Re: Stable backport request: skbuff: Proactively round up to kmalloc bucket size

[Kees Cook <kees@kernel.org>]

On May 23, 2023 8:52:53 PM PDT, "Daniel Díaz" <daniel.diaz@linaro.org> wrote:
>Hello!
>
>On Tue, 23 May 2023 at 00:28, Vlastimil Babka <vbabka@suse.cz> wrote:
>> On 5/22/23 20:23, Daniel Díaz wrote:
>> > Hello!
>> >
>> > Would the stable maintainers please consider backporting the following
>> > commit to the 6.1? We are trying to build gki_defconfig (plus a few
>> > extras) on Arm64 and test it under Qemu-arm64, but it fails to boot.
>> > Bisection has pointed here.
>>
>> You mean the bisection was done to find the first "good" commit between 6.1
>> and e.g. 6.3?
>>
>> As others said, this commit wasn't expected to be a fix to a known bug.
>> Maybe you found one that we didn't know of, or it might be accidentaly
>> masking some other bug.
>
>How interesting! Yes, we happened to run a bisection between v6.1 and
>v6.3 and we found where it started working with the following
>configuration:
>  https://storage.tuxsuite.com/public/linaro/daniel/builds/2QA2CHQUpqKe27FyMZrBNILVwXi/config

Ah yes, from CONFIG_UBSAN_BOUNDS=y and CONFIG_UBSAN_TRAP=y

This was a known issue in upstream, oddly only exposed on arm64. Something re-broke with __alloc_size after commit 93dd04ab0b2b had tried to work around it. I didn't think any kernel released with this broken, though, so perhaps what broke it got added to -stable?

>With that patch on top of v6.1.29 it boots fine under Qemu-arm64; as
>v6.1.y stands, it panics with this:

It should be fine to backport the patch, IMO.


-- 
Kees Cook