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 struct net_generic *ng;
73 unsigned int generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
74
75 ng = kzalloc(size: generic_size, GFP_KERNEL);
76 if (ng)
77 ng->s.len = max_gen_ptrs;
78
79 return ng;
80}
81
82static int net_assign_generic(struct net *net, unsigned int id, void *data)
83{
84 struct net_generic *ng, *old_ng;
85
86 BUG_ON(id < MIN_PERNET_OPS_ID);
87
88 old_ng = rcu_dereference_protected(net->gen,
89 lockdep_is_held(&pernet_ops_rwsem));
90 if (old_ng->s.len > id) {
91 old_ng->ptr[id] = data;
92 return 0;
93 }
94
95 ng = net_alloc_generic();
96 if (!ng)
97 return -ENOMEM;
98
99 /*
100 * Some synchronisation notes:
101 *
102 * The net_generic explores the net->gen array inside rcu
103 * read section. Besides once set the net->gen->ptr[x]
104 * pointer never changes (see rules in netns/generic.h).
105 *
106 * That said, we simply duplicate this array and schedule
107 * the old copy for kfree after a grace period.
108 */
109
110 memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
111 (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
112 ng->ptr[id] = data;
113
114 rcu_assign_pointer(net->gen, ng);
115 kfree_rcu(old_ng, s.rcu);
116 return 0;
117}
118
119static int ops_init(const struct pernet_operations *ops, struct net *net)
120{
121 struct net_generic *ng;
122 int err = -ENOMEM;
123 void *data = NULL;
124
125 if (ops->id && ops->size) {
126 data = kzalloc(size: ops->size, GFP_KERNEL);
127 if (!data)
128 goto out;
129
130 err = net_assign_generic(net, id: *ops->id, data);
131 if (err)
132 goto cleanup;
133 }
134 err = 0;
135 if (ops->init)
136 err = ops->init(net);
137 if (!err)
138 return 0;
139
140 if (ops->id && ops->size) {
141 ng = rcu_dereference_protected(net->gen,
142 lockdep_is_held(&pernet_ops_rwsem));
143 ng->ptr[*ops->id] = NULL;
144 }
145
146cleanup:
147 kfree(objp: data);
148
149out:
150 return err;
151}
152
153static void ops_pre_exit_list(const struct pernet_operations *ops,
154 struct list_head *net_exit_list)
155{
156 struct net *net;
157
158 if (ops->pre_exit) {
159 list_for_each_entry(net, net_exit_list, exit_list)
160 ops->pre_exit(net);
161 }
162}
163
164static void ops_exit_list(const struct pernet_operations *ops,
165 struct list_head *net_exit_list)
166{
167 struct net *net;
168 if (ops->exit) {
169 list_for_each_entry(net, net_exit_list, exit_list) {
170 ops->exit(net);
171 cond_resched();
172 }
173 }
174 if (ops->exit_batch)
175 ops->exit_batch(net_exit_list);
176}
177
178static void ops_free_list(const struct pernet_operations *ops,
179 struct list_head *net_exit_list)
180{
181 struct net *net;
182 if (ops->size && ops->id) {
183 list_for_each_entry(net, net_exit_list, exit_list)
184 kfree(objp: net_generic(net, id: *ops->id));
185 }
186}
187
188/* should be called with nsid_lock held */
189static int alloc_netid(struct net *net, struct net *peer, int reqid)
190{
191 int min = 0, max = 0;
192
193 if (reqid >= 0) {
194 min = reqid;
195 max = reqid + 1;
196 }
197
198 return idr_alloc(&net->netns_ids, ptr: peer, start: min, end: max, GFP_ATOMIC);
199}
200
201/* This function is used by idr_for_each(). If net is equal to peer, the
202 * function returns the id so that idr_for_each() stops. Because we cannot
203 * returns the id 0 (idr_for_each() will not stop), we return the magic value
204 * NET_ID_ZERO (-1) for it.
205 */
206#define NET_ID_ZERO -1
207static int net_eq_idr(int id, void *net, void *peer)
208{
209 if (net_eq(net1: net, net2: peer))
210 return id ? : NET_ID_ZERO;
211 return 0;
212}
213
214/* Must be called from RCU-critical section or with nsid_lock held */
215static int __peernet2id(const struct net *net, struct net *peer)
216{
217 int id = idr_for_each(&net->netns_ids, fn: net_eq_idr, data: peer);
218
219 /* Magic value for id 0. */
220 if (id == NET_ID_ZERO)
221 return 0;
222 if (id > 0)
223 return id;
224
225 return NETNSA_NSID_NOT_ASSIGNED;
226}
227
228static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
229 struct nlmsghdr *nlh, gfp_t gfp);
230/* This function returns the id of a peer netns. If no id is assigned, one will
231 * be allocated and returned.
232 */
233int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp)
234{
235 int id;
236
237 if (refcount_read(r: &net->ns.count) == 0)
238 return NETNSA_NSID_NOT_ASSIGNED;
239
240 spin_lock_bh(lock: &net->nsid_lock);
241 id = __peernet2id(net, peer);
242 if (id >= 0) {
243 spin_unlock_bh(lock: &net->nsid_lock);
244 return id;
245 }
246
247 /* When peer is obtained from RCU lists, we may race with
248 * its cleanup. Check whether it's alive, and this guarantees
249 * we never hash a peer back to net->netns_ids, after it has
250 * just been idr_remove()'d from there in cleanup_net().
251 */
252 if (!maybe_get_net(net: peer)) {
253 spin_unlock_bh(lock: &net->nsid_lock);
254 return NETNSA_NSID_NOT_ASSIGNED;
255 }
256
257 id = alloc_netid(net, peer, reqid: -1);
258 spin_unlock_bh(lock: &net->nsid_lock);
259
260 put_net(net: peer);
261 if (id < 0)
262 return NETNSA_NSID_NOT_ASSIGNED;
263
264 rtnl_net_notifyid(net, RTM_NEWNSID, id, portid: 0, NULL, gfp);
265
266 return id;
267}
268EXPORT_SYMBOL_GPL(peernet2id_alloc);
269
270/* This function returns, if assigned, the id of a peer netns. */
271int peernet2id(const struct net *net, struct net *peer)
272{
273 int id;
274
275 rcu_read_lock();
276 id = __peernet2id(net, peer);
277 rcu_read_unlock();
278
279 return id;
280}
281EXPORT_SYMBOL(peernet2id);
282
283/* This function returns true is the peer netns has an id assigned into the
284 * current netns.
285 */
286bool peernet_has_id(const struct net *net, struct net *peer)
287{
288 return peernet2id(net, peer) >= 0;
289}
290
291struct net *get_net_ns_by_id(const struct net *net, int id)
292{
293 struct net *peer;
294
295 if (id < 0)
296 return NULL;
297
298 rcu_read_lock();
299 peer = idr_find(&net->netns_ids, id);
300 if (peer)
301 peer = maybe_get_net(net: peer);
302 rcu_read_unlock();
303
304 return peer;
305}
306EXPORT_SYMBOL_GPL(get_net_ns_by_id);
307
308/* init code that must occur even if setup_net() is not called. */
309static __net_init void preinit_net(struct net *net)
310{
311 ref_tracker_dir_init(dir: &net->notrefcnt_tracker, quarantine_count: 128, name: "net notrefcnt");
312}
313
314/*
315 * setup_net runs the initializers for the network namespace object.
316 */
317static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
318{
319 /* Must be called with pernet_ops_rwsem held */
320 const struct pernet_operations *ops, *saved_ops;
321 int error = 0;
322 LIST_HEAD(net_exit_list);
323
324 refcount_set(r: &net->ns.count, n: 1);
325 ref_tracker_dir_init(dir: &net->refcnt_tracker, quarantine_count: 128, name: "net refcnt");
326
327 refcount_set(r: &net->passive, n: 1);
328 get_random_bytes(buf: &net->hash_mix, len: sizeof(u32));
329 preempt_disable();
330 net->net_cookie = gen_cookie_next(gc: &net_cookie);
331 preempt_enable();
332 net->dev_base_seq = 1;
333 net->user_ns = user_ns;
334 idr_init(idr: &net->netns_ids);
335 spin_lock_init(&net->nsid_lock);
336 mutex_init(&net->ipv4.ra_mutex);
337
338 list_for_each_entry(ops, &pernet_list, list) {
339 error = ops_init(ops, net);
340 if (error < 0)
341 goto out_undo;
342 }
343 down_write(sem: &net_rwsem);
344 list_add_tail_rcu(new: &net->list, head: &net_namespace_list);
345 up_write(sem: &net_rwsem);
346out:
347 return error;
348
349out_undo:
350 /* Walk through the list backwards calling the exit functions
351 * for the pernet modules whose init functions did not fail.
352 */
353 list_add(new: &net->exit_list, head: &net_exit_list);
354 saved_ops = ops;
355 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
356 ops_pre_exit_list(ops, net_exit_list: &net_exit_list);
357
358 synchronize_rcu();
359
360 ops = saved_ops;
361 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
362 ops_exit_list(ops, net_exit_list: &net_exit_list);
363
364 ops = saved_ops;
365 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
366 ops_free_list(ops, net_exit_list: &net_exit_list);
367
368 rcu_barrier();
369 goto out;
370}
371
372static int __net_init net_defaults_init_net(struct net *net)
373{
374 net->core.sysctl_somaxconn = SOMAXCONN;
375 net->core.sysctl_txrehash = SOCK_TXREHASH_ENABLED;
376
377 return 0;
378}
379
380static struct pernet_operations net_defaults_ops = {
381 .init = net_defaults_init_net,
382};
383
384static __init int net_defaults_init(void)
385{
386 if (register_pernet_subsys(&net_defaults_ops))
387 panic(fmt: "Cannot initialize net default settings");
388
389 return 0;
390}
391
392core_initcall(net_defaults_init);
393
394#ifdef CONFIG_NET_NS
395static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
396{
397 return inc_ucount(ns, current_euid(), type: UCOUNT_NET_NAMESPACES);
398}
399
400static void dec_net_namespaces(struct ucounts *ucounts)
401{
402 dec_ucount(ucounts, type: UCOUNT_NET_NAMESPACES);
403}
404
405static struct kmem_cache *net_cachep __ro_after_init;
406static struct workqueue_struct *netns_wq;
407
408static struct net *net_alloc(void)
409{
410 struct net *net = NULL;
411 struct net_generic *ng;
412
413 ng = net_alloc_generic();
414 if (!ng)
415 goto out;
416
417 net = kmem_cache_zalloc(k: net_cachep, GFP_KERNEL);
418 if (!net)
419 goto out_free;
420
421#ifdef CONFIG_KEYS
422 net->key_domain = kzalloc(size: sizeof(struct key_tag), GFP_KERNEL);
423 if (!net->key_domain)
424 goto out_free_2;
425 refcount_set(r: &net->key_domain->usage, n: 1);
426#endif
427
428 rcu_assign_pointer(net->gen, ng);
429out:
430 return net;
431
432#ifdef CONFIG_KEYS
433out_free_2:
434 kmem_cache_free(s: net_cachep, objp: net);
435 net = NULL;
436#endif
437out_free:
438 kfree(objp: ng);
439 goto out;
440}
441
442static void net_free(struct net *net)
443{
444 if (refcount_dec_and_test(r: &net->passive)) {
445 kfree(rcu_access_pointer(net->gen));
446
447 /* There should not be any trackers left there. */
448 ref_tracker_dir_exit(dir: &net->notrefcnt_tracker);
449
450 kmem_cache_free(s: net_cachep, objp: net);
451 }
452}
453
454void net_drop_ns(void *p)
455{
456 struct net *net = (struct net *)p;
457
458 if (net)
459 net_free(net);
460}
461
462struct net *copy_net_ns(unsigned long flags,
463 struct user_namespace *user_ns, struct net *old_net)
464{
465 struct ucounts *ucounts;
466 struct net *net;
467 int rv;
468
469 if (!(flags & CLONE_NEWNET))
470 return get_net(net: old_net);
471
472 ucounts = inc_net_namespaces(ns: user_ns);
473 if (!ucounts)
474 return ERR_PTR(error: -ENOSPC);
475
476 net = net_alloc();
477 if (!net) {
478 rv = -ENOMEM;
479 goto dec_ucounts;
480 }
481
482 preinit_net(net);
483 refcount_set(r: &net->passive, n: 1);
484 net->ucounts = ucounts;
485 get_user_ns(ns: user_ns);
486
487 rv = down_read_killable(sem: &pernet_ops_rwsem);
488 if (rv < 0)
489 goto put_userns;
490
491 rv = setup_net(net, user_ns);
492
493 up_read(sem: &pernet_ops_rwsem);
494
495 if (rv < 0) {
496put_userns:
497#ifdef CONFIG_KEYS
498 key_remove_domain(domain_tag: net->key_domain);
499#endif
500 put_user_ns(ns: user_ns);
501 net_free(net);
502dec_ucounts:
503 dec_net_namespaces(ucounts);
504 return ERR_PTR(error: rv);
505 }
506 return net;
507}
508
509/**
510 * net_ns_get_ownership - get sysfs ownership data for @net
511 * @net: network namespace in question (can be NULL)
512 * @uid: kernel user ID for sysfs objects
513 * @gid: kernel group ID for sysfs objects
514 *
515 * Returns the uid/gid pair of root in the user namespace associated with the
516 * given network namespace.
517 */
518void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
519{
520 if (net) {
521 kuid_t ns_root_uid = make_kuid(from: net->user_ns, uid: 0);
522 kgid_t ns_root_gid = make_kgid(from: net->user_ns, gid: 0);
523
524 if (uid_valid(uid: ns_root_uid))
525 *uid = ns_root_uid;
526
527 if (gid_valid(gid: ns_root_gid))
528 *gid = ns_root_gid;
529 } else {
530 *uid = GLOBAL_ROOT_UID;
531 *gid = GLOBAL_ROOT_GID;
532 }
533}
534EXPORT_SYMBOL_GPL(net_ns_get_ownership);
535
536static void unhash_nsid(struct net *net, struct net *last)
537{
538 struct net *tmp;
539 /* This function is only called from cleanup_net() work,
540 * and this work is the only process, that may delete
541 * a net from net_namespace_list. So, when the below
542 * is executing, the list may only grow. Thus, we do not
543 * use for_each_net_rcu() or net_rwsem.
544 */
545 for_each_net(tmp) {
546 int id;
547
548 spin_lock_bh(lock: &tmp->nsid_lock);
549 id = __peernet2id(net: tmp, peer: net);
550 if (id >= 0)
551 idr_remove(&tmp->netns_ids, id);
552 spin_unlock_bh(lock: &tmp->nsid_lock);
553 if (id >= 0)
554 rtnl_net_notifyid(net: tmp, RTM_DELNSID, id, portid: 0, NULL,
555 GFP_KERNEL);
556 if (tmp == last)
557 break;
558 }
559 spin_lock_bh(lock: &net->nsid_lock);
560 idr_destroy(&net->netns_ids);
561 spin_unlock_bh(lock: &net->nsid_lock);
562}
563
564static LLIST_HEAD(cleanup_list);
565
566static void cleanup_net(struct work_struct *work)
567{
568 const struct pernet_operations *ops;
569 struct net *net, *tmp, *last;
570 struct llist_node *net_kill_list;
571 LIST_HEAD(net_exit_list);
572
573 /* Atomically snapshot the list of namespaces to cleanup */
574 net_kill_list = llist_del_all(head: &cleanup_list);
575
576 down_read(sem: &pernet_ops_rwsem);
577
578 /* Don't let anyone else find us. */
579 down_write(sem: &net_rwsem);
580 llist_for_each_entry(net, net_kill_list, cleanup_list)
581 list_del_rcu(entry: &net->list);
582 /* Cache last net. After we unlock rtnl, no one new net
583 * added to net_namespace_list can assign nsid pointer
584 * to a net from net_kill_list (see peernet2id_alloc()).
585 * So, we skip them in unhash_nsid().
586 *
587 * Note, that unhash_nsid() does not delete nsid links
588 * between net_kill_list's nets, as they've already
589 * deleted from net_namespace_list. But, this would be
590 * useless anyway, as netns_ids are destroyed there.
591 */
592 last = list_last_entry(&net_namespace_list, struct net, list);
593 up_write(sem: &net_rwsem);
594
595 llist_for_each_entry(net, net_kill_list, cleanup_list) {
596 unhash_nsid(net, last);
597 list_add_tail(new: &net->exit_list, head: &net_exit_list);
598 }
599
600 /* Run all of the network namespace pre_exit methods */
601 list_for_each_entry_reverse(ops, &pernet_list, list)
602 ops_pre_exit_list(ops, net_exit_list: &net_exit_list);
603
604 /*
605 * Another CPU might be rcu-iterating the list, wait for it.
606 * This needs to be before calling the exit() notifiers, so
607 * the rcu_barrier() below isn't sufficient alone.
608 * Also the pre_exit() and exit() methods need this barrier.
609 */
610 synchronize_rcu();
611
612 /* Run all of the network namespace exit methods */
613 list_for_each_entry_reverse(ops, &pernet_list, list)
614 ops_exit_list(ops, net_exit_list: &net_exit_list);
615
616 /* Free the net generic variables */
617 list_for_each_entry_reverse(ops, &pernet_list, list)
618 ops_free_list(ops, net_exit_list: &net_exit_list);
619
620 up_read(sem: &pernet_ops_rwsem);
621
622 /* Ensure there are no outstanding rcu callbacks using this
623 * network namespace.
624 */
625 rcu_barrier();
626
627 /* Finally it is safe to free my network namespace structure */
628 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
629 list_del_init(entry: &net->exit_list);
630 dec_net_namespaces(ucounts: net->ucounts);
631#ifdef CONFIG_KEYS
632 key_remove_domain(domain_tag: net->key_domain);
633#endif
634 put_user_ns(ns: net->user_ns);
635 net_free(net);
636 }
637}
638
639/**
640 * net_ns_barrier - wait until concurrent net_cleanup_work is done
641 *
642 * cleanup_net runs from work queue and will first remove namespaces
643 * from the global list, then run net exit functions.
644 *
645 * Call this in module exit path to make sure that all netns
646 * ->exit ops have been invoked before the function is removed.
647 */
648void net_ns_barrier(void)
649{
650 down_write(sem: &pernet_ops_rwsem);
651 up_write(sem: &pernet_ops_rwsem);
652}
653EXPORT_SYMBOL(net_ns_barrier);
654
655static DECLARE_WORK(net_cleanup_work, cleanup_net);
656
657void __put_net(struct net *net)
658{
659 ref_tracker_dir_exit(dir: &net->refcnt_tracker);
660 /* Cleanup the network namespace in process context */
661 if (llist_add(new: &net->cleanup_list, head: &cleanup_list))
662 queue_work(wq: netns_wq, work: &net_cleanup_work);
663}
664EXPORT_SYMBOL_GPL(__put_net);
665
666/**
667 * get_net_ns - increment the refcount of the network namespace
668 * @ns: common namespace (net)
669 *
670 * Returns the net's common namespace.
671 */
672struct ns_common *get_net_ns(struct ns_common *ns)
673{
674 return &get_net(container_of(ns, struct net, ns))->ns;
675}
676EXPORT_SYMBOL_GPL(get_net_ns);
677
678struct net *get_net_ns_by_fd(int fd)
679{
680 struct fd f = fdget(fd);
681 struct net *net = ERR_PTR(error: -EINVAL);
682
683 if (!f.file)
684 return ERR_PTR(error: -EBADF);
685
686 if (proc_ns_file(file: f.file)) {
687 struct ns_common *ns = get_proc_ns(file_inode(f.file));
688 if (ns->ops == &netns_operations)
689 net = get_net(container_of(ns, struct net, ns));
690 }
691 fdput(fd: f);
692
693 return net;
694}
695EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
696#endif
697
698struct net *get_net_ns_by_pid(pid_t pid)
699{
700 struct task_struct *tsk;
701 struct net *net;
702
703 /* Lookup the network namespace */
704 net = ERR_PTR(error: -ESRCH);
705 rcu_read_lock();
706 tsk = find_task_by_vpid(nr: pid);
707 if (tsk) {
708 struct nsproxy *nsproxy;
709 task_lock(p: tsk);
710 nsproxy = tsk->nsproxy;
711 if (nsproxy)
712 net = get_net(net: nsproxy->net_ns);
713 task_unlock(p: tsk);
714 }
715 rcu_read_unlock();
716 return net;
717}
718EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
719
720static __net_init int net_ns_net_init(struct net *net)
721{
722#ifdef CONFIG_NET_NS
723 net->ns.ops = &netns_operations;
724#endif
725 return ns_alloc_inum(ns: &net->ns);
726}
727
728static __net_exit void net_ns_net_exit(struct net *net)
729{
730 ns_free_inum(&net->ns);
731}
732
733static struct pernet_operations __net_initdata net_ns_ops = {
734 .init = net_ns_net_init,
735 .exit = net_ns_net_exit,
736};
737
738static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
739 [NETNSA_NONE] = { .type = NLA_UNSPEC },
740 [NETNSA_NSID] = { .type = NLA_S32 },
741 [NETNSA_PID] = { .type = NLA_U32 },
742 [NETNSA_FD] = { .type = NLA_U32 },
743 [NETNSA_TARGET_NSID] = { .type = NLA_S32 },
744};
745
746static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
747 struct netlink_ext_ack *extack)
748{
749 struct net *net = sock_net(sk: skb->sk);
750 struct nlattr *tb[NETNSA_MAX + 1];
751 struct nlattr *nla;
752 struct net *peer;
753 int nsid, err;
754
755 err = nlmsg_parse_deprecated(nlh, hdrlen: sizeof(struct rtgenmsg), tb,
756 NETNSA_MAX, policy: rtnl_net_policy, extack);
757 if (err < 0)
758 return err;
759 if (!tb[NETNSA_NSID]) {
760 NL_SET_ERR_MSG(extack, "nsid is missing");
761 return -EINVAL;
762 }
763 nsid = nla_get_s32(nla: tb[NETNSA_NSID]);
764
765 if (tb[NETNSA_PID]) {
766 peer = get_net_ns_by_pid(nla_get_u32(nla: tb[NETNSA_PID]));
767 nla = tb[NETNSA_PID];
768 } else if (tb[NETNSA_FD]) {
769 peer = get_net_ns_by_fd(nla_get_u32(nla: tb[NETNSA_FD]));
770 nla = tb[NETNSA_FD];
771 } else {
772 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
773 return -EINVAL;
774 }
775 if (IS_ERR(ptr: peer)) {
776 NL_SET_BAD_ATTR(extack, nla);
777 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
778 return PTR_ERR(ptr: peer);
779 }
780
781 spin_lock_bh(lock: &net->nsid_lock);
782 if (__peernet2id(net, peer) >= 0) {
783 spin_unlock_bh(lock: &net->nsid_lock);
784 err = -EEXIST;
785 NL_SET_BAD_ATTR(extack, nla);
786 NL_SET_ERR_MSG(extack,
787 "Peer netns already has a nsid assigned");
788 goto out;
789 }
790
791 err = alloc_netid(net, peer, reqid: nsid);
792 spin_unlock_bh(lock: &net->nsid_lock);
793 if (err >= 0) {
794 rtnl_net_notifyid(net, RTM_NEWNSID, id: err, NETLINK_CB(skb).portid,
795 nlh, GFP_KERNEL);
796 err = 0;
797 } else if (err == -ENOSPC && nsid >= 0) {
798 err = -EEXIST;
799 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
800 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
801 }
802out:
803 put_net(net: peer);
804 return err;
805}
806
807static int rtnl_net_get_size(void)
808{
809 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
810 + nla_total_size(payload: sizeof(s32)) /* NETNSA_NSID */
811 + nla_total_size(payload: sizeof(s32)) /* NETNSA_CURRENT_NSID */
812 ;
813}
814
815struct net_fill_args {
816 u32 portid;
817 u32 seq;
818 int flags;
819 int cmd;
820 int nsid;
821 bool add_ref;
822 int ref_nsid;
823};
824
825static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
826{
827 struct nlmsghdr *nlh;
828 struct rtgenmsg *rth;
829
830 nlh = nlmsg_put(skb, portid: args->portid, seq: args->seq, type: args->cmd, payload: sizeof(*rth),
831 flags: args->flags);
832 if (!nlh)
833 return -EMSGSIZE;
834
835 rth = nlmsg_data(nlh);
836 rth->rtgen_family = AF_UNSPEC;
837
838 if (nla_put_s32(skb, attrtype: NETNSA_NSID, value: args->nsid))
839 goto nla_put_failure;
840
841 if (args->add_ref &&
842 nla_put_s32(skb, attrtype: NETNSA_CURRENT_NSID, value: args->ref_nsid))
843 goto nla_put_failure;
844
845 nlmsg_end(skb, nlh);
846 return 0;
847
848nla_put_failure:
849 nlmsg_cancel(skb, nlh);
850 return -EMSGSIZE;
851}
852
853static int rtnl_net_valid_getid_req(struct sk_buff *skb,
854 const struct nlmsghdr *nlh,
855 struct nlattr **tb,
856 struct netlink_ext_ack *extack)
857{
858 int i, err;
859
860 if (!netlink_strict_get_check(skb))
861 return nlmsg_parse_deprecated(nlh, hdrlen: sizeof(struct rtgenmsg),
862 tb, NETNSA_MAX, policy: rtnl_net_policy,
863 extack);
864
865 err = nlmsg_parse_deprecated_strict(nlh, hdrlen: sizeof(struct rtgenmsg), tb,
866 NETNSA_MAX, policy: rtnl_net_policy,
867 extack);
868 if (err)
869 return err;
870
871 for (i = 0; i <= NETNSA_MAX; i++) {
872 if (!tb[i])
873 continue;
874
875 switch (i) {
876 case NETNSA_PID:
877 case NETNSA_FD:
878 case NETNSA_NSID:
879 case NETNSA_TARGET_NSID:
880 break;
881 default:
882 NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
883 return -EINVAL;
884 }
885 }
886
887 return 0;
888}
889
890static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
891 struct netlink_ext_ack *extack)
892{
893 struct net *net = sock_net(sk: skb->sk);
894 struct nlattr *tb[NETNSA_MAX + 1];
895 struct net_fill_args fillargs = {
896 .portid = NETLINK_CB(skb).portid,
897 .seq = nlh->nlmsg_seq,
898 .cmd = RTM_NEWNSID,
899 };
900 struct net *peer, *target = net;
901 struct nlattr *nla;
902 struct sk_buff *msg;
903 int err;
904
905 err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
906 if (err < 0)
907 return err;
908 if (tb[NETNSA_PID]) {
909 peer = get_net_ns_by_pid(nla_get_u32(nla: tb[NETNSA_PID]));
910 nla = tb[NETNSA_PID];
911 } else if (tb[NETNSA_FD]) {
912 peer = get_net_ns_by_fd(nla_get_u32(nla: tb[NETNSA_FD]));
913 nla = tb[NETNSA_FD];
914 } else if (tb[NETNSA_NSID]) {
915 peer = get_net_ns_by_id(net, nla_get_s32(nla: tb[NETNSA_NSID]));
916 if (!peer)
917 peer = ERR_PTR(error: -ENOENT);
918 nla = tb[NETNSA_NSID];
919 } else {
920 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
921 return -EINVAL;
922 }
923
924 if (IS_ERR(ptr: peer)) {
925 NL_SET_BAD_ATTR(extack, nla);
926 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
927 return PTR_ERR(ptr: peer);
928 }
929
930 if (tb[NETNSA_TARGET_NSID]) {
931 int id = nla_get_s32(nla: tb[NETNSA_TARGET_NSID]);
932
933 target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, netnsid: id);
934 if (IS_ERR(ptr: target)) {
935 NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
936 NL_SET_ERR_MSG(extack,
937 "Target netns reference is invalid");
938 err = PTR_ERR(ptr: target);
939 goto out;
940 }
941 fillargs.add_ref = true;
942 fillargs.ref_nsid = peernet2id(net, peer);
943 }
944
945 msg = nlmsg_new(payload: rtnl_net_get_size(), GFP_KERNEL);
946 if (!msg) {
947 err = -ENOMEM;
948 goto out;
949 }
950
951 fillargs.nsid = peernet2id(target, peer);
952 err = rtnl_net_fill(skb: msg, args: &fillargs);
953 if (err < 0)
954 goto err_out;
955
956 err = rtnl_unicast(skb: msg, net, NETLINK_CB(skb).portid);
957 goto out;
958
959err_out:
960 nlmsg_free(skb: msg);
961out:
962 if (fillargs.add_ref)
963 put_net(net: target);
964 put_net(net: peer);
965 return err;
966}
967
968struct rtnl_net_dump_cb {
969 struct net *tgt_net;
970 struct net *ref_net;
971 struct sk_buff *skb;
972 struct net_fill_args fillargs;
973 int idx;
974 int s_idx;
975};
976
977/* Runs in RCU-critical section. */
978static int rtnl_net_dumpid_one(int id, void *peer, void *data)
979{
980 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
981 int ret;
982
983 if (net_cb->idx < net_cb->s_idx)
984 goto cont;
985
986 net_cb->fillargs.nsid = id;
987 if (net_cb->fillargs.add_ref)
988 net_cb->fillargs.ref_nsid = __peernet2id(net: net_cb->ref_net, peer);
989 ret = rtnl_net_fill(skb: net_cb->skb, args: &net_cb->fillargs);
990 if (ret < 0)
991 return ret;
992
993cont:
994 net_cb->idx++;
995 return 0;
996}
997
998static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
999 struct rtnl_net_dump_cb *net_cb,
1000 struct netlink_callback *cb)
1001{
1002 struct netlink_ext_ack *extack = cb->extack;
1003 struct nlattr *tb[NETNSA_MAX + 1];
1004 int err, i;
1005
1006 err = nlmsg_parse_deprecated_strict(nlh, hdrlen: sizeof(struct rtgenmsg), tb,
1007 NETNSA_MAX, policy: rtnl_net_policy,
1008 extack);
1009 if (err < 0)
1010 return err;
1011
1012 for (i = 0; i <= NETNSA_MAX; i++) {
1013 if (!tb[i])
1014 continue;
1015
1016 if (i == NETNSA_TARGET_NSID) {
1017 struct net *net;
1018
1019 net = rtnl_get_net_ns_capable(sk, netnsid: nla_get_s32(nla: tb[i]));
1020 if (IS_ERR(ptr: net)) {
1021 NL_SET_BAD_ATTR(extack, tb[i]);
1022 NL_SET_ERR_MSG(extack,
1023 "Invalid target network namespace id");
1024 return PTR_ERR(ptr: net);
1025 }
1026 net_cb->fillargs.add_ref = true;
1027 net_cb->ref_net = net_cb->tgt_net;
1028 net_cb->tgt_net = net;
1029 } else {
1030 NL_SET_BAD_ATTR(extack, tb[i]);
1031 NL_SET_ERR_MSG(extack,
1032 "Unsupported attribute in dump request");
1033 return -EINVAL;
1034 }
1035 }
1036
1037 return 0;
1038}
1039
1040static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1041{
1042 struct rtnl_net_dump_cb net_cb = {
1043 .tgt_net = sock_net(sk: skb->sk),
1044 .skb = skb,
1045 .fillargs = {
1046 .portid = NETLINK_CB(cb->skb).portid,
1047 .seq = cb->nlh->nlmsg_seq,
1048 .flags = NLM_F_MULTI,
1049 .cmd = RTM_NEWNSID,
1050 },
1051 .idx = 0,
1052 .s_idx = cb->args[0],
1053 };
1054 int err = 0;
1055
1056 if (cb->strict_check) {
1057 err = rtnl_valid_dump_net_req(nlh: cb->nlh, sk: skb->sk, net_cb: &net_cb, cb);
1058 if (err < 0)
1059 goto end;
1060 }
1061
1062 rcu_read_lock();
1063 idr_for_each(&net_cb.tgt_net->netns_ids, fn: rtnl_net_dumpid_one, data: &net_cb);
1064 rcu_read_unlock();
1065
1066 cb->args[0] = net_cb.idx;
1067end:
1068 if (net_cb.fillargs.add_ref)
1069 put_net(net: net_cb.tgt_net);
1070 return err < 0 ? err : skb->len;
1071}
1072
1073static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1074 struct nlmsghdr *nlh, gfp_t gfp)
1075{
1076 struct net_fill_args fillargs = {
1077 .portid = portid,
1078 .seq = nlh ? nlh->nlmsg_seq : 0,
1079 .cmd = cmd,
1080 .nsid = id,
1081 };
1082 struct sk_buff *msg;
1083 int err = -ENOMEM;
1084
1085 msg = nlmsg_new(payload: rtnl_net_get_size(), flags: gfp);
1086 if (!msg)
1087 goto out;
1088
1089 err = rtnl_net_fill(skb: msg, args: &fillargs);
1090 if (err < 0)
1091 goto err_out;
1092
1093 rtnl_notify(skb: msg, net, pid: portid, RTNLGRP_NSID, nlh, flags: gfp);
1094 return;
1095
1096err_out:
1097 nlmsg_free(skb: msg);
1098out:
1099 rtnl_set_sk_err(net, RTNLGRP_NSID, error: err);
1100}
1101
1102void __init net_ns_init(void)
1103{
1104 struct net_generic *ng;
1105
1106#ifdef CONFIG_NET_NS
1107 net_cachep = kmem_cache_create(name: "net_namespace", size: sizeof(struct net),
1108 SMP_CACHE_BYTES,
1109 SLAB_PANIC|SLAB_ACCOUNT, NULL);
1110
1111 /* Create workqueue for cleanup */
1112 netns_wq = create_singlethread_workqueue("netns");
1113 if (!netns_wq)
1114 panic(fmt: "Could not create netns workq");
1115#endif
1116
1117 ng = net_alloc_generic();
1118 if (!ng)
1119 panic(fmt: "Could not allocate generic netns");
1120
1121 rcu_assign_pointer(init_net.gen, ng);
1122
1123#ifdef CONFIG_KEYS
1124 init_net.key_domain = &init_net_key_domain;
1125#endif
1126 down_write(sem: &pernet_ops_rwsem);
1127 preinit_net(net: &init_net);
1128 if (setup_net(net: &init_net, user_ns: &init_user_ns))
1129 panic(fmt: "Could not setup the initial network namespace");
1130
1131 init_net_initialized = true;
1132 up_write(sem: &pernet_ops_rwsem);
1133
1134 if (register_pernet_subsys(&net_ns_ops))
1135 panic(fmt: "Could not register network namespace subsystems");
1136
1137 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
1138 flags: RTNL_FLAG_DOIT_UNLOCKED);
1139 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
1140 flags: RTNL_FLAG_DOIT_UNLOCKED);
1141}
1142
1143static void free_exit_list(struct pernet_operations *ops, struct list_head *net_exit_list)
1144{
1145 ops_pre_exit_list(ops, net_exit_list);
1146 synchronize_rcu();
1147 ops_exit_list(ops, net_exit_list);
1148 ops_free_list(ops, net_exit_list);
1149}
1150
1151#ifdef CONFIG_NET_NS
1152static int __register_pernet_operations(struct list_head *list,
1153 struct pernet_operations *ops)
1154{
1155 struct net *net;
1156 int error;
1157 LIST_HEAD(net_exit_list);
1158
1159 list_add_tail(new: &ops->list, head: list);
1160 if (ops->init || (ops->id && ops->size)) {
1161 /* We held write locked pernet_ops_rwsem, and parallel
1162 * setup_net() and cleanup_net() are not possible.
1163 */
1164 for_each_net(net) {
1165 error = ops_init(ops, net);
1166 if (error)
1167 goto out_undo;
1168 list_add_tail(new: &net->exit_list, head: &net_exit_list);
1169 }
1170 }
1171 return 0;
1172
1173out_undo:
1174 /* If I have an error cleanup all namespaces I initialized */
1175 list_del(entry: &ops->list);
1176 free_exit_list(ops, net_exit_list: &net_exit_list);
1177 return error;
1178}
1179
1180static void __unregister_pernet_operations(struct pernet_operations *ops)
1181{
1182 struct net *net;
1183 LIST_HEAD(net_exit_list);
1184
1185 list_del(entry: &ops->list);
1186 /* See comment in __register_pernet_operations() */
1187 for_each_net(net)
1188 list_add_tail(new: &net->exit_list, head: &net_exit_list);
1189
1190 free_exit_list(ops, net_exit_list: &net_exit_list);
1191}
1192
1193#else
1194
1195static int __register_pernet_operations(struct list_head *list,
1196 struct pernet_operations *ops)
1197{
1198 if (!init_net_initialized) {
1199 list_add_tail(&ops->list, list);
1200 return 0;
1201 }
1202
1203 return ops_init(ops, &init_net);
1204}
1205
1206static void __unregister_pernet_operations(struct pernet_operations *ops)
1207{
1208 if (!init_net_initialized) {
1209 list_del(&ops->list);
1210 } else {
1211 LIST_HEAD(net_exit_list);
1212 list_add(&init_net.exit_list, &net_exit_list);
1213 free_exit_list(ops, &net_exit_list);
1214 }
1215}
1216
1217#endif /* CONFIG_NET_NS */
1218
1219static DEFINE_IDA(net_generic_ids);
1220
1221static int register_pernet_operations(struct list_head *list,
1222 struct pernet_operations *ops)
1223{
1224 int error;
1225
1226 if (ops->id) {
1227 error = ida_alloc_min(ida: &net_generic_ids, MIN_PERNET_OPS_ID,
1228 GFP_KERNEL);
1229 if (error < 0)
1230 return error;
1231 *ops->id = error;
1232 max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
1233 }
1234 error = __register_pernet_operations(list, ops);
1235 if (error) {
1236 rcu_barrier();
1237 if (ops->id)
1238 ida_free(&net_generic_ids, id: *ops->id);
1239 }
1240
1241 return error;
1242}
1243
1244static void unregister_pernet_operations(struct pernet_operations *ops)
1245{
1246 __unregister_pernet_operations(ops);
1247 rcu_barrier();
1248 if (ops->id)
1249 ida_free(&net_generic_ids, id: *ops->id);
1250}
1251
1252/**
1253 * register_pernet_subsys - register a network namespace subsystem
1254 * @ops: pernet operations structure for the subsystem
1255 *
1256 * Register a subsystem which has init and exit functions
1257 * that are called when network namespaces are created and
1258 * destroyed respectively.
1259 *
1260 * When registered all network namespace init functions are
1261 * called for every existing network namespace. Allowing kernel
1262 * modules to have a race free view of the set of network namespaces.
1263 *
1264 * When a new network namespace is created all of the init
1265 * methods are called in the order in which they were registered.
1266 *
1267 * When a network namespace is destroyed all of the exit methods
1268 * are called in the reverse of the order with which they were
1269 * registered.
1270 */
1271int register_pernet_subsys(struct pernet_operations *ops)
1272{
1273 int error;
1274 down_write(sem: &pernet_ops_rwsem);
1275 error = register_pernet_operations(list: first_device, ops);
1276 up_write(sem: &pernet_ops_rwsem);
1277 return error;
1278}
1279EXPORT_SYMBOL_GPL(register_pernet_subsys);
1280
1281/**
1282 * unregister_pernet_subsys - unregister a network namespace subsystem
1283 * @ops: pernet operations structure to manipulate
1284 *
1285 * Remove the pernet operations structure from the list to be
1286 * used when network namespaces are created or destroyed. In
1287 * addition run the exit method for all existing network
1288 * namespaces.
1289 */
1290void unregister_pernet_subsys(struct pernet_operations *ops)
1291{
1292 down_write(sem: &pernet_ops_rwsem);
1293 unregister_pernet_operations(ops);
1294 up_write(sem: &pernet_ops_rwsem);
1295}
1296EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1297
1298/**
1299 * register_pernet_device - register a network namespace device
1300 * @ops: pernet operations structure for the subsystem
1301 *
1302 * Register a device which has init and exit functions
1303 * that are called when network namespaces are created and
1304 * destroyed respectively.
1305 *
1306 * When registered all network namespace init functions are
1307 * called for every existing network namespace. Allowing kernel
1308 * modules to have a race free view of the set of network namespaces.
1309 *
1310 * When a new network namespace is created all of the init
1311 * methods are called in the order in which they were registered.
1312 *
1313 * When a network namespace is destroyed all of the exit methods
1314 * are called in the reverse of the order with which they were
1315 * registered.
1316 */
1317int register_pernet_device(struct pernet_operations *ops)
1318{
1319 int error;
1320 down_write(sem: &pernet_ops_rwsem);
1321 error = register_pernet_operations(list: &pernet_list, ops);
1322 if (!error && (first_device == &pernet_list))
1323 first_device = &ops->list;
1324 up_write(sem: &pernet_ops_rwsem);
1325 return error;
1326}
1327EXPORT_SYMBOL_GPL(register_pernet_device);
1328
1329/**
1330 * unregister_pernet_device - unregister a network namespace netdevice
1331 * @ops: pernet operations structure to manipulate
1332 *
1333 * Remove the pernet operations structure from the list to be
1334 * used when network namespaces are created or destroyed. In
1335 * addition run the exit method for all existing network
1336 * namespaces.
1337 */
1338void unregister_pernet_device(struct pernet_operations *ops)
1339{
1340 down_write(sem: &pernet_ops_rwsem);
1341 if (&ops->list == first_device)
1342 first_device = first_device->next;
1343 unregister_pernet_operations(ops);
1344 up_write(sem: &pernet_ops_rwsem);
1345}
1346EXPORT_SYMBOL_GPL(unregister_pernet_device);
1347
1348#ifdef CONFIG_NET_NS
1349static struct ns_common *netns_get(struct task_struct *task)
1350{
1351 struct net *net = NULL;
1352 struct nsproxy *nsproxy;
1353
1354 task_lock(p: task);
1355 nsproxy = task->nsproxy;
1356 if (nsproxy)
1357 net = get_net(net: nsproxy->net_ns);
1358 task_unlock(p: task);
1359
1360 return net ? &net->ns : NULL;
1361}
1362
1363static inline struct net *to_net_ns(struct ns_common *ns)
1364{
1365 return container_of(ns, struct net, ns);
1366}
1367
1368static void netns_put(struct ns_common *ns)
1369{
1370 put_net(net: to_net_ns(ns));
1371}
1372
1373static int netns_install(struct nsset *nsset, struct ns_common *ns)
1374{
1375 struct nsproxy *nsproxy = nsset->nsproxy;
1376 struct net *net = to_net_ns(ns);
1377
1378 if (!ns_capable(ns: net->user_ns, CAP_SYS_ADMIN) ||
1379 !ns_capable(ns: nsset->cred->user_ns, CAP_SYS_ADMIN))
1380 return -EPERM;
1381
1382 put_net(net: nsproxy->net_ns);
1383 nsproxy->net_ns = get_net(net);
1384 return 0;
1385}
1386
1387static struct user_namespace *netns_owner(struct ns_common *ns)
1388{
1389 return to_net_ns(ns)->user_ns;
1390}
1391
1392const struct proc_ns_operations netns_operations = {
1393 .name = "net",
1394 .type = CLONE_NEWNET,
1395 .get = netns_get,
1396 .put = netns_put,
1397 .install = netns_install,
1398 .owner = netns_owner,
1399};
1400#endif
1401

source code of linux/net/core/net_namespace.c