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