1 // SPDX-License-Identifier: GPL-2.0-only 2 3 #include <linux/stat.h> 4 #include <linux/sysctl.h> 5 #include <linux/slab.h> 6 #include <linux/cred.h> 7 #include <linux/hash.h> 8 #include <linux/kmemleak.h> 9 #include <linux/user_namespace.h> 10 11 struct ucounts init_ucounts = { 12 .ns = &init_user_ns, 13 .uid = GLOBAL_ROOT_UID, 14 .count = ATOMIC_INIT(1), 15 }; 16 17 #define UCOUNTS_HASHTABLE_BITS 10 18 static struct hlist_head ucounts_hashtable[(1 << UCOUNTS_HASHTABLE_BITS)]; 19 static DEFINE_SPINLOCK(ucounts_lock); 20 21 #define ucounts_hashfn(ns, uid) \ 22 hash_long((unsigned long)__kuid_val(uid) + (unsigned long)(ns), \ 23 UCOUNTS_HASHTABLE_BITS) 24 #define ucounts_hashentry(ns, uid) \ 25 (ucounts_hashtable + ucounts_hashfn(ns, uid)) 26 27 28 #ifdef CONFIG_SYSCTL 29 static struct ctl_table_set * 30 set_lookup(struct ctl_table_root *root) 31 { 32 return ¤t_user_ns()->set; 33 } 34 35 static int set_is_seen(struct ctl_table_set *set) 36 { 37 return ¤t_user_ns()->set == set; 38 } 39 40 static int set_permissions(struct ctl_table_header *head, 41 struct ctl_table *table) 42 { 43 struct user_namespace *user_ns = 44 container_of(head->set, struct user_namespace, set); 45 int mode; 46 47 /* Allow users with CAP_SYS_RESOURCE unrestrained access */ 48 if (ns_capable(user_ns, CAP_SYS_RESOURCE)) 49 mode = (table->mode & S_IRWXU) >> 6; 50 else 51 /* Allow all others at most read-only access */ 52 mode = table->mode & S_IROTH; 53 return (mode << 6) | (mode << 3) | mode; 54 } 55 56 static struct ctl_table_root set_root = { 57 .lookup = set_lookup, 58 .permissions = set_permissions, 59 }; 60 61 static long ue_zero = 0; 62 static long ue_int_max = INT_MAX; 63 64 #define UCOUNT_ENTRY(name) \ 65 { \ 66 .procname = name, \ 67 .maxlen = sizeof(long), \ 68 .mode = 0644, \ 69 .proc_handler = proc_doulongvec_minmax, \ 70 .extra1 = &ue_zero, \ 71 .extra2 = &ue_int_max, \ 72 } 73 static struct ctl_table user_table[] = { 74 UCOUNT_ENTRY("max_user_namespaces"), 75 UCOUNT_ENTRY("max_pid_namespaces"), 76 UCOUNT_ENTRY("max_uts_namespaces"), 77 UCOUNT_ENTRY("max_ipc_namespaces"), 78 UCOUNT_ENTRY("max_net_namespaces"), 79 UCOUNT_ENTRY("max_mnt_namespaces"), 80 UCOUNT_ENTRY("max_cgroup_namespaces"), 81 UCOUNT_ENTRY("max_time_namespaces"), 82 #ifdef CONFIG_INOTIFY_USER 83 UCOUNT_ENTRY("max_inotify_instances"), 84 UCOUNT_ENTRY("max_inotify_watches"), 85 #endif 86 #ifdef CONFIG_FANOTIFY 87 UCOUNT_ENTRY("max_fanotify_groups"), 88 UCOUNT_ENTRY("max_fanotify_marks"), 89 #endif 90 { }, 91 { }, 92 { }, 93 { }, 94 { } 95 }; 96 #endif /* CONFIG_SYSCTL */ 97 98 bool setup_userns_sysctls(struct user_namespace *ns) 99 { 100 #ifdef CONFIG_SYSCTL 101 struct ctl_table *tbl; 102 103 BUILD_BUG_ON(ARRAY_SIZE(user_table) != UCOUNT_COUNTS + 1); 104 setup_sysctl_set(&ns->set, &set_root, set_is_seen); 105 tbl = kmemdup(user_table, sizeof(user_table), GFP_KERNEL); 106 if (tbl) { 107 int i; 108 for (i = 0; i < UCOUNT_COUNTS; i++) { 109 tbl[i].data = &ns->ucount_max[i]; 110 } 111 ns->sysctls = __register_sysctl_table(&ns->set, "user", tbl); 112 } 113 if (!ns->sysctls) { 114 kfree(tbl); 115 retire_sysctl_set(&ns->set); 116 return false; 117 } 118 #endif 119 return true; 120 } 121 122 void retire_userns_sysctls(struct user_namespace *ns) 123 { 124 #ifdef CONFIG_SYSCTL 125 struct ctl_table *tbl; 126 127 tbl = ns->sysctls->ctl_table_arg; 128 unregister_sysctl_table(ns->sysctls); 129 retire_sysctl_set(&ns->set); 130 kfree(tbl); 131 #endif 132 } 133 134 static struct ucounts *find_ucounts(struct user_namespace *ns, kuid_t uid, struct hlist_head *hashent) 135 { 136 struct ucounts *ucounts; 137 138 hlist_for_each_entry(ucounts, hashent, node) { 139 if (uid_eq(ucounts->uid, uid) && (ucounts->ns == ns)) 140 return ucounts; 141 } 142 return NULL; 143 } 144 145 static void hlist_add_ucounts(struct ucounts *ucounts) 146 { 147 struct hlist_head *hashent = ucounts_hashentry(ucounts->ns, ucounts->uid); 148 spin_lock_irq(&ucounts_lock); 149 hlist_add_head(&ucounts->node, hashent); 150 spin_unlock_irq(&ucounts_lock); 151 } 152 153 static inline bool get_ucounts_or_wrap(struct ucounts *ucounts) 154 { 155 /* Returns true on a successful get, false if the count wraps. */ 156 return !atomic_add_negative(1, &ucounts->count); 157 } 158 159 struct ucounts *get_ucounts(struct ucounts *ucounts) 160 { 161 if (!get_ucounts_or_wrap(ucounts)) { 162 put_ucounts(ucounts); 163 ucounts = NULL; 164 } 165 return ucounts; 166 } 167 168 struct ucounts *alloc_ucounts(struct user_namespace *ns, kuid_t uid) 169 { 170 struct hlist_head *hashent = ucounts_hashentry(ns, uid); 171 struct ucounts *ucounts, *new; 172 bool wrapped; 173 174 spin_lock_irq(&ucounts_lock); 175 ucounts = find_ucounts(ns, uid, hashent); 176 if (!ucounts) { 177 spin_unlock_irq(&ucounts_lock); 178 179 new = kzalloc(sizeof(*new), GFP_KERNEL); 180 if (!new) 181 return NULL; 182 183 new->ns = ns; 184 new->uid = uid; 185 atomic_set(&new->count, 1); 186 187 spin_lock_irq(&ucounts_lock); 188 ucounts = find_ucounts(ns, uid, hashent); 189 if (ucounts) { 190 kfree(new); 191 } else { 192 hlist_add_head(&new->node, hashent); 193 get_user_ns(new->ns); 194 spin_unlock_irq(&ucounts_lock); 195 return new; 196 } 197 } 198 wrapped = !get_ucounts_or_wrap(ucounts); 199 spin_unlock_irq(&ucounts_lock); 200 if (wrapped) { 201 put_ucounts(ucounts); 202 return NULL; 203 } 204 return ucounts; 205 } 206 207 void put_ucounts(struct ucounts *ucounts) 208 { 209 unsigned long flags; 210 211 if (atomic_dec_and_lock_irqsave(&ucounts->count, &ucounts_lock, flags)) { 212 hlist_del_init(&ucounts->node); 213 spin_unlock_irqrestore(&ucounts_lock, flags); 214 put_user_ns(ucounts->ns); 215 kfree(ucounts); 216 } 217 } 218 219 static inline bool atomic_long_inc_below(atomic_long_t *v, int u) 220 { 221 long c, old; 222 c = atomic_long_read(v); 223 for (;;) { 224 if (unlikely(c >= u)) 225 return false; 226 old = atomic_long_cmpxchg(v, c, c+1); 227 if (likely(old == c)) 228 return true; 229 c = old; 230 } 231 } 232 233 struct ucounts *inc_ucount(struct user_namespace *ns, kuid_t uid, 234 enum ucount_type type) 235 { 236 struct ucounts *ucounts, *iter, *bad; 237 struct user_namespace *tns; 238 ucounts = alloc_ucounts(ns, uid); 239 for (iter = ucounts; iter; iter = tns->ucounts) { 240 long max; 241 tns = iter->ns; 242 max = READ_ONCE(tns->ucount_max[type]); 243 if (!atomic_long_inc_below(&iter->ucount[type], max)) 244 goto fail; 245 } 246 return ucounts; 247 fail: 248 bad = iter; 249 for (iter = ucounts; iter != bad; iter = iter->ns->ucounts) 250 atomic_long_dec(&iter->ucount[type]); 251 252 put_ucounts(ucounts); 253 return NULL; 254 } 255 256 void dec_ucount(struct ucounts *ucounts, enum ucount_type type) 257 { 258 struct ucounts *iter; 259 for (iter = ucounts; iter; iter = iter->ns->ucounts) { 260 long dec = atomic_long_dec_if_positive(&iter->ucount[type]); 261 WARN_ON_ONCE(dec < 0); 262 } 263 put_ucounts(ucounts); 264 } 265 266 long inc_rlimit_ucounts(struct ucounts *ucounts, enum ucount_type type, long v) 267 { 268 struct ucounts *iter; 269 long max = LONG_MAX; 270 long ret = 0; 271 272 for (iter = ucounts; iter; iter = iter->ns->ucounts) { 273 long new = atomic_long_add_return(v, &iter->ucount[type]); 274 if (new < 0 || new > max) 275 ret = LONG_MAX; 276 else if (iter == ucounts) 277 ret = new; 278 max = READ_ONCE(iter->ns->ucount_max[type]); 279 } 280 return ret; 281 } 282 283 bool dec_rlimit_ucounts(struct ucounts *ucounts, enum ucount_type type, long v) 284 { 285 struct ucounts *iter; 286 long new = -1; /* Silence compiler warning */ 287 for (iter = ucounts; iter; iter = iter->ns->ucounts) { 288 long dec = atomic_long_sub_return(v, &iter->ucount[type]); 289 WARN_ON_ONCE(dec < 0); 290 if (iter == ucounts) 291 new = dec; 292 } 293 return (new == 0); 294 } 295 296 static void do_dec_rlimit_put_ucounts(struct ucounts *ucounts, 297 struct ucounts *last, enum ucount_type type) 298 { 299 struct ucounts *iter, *next; 300 for (iter = ucounts; iter != last; iter = next) { 301 long dec = atomic_long_sub_return(1, &iter->ucount[type]); 302 WARN_ON_ONCE(dec < 0); 303 next = iter->ns->ucounts; 304 if (dec == 0) 305 put_ucounts(iter); 306 } 307 } 308 309 void dec_rlimit_put_ucounts(struct ucounts *ucounts, enum ucount_type type) 310 { 311 do_dec_rlimit_put_ucounts(ucounts, NULL, type); 312 } 313 314 long inc_rlimit_get_ucounts(struct ucounts *ucounts, enum ucount_type type) 315 { 316 /* Caller must hold a reference to ucounts */ 317 struct ucounts *iter; 318 long max = LONG_MAX; 319 long dec, ret = 0; 320 321 for (iter = ucounts; iter; iter = iter->ns->ucounts) { 322 long new = atomic_long_add_return(1, &iter->ucount[type]); 323 if (new < 0 || new > max) 324 goto unwind; 325 if (iter == ucounts) 326 ret = new; 327 max = READ_ONCE(iter->ns->ucount_max[type]); 328 /* 329 * Grab an extra ucount reference for the caller when 330 * the rlimit count was previously 0. 331 */ 332 if (new != 1) 333 continue; 334 if (!get_ucounts(iter)) 335 goto dec_unwind; 336 } 337 return ret; 338 dec_unwind: 339 dec = atomic_long_sub_return(1, &iter->ucount[type]); 340 WARN_ON_ONCE(dec < 0); 341 unwind: 342 do_dec_rlimit_put_ucounts(ucounts, iter, type); 343 return 0; 344 } 345 346 bool is_ucounts_overlimit(struct ucounts *ucounts, enum ucount_type type, unsigned long rlimit) 347 { 348 struct ucounts *iter; 349 long max = rlimit; 350 if (rlimit > LONG_MAX) 351 max = LONG_MAX; 352 for (iter = ucounts; iter; iter = iter->ns->ucounts) { 353 long val = get_ucounts_value(iter, type); 354 if (val < 0 || val > max) 355 return true; 356 max = READ_ONCE(iter->ns->ucount_max[type]); 357 } 358 return false; 359 } 360 361 static __init int user_namespace_sysctl_init(void) 362 { 363 #ifdef CONFIG_SYSCTL 364 static struct ctl_table_header *user_header; 365 static struct ctl_table empty[1]; 366 /* 367 * It is necessary to register the user directory in the 368 * default set so that registrations in the child sets work 369 * properly. 370 */ 371 user_header = register_sysctl("user", empty); 372 kmemleak_ignore(user_header); 373 BUG_ON(!user_header); 374 BUG_ON(!setup_userns_sysctls(&init_user_ns)); 375 #endif 376 hlist_add_ucounts(&init_ucounts); 377 inc_rlimit_ucounts(&init_ucounts, UCOUNT_RLIMIT_NPROC, 1); 378 return 0; 379 } 380 subsys_initcall(user_namespace_sysctl_init); 381