xref: /openbmc/linux/net/core/net_namespace.c (revision 0a73d21e)
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 
30 static LIST_HEAD(pernet_list);
31 static struct list_head *first_device = &pernet_list;
32 DEFINE_MUTEX(net_mutex);
33 
34 LIST_HEAD(net_namespace_list);
35 EXPORT_SYMBOL_GPL(net_namespace_list);
36 
37 struct net init_net = {
38 	.count		= REFCOUNT_INIT(1),
39 	.dev_base_head	= LIST_HEAD_INIT(init_net.dev_base_head),
40 };
41 EXPORT_SYMBOL(init_net);
42 
43 static bool init_net_initialized;
44 
45 #define MIN_PERNET_OPS_ID	\
46 	((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *))
47 
48 #define INITIAL_NET_GEN_PTRS	13 /* +1 for len +2 for rcu_head */
49 
50 static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
51 
52 static struct net_generic *net_alloc_generic(void)
53 {
54 	struct net_generic *ng;
55 	unsigned int generic_size = offsetof(struct net_generic, ptr[max_gen_ptrs]);
56 
57 	ng = kzalloc(generic_size, GFP_KERNEL);
58 	if (ng)
59 		ng->s.len = max_gen_ptrs;
60 
61 	return ng;
62 }
63 
64 static int net_assign_generic(struct net *net, unsigned int id, void *data)
65 {
66 	struct net_generic *ng, *old_ng;
67 
68 	BUG_ON(!mutex_is_locked(&net_mutex));
69 	BUG_ON(id < MIN_PERNET_OPS_ID);
70 
71 	old_ng = rcu_dereference_protected(net->gen,
72 					   lockdep_is_held(&net_mutex));
73 	if (old_ng->s.len > id) {
74 		old_ng->ptr[id] = data;
75 		return 0;
76 	}
77 
78 	ng = net_alloc_generic();
79 	if (ng == NULL)
80 		return -ENOMEM;
81 
82 	/*
83 	 * Some synchronisation notes:
84 	 *
85 	 * The net_generic explores the net->gen array inside rcu
86 	 * read section. Besides once set the net->gen->ptr[x]
87 	 * pointer never changes (see rules in netns/generic.h).
88 	 *
89 	 * That said, we simply duplicate this array and schedule
90 	 * the old copy for kfree after a grace period.
91 	 */
92 
93 	memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
94 	       (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
95 	ng->ptr[id] = data;
96 
97 	rcu_assign_pointer(net->gen, ng);
98 	kfree_rcu(old_ng, s.rcu);
99 	return 0;
100 }
101 
102 static int ops_init(const struct pernet_operations *ops, struct net *net)
103 {
104 	int err = -ENOMEM;
105 	void *data = NULL;
106 
107 	if (ops->id && ops->size) {
108 		data = kzalloc(ops->size, GFP_KERNEL);
109 		if (!data)
110 			goto out;
111 
112 		err = net_assign_generic(net, *ops->id, data);
113 		if (err)
114 			goto cleanup;
115 	}
116 	err = 0;
117 	if (ops->init)
118 		err = ops->init(net);
119 	if (!err)
120 		return 0;
121 
122 cleanup:
123 	kfree(data);
124 
125 out:
126 	return err;
127 }
128 
129 static void ops_free(const struct pernet_operations *ops, struct net *net)
130 {
131 	if (ops->id && ops->size) {
132 		kfree(net_generic(net, *ops->id));
133 	}
134 }
135 
136 static void ops_exit_list(const struct pernet_operations *ops,
137 			  struct list_head *net_exit_list)
138 {
139 	struct net *net;
140 	if (ops->exit) {
141 		list_for_each_entry(net, net_exit_list, exit_list)
142 			ops->exit(net);
143 	}
144 	if (ops->exit_batch)
145 		ops->exit_batch(net_exit_list);
146 }
147 
148 static void ops_free_list(const struct pernet_operations *ops,
149 			  struct list_head *net_exit_list)
150 {
151 	struct net *net;
152 	if (ops->size && ops->id) {
153 		list_for_each_entry(net, net_exit_list, exit_list)
154 			ops_free(ops, net);
155 	}
156 }
157 
158 /* should be called with nsid_lock held */
159 static int alloc_netid(struct net *net, struct net *peer, int reqid)
160 {
161 	int min = 0, max = 0;
162 
163 	if (reqid >= 0) {
164 		min = reqid;
165 		max = reqid + 1;
166 	}
167 
168 	return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
169 }
170 
171 /* This function is used by idr_for_each(). If net is equal to peer, the
172  * function returns the id so that idr_for_each() stops. Because we cannot
173  * returns the id 0 (idr_for_each() will not stop), we return the magic value
174  * NET_ID_ZERO (-1) for it.
175  */
176 #define NET_ID_ZERO -1
177 static int net_eq_idr(int id, void *net, void *peer)
178 {
179 	if (net_eq(net, peer))
180 		return id ? : NET_ID_ZERO;
181 	return 0;
182 }
183 
184 /* Should be called with nsid_lock held. If a new id is assigned, the bool alloc
185  * is set to true, thus the caller knows that the new id must be notified via
186  * rtnl.
187  */
188 static int __peernet2id_alloc(struct net *net, struct net *peer, bool *alloc)
189 {
190 	int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
191 	bool alloc_it = *alloc;
192 
193 	*alloc = false;
194 
195 	/* Magic value for id 0. */
196 	if (id == NET_ID_ZERO)
197 		return 0;
198 	if (id > 0)
199 		return id;
200 
201 	if (alloc_it) {
202 		id = alloc_netid(net, peer, -1);
203 		*alloc = true;
204 		return id >= 0 ? id : NETNSA_NSID_NOT_ASSIGNED;
205 	}
206 
207 	return NETNSA_NSID_NOT_ASSIGNED;
208 }
209 
210 /* should be called with nsid_lock held */
211 static int __peernet2id(struct net *net, struct net *peer)
212 {
213 	bool no = false;
214 
215 	return __peernet2id_alloc(net, peer, &no);
216 }
217 
218 static void rtnl_net_notifyid(struct net *net, int cmd, int id);
219 /* This function returns the id of a peer netns. If no id is assigned, one will
220  * be allocated and returned.
221  */
222 int peernet2id_alloc(struct net *net, struct net *peer)
223 {
224 	bool alloc = false, alive = false;
225 	int id;
226 
227 	if (refcount_read(&net->count) == 0)
228 		return NETNSA_NSID_NOT_ASSIGNED;
229 	spin_lock_bh(&net->nsid_lock);
230 	/*
231 	 * When peer is obtained from RCU lists, we may race with
232 	 * its cleanup. Check whether it's alive, and this guarantees
233 	 * we never hash a peer back to net->netns_ids, after it has
234 	 * just been idr_remove()'d from there in cleanup_net().
235 	 */
236 	if (maybe_get_net(peer))
237 		alive = alloc = true;
238 	id = __peernet2id_alloc(net, peer, &alloc);
239 	spin_unlock_bh(&net->nsid_lock);
240 	if (alloc && id >= 0)
241 		rtnl_net_notifyid(net, RTM_NEWNSID, id);
242 	if (alive)
243 		put_net(peer);
244 	return id;
245 }
246 EXPORT_SYMBOL_GPL(peernet2id_alloc);
247 
248 /* This function returns, if assigned, the id of a peer netns. */
249 int peernet2id(struct net *net, struct net *peer)
250 {
251 	int id;
252 
253 	spin_lock_bh(&net->nsid_lock);
254 	id = __peernet2id(net, peer);
255 	spin_unlock_bh(&net->nsid_lock);
256 	return id;
257 }
258 EXPORT_SYMBOL(peernet2id);
259 
260 /* This function returns true is the peer netns has an id assigned into the
261  * current netns.
262  */
263 bool peernet_has_id(struct net *net, struct net *peer)
264 {
265 	return peernet2id(net, peer) >= 0;
266 }
267 
268 struct net *get_net_ns_by_id(struct net *net, int id)
269 {
270 	struct net *peer;
271 
272 	if (id < 0)
273 		return NULL;
274 
275 	rcu_read_lock();
276 	peer = idr_find(&net->netns_ids, id);
277 	if (peer)
278 		peer = maybe_get_net(peer);
279 	rcu_read_unlock();
280 
281 	return peer;
282 }
283 
284 /*
285  * setup_net runs the initializers for the network namespace object.
286  */
287 static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
288 {
289 	/* Must be called with net_mutex held */
290 	const struct pernet_operations *ops, *saved_ops;
291 	int error = 0;
292 	LIST_HEAD(net_exit_list);
293 
294 	refcount_set(&net->count, 1);
295 	refcount_set(&net->passive, 1);
296 	net->dev_base_seq = 1;
297 	net->user_ns = user_ns;
298 	idr_init(&net->netns_ids);
299 	spin_lock_init(&net->nsid_lock);
300 
301 	list_for_each_entry(ops, &pernet_list, list) {
302 		error = ops_init(ops, net);
303 		if (error < 0)
304 			goto out_undo;
305 	}
306 out:
307 	return error;
308 
309 out_undo:
310 	/* Walk through the list backwards calling the exit functions
311 	 * for the pernet modules whose init functions did not fail.
312 	 */
313 	list_add(&net->exit_list, &net_exit_list);
314 	saved_ops = ops;
315 	list_for_each_entry_continue_reverse(ops, &pernet_list, list)
316 		ops_exit_list(ops, &net_exit_list);
317 
318 	ops = saved_ops;
319 	list_for_each_entry_continue_reverse(ops, &pernet_list, list)
320 		ops_free_list(ops, &net_exit_list);
321 
322 	rcu_barrier();
323 	goto out;
324 }
325 
326 static int __net_init net_defaults_init_net(struct net *net)
327 {
328 	net->core.sysctl_somaxconn = SOMAXCONN;
329 	return 0;
330 }
331 
332 static struct pernet_operations net_defaults_ops = {
333 	.init = net_defaults_init_net,
334 };
335 
336 static __init int net_defaults_init(void)
337 {
338 	if (register_pernet_subsys(&net_defaults_ops))
339 		panic("Cannot initialize net default settings");
340 
341 	return 0;
342 }
343 
344 core_initcall(net_defaults_init);
345 
346 #ifdef CONFIG_NET_NS
347 static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
348 {
349 	return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
350 }
351 
352 static void dec_net_namespaces(struct ucounts *ucounts)
353 {
354 	dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
355 }
356 
357 static struct kmem_cache *net_cachep;
358 static struct workqueue_struct *netns_wq;
359 
360 static struct net *net_alloc(void)
361 {
362 	struct net *net = NULL;
363 	struct net_generic *ng;
364 
365 	ng = net_alloc_generic();
366 	if (!ng)
367 		goto out;
368 
369 	net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
370 	if (!net)
371 		goto out_free;
372 
373 	rcu_assign_pointer(net->gen, ng);
374 out:
375 	return net;
376 
377 out_free:
378 	kfree(ng);
379 	goto out;
380 }
381 
382 static void net_free(struct net *net)
383 {
384 	kfree(rcu_access_pointer(net->gen));
385 	kmem_cache_free(net_cachep, net);
386 }
387 
388 void net_drop_ns(void *p)
389 {
390 	struct net *ns = p;
391 	if (ns && refcount_dec_and_test(&ns->passive))
392 		net_free(ns);
393 }
394 
395 struct net *copy_net_ns(unsigned long flags,
396 			struct user_namespace *user_ns, struct net *old_net)
397 {
398 	struct ucounts *ucounts;
399 	struct net *net;
400 	int rv;
401 
402 	if (!(flags & CLONE_NEWNET))
403 		return get_net(old_net);
404 
405 	ucounts = inc_net_namespaces(user_ns);
406 	if (!ucounts)
407 		return ERR_PTR(-ENOSPC);
408 
409 	net = net_alloc();
410 	if (!net) {
411 		dec_net_namespaces(ucounts);
412 		return ERR_PTR(-ENOMEM);
413 	}
414 
415 	get_user_ns(user_ns);
416 
417 	rv = mutex_lock_killable(&net_mutex);
418 	if (rv < 0) {
419 		net_free(net);
420 		dec_net_namespaces(ucounts);
421 		put_user_ns(user_ns);
422 		return ERR_PTR(rv);
423 	}
424 
425 	net->ucounts = ucounts;
426 	rv = setup_net(net, user_ns);
427 	if (rv == 0) {
428 		rtnl_lock();
429 		list_add_tail_rcu(&net->list, &net_namespace_list);
430 		rtnl_unlock();
431 	}
432 	mutex_unlock(&net_mutex);
433 	if (rv < 0) {
434 		dec_net_namespaces(ucounts);
435 		put_user_ns(user_ns);
436 		net_drop_ns(net);
437 		return ERR_PTR(rv);
438 	}
439 	return net;
440 }
441 
442 static void unhash_nsid(struct net *net, struct net *last)
443 {
444 	struct net *tmp;
445 	/* This function is only called from cleanup_net() work,
446 	 * and this work is the only process, that may delete
447 	 * a net from net_namespace_list. So, when the below
448 	 * is executing, the list may only grow. Thus, we do not
449 	 * use for_each_net_rcu() or rtnl_lock().
450 	 */
451 	for_each_net(tmp) {
452 		int id;
453 
454 		spin_lock_bh(&tmp->nsid_lock);
455 		id = __peernet2id(tmp, net);
456 		if (id >= 0)
457 			idr_remove(&tmp->netns_ids, id);
458 		spin_unlock_bh(&tmp->nsid_lock);
459 		if (id >= 0)
460 			rtnl_net_notifyid(tmp, RTM_DELNSID, id);
461 		if (tmp == last)
462 			break;
463 	}
464 	spin_lock_bh(&net->nsid_lock);
465 	idr_destroy(&net->netns_ids);
466 	spin_unlock_bh(&net->nsid_lock);
467 }
468 
469 static DEFINE_SPINLOCK(cleanup_list_lock);
470 static LIST_HEAD(cleanup_list);  /* Must hold cleanup_list_lock to touch */
471 
472 static void cleanup_net(struct work_struct *work)
473 {
474 	const struct pernet_operations *ops;
475 	struct net *net, *tmp, *last;
476 	struct list_head net_kill_list;
477 	LIST_HEAD(net_exit_list);
478 
479 	/* Atomically snapshot the list of namespaces to cleanup */
480 	spin_lock_irq(&cleanup_list_lock);
481 	list_replace_init(&cleanup_list, &net_kill_list);
482 	spin_unlock_irq(&cleanup_list_lock);
483 
484 	mutex_lock(&net_mutex);
485 
486 	/* Don't let anyone else find us. */
487 	rtnl_lock();
488 	list_for_each_entry(net, &net_kill_list, cleanup_list)
489 		list_del_rcu(&net->list);
490 	/* Cache last net. After we unlock rtnl, no one new net
491 	 * added to net_namespace_list can assign nsid pointer
492 	 * to a net from net_kill_list (see peernet2id_alloc()).
493 	 * So, we skip them in unhash_nsid().
494 	 *
495 	 * Note, that unhash_nsid() does not delete nsid links
496 	 * between net_kill_list's nets, as they've already
497 	 * deleted from net_namespace_list. But, this would be
498 	 * useless anyway, as netns_ids are destroyed there.
499 	 */
500 	last = list_last_entry(&net_namespace_list, struct net, list);
501 	rtnl_unlock();
502 
503 	list_for_each_entry(net, &net_kill_list, cleanup_list) {
504 		unhash_nsid(net, last);
505 		list_add_tail(&net->exit_list, &net_exit_list);
506 	}
507 
508 	/*
509 	 * Another CPU might be rcu-iterating the list, wait for it.
510 	 * This needs to be before calling the exit() notifiers, so
511 	 * the rcu_barrier() below isn't sufficient alone.
512 	 */
513 	synchronize_rcu();
514 
515 	/* Run all of the network namespace exit methods */
516 	list_for_each_entry_reverse(ops, &pernet_list, list)
517 		ops_exit_list(ops, &net_exit_list);
518 
519 	/* Free the net generic variables */
520 	list_for_each_entry_reverse(ops, &pernet_list, list)
521 		ops_free_list(ops, &net_exit_list);
522 
523 	mutex_unlock(&net_mutex);
524 
525 	/* Ensure there are no outstanding rcu callbacks using this
526 	 * network namespace.
527 	 */
528 	rcu_barrier();
529 
530 	/* Finally it is safe to free my network namespace structure */
531 	list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
532 		list_del_init(&net->exit_list);
533 		dec_net_namespaces(net->ucounts);
534 		put_user_ns(net->user_ns);
535 		net_drop_ns(net);
536 	}
537 }
538 
539 /**
540  * net_ns_barrier - wait until concurrent net_cleanup_work is done
541  *
542  * cleanup_net runs from work queue and will first remove namespaces
543  * from the global list, then run net exit functions.
544  *
545  * Call this in module exit path to make sure that all netns
546  * ->exit ops have been invoked before the function is removed.
547  */
548 void net_ns_barrier(void)
549 {
550 	mutex_lock(&net_mutex);
551 	mutex_unlock(&net_mutex);
552 }
553 EXPORT_SYMBOL(net_ns_barrier);
554 
555 static DECLARE_WORK(net_cleanup_work, cleanup_net);
556 
557 void __put_net(struct net *net)
558 {
559 	/* Cleanup the network namespace in process context */
560 	unsigned long flags;
561 
562 	spin_lock_irqsave(&cleanup_list_lock, flags);
563 	list_add(&net->cleanup_list, &cleanup_list);
564 	spin_unlock_irqrestore(&cleanup_list_lock, flags);
565 
566 	queue_work(netns_wq, &net_cleanup_work);
567 }
568 EXPORT_SYMBOL_GPL(__put_net);
569 
570 struct net *get_net_ns_by_fd(int fd)
571 {
572 	struct file *file;
573 	struct ns_common *ns;
574 	struct net *net;
575 
576 	file = proc_ns_fget(fd);
577 	if (IS_ERR(file))
578 		return ERR_CAST(file);
579 
580 	ns = get_proc_ns(file_inode(file));
581 	if (ns->ops == &netns_operations)
582 		net = get_net(container_of(ns, struct net, ns));
583 	else
584 		net = ERR_PTR(-EINVAL);
585 
586 	fput(file);
587 	return net;
588 }
589 
590 #else
591 struct net *get_net_ns_by_fd(int fd)
592 {
593 	return ERR_PTR(-EINVAL);
594 }
595 #endif
596 EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
597 
598 struct net *get_net_ns_by_pid(pid_t pid)
599 {
600 	struct task_struct *tsk;
601 	struct net *net;
602 
603 	/* Lookup the network namespace */
604 	net = ERR_PTR(-ESRCH);
605 	rcu_read_lock();
606 	tsk = find_task_by_vpid(pid);
607 	if (tsk) {
608 		struct nsproxy *nsproxy;
609 		task_lock(tsk);
610 		nsproxy = tsk->nsproxy;
611 		if (nsproxy)
612 			net = get_net(nsproxy->net_ns);
613 		task_unlock(tsk);
614 	}
615 	rcu_read_unlock();
616 	return net;
617 }
618 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
619 
620 static __net_init int net_ns_net_init(struct net *net)
621 {
622 #ifdef CONFIG_NET_NS
623 	net->ns.ops = &netns_operations;
624 #endif
625 	return ns_alloc_inum(&net->ns);
626 }
627 
628 static __net_exit void net_ns_net_exit(struct net *net)
629 {
630 	ns_free_inum(&net->ns);
631 }
632 
633 static struct pernet_operations __net_initdata net_ns_ops = {
634 	.init = net_ns_net_init,
635 	.exit = net_ns_net_exit,
636 };
637 
638 static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
639 	[NETNSA_NONE]		= { .type = NLA_UNSPEC },
640 	[NETNSA_NSID]		= { .type = NLA_S32 },
641 	[NETNSA_PID]		= { .type = NLA_U32 },
642 	[NETNSA_FD]		= { .type = NLA_U32 },
643 };
644 
645 static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
646 			  struct netlink_ext_ack *extack)
647 {
648 	struct net *net = sock_net(skb->sk);
649 	struct nlattr *tb[NETNSA_MAX + 1];
650 	struct nlattr *nla;
651 	struct net *peer;
652 	int nsid, err;
653 
654 	err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
655 			  rtnl_net_policy, extack);
656 	if (err < 0)
657 		return err;
658 	if (!tb[NETNSA_NSID]) {
659 		NL_SET_ERR_MSG(extack, "nsid is missing");
660 		return -EINVAL;
661 	}
662 	nsid = nla_get_s32(tb[NETNSA_NSID]);
663 
664 	if (tb[NETNSA_PID]) {
665 		peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
666 		nla = tb[NETNSA_PID];
667 	} else if (tb[NETNSA_FD]) {
668 		peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
669 		nla = tb[NETNSA_FD];
670 	} else {
671 		NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
672 		return -EINVAL;
673 	}
674 	if (IS_ERR(peer)) {
675 		NL_SET_BAD_ATTR(extack, nla);
676 		NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
677 		return PTR_ERR(peer);
678 	}
679 
680 	spin_lock_bh(&net->nsid_lock);
681 	if (__peernet2id(net, peer) >= 0) {
682 		spin_unlock_bh(&net->nsid_lock);
683 		err = -EEXIST;
684 		NL_SET_BAD_ATTR(extack, nla);
685 		NL_SET_ERR_MSG(extack,
686 			       "Peer netns already has a nsid assigned");
687 		goto out;
688 	}
689 
690 	err = alloc_netid(net, peer, nsid);
691 	spin_unlock_bh(&net->nsid_lock);
692 	if (err >= 0) {
693 		rtnl_net_notifyid(net, RTM_NEWNSID, err);
694 		err = 0;
695 	} else if (err == -ENOSPC && nsid >= 0) {
696 		err = -EEXIST;
697 		NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
698 		NL_SET_ERR_MSG(extack, "The specified nsid is already used");
699 	}
700 out:
701 	put_net(peer);
702 	return err;
703 }
704 
705 static int rtnl_net_get_size(void)
706 {
707 	return NLMSG_ALIGN(sizeof(struct rtgenmsg))
708 	       + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
709 	       ;
710 }
711 
712 static int rtnl_net_fill(struct sk_buff *skb, u32 portid, u32 seq, int flags,
713 			 int cmd, struct net *net, int nsid)
714 {
715 	struct nlmsghdr *nlh;
716 	struct rtgenmsg *rth;
717 
718 	nlh = nlmsg_put(skb, portid, seq, cmd, sizeof(*rth), flags);
719 	if (!nlh)
720 		return -EMSGSIZE;
721 
722 	rth = nlmsg_data(nlh);
723 	rth->rtgen_family = AF_UNSPEC;
724 
725 	if (nla_put_s32(skb, NETNSA_NSID, nsid))
726 		goto nla_put_failure;
727 
728 	nlmsg_end(skb, nlh);
729 	return 0;
730 
731 nla_put_failure:
732 	nlmsg_cancel(skb, nlh);
733 	return -EMSGSIZE;
734 }
735 
736 static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
737 			  struct netlink_ext_ack *extack)
738 {
739 	struct net *net = sock_net(skb->sk);
740 	struct nlattr *tb[NETNSA_MAX + 1];
741 	struct nlattr *nla;
742 	struct sk_buff *msg;
743 	struct net *peer;
744 	int err, id;
745 
746 	err = nlmsg_parse(nlh, sizeof(struct rtgenmsg), tb, NETNSA_MAX,
747 			  rtnl_net_policy, extack);
748 	if (err < 0)
749 		return err;
750 	if (tb[NETNSA_PID]) {
751 		peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
752 		nla = tb[NETNSA_PID];
753 	} else if (tb[NETNSA_FD]) {
754 		peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
755 		nla = tb[NETNSA_FD];
756 	} else {
757 		NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
758 		return -EINVAL;
759 	}
760 
761 	if (IS_ERR(peer)) {
762 		NL_SET_BAD_ATTR(extack, nla);
763 		NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
764 		return PTR_ERR(peer);
765 	}
766 
767 	msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
768 	if (!msg) {
769 		err = -ENOMEM;
770 		goto out;
771 	}
772 
773 	id = peernet2id(net, peer);
774 	err = rtnl_net_fill(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq, 0,
775 			    RTM_NEWNSID, net, id);
776 	if (err < 0)
777 		goto err_out;
778 
779 	err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
780 	goto out;
781 
782 err_out:
783 	nlmsg_free(msg);
784 out:
785 	put_net(peer);
786 	return err;
787 }
788 
789 struct rtnl_net_dump_cb {
790 	struct net *net;
791 	struct sk_buff *skb;
792 	struct netlink_callback *cb;
793 	int idx;
794 	int s_idx;
795 };
796 
797 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
798 {
799 	struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
800 	int ret;
801 
802 	if (net_cb->idx < net_cb->s_idx)
803 		goto cont;
804 
805 	ret = rtnl_net_fill(net_cb->skb, NETLINK_CB(net_cb->cb->skb).portid,
806 			    net_cb->cb->nlh->nlmsg_seq, NLM_F_MULTI,
807 			    RTM_NEWNSID, net_cb->net, id);
808 	if (ret < 0)
809 		return ret;
810 
811 cont:
812 	net_cb->idx++;
813 	return 0;
814 }
815 
816 static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
817 {
818 	struct net *net = sock_net(skb->sk);
819 	struct rtnl_net_dump_cb net_cb = {
820 		.net = net,
821 		.skb = skb,
822 		.cb = cb,
823 		.idx = 0,
824 		.s_idx = cb->args[0],
825 	};
826 
827 	spin_lock_bh(&net->nsid_lock);
828 	idr_for_each(&net->netns_ids, rtnl_net_dumpid_one, &net_cb);
829 	spin_unlock_bh(&net->nsid_lock);
830 
831 	cb->args[0] = net_cb.idx;
832 	return skb->len;
833 }
834 
835 static void rtnl_net_notifyid(struct net *net, int cmd, int id)
836 {
837 	struct sk_buff *msg;
838 	int err = -ENOMEM;
839 
840 	msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
841 	if (!msg)
842 		goto out;
843 
844 	err = rtnl_net_fill(msg, 0, 0, 0, cmd, net, id);
845 	if (err < 0)
846 		goto err_out;
847 
848 	rtnl_notify(msg, net, 0, RTNLGRP_NSID, NULL, 0);
849 	return;
850 
851 err_out:
852 	nlmsg_free(msg);
853 out:
854 	rtnl_set_sk_err(net, RTNLGRP_NSID, err);
855 }
856 
857 static int __init net_ns_init(void)
858 {
859 	struct net_generic *ng;
860 
861 #ifdef CONFIG_NET_NS
862 	net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
863 					SMP_CACHE_BYTES,
864 					SLAB_PANIC, NULL);
865 
866 	/* Create workqueue for cleanup */
867 	netns_wq = create_singlethread_workqueue("netns");
868 	if (!netns_wq)
869 		panic("Could not create netns workq");
870 #endif
871 
872 	ng = net_alloc_generic();
873 	if (!ng)
874 		panic("Could not allocate generic netns");
875 
876 	rcu_assign_pointer(init_net.gen, ng);
877 
878 	mutex_lock(&net_mutex);
879 	if (setup_net(&init_net, &init_user_ns))
880 		panic("Could not setup the initial network namespace");
881 
882 	init_net_initialized = true;
883 
884 	rtnl_lock();
885 	list_add_tail_rcu(&init_net.list, &net_namespace_list);
886 	rtnl_unlock();
887 
888 	mutex_unlock(&net_mutex);
889 
890 	register_pernet_subsys(&net_ns_ops);
891 
892 	rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
893 		      RTNL_FLAG_DOIT_UNLOCKED);
894 	rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
895 		      RTNL_FLAG_DOIT_UNLOCKED);
896 
897 	return 0;
898 }
899 
900 pure_initcall(net_ns_init);
901 
902 #ifdef CONFIG_NET_NS
903 static int __register_pernet_operations(struct list_head *list,
904 					struct pernet_operations *ops)
905 {
906 	struct net *net;
907 	int error;
908 	LIST_HEAD(net_exit_list);
909 
910 	list_add_tail(&ops->list, list);
911 	if (ops->init || (ops->id && ops->size)) {
912 		for_each_net(net) {
913 			error = ops_init(ops, net);
914 			if (error)
915 				goto out_undo;
916 			list_add_tail(&net->exit_list, &net_exit_list);
917 		}
918 	}
919 	return 0;
920 
921 out_undo:
922 	/* If I have an error cleanup all namespaces I initialized */
923 	list_del(&ops->list);
924 	ops_exit_list(ops, &net_exit_list);
925 	ops_free_list(ops, &net_exit_list);
926 	return error;
927 }
928 
929 static void __unregister_pernet_operations(struct pernet_operations *ops)
930 {
931 	struct net *net;
932 	LIST_HEAD(net_exit_list);
933 
934 	list_del(&ops->list);
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 
943 static 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 
954 static 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 
968 static DEFINE_IDA(net_generic_ids);
969 
970 static int register_pernet_operations(struct list_head *list,
971 				      struct pernet_operations *ops)
972 {
973 	int error;
974 
975 	if (ops->id) {
976 again:
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 
997 static void unregister_pernet_operations(struct pernet_operations *ops)
998 {
999 
1000 	__unregister_pernet_operations(ops);
1001 	rcu_barrier();
1002 	if (ops->id)
1003 		ida_remove(&net_generic_ids, *ops->id);
1004 }
1005 
1006 /**
1007  *      register_pernet_subsys - register a network namespace subsystem
1008  *	@ops:  pernet operations structure for the subsystem
1009  *
1010  *	Register a subsystem which has init and exit functions
1011  *	that are called when network namespaces are created and
1012  *	destroyed respectively.
1013  *
1014  *	When registered all network namespace init functions are
1015  *	called for every existing network namespace.  Allowing kernel
1016  *	modules to have a race free view of the set of network namespaces.
1017  *
1018  *	When a new network namespace is created all of the init
1019  *	methods are called in the order in which they were registered.
1020  *
1021  *	When a network namespace is destroyed all of the exit methods
1022  *	are called in the reverse of the order with which they were
1023  *	registered.
1024  */
1025 int register_pernet_subsys(struct pernet_operations *ops)
1026 {
1027 	int error;
1028 	mutex_lock(&net_mutex);
1029 	error =  register_pernet_operations(first_device, ops);
1030 	mutex_unlock(&net_mutex);
1031 	return error;
1032 }
1033 EXPORT_SYMBOL_GPL(register_pernet_subsys);
1034 
1035 /**
1036  *      unregister_pernet_subsys - unregister a network namespace subsystem
1037  *	@ops: pernet operations structure to manipulate
1038  *
1039  *	Remove the pernet operations structure from the list to be
1040  *	used when network namespaces are created or destroyed.  In
1041  *	addition run the exit method for all existing network
1042  *	namespaces.
1043  */
1044 void unregister_pernet_subsys(struct pernet_operations *ops)
1045 {
1046 	mutex_lock(&net_mutex);
1047 	unregister_pernet_operations(ops);
1048 	mutex_unlock(&net_mutex);
1049 }
1050 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1051 
1052 /**
1053  *      register_pernet_device - register a network namespace device
1054  *	@ops:  pernet operations structure for the subsystem
1055  *
1056  *	Register a device which has init and exit functions
1057  *	that are called when network namespaces are created and
1058  *	destroyed respectively.
1059  *
1060  *	When registered all network namespace init functions are
1061  *	called for every existing network namespace.  Allowing kernel
1062  *	modules to have a race free view of the set of network namespaces.
1063  *
1064  *	When a new network namespace is created all of the init
1065  *	methods are called in the order in which they were registered.
1066  *
1067  *	When a network namespace is destroyed all of the exit methods
1068  *	are called in the reverse of the order with which they were
1069  *	registered.
1070  */
1071 int register_pernet_device(struct pernet_operations *ops)
1072 {
1073 	int error;
1074 	mutex_lock(&net_mutex);
1075 	error = register_pernet_operations(&pernet_list, ops);
1076 	if (!error && (first_device == &pernet_list))
1077 		first_device = &ops->list;
1078 	mutex_unlock(&net_mutex);
1079 	return error;
1080 }
1081 EXPORT_SYMBOL_GPL(register_pernet_device);
1082 
1083 /**
1084  *      unregister_pernet_device - unregister a network namespace netdevice
1085  *	@ops: pernet operations structure to manipulate
1086  *
1087  *	Remove the pernet operations structure from the list to be
1088  *	used when network namespaces are created or destroyed.  In
1089  *	addition run the exit method for all existing network
1090  *	namespaces.
1091  */
1092 void unregister_pernet_device(struct pernet_operations *ops)
1093 {
1094 	mutex_lock(&net_mutex);
1095 	if (&ops->list == first_device)
1096 		first_device = first_device->next;
1097 	unregister_pernet_operations(ops);
1098 	mutex_unlock(&net_mutex);
1099 }
1100 EXPORT_SYMBOL_GPL(unregister_pernet_device);
1101 
1102 #ifdef CONFIG_NET_NS
1103 static struct ns_common *netns_get(struct task_struct *task)
1104 {
1105 	struct net *net = NULL;
1106 	struct nsproxy *nsproxy;
1107 
1108 	task_lock(task);
1109 	nsproxy = task->nsproxy;
1110 	if (nsproxy)
1111 		net = get_net(nsproxy->net_ns);
1112 	task_unlock(task);
1113 
1114 	return net ? &net->ns : NULL;
1115 }
1116 
1117 static inline struct net *to_net_ns(struct ns_common *ns)
1118 {
1119 	return container_of(ns, struct net, ns);
1120 }
1121 
1122 static void netns_put(struct ns_common *ns)
1123 {
1124 	put_net(to_net_ns(ns));
1125 }
1126 
1127 static int netns_install(struct nsproxy *nsproxy, struct ns_common *ns)
1128 {
1129 	struct net *net = to_net_ns(ns);
1130 
1131 	if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1132 	    !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1133 		return -EPERM;
1134 
1135 	put_net(nsproxy->net_ns);
1136 	nsproxy->net_ns = get_net(net);
1137 	return 0;
1138 }
1139 
1140 static struct user_namespace *netns_owner(struct ns_common *ns)
1141 {
1142 	return to_net_ns(ns)->user_ns;
1143 }
1144 
1145 const struct proc_ns_operations netns_operations = {
1146 	.name		= "net",
1147 	.type		= CLONE_NEWNET,
1148 	.get		= netns_get,
1149 	.put		= netns_put,
1150 	.install	= netns_install,
1151 	.owner		= netns_owner,
1152 };
1153 #endif
1154