--- servers/slapd/overlays/syncprov.c	2007/09/24 03:02:32	1.200
+++ servers/slapd/overlays/syncprov.c	2011/01/13 21:23:05	1.331
@@ -1,8 +1,8 @@
-/* $OpenLDAP: pkg/ldap/servers/slapd/overlays/syncprov.c,v 1.199 2007/09/23 23:31:24 hyc Exp $ */
+/* $OpenLDAP: pkg/ldap/servers/slapd/overlays/syncprov.c,v 1.330 2011/01/12 13:44:03 hallvard Exp $ */
 /* syncprov.c - syncrepl provider */
 /* This work is part of OpenLDAP Software <http://www.openldap.org/>.
  *
- * Copyright 2004-2007 The OpenLDAP Foundation.
+ * Copyright 2004-2011 The OpenLDAP Foundation.
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -28,6 +28,10 @@
 #include "config.h"
 #include "ldap_rq.h"
 
+#ifdef LDAP_DEVEL
+#define	CHECK_CSN	1
+#endif
+
 /* A modify request on a particular entry */
 typedef struct modinst {
 	struct modinst *mi_next;
@@ -44,6 +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;
@@ -67,12 +72,11 @@ typedef struct syncops {
 #define	PS_WROTE_BASE		0x04
 #define	PS_FIND_BASE		0x08
 #define	PS_FIX_FILTER		0x10
+#define	PS_TASK_QUEUED		0x20
 
 	int		s_inuse;	/* reference count */
 	struct syncres *s_res;
 	struct syncres *s_restail;
-	struct re_s	*s_qtask;	/* task for playing psearch responses */
-#define	RUNQ_INTERVAL	36000	/* a long time */
 	ldap_pvt_thread_mutex_t	s_mutex;
 } syncops;
 
@@ -121,6 +125,7 @@ typedef struct sessionlog {
 typedef struct syncprov_info_t {
 	syncops		*si_ops;
 	BerVarray	si_ctxcsn;	/* ldapsync context */
+	struct berval	si_contextdn;
 	int		*si_sids;
 	int		si_numcsns;
 	int		si_chkops;	/* checkpointing info */
@@ -128,24 +133,37 @@ 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;
-	int sreference;	/* Is the entry a reference? */
+	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 */
@@ -393,9 +411,6 @@ static struct berval generic_filterstr =
 static int
 syncprov_findbase( Operation *op, fbase_cookie *fc )
 {
-	opcookie *opc = op->o_callback->sc_private;
-	slap_overinst *on = opc->son;
-
 	/* Use basic parameters from syncrepl search, but use
 	 * current op's threadctx / tmpmemctx
 	 */
@@ -404,7 +419,6 @@ syncprov_findbase( Operation *op, fbase_
 		slap_callback cb = {0};
 		Operation fop;
 		SlapReply frs = { REP_RESULT };
-		BackendInfo *bi;
 		int rc;
 
 		fc->fss->s_flags ^= PS_FIND_BASE;
@@ -412,11 +426,10 @@ syncprov_findbase( Operation *op, fbase_
 
 		fop = *fc->fss->s_op;
 
+		fop.o_bd = fop.o_bd->bd_self;
 		fop.o_hdr = op->o_hdr;
-		fop.o_bd = op->o_bd;
 		fop.o_time = op->o_time;
 		fop.o_tincr = op->o_tincr;
-		bi = op->o_bd->bd_info;
 
 		cb.sc_response = findbase_cb;
 		cb.sc_private = fc;
@@ -434,8 +447,7 @@ syncprov_findbase( Operation *op, fbase_
 		fop.ors_filter = &generic_filter;
 		fop.ors_filterstr = generic_filterstr;
 
-		rc = overlay_op_walk( &fop, &frs, op_search, on->on_info, on );
-		op->o_bd->bd_info = bi;
+		rc = fop.o_bd->be_search( &fop, &frs );
 	} else {
 		ldap_pvt_thread_mutex_unlock( &fc->fss->s_mutex );
 		fc->fbase = 1;
@@ -500,7 +512,8 @@ findmax_cb( Operation *op, SlapReply *rs
 		Attribute *a = attr_find( rs->sr_entry->e_attrs,
 			slap_schema.si_ad_entryCSN );
 
-		if ( a && ber_bvcmp( &a->a_vals[0], maxcsn ) > 0 ) {
+		if ( a && ber_bvcmp( &a->a_vals[0], maxcsn ) > 0 &&
+			slap_parse_csn_sid( &a->a_vals[0] ) == slap_serverID ) {
 			maxcsn->bv_len = a->a_vals[0].bv_len;
 			strcpy( maxcsn->bv_val, a->a_vals[0].bv_val );
 		}
@@ -581,8 +594,8 @@ syncprov_findcsn( Operation *op, find_cs
 	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;
@@ -590,7 +603,7 @@ syncprov_findcsn( Operation *op, find_cs
 	sync_control *srs = NULL;
 	struct slap_limits_set fc_limits;
 	int i, rc = LDAP_SUCCESS, findcsn_retry = 1;
-	int maxid = 0;
+	int maxid;
 
 	if ( mode != FIND_MAXCSN ) {
 		srs = op->o_controls[slap_cids.sc_LDAPsync];
@@ -616,16 +629,25 @@ again:
 	switch( mode ) {
 	case FIND_MAXCSN:
 		cf.f_choice = LDAP_FILTER_GE;
-		cf.f_av_value = si->si_ctxcsn[0];
-		/* If there are multiple CSNs, use the largest */
-		for ( i=1; i<si->si_numcsns; i++) {
-			if ( ber_bvcmp( &cf.f_av_value, &si->si_ctxcsn[i] ) < 0 ) {
-				cf.f_av_value = si->si_ctxcsn[i];
+		/* If there are multiple CSNs, use the one with our serverID */
+		for ( i=0; i<si->si_numcsns; i++) {
+			if ( slap_serverID == si->si_sids[i] ) {
 				maxid = i;
+				break;
 			}
 		}
-		fop.ors_filterstr.bv_len = sprintf( buf, "(entryCSN>=%s)",
-			cf.f_av_value.bv_val );
+		if ( i == si->si_numcsns ) {
+			/* No match: this is multimaster, and none of the content in the DB
+			 * originated locally. Treat like no CSN.
+			 */
+			return LDAP_NO_SUCH_OBJECT;
+		}
+		cf.f_av_value = si->si_ctxcsn[maxid];
+		fop.ors_filterstr.bv_len = snprintf( buf, sizeof( buf ),
+			"(entryCSN>=%s)", cf.f_av_value.bv_val );
+		if ( fop.ors_filterstr.bv_len >= sizeof( buf ) ) {
+			return LDAP_OTHER;
+		}
 		fop.ors_attrsonly = 0;
 		fop.ors_attrs = csn_anlist;
 		fop.ors_slimit = SLAP_NO_LIMIT;
@@ -649,16 +671,19 @@ again:
 		/* Look for exact match the first time */
 		if ( findcsn_retry ) {
 			cf.f_choice = LDAP_FILTER_EQUALITY;
-			fop.ors_filterstr.bv_len = sprintf( buf, "(entryCSN=%s)",
-				cf.f_av_value.bv_val );
+			fop.ors_filterstr.bv_len = snprintf( buf, sizeof( buf ),
+				"(entryCSN=%s)", cf.f_av_value.bv_val );
 		/* On retry, look for <= */
 		} else {
 			cf.f_choice = LDAP_FILTER_LE;
 			fop.ors_limit = &fc_limits;
 			memset( &fc_limits, 0, sizeof( fc_limits ));
 			fc_limits.lms_s_unchecked = 1;
-			fop.ors_filterstr.bv_len = sprintf( buf, "(entryCSN<=%s)",
-				cf.f_av_value.bv_val );
+			fop.ors_filterstr.bv_len = snprintf( buf, sizeof( buf ),
+				"(entryCSN<=%s)", cf.f_av_value.bv_val );
+		}
+		if ( fop.ors_filterstr.bv_len >= sizeof( buf ) ) {
+			return LDAP_OTHER;
 		}
 		fop.ors_attrsonly = 1;
 		fop.ors_attrs = slap_anlist_no_attrs;
@@ -690,13 +715,17 @@ again:
 		break;
 	}
 
-	fop.o_bd->bd_info = on->on_info->oi_orig;
+	fop.o_bd->bd_info = (BackendInfo *)on->on_info;
 	fop.o_bd->be_search( &fop, &frs );
 	fop.o_bd->bd_info = (BackendInfo *)on;
 
 	switch( mode ) {
 	case FIND_MAXCSN:
 		if ( ber_bvcmp( &si->si_ctxcsn[maxid], &maxcsn )) {
+#ifdef CHECK_CSN
+			Syntax *syn = slap_schema.si_ad_contextCSN->ad_type->sat_syntax;
+			assert( !syn->ssyn_validate( syn, &maxcsn ));
+#endif
 			ber_bvreplace( &si->si_ctxcsn[maxid], &maxcsn );
 			si->si_numops++;	/* ensure a checkpoint */
 		}
@@ -707,6 +736,7 @@ 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;
@@ -720,6 +750,36 @@ 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 )
 {
@@ -727,7 +787,8 @@ 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;
 	}
@@ -743,6 +804,12 @@ 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 );
@@ -751,14 +818,11 @@ 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};
 
@@ -766,48 +830,56 @@ syncprov_sendresp( Operation *op, opcook
 		return SLAPD_ABANDON;
 
 	ctrls[1] = NULL;
-	csns[0] = opc->sctxcsn;
-	BER_BVZERO( &csns[1] );
-	slap_compose_sync_cookie( op, &cookie, csns, so->s_rid, so->s_sid );
+	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 );
+	}
+
+#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 ) {
+		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 ) {
+		if ( opc->sreference && so->s_op->o_managedsait <= SLAP_CONTROL_IGNORED ) {
 			struct berval bv = BER_BVNULL;
 			rs.sr_ref = &bv;
 			rs.sr_err = send_search_reference( op, &rs );
@@ -820,74 +892,96 @@ syncprov_sendresp( Operation *op, opcook
 	}
 	/* 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;
 	}
 
 	return rs.sr_err;
 }
 
+static void
+syncprov_qstart( syncops *so );
+
 /* Play back queued responses */
 static int
-syncprov_qplay( Operation *op, slap_overinst *on, syncops *so )
+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;
 
 	opc.son = on;
 
-	for (;;) {
+	do {
 		ldap_pvt_thread_mutex_lock( &so->s_mutex );
 		sr = so->s_res;
 		if ( sr )
 			so->s_res = sr->s_next;
 		if ( !so->s_res )
 			so->s_restail = NULL;
-		ldap_pvt_thread_mutex_unlock( &so->s_mutex );
-
+		/* Exit loop with mutex held */
 		if ( !sr || so->s_op->o_abandon )
 			break;
+		ldap_pvt_thread_mutex_unlock( &so->s_mutex );
 
-		opc.sdn = sr->s_dn;
-		opc.sndn = sr->s_ndn;
-		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 );
+		if ( sr->s_mode == LDAP_SYNC_NEW_COOKIE ) {
+			SlapReply rs = { REP_INTERMEDIATE };
 
-		if ( e ) {
-			overlay_entry_release_ov( op, e, 0, on );
+			rc = syncprov_sendinfo( op, &rs, LDAP_TAG_SYNC_NEW_COOKIE,
+				&sr->s_csn, 0, NULL, 0 );
+		} else {
+			opc.sdn = sr->s_dn;
+			opc.sndn = sr->s_ndn;
+			opc.suuid = sr->s_uuid;
+			opc.sctxcsn = sr->s_csn;
+			opc.sreference = sr->s_isreference;
+			opc.se = sr->s_e;
+
+			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 );
+			}
 		}
 
 		ch_free( sr );
 
-		if ( rc )
-			break;
+		/* Exit loop with mutex held */
+		ldap_pvt_thread_mutex_lock( &so->s_mutex );
+
+	} while (0);
+
+	/* We now only send one change at a time, to prevent one
+	 * psearch from hogging all the CPU. Resubmit this task if
+	 * there are more responses queued and no errors occurred.
+	 */
+
+	if ( rc == 0 && so->s_res ) {
+		syncprov_qstart( so );
+	} else {
+		so->s_flags ^= PS_TASK_QUEUED;
 	}
+
+	ldap_pvt_thread_mutex_unlock( &so->s_mutex );
 	return rc;
 }
 
-/* runqueue task for playing back queued responses */
+/* task for playing back queued responses */
 static void *
 syncprov_qtask( void *ctx, void *arg )
 {
-	struct re_s *rtask = arg;
-	syncops *so = rtask->arg;
-	slap_overinst *on = so->s_op->o_private;
+	syncops *so = arg;
 	OperationBuffer opbuf;
 	Operation *op;
 	BackendDB be;
@@ -909,36 +1003,14 @@ syncprov_qtask( void *ctx, void *arg )
 	be = *so->s_op->o_bd;
 	be.be_flags |= SLAP_DBFLAG_OVERLAY;
 	op->o_bd = &be;
-	op->o_private = NULL;
+	LDAP_SLIST_FIRST(&op->o_extra) = NULL;
 	op->o_callback = NULL;
 
-	rc = syncprov_qplay( op, on, so );
+	rc = syncprov_qplay( op, so );
 
 	/* decrement use count... */
 	syncprov_free_syncop( so );
 
-	/* wait until we get explicitly scheduled again */
-	ldap_pvt_thread_mutex_lock( &slapd_rq.rq_mutex );
-	ldap_pvt_runqueue_stoptask( &slapd_rq, rtask );
-	if ( rc == 0 ) {
-		ldap_pvt_runqueue_resched( &slapd_rq, rtask, 1 );
-	} else {
-		/* bail out on any error */
-		ldap_pvt_runqueue_remove( &slapd_rq, rtask );
-	}
-	ldap_pvt_thread_mutex_unlock( &slapd_rq.rq_mutex );
-
-#if 0	/* FIXME: connection_close isn't exported from slapd.
-		 * should it be?
-		 */
-	if ( rc ) {
-		ldap_pvt_thread_mutex_lock( &op->o_conn->c_mutex );
-		if ( connection_state_closing( op->o_conn )) {
-			connection_close( op->o_conn );
-		}
-		ldap_pvt_thread_mutex_unlock( &op->o_conn->c_mutex );
-	}
-#endif
 	return NULL;
 }
 
@@ -946,27 +1018,10 @@ syncprov_qtask( void *ctx, void *arg )
 static void
 syncprov_qstart( syncops *so )
 {
-	int wake=0;
-	ldap_pvt_thread_mutex_lock( &slapd_rq.rq_mutex );
-	if ( !so->s_qtask ) {
-		so->s_qtask = ldap_pvt_runqueue_insert( &slapd_rq, RUNQ_INTERVAL,
-			syncprov_qtask, so, "syncprov_qtask",
-			so->s_op->o_conn->c_peer_name.bv_val );
-		++so->s_inuse;
-		wake = 1;
-	} else {
-		if (!ldap_pvt_runqueue_isrunning( &slapd_rq, so->s_qtask ) &&
-			!so->s_qtask->next_sched.tv_sec ) {
-			so->s_qtask->interval.tv_sec = 0;
-			ldap_pvt_runqueue_resched( &slapd_rq, so->s_qtask, 0 );
-			so->s_qtask->interval.tv_sec = RUNQ_INTERVAL;
-			++so->s_inuse;
-			wake = 1;
-		}
-	}
-	ldap_pvt_thread_mutex_unlock( &slapd_rq.rq_mutex );
-	if ( wake )
-		slap_wake_listener();
+	so->s_flags |= PS_TASK_QUEUED;
+	so->s_inuse++;
+	ldap_pvt_thread_pool_submit( &connection_pool, 
+		syncprov_qtask, so );
 }
 
 /* Queue a persistent search response */
@@ -974,16 +1029,26 @@ static int
 syncprov_qresp( opcookie *opc, syncops *so, int mode )
 {
 	syncres *sr;
-	int sid;
+	int srsize;
+	struct berval cookie = opc->sctxcsn;
 
-	/* Don't send changes back to their originator */
-	sid = slap_parse_csn_sid( &opc->sctxcsn );
-	if ( sid >= 0 && sid == so->s_sid )
-		return LDAP_SUCCESS;
+	if ( mode == LDAP_SYNC_NEW_COOKIE ) {
+		syncprov_info_t	*si = opc->son->on_bi.bi_private;
 
-	sr = ch_malloc(sizeof(syncres) + opc->suuid.bv_len + 1 +
-		opc->sdn.bv_len + 1 + opc->sndn.bv_len + 1 + opc->sctxcsn.bv_len + 1 );
+		slap_compose_sync_cookie( NULL, &cookie, si->si_ctxcsn,
+			so->s_rid, slap_serverID ? slap_serverID : -1);
+	}
+
+	srsize = sizeof(syncres) + opc->suuid.bv_len + 1 +
+		opc->sdn.bv_len + 1 + opc->sndn.bv_len + 1;
+	if ( cookie.bv_len )
+		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;
@@ -995,9 +1060,17 @@ syncprov_qresp( opcookie *opc, syncops *
 		 opc->sndn.bv_val ) + 1;
 	sr->s_uuid.bv_len = opc->suuid.bv_len;
 	AC_MEMCPY( sr->s_uuid.bv_val, opc->suuid.bv_val, opc->suuid.bv_len );
-	sr->s_csn.bv_val = sr->s_uuid.bv_val + sr->s_uuid.bv_len + 1;
-	sr->s_csn.bv_len = opc->sctxcsn.bv_len;
-	strcpy( sr->s_csn.bv_val, opc->sctxcsn.bv_val );
+	if ( cookie.bv_len ) {
+		sr->s_csn.bv_val = sr->s_uuid.bv_val + sr->s_uuid.bv_len + 1;
+		strcpy( sr->s_csn.bv_val, cookie.bv_val );
+	} else {
+		sr->s_csn.bv_val = NULL;
+	}
+	sr->s_csn.bv_len = cookie.bv_len;
+
+	if ( mode == LDAP_SYNC_NEW_COOKIE && cookie.bv_val ) {
+		ch_free( cookie.bv_val );
+	}
 
 	ldap_pvt_thread_mutex_lock( &so->s_mutex );
 	if ( !so->s_res ) {
@@ -1012,7 +1085,7 @@ syncprov_qresp( opcookie *opc, syncops *
 		so->s_flags ^= PS_WROTE_BASE;
 		so->s_flags |= PS_FIND_BASE;
 	}
-	if ( so->s_flags & PS_IS_DETACHED ) {
+	if (( so->s_flags & (PS_IS_DETACHED|PS_TASK_QUEUED)) == PS_IS_DETACHED ) {
 		syncprov_qstart( so );
 	}
 	ldap_pvt_thread_mutex_unlock( &so->s_mutex );
@@ -1119,9 +1192,12 @@ syncprov_matchops( Operation *op, opcook
 		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;
@@ -1129,6 +1205,13 @@ syncprov_matchops( Operation *op, opcook
 		}
 	} 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 ) {
@@ -1149,10 +1232,29 @@ syncprov_matchops( Operation *op, opcook
 	for (ss = si->si_ops, sprev = (syncops *)&si->si_ops; ss;
 		sprev = ss, ss=snext)
 	{
+		Operation op2;
+		Opheader oh;
 		syncmatches *sm;
 		int found = 0;
 
 		snext = ss->s_next;
+		if ( ss->s_op->o_abandon )
+			continue;
+
+		/* Don't send ops back to the originator */
+		if ( opc->osid > 0 && opc->osid == ss->s_sid ) {
+			Debug( LDAP_DEBUG_SYNC, "syncprov_matchops: skipping original sid %03x\n",
+				opc->osid, 0, 0 );
+			continue;
+		}
+
+		/* Don't send ops back to the messenger */
+		if ( opc->rsid > 0 && opc->rsid == ss->s_sid ) {
+			Debug( LDAP_DEBUG_SYNC, "syncprov_matchops: skipping relayed sid %03x\n",
+				opc->rsid, 0, 0 );
+			continue;
+		}
+
 		/* validate base */
 		fc.fss = ss;
 		fc.fbase = 0;
@@ -1170,7 +1272,6 @@ syncprov_matchops( Operation *op, opcook
 			continue;
 		}
 
-
 		/* If we're sending results now, look for this op in old matches */
 		if ( !saveit ) {
 			syncmatches *old;
@@ -1193,9 +1294,31 @@ syncprov_matchops( Operation *op, opcook
 			}
 		}
 
+		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( op, 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;
@@ -1212,6 +1335,14 @@ syncprov_matchops( Operation *op, opcook
 		} else if ( !saveit && found ) {
 			/* send DELETE */
 			syncprov_qresp( opc, ss, LDAP_SYNC_DELETE );
+		} else if ( !saveit ) {
+			syncprov_qresp( opc, ss, LDAP_SYNC_NEW_COOKIE );
+		}
+		if ( !saveit && found ) {
+			/* Decrement s_inuse, was incremented when called
+			 * with saveit == TRUE
+			 */
+			syncprov_free_syncop( ss );
 		}
 	}
 	ldap_pvt_thread_mutex_unlock( &si->si_ops_mutex );
@@ -1220,9 +1351,17 @@ syncprov_matchops( Operation *op, opcook
 		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 );
 	}
@@ -1237,7 +1376,12 @@ syncprov_op_cleanup( Operation *op, Slap
 	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;
@@ -1246,26 +1390,23 @@ syncprov_op_cleanup( Operation *op, Slap
 	}
 
 	/* 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 ))
@@ -1279,20 +1420,30 @@ syncprov_op_cleanup( Operation *op, Slap
 }
 
 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;
 
+	int i;
+	for ( i=0; i<si->si_numcsns; i++ ) {
+		assert( !syn->ssyn_validate( syn, si->si_ctxcsn+i ));
+	}
+#endif
 	mod.sml_numvals = si->si_numcsns;
 	mod.sml_values = si->si_ctxcsn;
 	mod.sml_nvalues = NULL;
 	mod.sml_desc = slap_schema.si_ad_contextCSN;
 	mod.sml_op = LDAP_MOD_REPLACE;
-	mod.sml_flags = 0;
+	mod.sml_flags = SLAP_MOD_INTERNAL;
 	mod.sml_next = NULL;
 
 	cb.sc_response = slap_null_cb;
@@ -1301,16 +1452,41 @@ syncprov_checkpoint( Operation *op, Slap
 	opm.o_callback = &cb;
 	opm.orm_modlist = &mod;
 	opm.orm_no_opattrs = 1;
-	opm.o_req_dn = op->o_bd->be_suffix[0];
-	opm.o_req_ndn = op->o_bd->be_nsuffix[0];
+	if ( SLAP_GLUE_SUBORDINATE( op->o_bd )) {
+		be = *on->on_info->oi_origdb;
+		opm.o_bd = &be;
+	}
+	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 );
 	}
-	opm.orm_no_opattrs = 0;
+#ifdef CHECK_CSN
+	for ( i=0; i<si->si_numcsns; i++ ) {
+		assert( !syn->ssyn_validate( syn, si->si_ctxcsn+i ));
+	}
+#endif
 }
 
 static void
@@ -1324,6 +1500,23 @@ 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 );
@@ -1342,16 +1535,30 @@ 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;
 		}
-		sl->sl_tail = se;
 		sl->sl_num++;
 		while ( sl->sl_num > sl->sl_size ) {
 			se = sl->sl_head;
 			sl->sl_head = se->se_next;
-			strcpy( sl->sl_mincsn.bv_val, se->se_csn.bv_val );
+			AC_MEMCPY( sl->sl_mincsn.bv_val, se->se_csn.bv_val, se->se_csn.bv_len );
 			sl->sl_mincsn.bv_len = se->se_csn.bv_len;
 			ch_free( se );
 			sl->sl_num--;
@@ -1378,7 +1585,7 @@ syncprov_playlog( Operation *op, SlapRep
 	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];
 
@@ -1437,6 +1644,8 @@ syncprov_playlog( Operation *op, SlapRep
 			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;
 		}
@@ -1477,7 +1686,6 @@ syncprov_playlog( Operation *op, SlapRep
 
 	if ( mmods ) {
 		Operation fop;
-		SlapReply frs = { REP_RESULT };
 		int rc;
 		Filter mf, af;
 		AttributeAssertion eq = ATTRIBUTEASSERTION_INIT;
@@ -1504,32 +1712,41 @@ syncprov_playlog( Operation *op, SlapRep
 		fop.ors_filter = &af;
 
 		cb.sc_response = playlog_cb;
-		fop.o_bd->bd_info = on->on_info->oi_orig;
+		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;
 	}
 	if ( ndel ) {
 		struct berval cookie;
 
-		slap_compose_sync_cookie( op, &cookie, delcsn, srs->sr_state.rid,
-			srs->sr_state.sid );
+		if ( delcsn[0].bv_len ) {
+			slap_compose_sync_cookie( op, &cookie, delcsn, srs->sr_state.rid,
+				slap_serverID ? slap_serverID : -1 );
+
+			Debug( LDAP_DEBUG_SYNC, "syncprov_playlog: cookie=%s\n", cookie.bv_val, 0, 0 );
+		}
+
 		uuids[ndel].bv_val = NULL;
-		syncprov_sendinfo( op, rs, LDAP_TAG_SYNC_ID_SET, &cookie, 0, uuids, 1 );
-		op->o_tmpfree( cookie.bv_val, op->o_tmpmemctx );
+		syncprov_sendinfo( op, rs, LDAP_TAG_SYNC_ID_SET,
+			delcsn[0].bv_len ? &cookie : NULL, 0, uuids, 1 );
+		if ( delcsn[0].bv_len ) {
+			op->o_tmpfree( cookie.bv_val, op->o_tmpmemctx );
+		}
 	}
 	op->o_tmpfree( uuids, op->o_tmpmemctx );
 }
@@ -1544,22 +1761,44 @@ syncprov_op_response( Operation *op, Sla
 
 	if ( rs->sr_err == LDAP_SUCCESS )
 	{
-		struct berval maxcsn = BER_BVNULL;
-		char cbuf[LDAP_LUTIL_CSNSTR_BUFSIZE];
-		int do_check = 0;
+		struct berval maxcsn;
+		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 );
-		slap_get_commit_csn( op, &maxcsn );
-		if ( !BER_BVISNULL( &maxcsn ) ) {
+
+		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
+			 * it is configured , not where the changes are made.
+			 * So look for a value in the glue db if we didn't
+			 * find any in this db.
+			 */
+			BackendDB *be = op->o_bd;
+			op->o_bd = select_backend( &be->be_nsuffix[0], 1);
+			maxcsn.bv_val = cbuf;
+			maxcsn.bv_len = sizeof(cbuf);
+			slap_get_commit_csn( op, &maxcsn, &foundit );
+			op->o_bd = be;
+		}
+		if ( !BER_BVISEMPTY( &maxcsn ) ) {
 			int i, sid;
-			strcpy( cbuf, maxcsn.bv_val );
+#ifdef CHECK_CSN
+			Syntax *syn = slap_schema.si_ad_contextCSN->ad_type->sat_syntax;
+			assert( !syn->ssyn_validate( syn, &maxcsn ));
+#endif
 			sid = slap_parse_csn_sid( &maxcsn );
 			for ( i=0; i<si->si_numcsns; i++ ) {
 				if ( sid == si->si_sids[i] ) {
 					if ( ber_bvcmp( &maxcsn, &si->si_ctxcsn[i] ) > 0 ) {
 						ber_bvreplace( &si->si_ctxcsn[i], &maxcsn );
+						csn_changed = 1;
 					}
 					break;
 				}
@@ -1567,6 +1806,7 @@ syncprov_op_response( Operation *op, Sla
 			/* It's a new SID for us */
 			if ( i == si->si_numcsns ) {
 				value_add_one( &si->si_ctxcsn, &maxcsn );
+				csn_changed = 1;
 				si->si_numcsns++;
 				si->si_sids = ch_realloc( si->si_sids, si->si_numcsns *
 					sizeof(int));
@@ -1575,37 +1815,106 @@ syncprov_op_response( Operation *op, Sla
 		}
 
 		/* Don't do any processing for consumer contextCSN updates */
-		if ( SLAP_SYNC_SHADOW( op->o_bd ) && 
-			op->o_msgid == SLAP_SYNC_UPDATE_MSGID ) {
+		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] ) {
+						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;
+				}
+			}
+			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 )) {
-				do_check = 1;
-				si->si_chklast = op->o_time;
+			/* 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_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 );
 		}
 
-		opc->sctxcsn.bv_len = maxcsn.bv_len;
-		opc->sctxcsn.bv_val = cbuf;
+		/* only update consumer ctx if this is a newer csn */
+		if ( csn_changed ) {
+			opc->sctxcsn = maxcsn;
+		}
 
 		/* Handle any persistent searches */
-		if ( si->si_ops ) {
+		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 ) {
 			switch(op->o_tag) {
 			case LDAP_REQ_ADD:
 			case LDAP_REQ_MODIFY:
@@ -1627,10 +1936,10 @@ syncprov_op_response( Operation *op, Sla
 		}
 
 		/* 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;
 }
@@ -1646,14 +1955,14 @@ syncprov_op_compare( Operation *op, Slap
 	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;
@@ -1662,6 +1971,7 @@ syncprov_op_compare( Operation *op, Slap
 
 		a.a_vals = si->si_ctxcsn;
 		a.a_nvals = a.a_vals;
+		a.a_numvals = si->si_numcsns;
 
 		rs->sr_err = access_allowed( op, &e, op->oq_compare.rs_ava->aa_desc,
 			&op->oq_compare.rs_ava->aa_value, ACL_COMPARE, NULL );
@@ -1708,12 +2018,20 @@ syncprov_op_mod( Operation *op, SlapRepl
 {
 	slap_overinst		*on = (slap_overinst *)op->o_bd->bd_info;
 	syncprov_info_t		*si = on->on_bi.bi_private;
+	slap_callback *cb;
+	opcookie *opc;
+	int have_psearches, cbsize;
 
-	slap_callback *cb = op->o_tmpcalloc(1, sizeof(slap_callback)+
-		sizeof(opcookie) +
-		(si->si_ops ? sizeof(modinst) : 0 ),
-		op->o_tmpmemctx);
-	opcookie *opc = (opcookie *)(cb+1);
+	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) +
+		(have_psearches ? sizeof(modinst) : 0 );
+
+	cb = op->o_tmpcalloc(1, cbsize, op->o_tmpmemctx);
+	opc = (opcookie *)(cb+1);
 	opc->son = on;
 	cb->sc_response = syncprov_op_response;
 	cb->sc_cleanup = syncprov_op_cleanup;
@@ -1721,10 +2039,22 @@ syncprov_op_mod( Operation *op, SlapRepl
 	cb->sc_next = op->o_callback;
 	op->o_callback = cb;
 
+	opc->osid = -1;
+	opc->rsid = -1;
+	if ( op->o_csn.bv_val ) {
+		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 there are active persistent searches, lock this operation.
 	 * See seqmod.c for the locking logic on its own.
 	 */
-	if ( si->si_ops ) {
+	if ( have_psearches ) {
 		modtarget *mt, mtdummy;
 		modinst *mi;
 
@@ -1737,13 +2067,31 @@ syncprov_op_mod( Operation *op, SlapRepl
 		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;
 			/* wait for this op to get to head of list */
 			while ( mt->mt_mods != mi ) {
 				ldap_pvt_thread_mutex_unlock( &mt->mt_mutex );
-				ldap_pvt_thread_yield();
+				/* FIXME: if dynamic config can delete overlays or
+				 * databases we'll have to check for cleanup here.
+				 * 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 );
 
 				/* clean up if the caller is giving up */
@@ -1769,9 +2117,10 @@ syncprov_op_mod( Operation *op, SlapRepl
 			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 (( si->si_ops || si->si_logs ) && op->o_tag != LDAP_REQ_ADD )
+	if (( have_psearches || si->si_logs ) && op->o_tag != LDAP_REQ_ADD )
 		syncprov_matchops( op, opc, 1 );
 
 	return SLAP_CB_CONTINUE;
@@ -1792,7 +2141,9 @@ typedef struct searchstate {
 	BerVarray ss_ctxcsn;
 	int *ss_sids;
 	int ss_numcsns;
-	int ss_present;
+#define	SS_PRESENT	0x01
+#define	SS_CHANGED	0x02
+	int ss_flags;
 } searchstate;
 
 static int
@@ -1809,6 +2160,7 @@ syncprov_search_cleanup( Operation *op,
 typedef struct SyncOperationBuffer {
 	Operation		sob_op;
 	Opheader		sob_hdr;
+	OpExtra			sob_oe;
 	AttributeName	sob_extra;	/* not always present */
 	/* Further data allocated here */
 } SyncOperationBuffer;
@@ -1837,6 +2189,7 @@ syncprov_detach_op( Operation *op, synco
 	sopbuf2 = ch_calloc( 1, size );
 	op2 = &sopbuf2->sob_op;
 	op2->o_hdr = &sopbuf2->sob_hdr;
+	LDAP_SLIST_FIRST(&op2->o_extra) = &sopbuf2->sob_oe;
 
 	/* Copy the fields we care about explicitly, leave the rest alone */
 	*op2->o_hdr = *op->o_hdr;
@@ -1844,7 +2197,9 @@ syncprov_detach_op( Operation *op, synco
 	op2->o_time = op->o_time;
 	op2->o_bd = on->on_info->oi_origdb;
 	op2->o_request = op->o_request;
-	op2->o_private = on;
+	op2->o_managedsait = op->o_managedsait;
+	LDAP_SLIST_FIRST(&op2->o_extra)->oe_key = on;
+	LDAP_SLIST_NEXT(LDAP_SLIST_FIRST(&op2->o_extra), oe_next) = NULL;
 
 	ptr = (char *) sopbuf2 + offsetof( SyncOperationBuffer, sob_extra );
 	if ( i ) {
@@ -1895,12 +2250,10 @@ syncprov_detach_op( Operation *op, synco
 	op2->o_do_not_cache = 1;
 
 	/* Add op2 to conn so abandon will find us */
-	ldap_pvt_thread_mutex_lock( &op->o_conn->c_mutex );
 	op->o_conn->c_n_ops_executing++;
 	op->o_conn->c_n_ops_completed--;
 	LDAP_STAILQ_INSERT_TAIL( &op->o_conn->c_ops, op2, o_next );
 	so->s_flags |= PS_IS_DETACHED;
-	ldap_pvt_thread_mutex_unlock( &op->o_conn->c_mutex );
 
 	/* Prevent anyone else from trying to send a result for this op */
 	op->o_abandon = 1;
@@ -1911,6 +2264,7 @@ syncprov_search_response( Operation *op,
 {
 	searchstate *ss = op->o_callback->sc_private;
 	slap_overinst *on = ss->ss_on;
+	syncprov_info_t *si = (syncprov_info_t *)on->on_bi.bi_private;
 	sync_control *srs = op->o_controls[slap_cids.sc_LDAPsync];
 
 	if ( rs->sr_type == REP_SEARCH || rs->sr_type == REP_SEARCHREF ) {
@@ -1940,16 +2294,20 @@ syncprov_search_response( Operation *op,
 					rs->sr_entry->e_name.bv_val, 0, 0 );
 				return LDAP_SUCCESS;
 			}
-			/* Make sure entry is less than the snapshot'd contextCSN */
-			for ( i=0; i<ss->ss_numcsns; i++ ) {
-				if ( sid == ss->ss_sids[i] && ber_bvcmp( &a->a_nvals[0],
-					&ss->ss_ctxcsn[i] ) > 0 ) {
-					Debug( LDAP_DEBUG_SYNC,
-						"Entry %s CSN %s greater than snapshot %s\n",
-						rs->sr_entry->e_name.bv_val,
-						a->a_nvals[0].bv_val,
-						ss->ss_ctxcsn[i].bv_val );
-					return LDAP_SUCCESS;
+
+			/* If not a persistent search */
+			if ( !ss->ss_so ) {
+				/* Make sure entry is less than the snapshot'd contextCSN */
+				for ( i=0; i<ss->ss_numcsns; i++ ) {
+					if ( sid == ss->ss_sids[i] && ber_bvcmp( &a->a_nvals[0],
+						&ss->ss_ctxcsn[i] ) > 0 ) {
+						Debug( LDAP_DEBUG_SYNC,
+							"Entry %s CSN %s greater than snapshot %s\n",
+							rs->sr_entry->e_name.bv_val,
+							a->a_nvals[0].bv_val,
+							ss->ss_ctxcsn[i].bv_val );
+						return LDAP_SUCCESS;
+					}
 				}
 			}
 
@@ -1972,42 +2330,70 @@ syncprov_search_response( Operation *op,
 		rs->sr_ctrls = op->o_tmpalloc( sizeof(LDAPControl *)*2,
 			op->o_tmpmemctx );
 		rs->sr_ctrls[1] = NULL;
-		rs->sr_err = syncprov_state_ctrl( op, rs, rs->sr_entry,
-			LDAP_SYNC_ADD, rs->sr_ctrls, 0, 0, NULL );
+		/* If we're in delta-sync mode, always send a cookie */
+		if ( si->si_nopres && si->si_usehint && a ) {
+			struct berval cookie;
+			slap_compose_sync_cookie( op, &cookie, a->a_nvals, srs->sr_state.rid, slap_serverID ? slap_serverID : -1 );
+			rs->sr_err = syncprov_state_ctrl( op, rs, rs->sr_entry,
+				LDAP_SYNC_ADD, rs->sr_ctrls, 0, 1, &cookie );
+		} else {
+			rs->sr_err = syncprov_state_ctrl( op, rs, rs->sr_entry,
+				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 ) &&
+			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 );
 
-		slap_compose_sync_cookie( op, &cookie, ss->ss_ctxcsn,
-			srs->sr_state.rid, srs->sr_state.sid );
+			Debug( LDAP_DEBUG_SYNC, "syncprov_search_response: cookie=%s\n", cookie.bv_val, 0, 0 );
+		}
 
-		/* Is this a regular refresh? */
+		/* Is this a regular refresh?
+		 * Note: refresh never gets here if there were no changes
+		 */
 		if ( !ss->ss_so ) {
 			rs->sr_ctrls = op->o_tmpalloc( sizeof(LDAPControl *)*2,
 				op->o_tmpmemctx );
 			rs->sr_ctrls[1] = NULL;
 			rs->sr_err = syncprov_done_ctrl( op, rs, rs->sr_ctrls,
-				0, 1, &cookie, ss->ss_present ?  LDAP_SYNC_REFRESH_PRESENTS :
+				0, 1, &cookie, ( ss->ss_flags & SS_PRESENT ) ?  LDAP_SYNC_REFRESH_PRESENTS :
 					LDAP_SYNC_REFRESH_DELETES );
 			op->o_tmpfree( cookie.bv_val, op->o_tmpmemctx );
 		} else {
 		/* It's RefreshAndPersist, transition to Persist phase */
-			syncprov_sendinfo( op, rs, ss->ss_present ?
+			syncprov_sendinfo( op, rs, ( ss->ss_flags & SS_PRESENT ) ?
 	 			LDAP_TAG_SYNC_REFRESH_PRESENT : LDAP_TAG_SYNC_REFRESH_DELETE,
-				&cookie, 1, NULL, 0 );
-			op->o_tmpfree( cookie.bv_val, op->o_tmpmemctx );
+				( ss->ss_flags & SS_CHANGED ) ? &cookie : NULL,
+				1, NULL, 0 );
+			if ( !BER_BVISNULL( &cookie ))
+				op->o_tmpfree( cookie.bv_val, op->o_tmpmemctx );
 
 			/* Detach this Op from frontend control */
-			ldap_pvt_thread_mutex_lock( &ss->ss_so->s_mutex );
-
-			/* Turn off the refreshing flag */
-			ss->ss_so->s_flags ^= PS_IS_REFRESHING;
+			ldap_pvt_thread_mutex_lock( &op->o_conn->c_mutex );
 
-			syncprov_detach_op( op, ss->ss_so, on );
+			/* But not if this connection was closed along the way */
+			if ( op->o_abandon ) {
+				ldap_pvt_thread_mutex_unlock( &op->o_conn->c_mutex );
+				/* syncprov_ab_cleanup will free this syncop */
+				return SLAPD_ABANDON;
 
-			/* If there are queued responses, fire them off */
-			if ( ss->ss_so->s_res )
-				syncprov_qstart( ss->ss_so );
-			ldap_pvt_thread_mutex_unlock( &ss->ss_so->s_mutex );
+			} else {
+				ldap_pvt_thread_mutex_lock( &ss->ss_so->s_mutex );
+				/* Turn off the refreshing flag */
+				ss->ss_so->s_flags ^= PS_IS_REFRESHING;
+
+				syncprov_detach_op( op, ss->ss_so, on );
+
+				ldap_pvt_thread_mutex_unlock( &op->o_conn->c_mutex );
+
+				/* If there are queued responses, fire them off */
+				if ( ss->ss_so->s_res )
+					syncprov_qstart( ss->ss_so );
+				ldap_pvt_thread_mutex_unlock( &ss->ss_so->s_mutex );
+			}
 
 			return LDAP_SUCCESS;
 		}
@@ -2022,13 +2408,14 @@ syncprov_op_search( Operation *op, SlapR
 	slap_overinst		*on = (slap_overinst *)op->o_bd->bd_info;
 	syncprov_info_t		*si = (syncprov_info_t *)on->on_bi.bi_private;
 	slap_callback	*cb;
-	int gotstate = 0, nochange = 0, do_present;
+	int gotstate = 0, changed = 0, do_present = 0;
 	syncops *sop = NULL;
 	searchstate *ss;
 	sync_control *srs;
 	BerVarray ctxcsn;
 	int i, *sids, numcsns;
 	struct berval mincsn;
+	int dirty = 0;
 
 	if ( !(op->o_sync_mode & SLAP_SYNC_REFRESH) ) return SLAP_CB_CONTINUE;
 
@@ -2037,10 +2424,7 @@ syncprov_op_search( Operation *op, SlapR
 		return rs->sr_err;
 	}
 
-	do_present = si->si_nopres ? 0 : 1;
-
 	srs = op->o_controls[slap_cids.sc_LDAPsync];
-	op->o_managedsait = SLAP_CONTROL_NONCRITICAL;
 
 	/* If this is a persistent search, set it up right away */
 	if ( op->o_sync_mode & SLAP_SYNC_PERSIST ) {
@@ -2076,6 +2460,20 @@ syncprov_op_search( Operation *op, SlapR
 		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 );
@@ -2093,6 +2491,7 @@ syncprov_op_search( Operation *op, SlapR
 		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 */
@@ -2106,6 +2505,9 @@ syncprov_op_search( Operation *op, SlapR
 		if ( !numcsns )
 			goto no_change;
 
+		if ( !si->si_nopres )
+			do_present = SS_PRESENT;
+
 		/* 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++) {
@@ -2136,21 +2538,42 @@ syncprov_op_search( Operation *op, SlapR
 
 		/* If nothing has changed, shortcut it */
 		if ( srs->sr_state.numcsns == numcsns ) {
-			int i, j, changed = 0;
+			int i, j, newer;
 			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 = 1;
+					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;
+					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;
+					}
 					break;
 				}
 				if ( changed )
 					break;
 			}
-			if ( !changed ) {
-no_change:		nochange = 1;
-				if ( !(op->o_sync_mode & SLAP_SYNC_PERSIST) ) {
+			if ( !changed && !dirty ) {
+				do_present = 0;
+no_change:		if ( !(op->o_sync_mode & SLAP_SYNC_PERSIST) ) {
 					LDAPControl	*ctrls[2];
 
 					ctrls[0] = NULL;
@@ -2165,6 +2588,9 @@ no_change:		nochange = 1;
 				}
 				goto shortcut;
 			}
+		} else {
+			/* consumer doesn't have the right number of CSNs */
+			changed = SS_CHANGED;
 		}
 		/* Do we have a sessionlog for this search? */
 		sl=si->si_logs;
@@ -2188,9 +2614,21 @@ no_change:		nochange = 1;
 			if ( si->si_usehint && srs->sr_rhint == 0 ) {
 				if ( ctxcsn )
 					ber_bvarray_free_x( ctxcsn, op->o_tmpmemctx );
-				send_ldap_error( op, rs, LDAP_SYNC_REFRESH_REQUIRED, "sync cookie is stale" );
-				return rs->sr_err;
+				if ( sids )
+					op->o_tmpfree( sids, op->o_tmpmemctx );
+				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 );
+				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;
 		} else {
 			gotstate = 1;
 			/* If changed and doing Present lookup, send Present UUIDs */
@@ -2198,10 +2636,14 @@ no_change:		nochange = 1;
 				LDAP_SUCCESS ) {
 				if ( ctxcsn )
 					ber_bvarray_free_x( ctxcsn, op->o_tmpmemctx );
-				send_ldap_result( op, rs );
-				return rs->sr_err;
+				if ( sids )
+					op->o_tmpfree( sids, op->o_tmpmemctx );
+				goto bailout;
 			}
 		}
+	} else {
+		/* No consumer state, assume something has changed */
+		changed = SS_CHANGED;
 	}
 
 shortcut:
@@ -2213,7 +2655,7 @@ shortcut:
 	}
 
 	/* If something changed, find the changes */
-	if ( gotstate && !nochange ) {
+	if ( gotstate && ( changed || dirty ) ) {
 		Filter *fand, *fava;
 
 		fand = op->o_tmpalloc( sizeof(Filter), op->o_tmpmemctx );
@@ -2229,10 +2671,14 @@ 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 */
@@ -2240,7 +2686,7 @@ shortcut:
 	ss = (searchstate *)(cb+1);
 	ss->ss_on = on;
 	ss->ss_so = sop;
-	ss->ss_present = do_present;
+	ss->ss_flags = do_present | changed;
 	ss->ss_ctxcsn = ctxcsn;
 	ss->ss_numcsns = numcsns;
 	ss->ss_sids = sids;
@@ -2254,7 +2700,7 @@ shortcut:
 	 * the refresh phase, just invoke the response callback to transition
 	 * us into persist phase
 	 */
-	if ( nochange ) {
+	if ( !changed && !dirty ) {
 		rs->sr_err = LDAP_SUCCESS;
 		rs->sr_nentries = 0;
 		send_ldap_result( op, rs );
@@ -2271,8 +2717,14 @@ syncprov_operational(
 	slap_overinst		*on = (slap_overinst *)op->o_bd->bd_info;
 	syncprov_info_t		*si = (syncprov_info_t *)on->on_bi.bi_private;
 
+	/* This prevents generating unnecessarily; frontend will strip
+	 * any statically stored copy.
+	 */
+	if ( op->o_sync != SLAP_CONTROL_NONE )
+		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 )) {
@@ -2294,10 +2746,7 @@ syncprov_operational(
 				}
 
 				if ( !ap ) {
-					if ( !rs->sr_flags & REP_ENTRY_MODIFIABLE ) {
-						rs->sr_entry = entry_dup( rs->sr_entry );
-						rs->sr_flags |=
-							REP_ENTRY_MODIFIABLE|REP_ENTRY_MUSTBEFREED;
+					if ( rs_ensure_entry_modifiable( op, rs, on )) {
 						a = attr_find( rs->sr_entry->e_attrs,
 							slap_schema.si_ad_contextCSN );
 					}
@@ -2351,7 +2800,11 @@ static ConfigOCs spocs[] = {
 		"NAME 'olcSyncProvConfig' "
 		"DESC 'SyncRepl Provider configuration' "
 		"SUP olcOverlayConfig "
-		"MAY ( olcSpCheckpoint $ olcSpSessionlog $ olcSpNoPresent ) )",
+		"MAY ( olcSpCheckpoint "
+			"$ olcSpSessionlog "
+			"$ olcSpNoPresent "
+			"$ olcSpReloadHint "
+		") )",
 			Cft_Overlay, spcfg },
 	{ NULL, 0, NULL }
 };
@@ -2368,10 +2821,17 @@ sp_cf_gen(ConfigArgs *c)
 		case SP_CHKPT:
 			if ( si->si_chkops || si->si_chktime ) {
 				struct berval bv;
-				bv.bv_len = sprintf( c->cr_msg, "%d %d",
-					si->si_chkops, si->si_chktime );
-				bv.bv_val = c->cr_msg;
-				value_add_one( &c->rvalue_vals, &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/60 );
+				if ( bv.bv_len >= sizeof( c->cr_msg ) ) {
+					rc = 1;
+				} else {
+					bv.bv_val = c->cr_msg;
+					value_add_one( &c->rvalue_vals, &bv );
+				}
 			} else {
 				rc = 1;
 			}
@@ -2471,7 +2931,7 @@ sp_cf_gen(ConfigArgs *c)
 		}
 		sl = si->si_logs;
 		if ( !sl ) {
-			sl = ch_malloc( sizeof( sessionlog ) + LDAP_LUTIL_CSNSTR_BUFSIZE );
+			sl = ch_malloc( sizeof( sessionlog ) + LDAP_PVT_CSNSTR_BUFSIZE );
 			sl->sl_mincsn.bv_val = (char *)(sl+1);
 			sl->sl_mincsn.bv_len = 0;
 			sl->sl_num = 0;
@@ -2541,13 +3001,19 @@ 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 ) {
@@ -2560,7 +3026,7 @@ syncprov_db_open(
 			si->si_sids = slap_parse_csn_sids( si->si_ctxcsn, a->a_numvals, NULL );
 		}
 		overlay_entry_release_ov( op, e, 0, on );
-		if ( si->si_ctxcsn ) {
+		if ( si->si_ctxcsn && !SLAP_DBCLEAN( be )) {
 			op->o_req_dn = be->be_suffix[0];
 			op->o_req_ndn = be->be_nsuffix[0];
 			op->ors_scope = LDAP_SCOPE_SUBTREE;
@@ -2571,7 +3037,7 @@ 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 )) {
@@ -2593,6 +3059,20 @@ syncprov_db_open(
 		si->si_numops++;
 	}
 
+	/* Initialize the sessionlog mincsn */
+	if ( si->si_logs && si->si_numcsns ) {
+		sessionlog *sl = si->si_logs;
+		int i;
+		/* If there are multiple, find the newest */
+		for ( i=0; i < si->si_numcsns; i++ ) {
+			if ( ber_bvcmp( &sl->sl_mincsn, &si->si_ctxcsn[i] ) < 0 ) {
+				AC_MEMCPY( sl->sl_mincsn.bv_val, si->si_ctxcsn[i].bv_val,
+					si->si_ctxcsn[i].bv_len );
+				sl->sl_mincsn.bv_len = si->si_ctxcsn[i].bv_len;
+			}
+		}
+	}
+
 out:
 	op->o_bd->bd_info = (BackendInfo *)on;
 	return 0;
@@ -2608,6 +3088,9 @@ 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;
@@ -2616,18 +3099,31 @@ 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;
 }
 
@@ -2652,6 +3148,7 @@ 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;
@@ -2683,12 +3180,14 @@ syncprov_db_destroy(
 				se = se_next;
 			}
 				
+			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 );
@@ -2786,10 +3285,14 @@ static int syncprov_parseCtrl (
 	sr->sr_rhint = rhint;
 	if (!BER_BVISNULL(&cookie)) {
 		ber_dupbv_x( &sr->sr_state.octet_str, &cookie, op->o_tmpmemctx );
+		/* If parse fails, pretend no cookie was sent */
 		if ( slap_parse_sync_cookie( &sr->sr_state, op->o_tmpmemctx ) ||
 			sr->sr_state.rid == -1 ) {
-			rs->sr_text = "Sync control : cookie parsing error";
-			return LDAP_PROTOCOL_ERROR;
+			if ( sr->sr_state.ctxcsn ) {
+				ber_bvarray_free_x( sr->sr_state.ctxcsn, op->o_tmpmemctx );
+				sr->sr_state.ctxcsn = NULL;
+			}
+			sr->sr_state.numcsns = 0;
 		}
 	}