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;
321 LIST_HEAD(net_exit_list);
322 LIST_HEAD(dev_kill_list);
325 refcount_set(&net->ns.count, 1);
326 ref_tracker_dir_init(&net->refcnt_tracker, 128, "net refcnt");
328 refcount_set(&net->passive, 1);
329 get_random_bytes(&net->hash_mix, sizeof(u32));
331 net->net_cookie = gen_cookie_next(&net_cookie);
333 net->dev_base_seq = 1;
334 net->user_ns = user_ns;
335 idr_init(&net->netns_ids);
336 spin_lock_init(&net->nsid_lock);
337 mutex_init(&net->ipv4.ra_mutex);
339 list_for_each_entry(ops, &pernet_list, list) {
340 error = ops_init(ops, net);
344 down_write(&net_rwsem);
345 list_add_tail_rcu(&net->list, &net_namespace_list);
346 up_write(&net_rwsem);
351 /* Walk through the list backwards calling the exit functions
352 * for the pernet modules whose init functions did not fail.
354 list_add(&net->exit_list, &net_exit_list);
356 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
357 ops_pre_exit_list(ops, &net_exit_list);
363 list_for_each_entry_continue_reverse(ops, &pernet_list, list) {
364 if (ops->exit_batch_rtnl)
365 ops->exit_batch_rtnl(&net_exit_list, &dev_kill_list);
367 unregister_netdevice_many(&dev_kill_list);
371 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
372 ops_exit_list(ops, &net_exit_list);
375 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
376 ops_free_list(ops, &net_exit_list);
382 static int __net_init net_defaults_init_net(struct net *net)
384 net->core.sysctl_somaxconn = SOMAXCONN;
385 /* Limits per socket sk_omem_alloc usage.
386 * TCP zerocopy regular usage needs 128 KB.
388 net->core.sysctl_optmem_max = 128 * 1024;
389 net->core.sysctl_txrehash = SOCK_TXREHASH_ENABLED;
394 static struct pernet_operations net_defaults_ops = {
395 .init = net_defaults_init_net,
398 static __init int net_defaults_init(void)
400 if (register_pernet_subsys(&net_defaults_ops))
401 panic("Cannot initialize net default settings");
406 core_initcall(net_defaults_init);
409 static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
411 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
414 static void dec_net_namespaces(struct ucounts *ucounts)
416 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
419 static struct kmem_cache *net_cachep __ro_after_init;
420 static struct workqueue_struct *netns_wq;
422 static struct net *net_alloc(void)
424 struct net *net = NULL;
425 struct net_generic *ng;
427 ng = net_alloc_generic();
431 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
436 net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
437 if (!net->key_domain)
439 refcount_set(&net->key_domain->usage, 1);
442 rcu_assign_pointer(net->gen, ng);
448 kmem_cache_free(net_cachep, net);
456 static void net_free(struct net *net)
458 if (refcount_dec_and_test(&net->passive)) {
459 kfree(rcu_access_pointer(net->gen));
461 /* There should not be any trackers left there. */
462 ref_tracker_dir_exit(&net->notrefcnt_tracker);
464 kmem_cache_free(net_cachep, net);
468 void net_drop_ns(void *p)
470 struct net *net = (struct net *)p;
476 struct net *copy_net_ns(unsigned long flags,
477 struct user_namespace *user_ns, struct net *old_net)
479 struct ucounts *ucounts;
483 if (!(flags & CLONE_NEWNET))
484 return get_net(old_net);
486 ucounts = inc_net_namespaces(user_ns);
488 return ERR_PTR(-ENOSPC);
497 refcount_set(&net->passive, 1);
498 net->ucounts = ucounts;
499 get_user_ns(user_ns);
501 rv = down_read_killable(&pernet_ops_rwsem);
505 rv = setup_net(net, user_ns);
507 up_read(&pernet_ops_rwsem);
512 key_remove_domain(net->key_domain);
514 put_user_ns(user_ns);
517 dec_net_namespaces(ucounts);
524 * net_ns_get_ownership - get sysfs ownership data for @net
525 * @net: network namespace in question (can be NULL)
526 * @uid: kernel user ID for sysfs objects
527 * @gid: kernel group ID for sysfs objects
529 * Returns the uid/gid pair of root in the user namespace associated with the
530 * given network namespace.
532 void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
535 kuid_t ns_root_uid = make_kuid(net->user_ns, 0);
536 kgid_t ns_root_gid = make_kgid(net->user_ns, 0);
538 if (uid_valid(ns_root_uid))
541 if (gid_valid(ns_root_gid))
544 *uid = GLOBAL_ROOT_UID;
545 *gid = GLOBAL_ROOT_GID;
548 EXPORT_SYMBOL_GPL(net_ns_get_ownership);
550 static void unhash_nsid(struct net *net, struct net *last)
553 /* This function is only called from cleanup_net() work,
554 * and this work is the only process, that may delete
555 * a net from net_namespace_list. So, when the below
556 * is executing, the list may only grow. Thus, we do not
557 * use for_each_net_rcu() or net_rwsem.
562 spin_lock_bh(&tmp->nsid_lock);
563 id = __peernet2id(tmp, net);
565 idr_remove(&tmp->netns_ids, id);
566 spin_unlock_bh(&tmp->nsid_lock);
568 rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL,
573 spin_lock_bh(&net->nsid_lock);
574 idr_destroy(&net->netns_ids);
575 spin_unlock_bh(&net->nsid_lock);
578 static LLIST_HEAD(cleanup_list);
580 static void cleanup_net(struct work_struct *work)
582 const struct pernet_operations *ops;
583 struct net *net, *tmp, *last;
584 struct llist_node *net_kill_list;
585 LIST_HEAD(net_exit_list);
586 LIST_HEAD(dev_kill_list);
588 /* Atomically snapshot the list of namespaces to cleanup */
589 net_kill_list = llist_del_all(&cleanup_list);
591 down_read(&pernet_ops_rwsem);
593 /* Don't let anyone else find us. */
594 down_write(&net_rwsem);
595 llist_for_each_entry(net, net_kill_list, cleanup_list)
596 list_del_rcu(&net->list);
597 /* Cache last net. After we unlock rtnl, no one new net
598 * added to net_namespace_list can assign nsid pointer
599 * to a net from net_kill_list (see peernet2id_alloc()).
600 * So, we skip them in unhash_nsid().
602 * Note, that unhash_nsid() does not delete nsid links
603 * between net_kill_list's nets, as they've already
604 * deleted from net_namespace_list. But, this would be
605 * useless anyway, as netns_ids are destroyed there.
607 last = list_last_entry(&net_namespace_list, struct net, list);
608 up_write(&net_rwsem);
610 llist_for_each_entry(net, net_kill_list, cleanup_list) {
611 unhash_nsid(net, last);
612 list_add_tail(&net->exit_list, &net_exit_list);
615 /* Run all of the network namespace pre_exit methods */
616 list_for_each_entry_reverse(ops, &pernet_list, list)
617 ops_pre_exit_list(ops, &net_exit_list);
620 * Another CPU might be rcu-iterating the list, wait for it.
621 * This needs to be before calling the exit() notifiers, so
622 * the rcu_barrier() below isn't sufficient alone.
623 * Also the pre_exit() and exit() methods need this barrier.
625 synchronize_rcu_expedited();
628 list_for_each_entry_reverse(ops, &pernet_list, list) {
629 if (ops->exit_batch_rtnl)
630 ops->exit_batch_rtnl(&net_exit_list, &dev_kill_list);
632 unregister_netdevice_many(&dev_kill_list);
635 /* Run all of the network namespace exit methods */
636 list_for_each_entry_reverse(ops, &pernet_list, list)
637 ops_exit_list(ops, &net_exit_list);
639 /* Free the net generic variables */
640 list_for_each_entry_reverse(ops, &pernet_list, list)
641 ops_free_list(ops, &net_exit_list);
643 up_read(&pernet_ops_rwsem);
645 /* Ensure there are no outstanding rcu callbacks using this
650 /* Finally it is safe to free my network namespace structure */
651 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
652 list_del_init(&net->exit_list);
653 dec_net_namespaces(net->ucounts);
655 key_remove_domain(net->key_domain);
657 put_user_ns(net->user_ns);
663 * net_ns_barrier - wait until concurrent net_cleanup_work is done
665 * cleanup_net runs from work queue and will first remove namespaces
666 * from the global list, then run net exit functions.
668 * Call this in module exit path to make sure that all netns
669 * ->exit ops have been invoked before the function is removed.
671 void net_ns_barrier(void)
673 down_write(&pernet_ops_rwsem);
674 up_write(&pernet_ops_rwsem);
676 EXPORT_SYMBOL(net_ns_barrier);
678 static DECLARE_WORK(net_cleanup_work, cleanup_net);
680 void __put_net(struct net *net)
682 ref_tracker_dir_exit(&net->refcnt_tracker);
683 /* Cleanup the network namespace in process context */
684 if (llist_add(&net->cleanup_list, &cleanup_list))
685 queue_work(netns_wq, &net_cleanup_work);
687 EXPORT_SYMBOL_GPL(__put_net);
690 * get_net_ns - increment the refcount of the network namespace
691 * @ns: common namespace (net)
693 * Returns the net's common namespace.
695 struct ns_common *get_net_ns(struct ns_common *ns)
697 return &get_net(container_of(ns, struct net, ns))->ns;
699 EXPORT_SYMBOL_GPL(get_net_ns);
701 struct net *get_net_ns_by_fd(int fd)
703 struct fd f = fdget(fd);
704 struct net *net = ERR_PTR(-EINVAL);
707 return ERR_PTR(-EBADF);
709 if (proc_ns_file(f.file)) {
710 struct ns_common *ns = get_proc_ns(file_inode(f.file));
711 if (ns->ops == &netns_operations)
712 net = get_net(container_of(ns, struct net, ns));
718 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
721 struct net *get_net_ns_by_pid(pid_t pid)
723 struct task_struct *tsk;
726 /* Lookup the network namespace */
727 net = ERR_PTR(-ESRCH);
729 tsk = find_task_by_vpid(pid);
731 struct nsproxy *nsproxy;
733 nsproxy = tsk->nsproxy;
735 net = get_net(nsproxy->net_ns);
741 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
743 static __net_init int net_ns_net_init(struct net *net)
746 net->ns.ops = &netns_operations;
748 return ns_alloc_inum(&net->ns);
751 static __net_exit void net_ns_net_exit(struct net *net)
753 ns_free_inum(&net->ns);
756 static struct pernet_operations __net_initdata net_ns_ops = {
757 .init = net_ns_net_init,
758 .exit = net_ns_net_exit,
761 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
762 [NETNSA_NONE] = { .type = NLA_UNSPEC },
763 [NETNSA_NSID] = { .type = NLA_S32 },
764 [NETNSA_PID] = { .type = NLA_U32 },
765 [NETNSA_FD] = { .type = NLA_U32 },
766 [NETNSA_TARGET_NSID] = { .type = NLA_S32 },
769 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
770 struct netlink_ext_ack *extack)
772 struct net *net = sock_net(skb->sk);
773 struct nlattr *tb[NETNSA_MAX + 1];
778 err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
779 NETNSA_MAX, rtnl_net_policy, extack);
782 if (!tb[NETNSA_NSID]) {
783 NL_SET_ERR_MSG(extack, "nsid is missing");
786 nsid = nla_get_s32(tb[NETNSA_NSID]);
788 if (tb[NETNSA_PID]) {
789 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
790 nla = tb[NETNSA_PID];
791 } else if (tb[NETNSA_FD]) {
792 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
795 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
799 NL_SET_BAD_ATTR(extack, nla);
800 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
801 return PTR_ERR(peer);
804 spin_lock_bh(&net->nsid_lock);
805 if (__peernet2id(net, peer) >= 0) {
806 spin_unlock_bh(&net->nsid_lock);
808 NL_SET_BAD_ATTR(extack, nla);
809 NL_SET_ERR_MSG(extack,
810 "Peer netns already has a nsid assigned");
814 err = alloc_netid(net, peer, nsid);
815 spin_unlock_bh(&net->nsid_lock);
817 rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid,
820 } else if (err == -ENOSPC && nsid >= 0) {
822 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
823 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
830 static int rtnl_net_get_size(void)
832 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
833 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
834 + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
838 struct net_fill_args {
848 static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
850 struct nlmsghdr *nlh;
851 struct rtgenmsg *rth;
853 nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
858 rth = nlmsg_data(nlh);
859 rth->rtgen_family = AF_UNSPEC;
861 if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
862 goto nla_put_failure;
865 nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
866 goto nla_put_failure;
872 nlmsg_cancel(skb, nlh);
876 static int rtnl_net_valid_getid_req(struct sk_buff *skb,
877 const struct nlmsghdr *nlh,
879 struct netlink_ext_ack *extack)
883 if (!netlink_strict_get_check(skb))
884 return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
885 tb, NETNSA_MAX, rtnl_net_policy,
888 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
889 NETNSA_MAX, rtnl_net_policy,
894 for (i = 0; i <= NETNSA_MAX; i++) {
902 case NETNSA_TARGET_NSID:
905 NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
913 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
914 struct netlink_ext_ack *extack)
916 struct net *net = sock_net(skb->sk);
917 struct nlattr *tb[NETNSA_MAX + 1];
918 struct net_fill_args fillargs = {
919 .portid = NETLINK_CB(skb).portid,
920 .seq = nlh->nlmsg_seq,
923 struct net *peer, *target = net;
928 err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
931 if (tb[NETNSA_PID]) {
932 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
933 nla = tb[NETNSA_PID];
934 } else if (tb[NETNSA_FD]) {
935 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
937 } else if (tb[NETNSA_NSID]) {
938 peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
940 peer = ERR_PTR(-ENOENT);
941 nla = tb[NETNSA_NSID];
943 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
948 NL_SET_BAD_ATTR(extack, nla);
949 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
950 return PTR_ERR(peer);
953 if (tb[NETNSA_TARGET_NSID]) {
954 int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
956 target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
957 if (IS_ERR(target)) {
958 NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
959 NL_SET_ERR_MSG(extack,
960 "Target netns reference is invalid");
961 err = PTR_ERR(target);
964 fillargs.add_ref = true;
965 fillargs.ref_nsid = peernet2id(net, peer);
968 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
974 fillargs.nsid = peernet2id(target, peer);
975 err = rtnl_net_fill(msg, &fillargs);
979 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
985 if (fillargs.add_ref)
991 struct rtnl_net_dump_cb {
995 struct net_fill_args fillargs;
1000 /* Runs in RCU-critical section. */
1001 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
1003 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
1006 if (net_cb->idx < net_cb->s_idx)
1009 net_cb->fillargs.nsid = id;
1010 if (net_cb->fillargs.add_ref)
1011 net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
1012 ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
1021 static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
1022 struct rtnl_net_dump_cb *net_cb,
1023 struct netlink_callback *cb)
1025 struct netlink_ext_ack *extack = cb->extack;
1026 struct nlattr *tb[NETNSA_MAX + 1];
1029 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
1030 NETNSA_MAX, rtnl_net_policy,
1035 for (i = 0; i <= NETNSA_MAX; i++) {
1039 if (i == NETNSA_TARGET_NSID) {
1042 net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
1044 NL_SET_BAD_ATTR(extack, tb[i]);
1045 NL_SET_ERR_MSG(extack,
1046 "Invalid target network namespace id");
1047 return PTR_ERR(net);
1049 net_cb->fillargs.add_ref = true;
1050 net_cb->ref_net = net_cb->tgt_net;
1051 net_cb->tgt_net = net;
1053 NL_SET_BAD_ATTR(extack, tb[i]);
1054 NL_SET_ERR_MSG(extack,
1055 "Unsupported attribute in dump request");
1063 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1065 struct rtnl_net_dump_cb net_cb = {
1066 .tgt_net = sock_net(skb->sk),
1069 .portid = NETLINK_CB(cb->skb).portid,
1070 .seq = cb->nlh->nlmsg_seq,
1071 .flags = NLM_F_MULTI,
1075 .s_idx = cb->args[0],
1079 if (cb->strict_check) {
1080 err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1086 idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1089 cb->args[0] = net_cb.idx;
1091 if (net_cb.fillargs.add_ref)
1092 put_net(net_cb.tgt_net);
1093 return err < 0 ? err : skb->len;
1096 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1097 struct nlmsghdr *nlh, gfp_t gfp)
1099 struct net_fill_args fillargs = {
1101 .seq = nlh ? nlh->nlmsg_seq : 0,
1105 struct sk_buff *msg;
1108 msg = nlmsg_new(rtnl_net_get_size(), gfp);
1112 err = rtnl_net_fill(msg, &fillargs);
1116 rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, gfp);
1122 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1125 #ifdef CONFIG_NET_NS
1126 static void __init netns_ipv4_struct_check(void)
1128 /* TX readonly hotpath cache lines */
1129 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1130 sysctl_tcp_early_retrans);
1131 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1132 sysctl_tcp_tso_win_divisor);
1133 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1134 sysctl_tcp_tso_rtt_log);
1135 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1136 sysctl_tcp_autocorking);
1137 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1138 sysctl_tcp_min_snd_mss);
1139 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1140 sysctl_tcp_notsent_lowat);
1141 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1142 sysctl_tcp_limit_output_bytes);
1143 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1144 sysctl_tcp_min_rtt_wlen);
1145 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1147 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1148 sysctl_ip_fwd_use_pmtu);
1149 CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_tx, 33);
1151 /* TXRX readonly hotpath cache lines */
1152 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_txrx,
1153 sysctl_tcp_moderate_rcvbuf);
1154 CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_txrx, 1);
1156 /* RX readonly hotpath cache line */
1157 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1158 sysctl_ip_early_demux);
1159 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1160 sysctl_tcp_early_demux);
1161 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1162 sysctl_tcp_reordering);
1163 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1165 CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_rx, 18);
1169 void __init net_ns_init(void)
1171 struct net_generic *ng;
1173 #ifdef CONFIG_NET_NS
1174 netns_ipv4_struct_check();
1175 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1177 SLAB_PANIC|SLAB_ACCOUNT, NULL);
1179 /* Create workqueue for cleanup */
1180 netns_wq = create_singlethread_workqueue("netns");
1182 panic("Could not create netns workq");
1185 ng = net_alloc_generic();
1187 panic("Could not allocate generic netns");
1189 rcu_assign_pointer(init_net.gen, ng);
1192 init_net.key_domain = &init_net_key_domain;
1194 down_write(&pernet_ops_rwsem);
1195 preinit_net(&init_net);
1196 if (setup_net(&init_net, &init_user_ns))
1197 panic("Could not setup the initial network namespace");
1199 init_net_initialized = true;
1200 up_write(&pernet_ops_rwsem);
1202 if (register_pernet_subsys(&net_ns_ops))
1203 panic("Could not register network namespace subsystems");
1205 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
1206 RTNL_FLAG_DOIT_UNLOCKED);
1207 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
1208 RTNL_FLAG_DOIT_UNLOCKED);
1211 static void free_exit_list(struct pernet_operations *ops, struct list_head *net_exit_list)
1213 ops_pre_exit_list(ops, net_exit_list);
1216 if (ops->exit_batch_rtnl) {
1217 LIST_HEAD(dev_kill_list);
1220 ops->exit_batch_rtnl(net_exit_list, &dev_kill_list);
1221 unregister_netdevice_many(&dev_kill_list);
1224 ops_exit_list(ops, net_exit_list);
1226 ops_free_list(ops, net_exit_list);
1229 #ifdef CONFIG_NET_NS
1230 static int __register_pernet_operations(struct list_head *list,
1231 struct pernet_operations *ops)
1235 LIST_HEAD(net_exit_list);
1237 list_add_tail(&ops->list, list);
1238 if (ops->init || (ops->id && ops->size)) {
1239 /* We held write locked pernet_ops_rwsem, and parallel
1240 * setup_net() and cleanup_net() are not possible.
1243 error = ops_init(ops, net);
1246 list_add_tail(&net->exit_list, &net_exit_list);
1252 /* If I have an error cleanup all namespaces I initialized */
1253 list_del(&ops->list);
1254 free_exit_list(ops, &net_exit_list);
1258 static void __unregister_pernet_operations(struct pernet_operations *ops)
1261 LIST_HEAD(net_exit_list);
1263 list_del(&ops->list);
1264 /* See comment in __register_pernet_operations() */
1266 list_add_tail(&net->exit_list, &net_exit_list);
1268 free_exit_list(ops, &net_exit_list);
1273 static int __register_pernet_operations(struct list_head *list,
1274 struct pernet_operations *ops)
1276 if (!init_net_initialized) {
1277 list_add_tail(&ops->list, list);
1281 return ops_init(ops, &init_net);
1284 static void __unregister_pernet_operations(struct pernet_operations *ops)
1286 if (!init_net_initialized) {
1287 list_del(&ops->list);
1289 LIST_HEAD(net_exit_list);
1290 list_add(&init_net.exit_list, &net_exit_list);
1291 free_exit_list(ops, &net_exit_list);
1295 #endif /* CONFIG_NET_NS */
1297 static DEFINE_IDA(net_generic_ids);
1299 static int register_pernet_operations(struct list_head *list,
1300 struct pernet_operations *ops)
1305 error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID,
1310 max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
1312 error = __register_pernet_operations(list, ops);
1316 ida_free(&net_generic_ids, *ops->id);
1322 static void unregister_pernet_operations(struct pernet_operations *ops)
1324 __unregister_pernet_operations(ops);
1327 ida_free(&net_generic_ids, *ops->id);
1331 * register_pernet_subsys - register a network namespace subsystem
1332 * @ops: pernet operations structure for the subsystem
1334 * Register a subsystem which has init and exit functions
1335 * that are called when network namespaces are created and
1336 * destroyed respectively.
1338 * When registered all network namespace init functions are
1339 * called for every existing network namespace. Allowing kernel
1340 * modules to have a race free view of the set of network namespaces.
1342 * When a new network namespace is created all of the init
1343 * methods are called in the order in which they were registered.
1345 * When a network namespace is destroyed all of the exit methods
1346 * are called in the reverse of the order with which they were
1349 int register_pernet_subsys(struct pernet_operations *ops)
1352 down_write(&pernet_ops_rwsem);
1353 error = register_pernet_operations(first_device, ops);
1354 up_write(&pernet_ops_rwsem);
1357 EXPORT_SYMBOL_GPL(register_pernet_subsys);
1360 * unregister_pernet_subsys - unregister a network namespace subsystem
1361 * @ops: pernet operations structure to manipulate
1363 * Remove the pernet operations structure from the list to be
1364 * used when network namespaces are created or destroyed. In
1365 * addition run the exit method for all existing network
1368 void unregister_pernet_subsys(struct pernet_operations *ops)
1370 down_write(&pernet_ops_rwsem);
1371 unregister_pernet_operations(ops);
1372 up_write(&pernet_ops_rwsem);
1374 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1377 * register_pernet_device - register a network namespace device
1378 * @ops: pernet operations structure for the subsystem
1380 * Register a device which has init and exit functions
1381 * that are called when network namespaces are created and
1382 * destroyed respectively.
1384 * When registered all network namespace init functions are
1385 * called for every existing network namespace. Allowing kernel
1386 * modules to have a race free view of the set of network namespaces.
1388 * When a new network namespace is created all of the init
1389 * methods are called in the order in which they were registered.
1391 * When a network namespace is destroyed all of the exit methods
1392 * are called in the reverse of the order with which they were
1395 int register_pernet_device(struct pernet_operations *ops)
1398 down_write(&pernet_ops_rwsem);
1399 error = register_pernet_operations(&pernet_list, ops);
1400 if (!error && (first_device == &pernet_list))
1401 first_device = &ops->list;
1402 up_write(&pernet_ops_rwsem);
1405 EXPORT_SYMBOL_GPL(register_pernet_device);
1408 * unregister_pernet_device - unregister a network namespace netdevice
1409 * @ops: pernet operations structure to manipulate
1411 * Remove the pernet operations structure from the list to be
1412 * used when network namespaces are created or destroyed. In
1413 * addition run the exit method for all existing network
1416 void unregister_pernet_device(struct pernet_operations *ops)
1418 down_write(&pernet_ops_rwsem);
1419 if (&ops->list == first_device)
1420 first_device = first_device->next;
1421 unregister_pernet_operations(ops);
1422 up_write(&pernet_ops_rwsem);
1424 EXPORT_SYMBOL_GPL(unregister_pernet_device);
1426 #ifdef CONFIG_NET_NS
1427 static struct ns_common *netns_get(struct task_struct *task)
1429 struct net *net = NULL;
1430 struct nsproxy *nsproxy;
1433 nsproxy = task->nsproxy;
1435 net = get_net(nsproxy->net_ns);
1438 return net ? &net->ns : NULL;
1441 static inline struct net *to_net_ns(struct ns_common *ns)
1443 return container_of(ns, struct net, ns);
1446 static void netns_put(struct ns_common *ns)
1448 put_net(to_net_ns(ns));
1451 static int netns_install(struct nsset *nsset, struct ns_common *ns)
1453 struct nsproxy *nsproxy = nsset->nsproxy;
1454 struct net *net = to_net_ns(ns);
1456 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1457 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
1460 put_net(nsproxy->net_ns);
1461 nsproxy->net_ns = get_net(net);
1465 static struct user_namespace *netns_owner(struct ns_common *ns)
1467 return to_net_ns(ns)->user_ns;
1470 const struct proc_ns_operations netns_operations = {
1472 .type = CLONE_NEWNET,
1475 .install = netns_install,
1476 .owner = netns_owner,