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 struct net_generic *ng;
73 unsigned int generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
75 ng = kzalloc(generic_size, GFP_KERNEL);
77 ng->s.len = max_gen_ptrs;
82 static int net_assign_generic(struct net *net, unsigned int id, void *data)
84 struct net_generic *ng, *old_ng;
86 BUG_ON(id < MIN_PERNET_OPS_ID);
88 old_ng = rcu_dereference_protected(net->gen,
89 lockdep_is_held(&pernet_ops_rwsem));
90 if (old_ng->s.len > id) {
91 old_ng->ptr[id] = data;
95 ng = net_alloc_generic();
100 * Some synchronisation notes:
102 * The net_generic explores the net->gen array inside rcu
103 * read section. Besides once set the net->gen->ptr[x]
104 * pointer never changes (see rules in netns/generic.h).
106 * That said, we simply duplicate this array and schedule
107 * the old copy for kfree after a grace period.
110 memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
111 (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
114 rcu_assign_pointer(net->gen, ng);
115 kfree_rcu(old_ng, s.rcu);
119 static int ops_init(const struct pernet_operations *ops, struct net *net)
121 struct net_generic *ng;
125 if (ops->id && ops->size) {
126 data = kzalloc(ops->size, GFP_KERNEL);
130 err = net_assign_generic(net, *ops->id, data);
136 err = ops->init(net);
140 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;
153 static void ops_pre_exit_list(const struct pernet_operations *ops,
154 struct list_head *net_exit_list)
159 list_for_each_entry(net, net_exit_list, exit_list)
164 static void ops_exit_list(const struct pernet_operations *ops,
165 struct list_head *net_exit_list)
169 list_for_each_entry(net, net_exit_list, exit_list) {
175 ops->exit_batch(net_exit_list);
178 static void ops_free_list(const struct pernet_operations *ops,
179 struct list_head *net_exit_list)
182 if (ops->size && ops->id) {
183 list_for_each_entry(net, net_exit_list, exit_list)
184 kfree(net_generic(net, *ops->id));
188 /* should be called with nsid_lock held */
189 static int alloc_netid(struct net *net, struct net *peer, int reqid)
191 int min = 0, max = 0;
198 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
201 /* This function is used by idr_for_each(). If net is equal to peer, the
202 * function returns the id so that idr_for_each() stops. Because we cannot
203 * returns the id 0 (idr_for_each() will not stop), we return the magic value
204 * NET_ID_ZERO (-1) for it.
206 #define NET_ID_ZERO -1
207 static int net_eq_idr(int id, void *net, void *peer)
209 if (net_eq(net, peer))
210 return id ? : NET_ID_ZERO;
214 /* Must be called from RCU-critical section or with nsid_lock held */
215 static int __peernet2id(const struct net *net, struct net *peer)
217 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
219 /* Magic value for id 0. */
220 if (id == NET_ID_ZERO)
225 return NETNSA_NSID_NOT_ASSIGNED;
228 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
229 struct nlmsghdr *nlh, gfp_t gfp);
230 /* This function returns the id of a peer netns. If no id is assigned, one will
231 * be allocated and returned.
233 int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp)
237 if (refcount_read(&net->ns.count) == 0)
238 return NETNSA_NSID_NOT_ASSIGNED;
240 spin_lock_bh(&net->nsid_lock);
241 id = __peernet2id(net, peer);
243 spin_unlock_bh(&net->nsid_lock);
247 /* When peer is obtained from RCU lists, we may race with
248 * its cleanup. Check whether it's alive, and this guarantees
249 * we never hash a peer back to net->netns_ids, after it has
250 * just been idr_remove()'d from there in cleanup_net().
252 if (!maybe_get_net(peer)) {
253 spin_unlock_bh(&net->nsid_lock);
254 return NETNSA_NSID_NOT_ASSIGNED;
257 id = alloc_netid(net, peer, -1);
258 spin_unlock_bh(&net->nsid_lock);
262 return NETNSA_NSID_NOT_ASSIGNED;
264 rtnl_net_notifyid(net, RTM_NEWNSID, id, 0, NULL, gfp);
268 EXPORT_SYMBOL_GPL(peernet2id_alloc);
270 /* This function returns, if assigned, the id of a peer netns. */
271 int peernet2id(const struct net *net, struct net *peer)
276 id = __peernet2id(net, peer);
281 EXPORT_SYMBOL(peernet2id);
283 /* This function returns true is the peer netns has an id assigned into the
286 bool peernet_has_id(const struct net *net, struct net *peer)
288 return peernet2id(net, peer) >= 0;
291 struct net *get_net_ns_by_id(const struct net *net, int id)
299 peer = idr_find(&net->netns_ids, id);
301 peer = maybe_get_net(peer);
306 EXPORT_SYMBOL_GPL(get_net_ns_by_id);
308 /* init code that must occur even if setup_net() is not called. */
309 static __net_init void preinit_net(struct net *net)
311 ref_tracker_dir_init(&net->notrefcnt_tracker, 128, "net notrefcnt");
315 * setup_net runs the initializers for the network namespace object.
317 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
319 /* Must be called with pernet_ops_rwsem held */
320 const struct pernet_operations *ops, *saved_ops;
322 LIST_HEAD(net_exit_list);
324 refcount_set(&net->ns.count, 1);
325 ref_tracker_dir_init(&net->refcnt_tracker, 128, "net refcnt");
327 refcount_set(&net->passive, 1);
328 get_random_bytes(&net->hash_mix, sizeof(u32));
330 net->net_cookie = gen_cookie_next(&net_cookie);
332 net->dev_base_seq = 1;
333 net->user_ns = user_ns;
334 idr_init(&net->netns_ids);
335 spin_lock_init(&net->nsid_lock);
336 mutex_init(&net->ipv4.ra_mutex);
338 list_for_each_entry(ops, &pernet_list, list) {
339 error = ops_init(ops, net);
343 down_write(&net_rwsem);
344 list_add_tail_rcu(&net->list, &net_namespace_list);
345 up_write(&net_rwsem);
350 /* Walk through the list backwards calling the exit functions
351 * for the pernet modules whose init functions did not fail.
353 list_add(&net->exit_list, &net_exit_list);
355 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
356 ops_pre_exit_list(ops, &net_exit_list);
361 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
362 ops_exit_list(ops, &net_exit_list);
365 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
366 ops_free_list(ops, &net_exit_list);
372 static int __net_init net_defaults_init_net(struct net *net)
374 net->core.sysctl_somaxconn = SOMAXCONN;
375 /* Limits per socket sk_omem_alloc usage.
376 * TCP zerocopy regular usage needs 128 KB.
378 net->core.sysctl_optmem_max = 128 * 1024;
379 net->core.sysctl_txrehash = SOCK_TXREHASH_ENABLED;
384 static struct pernet_operations net_defaults_ops = {
385 .init = net_defaults_init_net,
388 static __init int net_defaults_init(void)
390 if (register_pernet_subsys(&net_defaults_ops))
391 panic("Cannot initialize net default settings");
396 core_initcall(net_defaults_init);
399 static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
401 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
404 static void dec_net_namespaces(struct ucounts *ucounts)
406 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
409 static struct kmem_cache *net_cachep __ro_after_init;
410 static struct workqueue_struct *netns_wq;
412 static struct net *net_alloc(void)
414 struct net *net = NULL;
415 struct net_generic *ng;
417 ng = net_alloc_generic();
421 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
426 net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
427 if (!net->key_domain)
429 refcount_set(&net->key_domain->usage, 1);
432 rcu_assign_pointer(net->gen, ng);
438 kmem_cache_free(net_cachep, net);
446 static void net_free(struct net *net)
448 if (refcount_dec_and_test(&net->passive)) {
449 kfree(rcu_access_pointer(net->gen));
451 /* There should not be any trackers left there. */
452 ref_tracker_dir_exit(&net->notrefcnt_tracker);
454 kmem_cache_free(net_cachep, net);
458 void net_drop_ns(void *p)
460 struct net *net = (struct net *)p;
466 struct net *copy_net_ns(unsigned long flags,
467 struct user_namespace *user_ns, struct net *old_net)
469 struct ucounts *ucounts;
473 if (!(flags & CLONE_NEWNET))
474 return get_net(old_net);
476 ucounts = inc_net_namespaces(user_ns);
478 return ERR_PTR(-ENOSPC);
487 refcount_set(&net->passive, 1);
488 net->ucounts = ucounts;
489 get_user_ns(user_ns);
491 rv = down_read_killable(&pernet_ops_rwsem);
495 rv = setup_net(net, user_ns);
497 up_read(&pernet_ops_rwsem);
502 key_remove_domain(net->key_domain);
504 put_user_ns(user_ns);
507 dec_net_namespaces(ucounts);
514 * net_ns_get_ownership - get sysfs ownership data for @net
515 * @net: network namespace in question (can be NULL)
516 * @uid: kernel user ID for sysfs objects
517 * @gid: kernel group ID for sysfs objects
519 * Returns the uid/gid pair of root in the user namespace associated with the
520 * given network namespace.
522 void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
525 kuid_t ns_root_uid = make_kuid(net->user_ns, 0);
526 kgid_t ns_root_gid = make_kgid(net->user_ns, 0);
528 if (uid_valid(ns_root_uid))
531 if (gid_valid(ns_root_gid))
534 *uid = GLOBAL_ROOT_UID;
535 *gid = GLOBAL_ROOT_GID;
538 EXPORT_SYMBOL_GPL(net_ns_get_ownership);
540 static void unhash_nsid(struct net *net, struct net *last)
543 /* This function is only called from cleanup_net() work,
544 * and this work is the only process, that may delete
545 * a net from net_namespace_list. So, when the below
546 * is executing, the list may only grow. Thus, we do not
547 * use for_each_net_rcu() or net_rwsem.
552 spin_lock_bh(&tmp->nsid_lock);
553 id = __peernet2id(tmp, net);
555 idr_remove(&tmp->netns_ids, id);
556 spin_unlock_bh(&tmp->nsid_lock);
558 rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL,
563 spin_lock_bh(&net->nsid_lock);
564 idr_destroy(&net->netns_ids);
565 spin_unlock_bh(&net->nsid_lock);
568 static LLIST_HEAD(cleanup_list);
570 static void cleanup_net(struct work_struct *work)
572 const struct pernet_operations *ops;
573 struct net *net, *tmp, *last;
574 struct llist_node *net_kill_list;
575 LIST_HEAD(net_exit_list);
577 /* Atomically snapshot the list of namespaces to cleanup */
578 net_kill_list = llist_del_all(&cleanup_list);
580 down_read(&pernet_ops_rwsem);
582 /* Don't let anyone else find us. */
583 down_write(&net_rwsem);
584 llist_for_each_entry(net, net_kill_list, cleanup_list)
585 list_del_rcu(&net->list);
586 /* Cache last net. After we unlock rtnl, no one new net
587 * added to net_namespace_list can assign nsid pointer
588 * to a net from net_kill_list (see peernet2id_alloc()).
589 * So, we skip them in unhash_nsid().
591 * Note, that unhash_nsid() does not delete nsid links
592 * between net_kill_list's nets, as they've already
593 * deleted from net_namespace_list. But, this would be
594 * useless anyway, as netns_ids are destroyed there.
596 last = list_last_entry(&net_namespace_list, struct net, list);
597 up_write(&net_rwsem);
599 llist_for_each_entry(net, net_kill_list, cleanup_list) {
600 unhash_nsid(net, last);
601 list_add_tail(&net->exit_list, &net_exit_list);
604 /* Run all of the network namespace pre_exit methods */
605 list_for_each_entry_reverse(ops, &pernet_list, list)
606 ops_pre_exit_list(ops, &net_exit_list);
609 * Another CPU might be rcu-iterating the list, wait for it.
610 * This needs to be before calling the exit() notifiers, so
611 * the rcu_barrier() below isn't sufficient alone.
612 * Also the pre_exit() and exit() methods need this barrier.
616 /* Run all of the network namespace exit methods */
617 list_for_each_entry_reverse(ops, &pernet_list, list)
618 ops_exit_list(ops, &net_exit_list);
620 /* Free the net generic variables */
621 list_for_each_entry_reverse(ops, &pernet_list, list)
622 ops_free_list(ops, &net_exit_list);
624 up_read(&pernet_ops_rwsem);
626 /* Ensure there are no outstanding rcu callbacks using this
631 /* Finally it is safe to free my network namespace structure */
632 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
633 list_del_init(&net->exit_list);
634 dec_net_namespaces(net->ucounts);
636 key_remove_domain(net->key_domain);
638 put_user_ns(net->user_ns);
644 * net_ns_barrier - wait until concurrent net_cleanup_work is done
646 * cleanup_net runs from work queue and will first remove namespaces
647 * from the global list, then run net exit functions.
649 * Call this in module exit path to make sure that all netns
650 * ->exit ops have been invoked before the function is removed.
652 void net_ns_barrier(void)
654 down_write(&pernet_ops_rwsem);
655 up_write(&pernet_ops_rwsem);
657 EXPORT_SYMBOL(net_ns_barrier);
659 static DECLARE_WORK(net_cleanup_work, cleanup_net);
661 void __put_net(struct net *net)
663 ref_tracker_dir_exit(&net->refcnt_tracker);
664 /* Cleanup the network namespace in process context */
665 if (llist_add(&net->cleanup_list, &cleanup_list))
666 queue_work(netns_wq, &net_cleanup_work);
668 EXPORT_SYMBOL_GPL(__put_net);
671 * get_net_ns - increment the refcount of the network namespace
672 * @ns: common namespace (net)
674 * Returns the net's common namespace.
676 struct ns_common *get_net_ns(struct ns_common *ns)
678 return &get_net(container_of(ns, struct net, ns))->ns;
680 EXPORT_SYMBOL_GPL(get_net_ns);
682 struct net *get_net_ns_by_fd(int fd)
684 struct fd f = fdget(fd);
685 struct net *net = ERR_PTR(-EINVAL);
688 return ERR_PTR(-EBADF);
690 if (proc_ns_file(f.file)) {
691 struct ns_common *ns = get_proc_ns(file_inode(f.file));
692 if (ns->ops == &netns_operations)
693 net = get_net(container_of(ns, struct net, ns));
699 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
702 struct net *get_net_ns_by_pid(pid_t pid)
704 struct task_struct *tsk;
707 /* Lookup the network namespace */
708 net = ERR_PTR(-ESRCH);
710 tsk = find_task_by_vpid(pid);
712 struct nsproxy *nsproxy;
714 nsproxy = tsk->nsproxy;
716 net = get_net(nsproxy->net_ns);
722 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
724 static __net_init int net_ns_net_init(struct net *net)
727 net->ns.ops = &netns_operations;
729 return ns_alloc_inum(&net->ns);
732 static __net_exit void net_ns_net_exit(struct net *net)
734 ns_free_inum(&net->ns);
737 static struct pernet_operations __net_initdata net_ns_ops = {
738 .init = net_ns_net_init,
739 .exit = net_ns_net_exit,
742 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
743 [NETNSA_NONE] = { .type = NLA_UNSPEC },
744 [NETNSA_NSID] = { .type = NLA_S32 },
745 [NETNSA_PID] = { .type = NLA_U32 },
746 [NETNSA_FD] = { .type = NLA_U32 },
747 [NETNSA_TARGET_NSID] = { .type = NLA_S32 },
750 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
751 struct netlink_ext_ack *extack)
753 struct net *net = sock_net(skb->sk);
754 struct nlattr *tb[NETNSA_MAX + 1];
759 err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
760 NETNSA_MAX, rtnl_net_policy, extack);
763 if (!tb[NETNSA_NSID]) {
764 NL_SET_ERR_MSG(extack, "nsid is missing");
767 nsid = nla_get_s32(tb[NETNSA_NSID]);
769 if (tb[NETNSA_PID]) {
770 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
771 nla = tb[NETNSA_PID];
772 } else if (tb[NETNSA_FD]) {
773 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
776 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
780 NL_SET_BAD_ATTR(extack, nla);
781 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
782 return PTR_ERR(peer);
785 spin_lock_bh(&net->nsid_lock);
786 if (__peernet2id(net, peer) >= 0) {
787 spin_unlock_bh(&net->nsid_lock);
789 NL_SET_BAD_ATTR(extack, nla);
790 NL_SET_ERR_MSG(extack,
791 "Peer netns already has a nsid assigned");
795 err = alloc_netid(net, peer, nsid);
796 spin_unlock_bh(&net->nsid_lock);
798 rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid,
801 } else if (err == -ENOSPC && nsid >= 0) {
803 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
804 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
811 static int rtnl_net_get_size(void)
813 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
814 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
815 + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
819 struct net_fill_args {
829 static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
831 struct nlmsghdr *nlh;
832 struct rtgenmsg *rth;
834 nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
839 rth = nlmsg_data(nlh);
840 rth->rtgen_family = AF_UNSPEC;
842 if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
843 goto nla_put_failure;
846 nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
847 goto nla_put_failure;
853 nlmsg_cancel(skb, nlh);
857 static int rtnl_net_valid_getid_req(struct sk_buff *skb,
858 const struct nlmsghdr *nlh,
860 struct netlink_ext_ack *extack)
864 if (!netlink_strict_get_check(skb))
865 return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
866 tb, NETNSA_MAX, rtnl_net_policy,
869 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
870 NETNSA_MAX, rtnl_net_policy,
875 for (i = 0; i <= NETNSA_MAX; i++) {
883 case NETNSA_TARGET_NSID:
886 NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
894 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
895 struct netlink_ext_ack *extack)
897 struct net *net = sock_net(skb->sk);
898 struct nlattr *tb[NETNSA_MAX + 1];
899 struct net_fill_args fillargs = {
900 .portid = NETLINK_CB(skb).portid,
901 .seq = nlh->nlmsg_seq,
904 struct net *peer, *target = net;
909 err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
912 if (tb[NETNSA_PID]) {
913 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
914 nla = tb[NETNSA_PID];
915 } else if (tb[NETNSA_FD]) {
916 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
918 } else if (tb[NETNSA_NSID]) {
919 peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
921 peer = ERR_PTR(-ENOENT);
922 nla = tb[NETNSA_NSID];
924 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
929 NL_SET_BAD_ATTR(extack, nla);
930 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
931 return PTR_ERR(peer);
934 if (tb[NETNSA_TARGET_NSID]) {
935 int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
937 target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
938 if (IS_ERR(target)) {
939 NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
940 NL_SET_ERR_MSG(extack,
941 "Target netns reference is invalid");
942 err = PTR_ERR(target);
945 fillargs.add_ref = true;
946 fillargs.ref_nsid = peernet2id(net, peer);
949 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
955 fillargs.nsid = peernet2id(target, peer);
956 err = rtnl_net_fill(msg, &fillargs);
960 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
966 if (fillargs.add_ref)
972 struct rtnl_net_dump_cb {
976 struct net_fill_args fillargs;
981 /* Runs in RCU-critical section. */
982 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
984 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
987 if (net_cb->idx < net_cb->s_idx)
990 net_cb->fillargs.nsid = id;
991 if (net_cb->fillargs.add_ref)
992 net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
993 ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
1002 static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
1003 struct rtnl_net_dump_cb *net_cb,
1004 struct netlink_callback *cb)
1006 struct netlink_ext_ack *extack = cb->extack;
1007 struct nlattr *tb[NETNSA_MAX + 1];
1010 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
1011 NETNSA_MAX, rtnl_net_policy,
1016 for (i = 0; i <= NETNSA_MAX; i++) {
1020 if (i == NETNSA_TARGET_NSID) {
1023 net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
1025 NL_SET_BAD_ATTR(extack, tb[i]);
1026 NL_SET_ERR_MSG(extack,
1027 "Invalid target network namespace id");
1028 return PTR_ERR(net);
1030 net_cb->fillargs.add_ref = true;
1031 net_cb->ref_net = net_cb->tgt_net;
1032 net_cb->tgt_net = net;
1034 NL_SET_BAD_ATTR(extack, tb[i]);
1035 NL_SET_ERR_MSG(extack,
1036 "Unsupported attribute in dump request");
1044 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1046 struct rtnl_net_dump_cb net_cb = {
1047 .tgt_net = sock_net(skb->sk),
1050 .portid = NETLINK_CB(cb->skb).portid,
1051 .seq = cb->nlh->nlmsg_seq,
1052 .flags = NLM_F_MULTI,
1056 .s_idx = cb->args[0],
1060 if (cb->strict_check) {
1061 err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1067 idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1070 cb->args[0] = net_cb.idx;
1072 if (net_cb.fillargs.add_ref)
1073 put_net(net_cb.tgt_net);
1074 return err < 0 ? err : skb->len;
1077 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1078 struct nlmsghdr *nlh, gfp_t gfp)
1080 struct net_fill_args fillargs = {
1082 .seq = nlh ? nlh->nlmsg_seq : 0,
1086 struct sk_buff *msg;
1089 msg = nlmsg_new(rtnl_net_get_size(), gfp);
1093 err = rtnl_net_fill(msg, &fillargs);
1097 rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, gfp);
1103 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1106 #ifdef CONFIG_NET_NS
1107 static void __init netns_ipv4_struct_check(void)
1109 /* TX readonly hotpath cache lines */
1110 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1111 sysctl_tcp_early_retrans);
1112 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1113 sysctl_tcp_tso_win_divisor);
1114 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1115 sysctl_tcp_tso_rtt_log);
1116 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1117 sysctl_tcp_autocorking);
1118 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1119 sysctl_tcp_min_snd_mss);
1120 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1121 sysctl_tcp_notsent_lowat);
1122 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1123 sysctl_tcp_limit_output_bytes);
1124 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1125 sysctl_tcp_min_rtt_wlen);
1126 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1128 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1129 sysctl_ip_fwd_use_pmtu);
1130 CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_tx, 33);
1132 /* TXRX readonly hotpath cache lines */
1133 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_txrx,
1134 sysctl_tcp_moderate_rcvbuf);
1135 CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_txrx, 1);
1137 /* RX readonly hotpath cache line */
1138 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1139 sysctl_ip_early_demux);
1140 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1141 sysctl_tcp_early_demux);
1142 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1143 sysctl_tcp_reordering);
1144 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1146 CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_rx, 18);
1150 void __init net_ns_init(void)
1152 struct net_generic *ng;
1154 #ifdef CONFIG_NET_NS
1155 netns_ipv4_struct_check();
1156 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1158 SLAB_PANIC|SLAB_ACCOUNT, NULL);
1160 /* Create workqueue for cleanup */
1161 netns_wq = create_singlethread_workqueue("netns");
1163 panic("Could not create netns workq");
1166 ng = net_alloc_generic();
1168 panic("Could not allocate generic netns");
1170 rcu_assign_pointer(init_net.gen, ng);
1173 init_net.key_domain = &init_net_key_domain;
1175 down_write(&pernet_ops_rwsem);
1176 preinit_net(&init_net);
1177 if (setup_net(&init_net, &init_user_ns))
1178 panic("Could not setup the initial network namespace");
1180 init_net_initialized = true;
1181 up_write(&pernet_ops_rwsem);
1183 if (register_pernet_subsys(&net_ns_ops))
1184 panic("Could not register network namespace subsystems");
1186 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
1187 RTNL_FLAG_DOIT_UNLOCKED);
1188 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
1189 RTNL_FLAG_DOIT_UNLOCKED);
1192 static void free_exit_list(struct pernet_operations *ops, struct list_head *net_exit_list)
1194 ops_pre_exit_list(ops, net_exit_list);
1196 ops_exit_list(ops, net_exit_list);
1197 ops_free_list(ops, net_exit_list);
1200 #ifdef CONFIG_NET_NS
1201 static int __register_pernet_operations(struct list_head *list,
1202 struct pernet_operations *ops)
1206 LIST_HEAD(net_exit_list);
1208 list_add_tail(&ops->list, list);
1209 if (ops->init || (ops->id && ops->size)) {
1210 /* We held write locked pernet_ops_rwsem, and parallel
1211 * setup_net() and cleanup_net() are not possible.
1214 error = ops_init(ops, net);
1217 list_add_tail(&net->exit_list, &net_exit_list);
1223 /* If I have an error cleanup all namespaces I initialized */
1224 list_del(&ops->list);
1225 free_exit_list(ops, &net_exit_list);
1229 static void __unregister_pernet_operations(struct pernet_operations *ops)
1232 LIST_HEAD(net_exit_list);
1234 list_del(&ops->list);
1235 /* See comment in __register_pernet_operations() */
1237 list_add_tail(&net->exit_list, &net_exit_list);
1239 free_exit_list(ops, &net_exit_list);
1244 static int __register_pernet_operations(struct list_head *list,
1245 struct pernet_operations *ops)
1247 if (!init_net_initialized) {
1248 list_add_tail(&ops->list, list);
1252 return ops_init(ops, &init_net);
1255 static void __unregister_pernet_operations(struct pernet_operations *ops)
1257 if (!init_net_initialized) {
1258 list_del(&ops->list);
1260 LIST_HEAD(net_exit_list);
1261 list_add(&init_net.exit_list, &net_exit_list);
1262 free_exit_list(ops, &net_exit_list);
1266 #endif /* CONFIG_NET_NS */
1268 static DEFINE_IDA(net_generic_ids);
1270 static int register_pernet_operations(struct list_head *list,
1271 struct pernet_operations *ops)
1276 error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID,
1281 max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
1283 error = __register_pernet_operations(list, ops);
1287 ida_free(&net_generic_ids, *ops->id);
1293 static void unregister_pernet_operations(struct pernet_operations *ops)
1295 __unregister_pernet_operations(ops);
1298 ida_free(&net_generic_ids, *ops->id);
1302 * register_pernet_subsys - register a network namespace subsystem
1303 * @ops: pernet operations structure for the subsystem
1305 * Register a subsystem which has init and exit functions
1306 * that are called when network namespaces are created and
1307 * destroyed respectively.
1309 * When registered all network namespace init functions are
1310 * called for every existing network namespace. Allowing kernel
1311 * modules to have a race free view of the set of network namespaces.
1313 * When a new network namespace is created all of the init
1314 * methods are called in the order in which they were registered.
1316 * When a network namespace is destroyed all of the exit methods
1317 * are called in the reverse of the order with which they were
1320 int register_pernet_subsys(struct pernet_operations *ops)
1323 down_write(&pernet_ops_rwsem);
1324 error = register_pernet_operations(first_device, ops);
1325 up_write(&pernet_ops_rwsem);
1328 EXPORT_SYMBOL_GPL(register_pernet_subsys);
1331 * unregister_pernet_subsys - unregister a network namespace subsystem
1332 * @ops: pernet operations structure to manipulate
1334 * Remove the pernet operations structure from the list to be
1335 * used when network namespaces are created or destroyed. In
1336 * addition run the exit method for all existing network
1339 void unregister_pernet_subsys(struct pernet_operations *ops)
1341 down_write(&pernet_ops_rwsem);
1342 unregister_pernet_operations(ops);
1343 up_write(&pernet_ops_rwsem);
1345 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1348 * register_pernet_device - register a network namespace device
1349 * @ops: pernet operations structure for the subsystem
1351 * Register a device which has init and exit functions
1352 * that are called when network namespaces are created and
1353 * destroyed respectively.
1355 * When registered all network namespace init functions are
1356 * called for every existing network namespace. Allowing kernel
1357 * modules to have a race free view of the set of network namespaces.
1359 * When a new network namespace is created all of the init
1360 * methods are called in the order in which they were registered.
1362 * When a network namespace is destroyed all of the exit methods
1363 * are called in the reverse of the order with which they were
1366 int register_pernet_device(struct pernet_operations *ops)
1369 down_write(&pernet_ops_rwsem);
1370 error = register_pernet_operations(&pernet_list, ops);
1371 if (!error && (first_device == &pernet_list))
1372 first_device = &ops->list;
1373 up_write(&pernet_ops_rwsem);
1376 EXPORT_SYMBOL_GPL(register_pernet_device);
1379 * unregister_pernet_device - unregister a network namespace netdevice
1380 * @ops: pernet operations structure to manipulate
1382 * Remove the pernet operations structure from the list to be
1383 * used when network namespaces are created or destroyed. In
1384 * addition run the exit method for all existing network
1387 void unregister_pernet_device(struct pernet_operations *ops)
1389 down_write(&pernet_ops_rwsem);
1390 if (&ops->list == first_device)
1391 first_device = first_device->next;
1392 unregister_pernet_operations(ops);
1393 up_write(&pernet_ops_rwsem);
1395 EXPORT_SYMBOL_GPL(unregister_pernet_device);
1397 #ifdef CONFIG_NET_NS
1398 static struct ns_common *netns_get(struct task_struct *task)
1400 struct net *net = NULL;
1401 struct nsproxy *nsproxy;
1404 nsproxy = task->nsproxy;
1406 net = get_net(nsproxy->net_ns);
1409 return net ? &net->ns : NULL;
1412 static inline struct net *to_net_ns(struct ns_common *ns)
1414 return container_of(ns, struct net, ns);
1417 static void netns_put(struct ns_common *ns)
1419 put_net(to_net_ns(ns));
1422 static int netns_install(struct nsset *nsset, struct ns_common *ns)
1424 struct nsproxy *nsproxy = nsset->nsproxy;
1425 struct net *net = to_net_ns(ns);
1427 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1428 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
1431 put_net(nsproxy->net_ns);
1432 nsproxy->net_ns = get_net(net);
1436 static struct user_namespace *netns_owner(struct ns_common *ns)
1438 return to_net_ns(ns)->user_ns;
1441 const struct proc_ns_operations netns_operations = {
1443 .type = CLONE_NEWNET,
1446 .install = netns_install,
1447 .owner = netns_owner,