From: Hannes Reinecke Subject: Re: [PATCH] bio: return EINTR if copying to user space got interrupted Date: Fri, 12 Feb 2016 17:14:12 +0100 Message-ID: <56BE04D4.7080308@suse.de> References: <1455266355-44676-1-git-send-email-hare@suse.de> <56BE02BA.5080001@interlog.com> Mime-Version: 1.0 Content-Type: text/plain; charset=windows-1252; format=flowed Content-Transfer-Encoding: QUOTED-PRINTABLE Return-path: In-Reply-To: <56BE02BA.5080001@interlog.com> Sender: stable-owner@vger.kernel.org To: dgilbert@interlog.com, "Martin K. Petersen" Cc: James Bottomley , linux-scsi@vger.kernel.org, Ewan Milne , Jens Axboe , linux-block@vger.kernel.org, Johannes Thumshirn , stable@vger.kernel.org, #@suse.de, v.3.11+@suse.de List-Id: linux-scsi@vger.kernel.org On 02/12/2016 05:05 PM, Douglas Gilbert wrote: > On 16-02-12 03:39 AM, Hannes Reinecke wrote: >> Commit 35dc248383bbab0a7203fca4d722875bc81ef091 introduced a check f= or >> current->mm to see if we have a user space context and only copies d= ata >> if we do. Now if an IO gets interrupted by a signal data isn't copie= d >> into user space any more (as we don't have a user space context) but >> user space isn't notified about it. >> >> This patch modifies the behaviour to return -EINTR from bio_uncopy_u= ser() >> to notify userland that a signal has interrupted the syscall, otherw= ise >> it could lead to a situation where the caller may get a buffer with >> no data returned. > > Interesting, the "f091" commit has been in the kernel since 2013 > hence your reference to v.3.11 . I always had the feeling that > handling signals that interrupted SG_IO calls was skating on thin > ice. Yeah; that bug was really annoying, as occasionally receiving no data=20 whatsoever without any indication makes it really, really hard to debug= =2E Kudos go to Johannes and Ewan for pointing to the offending function. > Hence in ddpt (but not sg_dd nor sgp_dd) the code masks out > all signals (that it can) during the SG_IO calls then opens a > signal window briefly after a SG_IO ioctl has finished and before > the next one starts. This approach used by ddpt is borrowed from > dd (in coreutils) which masks signals during its read() and > write() calls. > > Any idea how accurate resid is in this scenario? > Bah. F*sk knows. That takes far more POSIX knowledge than I have. I would suspect that by masking out signals they'll be marked as pendin= g=20 for the process, and will be delivered once you unmask them. Or dropped, depending. In either case, once they are blocked out the kernel part shouldn't=20 receive a signal, and hence we should be able to receive the data prope= rly. But as noted I'm not a POSIX expert. Cheers, Hannes --=20 Dr. Hannes Reinecke zSeries & Storage hare@suse.de +49 911 74053 688 SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 N=FCrnberg GF: J. Hawn, J. Guild, F. Imend=F6rffer, HRB 16746 (AG N=FCrnberg)