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 #define UCOUNT_ENTRY(name)                              \
  61         {                                               \
  62                 .procname       = name,                 \
  63                 .maxlen         = sizeof(int),          \
  64                 .mode           = 0644,                 \
  65                 .proc_handler   = proc_dointvec_minmax, \
  66                 .extra1         = &zero,                \
  67                 .extra2         = &int_max,             \
  68         }
  69 static struct ctl_table user_table[] = {
  70         UCOUNT_ENTRY("max_user_namespaces"),
  71         UCOUNT_ENTRY("max_pid_namespaces"),
  72         UCOUNT_ENTRY("max_uts_namespaces"),
  73         UCOUNT_ENTRY("max_ipc_namespaces"),
  74         UCOUNT_ENTRY("max_net_namespaces"),
  75         UCOUNT_ENTRY("max_mnt_namespaces"),
  76         UCOUNT_ENTRY("max_cgroup_namespaces"),
  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         setup_sysctl_set(&ns->set, &set_root, set_is_seen);
  86         tbl = kmemdup(user_table, sizeof(user_table), GFP_KERNEL);
  87         if (tbl) {
  88                 int i;
  89                 for (i = 0; i < UCOUNT_COUNTS; i++) {
  90                         tbl[i].data = &ns->ucount_max[i];
  91                 }
  92                 ns->sysctls = __register_sysctl_table(&ns->set, "user", tbl);
  93         }
  94         if (!ns->sysctls) {
  95                 kfree(tbl);
  96                 retire_sysctl_set(&ns->set);
  97                 return false;
  98         }
  99 #endif
 100         return true;
 101 }
 102
 103 void retire_userns_sysctls(struct user_namespace *ns)
 104 {
 105 #ifdef CONFIG_SYSCTL
 106         struct ctl_table *tbl;
 107
 108         tbl = ns->sysctls->ctl_table_arg;
 109         unregister_sysctl_table(ns->sysctls);
 110         retire_sysctl_set(&ns->set);
 111         kfree(tbl);
 112 #endif
 113 }
 114
 115 static struct ucounts *find_ucounts(struct user_namespace *ns, kuid_t uid, struct hlist_head *hashent)
 116 {
 117         struct ucounts *ucounts;
 118
 119         hlist_for_each_entry(ucounts, hashent, node) {
 120                 if (uid_eq(ucounts->uid, uid) && (ucounts->ns == ns))
 121                         return ucounts;
 122         }
 123         return NULL;
 124 }
 125
 126 static struct ucounts *get_ucounts(struct user_namespace *ns, kuid_t uid)
 127 {
 128         struct hlist_head *hashent = ucounts_hashentry(ns, uid);
 129         struct ucounts *ucounts, *new;
 130
 131         spin_lock_irq(&ucounts_lock);
 132         ucounts = find_ucounts(ns, uid, hashent);
 133         if (!ucounts) {
 134                 spin_unlock_irq(&ucounts_lock);
 135
 136                 new = kzalloc(sizeof(*new), GFP_KERNEL);
 137                 if (!new)
 138                         return NULL;
 139
 140                 new->ns = ns;
 141                 new->uid = uid;
 142                 new->count = 0;
 143
 144                 spin_lock_irq(&ucounts_lock);
 145                 ucounts = find_ucounts(ns, uid, hashent);
 146                 if (ucounts) {
 147                         kfree(new);
 148                 } else {
 149                         hlist_add_head(&new->node, hashent);
 150                         ucounts = new;
 151                 }
 152         }
 153         if (ucounts->count == INT_MAX)
 154                 ucounts = NULL;
 155         else
 156                 ucounts->count += 1;
 157         spin_unlock_irq(&ucounts_lock);
 158         return ucounts;
 159 }
 160
 161 static void put_ucounts(struct ucounts *ucounts)
 162 {
 163         unsigned long flags;
 164
 165         spin_lock_irqsave(&ucounts_lock, flags);
 166         ucounts->count -= 1;
 167         if (!ucounts->count)
 168                 hlist_del_init(&ucounts->node);
 169         else
 170                 ucounts = NULL;
 171         spin_unlock_irqrestore(&ucounts_lock, flags);
 172
 173         kfree(ucounts);
 174 }
 175
 176 static inline bool atomic_inc_below(atomic_t *v, int u)
 177 {
 178         int c, old;
 179         c = atomic_read(v);
 180         for (;;) {
 181                 if (unlikely(c >= u))
 182                         return false;
 183                 old = atomic_cmpxchg(v, c, c+1);
 184                 if (likely(old == c))
 185                         return true;
 186                 c = old;
 187         }
 188 }
 189
 190 struct ucounts *inc_ucount(struct user_namespace *ns, kuid_t uid,
 191                            enum ucount_type type)
 192 {
 193         struct ucounts *ucounts, *iter, *bad;
 194         struct user_namespace *tns;
 195         ucounts = get_ucounts(ns, uid);
 196         for (iter = ucounts; iter; iter = tns->ucounts) {
 197                 int max;
 198                 tns = iter->ns;
 199                 max = READ_ONCE(tns->ucount_max[type]);
 200                 if (!atomic_inc_below(&iter->ucount[type], max))
 201                         goto fail;
 202         }
 203         return ucounts;
 204 fail:
 205         bad = iter;
 206         for (iter = ucounts; iter != bad; iter = iter->ns->ucounts)
 207                 atomic_dec(&iter->ucount[type]);
 208
 209         put_ucounts(ucounts);
 210         return NULL;
 211 }
 212
 213 void dec_ucount(struct ucounts *ucounts, enum ucount_type type)
 214 {
 215         struct ucounts *iter;
 216         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
 217                 int dec = atomic_dec_if_positive(&iter->ucount[type]);
 218                 WARN_ON_ONCE(dec < 0);
 219         }
 220         put_ucounts(ucounts);
 221 }
 222
 223 static __init int user_namespace_sysctl_init(void)
 224 {
 225 #ifdef CONFIG_SYSCTL
 226         static struct ctl_table_header *user_header;
 227         static struct ctl_table empty[1];
 228         /*
 229          * It is necessary to register the user directory in the
 230          * default set so that registrations in the child sets work
 231          * properly.
 232          */
 233         user_header = register_sysctl("user", empty);
 234         kmemleak_ignore(user_header);
 235         BUG_ON(!user_header);
 236         BUG_ON(!setup_userns_sysctls(&init_user_ns));
 237 #endif
 238         return 0;
 239 }
 240 subsys_initcall(user_namespace_sysctl_init);