Re: [PATCH] direct IO breaks root filesystem

[GOTO Masanori]

At Sun, 9 Dec 2001 21:55:57 -0800 (PST),
Linus Torvalds <[e-mail]> wrote:
> On Mon, 10 Dec 2001, GOTO Masanori wrote:
> >
> > The reason is that when kernel accesses /dev/sda with O_DIRECT,
> > blkdev_direct_IO() is called. But, it calls generic_direct_IO(),
> > and generic_direct_IO() calls brw_kiovec(..., inode->i_dev, ...).
> 
> That's a bad bug, yes.

Yes, dangerous bug...

> However, the bug is really in "generic_direct_IO()", and you should fix it
> there, instead of avoiding to use it altogether.
> 
> "generic_direct_IO()" should just get the device from "bh->b_dev (which is
> filled in correctly by "get_block()".

Oh, that's right, the patch becomes more simple (and works well).
Is this patch OK?

--- linux.vanilla/fs/buffer.c	Mon Dec 10 21:13:10 2001
+++ linux/fs/buffer.c	Mon Dec 10 20:59:34 2001
@@ -2003,12 +2003,12 @@
 {
 	int i, nr_blocks, retval;
 	unsigned long * blocks = iobuf->blocks;
+	struct buffer_head bh;
 
+	bh.b_dev = inode->i_dev;
 	nr_blocks = iobuf->length / blocksize;
 	/* build the blocklist */
 	for (i = 0; i < nr_blocks; i++, blocknr++) {
-		struct buffer_head bh;
-
 		bh.b_state = 0;
 		bh.b_dev = inode->i_dev;
 		bh.b_size = blocksize;
@@ -2034,7 +2034,7 @@
 		blocks[i] = bh.b_blocknr;
 	}
 
-	retval = brw_kiovec(rw, 1, &iobuf, inode->i_dev, iobuf->blocks, blocksize);
+	retval = brw_kiovec(rw, 1, &iobuf, bh.b_dev, iobuf->blocks, blocksize);
 
  out:
 	return retval;

-- gotom