Skip to contentSkip to conversation

2.4.18-pre9: iptables screwed?

SOLUTION Harald Welte 3 participants 6 messages Feb 8, 2002 – Feb 12, 2002 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw
Quoted 1 line · David Miller

Stelian has analyzed the bug already.

This is strange.

Quoted 7 lines

From: Stelian Pop <[email protected]>
To: "H. Peter Anvin" <[email protected]>
Cc: Linux Kernel Mailing List <[email protected]>
Subject: Re: 2.4.18-pre9: iptables screwed?
Reply-To: Stelian Pop <[email protected]>
In-Reply-To: <[email protected]>

Quoted 8 lines
Quoted 6 lines · H. Peter Anvin

I get the following error with iptables on 2.4.18-pre9:

sudo iptables-restore < /etc/sysconfig/iptables
iptables-restore: libiptc/libip4tc.c:384: do_check: Assertion `h->info.valid_hooks == (1 << 0 | 1 << 3)' failed. Abort (core dumped)

The code you are quoting is only defined if debugging is compiled into the iptables package. The default distribution of the iptables package does _not_ ship with debugging enabled.

The Makefile of all iptables versions between 1.1.1 (released way before the linux 2.4.0 kernel came out!) and 1.2.5 (current) have the following line in the Makefile:

COPT_FLAGS:=-O2 -DNDEBUG

reads: define no debug

Quoted 10 lines
Quoted 3 lines

However, if I apply the rules manually (using iptables), I have no problem; only if I'm using iptables-save or iptables-restore do I get a dump...

I have this since the netfilter update from pre6 or pre7...

It seems to be caused by a change in the logic for the mangle table:
the userspace tools check only for PREROUTING and OUTPUT chains (the 1 << 0 | 1 << 3 check), but the kernel code was recently updated to support more chains in this table (POSTROUTING etc).

This is true. We introduced this change after some testing since it is needed for complex policy routing scenarios. It's the so-called mangle5hooks.patch

Quoted 5 lines

So it would seem that we need to have a more recent version of the userspace tools (CVS maybe, since the latest released version has the same bug), or the netfilter people should check the userspace tools version before introducing this kind of incompatible change.

I'm running the same iptables-1.2.2 binary (compiled at a 2.4.x kernel in July 2001) with a mangle5hooks-patch'ed linux kernel.

just re-checked it again:
======================================================================
sunbeam# rpm -qi iptables
Name : iptables Relocations: (not relocateable)
Version : 1.2.2 Vendor: Conectiva
Release : 2cl Build Date: Sun 17 Jun 2001 08:17:20 PM CEST
Install date: Thu 08 Nov 2001 01:42:57 PM CET Build Host: mapi2.distro.conectiva
Group : Networking Source RPM: iptables-1.2.2-2cl.src.rpm
Size : 439232 License: GPL
URL : http://netfilter.samba.org
Summary : Packet filtering tool for kernel-2.4.x
Description :
This is the packet filtering tool for kernel-2.4.x. It is much more advanced than ipchains and can take full advantage of the new
features within the 2.4.x packet filtering code. It allows you to
set up masquerading, full NAT, stateful inspection rules, etc.
sunbeam# rpm -V iptables
sunbeam# cat foo
# Generated by iptables-save v1.2.2 on Fri Feb 8 10:35:05 2002
*mangle
:PREROUTING ACCEPT [36557505:30073123582]
:INPUT ACCEPT [31280258:26426457730]
:FORWARD ACCEPT [5276687:3646630572]
:OUTPUT ACCEPT [28690202:18841029987]
:POSTROUTING ACCEPT [34105840:22505172519]
:knf - [0:0]
-A PREROUTING -p tcp -m tcp --dport 25 -j knf
-A PREROUTING -p tcp -m tcp --dport 6667 -j knf
-A POSTROUTING -p tcp -m tcp --tcp-flags SYN,RST,ACK SYN -j TCPMSS --clamp-mss-to-pmtu -A knf -j MARK --set-mark 0xa
COMMIT
sunbeam# iptables-restore < foo
sunbeam# iptables -t nat -L -n
Chain PREROUTING (policy ACCEPT)
target prot opt source destination

Chain POSTROUTING (policy ACCEPT)
target prot opt source destination

MASQUERADE  all  --  192.168.100.0/24     0.0.0.0/0          
MASQUERADE  all  --  192.168.101.0/24     0.0.0.0/0          

Chain OUTPUT (policy ACCEPT)
target prot opt source destination
======================================================================

Because it was working on several systems, we have decided to forward this patch to the mainstream kerel.

We always want to make sure that nobody needs to update the iptables package during the 2.4.x stable kernel series. Because of this (sane) policy, we are keeping back a whole bunch of changes. We can't just silently abandon backwards compatibility.

Quoted 3 lines

(BTW, the quick and dirty fix for me was to hand edit /etc/sysconfig/iptables and remove all references to the mangle table, since I don't use it).

this is of coruse no possible 'solution'.

Quoted 1 line

Stelian Pop <[email protected]>

Signature · 5 lines
Live long and prosper
- Harald Welte / [email protected]               http://www.gnumonks.org/
============================================================================
GCS/E/IT d- s-: a-- C+++ UL++++$ P+++ L++++$ E--- W- N++ o? K- w--- O- M+ 
V-- PS++ PE-- Y++ PGP++ t+ 5-- !X !R tv-- b+++ !DI !D G+ e* h--- r++ y+(*)
#2 lore raw
Quoted 7 lines · Harald Welte
Quoted 1 line · David Miller

Stelian has analyzed the bug already.

This is strange.

Quoted 7 lines

The code you are quoting is only defined if debugging is compiled into the iptables package. The default distribution of the iptables package does _not_ ship with debugging enabled.

The Makefile of all iptables versions between 1.1.1 (released way before the linux 2.4.0 kernel came out!) and 1.2.5 (current) have the following line in the Makefile:

What is the first thing anyone would do if they had a problem with iptables? Turn on debugging, obviously.

1 - that explains why I have no problem, I build from source, since I'm
always trying features in the new versions for better firewalls.
2 - it really should work, debug should find errors, not cause them. 3 - Perhaps debug could be disabled but default, so instead of needing
NDEBUG in production, you would use -DDEBUG when you had a problem (and it might even work ;-).

Signature · 3 lines
bill davidsen <[email protected]>
  CTO, TMR Associates, Inc
Doing interesting things with little computers since 1979.
#3 lore raw
In reply to Harald Welte
Quoted 2 lines · Bill Davidsen

What is the first thing anyone would do if they had a problem with iptables? Turn on debugging, obviously.

yes, I totally agree. I'm not arguing that the current behaviour is OK - of course it should work with debugging enabled. All I said is that under normal circumstances a production system is not supposed to have debugging enabled.

Quoted 1 line

2 - it really should work, debug should find errors, not cause them.

of course. That is why debugging does extra checks, which are disabled in production use. Unfortunately we missed one check...

Quoted 3 lines

3 - Perhaps debug could be disabled but default, so instead of needing
NDEBUG in production, you would use -DDEBUG when you had a problem (and it might even work ;-).

Yes, this is what I'm doing for the iptables-1.2.6 release. The current -DNDEBUG stuff has historical reasons which are no longer valid.

Quoted 1 line

bill davidsen <[email protected]>

Signature · 5 lines
Live long and prosper
- Harald Welte / [email protected]               http://www.gnumonks.org/
============================================================================
GCS/E/IT d- s-: a-- C+++ UL++++$ P+++ L++++$ E--- W- N++ o? K- w--- O- M+ 
V-- PS++ PE-- Y++ PGP++ t+ 5-- !X !R tv-- b+++ !DI !D G+ e* h--- r++ y+(*)
#4 lore raw
In reply to a message not in the archive
Quoted 10 lines · Harald Welte
Quoted 6 lines
Quoted 6 lines

I get the following error with iptables on 2.4.18-pre9:

sudo iptables-restore < /etc/sysconfig/iptables
iptables-restore: libiptc/libip4tc.c:384: do_check: Assertion `h->info.valid_hooks == (1 << 0 | 1 << 3)' failed. Abort (core dumped)

The code you are quoting is only defined if debugging is compiled into the iptables package. The default distribution of the iptables package does _not_ ship with debugging enabled.

Right, upon further analysis the redhat RPM is overriding COPT_FLAGS and removes the -DNDEBUG from the compile line.

Quoted 4 lines

We always want to make sure that nobody needs to update the iptables package during the 2.4.x stable kernel series. Because of this (sane) policy, we are keeping back a whole bunch of changes. We can't just silently abandon backwards compatibility.

Hey, you keeped backwards compatibility _only_ for those compiling the non debug version of the package. I believe this remains however a half-bug...

Stelian.

Signature · 4 lines
Stelian Pop <[email protected]>
|---------------- Free Software Engineer -----------------|
| Alcôve - http://www.alcove.com - Tel: +33 1 49 22 68 00 |
|------------- Alcôve, liberating software ---------------|
#5 lore raw
In reply to a message not in the archive
Quoted 6 lines · Stelian Pop
Quoted 3 lines

The code you are quoting is only defined if debugging is compiled into the iptables package. The default distribution of the iptables package does _not_ ship with debugging enabled.

Right, upon further analysis the redhat RPM is overriding COPT_FLAGS and removes the -DNDEBUG from the compile line.

hm. bad coincidence :(

Quoted 8 lines
Quoted 4 lines

We always want to make sure that nobody needs to update the iptables package during the 2.4.x stable kernel series. Because of this (sane) policy, we are keeping back a whole bunch of changes. We can't just silently abandon backwards compatibility.

Hey, you keeped backwards compatibility _only_ for those compiling the non debug version of the package. I believe this remains however a half-bug...

Well, it is a bug in the debugging code, yes. We missed to updated the debugging code when changing the mangle table. The reason for this is that not even the developers are using the debugging code.

But when I do compatibility checks / tests for something I want to submit to the kernel, I use plain unmodified iptables sources - new and old.

And I really didn't expect anybody to run debugging code on production systems, sorry.

I'm not going to pull this change back out of the kernel because one (or more) distributors/vendors ship a compiled-with-debug iptables.

I don't think this was by RedHat's intention.

Quoted 2 lines

Stelian.
Stelian Pop <[email protected]>

Signature · 5 lines
Live long and prosper
- Harald Welte / [email protected]               http://www.gnumonks.org/
============================================================================
GCS/E/IT d- s-: a-- C+++ UL++++$ P+++ L++++$ E--- W- N++ o? K- w--- O- M+ 
V-- PS++ PE-- Y++ PGP++ t+ 5-- !X !R tv-- b+++ !DI !D G+ e* h--- r++ y+(*)
#6 lore raw
In reply to a message not in the archive
Quoted 3 lines · Harald Welte

Well, it is a bug in the debugging code, yes. We missed to updated the debugging code when changing the mangle table. The reason for this is that not even the developers are using the debugging code.

One may wonder then who is using this debugging code... :-) Especially when you must pass an option to the cc line to _disable_ it...

Quoted 2 lines

I'm not going to pull this change back out of the kernel because one (or more) distributors/vendors ship a compiled-with-debug iptables.

Of course, never said that.

Quoted 1 line

I don't think this was by RedHat's intention.

I don't think so, and anyway, the latest kernel is not supported officially by RedHat either. When (if) they release an updated kernel containing the mangle patch, they will need to update the iptables userspace tools too. That's all.

Signature · 4 lines
Stelian Pop <[email protected]>
|---------------- Free Software Engineer -----------------|
| Alcôve - http://www.alcove.com - Tel: +33 1 49 22 68 00 |
|------------- Alcôve, liberating software ---------------|