linux-kernel - Re: [PATCHv2] thunderbolt: do not double dequeue a request

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <vxocwwtfwg3tmjm62kcz33ypsg22afccd2ua5jqymbxaxwcigf@nnydc53vu3gv>
Date: Thu, 27 Mar 2025 23:02:04 +0900
From: Sergey Senozhatsky <senozhatsky@...omium.org>
To: Mika Westerberg <mika.westerberg@...ux.intel.com>
Cc: Sergey Senozhatsky <senozhatsky@...omium.org>, 
	Andreas Noever <andreas.noever@...il.com>, Michael Jamet <michael.jamet@...el.com>, 
	Mika Westerberg <westeri@...nel.org>, Yehezkel Bernat <YehezkelShB@...il.com>, 
	linux-usb@...r.kernel.org, linux-kernel@...r.kernel.org, stable@...r.kernel.org
Subject: Re: [PATCHv2] thunderbolt: do not double dequeue a request

Hi,

On (25/03/27 15:37), Mika Westerberg wrote:
> > Another possibility can be tb_cfg_request_sync():
> > 
> > tb_cfg_request_sync()
> >  tb_cfg_request()
> >   schedule_work(&req->work) -> tb_cfg_request_dequeue()
> >  tb_cfg_request_cancel()
> >   schedule_work(&req->work) -> tb_cfg_request_dequeue()
> 
> Not sure about this one because &req->work will only be scheduled once the
> second schedule_work() should not queue it again (as far as I can tell).

If the second schedule_work() happens after a timeout, that's what
!wait_for_completion_timeout() does, then the first schedule_work()
can already execute the work by that time, and then we can schedule
the work again (but the request is already dequeued).  Am I missing
something?

> > To address the issue, do not dequeue requests that don't
> > have TB_CFG_REQUEST_ACTIVE bit set.
> 
> Just to be sure. After this change you have not seen the issue anymore
> with your testing?

Haven't tried it yet.

We just found it today, it usually takes several weeks before
we can roll out the fix to our fleet and we prefer patches from
upstream/subsystem git, so that's why we reach out to the upstream.

The 0xdead000000000122 deference is a LIST_POISON on x86_64, which
is set explicitly in list_del(), so I'd say I'm fairly confident
that we have a double list_del() in tb_cfg_request_dequeue().