X-Recipient: archive-cygwin AT delorie DOT com
X-Spam-Check-By: sourceware.org
Date: Fri, 23 Nov 2012 13:19:42 +0100
From: Corinna Vinschen <corinna-cygwin AT cygwin DOT com>
To: cygwin AT cygwin DOT com
Subject: Re: SYSV semaphore bug (testcase attached)
Message-ID: <20121123121942.GP17347@calimero.vinschen.de>
Reply-To: cygwin AT cygwin DOT com
Mail-Followup-To: cygwin AT cygwin DOT com
References: <5F8AAC04F9616747BC4CC0E803D5907D012656 AT MLBXV09 DOT nih DOT gov>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <5F8AAC04F9616747BC4CC0E803D5907D012656@MLBXV09.nih.gov>
User-Agent: Mutt/1.5.21 (2010-09-15)
Mailing-List: contact cygwin-help AT cygwin DOT com; run by ezmlm
Precedence: bulk
List-Id: <cygwin.cygwin.com>
List-Unsubscribe: <mailto:cygwin-unsubscribe-archive-cygwin=delorie DOT com AT cygwin DOT com>
List-Subscribe: <mailto:cygwin-subscribe AT cygwin DOT com>
List-Archive: <http://sourceware.org/ml/cygwin/>
List-Post: <mailto:cygwin AT cygwin DOT com>
List-Help: <mailto:cygwin-help AT cygwin DOT com>, <http://sourceware.org/ml/#faqs>
Sender: cygwin-owner AT cygwin DOT com
Mail-Followup-To: cygwin AT cygwin DOT com
Delivered-To: mailing list cygwin AT cygwin DOT com

On Nov 19 16:04, Lavrentiev, Anton (NIH/NLM/NCBI) [C] wrote:
> Hi,
> 
> As I previously reported, there is a weird behavior of CYGWIN implementation
> of SYSV semaphores, and a bug exposition for one problem is attached below
> (I'm still looking into some other issues in there).
> 
> If the code is compiled with both BUG1 and BUG2 defined (as shown), it
> will abort at iteration 16384 (just the default semaphore overflow).
> 
> Undefining BUG2 causes the problem disappear (because there is no longer
> UNDO on the 1st semaphore).  Also, running the code with just one semaphore
> (undef BUG1) causes no problem.  Finally, replacing "#if 1" with "#if 0"
> to unlock the semaphore allows to run indefinitely with any combination of
> BUG1/2 (just remember to issue ipcs/ipcrm to start with a clean slate at
> all times).
> 
> Reviewing the code of CYGSERVER, there is an apparent bug in the semundo_clear()
> routine (at around line 536, which looks like "i++, sunptr++;", and advances
> both undo indexes even when "if (sunptr->un_id == semid)" (line 524)
> failed to match semid.  This means that for two (or more) semaphores, the
> undo index "i" moves ahead even when nothing was done while still searching.
> This causes the adjust pointer to miss the position to clear, and overflow
> the semaphore adjust value (line 1207, semop(), by the virtue of
> semundo_adjust()'s logic at about line 486).

This is original FreeBSD code, so I have a hard time to follow the idea
that it might be wrong.  I stared a long while into the source now, and
I compared that with the latest version of the upstream code.

The i counter is in lockstep with the sunptr pointer.  The only time
something happens is if sunptr->un_num (== suptr->un_ent[i].un_num)
equals semnum.  In that case, suptr->un_ent[i] is overwritten with
with the last element uptr->un_ent[suptr->un_cnt], and then the code
calls continue, thus NOT incrementing i and sunptr.  So the same
element, now containing the contents of suptr->un_ent[suptr->un_cnt],
is evaluated again.

Am I missing something?

Incidentally this seems to be wrong now in the new upstream code, which
uses for rather than while, thus i and sunptr are actually incremented
in the above case, too.


Corinna

-- 
Corinna Vinschen                  Please, send mails regarding Cygwin to
Cygwin Project Co-Leader          cygwin AT cygwin DOT com
Red Hat

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple