Package "postgresql-client-9.3"

  postgresql-9.3 (9.3.4-1) unstable; urgency=medium

  * New upstream bugfix release. Most notable change:

    Fix WAL replay of locking an already-updated tuple (Andres Freund,
    ��lvaro Herrera)

    This error caused updated rows to not be found by index scans, resulting
    in inconsistent query results depending on whether an index scan was used.
    Subsequent processing could result in constraint violations, since the
    previously updated row would not be found by later index searches, thus
    possibly allowing conflicting rows to be inserted. Since this error is in
    WAL replay, it would only manifest during crash recovery or on standby
    servers. The improperly-replayed case most commonly arises when a table
    row that is referenced by a foreign-key constraint is updated concurrently
    with creation of a referencing row.

  * Compile with -fno-omit-frame-pointer on amd64 to facilitate hierarchical
    profile generation. (Closes: #730134)
  * Remove obsolete configure option --with-tkconfig.

 -- Christoph Berg <email address hidden> Tue, 18 Mar 2014 07:19:37 +0100

  postgresql-9.3 (9.3.3-1bzr2) trusty; urgency=medium

  Upload current Debian packaging bzr to fix autopkgtest in LXC.

  * Add missing build-essential test depends, for 180_ecpg.t.
 -- Martin Pitt <email address hidden> Fri, 21 Feb 2014 16:14:59 -0800

  postgresql-9.3 (9.3.3-1bzr1) trusty; urgency=medium

  * Add missing gcc test depends, for 180_ecpg.t.
 -- Martin Pitt <email address hidden> Thu, 20 Feb 2014 22:26:18 -0800

  postgresql-9.3 (9.3.3-1) unstable; urgency=medium

  [ Christoph Berg ]
  * New upstream security/bugfix release.

    + Shore up GRANT ... WITH ADMIN OPTION restrictions (Noah Misch)

      Granting a role without ADMIN OPTION is supposed to prevent the grantee
      from adding or removing members from the granted role, but this
      restriction was easily bypassed by doing SET ROLE first. The security
      impact is mostly that a role member can revoke the access of others,
      contrary to the wishes of his grantor. Unapproved role member additions
      are a lesser concern, since an uncooperative role member could provide
      most of his rights to others anyway by creating views or SECURITY
      DEFINER functions. (CVE-2014-0060)

    + Prevent privilege escalation via manual calls to PL validator functions
      (Andres Freund)

      The primary role of PL validator functions is to be called implicitly
      during CREATE FUNCTION, but they are also normal SQL functions that a
      user can call explicitly. Calling a validator on a function actually
      written in some other language was not checked for and could be
      exploited for privilege-escalation purposes. The fix involves adding a
      call to a privilege-checking function in each validator function.
      Non-core procedural languages will also need to make this change to
      their own validator functions, if any. (CVE-2014-0061)

    + Avoid multiple name lookups during table and index DDL (Robert Haas,
      Andres Freund)

      If the name lookups come to different conclusions due to concurrent
      activity, we might perform some parts of the DDL on a different table
      than other parts. At least in the case of CREATE INDEX, this can be used
      to cause the permissions checks to be performed against a different
      table than the index creation, allowing for a privilege escalation
      attack. (CVE-2014-0062)

    + Prevent buffer overrun with long datetime strings (Noah Misch)

      The MAXDATELEN constant was too small for the longest possible value of
      type interval, allowing a buffer overrun in interval_out(). Although the
      datetime input functions were more careful about avoiding buffer
      overrun, the limit was short enough to cause them to reject some valid
      inputs, such as input containing a very long timezone name. The ecpg
      library contained these vulnerabilities along with some of its own.
      (CVE-2014-0063)

    + Prevent buffer overrun due to integer overflow in size calculations
      (Noah Misch, Heikki Linnakangas)

      Several functions, mostly type input functions, calculated an allocation
      size without checking for overflow. If overflow did occur, a too-small
      buffer would be allocated and then written past. (CVE-2014-0064)

    + Prevent overruns of fixed-size buffers (Peter Eisentraut, Jozef Mlich)

      Use strlcpy() and related functions to provide a clear guarantee that
      fixed-size buffers are not overrun. Unlike the preceding items, it is
      unclear whether these cases really represent live issues, since in most
      cases there appear to be previous constraints on the size of the input
      string. Nonetheless it seems prudent to silence all Coverity warnings of
      this type. (CVE-2014-0065)

    + Avoid crashing if crypt() returns NULL (Honza Horak, Bruce Momjian)

      There are relatively few scenarios in which crypt() could return NULL,
      but contrib/chkpass would crash if it did. One practical case in which
      this could be an issue is if libc is configured to refuse to execute
      unapproved hashing algorithms (e.g., "FIPS mode"). (CVE-2014-0066)

    + Document risks of make check in the regression testing instructions
      (Noah Misch, Tom Lane)

      Since the temporary server started by make check uses "trust"
      authentication, another user on the same machine could connect to it as
      database superuser, and then potentially exploit the privileges of the
      operating-system user who started the tests. A future release will
      probably incorporate changes in the testing procedure to prevent this
      risk, but some public discussion is needed first. So for the moment,
      just warn people against using make check when there are untrusted users
      on the same machine. (CVE-2014-0067)

    + Rework tuple freezing protocol (��lvaro Herrera, Andres Freund)

      The logic for tuple freezing was unable to handle some cases involving
      freezing of multixact IDs, with the practical effect that shared
      row-level locks might be forgotten once old enough.

      Fixing this required changing the WAL record format for tuple freezing.
      While this is no issue for standalone servers, when using replication it
      means that standby servers must be upgraded to 9.3.3 or later before
      their masters are. An older standby will be unable to interpret freeze
      records generated by a newer master, and will fail with a PANIC message.
      (In such a case, upgrading the standby should be sufficient to let it
      resume execution.)

  * The upstream tarballs no longer contain a plain HISTORY file, but point to
    the html documentation. Note the location of these files in our
    changelog.gz file.
  * Teach configure to find tclsh8.6 where tclsh is not available.

  [ Martin Pitt ]
  * Build with LINUX_OOM_SCORE_ADJ=0 instead of the older LINUX_OOM_ADJ=0. All
    relevant distro releases (>= squeeze/lucid) use kernels which support
    /proc/pid/oom_score_adj, so avoid the dmesg warnings. (Closes: #646245,
    LP: #991725)
  * Bump Standards-Version to 3.9.5 (no changes necessary).
  * Build with tcl8.6 where available (>= Jessie, >= trusty).

 -- Christoph Berg <email address hidden> Wed, 19 Feb 2014 10:15:39 +0100

  postgresql-9.3 (9.3.2-1ubuntu2) trusty; urgency=medium

  * No-change rebuild for Python 3.4.
 -- Matthias Klose <email address hidden> Wed, 19 Feb 2014 12:03:01 +0100