Skip to contentSkip to conversation

PF_PACKET: packets out of order

DISCUSSION Joerg Schumacher 2 participants 3 messages Oct 8, 2001 – Oct 9, 2001

lore
Share & tools
lore
#0 lore raw

Hi!

NeTraMet v44b10 uses pcap(3) and complains about timestamps jumping backwards. Looks like a PF_PACKET socket doesn't receive the packets in the correct order. Some timestamps from a "tcpdump -tt":

   RX:  1001465480.175100 [...] 
   TX:  1001465480.179111 [...] 
   RX:  1001465480.177315 [...] 
                   ^^^^^^

TX: 1001465480.180514 [...]
RX: 1001465480.179706 [...]

Some more figures (obtained with a quick hack using libpcap-2001.09.25 on two different machines):

2.2.19:
stats: 100000 packets received
0 packets dropped
good: 90222 packets
bad: 9778 packets
max delta: 15850 usec

2.4.1:
stats: 100000 packets received
0 packets dropped
good: 99538 packets
bad: 462 packets
max delta: 471 usec

Any plans to fix this in the kernel?

Regards,
Joerg

Signature · 3 lines
 Gaertner Datensysteme                         38114 Braunschweig 
 Joerg Schumacher                              Hamburger Str. 273a
 Tel: 0531-2335555                             Fax: 0531-2335556
#1 lore raw

From: Joerg Schumacher <[email protected]>
Date: Tue, 9 Oct 2001 01:32:16 +0200 (MET DST)

Hi!

NeTraMet v44b10 uses pcap(3) and complains about timestamps jumping backwards. Looks like a PF_PACKET socket doesn't receive the packets in the correct order. Some timestamps from a "tcpdump -tt":

      RX:  1001465480.175100 [...] 
      TX:  1001465480.179111 [...] 
      RX:  1001465480.177315 [...] 
                      ^^^^^^
      TX:  1001465480.180514 [...]
      RX:  1001465480.179706 [...]

The receive packets are in order, as are the transmit packets.

Anything which absolutely _requires_ all TX and RX packets to be in precise order, should really be fixed not to have such a weird restriction.

Franks a lot,
David S. Miller
[email protected]

#2 lore raw
In reply to Joerg Schumacher
Quoted 4 lines · David S. Miller

[...]
Anything which absolutely _requires_ all TX and RX packets to be in precise order, should really be fixed not to have such a weird restriction.

Do humans qualify as "anything" ;-)? Until I saw the "LastTime jump backward" warning in netramet I used to read the tcpdump output line
by line without checking the timestamps. Neither packet(7) nor
tcpdump(1) indicate that I should do so.

Regards,
Joerg.

Signature · 3 lines
 Gaertner Datensysteme                         38114 Braunschweig 
 Joerg Schumacher                              Hamburger Str. 273a
 Tel: 0531-2335555                             Fax: 0531-2335556