Loading...
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/fs.h>
14#include <linux/proc_ns.h>
15#include <linux/file.h>
16#include <linux/export.h>
17#include <linux/user_namespace.h>
18#include <linux/net_namespace.h>
19#include <linux/sched/task.h>
20
21#include <net/sock.h>
22#include <net/netlink.h>
23#include <net/net_namespace.h>
24#include <net/netns/generic.h>
25
26/*
27 * Our network namespace constructor/destructor lists
28 */
29
30static LIST_HEAD(pernet_list);
31static struct list_head *first_device = &pernet_list;
32
33LIST_HEAD(net_namespace_list);
34EXPORT_SYMBOL_GPL(net_namespace_list);
35
36/* Protects net_namespace_list. Nests iside rtnl_lock() */
37DECLARE_RWSEM(net_rwsem);
38EXPORT_SYMBOL_GPL(net_rwsem);
39
40struct net init_net = {
41 .count = REFCOUNT_INIT(1),
42 .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head),
43};
44EXPORT_SYMBOL(init_net);
45
46static bool init_net_initialized;
47/*
48 * pernet_ops_rwsem: protects: pernet_list, net_generic_ids,
49 * init_net_initialized and first_device pointer.
50 * This is internal net namespace object. Please, don't use it
51 * outside.
52 */
53DECLARE_RWSEM(pernet_ops_rwsem);
54EXPORT_SYMBOL_GPL(pernet_ops_rwsem);
55
56#define MIN_PERNET_OPS_ID \
57 ((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *))
58
59#define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
60
61static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
62
63static struct net_generic *net_alloc_generic(void)
64{
65 struct net_generic *ng;
66 unsigned int generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
67
68 ng = kzalloc(generic_size, GFP_KERNEL);
69 if (ng)
70 ng->s.len = max_gen_ptrs;
71
72 return ng;
73}
74
75static int net_assign_generic(struct net *net, unsigned int id, void *data)
76{
77 struct net_generic *ng, *old_ng;
78
79 BUG_ON(id < MIN_PERNET_OPS_ID);
80
81 old_ng = rcu_dereference_protected(net->gen,
82 lockdep_is_held(&pernet_ops_rwsem));
83 if (old_ng->s.len > id) {
84 old_ng->ptr[id] = data;
85 return 0;
86 }
87
88 ng = net_alloc_generic();
89 if (ng == NULL)
90 return -ENOMEM;
91
92 /*
93 * Some synchronisation notes:
94 *
95 * The net_generic explores the net->gen array inside rcu
96 * read section. Besides once set the net->gen->ptr[x]
97 * pointer never changes (see rules in netns/generic.h).
98 *
99 * That said, we simply duplicate this array and schedule
100 * the old copy for kfree after a grace period.
101 */
102
103 memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
104 (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
105 ng->ptr[id] = data;
106
107 rcu_assign_pointer(net->gen, ng);
108 kfree_rcu(old_ng, s.rcu);
109 return 0;
110}
111
112static int ops_init(const struct pernet_operations *ops, struct net *net)
113{
114 int err = -ENOMEM;
115 void *data = NULL;
116
117 if (ops->id && ops->size) {
118 data = kzalloc(ops->size, GFP_KERNEL);
119 if (!data)
120 goto out;
121
122 err = net_assign_generic(net, *ops->id, data);
123 if (err)
124 goto cleanup;
125 }
126 err = 0;
127 if (ops->init)
128 err = ops->init(net);
129 if (!err)
130 return 0;
131
132cleanup:
133 kfree(data);
134
135out:
136 return err;
137}
138
139static void ops_free(const struct pernet_operations *ops, struct net *net)
140{
141 if (ops->id && ops->size) {
142 kfree(net_generic(net, *ops->id));
143 }
144}
145
146static void ops_exit_list(const struct pernet_operations *ops,
147 struct list_head *net_exit_list)
148{
149 struct net *net;
150 if (ops->exit) {
151 list_for_each_entry(net, net_exit_list, exit_list)
152 ops->exit(net);
153 }
154 if (ops->exit_batch)
155 ops->exit_batch(net_exit_list);
156}
157
158static void ops_free_list(const struct pernet_operations *ops,
159 struct list_head *net_exit_list)
160{
161 struct net *net;
162 if (ops->size && ops->id) {
163 list_for_each_entry(net, net_exit_list, exit_list)
164 ops_free(ops, net);
165 }
166}
167
168/* should be called with nsid_lock held */
169static int alloc_netid(struct net *net, struct net *peer, int reqid)
170{
171 int min = 0, max = 0;
172
173 if (reqid >= 0) {
174 min = reqid;
175 max = reqid + 1;
176 }
177
178 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
179}
180
181/* This function is used by idr_for_each(). If net is equal to peer, the
182 * function returns the id so that idr_for_each() stops. Because we cannot
183 * returns the id 0 (idr_for_each() will not stop), we return the magic value
184 * NET_ID_ZERO (-1) for it.
185 */
186#define NET_ID_ZERO -1
187static int net_eq_idr(int id, void *net, void *peer)
188{
189 if (net_eq(net, peer))
190 return id ? : NET_ID_ZERO;
191 return 0;
192}
193
194/* Should be called with nsid_lock held. If a new id is assigned, the bool alloc
195 * is set to true, thus the caller knows that the new id must be notified via
196 * rtnl.
197 */
198static int __peernet2id_alloc(struct net *net, struct net *peer, bool *alloc)
199{
200 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
201 bool alloc_it = *alloc;
202
203 *alloc = false;
204
205 /* Magic value for id 0. */
206 if (id == NET_ID_ZERO)
207 return 0;
208 if (id > 0)
209 return id;
210
211 if (alloc_it) {
212 id = alloc_netid(net, peer, -1);
213 *alloc = true;
214 return id >= 0 ? id : NETNSA_NSID_NOT_ASSIGNED;
215 }
216
217 return NETNSA_NSID_NOT_ASSIGNED;
218}
219
220/* should be called with nsid_lock held */
221static int __peernet2id(struct net *net, struct net *peer)
222{
223 bool no = false;
224
225 return __peernet2id_alloc(net, peer, &no);
226}
227
228static void rtnl_net_notifyid(struct net *net, int cmd, int id);
229/* This function returns the id of a peer netns. If no id is assigned, one will
230 * be allocated and returned.
231 */
232int peernet2id_alloc(struct net *net, struct net *peer)
233{
234 bool alloc = false, alive = false;
235 int id;
236
237 if (refcount_read(&net->count) == 0)
238 return NETNSA_NSID_NOT_ASSIGNED;
239 spin_lock_bh(&net->nsid_lock);
240 /*
241 * When peer is obtained from RCU lists, we may race with
242 * its cleanup. Check whether it's alive, and this guarantees
243 * we never hash a peer back to net->netns_ids, after it has
244 * just been idr_remove()'d from there in cleanup_net().
245 */
246 if (maybe_get_net(peer))
247 alive = alloc = true;
248 id = __peernet2id_alloc(net, peer, &alloc);
249 spin_unlock_bh(&net->nsid_lock);
250 if (alloc && id >= 0)
251 rtnl_net_notifyid(net, RTM_NEWNSID, id);
252 if (alive)
253 put_net(peer);
254 return id;
255}
256EXPORT_SYMBOL_GPL(peernet2id_alloc);
257
258/* This function returns, if assigned, the id of a peer netns. */
259int peernet2id(struct net *net, struct net *peer)
260{
261 int id;
262
263 spin_lock_bh(&net->nsid_lock);
264 id = __peernet2id(net, peer);
265 spin_unlock_bh(&net->nsid_lock);
266 return id;
267}
268EXPORT_SYMBOL(peernet2id);
269
270/* This function returns true is the peer netns has an id assigned into the
271 * current netns.
272 */
273bool peernet_has_id(struct net *net, struct net *peer)
274{
275 return peernet2id(net, peer) >= 0;
276}
277
278struct net *get_net_ns_by_id(struct net *net, int id)
279{
280 struct net *peer;
281
282 if (id < 0)
283 return NULL;
284
285 rcu_read_lock();
286 peer = idr_find(&net->netns_ids, id);
287 if (peer)
288 peer = maybe_get_net(peer);
289 rcu_read_unlock();
290
291 return peer;
292}
293
294/*
295 * setup_net runs the initializers for the network namespace object.
296 */
297static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
298{
299 /* Must be called with pernet_ops_rwsem held */
300 const struct pernet_operations *ops, *saved_ops;
301 int error = 0;
302 LIST_HEAD(net_exit_list);
303
304 refcount_set(&net->count, 1);
305 refcount_set(&net->passive, 1);
306 net->dev_base_seq = 1;
307 net->user_ns = user_ns;
308 idr_init(&net->netns_ids);
309 spin_lock_init(&net->nsid_lock);
310 mutex_init(&net->ipv4.ra_mutex);
311
312 list_for_each_entry(ops, &pernet_list, list) {
313 error = ops_init(ops, net);
314 if (error < 0)
315 goto out_undo;
316 }
317 down_write(&net_rwsem);
318 list_add_tail_rcu(&net->list, &net_namespace_list);
319 up_write(&net_rwsem);
320out:
321 return error;
322
323out_undo:
324 /* Walk through the list backwards calling the exit functions
325 * for the pernet modules whose init functions did not fail.
326 */
327 list_add(&net->exit_list, &net_exit_list);
328 saved_ops = ops;
329 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
330 ops_exit_list(ops, &net_exit_list);
331
332 ops = saved_ops;
333 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
334 ops_free_list(ops, &net_exit_list);
335
336 rcu_barrier();
337 goto out;
338}
339
340static int __net_init net_defaults_init_net(struct net *net)
341{
342 net->core.sysctl_somaxconn = SOMAXCONN;
343 return 0;
344}
345
346static struct pernet_operations net_defaults_ops = {
347 .init = net_defaults_init_net,
348};
349
350static __init int net_defaults_init(void)
351{
352 if (register_pernet_subsys(&net_defaults_ops))
353 panic("Cannot initialize net default settings");
354
355 return 0;
356}
357
358core_initcall(net_defaults_init);
359
360#ifdef CONFIG_NET_NS
361static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
362{
363 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
364}
365
366static void dec_net_namespaces(struct ucounts *ucounts)
367{
368 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
369}
370
371static struct kmem_cache *net_cachep __ro_after_init;
372static struct workqueue_struct *netns_wq;
373
374static struct net *net_alloc(void)
375{
376 struct net *net = NULL;
377 struct net_generic *ng;
378
379 ng = net_alloc_generic();
380 if (!ng)
381 goto out;
382
383 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
384 if (!net)
385 goto out_free;
386
387 rcu_assign_pointer(net->gen, ng);
388out:
389 return net;
390
391out_free:
392 kfree(ng);
393 goto out;
394}
395
396static void net_free(struct net *net)
397{
398 kfree(rcu_access_pointer(net->gen));
399 kmem_cache_free(net_cachep, net);
400}
401
402void net_drop_ns(void *p)
403{
404 struct net *ns = p;
405 if (ns && refcount_dec_and_test(&ns->passive))
406 net_free(ns);
407}
408
409struct net *copy_net_ns(unsigned long flags,
410 struct user_namespace *user_ns, struct net *old_net)
411{
412 struct ucounts *ucounts;
413 struct net *net;
414 int rv;
415
416 if (!(flags & CLONE_NEWNET))
417 return get_net(old_net);
418
419 ucounts = inc_net_namespaces(user_ns);
420 if (!ucounts)
421 return ERR_PTR(-ENOSPC);
422
423 net = net_alloc();
424 if (!net) {
425 rv = -ENOMEM;
426 goto dec_ucounts;
427 }
428 refcount_set(&net->passive, 1);
429 net->ucounts = ucounts;
430 get_user_ns(user_ns);
431
432 rv = down_read_killable(&pernet_ops_rwsem);
433 if (rv < 0)
434 goto put_userns;
435
436 rv = setup_net(net, user_ns);
437
438 up_read(&pernet_ops_rwsem);
439
440 if (rv < 0) {
441put_userns:
442 put_user_ns(user_ns);
443 net_drop_ns(net);
444dec_ucounts:
445 dec_net_namespaces(ucounts);
446 return ERR_PTR(rv);
447 }
448 return net;
449}
450
451static void unhash_nsid(struct net *net, struct net *last)
452{
453 struct net *tmp;
454 /* This function is only called from cleanup_net() work,
455 * and this work is the only process, that may delete
456 * a net from net_namespace_list. So, when the below
457 * is executing, the list may only grow. Thus, we do not
458 * use for_each_net_rcu() or net_rwsem.
459 */
460 for_each_net(tmp) {
461 int id;
462
463 spin_lock_bh(&tmp->nsid_lock);
464 id = __peernet2id(tmp, net);
465 if (id >= 0)
466 idr_remove(&tmp->netns_ids, id);
467 spin_unlock_bh(&tmp->nsid_lock);
468 if (id >= 0)
469 rtnl_net_notifyid(tmp, RTM_DELNSID, id);
470 if (tmp == last)
471 break;
472 }
473 spin_lock_bh(&net->nsid_lock);
474 idr_destroy(&net->netns_ids);
475 spin_unlock_bh(&net->nsid_lock);
476}
477
478static LLIST_HEAD(cleanup_list);
479
480static void cleanup_net(struct work_struct *work)
481{
482 const struct pernet_operations *ops;
483 struct net *net, *tmp, *last;
484 struct llist_node *net_kill_list;
485 LIST_HEAD(net_exit_list);
486
487 /* Atomically snapshot the list of namespaces to cleanup */
488 net_kill_list = llist_del_all(&cleanup_list);
489
490 down_read(&pernet_ops_rwsem);
491
492 /* Don't let anyone else find us. */
493 down_write(&net_rwsem);
494 llist_for_each_entry(net, net_kill_list, cleanup_list)
495 list_del_rcu(&net->list);
496 /* Cache last net. After we unlock rtnl, no one new net
497 * added to net_namespace_list can assign nsid pointer
498 * to a net from net_kill_list (see peernet2id_alloc()).
499 * So, we skip them in unhash_nsid().
500 *
501 * Note, that unhash_nsid() does not delete nsid links
502 * between net_kill_list's nets, as they've already
503 * deleted from net_namespace_list. But, this would be
504 * useless anyway, as netns_ids are destroyed there.
505 */
506 last = list_last_entry(&net_namespace_list, struct net, list);
507 up_write(&net_rwsem);
508
509 llist_for_each_entry(net, net_kill_list, cleanup_list) {
510 unhash_nsid(net, last);
511 list_add_tail(&net->exit_list, &net_exit_list);
512 }
513
514 /*
515 * Another CPU might be rcu-iterating the list, wait for it.
516 * This needs to be before calling the exit() notifiers, so
517 * the rcu_barrier() below isn't sufficient alone.
518 */
519 synchronize_rcu();
520
521 /* Run all of the network namespace exit methods */
522 list_for_each_entry_reverse(ops, &pernet_list, list)
523 ops_exit_list(ops, &net_exit_list);
524
525 /* Free the net generic variables */
526 list_for_each_entry_reverse(ops, &pernet_list, list)
527 ops_free_list(ops, &net_exit_list);
528
529 up_read(&pernet_ops_rwsem);
530
531 /* Ensure there are no outstanding rcu callbacks using this
532 * network namespace.
533 */
534 rcu_barrier();
535
536 /* Finally it is safe to free my network namespace structure */
537 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
538 list_del_init(&net->exit_list);
539 dec_net_namespaces(net->ucounts);
540 put_user_ns(net->user_ns);
541 net_drop_ns(net);
542 }
543}
544
545/**
546 * net_ns_barrier - wait until concurrent net_cleanup_work is done
547 *
548 * cleanup_net runs from work queue and will first remove namespaces
549 * from the global list, then run net exit functions.
550 *
551 * Call this in module exit path to make sure that all netns
552 * ->exit ops have been invoked before the function is removed.
553 */
554void net_ns_barrier(void)
555{
556 down_write(&pernet_ops_rwsem);
557 up_write(&pernet_ops_rwsem);
558}
559EXPORT_SYMBOL(net_ns_barrier);
560
561static DECLARE_WORK(net_cleanup_work, cleanup_net);
562
563void __put_net(struct net *net)
564{
565 /* Cleanup the network namespace in process context */
566 if (llist_add(&net->cleanup_list, &cleanup_list))
567 queue_work(netns_wq, &net_cleanup_work);
568}
569EXPORT_SYMBOL_GPL(__put_net);
570
571struct net *get_net_ns_by_fd(int fd)
572{
573 struct file *file;
574 struct ns_common *ns;
575 struct net *net;
576
577 file = proc_ns_fget(fd);
578 if (IS_ERR(file))
579 return ERR_CAST(file);
580
581 ns = get_proc_ns(file_inode(file));
582 if (ns->ops == &netns_operations)
583 net = get_net(container_of(ns, struct net, ns));
584 else
585 net = ERR_PTR(-EINVAL);
586
587 fput(file);
588 return net;
589}
590
591#else
592struct net *get_net_ns_by_fd(int fd)
593{
594 return ERR_PTR(-EINVAL);
595}
596#endif
597EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
598
599struct net *get_net_ns_by_pid(pid_t pid)
600{
601 struct task_struct *tsk;
602 struct net *net;
603
604 /* Lookup the network namespace */
605 net = ERR_PTR(-ESRCH);
606 rcu_read_lock();
607 tsk = find_task_by_vpid(pid);
608 if (tsk) {
609 struct nsproxy *nsproxy;
610 task_lock(tsk);
611 nsproxy = tsk->nsproxy;
612 if (nsproxy)
613 net = get_net(nsproxy->net_ns);
614 task_unlock(tsk);
615 }
616 rcu_read_unlock();
617 return net;
618}
619EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
620
621static __net_init int net_ns_net_init(struct net *net)
622{
623#ifdef CONFIG_NET_NS
624 net->ns.ops = &netns_operations;
625#endif
626 return ns_alloc_inum(&net->ns);
627}
628
629static __net_exit void net_ns_net_exit(struct net *net)
630{
631 ns_free_inum(&net->ns);
632}
633
634static struct pernet_operations __net_initdata net_ns_ops = {
635 .init = net_ns_net_init,
636 .exit = net_ns_net_exit,
637};
638
639static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
640 [NETNSA_NONE] = { .type = NLA_UNSPEC },
641 [NETNSA_NSID] = { .type = NLA_S32 },
642 [NETNSA_PID] = { .type = NLA_U32 },
643 [NETNSA_FD] = { .type = NLA_U32 },
644};
645
646static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
647 struct netlink_ext_ack *extack)
648{
649 struct net *net = sock_net(skb->sk);
650 struct nlattr *tb[NETNSA_MAX + 1];
651 struct nlattr *nla;
652 struct net *peer;
653 int nsid, err;
654
655 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
656 rtnl_net_policy, extack);
657 if (err < 0)
658 return err;
659 if (!tb[NETNSA_NSID]) {
660 NL_SET_ERR_MSG(extack, "nsid is missing");
661 return -EINVAL;
662 }
663 nsid = nla_get_s32(tb[NETNSA_NSID]);
664
665 if (tb[NETNSA_PID]) {
666 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
667 nla = tb[NETNSA_PID];
668 } else if (tb[NETNSA_FD]) {
669 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
670 nla = tb[NETNSA_FD];
671 } else {
672 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
673 return -EINVAL;
674 }
675 if (IS_ERR(peer)) {
676 NL_SET_BAD_ATTR(extack, nla);
677 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
678 return PTR_ERR(peer);
679 }
680
681 spin_lock_bh(&net->nsid_lock);
682 if (__peernet2id(net, peer) >= 0) {
683 spin_unlock_bh(&net->nsid_lock);
684 err = -EEXIST;
685 NL_SET_BAD_ATTR(extack, nla);
686 NL_SET_ERR_MSG(extack,
687 "Peer netns already has a nsid assigned");
688 goto out;
689 }
690
691 err = alloc_netid(net, peer, nsid);
692 spin_unlock_bh(&net->nsid_lock);
693 if (err >= 0) {
694 rtnl_net_notifyid(net, RTM_NEWNSID, err);
695 err = 0;
696 } else if (err == -ENOSPC && nsid >= 0) {
697 err = -EEXIST;
698 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
699 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
700 }
701out:
702 put_net(peer);
703 return err;
704}
705
706static int rtnl_net_get_size(void)
707{
708 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
709 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
710 ;
711}
712
713static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags,
714 int cmd, struct net *net, int nsid)
715{
716 struct nlmsghdr *nlh;
717 struct rtgenmsg *rth;
718
719 nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags);
720 if (!nlh)
721 return -EMSGSIZE;
722
723 rth = nlmsg_data(nlh);
724 rth->rtgen_family = AF_UNSPEC;
725
726 if (nla_put_s32(skb, NETNSA_NSID, nsid))
727 goto nla_put_failure;
728
729 nlmsg_end(skb, nlh);
730 return 0;
731
732nla_put_failure:
733 nlmsg_cancel(skb, nlh);
734 return -EMSGSIZE;
735}
736
737static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
738 struct netlink_ext_ack *extack)
739{
740 struct net *net = sock_net(skb->sk);
741 struct nlattr *tb[NETNSA_MAX + 1];
742 struct nlattr *nla;
743 struct sk_buff *msg;
744 struct net *peer;
745 int err, id;
746
747 err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
748 rtnl_net_policy, extack);
749 if (err < 0)
750 return err;
751 if (tb[NETNSA_PID]) {
752 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
753 nla = tb[NETNSA_PID];
754 } else if (tb[NETNSA_FD]) {
755 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
756 nla = tb[NETNSA_FD];
757 } else {
758 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
759 return -EINVAL;
760 }
761
762 if (IS_ERR(peer)) {
763 NL_SET_BAD_ATTR(extack, nla);
764 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
765 return PTR_ERR(peer);
766 }
767
768 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
769 if (!msg) {
770 err = -ENOMEM;
771 goto out;
772 }
773
774 id = peernet2id(net, peer);
775 err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0,
776 RTM_NEWNSID, net, id);
777 if (err < 0)
778 goto err_out;
779
780 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
781 goto out;
782
783err_out:
784 nlmsg_free(msg);
785out:
786 put_net(peer);
787 return err;
788}
789
790struct rtnl_net_dump_cb {
791 struct net *net;
792 struct sk_buff *skb;
793 struct netlink_callback *cb;
794 int idx;
795 int s_idx;
796};
797
798static int rtnl_net_dumpid_one(int id, void *peer, void *data)
799{
800 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
801 int ret;
802
803 if (net_cb->idx < net_cb->s_idx)
804 goto cont;
805
806 ret = rtnl_net_fill(net_cb->skb, NETLINK_CB(net_cb->cb->skb).portid,
807 net_cb->cb->nlh->nlmsg_seq, NLM_F_MULTI,
808 RTM_NEWNSID, net_cb->net, id);
809 if (ret < 0)
810 return ret;
811
812cont:
813 net_cb->idx++;
814 return 0;
815}
816
817static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
818{
819 struct net *net = sock_net(skb->sk);
820 struct rtnl_net_dump_cb net_cb = {
821 .net = net,
822 .skb = skb,
823 .cb = cb,
824 .idx = 0,
825 .s_idx = cb->args[0],
826 };
827
828 spin_lock_bh(&net->nsid_lock);
829 idr_for_each(&net->netns_ids, rtnl_net_dumpid_one, &net_cb);
830 spin_unlock_bh(&net->nsid_lock);
831
832 cb->args[0] = net_cb.idx;
833 return skb->len;
834}
835
836static void rtnl_net_notifyid(struct net *net, int cmd, int id)
837{
838 struct sk_buff *msg;
839 int err = -ENOMEM;
840
841 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
842 if (!msg)
843 goto out;
844
845 err = rtnl_net_fill(msg, 0, 0, 0, cmd, net, id);
846 if (err < 0)
847 goto err_out;
848
849 rtnl_notify(msg, net, 0, RTNLGRP_NSID, NULL, 0);
850 return;
851
852err_out:
853 nlmsg_free(msg);
854out:
855 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
856}
857
858static int __init net_ns_init(void)
859{
860 struct net_generic *ng;
861
862#ifdef CONFIG_NET_NS
863 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
864 SMP_CACHE_BYTES,
865 SLAB_PANIC|SLAB_ACCOUNT, NULL);
866
867 /* Create workqueue for cleanup */
868 netns_wq = create_singlethread_workqueue("netns");
869 if (!netns_wq)
870 panic("Could not create netns workq");
871#endif
872
873 ng = net_alloc_generic();
874 if (!ng)
875 panic("Could not allocate generic netns");
876
877 rcu_assign_pointer(init_net.gen, ng);
878
879 down_write(&pernet_ops_rwsem);
880 if (setup_net(&init_net, &init_user_ns))
881 panic("Could not setup the initial network namespace");
882
883 init_net_initialized = true;
884 up_write(&pernet_ops_rwsem);
885
886 register_pernet_subsys(&net_ns_ops);
887
888 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
889 RTNL_FLAG_DOIT_UNLOCKED);
890 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
891 RTNL_FLAG_DOIT_UNLOCKED);
892
893 return 0;
894}
895
896pure_initcall(net_ns_init);
897
898#ifdef CONFIG_NET_NS
899static int __register_pernet_operations(struct list_head *list,
900 struct pernet_operations *ops)
901{
902 struct net *net;
903 int error;
904 LIST_HEAD(net_exit_list);
905
906 list_add_tail(&ops->list, list);
907 if (ops->init || (ops->id && ops->size)) {
908 /* We held write locked pernet_ops_rwsem, and parallel
909 * setup_net() and cleanup_net() are not possible.
910 */
911 for_each_net(net) {
912 error = ops_init(ops, net);
913 if (error)
914 goto out_undo;
915 list_add_tail(&net->exit_list, &net_exit_list);
916 }
917 }
918 return 0;
919
920out_undo:
921 /* If I have an error cleanup all namespaces I initialized */
922 list_del(&ops->list);
923 ops_exit_list(ops, &net_exit_list);
924 ops_free_list(ops, &net_exit_list);
925 return error;
926}
927
928static void __unregister_pernet_operations(struct pernet_operations *ops)
929{
930 struct net *net;
931 LIST_HEAD(net_exit_list);
932
933 list_del(&ops->list);
934 /* See comment in __register_pernet_operations() */
935 for_each_net(net)
936 list_add_tail(&net->exit_list, &net_exit_list);
937 ops_exit_list(ops, &net_exit_list);
938 ops_free_list(ops, &net_exit_list);
939}
940
941#else
942
943static int __register_pernet_operations(struct list_head *list,
944 struct pernet_operations *ops)
945{
946 if (!init_net_initialized) {
947 list_add_tail(&ops->list, list);
948 return 0;
949 }
950
951 return ops_init(ops, &init_net);
952}
953
954static void __unregister_pernet_operations(struct pernet_operations *ops)
955{
956 if (!init_net_initialized) {
957 list_del(&ops->list);
958 } else {
959 LIST_HEAD(net_exit_list);
960 list_add(&init_net.exit_list, &net_exit_list);
961 ops_exit_list(ops, &net_exit_list);
962 ops_free_list(ops, &net_exit_list);
963 }
964}
965
966#endif /* CONFIG_NET_NS */
967
968static DEFINE_IDA(net_generic_ids);
969
970static int register_pernet_operations(struct list_head *list,
971 struct pernet_operations *ops)
972{
973 int error;
974
975 if (ops->id) {
976again:
977 error = ida_get_new_above(&net_generic_ids, MIN_PERNET_OPS_ID, ops->id);
978 if (error < 0) {
979 if (error == -EAGAIN) {
980 ida_pre_get(&net_generic_ids, GFP_KERNEL);
981 goto again;
982 }
983 return error;
984 }
985 max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
986 }
987 error = __register_pernet_operations(list, ops);
988 if (error) {
989 rcu_barrier();
990 if (ops->id)
991 ida_remove(&net_generic_ids, *ops->id);
992 }
993
994 return error;
995}
996
997static void unregister_pernet_operations(struct pernet_operations *ops)
998{
999 __unregister_pernet_operations(ops);
1000 rcu_barrier();
1001 if (ops->id)
1002 ida_remove(&net_generic_ids, *ops->id);
1003}
1004
1005/**
1006 * register_pernet_subsys - register a network namespace subsystem
1007 * @ops: pernet operations structure for the subsystem
1008 *
1009 * Register a subsystem which has init and exit functions
1010 * that are called when network namespaces are created and
1011 * destroyed respectively.
1012 *
1013 * When registered all network namespace init functions are
1014 * called for every existing network namespace. Allowing kernel
1015 * modules to have a race free view of the set of network namespaces.
1016 *
1017 * When a new network namespace is created all of the init
1018 * methods are called in the order in which they were registered.
1019 *
1020 * When a network namespace is destroyed all of the exit methods
1021 * are called in the reverse of the order with which they were
1022 * registered.
1023 */
1024int register_pernet_subsys(struct pernet_operations *ops)
1025{
1026 int error;
1027 down_write(&pernet_ops_rwsem);
1028 error = register_pernet_operations(first_device, ops);
1029 up_write(&pernet_ops_rwsem);
1030 return error;
1031}
1032EXPORT_SYMBOL_GPL(register_pernet_subsys);
1033
1034/**
1035 * unregister_pernet_subsys - unregister a network namespace subsystem
1036 * @ops: pernet operations structure to manipulate
1037 *
1038 * Remove the pernet operations structure from the list to be
1039 * used when network namespaces are created or destroyed. In
1040 * addition run the exit method for all existing network
1041 * namespaces.
1042 */
1043void unregister_pernet_subsys(struct pernet_operations *ops)
1044{
1045 down_write(&pernet_ops_rwsem);
1046 unregister_pernet_operations(ops);
1047 up_write(&pernet_ops_rwsem);
1048}
1049EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1050
1051/**
1052 * register_pernet_device - register a network namespace device
1053 * @ops: pernet operations structure for the subsystem
1054 *
1055 * Register a device which has init and exit functions
1056 * that are called when network namespaces are created and
1057 * destroyed respectively.
1058 *
1059 * When registered all network namespace init functions are
1060 * called for every existing network namespace. Allowing kernel
1061 * modules to have a race free view of the set of network namespaces.
1062 *
1063 * When a new network namespace is created all of the init
1064 * methods are called in the order in which they were registered.
1065 *
1066 * When a network namespace is destroyed all of the exit methods
1067 * are called in the reverse of the order with which they were
1068 * registered.
1069 */
1070int register_pernet_device(struct pernet_operations *ops)
1071{
1072 int error;
1073 down_write(&pernet_ops_rwsem);
1074 error = register_pernet_operations(&pernet_list, ops);
1075 if (!error && (first_device == &pernet_list))
1076 first_device = &ops->list;
1077 up_write(&pernet_ops_rwsem);
1078 return error;
1079}
1080EXPORT_SYMBOL_GPL(register_pernet_device);
1081
1082/**
1083 * unregister_pernet_device - unregister a network namespace netdevice
1084 * @ops: pernet operations structure to manipulate
1085 *
1086 * Remove the pernet operations structure from the list to be
1087 * used when network namespaces are created or destroyed. In
1088 * addition run the exit method for all existing network
1089 * namespaces.
1090 */
1091void unregister_pernet_device(struct pernet_operations *ops)
1092{
1093 down_write(&pernet_ops_rwsem);
1094 if (&ops->list == first_device)
1095 first_device = first_device->next;
1096 unregister_pernet_operations(ops);
1097 up_write(&pernet_ops_rwsem);
1098}
1099EXPORT_SYMBOL_GPL(unregister_pernet_device);
1100
1101#ifdef CONFIG_NET_NS
1102static struct ns_common *netns_get(struct task_struct *task)
1103{
1104 struct net *net = NULL;
1105 struct nsproxy *nsproxy;
1106
1107 task_lock(task);
1108 nsproxy = task->nsproxy;
1109 if (nsproxy)
1110 net = get_net(nsproxy->net_ns);
1111 task_unlock(task);
1112
1113 return net ? &net->ns : NULL;
1114}
1115
1116static inline struct net *to_net_ns(struct ns_common *ns)
1117{
1118 return container_of(ns, struct net, ns);
1119}
1120
1121static void netns_put(struct ns_common *ns)
1122{
1123 put_net(to_net_ns(ns));
1124}
1125
1126static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
1127{
1128 struct net *net = to_net_ns(ns);
1129
1130 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1131 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1132 return -EPERM;
1133
1134 put_net(nsproxy->net_ns);
1135 nsproxy->net_ns = get_net(net);
1136 return 0;
1137}
1138
1139static struct user_namespace *netns_owner(struct ns_common *ns)
1140{
1141 return to_net_ns(ns)->user_ns;
1142}
1143
1144const struct proc_ns_operations netns_operations = {
1145 .name = "net",
1146 .type = CLONE_NEWNET,
1147 .get = netns_get,
1148 .put = netns_put,
1149 .install = netns_install,
1150 .owner = netns_owner,
1151};
1152#endif
1// SPDX-License-Identifier: GPL-2.0-only
2#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
3
4#include <linux/workqueue.h>
5#include <linux/rtnetlink.h>
6#include <linux/cache.h>
7#include <linux/slab.h>
8#include <linux/list.h>
9#include <linux/delay.h>
10#include <linux/sched.h>
11#include <linux/idr.h>
12#include <linux/rculist.h>
13#include <linux/nsproxy.h>
14#include <linux/fs.h>
15#include <linux/proc_ns.h>
16#include <linux/file.h>
17#include <linux/export.h>
18#include <linux/user_namespace.h>
19#include <linux/net_namespace.h>
20#include <linux/sched/task.h>
21#include <linux/uidgid.h>
22
23#include <net/sock.h>
24#include <net/netlink.h>
25#include <net/net_namespace.h>
26#include <net/netns/generic.h>
27
28/*
29 * Our network namespace constructor/destructor lists
30 */
31
32static LIST_HEAD(pernet_list);
33static struct list_head *first_device = &pernet_list;
34
35LIST_HEAD(net_namespace_list);
36EXPORT_SYMBOL_GPL(net_namespace_list);
37
38/* Protects net_namespace_list. Nests iside rtnl_lock() */
39DECLARE_RWSEM(net_rwsem);
40EXPORT_SYMBOL_GPL(net_rwsem);
41
42#ifdef CONFIG_KEYS
43static struct key_tag init_net_key_domain = { .usage = REFCOUNT_INIT(1) };
44#endif
45
46struct net init_net = {
47 .count = REFCOUNT_INIT(1),
48 .dev_base_head = LIST_HEAD_INIT(init_net.dev_base_head),
49#ifdef CONFIG_KEYS
50 .key_domain = &init_net_key_domain,
51#endif
52};
53EXPORT_SYMBOL(init_net);
54
55static bool init_net_initialized;
56/*
57 * pernet_ops_rwsem: protects: pernet_list, net_generic_ids,
58 * init_net_initialized and first_device pointer.
59 * This is internal net namespace object. Please, don't use it
60 * outside.
61 */
62DECLARE_RWSEM(pernet_ops_rwsem);
63EXPORT_SYMBOL_GPL(pernet_ops_rwsem);
64
65#define MIN_PERNET_OPS_ID \
66 ((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *))
67
68#define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
69
70static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
71
72static atomic64_t cookie_gen;
73
74u64 net_gen_cookie(struct net *net)
75{
76 while (1) {
77 u64 res = atomic64_read(&net->net_cookie);
78
79 if (res)
80 return res;
81 res = atomic64_inc_return(&cookie_gen);
82 atomic64_cmpxchg(&net->net_cookie, 0, res);
83 }
84}
85
86static struct net_generic *net_alloc_generic(void)
87{
88 struct net_generic *ng;
89 unsigned int generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
90
91 ng = kzalloc(generic_size, GFP_KERNEL);
92 if (ng)
93 ng->s.len = max_gen_ptrs;
94
95 return ng;
96}
97
98static int net_assign_generic(struct net *net, unsigned int id, void *data)
99{
100 struct net_generic *ng, *old_ng;
101
102 BUG_ON(id < MIN_PERNET_OPS_ID);
103
104 old_ng = rcu_dereference_protected(net->gen,
105 lockdep_is_held(&pernet_ops_rwsem));
106 if (old_ng->s.len > id) {
107 old_ng->ptr[id] = data;
108 return 0;
109 }
110
111 ng = net_alloc_generic();
112 if (ng == NULL)
113 return -ENOMEM;
114
115 /*
116 * Some synchronisation notes:
117 *
118 * The net_generic explores the net->gen array inside rcu
119 * read section. Besides once set the net->gen->ptr[x]
120 * pointer never changes (see rules in netns/generic.h).
121 *
122 * That said, we simply duplicate this array and schedule
123 * the old copy for kfree after a grace period.
124 */
125
126 memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
127 (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
128 ng->ptr[id] = data;
129
130 rcu_assign_pointer(net->gen, ng);
131 kfree_rcu(old_ng, s.rcu);
132 return 0;
133}
134
135static int ops_init(const struct pernet_operations *ops, struct net *net)
136{
137 int err = -ENOMEM;
138 void *data = NULL;
139
140 if (ops->id && ops->size) {
141 data = kzalloc(ops->size, GFP_KERNEL);
142 if (!data)
143 goto out;
144
145 err = net_assign_generic(net, *ops->id, data);
146 if (err)
147 goto cleanup;
148 }
149 err = 0;
150 if (ops->init)
151 err = ops->init(net);
152 if (!err)
153 return 0;
154
155cleanup:
156 kfree(data);
157
158out:
159 return err;
160}
161
162static void ops_free(const struct pernet_operations *ops, struct net *net)
163{
164 if (ops->id && ops->size) {
165 kfree(net_generic(net, *ops->id));
166 }
167}
168
169static void ops_pre_exit_list(const struct pernet_operations *ops,
170 struct list_head *net_exit_list)
171{
172 struct net *net;
173
174 if (ops->pre_exit) {
175 list_for_each_entry(net, net_exit_list, exit_list)
176 ops->pre_exit(net);
177 }
178}
179
180static void ops_exit_list(const struct pernet_operations *ops,
181 struct list_head *net_exit_list)
182{
183 struct net *net;
184 if (ops->exit) {
185 list_for_each_entry(net, net_exit_list, exit_list)
186 ops->exit(net);
187 }
188 if (ops->exit_batch)
189 ops->exit_batch(net_exit_list);
190}
191
192static void ops_free_list(const struct pernet_operations *ops,
193 struct list_head *net_exit_list)
194{
195 struct net *net;
196 if (ops->size && ops->id) {
197 list_for_each_entry(net, net_exit_list, exit_list)
198 ops_free(ops, net);
199 }
200}
201
202/* should be called with nsid_lock held */
203static int alloc_netid(struct net *net, struct net *peer, int reqid)
204{
205 int min = 0, max = 0;
206
207 if (reqid >= 0) {
208 min = reqid;
209 max = reqid + 1;
210 }
211
212 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
213}
214
215/* This function is used by idr_for_each(). If net is equal to peer, the
216 * function returns the id so that idr_for_each() stops. Because we cannot
217 * returns the id 0 (idr_for_each() will not stop), we return the magic value
218 * NET_ID_ZERO (-1) for it.
219 */
220#define NET_ID_ZERO -1
221static int net_eq_idr(int id, void *net, void *peer)
222{
223 if (net_eq(net, peer))
224 return id ? : NET_ID_ZERO;
225 return 0;
226}
227
228/* Must be called from RCU-critical section or with nsid_lock held */
229static int __peernet2id(const struct net *net, struct net *peer)
230{
231 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
232
233 /* Magic value for id 0. */
234 if (id == NET_ID_ZERO)
235 return 0;
236 if (id > 0)
237 return id;
238
239 return NETNSA_NSID_NOT_ASSIGNED;
240}
241
242static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
243 struct nlmsghdr *nlh, gfp_t gfp);
244/* This function returns the id of a peer netns. If no id is assigned, one will
245 * be allocated and returned.
246 */
247int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp)
248{
249 int id;
250
251 if (refcount_read(&net->count) == 0)
252 return NETNSA_NSID_NOT_ASSIGNED;
253
254 spin_lock_bh(&net->nsid_lock);
255 id = __peernet2id(net, peer);
256 if (id >= 0) {
257 spin_unlock_bh(&net->nsid_lock);
258 return id;
259 }
260
261 /* When peer is obtained from RCU lists, we may race with
262 * its cleanup. Check whether it's alive, and this guarantees
263 * we never hash a peer back to net->netns_ids, after it has
264 * just been idr_remove()'d from there in cleanup_net().
265 */
266 if (!maybe_get_net(peer)) {
267 spin_unlock_bh(&net->nsid_lock);
268 return NETNSA_NSID_NOT_ASSIGNED;
269 }
270
271 id = alloc_netid(net, peer, -1);
272 spin_unlock_bh(&net->nsid_lock);
273
274 put_net(peer);
275 if (id < 0)
276 return NETNSA_NSID_NOT_ASSIGNED;
277
278 rtnl_net_notifyid(net, RTM_NEWNSID, id, 0, NULL, gfp);
279
280 return id;
281}
282EXPORT_SYMBOL_GPL(peernet2id_alloc);
283
284/* This function returns, if assigned, the id of a peer netns. */
285int peernet2id(const struct net *net, struct net *peer)
286{
287 int id;
288
289 rcu_read_lock();
290 id = __peernet2id(net, peer);
291 rcu_read_unlock();
292
293 return id;
294}
295EXPORT_SYMBOL(peernet2id);
296
297/* This function returns true is the peer netns has an id assigned into the
298 * current netns.
299 */
300bool peernet_has_id(const struct net *net, struct net *peer)
301{
302 return peernet2id(net, peer) >= 0;
303}
304
305struct net *get_net_ns_by_id(const struct net *net, int id)
306{
307 struct net *peer;
308
309 if (id < 0)
310 return NULL;
311
312 rcu_read_lock();
313 peer = idr_find(&net->netns_ids, id);
314 if (peer)
315 peer = maybe_get_net(peer);
316 rcu_read_unlock();
317
318 return peer;
319}
320
321/*
322 * setup_net runs the initializers for the network namespace object.
323 */
324static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
325{
326 /* Must be called with pernet_ops_rwsem held */
327 const struct pernet_operations *ops, *saved_ops;
328 int error = 0;
329 LIST_HEAD(net_exit_list);
330
331 refcount_set(&net->count, 1);
332 refcount_set(&net->passive, 1);
333 get_random_bytes(&net->hash_mix, sizeof(u32));
334 net->dev_base_seq = 1;
335 net->user_ns = user_ns;
336 idr_init(&net->netns_ids);
337 spin_lock_init(&net->nsid_lock);
338 mutex_init(&net->ipv4.ra_mutex);
339
340 list_for_each_entry(ops, &pernet_list, list) {
341 error = ops_init(ops, net);
342 if (error < 0)
343 goto out_undo;
344 }
345 down_write(&net_rwsem);
346 list_add_tail_rcu(&net->list, &net_namespace_list);
347 up_write(&net_rwsem);
348out:
349 return error;
350
351out_undo:
352 /* Walk through the list backwards calling the exit functions
353 * for the pernet modules whose init functions did not fail.
354 */
355 list_add(&net->exit_list, &net_exit_list);
356 saved_ops = ops;
357 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
358 ops_pre_exit_list(ops, &net_exit_list);
359
360 synchronize_rcu();
361
362 ops = saved_ops;
363 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
364 ops_exit_list(ops, &net_exit_list);
365
366 ops = saved_ops;
367 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
368 ops_free_list(ops, &net_exit_list);
369
370 rcu_barrier();
371 goto out;
372}
373
374static int __net_init net_defaults_init_net(struct net *net)
375{
376 net->core.sysctl_somaxconn = SOMAXCONN;
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("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(), UCOUNT_NET_NAMESPACES);
398}
399
400static void dec_net_namespaces(struct ucounts *ucounts)
401{
402 dec_ucount(ucounts, 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(net_cachep, GFP_KERNEL);
418 if (!net)
419 goto out_free;
420
421#ifdef CONFIG_KEYS
422 net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
423 if (!net->key_domain)
424 goto out_free_2;
425 refcount_set(&net->key_domain->usage, 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(net_cachep, net);
435 net = NULL;
436#endif
437out_free:
438 kfree(ng);
439 goto out;
440}
441
442static void net_free(struct net *net)
443{
444 kfree(rcu_access_pointer(net->gen));
445 kmem_cache_free(net_cachep, net);
446}
447
448void net_drop_ns(void *p)
449{
450 struct net *ns = p;
451 if (ns && refcount_dec_and_test(&ns->passive))
452 net_free(ns);
453}
454
455struct net *copy_net_ns(unsigned long flags,
456 struct user_namespace *user_ns, struct net *old_net)
457{
458 struct ucounts *ucounts;
459 struct net *net;
460 int rv;
461
462 if (!(flags & CLONE_NEWNET))
463 return get_net(old_net);
464
465 ucounts = inc_net_namespaces(user_ns);
466 if (!ucounts)
467 return ERR_PTR(-ENOSPC);
468
469 net = net_alloc();
470 if (!net) {
471 rv = -ENOMEM;
472 goto dec_ucounts;
473 }
474 refcount_set(&net->passive, 1);
475 net->ucounts = ucounts;
476 get_user_ns(user_ns);
477
478 rv = down_read_killable(&pernet_ops_rwsem);
479 if (rv < 0)
480 goto put_userns;
481
482 rv = setup_net(net, user_ns);
483
484 up_read(&pernet_ops_rwsem);
485
486 if (rv < 0) {
487put_userns:
488 key_remove_domain(net->key_domain);
489 put_user_ns(user_ns);
490 net_drop_ns(net);
491dec_ucounts:
492 dec_net_namespaces(ucounts);
493 return ERR_PTR(rv);
494 }
495 return net;
496}
497
498/**
499 * net_ns_get_ownership - get sysfs ownership data for @net
500 * @net: network namespace in question (can be NULL)
501 * @uid: kernel user ID for sysfs objects
502 * @gid: kernel group ID for sysfs objects
503 *
504 * Returns the uid/gid pair of root in the user namespace associated with the
505 * given network namespace.
506 */
507void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
508{
509 if (net) {
510 kuid_t ns_root_uid = make_kuid(net->user_ns, 0);
511 kgid_t ns_root_gid = make_kgid(net->user_ns, 0);
512
513 if (uid_valid(ns_root_uid))
514 *uid = ns_root_uid;
515
516 if (gid_valid(ns_root_gid))
517 *gid = ns_root_gid;
518 } else {
519 *uid = GLOBAL_ROOT_UID;
520 *gid = GLOBAL_ROOT_GID;
521 }
522}
523EXPORT_SYMBOL_GPL(net_ns_get_ownership);
524
525static void unhash_nsid(struct net *net, struct net *last)
526{
527 struct net *tmp;
528 /* This function is only called from cleanup_net() work,
529 * and this work is the only process, that may delete
530 * a net from net_namespace_list. So, when the below
531 * is executing, the list may only grow. Thus, we do not
532 * use for_each_net_rcu() or net_rwsem.
533 */
534 for_each_net(tmp) {
535 int id;
536
537 spin_lock_bh(&tmp->nsid_lock);
538 id = __peernet2id(tmp, net);
539 if (id >= 0)
540 idr_remove(&tmp->netns_ids, id);
541 spin_unlock_bh(&tmp->nsid_lock);
542 if (id >= 0)
543 rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL,
544 GFP_KERNEL);
545 if (tmp == last)
546 break;
547 }
548 spin_lock_bh(&net->nsid_lock);
549 idr_destroy(&net->netns_ids);
550 spin_unlock_bh(&net->nsid_lock);
551}
552
553static LLIST_HEAD(cleanup_list);
554
555static void cleanup_net(struct work_struct *work)
556{
557 const struct pernet_operations *ops;
558 struct net *net, *tmp, *last;
559 struct llist_node *net_kill_list;
560 LIST_HEAD(net_exit_list);
561
562 /* Atomically snapshot the list of namespaces to cleanup */
563 net_kill_list = llist_del_all(&cleanup_list);
564
565 down_read(&pernet_ops_rwsem);
566
567 /* Don't let anyone else find us. */
568 down_write(&net_rwsem);
569 llist_for_each_entry(net, net_kill_list, cleanup_list)
570 list_del_rcu(&net->list);
571 /* Cache last net. After we unlock rtnl, no one new net
572 * added to net_namespace_list can assign nsid pointer
573 * to a net from net_kill_list (see peernet2id_alloc()).
574 * So, we skip them in unhash_nsid().
575 *
576 * Note, that unhash_nsid() does not delete nsid links
577 * between net_kill_list's nets, as they've already
578 * deleted from net_namespace_list. But, this would be
579 * useless anyway, as netns_ids are destroyed there.
580 */
581 last = list_last_entry(&net_namespace_list, struct net, list);
582 up_write(&net_rwsem);
583
584 llist_for_each_entry(net, net_kill_list, cleanup_list) {
585 unhash_nsid(net, last);
586 list_add_tail(&net->exit_list, &net_exit_list);
587 }
588
589 /* Run all of the network namespace pre_exit methods */
590 list_for_each_entry_reverse(ops, &pernet_list, list)
591 ops_pre_exit_list(ops, &net_exit_list);
592
593 /*
594 * Another CPU might be rcu-iterating the list, wait for it.
595 * This needs to be before calling the exit() notifiers, so
596 * the rcu_barrier() below isn't sufficient alone.
597 * Also the pre_exit() and exit() methods need this barrier.
598 */
599 synchronize_rcu();
600
601 /* Run all of the network namespace exit methods */
602 list_for_each_entry_reverse(ops, &pernet_list, list)
603 ops_exit_list(ops, &net_exit_list);
604
605 /* Free the net generic variables */
606 list_for_each_entry_reverse(ops, &pernet_list, list)
607 ops_free_list(ops, &net_exit_list);
608
609 up_read(&pernet_ops_rwsem);
610
611 /* Ensure there are no outstanding rcu callbacks using this
612 * network namespace.
613 */
614 rcu_barrier();
615
616 /* Finally it is safe to free my network namespace structure */
617 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
618 list_del_init(&net->exit_list);
619 dec_net_namespaces(net->ucounts);
620 key_remove_domain(net->key_domain);
621 put_user_ns(net->user_ns);
622 net_drop_ns(net);
623 }
624}
625
626/**
627 * net_ns_barrier - wait until concurrent net_cleanup_work is done
628 *
629 * cleanup_net runs from work queue and will first remove namespaces
630 * from the global list, then run net exit functions.
631 *
632 * Call this in module exit path to make sure that all netns
633 * ->exit ops have been invoked before the function is removed.
634 */
635void net_ns_barrier(void)
636{
637 down_write(&pernet_ops_rwsem);
638 up_write(&pernet_ops_rwsem);
639}
640EXPORT_SYMBOL(net_ns_barrier);
641
642static DECLARE_WORK(net_cleanup_work, cleanup_net);
643
644void __put_net(struct net *net)
645{
646 /* Cleanup the network namespace in process context */
647 if (llist_add(&net->cleanup_list, &cleanup_list))
648 queue_work(netns_wq, &net_cleanup_work);
649}
650EXPORT_SYMBOL_GPL(__put_net);
651
652struct net *get_net_ns_by_fd(int fd)
653{
654 struct file *file;
655 struct ns_common *ns;
656 struct net *net;
657
658 file = proc_ns_fget(fd);
659 if (IS_ERR(file))
660 return ERR_CAST(file);
661
662 ns = get_proc_ns(file_inode(file));
663 if (ns->ops == &netns_operations)
664 net = get_net(container_of(ns, struct net, ns));
665 else
666 net = ERR_PTR(-EINVAL);
667
668 fput(file);
669 return net;
670}
671
672#else
673struct net *get_net_ns_by_fd(int fd)
674{
675 return ERR_PTR(-EINVAL);
676}
677#endif
678EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
679
680struct net *get_net_ns_by_pid(pid_t pid)
681{
682 struct task_struct *tsk;
683 struct net *net;
684
685 /* Lookup the network namespace */
686 net = ERR_PTR(-ESRCH);
687 rcu_read_lock();
688 tsk = find_task_by_vpid(pid);
689 if (tsk) {
690 struct nsproxy *nsproxy;
691 task_lock(tsk);
692 nsproxy = tsk->nsproxy;
693 if (nsproxy)
694 net = get_net(nsproxy->net_ns);
695 task_unlock(tsk);
696 }
697 rcu_read_unlock();
698 return net;
699}
700EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
701
702static __net_init int net_ns_net_init(struct net *net)
703{
704#ifdef CONFIG_NET_NS
705 net->ns.ops = &netns_operations;
706#endif
707 return ns_alloc_inum(&net->ns);
708}
709
710static __net_exit void net_ns_net_exit(struct net *net)
711{
712 ns_free_inum(&net->ns);
713}
714
715static struct pernet_operations __net_initdata net_ns_ops = {
716 .init = net_ns_net_init,
717 .exit = net_ns_net_exit,
718};
719
720static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
721 [NETNSA_NONE] = { .type = NLA_UNSPEC },
722 [NETNSA_NSID] = { .type = NLA_S32 },
723 [NETNSA_PID] = { .type = NLA_U32 },
724 [NETNSA_FD] = { .type = NLA_U32 },
725 [NETNSA_TARGET_NSID] = { .type = NLA_S32 },
726};
727
728static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
729 struct netlink_ext_ack *extack)
730{
731 struct net *net = sock_net(skb->sk);
732 struct nlattr *tb[NETNSA_MAX + 1];
733 struct nlattr *nla;
734 struct net *peer;
735 int nsid, err;
736
737 err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
738 NETNSA_MAX, rtnl_net_policy, extack);
739 if (err < 0)
740 return err;
741 if (!tb[NETNSA_NSID]) {
742 NL_SET_ERR_MSG(extack, "nsid is missing");
743 return -EINVAL;
744 }
745 nsid = nla_get_s32(tb[NETNSA_NSID]);
746
747 if (tb[NETNSA_PID]) {
748 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
749 nla = tb[NETNSA_PID];
750 } else if (tb[NETNSA_FD]) {
751 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
752 nla = tb[NETNSA_FD];
753 } else {
754 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
755 return -EINVAL;
756 }
757 if (IS_ERR(peer)) {
758 NL_SET_BAD_ATTR(extack, nla);
759 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
760 return PTR_ERR(peer);
761 }
762
763 spin_lock_bh(&net->nsid_lock);
764 if (__peernet2id(net, peer) >= 0) {
765 spin_unlock_bh(&net->nsid_lock);
766 err = -EEXIST;
767 NL_SET_BAD_ATTR(extack, nla);
768 NL_SET_ERR_MSG(extack,
769 "Peer netns already has a nsid assigned");
770 goto out;
771 }
772
773 err = alloc_netid(net, peer, nsid);
774 spin_unlock_bh(&net->nsid_lock);
775 if (err >= 0) {
776 rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid,
777 nlh, GFP_KERNEL);
778 err = 0;
779 } else if (err == -ENOSPC && nsid >= 0) {
780 err = -EEXIST;
781 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
782 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
783 }
784out:
785 put_net(peer);
786 return err;
787}
788
789static int rtnl_net_get_size(void)
790{
791 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
792 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
793 + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
794 ;
795}
796
797struct net_fill_args {
798 u32 portid;
799 u32 seq;
800 int flags;
801 int cmd;
802 int nsid;
803 bool add_ref;
804 int ref_nsid;
805};
806
807static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
808{
809 struct nlmsghdr *nlh;
810 struct rtgenmsg *rth;
811
812 nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
813 args->flags);
814 if (!nlh)
815 return -EMSGSIZE;
816
817 rth = nlmsg_data(nlh);
818 rth->rtgen_family = AF_UNSPEC;
819
820 if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
821 goto nla_put_failure;
822
823 if (args->add_ref &&
824 nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
825 goto nla_put_failure;
826
827 nlmsg_end(skb, nlh);
828 return 0;
829
830nla_put_failure:
831 nlmsg_cancel(skb, nlh);
832 return -EMSGSIZE;
833}
834
835static int rtnl_net_valid_getid_req(struct sk_buff *skb,
836 const struct nlmsghdr *nlh,
837 struct nlattr **tb,
838 struct netlink_ext_ack *extack)
839{
840 int i, err;
841
842 if (!netlink_strict_get_check(skb))
843 return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
844 tb, NETNSA_MAX, rtnl_net_policy,
845 extack);
846
847 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
848 NETNSA_MAX, rtnl_net_policy,
849 extack);
850 if (err)
851 return err;
852
853 for (i = 0; i <= NETNSA_MAX; i++) {
854 if (!tb[i])
855 continue;
856
857 switch (i) {
858 case NETNSA_PID:
859 case NETNSA_FD:
860 case NETNSA_NSID:
861 case NETNSA_TARGET_NSID:
862 break;
863 default:
864 NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
865 return -EINVAL;
866 }
867 }
868
869 return 0;
870}
871
872static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
873 struct netlink_ext_ack *extack)
874{
875 struct net *net = sock_net(skb->sk);
876 struct nlattr *tb[NETNSA_MAX + 1];
877 struct net_fill_args fillargs = {
878 .portid = NETLINK_CB(skb).portid,
879 .seq = nlh->nlmsg_seq,
880 .cmd = RTM_NEWNSID,
881 };
882 struct net *peer, *target = net;
883 struct nlattr *nla;
884 struct sk_buff *msg;
885 int err;
886
887 err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
888 if (err < 0)
889 return err;
890 if (tb[NETNSA_PID]) {
891 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
892 nla = tb[NETNSA_PID];
893 } else if (tb[NETNSA_FD]) {
894 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
895 nla = tb[NETNSA_FD];
896 } else if (tb[NETNSA_NSID]) {
897 peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
898 if (!peer)
899 peer = ERR_PTR(-ENOENT);
900 nla = tb[NETNSA_NSID];
901 } else {
902 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
903 return -EINVAL;
904 }
905
906 if (IS_ERR(peer)) {
907 NL_SET_BAD_ATTR(extack, nla);
908 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
909 return PTR_ERR(peer);
910 }
911
912 if (tb[NETNSA_TARGET_NSID]) {
913 int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
914
915 target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
916 if (IS_ERR(target)) {
917 NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
918 NL_SET_ERR_MSG(extack,
919 "Target netns reference is invalid");
920 err = PTR_ERR(target);
921 goto out;
922 }
923 fillargs.add_ref = true;
924 fillargs.ref_nsid = peernet2id(net, peer);
925 }
926
927 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
928 if (!msg) {
929 err = -ENOMEM;
930 goto out;
931 }
932
933 fillargs.nsid = peernet2id(target, peer);
934 err = rtnl_net_fill(msg, &fillargs);
935 if (err < 0)
936 goto err_out;
937
938 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
939 goto out;
940
941err_out:
942 nlmsg_free(msg);
943out:
944 if (fillargs.add_ref)
945 put_net(target);
946 put_net(peer);
947 return err;
948}
949
950struct rtnl_net_dump_cb {
951 struct net *tgt_net;
952 struct net *ref_net;
953 struct sk_buff *skb;
954 struct net_fill_args fillargs;
955 int idx;
956 int s_idx;
957};
958
959/* Runs in RCU-critical section. */
960static int rtnl_net_dumpid_one(int id, void *peer, void *data)
961{
962 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
963 int ret;
964
965 if (net_cb->idx < net_cb->s_idx)
966 goto cont;
967
968 net_cb->fillargs.nsid = id;
969 if (net_cb->fillargs.add_ref)
970 net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
971 ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
972 if (ret < 0)
973 return ret;
974
975cont:
976 net_cb->idx++;
977 return 0;
978}
979
980static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
981 struct rtnl_net_dump_cb *net_cb,
982 struct netlink_callback *cb)
983{
984 struct netlink_ext_ack *extack = cb->extack;
985 struct nlattr *tb[NETNSA_MAX + 1];
986 int err, i;
987
988 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
989 NETNSA_MAX, rtnl_net_policy,
990 extack);
991 if (err < 0)
992 return err;
993
994 for (i = 0; i <= NETNSA_MAX; i++) {
995 if (!tb[i])
996 continue;
997
998 if (i == NETNSA_TARGET_NSID) {
999 struct net *net;
1000
1001 net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
1002 if (IS_ERR(net)) {
1003 NL_SET_BAD_ATTR(extack, tb[i]);
1004 NL_SET_ERR_MSG(extack,
1005 "Invalid target network namespace id");
1006 return PTR_ERR(net);
1007 }
1008 net_cb->fillargs.add_ref = true;
1009 net_cb->ref_net = net_cb->tgt_net;
1010 net_cb->tgt_net = net;
1011 } else {
1012 NL_SET_BAD_ATTR(extack, tb[i]);
1013 NL_SET_ERR_MSG(extack,
1014 "Unsupported attribute in dump request");
1015 return -EINVAL;
1016 }
1017 }
1018
1019 return 0;
1020}
1021
1022static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1023{
1024 struct rtnl_net_dump_cb net_cb = {
1025 .tgt_net = sock_net(skb->sk),
1026 .skb = skb,
1027 .fillargs = {
1028 .portid = NETLINK_CB(cb->skb).portid,
1029 .seq = cb->nlh->nlmsg_seq,
1030 .flags = NLM_F_MULTI,
1031 .cmd = RTM_NEWNSID,
1032 },
1033 .idx = 0,
1034 .s_idx = cb->args[0],
1035 };
1036 int err = 0;
1037
1038 if (cb->strict_check) {
1039 err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1040 if (err < 0)
1041 goto end;
1042 }
1043
1044 rcu_read_lock();
1045 idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1046 rcu_read_unlock();
1047
1048 cb->args[0] = net_cb.idx;
1049end:
1050 if (net_cb.fillargs.add_ref)
1051 put_net(net_cb.tgt_net);
1052 return err < 0 ? err : skb->len;
1053}
1054
1055static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1056 struct nlmsghdr *nlh, gfp_t gfp)
1057{
1058 struct net_fill_args fillargs = {
1059 .portid = portid,
1060 .seq = nlh ? nlh->nlmsg_seq : 0,
1061 .cmd = cmd,
1062 .nsid = id,
1063 };
1064 struct sk_buff *msg;
1065 int err = -ENOMEM;
1066
1067 msg = nlmsg_new(rtnl_net_get_size(), gfp);
1068 if (!msg)
1069 goto out;
1070
1071 err = rtnl_net_fill(msg, &fillargs);
1072 if (err < 0)
1073 goto err_out;
1074
1075 rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, gfp);
1076 return;
1077
1078err_out:
1079 nlmsg_free(msg);
1080out:
1081 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1082}
1083
1084static int __init net_ns_init(void)
1085{
1086 struct net_generic *ng;
1087
1088#ifdef CONFIG_NET_NS
1089 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1090 SMP_CACHE_BYTES,
1091 SLAB_PANIC|SLAB_ACCOUNT, NULL);
1092
1093 /* Create workqueue for cleanup */
1094 netns_wq = create_singlethread_workqueue("netns");
1095 if (!netns_wq)
1096 panic("Could not create netns workq");
1097#endif
1098
1099 ng = net_alloc_generic();
1100 if (!ng)
1101 panic("Could not allocate generic netns");
1102
1103 rcu_assign_pointer(init_net.gen, ng);
1104 net_gen_cookie(&init_net);
1105
1106 down_write(&pernet_ops_rwsem);
1107 if (setup_net(&init_net, &init_user_ns))
1108 panic("Could not setup the initial network namespace");
1109
1110 init_net_initialized = true;
1111 up_write(&pernet_ops_rwsem);
1112
1113 if (register_pernet_subsys(&net_ns_ops))
1114 panic("Could not register network namespace subsystems");
1115
1116 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
1117 RTNL_FLAG_DOIT_UNLOCKED);
1118 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
1119 RTNL_FLAG_DOIT_UNLOCKED);
1120
1121 return 0;
1122}
1123
1124pure_initcall(net_ns_init);
1125
1126#ifdef CONFIG_NET_NS
1127static int __register_pernet_operations(struct list_head *list,
1128 struct pernet_operations *ops)
1129{
1130 struct net *net;
1131 int error;
1132 LIST_HEAD(net_exit_list);
1133
1134 list_add_tail(&ops->list, list);
1135 if (ops->init || (ops->id && ops->size)) {
1136 /* We held write locked pernet_ops_rwsem, and parallel
1137 * setup_net() and cleanup_net() are not possible.
1138 */
1139 for_each_net(net) {
1140 error = ops_init(ops, net);
1141 if (error)
1142 goto out_undo;
1143 list_add_tail(&net->exit_list, &net_exit_list);
1144 }
1145 }
1146 return 0;
1147
1148out_undo:
1149 /* If I have an error cleanup all namespaces I initialized */
1150 list_del(&ops->list);
1151 ops_pre_exit_list(ops, &net_exit_list);
1152 synchronize_rcu();
1153 ops_exit_list(ops, &net_exit_list);
1154 ops_free_list(ops, &net_exit_list);
1155 return error;
1156}
1157
1158static void __unregister_pernet_operations(struct pernet_operations *ops)
1159{
1160 struct net *net;
1161 LIST_HEAD(net_exit_list);
1162
1163 list_del(&ops->list);
1164 /* See comment in __register_pernet_operations() */
1165 for_each_net(net)
1166 list_add_tail(&net->exit_list, &net_exit_list);
1167 ops_pre_exit_list(ops, &net_exit_list);
1168 synchronize_rcu();
1169 ops_exit_list(ops, &net_exit_list);
1170 ops_free_list(ops, &net_exit_list);
1171}
1172
1173#else
1174
1175static int __register_pernet_operations(struct list_head *list,
1176 struct pernet_operations *ops)
1177{
1178 if (!init_net_initialized) {
1179 list_add_tail(&ops->list, list);
1180 return 0;
1181 }
1182
1183 return ops_init(ops, &init_net);
1184}
1185
1186static void __unregister_pernet_operations(struct pernet_operations *ops)
1187{
1188 if (!init_net_initialized) {
1189 list_del(&ops->list);
1190 } else {
1191 LIST_HEAD(net_exit_list);
1192 list_add(&init_net.exit_list, &net_exit_list);
1193 ops_pre_exit_list(ops, &net_exit_list);
1194 synchronize_rcu();
1195 ops_exit_list(ops, &net_exit_list);
1196 ops_free_list(ops, &net_exit_list);
1197 }
1198}
1199
1200#endif /* CONFIG_NET_NS */
1201
1202static DEFINE_IDA(net_generic_ids);
1203
1204static int register_pernet_operations(struct list_head *list,
1205 struct pernet_operations *ops)
1206{
1207 int error;
1208
1209 if (ops->id) {
1210 error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID,
1211 GFP_KERNEL);
1212 if (error < 0)
1213 return error;
1214 *ops->id = error;
1215 max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
1216 }
1217 error = __register_pernet_operations(list, ops);
1218 if (error) {
1219 rcu_barrier();
1220 if (ops->id)
1221 ida_free(&net_generic_ids, *ops->id);
1222 }
1223
1224 return error;
1225}
1226
1227static void unregister_pernet_operations(struct pernet_operations *ops)
1228{
1229 __unregister_pernet_operations(ops);
1230 rcu_barrier();
1231 if (ops->id)
1232 ida_free(&net_generic_ids, *ops->id);
1233}
1234
1235/**
1236 * register_pernet_subsys - register a network namespace subsystem
1237 * @ops: pernet operations structure for the subsystem
1238 *
1239 * Register a subsystem which has init and exit functions
1240 * that are called when network namespaces are created and
1241 * destroyed respectively.
1242 *
1243 * When registered all network namespace init functions are
1244 * called for every existing network namespace. Allowing kernel
1245 * modules to have a race free view of the set of network namespaces.
1246 *
1247 * When a new network namespace is created all of the init
1248 * methods are called in the order in which they were registered.
1249 *
1250 * When a network namespace is destroyed all of the exit methods
1251 * are called in the reverse of the order with which they were
1252 * registered.
1253 */
1254int register_pernet_subsys(struct pernet_operations *ops)
1255{
1256 int error;
1257 down_write(&pernet_ops_rwsem);
1258 error = register_pernet_operations(first_device, ops);
1259 up_write(&pernet_ops_rwsem);
1260 return error;
1261}
1262EXPORT_SYMBOL_GPL(register_pernet_subsys);
1263
1264/**
1265 * unregister_pernet_subsys - unregister a network namespace subsystem
1266 * @ops: pernet operations structure to manipulate
1267 *
1268 * Remove the pernet operations structure from the list to be
1269 * used when network namespaces are created or destroyed. In
1270 * addition run the exit method for all existing network
1271 * namespaces.
1272 */
1273void unregister_pernet_subsys(struct pernet_operations *ops)
1274{
1275 down_write(&pernet_ops_rwsem);
1276 unregister_pernet_operations(ops);
1277 up_write(&pernet_ops_rwsem);
1278}
1279EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1280
1281/**
1282 * register_pernet_device - register a network namespace device
1283 * @ops: pernet operations structure for the subsystem
1284 *
1285 * Register a device which has init and exit functions
1286 * that are called when network namespaces are created and
1287 * destroyed respectively.
1288 *
1289 * When registered all network namespace init functions are
1290 * called for every existing network namespace. Allowing kernel
1291 * modules to have a race free view of the set of network namespaces.
1292 *
1293 * When a new network namespace is created all of the init
1294 * methods are called in the order in which they were registered.
1295 *
1296 * When a network namespace is destroyed all of the exit methods
1297 * are called in the reverse of the order with which they were
1298 * registered.
1299 */
1300int register_pernet_device(struct pernet_operations *ops)
1301{
1302 int error;
1303 down_write(&pernet_ops_rwsem);
1304 error = register_pernet_operations(&pernet_list, ops);
1305 if (!error && (first_device == &pernet_list))
1306 first_device = &ops->list;
1307 up_write(&pernet_ops_rwsem);
1308 return error;
1309}
1310EXPORT_SYMBOL_GPL(register_pernet_device);
1311
1312/**
1313 * unregister_pernet_device - unregister a network namespace netdevice
1314 * @ops: pernet operations structure to manipulate
1315 *
1316 * Remove the pernet operations structure from the list to be
1317 * used when network namespaces are created or destroyed. In
1318 * addition run the exit method for all existing network
1319 * namespaces.
1320 */
1321void unregister_pernet_device(struct pernet_operations *ops)
1322{
1323 down_write(&pernet_ops_rwsem);
1324 if (&ops->list == first_device)
1325 first_device = first_device->next;
1326 unregister_pernet_operations(ops);
1327 up_write(&pernet_ops_rwsem);
1328}
1329EXPORT_SYMBOL_GPL(unregister_pernet_device);
1330
1331#ifdef CONFIG_NET_NS
1332static struct ns_common *netns_get(struct task_struct *task)
1333{
1334 struct net *net = NULL;
1335 struct nsproxy *nsproxy;
1336
1337 task_lock(task);
1338 nsproxy = task->nsproxy;
1339 if (nsproxy)
1340 net = get_net(nsproxy->net_ns);
1341 task_unlock(task);
1342
1343 return net ? &net->ns : NULL;
1344}
1345
1346static inline struct net *to_net_ns(struct ns_common *ns)
1347{
1348 return container_of(ns, struct net, ns);
1349}
1350
1351static void netns_put(struct ns_common *ns)
1352{
1353 put_net(to_net_ns(ns));
1354}
1355
1356static int netns_install(struct nsset *nsset, struct ns_common *ns)
1357{
1358 struct nsproxy *nsproxy = nsset->nsproxy;
1359 struct net *net = to_net_ns(ns);
1360
1361 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1362 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
1363 return -EPERM;
1364
1365 put_net(nsproxy->net_ns);
1366 nsproxy->net_ns = get_net(net);
1367 return 0;
1368}
1369
1370static struct user_namespace *netns_owner(struct ns_common *ns)
1371{
1372 return to_net_ns(ns)->user_ns;
1373}
1374
1375const struct proc_ns_operations netns_operations = {
1376 .name = "net",
1377 .type = CLONE_NEWNET,
1378 .get = netns_get,
1379 .put = netns_put,
1380 .install = netns_install,
1381 .owner = netns_owner,
1382};
1383#endif