Netgate Discussion Forum
    • Categories
    • Recent
    • Tags
    • Popular
    • Users
    • Search
    • Register
    • Login

    Ipv6 unusable due lack of love from FreeBSD (prev: Support baby jumbo frames)

    Scheduled Pinned Locked Moved
    General pfSense Questions
    7
    50
    9.6k
    Loading More Posts
    • Oldest to Newest
    • Newest to Oldest
    • Most Votes
    Reply
    • Reply as topic
    Log in to reply
    This topic has been deleted. Only users with topic management privileges can see it.
    • stephenw10S
      stephenw10 Netgate Administrator
      last edited by

      Well we can agree that with IPv6 the behaviour is certainly not as expected (or as desired as least).
      However in IPv4 that's exactly the expected behaviour. There's not a bug there, a missing feature perhaps.

      Steve

      1 Reply Last reply Reply Quote 0
      • M
        M_Devil
        last edited by

        Yes, so

        • Baby jumbo frames is about missing feature without having a big problem now.

        • But as Doktornotor points out, there is a problem with ipv6 (https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=172648) breaking normal ipv6 usage.

        As in linked post, there are two workarounds:

        • don't use 'scrub reassemble tcp' in PF, or disable PF
        • sysctl net.inet.tcp.rfc1323=0

        Trying second now and seems (for now) it is working

        1 Reply Last reply Reply Quote 0
        • stephenw10S
          stephenw10 Netgate Administrator
          last edited by

          Let us know how that goes.
          Thanks
          Steve

          1 Reply Last reply Reply Quote 0
          • M
            M_Devil
            last edited by

            So, running with the sysctl net.inet.tcp.rfc1323=0 workaround for a week now.

            The idea is it should (as a workaround) solve the problem of ipv6 websites refusing (or after approx. 10 seconds) loading.
            The good news: the workaround is helping a lot, but it does not enable the same browsing experience as with ipv4. A few times a day still an ipv6 site is not loading in first attempt. After an ctrl-F5 is uasually loads the page.

            Did I understand correct this problem need to be solved by FreeBSD developers and not by pfSense team?
            If so, why does it take so much time to solve, because as far as I can tell, this should effect a lot of people, right?

            1 Reply Last reply Reply Quote 0
            • D
              doktornotor Banned
              last edited by

              Sigh. I don't know why IPv6 fragmentation got under this topic. That problem is NOT caused by missing baby jumbo. It's caused by pf discarding legitimate traffic.

              As for why does it take so long, when you look at the mailing lists, it took years to get fixed with ipfw, and it's no better with pf. Generally, getting kernel bugs fixed is a completely disastrous experience with FreeBSD. Stuff rotting in bugzilla for years.

              1 Reply Last reply Reply Quote 0
              • M
                M_Devil
                last edited by

                You are right. The discussed bug has nothing to do with Jumbo Frames. Did not know that before, as my (wrong) assumption was the bad browsing experience was causes by missing Jumbo Frames support.

                I can read your frustration with fixing kernel bugs by FreeBSD *) team. So does that mean you don't expect this bug (dropping ipv6 traffic) to be fixed soon?
                If so, that whould be ridiculous, because (in the assumtion I understand it good enough) ipv6 is currently useless and breaking normal Internet traffic.

                *) FreeBSD is all about stability?

                1 Reply Last reply Reply Quote 0
                • D
                  doktornotor Banned
                  last edited by

                  @M_Devil:

                  I can read your frustration with fixing kernel bugs by FreeBSD *) team. So does that mean you don't expect this bug (dropping ipv6 traffic) to be fixed soon?

                  https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=124933 - filed: 2008-06-24

                  Frustration is an extreme understatement!!!  >:( >:( >:( Expect to be fixed soon? LOL.

                  The other bug has been sitting there since 2012 and the only response it got was "Hey guys, I cannot reproduce this" after 32 months.

                  Since the topic got somehow lost anyway - PR 172648 can indeed be mitigated by sysctl net.inet.tcp.rfc1323=0. OTOH, this certainly does not help with PR 124933. That only can be avoided by dropping MTU and MSS values to avoid fragmentation altogether – which is probably the issue you are still seeing:

                  @M_Devil:

                  A few times a day still an ipv6 site is not loading in first attempt. After an ctrl-F5 is uasually loads the page.

                  (even with  sysctl net.inet.tcp.rfc1323=0). There are logs on https://redmine.pfsense.org/issues/2762, see if that matches what you see.

                  1 Reply Last reply Reply Quote 0
                  • M
                    M_Devil
                    last edited by

                    So, I am naive to expect from FreeBSD team with stability as one of their main goals to fix this.
                    Maybe I am not that naive and FreeBSD team is not delivering (or can't I say that because I am using a free product?).

                    The bug reports you refer to are so old, does me wonder if pfSense (depend on FreeBSD) is my future routing/firewall platform. Until now I felt like using an platform with ultimate secure/robust/stable OS as host, apparently I am wrong.

                    I don't read much about ipv6 being practically unusable with pfSense. Is this because almost nobody is using it, or… ?

                    1 Reply Last reply Reply Quote 0
                    • D
                      doktornotor Banned
                      last edited by

                      @M_Devil:

                      I don't read much about ipv6 being practically unusable with pfSense. Is this because almost nobody is using it, or… ?

                      To make something positive out of this thread… In order to minimize hitting these bugs, lets assume you are using some IPv6 tunnel ( like HE or SixXS) and are on PPPoE.

                      1/ Go to Interfaces, select the tunnel interface and set MTU=1472, MSS=1452.

                      Check that it worked (note the mtu and max-mss values):

                      
$ ifconfig | grep gif0 | grep mtu
gif0: flags=8051 <up,pointopoint,running,multicast>metric 0 mtu 1472</up,pointopoint,running,multicast> 

                      
$ pfctl -sa | grep scrub | grep gif0
scrub on gif0 all max-mss 1412 fragment reassemble

                      Note: The configured MSS value must be 20 bytes lower than MTU, resulting in max-mss being 60 bytes lower than the configured MTU – because pfSense developers think that people understand better that the input value is not taken as input value but subtracted by some (incorrect) number –  (i.e., -40 which is the header size for IPv4, instead of -60 which is the header size for IPv6). No, not kidding you. Sigh.  ::)

                      2/ Browse to the tunnel provider website and set your IPv6 tunnel to match your MTU set above:

                      Note:
                      I find https://www.o2.cz/ to be a good site for testing that the above worked. It does not load at all no matter what, unless you get rid of the fragmentation issue completely. If it still does not work and you get firewall log entries like the ones posted on pfSense bug #2762, start back with 1/ and lower the values by further 20 bytes, i.e., MTU=1452, MSS=1432, do the same for 2/ and try again.

                      3/ (Optional) To avoid more annoyances related to FreeBSD kernel/pf bugs related to IPv6, go to System - Advanced - System Tunables, click the + button and disable net.inet.tcp.rfc1323

                      Click Save. Check that it worked:

                      
$ sysctl net.inet.tcp.rfc1323
net.inet.tcp.rfc1323: 0

                      This gives you additional headroom, since net.inet.tcp.rfc1323 enabled adds 12 bytes of timestamps to each TCP packet. Note: this may affect your WAN throughput, see this article for a simple explanation. (For most people, lower speed trade-off would seem better than flaky/broken connectivity, though…) :P

                      Important waste of time avoidance note:
                      While people might be tempted to go System - Advanced - Firewall / NAT and disable firewall scrub there, this will be a completely useless exercise – scrub is required for the MSS clamping above to work, so it won't be disabled on the tunnel interface at all, as easily verified by running pfctl -sa | grep scrub – obviously will fix nothing at all.

                      1 Reply Last reply Reply Quote 0
                      • M
                        M_Devil
                        last edited by

                        Using native ipv6 over PPPoE.

                        Thank you for detailed description. MSS value was to high and MTU was missing.
                        Now lets see if it gets better.

                        1 Reply Last reply Reply Quote 0
                        • D
                          doktornotor Banned
                          last edited by

                          @M_Devil:

                          Using native ipv6 over PPPoE.

                          Well, that'd be pretty much the same, except that you could start with values 20 bytes higher – 1492/1472 (since you do not have the tunnel encapsulation overhead) and go lower only if that does not help.

                          1 Reply Last reply Reply Quote 0
                          • H
                            hda
                            last edited by

                            @M_Devil:

                            Using native ipv6 over PPPoE.

                            Not only have PPPoE WAN on 1492, but LAN on 1492 too, for RA. Then MSS is 1432 for IPv6. Test with wireshark and see the MSS 1440 when you leave LAN MTU on 1500 (or blank ?).

                            1 Reply Last reply Reply Quote 0
                            • M
                              M_Devil
                              last edited by

                              Need to put LAN on MTU=1492 because of RA? I use Jumbo frames (MTU=9000) on LAN side. No problems at all with ipv4. Not sure if I understand your suggested test, but I give it a try.

                              1 Reply Last reply Reply Quote 0
                              • D
                                doktornotor Banned
                                last edited by

                                @hda: Not needed at all on any PPPoE box I tried.

                                @M_Devil: With the jumbo nonsense, you're on a sure way to a giant kaboom, which has totally nothing to do with this topic.

                                1 Reply Last reply Reply Quote 0
                                • M
                                  M_Devil
                                  last edited by

                                  @doktornotor mtu change on LAN side was suggested by hda. Because I did not see the relevance for changing MTU on LAN side, I did mention my use of jumbo for LAN side (for inter vlan routing).

                                  1 Reply Last reply Reply Quote 0
                                  • H
                                    hda
                                    last edited by

                                    My native IPv6 connection isn't comfortable (ISP IPv6 lease renewal fails) with MTU=1492 & MSS=1472 only on WAN & all 'blank' on LAN's. I concluded the common denominator MTU=1492 for IPv6, as long as MTU negotiation (rfc4638) doesn't work everywhere or flawlessly. Found out MTU=1492 & 'blank' MSS for all interfaces with IPv6 works best or reliable.

                                    @M_Devil: RA advertizes the MTU=xxxx in its broadcast.

                                    1 Reply Last reply Reply Quote 0
                                    • M
                                      M_Devil
                                      last edited by

                                      @hda thank you for the clarification.

                                      Running MTU and MSS settings like you suggested and running fine. Can only hope pfSense/FreeBSD team can solve the root cause in a reasonable amount of time.

                                      1 Reply Last reply Reply Quote 0
                                      • H
                                        Harvy66
                                        last edited by

                                        I may be missing something, but I found some reference on https://calomel.org/freebsd_network_tuning.html

                                        With an interface MTU of 1500 bytes we suggest an
                                        net.inet.tcp.mssdflt of 1460 bytes. 1500 MTU minus 20 byte IP header minus 20
                                        byte TCP header is 1460. With net.inet.tcp.rfc1323 enabled, tcp timestamps
                                        are added to the packets and the mss is automatically reduced from 1460 bytes
                                        to 1448 bytes total payload. Note: if you are using PF with an outgoing scrub
                                        rule then PF will re-package the data using an MTU of 1460 by default, thus
                                        overriding this mssdflt setting

                                        It sounds like PF scrub doesn't adjust for rfc1323, so you need to manually set your MSS to 1448?

                                        1 Reply Last reply Reply Quote 0
                                        • D
                                          doktornotor Banned
                                          last edited by

                                          It suggests to set net.inet.tcp.mssdflt which is not what the MSS setting does in the GUI. Plus it tells you that it will  be ignored completely by scrub, so no idea really what's the point in messing with net.inet.tcp.mssdflt here.

                                          1 Reply Last reply Reply Quote 0
                                          • H
                                            Harvy66
                                            last edited by

                                            I guess what I was getting at is it ignores the default setting. The term "default" implies the value can be changed. Instead of using the default values, why not manually set the values?

                                            1 Reply Last reply Reply Quote 0
                                            • First post
                                              Last post
                                            Copyright 2025 Rubicon Communications LLC (Netgate). All rights reserved.