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 static int zero = 0;
  56 static int int_max = INT_MAX;
  57 #define UCOUNT_ENTRY(name)                              \
  58         {                                               \
  59                 .procname       = name,                 \
  60                 .maxlen         = sizeof(int),          \
  61                 .mode           = 0644,                 \
  62                 .proc_handler   = proc_dointvec_minmax, \
  63                 .extra1         = &zero,                \
  64                 .extra2         = &int_max,             \
  65         }
  66 static struct ctl_table user_table[] = {
  67         UCOUNT_ENTRY("max_user_namespaces"),
  68         UCOUNT_ENTRY("max_pid_namespaces"),
  69         UCOUNT_ENTRY("max_uts_namespaces"),
  70         UCOUNT_ENTRY("max_ipc_namespaces"),
  71         UCOUNT_ENTRY("max_net_namespaces"),
  72         UCOUNT_ENTRY("max_mnt_namespaces"),
  73         UCOUNT_ENTRY("max_cgroup_namespaces"),
  74 #ifdef CONFIG_INOTIFY_USER
  75         UCOUNT_ENTRY("max_inotify_instances"),
  76         UCOUNT_ENTRY("max_inotify_watches"),
  77 #endif
  78         { }
  79 };
  80 #endif /* CONFIG_SYSCTL */
  81
  82 bool setup_userns_sysctls(struct user_namespace *ns)
  83 {
  84 #ifdef CONFIG_SYSCTL
  85         struct ctl_table *tbl;
  86         setup_sysctl_set(&ns->set, &set_root, set_is_seen);
  87         tbl = kmemdup(user_table, sizeof(user_table), GFP_KERNEL);
  88         if (tbl) {
  89                 int i;
  90                 for (i = 0; i < UCOUNT_COUNTS; i++) {
  91                         tbl[i].data = &ns->ucount_max[i];
  92                 }
  93                 ns->sysctls = __register_sysctl_table(&ns->set, "user", tbl);
  94         }
  95         if (!ns->sysctls) {
  96                 kfree(tbl);
  97                 retire_sysctl_set(&ns->set);
  98                 return false;
  99         }
 100 #endif
 101         return true;
 102 }
 103
 104 void retire_userns_sysctls(struct user_namespace *ns)
 105 {
 106 #ifdef CONFIG_SYSCTL
 107         struct ctl_table *tbl;
 108
 109         tbl = ns->sysctls->ctl_table_arg;
 110         unregister_sysctl_table(ns->sysctls);
 111         retire_sysctl_set(&ns->set);
 112         kfree(tbl);
 113 #endif
 114 }
 115
 116 static struct ucounts *find_ucounts(struct user_namespace *ns, kuid_t uid, struct hlist_head *hashent)
 117 {
 118         struct ucounts *ucounts;
 119
 120         hlist_for_each_entry(ucounts, hashent, node) {
 121                 if (uid_eq(ucounts->uid, uid) && (ucounts->ns == ns))
 122                         return ucounts;
 123         }
 124         return NULL;
 125 }
 126
 127 static struct ucounts *get_ucounts(struct user_namespace *ns, kuid_t uid)
 128 {
 129         struct hlist_head *hashent = ucounts_hashentry(ns, uid);
 130         struct ucounts *ucounts, *new;
 131
 132         spin_lock_irq(&ucounts_lock);
 133         ucounts = find_ucounts(ns, uid, hashent);
 134         if (!ucounts) {
 135                 spin_unlock_irq(&ucounts_lock);
 136
 137                 new = kzalloc(sizeof(*new), GFP_KERNEL);
 138                 if (!new)
 139                         return NULL;
 140
 141                 new->ns = ns;
 142                 new->uid = uid;
 143                 new->count = 0;
 144
 145                 spin_lock_irq(&ucounts_lock);
 146                 ucounts = find_ucounts(ns, uid, hashent);
 147                 if (ucounts) {
 148                         kfree(new);
 149                 } else {
 150                         hlist_add_head(&new->node, hashent);
 151                         ucounts = new;
 152                 }
 153         }
 154         if (ucounts->count == INT_MAX)
 155                 ucounts = NULL;
 156         else
 157                 ucounts->count += 1;
 158         spin_unlock_irq(&ucounts_lock);
 159         return ucounts;
 160 }
 161
 162 static void put_ucounts(struct ucounts *ucounts)
 163 {
 164         unsigned long flags;
 165
 166         spin_lock_irqsave(&ucounts_lock, flags);
 167         ucounts->count -= 1;
 168         if (!ucounts->count)
 169                 hlist_del_init(&ucounts->node);
 170         else
 171                 ucounts = NULL;
 172         spin_unlock_irqrestore(&ucounts_lock, flags);
 173
 174         kfree(ucounts);
 175 }
 176
 177 static inline bool atomic_inc_below(atomic_t *v, int u)
 178 {
 179         int c, old;
 180         c = atomic_read(v);
 181         for (;;) {
 182                 if (unlikely(c >= u))
 183                         return false;
 184                 old = atomic_cmpxchg(v, c, c+1);
 185                 if (likely(old == c))
 186                         return true;
 187                 c = old;
 188         }
 189 }
 190
 191 struct ucounts *inc_ucount(struct user_namespace *ns, kuid_t uid,
 192                            enum ucount_type type)
 193 {
 194         struct ucounts *ucounts, *iter, *bad;
 195         struct user_namespace *tns;
 196         ucounts = get_ucounts(ns, uid);
 197         for (iter = ucounts; iter; iter = tns->ucounts) {
 198                 int max;
 199                 tns = iter->ns;
 200                 max = READ_ONCE(tns->ucount_max[type]);
 201                 if (!atomic_inc_below(&iter->ucount[type], max))
 202                         goto fail;
 203         }
 204         return ucounts;
 205 fail:
 206         bad = iter;
 207         for (iter = ucounts; iter != bad; iter = iter->ns->ucounts)
 208                 atomic_dec(&iter->ucount[type]);
 209
 210         put_ucounts(ucounts);
 211         return NULL;
 212 }
 213
 214 void dec_ucount(struct ucounts *ucounts, enum ucount_type type)
 215 {
 216         struct ucounts *iter;
 217         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
 218                 int dec = atomic_dec_if_positive(&iter->ucount[type]);
 219                 WARN_ON_ONCE(dec < 0);
 220         }
 221         put_ucounts(ucounts);
 222 }
 223
 224 static __init int user_namespace_sysctl_init(void)
 225 {
 226 #ifdef CONFIG_SYSCTL
 227         static struct ctl_table_header *user_header;
 228         static struct ctl_table empty[1];
 229         /*
 230          * It is necessary to register the user directory in the
 231          * default set so that registrations in the child sets work
 232          * properly.
 233          */
 234         user_header = register_sysctl("user", empty);
 235         kmemleak_ignore(user_header);
 236         BUG_ON(!user_header);
 237         BUG_ON(!setup_userns_sysctls(&init_user_ns));
 238 #endif
 239         return 0;
 240 }
 241 subsys_initcall(user_namespace_sysctl_init);