[LTP] [PATCH 1/1] net/broken_ip: Rewrite into new API
Alexey Kodanev
alexey.kodanev@oracle.com
Mon May 6 16:32:20 CEST 2019
Hi Petr,
On 02.05.2019 14:44, Petr Vorel wrote:
> Hi Alexey,
>
>> Acked-by: Alexey Kodanev <alexey.kodanev@oracle.com>
> ...
>>> +export ICMPV4_DATA_MAXSIZE=1472
>>> +export ICMPV6_DATA_MAXSIZE=1452
>
>> Since these variables are in the library, may be it would be better to tune
>> them depending on an MTU of the tests interfaces?
> Good idea. 1472 and 1452 are based on MTU = 1500 (IPv4: 1500 - 20 bytes for IPv4 header
> - 8 bytes for ICMP header; IPv6: 1500 - 40 bytes for IPv6 header - 8 bytes for ICMP header),
> which looks to me as the default for netns NICs and ethernet and wifi NICs,
> but we should avoid failures when different MTU used.
> I propose merging this version and add MTU detection support later.
>
OK. Perhaps we should rename them to indicate that they relate to ns-icmpv_sender,
because the maximum icmp data size is 65535 - headers, not those values.
> I guess we don't need to check -s values passed to tst_icmp() (I guess nobody
> set MTU < 1000, although it's not forbidden for IPv4). But I'd add this check
> also into ns-icmpv[46]_sender.c.
>
> + off course other cleanup like merging ns-icmpv[46]_sender.c into single code would be nice,
> but first I want to finish route rewriting.
>
>> Also we can send more with fragmentation, but I guess ns-icmpv doesn't
>> have proper support...
> Do you mean to send packet big enough to have fragmentation on lhost?
> + setsockopt IP_PMTUDISC_DONT?
>
> Kind regards,
> Petr
>
More information about the ltp
mailing list