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>
24 #include <net/netlink.h>
25 #include <net/net_namespace.h>
26 #include <net/netns/generic.h>
29 * Our network namespace constructor/destructor lists
32 static LIST_HEAD(pernet_list);
33 static struct list_head *first_device = &pernet_list;
35 LIST_HEAD(net_namespace_list);
36 EXPORT_SYMBOL_GPL(net_namespace_list);
38 /* Protects net_namespace_list. Nests iside rtnl_lock() */
39 DECLARE_RWSEM(net_rwsem);
40 EXPORT_SYMBOL_GPL(net_rwsem);
43 static struct key_tag init_net_key_domain = { .usage = REFCOUNT_INIT(1) };
46 struct net init_net = {
47 .count = REFCOUNT_INIT(1),
48 .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head),
50 .key_domain = &init_net_key_domain,
53 EXPORT_SYMBOL(init_net);
55 static bool init_net_initialized;
57 * pernet_ops_rwsem: protects: pernet_list, net_generic_ids,
58 * init_net_initialized and first_device pointer.
59 * This is internal net namespace object. Please, don't use it
62 DECLARE_RWSEM(pernet_ops_rwsem);
63 EXPORT_SYMBOL_GPL(pernet_ops_rwsem);
65 #define MIN_PERNET_OPS_ID \
66 ((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *))
68 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
70 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
72 static struct net_generic *net_alloc_generic(void)
74 struct net_generic *ng;
75 unsigned int generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
77 ng = kzalloc(generic_size, GFP_KERNEL);
79 ng->s.len = max_gen_ptrs;
84 static int net_assign_generic(struct net *net, unsigned int id, void *data)
86 struct net_generic *ng, *old_ng;
88 BUG_ON(id < MIN_PERNET_OPS_ID);
90 old_ng = rcu_dereference_protected(net->gen,
91 lockdep_is_held(&pernet_ops_rwsem));
92 if (old_ng->s.len > id) {
93 old_ng->ptr[id] = data;
97 ng = net_alloc_generic();
102 * Some synchronisation notes:
104 * The net_generic explores the net->gen array inside rcu
105 * read section. Besides once set the net->gen->ptr[x]
106 * pointer never changes (see rules in netns/generic.h).
108 * That said, we simply duplicate this array and schedule
109 * the old copy for kfree after a grace period.
112 memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
113 (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
116 rcu_assign_pointer(net->gen, ng);
117 kfree_rcu(old_ng, s.rcu);
121 static int ops_init(const struct pernet_operations *ops, struct net *net)
126 if (ops->id && ops->size) {
127 data = kzalloc(ops->size, GFP_KERNEL);
131 err = net_assign_generic(net, *ops->id, data);
137 err = ops->init(net);
148 static void ops_free(const struct pernet_operations *ops, struct net *net)
150 if (ops->id && ops->size) {
151 kfree(net_generic(net, *ops->id));
155 static void ops_pre_exit_list(const struct pernet_operations *ops,
156 struct list_head *net_exit_list)
161 list_for_each_entry(net, net_exit_list, exit_list)
166 static void ops_exit_list(const struct pernet_operations *ops,
167 struct list_head *net_exit_list)
171 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)
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 /* Should be called with nsid_lock held. If a new id is assigned, the bool alloc
215 * is set to true, thus the caller knows that the new id must be notified via
218 static int __peernet2id_alloc(struct net *net, struct net *peer, bool *alloc)
220 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
221 bool alloc_it = *alloc;
225 /* Magic value for id 0. */
226 if (id == NET_ID_ZERO)
232 id = alloc_netid(net, peer, -1);
234 return id >= 0 ? id : NETNSA_NSID_NOT_ASSIGNED;
237 return NETNSA_NSID_NOT_ASSIGNED;
240 /* should be called with nsid_lock held */
241 static int __peernet2id(struct net *net, struct net *peer)
245 return __peernet2id_alloc(net, peer, &no);
248 static void rtnl_net_notifyid(struct net *net, int cmd, int id);
249 /* This function returns the id of a peer netns. If no id is assigned, one will
250 * be allocated and returned.
252 int peernet2id_alloc(struct net *net, struct net *peer)
254 bool alloc = false, alive = false;
257 if (refcount_read(&net->count) == 0)
258 return NETNSA_NSID_NOT_ASSIGNED;
259 spin_lock_bh(&net->nsid_lock);
261 * When peer is obtained from RCU lists, we may race with
262 * its cleanup. Check whether it's alive, and this guarantees
263 * we never hash a peer back to net->netns_ids, after it has
264 * just been idr_remove()'d from there in cleanup_net().
266 if (maybe_get_net(peer))
267 alive = alloc = true;
268 id = __peernet2id_alloc(net, peer, &alloc);
269 spin_unlock_bh(&net->nsid_lock);
270 if (alloc && id >= 0)
271 rtnl_net_notifyid(net, RTM_NEWNSID, id);
276 EXPORT_SYMBOL_GPL(peernet2id_alloc);
278 /* This function returns, if assigned, the id of a peer netns. */
279 int peernet2id(struct net *net, struct net *peer)
283 spin_lock_bh(&net->nsid_lock);
284 id = __peernet2id(net, peer);
285 spin_unlock_bh(&net->nsid_lock);
288 EXPORT_SYMBOL(peernet2id);
290 /* This function returns true is the peer netns has an id assigned into the
293 bool peernet_has_id(struct net *net, struct net *peer)
295 return peernet2id(net, peer) >= 0;
298 struct net *get_net_ns_by_id(struct net *net, int id)
306 peer = idr_find(&net->netns_ids, id);
308 peer = maybe_get_net(peer);
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->count, 1);
325 refcount_set(&net->passive, 1);
326 get_random_bytes(&net->hash_mix, sizeof(u32));
327 net->dev_base_seq = 1;
328 net->user_ns = user_ns;
329 idr_init(&net->netns_ids);
330 spin_lock_init(&net->nsid_lock);
331 mutex_init(&net->ipv4.ra_mutex);
333 list_for_each_entry(ops, &pernet_list, list) {
334 error = ops_init(ops, net);
338 down_write(&net_rwsem);
339 list_add_tail_rcu(&net->list, &net_namespace_list);
340 up_write(&net_rwsem);
345 /* Walk through the list backwards calling the exit functions
346 * for the pernet modules whose init functions did not fail.
348 list_add(&net->exit_list, &net_exit_list);
350 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
351 ops_pre_exit_list(ops, &net_exit_list);
356 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
357 ops_exit_list(ops, &net_exit_list);
360 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
361 ops_free_list(ops, &net_exit_list);
367 static int __net_init net_defaults_init_net(struct net *net)
369 net->core.sysctl_somaxconn = SOMAXCONN;
373 static struct pernet_operations net_defaults_ops = {
374 .init = net_defaults_init_net,
377 static __init int net_defaults_init(void)
379 if (register_pernet_subsys(&net_defaults_ops))
380 panic("Cannot initialize net default settings");
385 core_initcall(net_defaults_init);
388 static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
390 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
393 static void dec_net_namespaces(struct ucounts *ucounts)
395 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
398 static struct kmem_cache *net_cachep __ro_after_init;
399 static struct workqueue_struct *netns_wq;
401 static struct net *net_alloc(void)
403 struct net *net = NULL;
404 struct net_generic *ng;
406 ng = net_alloc_generic();
410 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
415 net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
416 if (!net->key_domain)
418 refcount_set(&net->key_domain->usage, 1);
421 rcu_assign_pointer(net->gen, ng);
427 kmem_cache_free(net_cachep, net);
435 static void net_free(struct net *net)
437 kfree(rcu_access_pointer(net->gen));
438 kmem_cache_free(net_cachep, net);
441 void net_drop_ns(void *p)
444 if (ns && refcount_dec_and_test(&ns->passive))
448 struct net *copy_net_ns(unsigned long flags,
449 struct user_namespace *user_ns, struct net *old_net)
451 struct ucounts *ucounts;
455 if (!(flags & CLONE_NEWNET))
456 return get_net(old_net);
458 ucounts = inc_net_namespaces(user_ns);
460 return ERR_PTR(-ENOSPC);
467 refcount_set(&net->passive, 1);
468 net->ucounts = ucounts;
469 get_user_ns(user_ns);
471 rv = down_read_killable(&pernet_ops_rwsem);
475 rv = setup_net(net, user_ns);
477 up_read(&pernet_ops_rwsem);
481 put_user_ns(user_ns);
484 dec_net_namespaces(ucounts);
491 * net_ns_get_ownership - get sysfs ownership data for @net
492 * @net: network namespace in question (can be NULL)
493 * @uid: kernel user ID for sysfs objects
494 * @gid: kernel group ID for sysfs objects
496 * Returns the uid/gid pair of root in the user namespace associated with the
497 * given network namespace.
499 void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
502 kuid_t ns_root_uid = make_kuid(net->user_ns, 0);
503 kgid_t ns_root_gid = make_kgid(net->user_ns, 0);
505 if (uid_valid(ns_root_uid))
508 if (gid_valid(ns_root_gid))
511 *uid = GLOBAL_ROOT_UID;
512 *gid = GLOBAL_ROOT_GID;
515 EXPORT_SYMBOL_GPL(net_ns_get_ownership);
517 static void unhash_nsid(struct net *net, struct net *last)
520 /* This function is only called from cleanup_net() work,
521 * and this work is the only process, that may delete
522 * a net from net_namespace_list. So, when the below
523 * is executing, the list may only grow. Thus, we do not
524 * use for_each_net_rcu() or net_rwsem.
529 spin_lock_bh(&tmp->nsid_lock);
530 id = __peernet2id(tmp, net);
532 idr_remove(&tmp->netns_ids, id);
533 spin_unlock_bh(&tmp->nsid_lock);
535 rtnl_net_notifyid(tmp, RTM_DELNSID, id);
539 spin_lock_bh(&net->nsid_lock);
540 idr_destroy(&net->netns_ids);
541 spin_unlock_bh(&net->nsid_lock);
544 static LLIST_HEAD(cleanup_list);
546 static void cleanup_net(struct work_struct *work)
548 const struct pernet_operations *ops;
549 struct net *net, *tmp, *last;
550 struct llist_node *net_kill_list;
551 LIST_HEAD(net_exit_list);
553 /* Atomically snapshot the list of namespaces to cleanup */
554 net_kill_list = llist_del_all(&cleanup_list);
556 down_read(&pernet_ops_rwsem);
558 /* Don't let anyone else find us. */
559 down_write(&net_rwsem);
560 llist_for_each_entry(net, net_kill_list, cleanup_list)
561 list_del_rcu(&net->list);
562 /* Cache last net. After we unlock rtnl, no one new net
563 * added to net_namespace_list can assign nsid pointer
564 * to a net from net_kill_list (see peernet2id_alloc()).
565 * So, we skip them in unhash_nsid().
567 * Note, that unhash_nsid() does not delete nsid links
568 * between net_kill_list's nets, as they've already
569 * deleted from net_namespace_list. But, this would be
570 * useless anyway, as netns_ids are destroyed there.
572 last = list_last_entry(&net_namespace_list, struct net, list);
573 up_write(&net_rwsem);
575 llist_for_each_entry(net, net_kill_list, cleanup_list) {
576 unhash_nsid(net, last);
577 list_add_tail(&net->exit_list, &net_exit_list);
580 /* Run all of the network namespace pre_exit methods */
581 list_for_each_entry_reverse(ops, &pernet_list, list)
582 ops_pre_exit_list(ops, &net_exit_list);
585 * Another CPU might be rcu-iterating the list, wait for it.
586 * This needs to be before calling the exit() notifiers, so
587 * the rcu_barrier() below isn't sufficient alone.
588 * Also the pre_exit() and exit() methods need this barrier.
592 /* Run all of the network namespace exit methods */
593 list_for_each_entry_reverse(ops, &pernet_list, list)
594 ops_exit_list(ops, &net_exit_list);
596 /* Free the net generic variables */
597 list_for_each_entry_reverse(ops, &pernet_list, list)
598 ops_free_list(ops, &net_exit_list);
600 up_read(&pernet_ops_rwsem);
602 /* Ensure there are no outstanding rcu callbacks using this
607 /* Finally it is safe to free my network namespace structure */
608 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
609 list_del_init(&net->exit_list);
610 dec_net_namespaces(net->ucounts);
611 key_remove_domain(net->key_domain);
612 put_user_ns(net->user_ns);
618 * net_ns_barrier - wait until concurrent net_cleanup_work is done
620 * cleanup_net runs from work queue and will first remove namespaces
621 * from the global list, then run net exit functions.
623 * Call this in module exit path to make sure that all netns
624 * ->exit ops have been invoked before the function is removed.
626 void net_ns_barrier(void)
628 down_write(&pernet_ops_rwsem);
629 up_write(&pernet_ops_rwsem);
631 EXPORT_SYMBOL(net_ns_barrier);
633 static DECLARE_WORK(net_cleanup_work, cleanup_net);
635 void __put_net(struct net *net)
637 /* Cleanup the network namespace in process context */
638 if (llist_add(&net->cleanup_list, &cleanup_list))
639 queue_work(netns_wq, &net_cleanup_work);
641 EXPORT_SYMBOL_GPL(__put_net);
643 struct net *get_net_ns_by_fd(int fd)
646 struct ns_common *ns;
649 file = proc_ns_fget(fd);
651 return ERR_CAST(file);
653 ns = get_proc_ns(file_inode(file));
654 if (ns->ops == &netns_operations)
655 net = get_net(container_of(ns, struct net, ns));
657 net = ERR_PTR(-EINVAL);
664 struct net *get_net_ns_by_fd(int fd)
666 return ERR_PTR(-EINVAL);
669 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
671 struct net *get_net_ns_by_pid(pid_t pid)
673 struct task_struct *tsk;
676 /* Lookup the network namespace */
677 net = ERR_PTR(-ESRCH);
679 tsk = find_task_by_vpid(pid);
681 struct nsproxy *nsproxy;
683 nsproxy = tsk->nsproxy;
685 net = get_net(nsproxy->net_ns);
691 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
693 static __net_init int net_ns_net_init(struct net *net)
696 net->ns.ops = &netns_operations;
698 return ns_alloc_inum(&net->ns);
701 static __net_exit void net_ns_net_exit(struct net *net)
703 ns_free_inum(&net->ns);
706 static struct pernet_operations __net_initdata net_ns_ops = {
707 .init = net_ns_net_init,
708 .exit = net_ns_net_exit,
711 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
712 [NETNSA_NONE] = { .type = NLA_UNSPEC },
713 [NETNSA_NSID] = { .type = NLA_S32 },
714 [NETNSA_PID] = { .type = NLA_U32 },
715 [NETNSA_FD] = { .type = NLA_U32 },
716 [NETNSA_TARGET_NSID] = { .type = NLA_S32 },
719 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
720 struct netlink_ext_ack *extack)
722 struct net *net = sock_net(skb->sk);
723 struct nlattr *tb[NETNSA_MAX + 1];
728 err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
729 NETNSA_MAX, rtnl_net_policy, extack);
732 if (!tb[NETNSA_NSID]) {
733 NL_SET_ERR_MSG(extack, "nsid is missing");
736 nsid = nla_get_s32(tb[NETNSA_NSID]);
738 if (tb[NETNSA_PID]) {
739 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
740 nla = tb[NETNSA_PID];
741 } else if (tb[NETNSA_FD]) {
742 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
745 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
749 NL_SET_BAD_ATTR(extack, nla);
750 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
751 return PTR_ERR(peer);
754 spin_lock_bh(&net->nsid_lock);
755 if (__peernet2id(net, peer) >= 0) {
756 spin_unlock_bh(&net->nsid_lock);
758 NL_SET_BAD_ATTR(extack, nla);
759 NL_SET_ERR_MSG(extack,
760 "Peer netns already has a nsid assigned");
764 err = alloc_netid(net, peer, nsid);
765 spin_unlock_bh(&net->nsid_lock);
767 rtnl_net_notifyid(net, RTM_NEWNSID, err);
769 } else if (err == -ENOSPC && nsid >= 0) {
771 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
772 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
779 static int rtnl_net_get_size(void)
781 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
782 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
783 + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
787 struct net_fill_args {
797 static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
799 struct nlmsghdr *nlh;
800 struct rtgenmsg *rth;
802 nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
807 rth = nlmsg_data(nlh);
808 rth->rtgen_family = AF_UNSPEC;
810 if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
811 goto nla_put_failure;
814 nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
815 goto nla_put_failure;
821 nlmsg_cancel(skb, nlh);
825 static int rtnl_net_valid_getid_req(struct sk_buff *skb,
826 const struct nlmsghdr *nlh,
828 struct netlink_ext_ack *extack)
832 if (!netlink_strict_get_check(skb))
833 return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
834 tb, NETNSA_MAX, rtnl_net_policy,
837 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
838 NETNSA_MAX, rtnl_net_policy,
843 for (i = 0; i <= NETNSA_MAX; i++) {
851 case NETNSA_TARGET_NSID:
854 NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
862 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
863 struct netlink_ext_ack *extack)
865 struct net *net = sock_net(skb->sk);
866 struct nlattr *tb[NETNSA_MAX + 1];
867 struct net_fill_args fillargs = {
868 .portid = NETLINK_CB(skb).portid,
869 .seq = nlh->nlmsg_seq,
872 struct net *peer, *target = net;
877 err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
880 if (tb[NETNSA_PID]) {
881 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
882 nla = tb[NETNSA_PID];
883 } else if (tb[NETNSA_FD]) {
884 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
886 } else if (tb[NETNSA_NSID]) {
887 peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
889 peer = ERR_PTR(-ENOENT);
890 nla = tb[NETNSA_NSID];
892 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
897 NL_SET_BAD_ATTR(extack, nla);
898 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
899 return PTR_ERR(peer);
902 if (tb[NETNSA_TARGET_NSID]) {
903 int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
905 target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
906 if (IS_ERR(target)) {
907 NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
908 NL_SET_ERR_MSG(extack,
909 "Target netns reference is invalid");
910 err = PTR_ERR(target);
913 fillargs.add_ref = true;
914 fillargs.ref_nsid = peernet2id(net, peer);
917 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
923 fillargs.nsid = peernet2id(target, peer);
924 err = rtnl_net_fill(msg, &fillargs);
928 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
934 if (fillargs.add_ref)
940 struct rtnl_net_dump_cb {
944 struct net_fill_args fillargs;
949 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
951 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
954 if (net_cb->idx < net_cb->s_idx)
957 net_cb->fillargs.nsid = id;
958 if (net_cb->fillargs.add_ref)
959 net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
960 ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
969 static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
970 struct rtnl_net_dump_cb *net_cb,
971 struct netlink_callback *cb)
973 struct netlink_ext_ack *extack = cb->extack;
974 struct nlattr *tb[NETNSA_MAX + 1];
977 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
978 NETNSA_MAX, rtnl_net_policy,
983 for (i = 0; i <= NETNSA_MAX; i++) {
987 if (i == NETNSA_TARGET_NSID) {
990 net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
992 NL_SET_BAD_ATTR(extack, tb[i]);
993 NL_SET_ERR_MSG(extack,
994 "Invalid target network namespace id");
997 net_cb->fillargs.add_ref = true;
998 net_cb->ref_net = net_cb->tgt_net;
999 net_cb->tgt_net = net;
1001 NL_SET_BAD_ATTR(extack, tb[i]);
1002 NL_SET_ERR_MSG(extack,
1003 "Unsupported attribute in dump request");
1011 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1013 struct rtnl_net_dump_cb net_cb = {
1014 .tgt_net = sock_net(skb->sk),
1017 .portid = NETLINK_CB(cb->skb).portid,
1018 .seq = cb->nlh->nlmsg_seq,
1019 .flags = NLM_F_MULTI,
1023 .s_idx = cb->args[0],
1027 if (cb->strict_check) {
1028 err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1033 spin_lock_bh(&net_cb.tgt_net->nsid_lock);
1034 if (net_cb.fillargs.add_ref &&
1035 !net_eq(net_cb.ref_net, net_cb.tgt_net) &&
1036 !spin_trylock_bh(&net_cb.ref_net->nsid_lock)) {
1037 spin_unlock_bh(&net_cb.tgt_net->nsid_lock);
1041 idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1042 if (net_cb.fillargs.add_ref &&
1043 !net_eq(net_cb.ref_net, net_cb.tgt_net))
1044 spin_unlock_bh(&net_cb.ref_net->nsid_lock);
1045 spin_unlock_bh(&net_cb.tgt_net->nsid_lock);
1047 cb->args[0] = net_cb.idx;
1049 if (net_cb.fillargs.add_ref)
1050 put_net(net_cb.tgt_net);
1051 return err < 0 ? err : skb->len;
1054 static void rtnl_net_notifyid(struct net *net, int cmd, int id)
1056 struct net_fill_args fillargs = {
1060 struct sk_buff *msg;
1063 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
1067 err = rtnl_net_fill(msg, &fillargs);
1071 rtnl_notify(msg, net, 0, RTNLGRP_NSID, NULL, 0);
1077 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1080 static int __init net_ns_init(void)
1082 struct net_generic *ng;
1084 #ifdef CONFIG_NET_NS
1085 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1087 SLAB_PANIC|SLAB_ACCOUNT, NULL);
1089 /* Create workqueue for cleanup */
1090 netns_wq = create_singlethread_workqueue("netns");
1092 panic("Could not create netns workq");
1095 ng = net_alloc_generic();
1097 panic("Could not allocate generic netns");
1099 rcu_assign_pointer(init_net.gen, ng);
1101 down_write(&pernet_ops_rwsem);
1102 if (setup_net(&init_net, &init_user_ns))
1103 panic("Could not setup the initial network namespace");
1105 init_net_initialized = true;
1106 up_write(&pernet_ops_rwsem);
1108 if (register_pernet_subsys(&net_ns_ops))
1109 panic("Could not register network namespace subsystems");
1111 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
1112 RTNL_FLAG_DOIT_UNLOCKED);
1113 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
1114 RTNL_FLAG_DOIT_UNLOCKED);
1119 pure_initcall(net_ns_init);
1121 #ifdef CONFIG_NET_NS
1122 static int __register_pernet_operations(struct list_head *list,
1123 struct pernet_operations *ops)
1127 LIST_HEAD(net_exit_list);
1129 list_add_tail(&ops->list, list);
1130 if (ops->init || (ops->id && ops->size)) {
1131 /* We held write locked pernet_ops_rwsem, and parallel
1132 * setup_net() and cleanup_net() are not possible.
1135 error = ops_init(ops, net);
1138 list_add_tail(&net->exit_list, &net_exit_list);
1144 /* If I have an error cleanup all namespaces I initialized */
1145 list_del(&ops->list);
1146 ops_pre_exit_list(ops, &net_exit_list);
1148 ops_exit_list(ops, &net_exit_list);
1149 ops_free_list(ops, &net_exit_list);
1153 static void __unregister_pernet_operations(struct pernet_operations *ops)
1156 LIST_HEAD(net_exit_list);
1158 list_del(&ops->list);
1159 /* See comment in __register_pernet_operations() */
1161 list_add_tail(&net->exit_list, &net_exit_list);
1162 ops_pre_exit_list(ops, &net_exit_list);
1164 ops_exit_list(ops, &net_exit_list);
1165 ops_free_list(ops, &net_exit_list);
1170 static int __register_pernet_operations(struct list_head *list,
1171 struct pernet_operations *ops)
1173 if (!init_net_initialized) {
1174 list_add_tail(&ops->list, list);
1178 return ops_init(ops, &init_net);
1181 static void __unregister_pernet_operations(struct pernet_operations *ops)
1183 if (!init_net_initialized) {
1184 list_del(&ops->list);
1186 LIST_HEAD(net_exit_list);
1187 list_add(&init_net.exit_list, &net_exit_list);
1188 ops_pre_exit_list(ops, &net_exit_list);
1190 ops_exit_list(ops, &net_exit_list);
1191 ops_free_list(ops, &net_exit_list);
1195 #endif /* CONFIG_NET_NS */
1197 static DEFINE_IDA(net_generic_ids);
1199 static int register_pernet_operations(struct list_head *list,
1200 struct pernet_operations *ops)
1205 error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID,
1210 max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
1212 error = __register_pernet_operations(list, ops);
1216 ida_free(&net_generic_ids, *ops->id);
1222 static void unregister_pernet_operations(struct pernet_operations *ops)
1224 __unregister_pernet_operations(ops);
1227 ida_free(&net_generic_ids, *ops->id);
1231 * register_pernet_subsys - register a network namespace subsystem
1232 * @ops: pernet operations structure for the subsystem
1234 * Register a subsystem which has init and exit functions
1235 * that are called when network namespaces are created and
1236 * destroyed respectively.
1238 * When registered all network namespace init functions are
1239 * called for every existing network namespace. Allowing kernel
1240 * modules to have a race free view of the set of network namespaces.
1242 * When a new network namespace is created all of the init
1243 * methods are called in the order in which they were registered.
1245 * When a network namespace is destroyed all of the exit methods
1246 * are called in the reverse of the order with which they were
1249 int register_pernet_subsys(struct pernet_operations *ops)
1252 down_write(&pernet_ops_rwsem);
1253 error = register_pernet_operations(first_device, ops);
1254 up_write(&pernet_ops_rwsem);
1257 EXPORT_SYMBOL_GPL(register_pernet_subsys);
1260 * unregister_pernet_subsys - unregister a network namespace subsystem
1261 * @ops: pernet operations structure to manipulate
1263 * Remove the pernet operations structure from the list to be
1264 * used when network namespaces are created or destroyed. In
1265 * addition run the exit method for all existing network
1268 void unregister_pernet_subsys(struct pernet_operations *ops)
1270 down_write(&pernet_ops_rwsem);
1271 unregister_pernet_operations(ops);
1272 up_write(&pernet_ops_rwsem);
1274 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1277 * register_pernet_device - register a network namespace device
1278 * @ops: pernet operations structure for the subsystem
1280 * Register a device which has init and exit functions
1281 * that are called when network namespaces are created and
1282 * destroyed respectively.
1284 * When registered all network namespace init functions are
1285 * called for every existing network namespace. Allowing kernel
1286 * modules to have a race free view of the set of network namespaces.
1288 * When a new network namespace is created all of the init
1289 * methods are called in the order in which they were registered.
1291 * When a network namespace is destroyed all of the exit methods
1292 * are called in the reverse of the order with which they were
1295 int register_pernet_device(struct pernet_operations *ops)
1298 down_write(&pernet_ops_rwsem);
1299 error = register_pernet_operations(&pernet_list, ops);
1300 if (!error && (first_device == &pernet_list))
1301 first_device = &ops->list;
1302 up_write(&pernet_ops_rwsem);
1305 EXPORT_SYMBOL_GPL(register_pernet_device);
1308 * unregister_pernet_device - unregister a network namespace netdevice
1309 * @ops: pernet operations structure to manipulate
1311 * Remove the pernet operations structure from the list to be
1312 * used when network namespaces are created or destroyed. In
1313 * addition run the exit method for all existing network
1316 void unregister_pernet_device(struct pernet_operations *ops)
1318 down_write(&pernet_ops_rwsem);
1319 if (&ops->list == first_device)
1320 first_device = first_device->next;
1321 unregister_pernet_operations(ops);
1322 up_write(&pernet_ops_rwsem);
1324 EXPORT_SYMBOL_GPL(unregister_pernet_device);
1326 #ifdef CONFIG_NET_NS
1327 static struct ns_common *netns_get(struct task_struct *task)
1329 struct net *net = NULL;
1330 struct nsproxy *nsproxy;
1333 nsproxy = task->nsproxy;
1335 net = get_net(nsproxy->net_ns);
1338 return net ? &net->ns : NULL;
1341 static inline struct net *to_net_ns(struct ns_common *ns)
1343 return container_of(ns, struct net, ns);
1346 static void netns_put(struct ns_common *ns)
1348 put_net(to_net_ns(ns));
1351 static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
1353 struct net *net = to_net_ns(ns);
1355 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1356 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1359 put_net(nsproxy->net_ns);
1360 nsproxy->net_ns = get_net(net);
1364 static struct user_namespace *netns_owner(struct ns_common *ns)
1366 return to_net_ns(ns)->user_ns;
1369 const struct proc_ns_operations netns_operations = {
1371 .type = CLONE_NEWNET,
1374 .install = netns_install,
1375 .owner = netns_owner,