NetBSD Problem Report #47893

From Wolfgang.Stukenbrock@nagler-company.com  Wed Jun  5 11:41:00 2013
Return-Path: <Wolfgang.Stukenbrock@nagler-company.com>
Received: from mail.netbsd.org (mail.netbsd.org [149.20.53.66])
	(using TLSv1 with cipher DHE-RSA-AES256-SHA (256/256 bits))
	(Client CN "mail.NetBSD.org", Issuer "Postmaster NetBSD.org" (verified OK))
	by mollari.NetBSD.org (Postfix) with ESMTPS id 8AE1F71AE4
	for <gnats-bugs@gnats.NetBSD.org>; Wed,  5 Jun 2013 11:41:00 +0000 (UTC)
Message-Id: <20130605114051.B547C123B93@test-s0.nagler-company.com>
Date: Wed,  5 Jun 2013 13:40:51 +0200 (CEST)
From: Wolfgang.Stukenbrock@nagler-company.com
Reply-To: Wolfgang.Stukenbrock@nagler-company.com
To: gnats-bugs@gnats.NetBSD.org
Subject: opencrypto user level requests uses different mutex for same flag variable
X-Send-Pr-Version: 3.95

>Number:         47893
>Category:       kern
>Synopsis:       opencrypto user level requests uses different mutex for same flag variable
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Wed Jun 05 11:45:00 +0000 2013
>Originator:     Dr. Wolfgang Stukenbrock
>Release:        NetBSD 6.1
>Organization:
Dr. Nagler & Company GmbH
>Environment:


System: NetBSD test-s0 5.1.2 NetBSD 5.1.2 (NSW-WS) #3: Fri Dec 21 15:15:43 CET 2012 wgstuken@test-s0:/usr/src/sys/arch/amd64/compile/NSW-WS amd64
Architecture: x86_64
Machine: amd64
>Description:
	In opencrypto there is a special handling for userlevel request useing the CRYPTO_F_USER flag.
	It is set when the request is generated in cryptodev.c and passed to the crypto stuff for processing.
	On return to the ioctl function, this flag is removed again in order to allow bypassing of the
	return-queue to avoid additional context switches (in I think most of the calls).
	OK so far - sounds like a good idea.
	But now crypto.c has introduce additional mutex structures for the queue.
	Now the crypto_ret_q_mtx is aquired by crypto_done() for setting CRYPTO_F_DONE and checking this flag, but crypto_mtx is used
	by the ioctl function for resetting the CRYPTO_F_USER flag.
	We have a race condition here, because checking and resetting is no longer synchonized and it may happen that we loose
	the CRYPTO_F_DONE flag. This would end up in endless waiting of the user-thread. 
>How-To-Repeat:
	found while validating/integrating private changes/patches into 6.1
>Fix:
	I'm not useing user-level crypto (till now) and have no time to go deeper into this - sorry.
	The problems came up when the queue mutex structures are introduced in crypto.c.

>Unformatted:

Home	PR Database Search