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	#include <linux/proc_fs.h>
24
25	#include <net/sock.h>
26	#include <net/netlink.h>
27	#include <net/net_namespace.h>
28	#include <net/netns/generic.h>
29
30	/*
31	* Our network namespace constructor/destructor lists
32	*/
33
34	static LIST_HEAD(pernet_list);
35	static struct list_head *first_device = &pernet_list;
36
37	LIST_HEAD(net_namespace_list);
38	EXPORT_SYMBOL_GPL(net_namespace_list);
39
40	/* Protects net_namespace_list. Nests iside rtnl_lock() */
41	DECLARE_RWSEM(net_rwsem);
42	EXPORT_SYMBOL_GPL(net_rwsem);
43
44	#ifdef CONFIG_KEYS
45	static struct key_tag init_net_key_domain = { .usage = REFCOUNT_INIT(1) };
46	#endif
47
48	struct net init_net;
49	EXPORT_SYMBOL(init_net);
50
51	static bool init_net_initialized;
52	/*
53	* pernet_ops_rwsem: protects: pernet_list, net_generic_ids,
54	* init_net_initialized and first_device pointer.
55	* This is internal net namespace object. Please, don't use it
56	* outside.
57	*/
58	DECLARE_RWSEM(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
69	static struct net_generic *net_alloc_generic(void)
70	{
71	unsigned int gen_ptrs = READ_ONCE(max_gen_ptrs);
72	unsigned int generic_size;
73	struct net_generic *ng;
74
75	generic_size = offsetof(struct net_generic, ptr[gen_ptrs]);
76
77	ng = kzalloc(generic_size, GFP_KERNEL);
78	if (ng)
79	ng->s.len = gen_ptrs;
80
81	return ng;
82	}
83
84	static int net_assign_generic(struct net *net, unsigned int id, void *data)
85	{
86	struct net_generic *ng, *old_ng;
87
88	BUG_ON(id < MIN_PERNET_OPS_ID);
89
90	old_ng = rcu_dereference_protected(net->gen,
91	lockdep_is_held(&pernet_ops_rwsem));
92	if (old_ng->s.len > id) {
93	old_ng->ptr[id] = data;
94	return 0;
95	}
96
97	ng = net_alloc_generic();
98	if (!ng)
99	return -ENOMEM;
100
101	/*
102	* Some synchronisation notes:
103	*
104	* The net_generic explores the net->gen array inside rcu
105	* read section. Besides once set the net->gen->ptr[x]
106	* pointer never changes (see rules in netns/generic.h).
107	*
108	* That said, we simply duplicate this array and schedule
109	* the old copy for kfree after a grace period.
110	*/
111
112	memcpy(&ng->ptr[MIN_PERNET_OPS_ID], &old_ng->ptr[MIN_PERNET_OPS_ID],
113	(old_ng->s.len - MIN_PERNET_OPS_ID) * sizeof(void *));
114	ng->ptr[id] = data;
115
116	rcu_assign_pointer(net->gen, ng);
117	kfree_rcu(old_ng, s.rcu);
118	return 0;
119	}
120
121	static int ops_init(const struct pernet_operations *ops, struct net *net)
122	{
123	struct net_generic *ng;
124	int err = -ENOMEM;
125	void *data = NULL;
126
127	if (ops->id) {
128	data = kzalloc(ops->size, GFP_KERNEL);
129	if (!data)
130	goto out;
131
132	err = net_assign_generic(net, id: *ops->id, data);
133	if (err)
134	goto cleanup;
135	}
136	err = 0;
137	if (ops->init)
138	err = ops->init(net);
139	if (!err)
140	return 0;
141
142	if (ops->id) {
143	ng = rcu_dereference_protected(net->gen,
144	lockdep_is_held(&pernet_ops_rwsem));
145	ng->ptr[*ops->id] = NULL;
146	}
147
148	cleanup:
149	kfree(objp: data);
150
151	out:
152	return err;
153	}
154
155	static void ops_pre_exit_list(const struct pernet_operations *ops,
156	struct list_head *net_exit_list)
157	{
158	struct net *net;
159
160	if (ops->pre_exit) {
161	list_for_each_entry(net, net_exit_list, exit_list)
162	ops->pre_exit(net);
163	}
164	}
165
166	static void ops_exit_rtnl_list(const struct list_head *ops_list,
167	const struct pernet_operations *ops,
168	struct list_head *net_exit_list)
169	{
170	const struct pernet_operations *saved_ops = ops;
171	LIST_HEAD(dev_kill_list);
172	struct net *net;
173
174	rtnl_lock();
175
176	list_for_each_entry(net, net_exit_list, exit_list) {
177	__rtnl_net_lock(net);
178
179	ops = saved_ops;
180	list_for_each_entry_continue_reverse(ops, ops_list, list) {
181	if (ops->exit_rtnl)
182	ops->exit_rtnl(net, &dev_kill_list);
183	}
184
185	__rtnl_net_unlock(net);
186	}
187
188	unregister_netdevice_many(head: &dev_kill_list);
189
190	rtnl_unlock();
191	}
192
193	static void ops_exit_list(const struct pernet_operations *ops,
194	struct list_head *net_exit_list)
195	{
196	if (ops->exit) {
197	struct net *net;
198
199	list_for_each_entry(net, net_exit_list, exit_list) {
200	ops->exit(net);
201	cond_resched();
202	}
203	}
204
205	if (ops->exit_batch)
206	ops->exit_batch(net_exit_list);
207	}
208
209	static void ops_free_list(const struct pernet_operations *ops,
210	struct list_head *net_exit_list)
211	{
212	struct net *net;
213
214	if (ops->id) {
215	list_for_each_entry(net, net_exit_list, exit_list)
216	kfree(objp: net_generic(net, id: *ops->id));
217	}
218	}
219
220	static void ops_undo_list(const struct list_head *ops_list,
221	const struct pernet_operations *ops,
222	struct list_head *net_exit_list,
223	bool expedite_rcu)
224	{
225	const struct pernet_operations *saved_ops;
226	bool hold_rtnl = false;
227
228	if (!ops)
229	ops = list_entry(ops_list, typeof(*ops), list);
230
231	saved_ops = ops;
232
233	list_for_each_entry_continue_reverse(ops, ops_list, list) {
234	hold_rtnl |= !!ops->exit_rtnl;
235	ops_pre_exit_list(ops, net_exit_list);
236	}
237
238	/* Another CPU might be rcu-iterating the list, wait for it.
239	* This needs to be before calling the exit() notifiers, so the
240	* rcu_barrier() after ops_undo_list() isn't sufficient alone.
241	* Also the pre_exit() and exit() methods need this barrier.
242	*/
243	if (expedite_rcu)
244	synchronize_rcu_expedited();
245	else
246	synchronize_rcu();
247
248	if (hold_rtnl)
249	ops_exit_rtnl_list(ops_list, ops: saved_ops, net_exit_list);
250
251	ops = saved_ops;
252	list_for_each_entry_continue_reverse(ops, ops_list, list)
253	ops_exit_list(ops, net_exit_list);
254
255	ops = saved_ops;
256	list_for_each_entry_continue_reverse(ops, ops_list, list)
257	ops_free_list(ops, net_exit_list);
258	}
259
260	static void ops_undo_single(struct pernet_operations *ops,
261	struct list_head *net_exit_list)
262	{
263	LIST_HEAD(ops_list);
264
265	list_add(new: &ops->list, head: &ops_list);
266	ops_undo_list(ops_list: &ops_list, NULL, net_exit_list, expedite_rcu: false);
267	list_del(entry: &ops->list);
268	}
269
270	/* should be called with nsid_lock held */
271	static int alloc_netid(struct net *net, struct net *peer, int reqid)
272	{
273	int min = 0, max = 0;
274
275	if (reqid >= 0) {
276	min = reqid;
277	max = reqid + 1;
278	}
279
280	return idr_alloc(&net->netns_ids, ptr: peer, start: min, end: max, GFP_ATOMIC);
281	}
282
283	/* This function is used by idr_for_each(). If net is equal to peer, the
284	* function returns the id so that idr_for_each() stops. Because we cannot
285	* returns the id 0 (idr_for_each() will not stop), we return the magic value
286	* NET_ID_ZERO (-1) for it.
287	*/
288	#define NET_ID_ZERO -1
289	static int net_eq_idr(int id, void *net, void *peer)
290	{
291	if (net_eq(net1: net, net2: peer))
292	return id ? : NET_ID_ZERO;
293	return 0;
294	}
295
296	/* Must be called from RCU-critical section or with nsid_lock held */
297	static int __peernet2id(const struct net *net, struct net *peer)
298	{
299	int id = idr_for_each(&net->netns_ids, fn: net_eq_idr, data: peer);
300
301	/* Magic value for id 0. */
302	if (id == NET_ID_ZERO)
303	return 0;
304	if (id > 0)
305	return id;
306
307	return NETNSA_NSID_NOT_ASSIGNED;
308	}
309
310	static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
311	struct nlmsghdr *nlh, gfp_t gfp);
312	/* This function returns the id of a peer netns. If no id is assigned, one will
313	* be allocated and returned.
314	*/
315	int peernet2id_alloc(struct net *net, struct net *peer, gfp_t gfp)
316	{
317	int id;
318
319	if (refcount_read(r: &net->ns.count) == 0)
320	return NETNSA_NSID_NOT_ASSIGNED;
321
322	spin_lock_bh(lock: &net->nsid_lock);
323	id = __peernet2id(net, peer);
324	if (id >= 0) {
325	spin_unlock_bh(lock: &net->nsid_lock);
326	return id;
327	}
328
329	/* When peer is obtained from RCU lists, we may race with
330	* its cleanup. Check whether it's alive, and this guarantees
331	* we never hash a peer back to net->netns_ids, after it has
332	* just been idr_remove()'d from there in cleanup_net().
333	*/
334	if (!maybe_get_net(net: peer)) {
335	spin_unlock_bh(lock: &net->nsid_lock);
336	return NETNSA_NSID_NOT_ASSIGNED;
337	}
338
339	id = alloc_netid(net, peer, reqid: -1);
340	spin_unlock_bh(lock: &net->nsid_lock);
341
342	put_net(net: peer);
343	if (id < 0)
344	return NETNSA_NSID_NOT_ASSIGNED;
345
346	rtnl_net_notifyid(net, RTM_NEWNSID, id, portid: 0, NULL, gfp);
347
348	return id;
349	}
350	EXPORT_SYMBOL_GPL(peernet2id_alloc);
351
352	/* This function returns, if assigned, the id of a peer netns. */
353	int peernet2id(const struct net *net, struct net *peer)
354	{
355	int id;
356
357	rcu_read_lock();
358	id = __peernet2id(net, peer);
359	rcu_read_unlock();
360
361	return id;
362	}
363	EXPORT_SYMBOL(peernet2id);
364
365	/* This function returns true is the peer netns has an id assigned into the
366	* current netns.
367	*/
368	bool peernet_has_id(const struct net *net, struct net *peer)
369	{
370	return peernet2id(net, peer) >= 0;
371	}
372
373	struct net *get_net_ns_by_id(const struct net *net, int id)
374	{
375	struct net *peer;
376
377	if (id < 0)
378	return NULL;
379
380	rcu_read_lock();
381	peer = idr_find(&net->netns_ids, id);
382	if (peer)
383	peer = maybe_get_net(net: peer);
384	rcu_read_unlock();
385
386	return peer;
387	}
388	EXPORT_SYMBOL_GPL(get_net_ns_by_id);
389
390	static __net_init void preinit_net_sysctl(struct net *net)
391	{
392	net->core.sysctl_somaxconn = SOMAXCONN;
393	/* Limits per socket sk_omem_alloc usage.
394	* TCP zerocopy regular usage needs 128 KB.
395	*/
396	net->core.sysctl_optmem_max = 128 * 1024;
397	net->core.sysctl_txrehash = SOCK_TXREHASH_ENABLED;
398	net->core.sysctl_tstamp_allow_data = 1;
399	}
400
401	/* init code that must occur even if setup_net() is not called. */
402	static __net_init void preinit_net(struct net *net, struct user_namespace *user_ns)
403	{
404	refcount_set(r: &net->passive, n: 1);
405	refcount_set(r: &net->ns.count, n: 1);
406	ref_tracker_dir_init(dir: &net->refcnt_tracker, quarantine_count: 128, name: "net refcnt");
407	ref_tracker_dir_init(dir: &net->notrefcnt_tracker, quarantine_count: 128, name: "net notrefcnt");
408
409	get_random_bytes(buf: &net->hash_mix, len: sizeof(u32));
410	net->dev_base_seq = 1;
411	net->user_ns = user_ns;
412
413	idr_init(idr: &net->netns_ids);
414	spin_lock_init(&net->nsid_lock);
415	mutex_init(&net->ipv4.ra_mutex);
416
417	#ifdef CONFIG_DEBUG_NET_SMALL_RTNL
418	mutex_init(&net->rtnl_mutex);
419	lock_set_cmp_fn(&net->rtnl_mutex, rtnl_net_lock_cmp_fn, NULL);
420	#endif
421
422	INIT_LIST_HEAD(list: &net->ptype_all);
423	INIT_LIST_HEAD(list: &net->ptype_specific);
424	preinit_net_sysctl(net);
425	}
426
427	/*
428	* setup_net runs the initializers for the network namespace object.
429	*/
430	static __net_init int setup_net(struct net *net)
431	{
432	/* Must be called with pernet_ops_rwsem held */
433	const struct pernet_operations *ops;
434	LIST_HEAD(net_exit_list);
435	int error = 0;
436
437	preempt_disable();
438	net->net_cookie = gen_cookie_next(gc: &net_cookie);
439	preempt_enable();
440
441	list_for_each_entry(ops, &pernet_list, list) {
442	error = ops_init(ops, net);
443	if (error < 0)
444	goto out_undo;
445	}
446	down_write(sem: &net_rwsem);
447	list_add_tail_rcu(new: &net->list, head: &net_namespace_list);
448	up_write(sem: &net_rwsem);
449	out:
450	return error;
451
452	out_undo:
453	/* Walk through the list backwards calling the exit functions
454	* for the pernet modules whose init functions did not fail.
455	*/
456	list_add(new: &net->exit_list, head: &net_exit_list);
457	ops_undo_list(ops_list: &pernet_list, ops, net_exit_list: &net_exit_list, expedite_rcu: false);
458	rcu_barrier();
459	goto out;
460	}
461
462	#ifdef CONFIG_NET_NS
463	static struct ucounts *inc_net_namespaces(struct user_namespace *ns)
464	{
465	return inc_ucount(ns, current_euid(), type: UCOUNT_NET_NAMESPACES);
466	}
467
468	static void dec_net_namespaces(struct ucounts *ucounts)
469	{
470	dec_ucount(ucounts, type: UCOUNT_NET_NAMESPACES);
471	}
472
473	static struct kmem_cache *net_cachep __ro_after_init;
474	static struct workqueue_struct *netns_wq;
475
476	static struct net *net_alloc(void)
477	{
478	struct net *net = NULL;
479	struct net_generic *ng;
480
481	ng = net_alloc_generic();
482	if (!ng)
483	goto out;
484
485	net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
486	if (!net)
487	goto out_free;
488
489	#ifdef CONFIG_KEYS
490	net->key_domain = kzalloc(sizeof(struct key_tag), GFP_KERNEL);
491	if (!net->key_domain)
492	goto out_free_2;
493	refcount_set(r: &net->key_domain->usage, n: 1);
494	#endif
495
496	rcu_assign_pointer(net->gen, ng);
497	out:
498	return net;
499
500	#ifdef CONFIG_KEYS
501	out_free_2:
502	kmem_cache_free(s: net_cachep, objp: net);
503	net = NULL;
504	#endif
505	out_free:
506	kfree(objp: ng);
507	goto out;
508	}
509
510	static LLIST_HEAD(defer_free_list);
511
512	static void net_complete_free(void)
513	{
514	struct llist_node *kill_list;
515	struct net *net, *next;
516
517	/* Get the list of namespaces to free from last round. */
518	kill_list = llist_del_all(head: &defer_free_list);
519
520	llist_for_each_entry_safe(net, next, kill_list, defer_free_list)
521	kmem_cache_free(s: net_cachep, objp: net);
522
523	}
524
525	void net_passive_dec(struct net *net)
526	{
527	if (refcount_dec_and_test(r: &net->passive)) {
528	kfree(rcu_access_pointer(net->gen));
529
530	/* There should not be any trackers left there. */
531	ref_tracker_dir_exit(dir: &net->notrefcnt_tracker);
532
533	/* Wait for an extra rcu_barrier() before final free. */
534	llist_add(new: &net->defer_free_list, head: &defer_free_list);
535	}
536	}
537
538	void net_drop_ns(void *p)
539	{
540	struct net *net = (struct net *)p;
541
542	if (net)
543	net_passive_dec(net);
544	}
545
546	struct net *copy_net_ns(unsigned long flags,
547	struct user_namespace *user_ns, struct net *old_net)
548	{
549	struct ucounts *ucounts;
550	struct net *net;
551	int rv;
552
553	if (!(flags & CLONE_NEWNET))
554	return get_net(net: old_net);
555
556	ucounts = inc_net_namespaces(ns: user_ns);
557	if (!ucounts)
558	return ERR_PTR(error: -ENOSPC);
559
560	net = net_alloc();
561	if (!net) {
562	rv = -ENOMEM;
563	goto dec_ucounts;
564	}
565
566	preinit_net(net, user_ns);
567	net->ucounts = ucounts;
568	get_user_ns(ns: user_ns);
569
570	rv = down_read_killable(sem: &pernet_ops_rwsem);
571	if (rv < 0)
572	goto put_userns;
573
574	rv = setup_net(net);
575
576	up_read(sem: &pernet_ops_rwsem);
577
578	if (rv < 0) {
579	put_userns:
580	#ifdef CONFIG_KEYS
581	key_remove_domain(domain_tag: net->key_domain);
582	#endif
583	put_user_ns(ns: user_ns);
584	net_passive_dec(net);
585	dec_ucounts:
586	dec_net_namespaces(ucounts);
587	return ERR_PTR(error: rv);
588	}
589	return net;
590	}
591
592	/**
593	* net_ns_get_ownership - get sysfs ownership data for @net
594	* @net: network namespace in question (can be NULL)
595	* @uid: kernel user ID for sysfs objects
596	* @gid: kernel group ID for sysfs objects
597	*
598	* Returns the uid/gid pair of root in the user namespace associated with the
599	* given network namespace.
600	*/
601	void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
602	{
603	if (net) {
604	kuid_t ns_root_uid = make_kuid(from: net->user_ns, uid: 0);
605	kgid_t ns_root_gid = make_kgid(from: net->user_ns, gid: 0);
606
607	if (uid_valid(uid: ns_root_uid))
608	*uid = ns_root_uid;
609
610	if (gid_valid(gid: ns_root_gid))
611	*gid = ns_root_gid;
612	} else {
613	*uid = GLOBAL_ROOT_UID;
614	*gid = GLOBAL_ROOT_GID;
615	}
616	}
617	EXPORT_SYMBOL_GPL(net_ns_get_ownership);
618
619	static void unhash_nsid(struct net *net, struct net *last)
620	{
621	struct net *tmp;
622	/* This function is only called from cleanup_net() work,
623	* and this work is the only process, that may delete
624	* a net from net_namespace_list. So, when the below
625	* is executing, the list may only grow. Thus, we do not
626	* use for_each_net_rcu() or net_rwsem.
627	*/
628	for_each_net(tmp) {
629	int id;
630
631	spin_lock_bh(lock: &tmp->nsid_lock);
632	id = __peernet2id(net: tmp, peer: net);
633	if (id >= 0)
634	idr_remove(&tmp->netns_ids, id);
635	spin_unlock_bh(lock: &tmp->nsid_lock);
636	if (id >= 0)
637	rtnl_net_notifyid(net: tmp, RTM_DELNSID, id, portid: 0, NULL,
638	GFP_KERNEL);
639	if (tmp == last)
640	break;
641	}
642	spin_lock_bh(lock: &net->nsid_lock);
643	idr_destroy(&net->netns_ids);
644	spin_unlock_bh(lock: &net->nsid_lock);
645	}
646
647	static LLIST_HEAD(cleanup_list);
648
649	struct task_struct *cleanup_net_task;
650
651	static void cleanup_net(struct work_struct *work)
652	{
653	struct llist_node *net_kill_list;
654	struct net *net, *tmp, *last;
655	LIST_HEAD(net_exit_list);
656
657	WRITE_ONCE(cleanup_net_task, current);
658
659	/* Atomically snapshot the list of namespaces to cleanup */
660	net_kill_list = llist_del_all(head: &cleanup_list);
661
662	down_read(sem: &pernet_ops_rwsem);
663
664	/* Don't let anyone else find us. */
665	down_write(sem: &net_rwsem);
666	llist_for_each_entry(net, net_kill_list, cleanup_list)
667	list_del_rcu(entry: &net->list);
668	/* Cache last net. After we unlock rtnl, no one new net
669	* added to net_namespace_list can assign nsid pointer
670	* to a net from net_kill_list (see peernet2id_alloc()).
671	* So, we skip them in unhash_nsid().
672	*
673	* Note, that unhash_nsid() does not delete nsid links
674	* between net_kill_list's nets, as they've already
675	* deleted from net_namespace_list. But, this would be
676	* useless anyway, as netns_ids are destroyed there.
677	*/
678	last = list_last_entry(&net_namespace_list, struct net, list);
679	up_write(sem: &net_rwsem);
680
681	llist_for_each_entry(net, net_kill_list, cleanup_list) {
682	unhash_nsid(net, last);
683	list_add_tail(new: &net->exit_list, head: &net_exit_list);
684	}
685
686	ops_undo_list(ops_list: &pernet_list, NULL, net_exit_list: &net_exit_list, expedite_rcu: true);
687
688	up_read(sem: &pernet_ops_rwsem);
689
690	/* Ensure there are no outstanding rcu callbacks using this
691	* network namespace.
692	*/
693	rcu_barrier();
694
695	net_complete_free();
696
697	/* Finally it is safe to free my network namespace structure */
698	list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
699	list_del_init(entry: &net->exit_list);
700	dec_net_namespaces(ucounts: net->ucounts);
701	#ifdef CONFIG_KEYS
702	key_remove_domain(domain_tag: net->key_domain);
703	#endif
704	put_user_ns(ns: net->user_ns);
705	net_passive_dec(net);
706	}
707	WRITE_ONCE(cleanup_net_task, NULL);
708	}
709
710	/**
711	* net_ns_barrier - wait until concurrent net_cleanup_work is done
712	*
713	* cleanup_net runs from work queue and will first remove namespaces
714	* from the global list, then run net exit functions.
715	*
716	* Call this in module exit path to make sure that all netns
717	* ->exit ops have been invoked before the function is removed.
718	*/
719	void net_ns_barrier(void)
720	{
721	down_write(sem: &pernet_ops_rwsem);
722	up_write(sem: &pernet_ops_rwsem);
723	}
724	EXPORT_SYMBOL(net_ns_barrier);
725
726	static DECLARE_WORK(net_cleanup_work, cleanup_net);
727
728	void __put_net(struct net *net)
729	{
730	ref_tracker_dir_exit(dir: &net->refcnt_tracker);
731	/* Cleanup the network namespace in process context */
732	if (llist_add(new: &net->cleanup_list, head: &cleanup_list))
733	queue_work(wq: netns_wq, work: &net_cleanup_work);
734	}
735	EXPORT_SYMBOL_GPL(__put_net);
736
737	/**
738	* get_net_ns - increment the refcount of the network namespace
739	* @ns: common namespace (net)
740	*
741	* Returns the net's common namespace or ERR_PTR() if ref is zero.
742	*/
743	struct ns_common *get_net_ns(struct ns_common *ns)
744	{
745	struct net *net;
746
747	net = maybe_get_net(container_of(ns, struct net, ns));
748	if (net)
749	return &net->ns;
750	return ERR_PTR(error: -EINVAL);
751	}
752	EXPORT_SYMBOL_GPL(get_net_ns);
753
754	struct net *get_net_ns_by_fd(int fd)
755	{
756	CLASS(fd, f)(fd);
757
758	if (fd_empty(f))
759	return ERR_PTR(error: -EBADF);
760
761	if (proc_ns_file(fd_file(f))) {
762	struct ns_common *ns = get_proc_ns(file_inode(fd_file(f)));
763	if (ns->ops == &netns_operations)
764	return get_net(container_of(ns, struct net, ns));
765	}
766
767	return ERR_PTR(error: -EINVAL);
768	}
769	EXPORT_SYMBOL_GPL(get_net_ns_by_fd);
770	#endif
771
772	struct net *get_net_ns_by_pid(pid_t pid)
773	{
774	struct task_struct *tsk;
775	struct net *net;
776
777	/* Lookup the network namespace */
778	net = ERR_PTR(error: -ESRCH);
779	rcu_read_lock();
780	tsk = find_task_by_vpid(nr: pid);
781	if (tsk) {
782	struct nsproxy *nsproxy;
783	task_lock(p: tsk);
784	nsproxy = tsk->nsproxy;
785	if (nsproxy)
786	net = get_net(net: nsproxy->net_ns);
787	task_unlock(p: tsk);
788	}
789	rcu_read_unlock();
790	return net;
791	}
792	EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
793
794	static __net_init int net_ns_net_init(struct net *net)
795	{
796	#ifdef CONFIG_NET_NS
797	net->ns.ops = &netns_operations;
798	#endif
799	return ns_alloc_inum(ns: &net->ns);
800	}
801
802	static __net_exit void net_ns_net_exit(struct net *net)
803	{
804	ns_free_inum(&net->ns);
805	}
806
807	static struct pernet_operations __net_initdata net_ns_ops = {
808	.init = net_ns_net_init,
809	.exit = net_ns_net_exit,
810	};
811
812	static const struct nla_policy rtnl_net_policy[NETNSA_MAX + 1] = {
813	[NETNSA_NONE] = { .type = NLA_UNSPEC },
814	[NETNSA_NSID] = { .type = NLA_S32 },
815	[NETNSA_PID] = { .type = NLA_U32 },
816	[NETNSA_FD] = { .type = NLA_U32 },
817	[NETNSA_TARGET_NSID] = { .type = NLA_S32 },
818	};
819
820	static int rtnl_net_newid(struct sk_buff *skb, struct nlmsghdr *nlh,
821	struct netlink_ext_ack *extack)
822	{
823	struct net *net = sock_net(sk: skb->sk);
824	struct nlattr *tb[NETNSA_MAX + 1];
825	struct nlattr *nla;
826	struct net *peer;
827	int nsid, err;
828
829	err = nlmsg_parse_deprecated(nlh, hdrlen: sizeof(struct rtgenmsg), tb,
830	NETNSA_MAX, policy: rtnl_net_policy, extack);
831	if (err < 0)
832	return err;
833	if (!tb[NETNSA_NSID]) {
834	NL_SET_ERR_MSG(extack, "nsid is missing");
835	return -EINVAL;
836	}
837	nsid = nla_get_s32(nla: tb[NETNSA_NSID]);
838
839	if (tb[NETNSA_PID]) {
840	peer = get_net_ns_by_pid(nla_get_u32(nla: tb[NETNSA_PID]));
841	nla = tb[NETNSA_PID];
842	} else if (tb[NETNSA_FD]) {
843	peer = get_net_ns_by_fd(nla_get_u32(nla: tb[NETNSA_FD]));
844	nla = tb[NETNSA_FD];
845	} else {
846	NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
847	return -EINVAL;
848	}
849	if (IS_ERR(ptr: peer)) {
850	NL_SET_BAD_ATTR(extack, nla);
851	NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
852	return PTR_ERR(ptr: peer);
853	}
854
855	spin_lock_bh(lock: &net->nsid_lock);
856	if (__peernet2id(net, peer) >= 0) {
857	spin_unlock_bh(lock: &net->nsid_lock);
858	err = -EEXIST;
859	NL_SET_BAD_ATTR(extack, nla);
860	NL_SET_ERR_MSG(extack,
861	"Peer netns already has a nsid assigned");
862	goto out;
863	}
864
865	err = alloc_netid(net, peer, reqid: nsid);
866	spin_unlock_bh(lock: &net->nsid_lock);
867	if (err >= 0) {
868	rtnl_net_notifyid(net, RTM_NEWNSID, id: err, NETLINK_CB(skb).portid,
869	nlh, GFP_KERNEL);
870	err = 0;
871	} else if (err == -ENOSPC && nsid >= 0) {
872	err = -EEXIST;
873	NL_SET_BAD_ATTR(extack, tb[NETNSA_NSID]);
874	NL_SET_ERR_MSG(extack, "The specified nsid is already used");
875	}
876	out:
877	put_net(net: peer);
878	return err;
879	}
880
881	static int rtnl_net_get_size(void)
882	{
883	return NLMSG_ALIGN(sizeof(struct rtgenmsg))
884	+ nla_total_size(payload: sizeof(s32)) /* NETNSA_NSID */
885	+ nla_total_size(payload: sizeof(s32)) /* NETNSA_CURRENT_NSID */
886	;
887	}
888
889	struct net_fill_args {
890	u32 portid;
891	u32 seq;
892	int flags;
893	int cmd;
894	int nsid;
895	bool add_ref;
896	int ref_nsid;
897	};
898
899	static int rtnl_net_fill(struct sk_buff *skb, struct net_fill_args *args)
900	{
901	struct nlmsghdr *nlh;
902	struct rtgenmsg *rth;
903
904	nlh = nlmsg_put(skb, portid: args->portid, seq: args->seq, type: args->cmd, payload: sizeof(*rth),
905	flags: args->flags);
906	if (!nlh)
907	return -EMSGSIZE;
908
909	rth = nlmsg_data(nlh);
910	rth->rtgen_family = AF_UNSPEC;
911
912	if (nla_put_s32(skb, attrtype: NETNSA_NSID, value: args->nsid))
913	goto nla_put_failure;
914
915	if (args->add_ref &&
916	nla_put_s32(skb, attrtype: NETNSA_CURRENT_NSID, value: args->ref_nsid))
917	goto nla_put_failure;
918
919	nlmsg_end(skb, nlh);
920	return 0;
921
922	nla_put_failure:
923	nlmsg_cancel(skb, nlh);
924	return -EMSGSIZE;
925	}
926
927	static int rtnl_net_valid_getid_req(struct sk_buff *skb,
928	const struct nlmsghdr *nlh,
929	struct nlattr **tb,
930	struct netlink_ext_ack *extack)
931	{
932	int i, err;
933
934	if (!netlink_strict_get_check(skb))
935	return nlmsg_parse_deprecated(nlh, hdrlen: sizeof(struct rtgenmsg),
936	tb, NETNSA_MAX, policy: rtnl_net_policy,
937	extack);
938
939	err = nlmsg_parse_deprecated_strict(nlh, hdrlen: sizeof(struct rtgenmsg), tb,
940	NETNSA_MAX, policy: rtnl_net_policy,
941	extack);
942	if (err)
943	return err;
944
945	for (i = 0; i <= NETNSA_MAX; i++) {
946	if (!tb[i])
947	continue;
948
949	switch (i) {
950	case NETNSA_PID:
951	case NETNSA_FD:
952	case NETNSA_NSID:
953	case NETNSA_TARGET_NSID:
954	break;
955	default:
956	NL_SET_ERR_MSG(extack, "Unsupported attribute in peer netns getid request");
957	return -EINVAL;
958	}
959	}
960
961	return 0;
962	}
963
964	static int rtnl_net_getid(struct sk_buff *skb, struct nlmsghdr *nlh,
965	struct netlink_ext_ack *extack)
966	{
967	struct net *net = sock_net(sk: skb->sk);
968	struct nlattr *tb[NETNSA_MAX + 1];
969	struct net_fill_args fillargs = {
970	.portid = NETLINK_CB(skb).portid,
971	.seq = nlh->nlmsg_seq,
972	.cmd = RTM_NEWNSID,
973	};
974	struct net *peer, *target = net;
975	struct nlattr *nla;
976	struct sk_buff *msg;
977	int err;
978
979	err = rtnl_net_valid_getid_req(skb, nlh, tb, extack);
980	if (err < 0)
981	return err;
982	if (tb[NETNSA_PID]) {
983	peer = get_net_ns_by_pid(nla_get_u32(nla: tb[NETNSA_PID]));
984	nla = tb[NETNSA_PID];
985	} else if (tb[NETNSA_FD]) {
986	peer = get_net_ns_by_fd(nla_get_u32(nla: tb[NETNSA_FD]));
987	nla = tb[NETNSA_FD];
988	} else if (tb[NETNSA_NSID]) {
989	peer = get_net_ns_by_id(net, nla_get_s32(nla: tb[NETNSA_NSID]));
990	if (!peer)
991	peer = ERR_PTR(error: -ENOENT);
992	nla = tb[NETNSA_NSID];
993	} else {
994	NL_SET_ERR_MSG(extack, "Peer netns reference is missing");
995	return -EINVAL;
996	}
997
998	if (IS_ERR(ptr: peer)) {
999	NL_SET_BAD_ATTR(extack, nla);
1000	NL_SET_ERR_MSG(extack, "Peer netns reference is invalid");
1001	return PTR_ERR(ptr: peer);
1002	}
1003
1004	if (tb[NETNSA_TARGET_NSID]) {
1005	int id = nla_get_s32(nla: tb[NETNSA_TARGET_NSID]);
1006
1007	target = rtnl_get_net_ns_capable(NETLINK_CB(skb).sk, netnsid: id);
1008	if (IS_ERR(ptr: target)) {
1009	NL_SET_BAD_ATTR(extack, tb[NETNSA_TARGET_NSID]);
1010	NL_SET_ERR_MSG(extack,
1011	"Target netns reference is invalid");
1012	err = PTR_ERR(ptr: target);
1013	goto out;
1014	}
1015	fillargs.add_ref = true;
1016	fillargs.ref_nsid = peernet2id(net, peer);
1017	}
1018
1019	msg = nlmsg_new(payload: rtnl_net_get_size(), GFP_KERNEL);
1020	if (!msg) {
1021	err = -ENOMEM;
1022	goto out;
1023	}
1024
1025	fillargs.nsid = peernet2id(target, peer);
1026	err = rtnl_net_fill(skb: msg, args: &fillargs);
1027	if (err < 0)
1028	goto err_out;
1029
1030	err = rtnl_unicast(skb: msg, net, NETLINK_CB(skb).portid);
1031	goto out;
1032
1033	err_out:
1034	nlmsg_free(skb: msg);
1035	out:
1036	if (fillargs.add_ref)
1037	put_net(net: target);
1038	put_net(net: peer);
1039	return err;
1040	}
1041
1042	struct rtnl_net_dump_cb {
1043	struct net *tgt_net;
1044	struct net *ref_net;
1045	struct sk_buff *skb;
1046	struct net_fill_args fillargs;
1047	int idx;
1048	int s_idx;
1049	};
1050
1051	/* Runs in RCU-critical section. */
1052	static int rtnl_net_dumpid_one(int id, void *peer, void *data)
1053	{
1054	struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
1055	int ret;
1056
1057	if (net_cb->idx < net_cb->s_idx)
1058	goto cont;
1059
1060	net_cb->fillargs.nsid = id;
1061	if (net_cb->fillargs.add_ref)
1062	net_cb->fillargs.ref_nsid = __peernet2id(net: net_cb->ref_net, peer);
1063	ret = rtnl_net_fill(skb: net_cb->skb, args: &net_cb->fillargs);
1064	if (ret < 0)
1065	return ret;
1066
1067	cont:
1068	net_cb->idx++;
1069	return 0;
1070	}
1071
1072	static int rtnl_valid_dump_net_req(const struct nlmsghdr *nlh, struct sock *sk,
1073	struct rtnl_net_dump_cb *net_cb,
1074	struct netlink_callback *cb)
1075	{
1076	struct netlink_ext_ack *extack = cb->extack;
1077	struct nlattr *tb[NETNSA_MAX + 1];
1078	int err, i;
1079
1080	err = nlmsg_parse_deprecated_strict(nlh, hdrlen: sizeof(struct rtgenmsg), tb,
1081	NETNSA_MAX, policy: rtnl_net_policy,
1082	extack);
1083	if (err < 0)
1084	return err;
1085
1086	for (i = 0; i <= NETNSA_MAX; i++) {
1087	if (!tb[i])
1088	continue;
1089
1090	if (i == NETNSA_TARGET_NSID) {
1091	struct net *net;
1092
1093	net = rtnl_get_net_ns_capable(sk, netnsid: nla_get_s32(nla: tb[i]));
1094	if (IS_ERR(ptr: net)) {
1095	NL_SET_BAD_ATTR(extack, tb[i]);
1096	NL_SET_ERR_MSG(extack,
1097	"Invalid target network namespace id");
1098	return PTR_ERR(ptr: net);
1099	}
1100	net_cb->fillargs.add_ref = true;
1101	net_cb->ref_net = net_cb->tgt_net;
1102	net_cb->tgt_net = net;
1103	} else {
1104	NL_SET_BAD_ATTR(extack, tb[i]);
1105	NL_SET_ERR_MSG(extack,
1106	"Unsupported attribute in dump request");
1107	return -EINVAL;
1108	}
1109	}
1110
1111	return 0;
1112	}
1113
1114	static int rtnl_net_dumpid(struct sk_buff *skb, struct netlink_callback *cb)
1115	{
1116	struct rtnl_net_dump_cb net_cb = {
1117	.tgt_net = sock_net(sk: skb->sk),
1118	.skb = skb,
1119	.fillargs = {
1120	.portid = NETLINK_CB(cb->skb).portid,
1121	.seq = cb->nlh->nlmsg_seq,
1122	.flags = NLM_F_MULTI,
1123	.cmd = RTM_NEWNSID,
1124	},
1125	.idx = 0,
1126	.s_idx = cb->args[0],
1127	};
1128	int err = 0;
1129
1130	if (cb->strict_check) {
1131	err = rtnl_valid_dump_net_req(nlh: cb->nlh, sk: skb->sk, net_cb: &net_cb, cb);
1132	if (err < 0)
1133	goto end;
1134	}
1135
1136	rcu_read_lock();
1137	idr_for_each(&net_cb.tgt_net->netns_ids, fn: rtnl_net_dumpid_one, data: &net_cb);
1138	rcu_read_unlock();
1139
1140	cb->args[0] = net_cb.idx;
1141	end:
1142	if (net_cb.fillargs.add_ref)
1143	put_net(net: net_cb.tgt_net);
1144	return err;
1145	}
1146
1147	static void rtnl_net_notifyid(struct net *net, int cmd, int id, u32 portid,
1148	struct nlmsghdr *nlh, gfp_t gfp)
1149	{
1150	struct net_fill_args fillargs = {
1151	.portid = portid,
1152	.seq = nlh ? nlh->nlmsg_seq : 0,
1153	.cmd = cmd,
1154	.nsid = id,
1155	};
1156	struct sk_buff *msg;
1157	int err = -ENOMEM;
1158
1159	msg = nlmsg_new(payload: rtnl_net_get_size(), flags: gfp);
1160	if (!msg)
1161	goto out;
1162
1163	err = rtnl_net_fill(skb: msg, args: &fillargs);
1164	if (err < 0)
1165	goto err_out;
1166
1167	rtnl_notify(skb: msg, net, pid: portid, RTNLGRP_NSID, nlh, flags: gfp);
1168	return;
1169
1170	err_out:
1171	nlmsg_free(skb: msg);
1172	out:
1173	rtnl_set_sk_err(net, RTNLGRP_NSID, error: err);
1174	}
1175
1176	#ifdef CONFIG_NET_NS
1177	static void __init netns_ipv4_struct_check(void)
1178	{
1179	/* TX readonly hotpath cache lines */
1180	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1181	sysctl_tcp_early_retrans);
1182	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1183	sysctl_tcp_tso_win_divisor);
1184	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1185	sysctl_tcp_tso_rtt_log);
1186	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1187	sysctl_tcp_autocorking);
1188	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1189	sysctl_tcp_min_snd_mss);
1190	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1191	sysctl_tcp_notsent_lowat);
1192	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1193	sysctl_tcp_limit_output_bytes);
1194	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1195	sysctl_tcp_min_rtt_wlen);
1196	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1197	sysctl_tcp_wmem);
1198	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_tx,
1199	sysctl_ip_fwd_use_pmtu);
1200	CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_tx, 33);
1201
1202	/* TXRX readonly hotpath cache lines */
1203	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_txrx,
1204	sysctl_tcp_moderate_rcvbuf);
1205	CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_txrx, 1);
1206
1207	/* RX readonly hotpath cache line */
1208	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1209	sysctl_ip_early_demux);
1210	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1211	sysctl_tcp_early_demux);
1212	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1213	sysctl_tcp_l3mdev_accept);
1214	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1215	sysctl_tcp_reordering);
1216	CACHELINE_ASSERT_GROUP_MEMBER(struct netns_ipv4, netns_ipv4_read_rx,
1217	sysctl_tcp_rmem);
1218	CACHELINE_ASSERT_GROUP_SIZE(struct netns_ipv4, netns_ipv4_read_rx, 22);
1219	}
1220	#endif
1221
1222	static const struct rtnl_msg_handler net_ns_rtnl_msg_handlers[] __initconst = {
1223	{.msgtype = RTM_NEWNSID, .doit = rtnl_net_newid,
1224	.flags = RTNL_FLAG_DOIT_UNLOCKED},
1225	{.msgtype = RTM_GETNSID, .doit = rtnl_net_getid,
1226	.dumpit = rtnl_net_dumpid,
1227	.flags = RTNL_FLAG_DOIT_UNLOCKED | RTNL_FLAG_DUMP_UNLOCKED},
1228	};
1229
1230	void __init net_ns_init(void)
1231	{
1232	struct net_generic *ng;
1233
1234	#ifdef CONFIG_NET_NS
1235	netns_ipv4_struct_check();
1236	net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
1237	SMP_CACHE_BYTES,
1238	SLAB_PANIC|SLAB_ACCOUNT, NULL);
1239
1240	/* Create workqueue for cleanup */
1241	netns_wq = create_singlethread_workqueue("netns");
1242	if (!netns_wq)
1243	panic(fmt: "Could not create netns workq");
1244	#endif
1245
1246	ng = net_alloc_generic();
1247	if (!ng)
1248	panic(fmt: "Could not allocate generic netns");
1249
1250	rcu_assign_pointer(init_net.gen, ng);
1251
1252	#ifdef CONFIG_KEYS
1253	init_net.key_domain = &init_net_key_domain;
1254	#endif
1255	preinit_net(net: &init_net, user_ns: &init_user_ns);
1256
1257	down_write(sem: &pernet_ops_rwsem);
1258	if (setup_net(&init_net))
1259	panic(fmt: "Could not setup the initial network namespace");
1260
1261	init_net_initialized = true;
1262	up_write(sem: &pernet_ops_rwsem);
1263
1264	if (register_pernet_subsys(&net_ns_ops))
1265	panic(fmt: "Could not register network namespace subsystems");
1266
1267	rtnl_register_many(net_ns_rtnl_msg_handlers);
1268	}
1269
1270	#ifdef CONFIG_NET_NS
1271	static int __register_pernet_operations(struct list_head *list,
1272	struct pernet_operations *ops)
1273	{
1274	LIST_HEAD(net_exit_list);
1275	struct net *net;
1276	int error;
1277
1278	list_add_tail(new: &ops->list, head: list);
1279	if (ops->init || ops->id) {
1280	/* We held write locked pernet_ops_rwsem, and parallel
1281	* setup_net() and cleanup_net() are not possible.
1282	*/
1283	for_each_net(net) {
1284	error = ops_init(ops, net);
1285	if (error)
1286	goto out_undo;
1287	list_add_tail(new: &net->exit_list, head: &net_exit_list);
1288	}
1289	}
1290	return 0;
1291
1292	out_undo:
1293	/* If I have an error cleanup all namespaces I initialized */
1294	list_del(entry: &ops->list);
1295	ops_undo_single(ops, net_exit_list: &net_exit_list);
1296	return error;
1297	}
1298
1299	static void __unregister_pernet_operations(struct pernet_operations *ops)
1300	{
1301	LIST_HEAD(net_exit_list);
1302	struct net *net;
1303
1304	/* See comment in __register_pernet_operations() */
1305	for_each_net(net)
1306	list_add_tail(new: &net->exit_list, head: &net_exit_list);
1307
1308	list_del(entry: &ops->list);
1309	ops_undo_single(ops, net_exit_list: &net_exit_list);
1310	}
1311
1312	#else
1313
1314	static int __register_pernet_operations(struct list_head *list,
1315	struct pernet_operations *ops)
1316	{
1317	if (!init_net_initialized) {
1318	list_add_tail(&ops->list, list);
1319	return 0;
1320	}
1321
1322	return ops_init(ops, &init_net);
1323	}
1324
1325	static void __unregister_pernet_operations(struct pernet_operations *ops)
1326	{
1327	if (!init_net_initialized) {
1328	list_del(&ops->list);
1329	} else {
1330	LIST_HEAD(net_exit_list);
1331
1332	list_add(&init_net.exit_list, &net_exit_list);
1333	ops_undo_single(ops, &net_exit_list);
1334	}
1335	}
1336
1337	#endif /* CONFIG_NET_NS */
1338
1339	static DEFINE_IDA(net_generic_ids);
1340
1341	static int register_pernet_operations(struct list_head *list,
1342	struct pernet_operations *ops)
1343	{
1344	int error;
1345
1346	if (WARN_ON(!!ops->id ^ !!ops->size))
1347	return -EINVAL;
1348
1349	if (ops->id) {
1350	error = ida_alloc_min(ida: &net_generic_ids, MIN_PERNET_OPS_ID,
1351	GFP_KERNEL);
1352	if (error < 0)
1353	return error;
1354	*ops->id = error;
1355	/* This does not require READ_ONCE as writers already hold
1356	* pernet_ops_rwsem. But WRITE_ONCE is needed to protect
1357	* net_alloc_generic.
1358	*/
1359	WRITE_ONCE(max_gen_ptrs, max(max_gen_ptrs, *ops->id + 1));
1360	}
1361	error = __register_pernet_operations(list, ops);
1362	if (error) {
1363	rcu_barrier();
1364	if (ops->id)
1365	ida_free(&net_generic_ids, id: *ops->id);
1366	}
1367
1368	return error;
1369	}
1370
1371	static void unregister_pernet_operations(struct pernet_operations *ops)
1372	{
1373	__unregister_pernet_operations(ops);
1374	rcu_barrier();
1375	if (ops->id)
1376	ida_free(&net_generic_ids, id: *ops->id);
1377	}
1378
1379	/**
1380	* register_pernet_subsys - register a network namespace subsystem
1381	* @ops: pernet operations structure for the subsystem
1382	*
1383	* Register a subsystem which has init and exit functions
1384	* that are called when network namespaces are created and
1385	* destroyed respectively.
1386	*
1387	* When registered all network namespace init functions are
1388	* called for every existing network namespace. Allowing kernel
1389	* modules to have a race free view of the set of network namespaces.
1390	*
1391	* When a new network namespace is created all of the init
1392	* methods are called in the order in which they were registered.
1393	*
1394	* When a network namespace is destroyed all of the exit methods
1395	* are called in the reverse of the order with which they were
1396	* registered.
1397	*/
1398	int register_pernet_subsys(struct pernet_operations *ops)
1399	{
1400	int error;
1401	down_write(sem: &pernet_ops_rwsem);
1402	error = register_pernet_operations(list: first_device, ops);
1403	up_write(sem: &pernet_ops_rwsem);
1404	return error;
1405	}
1406	EXPORT_SYMBOL_GPL(register_pernet_subsys);
1407
1408	/**
1409	* unregister_pernet_subsys - unregister a network namespace subsystem
1410	* @ops: pernet operations structure to manipulate
1411	*
1412	* Remove the pernet operations structure from the list to be
1413	* used when network namespaces are created or destroyed. In
1414	* addition run the exit method for all existing network
1415	* namespaces.
1416	*/
1417	void unregister_pernet_subsys(struct pernet_operations *ops)
1418	{
1419	down_write(sem: &pernet_ops_rwsem);
1420	unregister_pernet_operations(ops);
1421	up_write(sem: &pernet_ops_rwsem);
1422	}
1423	EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
1424
1425	/**
1426	* register_pernet_device - register a network namespace device
1427	* @ops: pernet operations structure for the subsystem
1428	*
1429	* Register a device which has init and exit functions
1430	* that are called when network namespaces are created and
1431	* destroyed respectively.
1432	*
1433	* When registered all network namespace init functions are
1434	* called for every existing network namespace. Allowing kernel
1435	* modules to have a race free view of the set of network namespaces.
1436	*
1437	* When a new network namespace is created all of the init
1438	* methods are called in the order in which they were registered.
1439	*
1440	* When a network namespace is destroyed all of the exit methods
1441	* are called in the reverse of the order with which they were
1442	* registered.
1443	*/
1444	int register_pernet_device(struct pernet_operations *ops)
1445	{
1446	int error;
1447	down_write(sem: &pernet_ops_rwsem);
1448	error = register_pernet_operations(list: &pernet_list, ops);
1449	if (!error && (first_device == &pernet_list))
1450	first_device = &ops->list;
1451	up_write(sem: &pernet_ops_rwsem);
1452	return error;
1453	}
1454	EXPORT_SYMBOL_GPL(register_pernet_device);
1455
1456	/**
1457	* unregister_pernet_device - unregister a network namespace netdevice
1458	* @ops: pernet operations structure to manipulate
1459	*
1460	* Remove the pernet operations structure from the list to be
1461	* used when network namespaces are created or destroyed. In
1462	* addition run the exit method for all existing network
1463	* namespaces.
1464	*/
1465	void unregister_pernet_device(struct pernet_operations *ops)
1466	{
1467	down_write(sem: &pernet_ops_rwsem);
1468	if (&ops->list == first_device)
1469	first_device = first_device->next;
1470	unregister_pernet_operations(ops);
1471	up_write(sem: &pernet_ops_rwsem);
1472	}
1473	EXPORT_SYMBOL_GPL(unregister_pernet_device);
1474
1475	#ifdef CONFIG_NET_NS
1476	static struct ns_common *netns_get(struct task_struct *task)
1477	{
1478	struct net *net = NULL;
1479	struct nsproxy *nsproxy;
1480
1481	task_lock(p: task);
1482	nsproxy = task->nsproxy;
1483	if (nsproxy)
1484	net = get_net(net: nsproxy->net_ns);
1485	task_unlock(p: task);
1486
1487	return net ? &net->ns : NULL;
1488	}
1489
1490	static inline struct net *to_net_ns(struct ns_common *ns)
1491	{
1492	return container_of(ns, struct net, ns);
1493	}
1494
1495	static void netns_put(struct ns_common *ns)
1496	{
1497	put_net(net: to_net_ns(ns));
1498	}
1499
1500	static int netns_install(struct nsset *nsset, struct ns_common *ns)
1501	{
1502	struct nsproxy *nsproxy = nsset->nsproxy;
1503	struct net *net = to_net_ns(ns);
1504
1505	if (!ns_capable(ns: net->user_ns, CAP_SYS_ADMIN) ||
1506	!ns_capable(ns: nsset->cred->user_ns, CAP_SYS_ADMIN))
1507	return -EPERM;
1508
1509	put_net(net: nsproxy->net_ns);
1510	nsproxy->net_ns = get_net(net);
1511	return 0;
1512	}
1513
1514	static struct user_namespace *netns_owner(struct ns_common *ns)
1515	{
1516	return to_net_ns(ns)->user_ns;
1517	}
1518
1519	const struct proc_ns_operations netns_operations = {
1520	.name = "net",
1521	.type = CLONE_NEWNET,
1522	.get = netns_get,
1523	.put = netns_put,
1524	.install = netns_install,
1525	.owner = netns_owner,
1526	};
1527	#endif
1528