linux-kernel - Re: [PATCH v3 2/3] zram: fix deadlock with sysfs attribute usage and driver removal

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] [day] [month] [year] [list]

Date:   Tue, 22 Jun 2021 12:57:01 -0700
From:   Luis Chamberlain <mcgrof@...nel.org>
To:     Greg KH <gregkh@...uxfoundation.org>
Cc:     minchan@...nel.org, jeyu@...nel.org, ngupta@...are.org,
        sergey.senozhatsky.work@...il.com, axboe@...nel.dk,
        mbenes@...e.com, jpoimboe@...hat.com, tglx@...utronix.de,
        keescook@...omium.org, jikos@...nel.org, rostedt@...dmis.org,
        peterz@...radead.org, linux-block@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH v3 2/3] zram: fix deadlock with sysfs attribute usage and
 driver removal

On Tue, Jun 22, 2021 at 08:05:12PM +0200, Greg KH wrote:
> On Tue, Jun 22, 2021 at 10:27:12AM -0700, Luis Chamberlain wrote:
> > On Tue, Jun 22, 2021 at 07:16:34PM +0200, Greg KH wrote:
> > > On Tue, Jun 22, 2021 at 09:32:08AM -0700, Luis Chamberlain wrote:
> > > > On Tue, Jun 22, 2021 at 09:45:39AM +0200, Greg KH wrote:
> > > > > On Mon, Jun 21, 2021 at 04:36:34PM -0700, Luis Chamberlain wrote:
> > > > > > @@ -2048,13 +2048,19 @@ static ssize_t hot_add_show(struct class *class,
> > > > > >  {
> > > > > >  	int ret;
> > > > > >  
> > > > > > +	if (!try_module_get(THIS_MODULE))
> > > > > > +		return -ENODEV;
> > > > > > +
> > > > > 
> > > > > You can not increment/decrement your own module's reference count and
> > > > > expect it to work properly, as it is still a race.
> > > > 
> > > > The goal here is to prevent an rmmod call if this succeeds. If it
> > > > succeeds then any subsequent rmmod will fail. Can you explain how
> > > > this is still racy?
> > > 
> > > {sigh}
> > > 
> > > What happens if the driver core is just about to call hot_add_show() and
> > > the module is removed from the system.  It then calls to the memory
> > > location that hot_add_show() was previously at, but now that is not a
> > > valid pointer to code, and boom.
> > 
> > The new kobject_get() on patch 3/3 ensures that the device will be up
> > throughout the entire life of the store call, and thus prevent the
> > code being executed being removed, no?
> 
> I do not know, I no longer remember what is in that patch at the moment
> as it is long-gone from my queue.

It was the changes *you* recommended, a generic way to ensure the
lifetime of the derefernce is valid. I had used bdgrab()/bdget() and you
suggested we generalize it with the kobject_get() for the device and a
bus get. With that change, I confirm that the device will still be
present during the lifetime of the sysfs knobs call.

> Also, if the device will be "up" for the whole lifetime, why do you need
> to increment the module reference count?

The goal is to prevent a deadlock. The lifetime of the device is not
an issue in this deadlock case, the issue is a race with module removal
and that code path using a lock which is also used on a sysfs knob.

  Luis