X-Git-Url: https://git.sur5r.net/?a=blobdiff_plain;f=servers%2Fslapd%2Foverlays%2Fsyncprov.c;h=7114eb90f895527c08fd6af0d03bc4a416b43e8e;hb=ee3f05000055619ec6d4eee48b57fa5cb6085127;hp=31da3e1a0d429f9d46223849a42ef5847bccf576;hpb=4775157666342f4bf63de21a7f653f03024dcc0e;p=openldap

diff --git a/servers/slapd/overlays/syncprov.c b/servers/slapd/overlays/syncprov.c
index 31da3e1a0d..7114eb90f8 100644
--- a/servers/slapd/overlays/syncprov.c
+++ b/servers/slapd/overlays/syncprov.c
@@ -2,7 +2,7 @@
 /* syncprov.c - syncrepl provider */
 /* This work is part of OpenLDAP Software <http://www.openldap.org/>.
  *
- * Copyright 2004-2009 The OpenLDAP Foundation.
+ * Copyright 2004-2011 The OpenLDAP Foundation.
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -48,6 +48,7 @@ typedef struct modtarget {
 /* A queued result of a persistent search */
 typedef struct syncres {
 	struct syncres *s_next;
+	Entry *s_e;
 	struct berval s_dn;
 	struct berval s_ndn;
 	struct berval s_uuid;
@@ -112,7 +113,9 @@ typedef struct slog_entry {
 } slog_entry;
 
 typedef struct sessionlog {
-	struct berval	sl_mincsn;
+	BerVarray	sl_mincsn;
+	int		*sl_sids;
+	int		sl_numcsns;
 	int		sl_num;
 	int		sl_size;
 	slog_entry *sl_head;
@@ -123,6 +126,7 @@ typedef struct sessionlog {
 /* The main state for this overlay */
 typedef struct syncprov_info_t {
 	syncops		*si_ops;
+	struct berval	si_contextdn;
 	BerVarray	si_ctxcsn;	/* ldapsync context */
 	int		*si_sids;
 	int		si_numcsns;
@@ -131,25 +135,37 @@ typedef struct syncprov_info_t {
 	int		si_numops;	/* number of ops since last checkpoint */
 	int		si_nopres;	/* Skip present phase */
 	int		si_usehint;	/* use reload hint */
+	int		si_active;	/* True if there are active mods */
+	int		si_dirty;	/* True if the context is dirty, i.e changes
+						 * have been made without updating the csn. */
 	time_t	si_chklast;	/* time of last checkpoint */
 	Avlnode	*si_mods;	/* entries being modified */
 	sessionlog	*si_logs;
 	ldap_pvt_thread_rdwr_t	si_csn_rwlock;
 	ldap_pvt_thread_mutex_t	si_ops_mutex;
 	ldap_pvt_thread_mutex_t	si_mods_mutex;
+	ldap_pvt_thread_mutex_t	si_resp_mutex;
 } syncprov_info_t;
 
 typedef struct opcookie {
 	slap_overinst *son;
 	syncmatches *smatches;
+	modtarget *smt;
+	Entry *se;
 	struct berval sdn;	/* DN of entry, for deletes */
 	struct berval sndn;
 	struct berval suuid;	/* UUID of entry */
 	struct berval sctxcsn;
-	short ssid;	/* sid of op csn */
+	short osid;	/* sid of op csn */
+	short rsid;	/* sid of relay */
 	short sreference;	/* Is the entry a reference? */
 } opcookie;
 
+typedef struct mutexint {
+	ldap_pvt_thread_mutex_t mi_mutex;
+	int mi_int;
+} mutexint;
+
 typedef struct fbase_cookie {
 	struct berval *fdn;	/* DN of a modified entry, for scope testing */
 	syncops *fss;	/* persistent search we're testing against */
@@ -211,10 +227,10 @@ syncprov_state_ctrl(
 
 	if ( ret < 0 ) {
 		Debug( LDAP_DEBUG_TRACE,
-			"slap_build_sync_ctrl: ber_flatten2 failed\n",
-			0, 0, 0 );
+			"slap_build_sync_ctrl: ber_flatten2 failed (%d)\n",
+			ret, 0, 0 );
 		send_ldap_error( op, rs, LDAP_OTHER, "internal error" );
-		return ret;
+		return LDAP_OTHER;
 	}
 
 	return LDAP_SUCCESS;
@@ -257,10 +273,10 @@ syncprov_done_ctrl(
 
 	if ( ret < 0 ) {
 		Debug( LDAP_DEBUG_TRACE,
-			"syncprov_done_ctrl: ber_flatten2 failed\n",
-			0, 0, 0 );
+			"syncprov_done_ctrl: ber_flatten2 failed (%d)\n",
+			ret, 0, 0 );
 		send_ldap_error( op, rs, LDAP_OTHER, "internal error" );
-		return ret;
+		return LDAP_OTHER;
 	}
 
 	return LDAP_SUCCESS;
@@ -324,10 +340,10 @@ syncprov_sendinfo(
 
 	if ( ret < 0 ) {
 		Debug( LDAP_DEBUG_TRACE,
-			"syncprov_sendinfo: ber_flatten2 failed\n",
-			0, 0, 0 );
+			"syncprov_sendinfo: ber_flatten2 failed (%d)\n",
+			ret, 0, 0 );
 		send_ldap_error( op, rs, LDAP_OTHER, "internal error" );
-		return ret;
+		return LDAP_OTHER;
 	}
 
 	rs->sr_rspoid = LDAP_SYNC_INFO;
@@ -572,7 +588,7 @@ findpres_cb( Operation *op, SlapReply *rs )
 }
 
 static int
-syncprov_findcsn( Operation *op, find_csn_t mode )
+syncprov_findcsn( Operation *op, find_csn_t mode, struct berval *csn )
 {
 	slap_overinst		*on = (slap_overinst *)op->o_bd->bd_info;
 	syncprov_info_t		*si = on->on_bi.bi_private;
@@ -580,8 +596,8 @@ syncprov_findcsn( Operation *op, find_csn_t mode )
 	slap_callback cb = {0};
 	Operation fop;
 	SlapReply frs = { REP_RESULT };
-	char buf[LDAP_LUTIL_CSNSTR_BUFSIZE + STRLENOF("(entryCSN<=)")];
-	char cbuf[LDAP_LUTIL_CSNSTR_BUFSIZE];
+	char buf[LDAP_PVT_CSNSTR_BUFSIZE + STRLENOF("(entryCSN<=)")];
+	char cbuf[LDAP_PVT_CSNSTR_BUFSIZE];
 	struct berval maxcsn;
 	Filter cf;
 	AttributeAssertion eq = ATTRIBUTEASSERTION_INIT;
@@ -645,15 +661,12 @@ again:
 		break;
 	case FIND_CSN:
 		if ( BER_BVISEMPTY( &cf.f_av_value )) {
-			cf.f_av_value = srs->sr_state.ctxcsn[0];
-			/* If there are multiple CSNs, use the smallest */
-			for ( i=1; i<srs->sr_state.numcsns; i++ ) {
-				if ( ber_bvcmp( &cf.f_av_value, &srs->sr_state.ctxcsn[i] )
-					> 0 ) {
-					cf.f_av_value = srs->sr_state.ctxcsn[i];
-				}
-			}
+			cf.f_av_value = *csn;
 		}
+		fop.o_dn = op->o_bd->be_rootdn;
+		fop.o_ndn = op->o_bd->be_rootndn;
+		fop.o_req_dn = op->o_bd->be_suffix[0];
+		fop.o_req_ndn = op->o_bd->be_nsuffix[0];
 		/* Look for exact match the first time */
 		if ( findcsn_retry ) {
 			cf.f_choice = LDAP_FILTER_EQUALITY;
@@ -722,6 +735,7 @@ again:
 			/* If we didn't find an exact match, then try for <= */
 			if ( findcsn_retry ) {
 				findcsn_retry = 0;
+				rs_reinit( &frs, REP_RESULT );
 				goto again;
 			}
 			rc = LDAP_NO_SUCH_OBJECT;
@@ -735,6 +749,36 @@ again:
 	return rc;
 }
 
+/* Should find a place to cache these */
+static mutexint *get_mutexint()
+{
+	mutexint *mi = ch_malloc( sizeof( mutexint ));
+	ldap_pvt_thread_mutex_init( &mi->mi_mutex );
+	mi->mi_int = 1;
+	return mi;
+}
+
+static void inc_mutexint( mutexint *mi )
+{
+	ldap_pvt_thread_mutex_lock( &mi->mi_mutex );
+	mi->mi_int++;
+	ldap_pvt_thread_mutex_unlock( &mi->mi_mutex );
+}
+
+/* return resulting counter */
+static int dec_mutexint( mutexint *mi )
+{
+	int i;
+	ldap_pvt_thread_mutex_lock( &mi->mi_mutex );
+	i = --mi->mi_int;
+	ldap_pvt_thread_mutex_unlock( &mi->mi_mutex );
+	if ( !i ) {
+		ldap_pvt_thread_mutex_destroy( &mi->mi_mutex );
+		ch_free( mi );
+	}
+	return i;
+}
+
 static void
 syncprov_free_syncop( syncops *so )
 {
@@ -742,7 +786,8 @@ syncprov_free_syncop( syncops *so )
 	GroupAssertion *ga, *gnext;
 
 	ldap_pvt_thread_mutex_lock( &so->s_mutex );
-	if ( --so->s_inuse > 0 ) {
+	/* already being freed, or still in use */
+	if ( !so->s_inuse || --so->s_inuse > 0 ) {
 		ldap_pvt_thread_mutex_unlock( &so->s_mutex );
 		return;
 	}
@@ -758,6 +803,12 @@ syncprov_free_syncop( syncops *so )
 	ch_free( so->s_base.bv_val );
 	for ( sr=so->s_res; sr; sr=srnext ) {
 		srnext = sr->s_next;
+		if ( sr->s_e ) {
+			if ( !dec_mutexint( sr->s_e->e_private )) {
+				sr->s_e->e_private = NULL;
+				entry_free( sr->s_e );
+			}
+		}
 		ch_free( sr );
 	}
 	ldap_pvt_thread_mutex_destroy( &so->s_mutex );
@@ -766,14 +817,11 @@ syncprov_free_syncop( syncops *so )
 
 /* Send a persistent search response */
 static int
-syncprov_sendresp( Operation *op, opcookie *opc, syncops *so,
-	Entry **e, int mode )
+syncprov_sendresp( Operation *op, opcookie *opc, syncops *so, int mode )
 {
-	slap_overinst *on = opc->son;
-
 	SlapReply rs = { REP_SEARCH };
 	LDAPControl *ctrls[2];
-	struct berval cookie, csns[2];
+	struct berval cookie = BER_BVNULL, csns[2];
 	Entry e_uuid = {0};
 	Attribute a_uuid = {0};
 
@@ -781,49 +829,55 @@ syncprov_sendresp( Operation *op, opcookie *opc, syncops *so,
 		return SLAPD_ABANDON;
 
 	ctrls[1] = NULL;
-	csns[0] = opc->sctxcsn;
-	BER_BVZERO( &csns[1] );
-	slap_compose_sync_cookie( op, &cookie, csns, so->s_rid, slap_serverID ? slap_serverID : -1 );
+	if ( !BER_BVISNULL( &opc->sctxcsn )) {
+		csns[0] = opc->sctxcsn;
+		BER_BVZERO( &csns[1] );
+		slap_compose_sync_cookie( op, &cookie, csns, so->s_rid, slap_serverID ? slap_serverID : -1 );
+	}
 
-	Debug( LDAP_DEBUG_SYNC, "syncprov_sendresp: cookie=%s\n", cookie.bv_val, 0, 0 );
+#ifdef LDAP_DEBUG
+	if ( so->s_sid > 0 ) {
+		Debug( LDAP_DEBUG_SYNC, "syncprov_sendresp: to=%03x, cookie=%s\n",
+			so->s_sid, cookie.bv_val ? cookie.bv_val : "", 0 );
+	} else {
+		Debug( LDAP_DEBUG_SYNC, "syncprov_sendresp: cookie=%s\n",
+			cookie.bv_val ? cookie.bv_val : "", 0, 0 );
+	}
+#endif
 
 	e_uuid.e_attrs = &a_uuid;
 	a_uuid.a_desc = slap_schema.si_ad_entryUUID;
 	a_uuid.a_nvals = &opc->suuid;
 	rs.sr_err = syncprov_state_ctrl( op, &rs, &e_uuid,
 		mode, ctrls, 0, 1, &cookie );
-	op->o_tmpfree( cookie.bv_val, op->o_tmpmemctx );
+	if ( !BER_BVISNULL( &cookie )) {
+		op->o_tmpfree( cookie.bv_val, op->o_tmpmemctx );
+	}
 
 	rs.sr_ctrls = ctrls;
-	op->o_bd->bd_info = (BackendInfo *)on->on_info;
+	rs.sr_entry = &e_uuid;
+	if ( mode == LDAP_SYNC_ADD || mode == LDAP_SYNC_MODIFY ) {
+		e_uuid = *opc->se;
+		e_uuid.e_private = NULL;
+	}
+
 	switch( mode ) {
 	case LDAP_SYNC_ADD:
-		rs.sr_entry = *e;
-		if ( rs.sr_entry->e_private )
-			rs.sr_flags = REP_ENTRY_MUSTRELEASE;
 		if ( opc->sreference && so->s_op->o_managedsait <= SLAP_CONTROL_IGNORED ) {
 			rs.sr_ref = get_entry_referrals( op, rs.sr_entry );
 			rs.sr_err = send_search_reference( op, &rs );
 			ber_bvarray_free( rs.sr_ref );
-			if ( !rs.sr_entry )
-				*e = NULL;
 			break;
 		}
 		/* fallthru */
 	case LDAP_SYNC_MODIFY:
-		rs.sr_entry = *e;
-		if ( rs.sr_entry->e_private )
-			rs.sr_flags = REP_ENTRY_MUSTRELEASE;
 		rs.sr_attrs = op->ors_attrs;
 		rs.sr_err = send_search_entry( op, &rs );
-		if ( !rs.sr_entry )
-			*e = NULL;
 		break;
 	case LDAP_SYNC_DELETE:
 		e_uuid.e_attrs = NULL;
 		e_uuid.e_name = opc->sdn;
 		e_uuid.e_nname = opc->sndn;
-		rs.sr_entry = &e_uuid;
 		if ( opc->sreference && so->s_op->o_managedsait <= SLAP_CONTROL_IGNORED ) {
 			struct berval bv = BER_BVNULL;
 			rs.sr_ref = &bv;
@@ -837,7 +891,14 @@ syncprov_sendresp( Operation *op, opcookie *opc, syncops *so,
 	}
 	/* In case someone else freed it already? */
 	if ( rs.sr_ctrls ) {
-		op->o_tmpfree( rs.sr_ctrls[0], op->o_tmpmemctx );
+		int i;
+		for ( i=0; rs.sr_ctrls[i]; i++ ) {
+			if ( rs.sr_ctrls[i] == ctrls[0] ) {
+				op->o_tmpfree( ctrls[0]->ldctl_value.bv_val, op->o_tmpmemctx );
+				ctrls[0]->ldctl_value.bv_val = NULL;
+				break;
+			}
+		}
 		rs.sr_ctrls = NULL;
 	}
 
@@ -853,7 +914,6 @@ syncprov_qplay( Operation *op, syncops *so )
 {
 	slap_overinst *on = LDAP_SLIST_FIRST(&so->s_op->o_extra)->oe_key;
 	syncres *sr;
-	Entry *e;
 	opcookie opc;
 	int rc = 0;
 
@@ -872,9 +932,9 @@ syncprov_qplay( Operation *op, syncops *so )
 		ldap_pvt_thread_mutex_unlock( &so->s_mutex );
 
 		if ( sr->s_mode == LDAP_SYNC_NEW_COOKIE ) {
-		    SlapReply rs = { REP_INTERMEDIATE };
+			SlapReply rs = { REP_INTERMEDIATE };
 
-		    rc = syncprov_sendinfo( op, &rs, LDAP_TAG_SYNC_NEW_COOKIE,
+			rc = syncprov_sendinfo( op, &rs, LDAP_TAG_SYNC_NEW_COOKIE,
 				&sr->s_csn, 0, NULL, 0 );
 		} else {
 			opc.sdn = sr->s_dn;
@@ -882,22 +942,15 @@ syncprov_qplay( Operation *op, syncops *so )
 			opc.suuid = sr->s_uuid;
 			opc.sctxcsn = sr->s_csn;
 			opc.sreference = sr->s_isreference;
-			e = NULL;
-
-			if ( sr->s_mode != LDAP_SYNC_DELETE ) {
-				rc = overlay_entry_get_ov( op, &opc.sndn, NULL, NULL, 0, &e, on );
-				if ( rc ) {
-					Debug( LDAP_DEBUG_SYNC, "syncprov_qplay: failed to get %s, "
-						"error (%d), ignoring...\n", opc.sndn.bv_val, rc, 0 );
-					ch_free( sr );
-					rc = 0;
-					continue;
-				}
-			}
-			rc = syncprov_sendresp( op, &opc, so, &e, sr->s_mode );
+			opc.se = sr->s_e;
 
-			if ( e ) {
-				overlay_entry_release_ov( op, e, 0, on );
+			rc = syncprov_sendresp( op, &opc, so, sr->s_mode );
+
+		}
+		if ( sr->s_e ) {
+			if ( !dec_mutexint( sr->s_e->e_private )) {
+				sr->s_e->e_private = NULL;
+				entry_free ( sr->s_e );
 			}
 		}
 
@@ -991,6 +1044,10 @@ syncprov_qresp( opcookie *opc, syncops *so, int mode )
 		srsize += cookie.bv_len + 1;
 	sr = ch_malloc( srsize );
 	sr->s_next = NULL;
+	sr->s_e = opc->se;
+	/* bump refcount on this entry */
+	if ( opc->se )
+		inc_mutexint( opc->se->e_private );
 	sr->s_dn.bv_val = (char *)(sr + 1);
 	sr->s_dn.bv_len = opc->sdn.bv_len;
 	sr->s_mode = mode;
@@ -1109,7 +1166,6 @@ syncprov_matchops( Operation *op, opcookie *opc, int saveit )
 
 	fbase_cookie fc;
 	syncops *ss, *sprev, *snext;
-	struct sync_cookie *scook;
 	Entry *e = NULL;
 	Attribute *a;
 	int rc;
@@ -1135,9 +1191,12 @@ syncprov_matchops( Operation *op, opcookie *opc, int saveit )
 		rc = overlay_entry_get_ov( op, fc.fdn, NULL, NULL, 0, &e, on );
 		/* If we're sending responses now, make a copy and unlock the DB */
 		if ( e && !saveit ) {
-			Entry *e2 = entry_dup( e );
+			if ( !opc->se ) {
+				opc->se = entry_dup( e );
+				opc->se->e_private = get_mutexint();
+			}
 			overlay_entry_release_ov( op, e, 0, on );
-			e = e2;
+			e = opc->se;
 		}
 		if ( rc ) {
 			op->o_bd = b0;
@@ -1145,6 +1204,13 @@ syncprov_matchops( Operation *op, opcookie *opc, int saveit )
 		}
 	} else {
 		e = op->ora_e;
+		if ( !saveit ) {
+			if ( !opc->se ) {
+				opc->se = entry_dup( e );
+				opc->se->e_private = get_mutexint();
+			}
+			e = opc->se;
+		}
 	}
 
 	if ( saveit || op->o_tag == LDAP_REQ_ADD ) {
@@ -1161,7 +1227,6 @@ syncprov_matchops( Operation *op, opcookie *opc, int saveit )
 		ber_dupbv_x( &opc->sndn, &e->e_nname, op->o_tmpmemctx );
 	}
 
-	scook = op->o_controls ? op->o_controls[slap_cids.sc_LDAPsync] : NULL;
 	ldap_pvt_thread_mutex_lock( &si->si_ops_mutex );
 	for (ss = si->si_ops, sprev = (syncops *)&si->si_ops; ss;
 		sprev = ss, ss=snext)
@@ -1176,16 +1241,16 @@ syncprov_matchops( Operation *op, opcookie *opc, int saveit )
 			continue;
 
 		/* Don't send ops back to the originator */
-		if ( opc->ssid > 0 && opc->ssid == ss->s_sid ) {
+		if ( opc->osid > 0 && opc->osid == ss->s_sid ) {
 			Debug( LDAP_DEBUG_SYNC, "syncprov_matchops: skipping original sid %03x\n",
-				opc->ssid, 0, 0 );
+				opc->osid, 0, 0 );
 			continue;
 		}
 
 		/* Don't send ops back to the messenger */
-		if ( scook && scook->sid > 0 && scook->sid == ss->s_sid ) {
+		if ( opc->rsid > 0 && opc->rsid == ss->s_sid ) {
 			Debug( LDAP_DEBUG_SYNC, "syncprov_matchops: skipping relayed sid %03x\n",
-				scook->sid, 0, 0 );
+				opc->rsid, 0, 0 );
 			continue;
 		}
 
@@ -1206,7 +1271,6 @@ syncprov_matchops( Operation *op, opcookie *opc, int saveit )
 			continue;
 		}
 
-
 		/* If we're sending results now, look for this op in old matches */
 		if ( !saveit ) {
 			syncmatches *old;
@@ -1230,17 +1294,30 @@ syncprov_matchops( Operation *op, opcookie *opc, int saveit )
 		}
 
 		if ( fc.fscope ) {
+			ldap_pvt_thread_mutex_lock( &ss->s_mutex );
 			op2 = *ss->s_op;
 			oh = *op->o_hdr;
 			oh.oh_conn = ss->s_op->o_conn;
 			oh.oh_connid = ss->s_op->o_connid;
+			op2.o_bd = op->o_bd->bd_self;
 			op2.o_hdr = &oh;
 			op2.o_extra = op->o_extra;
+			op2.o_callback = NULL;
+			if (ss->s_flags & PS_FIX_FILTER) {
+				/* Skip the AND/GE clause that we stuck on in front. We
+				   would lose deletes/mods that happen during the refresh
+				   phase otherwise (ITS#6555) */
+				op2.ors_filter = ss->s_op->ors_filter->f_and->f_next;
+			}
+			ldap_pvt_thread_mutex_unlock( &ss->s_mutex );
+			rc = test_filter( &op2, e, op2.ors_filter );
 		}
 
+		Debug( LDAP_DEBUG_TRACE, "syncprov_matchops: sid %03x fscope %d rc %d\n",
+			ss->s_sid, fc.fscope, rc );
+
 		/* check if current o_req_dn is in scope and matches filter */
-		if ( fc.fscope && test_filter( &op2, e, ss->s_op->ors_filter ) ==
-			LDAP_COMPARE_TRUE ) {
+		if ( fc.fscope && rc == LDAP_COMPARE_TRUE ) {
 			if ( saveit ) {
 				sm = op->o_tmpalloc( sizeof(syncmatches), op->o_tmpmemctx );
 				sm->sm_next = opc->smatches;
@@ -1273,9 +1350,17 @@ syncprov_matchops( Operation *op, opcookie *opc, int saveit )
 		if ( !SLAP_ISOVERLAY( op->o_bd )) {
 			op->o_bd = &db;
 		}
-		overlay_entry_release_ov( op, e, 0, on );
+		if ( saveit )
+			overlay_entry_release_ov( op, e, 0, on );
 		op->o_bd = b0;
 	}
+	if ( opc->se && !saveit ) {
+		if ( !dec_mutexint( opc->se->e_private )) {
+			opc->se->e_private = NULL;
+			entry_free( opc->se );
+			opc->se = NULL;
+		}
+	}
 	if ( freefdn ) {
 		op->o_tmpfree( fc.fdn->bv_val, op->o_tmpmemctx );
 	}
@@ -1290,7 +1375,12 @@ syncprov_op_cleanup( Operation *op, SlapReply *rs )
 	slap_overinst *on = opc->son;
 	syncprov_info_t		*si = on->on_bi.bi_private;
 	syncmatches *sm, *snext;
-	modtarget *mt, mtdummy;
+	modtarget *mt;
+
+	ldap_pvt_thread_mutex_lock( &si->si_ops_mutex );
+	if ( si->si_active )
+		si->si_active--;
+	ldap_pvt_thread_mutex_unlock( &si->si_ops_mutex );
 
 	for (sm = opc->smatches; sm; sm=snext) {
 		snext = sm->sm_next;
@@ -1299,26 +1389,23 @@ syncprov_op_cleanup( Operation *op, SlapReply *rs )
 	}
 
 	/* Remove op from lock table */
-	mtdummy.mt_op = op;
-	ldap_pvt_thread_mutex_lock( &si->si_mods_mutex );
-	mt = avl_find( si->si_mods, &mtdummy, sp_avl_cmp );
+	mt = opc->smt;
 	if ( mt ) {
-		modinst *mi = mt->mt_mods;
-
-		/* If there are more, promote the next one */
 		ldap_pvt_thread_mutex_lock( &mt->mt_mutex );
-		if ( mi->mi_next ) {
-			mt->mt_mods = mi->mi_next;
+		mt->mt_mods = mt->mt_mods->mi_next;
+		/* If there are more, promote the next one */
+		if ( mt->mt_mods ) {
 			mt->mt_op = mt->mt_mods->mi_op;
 			ldap_pvt_thread_mutex_unlock( &mt->mt_mutex );
 		} else {
-			avl_delete( &si->si_mods, mt, sp_avl_cmp );
 			ldap_pvt_thread_mutex_unlock( &mt->mt_mutex );
+			ldap_pvt_thread_mutex_lock( &si->si_mods_mutex );
+			avl_delete( &si->si_mods, mt, sp_avl_cmp );
+			ldap_pvt_thread_mutex_unlock( &si->si_mods_mutex );
 			ldap_pvt_thread_mutex_destroy( &mt->mt_mutex );
 			ch_free( mt );
 		}
 	}
-	ldap_pvt_thread_mutex_unlock( &si->si_mods_mutex );
 	if ( !BER_BVISNULL( &opc->suuid ))
 		op->o_tmpfree( opc->suuid.bv_val, op->o_tmpmemctx );
 	if ( !BER_BVISNULL( &opc->sndn ))
@@ -1332,14 +1419,16 @@ syncprov_op_cleanup( Operation *op, SlapReply *rs )
 }
 
 static void
-syncprov_checkpoint( Operation *op, SlapReply *rs, slap_overinst *on )
+syncprov_checkpoint( Operation *op, slap_overinst *on )
 {
 	syncprov_info_t *si = (syncprov_info_t *)on->on_bi.bi_private;
 	Modifications mod;
 	Operation opm;
-	SlapReply rsm = { 0 };
+	SlapReply rsm = {REP_RESULT};
 	slap_callback cb = {0};
 	BackendDB be;
+	BackendInfo *bi;
+
 #ifdef CHECK_CSN
 	Syntax *syn = slap_schema.si_ad_contextCSN->ad_type->sat_syntax;
 
@@ -1366,12 +1455,29 @@ syncprov_checkpoint( Operation *op, SlapReply *rs, slap_overinst *on )
 		be = *on->on_info->oi_origdb;
 		opm.o_bd = &be;
 	}
-	opm.o_req_dn = opm.o_bd->be_suffix[0];
-	opm.o_req_ndn = opm.o_bd->be_nsuffix[0];
+	opm.o_req_dn = si->si_contextdn;
+	opm.o_req_ndn = si->si_contextdn;
+	bi = opm.o_bd->bd_info;
 	opm.o_bd->bd_info = on->on_info->oi_orig;
 	opm.o_managedsait = SLAP_CONTROL_NONCRITICAL;
 	opm.o_no_schema_check = 1;
 	opm.o_bd->be_modify( &opm, &rsm );
+
+	if ( rsm.sr_err == LDAP_NO_SUCH_OBJECT &&
+		SLAP_SYNC_SUBENTRY( opm.o_bd )) {
+		const char	*text;
+		char txtbuf[SLAP_TEXT_BUFLEN];
+		size_t textlen = sizeof txtbuf;
+		Entry *e = slap_create_context_csn_entry( opm.o_bd, NULL );
+		rs_reinit( &rsm, REP_RESULT );
+		slap_mods2entry( &mod, &e, 0, 1, &text, txtbuf, textlen);
+		opm.ora_e = e;
+		opm.o_bd->be_add( &opm, &rsm );
+		if ( e == opm.ora_e )
+			be_entry_release_w( &opm, opm.ora_e );
+	}
+	opm.o_bd->bd_info = bi;
+
 	if ( mod.sml_next != NULL ) {
 		slap_mods_free( mod.sml_next, 1 );
 	}
@@ -1393,6 +1499,23 @@ syncprov_add_slog( Operation *op )
 
 	sl = si->si_logs;
 	{
+		if ( BER_BVISEMPTY( &op->o_csn ) ) {
+			/* During the syncrepl refresh phase we can receive operations
+			 * without a csn.  We cannot reliably determine the consumers
+			 * state with respect to such operations, so we ignore them and
+			 * wipe out anything in the log if we see them.
+			 */
+			ldap_pvt_thread_mutex_lock( &sl->sl_mutex );
+			while ( se = sl->sl_head ) {
+				sl->sl_head = se->se_next;
+				ch_free( se );
+			}
+			sl->sl_tail = NULL;
+			sl->sl_num = 0;
+			ldap_pvt_thread_mutex_unlock( &sl->sl_mutex );
+			return;
+		}
+
 		/* Allocate a record. UUIDs are not NUL-terminated. */
 		se = ch_malloc( sizeof( slog_entry ) + opc->suuid.bv_len + 
 			op->o_csn.bv_len + 1 );
@@ -1411,17 +1534,47 @@ syncprov_add_slog( Operation *op )
 
 		ldap_pvt_thread_mutex_lock( &sl->sl_mutex );
 		if ( sl->sl_head ) {
-			sl->sl_tail->se_next = se;
+			/* Keep the list in csn order. */
+			if ( ber_bvcmp( &sl->sl_tail->se_csn, &se->se_csn ) <= 0 ) {
+				sl->sl_tail->se_next = se;
+				sl->sl_tail = se;
+			} else {
+				slog_entry **sep;
+
+				for ( sep = &sl->sl_head; *sep; sep = &(*sep)->se_next ) {
+					if ( ber_bvcmp( &se->se_csn, &(*sep)->se_csn ) < 0 ) {
+						se->se_next = *sep;
+						*sep = se;
+						break;
+					}
+				}
+			}
 		} else {
 			sl->sl_head = se;
+			sl->sl_tail = se;
+			if ( !sl->sl_mincsn ) {
+				sl->sl_numcsns = 1;
+				sl->sl_mincsn = ch_malloc( 2*sizeof( struct berval ));
+				sl->sl_sids = ch_malloc( sizeof( int ));
+				sl->sl_sids[0] = se->se_sid;
+				ber_dupbv( sl->sl_mincsn, &se->se_csn );
+				BER_BVZERO( &sl->sl_mincsn[1] );
+			}
 		}
-		sl->sl_tail = se;
 		sl->sl_num++;
 		while ( sl->sl_num > sl->sl_size ) {
+			int i, j;
 			se = sl->sl_head;
 			sl->sl_head = se->se_next;
-			strcpy( sl->sl_mincsn.bv_val, se->se_csn.bv_val );
-			sl->sl_mincsn.bv_len = se->se_csn.bv_len;
+			for ( i=0; i<sl->sl_numcsns; i++ )
+				if ( sl->sl_sids[i] >= se->se_sid )
+					break;
+			if  ( i == sl->sl_numcsns || sl->sl_sids[i] != se->se_sid ) {
+				slap_insert_csn_sids( (struct sync_cookie *)sl,
+					i, se->se_sid, &se->se_csn );
+			} else {
+				ber_bvreplace( &sl->sl_mincsn[i], &se->se_csn );
+			}
 			ch_free( se );
 			sl->sl_num--;
 		}
@@ -1447,7 +1600,7 @@ syncprov_playlog( Operation *op, SlapReply *rs, sessionlog *sl,
 	slap_overinst		*on = (slap_overinst *)op->o_bd->bd_info;
 	slog_entry *se;
 	int i, j, ndel, num, nmods, mmods;
-	char cbuf[LDAP_LUTIL_CSNSTR_BUFSIZE];
+	char cbuf[LDAP_PVT_CSNSTR_BUFSIZE];
 	BerVarray uuids;
 	struct berval delcsn[2];
 
@@ -1506,6 +1659,8 @@ syncprov_playlog( Operation *op, SlapReply *rs, sessionlog *sl,
 			delcsn[0].bv_len = se->se_csn.bv_len;
 			delcsn[0].bv_val[delcsn[0].bv_len] = '\0';
 		} else {
+			if ( se->se_tag == LDAP_REQ_ADD )
+				continue;
 			nmods++;
 			j = num - nmods;
 		}
@@ -1546,7 +1701,6 @@ syncprov_playlog( Operation *op, SlapReply *rs, sessionlog *sl,
 
 	if ( mmods ) {
 		Operation fop;
-		SlapReply frs = { REP_RESULT };
 		int rc;
 		Filter mf, af;
 		AttributeAssertion eq = ATTRIBUTEASSERTION_INIT;
@@ -1576,18 +1730,19 @@ syncprov_playlog( Operation *op, SlapReply *rs, sessionlog *sl,
 		fop.o_bd->bd_info = (BackendInfo *)on->on_info;
 
 		for ( i=ndel; i<num; i++ ) {
-			if ( uuids[i].bv_len == 0 ) continue;
+		  if ( uuids[i].bv_len != 0 ) {
+			SlapReply frs = { REP_RESULT };
 
 			mf.f_av_value = uuids[i];
 			cb.sc_private = NULL;
 			fop.ors_slimit = 1;
-			frs.sr_nentries = 0;
 			rc = fop.o_bd->be_search( &fop, &frs );
 
 			/* If entry was not found, add to delete list */
 			if ( !cb.sc_private ) {
 				uuids[ndel++] = uuids[i];
 			}
+		  }
 		}
 		fop.o_bd->bd_info = (BackendInfo *)on;
 	}
@@ -1622,67 +1777,17 @@ syncprov_op_response( Operation *op, SlapReply *rs )
 	if ( rs->sr_err == LDAP_SUCCESS )
 	{
 		struct berval maxcsn;
-		char cbuf[LDAP_LUTIL_CSNSTR_BUFSIZE];
+		char cbuf[LDAP_PVT_CSNSTR_BUFSIZE];
 		int do_check = 0, have_psearches, foundit, csn_changed = 0;
 
+		ldap_pvt_thread_mutex_lock( &si->si_resp_mutex );
+
 		/* Update our context CSN */
 		cbuf[0] = '\0';
 		maxcsn.bv_val = cbuf;
 		maxcsn.bv_len = sizeof(cbuf);
 		ldap_pvt_thread_rdwr_wlock( &si->si_csn_rwlock );
 
-#if 0
-		if ( op->o_dont_replicate &&
-				op->orm_modlist->sml_op == LDAP_MOD_REPLACE &&
-				op->orm_modlist->sml_desc == slap_schema.si_ad_contextCSN ) {
-			/* Catch contextCSN updates from syncrepl. We have to look at
-			 * all the attribute values, as there may be more than one csn
-			 * that changed, and only one can be passed in the csn queue.
-			 */
-			Modifications *mod = op->orm_modlist;
-			int i, j, sid;
-
-			for ( i=0; i<mod->sml_numvals; i++ ) {
-				sid = slap_parse_csn_sid( &mod->sml_values[i] );
-
-				for ( j=0; j<si->si_numcsns; j++ ) {
-					if ( sid == si->si_sids[j] ) {
-						if ( ber_bvcmp( &mod->sml_values[i], &si->si_ctxcsn[j] ) > 0 ) {
-							ber_bvreplace( &si->si_ctxcsn[j], &mod->sml_values[i] );
-							csn_changed = 1;
-						}
-						break;
-					}
-				}
-
-				if ( j == si->si_numcsns ) {
-					value_add_one( &si->si_ctxcsn, &mod->sml_values[i] );
-					si->si_numcsns++;
-					si->si_sids = ch_realloc( si->si_sids, si->si_numcsns *
-						sizeof(int));
-					si->si_sids[j] = sid;
-					csn_changed = 1;
-				}
-			}
-			ldap_pvt_thread_rdwr_wunlock( &si->si_csn_rwlock );
-
-			if ( csn_changed ) {
-				ldap_pvt_thread_mutex_lock( &si->si_ops_mutex );
-				have_psearches = ( si->si_ops != NULL );
-				ldap_pvt_thread_mutex_unlock( &si->si_ops_mutex );
-
-				if ( have_psearches ) {
-					for ( sm = opc->smatches; sm; sm=sm->sm_next ) {
-						if ( sm->sm_op->s_op->o_abandon )
-							continue;
-						syncprov_qresp( opc, sm->sm_op, LDAP_SYNC_NEW_COOKIE );
-					}
-				}
-			}
-			return SLAP_CB_CONTINUE;
-		}
-#endif
-
 		slap_get_commit_csn( op, &maxcsn, &foundit );
 		if ( BER_BVISEMPTY( &maxcsn ) && SLAP_GLUE_SUBORDINATE( op->o_bd )) {
 			/* syncrepl queues the CSN values in the db where
@@ -1705,6 +1810,8 @@ syncprov_op_response( Operation *op, SlapReply *rs )
 #endif
 			sid = slap_parse_csn_sid( &maxcsn );
 			for ( i=0; i<si->si_numcsns; i++ ) {
+				if ( sid < si->si_sids[i] )
+					break;
 				if ( sid == si->si_sids[i] ) {
 					if ( ber_bvcmp( &maxcsn, &si->si_ctxcsn[i] ) > 0 ) {
 						ber_bvreplace( &si->si_ctxcsn[i], &maxcsn );
@@ -1714,47 +1821,100 @@ syncprov_op_response( Operation *op, SlapReply *rs )
 				}
 			}
 			/* It's a new SID for us */
-			if ( i == si->si_numcsns ) {
-				value_add_one( &si->si_ctxcsn, &maxcsn );
+			if ( i == si->si_numcsns || sid != si->si_sids[i] ) {
+				slap_insert_csn_sids((struct sync_cookie *)&(si->si_ctxcsn),
+					i, sid, &maxcsn );
 				csn_changed = 1;
-				si->si_numcsns++;
-				si->si_sids = ch_realloc( si->si_sids, si->si_numcsns *
-					sizeof(int));
-				si->si_sids[i] = sid;
 			}
-		} else if ( !foundit ) {
-			/* internal ops that aren't meant to be replicated */
-			ldap_pvt_thread_rdwr_wunlock( &si->si_csn_rwlock );
-			return SLAP_CB_CONTINUE;
 		}
 
 		/* Don't do any processing for consumer contextCSN updates */
 		if ( op->o_dont_replicate ) {
+			if ( op->o_tag == LDAP_REQ_MODIFY &&
+				op->orm_modlist->sml_op == LDAP_MOD_REPLACE &&
+				op->orm_modlist->sml_desc == slap_schema.si_ad_contextCSN ) {
+			/* Catch contextCSN updates from syncrepl. We have to look at
+			 * all the attribute values, as there may be more than one csn
+			 * that changed, and only one can be passed in the csn queue.
+			 */
+			Modifications *mod = op->orm_modlist;
+			unsigned i;
+			int j, sid;
+
+			for ( i=0; i<mod->sml_numvals; i++ ) {
+				sid = slap_parse_csn_sid( &mod->sml_values[i] );
+				for ( j=0; j<si->si_numcsns; j++ ) {
+					if ( sid < si->si_sids[j] )
+						break;
+					if ( sid == si->si_sids[j] ) {
+						if ( ber_bvcmp( &mod->sml_values[i], &si->si_ctxcsn[j] ) > 0 ) {
+							ber_bvreplace( &si->si_ctxcsn[j], &mod->sml_values[i] );
+							csn_changed = 1;
+						}
+						break;
+					}
+				}
+
+				if ( j == si->si_numcsns || sid != si->si_sids[j] ) {
+					slap_insert_csn_sids( (struct sync_cookie *)&si->si_ctxcsn,
+						j, sid, &mod->sml_values[i] );
+					csn_changed = 1;
+				}
+			}
+			if ( csn_changed )
+				si->si_dirty = 0;
 			ldap_pvt_thread_rdwr_wunlock( &si->si_csn_rwlock );
-			return SLAP_CB_CONTINUE;
+
+			if ( csn_changed ) {
+				syncops *ss;
+				ldap_pvt_thread_mutex_lock( &si->si_ops_mutex );
+				for ( ss = si->si_ops; ss; ss = ss->s_next ) {
+					if ( ss->s_op->o_abandon )
+						continue;
+					/* Send the updated csn to all syncrepl consumers,
+					 * including the server from which it originated.
+					 * The syncrepl consumer and syncprov provider on
+					 * the originating server may be configured to store
+					 * their csn values in different entries.
+					 */
+					syncprov_qresp( opc, ss, LDAP_SYNC_NEW_COOKIE );
+				}
+				ldap_pvt_thread_mutex_unlock( &si->si_ops_mutex );
+			}
+			} else {
+			ldap_pvt_thread_rdwr_wunlock( &si->si_csn_rwlock );
+			}
+			goto leave;
 		}
 
 		si->si_numops++;
 		if ( si->si_chkops || si->si_chktime ) {
-			if ( si->si_chkops && si->si_numops >= si->si_chkops ) {
-				do_check = 1;
-				si->si_numops = 0;
-			}
-			if ( si->si_chktime &&
-				(op->o_time - si->si_chklast >= si->si_chktime )) {
-				if ( si->si_chklast ) {
+			/* Never checkpoint adding the context entry,
+			 * it will deadlock
+			 */
+			if ( op->o_tag != LDAP_REQ_ADD ||
+				!dn_match( &op->o_req_ndn, &si->si_contextdn )) {
+				if ( si->si_chkops && si->si_numops >= si->si_chkops ) {
 					do_check = 1;
-					si->si_chklast = op->o_time;
-				} else {
-					si->si_chklast = 1;
+					si->si_numops = 0;
+				}
+				if ( si->si_chktime &&
+					(op->o_time - si->si_chklast >= si->si_chktime )) {
+					if ( si->si_chklast ) {
+						do_check = 1;
+						si->si_chklast = op->o_time;
+					} else {
+						si->si_chklast = 1;
+					}
 				}
 			}
 		}
+		si->si_dirty = !csn_changed;
 		ldap_pvt_thread_rdwr_wunlock( &si->si_csn_rwlock );
 
 		if ( do_check ) {
 			ldap_pvt_thread_rdwr_rlock( &si->si_csn_rwlock );
-			syncprov_checkpoint( op, rs, on );
+			syncprov_checkpoint( op, on );
 			ldap_pvt_thread_rdwr_runlock( &si->si_csn_rwlock );
 		}
 
@@ -1789,10 +1949,10 @@ syncprov_op_response( Operation *op, SlapReply *rs )
 		}
 
 		/* Add any log records */
-		if ( si->si_logs && op->o_tag != LDAP_REQ_ADD ) {
+		if ( si->si_logs ) {
 			syncprov_add_slog( op );
 		}
-
+leave:		ldap_pvt_thread_mutex_unlock( &si->si_resp_mutex );
 	}
 	return SLAP_CB_CONTINUE;
 }
@@ -1808,14 +1968,14 @@ syncprov_op_compare( Operation *op, SlapReply *rs )
 	syncprov_info_t		*si = on->on_bi.bi_private;
 	int rc = SLAP_CB_CONTINUE;
 
-	if ( dn_match( &op->o_req_ndn, op->o_bd->be_nsuffix ) &&
+	if ( dn_match( &op->o_req_ndn, &si->si_contextdn ) &&
 		op->oq_compare.rs_ava->aa_desc == slap_schema.si_ad_contextCSN )
 	{
 		Entry e = {0};
 		Attribute a = {0};
 
-		e.e_name = op->o_bd->be_suffix[0];
-		e.e_nname = op->o_bd->be_nsuffix[0];
+		e.e_name = si->si_contextdn;
+		e.e_nname = si->si_contextdn;
 		e.e_attrs = &a;
 
 		a.a_desc = slap_schema.si_ad_contextCSN;
@@ -1877,6 +2037,7 @@ syncprov_op_mod( Operation *op, SlapReply *rs )
 
 	ldap_pvt_thread_mutex_lock( &si->si_ops_mutex );
 	have_psearches = ( si->si_ops != NULL );
+	si->si_active++;
 	ldap_pvt_thread_mutex_unlock( &si->si_ops_mutex );
 
 	cbsize = sizeof(slap_callback) + sizeof(opcookie) +
@@ -1891,12 +2052,21 @@ syncprov_op_mod( Operation *op, SlapReply *rs )
 	cb->sc_next = op->o_callback;
 	op->o_callback = cb;
 
+	opc->osid = -1;
+	opc->rsid = -1;
 	if ( op->o_csn.bv_val ) {
-		opc->ssid = slap_parse_csn_sid( &op->o_csn );
-	} else {
-		opc->ssid = -1;
+		opc->osid = slap_parse_csn_sid( &op->o_csn );
+	}
+	if ( op->o_controls ) {
+		struct sync_cookie *scook =
+		op->o_controls[slap_cids.sc_LDAPsync];
+		if ( scook )
+			opc->rsid = scook->sid;
 	}
 
+	if ( op->o_dont_replicate )
+		return SLAP_CB_CONTINUE;
+
 	/* If there are active persistent searches, lock this operation.
 	 * See seqmod.c for the locking logic on its own.
 	 */
@@ -1913,6 +2083,15 @@ syncprov_op_mod( Operation *op, SlapReply *rs )
 		mt = avl_find( si->si_mods, &mtdummy, sp_avl_cmp );
 		if ( mt ) {
 			ldap_pvt_thread_mutex_lock( &mt->mt_mutex );
+			if ( mt->mt_mods == NULL ) {
+				/* Cannot reuse this mt, as another thread is about
+				 * to release it in syncprov_op_cleanup.
+				 */
+				ldap_pvt_thread_mutex_unlock( &mt->mt_mutex );
+				mt = NULL;
+			}
+		}
+		if ( mt ) {
 			ldap_pvt_thread_mutex_unlock( &si->si_mods_mutex );
 			mt->mt_tail->mi_next = mi;
 			mt->mt_tail = mi;
@@ -1924,6 +2103,9 @@ syncprov_op_mod( Operation *op, SlapReply *rs )
 				 * Currently it's not an issue because there are
 				 * no dynamic config deletes...
 				 */
+				if ( slapd_shutdown )
+					return SLAPD_ABANDON;
+
 				if ( !ldap_pvt_thread_pool_pausecheck( &connection_pool ))
 					ldap_pvt_thread_yield();
 				ldap_pvt_thread_mutex_lock( &mt->mt_mutex );
@@ -1951,6 +2133,7 @@ syncprov_op_mod( Operation *op, SlapReply *rs )
 			avl_insert( &si->si_mods, mt, sp_avl_cmp, avl_dup_error );
 			ldap_pvt_thread_mutex_unlock( &si->si_mods_mutex );
 		}
+		opc->smt = mt;
 	}
 
 	if (( have_psearches || si->si_logs ) && op->o_tag != LDAP_REQ_ADD )
@@ -2163,6 +2346,7 @@ syncprov_search_response( Operation *op, SlapReply *rs )
 		rs->sr_ctrls = op->o_tmpalloc( sizeof(LDAPControl *)*2,
 			op->o_tmpmemctx );
 		rs->sr_ctrls[1] = NULL;
+		rs->sr_flags |= REP_CTRLS_MUSTBEFREED;
 		/* If we're in delta-sync mode, always send a cookie */
 		if ( si->si_nopres && si->si_usehint && a ) {
 			struct berval cookie;
@@ -2174,9 +2358,10 @@ syncprov_search_response( Operation *op, SlapReply *rs )
 				LDAP_SYNC_ADD, rs->sr_ctrls, 0, 0, NULL );
 		}
 	} else if ( rs->sr_type == REP_RESULT && rs->sr_err == LDAP_SUCCESS ) {
-		struct berval cookie;
+		struct berval cookie = BER_BVNULL;
 
-		if ( ss->ss_flags & SS_CHANGED ) {
+		if ( ( ss->ss_flags & SS_CHANGED ) &&
+			ss->ss_ctxcsn && !BER_BVISNULL( &ss->ss_ctxcsn[0] )) {
 			slap_compose_sync_cookie( op, &cookie, ss->ss_ctxcsn,
 				srs->sr_state.rid, slap_serverID ? slap_serverID : -1 );
 
@@ -2200,7 +2385,7 @@ syncprov_search_response( Operation *op, SlapReply *rs )
 	 			LDAP_TAG_SYNC_REFRESH_PRESENT : LDAP_TAG_SYNC_REFRESH_DELETE,
 				( ss->ss_flags & SS_CHANGED ) ? &cookie : NULL,
 				1, NULL, 0 );
-			if ( ss->ss_flags & SS_CHANGED )
+			if ( !BER_BVISNULL( &cookie ))
 				op->o_tmpfree( cookie.bv_val, op->o_tmpmemctx );
 
 			/* Detach this Op from frontend control */
@@ -2246,7 +2431,9 @@ syncprov_op_search( Operation *op, SlapReply *rs )
 	sync_control *srs;
 	BerVarray ctxcsn;
 	int i, *sids, numcsns;
-	struct berval mincsn;
+	struct berval mincsn, maxcsn;
+	int minsid, maxsid;
+	int dirty = 0;
 
 	if ( !(op->o_sync_mode & SLAP_SYNC_REFRESH) ) return SLAP_CB_CONTINUE;
 
@@ -2291,6 +2478,20 @@ syncprov_op_search( Operation *op, SlapReply *rs )
 		sop->s_inuse = 1;
 
 		ldap_pvt_thread_mutex_lock( &si->si_ops_mutex );
+		while ( si->si_active ) {
+			/* Wait for active mods to finish before proceeding, as they
+			 * may already have inspected the si_ops list looking for
+			 * consumers to replicate the change to.  Using the log
+			 * doesn't help, as we may finish playing it before the
+			 * active mods gets added to it.
+			 */
+			ldap_pvt_thread_mutex_unlock( &si->si_ops_mutex );
+			if ( slapd_shutdown )
+				return SLAPD_ABANDON;
+			if ( !ldap_pvt_thread_pool_pausecheck( &connection_pool ))
+				ldap_pvt_thread_yield();
+			ldap_pvt_thread_mutex_lock( &si->si_ops_mutex );
+		}
 		sop->s_next = si->si_ops;
 		si->si_ops = sop;
 		ldap_pvt_thread_mutex_unlock( &si->si_ops_mutex );
@@ -2308,6 +2509,7 @@ syncprov_op_search( Operation *op, SlapReply *rs )
 		ctxcsn = NULL;
 		sids = NULL;
 	}
+	dirty = si->si_dirty;
 	ldap_pvt_thread_rdwr_runlock( &si->si_csn_rwlock );
 	
 	/* If we have a cookie, handle the PRESENT lookups */
@@ -2326,75 +2528,133 @@ syncprov_op_search( Operation *op, SlapReply *rs )
 
 		/* If there are SIDs we don't recognize in the cookie, drop them */
 		for (i=0; i<srs->sr_state.numcsns; ) {
-			for (j=0; j<numcsns; j++) {
-				if ( srs->sr_state.sids[i] == sids[j] ) {
+			for (j=i; j<numcsns; j++) {
+				if ( srs->sr_state.sids[i] <= sids[j] ) {
 					break;
 				}
 			}
 			/* not found */
-			if ( j == numcsns ) {
-				struct berval tmp = srs->sr_state.ctxcsn[i];
-				j = srs->sr_state.numcsns - 1;
-				srs->sr_state.ctxcsn[i] = srs->sr_state.ctxcsn[j];
-				tmp.bv_len = 0;
-				srs->sr_state.ctxcsn[j] = tmp;
-				srs->sr_state.numcsns = j;
-				srs->sr_state.sids[i] = srs->sr_state.sids[j];
+			if ( j == numcsns || srs->sr_state.sids[i] != sids[j] ) {
+				char *tmp = srs->sr_state.ctxcsn[i].bv_val;
+				srs->sr_state.numcsns--;
+				for ( j=i; j<srs->sr_state.numcsns; j++ ) {
+					srs->sr_state.ctxcsn[j] = srs->sr_state.ctxcsn[j+1];
+					srs->sr_state.sids[j] = srs->sr_state.sids[j+1];
+				}
+				srs->sr_state.ctxcsn[j].bv_val = tmp;
+				srs->sr_state.ctxcsn[j].bv_len = 0;
 				continue;
 			}
 			i++;
 		}
 
-		/* Find the smallest CSN */
-		mincsn = srs->sr_state.ctxcsn[0];
-		for ( i=1; i<srs->sr_state.numcsns; i++ ) {
-			if ( ber_bvcmp( &mincsn, &srs->sr_state.ctxcsn[i] ) > 0 )
-				mincsn = srs->sr_state.ctxcsn[i];
+		if (srs->sr_state.numcsns != numcsns) {
+			/* consumer doesn't have the right number of CSNs */
+			changed = SS_CHANGED;
+			if ( srs->sr_state.ctxcsn ) {
+				ber_bvarray_free_x( srs->sr_state.ctxcsn, op->o_tmpmemctx );
+				srs->sr_state.ctxcsn = NULL;
+			}
+			if ( srs->sr_state.sids ) {
+				slap_sl_free( srs->sr_state.sids, op->o_tmpmemctx );
+				srs->sr_state.sids = NULL;
+			}
+			srs->sr_state.numcsns = 0;
+			goto shortcut;
+		}
+
+		/* Find the smallest CSN which differs from contextCSN */
+		mincsn.bv_len = 0;
+		maxcsn.bv_len = 0;
+		for ( i=0,j=0; i<srs->sr_state.numcsns; i++ ) {
+			int newer;
+			while ( srs->sr_state.sids[i] != sids[j] ) j++;
+			if ( BER_BVISEMPTY( &maxcsn ) || ber_bvcmp( &maxcsn,
+				&srs->sr_state.ctxcsn[i] ) < 0 ) {
+				maxcsn = srs->sr_state.ctxcsn[i];
+				maxsid = sids[j];
+			}
+			newer = ber_bvcmp( &srs->sr_state.ctxcsn[i], &ctxcsn[j] );
+			/* If our state is newer, tell consumer about changes */
+			if ( newer < 0) {
+				changed = SS_CHANGED;
+				if ( BER_BVISEMPTY( &mincsn ) || ber_bvcmp( &mincsn,
+					&srs->sr_state.ctxcsn[i] ) > 0 ) {
+					mincsn = srs->sr_state.ctxcsn[i];
+					minsid = sids[j];
+				}
+			} else if ( newer > 0 ) {
+			/* our state is older, complain to consumer */
+				rs->sr_err = LDAP_UNWILLING_TO_PERFORM;
+				rs->sr_text = "consumer state is newer than provider!";
+bailout:
+				if ( sop ) {
+					syncops **sp = &si->si_ops;
+
+					ldap_pvt_thread_mutex_lock( &si->si_ops_mutex );
+					while ( *sp != sop )
+						sp = &(*sp)->s_next;
+					*sp = sop->s_next;
+					ldap_pvt_thread_mutex_unlock( &si->si_ops_mutex );
+					ch_free( sop );
+				}
+				rs->sr_ctrls = NULL;
+				send_ldap_result( op, rs );
+				return rs->sr_err;
+			}
+ 		}
+		if ( BER_BVISEMPTY( &mincsn )) {
+			mincsn = maxcsn;
+			minsid = maxsid;
 		}
 
 		/* If nothing has changed, shortcut it */
-		if ( srs->sr_state.numcsns == numcsns ) {
-			int i, j;
-			for ( i=0; i<srs->sr_state.numcsns; i++ ) {
-				for ( j=0; j<numcsns; j++ ) {
-					if ( srs->sr_state.sids[i] != sids[j] )
-						continue;
-					if ( !bvmatch( &srs->sr_state.ctxcsn[i], &ctxcsn[j] ))
-						changed = SS_CHANGED;
-					break;
-				}
-				if ( changed )
-					break;
-			}
-			if ( !changed ) {
-				do_present = 0;
+		if ( !changed && !dirty ) {
+			do_present = 0;
 no_change:		if ( !(op->o_sync_mode & SLAP_SYNC_PERSIST) ) {
-					LDAPControl	*ctrls[2];
-
-					ctrls[0] = NULL;
-					ctrls[1] = NULL;
-					syncprov_done_ctrl( op, rs, ctrls, 0, 0,
-						NULL, LDAP_SYNC_REFRESH_DELETES );
-					rs->sr_ctrls = ctrls;
-					rs->sr_err = LDAP_SUCCESS;
-					send_ldap_result( op, rs );
-					rs->sr_ctrls = NULL;
-					return rs->sr_err;
-				}
-				goto shortcut;
+				LDAPControl	*ctrls[2];
+
+				ctrls[0] = NULL;
+				ctrls[1] = NULL;
+				syncprov_done_ctrl( op, rs, ctrls, 0, 0,
+					NULL, LDAP_SYNC_REFRESH_DELETES );
+				rs->sr_ctrls = ctrls;
+				rs->sr_err = LDAP_SUCCESS;
+				send_ldap_result( op, rs );
+				rs->sr_ctrls = NULL;
+				return rs->sr_err;
 			}
-		} else {
-			/* consumer doesn't have the right number of CSNs */
-			changed = SS_CHANGED;
+			goto shortcut;
 		}
+
 		/* Do we have a sessionlog for this search? */
 		sl=si->si_logs;
 		if ( sl ) {
+			int do_play = 0;
 			ldap_pvt_thread_mutex_lock( &sl->sl_mutex );
 			/* Are there any log entries, and is the consumer state
 			 * present in the session log?
 			 */
-			if ( sl->sl_num > 0 && ber_bvcmp( &mincsn, &sl->sl_mincsn ) >= 0 ) {
+			if ( sl->sl_num > 0 ) {
+				int i;
+				for ( i=0; i<sl->sl_numcsns; i++ ) {
+					/* SID not present == new enough */
+					if ( minsid < sl->sl_sids[i] ) {
+						do_play = 1;
+						break;
+					}
+					/* SID present and new enough */
+					if ( minsid == sl->sl_sids[i]
+						&& ber_bvcmp( &mincsn, &sl->sl_mincsn[i] ) >= 0 ) {
+						do_play = 1;
+						break;
+					}
+				}
+				/* SID not present == new enough */
+				if ( i == sl->sl_numcsns )
+					do_play = 1;
+			}
+			if ( do_play ) {
 				do_present = 0;
 				/* mutex is unlocked in playlog */
 				syncprov_playlog( op, rs, sl, srs, ctxcsn, numcsns, sids );
@@ -2403,7 +2663,7 @@ no_change:		if ( !(op->o_sync_mode & SLAP_SYNC_PERSIST) ) {
 			}
 		}
 		/* Is the CSN still present in the database? */
-		if ( syncprov_findcsn( op, FIND_CSN ) != LDAP_SUCCESS ) {
+		if ( syncprov_findcsn( op, FIND_CSN, &mincsn ) != LDAP_SUCCESS ) {
 			/* No, so a reload is required */
 			/* the 2.2 consumer doesn't send this hint */
 			if ( si->si_usehint && srs->sr_rhint == 0 ) {
@@ -2411,8 +2671,9 @@ no_change:		if ( !(op->o_sync_mode & SLAP_SYNC_PERSIST) ) {
 					ber_bvarray_free_x( ctxcsn, op->o_tmpmemctx );
 				if ( sids )
 					op->o_tmpfree( sids, op->o_tmpmemctx );
-				send_ldap_error( op, rs, LDAP_SYNC_REFRESH_REQUIRED, "sync cookie is stale" );
-				return rs->sr_err;
+				rs->sr_err = LDAP_SYNC_REFRESH_REQUIRED;
+				rs->sr_text = "sync cookie is stale";
+				goto bailout;
 			}
 			if ( srs->sr_state.ctxcsn ) {
 				ber_bvarray_free_x( srs->sr_state.ctxcsn, op->o_tmpmemctx );
@@ -2426,14 +2687,13 @@ no_change:		if ( !(op->o_sync_mode & SLAP_SYNC_PERSIST) ) {
 		} else {
 			gotstate = 1;
 			/* If changed and doing Present lookup, send Present UUIDs */
-			if ( do_present && syncprov_findcsn( op, FIND_PRESENT ) !=
+			if ( do_present && syncprov_findcsn( op, FIND_PRESENT, 0 ) !=
 				LDAP_SUCCESS ) {
 				if ( ctxcsn )
 					ber_bvarray_free_x( ctxcsn, op->o_tmpmemctx );
 				if ( sids )
 					op->o_tmpfree( sids, op->o_tmpmemctx );
-				send_ldap_result( op, rs );
-				return rs->sr_err;
+				goto bailout;
 			}
 		}
 	} else {
@@ -2450,7 +2710,7 @@ shortcut:
 	}
 
 	/* If something changed, find the changes */
-	if ( gotstate && changed ) {
+	if ( gotstate && ( changed || dirty ) ) {
 		Filter *fand, *fava;
 
 		fand = op->o_tmpalloc( sizeof(Filter), op->o_tmpmemctx );
@@ -2466,10 +2726,14 @@ shortcut:
 #endif
 		ber_dupbv_x( &fava->f_ava->aa_value, &mincsn, op->o_tmpmemctx );
 		fava->f_next = op->ors_filter;
+		if ( sop )
+			ldap_pvt_thread_mutex_lock( &sop->s_mutex );
 		op->ors_filter = fand;
 		filter2bv_x( op, op->ors_filter, &op->ors_filterstr );
-		if ( sop )
+		if ( sop ) {
 			sop->s_flags |= PS_FIX_FILTER;
+			ldap_pvt_thread_mutex_unlock( &sop->s_mutex );
+		}
 	}
 
 	/* Let our callback add needed info to returned entries */
@@ -2491,7 +2755,7 @@ shortcut:
 	 * the refresh phase, just invoke the response callback to transition
 	 * us into persist phase
 	 */
-	if ( !changed ) {
+	if ( !changed && !dirty ) {
 		rs->sr_err = LDAP_SUCCESS;
 		rs->sr_nentries = 0;
 		send_ldap_result( op, rs );
@@ -2515,7 +2779,7 @@ syncprov_operational(
 		return SLAP_CB_CONTINUE;
 
 	if ( rs->sr_entry &&
-		dn_match( &rs->sr_entry->e_nname, op->o_bd->be_nsuffix )) {
+		dn_match( &rs->sr_entry->e_nname, &si->si_contextdn )) {
 
 		if ( SLAP_OPATTRS( rs->sr_attr_flags ) ||
 			ad_inlist( slap_schema.si_ad_contextCSN, rs->sr_attrs )) {
@@ -2537,17 +2801,7 @@ syncprov_operational(
 				}
 
 				if ( !ap ) {
-					if ( !(rs->sr_flags & REP_ENTRY_MODIFIABLE) ) {
-						Entry *e = entry_dup( rs->sr_entry );
-						if ( rs->sr_flags & REP_ENTRY_MUSTRELEASE ) {
-							overlay_entry_release_ov( op, rs->sr_entry, 0, on );
-							rs->sr_flags ^= REP_ENTRY_MUSTRELEASE;
-						} else if ( rs->sr_flags & REP_ENTRY_MUSTBEFREED ) {
-							entry_free( rs->sr_entry );
-						}
-						rs->sr_entry = e;
-						rs->sr_flags |=
-							REP_ENTRY_MODIFIABLE|REP_ENTRY_MUSTBEFREED;
+					if ( rs_entry2modifiable( op, rs, on )) {
 						a = attr_find( rs->sr_entry->e_attrs,
 							slap_schema.si_ad_contextCSN );
 					}
@@ -2622,8 +2876,11 @@ sp_cf_gen(ConfigArgs *c)
 		case SP_CHKPT:
 			if ( si->si_chkops || si->si_chktime ) {
 				struct berval bv;
+				/* we assume si_chktime is a multiple of 60
+				 * because the parsed value was originally
+				 * multiplied by 60 */
 				bv.bv_len = snprintf( c->cr_msg, sizeof( c->cr_msg ),
-					"%d %d", si->si_chkops, si->si_chktime );
+					"%d %d", si->si_chkops, si->si_chktime/60 );
 				if ( bv.bv_len >= sizeof( c->cr_msg ) ) {
 					rc = 1;
 				} else {
@@ -2729,10 +2986,11 @@ sp_cf_gen(ConfigArgs *c)
 		}
 		sl = si->si_logs;
 		if ( !sl ) {
-			sl = ch_malloc( sizeof( sessionlog ) + LDAP_LUTIL_CSNSTR_BUFSIZE );
-			sl->sl_mincsn.bv_val = (char *)(sl+1);
-			sl->sl_mincsn.bv_len = 0;
+			sl = ch_malloc( sizeof( sessionlog ));
+			sl->sl_mincsn = NULL;
+			sl->sl_sids = NULL;
 			sl->sl_num = 0;
+			sl->sl_numcsns = 0;
 			sl->sl_head = sl->sl_tail = NULL;
 			ldap_pvt_thread_mutex_init( &sl->sl_mutex );
 			si->si_logs = sl;
@@ -2758,10 +3016,11 @@ syncprov_db_otask(
 	void *ptr
 )
 {
-	syncprov_findcsn( ptr, FIND_MAXCSN );
+	syncprov_findcsn( ptr, FIND_MAXCSN, 0 );
 	return NULL;
 }
 
+
 /* Read any existing contextCSN from the underlying db.
  * Then search for any entries newer than that. If no value exists,
  * just generate it. Cache whatever result.
@@ -2799,13 +3058,19 @@ syncprov_db_open(
 	}
 
 	thrctx = ldap_pvt_thread_pool_context();
-	connection_fake_init( &conn, &opbuf, thrctx );
+	connection_fake_init2( &conn, &opbuf, thrctx, 0 );
 	op = &opbuf.ob_op;
 	op->o_bd = be;
 	op->o_dn = be->be_rootdn;
 	op->o_ndn = be->be_rootndn;
 
-	rc = overlay_entry_get_ov( op, be->be_nsuffix, NULL,
+	if ( SLAP_SYNC_SUBENTRY( be )) {
+		build_new_dn( &si->si_contextdn, be->be_nsuffix,
+			(struct berval *)&slap_ldapsync_cn_bv, NULL );
+	} else {
+		si->si_contextdn = be->be_nsuffix[0];
+	}
+	rc = overlay_entry_get_ov( op, &si->si_contextdn, NULL,
 		slap_schema.si_ad_contextCSN, 0, &e, on );
 
 	if ( e ) {
@@ -2816,6 +3081,7 @@ syncprov_db_open(
 			ber_bvarray_dup_x( &si->si_ctxcsn, a->a_vals, NULL );
 			si->si_numcsns = a->a_numvals;
 			si->si_sids = slap_parse_csn_sids( si->si_ctxcsn, a->a_numvals, NULL );
+			slap_sort_csn_sids( si->si_ctxcsn, si->si_sids, si->si_numcsns, NULL );
 		}
 		overlay_entry_release_ov( op, e, 0, on );
 		if ( si->si_ctxcsn && !SLAP_DBCLEAN( be )) {
@@ -2829,7 +3095,7 @@ syncprov_db_open(
 
 	/* Didn't find a contextCSN, should we generate one? */
 	if ( !si->si_ctxcsn ) {
-		char csnbuf[ LDAP_LUTIL_CSNSTR_BUFSIZE ];
+		char csnbuf[ LDAP_PVT_CSNSTR_BUFSIZE ];
 		struct berval csn;
 
 		if ( SLAP_SYNC_SHADOW( op->o_bd )) {
@@ -2851,6 +3117,17 @@ syncprov_db_open(
 		si->si_numops++;
 	}
 
+	/* Initialize the sessionlog mincsn */
+	if ( si->si_logs && si->si_numcsns ) {
+		sessionlog *sl = si->si_logs;
+		int i;
+		ber_bvarray_dup_x( &sl->sl_mincsn, si->si_ctxcsn, NULL );
+		sl->sl_numcsns = si->si_numcsns;
+		sl->sl_sids = ch_malloc( si->si_numcsns * sizeof(int) );
+		for ( i=0; i < si->si_numcsns; i++ )
+			sl->sl_sids[i] = si->si_sids[i];
+	}
+
 out:
 	op->o_bd->bd_info = (BackendInfo *)on;
 	return 0;
@@ -2866,6 +3143,9 @@ syncprov_db_close(
 {
     slap_overinst   *on = (slap_overinst *) be->bd_info;
     syncprov_info_t *si = (syncprov_info_t *)on->on_bi.bi_private;
+#ifdef SLAP_CONFIG_DELETE
+	syncops *so, *sonext;
+#endif /* SLAP_CONFIG_DELETE */
 
 	if ( slapMode & SLAP_TOOL_MODE ) {
 		return 0;
@@ -2874,18 +3154,31 @@ syncprov_db_close(
 		Connection conn = {0};
 		OperationBuffer opbuf;
 		Operation *op;
-		SlapReply rs = {REP_RESULT};
 		void *thrctx;
 
 		thrctx = ldap_pvt_thread_pool_context();
-		connection_fake_init( &conn, &opbuf, thrctx );
+		connection_fake_init2( &conn, &opbuf, thrctx, 0 );
 		op = &opbuf.ob_op;
 		op->o_bd = be;
 		op->o_dn = be->be_rootdn;
 		op->o_ndn = be->be_rootndn;
-		syncprov_checkpoint( op, &rs, on );
+		syncprov_checkpoint( op, on );
 	}
 
+#ifdef SLAP_CONFIG_DELETE
+	ldap_pvt_thread_mutex_lock( &si->si_ops_mutex );
+	for ( so=si->si_ops, sonext=so;  so; so=sonext  ) {
+		SlapReply rs = {REP_RESULT};
+		rs.sr_err = LDAP_UNAVAILABLE;
+		send_ldap_result( so->s_op, &rs );
+		sonext=so->s_next;
+		syncprov_drop_psearch( so, 0);
+	}
+	si->si_ops=NULL;
+	ldap_pvt_thread_mutex_unlock( &si->si_ops_mutex );
+	overlay_unregister_control( be, LDAP_CONTROL_SYNC );
+#endif /* SLAP_CONFIG_DELETE */
+
     return 0;
 }
 
@@ -2910,6 +3203,7 @@ syncprov_db_init(
 	ldap_pvt_thread_rdwr_init( &si->si_csn_rwlock );
 	ldap_pvt_thread_mutex_init( &si->si_ops_mutex );
 	ldap_pvt_thread_mutex_init( &si->si_mods_mutex );
+	ldap_pvt_thread_mutex_init( &si->si_resp_mutex );
 
 	csn_anlist[0].an_desc = slap_schema.si_ad_entryCSN;
 	csn_anlist[0].an_name = slap_schema.si_ad_entryCSN->ad_cname;
@@ -2933,20 +3227,27 @@ syncprov_db_destroy(
 
 	if ( si ) {
 		if ( si->si_logs ) {
-			slog_entry *se = si->si_logs->sl_head;
+			sessionlog *sl = si->si_logs;
+			slog_entry *se = sl->sl_head;
 
 			while ( se ) {
 				slog_entry *se_next = se->se_next;
 				ch_free( se );
 				se = se_next;
 			}
-				
+			if ( sl->sl_mincsn )
+				ber_bvarray_free( sl->sl_mincsn );
+			if ( sl->sl_sids )
+				ch_free( sl->sl_sids );
+
+			ldap_pvt_thread_mutex_destroy(&si->si_logs->sl_mutex);
 			ch_free( si->si_logs );
 		}
 		if ( si->si_ctxcsn )
 			ber_bvarray_free( si->si_ctxcsn );
 		if ( si->si_sids )
 			ch_free( si->si_sids );
+		ldap_pvt_thread_mutex_destroy( &si->si_resp_mutex );
 		ldap_pvt_thread_mutex_destroy( &si->si_mods_mutex );
 		ldap_pvt_thread_mutex_destroy( &si->si_ops_mutex );
 		ldap_pvt_thread_rdwr_destroy( &si->si_csn_rwlock );