1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * The "user cache".
4 *
5 * (C) Copyright 1991-2000 Linus Torvalds
6 *
7 * We have a per-user structure to keep track of how many
8 * processes, files etc the user has claimed, in order to be
9 * able to have per-user limits for system resources.
10 */
11
12#include <linux/init.h>
13#include <linux/sched.h>
14#include <linux/slab.h>
15#include <linux/bitops.h>
16#include <linux/key.h>
17#include <linux/sched/user.h>
18#include <linux/interrupt.h>
19#include <linux/export.h>
20#include <linux/user_namespace.h>
21#include <linux/binfmts.h>
22#include <linux/proc_ns.h>
23
24#if IS_ENABLED(CONFIG_BINFMT_MISC)
25struct binfmt_misc init_binfmt_misc = {
26	.entries = LIST_HEAD_INIT(init_binfmt_misc.entries),
27	.enabled = true,
28	.entries_lock = __RW_LOCK_UNLOCKED(init_binfmt_misc.entries_lock),
29};
30EXPORT_SYMBOL_GPL(init_binfmt_misc);
31#endif
32
33/*
34 * userns count is 1 for root user, 1 for init_uts_ns,
35 * and 1 for... ?
36 */
37struct user_namespace init_user_ns = {
38	.uid_map = {
39		.nr_extents = 1,
40		{
41			.extent[0] = {
42				.first = 0,
43				.lower_first = 0,
44				.count = 4294967295U,
45			},
46		},
47	},
48	.gid_map = {
49		.nr_extents = 1,
50		{
51			.extent[0] = {
52				.first = 0,
53				.lower_first = 0,
54				.count = 4294967295U,
55			},
56		},
57	},
58	.projid_map = {
59		.nr_extents = 1,
60		{
61			.extent[0] = {
62				.first = 0,
63				.lower_first = 0,
64				.count = 4294967295U,
65			},
66		},
67	},
68	.ns.count = REFCOUNT_INIT(3),
69	.owner = GLOBAL_ROOT_UID,
70	.group = GLOBAL_ROOT_GID,
71	.ns.inum = PROC_USER_INIT_INO,
72#ifdef CONFIG_USER_NS
73	.ns.ops = &userns_operations,
74#endif
75	.flags = USERNS_INIT_FLAGS,
76#ifdef CONFIG_KEYS
77	.keyring_name_list = LIST_HEAD_INIT(init_user_ns.keyring_name_list),
78	.keyring_sem = __RWSEM_INITIALIZER(init_user_ns.keyring_sem),
79#endif
80#if IS_ENABLED(CONFIG_BINFMT_MISC)
81	.binfmt_misc = &init_binfmt_misc,
82#endif
83};
84EXPORT_SYMBOL_GPL(init_user_ns);
85
86/*
87 * UID task count cache, to get fast user lookup in "alloc_uid"
88 * when changing user ID's (ie setuid() and friends).
89 */
90
91#define UIDHASH_BITS	(IS_ENABLED(CONFIG_BASE_SMALL) ? 3 : 7)
92#define UIDHASH_SZ	(1 << UIDHASH_BITS)
93#define UIDHASH_MASK		(UIDHASH_SZ - 1)
94#define __uidhashfn(uid)	(((uid >> UIDHASH_BITS) + uid) & UIDHASH_MASK)
95#define uidhashentry(uid)	(uidhash_table + __uidhashfn((__kuid_val(uid))))
96
97static struct kmem_cache *uid_cachep;
98static struct hlist_head uidhash_table[UIDHASH_SZ];
99
100/*
101 * The uidhash_lock is mostly taken from process context, but it is
102 * occasionally also taken from softirq/tasklet context, when
103 * task-structs get RCU-freed. Hence all locking must be softirq-safe.
104 * But free_uid() is also called with local interrupts disabled, and running
105 * local_bh_enable() with local interrupts disabled is an error - we'll run
106 * softirq callbacks, and they can unconditionally enable interrupts, and
107 * the caller of free_uid() didn't expect that..
108 */
109static DEFINE_SPINLOCK(uidhash_lock);
110
111/* root_user.__count is 1, for init task cred */
112struct user_struct root_user = {
113	.__count	= REFCOUNT_INIT(1),
114	.uid		= GLOBAL_ROOT_UID,
115	.ratelimit	= RATELIMIT_STATE_INIT(root_user.ratelimit, 0, 0),
116};
117
118/*
119 * These routines must be called with the uidhash spinlock held!
120 */
121static void uid_hash_insert(struct user_struct *up, struct hlist_head *hashent)
122{
123	hlist_add_head(&up->uidhash_node, hashent);
124}
125
126static void uid_hash_remove(struct user_struct *up)
127{
128	hlist_del_init(&up->uidhash_node);
129}
130
131static struct user_struct *uid_hash_find(kuid_t uid, struct hlist_head *hashent)
132{
133	struct user_struct *user;
134
135	hlist_for_each_entry(user, hashent, uidhash_node) {
136		if (uid_eq(user->uid, uid)) {
137			refcount_inc(&user->__count);
138			return user;
139		}
140	}
141
142	return NULL;
143}
144
145static int user_epoll_alloc(struct user_struct *up)
146{
147#ifdef CONFIG_EPOLL
148	return percpu_counter_init(&up->epoll_watches, 0, GFP_KERNEL);
149#else
150	return 0;
151#endif
152}
153
154static void user_epoll_free(struct user_struct *up)
155{
156#ifdef CONFIG_EPOLL
157	percpu_counter_destroy(&up->epoll_watches);
158#endif
159}
160
161/* IRQs are disabled and uidhash_lock is held upon function entry.
162 * IRQ state (as stored in flags) is restored and uidhash_lock released
163 * upon function exit.
164 */
165static void free_user(struct user_struct *up, unsigned long flags)
166	__releases(&uidhash_lock)
167{
168	uid_hash_remove(up);
169	spin_unlock_irqrestore(&uidhash_lock, flags);
170	user_epoll_free(up);
171	kmem_cache_free(uid_cachep, up);
172}
173
174/*
175 * Locate the user_struct for the passed UID.  If found, take a ref on it.  The
176 * caller must undo that ref with free_uid().
177 *
178 * If the user_struct could not be found, return NULL.
179 */
180struct user_struct *find_user(kuid_t uid)
181{
182	struct user_struct *ret;
183	unsigned long flags;
184
185	spin_lock_irqsave(&uidhash_lock, flags);
186	ret = uid_hash_find(uid, uidhashentry(uid));
187	spin_unlock_irqrestore(&uidhash_lock, flags);
188	return ret;
189}
190
191void free_uid(struct user_struct *up)
192{
193	unsigned long flags;
194
195	if (!up)
196		return;
197
198	if (refcount_dec_and_lock_irqsave(&up->__count, &uidhash_lock, &flags))
199		free_user(up, flags);
200}
201EXPORT_SYMBOL_GPL(free_uid);
202
203struct user_struct *alloc_uid(kuid_t uid)
204{
205	struct hlist_head *hashent = uidhashentry(uid);
206	struct user_struct *up, *new;
207
208	spin_lock_irq(&uidhash_lock);
209	up = uid_hash_find(uid, hashent);
210	spin_unlock_irq(&uidhash_lock);
211
212	if (!up) {
213		new = kmem_cache_zalloc(uid_cachep, GFP_KERNEL);
214		if (!new)
215			return NULL;
216
217		new->uid = uid;
218		refcount_set(&new->__count, 1);
219		if (user_epoll_alloc(new)) {
220			kmem_cache_free(uid_cachep, new);
221			return NULL;
222		}
223		ratelimit_state_init(&new->ratelimit, HZ, 100);
224		ratelimit_set_flags(&new->ratelimit, RATELIMIT_MSG_ON_RELEASE);
225
226		/*
227		 * Before adding this, check whether we raced
228		 * on adding the same user already..
229		 */
230		spin_lock_irq(&uidhash_lock);
231		up = uid_hash_find(uid, hashent);
232		if (up) {
233			user_epoll_free(new);
234			kmem_cache_free(uid_cachep, new);
235		} else {
236			uid_hash_insert(new, hashent);
237			up = new;
238		}
239		spin_unlock_irq(&uidhash_lock);
240	}
241
242	return up;
243}
244
245static int __init uid_cache_init(void)
246{
247	int n;
248
249	uid_cachep = kmem_cache_create("uid_cache", sizeof(struct user_struct),
250			0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
251
252	for(n = 0; n < UIDHASH_SZ; ++n)
253		INIT_HLIST_HEAD(uidhash_table + n);
254
255	if (user_epoll_alloc(&root_user))
256		panic("root_user epoll percpu counter alloc failed");
257
258	/* Insert the root user immediately (init already runs as root) */
259	spin_lock_irq(&uidhash_lock);
260	uid_hash_insert(&root_user, uidhashentry(GLOBAL_ROOT_UID));
261	spin_unlock_irq(&uidhash_lock);
262
263	return 0;
264}
265subsys_initcall(uid_cache_init);
266