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#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