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 #define UCOUNT_ENTRY(name) \ 62 { \ 63 .procname = name, \ 64 .maxlen = sizeof(int), \ 65 .mode = 0644, \ 66 .proc_handler = proc_dointvec_minmax, \ 67 .extra1 = SYSCTL_ZERO, \ 68 .extra2 = SYSCTL_INT_MAX, \ 69 } 70 static struct ctl_table user_table[] = { 71 UCOUNT_ENTRY("max_user_namespaces"), 72 UCOUNT_ENTRY("max_pid_namespaces"), 73 UCOUNT_ENTRY("max_uts_namespaces"), 74 UCOUNT_ENTRY("max_ipc_namespaces"), 75 UCOUNT_ENTRY("max_net_namespaces"), 76 UCOUNT_ENTRY("max_mnt_namespaces"), 77 UCOUNT_ENTRY("max_cgroup_namespaces"), 78 UCOUNT_ENTRY("max_time_namespaces"), 79 #ifdef CONFIG_INOTIFY_USER 80 UCOUNT_ENTRY("max_inotify_instances"), 81 UCOUNT_ENTRY("max_inotify_watches"), 82 #endif 83 { }, 84 { }, 85 { }, 86 { }, 87 { } 88 }; 89 #endif /* CONFIG_SYSCTL */ 90 91 bool setup_userns_sysctls(struct user_namespace *ns) 92 { 93 #ifdef CONFIG_SYSCTL 94 struct ctl_table *tbl; 95 96 BUILD_BUG_ON(ARRAY_SIZE(user_table) != UCOUNT_COUNTS + 1); 97 setup_sysctl_set(&ns->set, &set_root, set_is_seen); 98 tbl = kmemdup(user_table, sizeof(user_table), GFP_KERNEL); 99 if (tbl) { 100 int i; 101 for (i = 0; i < UCOUNT_COUNTS; i++) { 102 tbl[i].data = &ns->ucount_max[i]; 103 } 104 ns->sysctls = __register_sysctl_table(&ns->set, "user", tbl); 105 } 106 if (!ns->sysctls) { 107 kfree(tbl); 108 retire_sysctl_set(&ns->set); 109 return false; 110 } 111 #endif 112 return true; 113 } 114 115 void retire_userns_sysctls(struct user_namespace *ns) 116 { 117 #ifdef CONFIG_SYSCTL 118 struct ctl_table *tbl; 119 120 tbl = ns->sysctls->ctl_table_arg; 121 unregister_sysctl_table(ns->sysctls); 122 retire_sysctl_set(&ns->set); 123 kfree(tbl); 124 #endif 125 } 126 127 static struct ucounts *find_ucounts(struct user_namespace *ns, kuid_t uid, struct hlist_head *hashent) 128 { 129 struct ucounts *ucounts; 130 131 hlist_for_each_entry(ucounts, hashent, node) { 132 if (uid_eq(ucounts->uid, uid) && (ucounts->ns == ns)) 133 return ucounts; 134 } 135 return NULL; 136 } 137 138 static void hlist_add_ucounts(struct ucounts *ucounts) 139 { 140 struct hlist_head *hashent = ucounts_hashentry(ucounts->ns, ucounts->uid); 141 spin_lock_irq(&ucounts_lock); 142 hlist_add_head(&ucounts->node, hashent); 143 spin_unlock_irq(&ucounts_lock); 144 } 145 146 struct ucounts *get_ucounts(struct ucounts *ucounts) 147 { 148 if (ucounts && atomic_add_negative(1, &ucounts->count)) { 149 put_ucounts(ucounts); 150 ucounts = NULL; 151 } 152 return ucounts; 153 } 154 155 struct ucounts *alloc_ucounts(struct user_namespace *ns, kuid_t uid) 156 { 157 struct hlist_head *hashent = ucounts_hashentry(ns, uid); 158 struct ucounts *ucounts, *new; 159 160 spin_lock_irq(&ucounts_lock); 161 ucounts = find_ucounts(ns, uid, hashent); 162 if (!ucounts) { 163 spin_unlock_irq(&ucounts_lock); 164 165 new = kzalloc(sizeof(*new), GFP_KERNEL); 166 if (!new) 167 return NULL; 168 169 new->ns = ns; 170 new->uid = uid; 171 atomic_set(&new->count, 1); 172 173 spin_lock_irq(&ucounts_lock); 174 ucounts = find_ucounts(ns, uid, hashent); 175 if (ucounts) { 176 kfree(new); 177 } else { 178 hlist_add_head(&new->node, hashent); 179 spin_unlock_irq(&ucounts_lock); 180 return new; 181 } 182 } 183 spin_unlock_irq(&ucounts_lock); 184 ucounts = get_ucounts(ucounts); 185 return ucounts; 186 } 187 188 void put_ucounts(struct ucounts *ucounts) 189 { 190 unsigned long flags; 191 192 if (atomic_dec_and_test(&ucounts->count)) { 193 spin_lock_irqsave(&ucounts_lock, flags); 194 hlist_del_init(&ucounts->node); 195 spin_unlock_irqrestore(&ucounts_lock, flags); 196 kfree(ucounts); 197 } 198 } 199 200 static inline bool atomic_long_inc_below(atomic_long_t *v, int u) 201 { 202 long c, old; 203 c = atomic_long_read(v); 204 for (;;) { 205 if (unlikely(c >= u)) 206 return false; 207 old = atomic_long_cmpxchg(v, c, c+1); 208 if (likely(old == c)) 209 return true; 210 c = old; 211 } 212 } 213 214 struct ucounts *inc_ucount(struct user_namespace *ns, kuid_t uid, 215 enum ucount_type type) 216 { 217 struct ucounts *ucounts, *iter, *bad; 218 struct user_namespace *tns; 219 ucounts = alloc_ucounts(ns, uid); 220 for (iter = ucounts; iter; iter = tns->ucounts) { 221 long max; 222 tns = iter->ns; 223 max = READ_ONCE(tns->ucount_max[type]); 224 if (!atomic_long_inc_below(&iter->ucount[type], max)) 225 goto fail; 226 } 227 return ucounts; 228 fail: 229 bad = iter; 230 for (iter = ucounts; iter != bad; iter = iter->ns->ucounts) 231 atomic_long_dec(&iter->ucount[type]); 232 233 put_ucounts(ucounts); 234 return NULL; 235 } 236 237 void dec_ucount(struct ucounts *ucounts, enum ucount_type type) 238 { 239 struct ucounts *iter; 240 for (iter = ucounts; iter; iter = iter->ns->ucounts) { 241 long dec = atomic_long_dec_if_positive(&iter->ucount[type]); 242 WARN_ON_ONCE(dec < 0); 243 } 244 put_ucounts(ucounts); 245 } 246 247 long inc_rlimit_ucounts(struct ucounts *ucounts, enum ucount_type type, long v) 248 { 249 struct ucounts *iter; 250 long ret = 0; 251 252 for (iter = ucounts; iter; iter = iter->ns->ucounts) { 253 long max = READ_ONCE(iter->ns->ucount_max[type]); 254 long new = atomic_long_add_return(v, &iter->ucount[type]); 255 if (new < 0 || new > max) 256 ret = LONG_MAX; 257 else if (iter == ucounts) 258 ret = new; 259 } 260 return ret; 261 } 262 263 bool dec_rlimit_ucounts(struct ucounts *ucounts, enum ucount_type type, long v) 264 { 265 struct ucounts *iter; 266 long new; 267 for (iter = ucounts; iter; iter = iter->ns->ucounts) { 268 long dec = atomic_long_add_return(-v, &iter->ucount[type]); 269 WARN_ON_ONCE(dec < 0); 270 if (iter == ucounts) 271 new = dec; 272 } 273 return (new == 0); 274 } 275 276 bool is_ucounts_overlimit(struct ucounts *ucounts, enum ucount_type type, unsigned long max) 277 { 278 struct ucounts *iter; 279 if (get_ucounts_value(ucounts, type) > max) 280 return true; 281 for (iter = ucounts; iter; iter = iter->ns->ucounts) { 282 max = READ_ONCE(iter->ns->ucount_max[type]); 283 if (get_ucounts_value(iter, type) > max) 284 return true; 285 } 286 return false; 287 } 288 289 static __init int user_namespace_sysctl_init(void) 290 { 291 #ifdef CONFIG_SYSCTL 292 static struct ctl_table_header *user_header; 293 static struct ctl_table empty[1]; 294 /* 295 * It is necessary to register the user directory in the 296 * default set so that registrations in the child sets work 297 * properly. 298 */ 299 user_header = register_sysctl("user", empty); 300 kmemleak_ignore(user_header); 301 BUG_ON(!user_header); 302 BUG_ON(!setup_userns_sysctls(&init_user_ns)); 303 #endif 304 hlist_add_ucounts(&init_ucounts); 305 inc_rlimit_ucounts(&init_ucounts, UCOUNT_RLIMIT_NPROC, 1); 306 return 0; 307 } 308 subsys_initcall(user_namespace_sysctl_init); 309