Loading...
1#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
2
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 <net/sock.h>
20#include <net/netlink.h>
21#include <net/net_namespace.h>
22#include <net/netns/generic.h>
23
24/*
25 * Our network namespace constructor/destructor lists
26 */
27
28static LIST_HEAD(pernet_list);
29static struct list_head *first_device = &pernet_list;
30DEFINE_MUTEX(net_mutex);
31
32LIST_HEAD(net_namespace_list);
33EXPORT_SYMBOL_GPL(net_namespace_list);
34
35struct net init_net = {
36 .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head),
37};
38EXPORT_SYMBOL(init_net);
39
40#define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
41
42static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
43
44static struct net_generic *net_alloc_generic(void)
45{
46 struct net_generic *ng;
47 size_t generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
48
49 ng = kzalloc(generic_size, GFP_KERNEL);
50 if (ng)
51 ng->len = max_gen_ptrs;
52
53 return ng;
54}
55
56static int net_assign_generic(struct net *net, int id, void *data)
57{
58 struct net_generic *ng, *old_ng;
59
60 BUG_ON(!mutex_is_locked(&net_mutex));
61 BUG_ON(id == 0);
62
63 old_ng = rcu_dereference_protected(net->gen,
64 lockdep_is_held(&net_mutex));
65 ng = old_ng;
66 if (old_ng->len >= id)
67 goto assign;
68
69 ng = net_alloc_generic();
70 if (ng == NULL)
71 return -ENOMEM;
72
73 /*
74 * Some synchronisation notes:
75 *
76 * The net_generic explores the net->gen array inside rcu
77 * read section. Besides once set the net->gen->ptr[x]
78 * pointer never changes (see rules in netns/generic.h).
79 *
80 * That said, we simply duplicate this array and schedule
81 * the old copy for kfree after a grace period.
82 */
83
84 memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
85
86 rcu_assign_pointer(net->gen, ng);
87 kfree_rcu(old_ng, rcu);
88assign:
89 ng->ptr[id - 1] = data;
90 return 0;
91}
92
93static int ops_init(const struct pernet_operations *ops, struct net *net)
94{
95 int err = -ENOMEM;
96 void *data = NULL;
97
98 if (ops->id && ops->size) {
99 data = kzalloc(ops->size, GFP_KERNEL);
100 if (!data)
101 goto out;
102
103 err = net_assign_generic(net, *ops->id, data);
104 if (err)
105 goto cleanup;
106 }
107 err = 0;
108 if (ops->init)
109 err = ops->init(net);
110 if (!err)
111 return 0;
112
113cleanup:
114 kfree(data);
115
116out:
117 return err;
118}
119
120static void ops_free(const struct pernet_operations *ops, struct net *net)
121{
122 if (ops->id && ops->size) {
123 int id = *ops->id;
124 kfree(net_generic(net, id));
125 }
126}
127
128static void ops_exit_list(const struct pernet_operations *ops,
129 struct list_head *net_exit_list)
130{
131 struct net *net;
132 if (ops->exit) {
133 list_for_each_entry(net, net_exit_list, exit_list)
134 ops->exit(net);
135 }
136 if (ops->exit_batch)
137 ops->exit_batch(net_exit_list);
138}
139
140static void ops_free_list(const struct pernet_operations *ops,
141 struct list_head *net_exit_list)
142{
143 struct net *net;
144 if (ops->size && ops->id) {
145 list_for_each_entry(net, net_exit_list, exit_list)
146 ops_free(ops, net);
147 }
148}
149
150/* should be called with nsid_lock held */
151static int alloc_netid(struct net *net, struct net *peer, int reqid)
152{
153 int min = 0, max = 0;
154
155 if (reqid >= 0) {
156 min = reqid;
157 max = reqid + 1;
158 }
159
160 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
161}
162
163/* This function is used by idr_for_each(). If net is equal to peer, the
164 * function returns the id so that idr_for_each() stops. Because we cannot
165 * returns the id 0 (idr_for_each() will not stop), we return the magic value
166 * NET_ID_ZERO (-1) for it.
167 */
168#define NET_ID_ZERO -1
169static int net_eq_idr(int id, void *net, void *peer)
170{
171 if (net_eq(net, peer))
172 return id ? : NET_ID_ZERO;
173 return 0;
174}
175
176/* Should be called with nsid_lock held. If a new id is assigned, the bool alloc
177 * is set to true, thus the caller knows that the new id must be notified via
178 * rtnl.
179 */
180static int __peernet2id_alloc(struct net *net, struct net *peer, bool *alloc)
181{
182 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
183 bool alloc_it = *alloc;
184
185 *alloc = false;
186
187 /* Magic value for id 0. */
188 if (id == NET_ID_ZERO)
189 return 0;
190 if (id > 0)
191 return id;
192
193 if (alloc_it) {
194 id = alloc_netid(net, peer, -1);
195 *alloc = true;
196 return id >= 0 ? id : NETNSA_NSID_NOT_ASSIGNED;
197 }
198
199 return NETNSA_NSID_NOT_ASSIGNED;
200}
201
202/* should be called with nsid_lock held */
203static int __peernet2id(struct net *net, struct net *peer)
204{
205 bool no = false;
206
207 return __peernet2id_alloc(net, peer, &no);
208}
209
210static void rtnl_net_notifyid(struct net *net, int cmd, int id);
211/* This function returns the id of a peer netns. If no id is assigned, one will
212 * be allocated and returned.
213 */
214int peernet2id_alloc(struct net *net, struct net *peer)
215{
216 unsigned long flags;
217 bool alloc;
218 int id;
219
220 spin_lock_irqsave(&net->nsid_lock, flags);
221 alloc = atomic_read(&peer->count) == 0 ? false : true;
222 id = __peernet2id_alloc(net, peer, &alloc);
223 spin_unlock_irqrestore(&net->nsid_lock, flags);
224 if (alloc && id >= 0)
225 rtnl_net_notifyid(net, RTM_NEWNSID, id);
226 return id;
227}
228EXPORT_SYMBOL(peernet2id_alloc);
229
230/* This function returns, if assigned, the id of a peer netns. */
231int peernet2id(struct net *net, struct net *peer)
232{
233 unsigned long flags;
234 int id;
235
236 spin_lock_irqsave(&net->nsid_lock, flags);
237 id = __peernet2id(net, peer);
238 spin_unlock_irqrestore(&net->nsid_lock, flags);
239 return id;
240}
241
242/* This function returns true is the peer netns has an id assigned into the
243 * current netns.
244 */
245bool peernet_has_id(struct net *net, struct net *peer)
246{
247 return peernet2id(net, peer) >= 0;
248}
249
250struct net *get_net_ns_by_id(struct net *net, int id)
251{
252 unsigned long flags;
253 struct net *peer;
254
255 if (id < 0)
256 return NULL;
257
258 rcu_read_lock();
259 spin_lock_irqsave(&net->nsid_lock, flags);
260 peer = idr_find(&net->netns_ids, id);
261 if (peer)
262 get_net(peer);
263 spin_unlock_irqrestore(&net->nsid_lock, flags);
264 rcu_read_unlock();
265
266 return peer;
267}
268
269/*
270 * setup_net runs the initializers for the network namespace object.
271 */
272static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
273{
274 /* Must be called with net_mutex held */
275 const struct pernet_operations *ops, *saved_ops;
276 int error = 0;
277 LIST_HEAD(net_exit_list);
278
279 atomic_set(&net->count, 1);
280 atomic_set(&net->passive, 1);
281 net->dev_base_seq = 1;
282 net->user_ns = user_ns;
283 idr_init(&net->netns_ids);
284 spin_lock_init(&net->nsid_lock);
285
286 list_for_each_entry(ops, &pernet_list, list) {
287 error = ops_init(ops, net);
288 if (error < 0)
289 goto out_undo;
290 }
291out:
292 return error;
293
294out_undo:
295 /* Walk through the list backwards calling the exit functions
296 * for the pernet modules whose init functions did not fail.
297 */
298 list_add(&net->exit_list, &net_exit_list);
299 saved_ops = ops;
300 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
301 ops_exit_list(ops, &net_exit_list);
302
303 ops = saved_ops;
304 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
305 ops_free_list(ops, &net_exit_list);
306
307 rcu_barrier();
308 goto out;
309}
310
311
312#ifdef CONFIG_NET_NS
313static struct kmem_cache *net_cachep;
314static struct workqueue_struct *netns_wq;
315
316static struct net *net_alloc(void)
317{
318 struct net *net = NULL;
319 struct net_generic *ng;
320
321 ng = net_alloc_generic();
322 if (!ng)
323 goto out;
324
325 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
326 if (!net)
327 goto out_free;
328
329 rcu_assign_pointer(net->gen, ng);
330out:
331 return net;
332
333out_free:
334 kfree(ng);
335 goto out;
336}
337
338static void net_free(struct net *net)
339{
340 kfree(rcu_access_pointer(net->gen));
341 kmem_cache_free(net_cachep, net);
342}
343
344void net_drop_ns(void *p)
345{
346 struct net *ns = p;
347 if (ns && atomic_dec_and_test(&ns->passive))
348 net_free(ns);
349}
350
351struct net *copy_net_ns(unsigned long flags,
352 struct user_namespace *user_ns, struct net *old_net)
353{
354 struct net *net;
355 int rv;
356
357 if (!(flags & CLONE_NEWNET))
358 return get_net(old_net);
359
360 net = net_alloc();
361 if (!net)
362 return ERR_PTR(-ENOMEM);
363
364 get_user_ns(user_ns);
365
366 mutex_lock(&net_mutex);
367 rv = setup_net(net, user_ns);
368 if (rv == 0) {
369 rtnl_lock();
370 list_add_tail_rcu(&net->list, &net_namespace_list);
371 rtnl_unlock();
372 }
373 mutex_unlock(&net_mutex);
374 if (rv < 0) {
375 put_user_ns(user_ns);
376 net_drop_ns(net);
377 return ERR_PTR(rv);
378 }
379 return net;
380}
381
382static DEFINE_SPINLOCK(cleanup_list_lock);
383static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
384
385static void cleanup_net(struct work_struct *work)
386{
387 const struct pernet_operations *ops;
388 struct net *net, *tmp;
389 struct list_head net_kill_list;
390 LIST_HEAD(net_exit_list);
391
392 /* Atomically snapshot the list of namespaces to cleanup */
393 spin_lock_irq(&cleanup_list_lock);
394 list_replace_init(&cleanup_list, &net_kill_list);
395 spin_unlock_irq(&cleanup_list_lock);
396
397 mutex_lock(&net_mutex);
398
399 /* Don't let anyone else find us. */
400 rtnl_lock();
401 list_for_each_entry(net, &net_kill_list, cleanup_list) {
402 list_del_rcu(&net->list);
403 list_add_tail(&net->exit_list, &net_exit_list);
404 for_each_net(tmp) {
405 int id;
406
407 spin_lock_irq(&tmp->nsid_lock);
408 id = __peernet2id(tmp, net);
409 if (id >= 0)
410 idr_remove(&tmp->netns_ids, id);
411 spin_unlock_irq(&tmp->nsid_lock);
412 if (id >= 0)
413 rtnl_net_notifyid(tmp, RTM_DELNSID, id);
414 }
415 spin_lock_irq(&net->nsid_lock);
416 idr_destroy(&net->netns_ids);
417 spin_unlock_irq(&net->nsid_lock);
418
419 }
420 rtnl_unlock();
421
422 /*
423 * Another CPU might be rcu-iterating the list, wait for it.
424 * This needs to be before calling the exit() notifiers, so
425 * the rcu_barrier() below isn't sufficient alone.
426 */
427 synchronize_rcu();
428
429 /* Run all of the network namespace exit methods */
430 list_for_each_entry_reverse(ops, &pernet_list, list)
431 ops_exit_list(ops, &net_exit_list);
432
433 /* Free the net generic variables */
434 list_for_each_entry_reverse(ops, &pernet_list, list)
435 ops_free_list(ops, &net_exit_list);
436
437 mutex_unlock(&net_mutex);
438
439 /* Ensure there are no outstanding rcu callbacks using this
440 * network namespace.
441 */
442 rcu_barrier();
443
444 /* Finally it is safe to free my network namespace structure */
445 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
446 list_del_init(&net->exit_list);
447 put_user_ns(net->user_ns);
448 net_drop_ns(net);
449 }
450}
451static DECLARE_WORK(net_cleanup_work, cleanup_net);
452
453void __put_net(struct net *net)
454{
455 /* Cleanup the network namespace in process context */
456 unsigned long flags;
457
458 spin_lock_irqsave(&cleanup_list_lock, flags);
459 list_add(&net->cleanup_list, &cleanup_list);
460 spin_unlock_irqrestore(&cleanup_list_lock, flags);
461
462 queue_work(netns_wq, &net_cleanup_work);
463}
464EXPORT_SYMBOL_GPL(__put_net);
465
466struct net *get_net_ns_by_fd(int fd)
467{
468 struct file *file;
469 struct ns_common *ns;
470 struct net *net;
471
472 file = proc_ns_fget(fd);
473 if (IS_ERR(file))
474 return ERR_CAST(file);
475
476 ns = get_proc_ns(file_inode(file));
477 if (ns->ops == &netns_operations)
478 net = get_net(container_of(ns, struct net, ns));
479 else
480 net = ERR_PTR(-EINVAL);
481
482 fput(file);
483 return net;
484}
485
486#else
487struct net *get_net_ns_by_fd(int fd)
488{
489 return ERR_PTR(-EINVAL);
490}
491#endif
492EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
493
494struct net *get_net_ns_by_pid(pid_t pid)
495{
496 struct task_struct *tsk;
497 struct net *net;
498
499 /* Lookup the network namespace */
500 net = ERR_PTR(-ESRCH);
501 rcu_read_lock();
502 tsk = find_task_by_vpid(pid);
503 if (tsk) {
504 struct nsproxy *nsproxy;
505 task_lock(tsk);
506 nsproxy = tsk->nsproxy;
507 if (nsproxy)
508 net = get_net(nsproxy->net_ns);
509 task_unlock(tsk);
510 }
511 rcu_read_unlock();
512 return net;
513}
514EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
515
516static __net_init int net_ns_net_init(struct net *net)
517{
518#ifdef CONFIG_NET_NS
519 net->ns.ops = &netns_operations;
520#endif
521 return ns_alloc_inum(&net->ns);
522}
523
524static __net_exit void net_ns_net_exit(struct net *net)
525{
526 ns_free_inum(&net->ns);
527}
528
529static struct pernet_operations __net_initdata net_ns_ops = {
530 .init = net_ns_net_init,
531 .exit = net_ns_net_exit,
532};
533
534static struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
535 [NETNSA_NONE] = { .type = NLA_UNSPEC },
536 [NETNSA_NSID] = { .type = NLA_S32 },
537 [NETNSA_PID] = { .type = NLA_U32 },
538 [NETNSA_FD] = { .type = NLA_U32 },
539};
540
541static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh)
542{
543 struct net *net = sock_net(skb->sk);
544 struct nlattr *tb[NETNSA_MAX + 1];
545 unsigned long flags;
546 struct net *peer;
547 int nsid, err;
548
549 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
550 rtnl_net_policy);
551 if (err < 0)
552 return err;
553 if (!tb[NETNSA_NSID])
554 return -EINVAL;
555 nsid = nla_get_s32(tb[NETNSA_NSID]);
556
557 if (tb[NETNSA_PID])
558 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
559 else if (tb[NETNSA_FD])
560 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
561 else
562 return -EINVAL;
563 if (IS_ERR(peer))
564 return PTR_ERR(peer);
565
566 spin_lock_irqsave(&net->nsid_lock, flags);
567 if (__peernet2id(net, peer) >= 0) {
568 spin_unlock_irqrestore(&net->nsid_lock, flags);
569 err = -EEXIST;
570 goto out;
571 }
572
573 err = alloc_netid(net, peer, nsid);
574 spin_unlock_irqrestore(&net->nsid_lock, flags);
575 if (err >= 0) {
576 rtnl_net_notifyid(net, RTM_NEWNSID, err);
577 err = 0;
578 }
579out:
580 put_net(peer);
581 return err;
582}
583
584static int rtnl_net_get_size(void)
585{
586 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
587 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
588 ;
589}
590
591static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags,
592 int cmd, struct net *net, int nsid)
593{
594 struct nlmsghdr *nlh;
595 struct rtgenmsg *rth;
596
597 nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags);
598 if (!nlh)
599 return -EMSGSIZE;
600
601 rth = nlmsg_data(nlh);
602 rth->rtgen_family = AF_UNSPEC;
603
604 if (nla_put_s32(skb, NETNSA_NSID, nsid))
605 goto nla_put_failure;
606
607 nlmsg_end(skb, nlh);
608 return 0;
609
610nla_put_failure:
611 nlmsg_cancel(skb, nlh);
612 return -EMSGSIZE;
613}
614
615static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh)
616{
617 struct net *net = sock_net(skb->sk);
618 struct nlattr *tb[NETNSA_MAX + 1];
619 struct sk_buff *msg;
620 struct net *peer;
621 int err, id;
622
623 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
624 rtnl_net_policy);
625 if (err < 0)
626 return err;
627 if (tb[NETNSA_PID])
628 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
629 else if (tb[NETNSA_FD])
630 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
631 else
632 return -EINVAL;
633
634 if (IS_ERR(peer))
635 return PTR_ERR(peer);
636
637 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
638 if (!msg) {
639 err = -ENOMEM;
640 goto out;
641 }
642
643 id = peernet2id(net, peer);
644 err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0,
645 RTM_NEWNSID, net, id);
646 if (err < 0)
647 goto err_out;
648
649 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
650 goto out;
651
652err_out:
653 nlmsg_free(msg);
654out:
655 put_net(peer);
656 return err;
657}
658
659struct rtnl_net_dump_cb {
660 struct net *net;
661 struct sk_buff *skb;
662 struct netlink_callback *cb;
663 int idx;
664 int s_idx;
665};
666
667static int rtnl_net_dumpid_one(int id, void *peer, void *data)
668{
669 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
670 int ret;
671
672 if (net_cb->idx < net_cb->s_idx)
673 goto cont;
674
675 ret = rtnl_net_fill(net_cb->skb, NETLINK_CB(net_cb->cb->skb).portid,
676 net_cb->cb->nlh->nlmsg_seq, NLM_F_MULTI,
677 RTM_NEWNSID, net_cb->net, id);
678 if (ret < 0)
679 return ret;
680
681cont:
682 net_cb->idx++;
683 return 0;
684}
685
686static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
687{
688 struct net *net = sock_net(skb->sk);
689 struct rtnl_net_dump_cb net_cb = {
690 .net = net,
691 .skb = skb,
692 .cb = cb,
693 .idx = 0,
694 .s_idx = cb->args[0],
695 };
696 unsigned long flags;
697
698 spin_lock_irqsave(&net->nsid_lock, flags);
699 idr_for_each(&net->netns_ids, rtnl_net_dumpid_one, &net_cb);
700 spin_unlock_irqrestore(&net->nsid_lock, flags);
701
702 cb->args[0] = net_cb.idx;
703 return skb->len;
704}
705
706static void rtnl_net_notifyid(struct net *net, int cmd, int id)
707{
708 struct sk_buff *msg;
709 int err = -ENOMEM;
710
711 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
712 if (!msg)
713 goto out;
714
715 err = rtnl_net_fill(msg, 0, 0, 0, cmd, net, id);
716 if (err < 0)
717 goto err_out;
718
719 rtnl_notify(msg, net, 0, RTNLGRP_NSID, NULL, 0);
720 return;
721
722err_out:
723 nlmsg_free(msg);
724out:
725 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
726}
727
728static int __init net_ns_init(void)
729{
730 struct net_generic *ng;
731
732#ifdef CONFIG_NET_NS
733 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
734 SMP_CACHE_BYTES,
735 SLAB_PANIC, NULL);
736
737 /* Create workqueue for cleanup */
738 netns_wq = create_singlethread_workqueue("netns");
739 if (!netns_wq)
740 panic("Could not create netns workq");
741#endif
742
743 ng = net_alloc_generic();
744 if (!ng)
745 panic("Could not allocate generic netns");
746
747 rcu_assign_pointer(init_net.gen, ng);
748
749 mutex_lock(&net_mutex);
750 if (setup_net(&init_net, &init_user_ns))
751 panic("Could not setup the initial network namespace");
752
753 rtnl_lock();
754 list_add_tail_rcu(&init_net.list, &net_namespace_list);
755 rtnl_unlock();
756
757 mutex_unlock(&net_mutex);
758
759 register_pernet_subsys(&net_ns_ops);
760
761 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL, NULL);
762 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
763 NULL);
764
765 return 0;
766}
767
768pure_initcall(net_ns_init);
769
770#ifdef CONFIG_NET_NS
771static int __register_pernet_operations(struct list_head *list,
772 struct pernet_operations *ops)
773{
774 struct net *net;
775 int error;
776 LIST_HEAD(net_exit_list);
777
778 list_add_tail(&ops->list, list);
779 if (ops->init || (ops->id && ops->size)) {
780 for_each_net(net) {
781 error = ops_init(ops, net);
782 if (error)
783 goto out_undo;
784 list_add_tail(&net->exit_list, &net_exit_list);
785 }
786 }
787 return 0;
788
789out_undo:
790 /* If I have an error cleanup all namespaces I initialized */
791 list_del(&ops->list);
792 ops_exit_list(ops, &net_exit_list);
793 ops_free_list(ops, &net_exit_list);
794 return error;
795}
796
797static void __unregister_pernet_operations(struct pernet_operations *ops)
798{
799 struct net *net;
800 LIST_HEAD(net_exit_list);
801
802 list_del(&ops->list);
803 for_each_net(net)
804 list_add_tail(&net->exit_list, &net_exit_list);
805 ops_exit_list(ops, &net_exit_list);
806 ops_free_list(ops, &net_exit_list);
807}
808
809#else
810
811static int __register_pernet_operations(struct list_head *list,
812 struct pernet_operations *ops)
813{
814 return ops_init(ops, &init_net);
815}
816
817static void __unregister_pernet_operations(struct pernet_operations *ops)
818{
819 LIST_HEAD(net_exit_list);
820 list_add(&init_net.exit_list, &net_exit_list);
821 ops_exit_list(ops, &net_exit_list);
822 ops_free_list(ops, &net_exit_list);
823}
824
825#endif /* CONFIG_NET_NS */
826
827static DEFINE_IDA(net_generic_ids);
828
829static int register_pernet_operations(struct list_head *list,
830 struct pernet_operations *ops)
831{
832 int error;
833
834 if (ops->id) {
835again:
836 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
837 if (error < 0) {
838 if (error == -EAGAIN) {
839 ida_pre_get(&net_generic_ids, GFP_KERNEL);
840 goto again;
841 }
842 return error;
843 }
844 max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
845 }
846 error = __register_pernet_operations(list, ops);
847 if (error) {
848 rcu_barrier();
849 if (ops->id)
850 ida_remove(&net_generic_ids, *ops->id);
851 }
852
853 return error;
854}
855
856static void unregister_pernet_operations(struct pernet_operations *ops)
857{
858
859 __unregister_pernet_operations(ops);
860 rcu_barrier();
861 if (ops->id)
862 ida_remove(&net_generic_ids, *ops->id);
863}
864
865/**
866 * register_pernet_subsys - register a network namespace subsystem
867 * @ops: pernet operations structure for the subsystem
868 *
869 * Register a subsystem which has init and exit functions
870 * that are called when network namespaces are created and
871 * destroyed respectively.
872 *
873 * When registered all network namespace init functions are
874 * called for every existing network namespace. Allowing kernel
875 * modules to have a race free view of the set of network namespaces.
876 *
877 * When a new network namespace is created all of the init
878 * methods are called in the order in which they were registered.
879 *
880 * When a network namespace is destroyed all of the exit methods
881 * are called in the reverse of the order with which they were
882 * registered.
883 */
884int register_pernet_subsys(struct pernet_operations *ops)
885{
886 int error;
887 mutex_lock(&net_mutex);
888 error = register_pernet_operations(first_device, ops);
889 mutex_unlock(&net_mutex);
890 return error;
891}
892EXPORT_SYMBOL_GPL(register_pernet_subsys);
893
894/**
895 * unregister_pernet_subsys - unregister a network namespace subsystem
896 * @ops: pernet operations structure to manipulate
897 *
898 * Remove the pernet operations structure from the list to be
899 * used when network namespaces are created or destroyed. In
900 * addition run the exit method for all existing network
901 * namespaces.
902 */
903void unregister_pernet_subsys(struct pernet_operations *ops)
904{
905 mutex_lock(&net_mutex);
906 unregister_pernet_operations(ops);
907 mutex_unlock(&net_mutex);
908}
909EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
910
911/**
912 * register_pernet_device - register a network namespace device
913 * @ops: pernet operations structure for the subsystem
914 *
915 * Register a device which has init and exit functions
916 * that are called when network namespaces are created and
917 * destroyed respectively.
918 *
919 * When registered all network namespace init functions are
920 * called for every existing network namespace. Allowing kernel
921 * modules to have a race free view of the set of network namespaces.
922 *
923 * When a new network namespace is created all of the init
924 * methods are called in the order in which they were registered.
925 *
926 * When a network namespace is destroyed all of the exit methods
927 * are called in the reverse of the order with which they were
928 * registered.
929 */
930int register_pernet_device(struct pernet_operations *ops)
931{
932 int error;
933 mutex_lock(&net_mutex);
934 error = register_pernet_operations(&pernet_list, ops);
935 if (!error && (first_device == &pernet_list))
936 first_device = &ops->list;
937 mutex_unlock(&net_mutex);
938 return error;
939}
940EXPORT_SYMBOL_GPL(register_pernet_device);
941
942/**
943 * unregister_pernet_device - unregister a network namespace netdevice
944 * @ops: pernet operations structure to manipulate
945 *
946 * Remove the pernet operations structure from the list to be
947 * used when network namespaces are created or destroyed. In
948 * addition run the exit method for all existing network
949 * namespaces.
950 */
951void unregister_pernet_device(struct pernet_operations *ops)
952{
953 mutex_lock(&net_mutex);
954 if (&ops->list == first_device)
955 first_device = first_device->next;
956 unregister_pernet_operations(ops);
957 mutex_unlock(&net_mutex);
958}
959EXPORT_SYMBOL_GPL(unregister_pernet_device);
960
961#ifdef CONFIG_NET_NS
962static struct ns_common *netns_get(struct task_struct *task)
963{
964 struct net *net = NULL;
965 struct nsproxy *nsproxy;
966
967 task_lock(task);
968 nsproxy = task->nsproxy;
969 if (nsproxy)
970 net = get_net(nsproxy->net_ns);
971 task_unlock(task);
972
973 return net ? &net->ns : NULL;
974}
975
976static inline struct net *to_net_ns(struct ns_common *ns)
977{
978 return container_of(ns, struct net, ns);
979}
980
981static void netns_put(struct ns_common *ns)
982{
983 put_net(to_net_ns(ns));
984}
985
986static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
987{
988 struct net *net = to_net_ns(ns);
989
990 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
991 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
992 return -EPERM;
993
994 put_net(nsproxy->net_ns);
995 nsproxy->net_ns = get_net(net);
996 return 0;
997}
998
999const struct proc_ns_operations netns_operations = {
1000 .name = "net",
1001 .type = CLONE_NEWNET,
1002 .get = netns_get,
1003 .put = netns_put,
1004 .install = netns_install,
1005};
1006#endif
1#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
2
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 <net/sock.h>
20#include <net/netlink.h>
21#include <net/net_namespace.h>
22#include <net/netns/generic.h>
23
24/*
25 * Our network namespace constructor/destructor lists
26 */
27
28static LIST_HEAD(pernet_list);
29static struct list_head *first_device = &pernet_list;
30DEFINE_MUTEX(net_mutex);
31
32LIST_HEAD(net_namespace_list);
33EXPORT_SYMBOL_GPL(net_namespace_list);
34
35struct net init_net = {
36 .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head),
37};
38EXPORT_SYMBOL(init_net);
39
40static bool init_net_initialized;
41
42#define MIN_PERNET_OPS_ID \
43 ((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *))
44
45#define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
46
47static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
48
49static struct net_generic *net_alloc_generic(void)
50{
51 struct net_generic *ng;
52 unsigned int generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
53
54 ng = kzalloc(generic_size, GFP_KERNEL);
55 if (ng)
56 ng->s.len = max_gen_ptrs;
57
58 return ng;
59}
60
61static int net_assign_generic(struct net *net, unsigned int id, void *data)
62{
63 struct net_generic *ng, *old_ng;
64
65 BUG_ON(!mutex_is_locked(&net_mutex));
66 BUG_ON(id < MIN_PERNET_OPS_ID);
67
68 old_ng = rcu_dereference_protected(net->gen,
69 lockdep_is_held(&net_mutex));
70 if (old_ng->s.len > id) {
71 old_ng->ptr[id] = data;
72 return 0;
73 }
74
75 ng = net_alloc_generic();
76 if (ng == NULL)
77 return -ENOMEM;
78
79 /*
80 * Some synchronisation notes:
81 *
82 * The net_generic explores the net->gen array inside rcu
83 * read section. Besides once set the net->gen->ptr[x]
84 * pointer never changes (see rules in netns/generic.h).
85 *
86 * That said, we simply duplicate this array and schedule
87 * the old copy for kfree after a grace period.
88 */
89
90 memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
91 (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
92 ng->ptr[id] = data;
93
94 rcu_assign_pointer(net->gen, ng);
95 kfree_rcu(old_ng, s.rcu);
96 return 0;
97}
98
99static int ops_init(const struct pernet_operations *ops, struct net *net)
100{
101 int err = -ENOMEM;
102 void *data = NULL;
103
104 if (ops->id && ops->size) {
105 data = kzalloc(ops->size, GFP_KERNEL);
106 if (!data)
107 goto out;
108
109 err = net_assign_generic(net, *ops->id, data);
110 if (err)
111 goto cleanup;
112 }
113 err = 0;
114 if (ops->init)
115 err = ops->init(net);
116 if (!err)
117 return 0;
118
119cleanup:
120 kfree(data);
121
122out:
123 return err;
124}
125
126static void ops_free(const struct pernet_operations *ops, struct net *net)
127{
128 if (ops->id && ops->size) {
129 kfree(net_generic(net, *ops->id));
130 }
131}
132
133static void ops_exit_list(const struct pernet_operations *ops,
134 struct list_head *net_exit_list)
135{
136 struct net *net;
137 if (ops->exit) {
138 list_for_each_entry(net, net_exit_list, exit_list)
139 ops->exit(net);
140 }
141 if (ops->exit_batch)
142 ops->exit_batch(net_exit_list);
143}
144
145static void ops_free_list(const struct pernet_operations *ops,
146 struct list_head *net_exit_list)
147{
148 struct net *net;
149 if (ops->size && ops->id) {
150 list_for_each_entry(net, net_exit_list, exit_list)
151 ops_free(ops, net);
152 }
153}
154
155/* should be called with nsid_lock held */
156static int alloc_netid(struct net *net, struct net *peer, int reqid)
157{
158 int min = 0, max = 0;
159
160 if (reqid >= 0) {
161 min = reqid;
162 max = reqid + 1;
163 }
164
165 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
166}
167
168/* This function is used by idr_for_each(). If net is equal to peer, the
169 * function returns the id so that idr_for_each() stops. Because we cannot
170 * returns the id 0 (idr_for_each() will not stop), we return the magic value
171 * NET_ID_ZERO (-1) for it.
172 */
173#define NET_ID_ZERO -1
174static int net_eq_idr(int id, void *net, void *peer)
175{
176 if (net_eq(net, peer))
177 return id ? : NET_ID_ZERO;
178 return 0;
179}
180
181/* Should be called with nsid_lock held. If a new id is assigned, the bool alloc
182 * is set to true, thus the caller knows that the new id must be notified via
183 * rtnl.
184 */
185static int __peernet2id_alloc(struct net *net, struct net *peer, bool *alloc)
186{
187 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
188 bool alloc_it = *alloc;
189
190 *alloc = false;
191
192 /* Magic value for id 0. */
193 if (id == NET_ID_ZERO)
194 return 0;
195 if (id > 0)
196 return id;
197
198 if (alloc_it) {
199 id = alloc_netid(net, peer, -1);
200 *alloc = true;
201 return id >= 0 ? id : NETNSA_NSID_NOT_ASSIGNED;
202 }
203
204 return NETNSA_NSID_NOT_ASSIGNED;
205}
206
207/* should be called with nsid_lock held */
208static int __peernet2id(struct net *net, struct net *peer)
209{
210 bool no = false;
211
212 return __peernet2id_alloc(net, peer, &no);
213}
214
215static void rtnl_net_notifyid(struct net *net, int cmd, int id);
216/* This function returns the id of a peer netns. If no id is assigned, one will
217 * be allocated and returned.
218 */
219int peernet2id_alloc(struct net *net, struct net *peer)
220{
221 bool alloc;
222 int id;
223
224 if (atomic_read(&net->count) == 0)
225 return NETNSA_NSID_NOT_ASSIGNED;
226 spin_lock_bh(&net->nsid_lock);
227 alloc = atomic_read(&peer->count) == 0 ? false : true;
228 id = __peernet2id_alloc(net, peer, &alloc);
229 spin_unlock_bh(&net->nsid_lock);
230 if (alloc && id >= 0)
231 rtnl_net_notifyid(net, RTM_NEWNSID, id);
232 return id;
233}
234
235/* This function returns, if assigned, the id of a peer netns. */
236int peernet2id(struct net *net, struct net *peer)
237{
238 int id;
239
240 spin_lock_bh(&net->nsid_lock);
241 id = __peernet2id(net, peer);
242 spin_unlock_bh(&net->nsid_lock);
243 return id;
244}
245EXPORT_SYMBOL(peernet2id);
246
247/* This function returns true is the peer netns has an id assigned into the
248 * current netns.
249 */
250bool peernet_has_id(struct net *net, struct net *peer)
251{
252 return peernet2id(net, peer) >= 0;
253}
254
255struct net *get_net_ns_by_id(struct net *net, int id)
256{
257 struct net *peer;
258
259 if (id < 0)
260 return NULL;
261
262 rcu_read_lock();
263 spin_lock_bh(&net->nsid_lock);
264 peer = idr_find(&net->netns_ids, id);
265 if (peer)
266 get_net(peer);
267 spin_unlock_bh(&net->nsid_lock);
268 rcu_read_unlock();
269
270 return peer;
271}
272
273/*
274 * setup_net runs the initializers for the network namespace object.
275 */
276static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
277{
278 /* Must be called with net_mutex held */
279 const struct pernet_operations *ops, *saved_ops;
280 int error = 0;
281 LIST_HEAD(net_exit_list);
282
283 atomic_set(&net->count, 1);
284 atomic_set(&net->passive, 1);
285 net->dev_base_seq = 1;
286 net->user_ns = user_ns;
287 idr_init(&net->netns_ids);
288 spin_lock_init(&net->nsid_lock);
289
290 list_for_each_entry(ops, &pernet_list, list) {
291 error = ops_init(ops, net);
292 if (error < 0)
293 goto out_undo;
294 }
295out:
296 return error;
297
298out_undo:
299 /* Walk through the list backwards calling the exit functions
300 * for the pernet modules whose init functions did not fail.
301 */
302 list_add(&net->exit_list, &net_exit_list);
303 saved_ops = ops;
304 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
305 ops_exit_list(ops, &net_exit_list);
306
307 ops = saved_ops;
308 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
309 ops_free_list(ops, &net_exit_list);
310
311 rcu_barrier();
312 goto out;
313}
314
315
316#ifdef CONFIG_NET_NS
317static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
318{
319 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
320}
321
322static void dec_net_namespaces(struct ucounts *ucounts)
323{
324 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
325}
326
327static struct kmem_cache *net_cachep;
328static struct workqueue_struct *netns_wq;
329
330static struct net *net_alloc(void)
331{
332 struct net *net = NULL;
333 struct net_generic *ng;
334
335 ng = net_alloc_generic();
336 if (!ng)
337 goto out;
338
339 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
340 if (!net)
341 goto out_free;
342
343 rcu_assign_pointer(net->gen, ng);
344out:
345 return net;
346
347out_free:
348 kfree(ng);
349 goto out;
350}
351
352static void net_free(struct net *net)
353{
354 kfree(rcu_access_pointer(net->gen));
355 kmem_cache_free(net_cachep, net);
356}
357
358void net_drop_ns(void *p)
359{
360 struct net *ns = p;
361 if (ns && atomic_dec_and_test(&ns->passive))
362 net_free(ns);
363}
364
365struct net *copy_net_ns(unsigned long flags,
366 struct user_namespace *user_ns, struct net *old_net)
367{
368 struct ucounts *ucounts;
369 struct net *net;
370 int rv;
371
372 if (!(flags & CLONE_NEWNET))
373 return get_net(old_net);
374
375 ucounts = inc_net_namespaces(user_ns);
376 if (!ucounts)
377 return ERR_PTR(-ENOSPC);
378
379 net = net_alloc();
380 if (!net) {
381 dec_net_namespaces(ucounts);
382 return ERR_PTR(-ENOMEM);
383 }
384
385 get_user_ns(user_ns);
386
387 rv = mutex_lock_killable(&net_mutex);
388 if (rv < 0) {
389 net_free(net);
390 dec_net_namespaces(ucounts);
391 put_user_ns(user_ns);
392 return ERR_PTR(rv);
393 }
394
395 net->ucounts = ucounts;
396 rv = setup_net(net, user_ns);
397 if (rv == 0) {
398 rtnl_lock();
399 list_add_tail_rcu(&net->list, &net_namespace_list);
400 rtnl_unlock();
401 }
402 mutex_unlock(&net_mutex);
403 if (rv < 0) {
404 dec_net_namespaces(ucounts);
405 put_user_ns(user_ns);
406 net_drop_ns(net);
407 return ERR_PTR(rv);
408 }
409 return net;
410}
411
412static DEFINE_SPINLOCK(cleanup_list_lock);
413static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
414
415static void cleanup_net(struct work_struct *work)
416{
417 const struct pernet_operations *ops;
418 struct net *net, *tmp;
419 struct list_head net_kill_list;
420 LIST_HEAD(net_exit_list);
421
422 /* Atomically snapshot the list of namespaces to cleanup */
423 spin_lock_irq(&cleanup_list_lock);
424 list_replace_init(&cleanup_list, &net_kill_list);
425 spin_unlock_irq(&cleanup_list_lock);
426
427 mutex_lock(&net_mutex);
428
429 /* Don't let anyone else find us. */
430 rtnl_lock();
431 list_for_each_entry(net, &net_kill_list, cleanup_list) {
432 list_del_rcu(&net->list);
433 list_add_tail(&net->exit_list, &net_exit_list);
434 for_each_net(tmp) {
435 int id;
436
437 spin_lock_bh(&tmp->nsid_lock);
438 id = __peernet2id(tmp, net);
439 if (id >= 0)
440 idr_remove(&tmp->netns_ids, id);
441 spin_unlock_bh(&tmp->nsid_lock);
442 if (id >= 0)
443 rtnl_net_notifyid(tmp, RTM_DELNSID, id);
444 }
445 spin_lock_bh(&net->nsid_lock);
446 idr_destroy(&net->netns_ids);
447 spin_unlock_bh(&net->nsid_lock);
448
449 }
450 rtnl_unlock();
451
452 /*
453 * Another CPU might be rcu-iterating the list, wait for it.
454 * This needs to be before calling the exit() notifiers, so
455 * the rcu_barrier() below isn't sufficient alone.
456 */
457 synchronize_rcu();
458
459 /* Run all of the network namespace exit methods */
460 list_for_each_entry_reverse(ops, &pernet_list, list)
461 ops_exit_list(ops, &net_exit_list);
462
463 /* Free the net generic variables */
464 list_for_each_entry_reverse(ops, &pernet_list, list)
465 ops_free_list(ops, &net_exit_list);
466
467 mutex_unlock(&net_mutex);
468
469 /* Ensure there are no outstanding rcu callbacks using this
470 * network namespace.
471 */
472 rcu_barrier();
473
474 /* Finally it is safe to free my network namespace structure */
475 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
476 list_del_init(&net->exit_list);
477 dec_net_namespaces(net->ucounts);
478 put_user_ns(net->user_ns);
479 net_drop_ns(net);
480 }
481}
482static DECLARE_WORK(net_cleanup_work, cleanup_net);
483
484void __put_net(struct net *net)
485{
486 /* Cleanup the network namespace in process context */
487 unsigned long flags;
488
489 spin_lock_irqsave(&cleanup_list_lock, flags);
490 list_add(&net->cleanup_list, &cleanup_list);
491 spin_unlock_irqrestore(&cleanup_list_lock, flags);
492
493 queue_work(netns_wq, &net_cleanup_work);
494}
495EXPORT_SYMBOL_GPL(__put_net);
496
497struct net *get_net_ns_by_fd(int fd)
498{
499 struct file *file;
500 struct ns_common *ns;
501 struct net *net;
502
503 file = proc_ns_fget(fd);
504 if (IS_ERR(file))
505 return ERR_CAST(file);
506
507 ns = get_proc_ns(file_inode(file));
508 if (ns->ops == &netns_operations)
509 net = get_net(container_of(ns, struct net, ns));
510 else
511 net = ERR_PTR(-EINVAL);
512
513 fput(file);
514 return net;
515}
516
517#else
518struct net *get_net_ns_by_fd(int fd)
519{
520 return ERR_PTR(-EINVAL);
521}
522#endif
523EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
524
525struct net *get_net_ns_by_pid(pid_t pid)
526{
527 struct task_struct *tsk;
528 struct net *net;
529
530 /* Lookup the network namespace */
531 net = ERR_PTR(-ESRCH);
532 rcu_read_lock();
533 tsk = find_task_by_vpid(pid);
534 if (tsk) {
535 struct nsproxy *nsproxy;
536 task_lock(tsk);
537 nsproxy = tsk->nsproxy;
538 if (nsproxy)
539 net = get_net(nsproxy->net_ns);
540 task_unlock(tsk);
541 }
542 rcu_read_unlock();
543 return net;
544}
545EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
546
547static __net_init int net_ns_net_init(struct net *net)
548{
549#ifdef CONFIG_NET_NS
550 net->ns.ops = &netns_operations;
551#endif
552 return ns_alloc_inum(&net->ns);
553}
554
555static __net_exit void net_ns_net_exit(struct net *net)
556{
557 ns_free_inum(&net->ns);
558}
559
560static struct pernet_operations __net_initdata net_ns_ops = {
561 .init = net_ns_net_init,
562 .exit = net_ns_net_exit,
563};
564
565static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
566 [NETNSA_NONE] = { .type = NLA_UNSPEC },
567 [NETNSA_NSID] = { .type = NLA_S32 },
568 [NETNSA_PID] = { .type = NLA_U32 },
569 [NETNSA_FD] = { .type = NLA_U32 },
570};
571
572static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh)
573{
574 struct net *net = sock_net(skb->sk);
575 struct nlattr *tb[NETNSA_MAX + 1];
576 struct net *peer;
577 int nsid, err;
578
579 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
580 rtnl_net_policy);
581 if (err < 0)
582 return err;
583 if (!tb[NETNSA_NSID])
584 return -EINVAL;
585 nsid = nla_get_s32(tb[NETNSA_NSID]);
586
587 if (tb[NETNSA_PID])
588 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
589 else if (tb[NETNSA_FD])
590 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
591 else
592 return -EINVAL;
593 if (IS_ERR(peer))
594 return PTR_ERR(peer);
595
596 spin_lock_bh(&net->nsid_lock);
597 if (__peernet2id(net, peer) >= 0) {
598 spin_unlock_bh(&net->nsid_lock);
599 err = -EEXIST;
600 goto out;
601 }
602
603 err = alloc_netid(net, peer, nsid);
604 spin_unlock_bh(&net->nsid_lock);
605 if (err >= 0) {
606 rtnl_net_notifyid(net, RTM_NEWNSID, err);
607 err = 0;
608 }
609out:
610 put_net(peer);
611 return err;
612}
613
614static int rtnl_net_get_size(void)
615{
616 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
617 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
618 ;
619}
620
621static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags,
622 int cmd, struct net *net, int nsid)
623{
624 struct nlmsghdr *nlh;
625 struct rtgenmsg *rth;
626
627 nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags);
628 if (!nlh)
629 return -EMSGSIZE;
630
631 rth = nlmsg_data(nlh);
632 rth->rtgen_family = AF_UNSPEC;
633
634 if (nla_put_s32(skb, NETNSA_NSID, nsid))
635 goto nla_put_failure;
636
637 nlmsg_end(skb, nlh);
638 return 0;
639
640nla_put_failure:
641 nlmsg_cancel(skb, nlh);
642 return -EMSGSIZE;
643}
644
645static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh)
646{
647 struct net *net = sock_net(skb->sk);
648 struct nlattr *tb[NETNSA_MAX + 1];
649 struct sk_buff *msg;
650 struct net *peer;
651 int err, id;
652
653 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
654 rtnl_net_policy);
655 if (err < 0)
656 return err;
657 if (tb[NETNSA_PID])
658 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
659 else if (tb[NETNSA_FD])
660 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
661 else
662 return -EINVAL;
663
664 if (IS_ERR(peer))
665 return PTR_ERR(peer);
666
667 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
668 if (!msg) {
669 err = -ENOMEM;
670 goto out;
671 }
672
673 id = peernet2id(net, peer);
674 err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0,
675 RTM_NEWNSID, net, id);
676 if (err < 0)
677 goto err_out;
678
679 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
680 goto out;
681
682err_out:
683 nlmsg_free(msg);
684out:
685 put_net(peer);
686 return err;
687}
688
689struct rtnl_net_dump_cb {
690 struct net *net;
691 struct sk_buff *skb;
692 struct netlink_callback *cb;
693 int idx;
694 int s_idx;
695};
696
697static int rtnl_net_dumpid_one(int id, void *peer, void *data)
698{
699 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
700 int ret;
701
702 if (net_cb->idx < net_cb->s_idx)
703 goto cont;
704
705 ret = rtnl_net_fill(net_cb->skb, NETLINK_CB(net_cb->cb->skb).portid,
706 net_cb->cb->nlh->nlmsg_seq, NLM_F_MULTI,
707 RTM_NEWNSID, net_cb->net, id);
708 if (ret < 0)
709 return ret;
710
711cont:
712 net_cb->idx++;
713 return 0;
714}
715
716static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
717{
718 struct net *net = sock_net(skb->sk);
719 struct rtnl_net_dump_cb net_cb = {
720 .net = net,
721 .skb = skb,
722 .cb = cb,
723 .idx = 0,
724 .s_idx = cb->args[0],
725 };
726
727 spin_lock_bh(&net->nsid_lock);
728 idr_for_each(&net->netns_ids, rtnl_net_dumpid_one, &net_cb);
729 spin_unlock_bh(&net->nsid_lock);
730
731 cb->args[0] = net_cb.idx;
732 return skb->len;
733}
734
735static void rtnl_net_notifyid(struct net *net, int cmd, int id)
736{
737 struct sk_buff *msg;
738 int err = -ENOMEM;
739
740 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
741 if (!msg)
742 goto out;
743
744 err = rtnl_net_fill(msg, 0, 0, 0, cmd, net, id);
745 if (err < 0)
746 goto err_out;
747
748 rtnl_notify(msg, net, 0, RTNLGRP_NSID, NULL, 0);
749 return;
750
751err_out:
752 nlmsg_free(msg);
753out:
754 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
755}
756
757static int __init net_ns_init(void)
758{
759 struct net_generic *ng;
760
761#ifdef CONFIG_NET_NS
762 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
763 SMP_CACHE_BYTES,
764 SLAB_PANIC, NULL);
765
766 /* Create workqueue for cleanup */
767 netns_wq = create_singlethread_workqueue("netns");
768 if (!netns_wq)
769 panic("Could not create netns workq");
770#endif
771
772 ng = net_alloc_generic();
773 if (!ng)
774 panic("Could not allocate generic netns");
775
776 rcu_assign_pointer(init_net.gen, ng);
777
778 mutex_lock(&net_mutex);
779 if (setup_net(&init_net, &init_user_ns))
780 panic("Could not setup the initial network namespace");
781
782 init_net_initialized = true;
783
784 rtnl_lock();
785 list_add_tail_rcu(&init_net.list, &net_namespace_list);
786 rtnl_unlock();
787
788 mutex_unlock(&net_mutex);
789
790 register_pernet_subsys(&net_ns_ops);
791
792 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL, NULL);
793 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
794 NULL);
795
796 return 0;
797}
798
799pure_initcall(net_ns_init);
800
801#ifdef CONFIG_NET_NS
802static int __register_pernet_operations(struct list_head *list,
803 struct pernet_operations *ops)
804{
805 struct net *net;
806 int error;
807 LIST_HEAD(net_exit_list);
808
809 list_add_tail(&ops->list, list);
810 if (ops->init || (ops->id && ops->size)) {
811 for_each_net(net) {
812 error = ops_init(ops, net);
813 if (error)
814 goto out_undo;
815 list_add_tail(&net->exit_list, &net_exit_list);
816 }
817 }
818 return 0;
819
820out_undo:
821 /* If I have an error cleanup all namespaces I initialized */
822 list_del(&ops->list);
823 ops_exit_list(ops, &net_exit_list);
824 ops_free_list(ops, &net_exit_list);
825 return error;
826}
827
828static void __unregister_pernet_operations(struct pernet_operations *ops)
829{
830 struct net *net;
831 LIST_HEAD(net_exit_list);
832
833 list_del(&ops->list);
834 for_each_net(net)
835 list_add_tail(&net->exit_list, &net_exit_list);
836 ops_exit_list(ops, &net_exit_list);
837 ops_free_list(ops, &net_exit_list);
838}
839
840#else
841
842static int __register_pernet_operations(struct list_head *list,
843 struct pernet_operations *ops)
844{
845 if (!init_net_initialized) {
846 list_add_tail(&ops->list, list);
847 return 0;
848 }
849
850 return ops_init(ops, &init_net);
851}
852
853static void __unregister_pernet_operations(struct pernet_operations *ops)
854{
855 if (!init_net_initialized) {
856 list_del(&ops->list);
857 } else {
858 LIST_HEAD(net_exit_list);
859 list_add(&init_net.exit_list, &net_exit_list);
860 ops_exit_list(ops, &net_exit_list);
861 ops_free_list(ops, &net_exit_list);
862 }
863}
864
865#endif /* CONFIG_NET_NS */
866
867static DEFINE_IDA(net_generic_ids);
868
869static int register_pernet_operations(struct list_head *list,
870 struct pernet_operations *ops)
871{
872 int error;
873
874 if (ops->id) {
875again:
876 error = ida_get_new_above(&net_generic_ids, MIN_PERNET_OPS_ID, ops->id);
877 if (error < 0) {
878 if (error == -EAGAIN) {
879 ida_pre_get(&net_generic_ids, GFP_KERNEL);
880 goto again;
881 }
882 return error;
883 }
884 max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
885 }
886 error = __register_pernet_operations(list, ops);
887 if (error) {
888 rcu_barrier();
889 if (ops->id)
890 ida_remove(&net_generic_ids, *ops->id);
891 }
892
893 return error;
894}
895
896static void unregister_pernet_operations(struct pernet_operations *ops)
897{
898
899 __unregister_pernet_operations(ops);
900 rcu_barrier();
901 if (ops->id)
902 ida_remove(&net_generic_ids, *ops->id);
903}
904
905/**
906 * register_pernet_subsys - register a network namespace subsystem
907 * @ops: pernet operations structure for the subsystem
908 *
909 * Register a subsystem which has init and exit functions
910 * that are called when network namespaces are created and
911 * destroyed respectively.
912 *
913 * When registered all network namespace init functions are
914 * called for every existing network namespace. Allowing kernel
915 * modules to have a race free view of the set of network namespaces.
916 *
917 * When a new network namespace is created all of the init
918 * methods are called in the order in which they were registered.
919 *
920 * When a network namespace is destroyed all of the exit methods
921 * are called in the reverse of the order with which they were
922 * registered.
923 */
924int register_pernet_subsys(struct pernet_operations *ops)
925{
926 int error;
927 mutex_lock(&net_mutex);
928 error = register_pernet_operations(first_device, ops);
929 mutex_unlock(&net_mutex);
930 return error;
931}
932EXPORT_SYMBOL_GPL(register_pernet_subsys);
933
934/**
935 * unregister_pernet_subsys - unregister a network namespace subsystem
936 * @ops: pernet operations structure to manipulate
937 *
938 * Remove the pernet operations structure from the list to be
939 * used when network namespaces are created or destroyed. In
940 * addition run the exit method for all existing network
941 * namespaces.
942 */
943void unregister_pernet_subsys(struct pernet_operations *ops)
944{
945 mutex_lock(&net_mutex);
946 unregister_pernet_operations(ops);
947 mutex_unlock(&net_mutex);
948}
949EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
950
951/**
952 * register_pernet_device - register a network namespace device
953 * @ops: pernet operations structure for the subsystem
954 *
955 * Register a device which has init and exit functions
956 * that are called when network namespaces are created and
957 * destroyed respectively.
958 *
959 * When registered all network namespace init functions are
960 * called for every existing network namespace. Allowing kernel
961 * modules to have a race free view of the set of network namespaces.
962 *
963 * When a new network namespace is created all of the init
964 * methods are called in the order in which they were registered.
965 *
966 * When a network namespace is destroyed all of the exit methods
967 * are called in the reverse of the order with which they were
968 * registered.
969 */
970int register_pernet_device(struct pernet_operations *ops)
971{
972 int error;
973 mutex_lock(&net_mutex);
974 error = register_pernet_operations(&pernet_list, ops);
975 if (!error && (first_device == &pernet_list))
976 first_device = &ops->list;
977 mutex_unlock(&net_mutex);
978 return error;
979}
980EXPORT_SYMBOL_GPL(register_pernet_device);
981
982/**
983 * unregister_pernet_device - unregister a network namespace netdevice
984 * @ops: pernet operations structure to manipulate
985 *
986 * Remove the pernet operations structure from the list to be
987 * used when network namespaces are created or destroyed. In
988 * addition run the exit method for all existing network
989 * namespaces.
990 */
991void unregister_pernet_device(struct pernet_operations *ops)
992{
993 mutex_lock(&net_mutex);
994 if (&ops->list == first_device)
995 first_device = first_device->next;
996 unregister_pernet_operations(ops);
997 mutex_unlock(&net_mutex);
998}
999EXPORT_SYMBOL_GPL(unregister_pernet_device);
1000
1001#ifdef CONFIG_NET_NS
1002static struct ns_common *netns_get(struct task_struct *task)
1003{
1004 struct net *net = NULL;
1005 struct nsproxy *nsproxy;
1006
1007 task_lock(task);
1008 nsproxy = task->nsproxy;
1009 if (nsproxy)
1010 net = get_net(nsproxy->net_ns);
1011 task_unlock(task);
1012
1013 return net ? &net->ns : NULL;
1014}
1015
1016static inline struct net *to_net_ns(struct ns_common *ns)
1017{
1018 return container_of(ns, struct net, ns);
1019}
1020
1021static void netns_put(struct ns_common *ns)
1022{
1023 put_net(to_net_ns(ns));
1024}
1025
1026static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
1027{
1028 struct net *net = to_net_ns(ns);
1029
1030 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1031 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1032 return -EPERM;
1033
1034 put_net(nsproxy->net_ns);
1035 nsproxy->net_ns = get_net(net);
1036 return 0;
1037}
1038
1039static struct user_namespace *netns_owner(struct ns_common *ns)
1040{
1041 return to_net_ns(ns)->user_ns;
1042}
1043
1044const struct proc_ns_operations netns_operations = {
1045 .name = "net",
1046 .type = CLONE_NEWNET,
1047 .get = netns_get,
1048 .put = netns_put,
1049 .install = netns_install,
1050 .owner = netns_owner,
1051};
1052#endif