linux-kernel - Re: [PATCH v1 1/1] Input: ALPS - bail out when device path can't fit buffer

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: <2blm4mirspwgcukwnybfgqhiozhtkcqjl2e7g2onxp6ms4ex4a@l4jayj4i6fti>
Date: Mon, 5 May 2025 22:34:56 -0700
From: Dmitry Torokhov <dmitry.torokhov@...il.com>
To: Andy Shevchenko <andriy.shevchenko@...ux.intel.com>
Cc: linux-input@...r.kernel.org, linux-kernel@...r.kernel.org, 
	Pali Rohár <pali@...nel.org>
Subject: Re: [PATCH v1 1/1] Input: ALPS - bail out when device path can't fit
 buffer

On Fri, May 02, 2025 at 04:52:51PM +0300, Andy Shevchenko wrote:
> On Tue, Apr 29, 2025 at 08:01:10AM +0300, Andy Shevchenko wrote:
> > Mon, Apr 28, 2025 at 04:30:13PM -0700, Dmitry Torokhov kirjoitti:
> > > On Tue, Apr 22, 2025 at 09:56:45PM +0300, Andy Shevchenko wrote:
> 
> ...
> 
> > > > +		n = snprintf(priv->phys2, sizeof(priv->phys2), "%s/input1",
> > > > +			     psmouse->ps2dev.serio->phys);
> > > > +		if (n >= sizeof(priv->phys2)) {
> > > > +			psmouse_err(psmouse,
> > > > +				    "failed to prepare path to the trackstick device\n");
> > > > +			error = -E2BIG;
> > > > +			goto init_fail;
> > > 
> > > So you just broke touchpad of some poor guy who had it working just fine 
> > > for many years. For maximum impact you should add BUG() or panic()
> > > here.
> > 
> > Ha-ha. You know that your speculation most likely so far from the truth.

If your code is not a noop that is precisely what happened.

> 
> And actually what you are telling about is not true at all. If the device was
> working it means that the file node name is not cut, and hence this patch won't
> anyhow change this behaviour. Otherwise, provide an example which can fail this
> and still be working in the user space.

"phys" is not a name of a device node. It is a string available via
/proc/bus/input/devices, sysfs /sys/class/input/input<N>/phys and also
EVIOCGPHYS ioctl. A driver is free to not set it at all and everything
will be working fine.

Actually, input devices themselves to not have device nodes, it is evdev
interface that provides /dev/input/event<N>.

> 
> > > In all seriousness, it is OK to have truncated phys, rarely anyone looks
> > > at it and if we get a report of it being truncated then we can consider
> > > addressing the size (or we can decide to live with it truncated).
> > 
> > In all seriousness, while I agree on the statement, the 4 drivers in Input
> > subsystem break the build. It's the biggest obstacle now to enable WERROR=y,
> > which is default, builds on `make W=1`. So, I already gave you chance to fix,
> > instead I hear nothing back for a months (to be precise 2 months and a day
> > passed from my first attempt that you didn't like), the problem still exists.
> > Please, address this the way you like.
> 
> For the reference, the first approach:
> https://lore.kernel.org/r/20250228121147.242115-1-andriy.shevchenko@linux.intel.com
> where I also asked about this one, ano got no answer.

Sorry I was busy with other projects.

> 
> I really don't want to try anything new as it seems a big pushback to whatever
> I propose. So, please consider fixing the issues rather sooner. I will be more
> than happy to test.

Have you considered that this warning is bogus and it should be disabled
instead? Or maybe GCC should see if there are followup writes to the
same buffer before emitting the warning? 

Thanks.

-- 
Dmitry