Late ACKs from torn-down HTTP connections



  • I noticed several entries like this in /var/log/filter.log:

    000077 rule 69/0(match): block in on fxp0: 209.85.171.165.80 > x.x.x.x.54337: F 0:0(0) ack 1 win 8190
    7. 971085 rule 69/0(match): block in on fxp0: 66.102.7.104.80 > x.x.x.x.53173: F 0:0(0) ack 1 win 8190

    grep 'fxp0..80.: F' /var/log/filter.log | wc -l
          20

    I think these are related to late ACKs – mostly from torn-down HTTP connections.  PF (and IPF) sometimes expire connections faster than the other end can tear them down, resulting in erroneous block msgs which clutter the logs.

    A long time ago, this was related to a known problem with IPF state code.

    http://www.sigmasoft.com/~openbsd/archive/openbsd-misc/199912/msg00906.html

    I remember PF having the same issue when PF first came out.  Maybe PF still has the issue.  (SonicWall firewalls have a similar problem -- except that they send irritating e-mails accusing you of trying to hax0r them whenever this event occurs.)

    As the above post suggests, it's easily solved with:

    block in    quick on $wan inet proto tcp from any port = 80 to any port > 1023 flags F/F
    block in    quick on $wan inet proto tcp from any port = 80 to any port > 1023 flags R/R

    However, the pfsense WebGUI doesn't let you create rules which refer to specific flags.

    Are these two rules candidates for inclusion into the default pfsense config?  Or perhaps a future version of the GUI could allow us to reference TCP flags?



  • Try a recent RELENG_1_2 snapshot which has the tcp.established timeout parameter fixed.  You might be able to up this value a bit.



  • I just noticed RELENG_1 snapshots seem to have been deprecated.  Should we start moving to 1_2?



  • Yep.



  • How/where do I access the tcp.established timeout parm?  Is it "State Timeout in seconds" in the Advanced Options of any rule?



  • @rcarr:

    How/where do I access the tcp.established timeout parm?  Is it "State Timeout in seconds" in the Advanced Options of any rule?

    Yep, that's the one.


Log in to reply