From mboxrd@z Thu Jan 1 00:00:00 1970 From: Christoph Hellwig Subject: Re: [PATCH (resend)] xfs: don't allow recursion into fs under write_begin Date: Tue, 25 May 2010 06:33:17 -0400 Message-ID: <20100525103317.GB2864@infradead.org> References: <4BF8056E.8080900@sandeen.net> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: xfs-oss , "linux-fsdevel@vger.kernel.org" , linux-mm@kvack.org, Michael Monnerie To: Eric Sandeen Return-path: Received: from bombadil.infradead.org ([18.85.46.34]:43927 "EHLO bombadil.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753246Ab0EYKdU (ORCPT ); Tue, 25 May 2010 06:33:20 -0400 Content-Disposition: inline In-Reply-To: <4BF8056E.8080900@sandeen.net> Sender: linux-fsdevel-owner@vger.kernel.org List-ID: On Sat, May 22, 2010 at 11:25:18AM -0500, Eric Sandeen wrote: > Michael Monnerie reported this fantastic stack overflow: > I don't think we can afford to let write_begin recurse into the fs, > so we can set AOP_FLAG_NOFS ... is this too big a hammer? I don't really like it. There's nothing XFS-specific here - it's the same problem with direct reclaim calling back into the FS and causing massive amounts of problems. If we want to fix this class of problems we just need to do the same thing ext4 and btrfs already do and refuse to call the allocator from reclaim context. Just curious, how much stack does the path up to generic_perform_write use?