[dpdk-dev] [PATCH] test/hash: fix buffer overflow

[Medvedkin, Vladimir]

Hi David,

On 14/10/2021 09:04, David Marchand wrote:
> Hello Vladimir,
> 
> On Wed, Oct 13, 2021 at 9:27 PM Medvedkin, Vladimir
> <vladimir.medvedkin at intel.com> wrote:
>>> With patch applied, ASan reports another issue.
>>> Did you test your fix with ASan?
>>>
>>
>> You're right, for some reason ASAN wasn't enabled.
>> I applied patch and built running .ci/linux-build.sh,
>> also I build with CFLAGS + LDFLAGS.
>>
>> Bruce suggested to use meson options instead of using CFLAGS, so
>> meson configure build -Db_sanitize=address -Db_lundef=false
>> works fine.
> 
> Well, yes, you can directly do this.
> I linked to my GHA patch in the bz, because I find it easier and
> reproducible to push fixes in GHA and get the result: no question
> about "did I enable ASan?" or "did I start the test correctly?".
> 
> FYI, b_lundef seems necessary only with clang, gcc should be fine without it.
> IIUC, those compilers went with different choices on how to pull
> libasan (clang went with static, gcc went with shared).
> Hopefully, we will have something easier to use in DPDK with Zhihong work.
> 

Thanks!

>>
>> I'll sent v2 for this.
> 
> Thanks, I'll look at it.
> 

I'm going to send v3, because just dividing the size of the key for 
jhash_32b() cases is not correct (because rte_hash will compare just a 
part of the key in this case), so I'll replace rte_jhash_32b with a 
wrapper function.

> 

-- 
Regards,
Vladimir