Avoid using posix_fallocate() in WAL mode, as it is not supported by all file-systems.

FossilOrigin-Name: 9c7523dabf4aee609287732ce787c9b9a9087e7f

1 files changed, 65 insertions, 44 deletions

diff --git a/src/os_unix.c b/src/os_unix.c
index 7448afe4c..1200ff352 100644
--- a/src/os_unix.c
+++ b/src/os_unix.c
@@ -3194,46 +3194,59 @@ static int unixRead(
 }
 
 /*
-** Seek to the offset in id->offset then read cnt bytes into pBuf.
-** Return the number of bytes actually read.  Update the offset.
-**
-** To avoid stomping the errno value on a failed write the lastErrno value
-** is set before returning.
-*/
-static int seekAndWrite(unixFile *id, i64 offset, const void *pBuf, int cnt){
-  int got;
-#if (!defined(USE_PREAD) && !defined(USE_PREAD64))
-  i64 newOffset;
-#endif
-  assert( cnt==(cnt&0x1ffff) );
-  cnt &= 0x1ffff;
+** Attempt to seek the file-descriptor passed as the first argument to
+** absolute offset iOff, then attempt to write nBuf bytes of data from
+** pBuf to it. If an error occurs, return -1 and set *piErrno. Otherwise, 
+** return the actual number of bytes written (which may be less than
+** nBuf).
+*/
+static int seekAndWriteFd(
+  int fd,                         /* File descriptor to write to */
+  i64 iOff,                       /* File offset to begin writing at */
+  const void *pBuf,               /* Copy data from this buffer to the file */
+  int nBuf,                       /* Size of buffer pBuf in bytes */
+  int *piErrno                    /* OUT: Error number if error occurs */
+){
+  int rc = 0;                     /* Value returned by system call */
+
+  assert( nBuf==(nBuf&0x1ffff) );
+  nBuf &= 0x1ffff;
   TIMER_START;
+
 #if defined(USE_PREAD)
-  do{ got = osPwrite(id->h, pBuf, cnt, offset); }while( got<0 && errno==EINTR );
+  do{ rc = osPwrite(fd, pBuf, nBuf, iOff); }while( rc<0 && errno==EINTR );
 #elif defined(USE_PREAD64)
-  do{ got = osPwrite64(id->h, pBuf, cnt, offset);}while( got<0 && errno==EINTR);
+  do{ rc = osPwrite64(fd, pBuf, nBuf, iOff);}while( rc<0 && errno==EINTR);
 #else
   do{
-    newOffset = lseek(id->h, offset, SEEK_SET);
-    SimulateIOError( newOffset-- );
-    if( newOffset!=offset ){
-      if( newOffset == -1 ){
-        ((unixFile*)id)->lastErrno = errno;
-      }else{
-        ((unixFile*)id)->lastErrno = 0;
-      }
+    i64 iSeek = lseek(fd, iOff, SEEK_SET);
+    SimulateIOError( iSeek-- );
+
+    if( iSeek!=iOff ){
+      if( piErrno ) *piErrno = (iSeek==-1 ? errno : 0);
       return -1;
     }
-    got = osWrite(id->h, pBuf, cnt);
-  }while( got<0 && errno==EINTR );
+    rc = osWrite(fd, pBuf, nBuf);
+  }while( rc<0 && errno==EINTR );
 #endif
+
   TIMER_END;
-  if( got<0 ){
-    ((unixFile*)id)->lastErrno = errno;
-  }
+  OSTRACE(("WRITE   %-3d %5d %7lld %llu\n", fd, rc, iOff, TIMER_ELAPSED));
+
+  if( rc<0 && piErrno ) *piErrno = errno;
+  return rc;
+}
+
 
-  OSTRACE(("WRITE   %-3d %5d %7lld %llu\n", id->h, got, offset, TIMER_ELAPSED));
-  return got;
+/*
+** Seek to the offset in id->offset then read cnt bytes into pBuf.
+** Return the number of bytes actually read.  Update the offset.
+**
+** To avoid stomping the errno value on a failed write the lastErrno value
+** is set before returning.
+*/
+static int seekAndWrite(unixFile *id, i64 offset, const void *pBuf, int cnt){
+  return seekAndWriteFd(id->h, offset, pBuf, cnt, &id->lastErrno);
 }
 
 
@@ -4322,24 +4335,32 @@ static int unixShmMap(
       if( sStat.st_size<nByte ){
         /* The requested memory region does not exist. If bExtend is set to
         ** false, exit early. *pp will be set to NULL and SQLITE_OK returned.
-        **
-        ** Alternatively, if bExtend is true, use ftruncate() to allocate
-        ** the requested memory region.
         */
-        if( !bExtend ) goto shmpage_out;
-#if defined(HAVE_POSIX_FALLOCATE) && HAVE_POSIX_FALLOCATE
-        if( osFallocate(pShmNode->h, sStat.st_size, nByte)!=0 ){
-          rc = unixLogError(SQLITE_IOERR_SHMSIZE, "fallocate",
-                            pShmNode->zFilename);
+        if( !bExtend ){
           goto shmpage_out;
         }
-#else
-        if( robust_ftruncate(pShmNode->h, nByte) ){
-          rc = unixLogError(SQLITE_IOERR_SHMSIZE, "ftruncate",
-                            pShmNode->zFilename);
-          goto shmpage_out;
+
+        /* Alternatively, if bExtend is true, extend the file. Do this by
+        ** writing a single byte to the end of each (OS) page being
+        ** allocated or extended. Technically, we need only write to the
+        ** last page in order to extend the file. But writing to all new
+        ** pages forces the OS to allocate them immediately, which reduces
+        ** the chances of SIGBUS while accessing the mapped region later on.
+        */
+        else{
+          static const int pgsz = 4096;
+          int iPg;
+
+          /* Write to the last byte of each newly allocated or extended page */
+          assert( (nByte % pgsz)==0 );
+          for(iPg=(sStat.st_size/pgsz); iPg<(nByte/pgsz); iPg++){
+            if( seekAndWriteFd(pShmNode->h, iPg*pgsz + pgsz-1, "", 1, 0)!=1 ){
+              const char *zFile = pShmNode->zFilename;
+              rc = unixLogError(SQLITE_IOERR_SHMSIZE, "write", zFile);
+              goto shmpage_out;
+            }
+          }
         }
-#endif
       }
     }