xfs
[Top] [All Lists]

Re: TAKE - Fix fsx corruption.

To: Rusell Cattelan <cattelan@xxxxxxxxxxxxxxxxxxxxx>, linux-xfs <linux-xfs@xxxxxxxxxxx>
Subject: Re: TAKE - Fix fsx corruption.
From: Andrew Morton <akpm@xxxxxxxxx>
Date: Tue, 29 Oct 2002 19:49:29 -0800
References: <200210300037.g9U0b3Qj021050@rose.americas.sgi.com>
Sender: linux-xfs-bounce@xxxxxxxxxxx
Rusell Cattelan wrote:
> 
>         - fsx (file system exerciser orginally from NeXT)
>           does all sorts of combinations of buffered reads/write map 
> read/writes and truncates
>           This was exposing a problem in our write_page logic on file systems 
> with block size < page size.

fsx-linux is great.  It caused me *no end* of grief in 2.5.13 ;)

I have a warmed-up version which does O_DIRECT testing as well.  Just give
it the `-Z' flag to force O_DIRECT access and the existing `-r' and `-w'
flags to set the alignments and `-R' and `-W' to supress mmap access.  (I
don't think testing mmap and O_DIRECT consistency at the same time is an
interesting thing to do.  But, to my surprise, it _does_ work).

So to test the fancy finer-than-blocksize direct-IO support which Linus
just merged, use:

        fsx-linux -r 1024 -w 512 -R -W -Z foo

Grab it from http://www.zip.com.au/~akpm/linux/patches/stuff/


<Prev in Thread] Current Thread [Next in Thread>