From: Stephen Hemminger Subject: Re: Bug: tc filter show .... raise segfault if more than one rule with action -j MARK exist Date: Tue, 13 May 2014 13:10:48 -0700 Message-ID: <20140513131048.3a831c14@nehalam.linuxnetplumber.net> References: <536B6F03.5040401@a-greve.de> <536DED2C.10301@a-greve.de> <536DEF16.6010406@a-greve.de> Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Cc: netdev@vger.kernel.org To: andreas.greve@a-greve.de Return-path: Received: from mail-pa0-f54.google.com ([209.85.220.54]:42838 "EHLO mail-pa0-f54.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751596AbaEMUKw (ORCPT ); Tue, 13 May 2014 16:10:52 -0400 Received: by mail-pa0-f54.google.com with SMTP id bj1so685331pad.13 for ; Tue, 13 May 2014 13:10:52 -0700 (PDT) In-Reply-To: <536DEF16.6010406@a-greve.de> Sender: netdev-owner@vger.kernel.org List-ID: On Sat, 10 May 2014 11:19:18 +0200 Andreas Greve wrote: > sorry its to early for me ;) > > here the missing attachment. > > On 05/10/14 11:11, Andreas Greve wrote: > > Dear Stephen Hemminger, Dear List Members, > > > > this is a corrected/optimized version of the patch > > Thanks to Stephen Hemminger, for giving me the hint to use structure > > assignment instead of > > memcpy(). > > > > Best wishes > > > > Andreas Greve > > > > On 05/08/14 13:48, Andreas Greve wrote: > >> Dear Stephen Hemminger, Dear List Members, > >> > >> the command > >> > >> tc filter show dev eth1 parent ffff: > >> > >> raise a segmentation fault if there are more than one filter rule > >> with action -j MARK exists. > >> > >> Reason: > >> In print_ipt(...) xtables will be initialzed with a pointer to the > >> static struct tcipt_globals at xtables_init_all(). Later on the > >> fields .opts and .options_offset of tcipt_globals are modified. The > >> call of xtables_free_opts(1) at the end of print_ipt(...) does not > >> restore the original values of tcipt_globals for the modified fields. > >> It only frees some allocated memory and sets .opts to NULL. This > >> leads to a segmentation fault when print_ipt() > >> is called for the next filter rule with action -j MARK. > >> > >> Fix: (patch attached) > >> Cloneing tcipt_globals on the stack as tmp_tcipt_globals and use it > >> instead of tcipt_globals, so tcipt_globals will be not modified. > >> > >> Tests: > >> I only have done limit tests for this fix ( action -j MARK ) on a > >> debian wheezy iproute2 package build from source based on > >> upstream/3.14.0. The system is a little bit special kernel > >> 3.12.17-xen on xen-4.4.1-pre. > >> > >> > >> Andreas Henriksson (debian iproute maintainer) give me an other idea > >> of how to fix this bug by modifying xtables_free_opts() so that it > >> sets .opts = .orig_opts instead .opts = NULL. This should avoid > >> having to duplicate the entire struct each time. > >> > >> I prefer the cloning because: > >> > >> 1) In my understanding the variable tcipt_globals is a complex > >> "constant" that should never be modified. ( Is that right ?) > >> > >> 2) If we want to do the fix in xtables_free_opts we have to change > >> the struct xtables_globals because we need for all fields .orig > >> fields I think this will make all things much more complicated. > >> > >> > >> additional Brainstorms: > >> > >> a ) Perhaps we have to think about of a deep clone of tcipt_globals > >> to protect the underlaying option array > >> b ) makes cloning of tcipt_globals sense in parse_ipt(...) > >> But for this two decisions my understanding of the whole system is > >> not enough. > >> > >> > >> I hope the Information helps Applied, but overly wordy comment simplified, and whitespace issues fixed.