Message-ID: <CAMSo37VxLfqg7UGY8CYAYz8nqoJ12T2mERbKNn67e_+yOD3cXQ@mail.gmail.com>
Date:   Thu, 20 Apr 2023 14:30:57 +0800
From:   Yongqin Liu <yongqin.liu@...aro.org>
To:     Zheng Hacker <hackerzheng666@...il.com>
Cc:     Greg KH <gregkh@...uxfoundation.org>,
        John Stultz <jstultz@...gle.com>,
        Zheng Wang <zyytlz.wz@....com>,
        Sumit Semwal <sumit.semwal@...aro.org>, arnd@...db.de,
        linux-kernel@...r.kernel.org, 1395428693sheep@...il.com,
        alex000young@...il.com, Mauro Carvalho Chehab <mchehab@...nel.org>
Subject: Re: [PATCH] misc: hisi_hikey_usb: Fix use after free bug in
 hisi_hikey_usb_remove due to race condition

Hi, Zheng

BTW, I just see cancel_delayed_work_sync is used in
the drivers/usb/common/usb-conn-gpio.c usb_conn_remove function.
    https://android.googlesource.com/kernel/common/+/refs/heads/android-mainline/drivers/usb/common/usb-conn-gpio.c#274

I know nothing about the cancel_delayed_work_sync and cancel_work_sync
functions,
just for your information in case cancel_delayed_work_sync might be
better to be used here.

Thanks,
Yongqin Liu
On Tue, 18 Apr 2023 at 21:18, Zheng Hacker <hackerzheng666@...il.com> wrote:
>
> Yongqin Liu <yongqin.liu@...aro.org> 于2023年4月18日周二 01:31写道：
> >
> > Hi, Zheng
> >
> > Sorry for the late reply.
> >
> > I tested this change with Android build based on the ACK
> > android-mainline branch.
> > The hisi_hikey_usb module could not be removed with error like this:
> >     console:/ # rmmod -f hisi_hikey_usb
> >     rmmod: failed to unload hisi_hikey_usb: Try again
> >     1|console:/ #
> > Sorry I am not able to reproduce any problem without this commit,
> > but I do not see any problem with this change applied either.
> >
> > If there is any specific things you want to check, please feel free let me know
> >
>
> Hi Yongqin,
>
> Thanks for your testing. I have no more questions about the issue.
>
> Best regards,
> Zheng
>
> > Thanks,
> > Yongqin Liu
> >
> >
> > On Fri, 14 Apr 2023 at 00:46, Zheng Hacker <hackerzheng666@...il.com> wrote:
> > >
> > > Greg KH <gregkh@...uxfoundation.org> 于2023年4月13日周四 23:56写道：
> > > >
> > > > On Thu, Apr 13, 2023 at 11:35:17PM +0800, Zheng Hacker wrote:
> > > > > Greg KH <gregkh@...uxfoundation.org> 于2023年4月13日周四 20:48写道：
> > > > > >
> > > > > > On Thu, Apr 13, 2023 at 07:12:07PM +0800, Zheng Hacker wrote:
> > > > > > > Yongqin Liu <yongqin.liu@...aro.org> 于2023年4月13日周四 18:55写道：
> > > > > > > >
> > > > > > > > Hi, Zheng
> > > > > > > >
> > > > > > > > On Thu, 13 Apr 2023 at 16:08, Zheng Hacker <hackerzheng666@...il.com> wrote:
> > > > > > > > >
> > > > > > > > > Friendly ping about the bug.
> > > > > > > >
> > > > > > > > Sorry, wasn't aware of this message before,
> > > > > > > >
> > > > > > > > Could you please help share the instructions to reproduce the problem
> > > > > > > > this change fixes?
> > > > > > > >
> > > > > > >
> > > > > > > Hi Yongqin,
> > > > > > >
> > > > > > > Thanks for your reply. This bug is found by static analysis. There is no PoC.
> > > > > > >
> > > > > > > >From my personal experience, triggering race condition bugs stably in
> > > > > > > the kernel needs some tricks.
> > > > > > > For example, you can insert some sleep-time code to slow down the
> > > > > > > thread until the related object is freed.
> > > > > > > Besides, you can use gdb to control the time window. Also, there are
> > > > > > > some other tricks as [1] said.
> > > > > > >
> > > > > > > As for the reproduction, this attack vector requires that the attacker
> > > > > > > can physically access the device.
> > > > > > > When he/she unplugs the usb, the remove function is triggered, and if
> > > > > > > the set callback is invoked, there might be a race condition.
> > > > > >
> > > > > > How does the removal of the USB device trigger a platform device
> > > > > > removal?
> > > > >
> > > > > Sorry I made a mistake. The USB device usually calls disconnect
> > > > > callback when it's unpluged.
> > > >
> > > > Yes, but you are changing the platform device disconnect, not the USB
> > > > device disconnect.
> > > >
> > > > > What I want to express here is When the driver-related device(here
> > > > > it's USB GPIO Hub) was removed, the remove function is triggered.
> > > >
> > > > And is this a patform device on a USB device?  If so, that's a bigger
> > > > problem that we need to fix as that is not allowed.
> > >
> > > No this is not a platform  device on a USB device.
> > >
> > > >
> > > > But in looking at the code, it does not seem to be that at all, this is
> > > > just a "normal" platform device.  So how can it ever be removed from the
> > > > system?  (and no, unloading the driver doesn't count, that can never
> > > > happen on a normal machine.)
> > > >
> > >
> > > Yes, I finally figured out your meaning. I know it's hard to unplug
> > > the platform device
> > > directly. All I want to express is that it's a theoretical method
> > > except  rmmod. I think it's better to fix the bug. But if the
> > > developers think it's practically impossible, I think there's no need
> > > to take further action.
> > >
> > > Sorry for wasting your time and thanks for your explanation.
> > >
> > > Best regards,
> > > Zheng
> > >
> > > > thanks,
> > > >
> > > > greg k-h
> >
> >
> >
> > --
> > Best Regards,
> > Yongqin Liu
> > ---------------------------------------------------------------
> > #mailing list
> > linaro-android@...ts.linaro.org
> > http://lists.linaro.org/mailman/listinfo/linaro-android



-- 
Best Regards,
Yongqin Liu
---------------------------------------------------------------
#mailing list
linaro-android@...ts.linaro.org
http://lists.linaro.org/mailman/listinfo/linaro-android

Powered by blists - more mailing lists