From mboxrd@z Thu Jan 1 00:00:00 1970 From: Nick Piggin Subject: Re: [rfc][patch 2/2] buffer, btrfs: fix page_mkwrite error cases Date: Tue, 3 Mar 2009 15:41:46 +0100 Message-ID: <20090303144146.GA18142@wotan.suse.de> References: <20090303103838.GC17042@wotan.suse.de> <20090303104114.GD17042@wotan.suse.de> <1236090363.782.6.camel@think.oraclecorp.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Linux Memory Management List , linux-fsdevel@vger.kernel.org To: Chris Mason Return-path: Received: from ns.suse.de ([195.135.220.2]:57158 "EHLO mx1.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751929AbZCCOls (ORCPT ); Tue, 3 Mar 2009 09:41:48 -0500 Content-Disposition: inline In-Reply-To: <1236090363.782.6.camel@think.oraclecorp.com> Sender: linux-fsdevel-owner@vger.kernel.org List-ID: On Tue, Mar 03, 2009 at 09:26:03AM -0500, Chris Mason wrote: > On Tue, 2009-03-03 at 11:41 +0100, Nick Piggin wrote: > > page_mkwrite is called with neither the page lock nor the ptl held. This > > means a page can be concurrently truncated or invalidated out from underneath > > it. Callers are supposed to prevent truncate races themselves, however > > previously the only thing they can do in case they hit one is to raise a > > SIGBUS. A sigbus is wrong for the case that the page has been invalidated > > or truncated within i_size (eg. hole punched). Callers may also have to > > perform memory allocations in this path, where again, SIGBUS would be wrong. > > > > The previous patch made it possible to properly specify errors. Convert > > the generic buffer.c code and btrfs to return sane error values > > (in the case of page removed from pagecache, VM_FAULT_NOPAGE will cause the > > fault handler to exit without doing anything, and the fault will be retried > > properly). > > > > Thanks Nick. I think the btrfs patch needs an extra } to compile, but > it looks fine. OK... btrfs is obviously untested :) I just got to btrfs and realised that probably most of the non-trivial ones will want fs maintainers to take a look. I *think* the following errors should mostly be right: !page->mapping ==> VM_FAULT_NOPAGE (just cause the VM to retry the fault) -ENOMEM ==> VM_FAULT_OOM any other error ==> VM_FAULT_SIGBUS But there are a lot of possible error paths in some fs'es... [ For anyone unclear, SIGBUS is to be used when the physical object under the virtual mapping is invalid/unavailable, as opposed to SIGSEGV for when the virtual mapping itself if invalid. Obviously fs is below the virtual memory layer, so SIGBUS / OOM are the only errors that make sense. ]