for_each_entry_reverse(ops, &pernet_list, list)
640 ops_exit_list(ops, &net_exit_list);
642 /* Free the net generic variables */
643 list_for_each_entry_reverse(ops, &pernet_list, list)
644 ops_free_list(ops, &net_exit_list);
646 up_read(&pernet_ops_rwsem);
648 /* Ensure there are no outstanding rcu callbacks using this
653 /* Finally it is safe to free my network namespace structure */
654 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
655 list_del_init(&net->exit_list);
656 dec_net_namespaces(net->ucounts);
658 key_remove_domain(net->key_domain);
660 put_user_ns(net->user_ns);
666 * net_ns_barrier - wait until concurrent net_cleanup_work is done
668 * cleanup_net runs from work queue and will first remove namespaces
669 * from the global list, then run net exit functions.
671 * Call this in module exit path to make sure that all netns
672 * ->exit ops have been invoked before the function is removed.
674 void net_ns_barrier(void)
676 down_write(&pernet_ops_rwsem);
677 up_write(&pernet_ops_rwsem);
679 EXPORT_SYMBOL(net_ns_barrier);
681 static DECLARE_WORK(net_cleanup_work, cleanup_net);
683 void __put_net(struct net *net)
685 ref_tracker_dir_exit(&net->refcnt_tracker);
686 /* Cleanup the network namespace in process context */
687 if (llist_add(&net->cleanup_list, &cleanup_list))
688 queue_work(netns_wq, &net_cleanup_work);
690 EXPORT_SYMBOL_GPL(__put_net);
693 * get_net_ns - increment the refcount of the network namespace
694 * @ns: common namespace (net)
696 * Returns the net's common namespace or ERR_PTR() if ref is zero.
698 struct ns_common *get_net_ns(struct ns_common *ns)
702 net = maybe_get_net(container_of(ns, struct net, ns));
705 return ERR_PTR(-EINVAL);
707 EXPORT_SYMBOL_GPL(get_net_ns);
709 struct net *get_net_ns_by_fd(int fd)
711 struct fd f = fdget(fd);
712 struct net *net = ERR_PTR(-EINVAL);
715 return ERR_PTR(-EBADF);
717 if (proc_ns_file(f.file)) {
718 struct ns_common *ns = get_proc_ns(file_inode(f.file));
719 if (ns->ops == &netns_operations)
720 net = get_net(container_of(ns, struct net, ns));
726 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
729 struct net *get_net_ns_by_pid(pid_t pid)
731 struct task_struct *tsk;
734 /* Lookup the network namespace */
735 net = ERR_PTR(-ESRCH);
737 tsk = find_task_by_vpid(pid);
739 struct nsproxy *nsproxy;
741 nsproxy = tsk->nsproxy;
743 net = get_net(nsproxy->net_ns);
749 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
751 static __net_init int net_ns_net_init(struct net *net)
754 net->ns.ops = &netns_operations;
756 return ns_alloc_inum(&net->ns);
759 static __net_exit void net_ns_net_exit(struct net *net)
761 ns_free_inum(&net->ns);
764 static struct pernet_operations __net_initdata net_ns_ops = {
765 .init = net_ns_net_init,
766 .exit = net_ns_net_exit,
769 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
770 [NETNSA_NONE] = { .type = NLA_UNSPEC },
771 [NETNSA_NSID] = { .type = NLA_S32 },
772 [NETNSA_PID] = { .type = NLA_U32 },
773 [NETNSA_FD] = { .type = NLA_U32 },
774 [NETNSA_TARGET_NSID] = { .type = NLA_S32 },
777 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
778 struct netlink_ext_ack *extack)
780 struct net *net = sock_net(skb->sk);
781 struct nlattr *tb[NETNSA_MAX + 1];
786 err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
787 NETNSA_MAX, rtnl_net_policy, extack);
790 if (!tb[NETNSA_NSID]) {
791 NL_SET_ERR_MSG(extack, "nsid is missing");
794 nsid = nla_get_s32(tb[NETNSA_NSID]);
796 if (tb[NETNSA_PID]) {
797 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
798 nla = tb[NETNSA_PID];
799 } else if (tb[NETNSA_FD]) {
800 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
803 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
807 NL_SET_BAD_ATTR(extack, nla);
808 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
809 return PTR_ERR(peer);
812 spin_lock_bh(&net->nsid_lock);
813 if (__peernet2id(net, peer) >= 0) {
814 spin_unlock_bh(&net->nsid_lock);
816 NL_SET_BAD_ATTR(extack, nla);
817 NL_SET_ERR_MSG(extack,
818 "Peer netns already has a nsid assigned");
822 err = alloc_netid(net, peer, nsid);
823 spin_unlock_bh(&net->nsid_lock);
825 rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid,
828 } else if (err == -ENOSPC && nsid >= 0) {
830 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
831 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
838 static int rtnl_net_get_size(void)
840 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
841 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
842 + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
846 struct net_fill_args {
856 static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
858 struct nlmsghdr *nlh;
859 struct rtgenmsg *rth;
861 nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
866 rth = nlmsg_data(nlh);
867 rth->rtgen_family = AF_UNSPEC;
869 if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
870 goto nla_put_failure;
873 nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
874 goto nla_put_failure;
880 nlmsg_cancel(skb, nlh);
884 static int rtnl_net_valid_getid_req(struct sk_buff *skb,
885 const struct nlmsghdr *nlh,
887 struct netlink_ext_ack *extack)
891 if (!netlink_strict_get_check(skb))
892 return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
893 tb, NETNSA_MAX, rtnl_net_policy,
896 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
897 NETNSA_MAX, rtnl_net_policy,
902 for (i = 0; i <= NETNSA_MAX; i++) {
910 case NETNSA_TARGET_NSID:
913 NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
921 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
922 struct netlink_ext_ack *extack)
924 struct net *net = sock_net(skb->sk);
925 struct nlattr *tb[NETNSA_MAX + 1];
926 struct net_fill_args fillargs = {
927 .portid = NETLINK_CB(skb).portid,
928 .seq = nlh->nlmsg_seq,
931 struct net *peer, *target = net;
936 err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
939 if (tb[NETNSA_PID]) {
940 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
941 nla = tb[NETNSA_PID];
942 } else if (tb[NETNSA_FD]) {
943 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
945 } else if (tb[NETNSA_NSID]) {
946 peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
948 peer = ERR_PTR(-ENOENT);
949 nla = tb[NETNSA_NSID];
951 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
956 NL_SET_BAD_ATTR(extack, nla);
957 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
958 return PTR_ERR(peer);
961 if (tb[NETNSA_TARGET_NSID]) {
962 int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
964 target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
965 if (IS_ERR(target)) {
966 NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
967 NL_SET_ERR_MSG(extack,
968 "Target netns reference is invalid");
969 err = PTR_ERR(target);
972 fillargs.add_ref = true;
973 fillargs.ref_nsid = peernet2id(net, peer);
976 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
982 fillargs.nsid = peernet2id(target, peer);
983 err = rtnl_net_fill(msg, &fillargs);
987 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
993 if (fillargs.add_ref)
999 struct rtnl_net_dump_cb {
1000 struct net *tgt_net;
1001 struct net *ref_net;
1002 struct sk_buff *skb;
1003 struct net_fill_args fillargs;
1008 /* Runs in RCU-critical section. */
1009 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
1011 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
1014 if (net_cb->idx < net_cb->s_idx)
1017 net_cb->fillargs.nsid = id;
1018 if (net_cb->fillargs.add_ref)
1019 net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
1020 ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
1029 static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
1030 struct rtnl_net_dump_cb *net_cb,
1031 struct netlink_callback *cb)
1033 struct netlink_ext_ack *extack = cb->extack;
1034 struct nlattr *tb[NETNSA_MAX + 1];
1037 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
1038 NETNSA_MAX, rtnl_net_policy,
1043 for (i = 0; i <= NETNSA_MAX; i++) {
1047 if (i == NETNSA_TARGET_NSID) {
1050 net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
1052 NL_SET_BAD_ATTR(extack, tb[i]);
1053 NL_SET_ERR_MSG(extack,
1054 "Invalid target network namespace id");
1055 return PTR_ERR(net);
1057 net_cb->fillargs.add_ref = true;
1058 net_cb->ref_net = net_cb->tgt_net;
1059 net_cb->tgt_net = net;
1061 NL_SET_BAD_ATTR(extack, tb[i]);
1062 NL_SET_ERR_MSG(extack,
1063 "Unsupported attribute in dump request");
1071 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1073 struct rtnl_net_dump_cb net_cb = {
1074 .tgt_net = sock_net(skb->sk),
1077 .portid = NETLINK_CB(cb->skb).portid,
1078 .seq = cb->nlh->nlmsg_seq,
1079 .flags = NLM_F_MULTI,
1083 .s_idx = cb->args[0],
1087 if (cb->strict_check) {
1088 err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1094 idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1097 cb->args[0] = net_cb.idx;
1099 if (net_cb.fillargs.add_ref)
1100 put_net(net_cb.tgt_net);
1104 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1105 struct nlmsghdr *nlh, gfp_t gfp)
1107 struct net_fill_args fillargs = {
1109 .seq = nlh ? nlh->nlmsg_seq : 0,
1113 struct sk_buff *msg;
1116 msg = nlmsg_new(rtnl_net_get_size(), gfp);
1120 err = rtnl_net_fill(msg, &fillargs);
1124 rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, gfp);
1130 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1133 #ifdef CONFIG_NET_NS
1134 static void __init netns_ipv4_struct_check(void)
1136 /* TX readonly hotpath cache lines */
1137 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1138 sysctl_tcp_early_retrans);
1139 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1140 sysctl_tcp_tso_win_divisor);
1141 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1142 sysctl_tcp_tso_rtt_log);
1143 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1144 sysctl_tcp_autocorking);
1145 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1146 sysctl_tcp_min_snd_mss);
1147 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1148 sysctl_tcp_notsent_lowat);
1149 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1150 sysctl_tcp_limit_output_bytes);
1151 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1152 sysctl_tcp_min_rtt_wlen);
1153 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1155 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1156 sysctl_ip_fwd_use_pmtu);
1157 CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_tx, 33);
1159 /* TXRX readonly hotpath cache lines */
1160 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_txrx,
1161 sysctl_tcp_moderate_rcvbuf);
1162 CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_txrx, 1);
1164 /* RX readonly hotpath cache line */
1165 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1166 sysctl_ip_early_demux);
1167 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1168 sysctl_tcp_early_demux);
1169 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1170 sysctl_tcp_reordering);
1171 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1173 CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_rx, 18);
1177 void __init net_ns_init(void)
1179 struct net_generic *ng;
1181 #ifdef CONFIG_NET_NS
1182 netns_ipv4_struct_check();
1183 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1185 SLAB_PANIC|SLAB_ACCOUNT, NULL);
1187 /* Create workqueue for cleanup */
1188 netns_wq = create_singlethread_workqueue("netns");
1190 panic("Could not create netns workq");
1193 ng = net_alloc_generic();
1195 panic("Could not allocate generic netns");
1197 rcu_assign_pointer(init_net.gen, ng);
1200 init_net.key_domain = &init_net_key_domain;
1202 down_write(&pernet_ops_rwsem);
1203 preinit_net(&init_net);
1204 if (setup_net(&init_net, &init_user_ns))
1205 panic("Could not setup the initial network namespace");
1207 init_net_initialized = true;
1208 up_write(&pernet_ops_rwsem);
1210 if (register_pernet_subsys(&net_ns_ops))
1211 panic("Could not register network namespace subsystems");
1213 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
1214 RTNL_FLAG_DOIT_UNLOCKED);
1215 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
1216 RTNL_FLAG_DOIT_UNLOCKED |
1217 RTNL_FLAG_DUMP_UNLOCKED);
1220 static void free_exit_list(struct pernet_operations *ops, struct list_head *net_exit_list)
1222 ops_pre_exit_list(ops, net_exit_list);
1225 if (ops->exit_batch_rtnl) {
1226 LIST_HEAD(dev_kill_list);
1229 ops->exit_batch_rtnl(net_exit_list, &dev_kill_list);
1230 unregister_netdevice_many(&dev_kill_list);
1233 ops_exit_list(ops, net_exit_list);
1235 ops_free_list(ops, net_exit_list);
1238 #ifdef CONFIG_NET_NS
1239 static int __register_pernet_operations(struct list_head *list,
1240 struct pernet_operations *ops)
1244 LIST_HEAD(net_exit_list);
1246 list_add_tail(&ops->list, list);
1247 if (ops->init || (ops->id && ops->size)) {
1248 /* We held write locked pernet_ops_rwsem, and parallel
1249 * setup_net() and cleanup_net() are not possible.
1252 error = ops_init(ops, net);
1255 list_add_tail(&net->exit_list, &net_exit_list);
1261 /* If I have an error cleanup all namespaces I initialized */
1262 list_del(&ops->list);
1263 free_exit_list(ops, &net_exit_list);
1267 static void __unregister_pernet_operations(struct pernet_operations *ops)
1270 LIST_HEAD(net_exit_list);
1272 list_del(&ops->list);
1273 /* See comment in __register_pernet_operations() */
1275 list_add_tail(&net->exit_list, &net_exit_list);
1277 free_exit_list(ops, &net_exit_list);
1282 static int __register_pernet_operations(struct list_head *list,
1283 struct pernet_operations *ops)
1285 if (!init_net_initialized) {
1286 list_add_tail(&ops->list, list);
1290 return ops_init(ops, &init_net);
1293 static void __unregister_pernet_operations(struct pernet_operations *ops)
1295 if (!init_net_initialized) {
1296 list_del(&ops->list);
1298 LIST_HEAD(net_exit_list);
1299 list_add(&init_net.exit_list, &net_exit_list);
1300 free_exit_list(ops, &net_exit_list);
1304 #endif /* CONFIG_NET_NS */
1306 static DEFINE_IDA(net_generic_ids);
1308 static int register_pernet_operations(struct list_head *list,
1309 struct pernet_operations *ops)
1314 error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID,
1319 /* This does not require READ_ONCE as writers already hold
1320 * pernet_ops_rwsem. But WRITE_ONCE is needed to protect
1321 * net_alloc_generic.
1323 WRITE_ONCE(max_gen_ptrs, max(max_gen_ptrs, *ops->id + 1));
1325 error = __register_pernet_operations(list, ops);
1329 ida_free(&net_generic_ids, *ops->id);
1335 static void unregister_pernet_operations(struct pernet_operations *ops)
1337 __unregister_pernet_operations(ops);
1340 ida_free(&net_generic_ids, *ops->id);
1344 * register_pernet_subsys - register a network namespace subsystem
1345 * @ops: pernet operations structure for the subsystem
1347 * Register a subsystem which has init and exit functions
1348 * that are called when network namespaces are created and
1349 * destroyed respectively.
1351 * When registered all network namespace init functions are
1352 * called for every existing network namespace. Allowing kernel
1353 * modules to have a race free view of the set of network namespaces.
1355 * When a new network namespace is created all of the init
1356 * methods are called in the order in which they were registered.
1358 * When a network namespace is destroyed all of the exit methods
1359 * are called in the reverse of the order with which they were
1362 int register_pernet_subsys(struct pernet_operations *ops)
1365 down_write(&pernet_ops_rwsem);
1366 error = register_pernet_operations(first_device, ops);
1367 up_write(&pernet_ops_rwsem);
1370 EXPORT_SYMBOL_GPL(register_pernet_subsys);
1373 * unregister_pernet_subsys - unregister a network namespace subsystem
1374 * @ops: pernet operations structure to manipulate
1376 * Remove the pernet operations structure from the list to be
1377 * used when network namespaces are created or destroyed. In
1378 * addition run the exit method for all existing network
1381 void unregister_pernet_subsys(struct pernet_operations *ops)
1383 down_write(&pernet_ops_rwsem);
1384 unregister_pernet_operations(ops);
1385 up_write(&pernet_ops_rwsem);
1387 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1390 * register_pernet_device - register a network namespace device
1391 * @ops: pernet operations structure for the subsystem
1393 * Register a device which has init and exit functions
1394 * that are called when network namespaces are created and
1395 * destroyed respectively.
1397 * When registered all network namespace init functions are
1398 * called for every existing network namespace. Allowing kernel
1399 * modules to have a race free view of the set of network namespaces.
1401 * When a new network namespace is created all of the init
1402 * methods are called in the order in which they were registered.
1404 * When a network namespace is destroyed all of the exit methods
1405 * are called in the reverse of the order with which they were
1408 int register_pernet_device(struct pernet_operations *ops)
1411 down_write(&pernet_ops_rwsem);
1412 error = register_pernet_operations(&pernet_list, ops);
1413 if (!error && (first_device == &pernet_list))
1414 first_device = &ops->list;
1415 up_write(&pernet_ops_rwsem);
1418 EXPORT_SYMBOL_GPL(register_pernet_device);
1421 * unregister_pernet_device - unregister a network namespace netdevice
1422 * @ops: pernet operations structure to manipulate
1424 * Remove the pernet operations structure from the list to be
1425 * used when network namespaces are created or destroyed. In
1426 * addition run the exit method for all existing network
1429 void unregister_pernet_device(struct pernet_operations *ops)
1431 down_write(&pernet_ops_rwsem);
1432 if (&ops->list == first_device)
1433 first_device = first_device->next;
1434 unregister_pernet_operations(ops);
1435 up_write(&pernet_ops_rwsem);
1437 EXPORT_SYMBOL_GPL(unregister_pernet_device);
1439 #ifdef CONFIG_NET_NS
1440 static struct ns_common *netns_get(struct task_struct *task)
1442 struct net *net = NULL;
1443 struct nsproxy *nsproxy;
1446 nsproxy = task->nsproxy;
1448 net = get_net(nsproxy->net_ns);
1451 return net ? &net->ns : NULL;
1454 static inline struct net *to_net_ns(struct ns_common *ns)
1456 return container_of(ns, struct net, ns);
1459 static void netns_put(struct ns_common *ns)
1461 put_net(to_net_ns(ns));
1464 static int netns_install(struct nsset *nsset, struct ns_common *ns)
1466 struct nsproxy *nsproxy = nsset->nsproxy;
1467 struct net *net = to_net_ns(ns);
1469 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1470 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
1473 put_net(nsproxy->net_ns);
1474 nsproxy->net_ns = get_net(net);
1478 static struct user_namespace *netns_owner(struct ns_common *ns)
1480 return to_net_ns(ns)->user_ns;
1483 const struct proc_ns_operations netns_operations = {
1485 .type = CLONE_NEWNET,
1488 .install = netns_install,
1489 .owner = netns_owner,