389-ds-base/ldap/servers/plugins/replication/repl5_inc_protocol.c

/** BEGIN COPYRIGHT BLOCK
 * Copyright (C) 2001 Sun Microsystems, Inc. Used by permission.
 * Copyright (C) 2005 Red Hat, Inc.
 * All rights reserved.
 *
 * License: GPL (version 3 or any later version).
 * See LICENSE for details. 
 * END COPYRIGHT BLOCK **/

#ifdef HAVE_CONFIG_H
#  include <config.h>
#endif


/* repl5_inc_protocol.c */
/*

 The Prot_Incremental object implements the DS 5.0 multi-master incremental
 replication protocol. 


Stuff to do:

- Need to figure out how asynchronous events end up in here. They are:
  - entry updated in replicated area.
  - backoff timeout
  - enter/leave.

Perhaps these events should be properties of the main protocol.
*/
#include <plstr.h>

#include "repl.h"
#include "repl5.h"
#include "repl5_ruv.h"
#include "repl5_prot_private.h"
#include "cl5_api.h"

extern int slapi_log_urp;

/*** from proto-slap.h ***/
void ava_done(struct ava *ava);

typedef struct repl5_inc_private
{
	char *ruv;	/* RUV on remote replica (use diff type for this? - ggood */
	Backoff_Timer *backoff;
	Repl_Protocol *rp;
	PRLock *lock;
	PRUint32 eventbits;
} repl5_inc_private;

/* Structures used to communicate with the result reading thread */

#ifndef UIDSTR_SIZE
#define UIDSTR_SIZE 35 /* size of the string representation of the id */
#endif

typedef struct repl5_inc_operation 
{
	int ldap_message_id;
	unsigned long operation_type;
	char csn_str[CSN_STRSIZE];
	char uniqueid[UIDSTR_SIZE+1];
	ReplicaId  replica_id;
	struct repl5_inc_operation *next;
} repl5_inc_operation;

typedef struct result_data
{
    Private_Repl_Protocol *prp;
    int rc;    
	PRLock *lock; /* Lock to protect access to this structure, the message id list and to force memory barriers */
	PRThread *result_tid; /* The async result thread */
	repl5_inc_operation *operation_list_head; /* List of IDs for outstanding operations */
	repl5_inc_operation *operation_list_tail; /* List of IDs for outstanding operations */
	int abort; /* Flag used to tell the sending thread asyncronously that it should abort (because an error came up in a result) */
	PRUint32 num_changes_sent;
	int stop_result_thread; /* Flag used to tell the result thread to exit */
	int last_message_id_sent;
	int last_message_id_received;
	int flowcontrol_detection;
	int result; /* The UPDATE_TRANSIENT_ERROR etc */
	int WaitForAsyncResults;
} result_data;

/* Various states the incremental protocol can pass through */
#define	STATE_START 0                   /* ONREPL - should we rename this - we don't use it just to start up? */
#define STATE_WAIT_WINDOW_OPEN 1
#define STATE_WAIT_CHANGES 2
#define STATE_READY_TO_ACQUIRE 3
#define STATE_BACKOFF_START 4           /* ONREPL - can we combine BACKOFF_START and BACKOFF states? */
#define STATE_BACKOFF 5
#define STATE_SENDING_UPDATES 6
#define STATE_STOP_FATAL_ERROR 7
#define STATE_STOP_FATAL_ERROR_PART2 8
#define STATE_STOP_NORMAL_TERMINATION 9

/* Events (synchronous and asynchronous; these are bits) */
#define EVENT_WINDOW_OPENED 1
#define EVENT_WINDOW_CLOSED 2
#define EVENT_TRIGGERING_CRITERIA_MET 4             /* ONREPL - should we rename this to EVENT_CHANGE_AVAILABLE */
#define EVENT_BACKOFF_EXPIRED 8
#define EVENT_REPLICATE_NOW 16
#define EVENT_PROTOCOL_SHUTDOWN 32
#define EVENT_AGMT_CHANGED 64

#define UPDATE_NO_MORE_UPDATES 201
#define UPDATE_TRANSIENT_ERROR 202
#define UPDATE_FATAL_ERROR 203
#define UPDATE_SCHEDULE_WINDOW_CLOSED 204
#define UPDATE_CONNECTION_LOST 205
#define UPDATE_TIMEOUT 206
#define UPDATE_YIELD 207

/* Return codes from examine_update_vector */
#define EXAMINE_RUV_PRISTINE_REPLICA 401
#define EXAMINE_RUV_GENERATION_MISMATCH 402
#define EXAMINE_RUV_REPLICA_TOO_OLD 403
#define EXAMINE_RUV_OK 404
#define EXAMINE_RUV_PARAM_ERROR 405

#define MAX_CHANGES_PER_SESSION	10000
/*
 * Maximum time to wait between replication sessions. If we
 * don't see any updates for a period equal to this interval,
 * we go ahead and start a replication session, just to be safe
 */
#define MAX_WAIT_BETWEEN_SESSIONS PR_SecondsToInterval(60 * 5) /* 5 minutes */

/*
 * tests if the protocol has been shutdown and we need to quit
 * event_occurred resets the bits in the bit flag, so whoever tests for shutdown
 * resets the flags, so the next one who tests for shutdown won't get it, so we
 * also look at the terminate flag
 */
#define PROTOCOL_IS_SHUTDOWN(prp) (event_occurred(prp, EVENT_PROTOCOL_SHUTDOWN) || prp->terminate)

/* mods should be LDAPMod **mods */
#define MODS_ARE_EMPTY(mods) ((mods == NULL) || (mods[0] == NULL))

/* Forward declarations */
static PRUint32 event_occurred(Private_Repl_Protocol *prp, PRUint32 event);
static void reset_events (Private_Repl_Protocol *prp);
static void protocol_sleep(Private_Repl_Protocol *prp, PRIntervalTime duration);
static int send_updates(Private_Repl_Protocol *prp, RUV *ruv, PRUint32 *num_changes_sent);
static void repl5_inc_backoff_expired(time_t timer_fire_time, void *arg);
static int examine_update_vector(Private_Repl_Protocol *prp, RUV *ruv);
static const char* state2name (int state);
static const char* event2name (int event);
static const char* op2string (int op);
static int repl5_inc_update_from_op_result(Private_Repl_Protocol *prp, ConnResult replay_crc, int connection_error, char *csn_str, char *uniqueid, ReplicaId replica_id, int* finished, PRUint32 *num_changes_sent);

/* Push a newly sent operation onto the tail of the list */
static void repl5_int_push_operation(result_data *rd, repl5_inc_operation *it)
{
	repl5_inc_operation *tail = NULL;
	PR_Lock(rd->lock);
	tail = rd->operation_list_tail;
	if (tail) 
	{
		tail->next = it;
	}
	if (NULL == rd->operation_list_head) 
	{
		rd->operation_list_head = it;
	}
	rd->operation_list_tail = it;
	PR_Unlock(rd->lock);
}

/* Pop the next operation in line to respond from the list */
/* The caller is expected to free the operation item */
static repl5_inc_operation *repl5_inc_pop_operation(result_data *rd)
{
	repl5_inc_operation *head = NULL;
	repl5_inc_operation *ret = NULL;
	PR_Lock(rd->lock);
	head = rd->operation_list_head;
	if (head) 
	{
		ret = head;
		rd->operation_list_head = head->next;
		if (rd->operation_list_tail == head)
		{
			rd->operation_list_tail = NULL;
		}
	}
	PR_Unlock(rd->lock);
	return ret;
}

static void
repl5_inc_op_free(repl5_inc_operation *op)
{
	slapi_ch_free((void**)&op);
}

static repl5_inc_operation *repl5_inc_operation_new()
{
	repl5_inc_operation *ret = NULL;
	ret = (repl5_inc_operation *) slapi_ch_calloc(1,sizeof(repl5_inc_operation));
	return ret;
}

/* Called when in compatibility mode, to get the next result from the wire 
 * The operation thread will not send a second operation until it has read the
 * previous result. */
static int
repl5_inc_get_next_result(result_data *rd)
{
	ConnResult conres = 0;
	int message_id = 0;
	/* Wait on the next result */
	conres = conn_read_result(rd->prp->conn, &message_id);
	/* Return it to the caller */
	return conres;
}

#if NEEDED_FOR_DEBUGGING
static void 
repl5_inc_log_operation_failure(int operation_code, int ldap_error, char* ldap_error_string, const char *agreement_name)
{
	char *op_string = slapi_op_type_to_string(operation_code);

    slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
		"%s: Received error %d: %s for %s operation\n",
		agreement_name,
		ldap_error, ldap_error_string ? ldap_error_string : "NULL",
		op_string ? op_string : "NULL");
}
#endif

/* Thread that collects results from async operations sent to the consumer */
static void
repl5_inc_result_threadmain(void *param)
{
	result_data *rd = (result_data*) param;
	ConnResult conres = 0;
	Repl_Connection *conn = rd->prp->conn;
	int finished = 0;
	int message_id = 0;

	slapi_log_error(SLAPI_LOG_REPL, NULL, "repl5_inc_result_threadmain starting\n");
	while (!finished) 
	{
		repl5_inc_operation *op = NULL;
		int connection_error = 0;
		char *csn_str = NULL; 
		char *uniqueid = NULL;
		ReplicaId replica_id = 0;
		int operation_code = 0;
		char *ldap_error_string = NULL;
		time_t time_now = 0;
		time_t start_time = time( NULL );
		int backoff_time = 1;

		/* Read the next result */
		/* We call the get result function with a short timeout (non-blocking) 
		 * this is so we don't block here forever, and can stop this thread when
		 * the time comes. However, we do need to implement blocking with timeout
		 * semantics here instead.
		 */

		while (!finished)
		{
			conres = conn_read_result_ex(conn, NULL, NULL, NULL, LDAP_RES_ANY, &message_id, 0);
			slapi_log_error(SLAPI_LOG_REPL, NULL, "repl5_inc_result_threadmain: read result for message_id %d\n", message_id);
			/* Timeout here means that we didn't block, not a real timeout */
			if (CONN_TIMEOUT == conres)
			{
				/* We need to a) check that the 'real' timeout hasn't expired and
				 * b) implement a backoff sleep to avoid spinning */
				/* Did the connection's timeout expire ? */
				time_now = time( NULL );
				if (conn_get_timeout(conn) <= ( time_now - start_time ))
				{
					/* We timed out */
					conres = CONN_TIMEOUT;
					break;
				}
				/* Otherwise we backoff */
				DS_Sleep(PR_MillisecondsToInterval(backoff_time));
				if (backoff_time < 1000) 
				{
					backoff_time <<= 1;
				}
				/* Should we stop ? */
				PR_Lock(rd->lock);
				if (rd->stop_result_thread) 
				{
					finished = 1;
				}
				PR_Unlock(rd->lock);
			} else
			{
				/* Something other than a timeout, so we exit the loop */
				break;
			}
		}
		if (conres != CONN_TIMEOUT)
		{
			int return_value;
			int should_finish = 0;
			if (message_id) 
			{
				rd->last_message_id_received = message_id;
			}
			/* Handle any error etc */

			/* Get the stored operation details from the queue, unless we timed out... */
			op = repl5_inc_pop_operation(rd);
			if (op) 
			{
				csn_str = op->csn_str;
				replica_id = op->replica_id;
				uniqueid = op->uniqueid;
			}

			conn_get_error_ex(conn, &operation_code, &connection_error, &ldap_error_string);
			slapi_log_error(SLAPI_LOG_REPL, NULL, "repl5_inc_result_threadmain: result %d, %d, %d, %d, %s\n", operation_code,connection_error,conres,message_id,ldap_error_string);
			return_value = repl5_inc_update_from_op_result(rd->prp, conres, connection_error, csn_str, uniqueid, replica_id, &should_finish, &(rd->num_changes_sent));
			if (return_value || should_finish)
			{
				slapi_log_error(SLAPI_LOG_REPL, NULL, "repl5_inc_result_threadmain: got op result %d should finish %d\n", return_value, should_finish);
				/* If so then we need to take steps to abort the update process */
				PR_Lock(rd->lock);
				rd->result = return_value;
				rd->abort = 1;
				PR_Unlock(rd->lock);
				/* We also need to log the error, including details stored from when the operation was sent */
				/* we cannot finish yet - we still need to waitfor the pending results, then
				   the main repl code will shut down this thread */
				/* we can finish if we have disconnected - in that case, there will be nothing
				   to read */
				if (return_value == UPDATE_CONNECTION_LOST) {
					finished = 1;
				}
			} else {
				/* old semantics had result set outside of lock */
				rd->result = return_value;
			}
		}
		/* Should we stop ? */
		PR_Lock(rd->lock);
		if (rd->stop_result_thread) 
		{
			finished = 1;
		}
		PR_Unlock(rd->lock);
		if (op) 
		{
			repl5_inc_op_free(op);
		}
	}
	slapi_log_error(SLAPI_LOG_REPL, NULL, "repl5_inc_result_threadmain exiting\n");
}

static result_data *
repl5_inc_rd_new(Private_Repl_Protocol *prp)
{
	result_data *res = NULL;
	res = (result_data *) slapi_ch_calloc(1,sizeof(result_data));
	if (res) {
		res->prp = prp;
		res->lock = PR_NewLock();
		if (NULL == res->lock) {
			slapi_ch_free((void **)&res);
			res = NULL;
		}
	}
	return res;
}

static void 
repl5_inc_rd_list_destroy(repl5_inc_operation *op)
{
	while (op) {
		repl5_inc_operation *next = op->next;
		repl5_inc_op_free(op);
		op = next;
	}
}

static void 
repl5_inc_rd_destroy(result_data **pres)
{
	result_data *res = *pres;
	if (res->lock) {
		PR_DestroyLock(res->lock);
	}
	/* Delete the linked list if we have one */
	/* Begin at the head */
	repl5_inc_rd_list_destroy(res->operation_list_head);
	slapi_ch_free((void **)pres);
}

static int 
repl5_inc_create_async_result_thread(result_data *rd)
{
	int retval = 0;
	PRThread *tid = NULL;
	/* Create a thread that reads results from the connection and stores status in the callback_data structure */
	tid = PR_CreateThread(PR_USER_THREAD, 
				repl5_inc_result_threadmain, (void*)rd,
				PR_PRIORITY_NORMAL, PR_GLOBAL_THREAD, PR_JOINABLE_THREAD, 
				SLAPD_DEFAULT_THREAD_STACKSIZE);
	if (NULL == tid) 
	{
		slapi_log_error(SLAPI_LOG_FATAL, NULL,
					"repl5_tot_create_async_result_thread failed. "
					SLAPI_COMPONENT_NAME_NSPR " error %d (%s)\n",
					PR_GetError(), slapd_pr_strerror( PR_GetError() ));
		retval = -1;
	} else {
		rd->result_tid = tid;
	}
	return retval; 
}

static int 
repl5_inc_destroy_async_result_thread(result_data *rd)
{
	int retval = 0;
	PRThread *tid = rd->result_tid;
	if (tid) {
		PR_Lock(rd->lock);
		rd->stop_result_thread = 1;
		PR_Unlock(rd->lock);
		(void)PR_JoinThread(tid);
	}
	return retval;
}

/* The interest of this routine is to give time to the consumer
 * to apply the sent updates and return the acks.
 * So the caller should not hold the replication connection lock
 * to let the RA.reader receives the acks.
 */
static void
repl5_inc_flow_control_results(Repl_Agmt *agmt, result_data *rd)
{
    PR_Lock(rd->lock);
    if ((rd->last_message_id_received <= rd->last_message_id_sent) &&
        ((rd->last_message_id_sent - rd->last_message_id_received) >= agmt_get_flowcontrolwindow(agmt))) {
        rd->flowcontrol_detection++;
        DS_Sleep(PR_MillisecondsToInterval(agmt_get_flowcontrolpause(agmt)));
    }
    PR_Unlock(rd->lock);
}

static int
repl5_inc_waitfor_async_results(result_data *rd)
{
	int done = 0;
	int loops = 0;
	int rc = UPDATE_NO_MORE_UPDATES;

	/* Keep pulling results off the LDAP connection until we catch up to the last message id stored in the rd */
	while (!done && !slapi_is_shutting_down())
	{
		/* Lock the structure to force memory barrier */
		PR_Lock(rd->lock);
		/* Are we caught up ? */
		slapi_log_error(SLAPI_LOG_REPL, NULL,
					"repl5_inc_waitfor_async_results: %d %d\n",
					rd->last_message_id_received, rd->last_message_id_sent);
		if (rd->last_message_id_received >= rd->last_message_id_sent) {
			/* If so then we're done */
			done = 1;
		} else if (rd->abort && (rd->result == UPDATE_CONNECTION_LOST)) {
			done = 1; /* no connection == no more results */
		}
		/*
		 * Return the last operation result
		 */
		rc = rd->result;
		PR_Unlock(rd->lock);
		if (!done) {
			/* If not then sleep a bit */
			DS_Sleep(PR_MillisecondsToInterval(rd->WaitForAsyncResults));
		}
		loops++;
		/* If we sleep forever then we can conclude that something bad happened, and bail... */
		/* Arbitrary 30 second delay : basically we should only expect to wait as long as it takes to process a few operations, which should be on the order of a second at most */
		if (!done && (loops > 300))
		{
			/* Log a warning */
			slapi_log_error(SLAPI_LOG_FATAL, NULL,
					"repl5_inc_waitfor_async_results timed out waiting for responses: %d %d\n",
					rd->last_message_id_received, rd->last_message_id_sent);
			done = 1;
		}
	}
	return rc;
}

/*
 * It's specifically ok to delete a protocol instance that
 * is currently running. The instance will be shut down, and
 * then resources will be freed. Since a graceful shutdown is
 * attempted, this function may take some time to complete.
 */
static void
repl5_inc_delete(Private_Repl_Protocol **prpp)
{
	repl5_inc_private *prp_priv = (repl5_inc_private *)(*prpp)->private;
	/* if backoff is set, delete it (from EQ, as well) */ 
	if (prp_priv->backoff) { 
		backoff_delete(&prp_priv->backoff); 
	} 
	/* First, stop the protocol if it isn't already stopped */
	if (!(*prpp)->stopped) {
		(*prpp)->stopped = 1;
		(*prpp)->stop(*prpp);
	}
	/* Then, delete all resources used by the protocol */
	if ((*prpp)->lock) {
		PR_DestroyLock((*prpp)->lock);
		(*prpp)->lock = NULL;
	}
	if ((*prpp)->cvar) {
		PR_DestroyCondVar((*prpp)->cvar);
		(*prpp)->cvar = NULL;
	}
	slapi_ch_free((void **)&(*prpp)->private);
	slapi_ch_free((void **)prpp);
}

/* helper function */
void
set_pause_and_busy_time(Private_Repl_Protocol *prp, long *pausetime, long *busywaittime)
{
  /* If neither are set, set busy time to its default */
  if (!*pausetime && !*busywaittime)
    {
      *busywaittime = repl5_get_backoff_min(prp);
    }
  /* pause time must be at least 1 more than the busy backoff time */
  if (*pausetime && !*busywaittime)
    {
      /*
       * user specified a pause time but no busy wait time - must
       * set busy wait time to 1 less than pause time - if pause
       * time is 1, we must set it to 2
       */
      if (*pausetime < 2)
        {
	  *pausetime = 2;
	}
      *busywaittime = *pausetime - 1;
    }
  else if (!*pausetime && *busywaittime)
    {
      /*
       * user specified a busy wait time but no pause time - must
       * set pause time to 1 more than busy wait time
       */
      *pausetime = *busywaittime + 1;
    }
  else if (*pausetime && *busywaittime && *pausetime <= *busywaittime)
    {
      /*
       * user specified both pause and busy wait times, but the pause
       * time was <= busy wait time - pause time must be at least
       * 1 more than the busy wait time
       */
      *pausetime = *busywaittime + 1;
    }
}

/*
 * Do the incremental protocol.
 * 
 * What's going on here? This thing is a state machine. It has the
 * following states:
 * 
 * State transition table:
 * 
 * Curr State       Condition/Event                        Next State
 * ----------       ------------                           -----------
 * START            schedule window is open                ACQUIRE_REPLICA
 *                  schedule window is closed              WAIT_WINDOW_OPEN
 * WAIT_WINDOW_OPEN schedule change                        START
 *                  replicate now                          ACQUIRE_REPLICA
 *                  schedule window opens                  ACQUIRE_REPLICA
 * ACQUIRE_REPLICA  acquired replica                        SEND_CHANGES                                               
 *                  failed to acquire - transient error    START_BACKOFF
 *                  failed to acquire - fatal error        STOP_FATAL_ERROR
 * SEND_CHANGES     can't update                           CONSUMER_NEEDS_REINIT
 *                  no changes to send                     WAIT_CHANGES
 *                  can't send - thransient error          START_BACKOF
 *                  can't send - window closed             WAIT_WINDOW_OPEN
 *                  can'r send - fatal error               STOP_FATAL_ERROR
 * START_BACKOF     replicate now                          ACQUIRE_REPLICA
 *                  schedule changes                       START
 *                  schedule window closes                 WAIT_WINDOW_OPEN
 *                  backoff expires & can acquire          SEND_CHANGES
 *                  backoff expires & can't acquire-trans  BACKOFF              
 *                  backoff expires & can't acquire-fatal  STOP_FATAL_ERROR
 * BACKOF           replicate now                          ACQUIRE_REPLICA
 *                  schedule changes                       START
 *                  schedule window closes                 WAIT_WINDOW_OPEN
 *                  backoff expires & can acquire          SEND_CHANGES
 *                  backoff expires & can't acquire-trans  BACKOFF              
 *                  backoff expires & can't acquire-fatal  STOP_FATAL_ERROR
 * WAIT_CHANGES     schedule window closes                 WAIT_WINDOW_OPEN
 *                  replicate_now                          ACQUIRE_REPLICA
 *                  change available                       ACQUIRE_REPLICA
 *                  schedule_change                        START
 */

/*
 * Main state machine for the incremental protocol. This routine will,
 * under normal circumstances, not return until the protocol is shut
 * down.
 */
static void
repl5_inc_run(Private_Repl_Protocol *prp)
{
  repl5_inc_private *prp_priv = (repl5_inc_private *)prp->private;
  Replica *replica = NULL;
  CSN *cons_schema_csn;
  RUV *ruv = NULL;
  PRUint32 num_changes_sent;
  /* use a different backoff timer strategy for ACQUIRE_REPLICA_BUSY errors */
  PRBool use_busy_backoff_timer = PR_FALSE;
  time_t next_fire_time;
  time_t now;
  long busywaittime = 0;
  long pausetime = 0;
  long loops = 0;
  int wait_change_timer_set = 0;
  int current_state = STATE_START;
  int next_state = STATE_START;
  int optype, ldaprc;
  int done;
  int e1;

  prp->stopped = 0;
  prp->terminate = 0;

  /* establish_protocol_callbacks(prp); */
  done = 0;
  do {
      int rc;

      /* Take action, based on current state, and compute new state. */
      switch (current_state)
      {
          case STATE_START:
              dev_debug("repl5_inc_run(STATE_START)");
              if (PROTOCOL_IS_SHUTDOWN(prp)){
                  done = 1;
                  break;
              }
              /*
               * Our initial state. See if we're in a schedule window. If
               * so, then we're ready to acquire the replica and see if it
               * needs any updates from us. If not, then wait for the window
               * to open.
               */
              if (agmt_schedule_in_window_now(prp->agmt)){
                  next_state = STATE_READY_TO_ACQUIRE;
              } else {
                  next_state = STATE_WAIT_WINDOW_OPEN;
              }
              /*  we can get here from other states because some events happened and were
               *  not cleared. For instance when we wake up in STATE_WAIT_CHANGES state.
               *  Since this is a fresh start state, we should clear all events */
              /*  ONREPL - this does not feel right - we should take another look
               *  at this state machine */
              reset_events (prp);
              /* Cancel any linger timer that might be in effect... */
              conn_cancel_linger(prp->conn);
              /* ... and disconnect, if currently connected */
              conn_disconnect(prp->conn);
              /* get the new pause time, if any */
              pausetime = agmt_get_pausetime(prp->agmt);
              /* get the new busy wait time, if any */
              busywaittime = agmt_get_busywaittime(prp->agmt);
              if (pausetime || busywaittime){
                  /* helper function to make sure they are set correctly */
                  set_pause_and_busy_time(prp, &pausetime, &busywaittime);
              }
              break;

          case STATE_WAIT_WINDOW_OPEN:
              /*
               * We're waiting for a schedule window to open. If one did,
               * or we receive a "replicate now" event, then start a protocol
               * session immediately. If the replication schedule changed, go
               * back to start.  Otherwise, go back to sleep.
               */
              dev_debug("repl5_inc_run(STATE_WAIT_WINDOW_OPEN)");
              if (PROTOCOL_IS_SHUTDOWN(prp)){
                  done = 1;
                  break;
              } else if (event_occurred(prp, EVENT_WINDOW_OPENED)){
                  next_state = STATE_READY_TO_ACQUIRE;
              } else if (event_occurred(prp, EVENT_REPLICATE_NOW)){
                  next_state = STATE_READY_TO_ACQUIRE;
              } else if (event_occurred(prp, EVENT_AGMT_CHANGED)){
                  next_state = STATE_START;
                  conn_set_agmt_changed(prp->conn);
              } else if (event_occurred(prp, EVENT_TRIGGERING_CRITERIA_MET)){ /* change available */
                  /* just ignore it and go to sleep */
                  protocol_sleep(prp, PR_INTERVAL_NO_TIMEOUT);
              } else if ((e1 = event_occurred(prp, EVENT_WINDOW_CLOSED)) ||
		                  event_occurred(prp, EVENT_BACKOFF_EXPIRED)){
                  /* this events - should not occur - log a warning and go to sleep */
                  slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
                      "%s: Incremental protocol: "
                      "event %s should not occur in state %s; going to sleep\n",
                  agmt_get_long_name(prp->agmt), e1 ? event2name(EVENT_WINDOW_CLOSED) :
                                     event2name(EVENT_BACKOFF_EXPIRED), state2name(current_state));
                  protocol_sleep(prp, PR_INTERVAL_NO_TIMEOUT);
              } else {
                  /* wait until window opens or an event occurs */
            	  slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name,
                      "%s: Incremental protocol: "
                      "waiting for update window to open\n", agmt_get_long_name(prp->agmt));
                  protocol_sleep(prp, PR_INTERVAL_NO_TIMEOUT);
              }
              break;

      case STATE_WAIT_CHANGES:
          /*
           * We're in a replication window, but we're waiting for more
           * changes to accumulate before we actually hook up and send
           * them.
           */
          dev_debug("repl5_inc_run(STATE_WAIT_CHANGES)");
          if (PROTOCOL_IS_SHUTDOWN(prp)){
              dev_debug("repl5_inc_run(STATE_WAIT_CHANGES): PROTOCOL_IS_SHUTING_DOWN -> end repl5_inc_run\n");
              done = 1;
              break;
          } else if (event_occurred(prp, EVENT_REPLICATE_NOW)){
              dev_debug("repl5_inc_run(STATE_WAIT_CHANGES): EVENT_REPLICATE_NOW received -> STATE_READY_TO_ACQUIRE\n");
              next_state = STATE_READY_TO_ACQUIRE;
              wait_change_timer_set = 0;
          } else if (event_occurred(prp, EVENT_AGMT_CHANGED)){
        	  dev_debug("repl5_inc_run(STATE_WAIT_CHANGES): EVENT_AGMT_CHANGED received -> STATE_START\n");
        	  next_state = STATE_START;
        	  conn_set_agmt_changed(prp->conn);
        	  wait_change_timer_set = 0;
          } else if (event_occurred(prp, EVENT_WINDOW_CLOSED)){
        	  dev_debug("repl5_inc_run(STATE_WAIT_CHANGES): EVENT_WINDOW_CLOSED received -> STATE_WAIT_WINDOW_OPEN\n");
        	  next_state = STATE_WAIT_WINDOW_OPEN;
        	  wait_change_timer_set = 0;
          } else if (event_occurred(prp, EVENT_TRIGGERING_CRITERIA_MET)){
              dev_debug("repl5_inc_run(STATE_WAIT_CHANGES): EVENT_TRIGGERING_CRITERIA_MET received -> STATE_READY_TO_ACQUIRE\n");
              next_state = STATE_READY_TO_ACQUIRE;
              wait_change_timer_set = 0;
          } else if ((e1 = event_occurred(prp, EVENT_WINDOW_OPENED)) || event_occurred(prp, EVENT_BACKOFF_EXPIRED)){
              /* this events - should not occur - log a warning and clear the event */
              slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name, "%s: Incremental protocol: "
                  "event %s should not occur in state %s\n",agmt_get_long_name(prp->agmt),
                  e1 ? event2name(EVENT_WINDOW_OPENED) : event2name(EVENT_BACKOFF_EXPIRED),
                  state2name(current_state));
              wait_change_timer_set = 0;
          } else {
              if (wait_change_timer_set){
                  /* We are here because our timer expired */
                  dev_debug("repl5_inc_run(STATE_WAIT_CHANGES): wait_change_timer_set expired -> STATE_START\n");
                  next_state = STATE_START;
                  wait_change_timer_set = 0;
              } else {
                  /*
                   * We are here because the last replication session
                   * finished or aborted.
                   */
                  wait_change_timer_set = 1;
                  protocol_sleep(prp, MAX_WAIT_BETWEEN_SESSIONS);
              }
          }
          break;

      case STATE_READY_TO_ACQUIRE:
          dev_debug("repl5_inc_run(STATE_READY_TO_ACQUIRE)");
          if (PROTOCOL_IS_SHUTDOWN(prp)){
              done = 1;
              break;
          }

          /* ONREPL - at this state we unconditionally acquire the replica
             ignoring all events. Not sure if this is good */
          object_acquire(prp->replica_object);
          rc = acquire_replica(prp, REPL_NSDS50_INCREMENTAL_PROTOCOL_OID, &ruv);
          use_busy_backoff_timer = PR_FALSE; /* default */
          if (rc == ACQUIRE_SUCCESS){
              next_state = STATE_SENDING_UPDATES;
          } else if (rc == ACQUIRE_REPLICA_BUSY){
              next_state = STATE_BACKOFF_START;
              use_busy_backoff_timer = PR_TRUE;
          } else if (rc == ACQUIRE_CONSUMER_WAS_UPTODATE){
              next_state = STATE_WAIT_CHANGES;
          } else if (rc == ACQUIRE_TRANSIENT_ERROR){
              next_state = STATE_BACKOFF_START;
          } else if (rc == ACQUIRE_FATAL_ERROR){
              next_state = STATE_STOP_FATAL_ERROR;
          }

          if (rc != ACQUIRE_SUCCESS){
              int optype, ldaprc;
              conn_get_error(prp->conn, &optype, &ldaprc);
              agmt_set_last_update_status(prp->agmt, ldaprc,
                  prp->last_acquire_response_code, "Unable to acquire replica");
          }

          object_release(prp->replica_object);
          break;

      case STATE_BACKOFF_START:
          dev_debug("repl5_inc_run(STATE_BACKOFF_START)");
          if (PROTOCOL_IS_SHUTDOWN(prp)){
	          done = 1;
              break;
          }
          if (event_occurred(prp, EVENT_REPLICATE_NOW)){
              next_state = STATE_READY_TO_ACQUIRE;
          } else if (event_occurred(prp, EVENT_AGMT_CHANGED)){
              next_state = STATE_START;
              conn_set_agmt_changed(prp->conn);
          } else if (event_occurred (prp, EVENT_WINDOW_CLOSED)){
              next_state = STATE_WAIT_WINDOW_OPEN;
          } else if (event_occurred (prp, EVENT_TRIGGERING_CRITERIA_MET)){
              /* consume and ignore */
          } else if ((e1 = event_occurred (prp, EVENT_WINDOW_OPENED)) ||
                      event_occurred (prp, EVENT_BACKOFF_EXPIRED)){
              /* This should never happen */
              /* this events - should not occur - log a warning and go to sleep */
              slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
                  "%s: Incremental protocol: event %s should not occur in state %s\n",
                  agmt_get_long_name(prp->agmt),
                  e1 ? event2name(EVENT_WINDOW_OPENED) : event2name(EVENT_BACKOFF_EXPIRED),
                  state2name(current_state));
          } else {
              /* Set up the backoff timer to wake us up at the appropriate time */
              /* if previous backoff set up, delete it. */
              if (prp_priv->backoff) {
                  backoff_delete(&prp_priv->backoff);
              }
              if (use_busy_backoff_timer){
                  /* we received a busy signal from the consumer, wait for a while */
                  if (!busywaittime){
                      busywaittime = repl5_get_backoff_min(prp);
                  }
                  prp_priv->backoff = backoff_new(BACKOFF_FIXED, busywaittime, busywaittime);
              } else {
                  prp_priv->backoff = backoff_new(BACKOFF_EXPONENTIAL, repl5_get_backoff_min(prp),
                		  repl5_get_backoff_max(prp));
              }
              next_state = STATE_BACKOFF;
              backoff_reset(prp_priv->backoff, repl5_inc_backoff_expired, (void *)prp);
              protocol_sleep(prp, PR_INTERVAL_NO_TIMEOUT);
              use_busy_backoff_timer = PR_FALSE;
          }
          break;

      case STATE_BACKOFF:
          /*
           * We're in a backoff state.
           */
          dev_debug("repl5_inc_run(STATE_BACKOFF)");
          if (PROTOCOL_IS_SHUTDOWN(prp)){
              if (prp_priv->backoff)
                  backoff_delete(&prp_priv->backoff);
              done = 1;
              break;
          } else if (event_occurred(prp, EVENT_REPLICATE_NOW)){
              next_state = STATE_READY_TO_ACQUIRE;
          } else if (event_occurred(prp, EVENT_AGMT_CHANGED)){
              next_state = STATE_START;
              conn_set_agmt_changed(prp->conn);
              /* Destroy the backoff timer, since we won't need it anymore */
              if (prp_priv->backoff)
                  backoff_delete(&prp_priv->backoff);
          } else if (event_occurred(prp, EVENT_WINDOW_CLOSED)){
              next_state = STATE_WAIT_WINDOW_OPEN;
              /* Destroy the backoff timer, since we won't need it anymore */
              if (prp_priv->backoff)
                  backoff_delete(&prp_priv->backoff);
          } else if (event_occurred(prp, EVENT_BACKOFF_EXPIRED)){
              rc = acquire_replica(prp, REPL_NSDS50_INCREMENTAL_PROTOCOL_OID, &ruv);
              use_busy_backoff_timer = PR_FALSE;
              if (rc == ACQUIRE_SUCCESS){
            	  next_state = STATE_SENDING_UPDATES;
              } else if (rc == ACQUIRE_REPLICA_BUSY){
                  next_state = STATE_BACKOFF;
                  use_busy_backoff_timer = PR_TRUE;
              } else if (rc == ACQUIRE_CONSUMER_WAS_UPTODATE){
                  next_state = STATE_WAIT_CHANGES;
              } else if (rc == ACQUIRE_TRANSIENT_ERROR){
                  next_state = STATE_BACKOFF;
              } else if (rc == ACQUIRE_FATAL_ERROR){
                  next_state = STATE_STOP_FATAL_ERROR;
              }
              if (rc != ACQUIRE_SUCCESS){
                  conn_get_error(prp->conn, &optype, &ldaprc);
                  agmt_set_last_update_status(prp->agmt, ldaprc, prp->last_acquire_response_code, "Unable to acquire replica");
              }
              /*
               * We either need to step the backoff timer, or
               * destroy it if we don't need it anymore
               */
              if (STATE_BACKOFF == next_state){
                  /* Step the backoff timer */
                  time(&now);
                  next_fire_time = backoff_step(prp_priv->backoff);
                  /* And go back to sleep */
                  slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name,
                      "%s: Replication session backing off for %ld seconds\n",
                      agmt_get_long_name(prp->agmt),next_fire_time - now);
                  protocol_sleep(prp, PR_INTERVAL_NO_TIMEOUT);
              } else {
                  /* Destroy the backoff timer, since we won't need it anymore */
                  backoff_delete(&prp_priv->backoff);
              }
         } else if (event_occurred(prp, EVENT_TRIGGERING_CRITERIA_MET)){
             /* changes are available */
             if ( prp_priv->backoff == NULL || backoff_expired (prp_priv->backoff, 60)){
                 /*
                  * Have seen cases that the agmt stuck here forever since
                  * somehow the backoff timer was not in event queue anymore.
                  * If the backoff timer has expired more than 60 seconds, destroy it.
                  */
                 if ( prp_priv->backoff )
                     backoff_delete(&prp_priv->backoff);
                 next_state = STATE_READY_TO_ACQUIRE;
             } else {
                 /* ignore changes and go to sleep */
                 protocol_sleep(prp, PR_INTERVAL_NO_TIMEOUT);
             }
         } else if (event_occurred(prp, EVENT_WINDOW_OPENED)){
             /* this should never happen - log an error and go to sleep */
             slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name, "%s: Incremental protocol: "
                 "event %s should not occur in state %s; going to sleep\n",
                 agmt_get_long_name(prp->agmt), event2name(EVENT_WINDOW_OPENED),
                 state2name(current_state));
             protocol_sleep(prp, PR_INTERVAL_NO_TIMEOUT);
         }
         break;

      case STATE_SENDING_UPDATES:
          dev_debug("repl5_inc_run(STATE_SENDING_UPDATES)");
          num_changes_sent = 0;
          /*
           * We've acquired the replica, and are ready to send any needed updates.
           */
          if (PROTOCOL_IS_SHUTDOWN(prp)){
              release_replica (prp);
              done = 1;
              agmt_set_update_in_progress(prp->agmt, PR_FALSE);
              agmt_set_last_update_end(prp->agmt, current_time());
              /* MAB: I don't find the following status correct. How do we know it has
               * been stopped by an admin and not by a total update request, for instance?
               * In any case, how is this protocol shutdown situation different from all the
               * other ones that are present in this state machine? */
              /* richm: We at least need to let monitors know that the protocol has been
               * shutdown - maybe they can figure out why */
              agmt_set_last_update_status(prp->agmt, 0, 0, "Protocol stopped");
              agmt_update_done(prp->agmt, 0);
              break;
          }

          agmt_set_last_update_status(prp->agmt, 0, 0, "Incremental update started");
          /* ONREPL - in this state we send changes no matter what other events occur.
           * This is because we can get because of the REPLICATE_NOW event which
           * has high priority. Is this ok? */
          /* First, push new schema to the consumer if needed */
          /* ONREPL - should we push schema after we examine the RUV? */
          /*
           * GGOOREPL - I don't see why we should wait until we've
           * examined the RUV.  The schema entry has its own CSN that is
           * used to decide if the remote schema needs to be updated.
           */
          cons_schema_csn = agmt_get_consumer_schema_csn ( prp->agmt );
          rc = conn_push_schema(prp->conn, &cons_schema_csn);
          if ( cons_schema_csn != agmt_get_consumer_schema_csn ( prp->agmt )){
              agmt_set_consumer_schema_csn ( prp->agmt, cons_schema_csn );
          }
          if (CONN_SCHEMA_UPDATED != rc && CONN_SCHEMA_NO_UPDATE_NEEDED != rc){
              slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
                  "%s: Warning: unable to replicate schema: rc=%d\n", agmt_get_long_name(prp->agmt), rc);
              /* But keep going */
          }
          dev_debug("repl5_inc_run(STATE_SENDING_UPDATES) -> examine_update_vector");
          rc = examine_update_vector(prp, ruv);
          /*
           *  Decide what to do next - proceed with incremental, backoff, or total update
           */
          switch (rc){
              case EXAMINE_RUV_PARAM_ERROR:
                  /* this is really bad - we have NULL prp! */
                  next_state = STATE_STOP_FATAL_ERROR;
                  break;
              case EXAMINE_RUV_PRISTINE_REPLICA:
                  slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
                      "%s: Replica has no update vector. It has never been initialized.\n",
                      agmt_get_long_name(prp->agmt));
                  agmt_set_last_update_status(prp->agmt, 0, rc, "Replica is not initialized");
                  next_state = STATE_BACKOFF_START;
                  break;
              case EXAMINE_RUV_GENERATION_MISMATCH:
                  slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
                      "%s: The remote replica has a different database generation ID than "
                      "the local database.  You may have to reinitialize the remote replica, "
                      "or the local replica.\n", agmt_get_long_name(prp->agmt));
                  agmt_set_last_update_status(prp->agmt, 0, rc, "Replica has different database "
                      "generation ID, remote replica may need to be initialized");
                  next_state = STATE_BACKOFF_START;
                  break;
              case EXAMINE_RUV_REPLICA_TOO_OLD:
                  slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
                      "%s: Replica update vector is too out of date to bring "
                      "into sync using the incremental protocol. The replica "
                      "must be reinitialized.\n", agmt_get_long_name(prp->agmt));
                  agmt_set_last_update_status(prp->agmt, 0, rc, "Replica needs to be reinitialized");
                  next_state = STATE_BACKOFF_START;
                  break;
              case EXAMINE_RUV_OK:
                  /* update our csn generator state with the consumer's ruv data */
                  dev_debug("repl5_inc_run(STATE_SENDING_UPDATES) -> examine_update_vector OK");
                  object_acquire(prp->replica_object);
                  replica = object_get_data(prp->replica_object);
                  rc = replica_update_csngen_state (replica, ruv);
                  object_release (prp->replica_object);
                  replica = NULL;
                  if (rc == CSN_LIMIT_EXCEEDED) /* too much skew */ {
                      slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
                          "%s: Incremental protocol: fatal error - too much time skew between replicas!\n",
                          agmt_get_long_name(prp->agmt));
                      next_state = STATE_STOP_FATAL_ERROR;
                  } else if (rc != 0) /* internal error */ {
                      slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
                          "%s: Incremental protocol: fatal internal error updating the CSN generator!\n",
                          agmt_get_long_name(prp->agmt));
                      next_state = STATE_STOP_FATAL_ERROR;
                  } else {
                      /*
                       *  Reset our update times and status
                       */
                      agmt_set_last_update_start(prp->agmt, current_time());
                      agmt_set_last_update_end(prp->agmt, 0);
                      agmt_set_update_in_progress(prp->agmt, PR_TRUE);
                      /*
                       *  Send the updates
                       */
                      rc = send_updates(prp, ruv, &num_changes_sent);
                      if (rc == UPDATE_NO_MORE_UPDATES){
                          dev_debug("repl5_inc_run(STATE_SENDING_UPDATES) -> send_updates = UPDATE_NO_MORE_UPDATES -> STATE_WAIT_CHANGES");
                          agmt_set_last_update_status(prp->agmt, 0, 0, "Incremental update succeeded");
                          next_state = STATE_WAIT_CHANGES;
                      } else if (rc == UPDATE_YIELD){
                          dev_debug("repl5_inc_run(STATE_SENDING_UPDATES) -> send_updates = UPDATE_YIELD -> STATE_BACKOFF_START");
                          agmt_set_last_update_status(prp->agmt, 0, 0, "Incremental update succeeded and yielded");
                          next_state = STATE_BACKOFF_START;
                      } else if (rc == UPDATE_TRANSIENT_ERROR){
                          dev_debug("repl5_inc_run(STATE_SENDING_UPDATES) -> send_updates = UPDATE_TRANSIENT_ERROR -> STATE_BACKOFF_START");
                          agmt_set_last_update_status(prp->agmt, 0, rc, "Incremental update transient error.  Backing off, will retry update later.");
                          next_state = STATE_BACKOFF_START;
                      } else if (rc == UPDATE_FATAL_ERROR){
                          dev_debug("repl5_inc_run(STATE_SENDING_UPDATES) -> send_updates = UPDATE_FATAL_ERROR -> STATE_STOP_FATAL_ERROR");
                          next_state = STATE_STOP_FATAL_ERROR;
                      } else if (rc == UPDATE_SCHEDULE_WINDOW_CLOSED){
                          dev_debug("repl5_inc_run(STATE_SENDING_UPDATES) -> send_updates = UPDATE_SCHEDULE_WINDOW_CLOSED -> STATE_WAIT_WINDOW_OPEN");
                          /*
                           * ONREPL - I don't think we should check this. We might be
                           * here because of replicate_now event - so we don't care
                           * about the schedule
                           */
                          next_state = STATE_WAIT_WINDOW_OPEN;
                          /* ONREPL - do we need to release the replica here ? */
                          conn_disconnect (prp->conn);
                      } else if (rc == UPDATE_CONNECTION_LOST){
                          dev_debug("repl5_inc_run(STATE_SENDING_UPDATES) -> send_updates = UPDATE_CONNECTION_LOST -> STATE_BACKOFF_START");
                          agmt_set_last_update_status(prp->agmt, 0, rc, "Incremental update connection error.  Backing off, will retry update later.");
                          next_state = STATE_BACKOFF_START;
                      } else if (rc == UPDATE_TIMEOUT){
                          dev_debug("repl5_inc_run(STATE_SENDING_UPDATES) -> send_updates = UPDATE_TIMEOUT -> STATE_BACKOFF_START");
                          agmt_set_last_update_status(prp->agmt, 0, rc, "Incremental update timeout error.  Backing off, will retry update later.");
                          next_state = STATE_BACKOFF_START;
                      }
                      /* Set the updates times based off the result of send_updates() */
                      if(rc == UPDATE_NO_MORE_UPDATES){
                          /* update successful, set the end time */
                    	  agmt_set_last_update_end(prp->agmt, current_time());
                      } else {
                          /* Failed to send updates, reset the start time to zero */
                          agmt_set_last_update_start(prp->agmt, 0);
                      }
                      agmt_set_update_in_progress(prp->agmt, PR_FALSE);
                  }
                  break;
          }

          if (NULL != ruv){
              ruv_destroy(&ruv); ruv = NULL;
          }
          agmt_update_done(prp->agmt, 0);
          /* If timed out, close the connection after released the replica */
          release_replica(prp);
          if (rc == UPDATE_TIMEOUT) {
              conn_disconnect(prp->conn);
          }
          if (rc == UPDATE_NO_MORE_UPDATES && num_changes_sent > 0){
              if (pausetime > 0){
                  /* richm - 20020219 - If we have acquired the consumer, and another master has gone
                   * into backoff waiting for us to release it, we may acquire the replica sooner
                   * than the other master has a chance to, and the other master may not be able
                   * to acquire the consumer for a long time (hours, days?) if this server is
                   * under a heavy load (see reliab06 et. al. system tests)
                   * So, this sleep gives the other master(s) a chance to acquire the consumer replica */
                  loops = pausetime;
                  /* the while loop is so that we don't just sleep and sleep if an
                   * event comes in that we should handle immediately (like shutdown) */
                  slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name,
                      "%s: Pausing updates for %ld seconds to allow other suppliers to update consumer\n",
                      agmt_get_long_name(prp->agmt), pausetime);
                  while (loops-- && !(PROTOCOL_IS_SHUTDOWN(prp))){
                      DS_Sleep(PR_SecondsToInterval(1));
                  }
              } else if (num_changes_sent > 10){
                  /* wait for consumer to write its ruv if the replication was busy */
                  /* When asked, consumer sends its ruv in cache to the supplier. */
                  /* DS_Sleep ( PR_SecondsToInterval(1) ); */
              }
          }
          break;

      case STATE_STOP_FATAL_ERROR:
          /*
           * We encountered some sort of a fatal error. Suspend.
           */
          /* XXXggood update state in replica */
          agmt_set_last_update_status(prp->agmt, -1, 0, "Incremental update has failed and requires administrator action");
          dev_debug("repl5_inc_run(STATE_STOP_FATAL_ERROR)");
          slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
              "%s: Incremental update failed and requires administrator action\n",
              agmt_get_long_name(prp->agmt));
          next_state = STATE_STOP_FATAL_ERROR_PART2;
          break;

      case STATE_STOP_FATAL_ERROR_PART2:
          if (PROTOCOL_IS_SHUTDOWN(prp)){
              done = 1;
              break;
          }
          /* MAB: This state is the FATAL state where we are supposed to get
           * as a result of a FATAL error on send_updates. But, as bug
           * states, send_updates was always returning TRANSIENT errors and never
           * FATAL... In other words, this code has never been tested before...
           *
           * As of 01/16/01, this piece of code was in a very dangerous state. In particular,
           * 1) it does not catch any events
           * 2) it is a terminal state (once reached it never transitions to a different state)
           *
           * Both things combined make this state to become a consuming infinite loop
           * that is useless after all (we are in a fatal place requiring manual admin jobs */

          /* MAB: The following lines fix problem number 1 above... When the code gets
           * into this state, it should only get a chance to get out of it by an
           * EVENT_AGMT_CHANGED event... All other events should be ignored */
          else if (event_occurred(prp, EVENT_AGMT_CHANGED)){
             dev_debug("repl5_inc_run(STATE_STOP_FATAL_ERROR): EVENT_AGMT_CHANGED received\n");
             /* Chance to recover for the EVENT_AGMT_CHANGED event.
              * This is not mandatory, but fixes problem 2 above */
             next_state = STATE_STOP_NORMAL_TERMINATION;
          } else {
              dev_debug("repl5_inc_run(STATE_STOP_FATAL_ERROR): Event received. Clearing it\n");
              reset_events (prp);
          }

          protocol_sleep (prp, PR_INTERVAL_NO_TIMEOUT);
          break;

      case STATE_STOP_NORMAL_TERMINATION:
          /*
           * We encountered some sort of a fatal error. Return.
           */
          /* XXXggood update state in replica */
          dev_debug("repl5_inc_run(STATE_STOP_NORMAL_TERMINATION)");
          done = 1;
          break;
      }

      slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name,"%s: State: %s -> %s\n",
	      agmt_get_long_name(prp->agmt),state2name(current_state), state2name(next_state));

      current_state = next_state;
  } while (!done);

  /* remove_protocol_callbacks(prp); */
  prp->stopped = 1;
  /* Cancel any linger timer that might be in effect... */
  conn_cancel_linger(prp->conn);
  /* ... and disconnect, if currently connected */
  conn_disconnect(prp->conn);
}

/*
 * Go to sleep until awakened.
 */
static void
protocol_sleep(Private_Repl_Protocol *prp, PRIntervalTime duration)
{
	PR_ASSERT(NULL != prp);
	PR_Lock(prp->lock);
    /* we should not go to sleep if there are events available to be processed.
       Otherwise, we can miss the event that suppose to wake us up */
    if (prp->eventbits == 0)
	    PR_WaitCondVar(prp->cvar, duration);
    else
    {
        slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name,
			"%s: Incremental protocol: can't go to sleep: event bits - %x\n",
			agmt_get_long_name(prp->agmt), prp->eventbits);
    }
	PR_Unlock(prp->lock);
}

/*
 * Notify the protocol about some event. Signal the condition
 * variable in case the protocol is sleeping. Multiple occurences
 * of a single event type are not remembered (e.g. no stack
 * of events is maintained).
 */
static void
event_notify(Private_Repl_Protocol *prp, PRUint32 event)
{
	PR_ASSERT(NULL != prp);
	PR_Lock(prp->lock);
	prp->eventbits |= event;
	PR_NotifyCondVar(prp->cvar);
	PR_Unlock(prp->lock);
}

/*
 * Test to see if an event occurred. The event is cleared when
 * read.
 */
static PRUint32
event_occurred(Private_Repl_Protocol *prp, PRUint32 event)
{
	PRUint32 return_value;
	PR_ASSERT(NULL != prp);
	PR_Lock(prp->lock);
	return_value = (prp->eventbits & event);
	prp->eventbits &= ~event; /* Clear event */
	PR_Unlock(prp->lock);
	return return_value;
}

static void
reset_events (Private_Repl_Protocol *prp)
{
	PR_ASSERT(NULL != prp);
	PR_Lock(prp->lock);
	prp->eventbits = 0;
	PR_Unlock(prp->lock);
}

/*
 * Replay the actual update to the consumer. Construct an appropriate LDAP
 * operation, attach the baggage LDAPv3 control that contains the CSN, etc.,
 * and send the operation to the consumer. 
 */
ConnResult
replay_update(Private_Repl_Protocol *prp, slapi_operation_parameters *op, int *message_id)
{
	ConnResult return_value = CONN_OPERATION_FAILED;
	LDAPControl *update_control;
	char *parentuniqueid;
	LDAPMod **modrdn_mods = NULL;
	char csn_str[CSN_STRSIZE]; /* For logging only */

	if (message_id) {
		/* if we get out of this function without setting message_id, it means
		   we didn't send an op, so no result needs to be processed */
		*message_id = 0;
	}

	/* Construct the replication info control that accompanies the operation */
	if (SLAPI_OPERATION_ADD == op->operation_type)
	{
		parentuniqueid = op->p.p_add.parentuniqueid;
	}
	else if (SLAPI_OPERATION_MODRDN == op->operation_type)
	{
		/*
		 * For modrdn operations, we need to send along modified attributes, e.g.
		 * modifytimestamp.
		 * And the superior_uniqueid !
		 */
		modrdn_mods = op->p.p_modrdn.modrdn_mods;
		parentuniqueid = op->p.p_modrdn.modrdn_newsuperior_address.uniqueid;
	}
	else 
	{
		parentuniqueid = NULL;
	}
	if (create_NSDS50ReplUpdateInfoControl(op->target_address.uniqueid,
		parentuniqueid, op->csn, modrdn_mods, &update_control) != LDAP_SUCCESS)
	{
		slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
			"%s: replay_update: Unable to create NSDS50ReplUpdateInfoControl "
			"for operation with csn %s. Skipping update.\n",
			agmt_get_long_name(prp->agmt), csn_as_string(op->csn, PR_FALSE, csn_str));
	}
	else
	{
		if (slapi_is_loglevel_set(SLAPI_LOG_REPL)) {
			slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name,
							"%s: replay_update: Sending %s operation (dn=\"%s\" csn=%s)\n",
							agmt_get_long_name(prp->agmt),
							op2string(op->operation_type), REPL_GET_DN(&op->target_address),
							csn_as_string(op->csn, PR_FALSE, csn_str));
		}
		/* What type of operation is it? */
		switch (op->operation_type)
		{
		case SLAPI_OPERATION_ADD:
		{
			LDAPMod **entryattrs;
			/* Convert entry to mods */
			(void)slapi_entry2mods (op->p.p_add.target_entry, 
									NULL /* &entrydn : We don't need it */, 
									&entryattrs);
			if (NULL == entryattrs)
			{
				slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
					"%s: replay_update: Cannot convert entry to LDAPMods.\n",
					agmt_get_long_name(prp->agmt));
				return_value = CONN_LOCAL_ERROR;
			}
			else
			{
				/* If fractional agreement, trim down the entry */
				if (agmt_is_fractional(prp->agmt))
				{
					repl5_strip_fractional_mods(prp->agmt,entryattrs);
				}
				if (MODS_ARE_EMPTY(entryattrs)) {
					if (slapi_is_loglevel_set(SLAPI_LOG_REPL)) {
						slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name,
										"%s: replay_update: %s operation (dn=\"%s\" csn=%s) "
										"not sent - empty\n",
										agmt_get_long_name(prp->agmt),
										op2string(op->operation_type),
										REPL_GET_DN(&op->target_address),
										csn_as_string(op->csn, PR_FALSE, csn_str));
					}
					return_value = CONN_OPERATION_SUCCESS;
				} else {
					return_value = conn_send_add(prp->conn, REPL_GET_DN(&op->target_address),
						entryattrs, update_control, message_id);
				}
				ldap_mods_free(entryattrs, 1);
			}
			break;
		}
		case SLAPI_OPERATION_MODIFY:
			/* If fractional agreement, trim down the mods */
			if (agmt_is_fractional(prp->agmt))
			{
				repl5_strip_fractional_mods(prp->agmt,op->p.p_modify.modify_mods);
			}
			if (MODS_ARE_EMPTY(op->p.p_modify.modify_mods)) {
				if (slapi_is_loglevel_set(SLAPI_LOG_REPL)) {
					slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name,
									"%s: replay_update: %ss operation (dn=\"%s\" csn=%s) "
									"not sent - empty\n",
									agmt_get_long_name(prp->agmt),
									op2string(op->operation_type),
									REPL_GET_DN(&op->target_address),
									csn_as_string(op->csn, PR_FALSE, csn_str));
				}
				return_value = CONN_OPERATION_SUCCESS;
			} else {
				return_value = conn_send_modify(prp->conn, REPL_GET_DN(&op->target_address),
					op->p.p_modify.modify_mods, update_control, message_id);
			}
			break;
		case SLAPI_OPERATION_DELETE:
			return_value = conn_send_delete(prp->conn, REPL_GET_DN(&op->target_address),
				update_control, message_id);
			break;
		case SLAPI_OPERATION_MODRDN:
			/* XXXggood need to pass modrdn mods in update control! */
			return_value = conn_send_rename(prp->conn, REPL_GET_DN(&op->target_address),
				op->p.p_modrdn.modrdn_newrdn,
				REPL_GET_DN(&op->p.p_modrdn.modrdn_newsuperior_address),
				op->p.p_modrdn.modrdn_deloldrdn,
				update_control, message_id);
			break;
		default:
			slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name, "%s: replay_update: Unknown "
				"operation type %lu found in changelog - skipping change.\n",
				agmt_get_long_name(prp->agmt), op->operation_type);
		}

		destroy_NSDS50ReplUpdateInfoControl(&update_control);
	}

	if (CONN_OPERATION_SUCCESS == return_value)
	{
		if (slapi_is_loglevel_set(SLAPI_LOG_REPL)) {
			slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name,
							"%s: replay_update: Consumer successfully sent operation with csn %s\n",
							agmt_get_long_name(prp->agmt), csn_as_string(op->csn, PR_FALSE, csn_str));
		}
	}
	else
	{
		if (slapi_is_loglevel_set(SLAPI_LOG_REPL)) {
			slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name,
							"%s: replay_update: Consumer could not replay operation with csn %s\n",
							agmt_get_long_name(prp->agmt), csn_as_string(op->csn, PR_FALSE, csn_str));
		}
	}
	return return_value;
}

static PRBool
is_dummy_operation (const slapi_operation_parameters *op)
{
    return (strcmp (op->target_address.uniqueid, START_ITERATION_ENTRY_UNIQUEID) == 0);
}

void
cl5_operation_parameters_done (struct slapi_operation_parameters *sop)
{
	if(sop!=NULL) {
		switch(sop->operation_type) 
		{
		case SLAPI_OPERATION_BIND:
			slapi_ch_free((void **)&(sop->p.p_bind.bind_saslmechanism));
			if (sop->p.p_bind.bind_creds)
				ber_bvecfree((struct berval**)&(sop->p.p_bind.bind_creds));
			if (sop->p.p_bind.bind_ret_saslcreds)
				ber_bvecfree((struct berval**)&(sop->p.p_bind.bind_ret_saslcreds));
			sop->p.p_bind.bind_creds = NULL;
			sop->p.p_bind.bind_ret_saslcreds = NULL;
			break;
		case SLAPI_OPERATION_COMPARE:
			ava_done((struct ava *)&(sop->p.p_compare.compare_ava));
			break;
		case SLAPI_OPERATION_SEARCH:
			slapi_ch_free((void **)&(sop->p.p_search.search_strfilter));
			charray_free(sop->p.p_search.search_attrs);
			slapi_filter_free(sop->p.p_search.search_filter,1);
			break;
		case SLAPI_OPERATION_MODRDN:
			sop->p.p_modrdn.modrdn_deloldrdn = 0;
			break;
		case SLAPI_OPERATION_EXTENDED:
			slapi_ch_free((void **)&(sop->p.p_extended.exop_oid));
			if (sop->p.p_extended.exop_value)
				ber_bvecfree((struct berval**)&(sop->p.p_extended.exop_value));
			sop->p.p_extended.exop_value = NULL;
			break;
		default:
			break;
		}
	}
	operation_parameters_done(sop);

}

/* Helper to update the agreement state based on a the result of a replay operation */
static int
repl5_inc_update_from_op_result(Private_Repl_Protocol *prp, ConnResult replay_crc, int connection_error, char *csn_str, char *uniqueid, ReplicaId replica_id, int* finished, PRUint32 *num_changes_sent)
{
	int return_value = 0;

	if (CONN_OPERATION_SUCCESS != replay_crc)
	{
		/* Figure out what to do next */
		if (CONN_OPERATION_FAILED == replay_crc)
		{
			/* Map ldap error code to return value */
			if (!ignore_error_and_keep_going(connection_error))
			{
				return_value = UPDATE_TRANSIENT_ERROR;
				*finished = 1;
			}
			else
			{
				agmt_inc_last_update_changecount (prp->agmt, replica_id, 1 /*skipped*/);
			}
			slapi_log_error(*finished ? SLAPI_LOG_FATAL : slapi_log_urp, repl_plugin_name,
				"%s: Consumer failed to replay change (uniqueid %s, CSN %s): %s (%d). %s.\n",
				agmt_get_long_name(prp->agmt),
				uniqueid, csn_str,
				ldap_err2string(connection_error), connection_error,
				*finished ? "Will retry later" : "Skipping");
		}
		else if (CONN_NOT_CONNECTED == replay_crc)
		{
			/* We lost the connection - enter backoff state */

			return_value = UPDATE_CONNECTION_LOST;
			*finished = 1;
			slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
				"%s: Consumer failed to replay change (uniqueid %s, CSN %s): "
				"%s(%d). Will retry later.\n",
				agmt_get_long_name(prp->agmt),
				uniqueid, csn_str,
				connection_error ? ldap_err2string(connection_error) : "Connection lost",
				connection_error);
		}
		else if (CONN_TIMEOUT == replay_crc)
		{
			return_value = UPDATE_TIMEOUT;
			*finished = 1;
			slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
				"%s: Consumer timed out to replay change (uniqueid %s, CSN %s): "
				"%s.\n",
				agmt_get_long_name(prp->agmt),
				uniqueid, csn_str,
				connection_error ? ldap_err2string(connection_error) : "Timeout");
		}
		else if (CONN_LOCAL_ERROR == replay_crc)
		{
			/*
			 * Something bad happened on the local server - enter
			 * backoff state.
			 */
			return_value = UPDATE_TRANSIENT_ERROR;
			*finished = 1;
			slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
				"%s: Failed to replay change (uniqueid %s, CSN %s): "
				"Local error. Will retry later.\n",
				agmt_get_long_name(prp->agmt),
				uniqueid, csn_str);
		}
		if (*finished){
			/*
			 * A serious error has occurred, the consumer might have closed
			 * the connection already, but we need to close the conn on the
			 * supplier side to properly set the conn structure as closed.
			 */
			conn_disconnect(prp->conn);
		}
	}
	else
	{
		/* Positive response received */
		(*num_changes_sent)++;
		agmt_inc_last_update_changecount (prp->agmt, replica_id, 0 /*replayed*/);
	}
	return return_value;
}

/*
 * Send a set of updates to the replica.  Assumes that (1) the replica
 * has already been acquired, (2) that the consumer's update vector has
 * been checked and (3) that it's ok to send incremental updates.
 * Returns:
 * UPDATE_NO_MORE_UPDATES - all updates were sent successfully
 * UPDATE_TRANSIENT_ERROR - some non-permanent error occurred. Try again later.
 * UPDATE_FATAL_ERROR - some bad, permanent error occurred.
 * UPDATE_SCHEDULE_WINDOW_CLOSED - the schedule window closed on us.
 */
static int
send_updates(Private_Repl_Protocol *prp, RUV *remote_update_vector, PRUint32 *num_changes_sent)
{
	CL5Entry entry;
	slapi_operation_parameters op;
	int return_value = 0;
	int rc;
	CL5ReplayIterator *changelog_iterator;
	int message_id = 0;
	result_data *rd = NULL;

	*num_changes_sent = 0;
	/*
	 * Iterate over the changelog. Retrieve each update,
	 * construct an appropriate LDAP operation,
	 * attaching the CSN, and send the change.
	 */
    
	rc = cl5CreateReplayIterator(prp, remote_update_vector, &changelog_iterator);
	if (CL5_SUCCESS != rc)
	{
		switch (rc)
		{
		case CL5_BAD_DATA: /* invalid parameter passed to the function */
			slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
				"%s: Invalid parameter passed to cl5CreateReplayIterator\n",
				agmt_get_long_name(prp->agmt));
			return_value = UPDATE_FATAL_ERROR;
			break;
		case CL5_BAD_FORMAT:     /* db data has unexpected format */
			slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
				"%s: Unexpected format encountered in changelog database\n",
				agmt_get_long_name(prp->agmt));
			return_value = UPDATE_FATAL_ERROR;
			break;
		case CL5_BAD_STATE: /* changelog is in an incorrect state for attempted operation */
			slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
				"%s: Changelog database was in an incorrect state\n",
				agmt_get_long_name(prp->agmt));
			return_value = UPDATE_FATAL_ERROR;
			break;
		case CL5_BAD_DBVERSION:  /* changelog has invalid dbversion */
			slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
				"%s: Incorrect dbversion found in changelog database\n",
				agmt_get_long_name(prp->agmt));
			return_value = UPDATE_FATAL_ERROR;
			break;
		case CL5_DB_ERROR:       /* database error */
			slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
				"%s: A changelog database error was encountered\n",
				agmt_get_long_name(prp->agmt));
			return_value = UPDATE_FATAL_ERROR;
			break;
		case CL5_NOTFOUND:       /* we have no changes to send */
			slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name,
				"%s: No changes to send\n",
				agmt_get_long_name(prp->agmt));
			return_value = UPDATE_NO_MORE_UPDATES;
			break;
		case CL5_MEMORY_ERROR:   /* memory allocation failed */
			slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
				"%s: Memory allocation error occurred\n",
				agmt_get_long_name(prp->agmt));
			return_value = UPDATE_FATAL_ERROR;
			break;
		case CL5_SYSTEM_ERROR:   /* NSPR error occurred: use PR_GetError for further info */
			slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
				"%s: An NSPR error (%d) occurred\n",
				agmt_get_long_name(prp->agmt), PR_GetError());
			return_value = UPDATE_TRANSIENT_ERROR;
			break;
		case CL5_CSN_ERROR:      /* CSN API failed */
			slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
				"%s: A CSN API failure was encountered\n",
				agmt_get_long_name(prp->agmt));
			return_value = UPDATE_TRANSIENT_ERROR;
			break;
		case CL5_RUV_ERROR:      /* RUV API failed */
			slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
				"%s: An RUV API failure occurred\n",
				agmt_get_long_name(prp->agmt));
			return_value = UPDATE_TRANSIENT_ERROR;
			break;
		case CL5_OBJSET_ERROR:   /* namedobjset api failed */
			slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
				"%s: A namedobject API failure occurred\n",
				agmt_get_long_name(prp->agmt));
			return_value = UPDATE_TRANSIENT_ERROR;
			break;
		case CL5_PURGED_DATA:    /* requested data has been purged */
			slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
				"%s: Data required to update replica has been purged. "
				"The replica must be reinitialized.\n",
				agmt_get_long_name(prp->agmt));
			return_value = UPDATE_FATAL_ERROR;
			break;
		case CL5_MISSING_DATA:   /* data should be in the changelog, but is missing */
			slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
				"%s: Missing data encountered\n",
				agmt_get_long_name(prp->agmt));
			return_value = UPDATE_FATAL_ERROR;
			break;
		case CL5_UNKNOWN_ERROR:   /* unclassified error */
			slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
				"%s: An unknown error was encountered\n",
				agmt_get_long_name(prp->agmt));
			return_value = UPDATE_TRANSIENT_ERROR;
			break;
		default:
			slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
				"%s: An unknown error (%d) occurred "
				"(cl5CreateReplayIterator)\n",
				agmt_get_long_name(prp->agmt), rc);
			return_value = UPDATE_TRANSIENT_ERROR;
		}
	}
	else
	{
		int finished = 0;
		ConnResult replay_crc;
		char csn_str[CSN_STRSIZE];
		PRBool subentry_update_needed = PR_FALSE;
		int skipped_updates = 0;
		int fractional_repl;
#define FRACTIONAL_SKIPPED_THRESHOLD 100

		/* Start the results reading thread */
		rd = repl5_inc_rd_new(prp);
		if (!prp->repl50consumer) 
		{
			rc = repl5_inc_create_async_result_thread(rd);
			if (rc) {
				slapi_log_error (SLAPI_LOG_FATAL, repl_plugin_name, "%s: repl5_inc_run: "
							 "repl5_tot_create_async_result_thread failed; error - %d\n", 
							 agmt_get_long_name(prp->agmt), rc);
				return_value = UPDATE_FATAL_ERROR;
			}
		}

		memset ( (void*)&op, 0, sizeof (op) );
		entry.op = &op;
		fractional_repl = agmt_is_fractional(prp->agmt);
		do {
			cl5_operation_parameters_done ( entry.op );
			memset ( (void*)entry.op, 0, sizeof (op) );
			rc = cl5GetNextOperationToReplay(changelog_iterator, &entry);
			switch (rc)
			{
			case CL5_SUCCESS:
				/* check that we don't return dummy entries */
				if (is_dummy_operation (entry.op))
				{
					slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name, 
						"%s: changelog iteration code returned a dummy entry with csn %s, "
						"skipping ...\n",
						agmt_get_long_name(prp->agmt), csn_as_string(entry.op->csn, PR_FALSE, csn_str));
					continue;
				}
				replay_crc = replay_update(prp, entry.op, &message_id);
				if (message_id) 
				{
					rd->last_message_id_sent = message_id;
				}
				/* If we're talking to an old non-async replica, we need to pick up the response here */
				if (CONN_OPERATION_SUCCESS != replay_crc)
				{
					int operation, error;
					conn_get_error(prp->conn, &operation, &error);
					csn_as_string(entry.op->csn, PR_FALSE, csn_str);
					/* Figure out what to do next */
					if (CONN_OPERATION_FAILED == replay_crc)
					{
						/* Map ldap error code to return value */
						if (!ignore_error_and_keep_going(error))
						{
							return_value = UPDATE_TRANSIENT_ERROR;
							finished = 1;
						}
						else
						{
							agmt_inc_last_update_changecount (prp->agmt, csn_get_replicaid(entry.op->csn), 1 /*skipped*/);
						}
						slapi_log_error(finished ? SLAPI_LOG_FATAL : slapi_log_urp, repl_plugin_name,
							"%s: Failed to send update operation to consumer (uniqueid %s, CSN %s): %s. %s.\n",
							agmt_get_long_name(prp->agmt),
							entry.op->target_address.uniqueid, csn_str,
							ldap_err2string(error),
							finished ? "Will retry later" : "Skipping");
					}
					else if (CONN_NOT_CONNECTED == replay_crc)
					{
						/* We lost the connection - enter backoff state */

						return_value = UPDATE_CONNECTION_LOST;
						finished = 1;
						slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
							"%s: Failed to send update operation to consumer (uniqueid %s, CSN %s): "
							"%s. Will retry later.\n",
							agmt_get_long_name(prp->agmt),
							entry.op->target_address.uniqueid, csn_str,
							error ? ldap_err2string(error) : "Connection lost");
					}
					else if (CONN_TIMEOUT == replay_crc)
					{
						return_value = UPDATE_TIMEOUT;
						finished = 1;
						slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
							"%s: Timed out sending update operation to consumer (uniqueid %s, CSN %s): "
							"%s.\n",
							agmt_get_long_name(prp->agmt),
							entry.op->target_address.uniqueid, csn_str,
							error ? ldap_err2string(error) : "Timeout");
					}
					else if (CONN_LOCAL_ERROR == replay_crc)
					{
						/*
						 * Something bad happened on the local server - enter 
						 * backoff state.
						 */
						return_value = UPDATE_TRANSIENT_ERROR;
						finished = 1;
						slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
							"%s: Failed to send update operation to consumer (uniqueid %s, CSN %s): "
							"Local error. Will retry later.\n",
							agmt_get_long_name(prp->agmt),
							entry.op->target_address.uniqueid, csn_str);
					}
						
				}
				else
				{
					char *uniqueid = NULL;
					ReplicaId replica_id = 0;

					csn_as_string(entry.op->csn, PR_FALSE, csn_str);
					replica_id = csn_get_replicaid(entry.op->csn);
					uniqueid = entry.op->target_address.uniqueid;
                    
					if (fractional_repl && message_id) 
					{
						/* This update was sent no need to update the subentry
						 * and restart counting the skipped updates
						 */
						subentry_update_needed = PR_FALSE;
						skipped_updates = 0;
					}

					if (prp->repl50consumer && message_id) 
					{
						int operation, error = 0;

						conn_get_error(prp->conn, &operation, &error);

						/* Get the response here */
						replay_crc = repl5_inc_get_next_result(rd);
						conn_get_error(prp->conn, &operation, &error);
						csn_as_string(entry.op->csn, PR_FALSE, csn_str);
						return_value = repl5_inc_update_from_op_result(prp, replay_crc, error, csn_str, uniqueid, replica_id, &finished, num_changes_sent);
					}
					else if (message_id) {
						/* Queue the details for pickup later in the response thread */
						repl5_inc_operation *sop = NULL;
						sop = repl5_inc_operation_new();
						PL_strncpyz(sop->csn_str, csn_str, sizeof(sop->csn_str));
						sop->ldap_message_id = message_id;
						sop->operation_type = entry.op->operation_type;
						sop->replica_id = replica_id;
						PL_strncpyz(sop->uniqueid, uniqueid, sizeof(sop->uniqueid));
						repl5_int_push_operation(rd,sop);
						repl5_inc_flow_control_results(prp->agmt, rd);
					} else {
						slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name,
							"%s: Skipping update operation with no message_id (uniqueid %s, CSN %s):\n",
							agmt_get_long_name(prp->agmt),
							entry.op->target_address.uniqueid, csn_str);
						agmt_inc_last_update_changecount (prp->agmt, csn_get_replicaid(entry.op->csn), 1 /*skipped*/);
						if (fractional_repl) 
						{
							skipped_updates++;
							if (skipped_updates > FRACTIONAL_SKIPPED_THRESHOLD) {
								slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name,
										"%s: skipped updates is too high (%d) if no other update is sent we will update the subentry\n",
										agmt_get_long_name(prp->agmt), skipped_updates);
								subentry_update_needed = PR_TRUE;
							}
						}
					}
				}
				break;
			case CL5_BAD_DATA:
				slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
					"%s: Invalid parameter passed to cl5GetNextOperationToReplay\n",
					agmt_get_long_name(prp->agmt));
				return_value = UPDATE_FATAL_ERROR;
				finished = 1;
				break;
			case CL5_NOTFOUND:
				slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name,
					"%s: No more updates to send (cl5GetNextOperationToReplay)\n",
					agmt_get_long_name(prp->agmt));
				return_value = UPDATE_NO_MORE_UPDATES;
				finished = 1;
				break;
			case CL5_DB_ERROR:
				slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
					"%s: A database error occurred (cl5GetNextOperationToReplay)\n",
					agmt_get_long_name(prp->agmt));
				return_value = UPDATE_FATAL_ERROR;
				finished = 1;
				break;
			case CL5_BAD_FORMAT:
				slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
					"%s: A malformed changelog entry was encountered (cl5GetNextOperationToReplay)\n",
					agmt_get_long_name(prp->agmt));
				break;
			case CL5_MEMORY_ERROR:
				slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
					"%s: A memory allocation error occurred (cl5GetNextOperationToRepla)\n",
					agmt_get_long_name(prp->agmt));
				return_value = UPDATE_FATAL_ERROR;
				break;
			case CL5_IGNORE_OP:
				break;
			default:
				slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
					"%s: Unknown error code (%d) returned from cl5GetNextOperationToReplay\n",
					agmt_get_long_name(prp->agmt), rc);
				return_value = UPDATE_TRANSIENT_ERROR;
				break;
			}
			/* Check for protocol shutdown */
			if (prp->terminate)
			{
				return_value = UPDATE_NO_MORE_UPDATES;
				finished = 1;
			}
			if (*num_changes_sent >= MAX_CHANGES_PER_SESSION)
			{
				return_value = UPDATE_YIELD;
				finished = 1;
			}
			PR_Lock(rd->lock);
			/* See if the result thread has hit a problem */
			if (!finished && rd->abort)
			{
				return_value = rd->result;
				finished = 1;
			}
			PR_Unlock(rd->lock);
		} while (!finished);

		if (fractional_repl && subentry_update_needed)
		{
			Replica *replica;
			ReplicaId rid = -1; /* Used to create the replica keep alive subentry */
			Slapi_DN *replarea_sdn = NULL;
			replica = (Replica*) object_get_data(prp->replica_object);
			if (replica)
			{
				rid = replica_get_rid(replica);
			}
			slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name,
					"%s: skipped updates was definitely too high (%d) update the subentry now\n",
					agmt_get_long_name(prp->agmt), skipped_updates);
			replarea_sdn = agmt_get_replarea(prp->agmt);
			if (!replarea_sdn) {
				slapi_log_error(SLAPI_LOG_FATAL, repl_plugin_name,
				                "send_updates: Unknown replication area due to agreement not found.");
				return_value = UPDATE_FATAL_ERROR;
			} else {
				replica_subentry_update(replarea_sdn, rid);
			}
		}
		/* Terminate the results reading thread */
		if (!prp->repl50consumer) 
		{
			/* We need to ensure that we wait until all the responses have been received from our operations */
			if (return_value != UPDATE_CONNECTION_LOST) {
				/*
				 * If we already have an error, there is no need to check the
				 * async result thread anymore.
				 */
				if (return_value == UPDATE_NO_MORE_UPDATES)
				{
					/*
					 * We need to double check that an error hasn't popped up from
					 * the async result thread since our last check.
					 */
					int final_result;

					rd->WaitForAsyncResults = agmt_get_WaitForAsyncResults(prp->agmt);
					if((final_result = repl5_inc_waitfor_async_results(rd))){
						return_value = final_result;
					}
				}
			}

			rc = repl5_inc_destroy_async_result_thread(rd);
			if (rc) {
				slapi_log_error (SLAPI_LOG_FATAL, repl_plugin_name, "%s: repl5_inc_run: "
							 "repl5_tot_destroy_async_result_thread failed; error - %d\n", 
							 agmt_get_long_name(prp->agmt), rc);
			}
			*num_changes_sent = rd->num_changes_sent;
		}
		PR_Lock(rd->lock);
		if (rd->flowcontrol_detection) {
			slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name,
					"%s: Incremental update flow control triggered %d times\n"
					"You may increase %s and/or decrease %s in the replica agreement configuration\n",
					agmt_get_long_name(prp->agmt),
					rd->flowcontrol_detection,
					type_nsds5ReplicaFlowControlPause,
					type_nsds5ReplicaFlowControlWindow);             
		}
		PR_Unlock(rd->lock);
		repl5_inc_rd_destroy(&rd);

		cl5_operation_parameters_done ( entry.op );
		cl5DestroyReplayIterator(&changelog_iterator);
	}
	return return_value;
}

/*
 * XXXggood this should probably be in the superclass, since the full update
 * protocol is going to need it too.
 */
static int
repl5_inc_stop(Private_Repl_Protocol *prp)
{
	PRIntervalTime start, maxwait, now;
	Replica *replica = NULL;
	PRUint64 timeout;
	int return_value;

	if((timeout = agmt_get_protocol_timeout(prp->agmt)) == 0){
		timeout = DEFAULT_PROTOCOL_TIMEOUT;
		if(prp->replica_object){
			object_acquire(prp->replica_object);
			replica = object_get_data(prp->replica_object);
			if((timeout = replica_get_protocol_timeout(replica)) == 0){
				timeout = DEFAULT_PROTOCOL_TIMEOUT;
			}
			object_release(prp->replica_object);
		}
	}

	maxwait = PR_SecondsToInterval(timeout);
	prp->terminate = 1;
	event_notify(prp, EVENT_PROTOCOL_SHUTDOWN);
	start = PR_IntervalNow();
	now = start;
	while (!prp->stopped && ((now - start) < maxwait))
	{
		DS_Sleep(PR_MillisecondsToInterval(100));
		now = PR_IntervalNow();
	}
	if (!prp->stopped)
	{
		/* Isn't listening. Do something drastic. */
		return_value = -1;
		slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name,
				"%s: repl5_inc_stop: protocol does not stop after %" NSPRIu64 " seconds\n",
				agmt_get_long_name(prp->agmt), timeout);
	}
	else
	{
		return_value = 0;
		slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name,
				"%s: repl5_inc_stop: protocol stopped after %d seconds\n",
				agmt_get_long_name(prp->agmt),
				PR_IntervalToSeconds(now-start));
	}
	if (slapi_is_loglevel_set(SLAPI_LOG_REPL)) {
		if (NULL == prp->replica_object) {
			slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name,
					"%s: repl5_inc_stop: protocol replica_object is NULL\n",
					agmt_get_long_name(prp->agmt));
		} else {
			Replica *replica;
			object_acquire(prp->replica_object);
			replica = object_get_data(prp->replica_object);
			if (NULL == replica) {
				slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name,
						"%s: repl5_inc_stop: replica is NULL\n",
						agmt_get_long_name(prp->agmt));
			} else {
				Object *ruv_obj = replica_get_ruv(replica);
				if (NULL == ruv_obj) {
					slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name,
							"%s: repl5_inc_stop: ruv_obj is NULL\n",
							agmt_get_long_name(prp->agmt));
				} else {
					RUV *ruv;
					object_acquire(ruv_obj);
					ruv = (RUV*)object_get_data (ruv_obj);
					if (NULL == ruv) {
						slapi_log_error(SLAPI_LOG_REPL, repl_plugin_name,
								"%s: repl5_inc_stop: ruv is NULL\n",
								agmt_get_long_name(prp->agmt));

					} else {
						ruv_dump(ruv, "Database RUV", NULL);
					}
					object_release(ruv_obj);
				}
			}
			object_release(prp->replica_object);
		}

	}
	return return_value;
}

static int
repl5_inc_status(Private_Repl_Protocol *prp)
{
	int return_value = 0;

	return return_value;
}

static void
repl5_inc_notify_update(Private_Repl_Protocol *prp)
{
	event_notify(prp, EVENT_TRIGGERING_CRITERIA_MET);
}

static void
repl5_inc_update_now(Private_Repl_Protocol *prp)
{
	event_notify(prp, EVENT_REPLICATE_NOW);
}

static void
repl5_inc_notify_agmt_changed(Private_Repl_Protocol *prp)
{
	event_notify(prp, EVENT_AGMT_CHANGED);
}

static void 
repl5_inc_notify_window_opened (Private_Repl_Protocol *prp)
{
    event_notify(prp, EVENT_WINDOW_OPENED);
}

static void 
repl5_inc_notify_window_closed (Private_Repl_Protocol *prp)
{
    event_notify(prp, EVENT_WINDOW_CLOSED);
}

Private_Repl_Protocol *
Repl_5_Inc_Protocol_new(Repl_Protocol *rp)
{
	repl5_inc_private *rip = NULL;
	Private_Repl_Protocol *prp = (Private_Repl_Protocol *)slapi_ch_malloc(sizeof(Private_Repl_Protocol));
	prp->delete = repl5_inc_delete;
	prp->run = repl5_inc_run;
	prp->stop = repl5_inc_stop;
	prp->status = repl5_inc_status;
	prp->notify_update = repl5_inc_notify_update;
	prp->notify_agmt_changed = repl5_inc_notify_agmt_changed;
    prp->notify_window_opened = repl5_inc_notify_window_opened;
    prp->notify_window_closed = repl5_inc_notify_window_closed;
	prp->update_now = repl5_inc_update_now;
	prp->replica_object = prot_get_replica_object(rp);
	if ((prp->lock = PR_NewLock()) == NULL)
	{
		goto loser;
	}
	if ((prp->cvar = PR_NewCondVar(prp->lock)) == NULL)
	{
		goto loser;
	}
	prp->stopped = 0;
	prp->terminate = 0;
	prp->eventbits = 0;
	prp->conn = prot_get_connection(rp);
	prp->agmt = prot_get_agreement(rp);
	prp->last_acquire_response_code = NSDS50_REPL_REPLICA_READY;
	rip = (void *)slapi_ch_malloc(sizeof(repl5_inc_private));
	rip->ruv = NULL;
	rip->backoff = NULL;
	rip->rp = rp;
	prp->private = (void *)rip;
    prp->replica_acquired = PR_FALSE;
	prp->repl50consumer = 0;
	prp->repl71consumer = 0;
	prp->repl90consumer = 0;
	return prp;
loser:
	repl5_inc_delete(&prp);
	return NULL;
}

static void
repl5_inc_backoff_expired(time_t timer_fire_time, void *arg)
{
	Private_Repl_Protocol *prp = (Private_Repl_Protocol *)arg;
	PR_ASSERT(NULL != prp);
	event_notify(prp, EVENT_BACKOFF_EXPIRED);
}

/*
 * Examine the update vector and determine our course of action.
 * There are 3 different possibilities, plus a catch-all error:
 * 1 - no update vector (ruv is NULL). The consumer's replica is
 *     pristine, so it needs to be initialized. Return
 *     EXAMINE_RUV_PRISTINE_REPLICA.
 * 2 - ruv is present, but its database generation ID doesn't
 *     match the local generation ID. This means that either
 *     the local replica must be reinitialized from the remote
 *     replica or vice-versa. Return
 *     EXAMINE_RUV_GENERATION_MISMATCH.
 * 3 - ruv is present, and we have all updates needed to bring
 *     the replica up to date using the incremental protocol.
 *     return EXAMINE_RUV_OK.
 * 4 - parameter error. Return EXAMINE_RUV_PARAM_ERROR
 */
static int
examine_update_vector(Private_Repl_Protocol *prp, RUV *remote_ruv)
{
	int return_value;

	PR_ASSERT(NULL != prp);
	if (NULL == prp)
	{
		return_value = EXAMINE_RUV_PARAM_ERROR;
	}
	else if (NULL == remote_ruv)
	{
		return_value = EXAMINE_RUV_PRISTINE_REPLICA;
	}
	else
	{
		char *local_gen = NULL;
		char *remote_gen = ruv_get_replica_generation(remote_ruv);
		Object *local_ruv_obj;
		RUV *local_ruv;
		Replica *replica;

		PR_ASSERT(NULL != prp->replica_object);
		replica = object_get_data(prp->replica_object);
		PR_ASSERT(NULL != replica);
		local_ruv_obj = replica_get_ruv (replica);
		if (NULL != local_ruv_obj)
		{
			local_ruv = (RUV*)object_get_data (local_ruv_obj);
			PR_ASSERT (local_ruv);
			local_gen = ruv_get_replica_generation(local_ruv);
			object_release (local_ruv_obj);
		}
		if (NULL == remote_gen || NULL == local_gen || strcmp(remote_gen, local_gen) != 0)
		{
			return_value = EXAMINE_RUV_GENERATION_MISMATCH;
		}
		else
		{
			return_value = EXAMINE_RUV_OK;
		}
		slapi_ch_free((void**)&remote_gen);
		slapi_ch_free((void**)&local_gen);
	}
	return return_value;
}

/* 
 * When we get an error from an LDAP operation, we call this
 * function to decide if we should just keep replaying
 * updates, or if we should stop, back off, and try again
 * later.
 * Returns PR_TRUE if we shoould keep going, PR_FALSE if
 * we should back off and try again later.
 *
 * In general, we keep going if the return code is consistent
 * with some sort of bug in URP that causes the consumer to
 * emit an error code that it shouldn't have, e.g. LDAP_ALREADY_EXISTS.
 * 
 * We stop if there's some indication that the server just completely
 * failed to process the operation, e.g. LDAP_OPERATIONS_ERROR.
 */
PRBool
ignore_error_and_keep_going(int error)
{
	int return_value = PR_FALSE;

	switch (error)
	{
	/* Cases where we keep going */
	case LDAP_SUCCESS:
	case LDAP_NO_SUCH_ATTRIBUTE:
	case LDAP_UNDEFINED_TYPE:
	case LDAP_CONSTRAINT_VIOLATION:
	case LDAP_TYPE_OR_VALUE_EXISTS:
	case LDAP_INVALID_SYNTAX:
	case LDAP_NO_SUCH_OBJECT:
	case LDAP_INVALID_DN_SYNTAX:
	case LDAP_IS_LEAF:
	case LDAP_INSUFFICIENT_ACCESS:
	case LDAP_NAMING_VIOLATION:
	case LDAP_OBJECT_CLASS_VIOLATION:
	case LDAP_NOT_ALLOWED_ON_NONLEAF:
	case LDAP_NOT_ALLOWED_ON_RDN:
	case LDAP_ALREADY_EXISTS:
	case LDAP_NO_OBJECT_CLASS_MODS:
		return_value = PR_TRUE;
		break;

	/* Cases where we stop and retry */
	case LDAP_OPERATIONS_ERROR:
	case LDAP_PROTOCOL_ERROR:
	case LDAP_TIMELIMIT_EXCEEDED:
	case LDAP_SIZELIMIT_EXCEEDED:
	case LDAP_STRONG_AUTH_NOT_SUPPORTED:
	case LDAP_STRONG_AUTH_REQUIRED:
	case LDAP_PARTIAL_RESULTS:
	case LDAP_REFERRAL:
	case LDAP_ADMINLIMIT_EXCEEDED:
	case LDAP_UNAVAILABLE_CRITICAL_EXTENSION:
	case LDAP_CONFIDENTIALITY_REQUIRED:
	case LDAP_SASL_BIND_IN_PROGRESS:
	case LDAP_INAPPROPRIATE_MATCHING:
	case LDAP_ALIAS_PROBLEM:
	case LDAP_ALIAS_DEREF_PROBLEM:
	case LDAP_INAPPROPRIATE_AUTH:
	case LDAP_INVALID_CREDENTIALS:
	case LDAP_BUSY:
	case LDAP_UNAVAILABLE:
	case LDAP_UNWILLING_TO_PERFORM:
	case LDAP_LOOP_DETECT:
	case LDAP_SORT_CONTROL_MISSING:
	case LDAP_INDEX_RANGE_ERROR:
	case LDAP_RESULTS_TOO_LARGE:
	case LDAP_AFFECTS_MULTIPLE_DSAS:
	case LDAP_OTHER:
	case LDAP_SERVER_DOWN:
	case LDAP_LOCAL_ERROR:
	case LDAP_ENCODING_ERROR:
	case LDAP_DECODING_ERROR:
	case LDAP_TIMEOUT:
	case LDAP_AUTH_UNKNOWN:
	case LDAP_FILTER_ERROR:
	case LDAP_USER_CANCELLED:
	case LDAP_PARAM_ERROR:
	case LDAP_NO_MEMORY:
	case LDAP_CONNECT_ERROR:
	case LDAP_NOT_SUPPORTED:
	case LDAP_CONTROL_NOT_FOUND:
	case LDAP_NO_RESULTS_RETURNED:
	case LDAP_MORE_RESULTS_TO_RETURN:
	case LDAP_CLIENT_LOOP:
	case LDAP_REFERRAL_LIMIT_EXCEEDED:
		return_value = PR_FALSE;
		break;
	}
	return return_value;
}

/* this function converts a state to its name - for debug output */
static const char* 
state2name (int state)
{
    switch (state)
    {
        case STATE_START:                       return "start";
        case STATE_WAIT_WINDOW_OPEN:            return "wait_for_window_to_open";
        case STATE_WAIT_CHANGES:                return "wait_for_changes";
        case STATE_READY_TO_ACQUIRE:            return "ready_to_acquire_replica";
        case STATE_BACKOFF_START:               return "start_backoff";
        case STATE_BACKOFF:                     return "backoff";
        case STATE_SENDING_UPDATES:             return "sending_updates";
        case STATE_STOP_FATAL_ERROR:            return "stop_fatal_error";
        case STATE_STOP_FATAL_ERROR_PART2:      return "stop_fatal_error";
        case STATE_STOP_NORMAL_TERMINATION:     return "stop_normal_termination";
        default:                                return "invalid_state";
    }
}

/* this function convert s an event to its name - for debug output */
static const char* 
event2name (int event)
{
    switch (event)
    {
        case EVENT_WINDOW_OPENED:           return "update_window_opened";
        case EVENT_WINDOW_CLOSED:           return "update_window_closed"; 
        case EVENT_TRIGGERING_CRITERIA_MET: return "data_modified";
        case EVENT_BACKOFF_EXPIRED:         return "backoff_timer_expired";  
        case EVENT_REPLICATE_NOW:           return "replicate_now";
        case EVENT_PROTOCOL_SHUTDOWN:       return "protocol_shutdown";
        case EVENT_AGMT_CHANGED:            return "agreement_changed";
        default:                            return "invalid_event";
    }
}

static const char*
op2string(int op)
{
	switch (op) {
	case SLAPI_OPERATION_ADD:
		return "add";
	case SLAPI_OPERATION_MODIFY:
		return "modify";
	case SLAPI_OPERATION_DELETE:
		return "delete";
	case SLAPI_OPERATION_MODRDN:
		return "rename";
	case SLAPI_OPERATION_EXTENDED:
		return "extended";
	}

	return "unknown";
}

void
repl5_set_backoff_min(Private_Repl_Protocol *prp, int min)
{
	Replica *replica;

	replica = (Replica *)object_get_data(prp->replica_object);
	if(replica){
		replica_set_backoff_min(replica, min);
	}
}

void
repl5_set_backoff_max(Private_Repl_Protocol *prp, int max)
{
	Replica *replica;

	replica = object_get_data(prp->replica_object);
	if(replica){
		replica_set_backoff_max(replica, max);
	}
}

int
repl5_get_backoff_min(Private_Repl_Protocol *prp)
{
	Replica *replica;

	replica = object_get_data(prp->replica_object);
	if(replica){
		return (int)replica_get_backoff_min(replica);
	}
	return PROTOCOL_BACKOFF_MINIMUM;
}

int
repl5_get_backoff_max(Private_Repl_Protocol *prp)
{
	Replica *replica;

	replica = object_get_data(prp->replica_object);
	if(replica){
		return (int)replica_get_backoff_max(replica);
	}
	return PROTOCOL_BACKOFF_MAXIMUM;
}