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_pre_exit_list(const struct pernet_operations *ops,
152 struct list_head *net_exit_list)
157 list_for_each_entry(net, net_exit_list, exit_list)
162 static void ops_exit_list(const struct pernet_operations *ops,
163 struct list_head *net_exit_list)
167 list_for_each_entry(net, net_exit_list, exit_list) {
173 ops->exit_batch(net_exit_list);
176 static void ops_free_list(const struct pernet_operations *ops,
177 struct list_head *net_exit_list)
180 if (ops->size && ops->id) {
181 list_for_each_entry(net, net_exit_list, exit_list)
182 kfree(net_generic(net, *ops->id));
186 /* should be called with nsid_lock held */
187 static int alloc_netid(struct net *net, struct net *peer, int reqid)
189 int min = 0, max = 0;
196 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
199 /* This function is used by idr_for_each(). If net is equal to peer, the
200 * function returns the id so that idr_for_each() stops. Because we cannot
201 * returns the id 0 (idr_for_each() will not stop), we return the magic value
202 * NET_ID_ZERO (-1) for it.
204 #define NET_ID_ZERO -1
205 static int net_eq_idr(int id, void *net, void *peer)
207 if (net_eq(net, peer))
208 return id ? : NET_ID_ZERO;
212 /* Must be called from RCU-critical section or with nsid_lock held */
213 static int __peernet2id(const struct net *net, struct net *peer)
215 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
217 /* Magic value for id 0. */
218 if (id == NET_ID_ZERO)
223 return NETNSA_NSID_NOT_ASSIGNED;
226 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
227 struct nlmsghdr *nlh, gfp_t gfp);
228 /* This function returns the id of a peer netns. If no id is assigned, one will
229 * be allocated and returned.
231 int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp)
235 if (refcount_read(&net->ns.count) == 0)
236 return NETNSA_NSID_NOT_ASSIGNED;
238 spin_lock_bh(&net->nsid_lock);
239 id = __peernet2id(net, peer);
241 spin_unlock_bh(&net->nsid_lock);
245 /* When peer is obtained from RCU lists, we may race with
246 * its cleanup. Check whether it's alive, and this guarantees
247 * we never hash a peer back to net->netns_ids, after it has
248 * just been idr_remove()'d from there in cleanup_net().
250 if (!maybe_get_net(peer)) {
251 spin_unlock_bh(&net->nsid_lock);
252 return NETNSA_NSID_NOT_ASSIGNED;
255 id = alloc_netid(net, peer, -1);
256 spin_unlock_bh(&net->nsid_lock);
260 return NETNSA_NSID_NOT_ASSIGNED;
262 rtnl_net_notifyid(net, RTM_NEWNSID, id, 0, NULL, gfp);
266 EXPORT_SYMBOL_GPL(peernet2id_alloc);
268 /* This function returns, if assigned, the id of a peer netns. */
269 int peernet2id(const struct net *net, struct net *peer)
274 id = __peernet2id(net, peer);
279 EXPORT_SYMBOL(peernet2id);
281 /* This function returns true is the peer netns has an id assigned into the
284 bool peernet_has_id(const struct net *net, struct net *peer)
286 return peernet2id(net, peer) >= 0;
289 struct net *get_net_ns_by_id(const struct net *net, int id)
297 peer = idr_find(&net->netns_ids, id);
299 peer = maybe_get_net(peer);
306 * setup_net runs the initializers for the network namespace object.
308 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
310 /* Must be called with pernet_ops_rwsem held */
311 const struct pernet_operations *ops, *saved_ops;
313 LIST_HEAD(net_exit_list);
315 refcount_set(&net->ns.count, 1);
316 ref_tracker_dir_init(&net->refcnt_tracker, 128);
318 refcount_set(&net->passive, 1);
319 get_random_bytes(&net->hash_mix, sizeof(u32));
321 net->net_cookie = gen_cookie_next(&net_cookie);
323 net->dev_base_seq = 1;
324 net->user_ns = user_ns;
325 idr_init(&net->netns_ids);
326 spin_lock_init(&net->nsid_lock);
327 mutex_init(&net->ipv4.ra_mutex);
329 list_for_each_entry(ops, &pernet_list, list) {
330 error = ops_init(ops, net);
334 down_write(&net_rwsem);
335 list_add_tail_rcu(&net->list, &net_namespace_list);
336 up_write(&net_rwsem);
341 /* Walk through the list backwards calling the exit functions
342 * for the pernet modules whose init functions did not fail.
344 list_add(&net->exit_list, &net_exit_list);
346 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
347 ops_pre_exit_list(ops, &net_exit_list);
352 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
353 ops_exit_list(ops, &net_exit_list);
356 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
357 ops_free_list(ops, &net_exit_list);
363 static int __net_init net_defaults_init_net(struct net *net)
365 net->core.sysctl_somaxconn = SOMAXCONN;
369 static struct pernet_operations net_defaults_ops = {
370 .init = net_defaults_init_net,
373 static __init int net_defaults_init(void)
375 if (register_pernet_subsys(&net_defaults_ops))
376 panic("Cannot initialize net default settings");
381 core_initcall(net_defaults_init);
384 static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
386 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
389 static void dec_net_namespaces(struct ucounts *ucounts)
391 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
394 static struct kmem_cache *net_cachep __ro_after_init;
395 static struct workqueue_struct *netns_wq;
397 static struct net *net_alloc(void)
399 struct net *net = NULL;
400 struct net_generic *ng;
402 ng = net_alloc_generic();
406 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
411 net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
412 if (!net->key_domain)
414 refcount_set(&net->key_domain->usage, 1);
417 rcu_assign_pointer(net->gen, ng);
423 kmem_cache_free(net_cachep, net);
431 static void net_free(struct net *net)
433 if (refcount_dec_and_test(&net->passive)) {
434 kfree(rcu_access_pointer(net->gen));
435 kmem_cache_free(net_cachep, net);
439 void net_drop_ns(void *p)
441 struct net *net = (struct net *)p;
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);
481 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);
615 key_remove_domain(net->key_domain);
617 put_user_ns(net->user_ns);
623 * net_ns_barrier - wait until concurrent net_cleanup_work is done
625 * cleanup_net runs from work queue and will first remove namespaces
626 * from the global list, then run net exit functions.
628 * Call this in module exit path to make sure that all netns
629 * ->exit ops have been invoked before the function is removed.
631 void net_ns_barrier(void)
633 down_write(&pernet_ops_rwsem);
634 up_write(&pernet_ops_rwsem);
636 EXPORT_SYMBOL(net_ns_barrier);
638 static DECLARE_WORK(net_cleanup_work, cleanup_net);
640 void __put_net(struct net *net)
642 ref_tracker_dir_exit(&net->refcnt_tracker);
643 /* Cleanup the network namespace in process context */
644 if (llist_add(&net->cleanup_list, &cleanup_list))
645 queue_work(netns_wq, &net_cleanup_work);
647 EXPORT_SYMBOL_GPL(__put_net);
650 * get_net_ns - increment the refcount of the network namespace
651 * @ns: common namespace (net)
653 * Returns the net's common namespace.
655 struct ns_common *get_net_ns(struct ns_common *ns)
657 return &get_net(container_of(ns, struct net, ns))->ns;
659 EXPORT_SYMBOL_GPL(get_net_ns);
661 struct net *get_net_ns_by_fd(int fd)
664 struct ns_common *ns;
667 file = proc_ns_fget(fd);
669 return ERR_CAST(file);
671 ns = get_proc_ns(file_inode(file));
672 if (ns->ops == &netns_operations)
673 net = get_net(container_of(ns, struct net, ns));
675 net = ERR_PTR(-EINVAL);
680 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
683 struct net *get_net_ns_by_pid(pid_t pid)
685 struct task_struct *tsk;
688 /* Lookup the network namespace */
689 net = ERR_PTR(-ESRCH);
691 tsk = find_task_by_vpid(pid);
693 struct nsproxy *nsproxy;
695 nsproxy = tsk->nsproxy;
697 net = get_net(nsproxy->net_ns);
703 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
705 static __net_init int net_ns_net_init(struct net *net)
708 net->ns.ops = &netns_operations;
710 return ns_alloc_inum(&net->ns);
713 static __net_exit void net_ns_net_exit(struct net *net)
715 ns_free_inum(&net->ns);
718 static struct pernet_operations __net_initdata net_ns_ops = {
719 .init = net_ns_net_init,
720 .exit = net_ns_net_exit,
723 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
724 [NETNSA_NONE] = { .type = NLA_UNSPEC },
725 [NETNSA_NSID] = { .type = NLA_S32 },
726 [NETNSA_PID] = { .type = NLA_U32 },
727 [NETNSA_FD] = { .type = NLA_U32 },
728 [NETNSA_TARGET_NSID] = { .type = NLA_S32 },
731 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
732 struct netlink_ext_ack *extack)
734 struct net *net = sock_net(skb->sk);
735 struct nlattr *tb[NETNSA_MAX + 1];
740 err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
741 NETNSA_MAX, rtnl_net_policy, extack);
744 if (!tb[NETNSA_NSID]) {
745 NL_SET_ERR_MSG(extack, "nsid is missing");
748 nsid = nla_get_s32(tb[NETNSA_NSID]);
750 if (tb[NETNSA_PID]) {
751 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
752 nla = tb[NETNSA_PID];
753 } else if (tb[NETNSA_FD]) {
754 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
757 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
761 NL_SET_BAD_ATTR(extack, nla);
762 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
763 return PTR_ERR(peer);
766 spin_lock_bh(&net->nsid_lock);
767 if (__peernet2id(net, peer) >= 0) {
768 spin_unlock_bh(&net->nsid_lock);
770 NL_SET_BAD_ATTR(extack, nla);
771 NL_SET_ERR_MSG(extack,
772 "Peer netns already has a nsid assigned");
776 err = alloc_netid(net, peer, nsid);
777 spin_unlock_bh(&net->nsid_lock);
779 rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid,
782 } else if (err == -ENOSPC && nsid >= 0) {
784 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
785 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
792 static int rtnl_net_get_size(void)
794 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
795 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
796 + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
800 struct net_fill_args {
810 static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
812 struct nlmsghdr *nlh;
813 struct rtgenmsg *rth;
815 nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
820 rth = nlmsg_data(nlh);
821 rth->rtgen_family = AF_UNSPEC;
823 if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
824 goto nla_put_failure;
827 nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
828 goto nla_put_failure;
834 nlmsg_cancel(skb, nlh);
838 static int rtnl_net_valid_getid_req(struct sk_buff *skb,
839 const struct nlmsghdr *nlh,
841 struct netlink_ext_ack *extack)
845 if (!netlink_strict_get_check(skb))
846 return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
847 tb, NETNSA_MAX, rtnl_net_policy,
850 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
851 NETNSA_MAX, rtnl_net_policy,
856 for (i = 0; i <= NETNSA_MAX; i++) {
864 case NETNSA_TARGET_NSID:
867 NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
875 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
876 struct netlink_ext_ack *extack)
878 struct net *net = sock_net(skb->sk);
879 struct nlattr *tb[NETNSA_MAX + 1];
880 struct net_fill_args fillargs = {
881 .portid = NETLINK_CB(skb).portid,
882 .seq = nlh->nlmsg_seq,
885 struct net *peer, *target = net;
890 err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
893 if (tb[NETNSA_PID]) {
894 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
895 nla = tb[NETNSA_PID];
896 } else if (tb[NETNSA_FD]) {
897 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
899 } else if (tb[NETNSA_NSID]) {
900 peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
902 peer = ERR_PTR(-ENOENT);
903 nla = tb[NETNSA_NSID];
905 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
910 NL_SET_BAD_ATTR(extack, nla);
911 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
912 return PTR_ERR(peer);
915 if (tb[NETNSA_TARGET_NSID]) {
916 int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
918 target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
919 if (IS_ERR(target)) {
920 NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
921 NL_SET_ERR_MSG(extack,
922 "Target netns reference is invalid");
923 err = PTR_ERR(target);
926 fillargs.add_ref = true;
927 fillargs.ref_nsid = peernet2id(net, peer);
930 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
936 fillargs.nsid = peernet2id(target, peer);
937 err = rtnl_net_fill(msg, &fillargs);
941 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
947 if (fillargs.add_ref)
953 struct rtnl_net_dump_cb {
957 struct net_fill_args fillargs;
962 /* Runs in RCU-critical section. */
963 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
965 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
968 if (net_cb->idx < net_cb->s_idx)
971 net_cb->fillargs.nsid = id;
972 if (net_cb->fillargs.add_ref)
973 net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
974 ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
983 static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
984 struct rtnl_net_dump_cb *net_cb,
985 struct netlink_callback *cb)
987 struct netlink_ext_ack *extack = cb->extack;
988 struct nlattr *tb[NETNSA_MAX + 1];
991 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
992 NETNSA_MAX, rtnl_net_policy,
997 for (i = 0; i <= NETNSA_MAX; i++) {
1001 if (i == NETNSA_TARGET_NSID) {
1004 net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
1006 NL_SET_BAD_ATTR(extack, tb[i]);
1007 NL_SET_ERR_MSG(extack,
1008 "Invalid target network namespace id");
1009 return PTR_ERR(net);
1011 net_cb->fillargs.add_ref = true;
1012 net_cb->ref_net = net_cb->tgt_net;
1013 net_cb->tgt_net = net;
1015 NL_SET_BAD_ATTR(extack, tb[i]);
1016 NL_SET_ERR_MSG(extack,
1017 "Unsupported attribute in dump request");
1025 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1027 struct rtnl_net_dump_cb net_cb = {
1028 .tgt_net = sock_net(skb->sk),
1031 .portid = NETLINK_CB(cb->skb).portid,
1032 .seq = cb->nlh->nlmsg_seq,
1033 .flags = NLM_F_MULTI,
1037 .s_idx = cb->args[0],
1041 if (cb->strict_check) {
1042 err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1048 idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
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 static void free_exit_list(struct pernet_operations *ops, struct list_head *net_exit_list)
1130 ops_pre_exit_list(ops, net_exit_list);
1132 ops_exit_list(ops, net_exit_list);
1133 ops_free_list(ops, net_exit_list);
1136 #ifdef CONFIG_NET_NS
1137 static int __register_pernet_operations(struct list_head *list,
1138 struct pernet_operations *ops)
1142 LIST_HEAD(net_exit_list);
1144 list_add_tail(&ops->list, list);
1145 if (ops->init || (ops->id && ops->size)) {
1146 /* We held write locked pernet_ops_rwsem, and parallel
1147 * setup_net() and cleanup_net() are not possible.
1150 error = ops_init(ops, net);
1153 list_add_tail(&net->exit_list, &net_exit_list);
1159 /* If I have an error cleanup all namespaces I initialized */
1160 list_del(&ops->list);
1161 free_exit_list(ops, &net_exit_list);
1165 static void __unregister_pernet_operations(struct pernet_operations *ops)
1168 LIST_HEAD(net_exit_list);
1170 list_del(&ops->list);
1171 /* See comment in __register_pernet_operations() */
1173 list_add_tail(&net->exit_list, &net_exit_list);
1175 free_exit_list(ops, &net_exit_list);
1180 static int __register_pernet_operations(struct list_head *list,
1181 struct pernet_operations *ops)
1183 if (!init_net_initialized) {
1184 list_add_tail(&ops->list, list);
1188 return ops_init(ops, &init_net);
1191 static void __unregister_pernet_operations(struct pernet_operations *ops)
1193 if (!init_net_initialized) {
1194 list_del(&ops->list);
1196 LIST_HEAD(net_exit_list);
1197 list_add(&init_net.exit_list, &net_exit_list);
1198 free_exit_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 nsset *nsset, struct ns_common *ns)
1360 struct nsproxy *nsproxy = nsset->nsproxy;
1361 struct net *net = to_net_ns(ns);
1363 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1364 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
1367 put_net(nsproxy->net_ns);
1368 nsproxy->net_ns = get_net(net);
1372 static struct user_namespace *netns_owner(struct ns_common *ns)
1374 return to_net_ns(ns)->user_ns;
1377 const struct proc_ns_operations netns_operations = {
1379 .type = CLONE_NEWNET,
1382 .install = netns_install,
1383 .owner = netns_owner,