libexpat released a new version 2.2.2 which seems to contain 2 or 3 security fixes. I'm not sure that Python is affected by these bugs.

https://github.com/libexpat/libexpat/blob/R_2_2_2/expat/Changes#L5

Release 2.2.2 Wed July 12 2017
        Security fixes:
             #43  Protect against compilation without any source of high
                    quality entropy enabled, e.g. with CMake build system;
                    commit ff0207e6076e9828e536b8d9cd45c9c92069b895
             #60  Windows with _UNICODE:
                    Unintended use of LoadLibraryW with a non-wide string
                    resulted in failure to load advapi32.dll and degradation
                    in quality of used entropy when compiled with _UNICODE for
                    Windows; you can launch existing binaries with
                    EXPAT_ENTROPY_DEBUG=1 in the environment to inspect the
                    quality of entropy used during runtime; commits
                    * 95b95032f907ef1cd17ee7a9a1768010a825d61d
                    * 73a5a2e9c081f49f2d775cf7ced864158b68dc80
   [MOX-006]      Fix non-NULL parser parameter validation in XML_Parse;
                    resulted in NULL dereference, previously;
                    commit ac256dafdffc9622ab0dc2c62fcecb0dfcfa71fe

        Bug fixes:
             #69  Fix improper use of unsigned long long integer literals

        Other changes:
             #73  Start requiring a C99 compiler
             #49  Fix "==" Bashism in configure script
             #50  Fix too eager getrandom detection for Debian GNU/kFreeBSD
             #52    and macOS
             #51  Address lack of stdint.h in Visual Studio 2003 to 2008
             #58  Address compile warnings
             #68  Fix "./buildconf.sh && ./configure" for some versions
                    of Dash for /bin/sh
             #72  CMake: Ease use of Expat in context of a parent project
                    with multipe CMakeLists.txt files
             #72  CMake: Resolve mistaken executable permissions
             #76  Address compile warning with -DNDEBUG (not recommended!)
             #77  Address compile warning about macro redefinition

        Special thanks to:
            Alexander Bluhm
            Ben Boeckel
            Cătălin Răceanu
            Kerin Millar
            László Böszörményi
            S. P. Zeidler
            Segev Finer
            Václav Slavík
            Victor Stinner
            Viktor Szakats
                 and
Radically Open Security

--

Previous issue for expat 2.2.1: issue #30694.

> #51  Address lack of stdint.h in Visual Studio 2003 to 2008

FYI this change only impacts Python 2.7, since Python 3.3 and newer requires Visual Studio 2010 or newer, and I already backported (cherry-picked) this specific commit in Python 2.7:
https://github.com/python/cpython/pull/2312/commits

> #58  Address compile warnings

That's my small contribution, so coming from CPython :-)
https://github.com/libexpat/libexpat/pull/58

> #76  Address compile warning with -DNDEBUG (not recommended!)

Nice contributions from Segev Finer, coming from CPython ;-)
https://github.com/libexpat/libexpat/issues/76

> #77  Address compile warning about macro redefinition

Another contribution of Segev Finer, already fixed downstream (in Python):
https://github.com/libexpat/libexpat/pull/77

About the 3 security fixes (is the last change a security fix?).

"""
             #43  Protect against compilation without any source of high
                    quality entropy enabled, e.g. with CMake build system;
                    commit ff0207e6076e9828e536b8d9cd45c9c92069b895
"""

Since Python uses its own entropy source, I don't think that this change impacts us.

https://github.com/libexpat/libexpat/commit/ff0207e6076e9828e536b8d9cd45c9c92069b895


"""
             #60  Windows with _UNICODE:
                    Unintended use of LoadLibraryW with a non-wide string
                    resulted in failure to load advapi32.dll and degradation
                    in quality of used entropy when compiled with _UNICODE for
                    Windows; you can launch existing binaries with
                    EXPAT_ENTROPY_DEBUG=1 in the environment to inspect the
                    quality of entropy used during runtime; commits
                    * 95b95032f907ef1cd17ee7a9a1768010a825d61d
                    * 73a5a2e9c081f49f2d775cf7ced864158b68dc80
"""

I don't understand the consequence of this specific bug.

https://github.com/libexpat/libexpat/commit/95b95032f907ef1cd17ee7a9a1768010a825d61d
https://github.com/libexpat/libexpat/commit/73a5a2e9c081f49f2d775cf7ced864158b68dc80


"""
   [MOX-006]      Fix non-NULL parser parameter validation in XML_Parse;
                    resulted in NULL dereference, previously;
                    commit ac256dafdffc9622ab0dc2c62fcecb0dfcfa71fe
"""

I'm not sure that it's possible to call XML_Parse() with NULL in Python.

https://github.com/libexpat/libexpat/commit/ac256dafdffc9622ab0dc2c62fcecb0dfcfa71fe

Expat 2.2.3 was released:

Release 2.2.3 Wed August 2 2017
        Security fixes:
             #82  CVE-2017-11742 -- Windows: Fix DLL hijacking vulnerability
                    using Steve Holme's LoadLibrary wrapper for/of cURL

        Bug fixes:
             #85  Fix a dangling pointer issue related to realloc

        Other changes:
                  Increase code coverage
             #91  Linux: Allow getrandom to fail if nonblocking pool has not
                    yet been initialized and read /dev/urandom then, instead.
                    This is in line with what recent Python does.
             #81  Pre-10.7/Lion macOS: Support entropy from arc4random
             #86  Check that a UTF-16 encoding in an XML declaration has the
                    right endianness
        #4 #5 #7  Recover correctly when some reallocations fail
                  Repair "./configure && make" for systems without any
                    provider of high quality entropy
                    and try reading /dev/urandom on those
                  Ensure that user-defined character encodings have converter
                    functions when they are needed
                  Fix mis-leading description of argument -c in xmlwf.1
                  Rely on macro HAVE_ARC4RANDOM_BUF (rather than __CloudABI__)
                    for CloudABI
            #100  Fix use of SIPHASH_MAIN in siphash.h
             #23  Test suite: Fix memory leaks
                  Version info bumped from 7:4:6 to 7:5:6

        Special thanks to:
            Chanho Park
            Joe Orton
            Pascal Cuoq
            Rhodri James
            Simon McVittie
            Vadim Zeitlin
            Viktor Szakats
                 and
Core Infrastructure Initiative

Previous update: bpo-30694.

cpython_rebuild_expat_dir.sh: Script used to update Modules/expat/ to 2.2.3. The script now uses the libexpat Git repository. Previously, I used tarballs.

> #82  CVE-2017-11742 -- Windows: Fix DLL hijacking vulnerability using Steve Holme's LoadLibrary wrapper for/of cURL

https://github.com/libexpat/libexpat/issues/82

I don't think that this bug affects Python since Python sets a hash secret.

Could the updating script be added into the CPython repository?

If libexpat is upgraded in Python 2.7, the new Modules/expat/loadlibrary.c should also be added to PC/VS9.0/ project files, as I did for PCbuild.

Note: PC/VS7.1/ and PC/VS8.0/ are likely broken and don't need to be updated, right?

New changeset 93d0cb58b4da2a88c56f472c6c19491cc7a390df by Victor Stinner in branch 'master':
bpo-30947: Update libexpat from 2.2.1 to 2.2.3 (#3106)
https://github.com/python/cpython/commit/93d0cb58b4da2a88c56f472c6c19491cc7a390df

New changeset 83e37e16f3065086d721d4e62a3788e01db3431c by Victor Stinner in branch '3.6':
bpo-30947: Update libexpat from 2.2.1 to 2.2.3 (#3106) (#3143)
https://github.com/python/cpython/commit/83e37e16f3065086d721d4e62a3788e01db3431c

New changeset ec4ab09b7c0b5070bdb27351f979cbecc4636245 by Victor Stinner in branch '2.7':
bpo-30947: Update libexpat from 2.2.1 to 2.2.3 (#3106) (#3145)
https://github.com/python/cpython/commit/ec4ab09b7c0b5070bdb27351f979cbecc4636245

Expat 2.2.3 has a bug: see bpo-31170 :-(

New changeset 297516ea509c72d8ebed3a9b3ce200f023aca0b7 by Ned Deily (Victor Stinner) in branch '3.3':
[3.3] bpo-30947, bpo-31170: Update expat from 2.2.1 to 2.2.4 (#3352)
https://github.com/python/cpython/commit/297516ea509c72d8ebed3a9b3ce200f023aca0b7

New changeset 86a713cb0c110b6798ca7f9e630fc511ee0a4028 by larryhastings (Victor Stinner) in branch '3.4':
[3.4][Security] bpo-30947, bpo-31170: Update expat from 2.2.1 to 2.2.4 (#3353)
https://github.com/python/cpython/commit/86a713cb0c110b6798ca7f9e630fc511ee0a4028

New changeset f2492bb6aae061aea47e21fc7e56b7ab9bfdf543 by larryhastings (Victor Stinner) in branch '3.5':
[3.5][Security] bpo-30947, bpo-31170: Update expat from 2.2.1 to 2.2.4 (#3354)
https://github.com/python/cpython/commit/f2492bb6aae061aea47e21fc7e56b7ab9bfdf543

libexpat has been upgraded from 2.2.1 to 2.2.4 in 2.7, 3.4, 3.5, 3.6 and master branches.

And in 3.3.

Victor, the PR for this BPO has introduced XML_POOR_ENTROPY. Neither the commit message nor the issue explains why. Which platform failed to compile without XML_POOR_ENTROPY?

Christian Heimes <lists@cheimes.de> added the comment:
> Victor, the PR for this BPO has introduced XML_POOR_ENTROPY. Neither the commit message nor the issue explains why. Which platform failed to compile without XML_POOR_ENTROPY?

And, maybe:

"Oh, compilation fails on Travis CI at:"
https://github.com/python/cpython/pull/3106#issuecomment-323118722

And my rationale is (extract of setup.py):


                # bpo-30947: Python uses best available entropy sources to
                # call XML_SetHashSalt(), expat entropy sources are not needed
                ('XML_POOR_ENTROPY', '1'),

But I'm wrong if I understood what you told me last week.

Do you remember which platform failed? It doesn't say on the GH PR either.

See #34623 for security bug. We only set good salt for pyexpat based parsers (sax, dom, pure Python etree), but not for the C-accelerated ElementTree implementation.

> Do you remember which platform failed? It doesn't say on the GH PR either.

It was the Linux job of Travis CI, something like an old Ubuntu LTS version.

Since Travis CI prevented me to merge anything and Python already has access to a safe PRNG, I didn't worry about that issue. The disabled code shouldn't be needed on Python.