Skip to contentSkip to conversation

Patch FIOFLUSH

PATCH Ross Biro 3 participants 3 messages May 21, 2003

Subsystems (2)
VFS Landlock security module
lore
Share & tools
lore

Ross Biro

author
#0 lore raw

Here's a patch against 2.4.18 that allows user space to flush a file from both the buffer cache and the page cache. The reason for flushing a file from the caches is to the read the file again to verify it made it to more permanent storage correctly. Someone may want to add similiar code to 2.5.

Ross

fs/ioctl.c +7 -0
diff -urdbB linux-2.4.18-58/fs/ioctl.c linux-2.4.18-59/fs/ioctl.c
@@ -39,6 +39,13 @@
 			return put_user(inode->i_sb->s_blocksize, (int *) arg);
 		case FIONREAD:
 			return put_user(inode->i_size - filp->f_pos, (int *) arg);
+
+                case FIOFLUSH:
+                        write_inode_now(inode, 1);
+                        invalidate_inode_buffers(inode);
+                        invalidate_inode_pages(inode);
+                        return 0;
+
 	}
 	if (filp->f_op && filp->f_op->ioctl)
 		return filp->f_op->ioctl(inode, filp, cmd, arg);
include/asm-i386/ioctls.h +1 -0
diff -urdbB linux-2.4.18-58/include/asm/ioctls.h linux-2.4.18-59/include/asm/ioctls.h
@@ -32,6 +32,7 @@
 #define TIOCGSOFTCAR	0x5419
 #define TIOCSSOFTCAR	0x541A
 #define FIONREAD	0x541B
+#define FIOFLUSH	_IO('F', 1) /* flush a file out of the caches */
 #define TIOCINQ		FIONREAD
 #define TIOCLINUX	0x541C
 #define TIOCCONS	0x541D
#1 lore raw
Quoted 6 lines · Ross Biro

Here's a patch against 2.4.18 that allows user space to flush a file from both the buffer cache and the page cache. The reason for flushing a file from the caches is to the read the file again to verify it made it to more permanent storage correctly. Someone may want to add similiar code to 2.5.

This came up at SGI some time ago and the right solution is _not_ a new
ioctl but fadvise(..., POSIX_FADV_DONTNEED). I'll submit a clean backport
once 2.4.21 is out (if that will ever happen)

#2 lore raw

On Wednesday 21 May 2003 19:38, Christoph Hellwig wrote:

Hi Christoph,

Quoted 2 lines

This came up at SGI some time ago and the right solution is _not_ a new
ioctl but fadvise(..., POSIX_FADV_DONTNEED). I'll submit a clean backport

could you please send it to me even 2.4.21 is not out? Thank you.

Quoted 1 line

once 2.4.21 is out (if that will ever happen)

ROTFL! :-)

ciao, Marc