Re: [Qemu-devel] [PATCH 25/41] slirp: Make IP packet ID consistent

[Jan Kiszka]

Filip Navara wrote:
> On Wed, Jun 24, 2009 at 2:42 PM, Jan Kiszka<address@hidden> wrote:
>> Currently, ip_id is always initialized to 0 on slirp startup (despite
>> the broken attempt to derive it from the clock). This is good for
>> reproducibility. But it is not preserved across save/restore. This patch
>> therefore drops the dead initialization code from ip_init and introduces
>> ip_id to the persistent slirp state.
>>
>> Signed-off-by: Jan Kiszka <address@hidden>
>> ---
>>
>>  slirp/ip_input.c |    1 -
>>  slirp/slirp.c    |    8 +++++++-
>>  2 files changed, 7 insertions(+), 2 deletions(-)
>>
>> diff --git a/slirp/ip_input.c b/slirp/ip_input.c
>> index 0356eb5..b07d3d5 100644
>> --- a/slirp/ip_input.c
>> +++ b/slirp/ip_input.c
>> @@ -59,7 +59,6 @@ void
>>  ip_init(void)
>>  {
>>        ipq.ip_link.next = ipq.ip_link.prev = &ipq.ip_link;
>> -       ip_id = tt.tv_sec & 0xffff;
> 
> You removed the ip_id initialization and now it's never initialized in
> the code. That sounds wrong.

Thanks for having a look! All this slirp code is really hairy and no fun
to dig through.

But now to your remark: That removal is in fact not changing the
behavior. ip_id is also initialized to 0 afterwards. To understand this
one has to track the messy use of 'tt' across slirp. It is in fact
carrying the time at some point (when select_fill is invoked), but not
yet on ip_init.

I was thinking about fixing the above initialization to actually take
the current time and derive ip_id, but then I thought it might be better
to keep this for network traffic reproducibility across qemu starts.

Jan

-- 
Siemens AG, Corporate Technology, CT SE 2
Corporate Competence Center Embedded Linux