Project

General

Profile

Actions

Bug #2762

closed

PF drops IPv6 packets with fragment header followed by a last fragment only

Added by Chris Buechler about 11 years ago. Updated over 8 years ago.

Status:
Resolved
Priority:
Normal
Assignee:
Category:
Operating System
Target version:
Start date:
01/18/2013
Due date:
% Done:

0%

Estimated time:
Plus Target Version:
Release Notes:
Affected Version:
2.1-IPv6
Affected Architecture:
All

Description

PF has the same problem as is described here for ipfw.
http://lists.freebsd.org/pipermail/freebsd-net/2011-February/027838.html

This used to be replicable by doing this:
telnet -6 www.allstream.com 80
but as of 201509, this site no longer exhibits this behavior.

They set a frag header, offset = 0, M bit = 0, in all their SYN ACKs for some reason. That's valid per RFC 2460. pcap showing is attached.

PF logs it as follows:

Jan 18 02:48:56 fw1 pf: 00:00:00.242205 rule 5/0(match): block in on em0: (flowlabel 0xeb8da, hlim 56, next-header Fragment (44) payload length: 48) 2607:f4e8:200:12:225:90ff:fe2a:a072 > 2610:160:11:a033::230: frag (0xb5736529:0|40) 80 > 40842: Flags [S.], seq 3303787714, ack 1052652245, win 65535, options [mss 1140,nop,wscale 4,sackOK,TS val 260605935 ecr 179963673], length 0
Jan 18 02:48:59 fw1 pf: 00:00:02.934772 rule 5/0(match): block in on em0: (flowlabel 0xeb8da, hlim 56, next-header Fragment (44) payload length: 48) 2607:f4e8:200:12:225:90ff:fe2a:a072 > 2610:160:11:a033::230: frag (0xaf40f4e7:0|40) 80 > 40842: Flags [S.], seq 3303787714, ack 1052652245, win 65535, options [mss 1140,nop,wscale 4,sackOK,TS val 260605935 ecr 179963973], length 0
Jan 18 02:49:02 fw1 pf: 00:00:02.999317 rule 5/0(match): block in on em0: (flowlabel 0xeb8da, hlim 56, next-header Fragment (44) payload length: 48) 2607:f4e8:200:12:225:90ff:fe2a:a072 > 2610:160:11:a033::230: frag (0xf2d6888d:0|40) 80 > 40842: Flags [S.], seq 3303787714, ack 1052652245, win 65535, options [mss 1140,nop,wscale 4,sackOK,TS val 260605935 ecr 179963973], length 0
Jan 18 02:49:02 fw1 pf: 00:00:00.205661 rule 5/0(match): block in on em0: (flowlabel 0xeb8da, hlim 56, next-header Fragment (44) payload length: 48) 2607:f4e8:200:12:225:90ff:fe2a:a072 > 2610:160:11:a033::230: frag (0x8009c1bf:0|40) 80 > 40842: Flags [S.], seq 3303787714, ack 1052652245, win 65535, options [mss 1140,nop,wscale 4,sackOK,TS val 260605935 ecr 179964293], length 0
Jan 18 02:49:05 fw1 pf: 00:00:02.999839 rule 5/0(match): block in on em0: (flowlabel 0xeb8da, hlim 56, next-header Fragment (44) payload length: 48) 2607:f4e8:200:12:225:90ff:fe2a:a072 > 2610:160:11:a033::230: frag (0xe718b255:0|40) 80 > 40842: Flags [S.], seq 3303787714, ack 1052652245, win 65535, options [mss 1140,nop,wscale 4,sackOK,TS val 260605935 ecr 179964293], length 0


Files

allstream.pcap (6.21 KB) allstream.pcap Chris Buechler, 01/18/2013 03:25 AM
firewall_rules_edit.diff (749 Bytes) firewall_rules_edit.diff Christian Felsing, 08/04/2015 03:23 AM
Actions #1

Updated by Ermal Luçi about 11 years ago

This is scrub in action.
Will see how to make this behave normally.

Actions #2

Updated by Sander Steffann over 10 years ago

PS: It is not broken or weird behaviour (according to the RFCs). RFC 6145 (translating IPv4 <-> IPv6) specifies:

   When the IPv4 sender does not set the DF bit, the translator SHOULD
   always include an IPv6 Fragment Header to indicate that the sender
   allows fragmentation.  The translator MAY provide a configuration
   function that allows the translator not to include the Fragment
   Header for the non-fragmented IPv6 packets.

Translators that follow this RFC will generate such atomic fragments, and pfSense will break communication with them.

Also see RFC 6946 (Processing of IPv6 "Atomic" Fragments)

Actions #3

Updated by Ermal Luçi over 10 years ago

The only option for now seems to create rules with allow-option advanced setting set.

Actions #4

Updated by Doktor Notor over 10 years ago

Sadly, I keep hitting this with http://snapshots.pfsense.org:

Jul 12 10:15:58 gw pf: 00:00:07.982521 rule 5/0(match): block in on gif0: (flowlabel 0x3dfd0, hlim 55, next-header Fragment (44) payload length: 36) 2610:1c0:1:25::51 > 2001:470:6f:xxx:yyy::zzz frag (0xbe5ada46:1432|28)
Jul 12 10:15:58 gw pf: 00:00:00.000058 rule 5/0(match): block in on gif0: (flowlabel 0x3dfd0, hlim 55, next-header Fragment (44) payload length: 36) 2610:1c0:1:25::51 > 2001:470:6f:xxx:yyy::zzz frag (0x86c96d61:1432|28)
Jul 12 10:15:58 gw pf: 00:00:00.000700 rule 5/0(match): block in on gif0: (flowlabel 0x3dfd0, hlim 55, next-header Fragment (44) payload length: 1440) 2610:1c0:1:25::51 > 2001:470:6f:xxx:yyy::zzz frag (0xbe5ada46:0|1432) 80 > 63352: Flags [.], ack 982233318, win 65535, length 1412
Jul 12 10:15:58 gw pf: 00:00:00.000123 rule 5/0(match): block in on gif0: (flowlabel 0x3dfd0, hlim 55, next-header Fragment (44) payload length: 1440) 2610:1c0:1:25::51 > 2001:470:6f:xxx:yyy::zzz frag (0x86c96d61:0|1432) 80 > 63352: Flags [.], ack 982233318, win 65535, length 1412
Jul 12 10:16:02 gw pf: 00:00:03.688589 rule 5/0(match): block in on gif0: (flowlabel 0x3dfd0, hlim 55, next-header Fragment (44) payload length: 1440) 2610:1c0:1:25::51 > 2001:470:6f:xxx:yyy::zzz frag (0x94e13b01:0|1432) 80 > 63352: Flags [.], ack 982233318, win 65535, length 1412
Jul 12 10:16:02 gw pf: 00:00:00.000058 rule 5/0(match): block in on gif0: (flowlabel 0x3dfd0, hlim 55, next-header Fragment (44) payload length: 36) 2610:1c0:1:25::51 > 2001:470:6f:xxx:yyy::zzz frag (0x94e13b01:1432|28)
Jul 12 10:16:11 gw pf: 00:00:03.440808 rule 5/0(match): block in on gif0: (flowlabel 0x3dfd0, hlim 55, next-header Fragment (44) payload length: 1440) 2610:1c0:1:25::51 > 2001:470:6f:xxx:yyy::zzz frag (0xfdfc4caf:0|1432) 80 > 63352: Flags [.], ack 982233318, win 65535, length 1412
Jul 12 10:16:11 gw pf: 00:00:00.000060 rule 5/0(match): block in on gif0: (flowlabel 0x3dfd0, hlim 55, next-header Fragment (44) payload length: 36) 2610:1c0:1:25::51 > 2001:470:6f:xxx:yyy::zzz frag (0xfdfc4caf:1432|28)
Jul 12 10:16:43 gw pf: 00:00:03.721811 rule 5/0(match): block in on gif0: (flowlabel 0x3dfd0, hlim 55, next-header Fragment (44) payload length: 1440) 2610:1c0:1:25::51 > 2001:470:6f:xxx:yyy::zzz frag (0xf4c1f620:0|1432) 80 > 63352: Flags [.], ack 982233318, win 65535, length 1412
Jul 12 10:16:43 gw pf: 00:00:00.000059 rule 5/0(match): block in on gif0: (flowlabel 0x3dfd0, hlim 55, next-header Fragment (44) payload length: 1440) 2610:1c0:1:25::51 > 2001:470:6f:xxx:yyy::zzz frag (0xbbcd08b4:0|1432) 80 > 63352: Flags [.], ack 982233318, win 65535, length 1412
Jul 12 10:16:43 gw pf: 00:00:00.000052 rule 5/0(match): block in on gif0: (flowlabel 0x3dfd0, hlim 55, next-header Fragment (44) payload length: 36) 2610:1c0:1:25::51 > 2001:470:6f:xxx:yyy::zzz frag (0xf4c1f620:1432|28)
Jul 12 10:16:43 gw pf: 00:00:00.000049 rule 5/0(match): block in on gif0: (flowlabel 0x3dfd0, hlim 55, next-header Fragment (44) payload length: 36) 2610:1c0:1:25::51 > 2001:470:6f:xxx:yyy::zzz frag (0xbbcd08b4:1432|28)
Jul 12 10:16:44 gw pf: 00:00:00.012077 rule 5/0(match): block in on gif0: (flowlabel 0x1f70f, hlim 55, next-header Fragment (44) payload length: 36) 2610:1c0:1:25::51 > 2001:470:6f:xxx:yyy::zzz frag (0xdc6772be:1432|28)
Jul 12 10:16:44 gw pf: 00:00:00.000603 rule 5/0(match): block in on gif0: (flowlabel 0x1f70f, hlim 55, next-header Fragment (44) payload length: 1440) 2610:1c0:1:25::51 > 2001:470:6f:xxx:yyy::zzz frag (0xdc6772be:0|1432) 80 > 63354: Flags [.], ack 3682091665, win 65535, length 1412
Jul 12 10:17:06 gw pf: 00:00:07.393108 rule 5/0(match): block in on gif0: (flowlabel 0x3dfd0, hlim 55, next-header Fragment (44) payload length: 36) 2610:1c0:1:25::51 > 2001:470:6f:xxx:yyy::zzz frag (0xaf182ff3:1432|28)
Jul 12 10:17:06 gw pf: 00:00:00.000549 rule 5/0(match): block in on gif0: (flowlabel 0x3dfd0, hlim 55, next-header Fragment (44) payload length: 1440) 2610:1c0:1:25::51 > 2001:470:6f:xxx:yyy::zzz frag (0xaf182ff3:0|1432) 80 > 63352: Flags [.], ack 982233318, win 65535, length 1412
Jul 12 10:17:48 gw pf: 00:00:02.428094 rule 5/0(match): block in on gif0: (flowlabel 0x1f70f, hlim 55, next-header Fragment (44) payload length: 1440) 2610:1c0:1:25::51 > 2001:470:6f:xxx:yyy::zzz frag (0x47368efe:0|1432) 80 > 63354: Flags [.], ack 3682091665, win 65535, length 1412
Jul 12 10:17:48 gw pf: 00:00:00.001648 rule 5/0(match): block in on gif0: (flowlabel 0x1f70f, hlim 55, next-header Fragment (44) payload length: 36) 2610:1c0:1:25::51 > 2001:470:6f:xxx:yyy::zzz frag (0x47368efe:1432|28)
Jul 12 10:18:52 gw pf: 00:00:07.936447 rule 5/0(match): block in on gif0: (flowlabel 0x1f70f, hlim 55, next-header Fragment (44) payload length: 1440) 2610:1c0:1:25::51 > 2001:470:6f:xxx:yyy::zzz frag (0x3a1cb4de:0|1432) 80 > 63354: Flags [.], ack 3682091665, win 65535, length 1412
Jul 12 10:18:52 gw pf: 00:00:00.000060 rule 5/0(match): block in on gif0: (flowlabel 0x1f70f, hlim 55, next-header Fragment (44) payload length: 36) 2610:1c0:1:25::51 > 2001:470:6f:xxx:yyy::zzz frag (0x3a1cb4de:1432|28)

Rather annoying. +1 on lets stop blocking this.

Actions #5

Updated by Ermal Luçi over 10 years ago

  • Target version changed from 2.1 to 2.2

This cannot be solved for now apart the workaround to allow fragments.

Actions #6

Updated by Doktor Notor over 10 years ago

I seriously don't really care whether it's a workaround or not... How I can prevent pf from dropping legitimate traffic!?

Actions #7

Updated by Doktor Notor about 10 years ago

Erm, guys, what's up with this?! Upstream apparently does NOT intend to fix this in any way, cf. http://www.freebsd.org/cgi/query-pr.cgi?pr=124933 and they do not intend to port pf-related updates from OpenBSD either (http://www.freebsd.org/cgi/query-pr.cgi?pr=167057&cat=kern and the fabulous bitchfest @ http://lists.freebsd.org/pipermail/freebsd-pf/2012-September/006740.html). This just "rocks".

How do I create something like

pass in on <iface> inet6 proto ipv6-frag all

via the GUI? This is breaking web pages, hit this all the time, oh the irony, with pfsense.org website itself. This issue also makes http://www.o2.cz completely unusable for me with IPv6 enabled!

Actions #8

Updated by Doktor Notor about 10 years ago

And another one on the broken scrub: http://www.freebsd.org/cgi/query-pr.cgi?pr=172648

Actions #9

Updated by Jim Thompson almost 10 years ago

  • Assignee set to Ermal Luçi
Actions #10

Updated by Jens Groh over 9 years ago

Just FYI:

The official bug (https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=172648) got another mention:

-> https://lists.freebsd.org/pipermail/freebsd-net/2014-November/040319.html

Perhaps some of this could be integrated/backported to v10 as this was developed against FreeBSD 11-current.

Actions #11

Updated by Jim Thompson over 9 years ago

Jens,

If you look at that thread, Ermal has the fix in-hand.

IJS...

Actions #12

Updated by Chris Buechler over 9 years ago

  • Target version changed from 2.2 to 2.2.1
Actions #13

Updated by Jim Thompson about 9 years ago

  • Target version changed from 2.2.1 to 2.3
  • Affected Architecture All added
  • Affected Architecture deleted ()

I think this is going to want more testing than what we can afford in the 2.2.1 timeframe. That said, if it gets fixed for 2.2.1, I'm not upset.

Actions #14

Updated by Klaus Steinberger almost 9 years ago

What is the time frame for fixing this? I was hit by this bug now by adding dnssec NSEC3 to my DNS which enlarged the payloads of my DNS Server over the MTU.

Actions #15

Updated by Christian Felsing over 8 years ago

I modified /usr/local/www/firewall_rules_edit.php with enclosed patch (pfSense 2.2.4)

After that, protocol "IPV6-FRAG" is offered in GUI. A test rule with a Sixxs host did desired modification in pf ruleset. pfctl -sa | less shows

pass in log quick on em0 inet6 proto ipv6-frag from 2a01:258:8:2::4 to any keep state label "USER_RULE"

and ping6 -n -s 2000 2a01:258:8:2::4 works fine.

Use this on your own risk, Due to application to WAN interface security issues may arise.

Actions #16

Updated by Jim Thompson over 8 years ago

  • Assignee changed from Ermal Luçi to Luiz Souza

reassigned to Luiz. Maybe this is fixed in 10.2 (I'm looking mostly at MFS 286079 / r285999)

Actions #17

Updated by Chris Buechler over 8 years ago

  • Description updated (diff)
  • Status changed from New to Feedback

the real world systems I'm aware of that exhibited this behavior no longer do, due to infrastructure changes on their end. Original site in this ticket doesn't, IPv6 IP in the pcap no longer alive. I'm not aware of a replicable real-world scenario, but is testable by replaying the pcap.

I suspect this was fixed in the general fixing of pf+IPv6 fragmentation in FreeBSD, but needs confirmation.

If anyone's aware of a real-world site exhibiting the behavior, let us know.

Actions #18

Updated by Klaus Steinberger over 8 years ago

Chris Buechler wrote:

the real world systems I'm aware of that exhibited this behavior no longer do, due to infrastructure changes on their end. Original site in this ticket doesn't, IPv6 IP in the pcap no longer alive. I'm not aware of a replicable real-world scenario, but is testable by replaying the pcap.

I suspect this was fixed in the general fixing of pf+IPv6 fragmentation in FreeBSD, but needs confirmation.

If anyone's aware of a real-world site exhibiting the behavior, let us know.

We do run a DNS Server with IPV6 and DNSSEC. to complete the the tests on dnsviz.net succesful we had to decrease the size of UDP answers in bind with these parameters in named.conf:
/* * Work around pfsense IPV6 fragmentation problem
*/
edns-udp-size 1024;
max-udp-size 1024;

If we get a fixed up pfsense version, we can verify the fix just with commenting out the parameters in bind and redoing the tests on dnsviz.net

Sincerly,
Klaus

Actions #19

Updated by Chris Buechler over 8 years ago

Klaus Steinberger wrote:

We do run a DNS Server with IPV6 and DNSSEC. to complete the the tests on dnsviz.net succesful we had to decrease the size of UDP answers

That's the general issues with IPv6 fragmentation. This bug ticket is specific to the circumstance described in the original post, which isn't actually fragmented traffic at all.

2.3 has the fixes included in FreeBSD for general IPv6 fragmentation handling with pf, so your issue should be resolved in 2.3. Snapshots will be publicly available soon. Definitely would appreciate any feedback from your usage case on the 2.3 board of the forum (when it exists).

Actions #20

Updated by Sander Steffann over 8 years ago

PS: for those who want to test with a website that sends fragments try www.cbs.nl. It has an RFC 6145 SIIT box in front of it that always adds a fragmentation header.

Relevant part of the RFC:
When the IPv4 sender does not set the DF bit, the translator SHOULD
always include an IPv6 Fragment Header to indicate that the sender
allows fragmentation.

So it's a reliable source of fragments (until they change the implementation)

Actions #21

Updated by Chris Buechler over 8 years ago

  • Status changed from Feedback to Resolved

Thanks Sander, that helps. This is definitely fixed in 2.3.

Actions

Also available in: Atom PDF