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