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 = {
48	.count		= REFCOUNT_INIT(1),
49	.dev_base_head	= LIST_HEAD_INIT(init_net.dev_base_head),
50#ifdef CONFIG_KEYS
51	.key_domain	= &init_net_key_domain,
52#endif
53};
54EXPORT_SYMBOL(init_net);
55
56static bool init_net_initialized;
57/*
58 * pernet_ops_rwsem: protects: pernet_list, net_generic_ids,
59 * init_net_initialized and first_device pointer.
60 * This is internal net namespace object. Please, don't use it
61 * outside.
62 */
63DECLARE_RWSEM(pernet_ops_rwsem);
64EXPORT_SYMBOL_GPL(pernet_ops_rwsem);
65
66#define MIN_PERNET_OPS_ID	\
67	((sizeof(struct net_generic) + sizeof(void *) - 1) / sizeof(void *))
68
69#define INITIAL_NET_GEN_PTRS	13 /* +1 for len +2 for rcu_head */
70
71static unsigned int max_gen_ptrs = INITIAL_NET_GEN_PTRS;
72
73DEFINE_COOKIE(net_cookie);
74
75u64 __net_gen_cookie(struct net *net)
76{
77	while (1) {
78		u64 res = atomic64_read(&net->net_cookie);
79
80		if (res)
81			return res;
82		res = gen_cookie_next(&net_cookie);
83		atomic64_cmpxchg(&net->net_cookie, 0, res);
84	}
85}
86
87static struct net_generic *net_alloc_generic(void)
88{
89	unsigned int gen_ptrs = READ_ONCE(max_gen_ptrs);
90	unsigned int generic_size;
91	struct net_generic *ng;
92
93	generic_size = offsetof(struct net_generic, ptr[gen_ptrs]);
94
95	ng = kzalloc(generic_size, GFP_KERNEL);
96	if (ng)
97		ng->s.len = gen_ptrs;
98
99	return ng;
100}
101
102static int net_assign_generic(struct net *net, unsigned int id, void *data)
103{
104	struct net_generic *ng, *old_ng;
105
106	BUG_ON(id < MIN_PERNET_OPS_ID);
107
108	old_ng = rcu_dereference_protected(net->gen,
109					   lockdep_is_held(&pernet_ops_rwsem));
110	if (old_ng->s.len > id) {
111		old_ng->ptr[id] = data;
112		return 0;
113	}
114
115	ng = net_alloc_generic();
116	if (ng == NULL)
117		return -ENOMEM;
118
119	/*
120	 * Some synchronisation notes:
121	 *
122	 * The net_generic explores the net->gen array inside rcu
123	 * read section. Besides once set the net->gen->ptr[x]
124	 * pointer never changes (see rules in netns/generic.h).
125	 *
126	 * That said, we simply duplicate this array and schedule
127	 * the old copy for kfree after a grace period.
128	 */
129
130	memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
131	       (old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
132	ng->ptr[id] = data;
133
134	rcu_assign_pointer(net->gen, ng);
135	kfree_rcu(old_ng, s.rcu);
136	return 0;
137}
138
139static int ops_init(const struct pernet_operations *ops, struct net *net)
140{
141	struct net_generic *ng;
142	int err = -ENOMEM;
143	void *data = NULL;
144
145	if (ops->id && ops->size) {
146		data = kzalloc(ops->size, GFP_KERNEL);
147		if (!data)
148			goto out;
149
150		err = net_assign_generic(net, *ops->id, data);
151		if (err)
152			goto cleanup;
153	}
154	err = 0;
155	if (ops->init)
156		err = ops->init(net);
157	if (!err)
158		return 0;
159
160	if (ops->id && ops->size) {
161		ng = rcu_dereference_protected(net->gen,
162					       lockdep_is_held(&pernet_ops_rwsem));
163		ng->ptr[*ops->id] = NULL;
164	}
165
166cleanup:
167	kfree(data);
168
169out:
170	return err;
171}
172
173static void ops_free(const struct pernet_operations *ops, struct net *net)
174{
175	if (ops->id && ops->size) {
176		kfree(net_generic(net, *ops->id));
177	}
178}
179
180static void ops_pre_exit_list(const struct pernet_operations *ops,
181			      struct list_head *net_exit_list)
182{
183	struct net *net;
184
185	if (ops->pre_exit) {
186		list_for_each_entry(net, net_exit_list, exit_list)
187			ops->pre_exit(net);
188	}
189}
190
191static void ops_exit_list(const struct pernet_operations *ops,
192			  struct list_head *net_exit_list)
193{
194	struct net *net;
195	if (ops->exit) {
196		list_for_each_entry(net, net_exit_list, exit_list) {
197			ops->exit(net);
198			cond_resched();
199		}
200	}
201	if (ops->exit_batch)
202		ops->exit_batch(net_exit_list);
203}
204
205static void ops_free_list(const struct pernet_operations *ops,
206			  struct list_head *net_exit_list)
207{
208	struct net *net;
209	if (ops->size && ops->id) {
210		list_for_each_entry(net, net_exit_list, exit_list)
211			ops_free(ops, net);
212	}
213}
214
215/* should be called with nsid_lock held */
216static int alloc_netid(struct net *net, struct net *peer, int reqid)
217{
218	int min = 0, max = 0;
219
220	if (reqid >= 0) {
221		min = reqid;
222		max = reqid + 1;
223	}
224
225	return idr_alloc(&net->netns_ids, peer, min, max, GFP_ATOMIC);
226}
227
228/* This function is used by idr_for_each(). If net is equal to peer, the
229 * function returns the id so that idr_for_each() stops. Because we cannot
230 * returns the id 0 (idr_for_each() will not stop), we return the magic value
231 * NET_ID_ZERO (-1) for it.
232 */
233#define NET_ID_ZERO -1
234static int net_eq_idr(int id, void *net, void *peer)
235{
236	if (net_eq(net, peer))
237		return id ? : NET_ID_ZERO;
238	return 0;
239}
240
241/* Must be called from RCU-critical section or with nsid_lock held */
242static int __peernet2id(const struct net *net, struct net *peer)
243{
244	int id = idr_for_each(&net->netns_ids, net_eq_idr, peer);
245
246	/* Magic value for id 0. */
247	if (id == NET_ID_ZERO)
248		return 0;
249	if (id > 0)
250		return id;
251
252	return NETNSA_NSID_NOT_ASSIGNED;
253}
254
255static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
256			      struct nlmsghdr *nlh, gfp_t gfp);
257/* This function returns the id of a peer netns. If no id is assigned, one will
258 * be allocated and returned.
259 */
260int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp)
261{
262	int id;
263
264	if (refcount_read(&net->count) == 0)
265		return NETNSA_NSID_NOT_ASSIGNED;
266
267	spin_lock_bh(&net->nsid_lock);
268	id = __peernet2id(net, peer);
269	if (id >= 0) {
270		spin_unlock_bh(&net->nsid_lock);
271		return id;
272	}
273
274	/* When peer is obtained from RCU lists, we may race with
275	 * its cleanup. Check whether it's alive, and this guarantees
276	 * we never hash a peer back to net->netns_ids, after it has
277	 * just been idr_remove()'d from there in cleanup_net().
278	 */
279	if (!maybe_get_net(peer)) {
280		spin_unlock_bh(&net->nsid_lock);
281		return NETNSA_NSID_NOT_ASSIGNED;
282	}
283
284	id = alloc_netid(net, peer, -1);
285	spin_unlock_bh(&net->nsid_lock);
286
287	put_net(peer);
288	if (id < 0)
289		return NETNSA_NSID_NOT_ASSIGNED;
290
291	rtnl_net_notifyid(net, RTM_NEWNSID, id, 0, NULL, gfp);
292
293	return id;
294}
295EXPORT_SYMBOL_GPL(peernet2id_alloc);
296
297/* This function returns, if assigned, the id of a peer netns. */
298int peernet2id(const struct net *net, struct net *peer)
299{
300	int id;
301
302	rcu_read_lock();
303	id = __peernet2id(net, peer);
304	rcu_read_unlock();
305
306	return id;
307}
308EXPORT_SYMBOL(peernet2id);
309
310/* This function returns true is the peer netns has an id assigned into the
311 * current netns.
312 */
313bool peernet_has_id(const struct net *net, struct net *peer)
314{
315	return peernet2id(net, peer) >= 0;
316}
317
318struct net *get_net_ns_by_id(const struct net *net, int id)
319{
320	struct net *peer;
321
322	if (id < 0)
323		return NULL;
324
325	rcu_read_lock();
326	peer = idr_find(&net->netns_ids, id);
327	if (peer)
328		peer = maybe_get_net(peer);
329	rcu_read_unlock();
330
331	return peer;
332}
333
334/*
335 * setup_net runs the initializers for the network namespace object.
336 */
337static __net_init int setup_net(struct net *net, struct user_namespace *user_ns)
338{
339	/* Must be called with pernet_ops_rwsem held */
340	const struct pernet_operations *ops, *saved_ops;
341	int error = 0;
342	LIST_HEAD(net_exit_list);
343
344	refcount_set(&net->count, 1);
345	refcount_set(&net->passive, 1);
346	get_random_bytes(&net->hash_mix, sizeof(u32));
347	net->dev_base_seq = 1;
348	net->user_ns = user_ns;
349	idr_init(&net->netns_ids);
350	spin_lock_init(&net->nsid_lock);
351	mutex_init(&net->ipv4.ra_mutex);
352
353	list_for_each_entry(ops, &pernet_list, list) {
354		error = ops_init(ops, net);
355		if (error < 0)
356			goto out_undo;
357	}
358	down_write(&net_rwsem);
359	list_add_tail_rcu(&net->list, &net_namespace_list);
360	up_write(&net_rwsem);
361out:
362	return error;
363
364out_undo:
365	/* Walk through the list backwards calling the exit functions
366	 * for the pernet modules whose init functions did not fail.
367	 */
368	list_add(&net->exit_list, &net_exit_list);
369	saved_ops = ops;
370	list_for_each_entry_continue_reverse(ops, &pernet_list, list)
371		ops_pre_exit_list(ops, &net_exit_list);
372
373	synchronize_rcu();
374
375	ops = saved_ops;
376	list_for_each_entry_continue_reverse(ops, &pernet_list, list)
377		ops_exit_list(ops, &net_exit_list);
378
379	ops = saved_ops;
380	list_for_each_entry_continue_reverse(ops, &pernet_list, list)
381		ops_free_list(ops, &net_exit_list);
382
383	rcu_barrier();
384	goto out;
385}
386
387static int __net_init net_defaults_init_net(struct net *net)
388{
389	net->core.sysctl_somaxconn = SOMAXCONN;
390	return 0;
391}
392
393static struct pernet_operations net_defaults_ops = {
394	.init = net_defaults_init_net,
395};
396
397static __init int net_defaults_init(void)
398{
399	if (register_pernet_subsys(&net_defaults_ops))
400		panic("Cannot initialize net default settings");
401
402	return 0;
403}
404
405core_initcall(net_defaults_init);
406
407#ifdef CONFIG_NET_NS
408static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
409{
410	return inc_ucount(ns, current_euid(), UCOUNT_NET_NAMESPACES);
411}
412
413static void dec_net_namespaces(struct ucounts *ucounts)
414{
415	dec_ucount(ucounts, UCOUNT_NET_NAMESPACES);
416}
417
418static struct kmem_cache *net_cachep __ro_after_init;
419static struct workqueue_struct *netns_wq;
420
421static struct net *net_alloc(void)
422{
423	struct net *net = NULL;
424	struct net_generic *ng;
425
426	ng = net_alloc_generic();
427	if (!ng)
428		goto out;
429
430	net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
431	if (!net)
432		goto out_free;
433
434#ifdef CONFIG_KEYS
435	net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
436	if (!net->key_domain)
437		goto out_free_2;
438	refcount_set(&net->key_domain->usage, 1);
439#endif
440
441	rcu_assign_pointer(net->gen, ng);
442out:
443	return net;
444
445#ifdef CONFIG_KEYS
446out_free_2:
447	kmem_cache_free(net_cachep, net);
448	net = NULL;
449#endif
450out_free:
451	kfree(ng);
452	goto out;
453}
454
455static void net_free(struct net *net)
456{
457	kfree(rcu_access_pointer(net->gen));
458	kmem_cache_free(net_cachep, net);
459}
460
461void net_drop_ns(void *p)
462{
463	struct net *ns = p;
464	if (ns && refcount_dec_and_test(&ns->passive))
465		net_free(ns);
466}
467
468struct net *copy_net_ns(unsigned long flags,
469			struct user_namespace *user_ns, struct net *old_net)
470{
471	struct ucounts *ucounts;
472	struct net *net;
473	int rv;
474
475	if (!(flags & CLONE_NEWNET))
476		return get_net(old_net);
477
478	ucounts = inc_net_namespaces(user_ns);
479	if (!ucounts)
480		return ERR_PTR(-ENOSPC);
481
482	net = net_alloc();
483	if (!net) {
484		rv = -ENOMEM;
485		goto dec_ucounts;
486	}
487	refcount_set(&net->passive, 1);
488	net->ucounts = ucounts;
489	get_user_ns(user_ns);
490
491	rv = down_read_killable(&pernet_ops_rwsem);
492	if (rv < 0)
493		goto put_userns;
494
495	rv = setup_net(net, user_ns);
496
497	up_read(&pernet_ops_rwsem);
498
499	if (rv < 0) {
500put_userns:
501#ifdef CONFIG_KEYS
502		key_remove_domain(net->key_domain);
503#endif
504		put_user_ns(user_ns);
505		net_drop_ns(net);
506dec_ucounts:
507		dec_net_namespaces(ucounts);
508		return ERR_PTR(rv);
509	}
510	return net;
511}
512
513/**
514 * net_ns_get_ownership - get sysfs ownership data for @net
515 * @net: network namespace in question (can be NULL)
516 * @uid: kernel user ID for sysfs objects
517 * @gid: kernel group ID for sysfs objects
518 *
519 * Returns the uid/gid pair of root in the user namespace associated with the
520 * given network namespace.
521 */
522void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
523{
524	if (net) {
525		kuid_t ns_root_uid = make_kuid(net->user_ns, 0);
526		kgid_t ns_root_gid = make_kgid(net->user_ns, 0);
527
528		if (uid_valid(ns_root_uid))
529			*uid = ns_root_uid;
530
531		if (gid_valid(ns_root_gid))
532			*gid = ns_root_gid;
533	} else {
534		*uid = GLOBAL_ROOT_UID;
535		*gid = GLOBAL_ROOT_GID;
536	}
537}
538EXPORT_SYMBOL_GPL(net_ns_get_ownership);
539
540static void unhash_nsid(struct net *net, struct net *last)
541{
542	struct net *tmp;
543	/* This function is only called from cleanup_net() work,
544	 * and this work is the only process, that may delete
545	 * a net from net_namespace_list. So, when the below
546	 * is executing, the list may only grow. Thus, we do not
547	 * use for_each_net_rcu() or net_rwsem.
548	 */
549	for_each_net(tmp) {
550		int id;
551
552		spin_lock_bh(&tmp->nsid_lock);
553		id = __peernet2id(tmp, net);
554		if (id >= 0)
555			idr_remove(&tmp->netns_ids, id);
556		spin_unlock_bh(&tmp->nsid_lock);
557		if (id >= 0)
558			rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL,
559					  GFP_KERNEL);
560		if (tmp == last)
561			break;
562	}
563	spin_lock_bh(&net->nsid_lock);
564	idr_destroy(&net->netns_ids);
565	spin_unlock_bh(&net->nsid_lock);
566}
567
568static LLIST_HEAD(cleanup_list);
569
570static void cleanup_net(struct work_struct *work)
571{
572	const struct pernet_operations *ops;
573	struct net *net, *tmp, *last;
574	struct llist_node *net_kill_list;
575	LIST_HEAD(net_exit_list);
576
577	/* Atomically snapshot the list of namespaces to cleanup */
578	net_kill_list = llist_del_all(&cleanup_list);
579
580	down_read(&pernet_ops_rwsem);
581
582	/* Don't let anyone else find us. */
583	down_write(&net_rwsem);
584	llist_for_each_entry(net, net_kill_list, cleanup_list)
585		list_del_rcu(&net->list);
586	/* Cache last net. After we unlock rtnl, no one new net
587	 * added to net_namespace_list can assign nsid pointer
588	 * to a net from net_kill_list (see peernet2id_alloc()).
589	 * So, we skip them in unhash_nsid().
590	 *
591	 * Note, that unhash_nsid() does not delete nsid links
592	 * between net_kill_list's nets, as they've already
593	 * deleted from net_namespace_list. But, this would be
594	 * useless anyway, as netns_ids are destroyed there.
595	 */
596	last = list_last_entry(&net_namespace_list, struct net, list);
597	up_write(&net_rwsem);
598
599	llist_for_each_entry(net, net_kill_list, cleanup_list) {
600		unhash_nsid(net, last);
601		list_add_tail(&net->exit_list, &net_exit_list);
602	}
603
604	/* Run all of the network namespace pre_exit methods */
605	list_for_each_entry_reverse(ops, &pernet_list, list)
606		ops_pre_exit_list(ops, &net_exit_list);
607
608	/*
609	 * Another CPU might be rcu-iterating the list, wait for it.
610	 * This needs to be before calling the exit() notifiers, so
611	 * the rcu_barrier() below isn't sufficient alone.
612	 * Also the pre_exit() and exit() methods need this barrier.
613	 */
614	synchronize_rcu();
615
616	/* Run all of the network namespace exit methods */
617	list_for_each_entry_reverse(ops, &pernet_list, list)
618		ops_exit_list(ops, &net_exit_list);
619
620	/* Free the net generic variables */
621	list_for_each_entry_reverse(ops, &pernet_list, list)
622		ops_free_list(ops, &net_exit_list);
623
624	up_read(&pernet_ops_rwsem);
625
626	/* Ensure there are no outstanding rcu callbacks using this
627	 * network namespace.
628	 */
629	rcu_barrier();
630
631	/* Finally it is safe to free my network namespace structure */
632	list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
633		list_del_init(&net->exit_list);
634		dec_net_namespaces(net->ucounts);
635#ifdef CONFIG_KEYS
636		key_remove_domain(net->key_domain);
637#endif
638		put_user_ns(net->user_ns);
639		net_drop_ns(net);
640	}
641}
642
643/**
644 * net_ns_barrier - wait until concurrent net_cleanup_work is done
645 *
646 * cleanup_net runs from work queue and will first remove namespaces
647 * from the global list, then run net exit functions.
648 *
649 * Call this in module exit path to make sure that all netns
650 * ->exit ops have been invoked before the function is removed.
651 */
652void net_ns_barrier(void)
653{
654	down_write(&pernet_ops_rwsem);
655	up_write(&pernet_ops_rwsem);
656}
657EXPORT_SYMBOL(net_ns_barrier);
658
659static DECLARE_WORK(net_cleanup_work, cleanup_net);
660
661void __put_net(struct net *net)
662{
663	/* Cleanup the network namespace in process context */
664	if (llist_add(&net->cleanup_list, &cleanup_list))
665		queue_work(netns_wq, &net_cleanup_work);
666}
667EXPORT_SYMBOL_GPL(__put_net);
668
669/**
670 * get_net_ns - increment the refcount of the network namespace
671 * @ns: common namespace (net)
672 *
673 * Returns the net's common namespace.
674 */
675struct ns_common *get_net_ns(struct ns_common *ns)
676{
677	return &get_net(container_of(ns, struct net, ns))->ns;
678}
679EXPORT_SYMBOL_GPL(get_net_ns);
680
681struct net *get_net_ns_by_fd(int fd)
682{
683	struct file *file;
684	struct ns_common *ns;
685	struct net *net;
686
687	file = proc_ns_fget(fd);
688	if (IS_ERR(file))
689		return ERR_CAST(file);
690
691	ns = get_proc_ns(file_inode(file));
692	if (ns->ops == &netns_operations)
693		net = get_net(container_of(ns, struct net, ns));
694	else
695		net = ERR_PTR(-EINVAL);
696
697	fput(file);
698	return net;
699}
700
701#else
702struct net *get_net_ns_by_fd(int fd)
703{
704	return ERR_PTR(-EINVAL);
705}
706#endif
707EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
708
709struct net *get_net_ns_by_pid(pid_t pid)
710{
711	struct task_struct *tsk;
712	struct net *net;
713
714	/* Lookup the network namespace */
715	net = ERR_PTR(-ESRCH);
716	rcu_read_lock();
717	tsk = find_task_by_vpid(pid);
718	if (tsk) {
719		struct nsproxy *nsproxy;
720		task_lock(tsk);
721		nsproxy = tsk->nsproxy;
722		if (nsproxy)
723			net = get_net(nsproxy->net_ns);
724		task_unlock(tsk);
725	}
726	rcu_read_unlock();
727	return net;
728}
729EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
730
731static __net_init int net_ns_net_init(struct net *net)
732{
733#ifdef CONFIG_NET_NS
734	net->ns.ops = &netns_operations;
735#endif
736	return ns_alloc_inum(&net->ns);
737}
738
739static __net_exit void net_ns_net_exit(struct net *net)
740{
741	ns_free_inum(&net->ns);
742}
743
744static struct pernet_operations __net_initdata net_ns_ops = {
745	.init = net_ns_net_init,
746	.exit = net_ns_net_exit,
747};
748
749static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
750	[NETNSA_NONE]		= { .type = NLA_UNSPEC },
751	[NETNSA_NSID]		= { .type = NLA_S32 },
752	[NETNSA_PID]		= { .type = NLA_U32 },
753	[NETNSA_FD]		= { .type = NLA_U32 },
754	[NETNSA_TARGET_NSID]	= { .type = NLA_S32 },
755};
756
757static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
758			  struct netlink_ext_ack *extack)
759{
760	struct net *net = sock_net(skb->sk);
761	struct nlattr *tb[NETNSA_MAX + 1];
762	struct nlattr *nla;
763	struct net *peer;
764	int nsid, err;
765
766	err = nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg), tb,
767				     NETNSA_MAX, rtnl_net_policy, extack);
768	if (err < 0)
769		return err;
770	if (!tb[NETNSA_NSID]) {
771		NL_SET_ERR_MSG(extack, "nsid is missing");
772		return -EINVAL;
773	}
774	nsid = nla_get_s32(tb[NETNSA_NSID]);
775
776	if (tb[NETNSA_PID]) {
777		peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
778		nla = tb[NETNSA_PID];
779	} else if (tb[NETNSA_FD]) {
780		peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
781		nla = tb[NETNSA_FD];
782	} else {
783		NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
784		return -EINVAL;
785	}
786	if (IS_ERR(peer)) {
787		NL_SET_BAD_ATTR(extack, nla);
788		NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
789		return PTR_ERR(peer);
790	}
791
792	spin_lock_bh(&net->nsid_lock);
793	if (__peernet2id(net, peer) >= 0) {
794		spin_unlock_bh(&net->nsid_lock);
795		err = -EEXIST;
796		NL_SET_BAD_ATTR(extack, nla);
797		NL_SET_ERR_MSG(extack,
798			       "Peer netns already has a nsid assigned");
799		goto out;
800	}
801
802	err = alloc_netid(net, peer, nsid);
803	spin_unlock_bh(&net->nsid_lock);
804	if (err >= 0) {
805		rtnl_net_notifyid(net, RTM_NEWNSID, err, NETLINK_CB(skb).portid,
806				  nlh, GFP_KERNEL);
807		err = 0;
808	} else if (err == -ENOSPC && nsid >= 0) {
809		err = -EEXIST;
810		NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
811		NL_SET_ERR_MSG(extack, "The specified nsid is already used");
812	}
813out:
814	put_net(peer);
815	return err;
816}
817
818static int rtnl_net_get_size(void)
819{
820	return NLMSG_ALIGN(sizeof(struct rtgenmsg))
821	       + nla_total_size(sizeof(s32)) /* NETNSA_NSID */
822	       + nla_total_size(sizeof(s32)) /* NETNSA_CURRENT_NSID */
823	       ;
824}
825
826struct net_fill_args {
827	u32 portid;
828	u32 seq;
829	int flags;
830	int cmd;
831	int nsid;
832	bool add_ref;
833	int ref_nsid;
834};
835
836static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
837{
838	struct nlmsghdr *nlh;
839	struct rtgenmsg *rth;
840
841	nlh = nlmsg_put(skb, args->portid, args->seq, args->cmd, sizeof(*rth),
842			args->flags);
843	if (!nlh)
844		return -EMSGSIZE;
845
846	rth = nlmsg_data(nlh);
847	rth->rtgen_family = AF_UNSPEC;
848
849	if (nla_put_s32(skb, NETNSA_NSID, args->nsid))
850		goto nla_put_failure;
851
852	if (args->add_ref &&
853	    nla_put_s32(skb, NETNSA_CURRENT_NSID, args->ref_nsid))
854		goto nla_put_failure;
855
856	nlmsg_end(skb, nlh);
857	return 0;
858
859nla_put_failure:
860	nlmsg_cancel(skb, nlh);
861	return -EMSGSIZE;
862}
863
864static int rtnl_net_valid_getid_req(struct sk_buff *skb,
865				    const struct nlmsghdr *nlh,
866				    struct nlattr **tb,
867				    struct netlink_ext_ack *extack)
868{
869	int i, err;
870
871	if (!netlink_strict_get_check(skb))
872		return nlmsg_parse_deprecated(nlh, sizeof(struct rtgenmsg),
873					      tb, NETNSA_MAX, rtnl_net_policy,
874					      extack);
875
876	err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
877					    NETNSA_MAX, rtnl_net_policy,
878					    extack);
879	if (err)
880		return err;
881
882	for (i = 0; i <= NETNSA_MAX; i++) {
883		if (!tb[i])
884			continue;
885
886		switch (i) {
887		case NETNSA_PID:
888		case NETNSA_FD:
889		case NETNSA_NSID:
890		case NETNSA_TARGET_NSID:
891			break;
892		default:
893			NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
894			return -EINVAL;
895		}
896	}
897
898	return 0;
899}
900
901static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
902			  struct netlink_ext_ack *extack)
903{
904	struct net *net = sock_net(skb->sk);
905	struct nlattr *tb[NETNSA_MAX + 1];
906	struct net_fill_args fillargs = {
907		.portid = NETLINK_CB(skb).portid,
908		.seq = nlh->nlmsg_seq,
909		.cmd = RTM_NEWNSID,
910	};
911	struct net *peer, *target = net;
912	struct nlattr *nla;
913	struct sk_buff *msg;
914	int err;
915
916	err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
917	if (err < 0)
918		return err;
919	if (tb[NETNSA_PID]) {
920		peer = get_net_ns_by_pid(nla_get_u32(tb[NETNSA_PID]));
921		nla = tb[NETNSA_PID];
922	} else if (tb[NETNSA_FD]) {
923		peer = get_net_ns_by_fd(nla_get_u32(tb[NETNSA_FD]));
924		nla = tb[NETNSA_FD];
925	} else if (tb[NETNSA_NSID]) {
926		peer = get_net_ns_by_id(net, nla_get_s32(tb[NETNSA_NSID]));
927		if (!peer)
928			peer = ERR_PTR(-ENOENT);
929		nla = tb[NETNSA_NSID];
930	} else {
931		NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
932		return -EINVAL;
933	}
934
935	if (IS_ERR(peer)) {
936		NL_SET_BAD_ATTR(extack, nla);
937		NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
938		return PTR_ERR(peer);
939	}
940
941	if (tb[NETNSA_TARGET_NSID]) {
942		int id = nla_get_s32(tb[NETNSA_TARGET_NSID]);
943
944		target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, id);
945		if (IS_ERR(target)) {
946			NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
947			NL_SET_ERR_MSG(extack,
948				       "Target netns reference is invalid");
949			err = PTR_ERR(target);
950			goto out;
951		}
952		fillargs.add_ref = true;
953		fillargs.ref_nsid = peernet2id(net, peer);
954	}
955
956	msg = nlmsg_new(rtnl_net_get_size(), GFP_KERNEL);
957	if (!msg) {
958		err = -ENOMEM;
959		goto out;
960	}
961
962	fillargs.nsid = peernet2id(target, peer);
963	err = rtnl_net_fill(msg, &fillargs);
964	if (err < 0)
965		goto err_out;
966
967	err = rtnl_unicast(msg, net, NETLINK_CB(skb).portid);
968	goto out;
969
970err_out:
971	nlmsg_free(msg);
972out:
973	if (fillargs.add_ref)
974		put_net(target);
975	put_net(peer);
976	return err;
977}
978
979struct rtnl_net_dump_cb {
980	struct net *tgt_net;
981	struct net *ref_net;
982	struct sk_buff *skb;
983	struct net_fill_args fillargs;
984	int idx;
985	int s_idx;
986};
987
988/* Runs in RCU-critical section. */
989static int rtnl_net_dumpid_one(int id, void *peer, void *data)
990{
991	struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
992	int ret;
993
994	if (net_cb->idx < net_cb->s_idx)
995		goto cont;
996
997	net_cb->fillargs.nsid = id;
998	if (net_cb->fillargs.add_ref)
999		net_cb->fillargs.ref_nsid = __peernet2id(net_cb->ref_net, peer);
1000	ret = rtnl_net_fill(net_cb->skb, &net_cb->fillargs);
1001	if (ret < 0)
1002		return ret;
1003
1004cont:
1005	net_cb->idx++;
1006	return 0;
1007}
1008
1009static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
1010				   struct rtnl_net_dump_cb *net_cb,
1011				   struct netlink_callback *cb)
1012{
1013	struct netlink_ext_ack *extack = cb->extack;
1014	struct nlattr *tb[NETNSA_MAX + 1];
1015	int err, i;
1016
1017	err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct rtgenmsg), tb,
1018					    NETNSA_MAX, rtnl_net_policy,
1019					    extack);
1020	if (err < 0)
1021		return err;
1022
1023	for (i = 0; i <= NETNSA_MAX; i++) {
1024		if (!tb[i])
1025			continue;
1026
1027		if (i == NETNSA_TARGET_NSID) {
1028			struct net *net;
1029
1030			net = rtnl_get_net_ns_capable(sk, nla_get_s32(tb[i]));
1031			if (IS_ERR(net)) {
1032				NL_SET_BAD_ATTR(extack, tb[i]);
1033				NL_SET_ERR_MSG(extack,
1034					       "Invalid target network namespace id");
1035				return PTR_ERR(net);
1036			}
1037			net_cb->fillargs.add_ref = true;
1038			net_cb->ref_net = net_cb->tgt_net;
1039			net_cb->tgt_net = net;
1040		} else {
1041			NL_SET_BAD_ATTR(extack, tb[i]);
1042			NL_SET_ERR_MSG(extack,
1043				       "Unsupported attribute in dump request");
1044			return -EINVAL;
1045		}
1046	}
1047
1048	return 0;
1049}
1050
1051static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1052{
1053	struct rtnl_net_dump_cb net_cb = {
1054		.tgt_net = sock_net(skb->sk),
1055		.skb = skb,
1056		.fillargs = {
1057			.portid = NETLINK_CB(cb->skb).portid,
1058			.seq = cb->nlh->nlmsg_seq,
1059			.flags = NLM_F_MULTI,
1060			.cmd = RTM_NEWNSID,
1061		},
1062		.idx = 0,
1063		.s_idx = cb->args[0],
1064	};
1065	int err = 0;
1066
1067	if (cb->strict_check) {
1068		err = rtnl_valid_dump_net_req(cb->nlh, skb->sk, &net_cb, cb);
1069		if (err < 0)
1070			goto end;
1071	}
1072
1073	rcu_read_lock();
1074	idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
1075	rcu_read_unlock();
1076
1077	cb->args[0] = net_cb.idx;
1078end:
1079	if (net_cb.fillargs.add_ref)
1080		put_net(net_cb.tgt_net);
1081	return err < 0 ? err : skb->len;
1082}
1083
1084static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1085			      struct nlmsghdr *nlh, gfp_t gfp)
1086{
1087	struct net_fill_args fillargs = {
1088		.portid = portid,
1089		.seq = nlh ? nlh->nlmsg_seq : 0,
1090		.cmd = cmd,
1091		.nsid = id,
1092	};
1093	struct sk_buff *msg;
1094	int err = -ENOMEM;
1095
1096	msg = nlmsg_new(rtnl_net_get_size(), gfp);
1097	if (!msg)
1098		goto out;
1099
1100	err = rtnl_net_fill(msg, &fillargs);
1101	if (err < 0)
1102		goto err_out;
1103
1104	rtnl_notify(msg, net, portid, RTNLGRP_NSID, nlh, gfp);
1105	return;
1106
1107err_out:
1108	nlmsg_free(msg);
1109out:
1110	rtnl_set_sk_err(net, RTNLGRP_NSID, err);
1111}
1112
1113static int __init net_ns_init(void)
1114{
1115	struct net_generic *ng;
1116
1117#ifdef CONFIG_NET_NS
1118	net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1119					SMP_CACHE_BYTES,
1120					SLAB_PANIC|SLAB_ACCOUNT, NULL);
1121
1122	/* Create workqueue for cleanup */
1123	netns_wq = create_singlethread_workqueue("netns");
1124	if (!netns_wq)
1125		panic("Could not create netns workq");
1126#endif
1127
1128	ng = net_alloc_generic();
1129	if (!ng)
1130		panic("Could not allocate generic netns");
1131
1132	rcu_assign_pointer(init_net.gen, ng);
1133
1134	preempt_disable();
1135	__net_gen_cookie(&init_net);
1136	preempt_enable();
1137
1138	down_write(&pernet_ops_rwsem);
1139	if (setup_net(&init_net, &init_user_ns))
1140		panic("Could not setup the initial network namespace");
1141
1142	init_net_initialized = true;
1143	up_write(&pernet_ops_rwsem);
1144
1145	if (register_pernet_subsys(&net_ns_ops))
1146		panic("Could not register network namespace subsystems");
1147
1148	rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL,
1149		      RTNL_FLAG_DOIT_UNLOCKED);
1150	rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
1151		      RTNL_FLAG_DOIT_UNLOCKED);
1152
1153	return 0;
1154}
1155
1156pure_initcall(net_ns_init);
1157
1158#ifdef CONFIG_NET_NS
1159static int __register_pernet_operations(struct list_head *list,
1160					struct pernet_operations *ops)
1161{
1162	struct net *net;
1163	int error;
1164	LIST_HEAD(net_exit_list);
1165
1166	list_add_tail(&ops->list, list);
1167	if (ops->init || (ops->id && ops->size)) {
1168		/* We held write locked pernet_ops_rwsem, and parallel
1169		 * setup_net() and cleanup_net() are not possible.
1170		 */
1171		for_each_net(net) {
1172			error = ops_init(ops, net);
1173			if (error)
1174				goto out_undo;
1175			list_add_tail(&net->exit_list, &net_exit_list);
1176		}
1177	}
1178	return 0;
1179
1180out_undo:
1181	/* If I have an error cleanup all namespaces I initialized */
1182	list_del(&ops->list);
1183	ops_pre_exit_list(ops, &net_exit_list);
1184	synchronize_rcu();
1185	ops_exit_list(ops, &net_exit_list);
1186	ops_free_list(ops, &net_exit_list);
1187	return error;
1188}
1189
1190static void __unregister_pernet_operations(struct pernet_operations *ops)
1191{
1192	struct net *net;
1193	LIST_HEAD(net_exit_list);
1194
1195	list_del(&ops->list);
1196	/* See comment in __register_pernet_operations() */
1197	for_each_net(net)
1198		list_add_tail(&net->exit_list, &net_exit_list);
1199	ops_pre_exit_list(ops, &net_exit_list);
1200	synchronize_rcu();
1201	ops_exit_list(ops, &net_exit_list);
1202	ops_free_list(ops, &net_exit_list);
1203}
1204
1205#else
1206
1207static int __register_pernet_operations(struct list_head *list,
1208					struct pernet_operations *ops)
1209{
1210	if (!init_net_initialized) {
1211		list_add_tail(&ops->list, list);
1212		return 0;
1213	}
1214
1215	return ops_init(ops, &init_net);
1216}
1217
1218static void __unregister_pernet_operations(struct pernet_operations *ops)
1219{
1220	if (!init_net_initialized) {
1221		list_del(&ops->list);
1222	} else {
1223		LIST_HEAD(net_exit_list);
1224		list_add(&init_net.exit_list, &net_exit_list);
1225		ops_pre_exit_list(ops, &net_exit_list);
1226		synchronize_rcu();
1227		ops_exit_list(ops, &net_exit_list);
1228		ops_free_list(ops, &net_exit_list);
1229	}
1230}
1231
1232#endif /* CONFIG_NET_NS */
1233
1234static DEFINE_IDA(net_generic_ids);
1235
1236static int register_pernet_operations(struct list_head *list,
1237				      struct pernet_operations *ops)
1238{
1239	int error;
1240
1241	if (ops->id) {
1242		error = ida_alloc_min(&net_generic_ids, MIN_PERNET_OPS_ID,
1243				GFP_KERNEL);
1244		if (error < 0)
1245			return error;
1246		*ops->id = error;
1247		/* This does not require READ_ONCE as writers already hold
1248		 * pernet_ops_rwsem. But WRITE_ONCE is needed to protect
1249		 * net_alloc_generic.
1250		 */
1251		WRITE_ONCE(max_gen_ptrs, max(max_gen_ptrs, *ops->id + 1));
1252	}
1253	error = __register_pernet_operations(list, ops);
1254	if (error) {
1255		rcu_barrier();
1256		if (ops->id)
1257			ida_free(&net_generic_ids, *ops->id);
1258	}
1259
1260	return error;
1261}
1262
1263static void unregister_pernet_operations(struct pernet_operations *ops)
1264{
1265	__unregister_pernet_operations(ops);
1266	rcu_barrier();
1267	if (ops->id)
1268		ida_free(&net_generic_ids, *ops->id);
1269}
1270
1271/**
1272 *      register_pernet_subsys - register a network namespace subsystem
1273 *	@ops:  pernet operations structure for the subsystem
1274 *
1275 *	Register a subsystem which has init and exit functions
1276 *	that are called when network namespaces are created and
1277 *	destroyed respectively.
1278 *
1279 *	When registered all network namespace init functions are
1280 *	called for every existing network namespace.  Allowing kernel
1281 *	modules to have a race free view of the set of network namespaces.
1282 *
1283 *	When a new network namespace is created all of the init
1284 *	methods are called in the order in which they were registered.
1285 *
1286 *	When a network namespace is destroyed all of the exit methods
1287 *	are called in the reverse of the order with which they were
1288 *	registered.
1289 */
1290int register_pernet_subsys(struct pernet_operations *ops)
1291{
1292	int error;
1293	down_write(&pernet_ops_rwsem);
1294	error =  register_pernet_operations(first_device, ops);
1295	up_write(&pernet_ops_rwsem);
1296	return error;
1297}
1298EXPORT_SYMBOL_GPL(register_pernet_subsys);
1299
1300/**
1301 *      unregister_pernet_subsys - unregister a network namespace subsystem
1302 *	@ops: pernet operations structure to manipulate
1303 *
1304 *	Remove the pernet operations structure from the list to be
1305 *	used when network namespaces are created or destroyed.  In
1306 *	addition run the exit method for all existing network
1307 *	namespaces.
1308 */
1309void unregister_pernet_subsys(struct pernet_operations *ops)
1310{
1311	down_write(&pernet_ops_rwsem);
1312	unregister_pernet_operations(ops);
1313	up_write(&pernet_ops_rwsem);
1314}
1315EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1316
1317/**
1318 *      register_pernet_device - register a network namespace device
1319 *	@ops:  pernet operations structure for the subsystem
1320 *
1321 *	Register a device which has init and exit functions
1322 *	that are called when network namespaces are created and
1323 *	destroyed respectively.
1324 *
1325 *	When registered all network namespace init functions are
1326 *	called for every existing network namespace.  Allowing kernel
1327 *	modules to have a race free view of the set of network namespaces.
1328 *
1329 *	When a new network namespace is created all of the init
1330 *	methods are called in the order in which they were registered.
1331 *
1332 *	When a network namespace is destroyed all of the exit methods
1333 *	are called in the reverse of the order with which they were
1334 *	registered.
1335 */
1336int register_pernet_device(struct pernet_operations *ops)
1337{
1338	int error;
1339	down_write(&pernet_ops_rwsem);
1340	error = register_pernet_operations(&pernet_list, ops);
1341	if (!error && (first_device == &pernet_list))
1342		first_device = &ops->list;
1343	up_write(&pernet_ops_rwsem);
1344	return error;
1345}
1346EXPORT_SYMBOL_GPL(register_pernet_device);
1347
1348/**
1349 *      unregister_pernet_device - unregister a network namespace netdevice
1350 *	@ops: pernet operations structure to manipulate
1351 *
1352 *	Remove the pernet operations structure from the list to be
1353 *	used when network namespaces are created or destroyed.  In
1354 *	addition run the exit method for all existing network
1355 *	namespaces.
1356 */
1357void unregister_pernet_device(struct pernet_operations *ops)
1358{
1359	down_write(&pernet_ops_rwsem);
1360	if (&ops->list == first_device)
1361		first_device = first_device->next;
1362	unregister_pernet_operations(ops);
1363	up_write(&pernet_ops_rwsem);
1364}
1365EXPORT_SYMBOL_GPL(unregister_pernet_device);
1366
1367#ifdef CONFIG_NET_NS
1368static struct ns_common *netns_get(struct task_struct *task)
1369{
1370	struct net *net = NULL;
1371	struct nsproxy *nsproxy;
1372
1373	task_lock(task);
1374	nsproxy = task->nsproxy;
1375	if (nsproxy)
1376		net = get_net(nsproxy->net_ns);
1377	task_unlock(task);
1378
1379	return net ? &net->ns : NULL;
1380}
1381
1382static inline struct net *to_net_ns(struct ns_common *ns)
1383{
1384	return container_of(ns, struct net, ns);
1385}
1386
1387static void netns_put(struct ns_common *ns)
1388{
1389	put_net(to_net_ns(ns));
1390}
1391
1392static int netns_install(struct nsset *nsset, struct ns_common *ns)
1393{
1394	struct nsproxy *nsproxy = nsset->nsproxy;
1395	struct net *net = to_net_ns(ns);
1396
1397	if (!ns_capable(net->user_ns, CAP_SYS_ADMIN) ||
1398	    !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN))
1399		return -EPERM;
1400
1401	put_net(nsproxy->net_ns);
1402	nsproxy->net_ns = get_net(net);
1403	return 0;
1404}
1405
1406static struct user_namespace *netns_owner(struct ns_common *ns)
1407{
1408	return to_net_ns(ns)->user_ns;
1409}
1410
1411const struct proc_ns_operations netns_operations = {
1412	.name		= "net",
1413	.type		= CLONE_NEWNET,
1414	.get		= netns_get,
1415	.put		= netns_put,
1416	.install	= netns_install,
1417	.owner		= netns_owner,
1418};
1419#endif
1420