Re: [BUG] hwmon: Widespread TOCTOU vulnerabilities in the hwmon subsystem

[Guenter Roeck <linux@roeck-us.net>]

On 6/6/25 14:30, Armin Wolf wrote:
> Am 06.06.25 um 09:03 schrieb Gui-Dong Han:
> 
>>> On Thu, Jun 05, 2025 at 07:33:24AM -0700, Guenter Roeck wrote:
>>>>> I would like to discuss these issues further and collaborate on the
>>>>> best way to address them comprehensively.
>>>>>
>>>> I'd suggest to start submitting patches, with the goal of minimizing
>>>> the scope of changes. Sometimes that may mean expanding the scope of
>>>> locks, sometimes it may mean converting macros to functions. When
>>>> converting to functions, it doesn't have to be inline functions: I'd
>>>> leave that up to the compiler to decide. None of that code is performance
>>>> critical.
>>>>
>>> Actualy, that makes me wonder if it would make sense to introduce
>>> subsystem-level locking. We could introduce a lock in struct
>>> hwmon_device_attribute and lock it whenever a show or store function
>>> executes in drivers/hwmon/hwmon.c. That would only help for drivers
>>> using the _with_info API, but it would simplify driver code a lot.
>>> Any thoughts on that ?
> 
> Hi,
> 
> i am against adding a subsystem lock just to work around buggy drivers. Many drivers
> should use fine-grained locking to avoid high latencies when reading a single attribute.
> 

The point would be driver code simplification and increasing robustness, not
working around buggy drivers.

Anyway, what high latency are you talking about ? Serialization of attribute
accesses would only increase latency if multiple processes read attributes from
the same driver at the same time, which is hardly a typical use case.

Guenter