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/cred.h>
  12 #include <linux/hash.h>
  13 #include <linux/user_namespace.h>
  14
  15 #define UCOUNTS_HASHTABLE_BITS 10
  16 static struct hlist_head ucounts_hashtable[(1 << UCOUNTS_HASHTABLE_BITS)];
  17 static DEFINE_SPINLOCK(ucounts_lock);
  18
  19 #define ucounts_hashfn(ns, uid)                                         \
  20         hash_long((unsigned long)__kuid_val(uid) + (unsigned long)(ns), \
  21                   UCOUNTS_HASHTABLE_BITS)
  22 #define ucounts_hashentry(ns, uid)      \
  23         (ucounts_hashtable + ucounts_hashfn(ns, uid))
  24
  25
  26 #ifdef CONFIG_SYSCTL
  27 static struct ctl_table_set *
  28 set_lookup(struct ctl_table_root *root)
  29 {
  30         return &current_user_ns()->set;
  31 }
  32
  33 static int set_is_seen(struct ctl_table_set *set)
  34 {
  35         return &current_user_ns()->set == set;
  36 }
  37
  38 static int set_permissions(struct ctl_table_header *head,
  39                                   struct ctl_table *table)
  40 {
  41         struct user_namespace *user_ns =
  42                 container_of(head->set, struct user_namespace, set);
  43         int mode;
  44
  45         /* Allow users with CAP_SYS_RESOURCE unrestrained access */
  46         if (ns_capable(user_ns, CAP_SYS_RESOURCE))
  47                 mode = (table->mode & S_IRWXU) >> 6;
  48         else
  49         /* Allow all others at most read-only access */
  50                 mode = table->mode & S_IROTH;
  51         return (mode << 6) | (mode << 3) | mode;
  52 }
  53
  54 static struct ctl_table_root set_root = {
  55         .lookup = set_lookup,
  56         .permissions = set_permissions,
  57 };
  58
  59 static int zero = 0;
  60 static int int_max = INT_MAX;
  61 #define UCOUNT_ENTRY(name)                              \
  62         {                                               \
  63                 .procname       = name,                 \
  64                 .maxlen         = sizeof(int),          \
  65                 .mode           = 0644,                 \
  66                 .proc_handler   = proc_dointvec_minmax, \
  67                 .extra1         = &zero,                \
  68                 .extra2         = &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 #ifdef CONFIG_INOTIFY_USER
  79         UCOUNT_ENTRY("max_inotify_instances"),
  80         UCOUNT_ENTRY("max_inotify_watches"),
  81 #endif
  82         { }
  83 };
  84 #endif /* CONFIG_SYSCTL */
  85
  86 bool setup_userns_sysctls(struct user_namespace *ns)
  87 {
  88 #ifdef CONFIG_SYSCTL
  89         struct ctl_table *tbl;
  90         setup_sysctl_set(&ns->set, &set_root, set_is_seen);
  91         tbl = kmemdup(user_table, sizeof(user_table), GFP_KERNEL);
  92         if (tbl) {
  93                 int i;
  94                 for (i = 0; i < UCOUNT_COUNTS; i++) {
  95                         tbl[i].data = &ns->ucount_max[i];
  96                 }
  97                 ns->sysctls = __register_sysctl_table(&ns->set, "user", tbl);
  98         }
  99         if (!ns->sysctls) {
 100                 kfree(tbl);
 101                 retire_sysctl_set(&ns->set);
 102                 return false;
 103         }
 104 #endif
 105         return true;
 106 }
 107
 108 void retire_userns_sysctls(struct user_namespace *ns)
 109 {
 110 #ifdef CONFIG_SYSCTL
 111         struct ctl_table *tbl;
 112
 113         tbl = ns->sysctls->ctl_table_arg;
 114         unregister_sysctl_table(ns->sysctls);
 115         retire_sysctl_set(&ns->set);
 116         kfree(tbl);
 117 #endif
 118 }
 119
 120 static struct ucounts *find_ucounts(struct user_namespace *ns, kuid_t uid, struct hlist_head *hashent)
 121 {
 122         struct ucounts *ucounts;
 123
 124         hlist_for_each_entry(ucounts, hashent, node) {
 125                 if (uid_eq(ucounts->uid, uid) && (ucounts->ns == ns))
 126                         return ucounts;
 127         }
 128         return NULL;
 129 }
 130
 131 static struct ucounts *get_ucounts(struct user_namespace *ns, kuid_t uid)
 132 {
 133         struct hlist_head *hashent = ucounts_hashentry(ns, uid);
 134         struct ucounts *ucounts, *new;
 135
 136         spin_lock_irq(&ucounts_lock);
 137         ucounts = find_ucounts(ns, uid, hashent);
 138         if (!ucounts) {
 139                 spin_unlock_irq(&ucounts_lock);
 140
 141                 new = kzalloc(sizeof(*new), GFP_KERNEL);
 142                 if (!new)
 143                         return NULL;
 144
 145                 new->ns = ns;
 146                 new->uid = uid;
 147                 new->count = 0;
 148
 149                 spin_lock_irq(&ucounts_lock);
 150                 ucounts = find_ucounts(ns, uid, hashent);
 151                 if (ucounts) {
 152                         kfree(new);
 153                 } else {
 154                         hlist_add_head(&new->node, hashent);
 155                         ucounts = new;
 156                 }
 157         }
 158         if (ucounts->count == INT_MAX)
 159                 ucounts = NULL;
 160         else
 161                 ucounts->count += 1;
 162         spin_unlock_irq(&ucounts_lock);
 163         return ucounts;
 164 }
 165
 166 static void put_ucounts(struct ucounts *ucounts)
 167 {
 168         unsigned long flags;
 169
 170         spin_lock_irqsave(&ucounts_lock, flags);
 171         ucounts->count -= 1;
 172         if (!ucounts->count)
 173                 hlist_del_init(&ucounts->node);
 174         else
 175                 ucounts = NULL;
 176         spin_unlock_irqrestore(&ucounts_lock, flags);
 177
 178         kfree(ucounts);
 179 }
 180
 181 static inline bool atomic_inc_below(atomic_t *v, int u)
 182 {
 183         int c, old;
 184         c = atomic_read(v);
 185         for (;;) {
 186                 if (unlikely(c >= u))
 187                         return false;
 188                 old = atomic_cmpxchg(v, c, c+1);
 189                 if (likely(old == c))
 190                         return true;
 191                 c = old;
 192         }
 193 }
 194
 195 struct ucounts *inc_ucount(struct user_namespace *ns, kuid_t uid,
 196                            enum ucount_type type)
 197 {
 198         struct ucounts *ucounts, *iter, *bad;
 199         struct user_namespace *tns;
 200         ucounts = get_ucounts(ns, uid);
 201         for (iter = ucounts; iter; iter = tns->ucounts) {
 202                 int max;
 203                 tns = iter->ns;
 204                 max = READ_ONCE(tns->ucount_max[type]);
 205                 if (!atomic_inc_below(&iter->ucount[type], max))
 206                         goto fail;
 207         }
 208         return ucounts;
 209 fail:
 210         bad = iter;
 211         for (iter = ucounts; iter != bad; iter = iter->ns->ucounts)
 212                 atomic_dec(&iter->ucount[type]);
 213
 214         put_ucounts(ucounts);
 215         return NULL;
 216 }
 217
 218 void dec_ucount(struct ucounts *ucounts, enum ucount_type type)
 219 {
 220         struct ucounts *iter;
 221         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
 222                 int dec = atomic_dec_if_positive(&iter->ucount[type]);
 223                 WARN_ON_ONCE(dec < 0);
 224         }
 225         put_ucounts(ucounts);
 226 }
 227
 228 static __init int user_namespace_sysctl_init(void)
 229 {
 230 #ifdef CONFIG_SYSCTL
 231         static struct ctl_table_header *user_header;
 232         static struct ctl_table empty[1];
 233         /*
 234          * It is necessary to register the user directory in the
 235          * default set so that registrations in the child sets work
 236          * properly.
 237          */
 238         user_header = register_sysctl("user", empty);
 239         kmemleak_ignore(user_header);
 240         BUG_ON(!user_header);
 241         BUG_ON(!setup_userns_sysctls(&init_user_ns));
 242 #endif
 243         return 0;
 244 }
 245 subsys_initcall(user_namespace_sysctl_init);