/* $OpenLDAP$ */
/* This work is part of OpenLDAP Software <http://www.openldap.org/>.
*
- * Copyright 2000-2005 The OpenLDAP Foundation.
+ * Copyright 2000-2006 The OpenLDAP Foundation.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
#include "back-bdb.h"
+#include "ldap_rq.h"
+
#ifdef BDB_HIER
#define bdb_cache_lru_add hdb_cache_lru_add
#endif
+static void bdb_cache_lru_add( struct bdb_info *bdb, EntryInfo *ei );
static int bdb_cache_delete_internal(Cache *cache, EntryInfo *e, int decr);
#ifdef LDAP_DEBUG
+#ifdef SLAPD_UNUSED
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 *
#else
int rc;
- if ( !lock ) return 0;
+ if ( !lock || lock->mode == DB_LOCK_NG ) return 0;
rc = LOCK_PUT ( env, lock );
return rc;
} \
(cache)->c_lruhead = (ei); \
(ei)->bei_lruprev = NULL; \
- if ( (cache)->c_lrutail == NULL ) { \
- (cache)->c_lrutail = (ei); \
+ if ( !ldap_pvt_thread_mutex_trylock( &(cache)->lru_tail_mutex )) { \
+ if ( (cache)->c_lrutail == NULL ) \
+ (cache)->c_lrutail = (ei); \
+ ldap_pvt_thread_mutex_unlock( &(cache)->lru_tail_mutex ); \
} \
} while(0)
} 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
/* 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 )
{
struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
EntryInfo ei, eip, *ei2 = NULL, *ein = NULL, *eir = NULL;
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_ckids++;
bdb_cache_entryinfo_unlock( ein );
}
+ addlru = 0;
- if ( !eir ) {
- addlru = 0;
- }
}
/* If this is the first time, save this node
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 );
+ if ( addlru ) {
+ ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_head_mutex );
+ bdb_cache_lru_add( bdb, ein );
+ }
+ addlru = 1;
+
/* Got the parent, link in and we're done. */
if ( ei2 ) {
bdb_cache_entryinfo_lock( ei2 );
}
#endif
-/* caller must have lru_mutex locked. mutex
+/* caller must have lru_head_mutex locked. mutex
* will be unlocked on return.
*/
static void
bdb_cache_lru_add(
struct bdb_info *bdb,
- u_int32_t locker,
EntryInfo *ei )
{
DB_LOCK lock, *lockp;
+ EntryInfo *elru, *elprev;
+ int count = 0;
- if ( locker ) {
+ LRU_ADD( &bdb->bi_cache, ei );
+ ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_head_mutex );
+
+ /* See if we're above the cache size limit */
+ if ( bdb->bi_cache.c_cursize <= bdb->bi_cache.c_maxsize )
+ return;
+
+ if ( bdb->bi_cache.c_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 ) {
- EntryInfo *elru, *elprev;
- int i = 0;
+ /* Don't bother if we can't get the lock */
+ if ( ldap_pvt_thread_mutex_trylock( &bdb->bi_cache.lru_tail_mutex ) )
+ return;
- /* Look for an unused entry to remove */
- for (elru = bdb->bi_cache.c_lrutail; elru; elru = elprev, i++ ) {
- elprev = elru->bei_lruprev;
+ /* Look for an unused entry to remove */
+ for (elru = bdb->bi_cache.c_lrutail; elru; elru = elprev ) {
+ elprev = elru->bei_lruprev;
- /* Too many probes, not enough idle, give up */
- if (i > 10)
- break;
+ /* If we can successfully writelock it, then
+ * the object is idle.
+ */
+ if ( bdb_cache_entry_db_lock( bdb->bi_dbenv,
+ bdb->bi_cache.c_locker, elru, 1, 1, lockp ) == 0 ) {
- /* If we can successfully writelock it, then
- * the object is idle.
- */
- if ( bdb_cache_entry_db_lock( bdb->bi_dbenv,
- bdb->bi_cache.c_locker, elru, 1, 1, lockp ) == 0 ) {
- int stop = 0;
- /* If there's no entry, or 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 | CACHE_ENTRY_DELETED ))) {
- bdb_cache_entry_db_unlock( bdb->bi_dbenv, lockp );
- continue;
- }
- LRU_DELETE( &bdb->bi_cache, elru );
+ /* If this node is in the process of linking into the cache,
+ * or this node is being deleted, skip it.
+ */
+ if ( elru->bei_state &
+ ( CACHE_ENTRY_NOT_LINKED | CACHE_ENTRY_DELETED )) {
+ bdb_cache_entry_db_unlock( bdb->bi_dbenv, lockp );
+ continue;
+ }
+ /* 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 );
bdb_entry_return( elru->bei_e );
#endif
elru->bei_e = NULL;
+ count++;
+ }
+ /* 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 );
+
+ if ( count >= bdb->bi_cache.c_minfree ) {
ldap_pvt_thread_rdwr_wlock( &bdb->bi_cache.c_rwlock );
- --bdb->bi_cache.c_cursize;
- if (bdb->bi_cache.c_cursize <= bdb->bi_cache.c_maxsize)
- stop = 1;
+ bdb->bi_cache.c_cursize -= count;
ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
- bdb_cache_entry_db_unlock( bdb->bi_dbenv, lockp );
- if (stop) break;
+ break;
}
}
}
- LRU_ADD( &bdb->bi_cache, ei );
- ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_mutex );
+
+ ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_tail_mutex );
}
EntryInfo *
/* See if the ID exists in the database; add it to the cache if so */
if ( !*eip ) {
#ifndef BDB_HIER
- rc = bdb_id2entry( op->o_bd, tid, id, &ep );
+ rc = bdb_id2entry( op->o_bd, tid, locker, id, &ep );
if ( rc == 0 ) {
rc = bdb_cache_find_ndn( op, tid,
&ep->e_nname, eip );
if ( *eip ) islocked = 1;
if ( rc ) {
+ ep->e_private = NULL;
#ifdef SLAP_ZONE_ALLOC
bdb_entry_return( bdb, ep, (*eip)->bei_zseq );
#else
}
}
#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 ) islocked = 1;
#endif
}
/* Ok, we found the info, do we have the entry? */
- if ( *eip && rc == 0 ) {
+ if ( rc == 0 ) {
if ( (*eip)->bei_state & CACHE_ENTRY_DELETED ) {
rc = DB_NOTFOUND;
} else {
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.
+ /* Give up original read lock, obtain write lock
*/
if ( rc == 0 ) {
- rc = bdb_cache_entry_db_relock( bdb->bi_dbenv, locker2,
+ rc = bdb_cache_entry_db_relock( bdb->bi_dbenv, locker,
*eip, 1, 0, lock );
}
if ( rc == 0 && !ep) {
- rc = bdb_id2entry( op->o_bd, ltid, id, &ep );
+ rc = bdb_id2entry( op->o_bd, tid, locker, id, &ep );
}
if ( rc == 0 ) {
ep->e_private = *eip;
/* 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_entryinfo_unlock( *eip );
}
if ( ep ) {
+ ep->e_private = NULL;
#ifdef SLAP_ZONE_ALLOC
bdb_entry_return( bdb, ep, (*eip)->bei_zseq );
#else
ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
}
- ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_mutex );
+ ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_head_mutex );
/* 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 );
+ ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_head_mutex );
} else {
-
/* if entry is on LRU list, remove from old spot */
if ( (*eip)->bei_lrunext || (*eip)->bei_lruprev ) {
+ ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_tail_mutex );
LRU_DELETE( &bdb->bi_cache, *eip );
+ ldap_pvt_thread_mutex_unlock( &bdb->bi_cache.lru_tail_mutex );
}
- /* lru_mutex is unlocked for us */
- bdb_cache_lru_add( bdb, locker, *eip );
+ /* lru_head_mutex is unlocked for us */
+ bdb_cache_lru_add( bdb, *eip );
}
}
#ifdef BDB_HIER
if ( nrdn->bv_len != e->e_nname.bv_len ) {
- char *ptr = strchr( rdn.bv_val, ',' );
+ char *ptr = ber_bvchr( &rdn, ',' );
+ assert( ptr != NULL );
rdn.bv_len = ptr - rdn.bv_val;
}
ber_dupbv( &ei.bei_rdn, &rdn );
ldap_pvt_thread_rdwr_wunlock( &bdb->bi_cache.c_rwlock );
/* set lru mutex */
- ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_mutex );
+ ldap_pvt_thread_mutex_lock( &bdb->bi_cache.lru_head_mutex );
- /* lru_mutex is unlocked for us */
- bdb_cache_lru_add( bdb, locker, new );
+ /* lru_head_mutex is unlocked for us */
+ bdb_cache_lru_add( bdb, new );
return rc;
}
avl_delete( &pei->bei_kids, (caddr_t) ei, bdb_rdn_cmp );
free( ei->bei_nrdn.bv_val );
ber_dupbv( &ei->bei_nrdn, nrdn );
+
+ if ( !pei->bei_kids )
+ pei->bei_state |= CACHE_ENTRY_NO_KIDS | CACHE_ENTRY_NO_GRANDKIDS;
+
#ifdef BDB_HIER
free( ei->bei_rdn.bv_val );
rdn = e->e_name;
if ( nrdn->bv_len != e->e_nname.bv_len ) {
- char *ptr = strchr(rdn.bv_val, ',');
+ char *ptr = ber_bvchr(&rdn, ',');
+ assert( ptr != NULL );
rdn.bv_len = ptr - rdn.bv_val;
}
ber_dupbv( &ei->bei_rdn, &rdn );
+ pei->bei_ckids--;
+ if ( pei->bei_dkids ) pei->bei_dkids--;
#endif
if (!ein) {
bdb_cache_entryinfo_unlock( pei );
bdb_cache_entryinfo_lock( ein );
}
+ /* parent now has kids */
+ if ( ein->bei_state & CACHE_ENTRY_NO_KIDS )
+ ein->bei_state ^= CACHE_ENTRY_NO_KIDS;
+
#ifdef BDB_HIER
+ /* parent might now have grandkids */
+ if ( ein->bei_state & CACHE_ENTRY_NO_GRANDKIDS &&
+ !(ei->bei_state & (CACHE_ENTRY_NO_KIDS)))
+ ein->bei_state ^= CACHE_ENTRY_NO_GRANDKIDS;
+
{
/* Record the generation number of this change */
ldap_pvt_thread_mutex_lock( &bdb->bi_modrdns_mutex );
ei->bei_modrdns = bdb->bi_modrdns;
ldap_pvt_thread_mutex_unlock( &bdb->bi_modrdns_mutex );
}
+ ein->bei_ckids++;
+ if ( ein->bei_dkids ) ein->bei_dkids++;
#endif
avl_insert( &ein->bei_kids, ei, bdb_rdn_cmp, avl_dup_error );
bdb_cache_entryinfo_unlock( ein );
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;
e->e_id, 0, 0 );
/* set lru mutex */
- ldap_pvt_thread_mutex_lock( &cache->lru_mutex );
+ ldap_pvt_thread_mutex_lock( &cache->lru_tail_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 );
+
/* free lru mutex */
- ldap_pvt_thread_mutex_unlock( &cache->lru_mutex );
+ ldap_pvt_thread_mutex_unlock( &cache->lru_tail_mutex );
/* Leave entry info locked */
{
int rc = 0; /* return code */
- /* set cache write lock */
- ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
-
/* Lock the parent's kids tree */
bdb_cache_entryinfo_lock( e->bei_parent );
{
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 ) {
rc = -1;
}
- if (rc != 0) {
- return rc;
- }
-
- cache->c_eiused--;
+ if ( rc == 0 ){
+ cache->c_eiused--;
- /* lru */
- LRU_DELETE( cache, e );
- if ( e->bei_e ) cache->c_cursize--;
+ /* lru */
+ LRU_DELETE( cache, e );
+ if ( e->bei_e ) cache->c_cursize--;
+ }
- /* free cache write lock */
- ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
bdb_cache_entryinfo_unlock( e->bei_parent );
- return( 0 );
+ return( rc );
}
static void
/* set cache write lock */
ldap_pvt_thread_rdwr_wlock( &cache->c_rwlock );
/* set lru mutex */
- ldap_pvt_thread_mutex_lock( &cache->lru_mutex );
+ ldap_pvt_thread_mutex_lock( &cache->lru_tail_mutex );
Debug( LDAP_DEBUG_TRACE, "====> bdb_cache_release_all\n", 0, 0, 0 );
}
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 );
+ ldap_pvt_thread_mutex_unlock( &cache->lru_tail_mutex );
/* free cache write lock */
ldap_pvt_thread_rdwr_wunlock( &cache->c_rwlock );
}
#ifdef LDAP_DEBUG
+#ifdef SLAPD_UNUSED
static void
bdb_lru_print( Cache *cache )
{
}
}
#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;
Debug( LDAP_DEBUG_ANY,
- "bdb_locker_id_free: %d err %s(%d)\n",
- lockid, db_strerror(rc), rc );
+ "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;
}
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 );
return rc;
}
} else {
- lockid = (int)data;
+ lockid = (long)data;
}
*locker = lockid;
return 0;
}
-#endif
+#endif /* BDB_REUSE_LOCKERS */
void
bdb_cache_delete_entry(