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 {
40 .extent[0] = {
41 .first = 0,
42 .lower_first = 0,
43 .count = 4294967295U,
44 },
45 .nr_extents = 1,
46 },
47 },
48 .gid_map = {
49 {
50 .extent[0] = {
51 .first = 0,
52 .lower_first = 0,
53 .count = 4294967295U,
54 },
55 .nr_extents = 1,
56 },
57 },
58 .projid_map = {
59 {
60 .extent[0] = {
61 .first = 0,
62 .lower_first = 0,
63 .count = 4294967295U,
64 },
65 .nr_extents = 1,
66 },
67 },
68 .ns.ns_type = ns_common_type(&init_user_ns),
69 .ns.__ns_ref = REFCOUNT_INIT(3),
70 .owner = GLOBAL_ROOT_UID,
71 .group = GLOBAL_ROOT_GID,
72 .ns.inum = ns_init_inum(&init_user_ns),
73#ifdef CONFIG_USER_NS
74 .ns.ops = &userns_operations,
75#endif
76 .flags = USERNS_INIT_FLAGS,
77#ifdef CONFIG_KEYS
78 .keyring_name_list = LIST_HEAD_INIT(init_user_ns.keyring_name_list),
79 .keyring_sem = __RWSEM_INITIALIZER(init_user_ns.keyring_sem),
80#endif
81#if IS_ENABLED(CONFIG_BINFMT_MISC)
82 .binfmt_misc = &init_binfmt_misc,
83#endif
84};
85EXPORT_SYMBOL_GPL(init_user_ns);
86
87/*
88 * UID task count cache, to get fast user lookup in "alloc_uid"
89 * when changing user ID's (ie setuid() and friends).
90 */
91
92#define UIDHASH_BITS (IS_ENABLED(CONFIG_BASE_SMALL) ? 3 : 7)
93#define UIDHASH_SZ (1 << UIDHASH_BITS)
94#define UIDHASH_MASK (UIDHASH_SZ - 1)
95#define __uidhashfn(uid) (((uid >> UIDHASH_BITS) + uid) & UIDHASH_MASK)
96#define uidhashentry(uid) (uidhash_table + __uidhashfn((__kuid_val(uid))))
97
98static struct kmem_cache *uid_cachep;
99static struct hlist_head uidhash_table[UIDHASH_SZ];
100
101/*
102 * The uidhash_lock is mostly taken from process context, but it is
103 * occasionally also taken from softirq/tasklet context, when
104 * task-structs get RCU-freed. Hence all locking must be softirq-safe.
105 * But free_uid() is also called with local interrupts disabled, and running
106 * local_bh_enable() with local interrupts disabled is an error - we'll run
107 * softirq callbacks, and they can unconditionally enable interrupts, and
108 * the caller of free_uid() didn't expect that..
109 */
110static DEFINE_SPINLOCK(uidhash_lock);
111
112/* root_user.__count is 1, for init task cred */
113struct user_struct root_user = {
114 .__count = REFCOUNT_INIT(1),
115 .uid = GLOBAL_ROOT_UID,
116 .ratelimit = RATELIMIT_STATE_INIT(root_user.ratelimit, 0, 0),
117};
118
119/*
120 * These routines must be called with the uidhash spinlock held!
121 */
122static void uid_hash_insert(struct user_struct *up, struct hlist_head *hashent)
123{
124 hlist_add_head(n: &up->uidhash_node, h: hashent);
125}
126
127static void uid_hash_remove(struct user_struct *up)
128{
129 hlist_del_init(n: &up->uidhash_node);
130}
131
132static struct user_struct *uid_hash_find(kuid_t uid, struct hlist_head *hashent)
133{
134 struct user_struct *user;
135
136 hlist_for_each_entry(user, hashent, uidhash_node) {
137 if (uid_eq(left: user->uid, right: uid)) {
138 refcount_inc(r: &user->__count);
139 return user;
140 }
141 }
142
143 return NULL;
144}
145
146static int user_epoll_alloc(struct user_struct *up)
147{
148#ifdef CONFIG_EPOLL
149 return percpu_counter_init(&up->epoll_watches, 0, GFP_KERNEL);
150#else
151 return 0;
152#endif
153}
154
155static void user_epoll_free(struct user_struct *up)
156{
157#ifdef CONFIG_EPOLL
158 percpu_counter_destroy(fbc: &up->epoll_watches);
159#endif
160}
161
162/* IRQs are disabled and uidhash_lock is held upon function entry.
163 * IRQ state (as stored in flags) is restored and uidhash_lock released
164 * upon function exit.
165 */
166static void free_user(struct user_struct *up, unsigned long flags)
167 __releases(&uidhash_lock)
168{
169 uid_hash_remove(up);
170 spin_unlock_irqrestore(lock: &uidhash_lock, flags);
171 user_epoll_free(up);
172 kmem_cache_free(s: uid_cachep, objp: up);
173}
174
175/*
176 * Locate the user_struct for the passed UID. If found, take a ref on it. The
177 * caller must undo that ref with free_uid().
178 *
179 * If the user_struct could not be found, return NULL.
180 */
181struct user_struct *find_user(kuid_t uid)
182{
183 struct user_struct *ret;
184 unsigned long flags;
185
186 spin_lock_irqsave(&uidhash_lock, flags);
187 ret = uid_hash_find(uid, uidhashentry(uid));
188 spin_unlock_irqrestore(lock: &uidhash_lock, flags);
189 return ret;
190}
191
192void free_uid(struct user_struct *up)
193{
194 unsigned long flags;
195
196 if (!up)
197 return;
198
199 if (refcount_dec_and_lock_irqsave(r: &up->__count, lock: &uidhash_lock, flags: &flags))
200 free_user(up, flags);
201}
202EXPORT_SYMBOL_GPL(free_uid);
203
204struct user_struct *alloc_uid(kuid_t uid)
205{
206 struct hlist_head *hashent = uidhashentry(uid);
207 struct user_struct *up, *new;
208
209 spin_lock_irq(lock: &uidhash_lock);
210 up = uid_hash_find(uid, hashent);
211 spin_unlock_irq(lock: &uidhash_lock);
212
213 if (!up) {
214 new = kmem_cache_zalloc(uid_cachep, GFP_KERNEL);
215 if (!new)
216 return NULL;
217
218 new->uid = uid;
219 refcount_set(r: &new->__count, n: 1);
220 if (user_epoll_alloc(up: new)) {
221 kmem_cache_free(s: uid_cachep, objp: new);
222 return NULL;
223 }
224 ratelimit_state_init(rs: &new->ratelimit, HZ, burst: 100);
225 ratelimit_set_flags(rs: &new->ratelimit, RATELIMIT_MSG_ON_RELEASE);
226
227 /*
228 * Before adding this, check whether we raced
229 * on adding the same user already..
230 */
231 spin_lock_irq(lock: &uidhash_lock);
232 up = uid_hash_find(uid, hashent);
233 if (up) {
234 user_epoll_free(up: new);
235 kmem_cache_free(s: uid_cachep, objp: new);
236 } else {
237 uid_hash_insert(up: new, hashent);
238 up = new;
239 }
240 spin_unlock_irq(lock: &uidhash_lock);
241 }
242
243 return up;
244}
245
246static int __init uid_cache_init(void)
247{
248 int n;
249
250 uid_cachep = kmem_cache_create("uid_cache", sizeof(struct user_struct),
251 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
252
253 for(n = 0; n < UIDHASH_SZ; ++n)
254 INIT_HLIST_HEAD(uidhash_table + n);
255
256 if (user_epoll_alloc(up: &root_user))
257 panic(fmt: "root_user epoll percpu counter alloc failed");
258
259 /* Insert the root user immediately (init already runs as root) */
260 spin_lock_irq(lock: &uidhash_lock);
261 uid_hash_insert(up: &root_user, uidhashentry(GLOBAL_ROOT_UID));
262 spin_unlock_irq(&uidhash_lock);
263
264 return 0;
265}
266subsys_initcall(uid_cache_init);
267