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 .ns.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 static struct net_generic *net_alloc_generic(void)
77 struct net_generic *ng;
78 unsigned int generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
80 ng = kzalloc(generic_size, GFP_KERNEL);
82 ng->s.len = max_gen_ptrs;
87 static int net_assign_generic(struct net *net, unsigned int id, void *data)
89 struct net_generic *ng, *old_ng;
91 BUG_ON(id < MIN_PERNET_OPS_ID);
93 old_ng = rcu_dereference_protected(net->gen,
94 lockdep_is_held(&pernet_ops_rwsem));
95 if (old_ng->s.len > id) {
96 old_ng->ptr[id] = data;
100 ng = net_alloc_generic();
105 * Some synchronisation notes:
107 * The net_generic explores the net->gen array inside rcu
108 * read section. Besides once set the net->gen->ptr[x]
109 * pointer never changes (see rules in netns/generic.h).
111 * That said, we simply duplicate this array and schedule
112 * the old copy for kfree after a grace period.
115 memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
116 (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
119 rcu_assign_pointer(net->gen, ng);
120 kfree_rcu(old_ng, s.rcu);
124 static int ops_init(const struct pernet_operations *ops, struct net *net)
129 if (ops->id && ops->size) {
130 data = kzalloc(ops->size, GFP_KERNEL);
134 err = net_assign_generic(net, *ops->id, data);
140 err = ops->init(net);
151 static void ops_free(const struct pernet_operations *ops, struct net *net)
153 if (ops->id && ops->size) {
154 kfree(net_generic(net, *ops->id));
158 static void ops_pre_exit_list(const struct pernet_operations *ops,
159 struct list_head *net_exit_list)
164 list_for_each_entry(net, net_exit_list, exit_list)
169 static void ops_exit_list(const struct pernet_operations *ops,
170 struct list_head *net_exit_list)
174 list_for_each_entry(net, net_exit_list, exit_list)
178 ops->exit_batch(net_exit_list);
181 static void ops_free_list(const struct pernet_operations *ops,
182 struct list_head *net_exit_list)
185 if (ops->size && ops->id) {
186 list_for_each_entry(net, net_exit_list, exit_list)
191 /* should be called with nsid_lock held */
192 static int alloc_netid(struct net *net, struct net *peer, int reqid)
194 int min = 0, max = 0;
201 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
204 /* This function is used by idr_for_each(). If net is equal to peer, the
205 * function returns the id so that idr_for_each() stops. Because we cannot
206 * returns the id 0 (idr_for_each() will not stop), we return the magic value
207 * NET_ID_ZERO (-1) for it.
209 #define NET_ID_ZERO -1
210 static int net_eq_idr(int id, void *net, void *peer)
212 if (net_eq(net, peer))
213 return id ? : NET_ID_ZERO;
217 /* Must be called from RCU-critical section or with nsid_lock held */
218 static int __peernet2id(const struct net *net, struct net *peer)
220 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
222 /* Magic value for id 0. */
223 if (id == NET_ID_ZERO)
228 return NETNSA_NSID_NOT_ASSIGNED;
231 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
232 struct nlmsghdr *nlh, gfp_t gfp);
233 /* This function returns the id of a peer netns. If no id is assigned, one will
234 * be allocated and returned.
236 int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp)
240 if (refcount_read(&net->ns.count) == 0)
241 return NETNSA_NSID_NOT_ASSIGNED;
243 spin_lock_bh(&net->nsid_lock);
244 id = __peernet2id(net, peer);
246 spin_unlock_bh(&net->nsid_lock);
250 /* When peer is obtained from RCU lists, we may race with
251 * its cleanup. Check whether it's alive, and this guarantees
252 * we never hash a peer back to net->netns_ids, after it has
253 * just been idr_remove()'d from there in cleanup_net().
255 if (!maybe_get_net(peer)) {
256 spin_unlock_bh(&net->nsid_lock);
257 return NETNSA_NSID_NOT_ASSIGNED;
260 id = alloc_netid(net, peer, -1);
261 spin_unlock_bh(&net->nsid_lock);
265 return NETNSA_NSID_NOT_ASSIGNED;
267 rtnl_net_notifyid(net, RTM_NEWNSID, id, 0, NULL, gfp);
271 EXPORT_SYMBOL_GPL(peernet2id_alloc);
273 /* This function returns, if assigned, the id of a peer netns. */
274 int peernet2id(const struct net *net, struct net *peer)
279 id = __peernet2id(net, peer);
284 EXPORT_SYMBOL(peernet2id);
286 /* This function returns true is the peer netns has an id assigned into the
289 bool peernet_has_id(const struct net *net, struct net *peer)
291 return peernet2id(net, peer) >= 0;
294 struct net *get_net_ns_by_id(const struct net *net, int id)
302 peer = idr_find(&net->netns_ids, id);
304 peer = maybe_get_net(peer);
311 * setup_net runs the initializers for the network namespace object.
313 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
315 /* Must be called with pernet_ops_rwsem held */
316 const struct pernet_operations *ops, *saved_ops;
318 LIST_HEAD(net_exit_list);
320 refcount_set(&net->ns.count, 1);
321 refcount_set(&net->passive, 1);
322 get_random_bytes(&net->hash_mix, sizeof(u32));
324 net->net_cookie = gen_cookie_next(&net_cookie);
326 net->dev_base_seq = 1;
327 net->user_ns = user_ns;
328 idr_init(&net->netns_ids);
329 spin_lock_init(&net->nsid_lock);
330 mutex_init(&net->ipv4.ra_mutex);
332 list_for_each_entry(ops, &pernet_list, list) {
333 error = ops_init(ops, net);
337 down_write(&net_rwsem);
338 list_add_tail_rcu(&net->list, &net_namespace_list);
339 up_write(&net_rwsem);
344 /* Walk through the list backwards calling the exit functions
345 * for the pernet modules whose init functions did not fail.
347 list_add(&net->exit_list, &net_exit_list);
349 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
350 ops_pre_exit_list(ops, &net_exit_list);
355 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
356 ops_exit_list(ops, &net_exit_list);
359 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
360 ops_free_list(ops, &net_exit_list);
366 static int __net_init net_defaults_init_net(struct net *net)
368 net->core.sysctl_somaxconn = SOMAXCONN;
372 static struct pernet_operations net_defaults_ops = {
373 .init = net_defaults_init_net,
376 static __init int net_defaults_init(void)
378 if (register_pernet_subsys(&net_defaults_ops))
379 panic("Cannot initialize net default settings");
384 core_initcall(net_defaults_init);
387 static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
389 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
392 static void dec_net_namespaces(struct ucounts *ucounts)
394 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
397 static struct kmem_cache *net_cachep __ro_after_init;
398 static struct workqueue_struct *netns_wq;
400 static struct net *net_alloc(void)
402 struct net *net = NULL;
403 struct net_generic *ng;
405 ng = net_alloc_generic();
409 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
414 net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
415 if (!net->key_domain)
417 refcount_set(&net->key_domain->usage, 1);
420 rcu_assign_pointer(net->gen, ng);
426 kmem_cache_free(net_cachep, net);
434 static void net_free(struct net *net)
436 kfree(rcu_access_pointer(net->gen));
437 kmem_cache_free(net_cachep, net);
440 void net_drop_ns(void *p)
443 if (ns && refcount_dec_and_test(&ns->passive))
447 struct net *copy_net_ns(unsigned long flags,
448 struct user_namespace *user_ns, struct net *old_net)
450 struct ucounts *ucounts;
454 if (!(flags & CLONE_NEWNET))
455 return get_net(old_net);
457 ucounts = inc_net_namespaces(user_ns);
459 return ERR_PTR(-ENOSPC);
466 refcount_set(&net->passive, 1);
467 net->ucounts = ucounts;
468 get_user_ns(user_ns);
470 rv = down_read_killable(&pernet_ops_rwsem);
474 rv = setup_net(net, user_ns);
476 up_read(&pernet_ops_rwsem);
480 key_remove_domain(net->key_domain);
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, 0, NULL,
540 spin_lock_bh(&net->nsid_lock);
541 idr_destroy(&net->netns_ids);
542 spin_unlock_bh(&net->nsid_lock);
545 static LLIST_HEAD(cleanup_list);
547 static void cleanup_net(struct work_struct *work)
549 const struct pernet_operations *ops;
550 struct net *net, *tmp, *last;
551 struct llist_node *net_kill_list;
552 LIST_HEAD(net_exit_list);
554 /* Atomically snapshot the list of namespaces to cleanup */
555 net_kill_list = llist_del_all(&cleanup_list);
557 down_read(&pernet_ops_rwsem);
559 /* Don't let anyone else find us. */
560 down_write(&net_rwsem);
561 llist_for_each_entry(net, net_kill_list, cleanup_list)
562 list_del_rcu(&net->list);
563 /* Cache last net. After we unlock rtnl, no one new net
564 * added to net_namespace_list can assign nsid pointer
565 * to a net from net_kill_list (see peernet2id_alloc()).
566 * So, we skip them in unhash_nsid().
568 * Note, that unhash_nsid() does not delete nsid links
569 * between net_kill_list's nets, as they've already
570 * deleted from net_namespace_list. But, this would be
571 * useless anyway, as netns_ids are destroyed there.
573 last = list_last_entry(&net_namespace_list, struct net, list);
574 up_write(&net_rwsem);
576 llist_for_each_entry(net, net_kill_list, cleanup_list) {
577 unhash_nsid(net, last);
578 list_add_tail(&net->exit_list, &net_exit_list);
581 /* Run all of the network namespace pre_exit methods */
582 list_for_each_entry_reverse(ops, &pernet_list, list)
583 ops_pre_exit_list(ops, &net_exit_list);
586 * Another CPU might be rcu-iterating the list, wait for it.
587 * This needs to be before calling the exit() notifiers, so
588 * the rcu_barrier() below isn't sufficient alone.
589 * Also the pre_exit() and exit() methods need this barrier.
593 /* Run all of the network namespace exit methods */
594 list_for_each_entry_reverse(ops, &pernet_list, list)
595 ops_exit_list(ops, &net_exit_list);
597 /* Free the net generic variables */
598 list_for_each_entry_reverse(ops, &pernet_list, list)
599 ops_free_list(ops, &net_exit_list);
601 up_read(&pernet_ops_rwsem);
603 /* Ensure there are no outstanding rcu callbacks using this
608 /* Finally it is safe to free my network namespace structure */
609 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
610 list_del_init(&net->exit_list);
611 dec_net_namespaces(net->ucounts);
612 key_remove_domain(net->key_domain);
613 put_user_ns(net->user_ns);
619 * net_ns_barrier - wait until concurrent net_cleanup_work is done
621 * cleanup_net runs from work queue and will first remove namespaces
622 * from the global list, then run net exit functions.
624 * Call this in module exit path to make sure that all netns
625 * ->exit ops have been invoked before the function is removed.
627 void net_ns_barrier(void)
629 down_write(&pernet_ops_rwsem);
630 up_write(&pernet_ops_rwsem);
632 EXPORT_SYMBOL(net_ns_barrier);
634 static DECLARE_WORK(net_cleanup_work, cleanup_net);
636 void __put_net(struct net *net)
638 /* Cleanup the network namespace in process context */
639 if (llist_add(&net->cleanup_list, &cleanup_list))
640 queue_work(netns_wq, &net_cleanup_work);
642 EXPORT_SYMBOL_GPL(__put_net);
644 struct net *get_net_ns_by_fd(int fd)
647 struct ns_common *ns;
650 file = proc_ns_fget(fd);
652 return ERR_CAST(file);
654 ns = get_proc_ns(file_inode(file));
655 if (ns->ops == &netns_operations)
656 net = get_net(container_of(ns, struct net, ns));
658 net = ERR_PTR(-EINVAL);
665 struct net *get_net_ns_by_fd(int fd)
667 return ERR_PTR(-EINVAL);
670 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
672 struct net *get_net_ns_by_pid(pid_t pid)
674 struct task_struct *tsk;
677 /* Lookup the network namespace */
678 net = ERR_PTR(-ESRCH);
680 tsk = find_task_by_vpid(pid);
682 struct nsproxy *nsproxy;
684 nsproxy = tsk->nsproxy;
686 net = get_net(nsproxy->net_ns);
692 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
694 static __net_init int net_ns_net_init(struct net *net)
697 net->ns.ops = &netns_operations;
699 return ns_alloc_inum(&net->ns);
702 static __net_exit void net_ns_net_exit(struct net *net)
704 ns_free_inum(&net->ns);
707 static struct pernet_operations __net_initdata net_ns_ops = {
708 .init = net_ns_net_init,
709 .exit = net_ns_net_exit,
712 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
713 [NETNSA_NONE] = { .type = NLA_UNSPEC },
714 [NETNSA_NSID] = { .type = NLA_S32 },
715 [NETNSA_PID] = { .type = NLA_U32 },
716 [NETNSA_FD] = { .type = NLA_U32 },
717 [NETNSA_TARGET_NSID] = { .type = NLA_S32 },
720 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
721 struct netlink_ext_ack *extack)
723 struct net *net = sock_net(skb->sk);
724 struct nlattr *tb[NETNSA_MAX + 1];
729 err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
730 NETNSA_MAX, rtnl_net_policy, extack);
733 if (!tb[NETNSA_NSID]) {
734 NL_SET_ERR_MSG(extack, "nsid is missing");
737 nsid = nla_get_s32(tb[NETNSA_NSID]);
739 if (tb[NETNSA_PID]) {
740 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
741 nla = tb[NETNSA_PID];
742 } else if (tb[NETNSA_FD]) {
743 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
746 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
750 NL_SET_BAD_ATTR(extack, nla);
751 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
752 return PTR_ERR(peer);
755 spin_lock_bh(&net->nsid_lock);
756 if (__peernet2id(net, peer) >= 0) {
757 spin_unlock_bh(&net->nsid_lock);
759 NL_SET_BAD_ATTR(extack, nla);
760 NL_SET_ERR_MSG(extack,
761 "Peer netns already has a nsid assigned");
765 err = alloc_netid(net, peer, nsid);
766 spin_unlock_bh(&net->nsid_lock);
768 rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid,
771 } else if (err == -ENOSPC && nsid >= 0) {
773 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
774 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
781 static int rtnl_net_get_size(void)
783 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
784 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
785 + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
789 struct net_fill_args {
799 static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
801 struct nlmsghdr *nlh;
802 struct rtgenmsg *rth;
804 nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
809 rth = nlmsg_data(nlh);
810 rth->rtgen_family = AF_UNSPEC;
812 if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
813 goto nla_put_failure;
816 nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
817 goto nla_put_failure;
823 nlmsg_cancel(skb, nlh);
827 static int rtnl_net_valid_getid_req(struct sk_buff *skb,
828 const struct nlmsghdr *nlh,
830 struct netlink_ext_ack *extack)
834 if (!netlink_strict_get_check(skb))
835 return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
836 tb, NETNSA_MAX, rtnl_net_policy,
839 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
840 NETNSA_MAX, rtnl_net_policy,
845 for (i = 0; i <= NETNSA_MAX; i++) {
853 case NETNSA_TARGET_NSID:
856 NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
864 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
865 struct netlink_ext_ack *extack)
867 struct net *net = sock_net(skb->sk);
868 struct nlattr *tb[NETNSA_MAX + 1];
869 struct net_fill_args fillargs = {
870 .portid = NETLINK_CB(skb).portid,
871 .seq = nlh->nlmsg_seq,
874 struct net *peer, *target = net;
879 err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
882 if (tb[NETNSA_PID]) {
883 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
884 nla = tb[NETNSA_PID];
885 } else if (tb[NETNSA_FD]) {
886 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
888 } else if (tb[NETNSA_NSID]) {
889 peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
891 peer = ERR_PTR(-ENOENT);
892 nla = tb[NETNSA_NSID];
894 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
899 NL_SET_BAD_ATTR(extack, nla);
900 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
901 return PTR_ERR(peer);
904 if (tb[NETNSA_TARGET_NSID]) {
905 int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
907 target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
908 if (IS_ERR(target)) {
909 NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
910 NL_SET_ERR_MSG(extack,
911 "Target netns reference is invalid");
912 err = PTR_ERR(target);
915 fillargs.add_ref = true;
916 fillargs.ref_nsid = peernet2id(net, peer);
919 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
925 fillargs.nsid = peernet2id(target, peer);
926 err = rtnl_net_fill(msg, &fillargs);
930 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
936 if (fillargs.add_ref)
942 struct rtnl_net_dump_cb {
946 struct net_fill_args fillargs;
951 /* Runs in RCU-critical section. */
952 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
954 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
957 if (net_cb->idx < net_cb->s_idx)
960 net_cb->fillargs.nsid = id;
961 if (net_cb->fillargs.add_ref)
962 net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
963 ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
972 static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
973 struct rtnl_net_dump_cb *net_cb,
974 struct netlink_callback *cb)
976 struct netlink_ext_ack *extack = cb->extack;
977 struct nlattr *tb[NETNSA_MAX + 1];
980 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
981 NETNSA_MAX, rtnl_net_policy,
986 for (i = 0; i <= NETNSA_MAX; i++) {
990 if (i == NETNSA_TARGET_NSID) {
993 net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
995 NL_SET_BAD_ATTR(extack, tb[i]);
996 NL_SET_ERR_MSG(extack,
997 "Invalid target network namespace id");
1000 net_cb->fillargs.add_ref = true;
1001 net_cb->ref_net = net_cb->tgt_net;
1002 net_cb->tgt_net = net;
1004 NL_SET_BAD_ATTR(extack, tb[i]);
1005 NL_SET_ERR_MSG(extack,
1006 "Unsupported attribute in dump request");
1014 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1016 struct rtnl_net_dump_cb net_cb = {
1017 .tgt_net = sock_net(skb->sk),
1020 .portid = NETLINK_CB(cb->skb).portid,
1021 .seq = cb->nlh->nlmsg_seq,
1022 .flags = NLM_F_MULTI,
1026 .s_idx = cb->args[0],
1030 if (cb->strict_check) {
1031 err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1037 idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1040 cb->args[0] = net_cb.idx;
1042 if (net_cb.fillargs.add_ref)
1043 put_net(net_cb.tgt_net);
1044 return err < 0 ? err : skb->len;
1047 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1048 struct nlmsghdr *nlh, gfp_t gfp)
1050 struct net_fill_args fillargs = {
1052 .seq = nlh ? nlh->nlmsg_seq : 0,
1056 struct sk_buff *msg;
1059 msg = nlmsg_new(rtnl_net_get_size(), gfp);
1063 err = rtnl_net_fill(msg, &fillargs);
1067 rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, gfp);
1073 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1076 static int __init net_ns_init(void)
1078 struct net_generic *ng;
1080 #ifdef CONFIG_NET_NS
1081 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1083 SLAB_PANIC|SLAB_ACCOUNT, NULL);
1085 /* Create workqueue for cleanup */
1086 netns_wq = create_singlethread_workqueue("netns");
1088 panic("Could not create netns workq");
1091 ng = net_alloc_generic();
1093 panic("Could not allocate generic netns");
1095 rcu_assign_pointer(init_net.gen, ng);
1097 down_write(&pernet_ops_rwsem);
1098 if (setup_net(&init_net, &init_user_ns))
1099 panic("Could not setup the initial network namespace");
1101 init_net_initialized = true;
1102 up_write(&pernet_ops_rwsem);
1104 if (register_pernet_subsys(&net_ns_ops))
1105 panic("Could not register network namespace subsystems");
1107 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
1108 RTNL_FLAG_DOIT_UNLOCKED);
1109 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
1110 RTNL_FLAG_DOIT_UNLOCKED);
1115 pure_initcall(net_ns_init);
1117 #ifdef CONFIG_NET_NS
1118 static int __register_pernet_operations(struct list_head *list,
1119 struct pernet_operations *ops)
1123 LIST_HEAD(net_exit_list);
1125 list_add_tail(&ops->list, list);
1126 if (ops->init || (ops->id && ops->size)) {
1127 /* We held write locked pernet_ops_rwsem, and parallel
1128 * setup_net() and cleanup_net() are not possible.
1131 error = ops_init(ops, net);
1134 list_add_tail(&net->exit_list, &net_exit_list);
1140 /* If I have an error cleanup all namespaces I initialized */
1141 list_del(&ops->list);
1142 ops_pre_exit_list(ops, &net_exit_list);
1144 ops_exit_list(ops, &net_exit_list);
1145 ops_free_list(ops, &net_exit_list);
1149 static void __unregister_pernet_operations(struct pernet_operations *ops)
1152 LIST_HEAD(net_exit_list);
1154 list_del(&ops->list);
1155 /* See comment in __register_pernet_operations() */
1157 list_add_tail(&net->exit_list, &net_exit_list);
1158 ops_pre_exit_list(ops, &net_exit_list);
1160 ops_exit_list(ops, &net_exit_list);
1161 ops_free_list(ops, &net_exit_list);
1166 static int __register_pernet_operations(struct list_head *list,
1167 struct pernet_operations *ops)
1169 if (!init_net_initialized) {
1170 list_add_tail(&ops->list, list);
1174 return ops_init(ops, &init_net);
1177 static void __unregister_pernet_operations(struct pernet_operations *ops)
1179 if (!init_net_initialized) {
1180 list_del(&ops->list);
1182 LIST_HEAD(net_exit_list);
1183 list_add(&init_net.exit_list, &net_exit_list);
1184 ops_pre_exit_list(ops, &net_exit_list);
1186 ops_exit_list(ops, &net_exit_list);
1187 ops_free_list(ops, &net_exit_list);
1191 #endif /* CONFIG_NET_NS */
1193 static DEFINE_IDA(net_generic_ids);
1195 static int register_pernet_operations(struct list_head *list,
1196 struct pernet_operations *ops)
1201 error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID,
1206 max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
1208 error = __register_pernet_operations(list, ops);
1212 ida_free(&net_generic_ids, *ops->id);
1218 static void unregister_pernet_operations(struct pernet_operations *ops)
1220 __unregister_pernet_operations(ops);
1223 ida_free(&net_generic_ids, *ops->id);
1227 * register_pernet_subsys - register a network namespace subsystem
1228 * @ops: pernet operations structure for the subsystem
1230 * Register a subsystem which has init and exit functions
1231 * that are called when network namespaces are created and
1232 * destroyed respectively.
1234 * When registered all network namespace init functions are
1235 * called for every existing network namespace. Allowing kernel
1236 * modules to have a race free view of the set of network namespaces.
1238 * When a new network namespace is created all of the init
1239 * methods are called in the order in which they were registered.
1241 * When a network namespace is destroyed all of the exit methods
1242 * are called in the reverse of the order with which they were
1245 int register_pernet_subsys(struct pernet_operations *ops)
1248 down_write(&pernet_ops_rwsem);
1249 error = register_pernet_operations(first_device, ops);
1250 up_write(&pernet_ops_rwsem);
1253 EXPORT_SYMBOL_GPL(register_pernet_subsys);
1256 * unregister_pernet_subsys - unregister a network namespace subsystem
1257 * @ops: pernet operations structure to manipulate
1259 * Remove the pernet operations structure from the list to be
1260 * used when network namespaces are created or destroyed. In
1261 * addition run the exit method for all existing network
1264 void unregister_pernet_subsys(struct pernet_operations *ops)
1266 down_write(&pernet_ops_rwsem);
1267 unregister_pernet_operations(ops);
1268 up_write(&pernet_ops_rwsem);
1270 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1273 * register_pernet_device - register a network namespace device
1274 * @ops: pernet operations structure for the subsystem
1276 * Register a device which has init and exit functions
1277 * that are called when network namespaces are created and
1278 * destroyed respectively.
1280 * When registered all network namespace init functions are
1281 * called for every existing network namespace. Allowing kernel
1282 * modules to have a race free view of the set of network namespaces.
1284 * When a new network namespace is created all of the init
1285 * methods are called in the order in which they were registered.
1287 * When a network namespace is destroyed all of the exit methods
1288 * are called in the reverse of the order with which they were
1291 int register_pernet_device(struct pernet_operations *ops)
1294 down_write(&pernet_ops_rwsem);
1295 error = register_pernet_operations(&pernet_list, ops);
1296 if (!error && (first_device == &pernet_list))
1297 first_device = &ops->list;
1298 up_write(&pernet_ops_rwsem);
1301 EXPORT_SYMBOL_GPL(register_pernet_device);
1304 * unregister_pernet_device - unregister a network namespace netdevice
1305 * @ops: pernet operations structure to manipulate
1307 * Remove the pernet operations structure from the list to be
1308 * used when network namespaces are created or destroyed. In
1309 * addition run the exit method for all existing network
1312 void unregister_pernet_device(struct pernet_operations *ops)
1314 down_write(&pernet_ops_rwsem);
1315 if (&ops->list == first_device)
1316 first_device = first_device->next;
1317 unregister_pernet_operations(ops);
1318 up_write(&pernet_ops_rwsem);
1320 EXPORT_SYMBOL_GPL(unregister_pernet_device);
1322 #ifdef CONFIG_NET_NS
1323 static struct ns_common *netns_get(struct task_struct *task)
1325 struct net *net = NULL;
1326 struct nsproxy *nsproxy;
1329 nsproxy = task->nsproxy;
1331 net = get_net(nsproxy->net_ns);
1334 return net ? &net->ns : NULL;
1337 static inline struct net *to_net_ns(struct ns_common *ns)
1339 return container_of(ns, struct net, ns);
1342 static void netns_put(struct ns_common *ns)
1344 put_net(to_net_ns(ns));
1347 static int netns_install(struct nsset *nsset, struct ns_common *ns)
1349 struct nsproxy *nsproxy = nsset->nsproxy;
1350 struct net *net = to_net_ns(ns);
1352 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1353 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
1356 put_net(nsproxy->net_ns);
1357 nsproxy->net_ns = get_net(net);
1361 static struct user_namespace *netns_owner(struct ns_common *ns)
1363 return to_net_ns(ns)->user_ns;
1366 const struct proc_ns_operations netns_operations = {
1368 .type = CLONE_NEWNET,
1371 .install = netns_install,
1372 .owner = netns_owner,