]> git.sur5r.net Git - openldap/blobdiff - servers/slapd/overlays/syncprov.c
Sync with HEAD for OL 2.4.5
[openldap] / servers / slapd / overlays / syncprov.c
index a6deda4a991a8ac3fafdd0f5c725cf52a4078733..a817700622eedc180dfc723b2a0a089660e3000e 100644 (file)
@@ -2,7 +2,7 @@
 /* syncprov.c - syncrepl provider */
 /* This work is part of OpenLDAP Software <http://www.openldap.org/>.
  *
- * Copyright 2004-2006 The OpenLDAP Foundation.
+ * Copyright 2004-2007 The OpenLDAP Foundation.
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -59,12 +59,14 @@ typedef struct syncops {
        ID              s_eid;          /* entryID of search base */
        Operation       *s_op;          /* search op */
        int             s_rid;
+       int             s_sid;
        struct berval s_filterstr;
        int             s_flags;        /* search status */
 #define        PS_IS_REFRESHING        0x01
 #define        PS_IS_DETACHED          0x02
 #define        PS_WROTE_BASE           0x04
 #define        PS_FIND_BASE            0x08
+#define        PS_FIX_FILTER           0x10
 
        int             s_inuse;        /* reference count */
        struct syncres *s_res;
@@ -102,6 +104,7 @@ typedef struct slog_entry {
        struct slog_entry *se_next;
        struct berval se_uuid;
        struct berval se_csn;
+       int     se_sid;
        ber_tag_t       se_tag;
 } slog_entry;
 
@@ -117,7 +120,9 @@ typedef struct sessionlog {
 /* The main state for this overlay */
 typedef struct syncprov_info_t {
        syncops         *si_ops;
-       struct berval   si_ctxcsn;      /* ldapsync context */
+       BerVarray       si_ctxcsn;      /* ldapsync context */
+       int             *si_sids;
+       int             si_numcsns;
        int             si_chkops;      /* checkpointing info */
        int             si_chktime;
        int             si_numops;      /* number of ops since last checkpoint */
@@ -126,10 +131,9 @@ typedef struct syncprov_info_t {
        time_t  si_chklast;     /* time of last checkpoint */
        Avlnode *si_mods;       /* entries being modified */
        sessionlog      *si_logs;
-       ldap_pvt_thread_mutex_t si_csn_mutex;
+       ldap_pvt_thread_rdwr_t  si_csn_rwlock;
        ldap_pvt_thread_mutex_t si_ops_mutex;
        ldap_pvt_thread_mutex_t si_mods_mutex;
-       char            si_ctxcsnbuf[LDAP_LUTIL_CSNSTR_BUFSIZE];
 } syncprov_info_t;
 
 typedef struct opcookie {
@@ -400,6 +404,7 @@ syncprov_findbase( Operation *op, fbase_cookie *fc )
                slap_callback cb = {0};
                Operation fop;
                SlapReply frs = { REP_RESULT };
+               BackendInfo *bi;
                int rc;
 
                fc->fss->s_flags ^= PS_FIND_BASE;
@@ -411,6 +416,7 @@ syncprov_findbase( Operation *op, fbase_cookie *fc )
                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;
@@ -428,9 +434,8 @@ syncprov_findbase( Operation *op, fbase_cookie *fc )
                fop.ors_filter = &generic_filter;
                fop.ors_filterstr = generic_filterstr;
 
-               fop.o_bd->bd_info = on->on_info->oi_orig;
-               rc = fop.o_bd->be_search( &fop, &frs );
-               fop.o_bd->bd_info = (BackendInfo *)on;
+               rc = overlay_op_walk( &fop, &frs, op_search, on->on_info, on );
+               op->o_bd->bd_info = bi;
        } else {
                ldap_pvt_thread_mutex_unlock( &fc->fss->s_mutex );
                fc->fbase = 1;
@@ -579,7 +584,7 @@ syncprov_findcsn( Operation *op, find_csn_t mode )
        char buf[LDAP_LUTIL_CSNSTR_BUFSIZE + STRLENOF("(entryCSN<=)")];
        char cbuf[LDAP_LUTIL_CSNSTR_BUFSIZE];
        struct berval maxcsn;
-       Filter cf, af;
+       Filter cf;
 #ifdef LDAP_COMP_MATCH
        AttributeAssertion eq = { NULL, BER_BVNULL, NULL };
 #else
@@ -589,13 +594,10 @@ syncprov_findcsn( Operation *op, find_csn_t mode )
        sync_control *srs = NULL;
        struct slap_limits_set fc_limits;
        int i, rc = LDAP_SUCCESS, findcsn_retry = 1;
+       int maxid = 0;
 
        if ( mode != FIND_MAXCSN ) {
                srs = op->o_controls[slap_cids.sc_LDAPsync];
-
-               if ( srs->sr_state.ctxcsn.bv_len >= LDAP_LUTIL_CSNSTR_BUFSIZE ) {
-                       return LDAP_OTHER;
-               }
        }
 
        fop = *op;
@@ -605,6 +607,7 @@ syncprov_findcsn( Operation *op, find_csn_t mode )
 
        cf.f_ava = &eq;
        cf.f_av_desc = slap_schema.si_ad_entryCSN;
+       BER_BVZERO( &cf.f_av_value );
        cf.f_next = NULL;
 
        fop.o_callback = &cb;
@@ -617,7 +620,14 @@ again:
        switch( mode ) {
        case FIND_MAXCSN:
                cf.f_choice = LDAP_FILTER_GE;
-               cf.f_av_value = si->si_ctxcsn;
+               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];
+                               maxid = i;
+                       }
+               }
                fop.ors_filterstr.bv_len = sprintf( buf, "(entryCSN>=%s)",
                        cf.f_av_value.bv_val );
                fop.ors_attrsonly = 0;
@@ -625,12 +635,21 @@ again:
                fop.ors_slimit = SLAP_NO_LIMIT;
                cb.sc_private = &maxcsn;
                cb.sc_response = findmax_cb;
-               strcpy( cbuf, si->si_ctxcsn.bv_val );
+               strcpy( cbuf, cf.f_av_value.bv_val );
                maxcsn.bv_val = cbuf;
-               maxcsn.bv_len = si->si_ctxcsn.bv_len;
+               maxcsn.bv_len = cf.f_av_value.bv_len;
                break;
        case FIND_CSN:
-               cf.f_av_value = srs->sr_state.ctxcsn;
+               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];
+                               }
+                       }
+               }
                /* Look for exact match the first time */
                if ( findcsn_retry ) {
                        cf.f_choice = LDAP_FILTER_EQUALITY;
@@ -640,6 +659,7 @@ again:
                } 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 );
@@ -651,14 +671,8 @@ again:
                cb.sc_response = findcsn_cb;
                break;
        case FIND_PRESENT:
-               af.f_choice = LDAP_FILTER_AND;
-               af.f_next = NULL;
-               af.f_and = &cf;
-               cf.f_choice = LDAP_FILTER_LE;
-               cf.f_av_value = srs->sr_state.ctxcsn;
-               cf.f_next = op->ors_filter;
-               fop.ors_filter = &af;
-               filter2bv_x( &fop, fop.ors_filter, &fop.ors_filterstr );
+               fop.ors_filter = op->ors_filter;
+               fop.ors_filterstr = op->ors_filterstr;
                fop.ors_attrsonly = 0;
                fop.ors_attrs = uuid_anlist;
                fop.ors_slimit = SLAP_NO_LIMIT;
@@ -686,8 +700,10 @@ again:
 
        switch( mode ) {
        case FIND_MAXCSN:
-               strcpy( si->si_ctxcsnbuf, maxcsn.bv_val );
-               si->si_ctxcsn.bv_len = maxcsn.bv_len;
+               if ( ber_bvcmp( &si->si_ctxcsn[maxid], &maxcsn )) {
+                       ber_bvreplace( &si->si_ctxcsn[maxid], &maxcsn );
+                       si->si_numops++;        /* ensure a checkpoint */
+               }
                break;
        case FIND_CSN:
                /* If matching CSN was not found, invalidate the context. */
@@ -702,7 +718,6 @@ again:
                break;
        case FIND_PRESENT:
                op->o_tmpfree( pcookie.uuids, op->o_tmpmemctx );
-               op->o_tmpfree( fop.ors_filterstr.bv_val, op->o_tmpmemctx );
                break;
        }
 
@@ -747,7 +762,7 @@ syncprov_sendresp( Operation *op, opcookie *opc, syncops *so,
 
        SlapReply rs = { REP_SEARCH };
        LDAPControl *ctrls[2];
-       struct berval cookie;
+       struct berval cookie, csns[2];
        Entry e_uuid = {0};
        Attribute a_uuid = {0};
 
@@ -755,7 +770,9 @@ syncprov_sendresp( Operation *op, opcookie *opc, syncops *so,
                return SLAPD_ABANDON;
 
        ctrls[1] = NULL;
-       slap_compose_sync_cookie( op, &cookie, &opc->sctxcsn, so->s_rid );
+       csns[0] = opc->sctxcsn;
+       BER_BVZERO( &csns[1] );
+       slap_compose_sync_cookie( op, &cookie, csns, so->s_rid, so->s_sid );
 
        e_uuid.e_attrs = &a_uuid;
        a_uuid.a_desc = slap_schema.si_ad_entryUUID;
@@ -773,7 +790,7 @@ syncprov_sendresp( Operation *op, opcookie *opc, syncops *so,
                        rs.sr_flags = REP_ENTRY_MUSTRELEASE;
                if ( opc->sreference ) {
                        rs.sr_ref = get_entry_referrals( op, rs.sr_entry );
-                       send_search_reference( op, &rs );
+                       rs.sr_err = send_search_reference( op, &rs );
                        ber_bvarray_free( rs.sr_ref );
                        if ( !rs.sr_entry )
                                *e = NULL;
@@ -785,7 +802,7 @@ syncprov_sendresp( Operation *op, opcookie *opc, syncops *so,
                if ( rs.sr_entry->e_private )
                        rs.sr_flags = REP_ENTRY_MUSTRELEASE;
                rs.sr_attrs = op->ors_attrs;
-               send_search_entry( op, &rs );
+               rs.sr_err = send_search_entry( op, &rs );
                if ( !rs.sr_entry )
                        *e = NULL;
                break;
@@ -797,9 +814,9 @@ syncprov_sendresp( Operation *op, opcookie *opc, syncops *so,
                if ( opc->sreference ) {
                        struct berval bv = BER_BVNULL;
                        rs.sr_ref = &bv;
-                       send_search_reference( op, &rs );
+                       rs.sr_err = send_search_reference( op, &rs );
                } else {
-                       send_search_entry( op, &rs );
+                       rs.sr_err = send_search_entry( op, &rs );
                }
                break;
        default:
@@ -821,10 +838,9 @@ syncprov_qplay( Operation *op, slap_overinst *on, syncops *so )
        syncres *sr;
        Entry *e;
        opcookie opc;
-       int rc;
+       int rc = 0;
 
        opc.son = on;
-       op->o_bd->bd_info = (BackendInfo *)on->on_info;
 
        for (;;) {
                ldap_pvt_thread_mutex_lock( &so->s_mutex );
@@ -846,16 +862,19 @@ syncprov_qplay( Operation *op, slap_overinst *on, syncops *so )
                e = NULL;
 
                if ( sr->s_mode != LDAP_SYNC_DELETE ) {
-                       rc = be_entry_get_rw( op, &opc.sndn, NULL, NULL, 0, &e );
+                       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 ( e ) {
-                       be_entry_release_rw( op, e, 0 );
+                       overlay_entry_release_ov( op, e, 0, on );
                }
 
                ch_free( sr );
@@ -863,7 +882,6 @@ syncprov_qplay( Operation *op, slap_overinst *on, syncops *so )
                if ( rc )
                        break;
        }
-       op->o_bd->bd_info = (BackendInfo *)on;
        return rc;
 }
 
@@ -877,16 +895,17 @@ syncprov_qtask( void *ctx, void *arg )
        OperationBuffer opbuf;
        Operation *op;
        BackendDB be;
+       int rc;
 
-       op = (Operation *) &opbuf;
+       op = &opbuf.ob_op;
        *op = *so->s_op;
-       op->o_hdr = (Opheader *)(op+1);
-       op->o_controls = (void **)(op->o_hdr+1);
-       memset( op->o_controls, 0, SLAP_MAX_CIDS * sizeof(void *));
+       op->o_hdr = &opbuf.ob_hdr;
+       op->o_controls = opbuf.ob_controls;
+       memset( op->o_controls, 0, sizeof(opbuf.ob_controls) );
 
        *op->o_hdr = *so->s_op->o_hdr;
 
-       op->o_tmpmemctx = slap_sl_mem_create(SLAP_SLAB_SIZE, SLAP_SLAB_STACK, ctx);
+       op->o_tmpmemctx = slap_sl_mem_create(SLAP_SLAB_SIZE, SLAP_SLAB_STACK, ctx, 1);
        op->o_tmpmfuncs = &slap_sl_mfuncs;
        op->o_threadctx = ctx;
 
@@ -897,25 +916,74 @@ syncprov_qtask( void *ctx, void *arg )
        op->o_private = NULL;
        op->o_callback = NULL;
 
-       syncprov_qplay( op, on, so );
+       rc = syncprov_qplay( op, on, 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, so->s_qtask );
-       ldap_pvt_runqueue_resched( &slapd_rq, so->s_qtask, 1 );
+       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;
 }
 
+/* Start the task to play back queued psearch responses */
+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();
+}
+
 /* Queue a persistent search response */
 static int
 syncprov_qresp( opcookie *opc, syncops *so, int mode )
 {
        syncres *sr;
+       int sid;
+
+       /* 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;
 
        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 );
@@ -949,22 +1017,7 @@ syncprov_qresp( opcookie *opc, syncops *so, int mode )
                so->s_flags |= PS_FIND_BASE;
        }
        if ( so->s_flags & PS_IS_DETACHED ) {
-               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;
-               } 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;
-                       }
-               }
-               ldap_pvt_thread_mutex_unlock( &slapd_rq.rq_mutex );
+               syncprov_qstart( so );
        }
        ldap_pvt_thread_mutex_unlock( &so->s_mutex );
        return LDAP_SUCCESS;
@@ -978,7 +1031,7 @@ syncprov_drop_psearch( syncops *so, int lock )
                        ldap_pvt_thread_mutex_lock( &so->s_op->o_conn->c_mutex );
                so->s_op->o_conn->c_n_ops_executing--;
                so->s_op->o_conn->c_n_ops_completed++;
-               LDAP_STAILQ_REMOVE( &so->s_op->o_conn->c_ops, so->s_op, slap_op,
+               LDAP_STAILQ_REMOVE( &so->s_op->o_conn->c_ops, so->s_op, Operation,
                        o_next );
                if ( lock )
                        ldap_pvt_thread_mutex_unlock( &so->s_op->o_conn->c_mutex );
@@ -1045,11 +1098,12 @@ syncprov_matchops( Operation *op, opcookie *opc, int saveit )
 
        fbase_cookie fc;
        syncops *ss, *sprev, *snext;
-       Entry *e;
+       Entry *e = NULL;
        Attribute *a;
        int rc;
        struct berval newdn;
        int freefdn = 0;
+       BackendDB *b0 = op->o_bd, db;
 
        fc.fdn = &op->o_req_ndn;
        /* compute new DN */
@@ -1062,16 +1116,21 @@ syncprov_matchops( Operation *op, opcookie *opc, int saveit )
                freefdn = 1;
        }
        if ( op->o_tag != LDAP_REQ_ADD ) {
-               op->o_bd->bd_info = (BackendInfo *)on->on_info;
-               rc = be_entry_get_rw( op, fc.fdn, NULL, NULL, 0, &e );
+               if ( !SLAP_ISOVERLAY( op->o_bd )) {
+                       db = *op->o_bd;
+                       op->o_bd = &db;
+               }
+               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 );
-                       be_entry_release_rw( op, e, 0 );
+                       overlay_entry_release_ov( op, e, 0, on );
                        e = e2;
                }
-               op->o_bd->bd_info = (BackendInfo *)on;
-               if ( rc ) return;
+               if ( rc ) {
+                       op->o_bd = b0;
+                       return;
+               }
        } else {
                e = op->ora_e;
        }
@@ -1162,13 +1221,16 @@ syncprov_matchops( Operation *op, opcookie *opc, int saveit )
        ldap_pvt_thread_mutex_unlock( &si->si_ops_mutex );
 
        if ( op->o_tag != LDAP_REQ_ADD && e ) {
-               op->o_bd->bd_info = (BackendInfo *)on->on_info;
-               be_entry_release_rw( op, e, 0 );
-               op->o_bd->bd_info = (BackendInfo *)on;
+               if ( !SLAP_ISOVERLAY( op->o_bd )) {
+                       op->o_bd = &db;
+               }
+               overlay_entry_release_ov( op, e, 0, on );
+               op->o_bd = b0;
        }
        if ( freefdn ) {
                op->o_tmpfree( fc.fdn->bv_val, op->o_tmpmemctx );
        }
+       op->o_bd = b0;
 }
 
 static int
@@ -1223,16 +1285,13 @@ syncprov_op_cleanup( Operation *op, SlapReply *rs )
 static void
 syncprov_checkpoint( Operation *op, SlapReply *rs, slap_overinst *on )
 {
-       syncprov_info_t         *si = on->on_bi.bi_private;
+       syncprov_info_t *si = (syncprov_info_t *)on->on_bi.bi_private;
        Modifications mod;
        Operation opm;
        SlapReply rsm = { 0 };
-       struct berval bv[2];
        slap_callback cb = {0};
 
-       mod.sml_values = bv;
-       bv[1].bv_val = NULL;
-       bv[0] = si->si_ctxcsn;
+       mod.sml_values = si->si_ctxcsn;
        mod.sml_nvalues = NULL;
        mod.sml_desc = slap_schema.si_ad_contextCSN;
        mod.sml_op = LDAP_MOD_REPLACE;
@@ -1244,20 +1303,21 @@ syncprov_checkpoint( Operation *op, SlapReply *rs, slap_overinst *on )
        opm.o_tag = LDAP_REQ_MODIFY;
        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];
        opm.o_bd->bd_info = on->on_info->oi_orig;
        opm.o_managedsait = SLAP_CONTROL_NONCRITICAL;
-       SLAP_DBFLAGS( opm.o_bd ) |= SLAP_DBFLAG_NOLASTMOD;
+       opm.o_no_schema_check = 1;
        opm.o_bd->be_modify( &opm, &rsm );
-       SLAP_DBFLAGS( opm.o_bd ) ^= SLAP_DBFLAG_NOLASTMOD;
        if ( mod.sml_next != NULL ) {
                slap_mods_free( mod.sml_next, 1 );
        }
+       opm.orm_no_opattrs = 0;
 }
 
 static void
-syncprov_add_slog( Operation *op, struct berval *csn )
+syncprov_add_slog( Operation *op )
 {
        opcookie *opc = op->o_callback->sc_private;
        slap_overinst *on = opc->son;
@@ -1269,7 +1329,7 @@ syncprov_add_slog( Operation *op, struct berval *csn )
        {
                /* Allocate a record. UUIDs are not NUL-terminated. */
                se = ch_malloc( sizeof( slog_entry ) + opc->suuid.bv_len + 
-                       csn->bv_len + 1 );
+                       op->o_csn.bv_len + 1 );
                se->se_next = NULL;
                se->se_tag = op->o_tag;
 
@@ -1278,9 +1338,10 @@ syncprov_add_slog( Operation *op, struct berval *csn )
                se->se_uuid.bv_len = opc->suuid.bv_len;
 
                se->se_csn.bv_val = se->se_uuid.bv_val + opc->suuid.bv_len;
-               AC_MEMCPY( se->se_csn.bv_val, csn->bv_val, csn->bv_len );
-               se->se_csn.bv_val[csn->bv_len] = '\0';
-               se->se_csn.bv_len = csn->bv_len;
+               AC_MEMCPY( se->se_csn.bv_val, op->o_csn.bv_val, op->o_csn.bv_len );
+               se->se_csn.bv_val[op->o_csn.bv_len] = '\0';
+               se->se_csn.bv_len = op->o_csn.bv_len;
+               se->se_sid = slap_parse_csn_sid( &se->se_csn );
 
                ldap_pvt_thread_mutex_lock( &sl->sl_mutex );
                if ( sl->sl_head ) {
@@ -1297,9 +1358,6 @@ syncprov_add_slog( Operation *op, struct berval *csn )
                        sl->sl_mincsn.bv_len = se->se_csn.bv_len;
                        ch_free( se );
                        sl->sl_num--;
-                       if ( !sl->sl_head ) {
-                               sl->sl_tail = NULL;
-                       }
                }
                ldap_pvt_thread_mutex_unlock( &sl->sl_mutex );
        }
@@ -1318,12 +1376,14 @@ playlog_cb( Operation *op, SlapReply *rs )
 /* enter with sl->sl_mutex locked, release before returning */
 static void
 syncprov_playlog( Operation *op, SlapReply *rs, sessionlog *sl,
-       struct berval *oldcsn, struct berval *ctxcsn )
+       sync_control *srs, BerVarray ctxcsn, int numcsns, int *sids )
 {
        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];
        BerVarray uuids;
+       struct berval delcsn[2];
 
        if ( !sl->sl_num ) {
                ldap_pvt_thread_mutex_unlock( &sl->sl_mutex );
@@ -1336,19 +1396,49 @@ syncprov_playlog( Operation *op, SlapReply *rs, sessionlog *sl,
 
        uuids = op->o_tmpalloc( (num+1) * sizeof( struct berval ) +
                num * UUID_LEN, op->o_tmpmemctx );
-
        uuids[0].bv_val = (char *)(uuids + num + 1);
 
+       delcsn[0].bv_len = 0;
+       delcsn[0].bv_val = cbuf;
+       BER_BVZERO(&delcsn[1]);
+
        /* Make a copy of the relevant UUIDs. Put the Deletes up front
         * and everything else at the end. Do this first so we can
         * unlock the list mutex.
         */
+       Debug( LDAP_DEBUG_SYNC, "srs csn %s\n",
+               srs->sr_state.ctxcsn[0].bv_val, 0, 0 );
        for ( se=sl->sl_head; se; se=se->se_next ) {
-               if ( ber_bvcmp( &se->se_csn, oldcsn ) < 0 ) continue;
-               if ( ber_bvcmp( &se->se_csn, ctxcsn ) > 0 ) break;
+               int k;
+               Debug( LDAP_DEBUG_SYNC, "log csn %s\n", se->se_csn.bv_val, 0, 0 );
+               ndel = 1;
+               for ( k=0; k<srs->sr_state.numcsns; k++ ) {
+                       if ( se->se_sid == srs->sr_state.sids[k] ) {
+                               ndel = ber_bvcmp( &se->se_csn, &srs->sr_state.ctxcsn[k] );
+                               break;
+                       }
+               }
+               if ( ndel <= 0 ) {
+                       Debug( LDAP_DEBUG_SYNC, "cmp %d, too old\n", ndel, 0, 0 );
+                       continue;
+               }
+               ndel = 0;
+               for ( k=0; k<numcsns; k++ ) {
+                       if ( se->se_sid == sids[k] ) {
+                               ndel = ber_bvcmp( &se->se_csn, &ctxcsn[k] );
+                               break;
+                       }
+               }
+               if ( ndel > 0 ) {
+                       Debug( LDAP_DEBUG_SYNC, "cmp %d, too new\n", ndel, 0, 0 );
+                       break;
+               }
                if ( se->se_tag == LDAP_REQ_DELETE ) {
                        j = i;
                        i++;
+                       AC_MEMCPY( cbuf, se->se_csn.bv_val, se->se_csn.bv_len );
+                       delcsn[0].bv_len = se->se_csn.bv_len;
+                       delcsn[0].bv_val[delcsn[0].bv_len] = '\0';
                } else {
                        nmods++;
                        j = num - nmods;
@@ -1429,6 +1519,7 @@ syncprov_playlog( Operation *op, SlapReply *rs, sessionlog *sl,
                        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 */
@@ -1439,9 +1530,15 @@ syncprov_playlog( Operation *op, SlapReply *rs, sessionlog *sl,
                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 );
                uuids[ndel].bv_val = NULL;
-               syncprov_sendinfo( op, rs, LDAP_TAG_SYNC_ID_SET, NULL, 0, uuids, 1 );
+               syncprov_sendinfo( op, rs, LDAP_TAG_SYNC_ID_SET, &cookie, 0, uuids, 1 );
+               op->o_tmpfree( cookie.bv_val, op->o_tmpmemctx );
        }
+       op->o_tmpfree( uuids, op->o_tmpmemctx );
 }
 
 static int
@@ -1454,31 +1551,45 @@ syncprov_op_response( Operation *op, SlapReply *rs )
 
        if ( rs->sr_err == LDAP_SUCCESS )
        {
-               struct berval maxcsn = BER_BVNULL, curcsn = BER_BVNULL;
+               struct berval maxcsn = BER_BVNULL;
                char cbuf[LDAP_LUTIL_CSNSTR_BUFSIZE];
+               int do_check = 0;
 
                /* Update our context CSN */
                cbuf[0] = '\0';
-               ldap_pvt_thread_mutex_lock( &si->si_csn_mutex );
-               slap_get_commit_csn( op, &maxcsn, &curcsn );
+               ldap_pvt_thread_rdwr_wlock( &si->si_csn_rwlock );
+               slap_get_commit_csn( op, &maxcsn );
                if ( !BER_BVISNULL( &maxcsn ) ) {
+                       int i, sid;
                        strcpy( cbuf, maxcsn.bv_val );
-                       if ( ber_bvcmp( &maxcsn, &si->si_ctxcsn ) > 0 ) {
-                               strcpy( si->si_ctxcsnbuf, cbuf );
-                               si->si_ctxcsn.bv_len = maxcsn.bv_len;
+                       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 );
+                                       }
+                                       break;
+                               }
+                       }
+                       /* It's a new SID for us */
+                       if ( i == si->si_numcsns ) {
+                               value_add_one( &si->si_ctxcsn, &maxcsn );
+                               si->si_numcsns++;
+                               si->si_sids = ch_realloc( si->si_sids, si->si_numcsns *
+                                       sizeof(int));
+                               si->si_sids[i] = sid;
                        }
                }
 
                /* Don't do any processing for consumer contextCSN updates */
                if ( SLAP_SYNC_SHADOW( op->o_bd ) && 
                        op->o_msgid == SLAP_SYNC_UPDATE_MSGID ) {
-                       ldap_pvt_thread_mutex_unlock( &si->si_csn_mutex );
+                       ldap_pvt_thread_rdwr_wunlock( &si->si_csn_rwlock );
                        return SLAP_CB_CONTINUE;
                }
 
                si->si_numops++;
                if ( si->si_chkops || si->si_chktime ) {
-                       int do_check=0;
                        if ( si->si_chkops && si->si_numops >= si->si_chkops ) {
                                do_check = 1;
                                si->si_numops = 0;
@@ -1488,11 +1599,14 @@ syncprov_op_response( Operation *op, SlapReply *rs )
                                do_check = 1;
                                si->si_chklast = op->o_time;
                        }
-                       if ( do_check ) {
-                               syncprov_checkpoint( op, rs, on );
-                       }
                }
-               ldap_pvt_thread_mutex_unlock( &si->si_csn_mutex );
+               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 );
+                       ldap_pvt_thread_rdwr_runlock( &si->si_csn_rwlock );
+               }
 
                opc->sctxcsn.bv_len = maxcsn.bv_len;
                opc->sctxcsn.bv_val = cbuf;
@@ -1521,7 +1635,7 @@ syncprov_op_response( Operation *op, SlapReply *rs )
 
                /* Add any log records */
                if ( si->si_logs && op->o_tag != LDAP_REQ_ADD ) {
-                       syncprov_add_slog( op, &curcsn );
+                       syncprov_add_slog( op );
                }
 
        }
@@ -1544,19 +1658,17 @@ syncprov_op_compare( Operation *op, SlapReply *rs )
        {
                Entry e = {0};
                Attribute a = {0};
-               struct berval bv[2];
 
                e.e_name = op->o_bd->be_suffix[0];
                e.e_nname = op->o_bd->be_nsuffix[0];
-
-               BER_BVZERO( &bv[1] );
-               bv[0] = si->si_ctxcsn;
+               e.e_attrs = &a;
 
                a.a_desc = slap_schema.si_ad_contextCSN;
-               a.a_vals = bv;
-               a.a_nvals = a.a_vals;
 
-               ldap_pvt_thread_mutex_lock( &si->si_csn_mutex );
+               ldap_pvt_thread_rdwr_rlock( &si->si_csn_rwlock );
+
+               a.a_vals = si->si_ctxcsn;
+               a.a_nvals = a.a_vals;
 
                rs->sr_err = access_allowed( op, &e, op->oq_compare.rs_ava->aa_desc,
                        &op->oq_compare.rs_ava->aa_value, ACL_COMPARE, NULL );
@@ -1585,7 +1697,7 @@ syncprov_op_compare( Operation *op, SlapReply *rs )
 
 return_results:;
 
-               ldap_pvt_thread_mutex_unlock( &si->si_csn_mutex );
+               ldap_pvt_thread_rdwr_runlock( &si->si_csn_rwlock );
 
                send_ldap_result( op, rs );
 
@@ -1684,9 +1796,10 @@ syncprov_op_extended( Operation *op, SlapReply *rs )
 typedef struct searchstate {
        slap_overinst *ss_on;
        syncops *ss_so;
+       BerVarray ss_ctxcsn;
+       int *ss_sids;
+       int ss_numcsns;
        int ss_present;
-       struct berval ss_ctxcsn;
-       char ss_csnbuf[LDAP_LUTIL_CSNSTR_BUFSIZE];
 } searchstate;
 
 static int
@@ -1700,9 +1813,17 @@ syncprov_search_cleanup( Operation *op, SlapReply *rs )
        return 0;
 }
 
+typedef struct SyncOperationBuffer {
+       Operation               sob_op;
+       Opheader                sob_hdr;
+       AttributeName   sob_extra;      /* not always present */
+       /* Further data allocated here */
+} SyncOperationBuffer;
+
 static void
 syncprov_detach_op( Operation *op, syncops *so, slap_overinst *on )
 {
+       SyncOperationBuffer *sopbuf2;
        Operation *op2;
        int i, alen = 0;
        size_t size;
@@ -1714,14 +1835,15 @@ syncprov_detach_op( Operation *op, syncops *so, slap_overinst *on )
                alen += op->ors_attrs[i].an_name.bv_len + 1;
        }
        /* Make a new copy of the operation */
-       size = sizeof(Operation) + sizeof(Opheader) +
+       size = offsetof( SyncOperationBuffer, sob_extra ) +
                (i ? ( (i+1) * sizeof(AttributeName) + alen) : 0) +
                op->o_req_dn.bv_len + 1 +
                op->o_req_ndn.bv_len + 1 +
                op->o_ndn.bv_len + 1 +
                so->s_filterstr.bv_len + 1;
-       op2 = (Operation *)ch_calloc( 1, size );
-       op2->o_hdr = (Opheader *)(op2+1);
+       sopbuf2 = ch_calloc( 1, size );
+       op2 = &sopbuf2->sob_op;
+       op2->o_hdr = &sopbuf2->sob_hdr;
 
        /* Copy the fields we care about explicitly, leave the rest alone */
        *op2->o_hdr = *op->o_hdr;
@@ -1731,18 +1853,18 @@ syncprov_detach_op( Operation *op, syncops *so, slap_overinst *on )
        op2->o_request = op->o_request;
        op2->o_private = on;
 
+       ptr = (char *) sopbuf2 + offsetof( SyncOperationBuffer, sob_extra );
        if ( i ) {
-               op2->ors_attrs = (AttributeName *)(op2->o_hdr + 1);
-               ptr = (char *)(op2->ors_attrs+i+1);
+               op2->ors_attrs = (AttributeName *) ptr;
+               ptr = (char *) &op2->ors_attrs[i+1];
                for (i=0; !BER_BVISNULL( &op->ors_attrs[i].an_name ); i++) {
                        op2->ors_attrs[i] = op->ors_attrs[i];
                        op2->ors_attrs[i].an_name.bv_val = ptr;
                        ptr = lutil_strcopy( ptr, op->ors_attrs[i].an_name.bv_val ) + 1;
                }
                BER_BVZERO( &op2->ors_attrs[i].an_name );
-       } else {
-               ptr = (char *)(op2->o_hdr + 1);
        }
+
        op2->o_authz = op->o_authz;
        op2->o_ndn.bv_val = ptr;
        ptr = lutil_strcopy(ptr, op->o_ndn.bv_val) + 1;
@@ -1756,7 +1878,15 @@ syncprov_detach_op( Operation *op, syncops *so, slap_overinst *on )
        op2->ors_filterstr.bv_val = ptr;
        strcpy( ptr, so->s_filterstr.bv_val );
        op2->ors_filterstr.bv_len = so->s_filterstr.bv_len;
-       op2->ors_filter = str2filter( ptr );
+
+       /* Skip the AND/GE clause that we stuck on in front */
+       if ( so->s_flags & PS_FIX_FILTER ) {
+               op2->ors_filter = op->ors_filter->f_and->f_next;
+               so->s_flags ^= PS_FIX_FILTER;
+       } else {
+               op2->ors_filter = op->ors_filter;
+       }
+       op2->ors_filter = filter_dup( op2->ors_filter, NULL );
        so->s_op = op2;
 
        /* Copy any cached group ACLs individually */
@@ -1778,6 +1908,9 @@ syncprov_detach_op( Operation *op, syncops *so, slap_overinst *on )
        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;
 }
 
 static int
@@ -1804,14 +1937,44 @@ syncprov_search_response( Operation *op, SlapReply *rs )
                        a = attr_find( rs->sr_operational_attrs, slap_schema.si_ad_entryCSN );
                }
                if ( a ) {
-                       /* Make sure entry is less than the snapshot'd contextCSN */
-                       if ( ber_bvcmp( &a->a_nvals[0], &ss->ss_ctxcsn ) > 0 )
+                       int i, sid;
+                       sid = slap_parse_csn_sid( &a->a_nvals[0] );
+
+                       /* Don't send changed entries back to the originator */
+                       if ( sid == srs->sr_state.sid && srs->sr_state.numcsns ) {
+                               Debug( LDAP_DEBUG_SYNC,
+                                       "Entry %s changed by peer, ignored\n",
+                                       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;
+                               }
+                       }
 
-                       /* Don't send the ctx entry twice */
-                       if ( !BER_BVISNULL( &srs->sr_state.ctxcsn ) &&
-                               bvmatch( &a->a_nvals[0], &srs->sr_state.ctxcsn ) )
-                               return LDAP_SUCCESS;
+                       /* Don't send old entries twice */
+                       if ( srs->sr_state.ctxcsn ) {
+                               for ( i=0; i<srs->sr_state.numcsns; i++ ) {
+                                       if ( sid == srs->sr_state.sids[i] &&
+                                               ber_bvcmp( &a->a_nvals[0],
+                                                       &srs->sr_state.ctxcsn[i] )<= 0 ) {
+                                               Debug( LDAP_DEBUG_SYNC,
+                                                       "Entry %s CSN %s older or equal to ctx %s\n",
+                                                       rs->sr_entry->e_name.bv_val,
+                                                       a->a_nvals[0].bv_val,
+                                                       srs->sr_state.ctxcsn[i].bv_val );
+                                               return LDAP_SUCCESS;
+                                       }
+                               }
+                       }
                }
                rs->sr_ctrls = op->o_tmpalloc( sizeof(LDAPControl *)*2,
                        op->o_tmpmemctx );
@@ -1821,8 +1984,8 @@ syncprov_search_response( Operation *op, SlapReply *rs )
        } else if ( rs->sr_type == REP_RESULT && rs->sr_err == LDAP_SUCCESS ) {
                struct berval cookie;
 
-               slap_compose_sync_cookie( op, &cookie, &ss->ss_ctxcsn,
-                       srs->sr_state.rid );
+               slap_compose_sync_cookie( op, &cookie, ss->ss_ctxcsn,
+                       srs->sr_state.rid, srs->sr_state.sid );
 
                /* Is this a regular refresh? */
                if ( !ss->ss_so ) {
@@ -1835,7 +1998,7 @@ syncprov_search_response( Operation *op, SlapReply *rs )
                        op->o_tmpfree( cookie.bv_val, op->o_tmpmemctx );
                } else {
                /* It's RefreshAndPersist, transition to Persist phase */
-                       syncprov_sendinfo( op, rs, ( ss->ss_present && rs->sr_nentries ) ?
+                       syncprov_sendinfo( op, rs, ss->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 );
@@ -1847,6 +2010,10 @@ syncprov_search_response( Operation *op, SlapReply *rs )
                        ss->ss_so->s_flags ^= PS_IS_REFRESHING;
 
                        syncprov_detach_op( op, ss->ss_so, on );
+
+                       /* 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;
@@ -1866,8 +2033,9 @@ syncprov_op_search( Operation *op, SlapReply *rs )
        syncops *sop = NULL;
        searchstate *ss;
        sync_control *srs;
-       struct berval ctxcsn;
-       char csnbuf[LDAP_LUTIL_CSNSTR_BUFSIZE];
+       BerVarray ctxcsn;
+       int i, *sids, numcsns;
+       struct berval mincsn;
 
        if ( !(op->o_sync_mode & SLAP_SYNC_REFRESH) ) return SLAP_CB_CONTINUE;
 
@@ -1896,10 +2064,12 @@ syncprov_op_search( Operation *op, SlapReply *rs )
                /* syncprov_findbase expects to be called as a callback... */
                sc.sc_private = &opc;
                opc.son = on;
+               ldap_pvt_thread_mutex_init( &so.s_mutex );
                cb = op->o_callback;
                op->o_callback = &sc;
                rs->sr_err = syncprov_findbase( op, &fc );
                op->o_callback = cb;
+               ldap_pvt_thread_mutex_destroy( &so.s_mutex );
 
                if ( rs->sr_err != LDAP_SUCCESS ) {
                        send_ldap_result( op, rs );
@@ -1909,6 +2079,7 @@ syncprov_op_search( Operation *op, SlapReply *rs )
                *sop = so;
                ldap_pvt_thread_mutex_init( &sop->s_mutex );
                sop->s_rid = srs->sr_state.rid;
+               sop->s_sid = srs->sr_state.sid;
                sop->s_inuse = 1;
 
                ldap_pvt_thread_mutex_lock( &si->si_ops_mutex );
@@ -1918,44 +2089,101 @@ syncprov_op_search( Operation *op, SlapReply *rs )
        }
 
        /* snapshot the ctxcsn */
-       ldap_pvt_thread_mutex_lock( &si->si_csn_mutex );
-       strcpy( csnbuf, si->si_ctxcsnbuf );
-       ctxcsn.bv_len = si->si_ctxcsn.bv_len;
-       ldap_pvt_thread_mutex_unlock( &si->si_csn_mutex );
-       ctxcsn.bv_val = csnbuf;
+       ldap_pvt_thread_rdwr_rlock( &si->si_csn_rwlock );
+       numcsns = si->si_numcsns;
+       if ( numcsns ) {
+               ber_bvarray_dup_x( &ctxcsn, si->si_ctxcsn, op->o_tmpmemctx );
+               sids = op->o_tmpalloc( numcsns * sizeof(int), op->o_tmpmemctx );
+               for ( i=0; i<numcsns; i++ )
+                       sids[i] = si->si_sids[i];
+       } else {
+               ctxcsn = NULL;
+               sids = NULL;
+       }
+       ldap_pvt_thread_rdwr_runlock( &si->si_csn_rwlock );
        
        /* If we have a cookie, handle the PRESENT lookups */
-       if ( !BER_BVISNULL( &srs->sr_state.ctxcsn )) {
+       if ( srs->sr_state.ctxcsn ) {
                sessionlog *sl;
+               int i, j;
 
-               /* The cookie was validated when it was parsed, just use it */
+               /* If we don't have any CSN of our own yet, pretend nothing
+                * has changed.
+                */
+               if ( !numcsns )
+                       goto no_change;
 
-               /* If just Refreshing and nothing has changed, shortcut it */
-               if ( bvmatch( &srs->sr_state.ctxcsn, &ctxcsn )) {
-                       nochange = 1;
-                       if ( !(op->o_sync_mode & SLAP_SYNC_PERSIST) ) {
-                               LDAPControl     *ctrls[2];
+               /* 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] ) {
+                                       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];
+                               continue;
+                       }
+                       i++;
+               }
 
-                               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;
+               /* 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 nothing has changed, shortcut it */
+               if ( srs->sr_state.numcsns == numcsns ) {
+                       int i, j, changed = 0;
+                       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;
+                                       break;
+                               }
+                               if ( changed )
+                                       break;
+                       }
+                       if ( !changed ) {
+no_change:             nochange = 1;
+                               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;
                        }
-                       goto shortcut;
                }
                /* Do we have a sessionlog for this search? */
                sl=si->si_logs;
                if ( sl ) {
                        ldap_pvt_thread_mutex_lock( &sl->sl_mutex );
-                       if ( ber_bvcmp( &srs->sr_state.ctxcsn, &sl->sl_mincsn ) >= 0 ) {
+                       /* 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 ) {
                                do_present = 0;
                                /* mutex is unlocked in playlog */
-                               syncprov_playlog( op, rs, sl, &srs->sr_state.ctxcsn, &ctxcsn );
+                               syncprov_playlog( op, rs, sl, srs, ctxcsn, numcsns, sids );
                        } else {
                                ldap_pvt_thread_mutex_unlock( &sl->sl_mutex );
                        }
@@ -1965,6 +2193,8 @@ syncprov_op_search( Operation *op, SlapReply *rs )
                        /* No, so a reload is required */
                        /* the 2.2 consumer doesn't send this hint */
                        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;
                        }
@@ -1973,6 +2203,8 @@ syncprov_op_search( Operation *op, SlapReply *rs )
                        /* If changed and doing Present lookup, send Present UUIDs */
                        if ( do_present && syncprov_findcsn( op, FIND_PRESENT ) !=
                                LDAP_SUCCESS ) {
+                               if ( ctxcsn )
+                                       ber_bvarray_free_x( ctxcsn, op->o_tmpmemctx );
                                send_ldap_result( op, rs );
                                return rs->sr_err;
                        }
@@ -2002,10 +2234,12 @@ shortcut:
 #ifdef LDAP_COMP_MATCH
                fava->f_ava->aa_cf = NULL;
 #endif
-               ber_dupbv_x( &fava->f_ava->aa_value, &srs->sr_state.ctxcsn, op->o_tmpmemctx );
+               ber_dupbv_x( &fava->f_ava->aa_value, &mincsn, op->o_tmpmemctx );
                fava->f_next = op->ors_filter;
                op->ors_filter = fand;
                filter2bv_x( op, op->ors_filter, &op->ors_filterstr );
+               if ( sop )
+                       sop->s_flags |= PS_FIX_FILTER;
        }
 
        /* Let our callback add needed info to returned entries */
@@ -2014,19 +2248,15 @@ shortcut:
        ss->ss_on = on;
        ss->ss_so = sop;
        ss->ss_present = do_present;
-       ss->ss_ctxcsn.bv_len = ctxcsn.bv_len;
-       ss->ss_ctxcsn.bv_val = ss->ss_csnbuf;
-       strcpy( ss->ss_ctxcsn.bv_val, ctxcsn.bv_val );
+       ss->ss_ctxcsn = ctxcsn;
+       ss->ss_numcsns = numcsns;
+       ss->ss_sids = sids;
        cb->sc_response = syncprov_search_response;
        cb->sc_cleanup = syncprov_search_cleanup;
        cb->sc_private = ss;
        cb->sc_next = op->o_callback;
        op->o_callback = cb;
 
-#if 0  /* I don't think we need to shortcircuit back-bdb any more */
-       op->o_sync_mode &= SLAP_CONTROL_MASK;
-#endif
-
        /* If this is a persistent search and no changes were reported during
         * the refresh phase, just invoke the response callback to transition
         * us into persist phase
@@ -2060,26 +2290,35 @@ syncprov_operational(
                                        break;
                        }
 
-                       if ( !a ) {
-                               for ( ap = &rs->sr_operational_attrs; *ap; ap=&(*ap)->a_next );
-
-                               a = ch_malloc( sizeof(Attribute));
-                               a->a_desc = slap_schema.si_ad_contextCSN;
-                               a->a_vals = ch_malloc( 2 * sizeof(struct berval));
-                               a->a_vals[1].bv_val = NULL;
-                               a->a_nvals = a->a_vals;
-                               a->a_next = NULL;
-                               a->a_flags = 0;
-                               *ap = a;
-                       }
+                       ldap_pvt_thread_rdwr_rlock( &si->si_csn_rwlock );
+                       if ( si->si_ctxcsn ) {
+                               if ( !a ) {
+                                       for ( ap = &rs->sr_operational_attrs; *ap;
+                                               ap=&(*ap)->a_next );
 
-                       ldap_pvt_thread_mutex_lock( &si->si_csn_mutex );
-                       if ( !ap ) {
-                               strcpy( a->a_vals[0].bv_val, si->si_ctxcsnbuf );
-                       } else {
-                               ber_dupbv( &a->a_vals[0], &si->si_ctxcsn );
+                                       a = attr_alloc( slap_schema.si_ad_contextCSN );
+                                       *ap = a;
+                               }
+
+                               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;
+                                               a = attr_find( rs->sr_entry->e_attrs,
+                                                       slap_schema.si_ad_contextCSN );
+                                       }
+                                       if ( a->a_nvals != a->a_vals ) {
+                                               ber_bvarray_free( a->a_nvals );
+                                       }
+                                       a->a_nvals = NULL;
+                                       ber_bvarray_free( a->a_vals );
+                                       a->a_vals = NULL;
+                               }
+                               ber_bvarray_dup_x( &a->a_vals, si->si_ctxcsn, NULL );
+                               ber_bvarray_dup_x( &a->a_nvals, si->si_ctxcsn, NULL );
                        }
-                       ldap_pvt_thread_mutex_unlock( &si->si_csn_mutex );
+                       ldap_pvt_thread_rdwr_runlock( &si->si_csn_rwlock );
                }
        }
        return SLAP_CB_CONTINUE;
@@ -2136,9 +2375,9 @@ sp_cf_gen(ConfigArgs *c)
                case SP_CHKPT:
                        if ( si->si_chkops || si->si_chktime ) {
                                struct berval bv;
-                               bv.bv_len = sprintf( c->msg, "%d %d",
+                               bv.bv_len = sprintf( c->cr_msg, "%d %d",
                                        si->si_chkops, si->si_chktime );
-                               bv.bv_val = c->msg;
+                               bv.bv_val = c->cr_msg;
                                value_add_one( &c->rvalue_vals, &bv );
                        } else {
                                rc = 1;
@@ -2197,27 +2436,31 @@ sp_cf_gen(ConfigArgs *c)
        switch ( c->type ) {
        case SP_CHKPT:
                if ( lutil_atoi( &si->si_chkops, c->argv[1] ) != 0 ) {
-                       sprintf( c->msg, "%s unable to parse checkpoint ops # \"%s\"",
+                       snprintf( c->cr_msg, sizeof( c->cr_msg ), "%s unable to parse checkpoint ops # \"%s\"",
                                c->argv[0], c->argv[1] );
-                       Debug( LDAP_DEBUG_CONFIG, "%s: %s\n", c->log, c->msg, 0 );
+                       Debug( LDAP_DEBUG_CONFIG|LDAP_DEBUG_NONE,
+                               "%s: %s\n", c->log, c->cr_msg, 0 );
                        return ARG_BAD_CONF;
                }
                if ( si->si_chkops <= 0 ) {
-                       sprintf( c->msg, "%s invalid checkpoint ops # \"%d\"",
+                       snprintf( c->cr_msg, sizeof( c->cr_msg ), "%s invalid checkpoint ops # \"%d\"",
                                c->argv[0], si->si_chkops );
-                       Debug( LDAP_DEBUG_CONFIG, "%s: %s\n", c->log, c->msg, 0 );
+                       Debug( LDAP_DEBUG_CONFIG|LDAP_DEBUG_NONE,
+                               "%s: %s\n", c->log, c->cr_msg, 0 );
                        return ARG_BAD_CONF;
                }
                if ( lutil_atoi( &si->si_chktime, c->argv[2] ) != 0 ) {
-                       sprintf( c->msg, "%s unable to parse checkpoint time \"%s\"",
+                       snprintf( c->cr_msg, sizeof( c->cr_msg ), "%s unable to parse checkpoint time \"%s\"",
                                c->argv[0], c->argv[1] );
-                       Debug( LDAP_DEBUG_CONFIG, "%s: %s\n", c->log, c->msg, 0 );
+                       Debug( LDAP_DEBUG_CONFIG|LDAP_DEBUG_NONE,
+                               "%s: %s\n", c->log, c->cr_msg, 0 );
                        return ARG_BAD_CONF;
                }
                if ( si->si_chktime <= 0 ) {
-                       sprintf( c->msg, "%s invalid checkpoint time \"%d\"",
+                       snprintf( c->cr_msg, sizeof( c->cr_msg ), "%s invalid checkpoint time \"%d\"",
                                c->argv[0], si->si_chkops );
-                       Debug( LDAP_DEBUG_CONFIG, "%s: %s\n", c->log, c->msg, 0 );
+                       Debug( LDAP_DEBUG_CONFIG|LDAP_DEBUG_NONE,
+                               "%s: %s\n", c->log, c->cr_msg, 0 );
                        return ARG_BAD_CONF;
                }
                si->si_chktime *= 60;
@@ -2227,9 +2470,10 @@ sp_cf_gen(ConfigArgs *c)
                int size = c->value_int;
 
                if ( size < 0 ) {
-                       sprintf( c->msg, "%s size %d is negative",
+                       snprintf( c->cr_msg, sizeof( c->cr_msg ), "%s size %d is negative",
                                c->argv[0], size );
-                       Debug( LDAP_DEBUG_CONFIG, "%s: %s\n", c->log, c->msg, 0 );
+                       Debug( LDAP_DEBUG_CONFIG|LDAP_DEBUG_NONE,
+                               "%s: %s\n", c->log, c->cr_msg, 0 );
                        return ARG_BAD_CONF;
                }
                sl = si->si_logs;
@@ -2273,21 +2517,27 @@ syncprov_db_otask(
  */
 static int
 syncprov_db_open(
-    BackendDB *be
+       BackendDB *be,
+       ConfigReply *cr
 )
 {
        slap_overinst   *on = (slap_overinst *) be->bd_info;
        syncprov_info_t *si = (syncprov_info_t *)on->on_bi.bi_private;
 
        Connection conn = { 0 };
-       OperationBuffer opbuf = { 0 };
-       char ctxcsnbuf[LDAP_LUTIL_CSNSTR_BUFSIZE];
-       Operation *op = (Operation *) &opbuf;
-       Entry *e;
+       OperationBuffer opbuf;
+       Operation *op;
+       Entry *e = NULL;
        Attribute *a;
        int rc;
        void *thrctx = NULL;
 
+       if ( !SLAP_LASTMOD( be )) {
+               Debug( LDAP_DEBUG_ANY,
+                       "syncprov_db_open: invalid config, lastmod must be enabled\n", 0, 0, 0 );
+               return -1;
+       }
+
        if ( slapMode & SLAP_TOOL_MODE ) {
                return 0;
        }
@@ -2298,62 +2548,62 @@ syncprov_db_open(
        }
 
        thrctx = ldap_pvt_thread_pool_context();
-       connection_fake_init( &conn, op, thrctx );
+       connection_fake_init( &conn, &opbuf, thrctx );
+       op = &opbuf.ob_op;
        op->o_bd = be;
        op->o_dn = be->be_rootdn;
        op->o_ndn = be->be_rootndn;
 
-       ctxcsnbuf[0] = '\0';
-
-       op->o_bd->bd_info = on->on_info->oi_orig;
-       rc = be_entry_get_rw( op, be->be_nsuffix, NULL,
-               slap_schema.si_ad_contextCSN, 0, &e );
+       rc = overlay_entry_get_ov( op, be->be_nsuffix, NULL,
+               slap_schema.si_ad_contextCSN, 0, &e, on );
 
        if ( e ) {
                ldap_pvt_thread_t tid;
 
                a = attr_find( e->e_attrs, slap_schema.si_ad_contextCSN );
                if ( a ) {
-                       si->si_ctxcsn.bv_len = a->a_nvals[0].bv_len;
-                       if ( si->si_ctxcsn.bv_len >= sizeof(si->si_ctxcsnbuf ))
-                               si->si_ctxcsn.bv_len = sizeof(si->si_ctxcsnbuf)-1;
-                       strncpy( si->si_ctxcsnbuf, a->a_nvals[0].bv_val,
-                               si->si_ctxcsn.bv_len );
-                       si->si_ctxcsnbuf[si->si_ctxcsn.bv_len] = '\0';
-                       strcpy( ctxcsnbuf, si->si_ctxcsnbuf );
-               }
-               be_entry_release_rw( op, e, 0 );
-               if ( !BER_BVISEMPTY( &si->si_ctxcsn ) ) {
-                       op->o_bd->bd_info = (BackendInfo *)on;
+                       int i;
+                       ber_bvarray_dup_x( &si->si_ctxcsn, a->a_vals, NULL );
+                       for ( i = 0; !BER_BVISNULL( &a->a_vals[i] ); i++ );
+                       si->si_numcsns = i;
+                       si->si_sids = slap_parse_csn_sids( si->si_ctxcsn, i, NULL );
+               }
+               overlay_entry_release_ov( op, e, 0, on );
+               if ( si->si_ctxcsn ) {
                        op->o_req_dn = be->be_suffix[0];
                        op->o_req_ndn = be->be_nsuffix[0];
                        op->ors_scope = LDAP_SCOPE_SUBTREE;
                        ldap_pvt_thread_create( &tid, 0, syncprov_db_otask, op );
                        ldap_pvt_thread_join( tid, NULL );
                }
-       } else if ( SLAP_SYNC_SHADOW( op->o_bd )) {
-               /* If we're also a consumer, and we didn't find the context entry,
-                * then don't generate anything, wait for our provider to send it
-                * to us.
-                */
-               goto out;
        }
 
-       if ( BER_BVISEMPTY( &si->si_ctxcsn ) ) {
-               si->si_ctxcsn.bv_len = sizeof( si->si_ctxcsnbuf );
-               slap_get_csn( op, &si->si_ctxcsn, 0 );
-       }
+       /* Didn't find a contextCSN, should we generate one? */
+       if ( !si->si_ctxcsn ) {
+               char csnbuf[ LDAP_LUTIL_CSNSTR_BUFSIZE ];
+               struct berval csn;
 
-       /* If our ctxcsn is different from what was read from the root
-        * entry, make sure we do a checkpoint on close
-        */
-       if ( strcmp( si->si_ctxcsnbuf, ctxcsnbuf )) {
+               if ( SLAP_SYNC_SHADOW( op->o_bd )) {
+               /* If we're also a consumer, then don't generate anything.
+                * Wait for our provider to send it to us, or for a local
+                * modify if we have multimaster.
+                */
+                       goto out;
+               }
+               csn.bv_val = csnbuf;
+               csn.bv_len = sizeof( csnbuf );
+               slap_get_csn( op, &csn, 0 );
+               value_add_one( &si->si_ctxcsn, &csn );
+               si->si_numcsns = 1;
+               si->si_sids = ch_malloc( sizeof(int) );
+               si->si_sids[0] = slap_serverID;
+
+               /* make sure we do a checkpoint on close */
                si->si_numops++;
        }
 
 out:
        op->o_bd->bd_info = (BackendInfo *)on;
-       ldap_pvt_thread_pool_context_reset( thrctx );
        return 0;
 }
 
@@ -2361,7 +2611,8 @@ out:
  */
 static int
 syncprov_db_close(
-    BackendDB *be
+       BackendDB *be,
+       ConfigReply *cr
 )
 {
     slap_overinst   *on = (slap_overinst *) be->bd_info;
@@ -2371,19 +2622,19 @@ syncprov_db_close(
                return 0;
        }
        if ( si->si_numops ) {
-               Connection conn;
+               Connection conn = {0};
                OperationBuffer opbuf;
-               Operation *op = (Operation *) &opbuf;
+               Operation *op;
                SlapReply rs = {REP_RESULT};
                void *thrctx;
 
                thrctx = ldap_pvt_thread_pool_context();
-               connection_fake_init( &conn, op, thrctx );
+               connection_fake_init( &conn, &opbuf, thrctx );
+               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 );
-               ldap_pvt_thread_pool_context_reset( thrctx );
        }
 
     return 0;
@@ -2391,18 +2642,25 @@ syncprov_db_close(
 
 static int
 syncprov_db_init(
-       BackendDB *be
+       BackendDB *be,
+       ConfigReply *cr
 )
 {
        slap_overinst   *on = (slap_overinst *)be->bd_info;
        syncprov_info_t *si;
 
+       if ( SLAP_ISGLOBALOVERLAY( be ) ) {
+               Debug( LDAP_DEBUG_ANY,
+                       "syncprov must be instantiated within a database.\n",
+                       0, 0, 0 );
+               return 1;
+       }
+
        si = ch_calloc(1, sizeof(syncprov_info_t));
        on->on_bi.bi_private = si;
-       ldap_pvt_thread_mutex_init( &si->si_csn_mutex );
+       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 );
-       si->si_ctxcsn.bv_val = si->si_ctxcsnbuf;
 
        csn_anlist[0].an_desc = slap_schema.si_ad_entryCSN;
        csn_anlist[0].an_name = slap_schema.si_ad_entryCSN->ad_cname;
@@ -2417,7 +2675,8 @@ syncprov_db_init(
 
 static int
 syncprov_db_destroy(
-       BackendDB *be
+       BackendDB *be,
+       ConfigReply *cr
 )
 {
        slap_overinst   *on = (slap_overinst *)be->bd_info;
@@ -2435,9 +2694,13 @@ syncprov_db_destroy(
                                
                        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_mods_mutex );
                ldap_pvt_thread_mutex_destroy( &si->si_ops_mutex );
-               ldap_pvt_thread_mutex_destroy( &si->si_csn_mutex );
+               ldap_pvt_thread_rdwr_destroy( &si->si_csn_rwlock );
                ch_free( si );
        }
 
@@ -2468,8 +2731,13 @@ static int syncprov_parseCtrl (
                return LDAP_PROTOCOL_ERROR;
        }
 
+       if ( BER_BVISNULL( &ctrl->ldctl_value ) ) {
+               rs->sr_text = "Sync control value is absent";
+               return LDAP_PROTOCOL_ERROR;
+       }
+
        if ( BER_BVISEMPTY( &ctrl->ldctl_value ) ) {
-               rs->sr_text = "Sync control value is empty (or absent)";
+               rs->sr_text = "Sync control value is empty";
                return LDAP_PROTOCOL_ERROR;
        }
 
@@ -2527,8 +2795,8 @@ static int syncprov_parseCtrl (
        sr->sr_rhint = rhint;
        if (!BER_BVISNULL(&cookie)) {
                ber_dupbv_x( &sr->sr_state.octet_str, &cookie, op->o_tmpmemctx );
-               slap_parse_sync_cookie( &sr->sr_state, op->o_tmpmemctx );
-               if ( sr->sr_state.rid == -1 ) {
+               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;
                }