kernel/ucount.c

   1 /*
   2  *  This program is free software; you can redistribute it and/or
   3  *  modify it under the terms of the GNU General Public License as
   4  *  published by the Free Software Foundation, version 2 of the
   5  *  License.
   6  */
   7
   8 #include <linux/stat.h>
   9 #include <linux/sysctl.h>
  10 #include <linux/slab.h>
  11 #include <linux/hash.h>
  12 #include <linux/user_namespace.h>
  13
  14 #define UCOUNTS_HASHTABLE_BITS 10
  15 static struct hlist_head ucounts_hashtable[(1 << UCOUNTS_HASHTABLE_BITS)];
  16 static DEFINE_SPINLOCK(ucounts_lock);
  17
  18 #define ucounts_hashfn(ns, uid)                                         \
  19         hash_long((unsigned long)__kuid_val(uid) + (unsigned long)(ns), \
  20                   UCOUNTS_HASHTABLE_BITS)
  21 #define ucounts_hashentry(ns, uid)      \
  22         (ucounts_hashtable + ucounts_hashfn(ns, uid))
  23
  24
  25 #ifdef CONFIG_SYSCTL
  26 static struct ctl_table_set *
  27 set_lookup(struct ctl_table_root *root)
  28 {
  29         return &current_user_ns()->set;
  30 }
  31
  32 static int set_is_seen(struct ctl_table_set *set)
  33 {
  34         return &current_user_ns()->set == set;
  35 }
  36
  37 static int set_permissions(struct ctl_table_header *head,
  38                                   struct ctl_table *table)
  39 {
  40         struct user_namespace *user_ns =
  41                 container_of(head->set, struct user_namespace, set);
  42         int mode;
  43
  44         /* Allow users with CAP_SYS_RESOURCE unrestrained access */
  45         if (ns_capable(user_ns, CAP_SYS_RESOURCE))
  46                 mode = (table->mode & S_IRWXU) >> 6;
  47         else
  48         /* Allow all others at most read-only access */
  49                 mode = table->mode & S_IROTH;
  50         return (mode << 6) | (mode << 3) | mode;
  51 }
  52
  53 static struct ctl_table_root set_root = {
  54         .lookup = set_lookup,
  55         .permissions = set_permissions,
  56 };
  57
  58 static int zero = 0;
  59 static int int_max = INT_MAX;
  60 static struct ctl_table user_table[] = {
  61         {
  62                 .procname       = "max_user_namespaces",
  63                 .data           = &init_user_ns.max_user_namespaces,
  64                 .maxlen         = sizeof(init_user_ns.max_user_namespaces),
  65                 .mode           = 0644,
  66                 .proc_handler   = proc_dointvec_minmax,
  67                 .extra1         = &zero,
  68                 .extra2         = &int_max,
  69         },
  70         { }
  71 };
  72 #endif /* CONFIG_SYSCTL */
  73
  74 bool setup_userns_sysctls(struct user_namespace *ns)
  75 {
  76 #ifdef CONFIG_SYSCTL
  77         struct ctl_table *tbl;
  78         setup_sysctl_set(&ns->set, &set_root, set_is_seen);
  79         tbl = kmemdup(user_table, sizeof(user_table), GFP_KERNEL);
  80         if (tbl) {
  81                 tbl[0].data = &ns->max_user_namespaces;
  82
  83                 ns->sysctls = __register_sysctl_table(&ns->set, "user", tbl);
  84         }
  85         if (!ns->sysctls) {
  86                 kfree(tbl);
  87                 retire_sysctl_set(&ns->set);
  88                 return false;
  89         }
  90 #endif
  91         return true;
  92 }
  93
  94 void retire_userns_sysctls(struct user_namespace *ns)
  95 {
  96 #ifdef CONFIG_SYSCTL
  97         struct ctl_table *tbl;
  98
  99         tbl = ns->sysctls->ctl_table_arg;
 100         unregister_sysctl_table(ns->sysctls);
 101         retire_sysctl_set(&ns->set);
 102         kfree(tbl);
 103 #endif
 104 }
 105
 106 static struct ucounts *find_ucounts(struct user_namespace *ns, kuid_t uid, struct hlist_head *hashent)
 107 {
 108         struct ucounts *ucounts;
 109
 110         hlist_for_each_entry(ucounts, hashent, node) {
 111                 if (uid_eq(ucounts->uid, uid) && (ucounts->ns == ns))
 112                         return ucounts;
 113         }
 114         return NULL;
 115 }
 116
 117 static struct ucounts *get_ucounts(struct user_namespace *ns, kuid_t uid)
 118 {
 119         struct hlist_head *hashent = ucounts_hashentry(ns, uid);
 120         struct ucounts *ucounts, *new;
 121
 122         spin_lock(&ucounts_lock);
 123         ucounts = find_ucounts(ns, uid, hashent);
 124         if (!ucounts) {
 125                 spin_unlock(&ucounts_lock);
 126
 127                 new = kzalloc(sizeof(*new), GFP_KERNEL);
 128                 if (!new)
 129                         return NULL;
 130
 131                 new->ns = ns;
 132                 new->uid = uid;
 133                 atomic_set(&new->count, 0);
 134
 135                 spin_lock(&ucounts_lock);
 136                 ucounts = find_ucounts(ns, uid, hashent);
 137                 if (ucounts) {
 138                         kfree(new);
 139                 } else {
 140                         hlist_add_head(&new->node, hashent);
 141                         ucounts = new;
 142                 }
 143         }
 144         if (!atomic_add_unless(&ucounts->count, 1, INT_MAX))
 145                 ucounts = NULL;
 146         spin_unlock(&ucounts_lock);
 147         return ucounts;
 148 }
 149
 150 static void put_ucounts(struct ucounts *ucounts)
 151 {
 152         if (atomic_dec_and_test(&ucounts->count)) {
 153                 spin_lock(&ucounts_lock);
 154                 hlist_del_init(&ucounts->node);
 155                 spin_unlock(&ucounts_lock);
 156
 157                 kfree(ucounts);
 158         }
 159 }
 160
 161 static inline bool atomic_inc_below(atomic_t *v, int u)
 162 {
 163         int c, old;
 164         c = atomic_read(v);
 165         for (;;) {
 166                 if (unlikely(c >= u))
 167                         return false;
 168                 old = atomic_cmpxchg(v, c, c+1);
 169                 if (likely(old == c))
 170                         return true;
 171                 c = old;
 172         }
 173 }
 174
 175 struct ucounts *inc_user_namespaces(struct user_namespace *ns, kuid_t uid)
 176 {
 177         struct ucounts *ucounts, *iter, *bad;
 178         struct user_namespace *tns;
 179         ucounts = get_ucounts(ns, uid);
 180         for (iter = ucounts; iter; iter = tns->ucounts) {
 181                 int max;
 182                 tns = iter->ns;
 183                 max = READ_ONCE(tns->max_user_namespaces);
 184                 if (!atomic_inc_below(&iter->user_namespaces, max))
 185                         goto fail;
 186         }
 187         return ucounts;
 188 fail:
 189         bad = iter;
 190         for (iter = ucounts; iter != bad; iter = iter->ns->ucounts)
 191                 atomic_dec(&iter->user_namespaces);
 192
 193         put_ucounts(ucounts);
 194         return NULL;
 195 }
 196
 197 void dec_user_namespaces(struct ucounts *ucounts)
 198 {
 199         struct ucounts *iter;
 200         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
 201                 int dec = atomic_dec_if_positive(&iter->user_namespaces);
 202                 WARN_ON_ONCE(dec < 0);
 203         }
 204         put_ucounts(ucounts);
 205 }
 206
 207
 208 static __init int user_namespace_sysctl_init(void)
 209 {
 210 #ifdef CONFIG_SYSCTL
 211         static struct ctl_table_header *user_header;
 212         static struct ctl_table empty[1];
 213         /*
 214          * It is necessary to register the user directory in the
 215          * default set so that registrations in the child sets work
 216          * properly.
 217          */
 218         user_header = register_sysctl("user", empty);
 219         BUG_ON(!user_header);
 220         BUG_ON(!setup_userns_sysctls(&init_user_ns));
 221 #endif
 222         return 0;
 223 }
 224 subsys_initcall(user_namespace_sysctl_init);
 225
 226