iwlwifi: split the regulatory rules when the bandwidth flags require it
[linux-2.6/btrfs-unstable.git] / net / core / net_namespace.c
blob8726d051f31dd2ddce96a8cd7b1ccf50435f0d92
1 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
3 #include <linux/workqueue.h>
4 #include <linux/rtnetlink.h>
5 #include <linux/cache.h>
6 #include <linux/slab.h>
7 #include <linux/list.h>
8 #include <linux/delay.h>
9 #include <linux/sched.h>
10 #include <linux/idr.h>
11 #include <linux/rculist.h>
12 #include <linux/nsproxy.h>
13 #include <linux/fs.h>
14 #include <linux/proc_ns.h>
15 #include <linux/file.h>
16 #include <linux/export.h>
17 #include <linux/user_namespace.h>
18 #include <linux/net_namespace.h>
19 #include <linux/sched/task.h>
21 #include <net/sock.h>
22 #include <net/netlink.h>
23 #include <net/net_namespace.h>
24 #include <net/netns/generic.h>
27 * Our network namespace constructor/destructor lists
30 static LIST_HEAD(pernet_list);
31 static struct list_head *first_device = &pernet_list;
32 DEFINE_MUTEX(net_mutex);
34 LIST_HEAD(net_namespace_list);
35 EXPORT_SYMBOL_GPL(net_namespace_list);
37 struct net init_net = {
38 .count = ATOMIC_INIT(1),
39 .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head),
41 EXPORT_SYMBOL(init_net);
43 static bool init_net_initialized;
45 #define MIN_PERNET_OPS_ID \
46 ((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *))
48 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
50 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
52 static struct net_generic *net_alloc_generic(void)
54 struct net_generic *ng;
55 unsigned int generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
57 ng = kzalloc(generic_size, GFP_KERNEL);
58 if (ng)
59 ng->s.len = max_gen_ptrs;
61 return ng;
64 static int net_assign_generic(struct net *net, unsigned int id, void *data)
66 struct net_generic *ng, *old_ng;
68 BUG_ON(!mutex_is_locked(&net_mutex));
69 BUG_ON(id < MIN_PERNET_OPS_ID);
71 old_ng = rcu_dereference_protected(net->gen,
72 lockdep_is_held(&net_mutex));
73 if (old_ng->s.len > id) {
74 old_ng->ptr[id] = data;
75 return 0;
78 ng = net_alloc_generic();
79 if (ng == NULL)
80 return -ENOMEM;
83 * Some synchronisation notes:
85 * The net_generic explores the net->gen array inside rcu
86 * read section. Besides once set the net->gen->ptr[x]
87 * pointer never changes (see rules in netns/generic.h).
89 * That said, we simply duplicate this array and schedule
90 * the old copy for kfree after a grace period.
93 memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
94 (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
95 ng->ptr[id] = data;
97 rcu_assign_pointer(net->gen, ng);
98 kfree_rcu(old_ng, s.rcu);
99 return 0;
102 static int ops_init(const struct pernet_operations *ops, struct net *net)
104 int err = -ENOMEM;
105 void *data = NULL;
107 if (ops->id && ops->size) {
108 data = kzalloc(ops->size, GFP_KERNEL);
109 if (!data)
110 goto out;
112 err = net_assign_generic(net, *ops->id, data);
113 if (err)
114 goto cleanup;
116 err = 0;
117 if (ops->init)
118 err = ops->init(net);
119 if (!err)
120 return 0;
122 cleanup:
123 kfree(data);
125 out:
126 return err;
129 static void ops_free(const struct pernet_operations *ops, struct net *net)
131 if (ops->id && ops->size) {
132 kfree(net_generic(net, *ops->id));
136 static void ops_exit_list(const struct pernet_operations *ops,
137 struct list_head *net_exit_list)
139 struct net *net;
140 if (ops->exit) {
141 list_for_each_entry(net, net_exit_list, exit_list)
142 ops->exit(net);
144 if (ops->exit_batch)
145 ops->exit_batch(net_exit_list);
148 static void ops_free_list(const struct pernet_operations *ops,
149 struct list_head *net_exit_list)
151 struct net *net;
152 if (ops->size && ops->id) {
153 list_for_each_entry(net, net_exit_list, exit_list)
154 ops_free(ops, net);
158 /* should be called with nsid_lock held */
159 static int alloc_netid(struct net *net, struct net *peer, int reqid)
161 int min = 0, max = 0;
163 if (reqid >= 0) {
164 min = reqid;
165 max = reqid + 1;
168 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
171 /* This function is used by idr_for_each(). If net is equal to peer, the
172 * function returns the id so that idr_for_each() stops. Because we cannot
173 * returns the id 0 (idr_for_each() will not stop), we return the magic value
174 * NET_ID_ZERO (-1) for it.
176 #define NET_ID_ZERO -1
177 static int net_eq_idr(int id, void *net, void *peer)
179 if (net_eq(net, peer))
180 return id ? : NET_ID_ZERO;
181 return 0;
184 /* Should be called with nsid_lock held. If a new id is assigned, the bool alloc
185 * is set to true, thus the caller knows that the new id must be notified via
186 * rtnl.
188 static int __peernet2id_alloc(struct net *net, struct net *peer, bool *alloc)
190 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
191 bool alloc_it = *alloc;
193 *alloc = false;
195 /* Magic value for id 0. */
196 if (id == NET_ID_ZERO)
197 return 0;
198 if (id > 0)
199 return id;
201 if (alloc_it) {
202 id = alloc_netid(net, peer, -1);
203 *alloc = true;
204 return id >= 0 ? id : NETNSA_NSID_NOT_ASSIGNED;
207 return NETNSA_NSID_NOT_ASSIGNED;
210 /* should be called with nsid_lock held */
211 static int __peernet2id(struct net *net, struct net *peer)
213 bool no = false;
215 return __peernet2id_alloc(net, peer, &no);
218 static void rtnl_net_notifyid(struct net *net, int cmd, int id);
219 /* This function returns the id of a peer netns. If no id is assigned, one will
220 * be allocated and returned.
222 int peernet2id_alloc(struct net *net, struct net *peer)
224 bool alloc;
225 int id;
227 if (atomic_read(&net->count) == 0)
228 return NETNSA_NSID_NOT_ASSIGNED;
229 spin_lock_bh(&net->nsid_lock);
230 alloc = atomic_read(&peer->count) == 0 ? false : true;
231 id = __peernet2id_alloc(net, peer, &alloc);
232 spin_unlock_bh(&net->nsid_lock);
233 if (alloc && id >= 0)
234 rtnl_net_notifyid(net, RTM_NEWNSID, id);
235 return id;
238 /* This function returns, if assigned, the id of a peer netns. */
239 int peernet2id(struct net *net, struct net *peer)
241 int id;
243 spin_lock_bh(&net->nsid_lock);
244 id = __peernet2id(net, peer);
245 spin_unlock_bh(&net->nsid_lock);
246 return id;
248 EXPORT_SYMBOL(peernet2id);
250 /* This function returns true is the peer netns has an id assigned into the
251 * current netns.
253 bool peernet_has_id(struct net *net, struct net *peer)
255 return peernet2id(net, peer) >= 0;
258 struct net *get_net_ns_by_id(struct net *net, int id)
260 struct net *peer;
262 if (id < 0)
263 return NULL;
265 rcu_read_lock();
266 spin_lock_bh(&net->nsid_lock);
267 peer = idr_find(&net->netns_ids, id);
268 if (peer)
269 get_net(peer);
270 spin_unlock_bh(&net->nsid_lock);
271 rcu_read_unlock();
273 return peer;
277 * setup_net runs the initializers for the network namespace object.
279 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
281 /* Must be called with net_mutex held */
282 const struct pernet_operations *ops, *saved_ops;
283 int error = 0;
284 LIST_HEAD(net_exit_list);
286 atomic_set(&net->count, 1);
287 refcount_set(&net->passive, 1);
288 net->dev_base_seq = 1;
289 net->user_ns = user_ns;
290 idr_init(&net->netns_ids);
291 spin_lock_init(&net->nsid_lock);
293 list_for_each_entry(ops, &pernet_list, list) {
294 error = ops_init(ops, net);
295 if (error < 0)
296 goto out_undo;
298 out:
299 return error;
301 out_undo:
302 /* Walk through the list backwards calling the exit functions
303 * for the pernet modules whose init functions did not fail.
305 list_add(&net->exit_list, &net_exit_list);
306 saved_ops = ops;
307 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
308 ops_exit_list(ops, &net_exit_list);
310 ops = saved_ops;
311 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
312 ops_free_list(ops, &net_exit_list);
314 rcu_barrier();
315 goto out;
318 static int __net_init net_defaults_init_net(struct net *net)
320 net->core.sysctl_somaxconn = SOMAXCONN;
321 return 0;
324 static struct pernet_operations net_defaults_ops = {
325 .init = net_defaults_init_net,
328 static __init int net_defaults_init(void)
330 if (register_pernet_subsys(&net_defaults_ops))
331 panic("Cannot initialize net default settings");
333 return 0;
336 core_initcall(net_defaults_init);
338 #ifdef CONFIG_NET_NS
339 static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
341 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
344 static void dec_net_namespaces(struct ucounts *ucounts)
346 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
349 static struct kmem_cache *net_cachep;
350 static struct workqueue_struct *netns_wq;
352 static struct net *net_alloc(void)
354 struct net *net = NULL;
355 struct net_generic *ng;
357 ng = net_alloc_generic();
358 if (!ng)
359 goto out;
361 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
362 if (!net)
363 goto out_free;
365 rcu_assign_pointer(net->gen, ng);
366 out:
367 return net;
369 out_free:
370 kfree(ng);
371 goto out;
374 static void net_free(struct net *net)
376 kfree(rcu_access_pointer(net->gen));
377 kmem_cache_free(net_cachep, net);
380 void net_drop_ns(void *p)
382 struct net *ns = p;
383 if (ns && refcount_dec_and_test(&ns->passive))
384 net_free(ns);
387 struct net *copy_net_ns(unsigned long flags,
388 struct user_namespace *user_ns, struct net *old_net)
390 struct ucounts *ucounts;
391 struct net *net;
392 int rv;
394 if (!(flags & CLONE_NEWNET))
395 return get_net(old_net);
397 ucounts = inc_net_namespaces(user_ns);
398 if (!ucounts)
399 return ERR_PTR(-ENOSPC);
401 net = net_alloc();
402 if (!net) {
403 dec_net_namespaces(ucounts);
404 return ERR_PTR(-ENOMEM);
407 get_user_ns(user_ns);
409 rv = mutex_lock_killable(&net_mutex);
410 if (rv < 0) {
411 net_free(net);
412 dec_net_namespaces(ucounts);
413 put_user_ns(user_ns);
414 return ERR_PTR(rv);
417 net->ucounts = ucounts;
418 rv = setup_net(net, user_ns);
419 if (rv == 0) {
420 rtnl_lock();
421 list_add_tail_rcu(&net->list, &net_namespace_list);
422 rtnl_unlock();
424 mutex_unlock(&net_mutex);
425 if (rv < 0) {
426 dec_net_namespaces(ucounts);
427 put_user_ns(user_ns);
428 net_drop_ns(net);
429 return ERR_PTR(rv);
431 return net;
434 static DEFINE_SPINLOCK(cleanup_list_lock);
435 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
437 static void cleanup_net(struct work_struct *work)
439 const struct pernet_operations *ops;
440 struct net *net, *tmp;
441 struct list_head net_kill_list;
442 LIST_HEAD(net_exit_list);
444 /* Atomically snapshot the list of namespaces to cleanup */
445 spin_lock_irq(&cleanup_list_lock);
446 list_replace_init(&cleanup_list, &net_kill_list);
447 spin_unlock_irq(&cleanup_list_lock);
449 mutex_lock(&net_mutex);
451 /* Don't let anyone else find us. */
452 rtnl_lock();
453 list_for_each_entry(net, &net_kill_list, cleanup_list) {
454 list_del_rcu(&net->list);
455 list_add_tail(&net->exit_list, &net_exit_list);
456 for_each_net(tmp) {
457 int id;
459 spin_lock_bh(&tmp->nsid_lock);
460 id = __peernet2id(tmp, net);
461 if (id >= 0)
462 idr_remove(&tmp->netns_ids, id);
463 spin_unlock_bh(&tmp->nsid_lock);
464 if (id >= 0)
465 rtnl_net_notifyid(tmp, RTM_DELNSID, id);
467 spin_lock_bh(&net->nsid_lock);
468 idr_destroy(&net->netns_ids);
469 spin_unlock_bh(&net->nsid_lock);
472 rtnl_unlock();
475 * Another CPU might be rcu-iterating the list, wait for it.
476 * This needs to be before calling the exit() notifiers, so
477 * the rcu_barrier() below isn't sufficient alone.
479 synchronize_rcu();
481 /* Run all of the network namespace exit methods */
482 list_for_each_entry_reverse(ops, &pernet_list, list)
483 ops_exit_list(ops, &net_exit_list);
485 /* Free the net generic variables */
486 list_for_each_entry_reverse(ops, &pernet_list, list)
487 ops_free_list(ops, &net_exit_list);
489 mutex_unlock(&net_mutex);
491 /* Ensure there are no outstanding rcu callbacks using this
492 * network namespace.
494 rcu_barrier();
496 /* Finally it is safe to free my network namespace structure */
497 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
498 list_del_init(&net->exit_list);
499 dec_net_namespaces(net->ucounts);
500 put_user_ns(net->user_ns);
501 net_drop_ns(net);
506 * net_ns_barrier - wait until concurrent net_cleanup_work is done
508 * cleanup_net runs from work queue and will first remove namespaces
509 * from the global list, then run net exit functions.
511 * Call this in module exit path to make sure that all netns
512 * ->exit ops have been invoked before the function is removed.
514 void net_ns_barrier(void)
516 mutex_lock(&net_mutex);
517 mutex_unlock(&net_mutex);
519 EXPORT_SYMBOL(net_ns_barrier);
521 static DECLARE_WORK(net_cleanup_work, cleanup_net);
523 void __put_net(struct net *net)
525 /* Cleanup the network namespace in process context */
526 unsigned long flags;
528 spin_lock_irqsave(&cleanup_list_lock, flags);
529 list_add(&net->cleanup_list, &cleanup_list);
530 spin_unlock_irqrestore(&cleanup_list_lock, flags);
532 queue_work(netns_wq, &net_cleanup_work);
534 EXPORT_SYMBOL_GPL(__put_net);
536 struct net *get_net_ns_by_fd(int fd)
538 struct file *file;
539 struct ns_common *ns;
540 struct net *net;
542 file = proc_ns_fget(fd);
543 if (IS_ERR(file))
544 return ERR_CAST(file);
546 ns = get_proc_ns(file_inode(file));
547 if (ns->ops == &netns_operations)
548 net = get_net(container_of(ns, struct net, ns));
549 else
550 net = ERR_PTR(-EINVAL);
552 fput(file);
553 return net;
556 #else
557 struct net *get_net_ns_by_fd(int fd)
559 return ERR_PTR(-EINVAL);
561 #endif
562 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
564 struct net *get_net_ns_by_pid(pid_t pid)
566 struct task_struct *tsk;
567 struct net *net;
569 /* Lookup the network namespace */
570 net = ERR_PTR(-ESRCH);
571 rcu_read_lock();
572 tsk = find_task_by_vpid(pid);
573 if (tsk) {
574 struct nsproxy *nsproxy;
575 task_lock(tsk);
576 nsproxy = tsk->nsproxy;
577 if (nsproxy)
578 net = get_net(nsproxy->net_ns);
579 task_unlock(tsk);
581 rcu_read_unlock();
582 return net;
584 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
586 static __net_init int net_ns_net_init(struct net *net)
588 #ifdef CONFIG_NET_NS
589 net->ns.ops = &netns_operations;
590 #endif
591 return ns_alloc_inum(&net->ns);
594 static __net_exit void net_ns_net_exit(struct net *net)
596 ns_free_inum(&net->ns);
599 static struct pernet_operations __net_initdata net_ns_ops = {
600 .init = net_ns_net_init,
601 .exit = net_ns_net_exit,
604 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
605 [NETNSA_NONE] = { .type = NLA_UNSPEC },
606 [NETNSA_NSID] = { .type = NLA_S32 },
607 [NETNSA_PID] = { .type = NLA_U32 },
608 [NETNSA_FD] = { .type = NLA_U32 },
611 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
612 struct netlink_ext_ack *extack)
614 struct net *net = sock_net(skb->sk);
615 struct nlattr *tb[NETNSA_MAX + 1];
616 struct nlattr *nla;
617 struct net *peer;
618 int nsid, err;
620 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
621 rtnl_net_policy, extack);
622 if (err < 0)
623 return err;
624 if (!tb[NETNSA_NSID]) {
625 NL_SET_ERR_MSG(extack, "nsid is missing");
626 return -EINVAL;
628 nsid = nla_get_s32(tb[NETNSA_NSID]);
630 if (tb[NETNSA_PID]) {
631 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
632 nla = tb[NETNSA_PID];
633 } else if (tb[NETNSA_FD]) {
634 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
635 nla = tb[NETNSA_FD];
636 } else {
637 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
638 return -EINVAL;
640 if (IS_ERR(peer)) {
641 NL_SET_BAD_ATTR(extack, nla);
642 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
643 return PTR_ERR(peer);
646 spin_lock_bh(&net->nsid_lock);
647 if (__peernet2id(net, peer) >= 0) {
648 spin_unlock_bh(&net->nsid_lock);
649 err = -EEXIST;
650 NL_SET_BAD_ATTR(extack, nla);
651 NL_SET_ERR_MSG(extack,
652 "Peer netns already has a nsid assigned");
653 goto out;
656 err = alloc_netid(net, peer, nsid);
657 spin_unlock_bh(&net->nsid_lock);
658 if (err >= 0) {
659 rtnl_net_notifyid(net, RTM_NEWNSID, err);
660 err = 0;
661 } else if (err == -ENOSPC && nsid >= 0) {
662 err = -EEXIST;
663 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
664 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
666 out:
667 put_net(peer);
668 return err;
671 static int rtnl_net_get_size(void)
673 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
674 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
678 static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags,
679 int cmd, struct net *net, int nsid)
681 struct nlmsghdr *nlh;
682 struct rtgenmsg *rth;
684 nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags);
685 if (!nlh)
686 return -EMSGSIZE;
688 rth = nlmsg_data(nlh);
689 rth->rtgen_family = AF_UNSPEC;
691 if (nla_put_s32(skb, NETNSA_NSID, nsid))
692 goto nla_put_failure;
694 nlmsg_end(skb, nlh);
695 return 0;
697 nla_put_failure:
698 nlmsg_cancel(skb, nlh);
699 return -EMSGSIZE;
702 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
703 struct netlink_ext_ack *extack)
705 struct net *net = sock_net(skb->sk);
706 struct nlattr *tb[NETNSA_MAX + 1];
707 struct nlattr *nla;
708 struct sk_buff *msg;
709 struct net *peer;
710 int err, id;
712 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
713 rtnl_net_policy, extack);
714 if (err < 0)
715 return err;
716 if (tb[NETNSA_PID]) {
717 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
718 nla = tb[NETNSA_PID];
719 } else if (tb[NETNSA_FD]) {
720 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
721 nla = tb[NETNSA_FD];
722 } else {
723 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
724 return -EINVAL;
727 if (IS_ERR(peer)) {
728 NL_SET_BAD_ATTR(extack, nla);
729 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
730 return PTR_ERR(peer);
733 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
734 if (!msg) {
735 err = -ENOMEM;
736 goto out;
739 id = peernet2id(net, peer);
740 err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0,
741 RTM_NEWNSID, net, id);
742 if (err < 0)
743 goto err_out;
745 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
746 goto out;
748 err_out:
749 nlmsg_free(msg);
750 out:
751 put_net(peer);
752 return err;
755 struct rtnl_net_dump_cb {
756 struct net *net;
757 struct sk_buff *skb;
758 struct netlink_callback *cb;
759 int idx;
760 int s_idx;
763 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
765 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
766 int ret;
768 if (net_cb->idx < net_cb->s_idx)
769 goto cont;
771 ret = rtnl_net_fill(net_cb->skb, NETLINK_CB(net_cb->cb->skb).portid,
772 net_cb->cb->nlh->nlmsg_seq, NLM_F_MULTI,
773 RTM_NEWNSID, net_cb->net, id);
774 if (ret < 0)
775 return ret;
777 cont:
778 net_cb->idx++;
779 return 0;
782 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
784 struct net *net = sock_net(skb->sk);
785 struct rtnl_net_dump_cb net_cb = {
786 .net = net,
787 .skb = skb,
788 .cb = cb,
789 .idx = 0,
790 .s_idx = cb->args[0],
793 spin_lock_bh(&net->nsid_lock);
794 idr_for_each(&net->netns_ids, rtnl_net_dumpid_one, &net_cb);
795 spin_unlock_bh(&net->nsid_lock);
797 cb->args[0] = net_cb.idx;
798 return skb->len;
801 static void rtnl_net_notifyid(struct net *net, int cmd, int id)
803 struct sk_buff *msg;
804 int err = -ENOMEM;
806 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
807 if (!msg)
808 goto out;
810 err = rtnl_net_fill(msg, 0, 0, 0, cmd, net, id);
811 if (err < 0)
812 goto err_out;
814 rtnl_notify(msg, net, 0, RTNLGRP_NSID, NULL, 0);
815 return;
817 err_out:
818 nlmsg_free(msg);
819 out:
820 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
823 static int __init net_ns_init(void)
825 struct net_generic *ng;
827 #ifdef CONFIG_NET_NS
828 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
829 SMP_CACHE_BYTES,
830 SLAB_PANIC, NULL);
832 /* Create workqueue for cleanup */
833 netns_wq = create_singlethread_workqueue("netns");
834 if (!netns_wq)
835 panic("Could not create netns workq");
836 #endif
838 ng = net_alloc_generic();
839 if (!ng)
840 panic("Could not allocate generic netns");
842 rcu_assign_pointer(init_net.gen, ng);
844 mutex_lock(&net_mutex);
845 if (setup_net(&init_net, &init_user_ns))
846 panic("Could not setup the initial network namespace");
848 init_net_initialized = true;
850 rtnl_lock();
851 list_add_tail_rcu(&init_net.list, &net_namespace_list);
852 rtnl_unlock();
854 mutex_unlock(&net_mutex);
856 register_pernet_subsys(&net_ns_ops);
858 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL, NULL);
859 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
860 NULL);
862 return 0;
865 pure_initcall(net_ns_init);
867 #ifdef CONFIG_NET_NS
868 static int __register_pernet_operations(struct list_head *list,
869 struct pernet_operations *ops)
871 struct net *net;
872 int error;
873 LIST_HEAD(net_exit_list);
875 list_add_tail(&ops->list, list);
876 if (ops->init || (ops->id && ops->size)) {
877 for_each_net(net) {
878 error = ops_init(ops, net);
879 if (error)
880 goto out_undo;
881 list_add_tail(&net->exit_list, &net_exit_list);
884 return 0;
886 out_undo:
887 /* If I have an error cleanup all namespaces I initialized */
888 list_del(&ops->list);
889 ops_exit_list(ops, &net_exit_list);
890 ops_free_list(ops, &net_exit_list);
891 return error;
894 static void __unregister_pernet_operations(struct pernet_operations *ops)
896 struct net *net;
897 LIST_HEAD(net_exit_list);
899 list_del(&ops->list);
900 for_each_net(net)
901 list_add_tail(&net->exit_list, &net_exit_list);
902 ops_exit_list(ops, &net_exit_list);
903 ops_free_list(ops, &net_exit_list);
906 #else
908 static int __register_pernet_operations(struct list_head *list,
909 struct pernet_operations *ops)
911 if (!init_net_initialized) {
912 list_add_tail(&ops->list, list);
913 return 0;
916 return ops_init(ops, &init_net);
919 static void __unregister_pernet_operations(struct pernet_operations *ops)
921 if (!init_net_initialized) {
922 list_del(&ops->list);
923 } else {
924 LIST_HEAD(net_exit_list);
925 list_add(&init_net.exit_list, &net_exit_list);
926 ops_exit_list(ops, &net_exit_list);
927 ops_free_list(ops, &net_exit_list);
931 #endif /* CONFIG_NET_NS */
933 static DEFINE_IDA(net_generic_ids);
935 static int register_pernet_operations(struct list_head *list,
936 struct pernet_operations *ops)
938 int error;
940 if (ops->id) {
941 again:
942 error = ida_get_new_above(&net_generic_ids, MIN_PERNET_OPS_ID, ops->id);
943 if (error < 0) {
944 if (error == -EAGAIN) {
945 ida_pre_get(&net_generic_ids, GFP_KERNEL);
946 goto again;
948 return error;
950 max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
952 error = __register_pernet_operations(list, ops);
953 if (error) {
954 rcu_barrier();
955 if (ops->id)
956 ida_remove(&net_generic_ids, *ops->id);
959 return error;
962 static void unregister_pernet_operations(struct pernet_operations *ops)
965 __unregister_pernet_operations(ops);
966 rcu_barrier();
967 if (ops->id)
968 ida_remove(&net_generic_ids, *ops->id);
972 * register_pernet_subsys - register a network namespace subsystem
973 * @ops: pernet operations structure for the subsystem
975 * Register a subsystem which has init and exit functions
976 * that are called when network namespaces are created and
977 * destroyed respectively.
979 * When registered all network namespace init functions are
980 * called for every existing network namespace. Allowing kernel
981 * modules to have a race free view of the set of network namespaces.
983 * When a new network namespace is created all of the init
984 * methods are called in the order in which they were registered.
986 * When a network namespace is destroyed all of the exit methods
987 * are called in the reverse of the order with which they were
988 * registered.
990 int register_pernet_subsys(struct pernet_operations *ops)
992 int error;
993 mutex_lock(&net_mutex);
994 error = register_pernet_operations(first_device, ops);
995 mutex_unlock(&net_mutex);
996 return error;
998 EXPORT_SYMBOL_GPL(register_pernet_subsys);
1001 * unregister_pernet_subsys - unregister a network namespace subsystem
1002 * @ops: pernet operations structure to manipulate
1004 * Remove the pernet operations structure from the list to be
1005 * used when network namespaces are created or destroyed. In
1006 * addition run the exit method for all existing network
1007 * namespaces.
1009 void unregister_pernet_subsys(struct pernet_operations *ops)
1011 mutex_lock(&net_mutex);
1012 unregister_pernet_operations(ops);
1013 mutex_unlock(&net_mutex);
1015 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1018 * register_pernet_device - register a network namespace device
1019 * @ops: pernet operations structure for the subsystem
1021 * Register a device which has init and exit functions
1022 * that are called when network namespaces are created and
1023 * destroyed respectively.
1025 * When registered all network namespace init functions are
1026 * called for every existing network namespace. Allowing kernel
1027 * modules to have a race free view of the set of network namespaces.
1029 * When a new network namespace is created all of the init
1030 * methods are called in the order in which they were registered.
1032 * When a network namespace is destroyed all of the exit methods
1033 * are called in the reverse of the order with which they were
1034 * registered.
1036 int register_pernet_device(struct pernet_operations *ops)
1038 int error;
1039 mutex_lock(&net_mutex);
1040 error = register_pernet_operations(&pernet_list, ops);
1041 if (!error && (first_device == &pernet_list))
1042 first_device = &ops->list;
1043 mutex_unlock(&net_mutex);
1044 return error;
1046 EXPORT_SYMBOL_GPL(register_pernet_device);
1049 * unregister_pernet_device - unregister a network namespace netdevice
1050 * @ops: pernet operations structure to manipulate
1052 * Remove the pernet operations structure from the list to be
1053 * used when network namespaces are created or destroyed. In
1054 * addition run the exit method for all existing network
1055 * namespaces.
1057 void unregister_pernet_device(struct pernet_operations *ops)
1059 mutex_lock(&net_mutex);
1060 if (&ops->list == first_device)
1061 first_device = first_device->next;
1062 unregister_pernet_operations(ops);
1063 mutex_unlock(&net_mutex);
1065 EXPORT_SYMBOL_GPL(unregister_pernet_device);
1067 #ifdef CONFIG_NET_NS
1068 static struct ns_common *netns_get(struct task_struct *task)
1070 struct net *net = NULL;
1071 struct nsproxy *nsproxy;
1073 task_lock(task);
1074 nsproxy = task->nsproxy;
1075 if (nsproxy)
1076 net = get_net(nsproxy->net_ns);
1077 task_unlock(task);
1079 return net ? &net->ns : NULL;
1082 static inline struct net *to_net_ns(struct ns_common *ns)
1084 return container_of(ns, struct net, ns);
1087 static void netns_put(struct ns_common *ns)
1089 put_net(to_net_ns(ns));
1092 static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
1094 struct net *net = to_net_ns(ns);
1096 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1097 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1098 return -EPERM;
1100 put_net(nsproxy->net_ns);
1101 nsproxy->net_ns = get_net(net);
1102 return 0;
1105 static struct user_namespace *netns_owner(struct ns_common *ns)
1107 return to_net_ns(ns)->user_ns;
1110 const struct proc_ns_operations netns_operations = {
1111 .name = "net",
1112 .type = CLONE_NEWNET,
1113 .get = netns_get,
1114 .put = netns_put,
1115 .install = netns_install,
1116 .owner = netns_owner,
1118 #endif