On 10/17/2019 12:52 PM, Jakub Grajciar -X (jgrajcia - PANTHEON TECHNOLOGIES at
Cisco) wrote:
>
>> Hi Jakub,
>>
>> Just to double check if anyone is looking into the bind() error issue which
>> is
>> since following commit, I am waiting for more input on it.
>>
>> Commit b923866c6974 ("net/memif: allow for full key size in socket name")
>> Cc: step...@networkplumber.org
>
> Definitely an issue, I must have messed something up while applying the patch
> for testing, as it's not working for me either. I looked into bind() and it
> seems that the size has upper limit. Are you able to revert the patch?
+1 to logically revert the patch, but actually I think it is easier to do an
incremental patch on top of current head to fix the issue.
The patch replaces hard-coded 'key_len' to a macro which is good to keep, also
number of lines needs to be changed for fix looks less than number of lines will
be changed by revert J
>
> As for the issue that patch addresses, there is an incorrect information in
> the docs that 'socket' parameter length is 256b. It should be 108b as that is
> the limitation on Linux.
I don't know the doc but code has it as 256b [1] and the issue patch addresses
looks like a valid issue. If you are OK to limit the 'key_len' to 108 [2], the
fix is really easy.
[1]
http://lxr.dpdk.org/dpdk/v19.08/source/drivers/net/memif/memif_socket.h#L84
[2]
If I remember correctly the suggested length was 99 for portability, it seems
different OS has this value different and 99 is the smallest ...
