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