1 // SPDX-License-Identifier: GPL-2.0-only
2 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
4 #include <linux/workqueue.h>
5 #include <linux/rtnetlink.h>
6 #include <linux/cache.h>
7 #include <linux/slab.h>
8 #include <linux/list.h>
9 #include <linux/delay.h>
10 #include <linux/sched.h>
11 #include <linux/idr.h>
12 #include <linux/rculist.h>
13 #include <linux/nsproxy.h>
15 #include <linux/proc_ns.h>
16 #include <linux/file.h>
17 #include <linux/export.h>
18 #include <linux/user_namespace.h>
19 #include <linux/net_namespace.h>
20 #include <linux/sched/task.h>
21 #include <linux/uidgid.h>
22 #include <linux/cookie.h>
25 #include <net/netlink.h>
26 #include <net/net_namespace.h>
27 #include <net/netns/generic.h>
30 * Our network namespace constructor/destructor lists
33 static LIST_HEAD(pernet_list);
34 static struct list_head *first_device = &pernet_list;
36 LIST_HEAD(net_namespace_list);
37 EXPORT_SYMBOL_GPL(net_namespace_list);
39 /* Protects net_namespace_list. Nests iside rtnl_lock() */
40 DECLARE_RWSEM(net_rwsem);
41 EXPORT_SYMBOL_GPL(net_rwsem);
44 static struct key_tag init_net_key_domain = { .usage = REFCOUNT_INIT(1) };
47 struct net init_net = {
48 .count = REFCOUNT_INIT(1),
49 .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head),
51 .key_domain = &init_net_key_domain,
54 EXPORT_SYMBOL(init_net);
56 static bool init_net_initialized;
58 * pernet_ops_rwsem: protects: pernet_list, net_generic_ids,
59 * init_net_initialized and first_device pointer.
60 * This is internal net namespace object. Please, don't use it
63 DECLARE_RWSEM(pernet_ops_rwsem);
64 EXPORT_SYMBOL_GPL(pernet_ops_rwsem);
66 #define MIN_PERNET_OPS_ID \
67 ((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *))
69 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
71 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
73 DEFINE_COOKIE(net_cookie);
75 u64 __net_gen_cookie(struct net *net)
78 u64 res = atomic64_read(&net->net_cookie);
82 res = gen_cookie_next(&net_cookie);
83 atomic64_cmpxchg(&net->net_cookie, 0, res);
87 static struct net_generic *net_alloc_generic(void)
89 struct net_generic *ng;
90 unsigned int generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
92 ng = kzalloc(generic_size, GFP_KERNEL);
94 ng->s.len = max_gen_ptrs;
99 static int net_assign_generic(struct net *net, unsigned int id, void *data)
101 struct net_generic *ng, *old_ng;
103 BUG_ON(id < MIN_PERNET_OPS_ID);
105 old_ng = rcu_dereference_protected(net->gen,
106 lockdep_is_held(&pernet_ops_rwsem));
107 if (old_ng->s.len > id) {
108 old_ng->ptr[id] = data;
112 ng = net_alloc_generic();
117 * Some synchronisation notes:
119 * The net_generic explores the net->gen array inside rcu
120 * read section. Besides once set the net->gen->ptr[x]
121 * pointer never changes (see rules in netns/generic.h).
123 * That said, we simply duplicate this array and schedule
124 * the old copy for kfree after a grace period.
127 memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
128 (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
131 rcu_assign_pointer(net->gen, ng);
132 kfree_rcu(old_ng, s.rcu);
136 static int ops_init(const struct pernet_operations *ops, struct net *net)
138 struct net_generic *ng;
142 if (ops->id && ops->size) {
143 data = kzalloc(ops->size, GFP_KERNEL);
147 err = net_assign_generic(net, *ops->id, data);
153 err = ops->init(net);
157 if (ops->id && ops->size) {
159 ng = rcu_dereference_protected(net->gen,
160 lockdep_is_held(&pernet_ops_rwsem));
161 ng->ptr[*ops->id] = NULL;
170 static void ops_free(const struct pernet_operations *ops, struct net *net)
172 if (ops->id && ops->size) {
173 kfree(net_generic(net, *ops->id));
177 static void ops_pre_exit_list(const struct pernet_operations *ops,
178 struct list_head *net_exit_list)
183 list_for_each_entry(net, net_exit_list, exit_list)
188 static void ops_exit_list(const struct pernet_operations *ops,
189 struct list_head *net_exit_list)
193 list_for_each_entry(net, net_exit_list, exit_list) {
199 ops->exit_batch(net_exit_list);
202 static void ops_free_list(const struct pernet_operations *ops,
203 struct list_head *net_exit_list)
206 if (ops->size && ops->id) {
207 list_for_each_entry(net, net_exit_list, exit_list)
212 /* should be called with nsid_lock held */
213 static int alloc_netid(struct net *net, struct net *peer, int reqid)
215 int min = 0, max = 0;
222 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
225 /* This function is used by idr_for_each(). If net is equal to peer, the
226 * function returns the id so that idr_for_each() stops. Because we cannot
227 * returns the id 0 (idr_for_each() will not stop), we return the magic value
228 * NET_ID_ZERO (-1) for it.
230 #define NET_ID_ZERO -1
231 static int net_eq_idr(int id, void *net, void *peer)
233 if (net_eq(net, peer))
234 return id ? : NET_ID_ZERO;
238 /* Must be called from RCU-critical section or with nsid_lock held */
239 static int __peernet2id(const struct net *net, struct net *peer)
241 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
243 /* Magic value for id 0. */
244 if (id == NET_ID_ZERO)
249 return NETNSA_NSID_NOT_ASSIGNED;
252 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
253 struct nlmsghdr *nlh, gfp_t gfp);
254 /* This function returns the id of a peer netns. If no id is assigned, one will
255 * be allocated and returned.
257 int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp)
261 if (refcount_read(&net->count) == 0)
262 return NETNSA_NSID_NOT_ASSIGNED;
264 spin_lock_bh(&net->nsid_lock);
265 id = __peernet2id(net, peer);
267 spin_unlock_bh(&net->nsid_lock);
271 /* When peer is obtained from RCU lists, we may race with
272 * its cleanup. Check whether it's alive, and this guarantees
273 * we never hash a peer back to net->netns_ids, after it has
274 * just been idr_remove()'d from there in cleanup_net().
276 if (!maybe_get_net(peer)) {
277 spin_unlock_bh(&net->nsid_lock);
278 return NETNSA_NSID_NOT_ASSIGNED;
281 id = alloc_netid(net, peer, -1);
282 spin_unlock_bh(&net->nsid_lock);
286 return NETNSA_NSID_NOT_ASSIGNED;
288 rtnl_net_notifyid(net, RTM_NEWNSID, id, 0, NULL, gfp);
292 EXPORT_SYMBOL_GPL(peernet2id_alloc);
294 /* This function returns, if assigned, the id of a peer netns. */
295 int peernet2id(const struct net *net, struct net *peer)
300 id = __peernet2id(net, peer);
305 EXPORT_SYMBOL(peernet2id);
307 /* This function returns true is the peer netns has an id assigned into the
310 bool peernet_has_id(const struct net *net, struct net *peer)
312 return peernet2id(net, peer) >= 0;
315 struct net *get_net_ns_by_id(const struct net *net, int id)
323 peer = idr_find(&net->netns_ids, id);
325 peer = maybe_get_net(peer);
332 * setup_net runs the initializers for the network namespace object.
334 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
336 /* Must be called with pernet_ops_rwsem held */
337 const struct pernet_operations *ops, *saved_ops;
339 LIST_HEAD(net_exit_list);
341 refcount_set(&net->count, 1);
342 refcount_set(&net->passive, 1);
343 get_random_bytes(&net->hash_mix, sizeof(u32));
344 net->dev_base_seq = 1;
345 net->user_ns = user_ns;
346 idr_init(&net->netns_ids);
347 spin_lock_init(&net->nsid_lock);
348 mutex_init(&net->ipv4.ra_mutex);
350 list_for_each_entry(ops, &pernet_list, list) {
351 error = ops_init(ops, net);
355 down_write(&net_rwsem);
356 list_add_tail_rcu(&net->list, &net_namespace_list);
357 up_write(&net_rwsem);
362 /* Walk through the list backwards calling the exit functions
363 * for the pernet modules whose init functions did not fail.
365 list_add(&net->exit_list, &net_exit_list);
367 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
368 ops_pre_exit_list(ops, &net_exit_list);
373 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
374 ops_exit_list(ops, &net_exit_list);
377 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
378 ops_free_list(ops, &net_exit_list);
384 static int __net_init net_defaults_init_net(struct net *net)
386 net->core.sysctl_somaxconn = SOMAXCONN;
390 static struct pernet_operations net_defaults_ops = {
391 .init = net_defaults_init_net,
394 static __init int net_defaults_init(void)
396 if (register_pernet_subsys(&net_defaults_ops))
397 panic("Cannot initialize net default settings");
402 core_initcall(net_defaults_init);
405 static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
407 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
410 static void dec_net_namespaces(struct ucounts *ucounts)
412 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
415 static struct kmem_cache *net_cachep __ro_after_init;
416 static struct workqueue_struct *netns_wq;
418 static struct net *net_alloc(void)
420 struct net *net = NULL;
421 struct net_generic *ng;
423 ng = net_alloc_generic();
427 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
432 net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
433 if (!net->key_domain)
435 refcount_set(&net->key_domain->usage, 1);
438 rcu_assign_pointer(net->gen, ng);
444 kmem_cache_free(net_cachep, net);
452 static void net_free(struct net *net)
454 kfree(rcu_access_pointer(net->gen));
455 kmem_cache_free(net_cachep, net);
458 void net_drop_ns(void *p)
461 if (ns && refcount_dec_and_test(&ns->passive))
465 struct net *copy_net_ns(unsigned long flags,
466 struct user_namespace *user_ns, struct net *old_net)
468 struct ucounts *ucounts;
472 if (!(flags & CLONE_NEWNET))
473 return get_net(old_net);
475 ucounts = inc_net_namespaces(user_ns);
477 return ERR_PTR(-ENOSPC);
484 refcount_set(&net->passive, 1);
485 net->ucounts = ucounts;
486 get_user_ns(user_ns);
488 rv = down_read_killable(&pernet_ops_rwsem);
492 rv = setup_net(net, user_ns);
494 up_read(&pernet_ops_rwsem);
499 key_remove_domain(net->key_domain);
501 put_user_ns(user_ns);
504 dec_net_namespaces(ucounts);
511 * net_ns_get_ownership - get sysfs ownership data for @net
512 * @net: network namespace in question (can be NULL)
513 * @uid: kernel user ID for sysfs objects
514 * @gid: kernel group ID for sysfs objects
516 * Returns the uid/gid pair of root in the user namespace associated with the
517 * given network namespace.
519 void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
522 kuid_t ns_root_uid = make_kuid(net->user_ns, 0);
523 kgid_t ns_root_gid = make_kgid(net->user_ns, 0);
525 if (uid_valid(ns_root_uid))
528 if (gid_valid(ns_root_gid))
531 *uid = GLOBAL_ROOT_UID;
532 *gid = GLOBAL_ROOT_GID;
535 EXPORT_SYMBOL_GPL(net_ns_get_ownership);
537 static void unhash_nsid(struct net *net, struct net *last)
540 /* This function is only called from cleanup_net() work,
541 * and this work is the only process, that may delete
542 * a net from net_namespace_list. So, when the below
543 * is executing, the list may only grow. Thus, we do not
544 * use for_each_net_rcu() or net_rwsem.
549 spin_lock_bh(&tmp->nsid_lock);
550 id = __peernet2id(tmp, net);
552 idr_remove(&tmp->netns_ids, id);
553 spin_unlock_bh(&tmp->nsid_lock);
555 rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL,
560 spin_lock_bh(&net->nsid_lock);
561 idr_destroy(&net->netns_ids);
562 spin_unlock_bh(&net->nsid_lock);
565 static LLIST_HEAD(cleanup_list);
567 static void cleanup_net(struct work_struct *work)
569 const struct pernet_operations *ops;
570 struct net *net, *tmp, *last;
571 struct llist_node *net_kill_list;
572 LIST_HEAD(net_exit_list);
574 /* Atomically snapshot the list of namespaces to cleanup */
575 net_kill_list = llist_del_all(&cleanup_list);
577 down_read(&pernet_ops_rwsem);
579 /* Don't let anyone else find us. */
580 down_write(&net_rwsem);
581 llist_for_each_entry(net, net_kill_list, cleanup_list)
582 list_del_rcu(&net->list);
583 /* Cache last net. After we unlock rtnl, no one new net
584 * added to net_namespace_list can assign nsid pointer
585 * to a net from net_kill_list (see peernet2id_alloc()).
586 * So, we skip them in unhash_nsid().
588 * Note, that unhash_nsid() does not delete nsid links
589 * between net_kill_list's nets, as they've already
590 * deleted from net_namespace_list. But, this would be
591 * useless anyway, as netns_ids are destroyed there.
593 last = list_last_entry(&net_namespace_list, struct net, list);
594 up_write(&net_rwsem);
596 llist_for_each_entry(net, net_kill_list, cleanup_list) {
597 unhash_nsid(net, last);
598 list_add_tail(&net->exit_list, &net_exit_list);
601 /* Run all of the network namespace pre_exit methods */
602 list_for_each_entry_reverse(ops, &pernet_list, list)
603 ops_pre_exit_list(ops, &net_exit_list);
606 * Another CPU might be rcu-iterating the list, wait for it.
607 * This needs to be before calling the exit() notifiers, so
608 * the rcu_barrier() below isn't sufficient alone.
609 * Also the pre_exit() and exit() methods need this barrier.
613 /* Run all of the network namespace exit methods */
614 list_for_each_entry_reverse(ops, &pernet_list, list)
615 ops_exit_list(ops, &net_exit_list);
617 /* Free the net generic variables */
618 list_for_each_entry_reverse(ops, &pernet_list, list)
619 ops_free_list(ops, &net_exit_list);
621 up_read(&pernet_ops_rwsem);
623 /* Ensure there are no outstanding rcu callbacks using this
628 /* Finally it is safe to free my network namespace structure */
629 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
630 list_del_init(&net->exit_list);
631 dec_net_namespaces(net->ucounts);
633 key_remove_domain(net->key_domain);
635 put_user_ns(net->user_ns);
641 * net_ns_barrier - wait until concurrent net_cleanup_work is done
643 * cleanup_net runs from work queue and will first remove namespaces
644 * from the global list, then run net exit functions.
646 * Call this in module exit path to make sure that all netns
647 * ->exit ops have been invoked before the function is removed.
649 void net_ns_barrier(void)
651 down_write(&pernet_ops_rwsem);
652 up_write(&pernet_ops_rwsem);
654 EXPORT_SYMBOL(net_ns_barrier);
656 static DECLARE_WORK(net_cleanup_work, cleanup_net);
658 void __put_net(struct net *net)
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)
681 struct ns_common *ns;
684 file = proc_ns_fget(fd);
686 return ERR_CAST(file);
688 ns = get_proc_ns(file_inode(file));
689 if (ns->ops == &netns_operations)
690 net = get_net(container_of(ns, struct net, ns));
692 net = ERR_PTR(-EINVAL);
699 struct net *get_net_ns_by_fd(int fd)
701 return ERR_PTR(-EINVAL);
704 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
706 struct net *get_net_ns_by_pid(pid_t pid)
708 struct task_struct *tsk;
711 /* Lookup the network namespace */
712 net = ERR_PTR(-ESRCH);
714 tsk = find_task_by_vpid(pid);
716 struct nsproxy *nsproxy;
718 nsproxy = tsk->nsproxy;
720 net = get_net(nsproxy->net_ns);
726 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
728 static __net_init int net_ns_net_init(struct net *net)
731 net->ns.ops = &netns_operations;
733 return ns_alloc_inum(&net->ns);
736 static __net_exit void net_ns_net_exit(struct net *net)
738 ns_free_inum(&net->ns);
741 static struct pernet_operations __net_initdata net_ns_ops = {
742 .init = net_ns_net_init,
743 .exit = net_ns_net_exit,
746 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
747 [NETNSA_NONE] = { .type = NLA_UNSPEC },
748 [NETNSA_NSID] = { .type = NLA_S32 },
749 [NETNSA_PID] = { .type = NLA_U32 },
750 [NETNSA_FD] = { .type = NLA_U32 },
751 [NETNSA_TARGET_NSID] = { .type = NLA_S32 },
754 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
755 struct netlink_ext_ack *extack)
757 struct net *net = sock_net(skb->sk);
758 struct nlattr *tb[NETNSA_MAX + 1];
763 err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
764 NETNSA_MAX, rtnl_net_policy, extack);
767 if (!tb[NETNSA_NSID]) {
768 NL_SET_ERR_MSG(extack, "nsid is missing");
771 nsid = nla_get_s32(tb[NETNSA_NSID]);
773 if (tb[NETNSA_PID]) {
774 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
775 nla = tb[NETNSA_PID];
776 } else if (tb[NETNSA_FD]) {
777 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
780 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
784 NL_SET_BAD_ATTR(extack, nla);
785 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
786 return PTR_ERR(peer);
789 spin_lock_bh(&net->nsid_lock);
790 if (__peernet2id(net, peer) >= 0) {
791 spin_unlock_bh(&net->nsid_lock);
793 NL_SET_BAD_ATTR(extack, nla);
794 NL_SET_ERR_MSG(extack,
795 "Peer netns already has a nsid assigned");
799 err = alloc_netid(net, peer, nsid);
800 spin_unlock_bh(&net->nsid_lock);
802 rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid,
805 } else if (err == -ENOSPC && nsid >= 0) {
807 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
808 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
815 static int rtnl_net_get_size(void)
817 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
818 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
819 + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
823 struct net_fill_args {
833 static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
835 struct nlmsghdr *nlh;
836 struct rtgenmsg *rth;
838 nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
843 rth = nlmsg_data(nlh);
844 rth->rtgen_family = AF_UNSPEC;
846 if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
847 goto nla_put_failure;
850 nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
851 goto nla_put_failure;
857 nlmsg_cancel(skb, nlh);
861 static int rtnl_net_valid_getid_req(struct sk_buff *skb,
862 const struct nlmsghdr *nlh,
864 struct netlink_ext_ack *extack)
868 if (!netlink_strict_get_check(skb))
869 return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
870 tb, NETNSA_MAX, rtnl_net_policy,
873 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
874 NETNSA_MAX, rtnl_net_policy,
879 for (i = 0; i <= NETNSA_MAX; i++) {
887 case NETNSA_TARGET_NSID:
890 NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
898 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
899 struct netlink_ext_ack *extack)
901 struct net *net = sock_net(skb->sk);
902 struct nlattr *tb[NETNSA_MAX + 1];
903 struct net_fill_args fillargs = {
904 .portid = NETLINK_CB(skb).portid,
905 .seq = nlh->nlmsg_seq,
908 struct net *peer, *target = net;
913 err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
916 if (tb[NETNSA_PID]) {
917 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
918 nla = tb[NETNSA_PID];
919 } else if (tb[NETNSA_FD]) {
920 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
922 } else if (tb[NETNSA_NSID]) {
923 peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
925 peer = ERR_PTR(-ENOENT);
926 nla = tb[NETNSA_NSID];
928 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
933 NL_SET_BAD_ATTR(extack, nla);
934 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
935 return PTR_ERR(peer);
938 if (tb[NETNSA_TARGET_NSID]) {
939 int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
941 target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
942 if (IS_ERR(target)) {
943 NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
944 NL_SET_ERR_MSG(extack,
945 "Target netns reference is invalid");
946 err = PTR_ERR(target);
949 fillargs.add_ref = true;
950 fillargs.ref_nsid = peernet2id(net, peer);
953 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
959 fillargs.nsid = peernet2id(target, peer);
960 err = rtnl_net_fill(msg, &fillargs);
964 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
970 if (fillargs.add_ref)
976 struct rtnl_net_dump_cb {
980 struct net_fill_args fillargs;
985 /* Runs in RCU-critical section. */
986 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
988 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
991 if (net_cb->idx < net_cb->s_idx)
994 net_cb->fillargs.nsid = id;
995 if (net_cb->fillargs.add_ref)
996 net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
997 ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
1006 static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
1007 struct rtnl_net_dump_cb *net_cb,
1008 struct netlink_callback *cb)
1010 struct netlink_ext_ack *extack = cb->extack;
1011 struct nlattr *tb[NETNSA_MAX + 1];
1014 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
1015 NETNSA_MAX, rtnl_net_policy,
1020 for (i = 0; i <= NETNSA_MAX; i++) {
1024 if (i == NETNSA_TARGET_NSID) {
1027 net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
1029 NL_SET_BAD_ATTR(extack, tb[i]);
1030 NL_SET_ERR_MSG(extack,
1031 "Invalid target network namespace id");
1032 return PTR_ERR(net);
1034 net_cb->fillargs.add_ref = true;
1035 net_cb->ref_net = net_cb->tgt_net;
1036 net_cb->tgt_net = net;
1038 NL_SET_BAD_ATTR(extack, tb[i]);
1039 NL_SET_ERR_MSG(extack,
1040 "Unsupported attribute in dump request");
1048 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1050 struct rtnl_net_dump_cb net_cb = {
1051 .tgt_net = sock_net(skb->sk),
1054 .portid = NETLINK_CB(cb->skb).portid,
1055 .seq = cb->nlh->nlmsg_seq,
1056 .flags = NLM_F_MULTI,
1060 .s_idx = cb->args[0],
1064 if (cb->strict_check) {
1065 err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1071 idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1074 cb->args[0] = net_cb.idx;
1076 if (net_cb.fillargs.add_ref)
1077 put_net(net_cb.tgt_net);
1078 return err < 0 ? err : skb->len;
1081 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1082 struct nlmsghdr *nlh, gfp_t gfp)
1084 struct net_fill_args fillargs = {
1086 .seq = nlh ? nlh->nlmsg_seq : 0,
1090 struct sk_buff *msg;
1093 msg = nlmsg_new(rtnl_net_get_size(), gfp);
1097 err = rtnl_net_fill(msg, &fillargs);
1101 rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, gfp);
1107 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1110 static int __init net_ns_init(void)
1112 struct net_generic *ng;
1114 #ifdef CONFIG_NET_NS
1115 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1117 SLAB_PANIC|SLAB_ACCOUNT, NULL);
1119 /* Create workqueue for cleanup */
1120 netns_wq = create_singlethread_workqueue("netns");
1122 panic("Could not create netns workq");
1125 ng = net_alloc_generic();
1127 panic("Could not allocate generic netns");
1129 rcu_assign_pointer(init_net.gen, ng);
1132 __net_gen_cookie(&init_net);
1135 down_write(&pernet_ops_rwsem);
1136 if (setup_net(&init_net, &init_user_ns))
1137 panic("Could not setup the initial network namespace");
1139 init_net_initialized = true;
1140 up_write(&pernet_ops_rwsem);
1142 if (register_pernet_subsys(&net_ns_ops))
1143 panic("Could not register network namespace subsystems");
1145 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
1146 RTNL_FLAG_DOIT_UNLOCKED);
1147 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
1148 RTNL_FLAG_DOIT_UNLOCKED);
1153 pure_initcall(net_ns_init);
1155 #ifdef CONFIG_NET_NS
1156 static int __register_pernet_operations(struct list_head *list,
1157 struct pernet_operations *ops)
1161 LIST_HEAD(net_exit_list);
1163 list_add_tail(&ops->list, list);
1164 if (ops->init || (ops->id && ops->size)) {
1165 /* We held write locked pernet_ops_rwsem, and parallel
1166 * setup_net() and cleanup_net() are not possible.
1169 error = ops_init(ops, net);
1172 list_add_tail(&net->exit_list, &net_exit_list);
1178 /* If I have an error cleanup all namespaces I initialized */
1179 list_del(&ops->list);
1180 ops_pre_exit_list(ops, &net_exit_list);
1182 ops_exit_list(ops, &net_exit_list);
1183 ops_free_list(ops, &net_exit_list);
1187 static void __unregister_pernet_operations(struct pernet_operations *ops)
1190 LIST_HEAD(net_exit_list);
1192 list_del(&ops->list);
1193 /* See comment in __register_pernet_operations() */
1195 list_add_tail(&net->exit_list, &net_exit_list);
1196 ops_pre_exit_list(ops, &net_exit_list);
1198 ops_exit_list(ops, &net_exit_list);
1199 ops_free_list(ops, &net_exit_list);
1204 static int __register_pernet_operations(struct list_head *list,
1205 struct pernet_operations *ops)
1207 if (!init_net_initialized) {
1208 list_add_tail(&ops->list, list);
1212 return ops_init(ops, &init_net);
1215 static void __unregister_pernet_operations(struct pernet_operations *ops)
1217 if (!init_net_initialized) {
1218 list_del(&ops->list);
1220 LIST_HEAD(net_exit_list);
1221 list_add(&init_net.exit_list, &net_exit_list);
1222 ops_pre_exit_list(ops, &net_exit_list);
1224 ops_exit_list(ops, &net_exit_list);
1225 ops_free_list(ops, &net_exit_list);
1229 #endif /* CONFIG_NET_NS */
1231 static DEFINE_IDA(net_generic_ids);
1233 static int register_pernet_operations(struct list_head *list,
1234 struct pernet_operations *ops)
1239 error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID,
1244 max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
1246 error = __register_pernet_operations(list, ops);
1250 ida_free(&net_generic_ids, *ops->id);
1256 static void unregister_pernet_operations(struct pernet_operations *ops)
1258 __unregister_pernet_operations(ops);
1261 ida_free(&net_generic_ids, *ops->id);
1265 * register_pernet_subsys - register a network namespace subsystem
1266 * @ops: pernet operations structure for the subsystem
1268 * Register a subsystem which has init and exit functions
1269 * that are called when network namespaces are created and
1270 * destroyed respectively.
1272 * When registered all network namespace init functions are
1273 * called for every existing network namespace. Allowing kernel
1274 * modules to have a race free view of the set of network namespaces.
1276 * When a new network namespace is created all of the init
1277 * methods are called in the order in which they were registered.
1279 * When a network namespace is destroyed all of the exit methods
1280 * are called in the reverse of the order with which they were
1283 int register_pernet_subsys(struct pernet_operations *ops)
1286 down_write(&pernet_ops_rwsem);
1287 error = register_pernet_operations(first_device, ops);
1288 up_write(&pernet_ops_rwsem);
1291 EXPORT_SYMBOL_GPL(register_pernet_subsys);
1294 * unregister_pernet_subsys - unregister a network namespace subsystem
1295 * @ops: pernet operations structure to manipulate
1297 * Remove the pernet operations structure from the list to be
1298 * used when network namespaces are created or destroyed. In
1299 * addition run the exit method for all existing network
1302 void unregister_pernet_subsys(struct pernet_operations *ops)
1304 down_write(&pernet_ops_rwsem);
1305 unregister_pernet_operations(ops);
1306 up_write(&pernet_ops_rwsem);
1308 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1311 * register_pernet_device - register a network namespace device
1312 * @ops: pernet operations structure for the subsystem
1314 * Register a device which has init and exit functions
1315 * that are called when network namespaces are created and
1316 * destroyed respectively.
1318 * When registered all network namespace init functions are
1319 * called for every existing network namespace. Allowing kernel
1320 * modules to have a race free view of the set of network namespaces.
1322 * When a new network namespace is created all of the init
1323 * methods are called in the order in which they were registered.
1325 * When a network namespace is destroyed all of the exit methods
1326 * are called in the reverse of the order with which they were
1329 int register_pernet_device(struct pernet_operations *ops)
1332 down_write(&pernet_ops_rwsem);
1333 error = register_pernet_operations(&pernet_list, ops);
1334 if (!error && (first_device == &pernet_list))
1335 first_device = &ops->list;
1336 up_write(&pernet_ops_rwsem);
1339 EXPORT_SYMBOL_GPL(register_pernet_device);
1342 * unregister_pernet_device - unregister a network namespace netdevice
1343 * @ops: pernet operations structure to manipulate
1345 * Remove the pernet operations structure from the list to be
1346 * used when network namespaces are created or destroyed. In
1347 * addition run the exit method for all existing network
1350 void unregister_pernet_device(struct pernet_operations *ops)
1352 down_write(&pernet_ops_rwsem);
1353 if (&ops->list == first_device)
1354 first_device = first_device->next;
1355 unregister_pernet_operations(ops);
1356 up_write(&pernet_ops_rwsem);
1358 EXPORT_SYMBOL_GPL(unregister_pernet_device);
1360 #ifdef CONFIG_NET_NS
1361 static struct ns_common *netns_get(struct task_struct *task)
1363 struct net *net = NULL;
1364 struct nsproxy *nsproxy;
1367 nsproxy = task->nsproxy;
1369 net = get_net(nsproxy->net_ns);
1372 return net ? &net->ns : NULL;
1375 static inline struct net *to_net_ns(struct ns_common *ns)
1377 return container_of(ns, struct net, ns);
1380 static void netns_put(struct ns_common *ns)
1382 put_net(to_net_ns(ns));
1385 static int netns_install(struct nsset *nsset, struct ns_common *ns)
1387 struct nsproxy *nsproxy = nsset->nsproxy;
1388 struct net *net = to_net_ns(ns);
1390 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1391 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
1394 put_net(nsproxy->net_ns);
1395 nsproxy->net_ns = get_net(net);
1399 static struct user_namespace *netns_owner(struct ns_common *ns)
1401 return to_net_ns(ns)->user_ns;
1404 const struct proc_ns_operations netns_operations = {
1406 .type = CLONE_NEWNET,
1409 .install = netns_install,
1410 .owner = netns_owner,