This commit was manufactured by cvs2svn to create branch 'vserver'.
[linux-2.6.git] / security / keys / key.c
diff --git a/security/keys/key.c b/security/keys/key.c
new file mode 100644 (file)
index 0000000..df264ad
--- /dev/null
@@ -0,0 +1,1039 @@
+/* key.c: basic authentication token and access key management
+ *
+ * Copyright (C) 2004 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/sched.h>
+#include <linux/slab.h>
+#include <linux/workqueue.h>
+#include <linux/err.h>
+#include "internal.h"
+
+static kmem_cache_t    *key_jar;
+static key_serial_t    key_serial_next = 3;
+struct rb_root         key_serial_tree; /* tree of keys indexed by serial */
+spinlock_t             key_serial_lock = SPIN_LOCK_UNLOCKED;
+
+struct rb_root key_user_tree; /* tree of quota records indexed by UID */
+spinlock_t     key_user_lock = SPIN_LOCK_UNLOCKED;
+
+static LIST_HEAD(key_types_list);
+static DECLARE_RWSEM(key_types_sem);
+
+static void key_cleanup(void *data);
+static DECLARE_WORK(key_cleanup_task, key_cleanup, NULL);
+
+/* we serialise key instantiation and link */
+DECLARE_RWSEM(key_construction_sem);
+
+/* any key who's type gets unegistered will be re-typed to this */
+struct key_type key_type_dead = {
+       .name           = "dead",
+};
+
+#ifdef KEY_DEBUGGING
+void __key_check(const struct key *key)
+{
+       printk("__key_check: key %p {%08x} should be {%08x}\n",
+              key, key->magic, KEY_DEBUG_MAGIC);
+       BUG();
+}
+#endif
+
+/*****************************************************************************/
+/*
+ * get the key quota record for a user, allocating a new record if one doesn't
+ * already exist
+ */
+struct key_user *key_user_lookup(uid_t uid)
+{
+       struct key_user *candidate = NULL, *user;
+       struct rb_node *parent = NULL;
+       struct rb_node **p = &key_user_tree.rb_node;
+
+ try_again:
+       spin_lock(&key_user_lock);
+
+       /* search the tree for a user record with a matching UID */
+       while (*p) {
+               parent = *p;
+               user = rb_entry(parent, struct key_user, node);
+
+               if (uid < user->uid)
+                       p = &(*p)->rb_left;
+               else if (uid > user->uid)
+                       p = &(*p)->rb_right;
+               else
+                       goto found;
+       }
+
+       /* if we get here, we failed to find a match in the tree */
+       if (!candidate) {
+               /* allocate a candidate user record if we don't already have
+                * one */
+               spin_unlock(&key_user_lock);
+
+               user = NULL;
+               candidate = kmalloc(sizeof(struct key_user), GFP_KERNEL);
+               if (unlikely(!candidate))
+                       goto out;
+
+               /* the allocation may have scheduled, so we need to repeat the
+                * search lest someone else added the record whilst we were
+                * asleep */
+               goto try_again;
+       }
+
+       /* if we get here, then the user record still hadn't appeared on the
+        * second pass - so we use the candidate record */
+       atomic_set(&candidate->usage, 1);
+       atomic_set(&candidate->nkeys, 0);
+       atomic_set(&candidate->nikeys, 0);
+       candidate->uid = uid;
+       candidate->qnkeys = 0;
+       candidate->qnbytes = 0;
+       spin_lock_init(&candidate->lock);
+       INIT_LIST_HEAD(&candidate->consq);
+
+       rb_link_node(&candidate->node, parent, p);
+       rb_insert_color(&candidate->node, &key_user_tree);
+       spin_unlock(&key_user_lock);
+       user = candidate;
+       goto out;
+
+       /* okay - we found a user record for this UID */
+ found:
+       atomic_inc(&user->usage);
+       spin_unlock(&key_user_lock);
+       if (candidate)
+               kfree(candidate);
+ out:
+       return user;
+
+} /* end key_user_lookup() */
+
+/*****************************************************************************/
+/*
+ * dispose of a user structure
+ */
+void key_user_put(struct key_user *user)
+{
+       if (atomic_dec_and_lock(&user->usage, &key_user_lock)) {
+               rb_erase(&user->node, &key_user_tree);
+               spin_unlock(&key_user_lock);
+
+               kfree(user);
+       }
+
+} /* end key_user_put() */
+
+/*****************************************************************************/
+/*
+ * insert a key with a fixed serial number
+ */
+static void __init __key_insert_serial(struct key *key)
+{
+       struct rb_node *parent, **p;
+       struct key *xkey;
+
+       parent = NULL;
+       p = &key_serial_tree.rb_node;
+
+       while (*p) {
+               parent = *p;
+               xkey = rb_entry(parent, struct key, serial_node);
+
+               if (key->serial < xkey->serial)
+                       p = &(*p)->rb_left;
+               else if (key->serial > xkey->serial)
+                       p = &(*p)->rb_right;
+               else
+                       BUG();
+       }
+
+       /* we've found a suitable hole - arrange for this key to occupy it */
+       rb_link_node(&key->serial_node, parent, p);
+       rb_insert_color(&key->serial_node, &key_serial_tree);
+
+} /* end __key_insert_serial() */
+
+/*****************************************************************************/
+/*
+ * assign a key the next unique serial number
+ * - we work through all the serial numbers between 2 and 2^31-1 in turn and
+ *   then wrap
+ */
+static inline void key_alloc_serial(struct key *key)
+{
+       struct rb_node *parent, **p;
+       struct key *xkey;
+
+       spin_lock(&key_serial_lock);
+
+       /* propose a likely serial number and look for a hole for it in the
+        * serial number tree */
+       key->serial = key_serial_next;
+       if (key->serial < 3)
+               key->serial = 3;
+       key_serial_next = key->serial + 1;
+
+       parent = NULL;
+       p = &key_serial_tree.rb_node;
+
+       while (*p) {
+               parent = *p;
+               xkey = rb_entry(parent, struct key, serial_node);
+
+               if (key->serial < xkey->serial)
+                       p = &(*p)->rb_left;
+               else if (key->serial > xkey->serial)
+                       p = &(*p)->rb_right;
+               else
+                       goto serial_exists;
+       }
+       goto insert_here;
+
+       /* we found a key with the proposed serial number - walk the tree from
+        * that point looking for the next unused serial number */
+ serial_exists:
+       for (;;) {
+               key->serial = key_serial_next;
+               if (key->serial < 2)
+                       key->serial = 2;
+               key_serial_next = key->serial + 1;
+
+               if (!parent->rb_parent)
+                       p = &key_serial_tree.rb_node;
+               else if (parent->rb_parent->rb_left == parent)
+                       p = &parent->rb_parent->rb_left;
+               else
+                       p = &parent->rb_parent->rb_right;
+
+               parent = rb_next(parent);
+               if (!parent)
+                       break;
+
+               xkey = rb_entry(parent, struct key, serial_node);
+               if (key->serial < xkey->serial)
+                       goto insert_here;
+       }
+
+       /* we've found a suitable hole - arrange for this key to occupy it */
+ insert_here:
+       rb_link_node(&key->serial_node, parent, p);
+       rb_insert_color(&key->serial_node, &key_serial_tree);
+
+       spin_unlock(&key_serial_lock);
+
+} /* end key_alloc_serial() */
+
+/*****************************************************************************/
+/*
+ * allocate a key of the specified type
+ * - update the user's quota to reflect the existence of the key
+ * - called from a key-type operation with key_types_sem read-locked by either
+ *   key_create_or_update() or by key_duplicate(); this prevents unregistration
+ *   of the key type
+ * - upon return the key is as yet uninstantiated; the caller needs to either
+ *   instantiate the key or discard it before returning
+ */
+struct key *key_alloc(struct key_type *type, const char *desc,
+                     uid_t uid, gid_t gid, key_perm_t perm,
+                     int not_in_quota)
+{
+       struct key_user *user = NULL;
+       struct key *key;
+       size_t desclen, quotalen;
+
+       key = ERR_PTR(-EINVAL);
+       if (!desc || !*desc)
+               goto error;
+
+       desclen = strlen(desc) + 1;
+       quotalen = desclen + type->def_datalen;
+
+       /* get hold of the key tracking for this user */
+       user = key_user_lookup(uid);
+       if (!user)
+               goto no_memory_1;
+
+       /* check that the user's quota permits allocation of another key and
+        * its description */
+       if (!not_in_quota) {
+               spin_lock(&user->lock);
+               if (user->qnkeys + 1 >= KEYQUOTA_MAX_KEYS &&
+                   user->qnbytes + quotalen >= KEYQUOTA_MAX_BYTES
+                   )
+                       goto no_quota;
+
+               user->qnkeys++;
+               user->qnbytes += quotalen;
+               spin_unlock(&user->lock);
+       }
+
+       /* allocate and initialise the key and its description */
+       key = kmem_cache_alloc(key_jar, SLAB_KERNEL);
+       if (!key)
+               goto no_memory_2;
+
+       if (desc) {
+               key->description = kmalloc(desclen, GFP_KERNEL);
+               if (!key->description)
+                       goto no_memory_3;
+
+               memcpy(key->description, desc, desclen);
+       }
+
+       atomic_set(&key->usage, 1);
+       rwlock_init(&key->lock);
+       init_rwsem(&key->sem);
+       key->type = type;
+       key->user = user;
+       key->quotalen = quotalen;
+       key->datalen = type->def_datalen;
+       key->uid = uid;
+       key->gid = gid;
+       key->perm = perm;
+       key->flags = 0;
+       key->expiry = 0;
+       key->payload.data = NULL;
+
+       if (!not_in_quota)
+               key->flags |= KEY_FLAG_IN_QUOTA;
+
+       memset(&key->type_data, 0, sizeof(key->type_data));
+
+#ifdef KEY_DEBUGGING
+       key->magic = KEY_DEBUG_MAGIC;
+#endif
+
+       /* publish the key by giving it a serial number */
+       atomic_inc(&user->nkeys);
+       key_alloc_serial(key);
+
+ error:
+       return key;
+
+ no_memory_3:
+       kmem_cache_free(key_jar, key);
+ no_memory_2:
+       if (!not_in_quota) {
+               spin_lock(&user->lock);
+               user->qnkeys--;
+               user->qnbytes -= quotalen;
+               spin_unlock(&user->lock);
+       }
+       key_user_put(user);
+ no_memory_1:
+       key = ERR_PTR(-ENOMEM);
+       goto error;
+
+ no_quota:
+       spin_unlock(&user->lock);
+       key_user_put(user);
+       key = ERR_PTR(-EDQUOT);
+       goto error;
+
+} /* end key_alloc() */
+
+EXPORT_SYMBOL(key_alloc);
+
+/*****************************************************************************/
+/*
+ * reserve an amount of quota for the key's payload
+ */
+int key_payload_reserve(struct key *key, size_t datalen)
+{
+       int delta = (int) datalen - key->datalen;
+       int ret = 0;
+
+       key_check(key);
+
+       /* contemplate the quota adjustment */
+       if (delta != 0 && key->flags & KEY_FLAG_IN_QUOTA) {
+               spin_lock(&key->user->lock);
+
+               if (delta > 0 &&
+                   key->user->qnbytes + delta > KEYQUOTA_MAX_BYTES
+                   ) {
+                       ret = -EDQUOT;
+               }
+               else {
+                       key->user->qnbytes += delta;
+                       key->quotalen += delta;
+               }
+               spin_unlock(&key->user->lock);
+       }
+
+       /* change the recorded data length if that didn't generate an error */
+       if (ret == 0)
+               key->datalen = datalen;
+
+       return ret;
+
+} /* end key_payload_reserve() */
+
+EXPORT_SYMBOL(key_payload_reserve);
+
+/*****************************************************************************/
+/*
+ * instantiate a key and link it into the target keyring atomically
+ * - called with the target keyring's semaphore writelocked
+ */
+static int __key_instantiate_and_link(struct key *key,
+                                     const void *data,
+                                     size_t datalen,
+                                     struct key *keyring)
+{
+       int ret, awaken;
+
+       key_check(key);
+       key_check(keyring);
+
+       awaken = 0;
+       ret = -EBUSY;
+
+       down_write(&key_construction_sem);
+
+       /* can't instantiate twice */
+       if (!(key->flags & KEY_FLAG_INSTANTIATED)) {
+               /* instantiate the key */
+               ret = key->type->instantiate(key, data, datalen);
+
+               if (ret == 0) {
+                       /* mark the key as being instantiated */
+                       write_lock(&key->lock);
+
+                       atomic_inc(&key->user->nikeys);
+                       key->flags |= KEY_FLAG_INSTANTIATED;
+
+                       if (key->flags & KEY_FLAG_USER_CONSTRUCT) {
+                               key->flags &= ~KEY_FLAG_USER_CONSTRUCT;
+                               awaken = 1;
+                       }
+
+                       write_unlock(&key->lock);
+
+                       /* and link it into the destination keyring */
+                       if (keyring)
+                               ret = __key_link(keyring, key);
+               }
+       }
+
+       up_write(&key_construction_sem);
+
+       /* wake up anyone waiting for a key to be constructed */
+       if (awaken)
+               wake_up_all(&request_key_conswq);
+
+       return ret;
+
+} /* end __key_instantiate_and_link() */
+
+/*****************************************************************************/
+/*
+ * instantiate a key and link it into the target keyring atomically
+ */
+int key_instantiate_and_link(struct key *key,
+                            const void *data,
+                            size_t datalen,
+                            struct key *keyring)
+{
+       int ret;
+
+       if (keyring)
+               down_write(&keyring->sem);
+
+       ret = __key_instantiate_and_link(key, data, datalen, keyring);
+
+       if (keyring)
+               up_write(&keyring->sem);
+
+       return ret;
+} /* end key_instantiate_and_link() */
+
+EXPORT_SYMBOL(key_instantiate_and_link);
+
+/*****************************************************************************/
+/*
+ * negatively instantiate a key and link it into the target keyring atomically
+ */
+int key_negate_and_link(struct key *key,
+                       unsigned timeout,
+                       struct key *keyring)
+{
+       struct timespec now;
+       int ret, awaken;
+
+       key_check(key);
+       key_check(keyring);
+
+       awaken = 0;
+       ret = -EBUSY;
+
+       if (keyring)
+               down_write(&keyring->sem);
+
+       down_write(&key_construction_sem);
+
+       /* can't instantiate twice */
+       if (!(key->flags & KEY_FLAG_INSTANTIATED)) {
+               /* mark the key as being negatively instantiated */
+               write_lock(&key->lock);
+
+               atomic_inc(&key->user->nikeys);
+               key->flags |= KEY_FLAG_INSTANTIATED | KEY_FLAG_NEGATIVE;
+               now = current_kernel_time();
+               key->expiry = now.tv_sec + timeout;
+
+               if (key->flags & KEY_FLAG_USER_CONSTRUCT) {
+                       key->flags &= ~KEY_FLAG_USER_CONSTRUCT;
+                       awaken = 1;
+               }
+
+               write_unlock(&key->lock);
+               ret = 0;
+
+               /* and link it into the destination keyring */
+               if (keyring)
+                       ret = __key_link(keyring, key);
+       }
+
+       up_write(&key_construction_sem);
+
+       if (keyring)
+               up_write(&keyring->sem);
+
+       /* wake up anyone waiting for a key to be constructed */
+       if (awaken)
+               wake_up_all(&request_key_conswq);
+
+       return ret;
+
+} /* end key_negate_and_link() */
+
+EXPORT_SYMBOL(key_negate_and_link);
+
+/*****************************************************************************/
+/*
+ * do cleaning up in process context so that we don't have to disable
+ * interrupts all over the place
+ */
+static void key_cleanup(void *data)
+{
+       struct rb_node *_n;
+       struct key *key;
+
+ go_again:
+       /* look for a dead key in the tree */
+       spin_lock(&key_serial_lock);
+
+       for (_n = rb_first(&key_serial_tree); _n; _n = rb_next(_n)) {
+               key = rb_entry(_n, struct key, serial_node);
+
+               if (atomic_read(&key->usage) == 0)
+                       goto found_dead_key;
+       }
+
+       spin_unlock(&key_serial_lock);
+       return;
+
+ found_dead_key:
+       /* we found a dead key - once we've removed it from the tree, we can
+        * drop the lock */
+       rb_erase(&key->serial_node, &key_serial_tree);
+       spin_unlock(&key_serial_lock);
+
+       /* deal with the user's key tracking and quota */
+       if (key->flags & KEY_FLAG_IN_QUOTA) {
+               spin_lock(&key->user->lock);
+               key->user->qnkeys--;
+               key->user->qnbytes -= key->quotalen;
+               spin_unlock(&key->user->lock);
+       }
+
+       atomic_dec(&key->user->nkeys);
+       if (key->flags & KEY_FLAG_INSTANTIATED)
+               atomic_dec(&key->user->nikeys);
+
+       key_user_put(key->user);
+
+       /* now throw away the key memory */
+       if (key->type->destroy)
+               key->type->destroy(key);
+
+       kfree(key->description);
+
+#ifdef KEY_DEBUGGING
+       key->magic = KEY_DEBUG_MAGIC_X;
+#endif
+       kmem_cache_free(key_jar, key);
+
+       /* there may, of course, be more than one key to destroy */
+       goto go_again;
+
+} /* end key_cleanup() */
+
+/*****************************************************************************/
+/*
+ * dispose of a reference to a key
+ * - when all the references are gone, we schedule the cleanup task to come and
+ *   pull it out of the tree in definite process context
+ */
+void key_put(struct key *key)
+{
+       if (key) {
+               key_check(key);
+
+               if (atomic_dec_and_test(&key->usage))
+                       schedule_work(&key_cleanup_task);
+       }
+
+} /* end key_put() */
+
+EXPORT_SYMBOL(key_put);
+
+/*****************************************************************************/
+/*
+ * find a key by its serial number
+ */
+struct key *key_lookup(key_serial_t id)
+{
+       struct rb_node *n;
+       struct key *key;
+
+       spin_lock(&key_serial_lock);
+
+       /* search the tree for the specified key */
+       n = key_serial_tree.rb_node;
+       while (n) {
+               key = rb_entry(n, struct key, serial_node);
+
+               if (id < key->serial)
+                       n = n->rb_left;
+               else if (id > key->serial)
+                       n = n->rb_right;
+               else
+                       goto found;
+       }
+
+ not_found:
+       key = ERR_PTR(-ENOKEY);
+       goto error;
+
+ found:
+       /* pretent doesn't exist if it's dead */
+       if (atomic_read(&key->usage) == 0 ||
+           (key->flags & KEY_FLAG_DEAD) ||
+           key->type == &key_type_dead)
+               goto not_found;
+
+       /* this races with key_put(), but that doesn't matter since key_put()
+        * doesn't actually change the key
+        */
+       atomic_inc(&key->usage);
+
+ error:
+       spin_unlock(&key_serial_lock);
+       return key;
+
+} /* end key_lookup() */
+
+/*****************************************************************************/
+/*
+ * find and lock the specified key type against removal
+ * - we return with the sem readlocked
+ */
+struct key_type *key_type_lookup(const char *type)
+{
+       struct key_type *ktype;
+
+       down_read(&key_types_sem);
+
+       /* look up the key type to see if it's one of the registered kernel
+        * types */
+       list_for_each_entry(ktype, &key_types_list, link) {
+               if (strcmp(ktype->name, type) == 0)
+                       goto found_kernel_type;
+       }
+
+       up_read(&key_types_sem);
+       ktype = ERR_PTR(-ENOKEY);
+
+ found_kernel_type:
+       return ktype;
+
+} /* end key_type_lookup() */
+
+/*****************************************************************************/
+/*
+ * unlock a key type
+ */
+void key_type_put(struct key_type *ktype)
+{
+       up_read(&key_types_sem);
+
+} /* end key_type_put() */
+
+/*****************************************************************************/
+/*
+ * attempt to update an existing key
+ * - the key has an incremented refcount
+ * - we need to put the key if we get an error
+ */
+static inline struct key *__key_update(struct key *key, const void *payload,
+                                      size_t plen)
+{
+       int ret;
+
+       /* need write permission on the key to update it */
+       ret = -EACCES;
+       if (!key_permission(key, KEY_WRITE))
+               goto error;
+
+       ret = -EEXIST;
+       if (!key->type->update)
+               goto error;
+
+       down_write(&key->sem);
+
+       ret = key->type->update(key, payload, plen);
+
+       if (ret == 0) {
+               /* updating a negative key instantiates it */
+               write_lock(&key->lock);
+               key->flags &= ~KEY_FLAG_NEGATIVE;
+               write_unlock(&key->lock);
+       }
+
+       up_write(&key->sem);
+
+       if (ret < 0)
+               goto error;
+ out:
+       return key;
+
+ error:
+       key_put(key);
+       key = ERR_PTR(ret);
+       goto out;
+
+} /* end __key_update() */
+
+/*****************************************************************************/
+/*
+ * search the specified keyring for a key of the same description; if one is
+ * found, update it, otherwise add a new one
+ */
+struct key *key_create_or_update(struct key *keyring,
+                                const char *type,
+                                const char *description,
+                                const void *payload,
+                                size_t plen,
+                                int not_in_quota)
+{
+       struct key_type *ktype;
+       struct key *key = NULL;
+       key_perm_t perm;
+       int ret;
+
+       key_check(keyring);
+
+       /* look up the key type to see if it's one of the registered kernel
+        * types */
+       ktype = key_type_lookup(type);
+       if (IS_ERR(ktype)) {
+               key = ERR_PTR(-ENODEV);
+               goto error;
+       }
+
+       ret = -EINVAL;
+       if (!ktype->match || !ktype->instantiate)
+               goto error_2;
+
+       /* search for an existing key of the same type and description in the
+        * destination keyring
+        */
+       down_write(&keyring->sem);
+
+       key = __keyring_search_one(keyring, ktype, description, 0);
+       if (!IS_ERR(key))
+               goto found_matching_key;
+
+       /* if we're going to allocate a new key, we're going to have to modify
+        * the keyring */
+       ret = -EACCES;
+       if (!key_permission(keyring, KEY_WRITE))
+               goto error_3;
+
+       /* decide on the permissions we want */
+       perm = KEY_USR_VIEW | KEY_USR_SEARCH | KEY_USR_LINK;
+
+       if (ktype->read)
+               perm |= KEY_USR_READ;
+
+       if (ktype == &key_type_keyring || ktype->update)
+               perm |= KEY_USR_WRITE;
+
+       /* allocate a new key */
+       key = key_alloc(ktype, description, current->fsuid, current->fsgid,
+                       perm, not_in_quota);
+       if (IS_ERR(key)) {
+               ret = PTR_ERR(key);
+               goto error_3;
+       }
+
+       /* instantiate it and link it into the target keyring */
+       ret = __key_instantiate_and_link(key, payload, plen, keyring);
+       if (ret < 0) {
+               key_put(key);
+               key = ERR_PTR(ret);
+       }
+
+ error_3:
+       up_write(&keyring->sem);
+ error_2:
+       key_type_put(ktype);
+ error:
+       return key;
+
+ found_matching_key:
+       /* we found a matching key, so we're going to try to update it
+        * - we can drop the locks first as we have the key pinned
+        */
+       up_write(&keyring->sem);
+       key_type_put(ktype);
+
+       key = __key_update(key, payload, plen);
+       goto error;
+
+} /* end key_create_or_update() */
+
+EXPORT_SYMBOL(key_create_or_update);
+
+/*****************************************************************************/
+/*
+ * update a key
+ */
+int key_update(struct key *key, const void *payload, size_t plen)
+{
+       int ret;
+
+       key_check(key);
+
+       /* the key must be writable */
+       ret = -EACCES;
+       if (!key_permission(key, KEY_WRITE))
+               goto error;
+
+       /* attempt to update it if supported */
+       ret = -EOPNOTSUPP;
+       if (key->type->update) {
+               down_write(&key->sem);
+               ret = key->type->update(key, payload, plen);
+
+               if (ret == 0) {
+                       /* updating a negative key instantiates it */
+                       write_lock(&key->lock);
+                       key->flags &= ~KEY_FLAG_NEGATIVE;
+                       write_unlock(&key->lock);
+               }
+
+               up_write(&key->sem);
+       }
+
+ error:
+       return ret;
+
+} /* end key_update() */
+
+EXPORT_SYMBOL(key_update);
+
+/*****************************************************************************/
+/*
+ * duplicate a key, potentially with a revised description
+ * - must be supported by the keytype (keyrings for instance can be duplicated)
+ */
+struct key *key_duplicate(struct key *source, const char *desc)
+{
+       struct key *key;
+       int ret;
+
+       key_check(source);
+
+       if (!desc)
+               desc = source->description;
+
+       down_read(&key_types_sem);
+
+       ret = -EINVAL;
+       if (!source->type->duplicate)
+               goto error;
+
+       /* allocate and instantiate a key */
+       key = key_alloc(source->type, desc, current->fsuid, current->fsgid,
+                       source->perm, 0);
+       if (IS_ERR(key))
+               goto error_k;
+
+       down_read(&source->sem);
+       ret = key->type->duplicate(key, source);
+       up_read(&source->sem);
+       if (ret < 0)
+               goto error2;
+
+       atomic_inc(&key->user->nikeys);
+
+       write_lock(&key->lock);
+       key->flags |= KEY_FLAG_INSTANTIATED;
+       write_unlock(&key->lock);
+
+ error_k:
+       up_read(&key_types_sem);
+ out:
+       return key;
+
+ error2:
+       key_put(key);
+ error:
+       up_read(&key_types_sem);
+       key = ERR_PTR(ret);
+       goto out;
+
+} /* end key_duplicate() */
+
+/*****************************************************************************/
+/*
+ * revoke a key
+ */
+void key_revoke(struct key *key)
+{
+       key_check(key);
+
+       /* make sure no one's trying to change or use the key when we mark
+        * it */
+       down_write(&key->sem);
+       write_lock(&key->lock);
+       key->flags |= KEY_FLAG_REVOKED;
+       write_unlock(&key->lock);
+       up_write(&key->sem);
+
+} /* end key_revoke() */
+
+EXPORT_SYMBOL(key_revoke);
+
+/*****************************************************************************/
+/*
+ * register a type of key
+ */
+int register_key_type(struct key_type *ktype)
+{
+       struct key_type *p;
+       int ret;
+
+       ret = -EEXIST;
+       down_write(&key_types_sem);
+
+       /* disallow key types with the same name */
+       list_for_each_entry(p, &key_types_list, link) {
+               if (strcmp(p->name, ktype->name) == 0)
+                       goto out;
+       }
+
+       /* store the type */
+       list_add(&ktype->link, &key_types_list);
+       ret = 0;
+
+ out:
+       up_write(&key_types_sem);
+       return ret;
+
+} /* end register_key_type() */
+
+EXPORT_SYMBOL(register_key_type);
+
+/*****************************************************************************/
+/*
+ * unregister a type of key
+ */
+void unregister_key_type(struct key_type *ktype)
+{
+       struct rb_node *_n;
+       struct key *key;
+
+       down_write(&key_types_sem);
+
+       /* withdraw the key type */
+       list_del_init(&ktype->link);
+
+       /* need to withdraw all keys of this type */
+       spin_lock(&key_serial_lock);
+
+       for (_n = rb_first(&key_serial_tree); _n; _n = rb_next(_n)) {
+               key = rb_entry(_n, struct key, serial_node);
+
+               if (key->type != ktype)
+                       continue;
+
+               write_lock(&key->lock);
+               key->type = &key_type_dead;
+               write_unlock(&key->lock);
+
+               /* there shouldn't be anyone looking at the description or
+                * payload now */
+               if (ktype->destroy)
+                       ktype->destroy(key);
+               memset(&key->payload, 0xbd, sizeof(key->payload));
+       }
+
+       spin_unlock(&key_serial_lock);
+       up_write(&key_types_sem);
+
+} /* end unregister_key_type() */
+
+EXPORT_SYMBOL(unregister_key_type);
+
+/*****************************************************************************/
+/*
+ * initialise the key management stuff
+ */
+void __init key_init(void)
+{
+       /* allocate a slab in which we can store keys */
+       key_jar = kmem_cache_create("key_jar", sizeof(struct key),
+                       0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
+
+       /* add the special key types */
+       list_add_tail(&key_type_keyring.link, &key_types_list);
+       list_add_tail(&key_type_dead.link, &key_types_list);
+       list_add_tail(&key_type_user.link, &key_types_list);
+
+       /* record the root user tracking */
+       rb_link_node(&root_key_user.node,
+                    NULL,
+                    &key_user_tree.rb_node);
+
+       rb_insert_color(&root_key_user.node,
+                       &key_user_tree);
+
+       /* record root's user standard keyrings */
+       key_check(&root_user_keyring);
+       key_check(&root_session_keyring);
+
+       __key_insert_serial(&root_user_keyring);
+       __key_insert_serial(&root_session_keyring);
+
+       keyring_publish_name(&root_user_keyring);
+       keyring_publish_name(&root_session_keyring);
+
+       /* link the two root keyrings together */
+       key_link(&root_session_keyring, &root_user_keyring);
+} /* end key_init() */