On Fri, 22 May 2020 01:32:00 +0200, Changming Liu wrote:
Hi Jaroslav, Takashi: Greetings, I'm a first year PhD student who is interested in using UBSan for linux. After some experiments, I found that in sound/core/hwdep.c function snd_hwdep_dsp_load there might be an undefined behavior that might cause unexpected result.
More specifically, in this function,info was fetched from user space and, info.index was checked if it's greater than or equal to 32. If not then it's used as number of left shift bits to string literal 1.
The problem is, since string literal 1 is by default signed int, so 1 << 31 cannot be represented as a valid integer and the result might be undefined across different platforms. So I guess change 1 to 1U might help?
Yes, that should work.
Due to the lack of knowledge of the interaction between this module and others, I'm not able to assess if this is critical or worth fixing. I'd appreciate if for your comment on this bug. This can help me understand UB a lot!
I don't think this is any serious problem. It's a bit flag that just indicates the status and the flag itself has any influence on the behavior of the rest in kernel. And, the hwdep DSP load feature itself is used very rarely, only by a couple of drivers. So it's pretty much never hit.
That said, we can simply fix the issue at any time, and it's "just to be sure". Could you submit a fix patch in the proper format?
thanks,
Takashi