linux-kernel - Re: KASAN: use-after-free Read in device_release_driver

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: <Pine.LNX.4.44L0.1908061009450.1571-100000@iolanthe.rowland.org>
Date:   Tue, 6 Aug 2019 10:19:37 -0400 (EDT)
From:   Alan Stern <stern@...land.harvard.edu>
To:     Oliver Neukum <oneukum@...e.com>
cc:     Andrey Konovalov <andreyknvl@...gle.com>,
        syzkaller-bugs <syzkaller-bugs@...glegroups.com>,
        syzbot <syzbot+1b2449b7b5dc240d107a@...kaller.appspotmail.com>,
        LKML <linux-kernel@...r.kernel.org>,
        USB list <linux-usb@...r.kernel.org>
Subject: Re: KASAN: use-after-free Read in device_release_driver_internal

On Tue, 6 Aug 2019, Oliver Neukum wrote:

> Am Donnerstag, den 01.08.2019, 14:47 -0400 schrieb Alan Stern:
> > 
> > I think this must be caused by an unbalanced refcount.  That is,
> > something must drop one more reference to the device than it takes.
> > That would explain why the invalid access occurs inside a single
> > bus_remove_device() call, between the klist_del() and
> > device_release_driver().
> > 
> > The kernel log indicates that the device was probed by rndis_wlan,
> > rndis_host, and cdc_acm, all of which got errors because of the
> > device's bogus descriptors.  Probably one of them is messing up the
> > refcount.
> 
> Hi,
> 
> you made me look at cdc-acm. I suspect
> 
> cae2bc768d176bfbdad7035bbcc3cdc973eb7984 ("usb: cdc-acm: Decrement tty port's refcount if probe() fail")
> 
> is buggy decrementing the refcount on the interface in destroy()
> even before the refcount is increased.
> 
> Unfortunately I cannot tell from the bug report how many and which
> interfaces the emulated test device has. Hence it is unclear to me,
> when exactly probe() would fail cdc-acm.

Only one interface (numbered 234!).

> If you agree. I am attaching a putative fix.

Your patch adds a line saying:

> +	usb_get_intf(acm->control); /* undone in destroy() */

but I don't see any destroy() function in that source file.  Did you 
mean acm_port_destruct()?

In any case, I don't know if this missing "get" would cause the 
problem, but it might well.

Alan Stern