int
hdb_dn2id_children(
Operation *op,
DB_TXN *txn,
Entry *e )
{
struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
DB *db = bdb->bi_dn2id->bdi_db;
DBT key, data;
DBC *cursor;
int rc;
ID id;
diskNode d;
DBTzero(&key);
key.size = sizeof(ID);
key.data = &e->e_id;
key.flags = DB_DBT_USERMEM;
BDB_ID2DISK( e->e_id, &id );
/* IDL cache is in host byte order */
if ( bdb->bi_idl_cache_size ) {
rc = bdb_idl_cache_get( bdb, db, &key, NULL );
if ( rc != LDAP_NO_SUCH_OBJECT ) {
return rc;
}
}
key.data = &id;
DBTzero(&data);
data.data = &d;
data.ulen = sizeof(d);
data.flags = DB_DBT_USERMEM | DB_DBT_PARTIAL;
data.dlen = sizeof(d);
rc = db->cursor( db, txn, &cursor, bdb->bi_db_opflags );
if ( rc ) return rc;
rc = cursor->c_get( cursor, &key, &data, DB_SET );
if ( rc == 0 ) {
db_recno_t dkids;
rc = cursor->c_count( cursor, &dkids, 0 );
if ( rc == 0 ) {
BEI(e)->bei_dkids = dkids;
if ( dkids < 2 ) rc = DB_NOTFOUND;
}
}
cursor->c_close( cursor );
return rc;
}
int
bdb_dn2id_children(
Operation *op,
DB_TXN *txn,
Entry *e )
{
DBT key, data;
struct bdb_info *bdb = (struct bdb_info *) op->o_bd->be_private;
DB *db = bdb->bi_dn2id->bdi_db;
ID id;
int rc;
Debug( LDAP_DEBUG_TRACE, "=> bdb_dn2id_children(\"%s\")\n",
e->e_nname.bv_val, 0, 0 );
DBTzero( &key );
key.size = e->e_nname.bv_len + 2;
key.data = op->o_tmpalloc( key.size, op->o_tmpmemctx );
((char *)key.data)[0] = DN_ONE_PREFIX;
AC_MEMCPY( &((char *)key.data)[1], e->e_nname.bv_val, key.size - 1 );
if ( bdb->bi_idl_cache_size ) {
rc = bdb_idl_cache_get( bdb, db, &key, NULL );
if ( rc != LDAP_NO_SUCH_OBJECT ) {
op->o_tmpfree( key.data, op->o_tmpmemctx );
return rc;
}
}
/* we actually could do a empty get... */
DBTzero( &data );
data.data = &id;
data.ulen = sizeof(id);
data.flags = DB_DBT_USERMEM;
data.doff = 0;
data.dlen = sizeof(id);
rc = db->get( db, txn, &key, &data, bdb->bi_db_opflags );
op->o_tmpfree( key.data, op->o_tmpmemctx );
Debug( LDAP_DEBUG_TRACE, "<= bdb_dn2id_children(\"%s\"): %s (%d)\n",
e->e_nname.bv_val,
rc == 0 ? "" : ( rc == DB_NOTFOUND ? "no " :
db_strerror(rc) ), rc );
return rc;
}
int
bdb_idl_fetch_key(
BackendDB *be,
DB *db,
DB_TXN *txn,
DBT *key,
ID *ids,
DBC **saved_cursor,
int get_flag )
{
struct bdb_info *bdb = (struct bdb_info *) be->be_private;
int rc;
DBT data, key2, *kptr;
DBC *cursor;
ID *i;
void *ptr;
size_t len;
int rc2;
int flags = bdb->bi_db_opflags | DB_MULTIPLE;
int opflag;
/* If using BerkeleyDB 4.0, the buf must be large enough to
* grab the entire IDL in one get(), otherwise BDB will leak
* resources on subsequent get's. We can safely call get()
* twice - once for the data, and once to get the DB_NOTFOUND
* result meaning there's no more data. See ITS#2040 for details.
* This bug is fixed in BDB 4.1 so a smaller buffer will work if
* stack space is too limited.
*
* configure now requires Berkeley DB 4.1.
*/
#if DB_VERSION_FULL < 0x04010000
# define BDB_ENOUGH 5
#else
/* We sometimes test with tiny IDLs, and BDB always wants buffers
* that are at least one page in size.
*/
# if BDB_IDL_DB_SIZE < 4096
# define BDB_ENOUGH 2048
# else
# define BDB_ENOUGH 1
# endif
#endif
ID buf[BDB_IDL_DB_SIZE*BDB_ENOUGH];
char keybuf[16];
Debug( LDAP_DEBUG_ARGS,
"bdb_idl_fetch_key: %s\n",
bdb_show_key( key, keybuf ), 0, 0 );
assert( ids != NULL );
if ( saved_cursor && *saved_cursor ) {
opflag = DB_NEXT;
} else if ( get_flag == LDAP_FILTER_GE ) {
opflag = DB_SET_RANGE;
} else if ( get_flag == LDAP_FILTER_LE ) {
opflag = DB_FIRST;
} else {
opflag = DB_SET;
}
/* only non-range lookups can use the IDL cache */
if ( bdb->bi_idl_cache_size && opflag == DB_SET ) {
rc = bdb_idl_cache_get( bdb, db, key, ids );
if ( rc != LDAP_NO_SUCH_OBJECT ) return rc;
}
DBTzero( &data );
data.data = buf;
data.ulen = sizeof(buf);
data.flags = DB_DBT_USERMEM;
/* If we're not reusing an existing cursor, get a new one */
if( opflag != DB_NEXT ) {
rc = db->cursor( db, txn, &cursor, bdb->bi_db_opflags );
if( rc != 0 ) {
Debug( LDAP_DEBUG_ANY, "=> bdb_idl_fetch_key: "
"cursor failed: %s (%d)\n", db_strerror(rc), rc, 0 );
return rc;
}
} else {
cursor = *saved_cursor;
}
/* If this is a LE lookup, save original key so we can determine
* when to stop. If this is a GE lookup, save the key since it
* will be overwritten.
*/
if ( get_flag == LDAP_FILTER_LE || get_flag == LDAP_FILTER_GE ) {
DBTzero( &key2 );
key2.flags = DB_DBT_USERMEM;
key2.ulen = sizeof(keybuf);
key2.data = keybuf;
key2.size = key->size;
AC_MEMCPY( keybuf, key->data, key->size );
kptr = &key2;
} else {
kptr = key;
}
len = key->size;
rc = cursor->c_get( cursor, kptr, &data, flags | opflag );
/* skip presence key on range inequality lookups */
while (rc == 0 && kptr->size != len) {
rc = cursor->c_get( cursor, kptr, &data, flags | DB_NEXT_NODUP );
}
/* If we're doing a LE compare and the new key is greater than
* our search key, we're done
*/
if (rc == 0 && get_flag == LDAP_FILTER_LE && memcmp( kptr->data,
key->data, key->size ) > 0 ) {
rc = DB_NOTFOUND;
}
if (rc == 0) {
i = ids;
while (rc == 0) {
u_int8_t *j;
DB_MULTIPLE_INIT( ptr, &data );
while (ptr) {
DB_MULTIPLE_NEXT(ptr, &data, j, len);
if (j) {
++i;
BDB_DISK2ID( j, i );
}
}
rc = cursor->c_get( cursor, key, &data, flags | DB_NEXT_DUP );
}
if ( rc == DB_NOTFOUND ) rc = 0;
ids[0] = i - ids;
/* On disk, a range is denoted by 0 in the first element */
if (ids[1] == 0) {
if (ids[0] != BDB_IDL_RANGE_SIZE) {
Debug( LDAP_DEBUG_ANY, "=> bdb_idl_fetch_key: "
"range size mismatch: expected %d, got %ld\n",
BDB_IDL_RANGE_SIZE, ids[0], 0 );
cursor->c_close( cursor );
return -1;
}
BDB_IDL_RANGE( ids, ids[2], ids[3] );
}
data.size = BDB_IDL_SIZEOF(ids);
}
if ( saved_cursor && rc == 0 ) {
if ( !*saved_cursor )
*saved_cursor = cursor;
rc2 = 0;
}
else
rc2 = cursor->c_close( cursor );
if (rc2) {
Debug( LDAP_DEBUG_ANY, "=> bdb_idl_fetch_key: "
"close failed: %s (%d)\n", db_strerror(rc2), rc2, 0 );
return rc2;
}
if( rc == DB_NOTFOUND ) {
return rc;
} else if( rc != 0 ) {
Debug( LDAP_DEBUG_ANY, "=> bdb_idl_fetch_key: "
"get failed: %s (%d)\n",
db_strerror(rc), rc, 0 );
return rc;
} else if ( data.size == 0 || data.size % sizeof( ID ) ) {
/* size not multiple of ID size */
Debug( LDAP_DEBUG_ANY, "=> bdb_idl_fetch_key: "
"odd size: expected %ld multiple, got %ld\n",
(long) sizeof( ID ), (long) data.size, 0 );
return -1;
} else if ( data.size != BDB_IDL_SIZEOF(ids) ) {
/* size mismatch */
Debug( LDAP_DEBUG_ANY, "=> bdb_idl_fetch_key: "
"get size mismatch: expected %ld, got %ld\n",
(long) ((1 + ids[0]) * sizeof( ID )), (long) data.size, 0 );
return -1;
}
if ( bdb->bi_idl_cache_max_size ) {
bdb_idl_cache_put( bdb, db, key, ids, rc );
}
return rc;
}
static int
hdb_dn2idl_internal(
struct dn2id_cookie *cx
)
{
BDB_IDL_ZERO( cx->tmp );
if ( cx->bdb->bi_idl_cache_size ) {
char *ptr = ((char *)&cx->id)-1;
cx->key.data = ptr;
cx->key.size = sizeof(ID)+1;
if ( cx->prefix == DN_SUBTREE_PREFIX ) {
ID *ids = cx->depth ? cx->tmp : cx->ids;
*ptr = cx->prefix;
cx->rc = bdb_idl_cache_get(cx->bdb, cx->db, &cx->key, ids);
if ( cx->rc == LDAP_SUCCESS ) {
if ( cx->depth ) {
bdb_idl_delete( cx->tmp, cx->id ); /* ITS#6983, drop our own ID */
bdb_idl_append( cx->ids, cx->tmp );
cx->need_sort = 1;
}
return cx->rc;
}
}
*ptr = DN_ONE_PREFIX;
cx->rc = bdb_idl_cache_get(cx->bdb, cx->db, &cx->key, cx->tmp);
if ( cx->rc == LDAP_SUCCESS ) {
goto gotit;
}
if ( cx->rc == DB_NOTFOUND ) {
return cx->rc;
}
}
bdb_cache_entryinfo_lock( cx->ei );
/* If number of kids in the cache differs from on-disk, load
* up all the kids from the database
*/
if ( cx->ei->bei_ckids+1 != cx->ei->bei_dkids ) {
EntryInfo ei;
db_recno_t dkids = cx->ei->bei_dkids;
ei.bei_parent = cx->ei;
/* Only one thread should load the cache */
while ( cx->ei->bei_state & CACHE_ENTRY_ONELEVEL ) {
bdb_cache_entryinfo_unlock( cx->ei );
ldap_pvt_thread_yield();
bdb_cache_entryinfo_lock( cx->ei );
if ( cx->ei->bei_ckids+1 == cx->ei->bei_dkids ) {
goto synced;
}
}
cx->ei->bei_state |= CACHE_ENTRY_ONELEVEL;
bdb_cache_entryinfo_unlock( cx->ei );
cx->rc = cx->db->cursor( cx->db, NULL, &cx->dbc,
cx->bdb->bi_db_opflags );
if ( cx->rc )
goto done_one;
cx->data.data = &cx->dbuf;
cx->data.ulen = sizeof(ID);
cx->data.dlen = sizeof(ID);
cx->data.flags = DB_DBT_USERMEM | DB_DBT_PARTIAL;
/* The first item holds the parent ID. Ignore it. */
cx->key.data = &cx->nid;
cx->key.size = sizeof(ID);
cx->rc = cx->dbc->c_get( cx->dbc, &cx->key, &cx->data, DB_SET );
if ( cx->rc ) {
cx->dbc->c_close( cx->dbc );
goto done_one;
}
/* If the on-disk count is zero we've never checked it.
* Count it now.
*/
if ( !dkids ) {
cx->dbc->c_count( cx->dbc, &dkids, 0 );
cx->ei->bei_dkids = dkids;
}
cx->data.data = cx->buf;
cx->data.ulen = BDB_IDL_UM_SIZE * sizeof(ID);
cx->data.flags = DB_DBT_USERMEM;
if ( dkids > 1 ) {
/* Fetch the rest of the IDs in a loop... */
while ( (cx->rc = cx->dbc->c_get( cx->dbc, &cx->key, &cx->data,
DB_MULTIPLE | DB_NEXT_DUP )) == 0 ) {
u_int8_t *j;
size_t len;
void *ptr;
DB_MULTIPLE_INIT( ptr, &cx->data );
while (ptr) {
DB_MULTIPLE_NEXT( ptr, &cx->data, j, len );
if (j) {
EntryInfo *ei2;
diskNode *d = (diskNode *)j;
short nrlen;
BDB_DISK2ID( j + len - sizeof(ID), &ei.bei_id );
nrlen = ((d->nrdnlen[0] ^ 0x80) << 8) | d->nrdnlen[1];
ei.bei_nrdn.bv_len = nrlen;
/* nrdn/rdn are set in-place.
* hdb_cache_load will copy them as needed
*/
ei.bei_nrdn.bv_val = d->nrdn;
ei.bei_rdn.bv_len = len - sizeof(diskNode)
- ei.bei_nrdn.bv_len;
ei.bei_rdn.bv_val = d->nrdn + ei.bei_nrdn.bv_len + 1;
bdb_idl_append_one( cx->tmp, ei.bei_id );
hdb_cache_load( cx->bdb, &ei, &ei2 );
}
}
}
}
cx->rc = cx->dbc->c_close( cx->dbc );
done_one:
bdb_cache_entryinfo_lock( cx->ei );
cx->ei->bei_state &= ~CACHE_ENTRY_ONELEVEL;
bdb_cache_entryinfo_unlock( cx->ei );
if ( cx->rc )
return cx->rc;
} else {
/* The in-memory cache is in sync with the on-disk data.
* do we have any kids?
*/
synced:
cx->rc = 0;
if ( cx->ei->bei_ckids > 0 ) {
/* Walk the kids tree; order is irrelevant since bdb_idl_sort
* will sort it later.
*/
avl_apply( cx->ei->bei_kids, apply_func,
cx->tmp, -1, AVL_POSTORDER );
}
bdb_cache_entryinfo_unlock( cx->ei );
}
if ( !BDB_IDL_IS_RANGE( cx->tmp ) && cx->tmp[0] > 3 )
bdb_idl_sort( cx->tmp, cx->buf );
if ( cx->bdb->bi_idl_cache_max_size && !BDB_IDL_IS_ZERO( cx->tmp )) {
char *ptr = ((char *)&cx->id)-1;
cx->key.data = ptr;
cx->key.size = sizeof(ID)+1;
*ptr = DN_ONE_PREFIX;
bdb_idl_cache_put( cx->bdb, cx->db, &cx->key, cx->tmp, cx->rc );
}
gotit:
if ( !BDB_IDL_IS_ZERO( cx->tmp )) {
if ( cx->prefix == DN_SUBTREE_PREFIX ) {
bdb_idl_append( cx->ids, cx->tmp );
cx->need_sort = 1;
if ( !(cx->ei->bei_state & CACHE_ENTRY_NO_GRANDKIDS)) {
ID *save, idcurs;
EntryInfo *ei = cx->ei;
int nokids = 1;
save = cx->op->o_tmpalloc( BDB_IDL_SIZEOF( cx->tmp ),
cx->op->o_tmpmemctx );
BDB_IDL_CPY( save, cx->tmp );
idcurs = 0;
cx->depth++;
for ( cx->id = bdb_idl_first( save, &idcurs );
cx->id != NOID;
cx->id = bdb_idl_next( save, &idcurs )) {
EntryInfo *ei2;
cx->ei = NULL;
if ( bdb_cache_find_id( cx->op, cx->txn, cx->id, &cx->ei,
ID_NOENTRY, NULL ))
continue;
if ( cx->ei ) {
ei2 = cx->ei;
if ( !( ei2->bei_state & CACHE_ENTRY_NO_KIDS )) {
BDB_ID2DISK( cx->id, &cx->nid );
hdb_dn2idl_internal( cx );
if ( !BDB_IDL_IS_ZERO( cx->tmp ))
nokids = 0;
}
bdb_cache_entryinfo_lock( ei2 );
ei2->bei_finders--;
bdb_cache_entryinfo_unlock( ei2 );
}
}
cx->depth--;
cx->op->o_tmpfree( save, cx->op->o_tmpmemctx );
if ( nokids ) {
bdb_cache_entryinfo_lock( ei );
ei->bei_state |= CACHE_ENTRY_NO_GRANDKIDS;
bdb_cache_entryinfo_unlock( ei );
}
}
/* Make sure caller knows it had kids! */
cx->tmp[0]=1;
cx->rc = 0;
} else {
BDB_IDL_CPY( cx->ids, cx->tmp );
}
}
return cx->rc;
}
