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);
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)
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);
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 put_user_ns(user_ns);
485 dec_net_namespaces(ucounts);
492 * net_ns_get_ownership - get sysfs ownership data for @net
493 * @net: network namespace in question (can be NULL)
494 * @uid: kernel user ID for sysfs objects
495 * @gid: kernel group ID for sysfs objects
497 * Returns the uid/gid pair of root in the user namespace associated with the
498 * given network namespace.
500 void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
503 kuid_t ns_root_uid = make_kuid(net->user_ns, 0);
504 kgid_t ns_root_gid = make_kgid(net->user_ns, 0);
506 if (uid_valid(ns_root_uid))
509 if (gid_valid(ns_root_gid))
512 *uid = GLOBAL_ROOT_UID;
513 *gid = GLOBAL_ROOT_GID;
516 EXPORT_SYMBOL_GPL(net_ns_get_ownership);
518 static void unhash_nsid(struct net *net, struct net *last)
521 /* This function is only called from cleanup_net() work,
522 * and this work is the only process, that may delete
523 * a net from net_namespace_list. So, when the below
524 * is executing, the list may only grow. Thus, we do not
525 * use for_each_net_rcu() or net_rwsem.
530 spin_lock_bh(&tmp->nsid_lock);
531 id = __peernet2id(tmp, net);
533 idr_remove(&tmp->netns_ids, id);
534 spin_unlock_bh(&tmp->nsid_lock);
536 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 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
953 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
956 if (net_cb->idx < net_cb->s_idx)
959 net_cb->fillargs.nsid = id;
960 if (net_cb->fillargs.add_ref)
961 net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
962 ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
971 static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
972 struct rtnl_net_dump_cb *net_cb,
973 struct netlink_callback *cb)
975 struct netlink_ext_ack *extack = cb->extack;
976 struct nlattr *tb[NETNSA_MAX + 1];
979 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
980 NETNSA_MAX, rtnl_net_policy,
985 for (i = 0; i <= NETNSA_MAX; i++) {
989 if (i == NETNSA_TARGET_NSID) {
992 net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
994 NL_SET_BAD_ATTR(extack, tb[i]);
995 NL_SET_ERR_MSG(extack,
996 "Invalid target network namespace id");
999 net_cb->fillargs.add_ref = true;
1000 net_cb->ref_net = net_cb->tgt_net;
1001 net_cb->tgt_net = net;
1003 NL_SET_BAD_ATTR(extack, tb[i]);
1004 NL_SET_ERR_MSG(extack,
1005 "Unsupported attribute in dump request");
1013 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1015 struct rtnl_net_dump_cb net_cb = {
1016 .tgt_net = sock_net(skb->sk),
1019 .portid = NETLINK_CB(cb->skb).portid,
1020 .seq = cb->nlh->nlmsg_seq,
1021 .flags = NLM_F_MULTI,
1025 .s_idx = cb->args[0],
1029 if (cb->strict_check) {
1030 err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1035 spin_lock_bh(&net_cb.tgt_net->nsid_lock);
1036 if (net_cb.fillargs.add_ref &&
1037 !net_eq(net_cb.ref_net, net_cb.tgt_net) &&
1038 !spin_trylock_bh(&net_cb.ref_net->nsid_lock)) {
1039 spin_unlock_bh(&net_cb.tgt_net->nsid_lock);
1043 idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1044 if (net_cb.fillargs.add_ref &&
1045 !net_eq(net_cb.ref_net, net_cb.tgt_net))
1046 spin_unlock_bh(&net_cb.ref_net->nsid_lock);
1047 spin_unlock_bh(&net_cb.tgt_net->nsid_lock);
1049 cb->args[0] = net_cb.idx;
1051 if (net_cb.fillargs.add_ref)
1052 put_net(net_cb.tgt_net);
1053 return err < 0 ? err : skb->len;
1056 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1057 struct nlmsghdr *nlh)
1059 struct net_fill_args fillargs = {
1061 .seq = nlh ? nlh->nlmsg_seq : 0,
1065 struct sk_buff *msg;
1068 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
1072 err = rtnl_net_fill(msg, &fillargs);
1076 rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, 0);
1082 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1085 static int __init net_ns_init(void)
1087 struct net_generic *ng;
1089 #ifdef CONFIG_NET_NS
1090 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1092 SLAB_PANIC|SLAB_ACCOUNT, NULL);
1094 /* Create workqueue for cleanup */
1095 netns_wq = create_singlethread_workqueue("netns");
1097 panic("Could not create netns workq");
1100 ng = net_alloc_generic();
1102 panic("Could not allocate generic netns");
1104 rcu_assign_pointer(init_net.gen, ng);
1106 down_write(&pernet_ops_rwsem);
1107 if (setup_net(&init_net, &init_user_ns))
1108 panic("Could not setup the initial network namespace");
1110 init_net_initialized = true;
1111 up_write(&pernet_ops_rwsem);
1113 if (register_pernet_subsys(&net_ns_ops))
1114 panic("Could not register network namespace subsystems");
1116 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
1117 RTNL_FLAG_DOIT_UNLOCKED);
1118 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
1119 RTNL_FLAG_DOIT_UNLOCKED);
1124 pure_initcall(net_ns_init);
1126 #ifdef CONFIG_NET_NS
1127 static int __register_pernet_operations(struct list_head *list,
1128 struct pernet_operations *ops)
1132 LIST_HEAD(net_exit_list);
1134 list_add_tail(&ops->list, list);
1135 if (ops->init || (ops->id && ops->size)) {
1136 /* We held write locked pernet_ops_rwsem, and parallel
1137 * setup_net() and cleanup_net() are not possible.
1140 error = ops_init(ops, net);
1143 list_add_tail(&net->exit_list, &net_exit_list);
1149 /* If I have an error cleanup all namespaces I initialized */
1150 list_del(&ops->list);
1151 ops_pre_exit_list(ops, &net_exit_list);
1153 ops_exit_list(ops, &net_exit_list);
1154 ops_free_list(ops, &net_exit_list);
1158 static void __unregister_pernet_operations(struct pernet_operations *ops)
1161 LIST_HEAD(net_exit_list);
1163 list_del(&ops->list);
1164 /* See comment in __register_pernet_operations() */
1166 list_add_tail(&net->exit_list, &net_exit_list);
1167 ops_pre_exit_list(ops, &net_exit_list);
1169 ops_exit_list(ops, &net_exit_list);
1170 ops_free_list(ops, &net_exit_list);
1175 static int __register_pernet_operations(struct list_head *list,
1176 struct pernet_operations *ops)
1178 if (!init_net_initialized) {
1179 list_add_tail(&ops->list, list);
1183 return ops_init(ops, &init_net);
1186 static void __unregister_pernet_operations(struct pernet_operations *ops)
1188 if (!init_net_initialized) {
1189 list_del(&ops->list);
1191 LIST_HEAD(net_exit_list);
1192 list_add(&init_net.exit_list, &net_exit_list);
1193 ops_pre_exit_list(ops, &net_exit_list);
1195 ops_exit_list(ops, &net_exit_list);
1196 ops_free_list(ops, &net_exit_list);
1200 #endif /* CONFIG_NET_NS */
1202 static DEFINE_IDA(net_generic_ids);
1204 static int register_pernet_operations(struct list_head *list,
1205 struct pernet_operations *ops)
1210 error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID,
1215 max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
1217 error = __register_pernet_operations(list, ops);
1221 ida_free(&net_generic_ids, *ops->id);
1227 static void unregister_pernet_operations(struct pernet_operations *ops)
1229 __unregister_pernet_operations(ops);
1232 ida_free(&net_generic_ids, *ops->id);
1236 * register_pernet_subsys - register a network namespace subsystem
1237 * @ops: pernet operations structure for the subsystem
1239 * Register a subsystem which has init and exit functions
1240 * that are called when network namespaces are created and
1241 * destroyed respectively.
1243 * When registered all network namespace init functions are
1244 * called for every existing network namespace. Allowing kernel
1245 * modules to have a race free view of the set of network namespaces.
1247 * When a new network namespace is created all of the init
1248 * methods are called in the order in which they were registered.
1250 * When a network namespace is destroyed all of the exit methods
1251 * are called in the reverse of the order with which they were
1254 int register_pernet_subsys(struct pernet_operations *ops)
1257 down_write(&pernet_ops_rwsem);
1258 error = register_pernet_operations(first_device, ops);
1259 up_write(&pernet_ops_rwsem);
1262 EXPORT_SYMBOL_GPL(register_pernet_subsys);
1265 * unregister_pernet_subsys - unregister a network namespace subsystem
1266 * @ops: pernet operations structure to manipulate
1268 * Remove the pernet operations structure from the list to be
1269 * used when network namespaces are created or destroyed. In
1270 * addition run the exit method for all existing network
1273 void unregister_pernet_subsys(struct pernet_operations *ops)
1275 down_write(&pernet_ops_rwsem);
1276 unregister_pernet_operations(ops);
1277 up_write(&pernet_ops_rwsem);
1279 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1282 * register_pernet_device - register a network namespace device
1283 * @ops: pernet operations structure for the subsystem
1285 * Register a device which has init and exit functions
1286 * that are called when network namespaces are created and
1287 * destroyed respectively.
1289 * When registered all network namespace init functions are
1290 * called for every existing network namespace. Allowing kernel
1291 * modules to have a race free view of the set of network namespaces.
1293 * When a new network namespace is created all of the init
1294 * methods are called in the order in which they were registered.
1296 * When a network namespace is destroyed all of the exit methods
1297 * are called in the reverse of the order with which they were
1300 int register_pernet_device(struct pernet_operations *ops)
1303 down_write(&pernet_ops_rwsem);
1304 error = register_pernet_operations(&pernet_list, ops);
1305 if (!error && (first_device == &pernet_list))
1306 first_device = &ops->list;
1307 up_write(&pernet_ops_rwsem);
1310 EXPORT_SYMBOL_GPL(register_pernet_device);
1313 * unregister_pernet_device - unregister a network namespace netdevice
1314 * @ops: pernet operations structure to manipulate
1316 * Remove the pernet operations structure from the list to be
1317 * used when network namespaces are created or destroyed. In
1318 * addition run the exit method for all existing network
1321 void unregister_pernet_device(struct pernet_operations *ops)
1323 down_write(&pernet_ops_rwsem);
1324 if (&ops->list == first_device)
1325 first_device = first_device->next;
1326 unregister_pernet_operations(ops);
1327 up_write(&pernet_ops_rwsem);
1329 EXPORT_SYMBOL_GPL(unregister_pernet_device);
1331 #ifdef CONFIG_NET_NS
1332 static struct ns_common *netns_get(struct task_struct *task)
1334 struct net *net = NULL;
1335 struct nsproxy *nsproxy;
1338 nsproxy = task->nsproxy;
1340 net = get_net(nsproxy->net_ns);
1343 return net ? &net->ns : NULL;
1346 static inline struct net *to_net_ns(struct ns_common *ns)
1348 return container_of(ns, struct net, ns);
1351 static void netns_put(struct ns_common *ns)
1353 put_net(to_net_ns(ns));
1356 static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
1358 struct net *net = to_net_ns(ns);
1360 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1361 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1364 put_net(nsproxy->net_ns);
1365 nsproxy->net_ns = get_net(net);
1369 static struct user_namespace *netns_owner(struct ns_common *ns)
1371 return to_net_ns(ns)->user_ns;
1374 const struct proc_ns_operations netns_operations = {
1376 .type = CLONE_NEWNET,
1379 .install = netns_install,
1380 .owner = netns_owner,