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 atomic64_t cookie_gen;
74 u64 net_gen_cookie(struct net *net)
77 u64 res = atomic64_read(&net->net_cookie);
81 res = atomic64_inc_return(&cookie_gen);
82 atomic64_cmpxchg(&net->net_cookie, 0, res);
86 static struct net_generic *net_alloc_generic(void)
88 struct net_generic *ng;
89 unsigned int generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
91 ng = kzalloc(generic_size, GFP_KERNEL);
93 ng->s.len = max_gen_ptrs;
98 static int net_assign_generic(struct net *net, unsigned int id, void *data)
100 struct net_generic *ng, *old_ng;
102 BUG_ON(id < MIN_PERNET_OPS_ID);
104 old_ng = rcu_dereference_protected(net->gen,
105 lockdep_is_held(&pernet_ops_rwsem));
106 if (old_ng->s.len > id) {
107 old_ng->ptr[id] = data;
111 ng = net_alloc_generic();
116 * Some synchronisation notes:
118 * The net_generic explores the net->gen array inside rcu
119 * read section. Besides once set the net->gen->ptr[x]
120 * pointer never changes (see rules in netns/generic.h).
122 * That said, we simply duplicate this array and schedule
123 * the old copy for kfree after a grace period.
126 memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
127 (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
130 rcu_assign_pointer(net->gen, ng);
131 kfree_rcu(old_ng, s.rcu);
135 static int ops_init(const struct pernet_operations *ops, struct net *net)
140 if (ops->id && ops->size) {
141 data = kzalloc(ops->size, GFP_KERNEL);
145 err = net_assign_generic(net, *ops->id, data);
151 err = ops->init(net);
162 static void ops_free(const struct pernet_operations *ops, struct net *net)
164 if (ops->id && ops->size) {
165 kfree(net_generic(net, *ops->id));
169 static void ops_pre_exit_list(const struct pernet_operations *ops,
170 struct list_head *net_exit_list)
175 list_for_each_entry(net, net_exit_list, exit_list)
180 static void ops_exit_list(const struct pernet_operations *ops,
181 struct list_head *net_exit_list)
185 list_for_each_entry(net, net_exit_list, exit_list)
189 ops->exit_batch(net_exit_list);
192 static void ops_free_list(const struct pernet_operations *ops,
193 struct list_head *net_exit_list)
196 if (ops->size && ops->id) {
197 list_for_each_entry(net, net_exit_list, exit_list)
202 /* should be called with nsid_lock held */
203 static int alloc_netid(struct net *net, struct net *peer, int reqid)
205 int min = 0, max = 0;
212 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
215 /* This function is used by idr_for_each(). If net is equal to peer, the
216 * function returns the id so that idr_for_each() stops. Because we cannot
217 * returns the id 0 (idr_for_each() will not stop), we return the magic value
218 * NET_ID_ZERO (-1) for it.
220 #define NET_ID_ZERO -1
221 static int net_eq_idr(int id, void *net, void *peer)
223 if (net_eq(net, peer))
224 return id ? : NET_ID_ZERO;
228 /* Must be called from RCU-critical section or with nsid_lock held */
229 static int __peernet2id(const struct net *net, struct net *peer)
231 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
233 /* Magic value for id 0. */
234 if (id == NET_ID_ZERO)
239 return NETNSA_NSID_NOT_ASSIGNED;
242 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
243 struct nlmsghdr *nlh, gfp_t gfp);
244 /* This function returns the id of a peer netns. If no id is assigned, one will
245 * be allocated and returned.
247 int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp)
251 if (refcount_read(&net->count) == 0)
252 return NETNSA_NSID_NOT_ASSIGNED;
254 spin_lock(&net->nsid_lock);
255 id = __peernet2id(net, peer);
257 spin_unlock(&net->nsid_lock);
261 /* When peer is obtained from RCU lists, we may race with
262 * its cleanup. Check whether it's alive, and this guarantees
263 * we never hash a peer back to net->netns_ids, after it has
264 * just been idr_remove()'d from there in cleanup_net().
266 if (!maybe_get_net(peer)) {
267 spin_unlock(&net->nsid_lock);
268 return NETNSA_NSID_NOT_ASSIGNED;
271 id = alloc_netid(net, peer, -1);
272 spin_unlock(&net->nsid_lock);
276 return NETNSA_NSID_NOT_ASSIGNED;
278 rtnl_net_notifyid(net, RTM_NEWNSID, id, 0, NULL, gfp);
282 EXPORT_SYMBOL_GPL(peernet2id_alloc);
284 /* This function returns, if assigned, the id of a peer netns. */
285 int peernet2id(const struct net *net, struct net *peer)
290 id = __peernet2id(net, peer);
295 EXPORT_SYMBOL(peernet2id);
297 /* This function returns true is the peer netns has an id assigned into the
300 bool peernet_has_id(const struct net *net, struct net *peer)
302 return peernet2id(net, peer) >= 0;
305 struct net *get_net_ns_by_id(const struct net *net, int id)
313 peer = idr_find(&net->netns_ids, id);
315 peer = maybe_get_net(peer);
322 * setup_net runs the initializers for the network namespace object.
324 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
326 /* Must be called with pernet_ops_rwsem held */
327 const struct pernet_operations *ops, *saved_ops;
329 LIST_HEAD(net_exit_list);
331 refcount_set(&net->count, 1);
332 refcount_set(&net->passive, 1);
333 get_random_bytes(&net->hash_mix, sizeof(u32));
334 net->dev_base_seq = 1;
335 net->user_ns = user_ns;
336 idr_init(&net->netns_ids);
337 spin_lock_init(&net->nsid_lock);
338 mutex_init(&net->ipv4.ra_mutex);
340 list_for_each_entry(ops, &pernet_list, list) {
341 error = ops_init(ops, net);
345 down_write(&net_rwsem);
346 list_add_tail_rcu(&net->list, &net_namespace_list);
347 up_write(&net_rwsem);
352 /* Walk through the list backwards calling the exit functions
353 * for the pernet modules whose init functions did not fail.
355 list_add(&net->exit_list, &net_exit_list);
357 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
358 ops_pre_exit_list(ops, &net_exit_list);
363 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
364 ops_exit_list(ops, &net_exit_list);
367 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
368 ops_free_list(ops, &net_exit_list);
374 static int __net_init net_defaults_init_net(struct net *net)
376 net->core.sysctl_somaxconn = SOMAXCONN;
380 static struct pernet_operations net_defaults_ops = {
381 .init = net_defaults_init_net,
384 static __init int net_defaults_init(void)
386 if (register_pernet_subsys(&net_defaults_ops))
387 panic("Cannot initialize net default settings");
392 core_initcall(net_defaults_init);
395 static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
397 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
400 static void dec_net_namespaces(struct ucounts *ucounts)
402 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
405 static struct kmem_cache *net_cachep __ro_after_init;
406 static struct workqueue_struct *netns_wq;
408 static struct net *net_alloc(void)
410 struct net *net = NULL;
411 struct net_generic *ng;
413 ng = net_alloc_generic();
417 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
422 net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
423 if (!net->key_domain)
425 refcount_set(&net->key_domain->usage, 1);
428 rcu_assign_pointer(net->gen, ng);
434 kmem_cache_free(net_cachep, net);
442 static void net_free(struct net *net)
444 kfree(rcu_access_pointer(net->gen));
445 kmem_cache_free(net_cachep, net);
448 void net_drop_ns(void *p)
451 if (ns && refcount_dec_and_test(&ns->passive))
455 struct net *copy_net_ns(unsigned long flags,
456 struct user_namespace *user_ns, struct net *old_net)
458 struct ucounts *ucounts;
462 if (!(flags & CLONE_NEWNET))
463 return get_net(old_net);
465 ucounts = inc_net_namespaces(user_ns);
467 return ERR_PTR(-ENOSPC);
474 refcount_set(&net->passive, 1);
475 net->ucounts = ucounts;
476 get_user_ns(user_ns);
478 rv = down_read_killable(&pernet_ops_rwsem);
482 rv = setup_net(net, user_ns);
484 up_read(&pernet_ops_rwsem);
488 key_remove_domain(net->key_domain);
489 put_user_ns(user_ns);
492 dec_net_namespaces(ucounts);
499 * net_ns_get_ownership - get sysfs ownership data for @net
500 * @net: network namespace in question (can be NULL)
501 * @uid: kernel user ID for sysfs objects
502 * @gid: kernel group ID for sysfs objects
504 * Returns the uid/gid pair of root in the user namespace associated with the
505 * given network namespace.
507 void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
510 kuid_t ns_root_uid = make_kuid(net->user_ns, 0);
511 kgid_t ns_root_gid = make_kgid(net->user_ns, 0);
513 if (uid_valid(ns_root_uid))
516 if (gid_valid(ns_root_gid))
519 *uid = GLOBAL_ROOT_UID;
520 *gid = GLOBAL_ROOT_GID;
523 EXPORT_SYMBOL_GPL(net_ns_get_ownership);
525 static void unhash_nsid(struct net *net, struct net *last)
528 /* This function is only called from cleanup_net() work,
529 * and this work is the only process, that may delete
530 * a net from net_namespace_list. So, when the below
531 * is executing, the list may only grow. Thus, we do not
532 * use for_each_net_rcu() or net_rwsem.
537 spin_lock(&tmp->nsid_lock);
538 id = __peernet2id(tmp, net);
540 idr_remove(&tmp->netns_ids, id);
541 spin_unlock(&tmp->nsid_lock);
543 rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL,
548 spin_lock(&net->nsid_lock);
549 idr_destroy(&net->netns_ids);
550 spin_unlock(&net->nsid_lock);
553 static LLIST_HEAD(cleanup_list);
555 static void cleanup_net(struct work_struct *work)
557 const struct pernet_operations *ops;
558 struct net *net, *tmp, *last;
559 struct llist_node *net_kill_list;
560 LIST_HEAD(net_exit_list);
562 /* Atomically snapshot the list of namespaces to cleanup */
563 net_kill_list = llist_del_all(&cleanup_list);
565 down_read(&pernet_ops_rwsem);
567 /* Don't let anyone else find us. */
568 down_write(&net_rwsem);
569 llist_for_each_entry(net, net_kill_list, cleanup_list)
570 list_del_rcu(&net->list);
571 /* Cache last net. After we unlock rtnl, no one new net
572 * added to net_namespace_list can assign nsid pointer
573 * to a net from net_kill_list (see peernet2id_alloc()).
574 * So, we skip them in unhash_nsid().
576 * Note, that unhash_nsid() does not delete nsid links
577 * between net_kill_list's nets, as they've already
578 * deleted from net_namespace_list. But, this would be
579 * useless anyway, as netns_ids are destroyed there.
581 last = list_last_entry(&net_namespace_list, struct net, list);
582 up_write(&net_rwsem);
584 llist_for_each_entry(net, net_kill_list, cleanup_list) {
585 unhash_nsid(net, last);
586 list_add_tail(&net->exit_list, &net_exit_list);
589 /* Run all of the network namespace pre_exit methods */
590 list_for_each_entry_reverse(ops, &pernet_list, list)
591 ops_pre_exit_list(ops, &net_exit_list);
594 * Another CPU might be rcu-iterating the list, wait for it.
595 * This needs to be before calling the exit() notifiers, so
596 * the rcu_barrier() below isn't sufficient alone.
597 * Also the pre_exit() and exit() methods need this barrier.
601 /* Run all of the network namespace exit methods */
602 list_for_each_entry_reverse(ops, &pernet_list, list)
603 ops_exit_list(ops, &net_exit_list);
605 /* Free the net generic variables */
606 list_for_each_entry_reverse(ops, &pernet_list, list)
607 ops_free_list(ops, &net_exit_list);
609 up_read(&pernet_ops_rwsem);
611 /* Ensure there are no outstanding rcu callbacks using this
616 /* Finally it is safe to free my network namespace structure */
617 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
618 list_del_init(&net->exit_list);
619 dec_net_namespaces(net->ucounts);
620 key_remove_domain(net->key_domain);
621 put_user_ns(net->user_ns);
627 * net_ns_barrier - wait until concurrent net_cleanup_work is done
629 * cleanup_net runs from work queue and will first remove namespaces
630 * from the global list, then run net exit functions.
632 * Call this in module exit path to make sure that all netns
633 * ->exit ops have been invoked before the function is removed.
635 void net_ns_barrier(void)
637 down_write(&pernet_ops_rwsem);
638 up_write(&pernet_ops_rwsem);
640 EXPORT_SYMBOL(net_ns_barrier);
642 static DECLARE_WORK(net_cleanup_work, cleanup_net);
644 void __put_net(struct net *net)
646 /* Cleanup the network namespace in process context */
647 if (llist_add(&net->cleanup_list, &cleanup_list))
648 queue_work(netns_wq, &net_cleanup_work);
650 EXPORT_SYMBOL_GPL(__put_net);
652 struct net *get_net_ns_by_fd(int fd)
655 struct ns_common *ns;
658 file = proc_ns_fget(fd);
660 return ERR_CAST(file);
662 ns = get_proc_ns(file_inode(file));
663 if (ns->ops == &netns_operations)
664 net = get_net(container_of(ns, struct net, ns));
666 net = ERR_PTR(-EINVAL);
673 struct net *get_net_ns_by_fd(int fd)
675 return ERR_PTR(-EINVAL);
678 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
680 struct net *get_net_ns_by_pid(pid_t pid)
682 struct task_struct *tsk;
685 /* Lookup the network namespace */
686 net = ERR_PTR(-ESRCH);
688 tsk = find_task_by_vpid(pid);
690 struct nsproxy *nsproxy;
692 nsproxy = tsk->nsproxy;
694 net = get_net(nsproxy->net_ns);
700 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
702 static __net_init int net_ns_net_init(struct net *net)
705 net->ns.ops = &netns_operations;
707 return ns_alloc_inum(&net->ns);
710 static __net_exit void net_ns_net_exit(struct net *net)
712 ns_free_inum(&net->ns);
715 static struct pernet_operations __net_initdata net_ns_ops = {
716 .init = net_ns_net_init,
717 .exit = net_ns_net_exit,
720 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
721 [NETNSA_NONE] = { .type = NLA_UNSPEC },
722 [NETNSA_NSID] = { .type = NLA_S32 },
723 [NETNSA_PID] = { .type = NLA_U32 },
724 [NETNSA_FD] = { .type = NLA_U32 },
725 [NETNSA_TARGET_NSID] = { .type = NLA_S32 },
728 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
729 struct netlink_ext_ack *extack)
731 struct net *net = sock_net(skb->sk);
732 struct nlattr *tb[NETNSA_MAX + 1];
737 err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
738 NETNSA_MAX, rtnl_net_policy, extack);
741 if (!tb[NETNSA_NSID]) {
742 NL_SET_ERR_MSG(extack, "nsid is missing");
745 nsid = nla_get_s32(tb[NETNSA_NSID]);
747 if (tb[NETNSA_PID]) {
748 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
749 nla = tb[NETNSA_PID];
750 } else if (tb[NETNSA_FD]) {
751 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
754 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
758 NL_SET_BAD_ATTR(extack, nla);
759 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
760 return PTR_ERR(peer);
763 spin_lock(&net->nsid_lock);
764 if (__peernet2id(net, peer) >= 0) {
765 spin_unlock(&net->nsid_lock);
767 NL_SET_BAD_ATTR(extack, nla);
768 NL_SET_ERR_MSG(extack,
769 "Peer netns already has a nsid assigned");
773 err = alloc_netid(net, peer, nsid);
774 spin_unlock(&net->nsid_lock);
776 rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid,
779 } else if (err == -ENOSPC && nsid >= 0) {
781 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
782 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
789 static int rtnl_net_get_size(void)
791 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
792 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
793 + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
797 struct net_fill_args {
807 static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
809 struct nlmsghdr *nlh;
810 struct rtgenmsg *rth;
812 nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
817 rth = nlmsg_data(nlh);
818 rth->rtgen_family = AF_UNSPEC;
820 if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
821 goto nla_put_failure;
824 nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
825 goto nla_put_failure;
831 nlmsg_cancel(skb, nlh);
835 static int rtnl_net_valid_getid_req(struct sk_buff *skb,
836 const struct nlmsghdr *nlh,
838 struct netlink_ext_ack *extack)
842 if (!netlink_strict_get_check(skb))
843 return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
844 tb, NETNSA_MAX, rtnl_net_policy,
847 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
848 NETNSA_MAX, rtnl_net_policy,
853 for (i = 0; i <= NETNSA_MAX; i++) {
861 case NETNSA_TARGET_NSID:
864 NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
872 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
873 struct netlink_ext_ack *extack)
875 struct net *net = sock_net(skb->sk);
876 struct nlattr *tb[NETNSA_MAX + 1];
877 struct net_fill_args fillargs = {
878 .portid = NETLINK_CB(skb).portid,
879 .seq = nlh->nlmsg_seq,
882 struct net *peer, *target = net;
887 err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
890 if (tb[NETNSA_PID]) {
891 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
892 nla = tb[NETNSA_PID];
893 } else if (tb[NETNSA_FD]) {
894 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
896 } else if (tb[NETNSA_NSID]) {
897 peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
899 peer = ERR_PTR(-ENOENT);
900 nla = tb[NETNSA_NSID];
902 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
907 NL_SET_BAD_ATTR(extack, nla);
908 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
909 return PTR_ERR(peer);
912 if (tb[NETNSA_TARGET_NSID]) {
913 int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
915 target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
916 if (IS_ERR(target)) {
917 NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
918 NL_SET_ERR_MSG(extack,
919 "Target netns reference is invalid");
920 err = PTR_ERR(target);
923 fillargs.add_ref = true;
924 fillargs.ref_nsid = peernet2id(net, peer);
927 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
933 fillargs.nsid = peernet2id(target, peer);
934 err = rtnl_net_fill(msg, &fillargs);
938 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
944 if (fillargs.add_ref)
950 struct rtnl_net_dump_cb {
954 struct net_fill_args fillargs;
959 /* Runs in RCU-critical section. */
960 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
962 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
965 if (net_cb->idx < net_cb->s_idx)
968 net_cb->fillargs.nsid = id;
969 if (net_cb->fillargs.add_ref)
970 net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
971 ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
980 static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
981 struct rtnl_net_dump_cb *net_cb,
982 struct netlink_callback *cb)
984 struct netlink_ext_ack *extack = cb->extack;
985 struct nlattr *tb[NETNSA_MAX + 1];
988 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
989 NETNSA_MAX, rtnl_net_policy,
994 for (i = 0; i <= NETNSA_MAX; i++) {
998 if (i == NETNSA_TARGET_NSID) {
1001 net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
1003 NL_SET_BAD_ATTR(extack, tb[i]);
1004 NL_SET_ERR_MSG(extack,
1005 "Invalid target network namespace id");
1006 return PTR_ERR(net);
1008 net_cb->fillargs.add_ref = true;
1009 net_cb->ref_net = net_cb->tgt_net;
1010 net_cb->tgt_net = net;
1012 NL_SET_BAD_ATTR(extack, tb[i]);
1013 NL_SET_ERR_MSG(extack,
1014 "Unsupported attribute in dump request");
1022 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1024 struct rtnl_net_dump_cb net_cb = {
1025 .tgt_net = sock_net(skb->sk),
1028 .portid = NETLINK_CB(cb->skb).portid,
1029 .seq = cb->nlh->nlmsg_seq,
1030 .flags = NLM_F_MULTI,
1034 .s_idx = cb->args[0],
1038 if (cb->strict_check) {
1039 err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1045 idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1048 cb->args[0] = net_cb.idx;
1050 if (net_cb.fillargs.add_ref)
1051 put_net(net_cb.tgt_net);
1052 return err < 0 ? err : skb->len;
1055 static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1056 struct nlmsghdr *nlh, gfp_t gfp)
1058 struct net_fill_args fillargs = {
1060 .seq = nlh ? nlh->nlmsg_seq : 0,
1064 struct sk_buff *msg;
1067 msg = nlmsg_new(rtnl_net_get_size(), gfp);
1071 err = rtnl_net_fill(msg, &fillargs);
1075 rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, gfp);
1081 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1084 static int __init net_ns_init(void)
1086 struct net_generic *ng;
1088 #ifdef CONFIG_NET_NS
1089 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1091 SLAB_PANIC|SLAB_ACCOUNT, NULL);
1093 /* Create workqueue for cleanup */
1094 netns_wq = create_singlethread_workqueue("netns");
1096 panic("Could not create netns workq");
1099 ng = net_alloc_generic();
1101 panic("Could not allocate generic netns");
1103 rcu_assign_pointer(init_net.gen, ng);
1104 net_gen_cookie(&init_net);
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 nsset *nsset, struct ns_common *ns)
1358 struct nsproxy *nsproxy = nsset->nsproxy;
1359 struct net *net = to_net_ns(ns);
1361 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1362 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
1365 put_net(nsproxy->net_ns);
1366 nsproxy->net_ns = get_net(net);
1370 static struct user_namespace *netns_owner(struct ns_common *ns)
1372 return to_net_ns(ns)->user_ns;
1375 const struct proc_ns_operations netns_operations = {
1377 .type = CLONE_NEWNET,
1380 .install = netns_install,
1381 .owner = netns_owner,