GNU bug report logs - #18784
Coultdn't compile emacs-24.4

Previous Next

Full log

View this message in rfc822 format

From: Ulrich Mueller <ulm <at> gentoo.org>
To: Eli Zaretskii <eliz <at> gnu.org>
Cc: Stefan Monnier <monnier <at> IRO.UMontreal.CA>, 18784 <at> debbugs.gnu.org
Subject: bug#18784: Coultdn't compile emacs-24.4
Date: Mon, 16 Mar 2015 21:04:08 +0100

>>>>> On Mon, 16 Mar 2015, Eli Zaretskii wrote:

>> I guess if PIE is used in order to place Emacs's code at randomized
>> addresses (different address every time an Emacs process is launched),
>> it can break the dump because our subr objects will end up with pointers
>> to addresses that aren't valid any more (might be other such problems,
>> but that's the first that comes to my mind).

> Then this is only an issue on systems where PIE == address
> randomization, isn't it?

That would be my guess too. Here are some gdb backtraces showing that
the problem occurs in unexelf.c:

   https://bugs.gentoo.org/494316#c13
   https://bugs.gentoo.org/526948#c9
   http://debbugs.gnu.org/13847#5

There is a long history of Emacs catching up with kernel hardening
in this area. It started with setting ADDR_NO_RANDOMIZE via Linux
personality(2), then setting the NORANDEXEC flag with paxctl(1) or
setfattr(1). Now it seems that we need -nopie in addition.
My impression is that these are all workarounds that don't address the
real issue.

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.