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 for Android: free password hash cracker in your pocket
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20200330172048.GA12976@madhuparna-HP-Notebook>
Date:   Mon, 30 Mar 2020 22:50:48 +0530
From:   Madhuparna Bhowmik <madhuparnabhowmik10@...il.com>
To:     Alan Stern <stern@...land.harvard.edu>
Cc:     madhuparnabhowmik10@...il.com, gregkh@...uxfoundation.org,
        hariprasad.kelam@...il.com, colin.king@...onical.com,
        tony.olech@...ndigitalsystems.com, linux-usb@...r.kernel.org,
        linux-kernel@...r.kernel.org, andrianov@...ras.ru
Subject: Re: Possible data-race related bug in u132_hcd module.

On Mon, Mar 30, 2020 at 12:03:31PM -0400, Alan Stern wrote:
> On Mon, 30 Mar 2020 madhuparnabhowmik10@...il.com wrote:
> 
> > Hi,
> > 
> > This bug is found by  Linux Driver Verification project (linuxtesting.org).
> > 
> > The bug is related to the parallel execution of u132_probe() function
> > and u132_hcd_exit() function in u132_hcd.c. In case the module is
> > unloaded when the probe function is executing there can be data race
> > as the mutex lock u132_module_lock is not used properly. 
> 
> Normally drivers do not have to worry about races between their probe 
> and exit routines.  The exit routine should unregister the driver from 
> its bus subsystem, and unregistration is supposed to wait until all 
> probe and remove functions have finished executing.
> 
> > i) Usage of mutex lock only when writing into the u132_exiting
> > variable in u132_hcd_exit(). The lock is not used when this variable
> > is read in u132_probe().
> 
> I'm not familiar with u132_hcd, but the probe routine shouldn't need to 
> use and "exiting" variable at all.
>
Even I am not sure why it should use this variable to check, that's why
I thought of asking in the mailing list. If the maintainers agree that
we can remove this variable I can send a patch doing it. This variable
is not used for any other purpose in the module, so removing it
shouldn't be a problem.
> > 
> > Moreover, this variable does not serve its purpose, as even if
> > locking is used while the u132_exiting variable is read in probe(),
> > the function may still miss that exit function is executing if it
> > acquires the mutex before exit() function does.
> > 
> > How to fix this?
> 
> Are you certain there really is a problem?
> 
If the variable u132_exiting is really used for the purpose of checking
if the module is exiting then there might be a problem. If at all it is
assumed that exit never races with probe function and it is always
called after probe finishes or something, then this variable is not even
required. And suppose there is possibility of a race condition then only
holding the mutex in exit but not probe does not make sense.

> > ii) Usage of mutex while adding entries in u132_static_list in probe
> > function but not in exit function while unregistering.
> > This should be easy to fix by holding the mutex in the exit function as well.
> 
> Why does the driver need a static list?
>
I do not know much about this module so I cannot answer this.
>From the point of synchronization, this list is initialized in init and
then only used in probe and exit function. And lock is only held in
probe, if at all it is assumed that exit cannot race with probe, there
is no need to use the mutex.

> > There can be other synchronization problems related to the usage of
> > u132_module_lock in this module, I have only spotted these so far.
> 
> You should look at other drivers for comparison.  They don't have to 
> face this kind of problem.  u132_hcd should be similar to them.
>
Yes, I checked out a few others and I could not see any usage of such
variable to check if module is exiting.

Thank you,
Madhuparna

> Alan Stern
> 
> 

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ