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 #define UCOUNTS_HASHTABLE_BITS 10
  12 static struct hlist_head ucounts_hashtable[(1 << UCOUNTS_HASHTABLE_BITS)];
  13 static DEFINE_SPINLOCK(ucounts_lock);
  14
  15 #define ucounts_hashfn(ns, uid)                                         \
  16         hash_long((unsigned long)__kuid_val(uid) + (unsigned long)(ns), \
  17                   UCOUNTS_HASHTABLE_BITS)
  18 #define ucounts_hashentry(ns, uid)      \
  19         (ucounts_hashtable + ucounts_hashfn(ns, uid))
  20
  21
  22 #ifdef CONFIG_SYSCTL
  23 static struct ctl_table_set *
  24 set_lookup(struct ctl_table_root *root)
  25 {
  26         return &current_user_ns()->set;
  27 }
  28
  29 static int set_is_seen(struct ctl_table_set *set)
  30 {
  31         return &current_user_ns()->set == set;
  32 }
  33
  34 static int set_permissions(struct ctl_table_header *head,
  35                                   struct ctl_table *table)
  36 {
  37         struct user_namespace *user_ns =
  38                 container_of(head->set, struct user_namespace, set);
  39         int mode;
  40
  41         /* Allow users with CAP_SYS_RESOURCE unrestrained access */
  42         if (ns_capable(user_ns, CAP_SYS_RESOURCE))
  43                 mode = (table->mode & S_IRWXU) >> 6;
  44         else
  45         /* Allow all others at most read-only access */
  46                 mode = table->mode & S_IROTH;
  47         return (mode << 6) | (mode << 3) | mode;
  48 }
  49
  50 static struct ctl_table_root set_root = {
  51         .lookup = set_lookup,
  52         .permissions = set_permissions,
  53 };
  54
  55 #define UCOUNT_ENTRY(name)                              \
  56         {                                               \
  57                 .procname       = name,                 \
  58                 .maxlen         = sizeof(int),          \
  59                 .mode           = 0644,                 \
  60                 .proc_handler   = proc_dointvec_minmax, \
  61                 .extra1         = SYSCTL_ZERO,          \
  62                 .extra2         = SYSCTL_INT_MAX,       \
  63         }
  64 static struct ctl_table user_table[] = {
  65         UCOUNT_ENTRY("max_user_namespaces"),
  66         UCOUNT_ENTRY("max_pid_namespaces"),
  67         UCOUNT_ENTRY("max_uts_namespaces"),
  68         UCOUNT_ENTRY("max_ipc_namespaces"),
  69         UCOUNT_ENTRY("max_net_namespaces"),
  70         UCOUNT_ENTRY("max_mnt_namespaces"),
  71         UCOUNT_ENTRY("max_cgroup_namespaces"),
  72         UCOUNT_ENTRY("max_time_namespaces"),
  73 #ifdef CONFIG_INOTIFY_USER
  74         UCOUNT_ENTRY("max_inotify_instances"),
  75         UCOUNT_ENTRY("max_inotify_watches"),
  76 #endif
  77         { }
  78 };
  79 #endif /* CONFIG_SYSCTL */
  80
  81 bool setup_userns_sysctls(struct user_namespace *ns)
  82 {
  83 #ifdef CONFIG_SYSCTL
  84         struct ctl_table *tbl;
  85
  86         BUILD_BUG_ON(ARRAY_SIZE(user_table) != UCOUNT_COUNTS + 1);
  87         setup_sysctl_set(&ns->set, &set_root, set_is_seen);
  88         tbl = kmemdup(user_table, sizeof(user_table), GFP_KERNEL);
  89         if (tbl) {
  90                 int i;
  91                 for (i = 0; i < UCOUNT_COUNTS; i++) {
  92                         tbl[i].data = &ns->ucount_max[i];
  93                 }
  94                 ns->sysctls = __register_sysctl_table(&ns->set, "user", tbl);
  95         }
  96         if (!ns->sysctls) {
  97                 kfree(tbl);
  98                 retire_sysctl_set(&ns->set);
  99                 return false;
 100         }
 101 #endif
 102         return true;
 103 }
 104
 105 void retire_userns_sysctls(struct user_namespace *ns)
 106 {
 107 #ifdef CONFIG_SYSCTL
 108         struct ctl_table *tbl;
 109
 110         tbl = ns->sysctls->ctl_table_arg;
 111         unregister_sysctl_table(ns->sysctls);
 112         retire_sysctl_set(&ns->set);
 113         kfree(tbl);
 114 #endif
 115 }
 116
 117 static struct ucounts *find_ucounts(struct user_namespace *ns, kuid_t uid, struct hlist_head *hashent)
 118 {
 119         struct ucounts *ucounts;
 120
 121         hlist_for_each_entry(ucounts, hashent, node) {
 122                 if (uid_eq(ucounts->uid, uid) && (ucounts->ns == ns))
 123                         return ucounts;
 124         }
 125         return NULL;
 126 }
 127
 128 static struct ucounts *get_ucounts(struct user_namespace *ns, kuid_t uid)
 129 {
 130         struct hlist_head *hashent = ucounts_hashentry(ns, uid);
 131         struct ucounts *ucounts, *new;
 132
 133         spin_lock_irq(&ucounts_lock);
 134         ucounts = find_ucounts(ns, uid, hashent);
 135         if (!ucounts) {
 136                 spin_unlock_irq(&ucounts_lock);
 137
 138                 new = kzalloc(sizeof(*new), GFP_KERNEL);
 139                 if (!new)
 140                         return NULL;
 141
 142                 new->ns = ns;
 143                 new->uid = uid;
 144                 new->count = 0;
 145
 146                 spin_lock_irq(&ucounts_lock);
 147                 ucounts = find_ucounts(ns, uid, hashent);
 148                 if (ucounts) {
 149                         kfree(new);
 150                 } else {
 151                         hlist_add_head(&new->node, hashent);
 152                         ucounts = new;
 153                 }
 154         }
 155         if (ucounts->count == INT_MAX)
 156                 ucounts = NULL;
 157         else
 158                 ucounts->count += 1;
 159         spin_unlock_irq(&ucounts_lock);
 160         return ucounts;
 161 }
 162
 163 static void put_ucounts(struct ucounts *ucounts)
 164 {
 165         unsigned long flags;
 166
 167         spin_lock_irqsave(&ucounts_lock, flags);
 168         ucounts->count -= 1;
 169         if (!ucounts->count)
 170                 hlist_del_init(&ucounts->node);
 171         else
 172                 ucounts = NULL;
 173         spin_unlock_irqrestore(&ucounts_lock, flags);
 174
 175         kfree(ucounts);
 176 }
 177
 178 static inline bool atomic_inc_below(atomic_t *v, int u)
 179 {
 180         int c, old;
 181         c = atomic_read(v);
 182         for (;;) {
 183                 if (unlikely(c >= u))
 184                         return false;
 185                 old = atomic_cmpxchg(v, c, c+1);
 186                 if (likely(old == c))
 187                         return true;
 188                 c = old;
 189         }
 190 }
 191
 192 struct ucounts *inc_ucount(struct user_namespace *ns, kuid_t uid,
 193                            enum ucount_type type)
 194 {
 195         struct ucounts *ucounts, *iter, *bad;
 196         struct user_namespace *tns;
 197         ucounts = get_ucounts(ns, uid);
 198         for (iter = ucounts; iter; iter = tns->ucounts) {
 199                 int max;
 200                 tns = iter->ns;
 201                 max = READ_ONCE(tns->ucount_max[type]);
 202                 if (!atomic_inc_below(&iter->ucount[type], max))
 203                         goto fail;
 204         }
 205         return ucounts;
 206 fail:
 207         bad = iter;
 208         for (iter = ucounts; iter != bad; iter = iter->ns->ucounts)
 209                 atomic_dec(&iter->ucount[type]);
 210
 211         put_ucounts(ucounts);
 212         return NULL;
 213 }
 214
 215 void dec_ucount(struct ucounts *ucounts, enum ucount_type type)
 216 {
 217         struct ucounts *iter;
 218         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
 219                 int dec = atomic_dec_if_positive(&iter->ucount[type]);
 220                 WARN_ON_ONCE(dec < 0);
 221         }
 222         put_ucounts(ucounts);
 223 }
 224
 225 static __init int user_namespace_sysctl_init(void)
 226 {
 227 #ifdef CONFIG_SYSCTL
 228         static struct ctl_table_header *user_header;
 229         static struct ctl_table empty[1];
 230         /*
 231          * It is necessary to register the user directory in the
 232          * default set so that registrations in the child sets work
 233          * properly.
 234          */
 235         user_header = register_sysctl("user", empty);
 236         kmemleak_ignore(user_header);
 237         BUG_ON(!user_header);
 238         BUG_ON(!setup_userns_sysctls(&init_user_ns));
 239 #endif
 240         return 0;
 241 }
 242 subsys_initcall(user_namespace_sysctl_init);