Re: [PATCH nft v2 2/2] Handle retriable errors from mnl functions

[Eugene Crosser <crosser@average.org>]

[-- Attachment #1.1: Type: text/plain, Size: 1098 bytes --]

On 27/01/2022 19:20, Pablo Neira Ayuso wrote:
> On Thu, Dec 09, 2021 at 07:26:07PM +0100, Eugene Crosser wrote:
>> rc == -1 and errno == EINTR mean:
>>
>> mnl_socket_recvfrom() - blindly rerun the function
>> mnl_cb_run()          - restart dump request from scratch
>>
>> This commit introduces handling of both these conditions
> 
> Sorry it took me a while to come back to this.
> 
> https://patchwork.ozlabs.org/project/netfilter-devel/patch/20220127181835.571673-1-pablo@netfilter.org/
> 
> This follows the same approach as src/mnl.c, no need to close the
> reopen the socket to drop the existing messages.

Thanks for getting back to it and producing the fix!

I think it is slightly less clean, because if some (not EINTR) error happens
while it is draining the queue (the second call to `mnl_socket_recvfrom()` with
`eintr == true`), `errno` will be overwritten and the error misrepresented. This
should not be a _practical_ problem because presumably the same error will be
raised upon retry, and this time it will be reported correctly.

Thanks again,

Eugene

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 488 bytes --]