ReadOnly: fix bug writing incorrect amount of data in delegated record
[sahlberg/ctdb.git] / server / ctdb_call.c
index bbe07717ed830ee84e95742cf88bdf0bbc5220e6..23d10b9210b4f39e132203f96852537b7dc560d0 100644 (file)
@@ -21,7 +21,7 @@
   protocol design and packet details
 */
 #include "includes.h"
-#include "lib/events/events.h"
+#include "lib/tevent/tevent.h"
 #include "lib/tdb/include/tdb.h"
 #include "lib/util/dlinklist.h"
 #include "system/network.h"
@@ -43,7 +43,6 @@
        return ctdb_db;
 }
 
-
 /*
   a varient of input packet that can be used in lock requeue
 */
@@ -69,6 +68,11 @@ static void ctdb_send_error(struct ctdb_context *ctdb,
        char *msg;
        int msglen, len;
 
+       if (ctdb->methods == NULL) {
+               DEBUG(DEBUG_INFO,(__location__ " Failed to send error. Transport is DOWN\n"));
+               return;
+       }
+
        va_start(ap, fmt);
        msg = talloc_vasprintf(ctdb, fmt, ap);
        if (msg == NULL) {
@@ -94,9 +98,30 @@ static void ctdb_send_error(struct ctdb_context *ctdb,
 }
 
 
-/*
-  send a redirect reply
-*/
+/**
+ * send a redirect reply
+ *
+ * The logic behind this function is this:
+ *
+ * A client wants to grab a record and sends a CTDB_REQ_CALL packet
+ * to its local ctdb (ctdb_request_call). If the node is not itself
+ * the record's DMASTER, it first redirects the packet to  the
+ * record's LMASTER. The LMASTER then redirects the call packet to
+ * the current DMASTER. But there is a race: The record may have
+ * been migrated off the DMASTER while the redirected packet is
+ * on the wire (or in the local queue). So in case the record has
+ * migrated off the new destinaton of the call packet, instead of
+ * going back to the LMASTER to get the new DMASTER, we try to
+ * reduce rountrips by fist chasing the record a couple of times
+ * before giving up the direct chase and finally going back to the
+ * LMASTER (again). Note that this works because auf this: When
+ * a record is migrated off a node, then the new DMASTER is stored
+ * in the record's copy on the former DMASTER.
+ *
+ * The maxiumum number of attempts for direct chase to make before
+ * going back to the LMASTER is configurable by the tunable
+ * "MaxRedirectCount".
+ */
 static void ctdb_call_send_redirect(struct ctdb_context *ctdb, 
                                    TDB_DATA key,
                                    struct ctdb_req_call *c, 
@@ -104,7 +129,7 @@ static void ctdb_call_send_redirect(struct ctdb_context *ctdb,
 {
        
        uint32_t lmaster = ctdb_lmaster(ctdb, &key);
-       if (ctdb->vnn == lmaster) {
+       if (ctdb->pnn == lmaster) {
                c->hdr.destnode = header->dmaster;
        } else if ((c->hopcount % ctdb->tunable.max_redirect_count) == 0) {
                c->hdr.destnode = lmaster;
@@ -133,15 +158,20 @@ static void ctdb_send_dmaster_reply(struct ctdb_db_context *ctdb_db,
        int ret, len;
        TALLOC_CTX *tmp_ctx;
 
-       if (ctdb->vnn != ctdb_lmaster(ctdb, &key)) {
-               DEBUG(0,(__location__ " Caller is not lmaster!\n"));
+       if (ctdb->pnn != ctdb_lmaster(ctdb, &key)) {
+               DEBUG(DEBUG_ALERT,(__location__ " Caller is not lmaster!\n"));
                return;
        }
 
        header->dmaster = new_dmaster;
        ret = ctdb_ltdb_store(ctdb_db, key, header, data);
        if (ret != 0) {
-               ctdb_fatal(ctdb, "ctdb_req_dmaster unable to update dmaster");
+               ctdb_fatal(ctdb, "ctdb_send_dmaster_reply unable to update dmaster");
+               return;
+       }
+
+       if (ctdb->methods == NULL) {
+               ctdb_fatal(ctdb, "ctdb_send_dmaster_reply cant update dmaster since transport is down");
                return;
        }
 
@@ -150,7 +180,7 @@ static void ctdb_send_dmaster_reply(struct ctdb_db_context *ctdb_db,
        tmp_ctx = talloc_new(ctdb);
 
        /* send the CTDB_REPLY_DMASTER */
-       len = offsetof(struct ctdb_reply_dmaster, data) + key.dsize + data.dsize;
+       len = offsetof(struct ctdb_reply_dmaster, data) + key.dsize + data.dsize + sizeof(uint32_t);
        r = ctdb_transport_allocate(ctdb, tmp_ctx, CTDB_REPLY_DMASTER, len,
                                    struct ctdb_reply_dmaster);
        CTDB_NO_MEMORY_FATAL(ctdb, r);
@@ -163,6 +193,7 @@ static void ctdb_send_dmaster_reply(struct ctdb_db_context *ctdb_db,
        r->db_id         = ctdb_db->db_id;
        memcpy(&r->data[0], key.dptr, key.dsize);
        memcpy(&r->data[key.dsize], data.dptr, data.dsize);
+       memcpy(&r->data[key.dsize+data.dsize], &header->flags, sizeof(uint32_t));
 
        ctdb_queue_packet(ctdb, &r->hdr);
 
@@ -186,13 +217,23 @@ static void ctdb_call_send_dmaster(struct ctdb_db_context *ctdb_db,
        int len;
        uint32_t lmaster = ctdb_lmaster(ctdb, key);
 
-       if (lmaster == ctdb->vnn) {
+       if (ctdb->methods == NULL) {
+               ctdb_fatal(ctdb, "Failed ctdb_call_send_dmaster since transport is down");
+               return;
+       }
+
+       if (data->dsize != 0) {
+               header->flags |= CTDB_REC_FLAG_MIGRATED_WITH_DATA;
+       }
+
+       if (lmaster == ctdb->pnn) {
                ctdb_send_dmaster_reply(ctdb_db, header, *key, *data, 
                                        c->hdr.srcnode, c->hdr.reqid);
                return;
        }
        
-       len = offsetof(struct ctdb_req_dmaster, data) + key->dsize + data->dsize;
+       len = offsetof(struct ctdb_req_dmaster, data) + key->dsize + data->dsize
+                       + sizeof(uint32_t);
        r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REQ_DMASTER, len, 
                                    struct ctdb_req_dmaster);
        CTDB_NO_MEMORY_FATAL(ctdb, r);
@@ -205,6 +246,7 @@ static void ctdb_call_send_dmaster(struct ctdb_db_context *ctdb_db,
        r->datalen       = data->dsize;
        memcpy(&r->data[0], key->dptr, key->dsize);
        memcpy(&r->data[key->dsize], data->dptr, data->dsize);
+       memcpy(&r->data[key->dsize + data->dsize], &header->flags, sizeof(uint32_t));
 
        header->dmaster = c->hdr.srcnode;
        if (ctdb_ltdb_store(ctdb_db, *key, header, *data) != 0) {
@@ -222,45 +264,86 @@ static void ctdb_call_send_dmaster(struct ctdb_db_context *ctdb_db,
 
   must be called with the chainlock held. This function releases the chainlock
 */
-static void ctdb_become_dmaster(struct ctdb_db_context *ctdb_db, 
-                               uint32_t reqid, TDB_DATA key, TDB_DATA data,
-                               uint64_t rsn)
+static void ctdb_become_dmaster(struct ctdb_db_context *ctdb_db,
+                               struct ctdb_req_header *hdr,
+                               TDB_DATA key, TDB_DATA data,
+                               uint64_t rsn, uint32_t record_flags)
 {
        struct ctdb_call_state *state;
        struct ctdb_context *ctdb = ctdb_db->ctdb;
        struct ctdb_ltdb_header header;
+       int ret;
 
-       DEBUG(2,("vnn %u dmaster response %08x\n", ctdb->vnn, ctdb_hash(&key)));
+       DEBUG(DEBUG_DEBUG,("pnn %u dmaster response %08x\n", ctdb->pnn, ctdb_hash(&key)));
 
        ZERO_STRUCT(header);
        header.rsn = rsn + 1;
-       header.dmaster = ctdb->vnn;
+       header.dmaster = ctdb->pnn;
+       header.flags = record_flags;
+
+       state = ctdb_reqid_find(ctdb, hdr->reqid, struct ctdb_call_state);
+
+       if (state) {
+               if (state->call->flags & CTDB_CALL_FLAG_VACUUM_MIGRATION) {
+                       /*
+                        * We temporarily add the VACUUM_MIGRATED flag to
+                        * the record flags, so that ctdb_ltdb_store can
+                        * decide whether the record should be stored or
+                        * deleted.
+                        */
+                       header.flags |= CTDB_REC_FLAG_VACUUM_MIGRATED;
+               }
+       }
 
        if (ctdb_ltdb_store(ctdb_db, key, &header, data) != 0) {
                ctdb_fatal(ctdb, "ctdb_reply_dmaster store failed\n");
-               ctdb_ltdb_unlock(ctdb_db, key);
+
+               ret = ctdb_ltdb_unlock(ctdb_db, key);
+               if (ret != 0) {
+                       DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
+               }
                return;
        }
 
-       state = ctdb_reqid_find(ctdb, reqid, struct ctdb_call_state);
 
        if (state == NULL) {
-               DEBUG(0,("vnn %u Invalid reqid %u in ctdb_become_dmaster\n",
-                        ctdb->vnn, reqid));
-               ctdb_ltdb_unlock(ctdb_db, key);
+               DEBUG(DEBUG_ERR,("pnn %u Invalid reqid %u in ctdb_become_dmaster from node %u\n",
+                        ctdb->pnn, hdr->reqid, hdr->srcnode));
+
+               ret = ctdb_ltdb_unlock(ctdb_db, key);
+               if (ret != 0) {
+                       DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
+               }
+               return;
+       }
+
+       if (key.dsize != state->call->key.dsize || memcmp(key.dptr, state->call->key.dptr, key.dsize)) {
+               DEBUG(DEBUG_ERR, ("Got bogus DMASTER packet reqid:%u from node %u. Key does not match key held in matching idr.\n", hdr->reqid, hdr->srcnode));
+
+               ret = ctdb_ltdb_unlock(ctdb_db, key);
+               if (ret != 0) {
+                       DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
+               }
                return;
        }
 
-       if (reqid != state->reqid) {
+       if (hdr->reqid != state->reqid) {
                /* we found a record  but it was the wrong one */
-               DEBUG(0, ("Dropped orphan in ctdb_become_dmaster with reqid:%u\n",reqid));
-               ctdb_ltdb_unlock(ctdb_db, key);
+               DEBUG(DEBUG_ERR, ("Dropped orphan in ctdb_become_dmaster with reqid:%u\n from node %u", hdr->reqid, hdr->srcnode));
+
+               ret = ctdb_ltdb_unlock(ctdb_db, key);
+               if (ret != 0) {
+                       DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
+               }
                return;
        }
 
-       ctdb_call_local(ctdb_db, &state->call, &header, state, &data, ctdb->vnn);
+       ctdb_call_local(ctdb_db, state->call, &header, state, &data, true);
 
-       ctdb_ltdb_unlock(ctdb_db, state->call.key);
+       ret = ctdb_ltdb_unlock(ctdb_db, state->call->key);
+       if (ret != 0) {
+               DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
+       }
 
        state->state = CTDB_CALL_DONE;
        if (state->async.fn) {
@@ -282,12 +365,19 @@ void ctdb_request_dmaster(struct ctdb_context *ctdb, struct ctdb_req_header *hdr
        TDB_DATA key, data, data2;
        struct ctdb_ltdb_header header;
        struct ctdb_db_context *ctdb_db;
+       uint32_t record_flags = 0;
+       size_t len;
        int ret;
 
        key.dptr = c->data;
        key.dsize = c->keylen;
        data.dptr = c->data + c->keylen;
        data.dsize = c->datalen;
+       len = offsetof(struct ctdb_req_dmaster, data) + key.dsize + data.dsize
+                       + sizeof(uint32_t);
+       if (len <= c->hdr.length) {
+               record_flags = *(uint32_t *)&c->data[c->keylen + c->datalen];
+       }
 
        ctdb_db = find_ctdb_db(ctdb, c->db_id);
        if (!ctdb_db) {
@@ -305,39 +395,60 @@ void ctdb_request_dmaster(struct ctdb_context *ctdb, struct ctdb_req_header *hdr
                return;
        }
        if (ret == -2) {
-               DEBUG(2,(__location__ " deferring ctdb_request_dmaster\n"));
+               DEBUG(DEBUG_INFO,(__location__ " deferring ctdb_request_dmaster\n"));
                return;
        }
 
-       if (ctdb_lmaster(ctdb, &key) != ctdb->vnn) {
-               DEBUG(0,("vnn %u dmaster request to non-lmaster lmaster=%u gen=%u curgen=%u\n",
-                        ctdb->vnn, ctdb_lmaster(ctdb, &key), 
+       if (ctdb_lmaster(ctdb, &key) != ctdb->pnn) {
+               DEBUG(DEBUG_ALERT,("pnn %u dmaster request to non-lmaster lmaster=%u gen=%u curgen=%u\n",
+                        ctdb->pnn, ctdb_lmaster(ctdb, &key), 
                         hdr->generation, ctdb->vnn_map->generation));
                ctdb_fatal(ctdb, "ctdb_req_dmaster to non-lmaster");
        }
 
-       DEBUG(2,("vnn %u dmaster request on %08x for %u from %u\n", 
-                ctdb->vnn, ctdb_hash(&key), c->dmaster, c->hdr.srcnode));
+       DEBUG(DEBUG_DEBUG,("pnn %u dmaster request on %08x for %u from %u\n", 
+                ctdb->pnn, ctdb_hash(&key), c->dmaster, c->hdr.srcnode));
 
        /* its a protocol error if the sending node is not the current dmaster */
        if (header.dmaster != hdr->srcnode) {
-               DEBUG(0,("vnn %u dmaster request non-master %u dmaster=%u key %08x dbid 0x%08x gen=%u curgen=%u\n",
-                        ctdb->vnn, hdr->srcnode, header.dmaster, ctdb_hash(&key),
-                        ctdb_db->db_id, hdr->generation, ctdb->vnn_map->generation));
-               ctdb_fatal(ctdb, "ctdb_req_dmaster from non-master");
-               return;
+               DEBUG(DEBUG_ALERT,("pnn %u dmaster request for new-dmaster %u from non-master %u real-dmaster=%u key %08x dbid 0x%08x gen=%u curgen=%u c->rsn=%llu header.rsn=%llu reqid=%u keyval=0x%08x\n",
+                        ctdb->pnn, c->dmaster, hdr->srcnode, header.dmaster, ctdb_hash(&key),
+                        ctdb_db->db_id, hdr->generation, ctdb->vnn_map->generation,
+                        (unsigned long long)c->rsn, (unsigned long long)header.rsn, c->hdr.reqid,
+                        (key.dsize >= 4)?(*(uint32_t *)key.dptr):0));
+               if (header.rsn != 0 || header.dmaster != ctdb->pnn) {
+                       DEBUG(DEBUG_ERR,("ctdb_req_dmaster from non-master. Force a recovery.\n"));
+
+                       ctdb->recovery_mode = CTDB_RECOVERY_ACTIVE;
+                       ctdb_ltdb_unlock(ctdb_db, key);
+                       return;
+               }
+       }
+
+       if (header.rsn > c->rsn) {
+               DEBUG(DEBUG_ALERT,("pnn %u dmaster request with older RSN new-dmaster %u from %u real-dmaster=%u key %08x dbid 0x%08x gen=%u curgen=%u c->rsn=%llu header.rsn=%llu reqid=%u\n",
+                        ctdb->pnn, c->dmaster, hdr->srcnode, header.dmaster, ctdb_hash(&key),
+                        ctdb_db->db_id, hdr->generation, ctdb->vnn_map->generation,
+                        (unsigned long long)c->rsn, (unsigned long long)header.rsn, c->hdr.reqid));
        }
 
        /* use the rsn from the sending node */
        header.rsn = c->rsn;
 
+       /* store the record flags from the sending node */
+       header.flags = record_flags;
+
        /* check if the new dmaster is the lmaster, in which case we
           skip the dmaster reply */
-       if (c->dmaster == ctdb->vnn) {
-               ctdb_become_dmaster(ctdb_db, hdr->reqid, key, data, c->rsn);
+       if (c->dmaster == ctdb->pnn) {
+               ctdb_become_dmaster(ctdb_db, hdr, key, data, c->rsn, record_flags);
        } else {
                ctdb_send_dmaster_reply(ctdb_db, &header, key, data, c->dmaster, hdr->reqid);
-               ctdb_ltdb_unlock(ctdb_db, key);
+
+               ret = ctdb_ltdb_unlock(ctdb_db, key);
+               if (ret != 0) {
+                       DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
+               }
        }
 }
 
@@ -352,9 +463,15 @@ void ctdb_request_call(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
        struct ctdb_reply_call *r;
        int ret, len;
        struct ctdb_ltdb_header header;
-       struct ctdb_call call;
+       struct ctdb_call *call;
        struct ctdb_db_context *ctdb_db;
 
+       if (ctdb->methods == NULL) {
+               DEBUG(DEBUG_INFO,(__location__ " Failed ctdb_request_call. Transport is DOWN\n"));
+               return;
+       }
+
+
        ctdb_db = find_ctdb_db(ctdb, c->db_id);
        if (!ctdb_db) {
                ctdb_send_error(ctdb, hdr, -1,
@@ -363,70 +480,196 @@ void ctdb_request_call(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
                return;
        }
 
-       call.call_id  = c->callid;
-       call.key.dptr = c->data;
-       call.key.dsize = c->keylen;
-       call.call_data.dptr = c->data + c->keylen;
-       call.call_data.dsize = c->calldatalen;
+       call = talloc(hdr, struct ctdb_call);
+       CTDB_NO_MEMORY_FATAL(ctdb, call);
+
+       call->call_id  = c->callid;
+       call->key.dptr = c->data;
+       call->key.dsize = c->keylen;
+       call->call_data.dptr = c->data + c->keylen;
+       call->call_data.dsize = c->calldatalen;
+       call->reply_data.dptr  = NULL;
+       call->reply_data.dsize = 0;
 
        /* determine if we are the dmaster for this key. This also
           fetches the record data (if any), thus avoiding a 2nd fetch of the data 
           if the call will be answered locally */
 
-       ret = ctdb_ltdb_lock_fetch_requeue(ctdb_db, call.key, &header, hdr, &data,
+       ret = ctdb_ltdb_lock_fetch_requeue(ctdb_db, call->key, &header, hdr, &data,
                                           ctdb_call_input_pkt, ctdb, False);
        if (ret == -1) {
                ctdb_send_error(ctdb, hdr, ret, "ltdb fetch failed in ctdb_request_call");
                return;
        }
        if (ret == -2) {
-               DEBUG(2,(__location__ " deferred ctdb_request_call\n"));
+               DEBUG(DEBUG_INFO,(__location__ " deferred ctdb_request_call\n"));
                return;
        }
 
-       /* if we are not the dmaster, then send a redirect to the
-          requesting node */
-       if (header.dmaster != ctdb->vnn) {
+       /* Dont do READONLY if we dont have a tracking database */
+       if ((c->flags & CTDB_WANT_READONLY) && !ctdb_db->readonly) {
+               c->flags &= ~CTDB_WANT_READONLY;
+       }
+
+       if (header.flags & CTDB_REC_RO_REVOKE_COMPLETE) {
+               header.flags &= ~(CTDB_REC_RO_HAVE_DELEGATIONS|CTDB_REC_RO_HAVE_READONLY|CTDB_REC_RO_REVOKING_READONLY|CTDB_REC_RO_REVOKE_COMPLETE);
+               if (ctdb_ltdb_store(ctdb_db, call->key, &header, data) != 0) {
+                       ctdb_fatal(ctdb, "Failed to write header with cleared REVOKE flag");
+               }
+               /* and clear out the tracking data */
+               if (tdb_delete(ctdb_db->rottdb, call->key) != 0) {
+                       DEBUG(DEBUG_ERR,(__location__ " Failed to clear out trackingdb record\n"));
+               }
+       }
+
+       /* if we are revoking, we must defer all other calls until the revoke
+        * had completed.
+        */
+       if (header.flags & CTDB_REC_RO_REVOKING_READONLY) {
                talloc_free(data.dptr);
-               ctdb_call_send_redirect(ctdb, call.key, c, &header);
-               ctdb_ltdb_unlock(ctdb_db, call.key);
+               ret = ctdb_ltdb_unlock(ctdb_db, call->key);
+
+               if (ctdb_add_revoke_deferred_call(ctdb, ctdb_db, call->key, hdr, ctdb_call_input_pkt, ctdb) != 0) {
+                       ctdb_fatal(ctdb, "Failed to add deferred call for revoke child");
+               }
+               talloc_free(call);
                return;
        }
 
-       if (c->hopcount > ctdb->statistics.max_hop_count) {
-               ctdb->statistics.max_hop_count = c->hopcount;
+       /* if we are not the dmaster and are not hosting any delegations,
+          then send a redirect to the requesting node */
+       if ((header.dmaster != ctdb->pnn) 
+           && (!(header.flags & CTDB_REC_RO_HAVE_DELEGATIONS)) ) {
+               talloc_free(data.dptr);
+               ctdb_call_send_redirect(ctdb, call->key, c, &header);
+
+               ret = ctdb_ltdb_unlock(ctdb_db, call->key);
+               if (ret != 0) {
+                       DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
+               }
+               return;
        }
 
-       /* if this nodes has done enough consecutive calls on the same record
-          then give them the record
-          or if the node requested an immediate migration
-       */
-       if ( c->hdr.srcnode != ctdb->vnn &&
-            ((header.laccessor == c->hdr.srcnode
-              && header.lacount >= ctdb->tunable.max_lacount)
-             || (c->flags & CTDB_IMMEDIATE_MIGRATION)) ) {
-               DEBUG(2,("vnn %u starting migration of %08x to %u\n", 
-                        ctdb->vnn, ctdb_hash(&call.key), c->hdr.srcnode));
-               ctdb_call_send_dmaster(ctdb_db, c, &header, &call.key, &data);
+       if ( (!(c->flags & CTDB_WANT_READONLY))
+       && (header.flags & (CTDB_REC_RO_HAVE_DELEGATIONS|CTDB_REC_RO_HAVE_READONLY)) ) {
+               header.flags   |= CTDB_REC_RO_REVOKING_READONLY;
+               if (ctdb_ltdb_store(ctdb_db, call->key, &header, data) != 0) {
+                       ctdb_fatal(ctdb, "Failed to store record with HAVE_DELEGATIONS set");
+               }
+               ret = ctdb_ltdb_unlock(ctdb_db, call->key);
+
+               if (ctdb_start_revoke_ro_record(ctdb, ctdb_db, call->key, &header, data) != 0) {
+                       ctdb_fatal(ctdb, "Failed to start record revoke");
+               }
                talloc_free(data.dptr);
-               ctdb_ltdb_unlock(ctdb_db, call.key);
+
+               if (ctdb_add_revoke_deferred_call(ctdb, ctdb_db, call->key, hdr, ctdb_call_input_pkt, ctdb) != 0) {
+                       ctdb_fatal(ctdb, "Failed to add deferred call for revoke child");
+               }
+               talloc_free(call);
+
+               return;
+       }               
+
+       /* If this is the first request for delegation. bump rsn and set
+        * the delegations flag
+        */
+       if ((c->flags & CTDB_WANT_READONLY)
+       &&  (c->callid == CTDB_FETCH_WITH_HEADER_FUNC)
+       &&  (!(header.flags & CTDB_REC_RO_HAVE_DELEGATIONS))) {
+               header.rsn     += 3;
+               header.flags   |= CTDB_REC_RO_HAVE_DELEGATIONS;
+               if (ctdb_ltdb_store(ctdb_db, call->key, &header, data) != 0) {
+                       ctdb_fatal(ctdb, "Failed to store record with HAVE_DELEGATIONS set");
+               }
+       }
+       if ((c->flags & CTDB_WANT_READONLY) 
+       &&  (call->call_id == CTDB_FETCH_WITH_HEADER_FUNC)) {
+               TDB_DATA tdata;
+
+               tdata = tdb_fetch(ctdb_db->rottdb, call->key);
+               if (ctdb_trackingdb_add_pnn(ctdb, &tdata, c->hdr.srcnode) != 0) {
+                       ctdb_fatal(ctdb, "Failed to add node to trackingdb");
+               }
+               if (tdb_store(ctdb_db->rottdb, call->key, tdata, TDB_REPLACE) != 0) {
+                       ctdb_fatal(ctdb, "Failed to store trackingdb data");
+               }
+               free(tdata.dptr);
+
+               ret = ctdb_ltdb_unlock(ctdb_db, call->key);
+               if (ret != 0) {
+                       DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
+               }
+
+               len = offsetof(struct ctdb_reply_call, data) + data.dsize + sizeof(struct ctdb_ltdb_header);
+               r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REPLY_CALL, len, 
+                                           struct ctdb_reply_call);
+               CTDB_NO_MEMORY_FATAL(ctdb, r);
+               r->hdr.destnode  = c->hdr.srcnode;
+               r->hdr.reqid     = c->hdr.reqid;
+               r->status        = 0;
+               r->datalen       = data.dsize + sizeof(struct ctdb_ltdb_header);
+               header.rsn      -= 2;
+               header.flags   |= CTDB_REC_RO_HAVE_READONLY;
+               header.flags   &= ~CTDB_REC_RO_HAVE_DELEGATIONS;
+               memcpy(&r->data[0], &header, sizeof(struct ctdb_ltdb_header));
+
+               if (data.dsize) {
+                       memcpy(&r->data[sizeof(struct ctdb_ltdb_header)], data.dptr, data.dsize);
+               }
+
+               ctdb_queue_packet(ctdb, &r->hdr);
+
+               talloc_free(r);
                return;
        }
 
-       ctdb_call_local(ctdb_db, &call, &header, hdr, &data, c->hdr.srcnode);
+       CTDB_UPDATE_STAT(ctdb, max_hop_count, c->hopcount);
+
+       /* Try if possible to migrate the record off to the caller node.
+        * From the clients perspective a fetch of the data is just as 
+        * expensive as a migration.
+        */
+       if (c->hdr.srcnode != ctdb->pnn) {
+               if (ctdb_db->transaction_active) {
+                       DEBUG(DEBUG_INFO, (__location__ " refusing migration"
+                             " of key %s while transaction is active\n",
+                             (char *)call->key.dptr));
+               } else {
+                       DEBUG(DEBUG_DEBUG,("pnn %u starting migration of %08x to %u\n",
+                                ctdb->pnn, ctdb_hash(&(call->key)), c->hdr.srcnode));
+                       ctdb_call_send_dmaster(ctdb_db, c, &header, &(call->key), &data);
+                       talloc_free(data.dptr);
+
+                       ret = ctdb_ltdb_unlock(ctdb_db, call->key);
+                       if (ret != 0) {
+                               DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
+                       }
+                       return;
+               }
+       }
+
+       ret = ctdb_call_local(ctdb_db, call, &header, hdr, &data, true);
+       if (ret != 0) {
+               DEBUG(DEBUG_ERR,(__location__ " ctdb_call_local failed\n"));
+               call->status = -1;
+       }
 
-       ctdb_ltdb_unlock(ctdb_db, call.key);
+       ret = ctdb_ltdb_unlock(ctdb_db, call->key);
+       if (ret != 0) {
+               DEBUG(DEBUG_ERR,(__location__ " ctdb_ltdb_unlock() failed with error %d\n", ret));
+       }
 
-       len = offsetof(struct ctdb_reply_call, data) + call.reply_data.dsize;
+       len = offsetof(struct ctdb_reply_call, data) + call->reply_data.dsize;
        r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REPLY_CALL, len, 
                                    struct ctdb_reply_call);
        CTDB_NO_MEMORY_FATAL(ctdb, r);
        r->hdr.destnode  = hdr->srcnode;
        r->hdr.reqid     = hdr->reqid;
-       r->status        = call.status;
-       r->datalen       = call.reply_data.dsize;
-       if (call.reply_data.dsize) {
-               memcpy(&r->data[0], call.reply_data.dptr, call.reply_data.dsize);
+       r->status        = call->status;
+       r->datalen       = call->reply_data.dsize;
+       if (call->reply_data.dsize) {
+               memcpy(&r->data[0], call->reply_data.dptr, call->reply_data.dsize);
        }
 
        ctdb_queue_packet(ctdb, &r->hdr);
@@ -447,19 +690,79 @@ void ctdb_reply_call(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 
        state = ctdb_reqid_find(ctdb, hdr->reqid, struct ctdb_call_state);
        if (state == NULL) {
-               DEBUG(0, (__location__ " reqid %u not found\n", hdr->reqid));
+               DEBUG(DEBUG_ERR, (__location__ " reqid %u not found\n", hdr->reqid));
                return;
        }
 
        if (hdr->reqid != state->reqid) {
                /* we found a record  but it was the wrong one */
-               DEBUG(0, ("Dropped orphaned call reply with reqid:%u\n",hdr->reqid));
+               DEBUG(DEBUG_ERR, ("Dropped orphaned call reply with reqid:%u\n",hdr->reqid));
                return;
        }
 
-       state->call.reply_data.dptr = c->data;
-       state->call.reply_data.dsize = c->datalen;
-       state->call.status = c->status;
+
+       /* read only delegation processing */
+       /* If we got a FETCH_WITH_HEADER we should check if this is a ro
+        * delegation since we may need to update the record header
+        */
+       if (state->c->callid == CTDB_FETCH_WITH_HEADER_FUNC) {
+               struct ctdb_db_context *ctdb_db = state->ctdb_db;
+               struct ctdb_ltdb_header *header = (struct ctdb_ltdb_header *)&c->data[0];
+               struct ctdb_ltdb_header oldheader;
+               TDB_DATA key, data, olddata;
+               int ret;
+
+               if (!(header->flags & CTDB_REC_RO_HAVE_READONLY)) {
+                       goto finished_ro;
+                       return;
+               }
+
+               key.dsize = state->c->keylen;
+               key.dptr  = state->c->data;
+               ret = ctdb_ltdb_lock_requeue(ctdb_db, key, hdr,
+                                    ctdb_call_input_pkt, ctdb, False);
+               if (ret == -2) {
+                       return;
+               }
+               if (ret != 0) {
+                       DEBUG(DEBUG_ERR,(__location__ " Failed to get lock in ctdb_reply_call\n"));
+                       return;
+               }
+
+               ret = ctdb_ltdb_fetch(ctdb_db, key, &oldheader, state, &olddata);
+               if (ret != 0) {
+                       DEBUG(DEBUG_ERR, ("Failed to fetch old record in ctdb_reply_call\n"));
+                       ctdb_ltdb_unlock(ctdb_db, key);
+                       goto finished_ro;
+               }                       
+
+               if (header->rsn <= oldheader.rsn) {
+                       ctdb_ltdb_unlock(ctdb_db, key);
+                       goto finished_ro;
+               }
+
+               if (c->datalen < sizeof(struct ctdb_ltdb_header)) {
+                       DEBUG(DEBUG_ERR,(__location__ " Got FETCH_WITH_HEADER reply with too little data: %d bytes\n", c->datalen));
+                       ctdb_ltdb_unlock(ctdb_db, key);
+                       goto finished_ro;
+               }
+
+               data.dsize = c->datalen - sizeof(struct ctdb_ltdb_header);
+               data.dptr  = &c->data[sizeof(struct ctdb_ltdb_header)];
+               ret = ctdb_ltdb_store(ctdb_db, key, header, data);
+               if (ret != 0) {
+                       DEBUG(DEBUG_ERR, ("Failed to store new record in ctdb_reply_call\n"));
+                       ctdb_ltdb_unlock(ctdb_db, key);
+                       goto finished_ro;
+               }                       
+
+               ctdb_ltdb_unlock(ctdb_db, key);
+       }
+finished_ro:
+
+       state->call->reply_data.dptr = c->data;
+       state->call->reply_data.dsize = c->datalen;
+       state->call->status = c->status;
 
        talloc_steal(state, c);
 
@@ -482,11 +785,13 @@ void ctdb_reply_dmaster(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
        struct ctdb_reply_dmaster *c = (struct ctdb_reply_dmaster *)hdr;
        struct ctdb_db_context *ctdb_db;
        TDB_DATA key, data;
+       uint32_t record_flags = 0;
+       size_t len;
        int ret;
 
        ctdb_db = find_ctdb_db(ctdb, c->db_id);
        if (ctdb_db == NULL) {
-               DEBUG(0,("Unknown db_id 0x%x in ctdb_reply_dmaster\n", c->db_id));
+               DEBUG(DEBUG_ERR,("Unknown db_id 0x%x in ctdb_reply_dmaster\n", c->db_id));
                return;
        }
        
@@ -494,6 +799,11 @@ void ctdb_reply_dmaster(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
        key.dsize = c->keylen;
        data.dptr = &c->data[key.dsize];
        data.dsize = c->datalen;
+       len = offsetof(struct ctdb_reply_dmaster, data) + key.dsize + data.dsize
+               + sizeof(uint32_t);
+       if (len <= c->hdr.length) {
+               record_flags = *(uint32_t *)&c->data[c->keylen + c->datalen];
+       }
 
        ret = ctdb_ltdb_lock_requeue(ctdb_db, key, hdr,
                                     ctdb_call_input_pkt, ctdb, False);
@@ -501,11 +811,11 @@ void ctdb_reply_dmaster(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
                return;
        }
        if (ret != 0) {
-               DEBUG(0,(__location__ " Failed to get lock in ctdb_reply_dmaster\n"));
+               DEBUG(DEBUG_ERR,(__location__ " Failed to get lock in ctdb_reply_dmaster\n"));
                return;
        }
 
-       ctdb_become_dmaster(ctdb_db, hdr->reqid, key, data, c->rsn);
+       ctdb_become_dmaster(ctdb_db, hdr, key, data, c->rsn, record_flags);
 }
 
 
@@ -519,14 +829,14 @@ void ctdb_reply_error(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
 
        state = ctdb_reqid_find(ctdb, hdr->reqid, struct ctdb_call_state);
        if (state == NULL) {
-               DEBUG(0,("vnn %u Invalid reqid %u in ctdb_reply_error\n",
-                        ctdb->vnn, hdr->reqid));
+               DEBUG(DEBUG_ERR,("pnn %u Invalid reqid %u in ctdb_reply_error\n",
+                        ctdb->pnn, hdr->reqid));
                return;
        }
 
        if (hdr->reqid != state->reqid) {
                /* we found a record  but it was the wrong one */
-               DEBUG(0, ("Dropped orphaned error reply with reqid:%u\n",hdr->reqid));
+               DEBUG(DEBUG_ERR, ("Dropped orphaned error reply with reqid:%u\n",hdr->reqid));
                return;
        }
 
@@ -569,10 +879,10 @@ static void ctdb_call_resend(struct ctdb_call_state *state)
        state->c->hdr.generation = state->generation;
 
        /* send the packet to ourselves, it will be redirected appropriately */
-       state->c->hdr.destnode = ctdb->vnn;
+       state->c->hdr.destnode = ctdb->pnn;
 
        ctdb_queue_packet(ctdb, &state->c->hdr);
-       DEBUG(0,("resent ctdb_call\n"));
+       DEBUG(DEBUG_NOTICE,("resent ctdb_call\n"));
 }
 
 /*
@@ -621,10 +931,12 @@ struct ctdb_call_state *ctdb_call_local_send(struct ctdb_db_context *ctdb_db,
        talloc_steal(state, data->dptr);
 
        state->state = CTDB_CALL_DONE;
-       state->call = *call;
+       state->call  = talloc(state, struct ctdb_call);
+       CTDB_NO_MEMORY_NULL(ctdb, state->call);
+       *(state->call) = *call;
        state->ctdb_db = ctdb_db;
 
-       ret = ctdb_call_local(ctdb_db, &state->call, header, state, data, ctdb->vnn);
+       ret = ctdb_call_local(ctdb_db, state->call, header, state, data, true);
 
        event_add_timed(ctdb->ev, state, timeval_zero(), call_local_trigger, state);
 
@@ -646,8 +958,16 @@ struct ctdb_call_state *ctdb_daemon_call_send_remote(struct ctdb_db_context *ctd
        struct ctdb_call_state *state;
        struct ctdb_context *ctdb = ctdb_db->ctdb;
 
+       if (ctdb->methods == NULL) {
+               DEBUG(DEBUG_INFO,(__location__ " Failed send packet. Transport is down\n"));
+               return NULL;
+       }
+
        state = talloc_zero(ctdb_db, struct ctdb_call_state);
        CTDB_NO_MEMORY_NULL(ctdb, state);
+       state->call = talloc(state, struct ctdb_call);
+       CTDB_NO_MEMORY_NULL(ctdb, state->call);
+
        state->reqid = ctdb_reqid_new(ctdb, state);
        state->ctdb_db = ctdb_db;
        talloc_set_destructor(state, ctdb_call_destructor);
@@ -669,9 +989,9 @@ struct ctdb_call_state *ctdb_daemon_call_send_remote(struct ctdb_db_context *ctd
        memcpy(&state->c->data[0], call->key.dptr, call->key.dsize);
        memcpy(&state->c->data[call->key.dsize], 
               call->call_data.dptr, call->call_data.dsize);
-       state->call                = *call;
-       state->call.call_data.dptr = &state->c->data[call->key.dsize];
-       state->call.key.dptr       = &state->c->data[0];
+       *(state->call)              = *call;
+       state->call->call_data.dptr = &state->c->data[call->key.dsize];
+       state->call->key.dptr       = &state->c->data[0];
 
        state->state  = CTDB_CALL_WAIT;
        state->generation = ctdb->vnn_map->generation;
@@ -700,16 +1020,16 @@ int ctdb_daemon_call_recv(struct ctdb_call_state *state, struct ctdb_call *call)
                return -1;
        }
 
-       if (state->call.reply_data.dsize) {
-               call->reply_data.dptr = talloc_memdup(state->ctdb_db->ctdb,
-                                                     state->call.reply_data.dptr,
-                                                     state->call.reply_data.dsize);
-               call->reply_data.dsize = state->call.reply_data.dsize;
+       if (state->call->reply_data.dsize) {
+               call->reply_data.dptr = talloc_memdup(call,
+                                                     state->call->reply_data.dptr,
+                                                     state->call->reply_data.dsize);
+               call->reply_data.dsize = state->call->reply_data.dsize;
        } else {
                call->reply_data.dptr = NULL;
                call->reply_data.dsize = 0;
        }
-       call->status = state->call.status;
+       call->status = state->call->status;
        talloc_free(state);
        return 0;
 }
@@ -722,6 +1042,11 @@ void ctdb_send_keepalive(struct ctdb_context *ctdb, uint32_t destnode)
 {
        struct ctdb_req_keepalive *r;
        
+       if (ctdb->methods == NULL) {
+               DEBUG(DEBUG_INFO,(__location__ " Failed to send keepalive. Transport is DOWN\n"));
+               return;
+       }
+
        r = ctdb_transport_allocate(ctdb, ctdb, CTDB_REQ_KEEPALIVE,
                                    sizeof(struct ctdb_req_keepalive), 
                                    struct ctdb_req_keepalive);
@@ -729,9 +1054,368 @@ void ctdb_send_keepalive(struct ctdb_context *ctdb, uint32_t destnode)
        r->hdr.destnode  = destnode;
        r->hdr.reqid     = 0;
        
-       ctdb->statistics.keepalive_packets_sent++;
+       CTDB_INCREMENT_STAT(ctdb, keepalive_packets_sent);
 
        ctdb_queue_packet(ctdb, &r->hdr);
 
        talloc_free(r);
 }
+
+
+
+struct revokechild_deferred_call {
+       struct ctdb_context *ctdb;
+       struct ctdb_req_header *hdr;
+       deferred_requeue_fn fn;
+       void *ctx;
+};
+
+struct revokechild_handle {
+       struct revokechild_handle *next, *prev;
+       struct ctdb_context *ctdb;
+       struct ctdb_db_context *ctdb_db;
+       struct fd_event *fde;
+       int status;
+       int fd[2];
+       pid_t child;
+       TDB_DATA key;
+};
+
+struct revokechild_requeue_handle {
+       struct ctdb_context *ctdb;
+       struct ctdb_req_header *hdr;
+       deferred_requeue_fn fn;
+       void *ctx;
+};
+
+static void deferred_call_requeue(struct event_context *ev, struct timed_event *te, 
+                      struct timeval t, void *private_data)
+{
+       struct revokechild_requeue_handle *requeue_handle = talloc_get_type(private_data, struct revokechild_requeue_handle);
+
+       requeue_handle->fn(requeue_handle->ctx, requeue_handle->hdr);
+       talloc_free(requeue_handle);
+}
+
+static int deferred_call_destructor(struct revokechild_deferred_call *deferred_call)
+{
+       struct ctdb_context *ctdb = deferred_call->ctdb;
+       struct revokechild_requeue_handle *requeue_handle = talloc(ctdb, struct revokechild_requeue_handle);
+       struct ctdb_req_call *c = (struct ctdb_req_call *)deferred_call->hdr;
+
+       requeue_handle->ctdb = ctdb;
+       requeue_handle->hdr  = deferred_call->hdr;
+       requeue_handle->fn   = deferred_call->fn;
+       requeue_handle->ctx  = deferred_call->ctx;
+       talloc_steal(requeue_handle, requeue_handle->hdr);
+
+       /* when revoking, any READONLY requests have 1 second grace to let read/write finish first */
+       event_add_timed(ctdb->ev, requeue_handle, timeval_current_ofs(c->flags & CTDB_WANT_READONLY ? 1 : 0, 0), deferred_call_requeue, requeue_handle);
+
+       return 0;
+}
+
+
+static int revokechild_destructor(struct revokechild_handle *rc)
+{
+       if (rc->fde != NULL) {
+               talloc_free(rc->fde);
+       }
+
+       if (rc->fd[0] != -1) {
+               close(rc->fd[0]);
+       }
+       if (rc->fd[1] != -1) {
+               close(rc->fd[1]);
+       }
+       kill(rc->child, SIGKILL);
+
+       DLIST_REMOVE(rc->ctdb_db->revokechild_active, rc);
+       return 0;
+}
+
+static void revokechild_handler(struct event_context *ev, struct fd_event *fde, 
+                            uint16_t flags, void *private_data)
+{
+       struct revokechild_handle *rc = talloc_get_type(private_data, 
+                                                    struct revokechild_handle);
+       int ret;
+       char c;
+
+       ret = read(rc->fd[0], &c, 1);
+       if (ret != 1) {
+               DEBUG(DEBUG_ERR,("Failed to read status from revokechild. errno:%d\n", errno));
+               rc->status = -1;
+               talloc_free(rc);
+               return;
+       }
+       if (c != 0) {
+               DEBUG(DEBUG_ERR,("revokechild returned failure. status:%d\n", c));
+               rc->status = -1;
+               talloc_free(rc);
+               return;
+       }
+
+       talloc_free(rc);
+}
+
+struct ctdb_revoke_state {
+       struct ctdb_db_context *ctdb_db;
+       TDB_DATA key;
+       struct ctdb_ltdb_header *header;
+       TDB_DATA data;
+       int count;
+       int status;
+       int finished;
+};
+
+static void update_record_cb(struct ctdb_client_control_state *state)
+{
+       struct ctdb_revoke_state *revoke_state;
+       int ret;
+       int32_t res;
+
+       if (state == NULL) {
+               return;
+       }
+       revoke_state = state->async.private_data;
+
+       state->async.fn = NULL;
+        ret = ctdb_control_recv(state->ctdb, state, state, NULL, &res, NULL);
+        if ((ret != 0) || (res != 0)) {
+               DEBUG(DEBUG_ERR,("Recv for revoke update record failed ret:%d res:%d\n", ret, res));
+               revoke_state->status = -1;
+       }
+
+       revoke_state->count--;
+       if (revoke_state->count <= 0) {
+               revoke_state->finished = 1;
+       }
+}
+
+static void revoke_send_cb(struct ctdb_context *ctdb, uint32_t pnn, void *private_data)
+{
+       struct ctdb_revoke_state *revoke_state = private_data;
+       struct ctdb_client_control_state *state;
+
+       state = ctdb_ctrl_updaterecord_send(ctdb, revoke_state, timeval_current_ofs(5,0), pnn, revoke_state->ctdb_db, revoke_state->key, revoke_state->header, revoke_state->data);
+       if (state == NULL) {
+               DEBUG(DEBUG_ERR,("Failure to send update record to revoke readonly delegation\n"));
+               revoke_state->status = -1;
+               return;
+       }
+       state->async.fn           = update_record_cb;
+       state->async.private_data = revoke_state;
+
+       revoke_state->count++;
+
+}
+
+static void ctdb_revoke_timeout_handler(struct event_context *ev, struct timed_event *te, 
+                             struct timeval yt, void *private_data)
+{
+       struct ctdb_revoke_state *state = private_data;
+
+       DEBUG(DEBUG_ERR,("Timed out waiting for revoke to finish\n"));
+       state->finished = 1;
+       state->status   = -1;
+}
+
+static int ctdb_revoke_all_delegations(struct ctdb_context *ctdb, struct ctdb_db_context *ctdb_db, TDB_DATA tdata, TDB_DATA key, struct ctdb_ltdb_header *header, TDB_DATA data)
+{
+       struct ctdb_revoke_state *state = talloc_zero(ctdb, struct ctdb_revoke_state);
+       int status;
+
+       state->ctdb_db = ctdb_db;
+       state->key     = key;
+       state->header  = header;
+       state->data    = data;
+       ctdb_trackingdb_traverse(ctdb, tdata, revoke_send_cb, state);
+
+       event_add_timed(ctdb->ev, state, timeval_current_ofs(5, 0), ctdb_revoke_timeout_handler, state);
+
+       while (state->finished == 0) {
+               event_loop_once(ctdb->ev);
+       }
+
+       status = state->status;
+
+       if (status == 0) {
+               struct ctdb_ltdb_header new_header;
+               TDB_DATA new_data;
+
+               if (ctdb_ltdb_lock(ctdb_db, key) != 0) {
+                       DEBUG(DEBUG_ERR,("Failed to chainlock the database in revokechild\n"));
+                       talloc_free(state);
+                       return -1;
+               }
+               if (ctdb_ltdb_fetch(ctdb_db, key, &new_header, state, &new_data) != 0) {
+                       ctdb_ltdb_unlock(ctdb_db, key);
+                       DEBUG(DEBUG_ERR,("Failed for fetch tdb record in revokechild\n"));
+                       talloc_free(state);
+                       return -1;
+               }
+               header->rsn++;
+               if (new_header.rsn > header->rsn) {
+                       ctdb_ltdb_unlock(ctdb_db, key);
+                       DEBUG(DEBUG_ERR,("RSN too high in tdb record in revokechild\n"));
+                       talloc_free(state);
+                       return -1;
+               }
+               if ( (new_header.flags & (CTDB_REC_RO_REVOKING_READONLY|CTDB_REC_RO_HAVE_DELEGATIONS)) != (CTDB_REC_RO_REVOKING_READONLY|CTDB_REC_RO_HAVE_DELEGATIONS) ) {
+                       ctdb_ltdb_unlock(ctdb_db, key);
+                       DEBUG(DEBUG_ERR,("Flags are wrong in tdb record in revokechild\n"));
+                       talloc_free(state);
+                       return -1;
+               }
+               new_header.rsn++;
+               new_header.flags |= CTDB_REC_RO_REVOKE_COMPLETE;
+               if (ctdb_ltdb_store(ctdb_db, key, &new_header, new_data) != 0) {
+                       ctdb_ltdb_unlock(ctdb_db, key);
+                       DEBUG(DEBUG_ERR,("Failed to write new record in revokechild\n"));
+                       talloc_free(state);
+                       return -1;
+               }
+               ctdb_ltdb_unlock(ctdb_db, key);
+       }
+
+       talloc_free(state);
+       return status;
+}
+
+
+int ctdb_start_revoke_ro_record(struct ctdb_context *ctdb, struct ctdb_db_context *ctdb_db, TDB_DATA key, struct ctdb_ltdb_header *header, TDB_DATA data)
+{
+       TDB_DATA tdata;
+       struct revokechild_handle *rc;
+       pid_t parent = getpid();
+       int ret;
+
+       header->flags &= ~(CTDB_REC_RO_REVOKING_READONLY|CTDB_REC_RO_HAVE_DELEGATIONS|CTDB_REC_RO_HAVE_READONLY);
+       header->rsn   -= 1;
+
+       if ((rc = talloc_zero(ctdb_db, struct revokechild_handle)) == NULL) {
+               DEBUG(DEBUG_ERR,("Failed to allocate revokechild_handle\n"));
+               return -1;
+       }
+
+       tdata = tdb_fetch(ctdb_db->rottdb, key);
+       if (tdata.dsize > 0) {
+               uint8_t *tmp;
+
+               tmp = tdata.dptr;
+               tdata.dptr = talloc_memdup(rc, tdata.dptr, tdata.dsize);
+               free(tmp);
+       }
+
+       rc->status    = 0;
+       rc->ctdb      = ctdb;
+       rc->ctdb_db   = ctdb_db;
+       rc->fd[0]     = -1;
+       rc->fd[1]     = -1;
+
+       talloc_set_destructor(rc, revokechild_destructor);
+
+       rc->key.dsize = key.dsize;
+       rc->key.dptr  = talloc_memdup(rc, key.dptr, key.dsize);
+       if (rc->key.dptr == NULL) {
+               DEBUG(DEBUG_ERR,("Failed to allocate key for revokechild_handle\n"));
+               talloc_free(rc);
+               return -1;
+       }
+
+       ret = pipe(rc->fd);
+       if (ret != 0) {
+               DEBUG(DEBUG_ERR,("Failed to allocate key for revokechild_handle\n"));
+               talloc_free(rc);
+               return -1;
+       }
+
+
+       rc->child = ctdb_fork(ctdb);
+       if (rc->child == (pid_t)-1) {
+               DEBUG(DEBUG_ERR,("Failed to fork child for revokechild\n"));
+               talloc_free(rc);
+               return -1;
+       }
+
+       if (rc->child == 0) {
+               char c = 0;
+               close(rc->fd[0]);
+               debug_extra = talloc_asprintf(NULL, "revokechild-%s:", ctdb_db->db_name);
+
+               if (switch_from_server_to_client(ctdb, "revokechild-%s", ctdb_db->db_name) != 0) {
+                       DEBUG(DEBUG_ERR,("Failed to switch from server to client for revokechild process\n"));
+                       c = 1;
+                       goto child_finished;
+               }
+
+               c = ctdb_revoke_all_delegations(ctdb, ctdb_db, tdata, key, header, data);
+
+child_finished:
+               write(rc->fd[1], &c, 1);
+               /* make sure we die when our parent dies */
+               while (kill(parent, 0) == 0 || errno != ESRCH) {
+                       sleep(5);
+               }
+               _exit(0);
+       }
+
+       close(rc->fd[1]);
+       rc->fd[1] = -1;
+       set_close_on_exec(rc->fd[0]);
+
+       /* This is an active revokechild child process */
+       DLIST_ADD_END(ctdb_db->revokechild_active, rc, NULL);
+
+       rc->fde = event_add_fd(ctdb->ev, rc, rc->fd[0],
+                                  EVENT_FD_READ, revokechild_handler,
+                                  (void *)rc);
+       if (rc->fde == NULL) {
+               DEBUG(DEBUG_ERR,("Failed to set up fd event for revokechild process\n"));
+               talloc_free(rc);
+       }
+       tevent_fd_set_auto_close(rc->fde);
+
+       return 0;
+}
+
+int ctdb_add_revoke_deferred_call(struct ctdb_context *ctdb, struct ctdb_db_context *ctdb_db, TDB_DATA key, struct ctdb_req_header *hdr, deferred_requeue_fn fn, void *call_context)
+{
+       struct revokechild_handle *rc;
+       struct revokechild_deferred_call *deferred_call;
+
+       for (rc = ctdb_db->revokechild_active; rc; rc = rc->next) {
+               if (rc->key.dsize == 0) {
+                       continue;
+               }
+               if (rc->key.dsize != key.dsize) {
+                       continue;
+               }
+               if (!memcmp(rc->key.dptr, key.dptr, key.dsize)) {
+                       break;
+               }
+       }
+
+       if (rc == NULL) {
+               DEBUG(DEBUG_ERR,("Failed to add deferred call to revoke list. revoke structure not found\n"));
+               return -1;
+       }
+
+       deferred_call = talloc(rc, struct revokechild_deferred_call);
+       if (deferred_call == NULL) {
+               DEBUG(DEBUG_ERR,("Failed to allocate deferred call structure for revoking record\n"));
+               return -1;
+       }
+
+       deferred_call->ctdb = ctdb;
+       deferred_call->hdr  = hdr;
+       deferred_call->fn   = fn;
+       deferred_call->ctx  = call_context;
+
+       talloc_set_destructor(deferred_call, deferred_call_destructor);
+       talloc_steal(deferred_call, hdr);
+
+       return 0;
+}