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#include <linux/cookie.h>
23
24#include <net/sock.h>
25#include <net/netlink.h>
26#include <net/net_namespace.h>
27#include <net/netns/generic.h>
28
29/*
30 * Our network namespace constructor/destructor lists
31 */
32
33static LIST_HEAD(pernet_list);
34static struct list_head *first_device = &pernet_list;
35
36LIST_HEAD(net_namespace_list);
37EXPORT_SYMBOL_GPL(net_namespace_list);
38
39/* Protects net_namespace_list. Nests iside rtnl_lock() */
40DECLARE_RWSEM(net_rwsem);
41EXPORT_SYMBOL_GPL(net_rwsem);
42
43#ifdef CONFIG_KEYS
44static struct key_tag init_net_key_domain = { .usage = REFCOUNT_INIT(1) };
45#endif
46
47struct net init_net;
48EXPORT_SYMBOL(init_net);
49
50static bool init_net_initialized;
51/*
52 * pernet_ops_rwsem: protects: pernet_list, net_generic_ids,
53 * init_net_initialized and first_device pointer.
54 * This is internal net namespace object. Please, don't use it
55 * outside.
56 */
57DECLARE_RWSEM(pernet_ops_rwsem);
58EXPORT_SYMBOL_GPL(pernet_ops_rwsem);
59
60#define MIN_PERNET_OPS_ID \
61 ((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *))
62
63#define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
64
65static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
66
67DEFINE_COOKIE(net_cookie);
68
69static struct net_generic *net_alloc_generic(void)
70{
71 struct net_generic *ng;
72 unsigned int generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
73
74 ng = kzalloc(generic_size, GFP_KERNEL);
75 if (ng)
76 ng->s.len = max_gen_ptrs;
77
78 return ng;
79}
80
81static int net_assign_generic(struct net *net, unsigned int id, void *data)
82{
83 struct net_generic *ng, *old_ng;
84
85 BUG_ON(id < MIN_PERNET_OPS_ID);
86
87 old_ng = rcu_dereference_protected(net->gen,
88 lockdep_is_held(&pernet_ops_rwsem));
89 if (old_ng->s.len > id) {
90 old_ng->ptr[id] = data;
91 return 0;
92 }
93
94 ng = net_alloc_generic();
95 if (!ng)
96 return -ENOMEM;
97
98 /*
99 * Some synchronisation notes:
100 *
101 * The net_generic explores the net->gen array inside rcu
102 * read section. Besides once set the net->gen->ptr[x]
103 * pointer never changes (see rules in netns/generic.h).
104 *
105 * That said, we simply duplicate this array and schedule
106 * the old copy for kfree after a grace period.
107 */
108
109 memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
110 (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
111 ng->ptr[id] = data;
112
113 rcu_assign_pointer(net->gen, ng);
114 kfree_rcu(old_ng, s.rcu);
115 return 0;
116}
117
118static int ops_init(const struct pernet_operations *ops, struct net *net)
119{
120 struct net_generic *ng;
121 int err = -ENOMEM;
122 void *data = NULL;
123
124 if (ops->id && ops->size) {
125 data = kzalloc(ops->size, GFP_KERNEL);
126 if (!data)
127 goto out;
128
129 err = net_assign_generic(net, *ops->id, data);
130 if (err)
131 goto cleanup;
132 }
133 err = 0;
134 if (ops->init)
135 err = ops->init(net);
136 if (!err)
137 return 0;
138
139 if (ops->id && ops->size) {
140 ng = rcu_dereference_protected(net->gen,
141 lockdep_is_held(&pernet_ops_rwsem));
142 ng->ptr[*ops->id] = NULL;
143 }
144
145cleanup:
146 kfree(data);
147
148out:
149 return err;
150}
151
152static void ops_pre_exit_list(const struct pernet_operations *ops,
153 struct list_head *net_exit_list)
154{
155 struct net *net;
156
157 if (ops->pre_exit) {
158 list_for_each_entry(net, net_exit_list, exit_list)
159 ops->pre_exit(net);
160 }
161}
162
163static void ops_exit_list(const struct pernet_operations *ops,
164 struct list_head *net_exit_list)
165{
166 struct net *net;
167 if (ops->exit) {
168 list_for_each_entry(net, net_exit_list, exit_list) {
169 ops->exit(net);
170 cond_resched();
171 }
172 }
173 if (ops->exit_batch)
174 ops->exit_batch(net_exit_list);
175}
176
177static void ops_free_list(const struct pernet_operations *ops,
178 struct list_head *net_exit_list)
179{
180 struct net *net;
181 if (ops->size && ops->id) {
182 list_for_each_entry(net, net_exit_list, exit_list)
183 kfree(net_generic(net, *ops->id));
184 }
185}
186
187/* should be called with nsid_lock held */
188static int alloc_netid(struct net *net, struct net *peer, int reqid)
189{
190 int min = 0, max = 0;
191
192 if (reqid >= 0) {
193 min = reqid;
194 max = reqid + 1;
195 }
196
197 return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
198}
199
200/* This function is used by idr_for_each(). If net is equal to peer, the
201 * function returns the id so that idr_for_each() stops. Because we cannot
202 * returns the id 0 (idr_for_each() will not stop), we return the magic value
203 * NET_ID_ZERO (-1) for it.
204 */
205#define NET_ID_ZERO -1
206static int net_eq_idr(int id, void *net, void *peer)
207{
208 if (net_eq(net, peer))
209 return id ? : NET_ID_ZERO;
210 return 0;
211}
212
213/* Must be called from RCU-critical section or with nsid_lock held */
214static int __peernet2id(const struct net *net, struct net *peer)
215{
216 int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
217
218 /* Magic value for id 0. */
219 if (id == NET_ID_ZERO)
220 return 0;
221 if (id > 0)
222 return id;
223
224 return NETNSA_NSID_NOT_ASSIGNED;
225}
226
227static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
228 struct nlmsghdr *nlh, gfp_t gfp);
229/* This function returns the id of a peer netns. If no id is assigned, one will
230 * be allocated and returned.
231 */
232int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp)
233{
234 int id;
235
236 if (refcount_read(&net->ns.count) == 0)
237 return NETNSA_NSID_NOT_ASSIGNED;
238
239 spin_lock_bh(&net->nsid_lock);
240 id = __peernet2id(net, peer);
241 if (id >= 0) {
242 spin_unlock_bh(&net->nsid_lock);
243 return id;
244 }
245
246 /* When peer is obtained from RCU lists, we may race with
247 * its cleanup. Check whether it's alive, and this guarantees
248 * we never hash a peer back to net->netns_ids, after it has
249 * just been idr_remove()'d from there in cleanup_net().
250 */
251 if (!maybe_get_net(peer)) {
252 spin_unlock_bh(&net->nsid_lock);
253 return NETNSA_NSID_NOT_ASSIGNED;
254 }
255
256 id = alloc_netid(net, peer, -1);
257 spin_unlock_bh(&net->nsid_lock);
258
259 put_net(peer);
260 if (id < 0)
261 return NETNSA_NSID_NOT_ASSIGNED;
262
263 rtnl_net_notifyid(net, RTM_NEWNSID, id, 0, NULL, gfp);
264
265 return id;
266}
267EXPORT_SYMBOL_GPL(peernet2id_alloc);
268
269/* This function returns, if assigned, the id of a peer netns. */
270int peernet2id(const struct net *net, struct net *peer)
271{
272 int id;
273
274 rcu_read_lock();
275 id = __peernet2id(net, peer);
276 rcu_read_unlock();
277
278 return id;
279}
280EXPORT_SYMBOL(peernet2id);
281
282/* This function returns true is the peer netns has an id assigned into the
283 * current netns.
284 */
285bool peernet_has_id(const struct net *net, struct net *peer)
286{
287 return peernet2id(net, peer) >= 0;
288}
289
290struct net *get_net_ns_by_id(const struct net *net, int id)
291{
292 struct net *peer;
293
294 if (id < 0)
295 return NULL;
296
297 rcu_read_lock();
298 peer = idr_find(&net->netns_ids, id);
299 if (peer)
300 peer = maybe_get_net(peer);
301 rcu_read_unlock();
302
303 return peer;
304}
305EXPORT_SYMBOL_GPL(get_net_ns_by_id);
306
307/* init code that must occur even if setup_net() is not called. */
308static __net_init void preinit_net(struct net *net)
309{
310 ref_tracker_dir_init(&net->notrefcnt_tracker, 128);
311}
312
313/*
314 * setup_net runs the initializers for the network namespace object.
315 */
316static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
317{
318 /* Must be called with pernet_ops_rwsem held */
319 const struct pernet_operations *ops, *saved_ops;
320 int error = 0;
321 LIST_HEAD(net_exit_list);
322
323 refcount_set(&net->ns.count, 1);
324 ref_tracker_dir_init(&net->refcnt_tracker, 128);
325
326 refcount_set(&net->passive, 1);
327 get_random_bytes(&net->hash_mix, sizeof(u32));
328 preempt_disable();
329 net->net_cookie = gen_cookie_next(&net_cookie);
330 preempt_enable();
331 net->dev_base_seq = 1;
332 net->user_ns = user_ns;
333 idr_init(&net->netns_ids);
334 spin_lock_init(&net->nsid_lock);
335 mutex_init(&net->ipv4.ra_mutex);
336
337 list_for_each_entry(ops, &pernet_list, list) {
338 error = ops_init(ops, net);
339 if (error < 0)
340 goto out_undo;
341 }
342 down_write(&net_rwsem);
343 list_add_tail_rcu(&net->list, &net_namespace_list);
344 up_write(&net_rwsem);
345out:
346 return error;
347
348out_undo:
349 /* Walk through the list backwards calling the exit functions
350 * for the pernet modules whose init functions did not fail.
351 */
352 list_add(&net->exit_list, &net_exit_list);
353 saved_ops = ops;
354 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
355 ops_pre_exit_list(ops, &net_exit_list);
356
357 synchronize_rcu();
358
359 ops = saved_ops;
360 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
361 ops_exit_list(ops, &net_exit_list);
362
363 ops = saved_ops;
364 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
365 ops_free_list(ops, &net_exit_list);
366
367 rcu_barrier();
368 goto out;
369}
370
371static int __net_init net_defaults_init_net(struct net *net)
372{
373 net->core.sysctl_somaxconn = SOMAXCONN;
374 net->core.sysctl_txrehash = SOCK_TXREHASH_ENABLED;
375
376 return 0;
377}
378
379static struct pernet_operations net_defaults_ops = {
380 .init = net_defaults_init_net,
381};
382
383static __init int net_defaults_init(void)
384{
385 if (register_pernet_subsys(&net_defaults_ops))
386 panic("Cannot initialize net default settings");
387
388 return 0;
389}
390
391core_initcall(net_defaults_init);
392
393#ifdef CONFIG_NET_NS
394static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
395{
396 return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
397}
398
399static void dec_net_namespaces(struct ucounts *ucounts)
400{
401 dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
402}
403
404static struct kmem_cache *net_cachep __ro_after_init;
405static struct workqueue_struct *netns_wq;
406
407static struct net *net_alloc(void)
408{
409 struct net *net = NULL;
410 struct net_generic *ng;
411
412 ng = net_alloc_generic();
413 if (!ng)
414 goto out;
415
416 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
417 if (!net)
418 goto out_free;
419
420#ifdef CONFIG_KEYS
421 net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
422 if (!net->key_domain)
423 goto out_free_2;
424 refcount_set(&net->key_domain->usage, 1);
425#endif
426
427 rcu_assign_pointer(net->gen, ng);
428out:
429 return net;
430
431#ifdef CONFIG_KEYS
432out_free_2:
433 kmem_cache_free(net_cachep, net);
434 net = NULL;
435#endif
436out_free:
437 kfree(ng);
438 goto out;
439}
440
441static void net_free(struct net *net)
442{
443 if (refcount_dec_and_test(&net->passive)) {
444 kfree(rcu_access_pointer(net->gen));
445
446 /* There should not be any trackers left there. */
447 ref_tracker_dir_exit(&net->notrefcnt_tracker);
448
449 kmem_cache_free(net_cachep, net);
450 }
451}
452
453void net_drop_ns(void *p)
454{
455 struct net *net = (struct net *)p;
456
457 if (net)
458 net_free(net);
459}
460
461struct net *copy_net_ns(unsigned long flags,
462 struct user_namespace *user_ns, struct net *old_net)
463{
464 struct ucounts *ucounts;
465 struct net *net;
466 int rv;
467
468 if (!(flags & CLONE_NEWNET))
469 return get_net(old_net);
470
471 ucounts = inc_net_namespaces(user_ns);
472 if (!ucounts)
473 return ERR_PTR(-ENOSPC);
474
475 net = net_alloc();
476 if (!net) {
477 rv = -ENOMEM;
478 goto dec_ucounts;
479 }
480
481 preinit_net(net);
482 refcount_set(&net->passive, 1);
483 net->ucounts = ucounts;
484 get_user_ns(user_ns);
485
486 rv = down_read_killable(&pernet_ops_rwsem);
487 if (rv < 0)
488 goto put_userns;
489
490 rv = setup_net(net, user_ns);
491
492 up_read(&pernet_ops_rwsem);
493
494 if (rv < 0) {
495put_userns:
496#ifdef CONFIG_KEYS
497 key_remove_domain(net->key_domain);
498#endif
499 put_user_ns(user_ns);
500 net_free(net);
501dec_ucounts:
502 dec_net_namespaces(ucounts);
503 return ERR_PTR(rv);
504 }
505 return net;
506}
507
508/**
509 * net_ns_get_ownership - get sysfs ownership data for @net
510 * @net: network namespace in question (can be NULL)
511 * @uid: kernel user ID for sysfs objects
512 * @gid: kernel group ID for sysfs objects
513 *
514 * Returns the uid/gid pair of root in the user namespace associated with the
515 * given network namespace.
516 */
517void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
518{
519 if (net) {
520 kuid_t ns_root_uid = make_kuid(net->user_ns, 0);
521 kgid_t ns_root_gid = make_kgid(net->user_ns, 0);
522
523 if (uid_valid(ns_root_uid))
524 *uid = ns_root_uid;
525
526 if (gid_valid(ns_root_gid))
527 *gid = ns_root_gid;
528 } else {
529 *uid = GLOBAL_ROOT_UID;
530 *gid = GLOBAL_ROOT_GID;
531 }
532}
533EXPORT_SYMBOL_GPL(net_ns_get_ownership);
534
535static void unhash_nsid(struct net *net, struct net *last)
536{
537 struct net *tmp;
538 /* This function is only called from cleanup_net() work,
539 * and this work is the only process, that may delete
540 * a net from net_namespace_list. So, when the below
541 * is executing, the list may only grow. Thus, we do not
542 * use for_each_net_rcu() or net_rwsem.
543 */
544 for_each_net(tmp) {
545 int id;
546
547 spin_lock_bh(&tmp->nsid_lock);
548 id = __peernet2id(tmp, net);
549 if (id >= 0)
550 idr_remove(&tmp->netns_ids, id);
551 spin_unlock_bh(&tmp->nsid_lock);
552 if (id >= 0)
553 rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL,
554 GFP_KERNEL);
555 if (tmp == last)
556 break;
557 }
558 spin_lock_bh(&net->nsid_lock);
559 idr_destroy(&net->netns_ids);
560 spin_unlock_bh(&net->nsid_lock);
561}
562
563static LLIST_HEAD(cleanup_list);
564
565static void cleanup_net(struct work_struct *work)
566{
567 const struct pernet_operations *ops;
568 struct net *net, *tmp, *last;
569 struct llist_node *net_kill_list;
570 LIST_HEAD(net_exit_list);
571
572 /* Atomically snapshot the list of namespaces to cleanup */
573 net_kill_list = llist_del_all(&cleanup_list);
574
575 down_read(&pernet_ops_rwsem);
576
577 /* Don't let anyone else find us. */
578 down_write(&net_rwsem);
579 llist_for_each_entry(net, net_kill_list, cleanup_list)
580 list_del_rcu(&net->list);
581 /* Cache last net. After we unlock rtnl, no one new net
582 * added to net_namespace_list can assign nsid pointer
583 * to a net from net_kill_list (see peernet2id_alloc()).
584 * So, we skip them in unhash_nsid().
585 *
586 * Note, that unhash_nsid() does not delete nsid links
587 * between net_kill_list's nets, as they've already
588 * deleted from net_namespace_list. But, this would be
589 * useless anyway, as netns_ids are destroyed there.
590 */
591 last = list_last_entry(&net_namespace_list, struct net, list);
592 up_write(&net_rwsem);
593
594 llist_for_each_entry(net, net_kill_list, cleanup_list) {
595 unhash_nsid(net, last);
596 list_add_tail(&net->exit_list, &net_exit_list);
597 }
598
599 /* Run all of the network namespace pre_exit methods */
600 list_for_each_entry_reverse(ops, &pernet_list, list)
601 ops_pre_exit_list(ops, &net_exit_list);
602
603 /*
604 * Another CPU might be rcu-iterating the list, wait for it.
605 * This needs to be before calling the exit() notifiers, so
606 * the rcu_barrier() below isn't sufficient alone.
607 * Also the pre_exit() and exit() methods need this barrier.
608 */
609 synchronize_rcu();
610
611 /* Run all of the network namespace exit methods */
612 list_for_each_entry_reverse(ops, &pernet_list, list)
613 ops_exit_list(ops, &net_exit_list);
614
615 /* Free the net generic variables */
616 list_for_each_entry_reverse(ops, &pernet_list, list)
617 ops_free_list(ops, &net_exit_list);
618
619 up_read(&pernet_ops_rwsem);
620
621 /* Ensure there are no outstanding rcu callbacks using this
622 * network namespace.
623 */
624 rcu_barrier();
625
626 /* Finally it is safe to free my network namespace structure */
627 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
628 list_del_init(&net->exit_list);
629 dec_net_namespaces(net->ucounts);
630#ifdef CONFIG_KEYS
631 key_remove_domain(net->key_domain);
632#endif
633 put_user_ns(net->user_ns);
634 net_free(net);
635 }
636}
637
638/**
639 * net_ns_barrier - wait until concurrent net_cleanup_work is done
640 *
641 * cleanup_net runs from work queue and will first remove namespaces
642 * from the global list, then run net exit functions.
643 *
644 * Call this in module exit path to make sure that all netns
645 * ->exit ops have been invoked before the function is removed.
646 */
647void net_ns_barrier(void)
648{
649 down_write(&pernet_ops_rwsem);
650 up_write(&pernet_ops_rwsem);
651}
652EXPORT_SYMBOL(net_ns_barrier);
653
654static DECLARE_WORK(net_cleanup_work, cleanup_net);
655
656void __put_net(struct net *net)
657{
658 ref_tracker_dir_exit(&net->refcnt_tracker);
659 /* Cleanup the network namespace in process context */
660 if (llist_add(&net->cleanup_list, &cleanup_list))
661 queue_work(netns_wq, &net_cleanup_work);
662}
663EXPORT_SYMBOL_GPL(__put_net);
664
665/**
666 * get_net_ns - increment the refcount of the network namespace
667 * @ns: common namespace (net)
668 *
669 * Returns the net's common namespace.
670 */
671struct ns_common *get_net_ns(struct ns_common *ns)
672{
673 return &get_net(container_of(ns, struct net, ns))->ns;
674}
675EXPORT_SYMBOL_GPL(get_net_ns);
676
677struct net *get_net_ns_by_fd(int fd)
678{
679 struct file *file;
680 struct ns_common *ns;
681 struct net *net;
682
683 file = proc_ns_fget(fd);
684 if (IS_ERR(file))
685 return ERR_CAST(file);
686
687 ns = get_proc_ns(file_inode(file));
688 if (ns->ops == &netns_operations)
689 net = get_net(container_of(ns, struct net, ns));
690 else
691 net = ERR_PTR(-EINVAL);
692
693 fput(file);
694 return net;
695}
696EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
697#endif
698
699struct net *get_net_ns_by_pid(pid_t pid)
700{
701 struct task_struct *tsk;
702 struct net *net;
703
704 /* Lookup the network namespace */
705 net = ERR_PTR(-ESRCH);
706 rcu_read_lock();
707 tsk = find_task_by_vpid(pid);
708 if (tsk) {
709 struct nsproxy *nsproxy;
710 task_lock(tsk);
711 nsproxy = tsk->nsproxy;
712 if (nsproxy)
713 net = get_net(nsproxy->net_ns);
714 task_unlock(tsk);
715 }
716 rcu_read_unlock();
717 return net;
718}
719EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
720
721static __net_init int net_ns_net_init(struct net *net)
722{
723#ifdef CONFIG_NET_NS
724 net->ns.ops = &netns_operations;
725#endif
726 return ns_alloc_inum(&net->ns);
727}
728
729static __net_exit void net_ns_net_exit(struct net *net)
730{
731 ns_free_inum(&net->ns);
732}
733
734static struct pernet_operations __net_initdata net_ns_ops = {
735 .init = net_ns_net_init,
736 .exit = net_ns_net_exit,
737};
738
739static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
740 [NETNSA_NONE] = { .type = NLA_UNSPEC },
741 [NETNSA_NSID] = { .type = NLA_S32 },
742 [NETNSA_PID] = { .type = NLA_U32 },
743 [NETNSA_FD] = { .type = NLA_U32 },
744 [NETNSA_TARGET_NSID] = { .type = NLA_S32 },
745};
746
747static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
748 struct netlink_ext_ack *extack)
749{
750 struct net *net = sock_net(skb->sk);
751 struct nlattr *tb[NETNSA_MAX + 1];
752 struct nlattr *nla;
753 struct net *peer;
754 int nsid, err;
755
756 err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
757 NETNSA_MAX, rtnl_net_policy, extack);
758 if (err < 0)
759 return err;
760 if (!tb[NETNSA_NSID]) {
761 NL_SET_ERR_MSG(extack, "nsid is missing");
762 return -EINVAL;
763 }
764 nsid = nla_get_s32(tb[NETNSA_NSID]);
765
766 if (tb[NETNSA_PID]) {
767 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
768 nla = tb[NETNSA_PID];
769 } else if (tb[NETNSA_FD]) {
770 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
771 nla = tb[NETNSA_FD];
772 } else {
773 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
774 return -EINVAL;
775 }
776 if (IS_ERR(peer)) {
777 NL_SET_BAD_ATTR(extack, nla);
778 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
779 return PTR_ERR(peer);
780 }
781
782 spin_lock_bh(&net->nsid_lock);
783 if (__peernet2id(net, peer) >= 0) {
784 spin_unlock_bh(&net->nsid_lock);
785 err = -EEXIST;
786 NL_SET_BAD_ATTR(extack, nla);
787 NL_SET_ERR_MSG(extack,
788 "Peer netns already has a nsid assigned");
789 goto out;
790 }
791
792 err = alloc_netid(net, peer, nsid);
793 spin_unlock_bh(&net->nsid_lock);
794 if (err >= 0) {
795 rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid,
796 nlh, GFP_KERNEL);
797 err = 0;
798 } else if (err == -ENOSPC && nsid >= 0) {
799 err = -EEXIST;
800 NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
801 NL_SET_ERR_MSG(extack, "The specified nsid is already used");
802 }
803out:
804 put_net(peer);
805 return err;
806}
807
808static int rtnl_net_get_size(void)
809{
810 return NLMSG_ALIGN(sizeof(struct rtgenmsg))
811 + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
812 + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
813 ;
814}
815
816struct net_fill_args {
817 u32 portid;
818 u32 seq;
819 int flags;
820 int cmd;
821 int nsid;
822 bool add_ref;
823 int ref_nsid;
824};
825
826static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
827{
828 struct nlmsghdr *nlh;
829 struct rtgenmsg *rth;
830
831 nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
832 args->flags);
833 if (!nlh)
834 return -EMSGSIZE;
835
836 rth = nlmsg_data(nlh);
837 rth->rtgen_family = AF_UNSPEC;
838
839 if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
840 goto nla_put_failure;
841
842 if (args->add_ref &&
843 nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
844 goto nla_put_failure;
845
846 nlmsg_end(skb, nlh);
847 return 0;
848
849nla_put_failure:
850 nlmsg_cancel(skb, nlh);
851 return -EMSGSIZE;
852}
853
854static int rtnl_net_valid_getid_req(struct sk_buff *skb,
855 const struct nlmsghdr *nlh,
856 struct nlattr **tb,
857 struct netlink_ext_ack *extack)
858{
859 int i, err;
860
861 if (!netlink_strict_get_check(skb))
862 return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
863 tb, NETNSA_MAX, rtnl_net_policy,
864 extack);
865
866 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
867 NETNSA_MAX, rtnl_net_policy,
868 extack);
869 if (err)
870 return err;
871
872 for (i = 0; i <= NETNSA_MAX; i++) {
873 if (!tb[i])
874 continue;
875
876 switch (i) {
877 case NETNSA_PID:
878 case NETNSA_FD:
879 case NETNSA_NSID:
880 case NETNSA_TARGET_NSID:
881 break;
882 default:
883 NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
884 return -EINVAL;
885 }
886 }
887
888 return 0;
889}
890
891static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
892 struct netlink_ext_ack *extack)
893{
894 struct net *net = sock_net(skb->sk);
895 struct nlattr *tb[NETNSA_MAX + 1];
896 struct net_fill_args fillargs = {
897 .portid = NETLINK_CB(skb).portid,
898 .seq = nlh->nlmsg_seq,
899 .cmd = RTM_NEWNSID,
900 };
901 struct net *peer, *target = net;
902 struct nlattr *nla;
903 struct sk_buff *msg;
904 int err;
905
906 err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
907 if (err < 0)
908 return err;
909 if (tb[NETNSA_PID]) {
910 peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
911 nla = tb[NETNSA_PID];
912 } else if (tb[NETNSA_FD]) {
913 peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
914 nla = tb[NETNSA_FD];
915 } else if (tb[NETNSA_NSID]) {
916 peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
917 if (!peer)
918 peer = ERR_PTR(-ENOENT);
919 nla = tb[NETNSA_NSID];
920 } else {
921 NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
922 return -EINVAL;
923 }
924
925 if (IS_ERR(peer)) {
926 NL_SET_BAD_ATTR(extack, nla);
927 NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
928 return PTR_ERR(peer);
929 }
930
931 if (tb[NETNSA_TARGET_NSID]) {
932 int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
933
934 target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
935 if (IS_ERR(target)) {
936 NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
937 NL_SET_ERR_MSG(extack,
938 "Target netns reference is invalid");
939 err = PTR_ERR(target);
940 goto out;
941 }
942 fillargs.add_ref = true;
943 fillargs.ref_nsid = peernet2id(net, peer);
944 }
945
946 msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
947 if (!msg) {
948 err = -ENOMEM;
949 goto out;
950 }
951
952 fillargs.nsid = peernet2id(target, peer);
953 err = rtnl_net_fill(msg, &fillargs);
954 if (err < 0)
955 goto err_out;
956
957 err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
958 goto out;
959
960err_out:
961 nlmsg_free(msg);
962out:
963 if (fillargs.add_ref)
964 put_net(target);
965 put_net(peer);
966 return err;
967}
968
969struct rtnl_net_dump_cb {
970 struct net *tgt_net;
971 struct net *ref_net;
972 struct sk_buff *skb;
973 struct net_fill_args fillargs;
974 int idx;
975 int s_idx;
976};
977
978/* Runs in RCU-critical section. */
979static int rtnl_net_dumpid_one(int id, void *peer, void *data)
980{
981 struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
982 int ret;
983
984 if (net_cb->idx < net_cb->s_idx)
985 goto cont;
986
987 net_cb->fillargs.nsid = id;
988 if (net_cb->fillargs.add_ref)
989 net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
990 ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
991 if (ret < 0)
992 return ret;
993
994cont:
995 net_cb->idx++;
996 return 0;
997}
998
999static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
1000 struct rtnl_net_dump_cb *net_cb,
1001 struct netlink_callback *cb)
1002{
1003 struct netlink_ext_ack *extack = cb->extack;
1004 struct nlattr *tb[NETNSA_MAX + 1];
1005 int err, i;
1006
1007 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
1008 NETNSA_MAX, rtnl_net_policy,
1009 extack);
1010 if (err < 0)
1011 return err;
1012
1013 for (i = 0; i <= NETNSA_MAX; i++) {
1014 if (!tb[i])
1015 continue;
1016
1017 if (i == NETNSA_TARGET_NSID) {
1018 struct net *net;
1019
1020 net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
1021 if (IS_ERR(net)) {
1022 NL_SET_BAD_ATTR(extack, tb[i]);
1023 NL_SET_ERR_MSG(extack,
1024 "Invalid target network namespace id");
1025 return PTR_ERR(net);
1026 }
1027 net_cb->fillargs.add_ref = true;
1028 net_cb->ref_net = net_cb->tgt_net;
1029 net_cb->tgt_net = net;
1030 } else {
1031 NL_SET_BAD_ATTR(extack, tb[i]);
1032 NL_SET_ERR_MSG(extack,
1033 "Unsupported attribute in dump request");
1034 return -EINVAL;
1035 }
1036 }
1037
1038 return 0;
1039}
1040
1041static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1042{
1043 struct rtnl_net_dump_cb net_cb = {
1044 .tgt_net = sock_net(skb->sk),
1045 .skb = skb,
1046 .fillargs = {
1047 .portid = NETLINK_CB(cb->skb).portid,
1048 .seq = cb->nlh->nlmsg_seq,
1049 .flags = NLM_F_MULTI,
1050 .cmd = RTM_NEWNSID,
1051 },
1052 .idx = 0,
1053 .s_idx = cb->args[0],
1054 };
1055 int err = 0;
1056
1057 if (cb->strict_check) {
1058 err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1059 if (err < 0)
1060 goto end;
1061 }
1062
1063 rcu_read_lock();
1064 idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1065 rcu_read_unlock();
1066
1067 cb->args[0] = net_cb.idx;
1068end:
1069 if (net_cb.fillargs.add_ref)
1070 put_net(net_cb.tgt_net);
1071 return err < 0 ? err : skb->len;
1072}
1073
1074static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1075 struct nlmsghdr *nlh, gfp_t gfp)
1076{
1077 struct net_fill_args fillargs = {
1078 .portid = portid,
1079 .seq = nlh ? nlh->nlmsg_seq : 0,
1080 .cmd = cmd,
1081 .nsid = id,
1082 };
1083 struct sk_buff *msg;
1084 int err = -ENOMEM;
1085
1086 msg = nlmsg_new(rtnl_net_get_size(), gfp);
1087 if (!msg)
1088 goto out;
1089
1090 err = rtnl_net_fill(msg, &fillargs);
1091 if (err < 0)
1092 goto err_out;
1093
1094 rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, gfp);
1095 return;
1096
1097err_out:
1098 nlmsg_free(msg);
1099out:
1100 rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1101}
1102
1103void __init net_ns_init(void)
1104{
1105 struct net_generic *ng;
1106
1107#ifdef CONFIG_NET_NS
1108 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1109 SMP_CACHE_BYTES,
1110 SLAB_PANIC|SLAB_ACCOUNT, NULL);
1111
1112 /* Create workqueue for cleanup */
1113 netns_wq = create_singlethread_workqueue("netns");
1114 if (!netns_wq)
1115 panic("Could not create netns workq");
1116#endif
1117
1118 ng = net_alloc_generic();
1119 if (!ng)
1120 panic("Could not allocate generic netns");
1121
1122 rcu_assign_pointer(init_net.gen, ng);
1123
1124#ifdef CONFIG_KEYS
1125 init_net.key_domain = &init_net_key_domain;
1126#endif
1127 down_write(&pernet_ops_rwsem);
1128 preinit_net(&init_net);
1129 if (setup_net(&init_net, &init_user_ns))
1130 panic("Could not setup the initial network namespace");
1131
1132 init_net_initialized = true;
1133 up_write(&pernet_ops_rwsem);
1134
1135 if (register_pernet_subsys(&net_ns_ops))
1136 panic("Could not register network namespace subsystems");
1137
1138 rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
1139 RTNL_FLAG_DOIT_UNLOCKED);
1140 rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
1141 RTNL_FLAG_DOIT_UNLOCKED);
1142}
1143
1144static void free_exit_list(struct pernet_operations *ops, struct list_head *net_exit_list)
1145{
1146 ops_pre_exit_list(ops, net_exit_list);
1147 synchronize_rcu();
1148 ops_exit_list(ops, net_exit_list);
1149 ops_free_list(ops, net_exit_list);
1150}
1151
1152#ifdef CONFIG_NET_NS
1153static int __register_pernet_operations(struct list_head *list,
1154 struct pernet_operations *ops)
1155{
1156 struct net *net;
1157 int error;
1158 LIST_HEAD(net_exit_list);
1159
1160 list_add_tail(&ops->list, list);
1161 if (ops->init || (ops->id && ops->size)) {
1162 /* We held write locked pernet_ops_rwsem, and parallel
1163 * setup_net() and cleanup_net() are not possible.
1164 */
1165 for_each_net(net) {
1166 error = ops_init(ops, net);
1167 if (error)
1168 goto out_undo;
1169 list_add_tail(&net->exit_list, &net_exit_list);
1170 }
1171 }
1172 return 0;
1173
1174out_undo:
1175 /* If I have an error cleanup all namespaces I initialized */
1176 list_del(&ops->list);
1177 free_exit_list(ops, &net_exit_list);
1178 return error;
1179}
1180
1181static void __unregister_pernet_operations(struct pernet_operations *ops)
1182{
1183 struct net *net;
1184 LIST_HEAD(net_exit_list);
1185
1186 list_del(&ops->list);
1187 /* See comment in __register_pernet_operations() */
1188 for_each_net(net)
1189 list_add_tail(&net->exit_list, &net_exit_list);
1190
1191 free_exit_list(ops, &net_exit_list);
1192}
1193
1194#else
1195
1196static int __register_pernet_operations(struct list_head *list,
1197 struct pernet_operations *ops)
1198{
1199 if (!init_net_initialized) {
1200 list_add_tail(&ops->list, list);
1201 return 0;
1202 }
1203
1204 return ops_init(ops, &init_net);
1205}
1206
1207static void __unregister_pernet_operations(struct pernet_operations *ops)
1208{
1209 if (!init_net_initialized) {
1210 list_del(&ops->list);
1211 } else {
1212 LIST_HEAD(net_exit_list);
1213 list_add(&init_net.exit_list, &net_exit_list);
1214 free_exit_list(ops, &net_exit_list);
1215 }
1216}
1217
1218#endif /* CONFIG_NET_NS */
1219
1220static DEFINE_IDA(net_generic_ids);
1221
1222static int register_pernet_operations(struct list_head *list,
1223 struct pernet_operations *ops)
1224{
1225 int error;
1226
1227 if (ops->id) {
1228 error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID,
1229 GFP_KERNEL);
1230 if (error < 0)
1231 return error;
1232 *ops->id = error;
1233 max_gen_ptrs = max(max_gen_ptrs, *ops->id + 1);
1234 }
1235 error = __register_pernet_operations(list, ops);
1236 if (error) {
1237 rcu_barrier();
1238 if (ops->id)
1239 ida_free(&net_generic_ids, *ops->id);
1240 }
1241
1242 return error;
1243}
1244
1245static void unregister_pernet_operations(struct pernet_operations *ops)
1246{
1247 __unregister_pernet_operations(ops);
1248 rcu_barrier();
1249 if (ops->id)
1250 ida_free(&net_generic_ids, *ops->id);
1251}
1252
1253/**
1254 * register_pernet_subsys - register a network namespace subsystem
1255 * @ops: pernet operations structure for the subsystem
1256 *
1257 * Register a subsystem which has init and exit functions
1258 * that are called when network namespaces are created and
1259 * destroyed respectively.
1260 *
1261 * When registered all network namespace init functions are
1262 * called for every existing network namespace. Allowing kernel
1263 * modules to have a race free view of the set of network namespaces.
1264 *
1265 * When a new network namespace is created all of the init
1266 * methods are called in the order in which they were registered.
1267 *
1268 * When a network namespace is destroyed all of the exit methods
1269 * are called in the reverse of the order with which they were
1270 * registered.
1271 */
1272int register_pernet_subsys(struct pernet_operations *ops)
1273{
1274 int error;
1275 down_write(&pernet_ops_rwsem);
1276 error = register_pernet_operations(first_device, ops);
1277 up_write(&pernet_ops_rwsem);
1278 return error;
1279}
1280EXPORT_SYMBOL_GPL(register_pernet_subsys);
1281
1282/**
1283 * unregister_pernet_subsys - unregister a network namespace subsystem
1284 * @ops: pernet operations structure to manipulate
1285 *
1286 * Remove the pernet operations structure from the list to be
1287 * used when network namespaces are created or destroyed. In
1288 * addition run the exit method for all existing network
1289 * namespaces.
1290 */
1291void unregister_pernet_subsys(struct pernet_operations *ops)
1292{
1293 down_write(&pernet_ops_rwsem);
1294 unregister_pernet_operations(ops);
1295 up_write(&pernet_ops_rwsem);
1296}
1297EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1298
1299/**
1300 * register_pernet_device - register a network namespace device
1301 * @ops: pernet operations structure for the subsystem
1302 *
1303 * Register a device which has init and exit functions
1304 * that are called when network namespaces are created and
1305 * destroyed respectively.
1306 *
1307 * When registered all network namespace init functions are
1308 * called for every existing network namespace. Allowing kernel
1309 * modules to have a race free view of the set of network namespaces.
1310 *
1311 * When a new network namespace is created all of the init
1312 * methods are called in the order in which they were registered.
1313 *
1314 * When a network namespace is destroyed all of the exit methods
1315 * are called in the reverse of the order with which they were
1316 * registered.
1317 */
1318int register_pernet_device(struct pernet_operations *ops)
1319{
1320 int error;
1321 down_write(&pernet_ops_rwsem);
1322 error = register_pernet_operations(&pernet_list, ops);
1323 if (!error && (first_device == &pernet_list))
1324 first_device = &ops->list;
1325 up_write(&pernet_ops_rwsem);
1326 return error;
1327}
1328EXPORT_SYMBOL_GPL(register_pernet_device);
1329
1330/**
1331 * unregister_pernet_device - unregister a network namespace netdevice
1332 * @ops: pernet operations structure to manipulate
1333 *
1334 * Remove the pernet operations structure from the list to be
1335 * used when network namespaces are created or destroyed. In
1336 * addition run the exit method for all existing network
1337 * namespaces.
1338 */
1339void unregister_pernet_device(struct pernet_operations *ops)
1340{
1341 down_write(&pernet_ops_rwsem);
1342 if (&ops->list == first_device)
1343 first_device = first_device->next;
1344 unregister_pernet_operations(ops);
1345 up_write(&pernet_ops_rwsem);
1346}
1347EXPORT_SYMBOL_GPL(unregister_pernet_device);
1348
1349#ifdef CONFIG_NET_NS
1350static struct ns_common *netns_get(struct task_struct *task)
1351{
1352 struct net *net = NULL;
1353 struct nsproxy *nsproxy;
1354
1355 task_lock(task);
1356 nsproxy = task->nsproxy;
1357 if (nsproxy)
1358 net = get_net(nsproxy->net_ns);
1359 task_unlock(task);
1360
1361 return net ? &net->ns : NULL;
1362}
1363
1364static inline struct net *to_net_ns(struct ns_common *ns)
1365{
1366 return container_of(ns, struct net, ns);
1367}
1368
1369static void netns_put(struct ns_common *ns)
1370{
1371 put_net(to_net_ns(ns));
1372}
1373
1374static int netns_install(struct nsset *nsset, struct ns_common *ns)
1375{
1376 struct nsproxy *nsproxy = nsset->nsproxy;
1377 struct net *net = to_net_ns(ns);
1378
1379 if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1380 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
1381 return -EPERM;
1382
1383 put_net(nsproxy->net_ns);
1384 nsproxy->net_ns = get_net(net);
1385 return 0;
1386}
1387
1388static struct user_namespace *netns_owner(struct ns_common *ns)
1389{
1390 return to_net_ns(ns)->user_ns;
1391}
1392
1393const struct proc_ns_operations netns_operations = {
1394 .name = "net",
1395 .type = CLONE_NEWNET,
1396 .get = netns_get,
1397 .put = netns_put,
1398 .install = netns_install,
1399 .owner = netns_owner,
1400};
1401#endif