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, u32 portid,
249 struct nlmsghdr *nlh, gfp_t gfp);
250 /* This function returns the id of a peer netns. If no id is assigned, one will
251 * be allocated and returned.
253 int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp)
255 bool alloc = false, alive = false;
258 if (refcount_read(&net->count) == 0)
259 return NETNSA_NSID_NOT_ASSIGNED;
260 spin_lock_bh(&net->nsid_lock);
262 * When peer is obtained from RCU lists, we may race with
263 * its cleanup. Check whether it's alive, and this guarantees
264 * we never hash a peer back to net->netns_ids, after it has
265 * just been idr_remove()'d from there in cleanup_net().
267 if (maybe_get_net(peer))
268 alive = alloc = true;
269 id = __peernet2id_alloc(net, peer, &alloc);
270 spin_unlock_bh(&net->nsid_lock);
271 if (alloc && id >= 0)
272 rtnl_net_notifyid(net, RTM_NEWNSID, id, 0, NULL, gfp);
277 EXPORT_SYMBOL_GPL(peernet2id_alloc);
279 /* This function returns, if assigned, the id of a peer netns. */
280 int peernet2id(struct net *net, struct net *peer)
284 spin_lock_bh(&net->nsid_lock);
285 id = __peernet2id(net, peer);
286 spin_unlock_bh(&net->nsid_lock);
289 EXPORT_SYMBOL(peernet2id);
291 /* This function returns true is the peer netns has an id assigned into the
294 bool peernet_has_id(struct net *net, struct net *peer)
296 return peernet2id(net, peer) >= 0;
299 struct net *get_net_ns_by_id(struct net *net, int id)
307 peer = idr_find(&net->netns_ids, id);
309 peer = maybe_get_net(peer);
316 * setup_net runs the initializers for the network namespace object.
318 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
320 /* Must be called with pernet_ops_rwsem held */
321 const struct pernet_operations *ops, *saved_ops;
323 LIST_HEAD(net_exit_list);
325 refcount_set(&net->count, 1);
326 refcount_set(&net->passive, 1);
327 get_random_bytes(&net->hash_mix, sizeof(u32));
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;
374 static struct pernet_operations net_defaults_ops = {
375 .init = net_defaults_init_net,
378 static __init int net_defaults_init(void)
380 if (register_pernet_subsys(&net_defaults_ops))
381 panic("Cannot initialize net default settings");
386 core_initcall(net_defaults_init);
389 static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
391 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
394 static void dec_net_namespaces(struct ucounts *ucounts)
396 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
399 static struct kmem_cache *net_cachep __ro_after_init;
400 static struct workqueue_struct *netns_wq;
402 static struct net *net_alloc(void)
404 struct net *net = NULL;
405 struct net_generic *ng;
407 ng = net_alloc_generic();
411 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
416 net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
417 if (!net->key_domain)
419 refcount_set(&net->key_domain->usage, 1);
422 rcu_assign_pointer(net->gen, ng);
428 kmem_cache_free(net_cachep, net);
436 static void net_free(struct net *net)
438 kfree(rcu_access_pointer(net->gen));
439 kmem_cache_free(net_cachep, net);
442 void net_drop_ns(void *p)
445 if (ns && refcount_dec_and_test(&ns->passive))
449 struct net *copy_net_ns(unsigned long flags,
450 struct user_namespace *user_ns, struct net *old_net)
452 struct ucounts *ucounts;
456 if (!(flags & CLONE_NEWNET))
457 return get_net(old_net);
459 ucounts = inc_net_namespaces(user_ns);
461 return ERR_PTR(-ENOSPC);
468 refcount_set(&net->passive, 1);
469 net->ucounts = ucounts;
470 get_user_ns(user_ns);
472 rv = down_read_killable(&pernet_ops_rwsem);
476 rv = setup_net(net, user_ns);
478 up_read(&pernet_ops_rwsem);
482 key_remove_domain(net->key_domain);
483 put_user_ns(user_ns);
486 dec_net_namespaces(ucounts);
493 * net_ns_get_ownership - get sysfs ownership data for @net
494 * @net: network namespace in question (can be NULL)
495 * @uid: kernel user ID for sysfs objects
496 * @gid: kernel group ID for sysfs objects
498 * Returns the uid/gid pair of root in the user namespace associated with the
499 * given network namespace.
501 void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
504 kuid_t ns_root_uid = make_kuid(net->user_ns, 0);
505 kgid_t ns_root_gid = make_kgid(net->user_ns, 0);
507 if (uid_valid(ns_root_uid))
510 if (gid_valid(ns_root_gid))
513 *uid = GLOBAL_ROOT_UID;
514 *gid = GLOBAL_ROOT_GID;
517 EXPORT_SYMBOL_GPL(net_ns_get_ownership);
519 static void unhash_nsid(struct net *net, struct net *last)
522 /* This function is only called from cleanup_net() work,
523 * and this work is the only process, that may delete
524 * a net from net_namespace_list. So, when the below
525 * is executing, the list may only grow. Thus, we do not
526 * use for_each_net_rcu() or net_rwsem.
531 spin_lock_bh(&tmp->nsid_lock);
532 id = __peernet2id(tmp, net);
534 idr_remove(&tmp->netns_ids, id);
535 spin_unlock_bh(&tmp->nsid_lock);
537 rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL,
542 spin_lock_bh(&net->nsid_lock);
543 idr_destroy(&net->netns_ids);
544 spin_unlock_bh(&net->nsid_lock);
547 static LLIST_HEAD(cleanup_list);
549 static void cleanup_net(struct work_struct *work)
551 const struct pernet_operations *ops;
552 struct net *net, *tmp, *last;
553 struct llist_node *net_kill_list;
554 LIST_HEAD(net_exit_list);
556 /* Atomically snapshot the list of namespaces to cleanup */
557 net_kill_list = llist_del_all(&cleanup_list);
559 down_read(&pernet_ops_rwsem);
561 /* Don't let anyone else find us. */
562 down_write(&net_rwsem);
563 llist_for_each_entry(net, net_kill_list, cleanup_list)
564 list_del_rcu(&net->list);
565 /* Cache last net. After we unlock rtnl, no one new net
566 * added to net_namespace_list can assign nsid pointer
567 * to a net from net_kill_list (see peernet2id_alloc()).
568 * So, we skip them in unhash_nsid().
570 * Note, that unhash_nsid() does not delete nsid links
571 * between net_kill_list's nets, as they've already
572 * deleted from net_namespace_list. But, this would be
573 * useless anyway, as netns_ids are destroyed there.
575 last = list_last_entry(&net_namespace_list, struct net, list);
576 up_write(&net_rwsem);
578 llist_for_each_entry(net, net_kill_list, cleanup_list) {
579 unhash_nsid(net, last);
580 list_add_tail(&net->exit_list, &net_exit_list);
583 /* Run all of the network namespace pre_exit methods */
584 list_for_each_entry_reverse(ops, &pernet_list, list)
585 ops_pre_exit_list(ops, &net_exit_list);
588 * Another CPU might be rcu-iterating the list, wait for it.
589 * This needs to be before calling the exit() notifiers, so
590 * the rcu_barrier() below isn't sufficient alone.
591 * Also the pre_exit() and exit() methods need this barrier.
595 /* Run all of the network namespace exit methods */
596 list_for_each_entry_reverse(ops, &pernet_list, list)
597 ops_exit_list(ops, &net_exit_list);
599 /* Free the net generic variables */
600 list_for_each_entry_reverse(ops, &pernet_list, list)
601 ops_free_list(ops, &net_exit_list);
603 up_read(&pernet_ops_rwsem);
605 /* Ensure there are no outstanding rcu callbacks using this
610 /* Finally it is safe to free my network namespace structure */
611 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
612 list_del_init(&net->exit_list);
613 dec_net_namespaces(net->ucounts);
614 key_remove_domain(net->key_domain);
615 put_user_ns(net->user_ns);
621 * net_ns_barrier - wait until concurrent net_cleanup_work is done
623 * cleanup_net runs from work queue and will first remove namespaces
624 * from the global list, then run net exit functions.
626 * Call this in module exit path to make sure that all netns
627 * ->exit ops have been invoked before the function is removed.
629 void net_ns_barrier(void)
631 down_write(&pernet_ops_rwsem);
632 up_write(&pernet_ops_rwsem);
634 EXPORT_SYMBOL(net_ns_barrier);
636 static DECLARE_WORK(net_cleanup_work, cleanup_net);
638 void __put_net(struct net *net)
640 /* Cleanup the network namespace in process context */
641 if (llist_add(&net->cleanup_list, &cleanup_list))
642 queue_work(netns_wq, &net_cleanup_work);
644 EXPORT_SYMBOL_GPL(__put_net);
646 struct net *get_net_ns_by_fd(int fd)
649 struct ns_common *ns;
652 file = proc_ns_fget(fd);
654 return ERR_CAST(file);
656 ns = get_proc_ns(file_inode(file));
657 if (ns->ops == &netns_operations)
658 net = get_net(container_of(ns, struct net, ns));
660 net = ERR_PTR(-EINVAL);
667 struct net *get_net_ns_by_fd(int fd)
669 return ERR_PTR(-EINVAL);
672 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
674 struct net *get_net_ns_by_pid(pid_t pid)
676 struct task_struct *tsk;
679 /* Lookup the network namespace */
680 net = ERR_PTR(-ESRCH);
682 tsk = find_task_by_vpid(pid);
684 struct nsproxy *nsproxy;
686 nsproxy = tsk->nsproxy;
688 net = get_net(nsproxy->net_ns);
694 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
696 static __net_init int net_ns_net_init(struct net *net)
699 net->ns.ops = &netns_operations;
701 return ns_alloc_inum(&net->ns);
704 static __net_exit void net_ns_net_exit(struct net *net)
706 ns_free_inum(&net->ns);
709 static struct pernet_operations __net_initdata net_ns_ops = {
710 .init = net_ns_net_init,
711 .exit = net_ns_net_exit,
714 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
715 [NETNSA_NONE] = { .type = NLA_UNSPEC },
716 [NETNSA_NSID] = { .type = NLA_S32 },
717 [NETNSA_PID] = { .type = NLA_U32 },
718 [NETNSA_FD] = { .type = NLA_U32 },
719 [NETNSA_TARGET_NSID] = { .type = NLA_S32 },
722 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
723 struct netlink_ext_ack *extack)
725 struct net *net = sock_net(skb->sk);
726 struct nlattr *tb[NETNSA_MAX + 1];
731 err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
732 NETNSA_MAX, rtnl_net_policy, extack);
735 if (!tb[NETNSA_NSID]) {
736 NL_SET_ERR_MSG(extack, "nsid is missing");
739 nsid = nla_get_s32(tb[NETNSA_NSID]);
741 if (tb[NETNSA_PID]) {
742 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
743 nla = tb[NETNSA_PID];
744 } else if (tb[NETNSA_FD]) {
745 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
748 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
752 NL_SET_BAD_ATTR(extack, nla);
753 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
754 return PTR_ERR(peer);
757 spin_lock_bh(&net->nsid_lock);
758 if (__peernet2id(net, peer) >= 0) {
759 spin_unlock_bh(&net->nsid_lock);
761 NL_SET_BAD_ATTR(extack, nla);
762 NL_SET_ERR_MSG(extack,
763 "Peer netns already has a nsid assigned");
767 err = alloc_netid(net, peer, nsid);
768 spin_unlock_bh(&net->nsid_lock);
770 rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid,
773 } else if (err == -ENOSPC && nsid >= 0) {
775 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
776 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
783 static int rtnl_net_get_size(void)
785 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
786 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
787 + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
791 struct net_fill_args {
801 static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
803 struct nlmsghdr *nlh;
804 struct rtgenmsg *rth;
806 nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
811 rth = nlmsg_data(nlh);
812 rth->rtgen_family = AF_UNSPEC;
814 if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
815 goto nla_put_failure;
818 nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
819 goto nla_put_failure;
825 nlmsg_cancel(skb, nlh);
829 static int rtnl_net_valid_getid_req(struct sk_buff *skb,
830 const struct nlmsghdr *nlh,
832 struct netlink_ext_ack *extack)
836 if (!netlink_strict_get_check(skb))
837 return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
838 tb, NETNSA_MAX, rtnl_net_policy,
841 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
842 NETNSA_MAX, rtnl_net_policy,
847 for (i = 0; i <= NETNSA_MAX; i++) {
855 case NETNSA_TARGET_NSID:
858 NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
866 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
867 struct netlink_ext_ack *extack)
869 struct net *net = sock_net(skb->sk);
870 struct nlattr *tb[NETNSA_MAX + 1];
871 struct net_fill_args fillargs = {
872 .portid = NETLINK_CB(skb).portid,
873 .seq = nlh->nlmsg_seq,
876 struct net *peer, *target = net;
881 err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
884 if (tb[NETNSA_PID]) {
885 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
886 nla = tb[NETNSA_PID];
887 } else if (tb[NETNSA_FD]) {
888 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
890 } else if (tb[NETNSA_NSID]) {
891 peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
893 peer = ERR_PTR(-ENOENT);
894 nla = tb[NETNSA_NSID];
896 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
901 NL_SET_BAD_ATTR(extack, nla);
902 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
903 return PTR_ERR(peer);
906 if (tb[NETNSA_TARGET_NSID]) {
907 int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
909 target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
910 if (IS_ERR(target)) {
911 NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
912 NL_SET_ERR_MSG(extack,
913 "Target netns reference is invalid");
914 err = PTR_ERR(target);
917 fillargs.add_ref = true;
918 fillargs.ref_nsid = peernet2id(net, peer);
921 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
927 fillargs.nsid = peernet2id(target, peer);
928 err = rtnl_net_fill(msg, &fillargs);
932 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
938 if (fillargs.add_ref)
944 struct rtnl_net_dump_cb {
948 struct net_fill_args fillargs;
953 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
955 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
958 if (net_cb->idx < net_cb->s_idx)
961 net_cb->fillargs.nsid = id;
962 if (net_cb->fillargs.add_ref)
963 net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
964 ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
973 static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
974 struct rtnl_net_dump_cb *net_cb,
975 struct netlink_callback *cb)
977 struct netlink_ext_ack *extack = cb->extack;
978 struct nlattr *tb[NETNSA_MAX + 1];
981 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
982 NETNSA_MAX, rtnl_net_policy,
987 for (i = 0; i <= NETNSA_MAX; i++) {
991 if (i == NETNSA_TARGET_NSID) {
994 net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
996 NL_SET_BAD_ATTR(extack, tb[i]);
997 NL_SET_ERR_MSG(extack,
998 "Invalid target network namespace id");
1001 net_cb->fillargs.add_ref = true;
1002 net_cb->ref_net = net_cb->tgt_net;
1003 net_cb->tgt_net = net;
1005 NL_SET_BAD_ATTR(extack, tb[i]);
1006 NL_SET_ERR_MSG(extack,
1007 "Unsupported attribute in dump request");
1015 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1017 struct rtnl_net_dump_cb net_cb = {
1018 .tgt_net = sock_net(skb->sk),
1021 .portid = NETLINK_CB(cb->skb).portid,
1022 .seq = cb->nlh->nlmsg_seq,
1023 .flags = NLM_F_MULTI,
1027 .s_idx = cb->args[0],
1031 if (cb->strict_check) {
1032 err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1037 spin_lock_bh(&net_cb.tgt_net->nsid_lock);
1038 if (net_cb.fillargs.add_ref &&
1039 !net_eq(net_cb.ref_net, net_cb.tgt_net) &&
1040 !spin_trylock_bh(&net_cb.ref_net->nsid_lock)) {
1041 spin_unlock_bh(&net_cb.tgt_net->nsid_lock);
1045 idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1046 if (net_cb.fillargs.add_ref &&
1047 !net_eq(net_cb.ref_net, net_cb.tgt_net))
1048 spin_unlock_bh(&net_cb.ref_net->nsid_lock);
1049 spin_unlock_bh(&net_cb.tgt_net->nsid_lock);
1051 cb->args[0] = net_cb.idx;
1053 if (net_cb.fillargs.add_ref)
1054 put_net(net_cb.tgt_net);
1055 return err < 0 ? err : skb->len;
1058 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1059 struct nlmsghdr *nlh, gfp_t gfp)
1061 struct net_fill_args fillargs = {
1063 .seq = nlh ? nlh->nlmsg_seq : 0,
1067 struct sk_buff *msg;
1070 msg = nlmsg_new(rtnl_net_get_size(), gfp);
1074 err = rtnl_net_fill(msg, &fillargs);
1078 rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, gfp);
1084 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1087 static int __init net_ns_init(void)
1089 struct net_generic *ng;
1091 #ifdef CONFIG_NET_NS
1092 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1094 SLAB_PANIC|SLAB_ACCOUNT, NULL);
1096 /* Create workqueue for cleanup */
1097 netns_wq = create_singlethread_workqueue("netns");
1099 panic("Could not create netns workq");
1102 ng = net_alloc_generic();
1104 panic("Could not allocate generic netns");
1106 rcu_assign_pointer(init_net.gen, ng);
1108 down_write(&pernet_ops_rwsem);
1109 if (setup_net(&init_net, &init_user_ns))
1110 panic("Could not setup the initial network namespace");
1112 init_net_initialized = true;
1113 up_write(&pernet_ops_rwsem);
1115 if (register_pernet_subsys(&net_ns_ops))
1116 panic("Could not register network namespace subsystems");
1118 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
1119 RTNL_FLAG_DOIT_UNLOCKED);
1120 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
1121 RTNL_FLAG_DOIT_UNLOCKED);
1126 pure_initcall(net_ns_init);
1128 #ifdef CONFIG_NET_NS
1129 static int __register_pernet_operations(struct list_head *list,
1130 struct pernet_operations *ops)
1134 LIST_HEAD(net_exit_list);
1136 list_add_tail(&ops->list, list);
1137 if (ops->init || (ops->id && ops->size)) {
1138 /* We held write locked pernet_ops_rwsem, and parallel
1139 * setup_net() and cleanup_net() are not possible.
1142 error = ops_init(ops, net);
1145 list_add_tail(&net->exit_list, &net_exit_list);
1151 /* If I have an error cleanup all namespaces I initialized */
1152 list_del(&ops->list);
1153 ops_pre_exit_list(ops, &net_exit_list);
1155 ops_exit_list(ops, &net_exit_list);
1156 ops_free_list(ops, &net_exit_list);
1160 static void __unregister_pernet_operations(struct pernet_operations *ops)
1163 LIST_HEAD(net_exit_list);
1165 list_del(&ops->list);
1166 /* See comment in __register_pernet_operations() */
1168 list_add_tail(&net->exit_list, &net_exit_list);
1169 ops_pre_exit_list(ops, &net_exit_list);
1171 ops_exit_list(ops, &net_exit_list);
1172 ops_free_list(ops, &net_exit_list);
1177 static int __register_pernet_operations(struct list_head *list,
1178 struct pernet_operations *ops)
1180 if (!init_net_initialized) {
1181 list_add_tail(&ops->list, list);
1185 return ops_init(ops, &init_net);
1188 static void __unregister_pernet_operations(struct pernet_operations *ops)
1190 if (!init_net_initialized) {
1191 list_del(&ops->list);
1193 LIST_HEAD(net_exit_list);
1194 list_add(&init_net.exit_list, &net_exit_list);
1195 ops_pre_exit_list(ops, &net_exit_list);
1197 ops_exit_list(ops, &net_exit_list);
1198 ops_free_list(ops, &net_exit_list);
1202 #endif /* CONFIG_NET_NS */
1204 static DEFINE_IDA(net_generic_ids);
1206 static int register_pernet_operations(struct list_head *list,
1207 struct pernet_operations *ops)
1212 error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID,
1217 max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
1219 error = __register_pernet_operations(list, ops);
1223 ida_free(&net_generic_ids, *ops->id);
1229 static void unregister_pernet_operations(struct pernet_operations *ops)
1231 __unregister_pernet_operations(ops);
1234 ida_free(&net_generic_ids, *ops->id);
1238 * register_pernet_subsys - register a network namespace subsystem
1239 * @ops: pernet operations structure for the subsystem
1241 * Register a subsystem which has init and exit functions
1242 * that are called when network namespaces are created and
1243 * destroyed respectively.
1245 * When registered all network namespace init functions are
1246 * called for every existing network namespace. Allowing kernel
1247 * modules to have a race free view of the set of network namespaces.
1249 * When a new network namespace is created all of the init
1250 * methods are called in the order in which they were registered.
1252 * When a network namespace is destroyed all of the exit methods
1253 * are called in the reverse of the order with which they were
1256 int register_pernet_subsys(struct pernet_operations *ops)
1259 down_write(&pernet_ops_rwsem);
1260 error = register_pernet_operations(first_device, ops);
1261 up_write(&pernet_ops_rwsem);
1264 EXPORT_SYMBOL_GPL(register_pernet_subsys);
1267 * unregister_pernet_subsys - unregister a network namespace subsystem
1268 * @ops: pernet operations structure to manipulate
1270 * Remove the pernet operations structure from the list to be
1271 * used when network namespaces are created or destroyed. In
1272 * addition run the exit method for all existing network
1275 void unregister_pernet_subsys(struct pernet_operations *ops)
1277 down_write(&pernet_ops_rwsem);
1278 unregister_pernet_operations(ops);
1279 up_write(&pernet_ops_rwsem);
1281 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1284 * register_pernet_device - register a network namespace device
1285 * @ops: pernet operations structure for the subsystem
1287 * Register a device which has init and exit functions
1288 * that are called when network namespaces are created and
1289 * destroyed respectively.
1291 * When registered all network namespace init functions are
1292 * called for every existing network namespace. Allowing kernel
1293 * modules to have a race free view of the set of network namespaces.
1295 * When a new network namespace is created all of the init
1296 * methods are called in the order in which they were registered.
1298 * When a network namespace is destroyed all of the exit methods
1299 * are called in the reverse of the order with which they were
1302 int register_pernet_device(struct pernet_operations *ops)
1305 down_write(&pernet_ops_rwsem);
1306 error = register_pernet_operations(&pernet_list, ops);
1307 if (!error && (first_device == &pernet_list))
1308 first_device = &ops->list;
1309 up_write(&pernet_ops_rwsem);
1312 EXPORT_SYMBOL_GPL(register_pernet_device);
1315 * unregister_pernet_device - unregister a network namespace netdevice
1316 * @ops: pernet operations structure to manipulate
1318 * Remove the pernet operations structure from the list to be
1319 * used when network namespaces are created or destroyed. In
1320 * addition run the exit method for all existing network
1323 void unregister_pernet_device(struct pernet_operations *ops)
1325 down_write(&pernet_ops_rwsem);
1326 if (&ops->list == first_device)
1327 first_device = first_device->next;
1328 unregister_pernet_operations(ops);
1329 up_write(&pernet_ops_rwsem);
1331 EXPORT_SYMBOL_GPL(unregister_pernet_device);
1333 #ifdef CONFIG_NET_NS
1334 static struct ns_common *netns_get(struct task_struct *task)
1336 struct net *net = NULL;
1337 struct nsproxy *nsproxy;
1340 nsproxy = task->nsproxy;
1342 net = get_net(nsproxy->net_ns);
1345 return net ? &net->ns : NULL;
1348 static inline struct net *to_net_ns(struct ns_common *ns)
1350 return container_of(ns, struct net, ns);
1353 static void netns_put(struct ns_common *ns)
1355 put_net(to_net_ns(ns));
1358 static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
1360 struct net *net = to_net_ns(ns);
1362 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1363 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1366 put_net(nsproxy->net_ns);
1367 nsproxy->net_ns = get_net(net);
1371 static struct user_namespace *netns_owner(struct ns_common *ns)
1373 return to_net_ns(ns)->user_ns;
1376 const struct proc_ns_operations netns_operations = {
1378 .type = CLONE_NEWNET,
1381 .install = netns_install,
1382 .owner = netns_owner,