1 /* dn2id.c - routines to deal with the dn2id index */
3 /* This work is part of OpenLDAP Software <http://www.openldap.org/>.
5 * Copyright 2000-2011 The OpenLDAP Foundation.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted only as authorized by the OpenLDAP
12 * A copy of this license is available in the file LICENSE in the
13 * top-level directory of the distribution or, alternatively, at
14 * <http://www.OpenLDAP.org/license.html>.
20 #include <ac/string.h>
26 /* Management routines for a hierarchically structured database.
28 * Instead of a ldbm-style dn2id database, we use a hierarchical one. Each
29 * entry in this database is a struct diskNode, keyed by entryID and with
30 * the data containing the RDN and entryID of the node's children. We use
31 * a B-Tree with sorted duplicates to store all the children of a node under
32 * the same key. Also, the first item under the key contains the entry's own
33 * rdn and the ID of the node's parent, to allow bottom-up tree traversal as
34 * well as top-down. To keep this info first in the list, the high bit of all
35 * subsequent nrdnlen's is always set. This means we can only accomodate
36 * RDNs up to length 32767, but that's fine since full DNs are already
39 * The diskNode is a variable length structure. This definition is not
40 * directly usable for in-memory manipulation.
42 typedef struct diskNode {
43 unsigned char nrdnlen[2];
45 char rdn[1]; /* variable placement */
46 unsigned char entryID[sizeof(ID)]; /* variable placement */
49 /* Sort function for the sorted duplicate data items of a dn2id key.
50 * Sorts based on normalized RDN, in length order.
54 const MDB_val *usrkey,
61 un = (diskNode *)usrkey->mv_data;
62 cn = (diskNode *)curkey->mv_data;
64 /* data is not aligned, cannot compare directly */
65 rc = un->nrdnlen[0] - cn->nrdnlen[0];
67 rc = un->nrdnlen[1] - cn->nrdnlen[1];
70 nrlen = (un->nrdnlen[0] << 8) | un->nrdnlen[1];
71 return strncmp( un->nrdn, cn->nrdn, nrlen );
75 /* This function constructs a full DN for a given entry.
82 int rlen = 0, nrlen = 0;
89 /* count length of all DN components */
90 for ( ei = BEI(e); ei && ei->bei_id; ei=ei->bei_parent ) {
91 rlen += ei->bei_rdn.bv_len + 1;
92 nrlen += ei->bei_nrdn.bv_len + 1;
93 if (ei->bei_modrdns > max) max = ei->bei_modrdns;
96 /* See if the entry DN was invalidated by a subtree rename */
98 if ( BEI(e)->bei_modrdns >= max ) {
101 /* We found a mismatch, tell the caller to lock it */
102 if ( checkit == 1 ) {
105 /* checkit == 2. do the fix. */
106 free( e->e_name.bv_val );
107 free( e->e_nname.bv_val );
110 e->e_name.bv_len = rlen - 1;
111 e->e_nname.bv_len = nrlen - 1;
112 e->e_name.bv_val = ch_malloc(rlen);
113 e->e_nname.bv_val = ch_malloc(nrlen);
114 ptr = e->e_name.bv_val;
115 nptr = e->e_nname.bv_val;
116 for ( ei = BEI(e); ei && ei->bei_id; ei=ei->bei_parent ) {
117 ptr = lutil_strcopy(ptr, ei->bei_rdn.bv_val);
118 nptr = lutil_strcopy(nptr, ei->bei_nrdn.bv_val);
119 if ( ei->bei_parent ) {
124 BEI(e)->bei_modrdns = max;
125 if ( ptr > e->e_name.bv_val ) ptr[-1] = '\0';
126 if ( nptr > e->e_nname.bv_val ) nptr[-1] = '\0';
132 /* We add two elements to the DN2ID database - a data item under the parent's
133 * entryID containing the child's RDN and entryID, and an item under the
134 * child's entryID containing the parent's entryID.
143 struct mdb_info *mdb = (struct mdb_info *) op->o_bd->be_private;
144 MDB_dbi dbi = mdb->mi_dn2id->mdi_dbi;
151 Debug( LDAP_DEBUG_TRACE, "=> mdb_dn2id_add 0x%lx: \"%s\"\n",
152 e->e_id, e->e_ndn, 0 );
154 nrlen = dn_rdnlen( op->o_bd, &e->e_nname );
156 rlen = dn_rdnlen( op->o_bd, &e->e_name );
158 nrlen = e->e_nname.bv_len;
159 rlen = e->e_name.bv_len;
162 d = op->o_tmpalloc(sizeof(diskNode) + rlen + nrlen, op->o_tmpmemctx);
163 d->nrdnlen[1] = nrlen & 0xff;
164 d->nrdnlen[0] = (nrlen >> 8) | 0x80;
165 ptr = lutil_strncopy( d->nrdn, e->e_nname.bv_val, nrlen );
167 ptr = lutil_strncopy( ptr, e->e_name.bv_val, rlen );
169 memcpy( ptr, &e->e_id, sizeof( ID ));
171 key.mv_size = sizeof(ID);
176 /* Need to make dummy root node once. Subsequent attempts
177 * will fail harmlessly.
180 diskNode dummy = {{0, 0}, "", "", ""};
181 data.mv_data = &dummy;
182 data.mv_size = sizeof(diskNode);
184 mdb_put( txn, dbi, &key, &data, MDB_NODUPDATA );
188 data.mv_size = sizeof(diskNode) + rlen + nrlen;
190 rc = mdb_put( txn, dbi, &key, &data, MDB_NODUPDATA );
194 memcpy( ptr, &pid, sizeof( ID ));
195 d->nrdnlen[0] ^= 0x80;
197 rc = mdb_put( txn, dbi, &key, &data, MDB_NODUPDATA );
200 op->o_tmpfree( d, op->o_tmpmemctx );
201 Debug( LDAP_DEBUG_TRACE, "<= mdb_dn2id_add 0x%lx: %d\n", e->e_id, rc, 0 );
213 struct mdb_info *mdb = (struct mdb_info *) op->o_bd->be_private;
214 MDB_dbi dbi = mdb->mi_dn2id->mdi_dbi;
220 Debug( LDAP_DEBUG_TRACE, "=> mdb_dn2id_delete 0x%lx: \"%s\"\n",
221 e->e_id, e->e_ndn, 0 );
223 key.mv_size = sizeof(ID);
227 nrlen = dn_rdnlen( op->o_bd, &e->e_nname );
228 data.mv_size = sizeof(diskNode) + nrlen - sizeof(ID) - 1;
230 d = op->o_tmpalloc( data.mv_size, op->o_tmpmemctx );
231 d->nrdnlen[1] = nrlen & 0xff;
232 d->nrdnlen[0] = (nrlen >> 8) | 0x80;
233 memcpy( d->nrdn, e->e_nname.bv_val, nrlen );
236 /* Delete our ID from the parent's list */
237 rc = mdb_del( txn, dbi, &key, &data, MDB_DEL_DUP );
239 /* Delete our ID from the tree. With sorted duplicates, this
240 * will leave any child nodes still hanging around. This is OK
241 * for modrdn, which will add our info back in later.
245 d->nrdnlen[0] ^= 0x80;
246 rc = mdb_del( txn, dbi, &key, &data, MDB_DEL_DUP );
250 op->o_tmpfree( d, op->o_tmpmemctx );
252 Debug( LDAP_DEBUG_TRACE, "<= mdb_dn2id_delete 0x%lx: %d\n", e->e_id, rc, 0 );
256 /* return last found ID in *id if no match */
263 struct berval *matched )
265 struct mdb_info *mdb = (struct mdb_info *) op->o_bd->be_private;
267 MDB_dbi dbi = mdb->mi_dn2id->mdi_dbi;
272 char dn[SLAP_LDAPDN_MAXLEN];
273 unsigned char dlen[2];
277 Debug( LDAP_DEBUG_TRACE, "=> mdb_dn2id(\"%s\")\n", in->bv_val, 0, 0 );
288 matched->bv_val = dn + sizeof(dn) - 1;
290 *matched->bv_val-- = '\0';
293 nrlen = tmp.bv_len - op->o_bd->be_nsuffix[0].bv_len;
295 tmp.bv_len = op->o_bd->be_nsuffix[0].bv_len;
297 key.mv_size = sizeof(ID);
299 rc = mdb_cursor_open( txn, dbi, &cursor );
306 data.mv_size = sizeof(diskNode) + tmp.bv_len;
307 d = op->o_tmpalloc( data.mv_size, op->o_tmpmemctx );
308 d->nrdnlen[1] = tmp.bv_len & 0xff;
309 d->nrdnlen[0] = (tmp.bv_len >> 8) | 0x80;
310 ptr = lutil_strncopy( d->nrdn, tmp.bv_val, tmp.bv_len );
313 rc = mdb_cursor_get( cursor, &key, &data, MDB_GET_BOTH );
314 op->o_tmpfree( d, op->o_tmpmemctx );
315 if ( rc == MDB_NOTFOUND ) {
316 if ( matched && matched->bv_len ) {
317 ptr = op->o_tmpalloc( matched->bv_len+1, op->o_tmpmemctx );
318 strcpy( ptr, matched->bv_val );
319 matched->bv_val = ptr;
323 mdb_cursor_close( cursor );
326 ptr = (char *) data.mv_data + data.mv_size - sizeof(ID);
327 memcpy( &nid, ptr, sizeof(ID));
329 /* grab the non-normalized RDN */
333 rlen = data.mv_size - sizeof(diskNode) - tmp.bv_len;
334 matched->bv_len += rlen;
335 matched->bv_val -= rlen + 1;
336 ptr = lutil_strcopy( matched->bv_val, d->rdn + tmp.bv_len );
342 if ( tmp.bv_val > in->bv_val ) {
343 for (ptr = tmp.bv_val - 2; ptr > in->bv_val &&
344 !DN_SEPARATOR(*ptr); ptr--) /* empty */;
345 if ( ptr >= in->bv_val ) {
346 if (DN_SEPARATOR(*ptr)) ptr++;
347 tmp.bv_len = tmp.bv_val - ptr - 1;
358 Debug( LDAP_DEBUG_TRACE, "<= mdb_dn2id: get failed: %s (%d)\n",
359 mdb_strerror( rc ), rc, 0 );
361 Debug( LDAP_DEBUG_TRACE, "<= mdb_dn2id: got id=0x%lx\n",
376 struct mdb_info *mdb = (struct mdb_info *) op->o_bd->be_private;
377 DB *db = mdb->bi_dn2id->bdi_db;
386 key.size = sizeof(ID);
388 key.ulen = sizeof(ID);
389 key.flags = DB_DBT_USERMEM;
390 MDB_ID2DISK( ei->bei_id, &nid );
393 data.flags = DB_DBT_USERMEM;
395 rc = db->cursor( db, txn, &cursor, mdb->bi_db_opflags );
398 data.ulen = sizeof(diskNode) + (SLAP_LDAPDN_MAXLEN * 2);
399 d = op->o_tmpalloc( data.ulen, op->o_tmpmemctx );
402 rc = cursor->c_get( cursor, &key, &data, DB_SET );
404 if (d->nrdnlen[0] & 0x80) {
408 ptr = (char *) data.data + data.size - sizeof(ID);
409 MDB_DISK2ID( ptr, idp );
410 ei->bei_nrdn.bv_len = (d->nrdnlen[0] << 8) | d->nrdnlen[1];
411 ber_str2bv( d->nrdn, ei->bei_nrdn.bv_len, 1, &ei->bei_nrdn );
412 ei->bei_rdn.bv_len = data.size - sizeof(diskNode) -
414 ptr = d->nrdn + ei->bei_nrdn.bv_len + 1;
415 ber_str2bv( ptr, ei->bei_rdn.bv_len, 1, &ei->bei_rdn );
416 /* How many children does this node have? */
417 cursor->c_count( cursor, &dkids, 0 );
418 ei->bei_dkids = dkids;
421 cursor->c_close( cursor );
422 op->o_tmpfree( d, op->o_tmpmemctx );
433 struct mdb_info *mdb = (struct mdb_info *) op->o_bd->be_private;
434 MDB_dbi dbi = mdb->mi_dn2id->mdi_dbi;
440 key.mv_size = sizeof(ID);
444 rc = mdb_cursor_open( txn, dbi, &cursor );
447 rc = mdb_cursor_get( cursor, &key, &data, MDB_SET );
450 rc = mdb_cursor_count( cursor, &dkids );
452 if ( dkids < 2 ) rc = MDB_NOTFOUND;
455 mdb_cursor_close( cursor );
465 struct berval *nname )
467 struct mdb_info *mdb = (struct mdb_info *) op->o_bd->be_private;
468 MDB_dbi dbi = mdb->mi_dn2id->mdi_dbi;
472 char dn[SLAP_LDAPDN_MAXLEN], ndn[SLAP_LDAPDN_MAXLEN], *ptr;
476 key.mv_size = sizeof(ID);
478 rc = mdb_cursor_open( txn, dbi, &cursor );
490 rc = mdb_cursor_get( cursor, &key, &data, MDB_SET );
493 ptr += data.mv_size - sizeof(ID);
494 memcpy( &id, ptr, sizeof(ID) );
496 nrlen = (d->nrdnlen[0] << 8) | d->nrdnlen[1];
501 /* copy name and trailing NUL */
502 memcpy( nptr, d->nrdn, nrlen+1 );
503 rlen = data.mv_size - sizeof(diskNode) - nrlen;
504 memcpy( dptr, d->nrdn+nrlen+1, rlen+1 );
509 name->bv_len = dptr - dn;
510 nname->bv_len = nptr - ndn;
511 name->bv_val = op->o_tmpalloc( name->bv_len + 1, op->o_tmpmemctx );
512 nname->bv_val = op->o_tmpalloc( nname->bv_len + 1, op->o_tmpmemctx );
513 memcpy( name->bv_val, dn, name->bv_len );
514 name->bv_val[name->bv_len] = '\0';
515 memcpy( nname->bv_val, ndn, nname->bv_len );
516 nname->bv_val[nname->bv_len] = '\0';
518 mdb_cursor_close( cursor );
524 * We can't just use mdb_idl_fetch_key because
525 * 1 - our data items are longer than just an entry ID
526 * 2 - our data items are sorted alphabetically by nrdn, not by ID.
528 * We descend the tree recursively, so we define this cookie
529 * to hold our necessary state information. The mdb_dn2idl_internal
530 * function uses this cookie when calling itself.
533 struct dn2id_cookie {
534 struct mdb_info *mdb;
559 EntryInfo *ei = data;
562 mdb_idl_append_one( idl, ei->bei_id );
568 struct dn2id_cookie *cx
571 MDB_IDL_ZERO( cx->tmp );
573 if ( cx->mdb->bi_idl_cache_size ) {
574 char *ptr = ((char *)&cx->id)-1;
577 cx->key.size = sizeof(ID)+1;
578 if ( cx->prefix == DN_SUBTREE_PREFIX ) {
579 ID *ids = cx->depth ? cx->tmp : cx->ids;
581 cx->rc = mdb_idl_cache_get(cx->mdb, cx->db, &cx->key, ids);
582 if ( cx->rc == LDAP_SUCCESS ) {
584 mdb_idl_append( cx->ids, cx->tmp );
590 *ptr = DN_ONE_PREFIX;
591 cx->rc = mdb_idl_cache_get(cx->mdb, cx->db, &cx->key, cx->tmp);
592 if ( cx->rc == LDAP_SUCCESS ) {
595 if ( cx->rc == DB_NOTFOUND ) {
600 mdb_cache_entryinfo_lock( cx->ei );
602 /* If number of kids in the cache differs from on-disk, load
603 * up all the kids from the database
605 if ( cx->ei->bei_ckids+1 != cx->ei->bei_dkids ) {
607 db_recno_t dkids = cx->ei->bei_dkids;
608 ei.bei_parent = cx->ei;
610 /* Only one thread should load the cache */
611 while ( cx->ei->bei_state & CACHE_ENTRY_ONELEVEL ) {
612 mdb_cache_entryinfo_unlock( cx->ei );
613 ldap_pvt_thread_yield();
614 mdb_cache_entryinfo_lock( cx->ei );
615 if ( cx->ei->bei_ckids+1 == cx->ei->bei_dkids ) {
620 cx->ei->bei_state |= CACHE_ENTRY_ONELEVEL;
622 mdb_cache_entryinfo_unlock( cx->ei );
624 cx->rc = cx->db->cursor( cx->db, NULL, &cx->dbc,
625 cx->mdb->bi_db_opflags );
629 cx->data.data = &cx->dbuf;
630 cx->data.ulen = sizeof(ID);
631 cx->data.dlen = sizeof(ID);
632 cx->data.flags = DB_DBT_USERMEM | DB_DBT_PARTIAL;
634 /* The first item holds the parent ID. Ignore it. */
635 cx->key.data = &cx->nid;
636 cx->key.size = sizeof(ID);
637 cx->rc = cx->dbc->c_get( cx->dbc, &cx->key, &cx->data, DB_SET );
639 cx->dbc->c_close( cx->dbc );
643 /* If the on-disk count is zero we've never checked it.
647 cx->dbc->c_count( cx->dbc, &dkids, 0 );
648 cx->ei->bei_dkids = dkids;
651 cx->data.data = cx->buf;
652 cx->data.ulen = MDB_IDL_UM_SIZE * sizeof(ID);
653 cx->data.flags = DB_DBT_USERMEM;
656 /* Fetch the rest of the IDs in a loop... */
657 while ( (cx->rc = cx->dbc->c_get( cx->dbc, &cx->key, &cx->data,
658 DB_MULTIPLE | DB_NEXT_DUP )) == 0 ) {
662 DB_MULTIPLE_INIT( ptr, &cx->data );
664 DB_MULTIPLE_NEXT( ptr, &cx->data, j, len );
667 diskNode *d = (diskNode *)j;
670 MDB_DISK2ID( j + len - sizeof(ID), &ei.bei_id );
671 nrlen = ((d->nrdnlen[0] ^ 0x80) << 8) | d->nrdnlen[1];
672 ei.bei_nrdn.bv_len = nrlen;
673 /* nrdn/rdn are set in-place.
674 * mdb_cache_load will copy them as needed
676 ei.bei_nrdn.bv_val = d->nrdn;
677 ei.bei_rdn.bv_len = len - sizeof(diskNode)
678 - ei.bei_nrdn.bv_len;
679 ei.bei_rdn.bv_val = d->nrdn + ei.bei_nrdn.bv_len + 1;
680 mdb_idl_append_one( cx->tmp, ei.bei_id );
681 mdb_cache_load( cx->mdb, &ei, &ei2 );
687 cx->rc = cx->dbc->c_close( cx->dbc );
689 mdb_cache_entryinfo_lock( cx->ei );
690 cx->ei->bei_state &= ~CACHE_ENTRY_ONELEVEL;
691 mdb_cache_entryinfo_unlock( cx->ei );
696 /* The in-memory cache is in sync with the on-disk data.
697 * do we have any kids?
701 if ( cx->ei->bei_ckids > 0 ) {
702 /* Walk the kids tree; order is irrelevant since mdb_idl_sort
703 * will sort it later.
705 avl_apply( cx->ei->bei_kids, apply_func,
706 cx->tmp, -1, AVL_POSTORDER );
708 mdb_cache_entryinfo_unlock( cx->ei );
711 if ( !MDB_IDL_IS_RANGE( cx->tmp ) && cx->tmp[0] > 3 )
712 mdb_idl_sort( cx->tmp, cx->buf );
713 if ( cx->mdb->bi_idl_cache_max_size && !MDB_IDL_IS_ZERO( cx->tmp )) {
714 char *ptr = ((char *)&cx->id)-1;
716 cx->key.size = sizeof(ID)+1;
717 *ptr = DN_ONE_PREFIX;
718 mdb_idl_cache_put( cx->mdb, cx->db, &cx->key, cx->tmp, cx->rc );
722 if ( !MDB_IDL_IS_ZERO( cx->tmp )) {
723 if ( cx->prefix == DN_SUBTREE_PREFIX ) {
724 mdb_idl_append( cx->ids, cx->tmp );
726 if ( !(cx->ei->bei_state & CACHE_ENTRY_NO_GRANDKIDS)) {
728 EntryInfo *ei = cx->ei;
730 save = cx->op->o_tmpalloc( MDB_IDL_SIZEOF( cx->tmp ),
731 cx->op->o_tmpmemctx );
732 MDB_IDL_CPY( save, cx->tmp );
736 for ( cx->id = mdb_idl_first( save, &idcurs );
738 cx->id = mdb_idl_next( save, &idcurs )) {
741 if ( mdb_cache_find_id( cx->op, cx->txn, cx->id, &cx->ei,
746 if ( !( ei2->bei_state & CACHE_ENTRY_NO_KIDS )) {
747 MDB_ID2DISK( cx->id, &cx->nid );
748 mdb_dn2idl_internal( cx );
749 if ( !MDB_IDL_IS_ZERO( cx->tmp ))
752 mdb_cache_entryinfo_lock( ei2 );
754 mdb_cache_entryinfo_unlock( ei2 );
758 cx->op->o_tmpfree( save, cx->op->o_tmpmemctx );
760 mdb_cache_entryinfo_lock( ei );
761 ei->bei_state |= CACHE_ENTRY_NO_GRANDKIDS;
762 mdb_cache_entryinfo_unlock( ei );
765 /* Make sure caller knows it had kids! */
770 MDB_IDL_CPY( cx->ids, cx->tmp );
785 struct mdb_info *mdb = (struct mdb_info *)op->o_bd->be_private;
786 struct dn2id_cookie cx;
788 Debug( LDAP_DEBUG_TRACE, "=> mdb_dn2idl(\"%s\")\n",
791 #ifndef MDB_MULTIPLE_SUFFIXES
792 if ( op->ors_scope != LDAP_SCOPE_ONELEVEL &&
794 ( ei->bei_parent->bei_id == 0 && op->o_bd->be_suffix[0].bv_len )))
796 MDB_IDL_ALL( mdb, ids );
802 MDB_ID2DISK( cx.id, &cx.nid );
805 cx.db = cx.mdb->bi_dn2id->bdi_db;
806 cx.prefix = (op->ors_scope == LDAP_SCOPE_ONELEVEL) ?
807 DN_ONE_PREFIX : DN_SUBTREE_PREFIX;
810 cx.buf = stack + MDB_IDL_UM_SIZE;
816 if ( cx.prefix == DN_SUBTREE_PREFIX ) {
822 if ( cx.ei->bei_state & CACHE_ENTRY_NO_KIDS )
826 cx.key.ulen = sizeof(ID);
827 cx.key.size = sizeof(ID);
828 cx.key.flags = DB_DBT_USERMEM;
832 mdb_dn2idl_internal(&cx);
833 if ( cx.need_sort ) {
834 char *ptr = ((char *)&cx.id)-1;
835 if ( !MDB_IDL_IS_RANGE( cx.ids ) && cx.ids[0] > 3 )
836 mdb_idl_sort( cx.ids, cx.tmp );
838 cx.key.size = sizeof(ID)+1;
841 if ( cx.mdb->bi_idl_cache_max_size )
842 mdb_idl_cache_put( cx.mdb, cx.db, &cx.key, cx.ids, cx.rc );
845 if ( cx.rc == DB_NOTFOUND )
846 cx.rc = LDAP_SUCCESS;