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 net->core.sysctl_txrehash = SOCK_TXREHASH_ENABLED;
380 static struct pernet_operations net_defaults_ops = {
381 .init = net_defaults_init_net,
384 static __init int net_defaults_init(void)
386 if (register_pernet_subsys(&net_defaults_ops))
387 panic("Cannot initialize net default settings");
392 core_initcall(net_defaults_init);
395 static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
397 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
400 static void dec_net_namespaces(struct ucounts *ucounts)
402 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
405 static struct kmem_cache *net_cachep __ro_after_init;
406 static struct workqueue_struct *netns_wq;
408 static struct net *net_alloc(void)
410 struct net *net = NULL;
411 struct net_generic *ng;
413 ng = net_alloc_generic();
417 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
422 net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
423 if (!net->key_domain)
425 refcount_set(&net->key_domain->usage, 1);
428 rcu_assign_pointer(net->gen, ng);
434 kmem_cache_free(net_cachep, net);
442 static void net_free(struct net *net)
444 if (refcount_dec_and_test(&net->passive)) {
445 kfree(rcu_access_pointer(net->gen));
447 /* There should not be any trackers left there. */
448 ref_tracker_dir_exit(&net->notrefcnt_tracker);
450 kmem_cache_free(net_cachep, net);
454 void net_drop_ns(void *p)
456 struct net *net = (struct net *)p;
462 struct net *copy_net_ns(unsigned long flags,
463 struct user_namespace *user_ns, struct net *old_net)
465 struct ucounts *ucounts;
469 if (!(flags & CLONE_NEWNET))
470 return get_net(old_net);
472 ucounts = inc_net_namespaces(user_ns);
474 return ERR_PTR(-ENOSPC);
483 refcount_set(&net->passive, 1);
484 net->ucounts = ucounts;
485 get_user_ns(user_ns);
487 rv = down_read_killable(&pernet_ops_rwsem);
491 rv = setup_net(net, user_ns);
493 up_read(&pernet_ops_rwsem);
498 key_remove_domain(net->key_domain);
500 put_user_ns(user_ns);
503 dec_net_namespaces(ucounts);
510 * net_ns_get_ownership - get sysfs ownership data for @net
511 * @net: network namespace in question (can be NULL)
512 * @uid: kernel user ID for sysfs objects
513 * @gid: kernel group ID for sysfs objects
515 * Returns the uid/gid pair of root in the user namespace associated with the
516 * given network namespace.
518 void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
521 kuid_t ns_root_uid = make_kuid(net->user_ns, 0);
522 kgid_t ns_root_gid = make_kgid(net->user_ns, 0);
524 if (uid_valid(ns_root_uid))
527 if (gid_valid(ns_root_gid))
530 *uid = GLOBAL_ROOT_UID;
531 *gid = GLOBAL_ROOT_GID;
534 EXPORT_SYMBOL_GPL(net_ns_get_ownership);
536 static void unhash_nsid(struct net *net, struct net *last)
539 /* This function is only called from cleanup_net() work,
540 * and this work is the only process, that may delete
541 * a net from net_namespace_list. So, when the below
542 * is executing, the list may only grow. Thus, we do not
543 * use for_each_net_rcu() or net_rwsem.
548 spin_lock_bh(&tmp->nsid_lock);
549 id = __peernet2id(tmp, net);
551 idr_remove(&tmp->netns_ids, id);
552 spin_unlock_bh(&tmp->nsid_lock);
554 rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL,
559 spin_lock_bh(&net->nsid_lock);
560 idr_destroy(&net->netns_ids);
561 spin_unlock_bh(&net->nsid_lock);
564 static LLIST_HEAD(cleanup_list);
566 static void cleanup_net(struct work_struct *work)
568 const struct pernet_operations *ops;
569 struct net *net, *tmp, *last;
570 struct llist_node *net_kill_list;
571 LIST_HEAD(net_exit_list);
573 /* Atomically snapshot the list of namespaces to cleanup */
574 net_kill_list = llist_del_all(&cleanup_list);
576 down_read(&pernet_ops_rwsem);
578 /* Don't let anyone else find us. */
579 down_write(&net_rwsem);
580 llist_for_each_entry(net, net_kill_list, cleanup_list)
581 list_del_rcu(&net->list);
582 /* Cache last net. After we unlock rtnl, no one new net
583 * added to net_namespace_list can assign nsid pointer
584 * to a net from net_kill_list (see peernet2id_alloc()).
585 * So, we skip them in unhash_nsid().
587 * Note, that unhash_nsid() does not delete nsid links
588 * between net_kill_list's nets, as they've already
589 * deleted from net_namespace_list. But, this would be
590 * useless anyway, as netns_ids are destroyed there.
592 last = list_last_entry(&net_namespace_list, struct net, list);
593 up_write(&net_rwsem);
595 llist_for_each_entry(net, net_kill_list, cleanup_list) {
596 unhash_nsid(net, last);
597 list_add_tail(&net->exit_list, &net_exit_list);
600 /* Run all of the network namespace pre_exit methods */
601 list_for_each_entry_reverse(ops, &pernet_list, list)
602 ops_pre_exit_list(ops, &net_exit_list);
605 * Another CPU might be rcu-iterating the list, wait for it.
606 * This needs to be before calling the exit() notifiers, so
607 * the rcu_barrier() below isn't sufficient alone.
608 * Also the pre_exit() and exit() methods need this barrier.
612 /* Run all of the network namespace exit methods */
613 list_for_each_entry_reverse(ops, &pernet_list, list)
614 ops_exit_list(ops, &net_exit_list);
616 /* Free the net generic variables */
617 list_for_each_entry_reverse(ops, &pernet_list, list)
618 ops_free_list(ops, &net_exit_list);
620 up_read(&pernet_ops_rwsem);
622 /* Ensure there are no outstanding rcu callbacks using this
627 /* Finally it is safe to free my network namespace structure */
628 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
629 list_del_init(&net->exit_list);
630 dec_net_namespaces(net->ucounts);
632 key_remove_domain(net->key_domain);
634 put_user_ns(net->user_ns);
640 * net_ns_barrier - wait until concurrent net_cleanup_work is done
642 * cleanup_net runs from work queue and will first remove namespaces
643 * from the global list, then run net exit functions.
645 * Call this in module exit path to make sure that all netns
646 * ->exit ops have been invoked before the function is removed.
648 void net_ns_barrier(void)
650 down_write(&pernet_ops_rwsem);
651 up_write(&pernet_ops_rwsem);
653 EXPORT_SYMBOL(net_ns_barrier);
655 static DECLARE_WORK(net_cleanup_work, cleanup_net);
657 void __put_net(struct net *net)
659 ref_tracker_dir_exit(&net->refcnt_tracker);
660 /* Cleanup the network namespace in process context */
661 if (llist_add(&net->cleanup_list, &cleanup_list))
662 queue_work(netns_wq, &net_cleanup_work);
664 EXPORT_SYMBOL_GPL(__put_net);
667 * get_net_ns - increment the refcount of the network namespace
668 * @ns: common namespace (net)
670 * Returns the net's common namespace.
672 struct ns_common *get_net_ns(struct ns_common *ns)
674 return &get_net(container_of(ns, struct net, ns))->ns;
676 EXPORT_SYMBOL_GPL(get_net_ns);
678 struct net *get_net_ns_by_fd(int fd)
680 struct fd f = fdget(fd);
681 struct net *net = ERR_PTR(-EINVAL);
684 return ERR_PTR(-EBADF);
686 if (proc_ns_file(f.file)) {
687 struct ns_common *ns = get_proc_ns(file_inode(f.file));
688 if (ns->ops == &netns_operations)
689 net = get_net(container_of(ns, struct net, ns));
695 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
698 struct net *get_net_ns_by_pid(pid_t pid)
700 struct task_struct *tsk;
703 /* Lookup the network namespace */
704 net = ERR_PTR(-ESRCH);
706 tsk = find_task_by_vpid(pid);
708 struct nsproxy *nsproxy;
710 nsproxy = tsk->nsproxy;
712 net = get_net(nsproxy->net_ns);
718 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
720 static __net_init int net_ns_net_init(struct net *net)
723 net->ns.ops = &netns_operations;
725 return ns_alloc_inum(&net->ns);
728 static __net_exit void net_ns_net_exit(struct net *net)
730 ns_free_inum(&net->ns);
733 static struct pernet_operations __net_initdata net_ns_ops = {
734 .init = net_ns_net_init,
735 .exit = net_ns_net_exit,
738 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
739 [NETNSA_NONE] = { .type = NLA_UNSPEC },
740 [NETNSA_NSID] = { .type = NLA_S32 },
741 [NETNSA_PID] = { .type = NLA_U32 },
742 [NETNSA_FD] = { .type = NLA_U32 },
743 [NETNSA_TARGET_NSID] = { .type = NLA_S32 },
746 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
747 struct netlink_ext_ack *extack)
749 struct net *net = sock_net(skb->sk);
750 struct nlattr *tb[NETNSA_MAX + 1];
755 err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
756 NETNSA_MAX, rtnl_net_policy, extack);
759 if (!tb[NETNSA_NSID]) {
760 NL_SET_ERR_MSG(extack, "nsid is missing");
763 nsid = nla_get_s32(tb[NETNSA_NSID]);
765 if (tb[NETNSA_PID]) {
766 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
767 nla = tb[NETNSA_PID];
768 } else if (tb[NETNSA_FD]) {
769 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
772 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
776 NL_SET_BAD_ATTR(extack, nla);
777 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
778 return PTR_ERR(peer);
781 spin_lock_bh(&net->nsid_lock);
782 if (__peernet2id(net, peer) >= 0) {
783 spin_unlock_bh(&net->nsid_lock);
785 NL_SET_BAD_ATTR(extack, nla);
786 NL_SET_ERR_MSG(extack,
787 "Peer netns already has a nsid assigned");
791 err = alloc_netid(net, peer, nsid);
792 spin_unlock_bh(&net->nsid_lock);
794 rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid,
797 } else if (err == -ENOSPC && nsid >= 0) {
799 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
800 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
807 static int rtnl_net_get_size(void)
809 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
810 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
811 + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
815 struct net_fill_args {
825 static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
827 struct nlmsghdr *nlh;
828 struct rtgenmsg *rth;
830 nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
835 rth = nlmsg_data(nlh);
836 rth->rtgen_family = AF_UNSPEC;
838 if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
839 goto nla_put_failure;
842 nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
843 goto nla_put_failure;
849 nlmsg_cancel(skb, nlh);
853 static int rtnl_net_valid_getid_req(struct sk_buff *skb,
854 const struct nlmsghdr *nlh,
856 struct netlink_ext_ack *extack)
860 if (!netlink_strict_get_check(skb))
861 return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
862 tb, NETNSA_MAX, rtnl_net_policy,
865 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
866 NETNSA_MAX, rtnl_net_policy,
871 for (i = 0; i <= NETNSA_MAX; i++) {
879 case NETNSA_TARGET_NSID:
882 NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
890 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
891 struct netlink_ext_ack *extack)
893 struct net *net = sock_net(skb->sk);
894 struct nlattr *tb[NETNSA_MAX + 1];
895 struct net_fill_args fillargs = {
896 .portid = NETLINK_CB(skb).portid,
897 .seq = nlh->nlmsg_seq,
900 struct net *peer, *target = net;
905 err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
908 if (tb[NETNSA_PID]) {
909 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
910 nla = tb[NETNSA_PID];
911 } else if (tb[NETNSA_FD]) {
912 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
914 } else if (tb[NETNSA_NSID]) {
915 peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
917 peer = ERR_PTR(-ENOENT);
918 nla = tb[NETNSA_NSID];
920 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
925 NL_SET_BAD_ATTR(extack, nla);
926 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
927 return PTR_ERR(peer);
930 if (tb[NETNSA_TARGET_NSID]) {
931 int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
933 target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
934 if (IS_ERR(target)) {
935 NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
936 NL_SET_ERR_MSG(extack,
937 "Target netns reference is invalid");
938 err = PTR_ERR(target);
941 fillargs.add_ref = true;
942 fillargs.ref_nsid = peernet2id(net, peer);
945 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
951 fillargs.nsid = peernet2id(target, peer);
952 err = rtnl_net_fill(msg, &fillargs);
956 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
962 if (fillargs.add_ref)
968 struct rtnl_net_dump_cb {
972 struct net_fill_args fillargs;
977 /* Runs in RCU-critical section. */
978 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
980 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
983 if (net_cb->idx < net_cb->s_idx)
986 net_cb->fillargs.nsid = id;
987 if (net_cb->fillargs.add_ref)
988 net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
989 ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
998 static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
999 struct rtnl_net_dump_cb *net_cb,
1000 struct netlink_callback *cb)
1002 struct netlink_ext_ack *extack = cb->extack;
1003 struct nlattr *tb[NETNSA_MAX + 1];
1006 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
1007 NETNSA_MAX, rtnl_net_policy,
1012 for (i = 0; i <= NETNSA_MAX; i++) {
1016 if (i == NETNSA_TARGET_NSID) {
1019 net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
1021 NL_SET_BAD_ATTR(extack, tb[i]);
1022 NL_SET_ERR_MSG(extack,
1023 "Invalid target network namespace id");
1024 return PTR_ERR(net);
1026 net_cb->fillargs.add_ref = true;
1027 net_cb->ref_net = net_cb->tgt_net;
1028 net_cb->tgt_net = net;
1030 NL_SET_BAD_ATTR(extack, tb[i]);
1031 NL_SET_ERR_MSG(extack,
1032 "Unsupported attribute in dump request");
1040 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1042 struct rtnl_net_dump_cb net_cb = {
1043 .tgt_net = sock_net(skb->sk),
1046 .portid = NETLINK_CB(cb->skb).portid,
1047 .seq = cb->nlh->nlmsg_seq,
1048 .flags = NLM_F_MULTI,
1052 .s_idx = cb->args[0],
1056 if (cb->strict_check) {
1057 err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1063 idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1066 cb->args[0] = net_cb.idx;
1068 if (net_cb.fillargs.add_ref)
1069 put_net(net_cb.tgt_net);
1070 return err < 0 ? err : skb->len;
1073 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1074 struct nlmsghdr *nlh, gfp_t gfp)
1076 struct net_fill_args fillargs = {
1078 .seq = nlh ? nlh->nlmsg_seq : 0,
1082 struct sk_buff *msg;
1085 msg = nlmsg_new(rtnl_net_get_size(), gfp);
1089 err = rtnl_net_fill(msg, &fillargs);
1093 rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, gfp);
1099 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1102 void __init net_ns_init(void)
1104 struct net_generic *ng;
1106 #ifdef CONFIG_NET_NS
1107 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1109 SLAB_PANIC|SLAB_ACCOUNT, NULL);
1111 /* Create workqueue for cleanup */
1112 netns_wq = create_singlethread_workqueue("netns");
1114 panic("Could not create netns workq");
1117 ng = net_alloc_generic();
1119 panic("Could not allocate generic netns");
1121 rcu_assign_pointer(init_net.gen, ng);
1124 init_net.key_domain = &init_net_key_domain;
1126 down_write(&pernet_ops_rwsem);
1127 preinit_net(&init_net);
1128 if (setup_net(&init_net, &init_user_ns))
1129 panic("Could not setup the initial network namespace");
1131 init_net_initialized = true;
1132 up_write(&pernet_ops_rwsem);
1134 if (register_pernet_subsys(&net_ns_ops))
1135 panic("Could not register network namespace subsystems");
1137 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
1138 RTNL_FLAG_DOIT_UNLOCKED);
1139 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
1140 RTNL_FLAG_DOIT_UNLOCKED);
1143 static void free_exit_list(struct pernet_operations *ops, struct list_head *net_exit_list)
1145 ops_pre_exit_list(ops, net_exit_list);
1147 ops_exit_list(ops, net_exit_list);
1148 ops_free_list(ops, net_exit_list);
1151 #ifdef CONFIG_NET_NS
1152 static int __register_pernet_operations(struct list_head *list,
1153 struct pernet_operations *ops)
1157 LIST_HEAD(net_exit_list);
1159 list_add_tail(&ops->list, list);
1160 if (ops->init || (ops->id && ops->size)) {
1161 /* We held write locked pernet_ops_rwsem, and parallel
1162 * setup_net() and cleanup_net() are not possible.
1165 error = ops_init(ops, net);
1168 list_add_tail(&net->exit_list, &net_exit_list);
1174 /* If I have an error cleanup all namespaces I initialized */
1175 list_del(&ops->list);
1176 free_exit_list(ops, &net_exit_list);
1180 static void __unregister_pernet_operations(struct pernet_operations *ops)
1183 LIST_HEAD(net_exit_list);
1185 list_del(&ops->list);
1186 /* See comment in __register_pernet_operations() */
1188 list_add_tail(&net->exit_list, &net_exit_list);
1190 free_exit_list(ops, &net_exit_list);
1195 static int __register_pernet_operations(struct list_head *list,
1196 struct pernet_operations *ops)
1198 if (!init_net_initialized) {
1199 list_add_tail(&ops->list, list);
1203 return ops_init(ops, &init_net);
1206 static void __unregister_pernet_operations(struct pernet_operations *ops)
1208 if (!init_net_initialized) {
1209 list_del(&ops->list);
1211 LIST_HEAD(net_exit_list);
1212 list_add(&init_net.exit_list, &net_exit_list);
1213 free_exit_list(ops, &net_exit_list);
1217 #endif /* CONFIG_NET_NS */
1219 static DEFINE_IDA(net_generic_ids);
1221 static int register_pernet_operations(struct list_head *list,
1222 struct pernet_operations *ops)
1227 error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID,
1232 max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
1234 error = __register_pernet_operations(list, ops);
1238 ida_free(&net_generic_ids, *ops->id);
1244 static void unregister_pernet_operations(struct pernet_operations *ops)
1246 __unregister_pernet_operations(ops);
1249 ida_free(&net_generic_ids, *ops->id);
1253 * register_pernet_subsys - register a network namespace subsystem
1254 * @ops: pernet operations structure for the subsystem
1256 * Register a subsystem which has init and exit functions
1257 * that are called when network namespaces are created and
1258 * destroyed respectively.
1260 * When registered all network namespace init functions are
1261 * called for every existing network namespace. Allowing kernel
1262 * modules to have a race free view of the set of network namespaces.
1264 * When a new network namespace is created all of the init
1265 * methods are called in the order in which they were registered.
1267 * When a network namespace is destroyed all of the exit methods
1268 * are called in the reverse of the order with which they were
1271 int register_pernet_subsys(struct pernet_operations *ops)
1274 down_write(&pernet_ops_rwsem);
1275 error = register_pernet_operations(first_device, ops);
1276 up_write(&pernet_ops_rwsem);
1279 EXPORT_SYMBOL_GPL(register_pernet_subsys);
1282 * unregister_pernet_subsys - unregister a network namespace subsystem
1283 * @ops: pernet operations structure to manipulate
1285 * Remove the pernet operations structure from the list to be
1286 * used when network namespaces are created or destroyed. In
1287 * addition run the exit method for all existing network
1290 void unregister_pernet_subsys(struct pernet_operations *ops)
1292 down_write(&pernet_ops_rwsem);
1293 unregister_pernet_operations(ops);
1294 up_write(&pernet_ops_rwsem);
1296 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1299 * register_pernet_device - register a network namespace device
1300 * @ops: pernet operations structure for the subsystem
1302 * Register a device which has init and exit functions
1303 * that are called when network namespaces are created and
1304 * destroyed respectively.
1306 * When registered all network namespace init functions are
1307 * called for every existing network namespace. Allowing kernel
1308 * modules to have a race free view of the set of network namespaces.
1310 * When a new network namespace is created all of the init
1311 * methods are called in the order in which they were registered.
1313 * When a network namespace is destroyed all of the exit methods
1314 * are called in the reverse of the order with which they were
1317 int register_pernet_device(struct pernet_operations *ops)
1320 down_write(&pernet_ops_rwsem);
1321 error = register_pernet_operations(&pernet_list, ops);
1322 if (!error && (first_device == &pernet_list))
1323 first_device = &ops->list;
1324 up_write(&pernet_ops_rwsem);
1327 EXPORT_SYMBOL_GPL(register_pernet_device);
1330 * unregister_pernet_device - unregister a network namespace netdevice
1331 * @ops: pernet operations structure to manipulate
1333 * Remove the pernet operations structure from the list to be
1334 * used when network namespaces are created or destroyed. In
1335 * addition run the exit method for all existing network
1338 void unregister_pernet_device(struct pernet_operations *ops)
1340 down_write(&pernet_ops_rwsem);
1341 if (&ops->list == first_device)
1342 first_device = first_device->next;
1343 unregister_pernet_operations(ops);
1344 up_write(&pernet_ops_rwsem);
1346 EXPORT_SYMBOL_GPL(unregister_pernet_device);
1348 #ifdef CONFIG_NET_NS
1349 static struct ns_common *netns_get(struct task_struct *task)
1351 struct net *net = NULL;
1352 struct nsproxy *nsproxy;
1355 nsproxy = task->nsproxy;
1357 net = get_net(nsproxy->net_ns);
1360 return net ? &net->ns : NULL;
1363 static inline struct net *to_net_ns(struct ns_common *ns)
1365 return container_of(ns, struct net, ns);
1368 static void netns_put(struct ns_common *ns)
1370 put_net(to_net_ns(ns));
1373 static int netns_install(struct nsset *nsset, struct ns_common *ns)
1375 struct nsproxy *nsproxy = nsset->nsproxy;
1376 struct net *net = to_net_ns(ns);
1378 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1379 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
1382 put_net(nsproxy->net_ns);
1383 nsproxy->net_ns = get_net(net);
1387 static struct user_namespace *netns_owner(struct ns_common *ns)
1389 return to_net_ns(ns)->user_ns;
1392 const struct proc_ns_operations netns_operations = {
1394 .type = CLONE_NEWNET,
1397 .install = netns_install,
1398 .owner = netns_owner,