kernel/ucount.c

   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 &current_user_ns()->set;
  33 }
  34
  35 static int set_is_seen(struct ctl_table_set *set)
  36 {
  37         return &current_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 struct ucounts *get_ucounts(struct ucounts *ucounts)
 154 {
 155         if (ucounts && atomic_add_negative(1, &ucounts->count)) {
 156                 put_ucounts(ucounts);
 157                 ucounts = NULL;
 158         }
 159         return ucounts;
 160 }
 161
 162 struct ucounts *alloc_ucounts(struct user_namespace *ns, kuid_t uid)
 163 {
 164         struct hlist_head *hashent = ucounts_hashentry(ns, uid);
 165         struct ucounts *ucounts, *new;
 166         long overflow;
 167
 168         spin_lock_irq(&ucounts_lock);
 169         ucounts = find_ucounts(ns, uid, hashent);
 170         if (!ucounts) {
 171                 spin_unlock_irq(&ucounts_lock);
 172
 173                 new = kzalloc(sizeof(*new), GFP_KERNEL);
 174                 if (!new)
 175                         return NULL;
 176
 177                 new->ns = ns;
 178                 new->uid = uid;
 179                 atomic_set(&new->count, 1);
 180
 181                 spin_lock_irq(&ucounts_lock);
 182                 ucounts = find_ucounts(ns, uid, hashent);
 183                 if (ucounts) {
 184                         kfree(new);
 185                 } else {
 186                         hlist_add_head(&new->node, hashent);
 187                         get_user_ns(new->ns);
 188                         spin_unlock_irq(&ucounts_lock);
 189                         return new;
 190                 }
 191         }
 192         overflow = atomic_add_negative(1, &ucounts->count);
 193         spin_unlock_irq(&ucounts_lock);
 194         if (overflow) {
 195                 put_ucounts(ucounts);
 196                 return NULL;
 197         }
 198         return ucounts;
 199 }
 200
 201 void put_ucounts(struct ucounts *ucounts)
 202 {
 203         unsigned long flags;
 204
 205         if (atomic_dec_and_lock_irqsave(&ucounts->count, &ucounts_lock, flags)) {
 206                 hlist_del_init(&ucounts->node);
 207                 spin_unlock_irqrestore(&ucounts_lock, flags);
 208                 put_user_ns(ucounts->ns);
 209                 kfree(ucounts);
 210         }
 211 }
 212
 213 static inline bool atomic_long_inc_below(atomic_long_t *v, int u)
 214 {
 215         long c, old;
 216         c = atomic_long_read(v);
 217         for (;;) {
 218                 if (unlikely(c >= u))
 219                         return false;
 220                 old = atomic_long_cmpxchg(v, c, c+1);
 221                 if (likely(old == c))
 222                         return true;
 223                 c = old;
 224         }
 225 }
 226
 227 struct ucounts *inc_ucount(struct user_namespace *ns, kuid_t uid,
 228                            enum ucount_type type)
 229 {
 230         struct ucounts *ucounts, *iter, *bad;
 231         struct user_namespace *tns;
 232         ucounts = alloc_ucounts(ns, uid);
 233         for (iter = ucounts; iter; iter = tns->ucounts) {
 234                 long max;
 235                 tns = iter->ns;
 236                 max = READ_ONCE(tns->ucount_max[type]);
 237                 if (!atomic_long_inc_below(&iter->ucount[type], max))
 238                         goto fail;
 239         }
 240         return ucounts;
 241 fail:
 242         bad = iter;
 243         for (iter = ucounts; iter != bad; iter = iter->ns->ucounts)
 244                 atomic_long_dec(&iter->ucount[type]);
 245
 246         put_ucounts(ucounts);
 247         return NULL;
 248 }
 249
 250 void dec_ucount(struct ucounts *ucounts, enum ucount_type type)
 251 {
 252         struct ucounts *iter;
 253         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
 254                 long dec = atomic_long_dec_if_positive(&iter->ucount[type]);
 255                 WARN_ON_ONCE(dec < 0);
 256         }
 257         put_ucounts(ucounts);
 258 }
 259
 260 long inc_rlimit_ucounts(struct ucounts *ucounts, enum ucount_type type, long v)
 261 {
 262         struct ucounts *iter;
 263         long max = LONG_MAX;
 264         long ret = 0;
 265
 266         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
 267                 long new = atomic_long_add_return(v, &iter->ucount[type]);
 268                 if (new < 0 || new > max)
 269                         ret = LONG_MAX;
 270                 else if (iter == ucounts)
 271                         ret = new;
 272                 max = READ_ONCE(iter->ns->ucount_max[type]);
 273         }
 274         return ret;
 275 }
 276
 277 bool dec_rlimit_ucounts(struct ucounts *ucounts, enum ucount_type type, long v)
 278 {
 279         struct ucounts *iter;
 280         long new = -1; /* Silence compiler warning */
 281         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
 282                 long dec = atomic_long_add_return(-v, &iter->ucount[type]);
 283                 WARN_ON_ONCE(dec < 0);
 284                 if (iter == ucounts)
 285                         new = dec;
 286         }
 287         return (new == 0);
 288 }
 289
 290 static void do_dec_rlimit_put_ucounts(struct ucounts *ucounts,
 291                                 struct ucounts *last, enum ucount_type type)
 292 {
 293         struct ucounts *iter, *next;
 294         for (iter = ucounts; iter != last; iter = next) {
 295                 long dec = atomic_long_add_return(-1, &iter->ucount[type]);
 296                 WARN_ON_ONCE(dec < 0);
 297                 next = iter->ns->ucounts;
 298                 if (dec == 0)
 299                         put_ucounts(iter);
 300         }
 301 }
 302
 303 void dec_rlimit_put_ucounts(struct ucounts *ucounts, enum ucount_type type)
 304 {
 305         do_dec_rlimit_put_ucounts(ucounts, NULL, type);
 306 }
 307
 308 long inc_rlimit_get_ucounts(struct ucounts *ucounts, enum ucount_type type)
 309 {
 310         /* Caller must hold a reference to ucounts */
 311         struct ucounts *iter;
 312         long max = LONG_MAX;
 313         long dec, ret = 0;
 314
 315         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
 316                 long new = atomic_long_add_return(1, &iter->ucount[type]);
 317                 if (new < 0 || new > max)
 318                         goto unwind;
 319                 if (iter == ucounts)
 320                         ret = new;
 321                 max = READ_ONCE(iter->ns->ucount_max[type]);
 322                 /*
 323                  * Grab an extra ucount reference for the caller when
 324                  * the rlimit count was previously 0.
 325                  */
 326                 if (new != 1)
 327                         continue;
 328                 if (!get_ucounts(iter))
 329                         goto dec_unwind;
 330         }
 331         return ret;
 332 dec_unwind:
 333         dec = atomic_long_add_return(-1, &iter->ucount[type]);
 334         WARN_ON_ONCE(dec < 0);
 335 unwind:
 336         do_dec_rlimit_put_ucounts(ucounts, iter, type);
 337         return 0;
 338 }
 339
 340 bool is_ucounts_overlimit(struct ucounts *ucounts, enum ucount_type type, unsigned long rlimit)
 341 {
 342         struct ucounts *iter;
 343         long max = rlimit;
 344         if (rlimit > LONG_MAX)
 345                 max = LONG_MAX;
 346         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
 347                 long val = get_ucounts_value(iter, type);
 348                 if (val < 0 || val > max)
 349                         return true;
 350                 max = READ_ONCE(iter->ns->ucount_max[type]);
 351         }
 352         return false;
 353 }
 354
 355 static __init int user_namespace_sysctl_init(void)
 356 {
 357 #ifdef CONFIG_SYSCTL
 358         static struct ctl_table_header *user_header;
 359         static struct ctl_table empty[1];
 360         /*
 361          * It is necessary to register the user directory in the
 362          * default set so that registrations in the child sets work
 363          * properly.
 364          */
 365         user_header = register_sysctl("user", empty);
 366         kmemleak_ignore(user_header);
 367         BUG_ON(!user_header);
 368         BUG_ON(!setup_userns_sysctls(&init_user_ns));
 369 #endif
 370         hlist_add_ucounts(&init_ucounts);
 371         inc_rlimit_ucounts(&init_ucounts, UCOUNT_RLIMIT_NPROC, 1);
 372         return 0;
 373 }
 374 subsys_initcall(user_namespace_sysctl_init);