linux-kernel - Re: [PATCH v2] usb: gadget: Fix double free of device descriptor pointers

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]

Message-ID: <87h7jvlsg9.fsf@kernel.org>
Date:   Sat, 24 Apr 2021 16:31:34 +0300
From:   Felipe Balbi <balbi@...nel.org>
To:     Wesley Cheng <wcheng@...eaurora.org>, gregkh@...uxfoundation.org,
        peter.chen@...nel.org
Cc:     linux-usb@...r.kernel.org, linux-kernel@...r.kernel.org,
        Hemant Kumar <hemantk@...eaurora.org>, stable@...r.kernel.org
Subject: Re: [PATCH v2] usb: gadget: Fix double free of device descriptor
 pointers


Hi,

Wesley Cheng <wcheng@...eaurora.org> writes:
> On 4/24/2021 1:05 AM, Felipe Balbi wrote:
>> 
>> Hi,
>> 
>> Wesley Cheng <wcheng@...eaurora.org> writes:
>>>>>> From: Hemant Kumar <hemantk@...eaurora.org>
>>>>>>
>>>>>> Upon driver unbind usb_free_all_descriptors() function frees all
>>>>>> speed descriptor pointers without setting them to NULL. In case
>>>>>> gadget speed changes (i.e from super speed plus to super speed)
>>>>>> after driver unbind only upto super speed descriptor pointers get
>>>>>> populated. Super speed plus desc still holds the stale (already
>>>>>> freed) pointer. Fix this issue by setting all descriptor pointers
>>>>>> to NULL after freeing them in usb_free_all_descriptors().
>>>>>
>>>>> could you describe this a little better? How can one trigger this case?
>>>>> Is the speed demotion happening after unbinding? It's not clear how to
>>>>> cause this bug.
>>>>>
>>>> Hi Felipe,
>>>>
>>>> Internally, we have a mechanism to switch the DWC3 core maximum speed
>>>> parameter dynamically for displayport use cases.  This issue happens
>>>> whenever we have a maximum speed change occur on the USB gadget, which
>>>> for DWC3 happens whenever we call gadget init.  When we switch in and
>>>> out of host mode, gadget init is being executed, leading to the change
>>>> in the USB gadget max speed parameter:
>>>>
>>>> dwc->gadget->max_speed		= dwc->maximum_speed;
>>>>
>>>> I know that configFS gadget has the max_speed sysfs file, which is a
>>>> similar mechanism, but I haven't tried to see if we can reproduce the
>>>> same issue with it.  Let me see if we can reproduce this with that
>>>> configfs speed setting.
>>>>
>>>> Thanks
>>>> Wesley Cheng
>>>>
>>>
>>> Hi Felipe,
>>>
>>> So I tried with doing it through the configFS max_speed, but it doesn't
>>> have the same effect, as the setting done in dwc3_gadget_init() will
>>> still be assigning the composite/UDC device's maximum speed to SSP/SS.
>>> This is what the usb_assign_descriptor() uses to determine whether or
>>> not to copy the SSP and SS descriptors.
>>>
>>> So in summary, at least for a DWC3 based subsystem, the only way to
>>> reproduce it is if there is a way to dynamically switch the DWC3 core
>>> max speed parameter.
>> 
>> Could it be that you have a bug in your out-of-tree changes? Perhaps
>> there's some assumption which your changes aren't guaranteeing.
>> 
> Hi Felipe,
>
> Unless there is a limitation on how the USB gadget max speed can be
> used, i.e. the USB gadget max speed MUST stay the same throughout a
> device's boot period, then our out of tree changes may have the wrong

no, not throughout boot, but it can't change while the device is
enumerated.

> assumptions.  However, I don't see that stated anywhere, and I still
> feel the current usb_assign_descriptors() and usb_free_all_descriptors()
> aren't aligned with one another.  One API decides which descriptors to
> copy based on a parameter, whereas the other just frees all of them
> irrespective.

that's a valid point.

-- 
balbi

Download attachment "signature.asc" of type "application/pgp-signature" (858 bytes)