GNU bug report logs - #16343
24.3; Failure in unexec with hardened Linux kernel

Previous Next

To add a comment to this bug, you must first unarchive it, by sending
a message to control AT debbugs.gnu.org, with unarchive 16343 in the body.
You can then email your comments to 16343 AT debbugs.gnu.org in the normal way.

Toggle the display of automated, internal messages from the tracker.

Message #5 received at submit <at> debbugs.gnu.org (full text, mbox):

From: Ulrich Mueller <ulm <at> gentoo.org>
To: bug-gnu-emacs <at> gnu.org
Cc: emacs <at> gentoo.org
Subject: 24.3; Failure in unexec with hardened Linux kernel
Date: Sat, 4 Jan 2014 22:56:07 +0100

Forwarding downstream bug <https://bugs.gentoo.org/456970>.

Emacs 24.3 fails to build on Linux 3.11.7 with grsecurity/PaX patches,
e.g., grsecurity-2.9.1-3.11.7-201311102306.patch from
<http://grsecurity.net/>. This configuration is used in Gentoo with
sys-kernel/hardened-sources-3.11.7-r1.

The build process fails in unexec, the same way it previously did in
bug #11398:

   Dumping under the name emacs
   **************************************************
   Warning: Your system has a gap between BSS and the
   heap (15854248 bytes).  This usually means that exec-shield
   or something similar is in effect.  The dump may
   fail because of this.  See the section about
   exec-shield in etc/PROBLEMS for more information.
   **************************************************
   /bin/sh: Zeile 6: 29064 Speicherzugriffsfehler  `/bin/pwd`/temacs --batch --load loadup bootstrap
   make[1]: *** [bootstrap-emacs] Fehler 1

The reason that it fails again is that the PaX kernel switched from
setting PaX flags in the program header to extended file attributes
(that is, XATTR_PAX_FLAGS=y instead of PT_PAX_FLAGS=y in the kernel's
configuration). Therefore running paxctl on temacs is no longer
sufficient, but setfattr needs to be called.

The patch included below was tested with Emacs 24.3 and fixes the
problem. I've rebased it on the bzr trunk as of today, though.

Please note that extended attributes for temacs are set when they are
supported and when the setfattr program is available, regardless if
the kernel is hardened or not. They will not harm, but simply be
ignored in the latter case. Also the temacs binary is not being
installed, so the installed files will not change. (Contrary to the
paxctl method, the emacs binary doesn't "inherit" extended attributes
from temacs, so there is no need to unset them.)


--- old/configure.ac	2014-01-01 08:31:29 +0000
+++ new/configure.ac	2014-01-04 20:49:13 +0000
@@ -990,6 +990,18 @@
   fi
 fi
 
+AC_PATH_PROG(SETFATTR, setfattr)
+if test "X$SETFATTR" != X; then
+  AC_MSG_CHECKING([whether extended attributes are supported])
+  touch conftest.tmp
+  if $SETFATTR -n user.pax.flags conftest.tmp >/dev/null 2>&1; then
+    AC_MSG_RESULT(yes)
+  else
+    AC_MSG_RESULT(no); SETFATTR=""
+  fi
+  rm -f conftest.tmp
+fi
+
 ## Need makeinfo >= 4.7 (?) to build the manuals.
 AC_PATH_PROG(MAKEINFO, makeinfo, no)
 dnl By this stage, configure has already checked for egrep and set EGREP,
--- old/src/Makefile.in	2014-01-01 07:43:34 +0000
+++ new/src/Makefile.in	2014-01-04 20:49:13 +0000
@@ -114,6 +114,9 @@
 ## memory randomization in temacs with "paxctl -r".  See bug#11398.
 PAXCTL = @PAXCTL@
 
+## If available, the full path to the setfattr program.
+SETFATTR = @SETFATTR@
+
 ## Some systems define this to request special libraries.
 LIBS_SYSTEM=@LIBS_SYSTEM@
 
@@ -494,6 +497,8 @@
 	$(TEMACS_POST_LINK)
 	test "$(CANNOT_DUMP)" = "yes" || \
 	  test "X$(PAXCTL)" = X || $(PAXCTL) -r temacs$(EXEEXT)
+	test "$(CANNOT_DUMP)" = "yes" || test "X$(SETFATTR)" = X || \
+	  $(SETFATTR) -n user.pax.flags -v r temacs$(EXEEXT)
 
 ## The following oldxmenu-related rules are only (possibly) used if
 ## HAVE_X11 && !USE_GTK, but there is no harm in always defining them.

Message #10 received at 16343-done <at> debbugs.gnu.org (full text, mbox):

From: Paul Eggert <paul.eggert <at> verizon.net>
To: Ulrich Mueller <ulm <at> gentoo.org>
Cc: 16343-done <at> debbugs.gnu.org
Subject: 24.3; Failure in unexec with hardened Linux kernel
Date: Sat, 04 Jan 2014 17:03:22 -0800

Thanks very much for reporting that.  I installed a somewhat
different patch: a bit more conservative, it's used only on
GNU/Linux and doesn't attempt to find the full pathname of
setfattr, so 'configure' should run a bit faster.  Please
let me know if it doesn't work for you.  In the meantime I'm
marking the bug as done.

Here's what I installed into the trunk:

=== modified file 'ChangeLog'
--- ChangeLog	2014-01-03 01:59:58 +0000
+++ ChangeLog	2014-01-05 00:53:37 +0000
@@ -1,3 +1,10 @@
+2014-01-05  Paul Eggert  <eggert <at> cs.ucla.edu>
+
+	Port to GNU/Linux with recent grsecurity/PaX patches (Bug#16343).
+	Problem and proposed patch reported by Ulrich Mueller;
+	this patch uses a somewhat-different approach.
+	* configure.ac (SETFATTR): New variable.
+
 2014-01-03  Paul Eggert  <eggert <at> cs.ucla.edu>
 
 	Merge from gnulib, incorporating:

=== modified file 'configure.ac'
--- configure.ac	2014-01-01 08:31:29 +0000
+++ configure.ac	2014-01-05 00:59:55 +0000
@@ -988,6 +988,24 @@
       [if $PAXCTL -v conftest$EXEEXT >/dev/null 2>&1; then AC_MSG_RESULT(yes)
       else AC_MSG_RESULT(no); PAXCTL=""; fi])
   fi
+
+  if test "${SETFATTR+set}" != set; then
+    AC_CACHE_CHECK([for setfattr],
+      [emacs_cv_prog_setfattr],
+      [touch conftest.tmp
+       if (setfattr -n user.pax.flags conftest.tmp) >/dev/null 2>&1; then
+	 emacs_cv_prog_setfattr=yes
+       else
+	 emacs_cv_prog_setfattr=no
+       fi])
+    if test "$emacs_cv_prog_setfattr" = yes; then
+      SETFATTR=setfattr
+    else
+      SETFATTR=
+    fi
+    rm -f conftest.tmp
+    AC_SUBST([SETFATTR])
+  fi
 fi
 
 ## Need makeinfo >= 4.7 (?) to build the manuals.

=== modified file 'src/ChangeLog'
--- src/ChangeLog	2014-01-04 09:31:30 +0000
+++ src/ChangeLog	2014-01-05 00:54:04 +0000
@@ -1,3 +1,9 @@
+2014-01-05  Paul Eggert  <eggert <at> cs.ucla.edu>
+
+	Port to GNU/Linux with recent grsecurity/PaX patches (Bug#16343).
+	* Makefile.in (SETFATTR): New macro.
+	(temacs$(EXEEXT)): Use it.
+
 2014-01-04  Martin Rudalics  <rudalics <at> gmx.at>
 
 	Fix maximization behavior on Windows (Bug#16300).

=== modified file 'src/Makefile.in'
--- src/Makefile.in	2014-01-01 07:43:34 +0000
+++ src/Makefile.in	2014-01-05 00:52:09 +0000
@@ -108,11 +108,12 @@
 ## Flags to pass to ld only for temacs.
 TEMACS_LDFLAGS = $(LD_SWITCH_SYSTEM) $(LD_SWITCH_SYSTEM_TEMACS)
 
-## If available, the full path to the paxctl program.
+## If available, the names of the paxctl and setfattr programs.
 ## On grsecurity/PaX systems, unexec will fail due to a gap between
-## the bss section and the heap.  This can be prevented by disabling
-## memory randomization in temacs with "paxctl -r".  See bug#11398.
+## the bss section and the heap.  Older versions nee paxctl to work
+## around this, newer ones setfattr.  See Bug#11398 and Bug#16343.
 PAXCTL = @PAXCTL@
+SETFATTR = @SETFATTR@
 
 ## Some systems define this to request special libraries.
 LIBS_SYSTEM=@LIBS_SYSTEM@
@@ -494,6 +495,8 @@
 	$(TEMACS_POST_LINK)
 	test "$(CANNOT_DUMP)" = "yes" || \
 	  test "X$(PAXCTL)" = X || $(PAXCTL) -r temacs$(EXEEXT)
+	test "$(CANNOT_DUMP)" = "yes" || test -z "$(SETFATTR)" || \
+	  $(SETFATTR) -n user.pax.flags -v r $@
 
 ## The following oldxmenu-related rules are only (possibly) used if
 ## HAVE_X11 && !USE_GTK, but there is no harm in always defining them.

Message #13 received at 16343-done <at> debbugs.gnu.org (full text, mbox):

From: Ulrich Mueller <ulm <at> gentoo.org>
To: Paul Eggert <paul.eggert <at> verizon.net>
Cc: 16343-done <at> debbugs.gnu.org
Subject: Re: 24.3; Failure in unexec with hardened Linux kernel
Date: Sun, 5 Jan 2014 04:20:48 +0100

I confirm that it works fine here. Thank you.

Small typo in src/Makefile.in:

> +## the bss section and the heap.  Older versions nee paxctl to work

s/nee/need/

Message #16 received at 16343-done <at> debbugs.gnu.org (full text, mbox):

From: Paul Eggert <eggert <at> cs.ucla.edu>
To: Ulrich Mueller <ulm <at> gentoo.org>
Cc: 16343-done <at> debbugs.gnu.org
Subject: Re: 24.3; Failure in unexec with hardened Linux kernel
Date: Sat, 04 Jan 2014 20:57:30 -0800

Ulrich Mueller wrote:
> Small typo in src/Makefile.in:

Thanks, I fixed that in trunk bzr 115871.

Previous Next

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