/* cache.c - routines to maintain an in-core cache of entries */
/* $OpenLDAP$ */
-/*
- * Copyright 1998-2003 The OpenLDAP Foundation, All Rights Reserved.
- * COPYING RESTRICTIONS APPLY, see COPYRIGHT file
+/* This work is part of OpenLDAP Software <http://www.openldap.org/>.
+ *
+ * Copyright 2000-2005 The OpenLDAP Foundation.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted only as authorized by the OpenLDAP
+ * Public License.
+ *
+ * A copy of this license is available in the file LICENSE in the
+ * top-level directory of the distribution or, alternatively, at
+ * <http://www.OpenLDAP.org/license.html>.
*/
#include "portable.h"
#include "back-bdb.h"
-static int bdb_cache_delete_internal(Cache *cache, EntryInfo *e);
+#ifdef BDB_HIER
+#define bdb_cache_lru_add hdb_cache_lru_add
+#endif
+
+static int bdb_cache_delete_internal(Cache *cache, EntryInfo *e, int decr);
#ifdef LDAP_DEBUG
static void bdb_lru_print(Cache *cache);
#endif
+static int bdb_txn_get( Operation *op, DB_ENV *env, DB_TXN **txn, int reset );
+
+/* 4.2.52 */
+#if DB_VERSION_FULL == 0x04020034
+#define READ_TXN_FLAG ReadFlag
+static int ReadFlag = DB_TXN_NOT_DURABLE;
+#else
+#define READ_TXN_FLAG 0
+#endif
+
static EntryInfo *
bdb_cache_entryinfo_new( Cache *cache )
{
if ( !lock ) return 0;
- lockobj.data = ei;
- lockobj.size = sizeof(ei->bei_parent) + sizeof(ei->bei_id);
+ lockobj.data = &ei->bei_id;
+ lockobj.size = sizeof(ei->bei_id) + 1;
list[0].op = DB_LOCK_PUT;
list[0].lock = *lock;
rc = env->lock_vec(env, locker, tryOnly ? DB_LOCK_NOWAIT : 0,
list, 2, NULL );
- if (rc) {
-#ifdef NEW_LOGGING
- LDAP_LOG( CACHE, DETAIL1,
- "bdb_cache_entry_db_relock: entry %ld, rw %d, rc %d\n",
- ei->bei_id, rw, rc );
-#else
+ if (rc && !tryOnly) {
Debug( LDAP_DEBUG_TRACE,
"bdb_cache_entry_db_relock: entry %ld, rw %d, rc %d\n",
ei->bei_id, rw, rc );
-#endif
} else {
*lock = list[1].lock;
}
return rc;
#endif
}
+
static int
-bdb_cache_entry_db_lock
-( DB_ENV *env, u_int32_t locker, EntryInfo *ei, int rw, int tryOnly, DB_LOCK *lock )
+bdb_cache_entry_db_lock( DB_ENV *env, u_int32_t locker, EntryInfo *ei,
+ int rw, int tryOnly, DB_LOCK *lock )
{
#ifdef NO_THREADS
return 0;
else
db_rw = DB_LOCK_READ;
- lockobj.data = ei;
- lockobj.size = sizeof(ei->bei_parent) + sizeof(ei->bei_id);
+ lockobj.data = &ei->bei_id;
+ lockobj.size = sizeof(ei->bei_id) + 1;
rc = LOCK_GET(env, locker, tryOnly ? DB_LOCK_NOWAIT : 0,
&lockobj, db_rw, lock);
- if (rc) {
-#ifdef NEW_LOGGING
- LDAP_LOG( CACHE, DETAIL1,
- "bdb_cache_entry_db_lock: entry %ld, rw %d, rc %d\n",
- ei->bei_id, rw, rc );
-#else
+ if (rc && !tryOnly) {
Debug( LDAP_DEBUG_TRACE,
"bdb_cache_entry_db_lock: entry %ld, rw %d, rc %d\n",
ei->bei_id, rw, rc );
-#endif
}
return rc;
#endif /* NO_THREADS */
}
int
-bdb_cache_entry_db_unlock
-( DB_ENV *env, DB_LOCK *lock )
+bdb_cache_entry_db_unlock ( DB_ENV *env, DB_LOCK *lock )
{
#ifdef NO_THREADS
return 0;
#else
int rc;
+ if ( !lock ) return 0;
+
rc = LOCK_PUT ( env, lock );
return rc;
#endif
} else { \
(cache)->c_lrutail = (ei)->bei_lruprev; \
} \
+ (ei)->bei_lrunext = (ei)->bei_lruprev = NULL; \
} while(0)
#define LRU_ADD( cache, ei ) do { \
{
const EntryInfo *e1 = v_e1, *e2 = v_e2;
int rc = e1->bei_nrdn.bv_len - e2->bei_nrdn.bv_len;
- if (rc == 0) rc = strncmp( e1->bei_nrdn.bv_val, e2->bei_nrdn.bv_val,
- e1->bei_nrdn.bv_len );
+ if (rc == 0) {
+ rc = strncmp( e1->bei_nrdn.bv_val, e2->bei_nrdn.bv_val,
+ e1->bei_nrdn.bv_len );
+ }
return rc;
}
bdb_entryinfo_add_internal(
struct bdb_info *bdb,
EntryInfo *ei,
- EntryInfo **res
-)
+ EntryInfo **res )
{
EntryInfo *ei2 = NULL;
#ifdef BDB_HIER
ei2->bei_rdn = ei->bei_rdn;
#endif
+#ifdef SLAP_ZONE_ALLOC
+ ei2->bei_bdb = bdb;
+#endif
/* Add to cache ID tree */
if (avl_insert( &bdb->bi_cache.c_idtree, ei2, bdb_id_cmp, avl_dup_error )) {
ei->bei_rdn.bv_val = NULL;
#endif
} else {
+ bdb->bi_cache.c_eiused++;
ber_dupbv( &ei2->bei_nrdn, &ei->bei_nrdn );
+
+ /* This is a new leaf node. But if parent had no kids, then it was
+ * a leaf and we would be decrementing that. So, only increment if
+ * the parent already has kids.
+ */
+ if ( ei->bei_parent->bei_kids || !ei->bei_parent->bei_id )
+ bdb->bi_cache.c_leaves++;
avl_insert( &ei->bei_parent->bei_kids, ei2, bdb_rdn_cmp,
avl_dup_error );
#ifdef BDB_HIER
Operation *op,
DB_TXN *txn,
struct berval *ndn,
- EntryInfo **res
-)
+ EntryInfo **res )
{
struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
EntryInfo ei, *eip, *ei2;
ptr = ndn->bv_val + ndn->bv_len - op->o_bd->be_nsuffix[0].bv_len;
ei.bei_nrdn.bv_val = ptr;
ei.bei_nrdn.bv_len = op->o_bd->be_nsuffix[0].bv_len;
+ /* Skip to next rdn if suffix is empty */
+ if ( ei.bei_nrdn.bv_len == 0 ) {
+ for (ptr = ei.bei_nrdn.bv_val - 2; ptr > ndn->bv_val
+ && !DN_SEPARATOR(*ptr); ptr--) /* empty */;
+ if ( ptr >= ndn->bv_val ) {
+ if (DN_SEPARATOR(*ptr)) ptr++;
+ ei.bei_nrdn.bv_len = ei.bei_nrdn.bv_val - ptr;
+ ei.bei_nrdn.bv_val = ptr;
+ }
+ }
eip = &bdb->bi_cache.c_dntree;
}
if ( !ei2 ) {
int len = ei.bei_nrdn.bv_len;
- ei.bei_nrdn.bv_len = ndn->bv_len - (ei.bei_nrdn.bv_val - ndn->bv_val);
+ if ( BER_BVISEMPTY( ndn )) {
+ *res = eip;
+ return LDAP_SUCCESS;
+ }
+
+ ei.bei_nrdn.bv_len = ndn->bv_len -
+ (ei.bei_nrdn.bv_val - ndn->bv_val);
bdb_cache_entryinfo_unlock( eip );
rc = bdb_dn2id( op, txn, &ei.bei_nrdn, &ei );
/* Advance to next lower RDN */
for (ptr = ei.bei_nrdn.bv_val - 2; ptr > ndn->bv_val
- && !DN_SEPARATOR(*ptr); ptr--);
+ && !DN_SEPARATOR(*ptr); ptr--) /* empty */;
if ( ptr >= ndn->bv_val ) {
if (DN_SEPARATOR(*ptr)) ptr++;
ei.bei_nrdn.bv_len = ei.bei_nrdn.bv_val - ptr - 1;
/* Walk up the tree from a child node, looking for an ID that's already
* been linked into the cache.
*/
-static int
+int
hdb_cache_find_parent(
Operation *op,
DB_TXN *txn,
+ u_int32_t locker,
ID id,
- EntryInfo **res
-)
+ EntryInfo **res )
{
struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
EntryInfo ei, eip, *ei2 = NULL, *ein = NULL, *eir = NULL;
- char ndn[SLAP_LDAPDN_MAXLEN];
- ID parent;
int rc;
- int addlru = 1;
+ int addlru = 0;
ei.bei_id = id;
ei.bei_kids = NULL;
+ ei.bei_ckids = 0;
for (;;) {
- rc = hdb_dn2id_parent( op, txn, &ei, &eip.bei_id );
+ rc = hdb_dn2id_parent( op, txn, locker, &ei, &eip.bei_id );
if ( rc ) break;
/* Save the previous node, if any */
ein->bei_kids = ei.bei_kids;
ein->bei_nrdn = ei.bei_nrdn;
ein->bei_rdn = ei.bei_rdn;
+ ein->bei_ckids = ei.bei_ckids;
+#ifdef SLAP_ZONE_ALLOC
+ ein->bei_bdb = bdb;
+#endif
+ ei.bei_ckids = 0;
/* This node is not fully connected yet */
ein->bei_state = CACHE_ENTRY_NOT_LINKED;
bdb_cache_entryinfo_lock( ein );
avl_insert( &ein->bei_kids, (caddr_t)ei2,
bdb_rdn_cmp, avl_dup_error );
+ ein->bei_ckids++;
bdb_cache_entryinfo_unlock( ein );
}
+ addlru = 0;
- if ( !eir ) {
- addlru = 0;
- }
}
+ if ( addlru ) {
+ ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_mutex );
+ LRU_ADD( &bdb->bi_cache, ein );
+ ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_mutex );
+ }
+ addlru = 1;
/* If this is the first time, save this node
* to be returned later.
} else {
ei2 = &bdb->bi_cache.c_dntree;
}
+ bdb->bi_cache.c_eiused++;
+ if ( ei2 && ( ei2->bei_kids || !ei2->bei_id ))
+ bdb->bi_cache.c_leaves++;
ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
/* Got the parent, link in and we're done. */
ein->bei_parent = ei2;
avl_insert( &ei2->bei_kids, (caddr_t)ein, bdb_rdn_cmp,
avl_dup_error);
+ ei2->bei_ckids++;
bdb_cache_entryinfo_unlock( ei2 );
bdb_cache_entryinfo_lock( eir );
}
ei.bei_kids = NULL;
ei.bei_id = eip.bei_id;
+ ei.bei_ckids = 1;
avl_insert( &ei.bei_kids, (caddr_t)ein, bdb_rdn_cmp,
avl_dup_error );
}
bdb_cache_lru_add(
struct bdb_info *bdb,
u_int32_t locker,
- EntryInfo *ei
-)
+ EntryInfo *ei )
{
- DB_LOCK lock;
+ DB_LOCK lock, *lockp;
+
+ if ( locker ) {
+ lockp = &lock;
+ } else {
+ lockp = NULL;
+ }
/* See if we're above the cache size limit */
if ( bdb->bi_cache.c_cursize > bdb->bi_cache.c_maxsize ) {
elprev = elru->bei_lruprev;
/* Too many probes, not enough idle, give up */
- if (i > 10) break;
+ if (i > 10)
+ break;
/* If we can successfully writelock it, then
* the object is idle.
*/
- if ( bdb_cache_entry_db_lock( bdb->bi_dbenv, locker, elru, 1, 1,
- &lock ) == 0 ) {
- /* If there's no entry, or this node is in
- * the process of linking into the cache,
- * skip it.
+ if ( bdb_cache_entry_db_lock( bdb->bi_dbenv,
+ bdb->bi_cache.c_locker, elru, 1, 1, lockp ) == 0 ) {
+
+ int stop = 0, decr = 0;
+
+ /* If this node is in the process of linking into the cache,
+ * or this node is being deleted, skip it.
*/
- if ( !elru->bei_e || (elru->bei_state & CACHE_ENTRY_NOT_LINKED) ) {
- bdb_cache_entry_db_unlock( bdb->bi_dbenv, &lock );
+ if ( elru->bei_state &
+ ( CACHE_ENTRY_NOT_LINKED | CACHE_ENTRY_DELETED )) {
+ bdb_cache_entry_db_unlock( bdb->bi_dbenv, lockp );
continue;
}
- LRU_DELETE( &bdb->bi_cache, elru );
- elru->bei_e->e_private = NULL;
- bdb_entry_return( elru->bei_e );
- elru->bei_e = NULL;
- bdb_cache_entry_db_unlock( bdb->bi_dbenv, &lock );
- --bdb->bi_cache.c_cursize;
- if (bdb->bi_cache.c_cursize < bdb->bi_cache.c_maxsize)
- break;
+ /* Free entry for this node if it's present */
+ if ( elru->bei_e ) {
+ elru->bei_e->e_private = NULL;
+#ifdef SLAP_ZONE_ALLOC
+ bdb_entry_return( bdb, elru->bei_e, elru->bei_zseq );
+#else
+ bdb_entry_return( elru->bei_e );
+#endif
+ elru->bei_e = NULL;
+ decr = 1;
+ }
+ /* ITS#4010 if we're in slapcat, and this node is a leaf
+ * node, free it.
+ *
+ * FIXME: we need to do this for slapd as well, (which is
+ * why we compute bi_cache.c_leaves now) but at the moment
+ * we can't because it causes unresolvable deadlocks.
+ */
+ if ( slapMode & SLAP_TOOL_READONLY ) {
+ if ( !elru->bei_kids ) {
+ /* This does LRU_DELETE for us */
+ bdb_cache_delete_internal( &bdb->bi_cache, elru, 0 );
+ bdb_cache_delete_cleanup( &bdb->bi_cache, elru );
+ }
+ /* Leave node on LRU list for a future pass */
+ } else {
+ LRU_DELETE( &bdb->bi_cache, elru );
+ }
+ bdb_cache_entry_db_unlock( bdb->bi_dbenv, lockp );
+
+ ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
+ if ( decr )
+ --bdb->bi_cache.c_cursize;
+ if (bdb->bi_cache.c_cursize <= bdb->bi_cache.c_maxsize)
+ stop = 1;
+ ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
+ if (stop) break;
}
}
}
ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_mutex );
}
+EntryInfo *
+bdb_cache_find_info(
+ struct bdb_info *bdb,
+ ID id )
+{
+ EntryInfo ei = { 0 },
+ *ei2;
+
+ ei.bei_id = id;
+
+ ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
+ ei2 = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
+ (caddr_t) &ei, bdb_id_cmp );
+ ldap_pvt_thread_rdwr_runlock( &bdb->bi_cache.c_rwlock );
+ return ei2;
+}
+
/*
* cache_find_id - find an entry in the cache, given id.
* The entry is locked for Read upon return. Call with islocked TRUE if
EntryInfo **eip,
int islocked,
u_int32_t locker,
- DB_LOCK *lock
-)
+ DB_LOCK *lock )
{
struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
Entry *ep = NULL;
- int rc = 0;
- EntryInfo ei;
- int lru_del = 0;
+ int rc = 0, load = 0;
+ EntryInfo ei = { 0 };
ei.bei_id = id;
+#ifdef SLAP_ZONE_ALLOC
+ slap_zh_rlock(bdb->bi_cache.c_zctx);
+#endif
/* If we weren't given any info, see if we have it already cached */
if ( !*eip ) {
-again: ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
+again: ldap_pvt_thread_rdwr_rlock( &bdb->bi_cache.c_rwlock );
*eip = (EntryInfo *) avl_find( bdb->bi_cache.c_idtree,
- (caddr_t) &ei, bdb_id_cmp );
+ (caddr_t) &ei, bdb_id_cmp );
if ( *eip ) {
/* If the lock attempt fails, the info is in use */
if ( ldap_pvt_thread_mutex_trylock(
if ( rc == 0 ) {
rc = bdb_cache_find_ndn( op, tid,
&ep->e_nname, eip );
- if ( *eip )
- islocked = 1;
+ if ( *eip ) islocked = 1;
if ( rc ) {
+#ifdef SLAP_ZONE_ALLOC
+ bdb_entry_return( bdb, ep, (*eip)->bei_zseq );
+#else
bdb_entry_return( ep );
+#endif
ep = NULL;
}
}
#else
- rc = hdb_cache_find_parent(op, tid, id, eip );
- if ( rc == 0 && *eip )
- islocked = 1;
+ rc = hdb_cache_find_parent(op, tid, locker, id, eip );
+ if ( rc == 0 && *eip ) islocked = 1;
#endif
}
if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
rc = DB_NOTFOUND;
} else {
- bdb_cache_entry_db_lock( bdb->bi_dbenv, locker,
- *eip, 0, 0, lock );
- if ( !(*eip)->bei_e ) {
- if (!ep) {
- rc = bdb_id2entry( op->o_bd, tid, id, &ep );
- }
- if ( rc == 0 ) {
- bdb_cache_entry_db_relock( bdb->bi_dbenv, locker,
- *eip, 1, 0, lock );
- /* Make sure no other modifier beat us to it */
- if ( (*eip)->bei_e ) {
- bdb_entry_return( ep );
- ep = NULL;
+ /* Make sure only one thread tries to load the entry */
+load1:
+#ifdef SLAP_ZONE_ALLOC
+ if ((*eip)->bei_e && !slap_zn_validate(
+ bdb->bi_cache.c_zctx, (*eip)->bei_e, (*eip)->bei_zseq)) {
+ (*eip)->bei_e = NULL;
+ (*eip)->bei_zseq = 0;
+ }
+#endif
+ if ( !(*eip)->bei_e && !((*eip)->bei_state & CACHE_ENTRY_LOADING)) {
+ load = 1;
+ (*eip)->bei_state |= CACHE_ENTRY_LOADING;
+ }
+ if ( islocked ) {
+ bdb_cache_entryinfo_unlock( *eip );
+ islocked = 0;
+ }
+ rc = bdb_cache_entry_db_lock( bdb->bi_dbenv, locker, *eip, 0, 0, lock );
+ if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
+ rc = DB_NOTFOUND;
+ bdb_cache_entry_db_unlock( bdb->bi_dbenv, lock );
+ } else if ( rc == 0 ) {
+ if ( load ) {
+ DB_TXN *ltid;
+ u_int32_t locker2 = locker;
+
+ /* We don't wrap entire read operations in txn's, but
+ * we need our cache entry lock and any DB page locks
+ * to be associated, in order for deadlock detection
+ * to work properly. So if we need to read from the DB,
+ * we use a long-lived per-thread txn for this step.
+ */
+ if ( !ep && !tid ) {
+ rc = bdb_txn_get( op, bdb->bi_dbenv, <id, 0 );
+ if ( ltid )
+ locker2 = TXN_ID( ltid );
} else {
+ ltid = tid;
+ }
+ /* Give up original read lock, obtain write lock with
+ * (possibly) new locker ID.
+ */
+ if ( rc == 0 ) {
+ rc = bdb_cache_entry_db_relock( bdb->bi_dbenv, locker2,
+ *eip, 1, 0, lock );
+ }
+ if ( rc == 0 && !ep) {
+ rc = bdb_id2entry( op->o_bd, ltid, id, &ep );
+ }
+ if ( rc == 0 ) {
ep->e_private = *eip;
#ifdef BDB_HIER
bdb_fix_dn( ep, 0 );
#endif
(*eip)->bei_e = ep;
+#ifdef SLAP_ZONE_ALLOC
+ (*eip)->bei_zseq = *((ber_len_t *)ep - 2);
+#endif
+ ep = NULL;
}
- bdb_cache_entry_db_relock( bdb->bi_dbenv, locker,
- *eip, 0, 0, lock );
- }
- } else {
- /* If we had the entry already, this item
- * is on the LRU list.
- */
- lru_del = 1;
+ (*eip)->bei_state ^= CACHE_ENTRY_LOADING;
+ if ( rc == 0 ) {
+ /* If we succeeded, downgrade back to a readlock. */
+ rc = bdb_cache_entry_db_relock( bdb->bi_dbenv, locker,
+ *eip, 0, 0, lock );
+ } else {
+ /* Otherwise, release the lock. */
+ bdb_cache_entry_db_unlock( bdb->bi_dbenv, lock );
+ }
+ if ( locker2 != locker ) {
+ /* If we're using the per-thread txn, release all
+ * of its page locks now.
+ */
+ DB_LOCKREQ list;
+ list.op = DB_LOCK_PUT_ALL;
+ list.obj = NULL;
+ bdb->bi_dbenv->lock_vec( bdb->bi_dbenv, locker2,
+ 0, &list, 1, NULL );
+ /* If this txn was deadlocked, we must abort it
+ * and invalidate this per-thread txn.
+ */
+ if ( rc == DB_LOCK_DEADLOCK ) {
+ bdb_txn_get( op, bdb->bi_dbenv, <id, 1 );
+ }
+ }
+ } else if ( !(*eip)->bei_e ) {
+ /* Some other thread is trying to load the entry,
+ * give it a chance to finish.
+ */
+ bdb_cache_entry_db_unlock( bdb->bi_dbenv, lock );
+ ldap_pvt_thread_yield();
+ bdb_cache_entryinfo_lock( *eip );
+ islocked = 1;
+ goto load1;
#ifdef BDB_HIER
- rc = bdb_fix_dn( (*eip)->bei_e, 1 );
- if ( rc ) {
- bdb_cache_entry_db_relock( bdb->bi_dbenv,
- locker, *eip, 1, 0, lock );
- /* check again in case other modifier did it already */
- if ( bdb_fix_dn( (*eip)->bei_e, 1 ) )
- rc = bdb_fix_dn( (*eip)->bei_e, 2 );
- bdb_cache_entry_db_relock( bdb->bi_dbenv,
- locker, *eip, 0, 0, lock );
- }
+ } else {
+ /* Check for subtree renames
+ */
+ rc = bdb_fix_dn( (*eip)->bei_e, 1 );
+ if ( rc ) {
+ bdb_cache_entry_db_relock( bdb->bi_dbenv,
+ locker, *eip, 1, 0, lock );
+ /* check again in case other modifier did it already */
+ if ( bdb_fix_dn( (*eip)->bei_e, 1 ) )
+ rc = bdb_fix_dn( (*eip)->bei_e, 2 );
+ bdb_cache_entry_db_relock( bdb->bi_dbenv,
+ locker, *eip, 0, 0, lock );
+ }
#endif
+ }
+
}
}
}
+ if ( islocked ) {
+ bdb_cache_entryinfo_unlock( *eip );
+ }
+ if ( ep ) {
+#ifdef SLAP_ZONE_ALLOC
+ bdb_entry_return( bdb, ep, (*eip)->bei_zseq );
+#else
+ bdb_entry_return( ep );
+#endif
+ }
if ( rc == 0 ) {
- /* set lru mutex */
+
+ if ( load ) {
+ ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
+ bdb->bi_cache.c_cursize++;
+ ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
+ }
+
ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_mutex );
- /* if entry is old, remove from old spot on LRU list */
- if ( lru_del ) {
- LRU_DELETE( &bdb->bi_cache, *eip );
+
+ /* If the LRU list has only one entry and this is it, it
+ * doesn't need to be added again.
+ */
+ if ( bdb->bi_cache.c_lruhead == bdb->bi_cache.c_lrutail &&
+ bdb->bi_cache.c_lruhead == *eip ) {
+ ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_mutex );
} else {
- /* if entry is new, bump cache size */
- bdb->bi_cache.c_cursize++;
+
+ /* if entry is on LRU list, remove from old spot */
+ if ( (*eip)->bei_lrunext || (*eip)->bei_lruprev ) {
+ LRU_DELETE( &bdb->bi_cache, *eip );
+ }
+ /* lru_mutex is unlocked for us */
+ bdb_cache_lru_add( bdb, locker, *eip );
}
- /* lru_mutex is unlocked for us */
- bdb_cache_lru_add( bdb, locker, *eip );
}
- if ( islocked ) {
- bdb_cache_entryinfo_unlock( *eip );
+#ifdef SLAP_ZONE_ALLOC
+ if (rc == 0 && (*eip)->bei_e) {
+ slap_zn_rlock(bdb->bi_cache.c_zctx, (*eip)->bei_e);
}
+ slap_zh_runlock(bdb->bi_cache.c_zctx);
+#endif
return rc;
}
bdb_cache_children(
Operation *op,
DB_TXN *txn,
- Entry *e
-)
+ Entry *e )
{
int rc;
}
rc = bdb_dn2id_children( op, txn, e );
if ( rc == DB_NOTFOUND ) {
- BEI(e)->bei_state |= CACHE_ENTRY_NO_KIDS;
+ BEI(e)->bei_state |= CACHE_ENTRY_NO_KIDS | CACHE_ENTRY_NO_GRANDKIDS;
}
return rc;
}
EntryInfo *eip,
Entry *e,
struct berval *nrdn,
- u_int32_t locker
-)
+ u_int32_t locker )
{
EntryInfo *new, ei;
- struct berval rdn = e->e_name;
+ DB_LOCK lock;
int rc;
+#ifdef BDB_HIER
+ struct berval rdn = e->e_name;
+#endif
ei.bei_id = e->e_id;
ei.bei_parent = eip;
ei.bei_nrdn = *nrdn;
+ ei.bei_lockpad = 0;
+
+ /* Lock this entry so that bdb_add can run to completion.
+ * It can only fail if BDB has run out of lock resources.
+ */
+ rc = bdb_cache_entry_db_lock( bdb->bi_dbenv, locker, &ei, 1, 0, &lock );
+ if ( rc ) {
+ bdb_cache_entryinfo_unlock( eip );
+ return rc;
+ }
+
#ifdef BDB_HIER
if ( nrdn->bv_len != e->e_nname.bv_len ) {
char *ptr = strchr( rdn.bv_val, ',' );
ber_dupbv( &ei.bei_rdn, &rdn );
if ( eip->bei_dkids ) eip->bei_dkids++;
#endif
+
rc = bdb_entryinfo_add_internal( bdb, &ei, &new );
+ /* bdb_csn_commit can cause this when adding the database root entry */
+ if ( new->bei_e ) {
+ new->bei_e->e_private = NULL;
+#ifdef SLAP_ZONE_ALLOC
+ bdb_entry_return( bdb, new->bei_e, new->bei_zseq );
+#else
+ bdb_entry_return( new->bei_e );
+#endif
+ }
new->bei_e = e;
e->e_private = new;
- new->bei_state = CACHE_ENTRY_NO_KIDS;
+ new->bei_state = CACHE_ENTRY_NO_KIDS | CACHE_ENTRY_NO_GRANDKIDS;
eip->bei_state &= ~CACHE_ENTRY_NO_KIDS;
+ if (eip->bei_parent) {
+ eip->bei_parent->bei_state &= ~CACHE_ENTRY_NO_GRANDKIDS;
+ }
+ bdb_cache_entryinfo_unlock( eip );
+
+ ++bdb->bi_cache.c_cursize;
+ ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
/* set lru mutex */
ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_mutex );
- ++bdb->bi_cache.c_cursize;
+
/* lru_mutex is unlocked for us */
bdb_cache_lru_add( bdb, locker, new );
- bdb_cache_entryinfo_unlock( eip );
- ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
return rc;
}
Attribute *newAttrs,
DB_ENV *env,
u_int32_t locker,
- DB_LOCK *lock
-)
+ DB_LOCK *lock )
{
EntryInfo *ei = BEI(e);
-
+ int rc;
/* Get write lock on data */
- bdb_cache_entry_db_relock( env, locker, ei, 1, 0, lock );
+ rc = bdb_cache_entry_db_relock( env, locker, ei, 1, 0, lock );
/* If we've done repeated mods on a cached entry, then e_attrs
* is no longer contiguous with the entry, and must be freed.
*/
- if ( (void *)e->e_attrs != (void *)(e+1) ) {
- attrs_free( e->e_attrs );
+ if ( ! rc ) {
+ if ( (void *)e->e_attrs != (void *)(e+1) ) {
+ attrs_free( e->e_attrs );
+ }
+ e->e_attrs = newAttrs;
}
- e->e_attrs = newAttrs;
-
- return 0;
+ return rc;
}
/*
*/
int
bdb_cache_modrdn(
+ struct bdb_info *bdb,
Entry *e,
struct berval *nrdn,
Entry *new,
EntryInfo *ein,
- DB_ENV *env,
u_int32_t locker,
- DB_LOCK *lock
-)
+ DB_LOCK *lock )
{
EntryInfo *ei = BEI(e), *pei;
+ int rc;
+#ifdef BDB_HIER
struct berval rdn;
- int rc = 0;
+#endif
/* Get write lock on data */
- bdb_cache_entry_db_relock( env, locker, ei, 1, 0, lock );
+ rc = bdb_cache_entry_db_relock( bdb->bi_dbenv, locker, ei, 1, 0, lock );
+ if ( rc ) return rc;
/* If we've done repeated mods on a cached entry, then e_attrs
* is no longer contiguous with the entry, and must be freed.
attrs_free( e->e_attrs );
}
e->e_attrs = new->e_attrs;
- if( e->e_nname.bv_val < e->e_bv.bv_val || e->e_nname.bv_val >
- e->e_bv.bv_val + e->e_bv.bv_len ) {
+ if( e->e_nname.bv_val < e->e_bv.bv_val ||
+ e->e_nname.bv_val > e->e_bv.bv_val + e->e_bv.bv_len )
+ {
ch_free(e->e_name.bv_val);
ch_free(e->e_nname.bv_val);
}
bdb_cache_entryinfo_lock( ein );
}
#ifdef BDB_HIER
- { int max = ei->bei_modrdns;
- /* Record the generation number of this change */
- for ( pei = ein; pei->bei_parent; pei = pei->bei_parent ) {
- if ( pei->bei_modrdns > max )
- max = pei->bei_modrdns;
- }
- ei->bei_modrdns = max + 1;
+ {
+ /* Record the generation number of this change */
+ ldap_pvt_thread_mutex_lock( &bdb->bi_modrdns_mutex );
+ bdb->bi_modrdns++;
+ ei->bei_modrdns = bdb->bi_modrdns;
+ ldap_pvt_thread_mutex_unlock( &bdb->bi_modrdns_mutex );
}
#endif
avl_insert( &ein->bei_kids, ei, bdb_rdn_cmp, avl_dup_error );
Entry *e,
DB_ENV *env,
u_int32_t locker,
- DB_LOCK *lock
-)
+ DB_LOCK *lock )
{
EntryInfo *ei = BEI(e);
int rc;
- assert( e->e_private );
+ assert( e->e_private != NULL );
/* Set this early, warn off any queriers */
ei->bei_state |= CACHE_ENTRY_DELETED;
bdb_cache_entryinfo_lock( ei );
/* Get write lock on the data */
- bdb_cache_entry_db_relock( env, locker, ei, 1, 0, lock );
-
- /* set cache write lock */
- ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
-
- /* Lock the parent's kids tree */
- bdb_cache_entryinfo_lock( ei->bei_parent );
+ rc = bdb_cache_entry_db_relock( env, locker, ei, 1, 0, lock );
+ if ( rc ) {
+ /* couldn't lock, undo and give up */
+ ei->bei_state ^= CACHE_ENTRY_DELETED;
+ bdb_cache_entryinfo_unlock( ei );
+ return rc;
+ }
-#ifdef NEW_LOGGING
- LDAP_LOG( CACHE, ENTRY,
- "bdb_cache_delete: delete %ld.\n", e->e_id, 0, 0 );
-#else
Debug( LDAP_DEBUG_TRACE, "====> bdb_cache_delete( %ld )\n",
e->e_id, 0, 0 );
-#endif
/* set lru mutex */
ldap_pvt_thread_mutex_lock( &cache->lru_mutex );
- rc = bdb_cache_delete_internal( cache, e->e_private );
- /* free lru mutex */
- ldap_pvt_thread_mutex_unlock( &cache->lru_mutex );
+
+ /* set cache write lock */
+ ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
+
+ rc = bdb_cache_delete_internal( cache, e->e_private, 1 );
/* free cache write lock */
ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
- bdb_cache_entryinfo_unlock( ei->bei_parent );
+
+ /* free lru mutex */
+ ldap_pvt_thread_mutex_unlock( &cache->lru_mutex );
/* Leave entry info locked */
void
bdb_cache_delete_cleanup(
Cache *cache,
- Entry *e
-)
+ EntryInfo *ei )
{
- EntryInfo *ei = BEI(e);
-
- ei->bei_e = NULL;
- e->e_private = NULL;
- bdb_entry_return( e );
+ if ( ei->bei_e ) {
+ ei->bei_e->e_private = NULL;
+#ifdef SLAP_ZONE_ALLOC
+ bdb_entry_return( ei->bei_bdb, ei->bei_e, ei->bei_zseq );
+#else
+ bdb_entry_return( ei->bei_e );
+#endif
+ ei->bei_e = NULL;
+ }
free( ei->bei_nrdn.bv_val );
ei->bei_nrdn.bv_val = NULL;
static int
bdb_cache_delete_internal(
Cache *cache,
- EntryInfo *e
-)
+ EntryInfo *e,
+ int decr )
{
int rc = 0; /* return code */
+ /* Lock the parent's kids tree */
+ bdb_cache_entryinfo_lock( e->bei_parent );
+
#ifdef BDB_HIER
e->bei_parent->bei_ckids--;
- if ( e->bei_parent->bei_dkids ) e->bei_parent->bei_dkids--;
+ if ( decr && e->bei_parent->bei_dkids ) e->bei_parent->bei_dkids--;
#endif
/* dn tree */
- if ( avl_delete( &e->bei_parent->bei_kids, (caddr_t) e, bdb_rdn_cmp ) == NULL )
+ if ( avl_delete( &e->bei_parent->bei_kids, (caddr_t) e, bdb_rdn_cmp )
+ == NULL )
{
rc = -1;
}
+ if ( e->bei_parent->bei_kids )
+ cache->c_leaves--;
/* id tree */
- if ( avl_delete( &cache->c_idtree, (caddr_t) e, bdb_id_cmp ) == NULL )
- {
+ if ( avl_delete( &cache->c_idtree, (caddr_t) e, bdb_id_cmp ) == NULL ) {
rc = -1;
}
- if (rc != 0) {
- return rc;
+ if ( rc == 0 ){
+ cache->c_eiused--;
+
+ /* lru */
+ LRU_DELETE( cache, e );
+ if ( e->bei_e ) cache->c_cursize--;
}
- /* lru */
- LRU_DELETE( cache, e );
- cache->c_cursize--;
+ bdb_cache_entryinfo_unlock( e->bei_parent );
- /*
- * flag entry to be freed later by a call to cache_return_entry()
- */
- e->bei_state |= CACHE_ENTRY_DELETED;
-
- return( 0 );
+ return( rc );
}
static void
}
if ( ei->bei_e ) {
ei->bei_e->e_private = NULL;
+#ifdef SLAP_ZONE_ALLOC
+ bdb_entry_return( ei->bei_bdb, ei->bei_e, ei->bei_zseq );
+#else
bdb_entry_return( ei->bei_e );
+#endif
}
bdb_cache_entryinfo_destroy( ei );
}
/* set lru mutex */
ldap_pvt_thread_mutex_lock( &cache->lru_mutex );
-#ifdef NEW_LOGGING
- LDAP_LOG( CACHE, ENTRY, "bdb_cache_release_all: enter\n", 0, 0, 0 );
-#else
Debug( LDAP_DEBUG_TRACE, "====> bdb_cache_release_all\n", 0, 0, 0 );
-#endif
avl_free( cache->c_dntree.bei_kids, NULL );
avl_free( cache->c_idtree, bdb_entryinfo_release );
+ for (;cache->c_eifree;cache->c_eifree = cache->c_lruhead) {
+ cache->c_lruhead = cache->c_eifree->bei_lrunext;
+ bdb_cache_entryinfo_destroy(cache->c_eifree);
+ }
+ cache->c_cursize = 0;
+ cache->c_eiused = 0;
+ cache->c_leaves = 0;
+ cache->c_idtree = NULL;
cache->c_lruhead = NULL;
cache->c_lrutail = NULL;
+ cache->c_dntree.bei_kids = NULL;
/* free lru mutex */
ldap_pvt_thread_mutex_unlock( &cache->lru_mutex );
}
#endif
+static void
+bdb_txn_free( void *key, void *data )
+{
+ DB_TXN *txn = data;
+ TXN_ABORT( txn );
+}
+
+/* Obtain a long-lived transaction for the current thread.
+ * If reset == 1, remove the current transaction. */
+static int
+bdb_txn_get( Operation *op, DB_ENV *env, DB_TXN **txn, int reset )
+{
+ int i, rc;
+ void *ctx, *data = NULL;
+
+ if ( slapMode & SLAP_TOOL_MODE ) {
+ *txn = NULL;
+ return 0;
+ }
+
+ /* If no op was provided, try to find the ctx anyway... */
+ if ( op ) {
+ ctx = op->o_threadctx;
+ } else {
+ ctx = ldap_pvt_thread_pool_context();
+ }
+
+ /* Shouldn't happen unless we're single-threaded */
+ if ( !ctx ) {
+ *txn = NULL;
+ return 0;
+ }
+
+ if ( reset ) {
+ TXN_ABORT( *txn );
+ return ldap_pvt_thread_pool_setkey( ctx, ((char *)env)+1, NULL, NULL );
+ }
+
+ if ( ldap_pvt_thread_pool_getkey( ctx, ((char *)env)+1, &data, NULL ) ||
+ data == NULL ) {
+ for ( i=0, rc=1; rc != 0 && i<4; i++ ) {
+ rc = TXN_BEGIN( env, NULL, txn, READ_TXN_FLAG );
+#if DB_VERSION_FULL == 0x04020034
+ if ( rc == EINVAL && READ_TXN_FLAG ) {
+ READ_TXN_FLAG = 0;
+ Debug( LDAP_DEBUG_ANY,
+ "bdb_txn_get: BerkeleyDB 4.2.52 library needs TXN patch!\n",
+ 0, 0, 0 );
+ i--;
+ continue;
+ }
+#endif
+ if (rc) ldap_pvt_thread_yield();
+ }
+ if ( rc != 0) {
+ return rc;
+ }
+ if ( ( rc = ldap_pvt_thread_pool_setkey( ctx, ((char *)env)+1,
+ *txn, bdb_txn_free ) ) ) {
+ TXN_ABORT( *txn );
+ Debug( LDAP_DEBUG_ANY, "bdb_txn_get: err %s(%d)\n",
+ db_strerror(rc), rc, 0 );
+
+ return rc;
+ }
+ } else {
+ *txn = data;
+ }
+ return 0;
+}
+
#ifdef BDB_REUSE_LOCKERS
static void
bdb_locker_id_free( void *key, void *data )
{
DB_ENV *env = key;
- int lockid = (int) data;
+ u_int32_t lockid = (long)data;
int rc;
-
rc = XLOCK_ID_FREE( env, lockid );
if ( rc == EINVAL ) {
DB_LOCKREQ lr;
-#ifdef NEW_LOGGING
- LDAP_LOG( BACK_BDB, ERR,
- "bdb_locker_id_free: %d err %s(%d)\n",
- lockid, db_strerror(rc), rc );
-#else
Debug( LDAP_DEBUG_ANY,
- "bdb_locker_id_free: %d err %s(%d)\n",
- lockid, db_strerror(rc), rc );
-#endif
- memset( &lr, 0, sizeof(lr) );
-
+ "bdb_locker_id_free: %lu err %s(%d)\n",
+ (unsigned long) lockid, db_strerror(rc), rc );
/* release all locks held by this locker. */
lr.op = DB_LOCK_PUT_ALL;
+ lr.obj = NULL;
env->lock_vec( env, lockid, 0, &lr, 1, NULL );
XLOCK_ID_FREE( env, lockid );
}
}
int
-bdb_locker_id( Operation *op, DB_ENV *env, int *locker )
+bdb_locker_id( Operation *op, DB_ENV *env, u_int32_t *locker )
{
- int i, rc, lockid;
+ int i, rc;
+ u_int32_t lockid;
void *data;
void *ctx;
if ( rc != 0) {
return rc;
}
- data = (void *)lockid;
+ data = (void *)((long)lockid);
if ( ( rc = ldap_pvt_thread_pool_setkey( ctx, env,
data, bdb_locker_id_free ) ) ) {
XLOCK_ID_FREE( env, lockid );
-#ifdef NEW_LOGGING
- LDAP_LOG( BACK_BDB, ERR, "bdb_locker_id: err %s(%d)\n",
- db_strerror(rc), rc, 0 );
-#else
Debug( LDAP_DEBUG_ANY, "bdb_locker_id: err %s(%d)\n",
db_strerror(rc), rc, 0 );
-#endif
return rc;
}
} else {
- lockid = (int)data;
+ lockid = (long)data;
}
*locker = lockid;
return 0;
}
-#endif
+#endif /* BDB_REUSE_LOCKERS */
void
bdb_cache_delete_entry(
DB_LOCK *lock )
{
ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
- if ( bdb_cache_entry_db_lock( bdb->bi_dbenv, locker, ei, 1, 1, lock ) == 0 ) {
+ if ( bdb_cache_entry_db_lock( bdb->bi_dbenv, bdb->bi_cache.c_locker, ei, 1, 1, lock ) == 0 )
+ {
if ( ei->bei_e && !(ei->bei_state & CACHE_ENTRY_NOT_LINKED )) {
LRU_DELETE( &bdb->bi_cache, ei );
ei->bei_e->e_private = NULL;
+#ifdef SLAP_ZONE_ALLOC
+ bdb_entry_return( bdb, ei->bei_e, ei->bei_zseq );
+#else
bdb_entry_return( ei->bei_e );
+#endif
ei->bei_e = NULL;
--bdb->bi_cache.c_cursize;
}