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) };
48 EXPORT_SYMBOL(init_net);
50 static bool init_net_initialized;
52 * pernet_ops_rwsem: protects: pernet_list, net_generic_ids,
53 * init_net_initialized and first_device pointer.
54 * This is internal net namespace object. Please, don't use it
57 DECLARE_RWSEM(pernet_ops_rwsem);
58 EXPORT_SYMBOL_GPL(pernet_ops_rwsem);
60 #define MIN_PERNET_OPS_ID \
61 ((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *))
63 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
65 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
67 DEFINE_COOKIE(net_cookie);
69 static struct net_generic *net_alloc_generic(void)
71 unsigned int gen_ptrs = READ_ONCE(max_gen_ptrs);
72 unsigned int generic_size;
73 struct net_generic *ng;
75 generic_size = offsetof(struct net_generic, ptr[gen_ptrs]);
77 ng = kzalloc(generic_size, GFP_KERNEL);
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)
123 struct net_generic *ng;
127 if (ops->id && ops->size) {
128 data = kzalloc(ops->size, GFP_KERNEL);
132 err = net_assign_generic(net, *ops->id, data);
138 err = ops->init(net);
142 if (ops->id && ops->size) {
143 ng = rcu_dereference_protected(net->gen,
144 lockdep_is_held(&pernet_ops_rwsem));
145 ng->ptr[*ops->id] = NULL;
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) {
177 ops->exit_batch(net_exit_list);
180 static void ops_free_list(const struct pernet_operations *ops,
181 struct list_head *net_exit_list)
184 if (ops->size && ops->id) {
185 list_for_each_entry(net, net_exit_list, exit_list)
186 kfree(net_generic(net, *ops->id));
190 /* should be called with nsid_lock held */
191 static int alloc_netid(struct net *net, struct net *peer, int reqid)
193 int min = 0, max = 0;
200 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
203 /* This function is used by idr_for_each(). If net is equal to peer, the
204 * function returns the id so that idr_for_each() stops. Because we cannot
205 * returns the id 0 (idr_for_each() will not stop), we return the magic value
206 * NET_ID_ZERO (-1) for it.
208 #define NET_ID_ZERO -1
209 static int net_eq_idr(int id, void *net, void *peer)
211 if (net_eq(net, peer))
212 return id ? : NET_ID_ZERO;
216 /* Must be called from RCU-critical section or with nsid_lock held */
217 static int __peernet2id(const struct net *net, struct net *peer)
219 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
221 /* Magic value for id 0. */
222 if (id == NET_ID_ZERO)
227 return NETNSA_NSID_NOT_ASSIGNED;
230 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
231 struct nlmsghdr *nlh, gfp_t gfp);
232 /* This function returns the id of a peer netns. If no id is assigned, one will
233 * be allocated and returned.
235 int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp)
239 if (refcount_read(&net->ns.count) == 0)
240 return NETNSA_NSID_NOT_ASSIGNED;
242 spin_lock_bh(&net->nsid_lock);
243 id = __peernet2id(net, peer);
245 spin_unlock_bh(&net->nsid_lock);
249 /* When peer is obtained from RCU lists, we may race with
250 * its cleanup. Check whether it's alive, and this guarantees
251 * we never hash a peer back to net->netns_ids, after it has
252 * just been idr_remove()'d from there in cleanup_net().
254 if (!maybe_get_net(peer)) {
255 spin_unlock_bh(&net->nsid_lock);
256 return NETNSA_NSID_NOT_ASSIGNED;
259 id = alloc_netid(net, peer, -1);
260 spin_unlock_bh(&net->nsid_lock);
264 return NETNSA_NSID_NOT_ASSIGNED;
266 rtnl_net_notifyid(net, RTM_NEWNSID, id, 0, NULL, gfp);
270 EXPORT_SYMBOL_GPL(peernet2id_alloc);
272 /* This function returns, if assigned, the id of a peer netns. */
273 int peernet2id(const struct net *net, struct net *peer)
278 id = __peernet2id(net, peer);
283 EXPORT_SYMBOL(peernet2id);
285 /* This function returns true is the peer netns has an id assigned into the
288 bool peernet_has_id(const struct net *net, struct net *peer)
290 return peernet2id(net, peer) >= 0;
293 struct net *get_net_ns_by_id(const struct net *net, int id)
301 peer = idr_find(&net->netns_ids, id);
303 peer = maybe_get_net(peer);
308 EXPORT_SYMBOL_GPL(get_net_ns_by_id);
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 ref_tracker_dir_init(&net->refcnt_tracker, 128);
323 refcount_set(&net->passive, 1);
324 get_random_bytes(&net->hash_mix, sizeof(u32));
326 net->net_cookie = gen_cookie_next(&net_cookie);
328 net->dev_base_seq = 1;
329 net->user_ns = user_ns;
330 idr_init(&net->netns_ids);
331 spin_lock_init(&net->nsid_lock);
332 mutex_init(&net->ipv4.ra_mutex);
334 list_for_each_entry(ops, &pernet_list, list) {
335 error = ops_init(ops, net);
339 down_write(&net_rwsem);
340 list_add_tail_rcu(&net->list, &net_namespace_list);
341 up_write(&net_rwsem);
346 /* Walk through the list backwards calling the exit functions
347 * for the pernet modules whose init functions did not fail.
349 list_add(&net->exit_list, &net_exit_list);
351 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
352 ops_pre_exit_list(ops, &net_exit_list);
357 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
358 ops_exit_list(ops, &net_exit_list);
361 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
362 ops_free_list(ops, &net_exit_list);
368 static int __net_init net_defaults_init_net(struct net *net)
370 net->core.sysctl_somaxconn = SOMAXCONN;
371 net->core.sysctl_txrehash = SOCK_TXREHASH_ENABLED;
376 static struct pernet_operations net_defaults_ops = {
377 .init = net_defaults_init_net,
380 static __init int net_defaults_init(void)
382 if (register_pernet_subsys(&net_defaults_ops))
383 panic("Cannot initialize net default settings");
388 core_initcall(net_defaults_init);
391 static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
393 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
396 static void dec_net_namespaces(struct ucounts *ucounts)
398 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
401 static struct kmem_cache *net_cachep __ro_after_init;
402 static struct workqueue_struct *netns_wq;
404 static struct net *net_alloc(void)
406 struct net *net = NULL;
407 struct net_generic *ng;
409 ng = net_alloc_generic();
413 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
418 net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
419 if (!net->key_domain)
421 refcount_set(&net->key_domain->usage, 1);
424 rcu_assign_pointer(net->gen, ng);
430 kmem_cache_free(net_cachep, net);
438 static void net_free(struct net *net)
440 if (refcount_dec_and_test(&net->passive)) {
441 kfree(rcu_access_pointer(net->gen));
442 kmem_cache_free(net_cachep, net);
446 void net_drop_ns(void *p)
448 struct net *net = (struct net *)p;
454 struct net *copy_net_ns(unsigned long flags,
455 struct user_namespace *user_ns, struct net *old_net)
457 struct ucounts *ucounts;
461 if (!(flags & CLONE_NEWNET))
462 return get_net(old_net);
464 ucounts = inc_net_namespaces(user_ns);
466 return ERR_PTR(-ENOSPC);
473 refcount_set(&net->passive, 1);
474 net->ucounts = ucounts;
475 get_user_ns(user_ns);
477 rv = down_read_killable(&pernet_ops_rwsem);
481 rv = setup_net(net, user_ns);
483 up_read(&pernet_ops_rwsem);
488 key_remove_domain(net->key_domain);
490 put_user_ns(user_ns);
493 dec_net_namespaces(ucounts);
500 * net_ns_get_ownership - get sysfs ownership data for @net
501 * @net: network namespace in question (can be NULL)
502 * @uid: kernel user ID for sysfs objects
503 * @gid: kernel group ID for sysfs objects
505 * Returns the uid/gid pair of root in the user namespace associated with the
506 * given network namespace.
508 void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
511 kuid_t ns_root_uid = make_kuid(net->user_ns, 0);
512 kgid_t ns_root_gid = make_kgid(net->user_ns, 0);
514 if (uid_valid(ns_root_uid))
517 if (gid_valid(ns_root_gid))
520 *uid = GLOBAL_ROOT_UID;
521 *gid = GLOBAL_ROOT_GID;
524 EXPORT_SYMBOL_GPL(net_ns_get_ownership);
526 static void unhash_nsid(struct net *net, struct net *last)
529 /* This function is only called from cleanup_net() work,
530 * and this work is the only process, that may delete
531 * a net from net_namespace_list. So, when the below
532 * is executing, the list may only grow. Thus, we do not
533 * use for_each_net_rcu() or net_rwsem.
538 spin_lock_bh(&tmp->nsid_lock);
539 id = __peernet2id(tmp, net);
541 idr_remove(&tmp->netns_ids, id);
542 spin_unlock_bh(&tmp->nsid_lock);
544 rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL,
549 spin_lock_bh(&net->nsid_lock);
550 idr_destroy(&net->netns_ids);
551 spin_unlock_bh(&net->nsid_lock);
554 static LLIST_HEAD(cleanup_list);
556 static void cleanup_net(struct work_struct *work)
558 const struct pernet_operations *ops;
559 struct net *net, *tmp, *last;
560 struct llist_node *net_kill_list;
561 LIST_HEAD(net_exit_list);
563 /* Atomically snapshot the list of namespaces to cleanup */
564 net_kill_list = llist_del_all(&cleanup_list);
566 down_read(&pernet_ops_rwsem);
568 /* Don't let anyone else find us. */
569 down_write(&net_rwsem);
570 llist_for_each_entry(net, net_kill_list, cleanup_list)
571 list_del_rcu(&net->list);
572 /* Cache last net. After we unlock rtnl, no one new net
573 * added to net_namespace_list can assign nsid pointer
574 * to a net from net_kill_list (see peernet2id_alloc()).
575 * So, we skip them in unhash_nsid().
577 * Note, that unhash_nsid() does not delete nsid links
578 * between net_kill_list's nets, as they've already
579 * deleted from net_namespace_list. But, this would be
580 * useless anyway, as netns_ids are destroyed there.
582 last = list_last_entry(&net_namespace_list, struct net, list);
583 up_write(&net_rwsem);
585 llist_for_each_entry(net, net_kill_list, cleanup_list) {
586 unhash_nsid(net, last);
587 list_add_tail(&net->exit_list, &net_exit_list);
590 /* Run all of the network namespace pre_exit methods */
591 list_for_each_entry_reverse(ops, &pernet_list, list)
592 ops_pre_exit_list(ops, &net_exit_list);
595 * Another CPU might be rcu-iterating the list, wait for it.
596 * This needs to be before calling the exit() notifiers, so
597 * the rcu_barrier() below isn't sufficient alone.
598 * Also the pre_exit() and exit() methods need this barrier.
602 /* Run all of the network namespace exit methods */
603 list_for_each_entry_reverse(ops, &pernet_list, list)
604 ops_exit_list(ops, &net_exit_list);
606 /* Free the net generic variables */
607 list_for_each_entry_reverse(ops, &pernet_list, list)
608 ops_free_list(ops, &net_exit_list);
610 up_read(&pernet_ops_rwsem);
612 /* Ensure there are no outstanding rcu callbacks using this
617 /* Finally it is safe to free my network namespace structure */
618 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
619 list_del_init(&net->exit_list);
620 dec_net_namespaces(net->ucounts);
622 key_remove_domain(net->key_domain);
624 put_user_ns(net->user_ns);
630 * net_ns_barrier - wait until concurrent net_cleanup_work is done
632 * cleanup_net runs from work queue and will first remove namespaces
633 * from the global list, then run net exit functions.
635 * Call this in module exit path to make sure that all netns
636 * ->exit ops have been invoked before the function is removed.
638 void net_ns_barrier(void)
640 down_write(&pernet_ops_rwsem);
641 up_write(&pernet_ops_rwsem);
643 EXPORT_SYMBOL(net_ns_barrier);
645 static DECLARE_WORK(net_cleanup_work, cleanup_net);
647 void __put_net(struct net *net)
649 ref_tracker_dir_exit(&net->refcnt_tracker);
650 /* Cleanup the network namespace in process context */
651 if (llist_add(&net->cleanup_list, &cleanup_list))
652 queue_work(netns_wq, &net_cleanup_work);
654 EXPORT_SYMBOL_GPL(__put_net);
657 * get_net_ns - increment the refcount of the network namespace
658 * @ns: common namespace (net)
660 * Returns the net's common namespace.
662 struct ns_common *get_net_ns(struct ns_common *ns)
664 return &get_net(container_of(ns, struct net, ns))->ns;
666 EXPORT_SYMBOL_GPL(get_net_ns);
668 struct net *get_net_ns_by_fd(int fd)
671 struct ns_common *ns;
674 file = proc_ns_fget(fd);
676 return ERR_CAST(file);
678 ns = get_proc_ns(file_inode(file));
679 if (ns->ops == &netns_operations)
680 net = get_net(container_of(ns, struct net, ns));
682 net = ERR_PTR(-EINVAL);
687 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
690 struct net *get_net_ns_by_pid(pid_t pid)
692 struct task_struct *tsk;
695 /* Lookup the network namespace */
696 net = ERR_PTR(-ESRCH);
698 tsk = find_task_by_vpid(pid);
700 struct nsproxy *nsproxy;
702 nsproxy = tsk->nsproxy;
704 net = get_net(nsproxy->net_ns);
710 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
712 static __net_init int net_ns_net_init(struct net *net)
715 net->ns.ops = &netns_operations;
717 return ns_alloc_inum(&net->ns);
720 static __net_exit void net_ns_net_exit(struct net *net)
722 ns_free_inum(&net->ns);
725 static struct pernet_operations __net_initdata net_ns_ops = {
726 .init = net_ns_net_init,
727 .exit = net_ns_net_exit,
730 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
731 [NETNSA_NONE] = { .type = NLA_UNSPEC },
732 [NETNSA_NSID] = { .type = NLA_S32 },
733 [NETNSA_PID] = { .type = NLA_U32 },
734 [NETNSA_FD] = { .type = NLA_U32 },
735 [NETNSA_TARGET_NSID] = { .type = NLA_S32 },
738 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
739 struct netlink_ext_ack *extack)
741 struct net *net = sock_net(skb->sk);
742 struct nlattr *tb[NETNSA_MAX + 1];
747 err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
748 NETNSA_MAX, rtnl_net_policy, extack);
751 if (!tb[NETNSA_NSID]) {
752 NL_SET_ERR_MSG(extack, "nsid is missing");
755 nsid = nla_get_s32(tb[NETNSA_NSID]);
757 if (tb[NETNSA_PID]) {
758 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
759 nla = tb[NETNSA_PID];
760 } else if (tb[NETNSA_FD]) {
761 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
764 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
768 NL_SET_BAD_ATTR(extack, nla);
769 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
770 return PTR_ERR(peer);
773 spin_lock_bh(&net->nsid_lock);
774 if (__peernet2id(net, peer) >= 0) {
775 spin_unlock_bh(&net->nsid_lock);
777 NL_SET_BAD_ATTR(extack, nla);
778 NL_SET_ERR_MSG(extack,
779 "Peer netns already has a nsid assigned");
783 err = alloc_netid(net, peer, nsid);
784 spin_unlock_bh(&net->nsid_lock);
786 rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid,
789 } else if (err == -ENOSPC && nsid >= 0) {
791 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
792 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
799 static int rtnl_net_get_size(void)
801 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
802 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
803 + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
807 struct net_fill_args {
817 static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
819 struct nlmsghdr *nlh;
820 struct rtgenmsg *rth;
822 nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
827 rth = nlmsg_data(nlh);
828 rth->rtgen_family = AF_UNSPEC;
830 if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
831 goto nla_put_failure;
834 nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
835 goto nla_put_failure;
841 nlmsg_cancel(skb, nlh);
845 static int rtnl_net_valid_getid_req(struct sk_buff *skb,
846 const struct nlmsghdr *nlh,
848 struct netlink_ext_ack *extack)
852 if (!netlink_strict_get_check(skb))
853 return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
854 tb, NETNSA_MAX, rtnl_net_policy,
857 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
858 NETNSA_MAX, rtnl_net_policy,
863 for (i = 0; i <= NETNSA_MAX; i++) {
871 case NETNSA_TARGET_NSID:
874 NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
882 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
883 struct netlink_ext_ack *extack)
885 struct net *net = sock_net(skb->sk);
886 struct nlattr *tb[NETNSA_MAX + 1];
887 struct net_fill_args fillargs = {
888 .portid = NETLINK_CB(skb).portid,
889 .seq = nlh->nlmsg_seq,
892 struct net *peer, *target = net;
897 err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
900 if (tb[NETNSA_PID]) {
901 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
902 nla = tb[NETNSA_PID];
903 } else if (tb[NETNSA_FD]) {
904 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
906 } else if (tb[NETNSA_NSID]) {
907 peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
909 peer = ERR_PTR(-ENOENT);
910 nla = tb[NETNSA_NSID];
912 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
917 NL_SET_BAD_ATTR(extack, nla);
918 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
919 return PTR_ERR(peer);
922 if (tb[NETNSA_TARGET_NSID]) {
923 int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
925 target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
926 if (IS_ERR(target)) {
927 NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
928 NL_SET_ERR_MSG(extack,
929 "Target netns reference is invalid");
930 err = PTR_ERR(target);
933 fillargs.add_ref = true;
934 fillargs.ref_nsid = peernet2id(net, peer);
937 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
943 fillargs.nsid = peernet2id(target, peer);
944 err = rtnl_net_fill(msg, &fillargs);
948 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
954 if (fillargs.add_ref)
960 struct rtnl_net_dump_cb {
964 struct net_fill_args fillargs;
969 /* Runs in RCU-critical section. */
970 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
972 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
975 if (net_cb->idx < net_cb->s_idx)
978 net_cb->fillargs.nsid = id;
979 if (net_cb->fillargs.add_ref)
980 net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
981 ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
990 static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
991 struct rtnl_net_dump_cb *net_cb,
992 struct netlink_callback *cb)
994 struct netlink_ext_ack *extack = cb->extack;
995 struct nlattr *tb[NETNSA_MAX + 1];
998 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
999 NETNSA_MAX, rtnl_net_policy,
1004 for (i = 0; i <= NETNSA_MAX; i++) {
1008 if (i == NETNSA_TARGET_NSID) {
1011 net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
1013 NL_SET_BAD_ATTR(extack, tb[i]);
1014 NL_SET_ERR_MSG(extack,
1015 "Invalid target network namespace id");
1016 return PTR_ERR(net);
1018 net_cb->fillargs.add_ref = true;
1019 net_cb->ref_net = net_cb->tgt_net;
1020 net_cb->tgt_net = net;
1022 NL_SET_BAD_ATTR(extack, tb[i]);
1023 NL_SET_ERR_MSG(extack,
1024 "Unsupported attribute in dump request");
1032 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1034 struct rtnl_net_dump_cb net_cb = {
1035 .tgt_net = sock_net(skb->sk),
1038 .portid = NETLINK_CB(cb->skb).portid,
1039 .seq = cb->nlh->nlmsg_seq,
1040 .flags = NLM_F_MULTI,
1044 .s_idx = cb->args[0],
1048 if (cb->strict_check) {
1049 err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1055 idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1058 cb->args[0] = net_cb.idx;
1060 if (net_cb.fillargs.add_ref)
1061 put_net(net_cb.tgt_net);
1062 return err < 0 ? err : skb->len;
1065 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1066 struct nlmsghdr *nlh, gfp_t gfp)
1068 struct net_fill_args fillargs = {
1070 .seq = nlh ? nlh->nlmsg_seq : 0,
1074 struct sk_buff *msg;
1077 msg = nlmsg_new(rtnl_net_get_size(), gfp);
1081 err = rtnl_net_fill(msg, &fillargs);
1085 rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, gfp);
1091 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1094 void __init net_ns_init(void)
1096 struct net_generic *ng;
1098 #ifdef CONFIG_NET_NS
1099 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1101 SLAB_PANIC|SLAB_ACCOUNT, NULL);
1103 /* Create workqueue for cleanup */
1104 netns_wq = create_singlethread_workqueue("netns");
1106 panic("Could not create netns workq");
1109 ng = net_alloc_generic();
1111 panic("Could not allocate generic netns");
1113 rcu_assign_pointer(init_net.gen, ng);
1116 init_net.key_domain = &init_net_key_domain;
1118 down_write(&pernet_ops_rwsem);
1119 if (setup_net(&init_net, &init_user_ns))
1120 panic("Could not setup the initial network namespace");
1122 init_net_initialized = true;
1123 up_write(&pernet_ops_rwsem);
1125 if (register_pernet_subsys(&net_ns_ops))
1126 panic("Could not register network namespace subsystems");
1128 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
1129 RTNL_FLAG_DOIT_UNLOCKED);
1130 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
1131 RTNL_FLAG_DOIT_UNLOCKED);
1134 static void free_exit_list(struct pernet_operations *ops, struct list_head *net_exit_list)
1136 ops_pre_exit_list(ops, net_exit_list);
1138 ops_exit_list(ops, net_exit_list);
1139 ops_free_list(ops, net_exit_list);
1142 #ifdef CONFIG_NET_NS
1143 static int __register_pernet_operations(struct list_head *list,
1144 struct pernet_operations *ops)
1148 LIST_HEAD(net_exit_list);
1150 list_add_tail(&ops->list, list);
1151 if (ops->init || (ops->id && ops->size)) {
1152 /* We held write locked pernet_ops_rwsem, and parallel
1153 * setup_net() and cleanup_net() are not possible.
1156 error = ops_init(ops, net);
1159 list_add_tail(&net->exit_list, &net_exit_list);
1165 /* If I have an error cleanup all namespaces I initialized */
1166 list_del(&ops->list);
1167 free_exit_list(ops, &net_exit_list);
1171 static void __unregister_pernet_operations(struct pernet_operations *ops)
1174 LIST_HEAD(net_exit_list);
1176 list_del(&ops->list);
1177 /* See comment in __register_pernet_operations() */
1179 list_add_tail(&net->exit_list, &net_exit_list);
1181 free_exit_list(ops, &net_exit_list);
1186 static int __register_pernet_operations(struct list_head *list,
1187 struct pernet_operations *ops)
1189 if (!init_net_initialized) {
1190 list_add_tail(&ops->list, list);
1194 return ops_init(ops, &init_net);
1197 static void __unregister_pernet_operations(struct pernet_operations *ops)
1199 if (!init_net_initialized) {
1200 list_del(&ops->list);
1202 LIST_HEAD(net_exit_list);
1203 list_add(&init_net.exit_list, &net_exit_list);
1204 free_exit_list(ops, &net_exit_list);
1208 #endif /* CONFIG_NET_NS */
1210 static DEFINE_IDA(net_generic_ids);
1212 static int register_pernet_operations(struct list_head *list,
1213 struct pernet_operations *ops)
1218 error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID,
1223 /* This does not require READ_ONCE as writers already hold
1224 * pernet_ops_rwsem. But WRITE_ONCE is needed to protect
1225 * net_alloc_generic.
1227 WRITE_ONCE(max_gen_ptrs, max(max_gen_ptrs, *ops->id + 1));
1229 error = __register_pernet_operations(list, ops);
1233 ida_free(&net_generic_ids, *ops->id);
1239 static void unregister_pernet_operations(struct pernet_operations *ops)
1241 __unregister_pernet_operations(ops);
1244 ida_free(&net_generic_ids, *ops->id);
1248 * register_pernet_subsys - register a network namespace subsystem
1249 * @ops: pernet operations structure for the subsystem
1251 * Register a subsystem which has init and exit functions
1252 * that are called when network namespaces are created and
1253 * destroyed respectively.
1255 * When registered all network namespace init functions are
1256 * called for every existing network namespace. Allowing kernel
1257 * modules to have a race free view of the set of network namespaces.
1259 * When a new network namespace is created all of the init
1260 * methods are called in the order in which they were registered.
1262 * When a network namespace is destroyed all of the exit methods
1263 * are called in the reverse of the order with which they were
1266 int register_pernet_subsys(struct pernet_operations *ops)
1269 down_write(&pernet_ops_rwsem);
1270 error = register_pernet_operations(first_device, ops);
1271 up_write(&pernet_ops_rwsem);
1274 EXPORT_SYMBOL_GPL(register_pernet_subsys);
1277 * unregister_pernet_subsys - unregister a network namespace subsystem
1278 * @ops: pernet operations structure to manipulate
1280 * Remove the pernet operations structure from the list to be
1281 * used when network namespaces are created or destroyed. In
1282 * addition run the exit method for all existing network
1285 void unregister_pernet_subsys(struct pernet_operations *ops)
1287 down_write(&pernet_ops_rwsem);
1288 unregister_pernet_operations(ops);
1289 up_write(&pernet_ops_rwsem);
1291 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1294 * register_pernet_device - register a network namespace device
1295 * @ops: pernet operations structure for the subsystem
1297 * Register a device which has init and exit functions
1298 * that are called when network namespaces are created and
1299 * destroyed respectively.
1301 * When registered all network namespace init functions are
1302 * called for every existing network namespace. Allowing kernel
1303 * modules to have a race free view of the set of network namespaces.
1305 * When a new network namespace is created all of the init
1306 * methods are called in the order in which they were registered.
1308 * When a network namespace is destroyed all of the exit methods
1309 * are called in the reverse of the order with which they were
1312 int register_pernet_device(struct pernet_operations *ops)
1315 down_write(&pernet_ops_rwsem);
1316 error = register_pernet_operations(&pernet_list, ops);
1317 if (!error && (first_device == &pernet_list))
1318 first_device = &ops->list;
1319 up_write(&pernet_ops_rwsem);
1322 EXPORT_SYMBOL_GPL(register_pernet_device);
1325 * unregister_pernet_device - unregister a network namespace netdevice
1326 * @ops: pernet operations structure to manipulate
1328 * Remove the pernet operations structure from the list to be
1329 * used when network namespaces are created or destroyed. In
1330 * addition run the exit method for all existing network
1333 void unregister_pernet_device(struct pernet_operations *ops)
1335 down_write(&pernet_ops_rwsem);
1336 if (&ops->list == first_device)
1337 first_device = first_device->next;
1338 unregister_pernet_operations(ops);
1339 up_write(&pernet_ops_rwsem);
1341 EXPORT_SYMBOL_GPL(unregister_pernet_device);
1343 #ifdef CONFIG_NET_NS
1344 static struct ns_common *netns_get(struct task_struct *task)
1346 struct net *net = NULL;
1347 struct nsproxy *nsproxy;
1350 nsproxy = task->nsproxy;
1352 net = get_net(nsproxy->net_ns);
1355 return net ? &net->ns : NULL;
1358 static inline struct net *to_net_ns(struct ns_common *ns)
1360 return container_of(ns, struct net, ns);
1363 static void netns_put(struct ns_common *ns)
1365 put_net(to_net_ns(ns));
1368 static int netns_install(struct nsset *nsset, struct ns_common *ns)
1370 struct nsproxy *nsproxy = nsset->nsproxy;
1371 struct net *net = to_net_ns(ns);
1373 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1374 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
1377 put_net(nsproxy->net_ns);
1378 nsproxy->net_ns = get_net(net);
1382 static struct user_namespace *netns_owner(struct ns_common *ns)
1384 return to_net_ns(ns)->user_ns;
1387 const struct proc_ns_operations netns_operations = {
1389 .type = CLONE_NEWNET,
1392 .install = netns_install,
1393 .owner = netns_owner,