From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <fio-owner@vger.kernel.org>
Received: from brick.kernel.dk ([93.163.65.50]:50597 "EHLO kernel.dk"
	rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP
	id S1752074AbZDRRxK (ORCPT <rfc822;fio@vger.kernel.org>);
	Sat, 18 Apr 2009 13:53:10 -0400
Date: Sat, 18 Apr 2009 19:53:09 +0200
From: Jens Axboe <jens.axboe@oracle.com>
Subject: Re: Issue running random reads with verify on a raw device
Message-ID: <20090418175308.GE4593@kernel.dk>
References: <66dfd3fe0904101557g99d6b63r5a24a84cfe01ffbf@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <66dfd3fe0904101557g99d6b63r5a24a84cfe01ffbf@mail.gmail.com>
Sender: fio-owner@vger.kernel.org
List-Id: fio@vger.kernel.org
To: Radha Ramachandran <radha@google.com>
Cc: fio@vger.kernel.org

On Fri, Apr 10 2009, Radha Ramachandran wrote:
> Hi,
> We have a test that runs the test on a raw device say of size 100GB
> > Sequential write of a pattern to the raw device with size=10GB (so even though the device is 100GB big we try to write only to the first 10GB)
> > Random read and verify of the pattern to the raw device with size=10GB, (with randommap, so we maintain the bitmap of previously visited blocks)
> 
> This test almost always lands up trying to do the read beyond the
> 10GB. So I see this code in io_u.c:
> 
> static int get_next_free_block(struct thread_data *td, struct fio_file *f,
> 			       enum fio_ddir ddir, unsigned long long *b) {
> ...
>        while ((*b) * min_bs < f->real_file_size) {   ========= This
> code looks for an unvisited block within the real_file_size which in
> this case is 100GB, and will exceed 10GB in a lot of cases causing the
> verify to fail.
> 
> 
> So i fixed this by checking if the block is within the real_file_size
> and the io_size, so this wld work in cases where:
> 1. Actual file size < io_size
> 2. io_size < Actual file size
> 
> 
> diff -crB io_u.c io_u.c_fixed
> *** io_u.c      Fri Apr 10 15:47:02 2009
> --- io_u.c_fixed        Fri Apr 10 15:48:02 2009
> ***************
> *** 113,119 ****
> 
>         i = f->last_free_lookup;
>         *b = (i * BLOCKS_PER_MAP);
> !       while ((*b) * min_bs < f->real_file_size) {
>                 if (f->file_map[i] != (unsigned int) -1) {
>                         *b += ffz(f->file_map[i]);
>                         if (*b > last_block(td, f, ddir))
> --- 113,120 ----
> 
>         i = f->last_free_lookup;
>         *b = (i * BLOCKS_PER_MAP);
> !       while (((*b) * min_bs < f->real_file_size) &&
> !               ((*b) * min_bs < f->io_size)) {
>                 if (f->file_map[i] != (unsigned int) -1) {
>                         *b += ffz(f->file_map[i]);
>                         if (*b > last_block(td, f, ddir))
> 
> 
> Is there some case I missed?

Yeah I don't think this will work for using an offset within the file,
as it's legal for the offset to be larger than the io_size. If you can
check and add the offset check, I think it should be ok.

-- 
Jens Axboe