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                         spin_unlock_irq(&ucounts_lock);
 188                         return new;
 189                 }
 190         }
 191         overflow = atomic_add_negative(1, &ucounts->count);
 192         spin_unlock_irq(&ucounts_lock);
 193         if (overflow) {
 194                 put_ucounts(ucounts);
 195                 return NULL;
 196         }
 197         return ucounts;
 198 }
 199
 200 void put_ucounts(struct ucounts *ucounts)
 201 {
 202         unsigned long flags;
 203
 204         if (atomic_dec_and_lock_irqsave(&ucounts->count, &ucounts_lock, flags)) {
 205                 hlist_del_init(&ucounts->node);
 206                 spin_unlock_irqrestore(&ucounts_lock, flags);
 207                 kfree(ucounts);
 208         }
 209 }
 210
 211 static inline bool atomic_long_inc_below(atomic_long_t *v, int u)
 212 {
 213         long c, old;
 214         c = atomic_long_read(v);
 215         for (;;) {
 216                 if (unlikely(c >= u))
 217                         return false;
 218                 old = atomic_long_cmpxchg(v, c, c+1);
 219                 if (likely(old == c))
 220                         return true;
 221                 c = old;
 222         }
 223 }
 224
 225 struct ucounts *inc_ucount(struct user_namespace *ns, kuid_t uid,
 226                            enum ucount_type type)
 227 {
 228         struct ucounts *ucounts, *iter, *bad;
 229         struct user_namespace *tns;
 230         ucounts = alloc_ucounts(ns, uid);
 231         for (iter = ucounts; iter; iter = tns->ucounts) {
 232                 long max;
 233                 tns = iter->ns;
 234                 max = READ_ONCE(tns->ucount_max[type]);
 235                 if (!atomic_long_inc_below(&iter->ucount[type], max))
 236                         goto fail;
 237         }
 238         return ucounts;
 239 fail:
 240         bad = iter;
 241         for (iter = ucounts; iter != bad; iter = iter->ns->ucounts)
 242                 atomic_long_dec(&iter->ucount[type]);
 243
 244         put_ucounts(ucounts);
 245         return NULL;
 246 }
 247
 248 void dec_ucount(struct ucounts *ucounts, enum ucount_type type)
 249 {
 250         struct ucounts *iter;
 251         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
 252                 long dec = atomic_long_dec_if_positive(&iter->ucount[type]);
 253                 WARN_ON_ONCE(dec < 0);
 254         }
 255         put_ucounts(ucounts);
 256 }
 257
 258 long inc_rlimit_ucounts(struct ucounts *ucounts, enum ucount_type type, long v)
 259 {
 260         struct ucounts *iter;
 261         long ret = 0;
 262
 263         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
 264                 long max = READ_ONCE(iter->ns->ucount_max[type]);
 265                 long new = atomic_long_add_return(v, &iter->ucount[type]);
 266                 if (new < 0 || new > max)
 267                         ret = LONG_MAX;
 268                 else if (iter == ucounts)
 269                         ret = new;
 270         }
 271         return ret;
 272 }
 273
 274 bool dec_rlimit_ucounts(struct ucounts *ucounts, enum ucount_type type, long v)
 275 {
 276         struct ucounts *iter;
 277         long new = -1; /* Silence compiler warning */
 278         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
 279                 long dec = atomic_long_add_return(-v, &iter->ucount[type]);
 280                 WARN_ON_ONCE(dec < 0);
 281                 if (iter == ucounts)
 282                         new = dec;
 283         }
 284         return (new == 0);
 285 }
 286
 287 bool is_ucounts_overlimit(struct ucounts *ucounts, enum ucount_type type, unsigned long max)
 288 {
 289         struct ucounts *iter;
 290         if (get_ucounts_value(ucounts, type) > max)
 291                 return true;
 292         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
 293                 max = READ_ONCE(iter->ns->ucount_max[type]);
 294                 if (get_ucounts_value(iter, type) > max)
 295                         return true;
 296         }
 297         return false;
 298 }
 299
 300 static __init int user_namespace_sysctl_init(void)
 301 {
 302 #ifdef CONFIG_SYSCTL
 303         static struct ctl_table_header *user_header;
 304         static struct ctl_table empty[1];
 305         /*
 306          * It is necessary to register the user directory in the
 307          * default set so that registrations in the child sets work
 308          * properly.
 309          */
 310         user_header = register_sysctl("user", empty);
 311         kmemleak_ignore(user_header);
 312         BUG_ON(!user_header);
 313         BUG_ON(!setup_userns_sysctls(&init_user_ns));
 314 #endif
 315         hlist_add_ucounts(&init_ucounts);
 316         inc_rlimit_ucounts(&init_ucounts, UCOUNT_RLIMIT_NPROC, 1);
 317         return 0;
 318 }
 319 subsys_initcall(user_namespace_sysctl_init);