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