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 <linux/sched/task.h>
20
21#include <net/sock.h>
22#include <net/netlink.h>
23#include <net/net_namespace.h>
24#include <net/netns/generic.h>
25
26/*
27 * Our network namespace constructor/destructor lists
28 */
29
30static LIST_HEAD(pernet_list);
31static struct list_head *first_device = &pernet_list;
32
33LIST_HEAD(net_namespace_list);
34EXPORT_SYMBOL_GPL(net_namespace_list);
35
36/* Protects net_namespace_list. Nests iside rtnl_lock() */
37DECLARE_RWSEM(net_rwsem);
38EXPORT_SYMBOL_GPL(net_rwsem);
39
40struct net init_net = {
41 .count = REFCOUNT_INIT(1),
42 .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head),
43};
44EXPORT_SYMBOL(init_net);
45
46static bool init_net_initialized;
47/*
48 * pernet_ops_rwsem: protects: pernet_list, net_generic_ids,
49 * init_net_initialized and first_device pointer.
50 * This is internal net namespace object. Please, don't use it
51 * outside.
52 */
53DECLARE_RWSEM(pernet_ops_rwsem);
54EXPORT_SYMBOL_GPL(pernet_ops_rwsem);
55
56#define MIN_PERNET_OPS_ID \
57 ((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *))
58
59#define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
60
61static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
62
63static struct net_generic *net_alloc_generic(void)
64{
65 struct net_generic *ng;
66 unsigned int generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
67
68 ng = kzalloc(generic_size, GFP_KERNEL);
69 if (ng)
70 ng->s.len = max_gen_ptrs;
71
72 return ng;
73}
74
75static int net_assign_generic(struct net *net, unsigned int id, void *data)
76{
77 struct net_generic *ng, *old_ng;
78
79 BUG_ON(id < MIN_PERNET_OPS_ID);
80
81 old_ng = rcu_dereference_protected(net->gen,
82 lockdep_is_held(&pernet_ops_rwsem));
83 if (old_ng->s.len > id) {
84 old_ng->ptr[id] = data;
85 return 0;
86 }
87
88 ng = net_alloc_generic();
89 if (ng == NULL)
90 return -ENOMEM;
91
92 /*
93 * Some synchronisation notes:
94 *
95 * The net_generic explores the net->gen array inside rcu
96 * read section. Besides once set the net->gen->ptr[x]
97 * pointer never changes (see rules in netns/generic.h).
98 *
99 * That said, we simply duplicate this array and schedule
100 * the old copy for kfree after a grace period.
101 */
102
103 memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
104 (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
105 ng->ptr[id] = data;
106
107 rcu_assign_pointer(net->gen, ng);
108 kfree_rcu(old_ng, s.rcu);
109 return 0;
110}
111
112static int ops_init(const struct pernet_operations *ops, struct net *net)
113{
114 int err = -ENOMEM;
115 void *data = NULL;
116
117 if (ops->id && ops->size) {
118 data = kzalloc(ops->size, GFP_KERNEL);
119 if (!data)
120 goto out;
121
122 err = net_assign_generic(net, *ops->id, data);
123 if (err)
124 goto cleanup;
125 }
126 err = 0;
127 if (ops->init)
128 err = ops->init(net);
129 if (!err)
130 return 0;
131
132cleanup:
133 kfree(data);
134
135out:
136 return err;
137}
138
139static void ops_free(const struct pernet_operations *ops, struct net *net)
140{
141 if (ops->id && ops->size) {
142 kfree(net_generic(net, *ops->id));
143 }
144}
145
146static void ops_exit_list(const struct pernet_operations *ops,
147 struct list_head *net_exit_list)
148{
149 struct net *net;
150 if (ops->exit) {
151 list_for_each_entry(net, net_exit_list, exit_list)
152 ops->exit(net);
153 }
154 if (ops->exit_batch)
155 ops->exit_batch(net_exit_list);
156}
157
158static void ops_free_list(const struct pernet_operations *ops,
159 struct list_head *net_exit_list)
160{
161 struct net *net;
162 if (ops->size && ops->id) {
163 list_for_each_entry(net, net_exit_list, exit_list)
164 ops_free(ops, net);
165 }
166}
167
168/* should be called with nsid_lock held */
169static int alloc_netid(struct net *net, struct net *peer, int reqid)
170{
171 int min = 0, max = 0;
172
173 if (reqid >= 0) {
174 min = reqid;
175 max = reqid + 1;
176 }
177
178 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
179}
180
181/* This function is used by idr_for_each(). If net is equal to peer, the
182 * function returns the id so that idr_for_each() stops. Because we cannot
183 * returns the id 0 (idr_for_each() will not stop), we return the magic value
184 * NET_ID_ZERO (-1) for it.
185 */
186#define NET_ID_ZERO -1
187static int net_eq_idr(int id, void *net, void *peer)
188{
189 if (net_eq(net, peer))
190 return id ? : NET_ID_ZERO;
191 return 0;
192}
193
194/* Should be called with nsid_lock held. If a new id is assigned, the bool alloc
195 * is set to true, thus the caller knows that the new id must be notified via
196 * rtnl.
197 */
198static int __peernet2id_alloc(struct net *net, struct net *peer, bool *alloc)
199{
200 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
201 bool alloc_it = *alloc;
202
203 *alloc = false;
204
205 /* Magic value for id 0. */
206 if (id == NET_ID_ZERO)
207 return 0;
208 if (id > 0)
209 return id;
210
211 if (alloc_it) {
212 id = alloc_netid(net, peer, -1);
213 *alloc = true;
214 return id >= 0 ? id : NETNSA_NSID_NOT_ASSIGNED;
215 }
216
217 return NETNSA_NSID_NOT_ASSIGNED;
218}
219
220/* should be called with nsid_lock held */
221static int __peernet2id(struct net *net, struct net *peer)
222{
223 bool no = false;
224
225 return __peernet2id_alloc(net, peer, &no);
226}
227
228static void rtnl_net_notifyid(struct net *net, int cmd, int id);
229/* This function returns the id of a peer netns. If no id is assigned, one will
230 * be allocated and returned.
231 */
232int peernet2id_alloc(struct net *net, struct net *peer)
233{
234 bool alloc = false, alive = false;
235 int id;
236
237 if (refcount_read(&net->count) == 0)
238 return NETNSA_NSID_NOT_ASSIGNED;
239 spin_lock_bh(&net->nsid_lock);
240 /*
241 * When peer is obtained from RCU lists, we may race with
242 * its cleanup. Check whether it's alive, and this guarantees
243 * we never hash a peer back to net->netns_ids, after it has
244 * just been idr_remove()'d from there in cleanup_net().
245 */
246 if (maybe_get_net(peer))
247 alive = alloc = true;
248 id = __peernet2id_alloc(net, peer, &alloc);
249 spin_unlock_bh(&net->nsid_lock);
250 if (alloc && id >= 0)
251 rtnl_net_notifyid(net, RTM_NEWNSID, id);
252 if (alive)
253 put_net(peer);
254 return id;
255}
256EXPORT_SYMBOL_GPL(peernet2id_alloc);
257
258/* This function returns, if assigned, the id of a peer netns. */
259int peernet2id(struct net *net, struct net *peer)
260{
261 int id;
262
263 spin_lock_bh(&net->nsid_lock);
264 id = __peernet2id(net, peer);
265 spin_unlock_bh(&net->nsid_lock);
266 return id;
267}
268EXPORT_SYMBOL(peernet2id);
269
270/* This function returns true is the peer netns has an id assigned into the
271 * current netns.
272 */
273bool peernet_has_id(struct net *net, struct net *peer)
274{
275 return peernet2id(net, peer) >= 0;
276}
277
278struct net *get_net_ns_by_id(struct net *net, int id)
279{
280 struct net *peer;
281
282 if (id < 0)
283 return NULL;
284
285 rcu_read_lock();
286 peer = idr_find(&net->netns_ids, id);
287 if (peer)
288 peer = maybe_get_net(peer);
289 rcu_read_unlock();
290
291 return peer;
292}
293
294/*
295 * setup_net runs the initializers for the network namespace object.
296 */
297static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
298{
299 /* Must be called with pernet_ops_rwsem held */
300 const struct pernet_operations *ops, *saved_ops;
301 int error = 0;
302 LIST_HEAD(net_exit_list);
303
304 refcount_set(&net->count, 1);
305 refcount_set(&net->passive, 1);
306 net->dev_base_seq = 1;
307 net->user_ns = user_ns;
308 idr_init(&net->netns_ids);
309 spin_lock_init(&net->nsid_lock);
310 mutex_init(&net->ipv4.ra_mutex);
311
312 list_for_each_entry(ops, &pernet_list, list) {
313 error = ops_init(ops, net);
314 if (error < 0)
315 goto out_undo;
316 }
317 down_write(&net_rwsem);
318 list_add_tail_rcu(&net->list, &net_namespace_list);
319 up_write(&net_rwsem);
320out:
321 return error;
322
323out_undo:
324 /* Walk through the list backwards calling the exit functions
325 * for the pernet modules whose init functions did not fail.
326 */
327 list_add(&net->exit_list, &net_exit_list);
328 saved_ops = ops;
329 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
330 ops_exit_list(ops, &net_exit_list);
331
332 ops = saved_ops;
333 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
334 ops_free_list(ops, &net_exit_list);
335
336 rcu_barrier();
337 goto out;
338}
339
340static int __net_init net_defaults_init_net(struct net *net)
341{
342 net->core.sysctl_somaxconn = SOMAXCONN;
343 return 0;
344}
345
346static struct pernet_operations net_defaults_ops = {
347 .init = net_defaults_init_net,
348};
349
350static __init int net_defaults_init(void)
351{
352 if (register_pernet_subsys(&net_defaults_ops))
353 panic("Cannot initialize net default settings");
354
355 return 0;
356}
357
358core_initcall(net_defaults_init);
359
360#ifdef CONFIG_NET_NS
361static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
362{
363 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
364}
365
366static void dec_net_namespaces(struct ucounts *ucounts)
367{
368 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
369}
370
371static struct kmem_cache *net_cachep __ro_after_init;
372static struct workqueue_struct *netns_wq;
373
374static struct net *net_alloc(void)
375{
376 struct net *net = NULL;
377 struct net_generic *ng;
378
379 ng = net_alloc_generic();
380 if (!ng)
381 goto out;
382
383 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
384 if (!net)
385 goto out_free;
386
387 rcu_assign_pointer(net->gen, ng);
388out:
389 return net;
390
391out_free:
392 kfree(ng);
393 goto out;
394}
395
396static void net_free(struct net *net)
397{
398 kfree(rcu_access_pointer(net->gen));
399 kmem_cache_free(net_cachep, net);
400}
401
402void net_drop_ns(void *p)
403{
404 struct net *ns = p;
405 if (ns && refcount_dec_and_test(&ns->passive))
406 net_free(ns);
407}
408
409struct net *copy_net_ns(unsigned long flags,
410 struct user_namespace *user_ns, struct net *old_net)
411{
412 struct ucounts *ucounts;
413 struct net *net;
414 int rv;
415
416 if (!(flags & CLONE_NEWNET))
417 return get_net(old_net);
418
419 ucounts = inc_net_namespaces(user_ns);
420 if (!ucounts)
421 return ERR_PTR(-ENOSPC);
422
423 net = net_alloc();
424 if (!net) {
425 rv = -ENOMEM;
426 goto dec_ucounts;
427 }
428 refcount_set(&net->passive, 1);
429 net->ucounts = ucounts;
430 get_user_ns(user_ns);
431
432 rv = down_read_killable(&pernet_ops_rwsem);
433 if (rv < 0)
434 goto put_userns;
435
436 rv = setup_net(net, user_ns);
437
438 up_read(&pernet_ops_rwsem);
439
440 if (rv < 0) {
441put_userns:
442 put_user_ns(user_ns);
443 net_drop_ns(net);
444dec_ucounts:
445 dec_net_namespaces(ucounts);
446 return ERR_PTR(rv);
447 }
448 return net;
449}
450
451static void unhash_nsid(struct net *net, struct net *last)
452{
453 struct net *tmp;
454 /* This function is only called from cleanup_net() work,
455 * and this work is the only process, that may delete
456 * a net from net_namespace_list. So, when the below
457 * is executing, the list may only grow. Thus, we do not
458 * use for_each_net_rcu() or net_rwsem.
459 */
460 for_each_net(tmp) {
461 int id;
462
463 spin_lock_bh(&tmp->nsid_lock);
464 id = __peernet2id(tmp, net);
465 if (id >= 0)
466 idr_remove(&tmp->netns_ids, id);
467 spin_unlock_bh(&tmp->nsid_lock);
468 if (id >= 0)
469 rtnl_net_notifyid(tmp, RTM_DELNSID, id);
470 if (tmp == last)
471 break;
472 }
473 spin_lock_bh(&net->nsid_lock);
474 idr_destroy(&net->netns_ids);
475 spin_unlock_bh(&net->nsid_lock);
476}
477
478static LLIST_HEAD(cleanup_list);
479
480static void cleanup_net(struct work_struct *work)
481{
482 const struct pernet_operations *ops;
483 struct net *net, *tmp, *last;
484 struct llist_node *net_kill_list;
485 LIST_HEAD(net_exit_list);
486
487 /* Atomically snapshot the list of namespaces to cleanup */
488 net_kill_list = llist_del_all(&cleanup_list);
489
490 down_read(&pernet_ops_rwsem);
491
492 /* Don't let anyone else find us. */
493 down_write(&net_rwsem);
494 llist_for_each_entry(net, net_kill_list, cleanup_list)
495 list_del_rcu(&net->list);
496 /* Cache last net. After we unlock rtnl, no one new net
497 * added to net_namespace_list can assign nsid pointer
498 * to a net from net_kill_list (see peernet2id_alloc()).
499 * So, we skip them in unhash_nsid().
500 *
501 * Note, that unhash_nsid() does not delete nsid links
502 * between net_kill_list's nets, as they've already
503 * deleted from net_namespace_list. But, this would be
504 * useless anyway, as netns_ids are destroyed there.
505 */
506 last = list_last_entry(&net_namespace_list, struct net, list);
507 up_write(&net_rwsem);
508
509 llist_for_each_entry(net, net_kill_list, cleanup_list) {
510 unhash_nsid(net, last);
511 list_add_tail(&net->exit_list, &net_exit_list);
512 }
513
514 /*
515 * Another CPU might be rcu-iterating the list, wait for it.
516 * This needs to be before calling the exit() notifiers, so
517 * the rcu_barrier() below isn't sufficient alone.
518 */
519 synchronize_rcu();
520
521 /* Run all of the network namespace exit methods */
522 list_for_each_entry_reverse(ops, &pernet_list, list)
523 ops_exit_list(ops, &net_exit_list);
524
525 /* Free the net generic variables */
526 list_for_each_entry_reverse(ops, &pernet_list, list)
527 ops_free_list(ops, &net_exit_list);
528
529 up_read(&pernet_ops_rwsem);
530
531 /* Ensure there are no outstanding rcu callbacks using this
532 * network namespace.
533 */
534 rcu_barrier();
535
536 /* Finally it is safe to free my network namespace structure */
537 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
538 list_del_init(&net->exit_list);
539 dec_net_namespaces(net->ucounts);
540 put_user_ns(net->user_ns);
541 net_drop_ns(net);
542 }
543}
544
545/**
546 * net_ns_barrier - wait until concurrent net_cleanup_work is done
547 *
548 * cleanup_net runs from work queue and will first remove namespaces
549 * from the global list, then run net exit functions.
550 *
551 * Call this in module exit path to make sure that all netns
552 * ->exit ops have been invoked before the function is removed.
553 */
554void net_ns_barrier(void)
555{
556 down_write(&pernet_ops_rwsem);
557 up_write(&pernet_ops_rwsem);
558}
559EXPORT_SYMBOL(net_ns_barrier);
560
561static DECLARE_WORK(net_cleanup_work, cleanup_net);
562
563void __put_net(struct net *net)
564{
565 /* Cleanup the network namespace in process context */
566 if (llist_add(&net->cleanup_list, &cleanup_list))
567 queue_work(netns_wq, &net_cleanup_work);
568}
569EXPORT_SYMBOL_GPL(__put_net);
570
571struct net *get_net_ns_by_fd(int fd)
572{
573 struct file *file;
574 struct ns_common *ns;
575 struct net *net;
576
577 file = proc_ns_fget(fd);
578 if (IS_ERR(file))
579 return ERR_CAST(file);
580
581 ns = get_proc_ns(file_inode(file));
582 if (ns->ops == &netns_operations)
583 net = get_net(container_of(ns, struct net, ns));
584 else
585 net = ERR_PTR(-EINVAL);
586
587 fput(file);
588 return net;
589}
590
591#else
592struct net *get_net_ns_by_fd(int fd)
593{
594 return ERR_PTR(-EINVAL);
595}
596#endif
597EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
598
599struct net *get_net_ns_by_pid(pid_t pid)
600{
601 struct task_struct *tsk;
602 struct net *net;
603
604 /* Lookup the network namespace */
605 net = ERR_PTR(-ESRCH);
606 rcu_read_lock();
607 tsk = find_task_by_vpid(pid);
608 if (tsk) {
609 struct nsproxy *nsproxy;
610 task_lock(tsk);
611 nsproxy = tsk->nsproxy;
612 if (nsproxy)
613 net = get_net(nsproxy->net_ns);
614 task_unlock(tsk);
615 }
616 rcu_read_unlock();
617 return net;
618}
619EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
620
621static __net_init int net_ns_net_init(struct net *net)
622{
623#ifdef CONFIG_NET_NS
624 net->ns.ops = &netns_operations;
625#endif
626 return ns_alloc_inum(&net->ns);
627}
628
629static __net_exit void net_ns_net_exit(struct net *net)
630{
631 ns_free_inum(&net->ns);
632}
633
634static struct pernet_operations __net_initdata net_ns_ops = {
635 .init = net_ns_net_init,
636 .exit = net_ns_net_exit,
637};
638
639static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
640 [NETNSA_NONE] = { .type = NLA_UNSPEC },
641 [NETNSA_NSID] = { .type = NLA_S32 },
642 [NETNSA_PID] = { .type = NLA_U32 },
643 [NETNSA_FD] = { .type = NLA_U32 },
644};
645
646static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
647 struct netlink_ext_ack *extack)
648{
649 struct net *net = sock_net(skb->sk);
650 struct nlattr *tb[NETNSA_MAX + 1];
651 struct nlattr *nla;
652 struct net *peer;
653 int nsid, err;
654
655 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
656 rtnl_net_policy, extack);
657 if (err < 0)
658 return err;
659 if (!tb[NETNSA_NSID]) {
660 NL_SET_ERR_MSG(extack, "nsid is missing");
661 return -EINVAL;
662 }
663 nsid = nla_get_s32(tb[NETNSA_NSID]);
664
665 if (tb[NETNSA_PID]) {
666 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
667 nla = tb[NETNSA_PID];
668 } else if (tb[NETNSA_FD]) {
669 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
670 nla = tb[NETNSA_FD];
671 } else {
672 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
673 return -EINVAL;
674 }
675 if (IS_ERR(peer)) {
676 NL_SET_BAD_ATTR(extack, nla);
677 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
678 return PTR_ERR(peer);
679 }
680
681 spin_lock_bh(&net->nsid_lock);
682 if (__peernet2id(net, peer) >= 0) {
683 spin_unlock_bh(&net->nsid_lock);
684 err = -EEXIST;
685 NL_SET_BAD_ATTR(extack, nla);
686 NL_SET_ERR_MSG(extack,
687 "Peer netns already has a nsid assigned");
688 goto out;
689 }
690
691 err = alloc_netid(net, peer, nsid);
692 spin_unlock_bh(&net->nsid_lock);
693 if (err >= 0) {
694 rtnl_net_notifyid(net, RTM_NEWNSID, err);
695 err = 0;
696 } else if (err == -ENOSPC && nsid >= 0) {
697 err = -EEXIST;
698 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
699 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
700 }
701out:
702 put_net(peer);
703 return err;
704}
705
706static int rtnl_net_get_size(void)
707{
708 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
709 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
710 ;
711}
712
713static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags,
714 int cmd, struct net *net, int nsid)
715{
716 struct nlmsghdr *nlh;
717 struct rtgenmsg *rth;
718
719 nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags);
720 if (!nlh)
721 return -EMSGSIZE;
722
723 rth = nlmsg_data(nlh);
724 rth->rtgen_family = AF_UNSPEC;
725
726 if (nla_put_s32(skb, NETNSA_NSID, nsid))
727 goto nla_put_failure;
728
729 nlmsg_end(skb, nlh);
730 return 0;
731
732nla_put_failure:
733 nlmsg_cancel(skb, nlh);
734 return -EMSGSIZE;
735}
736
737static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
738 struct netlink_ext_ack *extack)
739{
740 struct net *net = sock_net(skb->sk);
741 struct nlattr *tb[NETNSA_MAX + 1];
742 struct nlattr *nla;
743 struct sk_buff *msg;
744 struct net *peer;
745 int err, id;
746
747 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
748 rtnl_net_policy, extack);
749 if (err < 0)
750 return err;
751 if (tb[NETNSA_PID]) {
752 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
753 nla = tb[NETNSA_PID];
754 } else if (tb[NETNSA_FD]) {
755 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
756 nla = tb[NETNSA_FD];
757 } else {
758 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
759 return -EINVAL;
760 }
761
762 if (IS_ERR(peer)) {
763 NL_SET_BAD_ATTR(extack, nla);
764 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
765 return PTR_ERR(peer);
766 }
767
768 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
769 if (!msg) {
770 err = -ENOMEM;
771 goto out;
772 }
773
774 id = peernet2id(net, peer);
775 err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0,
776 RTM_NEWNSID, net, id);
777 if (err < 0)
778 goto err_out;
779
780 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
781 goto out;
782
783err_out:
784 nlmsg_free(msg);
785out:
786 put_net(peer);
787 return err;
788}
789
790struct rtnl_net_dump_cb {
791 struct net *net;
792 struct sk_buff *skb;
793 struct netlink_callback *cb;
794 int idx;
795 int s_idx;
796};
797
798static int rtnl_net_dumpid_one(int id, void *peer, void *data)
799{
800 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
801 int ret;
802
803 if (net_cb->idx < net_cb->s_idx)
804 goto cont;
805
806 ret = rtnl_net_fill(net_cb->skb, NETLINK_CB(net_cb->cb->skb).portid,
807 net_cb->cb->nlh->nlmsg_seq, NLM_F_MULTI,
808 RTM_NEWNSID, net_cb->net, id);
809 if (ret < 0)
810 return ret;
811
812cont:
813 net_cb->idx++;
814 return 0;
815}
816
817static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
818{
819 struct net *net = sock_net(skb->sk);
820 struct rtnl_net_dump_cb net_cb = {
821 .net = net,
822 .skb = skb,
823 .cb = cb,
824 .idx = 0,
825 .s_idx = cb->args[0],
826 };
827
828 spin_lock_bh(&net->nsid_lock);
829 idr_for_each(&net->netns_ids, rtnl_net_dumpid_one, &net_cb);
830 spin_unlock_bh(&net->nsid_lock);
831
832 cb->args[0] = net_cb.idx;
833 return skb->len;
834}
835
836static void rtnl_net_notifyid(struct net *net, int cmd, int id)
837{
838 struct sk_buff *msg;
839 int err = -ENOMEM;
840
841 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
842 if (!msg)
843 goto out;
844
845 err = rtnl_net_fill(msg, 0, 0, 0, cmd, net, id);
846 if (err < 0)
847 goto err_out;
848
849 rtnl_notify(msg, net, 0, RTNLGRP_NSID, NULL, 0);
850 return;
851
852err_out:
853 nlmsg_free(msg);
854out:
855 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
856}
857
858static int __init net_ns_init(void)
859{
860 struct net_generic *ng;
861
862#ifdef CONFIG_NET_NS
863 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
864 SMP_CACHE_BYTES,
865 SLAB_PANIC|SLAB_ACCOUNT, NULL);
866
867 /* Create workqueue for cleanup */
868 netns_wq = create_singlethread_workqueue("netns");
869 if (!netns_wq)
870 panic("Could not create netns workq");
871#endif
872
873 ng = net_alloc_generic();
874 if (!ng)
875 panic("Could not allocate generic netns");
876
877 rcu_assign_pointer(init_net.gen, ng);
878
879 down_write(&pernet_ops_rwsem);
880 if (setup_net(&init_net, &init_user_ns))
881 panic("Could not setup the initial network namespace");
882
883 init_net_initialized = true;
884 up_write(&pernet_ops_rwsem);
885
886 register_pernet_subsys(&net_ns_ops);
887
888 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
889 RTNL_FLAG_DOIT_UNLOCKED);
890 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
891 RTNL_FLAG_DOIT_UNLOCKED);
892
893 return 0;
894}
895
896pure_initcall(net_ns_init);
897
898#ifdef CONFIG_NET_NS
899static int __register_pernet_operations(struct list_head *list,
900 struct pernet_operations *ops)
901{
902 struct net *net;
903 int error;
904 LIST_HEAD(net_exit_list);
905
906 list_add_tail(&ops->list, list);
907 if (ops->init || (ops->id && ops->size)) {
908 /* We held write locked pernet_ops_rwsem, and parallel
909 * setup_net() and cleanup_net() are not possible.
910 */
911 for_each_net(net) {
912 error = ops_init(ops, net);
913 if (error)
914 goto out_undo;
915 list_add_tail(&net->exit_list, &net_exit_list);
916 }
917 }
918 return 0;
919
920out_undo:
921 /* If I have an error cleanup all namespaces I initialized */
922 list_del(&ops->list);
923 ops_exit_list(ops, &net_exit_list);
924 ops_free_list(ops, &net_exit_list);
925 return error;
926}
927
928static void __unregister_pernet_operations(struct pernet_operations *ops)
929{
930 struct net *net;
931 LIST_HEAD(net_exit_list);
932
933 list_del(&ops->list);
934 /* See comment in __register_pernet_operations() */
935 for_each_net(net)
936 list_add_tail(&net->exit_list, &net_exit_list);
937 ops_exit_list(ops, &net_exit_list);
938 ops_free_list(ops, &net_exit_list);
939}
940
941#else
942
943static int __register_pernet_operations(struct list_head *list,
944 struct pernet_operations *ops)
945{
946 if (!init_net_initialized) {
947 list_add_tail(&ops->list, list);
948 return 0;
949 }
950
951 return ops_init(ops, &init_net);
952}
953
954static void __unregister_pernet_operations(struct pernet_operations *ops)
955{
956 if (!init_net_initialized) {
957 list_del(&ops->list);
958 } else {
959 LIST_HEAD(net_exit_list);
960 list_add(&init_net.exit_list, &net_exit_list);
961 ops_exit_list(ops, &net_exit_list);
962 ops_free_list(ops, &net_exit_list);
963 }
964}
965
966#endif /* CONFIG_NET_NS */
967
968static DEFINE_IDA(net_generic_ids);
969
970static int register_pernet_operations(struct list_head *list,
971 struct pernet_operations *ops)
972{
973 int error;
974
975 if (ops->id) {
976again:
977 error = ida_get_new_above(&net_generic_ids, MIN_PERNET_OPS_ID, ops->id);
978 if (error < 0) {
979 if (error == -EAGAIN) {
980 ida_pre_get(&net_generic_ids, GFP_KERNEL);
981 goto again;
982 }
983 return error;
984 }
985 max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
986 }
987 error = __register_pernet_operations(list, ops);
988 if (error) {
989 rcu_barrier();
990 if (ops->id)
991 ida_remove(&net_generic_ids, *ops->id);
992 }
993
994 return error;
995}
996
997static void unregister_pernet_operations(struct pernet_operations *ops)
998{
999 __unregister_pernet_operations(ops);
1000 rcu_barrier();
1001 if (ops->id)
1002 ida_remove(&net_generic_ids, *ops->id);
1003}
1004
1005/**
1006 * register_pernet_subsys - register a network namespace subsystem
1007 * @ops: pernet operations structure for the subsystem
1008 *
1009 * Register a subsystem which has init and exit functions
1010 * that are called when network namespaces are created and
1011 * destroyed respectively.
1012 *
1013 * When registered all network namespace init functions are
1014 * called for every existing network namespace. Allowing kernel
1015 * modules to have a race free view of the set of network namespaces.
1016 *
1017 * When a new network namespace is created all of the init
1018 * methods are called in the order in which they were registered.
1019 *
1020 * When a network namespace is destroyed all of the exit methods
1021 * are called in the reverse of the order with which they were
1022 * registered.
1023 */
1024int register_pernet_subsys(struct pernet_operations *ops)
1025{
1026 int error;
1027 down_write(&pernet_ops_rwsem);
1028 error = register_pernet_operations(first_device, ops);
1029 up_write(&pernet_ops_rwsem);
1030 return error;
1031}
1032EXPORT_SYMBOL_GPL(register_pernet_subsys);
1033
1034/**
1035 * unregister_pernet_subsys - unregister a network namespace subsystem
1036 * @ops: pernet operations structure to manipulate
1037 *
1038 * Remove the pernet operations structure from the list to be
1039 * used when network namespaces are created or destroyed. In
1040 * addition run the exit method for all existing network
1041 * namespaces.
1042 */
1043void unregister_pernet_subsys(struct pernet_operations *ops)
1044{
1045 down_write(&pernet_ops_rwsem);
1046 unregister_pernet_operations(ops);
1047 up_write(&pernet_ops_rwsem);
1048}
1049EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1050
1051/**
1052 * register_pernet_device - register a network namespace device
1053 * @ops: pernet operations structure for the subsystem
1054 *
1055 * Register a device which has init and exit functions
1056 * that are called when network namespaces are created and
1057 * destroyed respectively.
1058 *
1059 * When registered all network namespace init functions are
1060 * called for every existing network namespace. Allowing kernel
1061 * modules to have a race free view of the set of network namespaces.
1062 *
1063 * When a new network namespace is created all of the init
1064 * methods are called in the order in which they were registered.
1065 *
1066 * When a network namespace is destroyed all of the exit methods
1067 * are called in the reverse of the order with which they were
1068 * registered.
1069 */
1070int register_pernet_device(struct pernet_operations *ops)
1071{
1072 int error;
1073 down_write(&pernet_ops_rwsem);
1074 error = register_pernet_operations(&pernet_list, ops);
1075 if (!error && (first_device == &pernet_list))
1076 first_device = &ops->list;
1077 up_write(&pernet_ops_rwsem);
1078 return error;
1079}
1080EXPORT_SYMBOL_GPL(register_pernet_device);
1081
1082/**
1083 * unregister_pernet_device - unregister a network namespace netdevice
1084 * @ops: pernet operations structure to manipulate
1085 *
1086 * Remove the pernet operations structure from the list to be
1087 * used when network namespaces are created or destroyed. In
1088 * addition run the exit method for all existing network
1089 * namespaces.
1090 */
1091void unregister_pernet_device(struct pernet_operations *ops)
1092{
1093 down_write(&pernet_ops_rwsem);
1094 if (&ops->list == first_device)
1095 first_device = first_device->next;
1096 unregister_pernet_operations(ops);
1097 up_write(&pernet_ops_rwsem);
1098}
1099EXPORT_SYMBOL_GPL(unregister_pernet_device);
1100
1101#ifdef CONFIG_NET_NS
1102static struct ns_common *netns_get(struct task_struct *task)
1103{
1104 struct net *net = NULL;
1105 struct nsproxy *nsproxy;
1106
1107 task_lock(task);
1108 nsproxy = task->nsproxy;
1109 if (nsproxy)
1110 net = get_net(nsproxy->net_ns);
1111 task_unlock(task);
1112
1113 return net ? &net->ns : NULL;
1114}
1115
1116static inline struct net *to_net_ns(struct ns_common *ns)
1117{
1118 return container_of(ns, struct net, ns);
1119}
1120
1121static void netns_put(struct ns_common *ns)
1122{
1123 put_net(to_net_ns(ns));
1124}
1125
1126static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
1127{
1128 struct net *net = to_net_ns(ns);
1129
1130 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1131 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1132 return -EPERM;
1133
1134 put_net(nsproxy->net_ns);
1135 nsproxy->net_ns = get_net(net);
1136 return 0;
1137}
1138
1139static struct user_namespace *netns_owner(struct ns_common *ns)
1140{
1141 return to_net_ns(ns)->user_ns;
1142}
1143
1144const struct proc_ns_operations netns_operations = {
1145 .name = "net",
1146 .type = CLONE_NEWNET,
1147 .get = netns_get,
1148 .put = netns_put,
1149 .install = netns_install,
1150 .owner = netns_owner,
1151};
1152#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/proc_fs.h>
14#include <linux/file.h>
15#include <linux/export.h>
16#include <net/net_namespace.h>
17#include <net/netns/generic.h>
18
19/*
20 * Our network namespace constructor/destructor lists
21 */
22
23static LIST_HEAD(pernet_list);
24static struct list_head *first_device = &pernet_list;
25static DEFINE_MUTEX(net_mutex);
26
27LIST_HEAD(net_namespace_list);
28EXPORT_SYMBOL_GPL(net_namespace_list);
29
30struct net init_net = {
31 .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head),
32};
33EXPORT_SYMBOL(init_net);
34
35#define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
36
37static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
38
39static struct net_generic *net_alloc_generic(void)
40{
41 struct net_generic *ng;
42 size_t generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
43
44 ng = kzalloc(generic_size, GFP_KERNEL);
45 if (ng)
46 ng->len = max_gen_ptrs;
47
48 return ng;
49}
50
51static int net_assign_generic(struct net *net, int id, void *data)
52{
53 struct net_generic *ng, *old_ng;
54
55 BUG_ON(!mutex_is_locked(&net_mutex));
56 BUG_ON(id == 0);
57
58 old_ng = rcu_dereference_protected(net->gen,
59 lockdep_is_held(&net_mutex));
60 ng = old_ng;
61 if (old_ng->len >= id)
62 goto assign;
63
64 ng = net_alloc_generic();
65 if (ng == NULL)
66 return -ENOMEM;
67
68 /*
69 * Some synchronisation notes:
70 *
71 * The net_generic explores the net->gen array inside rcu
72 * read section. Besides once set the net->gen->ptr[x]
73 * pointer never changes (see rules in netns/generic.h).
74 *
75 * That said, we simply duplicate this array and schedule
76 * the old copy for kfree after a grace period.
77 */
78
79 memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
80
81 rcu_assign_pointer(net->gen, ng);
82 kfree_rcu(old_ng, rcu);
83assign:
84 ng->ptr[id - 1] = data;
85 return 0;
86}
87
88static int ops_init(const struct pernet_operations *ops, struct net *net)
89{
90 int err = -ENOMEM;
91 void *data = NULL;
92
93 if (ops->id && ops->size) {
94 data = kzalloc(ops->size, GFP_KERNEL);
95 if (!data)
96 goto out;
97
98 err = net_assign_generic(net, *ops->id, data);
99 if (err)
100 goto cleanup;
101 }
102 err = 0;
103 if (ops->init)
104 err = ops->init(net);
105 if (!err)
106 return 0;
107
108cleanup:
109 kfree(data);
110
111out:
112 return err;
113}
114
115static void ops_free(const struct pernet_operations *ops, struct net *net)
116{
117 if (ops->id && ops->size) {
118 int id = *ops->id;
119 kfree(net_generic(net, id));
120 }
121}
122
123static void ops_exit_list(const struct pernet_operations *ops,
124 struct list_head *net_exit_list)
125{
126 struct net *net;
127 if (ops->exit) {
128 list_for_each_entry(net, net_exit_list, exit_list)
129 ops->exit(net);
130 }
131 if (ops->exit_batch)
132 ops->exit_batch(net_exit_list);
133}
134
135static void ops_free_list(const struct pernet_operations *ops,
136 struct list_head *net_exit_list)
137{
138 struct net *net;
139 if (ops->size && ops->id) {
140 list_for_each_entry(net, net_exit_list, exit_list)
141 ops_free(ops, net);
142 }
143}
144
145/*
146 * setup_net runs the initializers for the network namespace object.
147 */
148static __net_init int setup_net(struct net *net)
149{
150 /* Must be called with net_mutex held */
151 const struct pernet_operations *ops, *saved_ops;
152 int error = 0;
153 LIST_HEAD(net_exit_list);
154
155 atomic_set(&net->count, 1);
156 atomic_set(&net->passive, 1);
157 net->dev_base_seq = 1;
158
159#ifdef NETNS_REFCNT_DEBUG
160 atomic_set(&net->use_count, 0);
161#endif
162
163 list_for_each_entry(ops, &pernet_list, list) {
164 error = ops_init(ops, net);
165 if (error < 0)
166 goto out_undo;
167 }
168out:
169 return error;
170
171out_undo:
172 /* Walk through the list backwards calling the exit functions
173 * for the pernet modules whose init functions did not fail.
174 */
175 list_add(&net->exit_list, &net_exit_list);
176 saved_ops = ops;
177 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
178 ops_exit_list(ops, &net_exit_list);
179
180 ops = saved_ops;
181 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
182 ops_free_list(ops, &net_exit_list);
183
184 rcu_barrier();
185 goto out;
186}
187
188
189#ifdef CONFIG_NET_NS
190static struct kmem_cache *net_cachep;
191static struct workqueue_struct *netns_wq;
192
193static struct net *net_alloc(void)
194{
195 struct net *net = NULL;
196 struct net_generic *ng;
197
198 ng = net_alloc_generic();
199 if (!ng)
200 goto out;
201
202 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
203 if (!net)
204 goto out_free;
205
206 rcu_assign_pointer(net->gen, ng);
207out:
208 return net;
209
210out_free:
211 kfree(ng);
212 goto out;
213}
214
215static void net_free(struct net *net)
216{
217#ifdef NETNS_REFCNT_DEBUG
218 if (unlikely(atomic_read(&net->use_count) != 0)) {
219 pr_emerg("network namespace not free! Usage: %d\n",
220 atomic_read(&net->use_count));
221 return;
222 }
223#endif
224 kfree(net->gen);
225 kmem_cache_free(net_cachep, net);
226}
227
228void net_drop_ns(void *p)
229{
230 struct net *ns = p;
231 if (ns && atomic_dec_and_test(&ns->passive))
232 net_free(ns);
233}
234
235struct net *copy_net_ns(unsigned long flags, struct net *old_net)
236{
237 struct net *net;
238 int rv;
239
240 if (!(flags & CLONE_NEWNET))
241 return get_net(old_net);
242
243 net = net_alloc();
244 if (!net)
245 return ERR_PTR(-ENOMEM);
246 mutex_lock(&net_mutex);
247 rv = setup_net(net);
248 if (rv == 0) {
249 rtnl_lock();
250 list_add_tail_rcu(&net->list, &net_namespace_list);
251 rtnl_unlock();
252 }
253 mutex_unlock(&net_mutex);
254 if (rv < 0) {
255 net_drop_ns(net);
256 return ERR_PTR(rv);
257 }
258 return net;
259}
260
261static DEFINE_SPINLOCK(cleanup_list_lock);
262static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
263
264static void cleanup_net(struct work_struct *work)
265{
266 const struct pernet_operations *ops;
267 struct net *net, *tmp;
268 LIST_HEAD(net_kill_list);
269 LIST_HEAD(net_exit_list);
270
271 /* Atomically snapshot the list of namespaces to cleanup */
272 spin_lock_irq(&cleanup_list_lock);
273 list_replace_init(&cleanup_list, &net_kill_list);
274 spin_unlock_irq(&cleanup_list_lock);
275
276 mutex_lock(&net_mutex);
277
278 /* Don't let anyone else find us. */
279 rtnl_lock();
280 list_for_each_entry(net, &net_kill_list, cleanup_list) {
281 list_del_rcu(&net->list);
282 list_add_tail(&net->exit_list, &net_exit_list);
283 }
284 rtnl_unlock();
285
286 /*
287 * Another CPU might be rcu-iterating the list, wait for it.
288 * This needs to be before calling the exit() notifiers, so
289 * the rcu_barrier() below isn't sufficient alone.
290 */
291 synchronize_rcu();
292
293 /* Run all of the network namespace exit methods */
294 list_for_each_entry_reverse(ops, &pernet_list, list)
295 ops_exit_list(ops, &net_exit_list);
296
297 /* Free the net generic variables */
298 list_for_each_entry_reverse(ops, &pernet_list, list)
299 ops_free_list(ops, &net_exit_list);
300
301 mutex_unlock(&net_mutex);
302
303 /* Ensure there are no outstanding rcu callbacks using this
304 * network namespace.
305 */
306 rcu_barrier();
307
308 /* Finally it is safe to free my network namespace structure */
309 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
310 list_del_init(&net->exit_list);
311 net_drop_ns(net);
312 }
313}
314static DECLARE_WORK(net_cleanup_work, cleanup_net);
315
316void __put_net(struct net *net)
317{
318 /* Cleanup the network namespace in process context */
319 unsigned long flags;
320
321 spin_lock_irqsave(&cleanup_list_lock, flags);
322 list_add(&net->cleanup_list, &cleanup_list);
323 spin_unlock_irqrestore(&cleanup_list_lock, flags);
324
325 queue_work(netns_wq, &net_cleanup_work);
326}
327EXPORT_SYMBOL_GPL(__put_net);
328
329struct net *get_net_ns_by_fd(int fd)
330{
331 struct proc_inode *ei;
332 struct file *file;
333 struct net *net;
334
335 file = proc_ns_fget(fd);
336 if (IS_ERR(file))
337 return ERR_CAST(file);
338
339 ei = PROC_I(file->f_dentry->d_inode);
340 if (ei->ns_ops == &netns_operations)
341 net = get_net(ei->ns);
342 else
343 net = ERR_PTR(-EINVAL);
344
345 fput(file);
346 return net;
347}
348
349#else
350struct net *copy_net_ns(unsigned long flags, struct net *old_net)
351{
352 if (flags & CLONE_NEWNET)
353 return ERR_PTR(-EINVAL);
354 return old_net;
355}
356
357struct net *get_net_ns_by_fd(int fd)
358{
359 return ERR_PTR(-EINVAL);
360}
361#endif
362
363struct net *get_net_ns_by_pid(pid_t pid)
364{
365 struct task_struct *tsk;
366 struct net *net;
367
368 /* Lookup the network namespace */
369 net = ERR_PTR(-ESRCH);
370 rcu_read_lock();
371 tsk = find_task_by_vpid(pid);
372 if (tsk) {
373 struct nsproxy *nsproxy;
374 nsproxy = task_nsproxy(tsk);
375 if (nsproxy)
376 net = get_net(nsproxy->net_ns);
377 }
378 rcu_read_unlock();
379 return net;
380}
381EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
382
383static int __init net_ns_init(void)
384{
385 struct net_generic *ng;
386
387#ifdef CONFIG_NET_NS
388 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
389 SMP_CACHE_BYTES,
390 SLAB_PANIC, NULL);
391
392 /* Create workqueue for cleanup */
393 netns_wq = create_singlethread_workqueue("netns");
394 if (!netns_wq)
395 panic("Could not create netns workq");
396#endif
397
398 ng = net_alloc_generic();
399 if (!ng)
400 panic("Could not allocate generic netns");
401
402 rcu_assign_pointer(init_net.gen, ng);
403
404 mutex_lock(&net_mutex);
405 if (setup_net(&init_net))
406 panic("Could not setup the initial network namespace");
407
408 rtnl_lock();
409 list_add_tail_rcu(&init_net.list, &net_namespace_list);
410 rtnl_unlock();
411
412 mutex_unlock(&net_mutex);
413
414 return 0;
415}
416
417pure_initcall(net_ns_init);
418
419#ifdef CONFIG_NET_NS
420static int __register_pernet_operations(struct list_head *list,
421 struct pernet_operations *ops)
422{
423 struct net *net;
424 int error;
425 LIST_HEAD(net_exit_list);
426
427 list_add_tail(&ops->list, list);
428 if (ops->init || (ops->id && ops->size)) {
429 for_each_net(net) {
430 error = ops_init(ops, net);
431 if (error)
432 goto out_undo;
433 list_add_tail(&net->exit_list, &net_exit_list);
434 }
435 }
436 return 0;
437
438out_undo:
439 /* If I have an error cleanup all namespaces I initialized */
440 list_del(&ops->list);
441 ops_exit_list(ops, &net_exit_list);
442 ops_free_list(ops, &net_exit_list);
443 return error;
444}
445
446static void __unregister_pernet_operations(struct pernet_operations *ops)
447{
448 struct net *net;
449 LIST_HEAD(net_exit_list);
450
451 list_del(&ops->list);
452 for_each_net(net)
453 list_add_tail(&net->exit_list, &net_exit_list);
454 ops_exit_list(ops, &net_exit_list);
455 ops_free_list(ops, &net_exit_list);
456}
457
458#else
459
460static int __register_pernet_operations(struct list_head *list,
461 struct pernet_operations *ops)
462{
463 return ops_init(ops, &init_net);
464}
465
466static void __unregister_pernet_operations(struct pernet_operations *ops)
467{
468 LIST_HEAD(net_exit_list);
469 list_add(&init_net.exit_list, &net_exit_list);
470 ops_exit_list(ops, &net_exit_list);
471 ops_free_list(ops, &net_exit_list);
472}
473
474#endif /* CONFIG_NET_NS */
475
476static DEFINE_IDA(net_generic_ids);
477
478static int register_pernet_operations(struct list_head *list,
479 struct pernet_operations *ops)
480{
481 int error;
482
483 if (ops->id) {
484again:
485 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
486 if (error < 0) {
487 if (error == -EAGAIN) {
488 ida_pre_get(&net_generic_ids, GFP_KERNEL);
489 goto again;
490 }
491 return error;
492 }
493 max_gen_ptrs = max_t(unsigned int, max_gen_ptrs, *ops->id);
494 }
495 error = __register_pernet_operations(list, ops);
496 if (error) {
497 rcu_barrier();
498 if (ops->id)
499 ida_remove(&net_generic_ids, *ops->id);
500 }
501
502 return error;
503}
504
505static void unregister_pernet_operations(struct pernet_operations *ops)
506{
507
508 __unregister_pernet_operations(ops);
509 rcu_barrier();
510 if (ops->id)
511 ida_remove(&net_generic_ids, *ops->id);
512}
513
514/**
515 * register_pernet_subsys - register a network namespace subsystem
516 * @ops: pernet operations structure for the subsystem
517 *
518 * Register a subsystem which has init and exit functions
519 * that are called when network namespaces are created and
520 * destroyed respectively.
521 *
522 * When registered all network namespace init functions are
523 * called for every existing network namespace. Allowing kernel
524 * modules to have a race free view of the set of network namespaces.
525 *
526 * When a new network namespace is created all of the init
527 * methods are called in the order in which they were registered.
528 *
529 * When a network namespace is destroyed all of the exit methods
530 * are called in the reverse of the order with which they were
531 * registered.
532 */
533int register_pernet_subsys(struct pernet_operations *ops)
534{
535 int error;
536 mutex_lock(&net_mutex);
537 error = register_pernet_operations(first_device, ops);
538 mutex_unlock(&net_mutex);
539 return error;
540}
541EXPORT_SYMBOL_GPL(register_pernet_subsys);
542
543/**
544 * unregister_pernet_subsys - unregister a network namespace subsystem
545 * @ops: pernet operations structure to manipulate
546 *
547 * Remove the pernet operations structure from the list to be
548 * used when network namespaces are created or destroyed. In
549 * addition run the exit method for all existing network
550 * namespaces.
551 */
552void unregister_pernet_subsys(struct pernet_operations *ops)
553{
554 mutex_lock(&net_mutex);
555 unregister_pernet_operations(ops);
556 mutex_unlock(&net_mutex);
557}
558EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
559
560/**
561 * register_pernet_device - register a network namespace device
562 * @ops: pernet operations structure for the subsystem
563 *
564 * Register a device which has init and exit functions
565 * that are called when network namespaces are created and
566 * destroyed respectively.
567 *
568 * When registered all network namespace init functions are
569 * called for every existing network namespace. Allowing kernel
570 * modules to have a race free view of the set of network namespaces.
571 *
572 * When a new network namespace is created all of the init
573 * methods are called in the order in which they were registered.
574 *
575 * When a network namespace is destroyed all of the exit methods
576 * are called in the reverse of the order with which they were
577 * registered.
578 */
579int register_pernet_device(struct pernet_operations *ops)
580{
581 int error;
582 mutex_lock(&net_mutex);
583 error = register_pernet_operations(&pernet_list, ops);
584 if (!error && (first_device == &pernet_list))
585 first_device = &ops->list;
586 mutex_unlock(&net_mutex);
587 return error;
588}
589EXPORT_SYMBOL_GPL(register_pernet_device);
590
591/**
592 * unregister_pernet_device - unregister a network namespace netdevice
593 * @ops: pernet operations structure to manipulate
594 *
595 * Remove the pernet operations structure from the list to be
596 * used when network namespaces are created or destroyed. In
597 * addition run the exit method for all existing network
598 * namespaces.
599 */
600void unregister_pernet_device(struct pernet_operations *ops)
601{
602 mutex_lock(&net_mutex);
603 if (&ops->list == first_device)
604 first_device = first_device->next;
605 unregister_pernet_operations(ops);
606 mutex_unlock(&net_mutex);
607}
608EXPORT_SYMBOL_GPL(unregister_pernet_device);
609
610#ifdef CONFIG_NET_NS
611static void *netns_get(struct task_struct *task)
612{
613 struct net *net = NULL;
614 struct nsproxy *nsproxy;
615
616 rcu_read_lock();
617 nsproxy = task_nsproxy(task);
618 if (nsproxy)
619 net = get_net(nsproxy->net_ns);
620 rcu_read_unlock();
621
622 return net;
623}
624
625static void netns_put(void *ns)
626{
627 put_net(ns);
628}
629
630static int netns_install(struct nsproxy *nsproxy, void *ns)
631{
632 put_net(nsproxy->net_ns);
633 nsproxy->net_ns = get_net(ns);
634 return 0;
635}
636
637const struct proc_ns_operations netns_operations = {
638 .name = "net",
639 .type = CLONE_NEWNET,
640 .get = netns_get,
641 .put = netns_put,
642 .install = netns_install,
643};
644#endif