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