Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
3
4#include <linux/workqueue.h>
5#include <linux/rtnetlink.h>
6#include <linux/cache.h>
7#include <linux/slab.h>
8#include <linux/list.h>
9#include <linux/delay.h>
10#include <linux/sched.h>
11#include <linux/idr.h>
12#include <linux/rculist.h>
13#include <linux/nsproxy.h>
14#include <linux/fs.h>
15#include <linux/proc_ns.h>
16#include <linux/file.h>
17#include <linux/export.h>
18#include <linux/user_namespace.h>
19#include <linux/net_namespace.h>
20#include <linux/sched/task.h>
21#include <linux/uidgid.h>
22#include <linux/cookie.h>
23#include <linux/proc_fs.h>
24
25#include <net/sock.h>
26#include <net/netlink.h>
27#include <net/net_namespace.h>
28#include <net/netns/generic.h>
29
30/*
31 * Our network namespace constructor/destructor lists
32 */
33
34static LIST_HEAD(pernet_list);
35static struct list_head *first_device = &pernet_list;
36
37LIST_HEAD(net_namespace_list);
38EXPORT_SYMBOL_GPL(net_namespace_list);
39
40/* Protects net_namespace_list. Nests iside rtnl_lock() */
41DECLARE_RWSEM(net_rwsem);
42EXPORT_SYMBOL_GPL(net_rwsem);
43
44#ifdef CONFIG_KEYS
45static struct key_tag init_net_key_domain = { .usage = REFCOUNT_INIT(1) };
46#endif
47
48struct net init_net;
49EXPORT_SYMBOL(init_net);
50
51static bool init_net_initialized;
52/*
53 * pernet_ops_rwsem: protects: pernet_list, net_generic_ids,
54 * init_net_initialized and first_device pointer.
55 * This is internal net namespace object. Please, don't use it
56 * outside.
57 */
58DECLARE_RWSEM(pernet_ops_rwsem);
59
60#define MIN_PERNET_OPS_ID \
61 ((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *))
62
63#define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
64
65static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
66
67DEFINE_COOKIE(net_cookie);
68
69static struct net_generic *net_alloc_generic(void)
70{
71 unsigned int gen_ptrs = READ_ONCE(max_gen_ptrs);
72 unsigned int generic_size;
73 struct net_generic *ng;
74
75 generic_size = offsetof(struct net_generic, ptr[gen_ptrs]);
76
77 ng = kzalloc(generic_size, GFP_KERNEL);
78 if (ng)
79 ng->s.len = gen_ptrs;
80
81 return ng;
82}
83
84static int net_assign_generic(struct net *net, unsigned int id, void *data)
85{
86 struct net_generic *ng, *old_ng;
87
88 BUG_ON(id < MIN_PERNET_OPS_ID);
89
90 old_ng = rcu_dereference_protected(net->gen,
91 lockdep_is_held(&pernet_ops_rwsem));
92 if (old_ng->s.len > id) {
93 old_ng->ptr[id] = data;
94 return 0;
95 }
96
97 ng = net_alloc_generic();
98 if (!ng)
99 return -ENOMEM;
100
101 /*
102 * Some synchronisation notes:
103 *
104 * The net_generic explores the net->gen array inside rcu
105 * read section. Besides once set the net->gen->ptr[x]
106 * pointer never changes (see rules in netns/generic.h).
107 *
108 * That said, we simply duplicate this array and schedule
109 * the old copy for kfree after a grace period.
110 */
111
112 memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
113 (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
114 ng->ptr[id] = data;
115
116 rcu_assign_pointer(net->gen, ng);
117 kfree_rcu(old_ng, s.rcu);
118 return 0;
119}
120
121static int ops_init(const struct pernet_operations *ops, struct net *net)
122{
123 struct net_generic *ng;
124 int err = -ENOMEM;
125 void *data = NULL;
126
127 if (ops->id) {
128 data = kzalloc(ops->size, GFP_KERNEL);
129 if (!data)
130 goto out;
131
132 err = net_assign_generic(net, *ops->id, data);
133 if (err)
134 goto cleanup;
135 }
136 err = 0;
137 if (ops->init)
138 err = ops->init(net);
139 if (!err)
140 return 0;
141
142 if (ops->id) {
143 ng = rcu_dereference_protected(net->gen,
144 lockdep_is_held(&pernet_ops_rwsem));
145 ng->ptr[*ops->id] = NULL;
146 }
147
148cleanup:
149 kfree(data);
150
151out:
152 return err;
153}
154
155static void ops_pre_exit_list(const struct pernet_operations *ops,
156 struct list_head *net_exit_list)
157{
158 struct net *net;
159
160 if (ops->pre_exit) {
161 list_for_each_entry(net, net_exit_list, exit_list)
162 ops->pre_exit(net);
163 }
164}
165
166static void ops_exit_list(const struct pernet_operations *ops,
167 struct list_head *net_exit_list)
168{
169 struct net *net;
170 if (ops->exit) {
171 list_for_each_entry(net, net_exit_list, exit_list) {
172 ops->exit(net);
173 cond_resched();
174 }
175 }
176 if (ops->exit_batch)
177 ops->exit_batch(net_exit_list);
178}
179
180static void ops_free_list(const struct pernet_operations *ops,
181 struct list_head *net_exit_list)
182{
183 struct net *net;
184
185 if (ops->id) {
186 list_for_each_entry(net, net_exit_list, exit_list)
187 kfree(net_generic(net, *ops->id));
188 }
189}
190
191/* should be called with nsid_lock held */
192static int alloc_netid(struct net *net, struct net *peer, int reqid)
193{
194 int min = 0, max = 0;
195
196 if (reqid >= 0) {
197 min = reqid;
198 max = reqid + 1;
199 }
200
201 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
202}
203
204/* This function is used by idr_for_each(). If net is equal to peer, the
205 * function returns the id so that idr_for_each() stops. Because we cannot
206 * returns the id 0 (idr_for_each() will not stop), we return the magic value
207 * NET_ID_ZERO (-1) for it.
208 */
209#define NET_ID_ZERO -1
210static int net_eq_idr(int id, void *net, void *peer)
211{
212 if (net_eq(net, peer))
213 return id ? : NET_ID_ZERO;
214 return 0;
215}
216
217/* Must be called from RCU-critical section or with nsid_lock held */
218static int __peernet2id(const struct net *net, struct net *peer)
219{
220 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
221
222 /* Magic value for id 0. */
223 if (id == NET_ID_ZERO)
224 return 0;
225 if (id > 0)
226 return id;
227
228 return NETNSA_NSID_NOT_ASSIGNED;
229}
230
231static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
232 struct nlmsghdr *nlh, gfp_t gfp);
233/* This function returns the id of a peer netns. If no id is assigned, one will
234 * be allocated and returned.
235 */
236int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp)
237{
238 int id;
239
240 if (refcount_read(&net->ns.count) == 0)
241 return NETNSA_NSID_NOT_ASSIGNED;
242
243 spin_lock_bh(&net->nsid_lock);
244 id = __peernet2id(net, peer);
245 if (id >= 0) {
246 spin_unlock_bh(&net->nsid_lock);
247 return id;
248 }
249
250 /* When peer is obtained from RCU lists, we may race with
251 * its cleanup. Check whether it's alive, and this guarantees
252 * we never hash a peer back to net->netns_ids, after it has
253 * just been idr_remove()'d from there in cleanup_net().
254 */
255 if (!maybe_get_net(peer)) {
256 spin_unlock_bh(&net->nsid_lock);
257 return NETNSA_NSID_NOT_ASSIGNED;
258 }
259
260 id = alloc_netid(net, peer, -1);
261 spin_unlock_bh(&net->nsid_lock);
262
263 put_net(peer);
264 if (id < 0)
265 return NETNSA_NSID_NOT_ASSIGNED;
266
267 rtnl_net_notifyid(net, RTM_NEWNSID, id, 0, NULL, gfp);
268
269 return id;
270}
271EXPORT_SYMBOL_GPL(peernet2id_alloc);
272
273/* This function returns, if assigned, the id of a peer netns. */
274int peernet2id(const struct net *net, struct net *peer)
275{
276 int id;
277
278 rcu_read_lock();
279 id = __peernet2id(net, peer);
280 rcu_read_unlock();
281
282 return id;
283}
284EXPORT_SYMBOL(peernet2id);
285
286/* This function returns true is the peer netns has an id assigned into the
287 * current netns.
288 */
289bool peernet_has_id(const struct net *net, struct net *peer)
290{
291 return peernet2id(net, peer) >= 0;
292}
293
294struct net *get_net_ns_by_id(const struct net *net, int id)
295{
296 struct net *peer;
297
298 if (id < 0)
299 return NULL;
300
301 rcu_read_lock();
302 peer = idr_find(&net->netns_ids, id);
303 if (peer)
304 peer = maybe_get_net(peer);
305 rcu_read_unlock();
306
307 return peer;
308}
309EXPORT_SYMBOL_GPL(get_net_ns_by_id);
310
311static __net_init void preinit_net_sysctl(struct net *net)
312{
313 net->core.sysctl_somaxconn = SOMAXCONN;
314 /* Limits per socket sk_omem_alloc usage.
315 * TCP zerocopy regular usage needs 128 KB.
316 */
317 net->core.sysctl_optmem_max = 128 * 1024;
318 net->core.sysctl_txrehash = SOCK_TXREHASH_ENABLED;
319 net->core.sysctl_tstamp_allow_data = 1;
320}
321
322/* init code that must occur even if setup_net() is not called. */
323static __net_init void preinit_net(struct net *net, struct user_namespace *user_ns)
324{
325 refcount_set(&net->passive, 1);
326 refcount_set(&net->ns.count, 1);
327 ref_tracker_dir_init(&net->refcnt_tracker, 128, "net refcnt");
328 ref_tracker_dir_init(&net->notrefcnt_tracker, 128, "net notrefcnt");
329
330 get_random_bytes(&net->hash_mix, sizeof(u32));
331 net->dev_base_seq = 1;
332 net->user_ns = user_ns;
333
334 idr_init(&net->netns_ids);
335 spin_lock_init(&net->nsid_lock);
336 mutex_init(&net->ipv4.ra_mutex);
337
338#ifdef CONFIG_DEBUG_NET_SMALL_RTNL
339 mutex_init(&net->rtnl_mutex);
340 lock_set_cmp_fn(&net->rtnl_mutex, rtnl_net_lock_cmp_fn, NULL);
341#endif
342
343 preinit_net_sysctl(net);
344}
345
346/*
347 * setup_net runs the initializers for the network namespace object.
348 */
349static __net_init int setup_net(struct net *net)
350{
351 /* Must be called with pernet_ops_rwsem held */
352 const struct pernet_operations *ops, *saved_ops;
353 LIST_HEAD(net_exit_list);
354 LIST_HEAD(dev_kill_list);
355 int error = 0;
356
357 preempt_disable();
358 net->net_cookie = gen_cookie_next(&net_cookie);
359 preempt_enable();
360
361 list_for_each_entry(ops, &pernet_list, list) {
362 error = ops_init(ops, net);
363 if (error < 0)
364 goto out_undo;
365 }
366 down_write(&net_rwsem);
367 list_add_tail_rcu(&net->list, &net_namespace_list);
368 up_write(&net_rwsem);
369out:
370 return error;
371
372out_undo:
373 /* Walk through the list backwards calling the exit functions
374 * for the pernet modules whose init functions did not fail.
375 */
376 list_add(&net->exit_list, &net_exit_list);
377 saved_ops = ops;
378 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
379 ops_pre_exit_list(ops, &net_exit_list);
380
381 synchronize_rcu();
382
383 ops = saved_ops;
384 rtnl_lock();
385 list_for_each_entry_continue_reverse(ops, &pernet_list, list) {
386 if (ops->exit_batch_rtnl)
387 ops->exit_batch_rtnl(&net_exit_list, &dev_kill_list);
388 }
389 unregister_netdevice_many(&dev_kill_list);
390 rtnl_unlock();
391
392 ops = saved_ops;
393 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
394 ops_exit_list(ops, &net_exit_list);
395
396 ops = saved_ops;
397 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
398 ops_free_list(ops, &net_exit_list);
399
400 rcu_barrier();
401 goto out;
402}
403
404#ifdef CONFIG_NET_NS
405static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
406{
407 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
408}
409
410static void dec_net_namespaces(struct ucounts *ucounts)
411{
412 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
413}
414
415static struct kmem_cache *net_cachep __ro_after_init;
416static struct workqueue_struct *netns_wq;
417
418static struct net *net_alloc(void)
419{
420 struct net *net = NULL;
421 struct net_generic *ng;
422
423 ng = net_alloc_generic();
424 if (!ng)
425 goto out;
426
427 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
428 if (!net)
429 goto out_free;
430
431#ifdef CONFIG_KEYS
432 net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
433 if (!net->key_domain)
434 goto out_free_2;
435 refcount_set(&net->key_domain->usage, 1);
436#endif
437
438 rcu_assign_pointer(net->gen, ng);
439out:
440 return net;
441
442#ifdef CONFIG_KEYS
443out_free_2:
444 kmem_cache_free(net_cachep, net);
445 net = NULL;
446#endif
447out_free:
448 kfree(ng);
449 goto out;
450}
451
452static LLIST_HEAD(defer_free_list);
453
454static void net_complete_free(void)
455{
456 struct llist_node *kill_list;
457 struct net *net, *next;
458
459 /* Get the list of namespaces to free from last round. */
460 kill_list = llist_del_all(&defer_free_list);
461
462 llist_for_each_entry_safe(net, next, kill_list, defer_free_list)
463 kmem_cache_free(net_cachep, net);
464
465}
466
467void net_passive_dec(struct net *net)
468{
469 if (refcount_dec_and_test(&net->passive)) {
470 kfree(rcu_access_pointer(net->gen));
471
472 /* There should not be any trackers left there. */
473 ref_tracker_dir_exit(&net->notrefcnt_tracker);
474
475 /* Wait for an extra rcu_barrier() before final free. */
476 llist_add(&net->defer_free_list, &defer_free_list);
477 }
478}
479
480void net_drop_ns(void *p)
481{
482 struct net *net = (struct net *)p;
483
484 if (net)
485 net_passive_dec(net);
486}
487
488struct net *copy_net_ns(unsigned long flags,
489 struct user_namespace *user_ns, struct net *old_net)
490{
491 struct ucounts *ucounts;
492 struct net *net;
493 int rv;
494
495 if (!(flags & CLONE_NEWNET))
496 return get_net(old_net);
497
498 ucounts = inc_net_namespaces(user_ns);
499 if (!ucounts)
500 return ERR_PTR(-ENOSPC);
501
502 net = net_alloc();
503 if (!net) {
504 rv = -ENOMEM;
505 goto dec_ucounts;
506 }
507
508 preinit_net(net, user_ns);
509 net->ucounts = ucounts;
510 get_user_ns(user_ns);
511
512 rv = down_read_killable(&pernet_ops_rwsem);
513 if (rv < 0)
514 goto put_userns;
515
516 rv = setup_net(net);
517
518 up_read(&pernet_ops_rwsem);
519
520 if (rv < 0) {
521put_userns:
522#ifdef CONFIG_KEYS
523 key_remove_domain(net->key_domain);
524#endif
525 put_user_ns(user_ns);
526 net_passive_dec(net);
527dec_ucounts:
528 dec_net_namespaces(ucounts);
529 return ERR_PTR(rv);
530 }
531 return net;
532}
533
534/**
535 * net_ns_get_ownership - get sysfs ownership data for @net
536 * @net: network namespace in question (can be NULL)
537 * @uid: kernel user ID for sysfs objects
538 * @gid: kernel group ID for sysfs objects
539 *
540 * Returns the uid/gid pair of root in the user namespace associated with the
541 * given network namespace.
542 */
543void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
544{
545 if (net) {
546 kuid_t ns_root_uid = make_kuid(net->user_ns, 0);
547 kgid_t ns_root_gid = make_kgid(net->user_ns, 0);
548
549 if (uid_valid(ns_root_uid))
550 *uid = ns_root_uid;
551
552 if (gid_valid(ns_root_gid))
553 *gid = ns_root_gid;
554 } else {
555 *uid = GLOBAL_ROOT_UID;
556 *gid = GLOBAL_ROOT_GID;
557 }
558}
559EXPORT_SYMBOL_GPL(net_ns_get_ownership);
560
561static void unhash_nsid(struct net *net, struct net *last)
562{
563 struct net *tmp;
564 /* This function is only called from cleanup_net() work,
565 * and this work is the only process, that may delete
566 * a net from net_namespace_list. So, when the below
567 * is executing, the list may only grow. Thus, we do not
568 * use for_each_net_rcu() or net_rwsem.
569 */
570 for_each_net(tmp) {
571 int id;
572
573 spin_lock_bh(&tmp->nsid_lock);
574 id = __peernet2id(tmp, net);
575 if (id >= 0)
576 idr_remove(&tmp->netns_ids, id);
577 spin_unlock_bh(&tmp->nsid_lock);
578 if (id >= 0)
579 rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL,
580 GFP_KERNEL);
581 if (tmp == last)
582 break;
583 }
584 spin_lock_bh(&net->nsid_lock);
585 idr_destroy(&net->netns_ids);
586 spin_unlock_bh(&net->nsid_lock);
587}
588
589static LLIST_HEAD(cleanup_list);
590
591static void cleanup_net(struct work_struct *work)
592{
593 const struct pernet_operations *ops;
594 struct net *net, *tmp, *last;
595 struct llist_node *net_kill_list;
596 LIST_HEAD(net_exit_list);
597 LIST_HEAD(dev_kill_list);
598
599 /* Atomically snapshot the list of namespaces to cleanup */
600 net_kill_list = llist_del_all(&cleanup_list);
601
602 down_read(&pernet_ops_rwsem);
603
604 /* Don't let anyone else find us. */
605 down_write(&net_rwsem);
606 llist_for_each_entry(net, net_kill_list, cleanup_list)
607 list_del_rcu(&net->list);
608 /* Cache last net. After we unlock rtnl, no one new net
609 * added to net_namespace_list can assign nsid pointer
610 * to a net from net_kill_list (see peernet2id_alloc()).
611 * So, we skip them in unhash_nsid().
612 *
613 * Note, that unhash_nsid() does not delete nsid links
614 * between net_kill_list's nets, as they've already
615 * deleted from net_namespace_list. But, this would be
616 * useless anyway, as netns_ids are destroyed there.
617 */
618 last = list_last_entry(&net_namespace_list, struct net, list);
619 up_write(&net_rwsem);
620
621 llist_for_each_entry(net, net_kill_list, cleanup_list) {
622 unhash_nsid(net, last);
623 list_add_tail(&net->exit_list, &net_exit_list);
624 }
625
626 /* Run all of the network namespace pre_exit methods */
627 list_for_each_entry_reverse(ops, &pernet_list, list)
628 ops_pre_exit_list(ops, &net_exit_list);
629
630 /*
631 * Another CPU might be rcu-iterating the list, wait for it.
632 * This needs to be before calling the exit() notifiers, so
633 * the rcu_barrier() below isn't sufficient alone.
634 * Also the pre_exit() and exit() methods need this barrier.
635 */
636 synchronize_rcu_expedited();
637
638 rtnl_lock();
639 list_for_each_entry_reverse(ops, &pernet_list, list) {
640 if (ops->exit_batch_rtnl)
641 ops->exit_batch_rtnl(&net_exit_list, &dev_kill_list);
642 }
643 unregister_netdevice_many(&dev_kill_list);
644 rtnl_unlock();
645
646 /* Run all of the network namespace exit methods */
647 list_for_each_entry_reverse(ops, &pernet_list, list)
648 ops_exit_list(ops, &net_exit_list);
649
650 /* Free the net generic variables */
651 list_for_each_entry_reverse(ops, &pernet_list, list)
652 ops_free_list(ops, &net_exit_list);
653
654 up_read(&pernet_ops_rwsem);
655
656 /* Ensure there are no outstanding rcu callbacks using this
657 * network namespace.
658 */
659 rcu_barrier();
660
661 net_complete_free();
662
663 /* Finally it is safe to free my network namespace structure */
664 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
665 list_del_init(&net->exit_list);
666 dec_net_namespaces(net->ucounts);
667#ifdef CONFIG_KEYS
668 key_remove_domain(net->key_domain);
669#endif
670 put_user_ns(net->user_ns);
671 net_passive_dec(net);
672 }
673}
674
675/**
676 * net_ns_barrier - wait until concurrent net_cleanup_work is done
677 *
678 * cleanup_net runs from work queue and will first remove namespaces
679 * from the global list, then run net exit functions.
680 *
681 * Call this in module exit path to make sure that all netns
682 * ->exit ops have been invoked before the function is removed.
683 */
684void net_ns_barrier(void)
685{
686 down_write(&pernet_ops_rwsem);
687 up_write(&pernet_ops_rwsem);
688}
689EXPORT_SYMBOL(net_ns_barrier);
690
691static DECLARE_WORK(net_cleanup_work, cleanup_net);
692
693void __put_net(struct net *net)
694{
695 ref_tracker_dir_exit(&net->refcnt_tracker);
696 /* Cleanup the network namespace in process context */
697 if (llist_add(&net->cleanup_list, &cleanup_list))
698 queue_work(netns_wq, &net_cleanup_work);
699}
700EXPORT_SYMBOL_GPL(__put_net);
701
702/**
703 * get_net_ns - increment the refcount of the network namespace
704 * @ns: common namespace (net)
705 *
706 * Returns the net's common namespace or ERR_PTR() if ref is zero.
707 */
708struct ns_common *get_net_ns(struct ns_common *ns)
709{
710 struct net *net;
711
712 net = maybe_get_net(container_of(ns, struct net, ns));
713 if (net)
714 return &net->ns;
715 return ERR_PTR(-EINVAL);
716}
717EXPORT_SYMBOL_GPL(get_net_ns);
718
719struct net *get_net_ns_by_fd(int fd)
720{
721 CLASS(fd, f)(fd);
722
723 if (fd_empty(f))
724 return ERR_PTR(-EBADF);
725
726 if (proc_ns_file(fd_file(f))) {
727 struct ns_common *ns = get_proc_ns(file_inode(fd_file(f)));
728 if (ns->ops == &netns_operations)
729 return get_net(container_of(ns, struct net, ns));
730 }
731
732 return ERR_PTR(-EINVAL);
733}
734EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
735#endif
736
737struct net *get_net_ns_by_pid(pid_t pid)
738{
739 struct task_struct *tsk;
740 struct net *net;
741
742 /* Lookup the network namespace */
743 net = ERR_PTR(-ESRCH);
744 rcu_read_lock();
745 tsk = find_task_by_vpid(pid);
746 if (tsk) {
747 struct nsproxy *nsproxy;
748 task_lock(tsk);
749 nsproxy = tsk->nsproxy;
750 if (nsproxy)
751 net = get_net(nsproxy->net_ns);
752 task_unlock(tsk);
753 }
754 rcu_read_unlock();
755 return net;
756}
757EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
758
759static __net_init int net_ns_net_init(struct net *net)
760{
761#ifdef CONFIG_NET_NS
762 net->ns.ops = &netns_operations;
763#endif
764 return ns_alloc_inum(&net->ns);
765}
766
767static __net_exit void net_ns_net_exit(struct net *net)
768{
769 ns_free_inum(&net->ns);
770}
771
772static struct pernet_operations __net_initdata net_ns_ops = {
773 .init = net_ns_net_init,
774 .exit = net_ns_net_exit,
775};
776
777static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
778 [NETNSA_NONE] = { .type = NLA_UNSPEC },
779 [NETNSA_NSID] = { .type = NLA_S32 },
780 [NETNSA_PID] = { .type = NLA_U32 },
781 [NETNSA_FD] = { .type = NLA_U32 },
782 [NETNSA_TARGET_NSID] = { .type = NLA_S32 },
783};
784
785static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
786 struct netlink_ext_ack *extack)
787{
788 struct net *net = sock_net(skb->sk);
789 struct nlattr *tb[NETNSA_MAX + 1];
790 struct nlattr *nla;
791 struct net *peer;
792 int nsid, err;
793
794 err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
795 NETNSA_MAX, rtnl_net_policy, extack);
796 if (err < 0)
797 return err;
798 if (!tb[NETNSA_NSID]) {
799 NL_SET_ERR_MSG(extack, "nsid is missing");
800 return -EINVAL;
801 }
802 nsid = nla_get_s32(tb[NETNSA_NSID]);
803
804 if (tb[NETNSA_PID]) {
805 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
806 nla = tb[NETNSA_PID];
807 } else if (tb[NETNSA_FD]) {
808 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
809 nla = tb[NETNSA_FD];
810 } else {
811 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
812 return -EINVAL;
813 }
814 if (IS_ERR(peer)) {
815 NL_SET_BAD_ATTR(extack, nla);
816 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
817 return PTR_ERR(peer);
818 }
819
820 spin_lock_bh(&net->nsid_lock);
821 if (__peernet2id(net, peer) >= 0) {
822 spin_unlock_bh(&net->nsid_lock);
823 err = -EEXIST;
824 NL_SET_BAD_ATTR(extack, nla);
825 NL_SET_ERR_MSG(extack,
826 "Peer netns already has a nsid assigned");
827 goto out;
828 }
829
830 err = alloc_netid(net, peer, nsid);
831 spin_unlock_bh(&net->nsid_lock);
832 if (err >= 0) {
833 rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid,
834 nlh, GFP_KERNEL);
835 err = 0;
836 } else if (err == -ENOSPC && nsid >= 0) {
837 err = -EEXIST;
838 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
839 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
840 }
841out:
842 put_net(peer);
843 return err;
844}
845
846static int rtnl_net_get_size(void)
847{
848 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
849 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
850 + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
851 ;
852}
853
854struct net_fill_args {
855 u32 portid;
856 u32 seq;
857 int flags;
858 int cmd;
859 int nsid;
860 bool add_ref;
861 int ref_nsid;
862};
863
864static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
865{
866 struct nlmsghdr *nlh;
867 struct rtgenmsg *rth;
868
869 nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
870 args->flags);
871 if (!nlh)
872 return -EMSGSIZE;
873
874 rth = nlmsg_data(nlh);
875 rth->rtgen_family = AF_UNSPEC;
876
877 if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
878 goto nla_put_failure;
879
880 if (args->add_ref &&
881 nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
882 goto nla_put_failure;
883
884 nlmsg_end(skb, nlh);
885 return 0;
886
887nla_put_failure:
888 nlmsg_cancel(skb, nlh);
889 return -EMSGSIZE;
890}
891
892static int rtnl_net_valid_getid_req(struct sk_buff *skb,
893 const struct nlmsghdr *nlh,
894 struct nlattr **tb,
895 struct netlink_ext_ack *extack)
896{
897 int i, err;
898
899 if (!netlink_strict_get_check(skb))
900 return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
901 tb, NETNSA_MAX, rtnl_net_policy,
902 extack);
903
904 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
905 NETNSA_MAX, rtnl_net_policy,
906 extack);
907 if (err)
908 return err;
909
910 for (i = 0; i <= NETNSA_MAX; i++) {
911 if (!tb[i])
912 continue;
913
914 switch (i) {
915 case NETNSA_PID:
916 case NETNSA_FD:
917 case NETNSA_NSID:
918 case NETNSA_TARGET_NSID:
919 break;
920 default:
921 NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
922 return -EINVAL;
923 }
924 }
925
926 return 0;
927}
928
929static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
930 struct netlink_ext_ack *extack)
931{
932 struct net *net = sock_net(skb->sk);
933 struct nlattr *tb[NETNSA_MAX + 1];
934 struct net_fill_args fillargs = {
935 .portid = NETLINK_CB(skb).portid,
936 .seq = nlh->nlmsg_seq,
937 .cmd = RTM_NEWNSID,
938 };
939 struct net *peer, *target = net;
940 struct nlattr *nla;
941 struct sk_buff *msg;
942 int err;
943
944 err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
945 if (err < 0)
946 return err;
947 if (tb[NETNSA_PID]) {
948 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
949 nla = tb[NETNSA_PID];
950 } else if (tb[NETNSA_FD]) {
951 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
952 nla = tb[NETNSA_FD];
953 } else if (tb[NETNSA_NSID]) {
954 peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
955 if (!peer)
956 peer = ERR_PTR(-ENOENT);
957 nla = tb[NETNSA_NSID];
958 } else {
959 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
960 return -EINVAL;
961 }
962
963 if (IS_ERR(peer)) {
964 NL_SET_BAD_ATTR(extack, nla);
965 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
966 return PTR_ERR(peer);
967 }
968
969 if (tb[NETNSA_TARGET_NSID]) {
970 int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
971
972 target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
973 if (IS_ERR(target)) {
974 NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
975 NL_SET_ERR_MSG(extack,
976 "Target netns reference is invalid");
977 err = PTR_ERR(target);
978 goto out;
979 }
980 fillargs.add_ref = true;
981 fillargs.ref_nsid = peernet2id(net, peer);
982 }
983
984 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
985 if (!msg) {
986 err = -ENOMEM;
987 goto out;
988 }
989
990 fillargs.nsid = peernet2id(target, peer);
991 err = rtnl_net_fill(msg, &fillargs);
992 if (err < 0)
993 goto err_out;
994
995 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
996 goto out;
997
998err_out:
999 nlmsg_free(msg);
1000out:
1001 if (fillargs.add_ref)
1002 put_net(target);
1003 put_net(peer);
1004 return err;
1005}
1006
1007struct rtnl_net_dump_cb {
1008 struct net *tgt_net;
1009 struct net *ref_net;
1010 struct sk_buff *skb;
1011 struct net_fill_args fillargs;
1012 int idx;
1013 int s_idx;
1014};
1015
1016/* Runs in RCU-critical section. */
1017static int rtnl_net_dumpid_one(int id, void *peer, void *data)
1018{
1019 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
1020 int ret;
1021
1022 if (net_cb->idx < net_cb->s_idx)
1023 goto cont;
1024
1025 net_cb->fillargs.nsid = id;
1026 if (net_cb->fillargs.add_ref)
1027 net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
1028 ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
1029 if (ret < 0)
1030 return ret;
1031
1032cont:
1033 net_cb->idx++;
1034 return 0;
1035}
1036
1037static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
1038 struct rtnl_net_dump_cb *net_cb,
1039 struct netlink_callback *cb)
1040{
1041 struct netlink_ext_ack *extack = cb->extack;
1042 struct nlattr *tb[NETNSA_MAX + 1];
1043 int err, i;
1044
1045 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
1046 NETNSA_MAX, rtnl_net_policy,
1047 extack);
1048 if (err < 0)
1049 return err;
1050
1051 for (i = 0; i <= NETNSA_MAX; i++) {
1052 if (!tb[i])
1053 continue;
1054
1055 if (i == NETNSA_TARGET_NSID) {
1056 struct net *net;
1057
1058 net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
1059 if (IS_ERR(net)) {
1060 NL_SET_BAD_ATTR(extack, tb[i]);
1061 NL_SET_ERR_MSG(extack,
1062 "Invalid target network namespace id");
1063 return PTR_ERR(net);
1064 }
1065 net_cb->fillargs.add_ref = true;
1066 net_cb->ref_net = net_cb->tgt_net;
1067 net_cb->tgt_net = net;
1068 } else {
1069 NL_SET_BAD_ATTR(extack, tb[i]);
1070 NL_SET_ERR_MSG(extack,
1071 "Unsupported attribute in dump request");
1072 return -EINVAL;
1073 }
1074 }
1075
1076 return 0;
1077}
1078
1079static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1080{
1081 struct rtnl_net_dump_cb net_cb = {
1082 .tgt_net = sock_net(skb->sk),
1083 .skb = skb,
1084 .fillargs = {
1085 .portid = NETLINK_CB(cb->skb).portid,
1086 .seq = cb->nlh->nlmsg_seq,
1087 .flags = NLM_F_MULTI,
1088 .cmd = RTM_NEWNSID,
1089 },
1090 .idx = 0,
1091 .s_idx = cb->args[0],
1092 };
1093 int err = 0;
1094
1095 if (cb->strict_check) {
1096 err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1097 if (err < 0)
1098 goto end;
1099 }
1100
1101 rcu_read_lock();
1102 idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1103 rcu_read_unlock();
1104
1105 cb->args[0] = net_cb.idx;
1106end:
1107 if (net_cb.fillargs.add_ref)
1108 put_net(net_cb.tgt_net);
1109 return err;
1110}
1111
1112static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1113 struct nlmsghdr *nlh, gfp_t gfp)
1114{
1115 struct net_fill_args fillargs = {
1116 .portid = portid,
1117 .seq = nlh ? nlh->nlmsg_seq : 0,
1118 .cmd = cmd,
1119 .nsid = id,
1120 };
1121 struct sk_buff *msg;
1122 int err = -ENOMEM;
1123
1124 msg = nlmsg_new(rtnl_net_get_size(), gfp);
1125 if (!msg)
1126 goto out;
1127
1128 err = rtnl_net_fill(msg, &fillargs);
1129 if (err < 0)
1130 goto err_out;
1131
1132 rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, gfp);
1133 return;
1134
1135err_out:
1136 nlmsg_free(msg);
1137out:
1138 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1139}
1140
1141#ifdef CONFIG_NET_NS
1142static void __init netns_ipv4_struct_check(void)
1143{
1144 /* TX readonly hotpath cache lines */
1145 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1146 sysctl_tcp_early_retrans);
1147 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1148 sysctl_tcp_tso_win_divisor);
1149 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1150 sysctl_tcp_tso_rtt_log);
1151 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1152 sysctl_tcp_autocorking);
1153 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1154 sysctl_tcp_min_snd_mss);
1155 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1156 sysctl_tcp_notsent_lowat);
1157 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1158 sysctl_tcp_limit_output_bytes);
1159 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1160 sysctl_tcp_min_rtt_wlen);
1161 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1162 sysctl_tcp_wmem);
1163 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1164 sysctl_ip_fwd_use_pmtu);
1165 CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_tx, 33);
1166
1167 /* TXRX readonly hotpath cache lines */
1168 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_txrx,
1169 sysctl_tcp_moderate_rcvbuf);
1170 CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_txrx, 1);
1171
1172 /* RX readonly hotpath cache line */
1173 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1174 sysctl_ip_early_demux);
1175 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1176 sysctl_tcp_early_demux);
1177 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1178 sysctl_tcp_l3mdev_accept);
1179 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1180 sysctl_tcp_reordering);
1181 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1182 sysctl_tcp_rmem);
1183 CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_rx, 22);
1184}
1185#endif
1186
1187static const struct rtnl_msg_handler net_ns_rtnl_msg_handlers[] __initconst = {
1188 {.msgtype = RTM_NEWNSID, .doit = rtnl_net_newid,
1189 .flags = RTNL_FLAG_DOIT_UNLOCKED},
1190 {.msgtype = RTM_GETNSID, .doit = rtnl_net_getid,
1191 .dumpit = rtnl_net_dumpid,
1192 .flags = RTNL_FLAG_DOIT_UNLOCKED | RTNL_FLAG_DUMP_UNLOCKED},
1193};
1194
1195void __init net_ns_init(void)
1196{
1197 struct net_generic *ng;
1198
1199#ifdef CONFIG_NET_NS
1200 netns_ipv4_struct_check();
1201 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1202 SMP_CACHE_BYTES,
1203 SLAB_PANIC|SLAB_ACCOUNT, NULL);
1204
1205 /* Create workqueue for cleanup */
1206 netns_wq = create_singlethread_workqueue("netns");
1207 if (!netns_wq)
1208 panic("Could not create netns workq");
1209#endif
1210
1211 ng = net_alloc_generic();
1212 if (!ng)
1213 panic("Could not allocate generic netns");
1214
1215 rcu_assign_pointer(init_net.gen, ng);
1216
1217#ifdef CONFIG_KEYS
1218 init_net.key_domain = &init_net_key_domain;
1219#endif
1220 preinit_net(&init_net, &init_user_ns);
1221
1222 down_write(&pernet_ops_rwsem);
1223 if (setup_net(&init_net))
1224 panic("Could not setup the initial network namespace");
1225
1226 init_net_initialized = true;
1227 up_write(&pernet_ops_rwsem);
1228
1229 if (register_pernet_subsys(&net_ns_ops))
1230 panic("Could not register network namespace subsystems");
1231
1232 rtnl_register_many(net_ns_rtnl_msg_handlers);
1233}
1234
1235static void free_exit_list(struct pernet_operations *ops, struct list_head *net_exit_list)
1236{
1237 ops_pre_exit_list(ops, net_exit_list);
1238 synchronize_rcu();
1239
1240 if (ops->exit_batch_rtnl) {
1241 LIST_HEAD(dev_kill_list);
1242
1243 rtnl_lock();
1244 ops->exit_batch_rtnl(net_exit_list, &dev_kill_list);
1245 unregister_netdevice_many(&dev_kill_list);
1246 rtnl_unlock();
1247 }
1248 ops_exit_list(ops, net_exit_list);
1249
1250 ops_free_list(ops, net_exit_list);
1251}
1252
1253#ifdef CONFIG_NET_NS
1254static int __register_pernet_operations(struct list_head *list,
1255 struct pernet_operations *ops)
1256{
1257 struct net *net;
1258 int error;
1259 LIST_HEAD(net_exit_list);
1260
1261 list_add_tail(&ops->list, list);
1262 if (ops->init || ops->id) {
1263 /* We held write locked pernet_ops_rwsem, and parallel
1264 * setup_net() and cleanup_net() are not possible.
1265 */
1266 for_each_net(net) {
1267 error = ops_init(ops, net);
1268 if (error)
1269 goto out_undo;
1270 list_add_tail(&net->exit_list, &net_exit_list);
1271 }
1272 }
1273 return 0;
1274
1275out_undo:
1276 /* If I have an error cleanup all namespaces I initialized */
1277 list_del(&ops->list);
1278 free_exit_list(ops, &net_exit_list);
1279 return error;
1280}
1281
1282static void __unregister_pernet_operations(struct pernet_operations *ops)
1283{
1284 struct net *net;
1285 LIST_HEAD(net_exit_list);
1286
1287 list_del(&ops->list);
1288 /* See comment in __register_pernet_operations() */
1289 for_each_net(net)
1290 list_add_tail(&net->exit_list, &net_exit_list);
1291
1292 free_exit_list(ops, &net_exit_list);
1293}
1294
1295#else
1296
1297static int __register_pernet_operations(struct list_head *list,
1298 struct pernet_operations *ops)
1299{
1300 if (!init_net_initialized) {
1301 list_add_tail(&ops->list, list);
1302 return 0;
1303 }
1304
1305 return ops_init(ops, &init_net);
1306}
1307
1308static void __unregister_pernet_operations(struct pernet_operations *ops)
1309{
1310 if (!init_net_initialized) {
1311 list_del(&ops->list);
1312 } else {
1313 LIST_HEAD(net_exit_list);
1314 list_add(&init_net.exit_list, &net_exit_list);
1315 free_exit_list(ops, &net_exit_list);
1316 }
1317}
1318
1319#endif /* CONFIG_NET_NS */
1320
1321static DEFINE_IDA(net_generic_ids);
1322
1323static int register_pernet_operations(struct list_head *list,
1324 struct pernet_operations *ops)
1325{
1326 int error;
1327
1328 if (WARN_ON(!!ops->id ^ !!ops->size))
1329 return -EINVAL;
1330
1331 if (ops->id) {
1332 error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID,
1333 GFP_KERNEL);
1334 if (error < 0)
1335 return error;
1336 *ops->id = error;
1337 /* This does not require READ_ONCE as writers already hold
1338 * pernet_ops_rwsem. But WRITE_ONCE is needed to protect
1339 * net_alloc_generic.
1340 */
1341 WRITE_ONCE(max_gen_ptrs, max(max_gen_ptrs, *ops->id + 1));
1342 }
1343 error = __register_pernet_operations(list, ops);
1344 if (error) {
1345 rcu_barrier();
1346 if (ops->id)
1347 ida_free(&net_generic_ids, *ops->id);
1348 }
1349
1350 return error;
1351}
1352
1353static void unregister_pernet_operations(struct pernet_operations *ops)
1354{
1355 __unregister_pernet_operations(ops);
1356 rcu_barrier();
1357 if (ops->id)
1358 ida_free(&net_generic_ids, *ops->id);
1359}
1360
1361/**
1362 * register_pernet_subsys - register a network namespace subsystem
1363 * @ops: pernet operations structure for the subsystem
1364 *
1365 * Register a subsystem which has init and exit functions
1366 * that are called when network namespaces are created and
1367 * destroyed respectively.
1368 *
1369 * When registered all network namespace init functions are
1370 * called for every existing network namespace. Allowing kernel
1371 * modules to have a race free view of the set of network namespaces.
1372 *
1373 * When a new network namespace is created all of the init
1374 * methods are called in the order in which they were registered.
1375 *
1376 * When a network namespace is destroyed all of the exit methods
1377 * are called in the reverse of the order with which they were
1378 * registered.
1379 */
1380int register_pernet_subsys(struct pernet_operations *ops)
1381{
1382 int error;
1383 down_write(&pernet_ops_rwsem);
1384 error = register_pernet_operations(first_device, ops);
1385 up_write(&pernet_ops_rwsem);
1386 return error;
1387}
1388EXPORT_SYMBOL_GPL(register_pernet_subsys);
1389
1390/**
1391 * unregister_pernet_subsys - unregister a network namespace subsystem
1392 * @ops: pernet operations structure to manipulate
1393 *
1394 * Remove the pernet operations structure from the list to be
1395 * used when network namespaces are created or destroyed. In
1396 * addition run the exit method for all existing network
1397 * namespaces.
1398 */
1399void unregister_pernet_subsys(struct pernet_operations *ops)
1400{
1401 down_write(&pernet_ops_rwsem);
1402 unregister_pernet_operations(ops);
1403 up_write(&pernet_ops_rwsem);
1404}
1405EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1406
1407/**
1408 * register_pernet_device - register a network namespace device
1409 * @ops: pernet operations structure for the subsystem
1410 *
1411 * Register a device which has init and exit functions
1412 * that are called when network namespaces are created and
1413 * destroyed respectively.
1414 *
1415 * When registered all network namespace init functions are
1416 * called for every existing network namespace. Allowing kernel
1417 * modules to have a race free view of the set of network namespaces.
1418 *
1419 * When a new network namespace is created all of the init
1420 * methods are called in the order in which they were registered.
1421 *
1422 * When a network namespace is destroyed all of the exit methods
1423 * are called in the reverse of the order with which they were
1424 * registered.
1425 */
1426int register_pernet_device(struct pernet_operations *ops)
1427{
1428 int error;
1429 down_write(&pernet_ops_rwsem);
1430 error = register_pernet_operations(&pernet_list, ops);
1431 if (!error && (first_device == &pernet_list))
1432 first_device = &ops->list;
1433 up_write(&pernet_ops_rwsem);
1434 return error;
1435}
1436EXPORT_SYMBOL_GPL(register_pernet_device);
1437
1438/**
1439 * unregister_pernet_device - unregister a network namespace netdevice
1440 * @ops: pernet operations structure to manipulate
1441 *
1442 * Remove the pernet operations structure from the list to be
1443 * used when network namespaces are created or destroyed. In
1444 * addition run the exit method for all existing network
1445 * namespaces.
1446 */
1447void unregister_pernet_device(struct pernet_operations *ops)
1448{
1449 down_write(&pernet_ops_rwsem);
1450 if (&ops->list == first_device)
1451 first_device = first_device->next;
1452 unregister_pernet_operations(ops);
1453 up_write(&pernet_ops_rwsem);
1454}
1455EXPORT_SYMBOL_GPL(unregister_pernet_device);
1456
1457#ifdef CONFIG_NET_NS
1458static struct ns_common *netns_get(struct task_struct *task)
1459{
1460 struct net *net = NULL;
1461 struct nsproxy *nsproxy;
1462
1463 task_lock(task);
1464 nsproxy = task->nsproxy;
1465 if (nsproxy)
1466 net = get_net(nsproxy->net_ns);
1467 task_unlock(task);
1468
1469 return net ? &net->ns : NULL;
1470}
1471
1472static inline struct net *to_net_ns(struct ns_common *ns)
1473{
1474 return container_of(ns, struct net, ns);
1475}
1476
1477static void netns_put(struct ns_common *ns)
1478{
1479 put_net(to_net_ns(ns));
1480}
1481
1482static int netns_install(struct nsset *nsset, struct ns_common *ns)
1483{
1484 struct nsproxy *nsproxy = nsset->nsproxy;
1485 struct net *net = to_net_ns(ns);
1486
1487 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1488 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
1489 return -EPERM;
1490
1491 put_net(nsproxy->net_ns);
1492 nsproxy->net_ns = get_net(net);
1493 return 0;
1494}
1495
1496static struct user_namespace *netns_owner(struct ns_common *ns)
1497{
1498 return to_net_ns(ns)->user_ns;
1499}
1500
1501const struct proc_ns_operations netns_operations = {
1502 .name = "net",
1503 .type = CLONE_NEWNET,
1504 .get = netns_get,
1505 .put = netns_put,
1506 .install = netns_install,
1507 .owner = netns_owner,
1508};
1509#endif
1// SPDX-License-Identifier: GPL-2.0-only
2#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
3
4#include <linux/workqueue.h>
5#include <linux/rtnetlink.h>
6#include <linux/cache.h>
7#include <linux/slab.h>
8#include <linux/list.h>
9#include <linux/delay.h>
10#include <linux/sched.h>
11#include <linux/idr.h>
12#include <linux/rculist.h>
13#include <linux/nsproxy.h>
14#include <linux/fs.h>
15#include <linux/proc_ns.h>
16#include <linux/file.h>
17#include <linux/export.h>
18#include <linux/user_namespace.h>
19#include <linux/net_namespace.h>
20#include <linux/sched/task.h>
21#include <linux/uidgid.h>
22#include <linux/cookie.h>
23#include <linux/proc_fs.h>
24
25#include <net/sock.h>
26#include <net/netlink.h>
27#include <net/net_namespace.h>
28#include <net/netns/generic.h>
29
30/*
31 * Our network namespace constructor/destructor lists
32 */
33
34static LIST_HEAD(pernet_list);
35static struct list_head *first_device = &pernet_list;
36
37LIST_HEAD(net_namespace_list);
38EXPORT_SYMBOL_GPL(net_namespace_list);
39
40/* Protects net_namespace_list. Nests iside rtnl_lock() */
41DECLARE_RWSEM(net_rwsem);
42EXPORT_SYMBOL_GPL(net_rwsem);
43
44#ifdef CONFIG_KEYS
45static struct key_tag init_net_key_domain = { .usage = REFCOUNT_INIT(1) };
46#endif
47
48struct net init_net;
49EXPORT_SYMBOL(init_net);
50
51static bool init_net_initialized;
52/*
53 * pernet_ops_rwsem: protects: pernet_list, net_generic_ids,
54 * init_net_initialized and first_device pointer.
55 * This is internal net namespace object. Please, don't use it
56 * outside.
57 */
58DECLARE_RWSEM(pernet_ops_rwsem);
59EXPORT_SYMBOL_GPL(pernet_ops_rwsem);
60
61#define MIN_PERNET_OPS_ID \
62 ((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *))
63
64#define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
65
66static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
67
68DEFINE_COOKIE(net_cookie);
69
70static struct net_generic *net_alloc_generic(void)
71{
72 unsigned int gen_ptrs = READ_ONCE(max_gen_ptrs);
73 unsigned int generic_size;
74 struct net_generic *ng;
75
76 generic_size = offsetof(struct net_generic, ptr[gen_ptrs]);
77
78 ng = kzalloc(generic_size, GFP_KERNEL);
79 if (ng)
80 ng->s.len = gen_ptrs;
81
82 return ng;
83}
84
85static int net_assign_generic(struct net *net, unsigned int id, void *data)
86{
87 struct net_generic *ng, *old_ng;
88
89 BUG_ON(id < MIN_PERNET_OPS_ID);
90
91 old_ng = rcu_dereference_protected(net->gen,
92 lockdep_is_held(&pernet_ops_rwsem));
93 if (old_ng->s.len > id) {
94 old_ng->ptr[id] = data;
95 return 0;
96 }
97
98 ng = net_alloc_generic();
99 if (!ng)
100 return -ENOMEM;
101
102 /*
103 * Some synchronisation notes:
104 *
105 * The net_generic explores the net->gen array inside rcu
106 * read section. Besides once set the net->gen->ptr[x]
107 * pointer never changes (see rules in netns/generic.h).
108 *
109 * That said, we simply duplicate this array and schedule
110 * the old copy for kfree after a grace period.
111 */
112
113 memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
114 (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
115 ng->ptr[id] = data;
116
117 rcu_assign_pointer(net->gen, ng);
118 kfree_rcu(old_ng, s.rcu);
119 return 0;
120}
121
122static int ops_init(const struct pernet_operations *ops, struct net *net)
123{
124 struct net_generic *ng;
125 int err = -ENOMEM;
126 void *data = NULL;
127
128 if (ops->id && ops->size) {
129 data = kzalloc(ops->size, GFP_KERNEL);
130 if (!data)
131 goto out;
132
133 err = net_assign_generic(net, *ops->id, data);
134 if (err)
135 goto cleanup;
136 }
137 err = 0;
138 if (ops->init)
139 err = ops->init(net);
140 if (!err)
141 return 0;
142
143 if (ops->id && ops->size) {
144 ng = rcu_dereference_protected(net->gen,
145 lockdep_is_held(&pernet_ops_rwsem));
146 ng->ptr[*ops->id] = NULL;
147 }
148
149cleanup:
150 kfree(data);
151
152out:
153 return err;
154}
155
156static void ops_pre_exit_list(const struct pernet_operations *ops,
157 struct list_head *net_exit_list)
158{
159 struct net *net;
160
161 if (ops->pre_exit) {
162 list_for_each_entry(net, net_exit_list, exit_list)
163 ops->pre_exit(net);
164 }
165}
166
167static void ops_exit_list(const struct pernet_operations *ops,
168 struct list_head *net_exit_list)
169{
170 struct net *net;
171 if (ops->exit) {
172 list_for_each_entry(net, net_exit_list, exit_list) {
173 ops->exit(net);
174 cond_resched();
175 }
176 }
177 if (ops->exit_batch)
178 ops->exit_batch(net_exit_list);
179}
180
181static void ops_free_list(const struct pernet_operations *ops,
182 struct list_head *net_exit_list)
183{
184 struct net *net;
185 if (ops->size && ops->id) {
186 list_for_each_entry(net, net_exit_list, exit_list)
187 kfree(net_generic(net, *ops->id));
188 }
189}
190
191/* should be called with nsid_lock held */
192static int alloc_netid(struct net *net, struct net *peer, int reqid)
193{
194 int min = 0, max = 0;
195
196 if (reqid >= 0) {
197 min = reqid;
198 max = reqid + 1;
199 }
200
201 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
202}
203
204/* This function is used by idr_for_each(). If net is equal to peer, the
205 * function returns the id so that idr_for_each() stops. Because we cannot
206 * returns the id 0 (idr_for_each() will not stop), we return the magic value
207 * NET_ID_ZERO (-1) for it.
208 */
209#define NET_ID_ZERO -1
210static int net_eq_idr(int id, void *net, void *peer)
211{
212 if (net_eq(net, peer))
213 return id ? : NET_ID_ZERO;
214 return 0;
215}
216
217/* Must be called from RCU-critical section or with nsid_lock held */
218static int __peernet2id(const struct net *net, struct net *peer)
219{
220 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
221
222 /* Magic value for id 0. */
223 if (id == NET_ID_ZERO)
224 return 0;
225 if (id > 0)
226 return id;
227
228 return NETNSA_NSID_NOT_ASSIGNED;
229}
230
231static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
232 struct nlmsghdr *nlh, gfp_t gfp);
233/* This function returns the id of a peer netns. If no id is assigned, one will
234 * be allocated and returned.
235 */
236int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp)
237{
238 int id;
239
240 if (refcount_read(&net->ns.count) == 0)
241 return NETNSA_NSID_NOT_ASSIGNED;
242
243 spin_lock_bh(&net->nsid_lock);
244 id = __peernet2id(net, peer);
245 if (id >= 0) {
246 spin_unlock_bh(&net->nsid_lock);
247 return id;
248 }
249
250 /* When peer is obtained from RCU lists, we may race with
251 * its cleanup. Check whether it's alive, and this guarantees
252 * we never hash a peer back to net->netns_ids, after it has
253 * just been idr_remove()'d from there in cleanup_net().
254 */
255 if (!maybe_get_net(peer)) {
256 spin_unlock_bh(&net->nsid_lock);
257 return NETNSA_NSID_NOT_ASSIGNED;
258 }
259
260 id = alloc_netid(net, peer, -1);
261 spin_unlock_bh(&net->nsid_lock);
262
263 put_net(peer);
264 if (id < 0)
265 return NETNSA_NSID_NOT_ASSIGNED;
266
267 rtnl_net_notifyid(net, RTM_NEWNSID, id, 0, NULL, gfp);
268
269 return id;
270}
271EXPORT_SYMBOL_GPL(peernet2id_alloc);
272
273/* This function returns, if assigned, the id of a peer netns. */
274int peernet2id(const struct net *net, struct net *peer)
275{
276 int id;
277
278 rcu_read_lock();
279 id = __peernet2id(net, peer);
280 rcu_read_unlock();
281
282 return id;
283}
284EXPORT_SYMBOL(peernet2id);
285
286/* This function returns true is the peer netns has an id assigned into the
287 * current netns.
288 */
289bool peernet_has_id(const struct net *net, struct net *peer)
290{
291 return peernet2id(net, peer) >= 0;
292}
293
294struct net *get_net_ns_by_id(const struct net *net, int id)
295{
296 struct net *peer;
297
298 if (id < 0)
299 return NULL;
300
301 rcu_read_lock();
302 peer = idr_find(&net->netns_ids, id);
303 if (peer)
304 peer = maybe_get_net(peer);
305 rcu_read_unlock();
306
307 return peer;
308}
309EXPORT_SYMBOL_GPL(get_net_ns_by_id);
310
311/* init code that must occur even if setup_net() is not called. */
312static __net_init void preinit_net(struct net *net)
313{
314 ref_tracker_dir_init(&net->notrefcnt_tracker, 128, "net notrefcnt");
315}
316
317/*
318 * setup_net runs the initializers for the network namespace object.
319 */
320static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
321{
322 /* Must be called with pernet_ops_rwsem held */
323 const struct pernet_operations *ops, *saved_ops;
324 LIST_HEAD(net_exit_list);
325 LIST_HEAD(dev_kill_list);
326 int error = 0;
327
328 refcount_set(&net->ns.count, 1);
329 ref_tracker_dir_init(&net->refcnt_tracker, 128, "net refcnt");
330
331 refcount_set(&net->passive, 1);
332 get_random_bytes(&net->hash_mix, sizeof(u32));
333 preempt_disable();
334 net->net_cookie = gen_cookie_next(&net_cookie);
335 preempt_enable();
336 net->dev_base_seq = 1;
337 net->user_ns = user_ns;
338 idr_init(&net->netns_ids);
339 spin_lock_init(&net->nsid_lock);
340 mutex_init(&net->ipv4.ra_mutex);
341
342 list_for_each_entry(ops, &pernet_list, list) {
343 error = ops_init(ops, net);
344 if (error < 0)
345 goto out_undo;
346 }
347 down_write(&net_rwsem);
348 list_add_tail_rcu(&net->list, &net_namespace_list);
349 up_write(&net_rwsem);
350out:
351 return error;
352
353out_undo:
354 /* Walk through the list backwards calling the exit functions
355 * for the pernet modules whose init functions did not fail.
356 */
357 list_add(&net->exit_list, &net_exit_list);
358 saved_ops = ops;
359 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
360 ops_pre_exit_list(ops, &net_exit_list);
361
362 synchronize_rcu();
363
364 ops = saved_ops;
365 rtnl_lock();
366 list_for_each_entry_continue_reverse(ops, &pernet_list, list) {
367 if (ops->exit_batch_rtnl)
368 ops->exit_batch_rtnl(&net_exit_list, &dev_kill_list);
369 }
370 unregister_netdevice_many(&dev_kill_list);
371 rtnl_unlock();
372
373 ops = saved_ops;
374 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
375 ops_exit_list(ops, &net_exit_list);
376
377 ops = saved_ops;
378 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
379 ops_free_list(ops, &net_exit_list);
380
381 rcu_barrier();
382 goto out;
383}
384
385static int __net_init net_defaults_init_net(struct net *net)
386{
387 net->core.sysctl_somaxconn = SOMAXCONN;
388 /* Limits per socket sk_omem_alloc usage.
389 * TCP zerocopy regular usage needs 128 KB.
390 */
391 net->core.sysctl_optmem_max = 128 * 1024;
392 net->core.sysctl_txrehash = SOCK_TXREHASH_ENABLED;
393
394 return 0;
395}
396
397static struct pernet_operations net_defaults_ops = {
398 .init = net_defaults_init_net,
399};
400
401static __init int net_defaults_init(void)
402{
403 if (register_pernet_subsys(&net_defaults_ops))
404 panic("Cannot initialize net default settings");
405
406 return 0;
407}
408
409core_initcall(net_defaults_init);
410
411#ifdef CONFIG_NET_NS
412static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
413{
414 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
415}
416
417static void dec_net_namespaces(struct ucounts *ucounts)
418{
419 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
420}
421
422static struct kmem_cache *net_cachep __ro_after_init;
423static struct workqueue_struct *netns_wq;
424
425static struct net *net_alloc(void)
426{
427 struct net *net = NULL;
428 struct net_generic *ng;
429
430 ng = net_alloc_generic();
431 if (!ng)
432 goto out;
433
434 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
435 if (!net)
436 goto out_free;
437
438#ifdef CONFIG_KEYS
439 net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
440 if (!net->key_domain)
441 goto out_free_2;
442 refcount_set(&net->key_domain->usage, 1);
443#endif
444
445 rcu_assign_pointer(net->gen, ng);
446out:
447 return net;
448
449#ifdef CONFIG_KEYS
450out_free_2:
451 kmem_cache_free(net_cachep, net);
452 net = NULL;
453#endif
454out_free:
455 kfree(ng);
456 goto out;
457}
458
459static void net_free(struct net *net)
460{
461 if (refcount_dec_and_test(&net->passive)) {
462 kfree(rcu_access_pointer(net->gen));
463
464 /* There should not be any trackers left there. */
465 ref_tracker_dir_exit(&net->notrefcnt_tracker);
466
467 kmem_cache_free(net_cachep, net);
468 }
469}
470
471void net_drop_ns(void *p)
472{
473 struct net *net = (struct net *)p;
474
475 if (net)
476 net_free(net);
477}
478
479struct net *copy_net_ns(unsigned long flags,
480 struct user_namespace *user_ns, struct net *old_net)
481{
482 struct ucounts *ucounts;
483 struct net *net;
484 int rv;
485
486 if (!(flags & CLONE_NEWNET))
487 return get_net(old_net);
488
489 ucounts = inc_net_namespaces(user_ns);
490 if (!ucounts)
491 return ERR_PTR(-ENOSPC);
492
493 net = net_alloc();
494 if (!net) {
495 rv = -ENOMEM;
496 goto dec_ucounts;
497 }
498
499 preinit_net(net);
500 refcount_set(&net->passive, 1);
501 net->ucounts = ucounts;
502 get_user_ns(user_ns);
503
504 rv = down_read_killable(&pernet_ops_rwsem);
505 if (rv < 0)
506 goto put_userns;
507
508 rv = setup_net(net, user_ns);
509
510 up_read(&pernet_ops_rwsem);
511
512 if (rv < 0) {
513put_userns:
514#ifdef CONFIG_KEYS
515 key_remove_domain(net->key_domain);
516#endif
517 put_user_ns(user_ns);
518 net_free(net);
519dec_ucounts:
520 dec_net_namespaces(ucounts);
521 return ERR_PTR(rv);
522 }
523 return net;
524}
525
526/**
527 * net_ns_get_ownership - get sysfs ownership data for @net
528 * @net: network namespace in question (can be NULL)
529 * @uid: kernel user ID for sysfs objects
530 * @gid: kernel group ID for sysfs objects
531 *
532 * Returns the uid/gid pair of root in the user namespace associated with the
533 * given network namespace.
534 */
535void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
536{
537 if (net) {
538 kuid_t ns_root_uid = make_kuid(net->user_ns, 0);
539 kgid_t ns_root_gid = make_kgid(net->user_ns, 0);
540
541 if (uid_valid(ns_root_uid))
542 *uid = ns_root_uid;
543
544 if (gid_valid(ns_root_gid))
545 *gid = ns_root_gid;
546 } else {
547 *uid = GLOBAL_ROOT_UID;
548 *gid = GLOBAL_ROOT_GID;
549 }
550}
551EXPORT_SYMBOL_GPL(net_ns_get_ownership);
552
553static void unhash_nsid(struct net *net, struct net *last)
554{
555 struct net *tmp;
556 /* This function is only called from cleanup_net() work,
557 * and this work is the only process, that may delete
558 * a net from net_namespace_list. So, when the below
559 * is executing, the list may only grow. Thus, we do not
560 * use for_each_net_rcu() or net_rwsem.
561 */
562 for_each_net(tmp) {
563 int id;
564
565 spin_lock_bh(&tmp->nsid_lock);
566 id = __peernet2id(tmp, net);
567 if (id >= 0)
568 idr_remove(&tmp->netns_ids, id);
569 spin_unlock_bh(&tmp->nsid_lock);
570 if (id >= 0)
571 rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL,
572 GFP_KERNEL);
573 if (tmp == last)
574 break;
575 }
576 spin_lock_bh(&net->nsid_lock);
577 idr_destroy(&net->netns_ids);
578 spin_unlock_bh(&net->nsid_lock);
579}
580
581static LLIST_HEAD(cleanup_list);
582
583static void cleanup_net(struct work_struct *work)
584{
585 const struct pernet_operations *ops;
586 struct net *net, *tmp, *last;
587 struct llist_node *net_kill_list;
588 LIST_HEAD(net_exit_list);
589 LIST_HEAD(dev_kill_list);
590
591 /* Atomically snapshot the list of namespaces to cleanup */
592 net_kill_list = llist_del_all(&cleanup_list);
593
594 down_read(&pernet_ops_rwsem);
595
596 /* Don't let anyone else find us. */
597 down_write(&net_rwsem);
598 llist_for_each_entry(net, net_kill_list, cleanup_list)
599 list_del_rcu(&net->list);
600 /* Cache last net. After we unlock rtnl, no one new net
601 * added to net_namespace_list can assign nsid pointer
602 * to a net from net_kill_list (see peernet2id_alloc()).
603 * So, we skip them in unhash_nsid().
604 *
605 * Note, that unhash_nsid() does not delete nsid links
606 * between net_kill_list's nets, as they've already
607 * deleted from net_namespace_list. But, this would be
608 * useless anyway, as netns_ids are destroyed there.
609 */
610 last = list_last_entry(&net_namespace_list, struct net, list);
611 up_write(&net_rwsem);
612
613 llist_for_each_entry(net, net_kill_list, cleanup_list) {
614 unhash_nsid(net, last);
615 list_add_tail(&net->exit_list, &net_exit_list);
616 }
617
618 /* Run all of the network namespace pre_exit methods */
619 list_for_each_entry_reverse(ops, &pernet_list, list)
620 ops_pre_exit_list(ops, &net_exit_list);
621
622 /*
623 * Another CPU might be rcu-iterating the list, wait for it.
624 * This needs to be before calling the exit() notifiers, so
625 * the rcu_barrier() below isn't sufficient alone.
626 * Also the pre_exit() and exit() methods need this barrier.
627 */
628 synchronize_rcu_expedited();
629
630 rtnl_lock();
631 list_for_each_entry_reverse(ops, &pernet_list, list) {
632 if (ops->exit_batch_rtnl)
633 ops->exit_batch_rtnl(&net_exit_list, &dev_kill_list);
634 }
635 unregister_netdevice_many(&dev_kill_list);
636 rtnl_unlock();
637
638 /* Run all of the network namespace exit methods */
639 list_for_each_entry_reverse(ops, &pernet_list, list)
640 ops_exit_list(ops, &net_exit_list);
641
642 /* Free the net generic variables */
643 list_for_each_entry_reverse(ops, &pernet_list, list)
644 ops_free_list(ops, &net_exit_list);
645
646 up_read(&pernet_ops_rwsem);
647
648 /* Ensure there are no outstanding rcu callbacks using this
649 * network namespace.
650 */
651 rcu_barrier();
652
653 /* Finally it is safe to free my network namespace structure */
654 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
655 list_del_init(&net->exit_list);
656 dec_net_namespaces(net->ucounts);
657#ifdef CONFIG_KEYS
658 key_remove_domain(net->key_domain);
659#endif
660 put_user_ns(net->user_ns);
661 net_free(net);
662 }
663}
664
665/**
666 * net_ns_barrier - wait until concurrent net_cleanup_work is done
667 *
668 * cleanup_net runs from work queue and will first remove namespaces
669 * from the global list, then run net exit functions.
670 *
671 * Call this in module exit path to make sure that all netns
672 * ->exit ops have been invoked before the function is removed.
673 */
674void net_ns_barrier(void)
675{
676 down_write(&pernet_ops_rwsem);
677 up_write(&pernet_ops_rwsem);
678}
679EXPORT_SYMBOL(net_ns_barrier);
680
681static DECLARE_WORK(net_cleanup_work, cleanup_net);
682
683void __put_net(struct net *net)
684{
685 ref_tracker_dir_exit(&net->refcnt_tracker);
686 /* Cleanup the network namespace in process context */
687 if (llist_add(&net->cleanup_list, &cleanup_list))
688 queue_work(netns_wq, &net_cleanup_work);
689}
690EXPORT_SYMBOL_GPL(__put_net);
691
692/**
693 * get_net_ns - increment the refcount of the network namespace
694 * @ns: common namespace (net)
695 *
696 * Returns the net's common namespace.
697 */
698struct ns_common *get_net_ns(struct ns_common *ns)
699{
700 return &get_net(container_of(ns, struct net, ns))->ns;
701}
702EXPORT_SYMBOL_GPL(get_net_ns);
703
704struct net *get_net_ns_by_fd(int fd)
705{
706 struct fd f = fdget(fd);
707 struct net *net = ERR_PTR(-EINVAL);
708
709 if (!f.file)
710 return ERR_PTR(-EBADF);
711
712 if (proc_ns_file(f.file)) {
713 struct ns_common *ns = get_proc_ns(file_inode(f.file));
714 if (ns->ops == &netns_operations)
715 net = get_net(container_of(ns, struct net, ns));
716 }
717 fdput(f);
718
719 return net;
720}
721EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
722#endif
723
724struct net *get_net_ns_by_pid(pid_t pid)
725{
726 struct task_struct *tsk;
727 struct net *net;
728
729 /* Lookup the network namespace */
730 net = ERR_PTR(-ESRCH);
731 rcu_read_lock();
732 tsk = find_task_by_vpid(pid);
733 if (tsk) {
734 struct nsproxy *nsproxy;
735 task_lock(tsk);
736 nsproxy = tsk->nsproxy;
737 if (nsproxy)
738 net = get_net(nsproxy->net_ns);
739 task_unlock(tsk);
740 }
741 rcu_read_unlock();
742 return net;
743}
744EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
745
746static __net_init int net_ns_net_init(struct net *net)
747{
748#ifdef CONFIG_NET_NS
749 net->ns.ops = &netns_operations;
750#endif
751 return ns_alloc_inum(&net->ns);
752}
753
754static __net_exit void net_ns_net_exit(struct net *net)
755{
756 ns_free_inum(&net->ns);
757}
758
759static struct pernet_operations __net_initdata net_ns_ops = {
760 .init = net_ns_net_init,
761 .exit = net_ns_net_exit,
762};
763
764static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
765 [NETNSA_NONE] = { .type = NLA_UNSPEC },
766 [NETNSA_NSID] = { .type = NLA_S32 },
767 [NETNSA_PID] = { .type = NLA_U32 },
768 [NETNSA_FD] = { .type = NLA_U32 },
769 [NETNSA_TARGET_NSID] = { .type = NLA_S32 },
770};
771
772static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
773 struct netlink_ext_ack *extack)
774{
775 struct net *net = sock_net(skb->sk);
776 struct nlattr *tb[NETNSA_MAX + 1];
777 struct nlattr *nla;
778 struct net *peer;
779 int nsid, err;
780
781 err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
782 NETNSA_MAX, rtnl_net_policy, extack);
783 if (err < 0)
784 return err;
785 if (!tb[NETNSA_NSID]) {
786 NL_SET_ERR_MSG(extack, "nsid is missing");
787 return -EINVAL;
788 }
789 nsid = nla_get_s32(tb[NETNSA_NSID]);
790
791 if (tb[NETNSA_PID]) {
792 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
793 nla = tb[NETNSA_PID];
794 } else if (tb[NETNSA_FD]) {
795 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
796 nla = tb[NETNSA_FD];
797 } else {
798 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
799 return -EINVAL;
800 }
801 if (IS_ERR(peer)) {
802 NL_SET_BAD_ATTR(extack, nla);
803 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
804 return PTR_ERR(peer);
805 }
806
807 spin_lock_bh(&net->nsid_lock);
808 if (__peernet2id(net, peer) >= 0) {
809 spin_unlock_bh(&net->nsid_lock);
810 err = -EEXIST;
811 NL_SET_BAD_ATTR(extack, nla);
812 NL_SET_ERR_MSG(extack,
813 "Peer netns already has a nsid assigned");
814 goto out;
815 }
816
817 err = alloc_netid(net, peer, nsid);
818 spin_unlock_bh(&net->nsid_lock);
819 if (err >= 0) {
820 rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid,
821 nlh, GFP_KERNEL);
822 err = 0;
823 } else if (err == -ENOSPC && nsid >= 0) {
824 err = -EEXIST;
825 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
826 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
827 }
828out:
829 put_net(peer);
830 return err;
831}
832
833static int rtnl_net_get_size(void)
834{
835 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
836 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
837 + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
838 ;
839}
840
841struct net_fill_args {
842 u32 portid;
843 u32 seq;
844 int flags;
845 int cmd;
846 int nsid;
847 bool add_ref;
848 int ref_nsid;
849};
850
851static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
852{
853 struct nlmsghdr *nlh;
854 struct rtgenmsg *rth;
855
856 nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
857 args->flags);
858 if (!nlh)
859 return -EMSGSIZE;
860
861 rth = nlmsg_data(nlh);
862 rth->rtgen_family = AF_UNSPEC;
863
864 if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
865 goto nla_put_failure;
866
867 if (args->add_ref &&
868 nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
869 goto nla_put_failure;
870
871 nlmsg_end(skb, nlh);
872 return 0;
873
874nla_put_failure:
875 nlmsg_cancel(skb, nlh);
876 return -EMSGSIZE;
877}
878
879static int rtnl_net_valid_getid_req(struct sk_buff *skb,
880 const struct nlmsghdr *nlh,
881 struct nlattr **tb,
882 struct netlink_ext_ack *extack)
883{
884 int i, err;
885
886 if (!netlink_strict_get_check(skb))
887 return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
888 tb, NETNSA_MAX, rtnl_net_policy,
889 extack);
890
891 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
892 NETNSA_MAX, rtnl_net_policy,
893 extack);
894 if (err)
895 return err;
896
897 for (i = 0; i <= NETNSA_MAX; i++) {
898 if (!tb[i])
899 continue;
900
901 switch (i) {
902 case NETNSA_PID:
903 case NETNSA_FD:
904 case NETNSA_NSID:
905 case NETNSA_TARGET_NSID:
906 break;
907 default:
908 NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
909 return -EINVAL;
910 }
911 }
912
913 return 0;
914}
915
916static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
917 struct netlink_ext_ack *extack)
918{
919 struct net *net = sock_net(skb->sk);
920 struct nlattr *tb[NETNSA_MAX + 1];
921 struct net_fill_args fillargs = {
922 .portid = NETLINK_CB(skb).portid,
923 .seq = nlh->nlmsg_seq,
924 .cmd = RTM_NEWNSID,
925 };
926 struct net *peer, *target = net;
927 struct nlattr *nla;
928 struct sk_buff *msg;
929 int err;
930
931 err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
932 if (err < 0)
933 return err;
934 if (tb[NETNSA_PID]) {
935 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
936 nla = tb[NETNSA_PID];
937 } else if (tb[NETNSA_FD]) {
938 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
939 nla = tb[NETNSA_FD];
940 } else if (tb[NETNSA_NSID]) {
941 peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
942 if (!peer)
943 peer = ERR_PTR(-ENOENT);
944 nla = tb[NETNSA_NSID];
945 } else {
946 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
947 return -EINVAL;
948 }
949
950 if (IS_ERR(peer)) {
951 NL_SET_BAD_ATTR(extack, nla);
952 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
953 return PTR_ERR(peer);
954 }
955
956 if (tb[NETNSA_TARGET_NSID]) {
957 int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
958
959 target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
960 if (IS_ERR(target)) {
961 NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
962 NL_SET_ERR_MSG(extack,
963 "Target netns reference is invalid");
964 err = PTR_ERR(target);
965 goto out;
966 }
967 fillargs.add_ref = true;
968 fillargs.ref_nsid = peernet2id(net, peer);
969 }
970
971 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
972 if (!msg) {
973 err = -ENOMEM;
974 goto out;
975 }
976
977 fillargs.nsid = peernet2id(target, peer);
978 err = rtnl_net_fill(msg, &fillargs);
979 if (err < 0)
980 goto err_out;
981
982 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
983 goto out;
984
985err_out:
986 nlmsg_free(msg);
987out:
988 if (fillargs.add_ref)
989 put_net(target);
990 put_net(peer);
991 return err;
992}
993
994struct rtnl_net_dump_cb {
995 struct net *tgt_net;
996 struct net *ref_net;
997 struct sk_buff *skb;
998 struct net_fill_args fillargs;
999 int idx;
1000 int s_idx;
1001};
1002
1003/* Runs in RCU-critical section. */
1004static int rtnl_net_dumpid_one(int id, void *peer, void *data)
1005{
1006 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
1007 int ret;
1008
1009 if (net_cb->idx < net_cb->s_idx)
1010 goto cont;
1011
1012 net_cb->fillargs.nsid = id;
1013 if (net_cb->fillargs.add_ref)
1014 net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
1015 ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
1016 if (ret < 0)
1017 return ret;
1018
1019cont:
1020 net_cb->idx++;
1021 return 0;
1022}
1023
1024static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
1025 struct rtnl_net_dump_cb *net_cb,
1026 struct netlink_callback *cb)
1027{
1028 struct netlink_ext_ack *extack = cb->extack;
1029 struct nlattr *tb[NETNSA_MAX + 1];
1030 int err, i;
1031
1032 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
1033 NETNSA_MAX, rtnl_net_policy,
1034 extack);
1035 if (err < 0)
1036 return err;
1037
1038 for (i = 0; i <= NETNSA_MAX; i++) {
1039 if (!tb[i])
1040 continue;
1041
1042 if (i == NETNSA_TARGET_NSID) {
1043 struct net *net;
1044
1045 net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
1046 if (IS_ERR(net)) {
1047 NL_SET_BAD_ATTR(extack, tb[i]);
1048 NL_SET_ERR_MSG(extack,
1049 "Invalid target network namespace id");
1050 return PTR_ERR(net);
1051 }
1052 net_cb->fillargs.add_ref = true;
1053 net_cb->ref_net = net_cb->tgt_net;
1054 net_cb->tgt_net = net;
1055 } else {
1056 NL_SET_BAD_ATTR(extack, tb[i]);
1057 NL_SET_ERR_MSG(extack,
1058 "Unsupported attribute in dump request");
1059 return -EINVAL;
1060 }
1061 }
1062
1063 return 0;
1064}
1065
1066static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1067{
1068 struct rtnl_net_dump_cb net_cb = {
1069 .tgt_net = sock_net(skb->sk),
1070 .skb = skb,
1071 .fillargs = {
1072 .portid = NETLINK_CB(cb->skb).portid,
1073 .seq = cb->nlh->nlmsg_seq,
1074 .flags = NLM_F_MULTI,
1075 .cmd = RTM_NEWNSID,
1076 },
1077 .idx = 0,
1078 .s_idx = cb->args[0],
1079 };
1080 int err = 0;
1081
1082 if (cb->strict_check) {
1083 err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1084 if (err < 0)
1085 goto end;
1086 }
1087
1088 rcu_read_lock();
1089 idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1090 rcu_read_unlock();
1091
1092 cb->args[0] = net_cb.idx;
1093end:
1094 if (net_cb.fillargs.add_ref)
1095 put_net(net_cb.tgt_net);
1096 return err < 0 ? err : skb->len;
1097}
1098
1099static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1100 struct nlmsghdr *nlh, gfp_t gfp)
1101{
1102 struct net_fill_args fillargs = {
1103 .portid = portid,
1104 .seq = nlh ? nlh->nlmsg_seq : 0,
1105 .cmd = cmd,
1106 .nsid = id,
1107 };
1108 struct sk_buff *msg;
1109 int err = -ENOMEM;
1110
1111 msg = nlmsg_new(rtnl_net_get_size(), gfp);
1112 if (!msg)
1113 goto out;
1114
1115 err = rtnl_net_fill(msg, &fillargs);
1116 if (err < 0)
1117 goto err_out;
1118
1119 rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, gfp);
1120 return;
1121
1122err_out:
1123 nlmsg_free(msg);
1124out:
1125 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1126}
1127
1128#ifdef CONFIG_NET_NS
1129static void __init netns_ipv4_struct_check(void)
1130{
1131 /* TX readonly hotpath cache lines */
1132 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1133 sysctl_tcp_early_retrans);
1134 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1135 sysctl_tcp_tso_win_divisor);
1136 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1137 sysctl_tcp_tso_rtt_log);
1138 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1139 sysctl_tcp_autocorking);
1140 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1141 sysctl_tcp_min_snd_mss);
1142 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1143 sysctl_tcp_notsent_lowat);
1144 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1145 sysctl_tcp_limit_output_bytes);
1146 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1147 sysctl_tcp_min_rtt_wlen);
1148 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1149 sysctl_tcp_wmem);
1150 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1151 sysctl_ip_fwd_use_pmtu);
1152 CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_tx, 33);
1153
1154 /* TXRX readonly hotpath cache lines */
1155 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_txrx,
1156 sysctl_tcp_moderate_rcvbuf);
1157 CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_txrx, 1);
1158
1159 /* RX readonly hotpath cache line */
1160 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1161 sysctl_ip_early_demux);
1162 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1163 sysctl_tcp_early_demux);
1164 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1165 sysctl_tcp_reordering);
1166 CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1167 sysctl_tcp_rmem);
1168 CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_rx, 18);
1169}
1170#endif
1171
1172void __init net_ns_init(void)
1173{
1174 struct net_generic *ng;
1175
1176#ifdef CONFIG_NET_NS
1177 netns_ipv4_struct_check();
1178 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1179 SMP_CACHE_BYTES,
1180 SLAB_PANIC|SLAB_ACCOUNT, NULL);
1181
1182 /* Create workqueue for cleanup */
1183 netns_wq = create_singlethread_workqueue("netns");
1184 if (!netns_wq)
1185 panic("Could not create netns workq");
1186#endif
1187
1188 ng = net_alloc_generic();
1189 if (!ng)
1190 panic("Could not allocate generic netns");
1191
1192 rcu_assign_pointer(init_net.gen, ng);
1193
1194#ifdef CONFIG_KEYS
1195 init_net.key_domain = &init_net_key_domain;
1196#endif
1197 down_write(&pernet_ops_rwsem);
1198 preinit_net(&init_net);
1199 if (setup_net(&init_net, &init_user_ns))
1200 panic("Could not setup the initial network namespace");
1201
1202 init_net_initialized = true;
1203 up_write(&pernet_ops_rwsem);
1204
1205 if (register_pernet_subsys(&net_ns_ops))
1206 panic("Could not register network namespace subsystems");
1207
1208 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
1209 RTNL_FLAG_DOIT_UNLOCKED);
1210 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
1211 RTNL_FLAG_DOIT_UNLOCKED);
1212}
1213
1214static void free_exit_list(struct pernet_operations *ops, struct list_head *net_exit_list)
1215{
1216 ops_pre_exit_list(ops, net_exit_list);
1217 synchronize_rcu();
1218
1219 if (ops->exit_batch_rtnl) {
1220 LIST_HEAD(dev_kill_list);
1221
1222 rtnl_lock();
1223 ops->exit_batch_rtnl(net_exit_list, &dev_kill_list);
1224 unregister_netdevice_many(&dev_kill_list);
1225 rtnl_unlock();
1226 }
1227 ops_exit_list(ops, net_exit_list);
1228
1229 ops_free_list(ops, net_exit_list);
1230}
1231
1232#ifdef CONFIG_NET_NS
1233static int __register_pernet_operations(struct list_head *list,
1234 struct pernet_operations *ops)
1235{
1236 struct net *net;
1237 int error;
1238 LIST_HEAD(net_exit_list);
1239
1240 list_add_tail(&ops->list, list);
1241 if (ops->init || (ops->id && ops->size)) {
1242 /* We held write locked pernet_ops_rwsem, and parallel
1243 * setup_net() and cleanup_net() are not possible.
1244 */
1245 for_each_net(net) {
1246 error = ops_init(ops, net);
1247 if (error)
1248 goto out_undo;
1249 list_add_tail(&net->exit_list, &net_exit_list);
1250 }
1251 }
1252 return 0;
1253
1254out_undo:
1255 /* If I have an error cleanup all namespaces I initialized */
1256 list_del(&ops->list);
1257 free_exit_list(ops, &net_exit_list);
1258 return error;
1259}
1260
1261static void __unregister_pernet_operations(struct pernet_operations *ops)
1262{
1263 struct net *net;
1264 LIST_HEAD(net_exit_list);
1265
1266 list_del(&ops->list);
1267 /* See comment in __register_pernet_operations() */
1268 for_each_net(net)
1269 list_add_tail(&net->exit_list, &net_exit_list);
1270
1271 free_exit_list(ops, &net_exit_list);
1272}
1273
1274#else
1275
1276static int __register_pernet_operations(struct list_head *list,
1277 struct pernet_operations *ops)
1278{
1279 if (!init_net_initialized) {
1280 list_add_tail(&ops->list, list);
1281 return 0;
1282 }
1283
1284 return ops_init(ops, &init_net);
1285}
1286
1287static void __unregister_pernet_operations(struct pernet_operations *ops)
1288{
1289 if (!init_net_initialized) {
1290 list_del(&ops->list);
1291 } else {
1292 LIST_HEAD(net_exit_list);
1293 list_add(&init_net.exit_list, &net_exit_list);
1294 free_exit_list(ops, &net_exit_list);
1295 }
1296}
1297
1298#endif /* CONFIG_NET_NS */
1299
1300static DEFINE_IDA(net_generic_ids);
1301
1302static int register_pernet_operations(struct list_head *list,
1303 struct pernet_operations *ops)
1304{
1305 int error;
1306
1307 if (ops->id) {
1308 error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID,
1309 GFP_KERNEL);
1310 if (error < 0)
1311 return error;
1312 *ops->id = error;
1313 /* This does not require READ_ONCE as writers already hold
1314 * pernet_ops_rwsem. But WRITE_ONCE is needed to protect
1315 * net_alloc_generic.
1316 */
1317 WRITE_ONCE(max_gen_ptrs, max(max_gen_ptrs, *ops->id + 1));
1318 }
1319 error = __register_pernet_operations(list, ops);
1320 if (error) {
1321 rcu_barrier();
1322 if (ops->id)
1323 ida_free(&net_generic_ids, *ops->id);
1324 }
1325
1326 return error;
1327}
1328
1329static void unregister_pernet_operations(struct pernet_operations *ops)
1330{
1331 __unregister_pernet_operations(ops);
1332 rcu_barrier();
1333 if (ops->id)
1334 ida_free(&net_generic_ids, *ops->id);
1335}
1336
1337/**
1338 * register_pernet_subsys - register a network namespace subsystem
1339 * @ops: pernet operations structure for the subsystem
1340 *
1341 * Register a subsystem which has init and exit functions
1342 * that are called when network namespaces are created and
1343 * destroyed respectively.
1344 *
1345 * When registered all network namespace init functions are
1346 * called for every existing network namespace. Allowing kernel
1347 * modules to have a race free view of the set of network namespaces.
1348 *
1349 * When a new network namespace is created all of the init
1350 * methods are called in the order in which they were registered.
1351 *
1352 * When a network namespace is destroyed all of the exit methods
1353 * are called in the reverse of the order with which they were
1354 * registered.
1355 */
1356int register_pernet_subsys(struct pernet_operations *ops)
1357{
1358 int error;
1359 down_write(&pernet_ops_rwsem);
1360 error = register_pernet_operations(first_device, ops);
1361 up_write(&pernet_ops_rwsem);
1362 return error;
1363}
1364EXPORT_SYMBOL_GPL(register_pernet_subsys);
1365
1366/**
1367 * unregister_pernet_subsys - unregister a network namespace subsystem
1368 * @ops: pernet operations structure to manipulate
1369 *
1370 * Remove the pernet operations structure from the list to be
1371 * used when network namespaces are created or destroyed. In
1372 * addition run the exit method for all existing network
1373 * namespaces.
1374 */
1375void unregister_pernet_subsys(struct pernet_operations *ops)
1376{
1377 down_write(&pernet_ops_rwsem);
1378 unregister_pernet_operations(ops);
1379 up_write(&pernet_ops_rwsem);
1380}
1381EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1382
1383/**
1384 * register_pernet_device - register a network namespace device
1385 * @ops: pernet operations structure for the subsystem
1386 *
1387 * Register a device which has init and exit functions
1388 * that are called when network namespaces are created and
1389 * destroyed respectively.
1390 *
1391 * When registered all network namespace init functions are
1392 * called for every existing network namespace. Allowing kernel
1393 * modules to have a race free view of the set of network namespaces.
1394 *
1395 * When a new network namespace is created all of the init
1396 * methods are called in the order in which they were registered.
1397 *
1398 * When a network namespace is destroyed all of the exit methods
1399 * are called in the reverse of the order with which they were
1400 * registered.
1401 */
1402int register_pernet_device(struct pernet_operations *ops)
1403{
1404 int error;
1405 down_write(&pernet_ops_rwsem);
1406 error = register_pernet_operations(&pernet_list, ops);
1407 if (!error && (first_device == &pernet_list))
1408 first_device = &ops->list;
1409 up_write(&pernet_ops_rwsem);
1410 return error;
1411}
1412EXPORT_SYMBOL_GPL(register_pernet_device);
1413
1414/**
1415 * unregister_pernet_device - unregister a network namespace netdevice
1416 * @ops: pernet operations structure to manipulate
1417 *
1418 * Remove the pernet operations structure from the list to be
1419 * used when network namespaces are created or destroyed. In
1420 * addition run the exit method for all existing network
1421 * namespaces.
1422 */
1423void unregister_pernet_device(struct pernet_operations *ops)
1424{
1425 down_write(&pernet_ops_rwsem);
1426 if (&ops->list == first_device)
1427 first_device = first_device->next;
1428 unregister_pernet_operations(ops);
1429 up_write(&pernet_ops_rwsem);
1430}
1431EXPORT_SYMBOL_GPL(unregister_pernet_device);
1432
1433#ifdef CONFIG_NET_NS
1434static struct ns_common *netns_get(struct task_struct *task)
1435{
1436 struct net *net = NULL;
1437 struct nsproxy *nsproxy;
1438
1439 task_lock(task);
1440 nsproxy = task->nsproxy;
1441 if (nsproxy)
1442 net = get_net(nsproxy->net_ns);
1443 task_unlock(task);
1444
1445 return net ? &net->ns : NULL;
1446}
1447
1448static inline struct net *to_net_ns(struct ns_common *ns)
1449{
1450 return container_of(ns, struct net, ns);
1451}
1452
1453static void netns_put(struct ns_common *ns)
1454{
1455 put_net(to_net_ns(ns));
1456}
1457
1458static int netns_install(struct nsset *nsset, struct ns_common *ns)
1459{
1460 struct nsproxy *nsproxy = nsset->nsproxy;
1461 struct net *net = to_net_ns(ns);
1462
1463 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1464 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
1465 return -EPERM;
1466
1467 put_net(nsproxy->net_ns);
1468 nsproxy->net_ns = get_net(net);
1469 return 0;
1470}
1471
1472static struct user_namespace *netns_owner(struct ns_common *ns)
1473{
1474 return to_net_ns(ns)->user_ns;
1475}
1476
1477const struct proc_ns_operations netns_operations = {
1478 .name = "net",
1479 .type = CLONE_NEWNET,
1480 .get = netns_get,
1481 .put = netns_put,
1482 .install = netns_install,
1483 .owner = netns_owner,
1484};
1485#endif