Re: [PATCH] direct IO breaks root filesystem

[GOTO Masanori]

At Wed, 12 Dec 2001 11:46:29 +0900,
Tachino Nobuhiro <[e-mail]> wrote:
> At Tue, 11 Dec 2001 17:46:01 -0800 (PST),
> Linus Torvalds wrote:
> > 
> > 
> > On Wed, 12 Dec 2001, Tachino Nobuhiro wrote:
> > >
> > > But my patch fixes another bug. Current /dev/ram* does not return -ENOSPC
> > > at the end of device size because generic_file_write() also checks whether
> > > mapping->host is a block device. So I think the patch is required.
> > 
> > I'll agree with your one-liner: it's good practice anyway to initialize
> > any fields that could ever be looked at. I actually already applied it to
> > my tree, I just want to make sure that people don't apply the other
> > patch..
> > 
> > 		Linus
> 
> Thank you.
> 
> I think the patch should be applied to 2.4 because "dd if=/dev/zero of=/dev/ram1"
> can cause system hang easily.

Linus, Tachino, your patch are both right.
I was not aware mapping inode used block file inode... Umm.

In my test, it works fine for both block device and file, and resolves
"inode->i_dev" and block size problems.
Thank you for your (simple and complete) patch! I'm now happy :)

> Marcelo, please consider applying the patch.

Yes. Would you remove my previous patches and apply these patches ?

Thanks,
-- gotom