SQLite

** The pager is used to access a database disk file.  It implements
** atomic commit and rollback through the use of a journal file that
** is separate from the database file.  The pager also implements file
** locking to prevent two processes from writing the same database
** file simultaneously, or one process from reading the database while
** another is writing.
**
** @(#) $Id: pager.c,v 1.57 2002/11/10 23:32:57 drh Exp $
*/
#include "os.h"         /* Must be first to enable large file support */
#include "sqliteInt.h"
#include "pager.h"
#include <assert.h>
#include <string.h>

  Pager *pPager;                 /* The pager to which this page belongs */
  Pgno pgno;                     /* The page number for this page */
  PgHdr *pNextHash, *pPrevHash;  /* Hash collision chain for PgHdr.pgno */
  int nRef;                      /* Number of users of this page */
  PgHdr *pNextFree, *pPrevFree;  /* Freelist of pages where nRef==0 */
  PgHdr *pNextAll, *pPrevAll;    /* A list of all pages */
  PgHdr *pNextCkpt, *pPrevCkpt;  /* List of pages in the checkpoint journal */
  PgHdr *pSort;                  /* Next in list of pages to be written */
  u8 inJournal;                  /* TRUE if has been written to journal */
  u8 inCkpt;                     /* TRUE if written to the checkpoint journal */
  u8 dirty;                      /* TRUE if we need to write back changes */
  u8 alwaysRollback;             /* Disable dont_rollback() for this page */
  /* SQLITE_PAGE_SIZE bytes of page data follow this header */
  /* Pager.nExtra bytes of local data follow the page data */
};

** a reference to the page data.
*/
int sqlitepager_ref(void *pData){
  PgHdr *pPg = DATA_TO_PGHDR(pData);
  page_ref(pPg);
  return SQLITE_OK;
}

/*
** The parameters are pointers to the head of two sorted lists
** of page headers.  Merge these two lists together and return
** a single sorted list.  This routine forms the core of the 
** merge-sort algorithm that sorts dirty pages into accending
** order prior to writing them back to the disk.
**
** In the case of a tie, left sorts in front of right.
**
** Headers are sorted in order of ascending page number.
*/
static PgHdr *page_merge(PgHdr *pLeft, PgHdr *pRight){
  PgHdr sHead;
  PgHdr *pTail;
  pTail = &sHead;
  pTail->pSort = 0;
  while( pLeft && pRight ){
    if( pLeft->pgno<=pRight->pgno ){
      pTail->pSort = pLeft;
      pLeft = pLeft->pSort;
    }else{
      pTail->pSort = pRight;
      pRight = pRight->pSort;
    }
    pTail = pTail->pSort;
  }
  if( pLeft ){
    pTail->pSort = pLeft;
  }else if( pRight ){
    pTail->pSort = pRight;
  }
  return sHead.pSort;
}


/*
** Sync the journal and then write all free dirty pages to the database
** file.
**
** Writing all free dirty pages to the database after the sync is a
** non-obvious optimization.  fsync() is an expensive operation so we
** want to minimize the number ot times it is called. After an fsync() call,
** we are free to write dirty pages back to the database.  It is best
** to go ahead and write as many dirty pages as possible to minimize 
** the risk of having to do another fsync() later on.  Writing dirty
** free pages in this way was observed to make database operations go
** up to 10 times faster.
**
** If we are writing to temporary database, there is no need to preserve
** the integrity of the journal file, so we can save time and skip the
** fsync().
**
** This routine goes to the extra trouble of sorting all the dirty
** pages by their page number prior to writing them.  Tests show that
** writing pages in order by page number gives a modest speed improvement
** under Linux.  
*/
static int syncAllPages(Pager *pPager){
  PgHdr *pPg;
  PgHdr *pToWrite;
# define NSORT 28
  Pgno lastPgno;
  int i;
  PgHdr *apSorter[NSORT];
  int rc = SQLITE_OK;

  /* Sync the journal before modifying the main database
  ** (assuming there is a journal and it needs to be synced.)
  */
  if( pPager->needSync ){
    if( !pPager->tempFile ){
      rc = sqliteOsSync(&pPager->jfd);
      if( rc!=0 ) return rc;
    }
    pPager->needSync = 0;
  }

  /* Create a list of all dirty pages



  */
  pToWrite = 0;
  for(pPg=pPager->pFirst; pPg; pPg=pPg->pNextFree){
    if( pPg->dirty ){
      pPg->pSort = pToWrite;
      pToWrite = pPg;
    }
  }

  /* Sort the list of dirty pages into accending order by
  ** page number
  */
  for(i=0; i<NSORT; i++){
    apSorter[i] = 0;
  }
  while( pToWrite ){
    pPg = pToWrite;
    pToWrite = pPg->pSort;
    pPg->pSort = 0;
    for(i=0; i<NSORT-1; i++){
      if( apSorter[i]==0 ){
        apSorter[i] = pPg;
        break;
      }else{
        pPg = page_merge(apSorter[i], pPg);
        apSorter[i] = 0;
      }
    }
    if( i>=NSORT-1 ){
      apSorter[NSORT-1] = page_merge(apSorter[NSORT-1],pPg);
    }
  }
  pToWrite = 0;
  for(i=0; i<NSORT; i++){
    pToWrite = page_merge(apSorter[i], pToWrite);
  }

  /* Write all dirty pages back to the database and mark
  ** them all clean.
  */
  lastPgno = 0;
  for(pPg=pToWrite; pPg; pPg=pPg->pSort){
    if( lastPgno==0 || pPg->pgno!=lastPgno-1 ){
      sqliteOsSeek(&pPager->fd, (pPg->pgno-1)*SQLITE_PAGE_SIZE);
    }
    rc = sqliteOsWrite(&pPager->fd, PGHDR_TO_DATA(pPg), SQLITE_PAGE_SIZE);
    if( rc!=SQLITE_OK ) break;
    pPg->dirty = 0;
    lastPgno = pPg->pgno;

  }
  return rc;
}

/*
** Acquire a page.
**