ebtables crashed with SIGSEGV in ebt_initialize_entry()

StacktraceTop:
 ebt_initialize_entry () from /tmp/tmpl5BBB0/lib/ebtables/libebtc.so
 do_command () from /tmp/tmpl5BBB0/lib/ebtables/libebtc.so
 ?? ()
 __libc_start_main (main=0x4005b0, argc=4, ubp_av=0x7fffea400b38, init=<optimized out>, fini=<optimized out>, rtld_fini=<optimized out>, stack_end=0x7fffea400b28) at libc-start.c:226
 ?? ()

up dated the softwear 17 hours age and it worked allwright turned on this morning and tryed to update again

Status changed to 'Confirmed' because the bug affects multiple users.

i thik it is the same bug it times out on up date and hangs the up date and locks the up date manger.
i have to restar computer to unhang .
then the updates load

I think you could be right about the update manager failing (can't say
because no crash page appears). In any even I don't need to reboot -
it just goes away!

marty

On Fri, Dec 2, 2011 at 10:14 PM, john morgan <email address hidden> wrote:
> i thik it is  the same bug it times out on up date and hangs the up date and  locks the up date manger.
> i  have to restar computer to unhang .
> then the updates load
>
> --
> You received this bug notification because you are subscribed to the bug
> report.
> https://bugs.launchpad.net/bugs/899315
>
> Title:
>  ebtables crashed with SIGSEGV in ebt_initialize_entry()
>
> To manage notifications about this bug go to:
> https://bugs.launchpad.net/ubuntu/+source/ebtables/+bug/899315/+subscriptions

--
Marty Felker
Floyd, VA

The segfault is cause by the as-needed compiler change. This also happened at SUSE, see https://bugzilla.novell.com/show_bug.cgi?id=567267.
The attached debdiff solves the issue for me. It builds, installs and runs fine (tested on amd64).
I also added a small change to get a dbgsym package for ebtables. The dbgsym package could not be created before, because the binaries were stripped in the upstream build system.

@Albert Damen
Thank you for your patch. However, it would be better to fix the linking order in the upstream Makefile rather than disabling --as-needed. Would it be possible for you to update your patch to do this? I'm unsubscribing ubuntu-sponsors for the moment. Please resubscribe when ready. If you do not wish to do this work, simply unassign yourself.

Hi Micah,
I have tried several ways to get the package working without --no-as-needed, but no joy.
To my understanding the problem is:
- the plugins are needed to initialize structures in ebtables/libebtc. They do this via their magic _init functions, which run when the plugins are loaded, by calling ebt_register_target. ebt_register_target allocates memory and sets a pointer in libebtc (ebt_targets).
- ebtables does not use any symbols from the plugins, so linking with --as-needed will not actually link ebtables with the plugins. So, when ebtables is built with --as-needed, the plugins are not loaded, the structs are not initialized and the ebt_targets pointer is left to 0x0 --> crash when ebt_targets is used.

So in my opinion we need a way to force the plugins to be loaded when ebtables is run. Forcing the linking with --no-as-needed works, but maybe there are other ways?

As a side note, I wonder if using --no-as-needed is a big problem in this specific case. It only adds linking to plugins built from the same source package and shipped in the same binary package as ebtables itself. So this does not seem to cause problems with additional dependencies or complicate transitions?

This bug was fixed in the package ebtables - 2.0.9.2-2ubuntu2

---------------
ebtables (2.0.9.2-2ubuntu2) precise; urgency=low

  * Add patch 05link_with_no-as-needed.patch to link ebtables with
    --no-as-needed and adjust the link order to fix crash when running
    ebtables (LP: #899315)
  * debian/rules: add -g option to DEBCFLAGS, so symbols can be extracted
    into an ebtables-dbgsym package
 -- Albert Damen <email address hidden> Sun, 04 Dec 2011 21:19:58 +0100

LinkedIn
------------

Bug,

I'd like to add you to my professional network on LinkedIn.

- Marty

Marty Felker
Owner at Black Ridge Computer
Roanoke, Virginia Area

Confirm that you know Marty Felker:
https://www.linkedin.com/e/-a9fv1b-hasnqhxo-6r/isd/10067226998/JNrgcKN2/?hs=false&tok=10LeN6fREGNlw1

--
You are receiving Invitation to Connect emails. Click to unsubscribe:
http://www.linkedin.com/e/-a9fv1b-hasnqhxo-6r/VRgt3F5THzv1sWf4joE6Y4PTlkCEF63l1X6pAfB/goo/899315%40bugs%2Elaunchpad%2Enet/20061/I3354391771_1/?hs=false&tok=3RNaQ4ci4GNlw1

(c) 2012 LinkedIn Corporation. 2029 Stierlin Ct, Mountain View, CA 94043, USA.