* fhandler.cc (fhandler_base::open_9x): Handle O_SYNC and O_DIRECT

	flags.
	(fhandler_base::open): Ditto.
	* fhandler_floppy.cc (fhandler_dev_floppy::open): Don't allocate devbuf
	in O_DIRECT case.
	* fhandler_raw.cc (fhandler_dev_raw::ioctl): Don't allow buffer
	changes in O_DIRECT case.  Allow returning a buffer size 0, which
	indicates O_DIRECT.
	* fhandler_tape.cc (fhandler_dev_tape::open): Use O_SYNC flag to
	hand down the !buffer_writes case.  Don't allocate devbuf in O_DIRECT
	case.
	(fhandler_dev_tape::raw_read): Don't mess with devbuf if it's NULL.
	* include/fcntl.h: Define _FDIRECT, O_DIRECT, O_DSYNC and O_RSYNC.
	* include/cygwin/version.h: Bump API minor version.

7 files changed, 124 insertions, 75 deletions

diff --git a/winsup/cygwin/ChangeLog b/winsup/cygwin/ChangeLog
index f4aa368fd..2591eab4d 100644
--- a/winsup/cygwin/ChangeLog
+++ b/winsup/cygwin/ChangeLog
@@ -1,3 +1,20 @@
+2005-12-14  Corinna Vinschen  <corinna@vinschen.de>
+
+	* fhandler.cc (fhandler_base::open_9x): Handle O_SYNC and O_DIRECT
+	flags.
+	(fhandler_base::open): Ditto.
+	* fhandler_floppy.cc (fhandler_dev_floppy::open): Don't allocate devbuf
+	in O_DIRECT case.
+	* fhandler_raw.cc (fhandler_dev_raw::ioctl): Don't allow buffer
+	changes in O_DIRECT case.  Allow returning a buffer size 0, which
+	indicates O_DIRECT.
+	* fhandler_tape.cc (fhandler_dev_tape::open): Use O_SYNC flag to
+	hand down the !buffer_writes case.  Don't allocate devbuf in O_DIRECT
+	case.
+	(fhandler_dev_tape::raw_read): Don't mess with devbuf if it's NULL.
+	* include/fcntl.h: Define _FDIRECT, O_DIRECT, O_DSYNC and O_RSYNC.
+	* include/cygwin/version.h: Bump API minor version.
+
 2005-12-13  Christopher Faylor  <cgf@timesys.com>
 
 	* cygtls.cc (_cygtls::remove): Don't bother if we're exiting.
diff --git a/winsup/cygwin/fhandler.cc b/winsup/cygwin/fhandler.cc
index 6931cb6bc..0a119c845 100644
--- a/winsup/cygwin/fhandler.cc
+++ b/winsup/cygwin/fhandler.cc
@@ -492,6 +492,10 @@ fhandler_base::open_9x (int flags, mode_t mode)
   file_attributes = FILE_ATTRIBUTE_NORMAL;
   if (flags & O_DIROPEN)
     file_attributes |= FILE_FLAG_BACKUP_SEMANTICS;
+  if (flags & O_SYNC)
+    file_attributes |= FILE_FLAG_WRITE_THROUGH;
+  if (flags & O_DIRECT)
+    file_attributes |= FILE_FLAG_NO_BUFFERING;
   if (get_major () == DEV_SERIAL_MAJOR)
     file_attributes |= FILE_FLAG_OVERLAPPED;
 
@@ -599,18 +603,16 @@ fhandler_base::open (int flags, mode_t mode)
 	break;
       default:
 	create_options = 0;
-	if (get_major () == DEV_TAPE_MAJOR && (flags & O_TEXT))
-	  {
-	    /* O_TEXT is used to indicate write-through on tape devices */
-	    create_options |= FILE_WRITE_THROUGH;
-	    flags &= ~O_TEXT;
-	  }
 	if ((flags & (O_RDONLY | O_WRONLY | O_RDWR)) == O_RDONLY)
 	  access = GENERIC_READ;
 	else if ((flags & (O_RDONLY | O_WRONLY | O_RDWR)) == O_WRONLY)
 	  access = GENERIC_WRITE | FILE_READ_ATTRIBUTES;
 	else
 	  access = GENERIC_READ | GENERIC_WRITE;
+	if (flags & O_SYNC)
+	  create_options |= FILE_WRITE_THROUGH;
+	if (flags & O_DIRECT)
+	  create_options |= FILE_NO_INTERMEDIATE_BUFFERING;
 	if (get_major () != DEV_SERIAL_MAJOR && get_major () != DEV_TAPE_MAJOR)
 	  {
 	    create_options |= FILE_SYNCHRONOUS_IO_NONALERT;
diff --git a/winsup/cygwin/fhandler_floppy.cc b/winsup/cygwin/fhandler_floppy.cc
index 3bc9f5191..a8b99b266 100644
--- a/winsup/cygwin/fhandler_floppy.cc
+++ b/winsup/cygwin/fhandler_floppy.cc
@@ -161,9 +161,10 @@ fhandler_dev_floppy::open (int flags, mode_t)
      relatively big value increases performance by means.  The new ioctl call
      with 'rdevio.h' header file supports changing this value.
 
-     Let's try to be smart: Let's take a multiple of typical tar and cpio
-     buffer sizes by default. */
-  devbufsiz = 61440L;
+     As default buffer size, we're using some value which is a multiple of
+     the typical tar and cpio buffer sizes, Except O_DIRECT is set, in which
+     case we're not buffering at all. */
+  devbufsiz = (flags & O_DIRECT) ? 0L : 61440L;
   int ret =  fhandler_dev_raw::open (flags);
 
   if (ret && get_drive_info (NULL))
diff --git a/winsup/cygwin/fhandler_raw.cc b/winsup/cygwin/fhandler_raw.cc
index 00d023e7e..0cfb51e6c 100644
--- a/winsup/cygwin/fhandler_raw.cc
+++ b/winsup/cygwin/fhandler_raw.cc
@@ -159,12 +159,14 @@ fhandler_dev_raw::ioctl (unsigned int cmd, void *buf)
 	      }
 	    else if ((devbuf && ((op->rd_parm <= 1 && (devbufend - devbufstart))
 				 || op->rd_parm < devbufend - devbufstart))
-		     || (op->rd_parm > 1 && (op->rd_parm % 512)))
+		     || (op->rd_parm > 1 && (op->rd_parm % 512))
+		     || (get_flags () & O_DIRECT))
 	      /* The conditions for a *valid* parameter are these:
 	         - If there's still data in the current buffer, it must
 		   fit in the new buffer.
 		 - The new size is either 0 or 1, both indicating unbufferd
-		   I/O, or the new buffersize must be a multiple of 512. */
+		   I/O, or the new buffersize must be a multiple of 512.
+		 - In the O_DIRECT case, the whole request is invalid. */
 	      ret = ERROR_INVALID_PARAMETER;
 	    else if (!devbuf || op->rd_parm != devbufsiz)
 	      {
@@ -198,7 +200,7 @@ fhandler_dev_raw::ioctl (unsigned int cmd, void *buf)
       if (!get)
 	ret = ERROR_INVALID_PARAMETER;
       else
-	get->bufsiz = devbufsiz ?: 1L;
+	get->bufsiz = devbufsiz;
     }
   else
     return fhandler_base::ioctl (cmd, buf);
diff --git a/winsup/cygwin/fhandler_tape.cc b/winsup/cygwin/fhandler_tape.cc
index d3fd3eee2..0b4e53773 100644
--- a/winsup/cygwin/fhandler_tape.cc
+++ b/winsup/cygwin/fhandler_tape.cc
@@ -1216,11 +1216,16 @@ fhandler_dev_tape::open (int flags, mode_t)
       __seterrno ();
       return 0;
     }
-  /* The O_TEXT flag is used to indicate write-through (non buffered writes)
-     to the underlying fhandler_dev_raw::open call. */
-  flags &= ~O_TEXT;
+
+  /* The O_SYNC flag is not supported by the tape driver.  Use the
+     MT_ST_BUFFER_WRITES and MT_ST_ASYNC_WRITES flags in the drive
+     settings instead.  In turn, the MT_ST_BUFFER_WRITES is translated
+     into O_SYNC, which controls the FILE_WRITE_THROUGH flag in the
+     NtCreateFile call in fhandler_base::open. */
+  flags &= ~O_SYNC;
   if (!mt->drive (driveno ())->buffer_writes ())
-    flags |= O_TEXT;
+    flags |= O_SYNC;
+
   ret = fhandler_dev_raw::open (flags);
   if (ret)
     {
@@ -1231,8 +1236,11 @@ fhandler_dev_tape::open (int flags, mode_t)
 	mt->drive (driveno ())->set_pos (get_handle (),
 					 TAPE_SPACE_FILEMARKS, 1, true);
 
-      devbufsiz = mt->drive (driveno ())->dp ()->MaximumBlockSize;
-      devbuf = new char [devbufsiz];
+      if (!(flags & O_DIRECT))
+        {
+	  devbufsiz = mt->drive (driveno ())->dp ()->MaximumBlockSize;
+	  devbuf = new char [devbufsiz];
+	}
       devbufstart = devbufend = 0;
     }
   else
@@ -1283,70 +1291,81 @@ fhandler_dev_tape::raw_read (void *ptr, size_t &ulen)
       return;
     }
   block_size = mt->drive (driveno ())->mp ()->BlockSize;
-  if (devbufend > devbufstart)
+  if (devbuf)
     {
-      bytes_to_read = min (len, devbufend - devbufstart);
-      debug_printf ("read %d bytes from buffer (rest %d)",
-		    bytes_to_read, devbufend - devbufstart - bytes_to_read);
-      memcpy (buf, devbuf + devbufstart, bytes_to_read);
-      len -= bytes_to_read;
-      bytes_read += bytes_to_read;
-      buf += bytes_to_read;
-      devbufstart += bytes_to_read;
-      if (devbufstart == devbufend)
-	devbufstart = devbufend = 0;
-      /* If a switch to variable block_size occured, just return the buffer
-	 remains until the buffer is empty, then proceed with usual variable
-	 block size handling (one block per read call). */
-      if (!block_size)
-	len = 0;
-    }
-  if (len > 0)
-    {
-      if (!mt_evt && !(mt_evt = CreateEvent (&sec_none, TRUE, FALSE, NULL)))
-	debug_printf ("Creating event failed, %E");
-      size_t block_fit = !block_size ? len : rounddown(len,  block_size);
-      if (block_fit)
+      if (devbufend > devbufstart)
 	{
-	  debug_printf ("read %d bytes from tape (rest %d)",
-			block_fit, len - block_fit);
-	  ret = mt->drive (driveno ())->read (get_handle (), mt_evt, buf,
-					      block_fit);
-	  if (ret)
-	    __seterrno_from_win_error (ret);
-	  else if (block_fit)
-	    {
-	      len -= block_fit;
-	      bytes_read += block_fit;
-	      buf += block_fit;
-	      /* Only one block in each read call, please. */
-	      if (!block_size)
-		len = 0;
-	    }
-	  else {
+	  bytes_to_read = min (len, devbufend - devbufstart);
+	  debug_printf ("read %d bytes from buffer (rest %d)",
+			bytes_to_read, devbufend - devbufstart - bytes_to_read);
+	  memcpy (buf, devbuf + devbufstart, bytes_to_read);
+	  len -= bytes_to_read;
+	  bytes_read += bytes_to_read;
+	  buf += bytes_to_read;
+	  devbufstart += bytes_to_read;
+	  if (devbufstart == devbufend)
+	    devbufstart = devbufend = 0;
+	  /* If a switch to variable block_size occured, just return the buffer
+	     remains until the buffer is empty, then proceed with usual variable
+	     block size handling (one block per read call). */
+	  if (!block_size)
 	    len = 0;
-	    if (bytes_read)
-	      lastblk_to_read (true);
-	  }
 	}
-      if (!ret && len > 0)
+      if (len > 0)
 	{
-	  debug_printf ("read %d bytes from tape (one block)", block_size);
-	  ret = mt->drive (driveno ())->read (get_handle (), mt_evt, devbuf,
-					      block_size);
-	  if (ret)
-	    __seterrno_from_win_error (ret);
-	  else if (block_size)
+	  if (!mt_evt && !(mt_evt = CreateEvent (&sec_none, TRUE, FALSE, NULL)))
+	    debug_printf ("Creating event failed, %E");
+	  size_t block_fit = !block_size ? len : rounddown(len,  block_size);
+	  if (block_fit)
+	    {
+	      debug_printf ("read %d bytes from tape (rest %d)",
+			    block_fit, len - block_fit);
+	      ret = mt->drive (driveno ())->read (get_handle (), mt_evt, buf,
+						  block_fit);
+	      if (ret)
+		__seterrno_from_win_error (ret);
+	      else if (block_fit)
+		{
+		  len -= block_fit;
+		  bytes_read += block_fit;
+		  buf += block_fit;
+		  /* Only one block in each read call, please. */
+		  if (!block_size)
+		    len = 0;
+		}
+	      else {
+		len = 0;
+		if (bytes_read)
+		  lastblk_to_read (true);
+	      }
+	    }
+	  if (!ret && len > 0)
 	    {
-	      devbufstart = len;
-	      devbufend = block_size;
-	      bytes_read += len;
-	      memcpy (buf, devbuf, len);
+	      debug_printf ("read %d bytes from tape (one block)", block_size);
+	      ret = mt->drive (driveno ())->read (get_handle (), mt_evt, devbuf,
+						  block_size);
+	      if (ret)
+		__seterrno_from_win_error (ret);
+	      else if (block_size)
+		{
+		  devbufstart = len;
+		  devbufend = block_size;
+		  bytes_read += len;
+		  memcpy (buf, devbuf, len);
+		}
+	      else if (bytes_read)
+		lastblk_to_read (true);
 	    }
-	  else if (bytes_read)
-	    lastblk_to_read (true);
 	}
     }
+  else
+    {
+      if (!mt_evt && !(mt_evt = CreateEvent (&sec_none, TRUE, FALSE, NULL)))
+	debug_printf ("Creating event failed, %E");
+      bytes_read = ulen;
+      ret = mt->drive (driveno ())->read (get_handle (), mt_evt, ptr,
+					  bytes_read);
+    }
   ulen = (ret ? (size_t) -1 : bytes_read);
   unlock ();
 }
diff --git a/winsup/cygwin/include/cygwin/version.h b/winsup/cygwin/include/cygwin/version.h
index db65e43a4..048b06e9d 100644
--- a/winsup/cygwin/include/cygwin/version.h
+++ b/winsup/cygwin/include/cygwin/version.h
@@ -282,12 +282,13 @@ details. */
            for this?
       147: Eliminate problematic d_ino from dirent structure.  unsetenv now returns
            int, as per linux.
+      148: Add open(2) flags O_SYNC, O_RSYNC, O_DSYNC and O_DIRECT.
      */
 
      /* Note that we forgot to bump the api for ualarm, strtoll, strtoull */
 
 #define CYGWIN_VERSION_API_MAJOR 0
-#define CYGWIN_VERSION_API_MINOR 147
+#define CYGWIN_VERSION_API_MINOR 148
 
      /* There is also a compatibity version number associated with the
 	shared memory regions.  It is incremented when incompatible
diff --git a/winsup/cygwin/include/fcntl.h b/winsup/cygwin/include/fcntl.h
index 774966e9a..33b0692b3 100644
--- a/winsup/cygwin/include/fcntl.h
+++ b/winsup/cygwin/include/fcntl.h
@@ -1,6 +1,6 @@
 /* fcntl.h
 
-   Copyright 1996, 1998, 2001 Red Hat, Inc.
+   Copyright 1996, 1998, 2001, 2005 Red Hat, Inc.
 
 This file is part of Cygwin.
 
@@ -14,4 +14,11 @@ details. */
 #include <sys/fcntl.h>
 #define O_NDELAY	_FNDELAY
 
+/* sys/fcntl defines values up to 0x40000 (O_NOINHERIT). */
+#define _FDIRECT	0x80000
+
+#define O_DIRECT	_FDIRECT
+#define O_DSYNC		_FSYNC
+#define O_RSYNC		_FSYNC
+
 #endif /* _FCNTL_H */