lsm_env *pRealEnv = tdb_lsm_env();

  if( iLock==2 && eType==LSM_LOCK_EXCL && p->pDb->bNoRecovery ){
    return LSM_BUSY;
  }
  return pRealEnv->xLock(p->pReal, iLock, eType);
}











static int testEnvShmMap(lsm_file *pFile, int iRegion, int sz, void **pp){
  LsmFile *p = (LsmFile *)pFile;
  lsm_env *pRealEnv = tdb_lsm_env();
  return pRealEnv->xShmMap(p->pReal, iRegion, sz, pp);
}

................................................................................
  pDb->env.xSync = testEnvSync;
  pDb->env.xSectorSize = testEnvSectorSize;
  pDb->env.xRemap = testEnvRemap;
  pDb->env.xFileid = testEnvFileid;
  pDb->env.xClose = testEnvClose;
  pDb->env.xUnlink = testEnvUnlink;
  pDb->env.xLock = testEnvLock;

  pDb->env.xShmBarrier = testEnvShmBarrier;
  pDb->env.xShmMap = testEnvShmMap;
  pDb->env.xShmUnmap = testEnvShmUnmap;
  pDb->env.xSleep = testEnvSleep;

  rc = lsm_new(&pDb->env, &pDb->db);
  if( rc==LSM_OK ){

  lsm_env *pRealEnv = tdb_lsm_env();

  if( iLock==2 && eType==LSM_LOCK_EXCL && p->pDb->bNoRecovery ){
    return LSM_BUSY;
  }
  return pRealEnv->xLock(p->pReal, iLock, eType);
}

static int testEnvTestLock(lsm_file *pFile, int iLock, int nLock, int eType){
  LsmFile *p = (LsmFile *)pFile;
  lsm_env *pRealEnv = tdb_lsm_env();

  if( iLock==2 && eType==LSM_LOCK_EXCL && p->pDb->bNoRecovery ){
    return LSM_BUSY;
  }
  return pRealEnv->xTestLock(p->pReal, iLock, nLock, eType);
}

static int testEnvShmMap(lsm_file *pFile, int iRegion, int sz, void **pp){
  LsmFile *p = (LsmFile *)pFile;
  lsm_env *pRealEnv = tdb_lsm_env();
  return pRealEnv->xShmMap(p->pReal, iRegion, sz, pp);
}

................................................................................
  pDb->env.xSync = testEnvSync;
  pDb->env.xSectorSize = testEnvSectorSize;
  pDb->env.xRemap = testEnvRemap;
  pDb->env.xFileid = testEnvFileid;
  pDb->env.xClose = testEnvClose;
  pDb->env.xUnlink = testEnvUnlink;
  pDb->env.xLock = testEnvLock;
  pDb->env.xTestLock = testEnvTestLock;
  pDb->env.xShmBarrier = testEnvShmBarrier;
  pDb->env.xShmMap = testEnvShmMap;
  pDb->env.xShmUnmap = testEnvShmUnmap;
  pDb->env.xSleep = testEnvSleep;

  rc = lsm_new(&pDb->env, &pDb->db);
  if( rc==LSM_OK ){

  int (*xSync)(lsm_file *);
  int (*xSectorSize)(lsm_file *);
  int (*xRemap)(lsm_file *, lsm_i64, void **, lsm_i64*);
  int (*xFileid)(lsm_file *, void *pBuf, int *pnBuf);
  int (*xClose)(lsm_file *);
  int (*xUnlink)(lsm_env*, const char *);
  int (*xLock)(lsm_file*, int, int);

  int (*xShmMap)(lsm_file*, int, int, void **);
  void (*xShmBarrier)(void);
  int (*xShmUnmap)(lsm_file*, int);
  /****** memory allocation ****************************************/
  void *pMemCtx;
  void *(*xMalloc)(lsm_env*, int);            /* malloc(3) function */
  void *(*xRealloc)(lsm_env*, void *, int);   /* realloc(3) function */

  int (*xSync)(lsm_file *);
  int (*xSectorSize)(lsm_file *);
  int (*xRemap)(lsm_file *, lsm_i64, void **, lsm_i64*);
  int (*xFileid)(lsm_file *, void *pBuf, int *pnBuf);
  int (*xClose)(lsm_file *);
  int (*xUnlink)(lsm_env*, const char *);
  int (*xLock)(lsm_file*, int, int);
  int (*xTestLock)(lsm_file*, int, int, int);
  int (*xShmMap)(lsm_file*, int, int, void **);
  void (*xShmBarrier)(void);
  int (*xShmUnmap)(lsm_file*, int);
  /****** memory allocation ****************************************/
  void *pMemCtx;
  void *(*xMalloc)(lsm_env*, int);            /* malloc(3) function */
  void *(*xRealloc)(lsm_env*, void *, int);   /* realloc(3) function */

/* The number of available read locks. */
#define LSM_LOCK_NREADER   6

/* The number of available read-write client locks. */
#define LSM_LOCK_NRWCLIENT   16

/* Lock definitions. */
#define LSM_LOCK_DMS1         1
#define LSM_LOCK_DMS2         2

#define LSM_LOCK_WRITER       3
#define LSM_LOCK_WORKER       4
#define LSM_LOCK_CHECKPOINTER 5
#define LSM_LOCK_READER(i)    ((i) + LSM_LOCK_CHECKPOINTER + 1)
#define LSM_LOCK_RWCLIENT(i)  ((i) + LSM_LOCK_READER(LSM_LOCK_NREADER))

/*
** Hard limit on the number of free-list entries that may be stored in 
** a checkpoint (the remainder are stored as a system record in the LSM).
** See also LSM_CONFIG_MAX_FREELIST.
................................................................................
  u32 nWrite;                     /* Total number of pages written to disk */
};
#define LSM_INITIAL_SNAPSHOT_ID 11

/*
** Functions from file "lsm_ckpt.c".
*/
int lsmCheckpointWrite(lsm_db *, int, u32 *);
int lsmCheckpointLevels(lsm_db *, int, void **, int *);
int lsmCheckpointLoadLevels(lsm_db *pDb, void *pVal, int nVal);

int lsmCheckpointRecover(lsm_db *);
int lsmCheckpointDeserialize(lsm_db *, int, u32 *, Snapshot **);

int lsmCheckpointLoadWorker(lsm_db *pDb);
................................................................................
int lsmInfoArrayStructure(lsm_db *pDb, int bBlock, Pgno iFirst, char **pzOut);
int lsmInfoArrayPages(lsm_db *pDb, Pgno iFirst, char **pzOut);
int lsmConfigMmap(lsm_db *pDb, int *piParam);

int lsmEnvOpen(lsm_env *, const char *, int, lsm_file **);
int lsmEnvClose(lsm_env *pEnv, lsm_file *pFile);
int lsmEnvLock(lsm_env *pEnv, lsm_file *pFile, int iLock, int eLock);


int lsmEnvShmMap(lsm_env *, lsm_file *, int, int, void **); 
void lsmEnvShmBarrier(lsm_env *);
void lsmEnvShmUnmap(lsm_env *, lsm_file *, int);

void lsmEnvSleep(lsm_env *, int);

/* The number of available read locks. */
#define LSM_LOCK_NREADER   6

/* The number of available read-write client locks. */
#define LSM_LOCK_NRWCLIENT   16

/* Lock definitions. */
#define LSM_LOCK_DMS1         1   /* Serialize connect/disconnect ops */
#define LSM_LOCK_DMS2         2   /* Read-write connections */
#define LSM_LOCK_DMS3         3   /* Read-only connections */
#define LSM_LOCK_WRITER       4
#define LSM_LOCK_WORKER       5
#define LSM_LOCK_CHECKPOINTER 6
#define LSM_LOCK_READER(i)    ((i) + LSM_LOCK_CHECKPOINTER + 1)
#define LSM_LOCK_RWCLIENT(i)  ((i) + LSM_LOCK_READER(LSM_LOCK_NREADER))

/*
** Hard limit on the number of free-list entries that may be stored in 
** a checkpoint (the remainder are stored as a system record in the LSM).
** See also LSM_CONFIG_MAX_FREELIST.
................................................................................
  u32 nWrite;                     /* Total number of pages written to disk */
};
#define LSM_INITIAL_SNAPSHOT_ID 11

/*
** Functions from file "lsm_ckpt.c".
*/
int lsmCheckpointWrite(lsm_db *, int, int, u32 *);
int lsmCheckpointLevels(lsm_db *, int, void **, int *);
int lsmCheckpointLoadLevels(lsm_db *pDb, void *pVal, int nVal);

int lsmCheckpointRecover(lsm_db *);
int lsmCheckpointDeserialize(lsm_db *, int, u32 *, Snapshot **);

int lsmCheckpointLoadWorker(lsm_db *pDb);
................................................................................
int lsmInfoArrayStructure(lsm_db *pDb, int bBlock, Pgno iFirst, char **pzOut);
int lsmInfoArrayPages(lsm_db *pDb, Pgno iFirst, char **pzOut);
int lsmConfigMmap(lsm_db *pDb, int *piParam);

int lsmEnvOpen(lsm_env *, const char *, int, lsm_file **);
int lsmEnvClose(lsm_env *pEnv, lsm_file *pFile);
int lsmEnvLock(lsm_env *pEnv, lsm_file *pFile, int iLock, int eLock);
int lsmEnvTestLock(lsm_env *pEnv, lsm_file *pFile, int iLock, int nLock, int);

int lsmEnvShmMap(lsm_env *, lsm_file *, int, int, void **); 
void lsmEnvShmBarrier(lsm_env *);
void lsmEnvShmUnmap(lsm_env *, lsm_file *, int);

void lsmEnvSleep(lsm_env *, int);

  ShmHeader *pShm = pDb->pShmhdr;
  int nInt1;
  int nInt2;

  /* Must be holding the WORKER lock to do this. Or DMS2. */
  assert( 
      lsmShmAssertLock(pDb, LSM_LOCK_WORKER, LSM_LOCK_EXCL) 
   || lsmShmAssertLock(pDb, LSM_LOCK_DMS2, LSM_LOCK_EXCL) 
  );

  /* Check that the two snapshots match. If not, repair them. */
  nInt1 = pShm->aSnap1[CKPT_HDR_NCKPT];
  nInt2 = pShm->aSnap2[CKPT_HDR_NCKPT];
  if( nInt1!=nInt2 || memcmp(pShm->aSnap1, pShm->aSnap2, nInt2*sizeof(u32)) ){
    if( ckptChecksumOk(pShm->aSnap1) ){

  ShmHeader *pShm = pDb->pShmhdr;
  int nInt1;
  int nInt2;

  /* Must be holding the WORKER lock to do this. Or DMS2. */
  assert( 
      lsmShmAssertLock(pDb, LSM_LOCK_WORKER, LSM_LOCK_EXCL) 
   || lsmShmAssertLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_EXCL) 
  );

  /* Check that the two snapshots match. If not, repair them. */
  nInt1 = pShm->aSnap1[CKPT_HDR_NCKPT];
  nInt2 = pShm->aSnap2[CKPT_HDR_NCKPT];
  if( nInt1!=nInt2 || memcmp(pShm->aSnap1, pShm->aSnap2, nInt2*sizeof(u32)) ){
    if( ckptChecksumOk(pShm->aSnap1) ){

  return pEnv->xRemap(pFile, szMin, ppMap, pszMap);
}

int lsmEnvLock(lsm_env *pEnv, lsm_file *pFile, int iLock, int eLock){
  if( pFile==0 ) return LSM_OK;
  return pEnv->xLock(pFile, iLock, eLock);
}











int lsmEnvShmMap(
  lsm_env *pEnv, 
  lsm_file *pFile, 
  int iChunk, 
  int sz, 
  void **ppOut
................................................................................
*/
int lsmFsTruncateLog(FileSystem *pFS, i64 nByte){
  if( pFS->fdLog==0 ) return LSM_OK;
  return lsmEnvTruncate(pFS->pEnv, pFS->fdLog, nByte);
}

/*
** Truncate the log file to nByte bytes in size.
*/
int lsmFsTruncateDb(FileSystem *pFS, i64 nByte){
  if( pFS->fdDb==0 ) return LSM_OK;
  return lsmEnvTruncate(pFS->pEnv, pFS->fdDb, nByte);
}

/*

  return pEnv->xRemap(pFile, szMin, ppMap, pszMap);
}

int lsmEnvLock(lsm_env *pEnv, lsm_file *pFile, int iLock, int eLock){
  if( pFile==0 ) return LSM_OK;
  return pEnv->xLock(pFile, iLock, eLock);
}

int lsmEnvTestLock(
  lsm_env *pEnv, 
  lsm_file *pFile, 
  int iLock, 
  int nLock, 
  int eLock
){
  return pEnv->xTestLock(pFile, iLock, nLock, eLock);
}

int lsmEnvShmMap(
  lsm_env *pEnv, 
  lsm_file *pFile, 
  int iChunk, 
  int sz, 
  void **ppOut
................................................................................
*/
int lsmFsTruncateLog(FileSystem *pFS, i64 nByte){
  if( pFS->fdLog==0 ) return LSM_OK;
  return lsmEnvTruncate(pFS->pEnv, pFS->fdLog, nByte);
}

/*
** Truncate the db file to nByte bytes in size.
*/
int lsmFsTruncateDb(FileSystem *pFS, i64 nByte){
  if( pFS->fdDb==0 ) return LSM_OK;
  return lsmEnvTruncate(pFS->pEnv, pFS->fdDb, nByte);
}

/*

** to as small a size as possible without truncating away any blocks that
** contain data.
*/
static int dbTruncateFile(lsm_db *pDb){
  int rc;

  assert( pDb->pWorker==0 );
  assert( lsmShmAssertLock(pDb, LSM_LOCK_DMS2, LSM_LOCK_EXCL) );
  rc = lsmCheckpointLoadWorker(pDb);

  if( rc==LSM_OK ){
    DbTruncateCtx ctx;

    /* Walk the database free-block-list in reverse order. Set ctx.nBlock
    ** to the block number of the last block in the database that actually
................................................................................
  pDb->pWorker = 0;
  return rc;
}

static void doDbDisconnect(lsm_db *pDb){
  int rc;




  /* Block for an exclusive lock on DMS1. This lock serializes all calls
  ** to doDbConnect() and doDbDisconnect() across all processes.  */
  rc = lsmShmLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_EXCL, 1);
  if( rc==LSM_OK ){

    /* Try an exclusive lock on DMS2. If successful, this is the last
    ** connection to the database. In this case flush the contents of the
    ** in-memory tree to disk and write a checkpoint.  */
    rc = lsmShmLock(pDb, LSM_LOCK_DMS2, LSM_LOCK_EXCL, 0);
    if( rc==LSM_OK ){


      /* Flush the in-memory tree, if required. If there is data to flush,
      ** this will create a new client snapshot in Database.pClient. The
      ** checkpoint (serialization) of this snapshot may be written to disk
      ** by the following block.  
      **
      ** There is no need to mess around with WRITER locks or anything at
      ** this point. The lock on DMS2 guarantees that pDb has exclusive
      ** access to the db at this point.




      */
      rc = lsmTreeLoadHeader(pDb, 0);
      if( rc==LSM_OK && (lsmTreeHasOld(pDb) || lsmTreeSize(pDb)>0) ){
        rc = lsmFlushTreeToDisk(pDb);
      }












      /* Write a checkpoint to disk. */
      if( rc==LSM_OK ){
        rc = lsmCheckpointWrite(pDb, 1, 0);
      }

      /* If the checkpoint was written successfully, delete the log file
      ** and, if possible, truncate the database file.  */
      if( rc==LSM_OK ){
        Database *p = pDb->pDatabase;

        dbTruncateFile(pDb);
        lsmFsCloseAndDeleteLog(pDb->pFS);

        if( p->pFile && p->bMultiProc ) lsmEnvShmUnmap(pDb->pEnv, p->pFile, 1);
      }
    }
  }



  if( pDb->iRwclient>=0 ){
    lsmShmLock(pDb, LSM_LOCK_RWCLIENT(pDb->iRwclient), LSM_LOCK_UNLOCK, 0);
  }

  lsmShmLock(pDb, LSM_LOCK_DMS2, LSM_LOCK_UNLOCK, 0);
  lsmShmLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_UNLOCK, 0);

  pDb->pShmhdr = 0;
}

static int doDbConnect(lsm_db *pDb){
  const int nUsMax = 100000;      /* Max value for nUs */
  int nUs = 1000;                 /* us to wait between DMS1 attempts */
  int rc;

  /* Obtain a pointer to the shared-memory header */
  assert( pDb->pShmhdr==0 );

  rc = lsmShmCacheChunks(pDb, 1);
  if( rc!=LSM_OK ) return rc;
  pDb->pShmhdr = (ShmHeader *)pDb->apShm[0];

  /* Block for an exclusive lock on DMS1. This lock serializes all calls
  ** to doDbConnect() and doDbDisconnect() across all processes.  */
  while( 1 ){
................................................................................
    if( nUs>nUsMax ) nUs = nUsMax;
  }
  if( rc!=LSM_OK ){
    pDb->pShmhdr = 0;
    return rc;
  }

  /* Try an exclusive lock on DMS2. If successful, this is the first and 
  ** only connection to the database. In this case initialize the 
  ** shared-memory and run log file recovery.  */

  rc = lsmShmLock(pDb, LSM_LOCK_DMS2, LSM_LOCK_EXCL, 0);
  if( rc==LSM_OK ){
    memset(pDb->pShmhdr, 0, sizeof(ShmHeader));
    rc = lsmCheckpointRecover(pDb);
    if( rc==LSM_OK ){
      rc = lsmLogRecover(pDb);
    }





  }else if( rc==LSM_BUSY ){
    rc = LSM_OK;
  }

  /* Take a shared lock on DMS2. In multi-process mode this lock "cannot" 
  ** fail, as connections may only hold an exclusive lock on DMS2 if they 
  ** first hold an exclusive lock on DMS1. And this connection is currently 
................................................................................
  if( rc==LSM_OK ){
    rc = lsmShmLock(pDb, LSM_LOCK_DMS2, LSM_LOCK_SHARED, 0);
  }

  /* If anything went wrong, unlock DMS2. Otherwise, try to take an exclusive
  ** lock on one of the LSM_LOCK_RWCLIENT() locks. Unlock DMS1 in any case. */
  if( rc!=LSM_OK ){
    lsmShmLock(pDb, LSM_LOCK_DMS2, LSM_LOCK_UNLOCK, 0);
    pDb->pShmhdr = 0;
  }else{
    int i;
    for(i=0; i<LSM_LOCK_NRWCLIENT; i++){
      int rc2 = lsmShmLock(pDb, LSM_LOCK_RWCLIENT(i), LSM_LOCK_EXCL, 0);
      if( rc2==LSM_OK ) pDb->iRwclient = i;
      if( rc2!=LSM_BUSY ){
................................................................................
** database itself.
**
** The WORKER lock must not be held when this is called. This is because
** this function may indirectly call fsync(). And the WORKER lock should
** not be held that long (in case it is required by a client flushing an
** in-memory tree to disk).
*/
int lsmCheckpointWrite(lsm_db *pDb, int bTruncate, u32 *pnWrite){
  int rc;                         /* Return Code */
  u32 nWrite = 0;

  assert( pDb->pWorker==0 );
  assert( 1 || pDb->pClient==0 );
  assert( lsmShmAssertLock(pDb, LSM_LOCK_WORKER, LSM_LOCK_UNLOCK) );

................................................................................
      );
#endif
    }

    if( rc==LSM_OK && bTruncate ){
      rc = lsmFsTruncateDb(pDb->pFS, (i64)nBlock*lsmFsBlockSize(pDb->pFS));
    }



  }

  lsmShmLock(pDb, LSM_LOCK_CHECKPOINTER, LSM_LOCK_UNLOCK, 0);
  if( pnWrite && rc==LSM_OK ) *pnWrite = nWrite;
  return rc;
}

................................................................................
int lsmBeginRoTrans(lsm_db *db){
  int rc = LSM_OK;

  assert( db->bReadonly && db->pShmhdr==0 );
  assert( db->iReader<0 );

  if( db->bRoTrans==0 ){
    if( 1 ){










      rc = lsmShmLock(db, LSM_LOCK_CHECKPOINTER, LSM_LOCK_SHARED, 0);


      if( rc==LSM_OK ){
        db->bRoTrans = 1;
        rc = lsmShmCacheChunks(db, 1);
        if( rc==LSM_OK ){
          db->pShmhdr = (ShmHeader *)db->apShm[0];
          memset(db->pShmhdr, 0, sizeof(ShmHeader));
          rc = lsmCheckpointRecover(db);
          if( rc==LSM_OK ){
            rc = lsmLogRecover(db);
          }
        }
      }
    }else{
      /* lock(DMS2, SHARED) etc. */








    }

    if( rc==LSM_OK ){
      rc = lsmBeginReadTrans(db);
    }
  }

................................................................................
static int lockSharedFile(lsm_env *pEnv, Database *p, int iLock, int eOp){
  int rc = LSM_OK;
  if( p->bMultiProc ){
    rc = lsmEnvLock(pEnv, p->pFile, iLock, eOp);
  }
  return rc;
}












































/*
** Attempt to obtain the lock identified by the iLock and bExcl parameters.
** If successful, return LSM_OK. If the lock cannot be obtained because 
** there exists some other conflicting lock, return LSM_BUSY. If some other
** error occurs, return an LSM error code.
**
................................................................................

int lsm_checkpoint(lsm_db *pDb, int *pnKB){
  int rc;                         /* Return code */
  u32 nWrite = 0;                 /* Number of pages checkpointed */

  /* Attempt the checkpoint. If successful, nWrite is set to the number of
  ** pages written between this and the previous checkpoint.  */
  rc = lsmCheckpointWrite(pDb, 0, &nWrite);

  /* If required, calculate the output variable (KB of data checkpointed). 
  ** Set it to zero if an error occured.  */
  if( pnKB ){
    int nKB = 0;
    if( rc==LSM_OK && nWrite ){
      nKB = (((i64)nWrite * lsmFsPageSize(pDb->pFS)) + 1023) / 1024;

** to as small a size as possible without truncating away any blocks that
** contain data.
*/
static int dbTruncateFile(lsm_db *pDb){
  int rc;

  assert( pDb->pWorker==0 );
  assert( lsmShmAssertLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_EXCL) );
  rc = lsmCheckpointLoadWorker(pDb);

  if( rc==LSM_OK ){
    DbTruncateCtx ctx;

    /* Walk the database free-block-list in reverse order. Set ctx.nBlock
    ** to the block number of the last block in the database that actually
................................................................................
  pDb->pWorker = 0;
  return rc;
}

static void doDbDisconnect(lsm_db *pDb){
  int rc;

  if( pDb->bReadonly ){
    lsmShmLock(pDb, LSM_LOCK_DMS3, LSM_LOCK_UNLOCK, 0);
  }else{
    /* Block for an exclusive lock on DMS1. This lock serializes all calls
    ** to doDbConnect() and doDbDisconnect() across all processes.  */
    rc = lsmShmLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_EXCL, 1);
    if( rc==LSM_OK ){

      /* Try an exclusive lock on DMS2. If successful, this is the last
      ** connection to the database. In this case flush the contents of the
      ** in-memory tree to disk and write a checkpoint.  */
      rc = lsmShmTestLock(pDb, LSM_LOCK_DMS2, 1, LSM_LOCK_EXCL);
      if( rc==LSM_OK ){
        int bReadonly = 0;        /* True if there exist read-only conns. */

        /* Flush the in-memory tree, if required. If there is data to flush,
        ** this will create a new client snapshot in Database.pClient. The
        ** checkpoint (serialization) of this snapshot may be written to disk
        ** by the following block.  
        **



        ** There is no need to take a WRITER lock here. That there are no 
        ** other locks on DMS2 guarantees that there are no other read-write
        ** connections at this time (and the lock on DMS1 guarantees that
        ** no new ones may appear).
        */
        rc = lsmTreeLoadHeader(pDb, 0);
        if( rc==LSM_OK && (lsmTreeHasOld(pDb) || lsmTreeSize(pDb)>0) ){
          rc = lsmFlushTreeToDisk(pDb);
        }

        /* Now check if there are any read-only connections. If there are,
        ** then do not truncate the db file or unlink the shared-memory 
        ** region.  */
        if( rc==LSM_OK ){
          rc = lsmShmTestLock(pDb, LSM_LOCK_DMS3, 1, LSM_LOCK_EXCL);
          if( rc==LSM_BUSY ){
            bReadonly = 1;
            rc = LSM_OK;
          }
        }

        /* Write a checkpoint to disk. */
        if( rc==LSM_OK ){
          rc = lsmCheckpointWrite(pDb, (bReadonly==0), 1, 0);
        }

        /* If the checkpoint was written successfully, delete the log file
        ** and, if possible, truncate the database file.  */
        if( rc==LSM_OK ){
          Database *p = pDb->pDatabase;
          if( bReadonly==0 ){
            dbTruncateFile(pDb);

            if( p->pFile && p->bMultiProc ){
              lsmEnvShmUnmap(pDb->pEnv, p->pFile, 1);
            }
          }
        }
      }
    }

    if( pDb->iRwclient>=0 ){
      lsmShmLock(pDb, LSM_LOCK_RWCLIENT(pDb->iRwclient), LSM_LOCK_UNLOCK, 0);
    }

    lsmShmLock(pDb, LSM_LOCK_DMS2, LSM_LOCK_UNLOCK, 0);
    lsmShmLock(pDb, LSM_LOCK_DMS1, LSM_LOCK_UNLOCK, 0);
  }
  pDb->pShmhdr = 0;
}

static int doDbConnect(lsm_db *pDb){
  const int nUsMax = 100000;      /* Max value for nUs */
  int nUs = 1000;                 /* us to wait between DMS1 attempts */
  int rc;

  /* Obtain a pointer to the shared-memory header */
  assert( pDb->pShmhdr==0 );
  assert( pDb->bReadonly==0 );
  rc = lsmShmCacheChunks(pDb, 1);
  if( rc!=LSM_OK ) return rc;
  pDb->pShmhdr = (ShmHeader *)pDb->apShm[0];

  /* Block for an exclusive lock on DMS1. This lock serializes all calls
  ** to doDbConnect() and doDbDisconnect() across all processes.  */
  while( 1 ){
................................................................................
    if( nUs>nUsMax ) nUs = nUsMax;
  }
  if( rc!=LSM_OK ){
    pDb->pShmhdr = 0;
    return rc;
  }

  /* Try an exclusive lock on DMS2/DMS3. If successful, this is the first 
  ** and only connection to the database. In this case initialize the 
  ** shared-memory and run log file recovery.  */
  assert( LSM_LOCK_DMS3==1+LSM_LOCK_DMS2 );
  rc = lsmShmTestLock(pDb, LSM_LOCK_DMS2, 2, LSM_LOCK_EXCL);
  if( rc==LSM_OK ){
    memset(pDb->pShmhdr, 0, sizeof(ShmHeader));
    rc = lsmCheckpointRecover(pDb);
    if( rc==LSM_OK ){
      rc = lsmLogRecover(pDb);
    }
    if( rc==LSM_OK ){
      ShmHeader *pShm = pDb->pShmhdr;
      pShm->aReader[0].iLsmId = lsmCheckpointId(pShm->aSnap1, 0);
      pShm->aReader[0].iTreeId = pDb->treehdr.iUsedShmid;
    }
  }else if( rc==LSM_BUSY ){
    rc = LSM_OK;
  }

  /* Take a shared lock on DMS2. In multi-process mode this lock "cannot" 
  ** fail, as connections may only hold an exclusive lock on DMS2 if they 
  ** first hold an exclusive lock on DMS1. And this connection is currently 
................................................................................
  if( rc==LSM_OK ){
    rc = lsmShmLock(pDb, LSM_LOCK_DMS2, LSM_LOCK_SHARED, 0);
  }

  /* If anything went wrong, unlock DMS2. Otherwise, try to take an exclusive
  ** lock on one of the LSM_LOCK_RWCLIENT() locks. Unlock DMS1 in any case. */
  if( rc!=LSM_OK ){

    pDb->pShmhdr = 0;
  }else{
    int i;
    for(i=0; i<LSM_LOCK_NRWCLIENT; i++){
      int rc2 = lsmShmLock(pDb, LSM_LOCK_RWCLIENT(i), LSM_LOCK_EXCL, 0);
      if( rc2==LSM_OK ) pDb->iRwclient = i;
      if( rc2!=LSM_BUSY ){
................................................................................
** database itself.
**
** The WORKER lock must not be held when this is called. This is because
** this function may indirectly call fsync(). And the WORKER lock should
** not be held that long (in case it is required by a client flushing an
** in-memory tree to disk).
*/
int lsmCheckpointWrite(lsm_db *pDb, int bTruncate, int bDellog, u32 *pnWrite){
  int rc;                         /* Return Code */
  u32 nWrite = 0;

  assert( pDb->pWorker==0 );
  assert( 1 || pDb->pClient==0 );
  assert( lsmShmAssertLock(pDb, LSM_LOCK_WORKER, LSM_LOCK_UNLOCK) );

................................................................................
      );
#endif
    }

    if( rc==LSM_OK && bTruncate ){
      rc = lsmFsTruncateDb(pDb->pFS, (i64)nBlock*lsmFsBlockSize(pDb->pFS));
    }
    if( rc==LSM_OK && bDellog ){
      lsmFsCloseAndDeleteLog(pDb->pFS);
    }
  }

  lsmShmLock(pDb, LSM_LOCK_CHECKPOINTER, LSM_LOCK_UNLOCK, 0);
  if( pnWrite && rc==LSM_OK ) *pnWrite = nWrite;
  return rc;
}

................................................................................
int lsmBeginRoTrans(lsm_db *db){
  int rc = LSM_OK;

  assert( db->bReadonly && db->pShmhdr==0 );
  assert( db->iReader<0 );

  if( db->bRoTrans==0 ){


    /* Attempt a shared-lock on DMS1. */
    rc = lsmShmLock(db, LSM_LOCK_DMS1, LSM_LOCK_SHARED, 0);
    if( rc!=LSM_OK ) return rc;

    rc = lsmShmTestLock(
        db, LSM_LOCK_RWCLIENT(0), LSM_LOCK_NREADER, LSM_LOCK_SHARED
    );
    if( rc==LSM_OK ){
      /* System is not live */
      rc = lsmShmLock(db, LSM_LOCK_CHECKPOINTER, LSM_LOCK_SHARED, 0);
      lsmShmLock(db, LSM_LOCK_DMS1, LSM_LOCK_UNLOCK, 0);

      if( rc==LSM_OK ){
        db->bRoTrans = 1;
        rc = lsmShmCacheChunks(db, 1);
        if( rc==LSM_OK ){
          db->pShmhdr = (ShmHeader *)db->apShm[0];
          memset(db->pShmhdr, 0, sizeof(ShmHeader));
          rc = lsmCheckpointRecover(db);
          if( rc==LSM_OK ){
            rc = lsmLogRecover(db);
          }
        }
      }
    }else if( rc==LSM_BUSY ){
      /* System is live! */
      rc = lsmShmLock(db, LSM_LOCK_DMS3, LSM_LOCK_SHARED, 0);
      lsmShmLock(db, LSM_LOCK_DMS1, LSM_LOCK_UNLOCK, 0);
      if( rc==LSM_OK ){
        rc = lsmShmCacheChunks(db, 1);
        if( rc==LSM_OK ){
          db->pShmhdr = (ShmHeader *)db->apShm[0];
        }
      }
    }

    if( rc==LSM_OK ){
      rc = lsmBeginReadTrans(db);
    }
  }

................................................................................
static int lockSharedFile(lsm_env *pEnv, Database *p, int iLock, int eOp){
  int rc = LSM_OK;
  if( p->bMultiProc ){
    rc = lsmEnvLock(pEnv, p->pFile, iLock, eOp);
  }
  return rc;
}

/*
** Test if it would be possible for connection db to obtain a lock of type
** eType on the nLock locks starting at iLock. If so, return LSM_OK. If it
** would not be possible to obtain the lock due to a lock held by another
** connection, return LSM_BUSY. If an IO or other error occurs (i.e. in the 
** lsm_env.xTestLock function), return some other LSM error code.
**
** Note that this function never actually locks the database - it merely
** queries the system to see if there exists a lock that would prevent
** it from doing so.
*/
int lsmShmTestLock(
  lsm_db *db,
  int iLock,
  int nLock,
  int eOp
){
  int rc = LSM_OK;
  lsm_db *pIter;
  Database *p = db->pDatabase;
  int i;
  u64 mask = 0;

  for(i=iLock; i<(iLock+nLock); i++){
    mask |= ((u64)1 << (iLock-1));
    if( eOp==LSM_LOCK_EXCL ) mask |= ((u64)1 << (iLock+32-1));
  }

  lsmMutexEnter(db->pEnv, p->pClientMutex);
  for(pIter=p->pConn; pIter; pIter=pIter->pNext){
    if( pIter!=db && (pIter->mLock & mask) ) break;
  }

  if( pIter ){
    rc = LSM_BUSY;
  }else if( p->bMultiProc ){
    rc = lsmEnvTestLock(db->pEnv, p->pFile, iLock, nLock, eOp);
  }

  lsmMutexLeave(db->pEnv, p->pClientMutex);
  return rc;
}

/*
** Attempt to obtain the lock identified by the iLock and bExcl parameters.
** If successful, return LSM_OK. If the lock cannot be obtained because 
** there exists some other conflicting lock, return LSM_BUSY. If some other
** error occurs, return an LSM error code.
**
................................................................................

int lsm_checkpoint(lsm_db *pDb, int *pnKB){
  int rc;                         /* Return code */
  u32 nWrite = 0;                 /* Number of pages checkpointed */

  /* Attempt the checkpoint. If successful, nWrite is set to the number of
  ** pages written between this and the previous checkpoint.  */
  rc = lsmCheckpointWrite(pDb, 0, 0, &nWrite);

  /* If required, calculate the output variable (KB of data checkpointed). 
  ** Set it to zero if an error occured.  */
  if( pnKB ){
    int nKB = 0;
    if( rc==LSM_OK && nWrite ){
      nKB = (((i64)nWrite * lsmFsPageSize(pDb->pFS)) + 1023) / 1024;

    }else{
      rc = LSM_IOERR_BKPT;
    }
  }

  return rc;
}




























int lsmPosixOsShmMap(lsm_file *pFile, int iChunk, int sz, void **ppShm){
  PosixFile *p = (PosixFile *)pFile;

  *ppShm = 0;
  assert( sz==LSM_SHM_CHUNK_SIZE );
  if( iChunk>=p->nShm ){
................................................................................
    lsmPosixOsSync,          /* xSync */
    lsmPosixOsSectorSize,    /* xSectorSize */
    lsmPosixOsRemap,         /* xRemap */
    lsmPosixOsFileid,        /* xFileid */
    lsmPosixOsClose,         /* xClose */
    lsmPosixOsUnlink,        /* xUnlink */
    lsmPosixOsLock,          /* xLock */

    lsmPosixOsShmMap,        /* xShmMap */
    lsmPosixOsShmBarrier,    /* xShmBarrier */
    lsmPosixOsShmUnmap,      /* xShmUnmap */
    /***** memory allocation *********/
    0,                       /* pMemCtx */
    lsmPosixOsMalloc,        /* xMalloc */
    lsmPosixOsRealloc,       /* xRealloc */

    }else{
      rc = LSM_IOERR_BKPT;
    }
  }

  return rc;
}

int lsmPosixOsTestLock(lsm_file *pFile, int iLock, int nLock, int eType){
  int rc = LSM_OK;
  PosixFile *p = (PosixFile *)pFile;
  static const short aType[3] = { 0, F_RDLCK, F_WRLCK };
  struct flock lock;

  assert( eType==LSM_LOCK_SHARED || eType==LSM_LOCK_EXCL );
  assert( aType[LSM_LOCK_SHARED]==F_RDLCK );
  assert( aType[LSM_LOCK_EXCL]==F_WRLCK );
  assert( eType>=0 && eType<array_size(aType) );
  assert( iLock>0 && iLock<=32 );

  memset(&lock, 0, sizeof(lock));
  lock.l_whence = SEEK_SET;
  lock.l_len = nLock;
  lock.l_type = aType[eType];
  lock.l_start = (4096-iLock);

  if( fcntl(p->fd, F_GETLK, &lock) ){
    rc = LSM_IOERR_BKPT;
  }else if( lock.l_type!=F_UNLCK ){
    rc = LSM_BUSY;
  }

  return rc;
}

int lsmPosixOsShmMap(lsm_file *pFile, int iChunk, int sz, void **ppShm){
  PosixFile *p = (PosixFile *)pFile;

  *ppShm = 0;
  assert( sz==LSM_SHM_CHUNK_SIZE );
  if( iChunk>=p->nShm ){
................................................................................
    lsmPosixOsSync,          /* xSync */
    lsmPosixOsSectorSize,    /* xSectorSize */
    lsmPosixOsRemap,         /* xRemap */
    lsmPosixOsFileid,        /* xFileid */
    lsmPosixOsClose,         /* xClose */
    lsmPosixOsUnlink,        /* xUnlink */
    lsmPosixOsLock,          /* xLock */
    lsmPosixOsTestLock,      /* xTestLock */
    lsmPosixOsShmMap,        /* xShmMap */
    lsmPosixOsShmBarrier,    /* xShmBarrier */
    lsmPosixOsShmUnmap,      /* xShmUnmap */
    /***** memory allocation *********/
    0,                       /* pMemCtx */
    lsmPosixOsMalloc,        /* xMalloc */
    lsmPosixOsRealloc,       /* xRealloc */

  forcedelete $file
  lsm_open db $file
  db write a alpha
  db write b bravo
  db write c charlie
  db close
}












#-------------------------------------------------------------------------
# When the database system is shut down (i.e. when the last connection
# disconnects), an attempt is made to truncate the database file to the
# minimum number of blocks required.
# 
# This test case checks that this process does not actually cause the
................................................................................
} {}
do_test 1.3 {
  expr [file size test.db] < (64*1024)
} 1

#-------------------------------------------------------------------------
# Test that if an attempt is made to open a read-write connection to a 
# database that the client does not have permission to write to is attempted
# an error is reported. In order to open a read-write connection to a
# database, the client requires:
#
#   * read-write access to the db file,
#   * read-write access to the log file,
#   * for multi-process mode, read-write access to the shm file.
#
# In the above, "read-write access" includes the ability to create the db,
# log or shm file if it does not exist.
#
# These tests verify that the lsm_open() command returns LSM_IOERR. At some
# point in the future this will be improved. Likely when sqlite4 level tests 
# for opening read-only databases are added.
# 

foreach {tn filename setup} {

  1 test.dir/test.db {
    # Create a directory "test.dir".
    forcedelete test.dir
    file mkdir test.dir

................................................................................
    create_abc_db test.dir/test.db
  
    # Now make test.dir read-only.
    file attr test.dir -perm r-xr-xr-x
  }

} {

  do_test 2.$tn.1 {
    eval $setup
    set rc [catch {lsm_open db $filename} msg]
    list $rc $msg
  } {1 {error in lsm_open() - 10}}

  do_test 2.$tn.2 {
    eval $setup
    lsm_open db $filename {readonly 1}
    set res [list [db_fetch db a] [db_fetch db b] [db_fetch db c]]
    db close
    set res
  } {alpha bravo charlie}

}













































finish_test

  forcedelete $file
  lsm_open db $file
  db write a alpha
  db write b bravo
  db write c charlie
  db close
}

proc create_abc_log {file} {
  forcedelete $file ${file}-2
  lsm_open db ${file}-2
  db write a alpha
  db write b bravo
  db write c charlie
  file copy ${file}-2 $file
  file copy ${file}-2-log $file-log
  db close
}

#-------------------------------------------------------------------------
# When the database system is shut down (i.e. when the last connection
# disconnects), an attempt is made to truncate the database file to the
# minimum number of blocks required.
# 
# This test case checks that this process does not actually cause the
................................................................................
} {}
do_test 1.3 {
  expr [file size test.db] < (64*1024)
} 1

#-------------------------------------------------------------------------
# Test that if an attempt is made to open a read-write connection to a 
# non-live database that the client does not have permission to write to is
# attempted an error is reported. In order to open a read-write connection 
# to a database, the client requires:
#
#   * read-write access to the db file,
#   * read-write access to the log file,
#   * for multi-process mode, read-write access to the shm file.
#
# In the above, "read-write access" includes the ability to create the db,
# log or shm file if it does not exist.
#
# These tests verify that the lsm_open() command returns LSM_IOERR. At some
# point in the future this will be improved. Likely when sqlite4 level tests 
# for opening read-only databases are added.
# 

foreach {tn filename setup} {

  1 test.dir/test.db {
    # Create a directory "test.dir".
    forcedelete test.dir
    file mkdir test.dir

................................................................................
    create_abc_db test.dir/test.db
  
    # Now make test.dir read-only.
    file attr test.dir -perm r-xr-xr-x
  }

} {

  do_test 2.$tn.1 {
    eval $setup
    set rc [catch {lsm_open db $filename} msg]
    list $rc $msg
  } {1 {error in lsm_open() - 10}}

  do_test 2.$tn.2 {
    eval $setup
    lsm_open db $filename {readonly 1}
    set res [list [db_fetch db a] [db_fetch db b] [db_fetch db c]]
    db close
    set res
  } {alpha bravo charlie}
}

#-------------------------------------------------------------------------
# Try having a read-only connection connect to a non-live system where the
# log file contains content. In this scenario the read-only client must 
# read the contents from the log file at the start of each read-transaction. 
#
do_test 3.1 {
  create_abc_log test.db
  list [file size test.db] [file size test.db-log]
} {0 56}
do_test 3.2 {
  lsm_open db $filename {readonly 1}
  set res [list [db_fetch db a] [db_fetch db b] [db_fetch db c]]
  db close
  set res
} {alpha bravo charlie}
do_test 3.3 {
  list [file size test.db] [file size test.db-log]
} {0 56}

# Now make the same db live and check the read-only connection can still
# read it.
do_test 3.4 { file exists test.db-shm } 0
do_test 3.5 { 
  lsm_open db_rw test.db
  file exists test.db-shm
} 1
do_test 3.6 {
  lsm_open db test.db {readonly 1}
  list [db_fetch db a] [db_fetch db b] [db_fetch db c]
} {alpha bravo charlie}

# Close the read-write connection. This should cause a checkpoint and delete
# the log file, even though the system remains live.
do_test 3.7 { 
  db_rw close
  list [file exists test.db-log] [file exists test.db-shm]
} {0 1}

# Now close the read-only connection. The system is now non-live, but the 
# *-shm remains in the file-system (the readonly connection cannot unlink it).
do_test 3.8 { 
  db close
  list [file exists test.db-log] [file exists test.db-shm]
} {0 1}

finish_test