From: Baruch Siach <baruch@tkos.co.il>
Subject: Re: [PATCH] e4defrag: fix build when posix_fadvise is missing
Date: Thu, 2 Jan 2014 08:49:00 +0200
Message-ID: <20140102064900.GJ6589@tarshish>
References: <33245e3808058c72b66931ac14aea8d5dc6d1ba5.1388572103.git.baruch@tkos.co.il>
 <20140101172205.GC17519@thunk.org>
 <20140101173146.GM4470@tarshish>
 <20140101173744.GD17519@thunk.org>
 <20140101190800.GA6589@tarshish>
 <20140101204650.GE17519@thunk.org>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Cc: linux-ext4@vger.kernel.org
To: Theodore Ts'o <tytso@mit.edu>
Content-Disposition: inline
In-Reply-To: <20140101204650.GE17519@thunk.org>
Sender: linux-ext4-owner@vger.kernel.org

Hi Ted,

On Wed, Jan 01, 2014 at 03:46:50PM -0500, Theodore Ts'o wrote:
> On Wed, Jan 01, 2014 at 09:08:00PM +0200, Baruch Siach wrote:
> > Yes. uClibc does provide posix_fadvise64() when __NR_fadvise64_64 is 
> > available.
> > 
> > > Maybe the better fix would be to try to use posix_fadvise64 if it is exists, 
> > > with posix_fadvise() and the locally defined posix_fadvise() as fallbacks.
> > 
> > OK. So you suggest to add another autoconf test for posix_fadvise64 and use 
> > that as a fall back before trying direct syscall()?
> 
> Yes, I think that's the more general and hence the more correct
> answer.  In fact what we have right now is arguably wrong, since on a
> 32-bit platform where off_t is 32-bits, and which defines
> posix_fadvise, we wouldn't correctly handle files that are larger than
> 2GB.  So if posix_fadvise64 exists, we should be using in preference
> to posix_fadvise(), even if both are provided by the C library.
> 
> Would you mind sending such a revised patch?

Just sent. Thanks for reviewing.

Turns out that this is actually an uClibc bug. There used to be a 
posix_fadvise definition for xtensa as a wrapper around the __NR_fadvise64_64 
system call. This broke with uClibc commit ee84b8b400 (linux: posix_fadvise: 
use new SYSCALL_ALIGN_64BIT). The patch I have sent can be used as a 
workaround for now. I'll send a separate patch for uClibc later.

baruch

-- 
     http://baruch.siach.name/blog/                  ~. .~   Tk Open Systems
=}------------------------------------------------ooO--U--Ooo------------{=
   - baruch@tkos.co.il - tel: +972.2.679.5364, http://www.tkos.co.il -