New upstream microreleases 9.1.16, 9.3.7, 9.4.2

[Martin Pitt]

PostgreSQL will push out new microreleases on Friday, 2015-05-22. The tarballs for the updates are not public yet, but the fixes are visible in the upstream git, so there's no need to treat this as embargoed, but there should still be a coordinated release. These fix a couple of security issues, as well as the usual set of bug fixes.

Upstream announcement: http://www.postgresql.org/about/news/1587/

[Martin Pitt (pitti)]

As usual, wily will be fixed via syncing from Debian unstable. Christoph already prepared 9.4.2-1 in git.

[Martin Pitt (pitti)]

I have prepared and tested updates for all releases. Trusty and newer are trivial (just debian/changelog), all upstream and the postgresql-common test suites pass (via autopkgtest).

Precise was also fairly trivial (no patch changes etc.), but I backported the autopkgtest. That way a new version can be tested with

 adt-run *.deb -B postgresql-9.1_9.1.16-0ubuntu0.12.04.dsc --- schroot precise

which is a lot easier than having to manually install all packages, create locales etc., and run the tests. Note that the tests fail in 090_multicluster.t, all tests up to that are alright. This is nothing new, they've failed forever (at least in schroot), one of these days I'll track this down. trusty and newer all succeed.

I'll put the packages on people.u.c. on Friday. We won't release this on a Friday anyway? If you want them earlier, please tell me and we figure out how to transfer them privately.

[Marc Deslauriers (mdeslaur)]

Friday is fine, we won't release them until monday anyway.

Thanks!

[Martin Pitt (pitti)]

All packages are in http://people.canonical.com/~pitti/tmp/psql-1457093/

[Launchpad Janitor (janitor)]

This bug was fixed in the package postgresql-9.1 - 9.1.16-0ubuntu0.14.04

---------------
postgresql-9.1 (9.1.16-0ubuntu0.14.04) trusty-security; urgency=medium

  * New upstream security/bug fix release (LP: #1457093)
    - Improve detection of system-call failures
      Our replacement implementation of snprintf() failed to check for errors
      reported by the underlying system library calls; the main case that
      might be missed is out-of-memory situations. In the worst case this
      might lead to information exposure, due to our code assuming that a
      buffer had been overwritten when it hadn't been. Also, there were a few
      places in which security-relevant calls of other system library
      functions did not check for failure.
      It remains possible that some calls of the *printf() family of functions
      are vulnerable to information disclosure if an out-of-memory error
      occurs at just the wrong time. We judge the risk to not be large, but
      will continue analysis in this area. (CVE-2015-3166)
   - Note: The other vulnerabilities fixed in 9.1.16 don't affect this version
     as we build the PL/Perl package only.

 -- Martin Pitt <email address hidden> Wed, 20 May 2015 23:16:18 +0200

[Launchpad Janitor (janitor)]

This bug was fixed in the package postgresql-9.1 - 9.1.16-0ubuntu0.12.04

---------------
postgresql-9.1 (9.1.16-0ubuntu0.12.04) precise-security; urgency=medium

  * New upstream security/bug fix release (LP: #1457093)
    - Avoid possible crash when client disconnects just before the
      authentication timeout expires.
      If the timeout interrupt fired partway through the session shutdown
      sequence, SSL-related state would be freed twice, typically causing a
      crash and hence denial of service to other sessions. Experimentation
      shows that an unauthenticated remote attacker could trigger the bug
      somewhat consistently, hence treat as security issue. (CVE-2015-3165)

    - Improve detection of system-call failures
      Our replacement implementation of snprintf() failed to check for errors
      reported by the underlying system library calls; the main case that
      might be missed is out-of-memory situations. In the worst case this
      might lead to information exposure, due to our code assuming that a
      buffer had been overwritten when it hadn't been. Also, there were a few
      places in which security-relevant calls of other system library
      functions did not check for failure.
      It remains possible that some calls of the *printf() family of functions
      are vulnerable to information disclosure if an out-of-memory error
      occurs at just the wrong time. We judge the risk to not be large, but
      will continue analysis in this area. (CVE-2015-3166)

    - In contrib/pgcrypto, uniformly report decryption failures as Wrong key
      or corrupt data
      Previously, some cases of decryption with an incorrect key could report
      other error message texts. It has been shown that such variance in
      error reports can aid attackers in recovering keys from other systems.
      While it's unknown whether pgcrypto's specific behaviors are likewise
      exploitable, it seems better to avoid the risk by using a
      one-size-fits-all message. (CVE-2015-3167)

    - Protect against wraparound of multixact member IDs
      Under certain usage patterns, the existing defenses against this might
      be insufficient, allowing pg_multixact/members files to be removed too
      early, resulting in data loss.
      The fix for this includes modifying the server to fail transactions that
      would result in overwriting old multixact member ID data, and improving
      autovacuum to ensure it will act proactively to prevent multixact member
      ID wraparound, as it does for transaction ID wraparound.

   - See release notes for details about other fixes.

  * Backport the autopkgtest, as running the postgresql-common integration
    test suite is a lot simpler that way. Add manual creation of required
    locales, as precise's postgresql-common test suite does not yet do that by
    itself.

 -- Martin Pitt <email address hidden> Wed, 20 May 2015 23:25:56 +0200

[Launchpad Janitor (janitor)]

This bug was fixed in the package postgresql-9.4 - 9.4.2-0ubuntu0.15.04

---------------
postgresql-9.4 (9.4.2-0ubuntu0.15.04) vivid-security; urgency=medium

  * New upstream security/bug fix release (LP: #1457093)
    - Avoid possible crash when client disconnects just before the
      authentication timeout expires.
      If the timeout interrupt fired partway through the session shutdown
      sequence, SSL-related state would be freed twice, typically causing a
      crash and hence denial of service to other sessions. Experimentation
      shows that an unauthenticated remote attacker could trigger the bug
      somewhat consistently, hence treat as security issue. (CVE-2015-3165)

    - Improve detection of system-call failures
      Our replacement implementation of snprintf() failed to check for errors
      reported by the underlying system library calls; the main case that
      might be missed is out-of-memory situations. In the worst case this
      might lead to information exposure, due to our code assuming that a
      buffer had been overwritten when it hadn't been. Also, there were a few
      places in which security-relevant calls of other system library
      functions did not check for failure.
      It remains possible that some calls of the *printf() family of functions
      are vulnerable to information disclosure if an out-of-memory error
      occurs at just the wrong time. We judge the risk to not be large, but
      will continue analysis in this area. (CVE-2015-3166)

    - In contrib/pgcrypto, uniformly report decryption failures as Wrong key
      or corrupt data
      Previously, some cases of decryption with an incorrect key could report
      other error message texts. It has been shown that such variance in
      error reports can aid attackers in recovering keys from other systems.
      While it's unknown whether pgcrypto's specific behaviors are likewise
      exploitable, it seems better to avoid the risk by using a
      one-size-fits-all message. (CVE-2015-3167)

    - Protect against wraparound of multixact member IDs
      Under certain usage patterns, the existing defenses against this might
      be insufficient, allowing pg_multixact/members files to be removed too
      early, resulting in data loss.
      The fix for this includes modifying the server to fail transactions that
      would result in overwriting old multixact member ID data, and improving
      autovacuum to ensure it will act proactively to prevent multixact member
      ID wraparound, as it does for transaction ID wraparound.

   - See release notes for details about other fixes.

 -- Martin Pitt <email address hidden> Wed, 20 May 2015 17:44:27 +0200

[Launchpad Janitor (janitor)]

This bug was fixed in the package postgresql-9.3 - 9.3.7-0ubuntu0.14.04

---------------
postgresql-9.3 (9.3.7-0ubuntu0.14.04) trusty-security; urgency=medium

  * New upstream security/bug fix release (LP: #1457093)
    - Avoid possible crash when client disconnects just before the
      authentication timeout expires.
      If the timeout interrupt fired partway through the session shutdown
      sequence, SSL-related state would be freed twice, typically causing a
      crash and hence denial of service to other sessions. Experimentation
      shows that an unauthenticated remote attacker could trigger the bug
      somewhat consistently, hence treat as security issue. (CVE-2015-3165)

    - Improve detection of system-call failures
      Our replacement implementation of snprintf() failed to check for errors
      reported by the underlying system library calls; the main case that
      might be missed is out-of-memory situations. In the worst case this
      might lead to information exposure, due to our code assuming that a
      buffer had been overwritten when it hadn't been. Also, there were a few
      places in which security-relevant calls of other system library
      functions did not check for failure.
      It remains possible that some calls of the *printf() family of functions
      are vulnerable to information disclosure if an out-of-memory error
      occurs at just the wrong time. We judge the risk to not be large, but
      will continue analysis in this area. (CVE-2015-3166)

    - In contrib/pgcrypto, uniformly report decryption failures as Wrong key
      or corrupt data
      Previously, some cases of decryption with an incorrect key could report
      other error message texts. It has been shown that such variance in
      error reports can aid attackers in recovering keys from other systems.
      While it's unknown whether pgcrypto's specific behaviors are likewise
      exploitable, it seems better to avoid the risk by using a
      one-size-fits-all message. (CVE-2015-3167)

    - Protect against wraparound of multixact member IDs
      Under certain usage patterns, the existing defenses against this might
      be insufficient, allowing pg_multixact/members files to be removed too
      early, resulting in data loss.
      The fix for this includes modifying the server to fail transactions that
      would result in overwriting old multixact member ID data, and improving
      autovacuum to ensure it will act proactively to prevent multixact member
      ID wraparound, as it does for transaction ID wraparound.

   - See release notes for details about other fixes.

 -- Martin Pitt <email address hidden> Wed, 20 May 2015 23:08:58 +0200

[Launchpad Janitor (janitor)]

This bug was fixed in the package postgresql-9.4 - 9.4.2-0ubuntu0.14.10

---------------
postgresql-9.4 (9.4.2-0ubuntu0.14.10) utopic-security; urgency=medium

  * New upstream security/bug fix release (LP: #1457093)
    - Avoid possible crash when client disconnects just before the
      authentication timeout expires.
      If the timeout interrupt fired partway through the session shutdown
      sequence, SSL-related state would be freed twice, typically causing a
      crash and hence denial of service to other sessions. Experimentation
      shows that an unauthenticated remote attacker could trigger the bug
      somewhat consistently, hence treat as security issue. (CVE-2015-3165)

    - Improve detection of system-call failures
      Our replacement implementation of snprintf() failed to check for errors
      reported by the underlying system library calls; the main case that
      might be missed is out-of-memory situations. In the worst case this
      might lead to information exposure, due to our code assuming that a
      buffer had been overwritten when it hadn't been. Also, there were a few
      places in which security-relevant calls of other system library
      functions did not check for failure.
      It remains possible that some calls of the *printf() family of functions
      are vulnerable to information disclosure if an out-of-memory error
      occurs at just the wrong time. We judge the risk to not be large, but
      will continue analysis in this area. (CVE-2015-3166)

    - In contrib/pgcrypto, uniformly report decryption failures as Wrong key
      or corrupt data
      Previously, some cases of decryption with an incorrect key could report
      other error message texts. It has been shown that such variance in
      error reports can aid attackers in recovering keys from other systems.
      While it's unknown whether pgcrypto's specific behaviors are likewise
      exploitable, it seems better to avoid the risk by using a
      one-size-fits-all message. (CVE-2015-3167)

    - Protect against wraparound of multixact member IDs
      Under certain usage patterns, the existing defenses against this might
      be insufficient, allowing pg_multixact/members files to be removed too
      early, resulting in data loss.
      The fix for this includes modifying the server to fail transactions that
      would result in overwriting old multixact member ID data, and improving
      autovacuum to ensure it will act proactively to prevent multixact member
      ID wraparound, as it does for transaction ID wraparound.

   - See release notes for details about other fixes.

 -- Martin Pitt <email address hidden> Wed, 20 May 2015 23:01:00 +0200

[Martin Pitt (pitti)]

https://launchpad.net/ubuntu/+source/postgresql-9.4/9.4.2-1 is in wily now.