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/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