ubuntu/+source/postgresql-9.1:ubuntu/oneiric-security

Last commit made on 2013-04-04
Get this branch:
git clone -b ubuntu/oneiric-security https://git.launchpad.net/ubuntu/+source/postgresql-9.1
Members of Ubuntu Server Dev import team can upload to this branch. Log in for directions.

Branch merges

Branch information

Name:
ubuntu/oneiric-security
Repository:
lp:ubuntu/+source/postgresql-9.1

Recent commits

4a1ef62... by Martin Pitt on 2013-04-02

Import patches-unapplied version 9.1.9-0ubuntu11.10 to ubuntu/oneiric-security

Imported using git-ubuntu import.

Changelog parent: d2309e0e3c7cd1592f424a6a12b87f85c4568649

New changelog entries:
  * New upstream security/bug fix release: (LP: #1163184)
    - Fix insecure parsing of server command-line switches.
      A connection request containing a database name that begins with
      "-" could be crafted to damage or destroy files within the server's
      data directory, even if the request is eventually rejected.
      [CVE-2013-1899]
    - Reset OpenSSL randomness state in each postmaster child process.
      This avoids a scenario wherein random numbers generated by
      "contrib/pgcrypto" functions might be relatively easy for another
      database user to guess. The risk is only significant when the
      postmaster is configured with ssl = on but most connections don't
      use SSL encryption. [CVE-2013-1900]
    - Make REPLICATION privilege checks test current user not
      authenticated user.
      An unprivileged database user could exploit this mistake to call
      pg_start_backup() or pg_stop_backup(), thus possibly interfering
      with creation of routine backups. [CVE-2013-1901]
    - Fix GiST indexes to not use "fuzzy" geometric comparisons when it's
      not appropriate to do so.
      The core geometric types perform comparisons using "fuzzy"
      equality, but gist_box_same must do exact comparisons, else GiST
      indexes using it might become inconsistent. After installing this
      update, users should "REINDEX" any GiST indexes on box, polygon,
      circle, or point columns, since all of these use gist_box_same.
    - Fix erroneous range-union and penalty logic in GiST indexes that
      use "contrib/btree_gist" for variable-width data types, that is
      text, bytea, bit, and numeric columns.
      These errors could result in inconsistent indexes in which some
      keys that are present would not be found by searches, and also in
      useless index bloat. Users are advised to "REINDEX" such indexes
      after installing this update.
    - Fix bugs in GiST page splitting code for multi-column indexes.
      These errors could result in inconsistent indexes in which some
      keys that are present would not be found by searches, and also in
      indexes that are unnecessarily inefficient to search. Users are
      advised to "REINDEX" multi-column GiST indexes after installing
      this update.
    - See HISTORY/changelog.gz for details about the other bug fixes.

d2309e0... by Martin Pitt on 2013-02-05

Import patches-unapplied version 9.1.8-0ubuntu11.10 to ubuntu/oneiric-security

Imported using git-ubuntu import.

Changelog parent: 214cb144670bf0740290291f61f27811c53917dd

New changelog entries:
  * New upstream security/bug fix release: (LP: #1116336)
    - Prevent execution of enum_recv from SQL
      The function was misdeclared, allowing a simple SQL command to crash the
      server. In principle an attacker might be able to use it to examine the
      contents of server memory. Our thanks to Sumit Soni (via Secunia SVCRP)
      for reporting this issue. (CVE-2013-0255)
    - See HISTORY/changelog.gz for the other bug fixes.

214cb14... by Martin Pitt on 2012-12-10

Import patches-unapplied version 9.1.7-0ubuntu11.10 to ubuntu/oneiric-proposed

Imported using git-ubuntu import.

Changelog parent: 2eae95ab6e42daf85520de355afc8a7e6cb678f6

New changelog entries:
  * New upstream bug fix release: (LP: #1088393)
    - Fix multiple bugs associated with "CREATE INDEX CONCURRENTLY".
      Fix "CREATE INDEX CONCURRENTLY" to use in-place updates when
      changing the state of an index's pg_index row. This prevents race
      conditions that could cause concurrent sessions to miss updating
      the target index, thus resulting in corrupt concurrently-created
      indexes.
      Also, fix various other operations to ensure that they ignore
      invalid indexes resulting from a failed "CREATE INDEX CONCURRENTLY"
      command. The most important of these is "VACUUM", because an
      auto-vacuum could easily be launched on the table before corrective
      action can be taken to fix or remove the invalid index.
    - Fix buffer locking during WAL replay.
      The WAL replay code was insufficiently careful about locking
      buffers when replaying WAL records that affect more than one page.
      This could result in hot standby queries transiently seeing
      inconsistent states, resulting in wrong answers or unexpected
      failures.
    - See HISTORY/changelog.gz for the other bug fixes.

2eae95a... by Martin Pitt on 2012-09-25

Import patches-unapplied version 9.1.6-0ubuntu11.10 to ubuntu/oneiric-proposed

Imported using git-ubuntu import.

Changelog parent: 6586ad21550d3adf5f6430edf8b95a1bdd09ed31

New changelog entries:
  * New upstream bug fix release: (LP: #1055944)
    - Fix persistence marking of shared buffers during WAL replay.
      This mistake can result in buffers not being written out during
      checkpoints, resulting in data corruption if the server later
      crashes without ever having written those buffers. Corruption can
      occur on any server following crash recovery, but it is
      significantly more likely to occur on standby slave servers since
      those perform much more WAL replay. There is a low probability of
      corruption of btree and GIN indexes. There is a much higher
      probability of corruption of table "visibility maps". Fortunately,
      visibility maps are non-critical data in 9.1, so the worst
      consequence of such corruption in 9.1 installations is transient
      inefficiency of vacuuming. Table data proper cannot be corrupted by
      this bug.
      While no index corruption due to this bug is known to have occurred
      in the field, as a precautionary measure it is recommended that
      production installations "REINDEX" all btree and GIN indexes at a
      convenient time after upgrading to 9.1.6.
      Also, if you intend to do an in-place upgrade to 9.2.X, before
      doing so it is recommended to perform a "VACUUM" of all tables
      while having vacuum_freeze_table_age set to zero. This will ensure
      that any lingering wrong data in the visibility maps is corrected
      before 9.2.X can depend on it. vacuum_cost_delay can be adjusted to
      reduce the performance impact of vacuuming, while causing it to
      take longer to finish.
    - See HISTORY/changelog.gz for the other bug fixes.

6586ad2... by Jamie Strandboge on 2012-08-16

Import patches-unapplied version 9.1.5-0ubuntu11.10 to ubuntu/oneiric-security

Imported using git-ubuntu import.

Changelog parent: a010f36bfa870eff7be423c21539af8cadb6da48

New changelog entries:
  * New upstream bug fix/security release:
   - Prevent access to external files/URLs via XML entity references
     (Noah Misch, Tom Lane)
     xml_parse() would attempt to fetch external files or URLs as needed
     to resolve DTD and entity references in an XML value, thus allowing
     unprivileged database users to attempt to fetch data with the
     privileges of the database server. While the external data wouldn't
     get returned directly to the user, portions of it could be exposed
     in error messages if the data didn't parse as valid XML; and in any
     case the mere ability to check existence of a file might be useful
     to an attacker. (CVE-2012-3489)
   - Prevent access to external files/URLs via "contrib/xml2"'s
     xslt_process() (Peter Eisentraut)
     libxslt offers the ability to read and write both files and URLs
     through stylesheet commands, thus allowing unprivileged database
     users to both read and write data with the privileges of the
     database server. Disable that through proper use of libxslt's
     security options. (CVE-2012-3488)
     Also, remove xslt_process()'s ability to fetch documents and
     stylesheets from external files/URLs. While this was a documented
     "feature", it was long regarded as a bad idea. The fix for
     CVE-2012-3489 broke that capability, and rather than expend effort
     on trying to fix it, we're just going to summarily remove it.
   - Prevent too-early recycling of btree index pages (Noah Misch)
     When we allowed read-only transactions to skip assigning XIDs, we
     introduced the possibility that a deleted btree page could be
     recycled while a read-only transaction was still in flight to it.
     This would result in incorrect index search results. The
     probability of such an error occurring in the field seems very low
     because of the timing requirements, but nonetheless it should be
     fixed.
   - Fix crash-safety bug with newly-created-or-reset sequences (Tom
     Lane)
     If "ALTER SEQUENCE" was executed on a freshly created or reset
     sequence, and then precisely one nextval() call was made on it, and
     then the server crashed, WAL replay would restore the sequence to a
     state in which it appeared that no nextval() had been done, thus
     allowing the first sequence value to be returned again by the next
     nextval() call. In particular this could manifest for serial
     columns, since creation of a serial column's sequence includes an
     "ALTER SEQUENCE OWNED BY" step.
   - Fix race condition in enum-type value comparisons (Robert Haas, Tom
     Lane)
     Comparisons could fail when encountering an enum value added since
     the current query started.
   - Fix txid_current() to report the correct epoch when not in hot
     standby (Heikki Linnakangas)
     This fixes a regression introduced in the previous minor release.
   - Prevent selection of unsuitable replication connections as the
     synchronous standby (Fujii Masao)
     The master might improperly choose pseudo-servers such as
     pg_receivexlog or pg_basebackup as the synchronous standby, and
     then wait indefinitely for them.
   - Fix bug in startup of Hot Standby when a master transaction has
     many subtransactions (Andres Freund)
     This mistake led to failures reported as "out-of-order XID
     insertion in KnownAssignedXids".
   - Ensure the "backup_label" file is fsync'd after pg_start_backup()
     (Dave Kerr)
   - Fix timeout handling in walsender processes (Tom Lane)
     WAL sender background processes neglected to establish a SIGALRM
     handler, meaning they would wait forever in some corner cases where
     a timeout ought to happen.
   - Wake walsenders after each background flush by walwriter (Andres
     Freund, Simon Riggs)
     This greatly reduces replication delay when the workload contains
     only asynchronously-committed transactions.
   - Fix LISTEN/NOTIFY to cope better with I/O problems, such as out of
     disk space (Tom Lane)
     After a write failure, all subsequent attempts to send more NOTIFY
     messages would fail with messages like "Could not read from file
     "pg_notify/nnnn" at offset nnnnn: Success".
   - Only allow autovacuum to be auto-canceled by a directly blocked
     process (Tom Lane)
     The original coding could allow inconsistent behavior in some
     cases; in particular, an autovacuum could get canceled after less
     than deadlock_timeout grace period.
   - Improve logging of autovacuum cancels (Robert Haas)
   - Fix log collector so that log_truncate_on_rotation works during the
     very first log rotation after server start (Tom Lane)
   - Fix WITH attached to a nested set operation
     (UNION/INTERSECT/EXCEPT) (Tom Lane)
   - Ensure that a whole-row reference to a subquery doesn't include any
     extra GROUP BY or ORDER BY columns (Tom Lane)
   - Fix dependencies generated during ALTER TABLE ... ADD CONSTRAINT
     USING INDEX (Tom Lane)
     This command left behind a redundant pg_depend entry for the index,
     which could confuse later operations, notably ALTER TABLE ... ALTER
     COLUMN TYPE on one of the indexed columns.
   - Fix "REASSIGN OWNED" to work on extensions (Alvaro Herrera)
   - Disallow copying whole-row references in CHECK constraints and
     index definitions during "CREATE TABLE" (Tom Lane)
     This situation can arise in "CREATE TABLE" with LIKE or INHERITS.
     The copied whole-row variable was incorrectly labeled with the row
     type of the original table not the new one. Rejecting the case
     seems reasonable for LIKE, since the row types might well diverge
     later. For INHERITS we should ideally allow it, with an implicit
     coercion to the parent table's row type; but that will require more
     work than seems safe to back-patch.
   - Fix memory leak in ARRAY(SELECT ...) subqueries (Heikki
     Linnakangas, Tom Lane)
   - Fix planner to pass correct collation to operator selectivity
     estimators (Tom Lane)
     This was not previously required by any core selectivity estimation
     function, but third-party code might need it.
   - Fix extraction of common prefixes from regular expressions (Tom
     Lane)
     The code could get confused by quantified parenthesized
     subexpressions, such as ^(foo)?bar. This would lead to incorrect
     index optimization of searches for such patterns.
   - Fix bugs with parsing signed "hh":"mm" and "hh":"mm":"ss" fields in
     interval constants (Amit Kapila, Tom Lane)
   - Fix pg_dump to better handle views containing partial GROUP BY
     lists (Tom Lane)
     A view that lists only a primary key column in GROUP BY, but uses
     other table columns as if they were grouped, gets marked as
     depending on the primary key. Improper handling of such primary key
     dependencies in pg_dump resulted in poorly-ordered dumps, which at
     best would be inefficient to restore and at worst could result in
     outright failure of a parallel pg_restore run.
   - In PL/Perl, avoid setting UTF8 flag when in SQL_ASCII encoding
     (Alex Hunsaker, Kyotaro Horiguchi, Alvaro Herrera)
   - Use Postgres' encoding conversion functions, not Python's, when
     converting a Python Unicode string to the server encoding in
     PL/Python (Jan Urbanski)
     This avoids some corner-case problems, notably that Python doesn't
     support all the encodings Postgres does. A notable functional
     change is that if the server encoding is SQL_ASCII, you will get
     the UTF-8 representation of the string; formerly, any non-ASCII
     characters in the string would result in an error.
   - Fix mapping of PostgreSQL encodings to Python encodings in
     PL/Python (Jan Urbanski)
   - Report errors properly in "contrib/xml2"'s xslt_process() (Tom
     Lane)
   - Update time zone data files to tzdata release 2012e for DST law
     changes in Morocco and Tokelau

a010f36... by Martin Pitt on 2012-06-04

Import patches-unapplied version 9.1.4-0ubuntu11.10 to ubuntu/oneiric-security

Imported using git-ubuntu import.

Changelog parent: 441ebd716d37f00051bfcb2e43438c03fc374dcc

New changelog entries:
  * New upstream bug fix/security release: (LP: #1008317)
    - Fix incorrect password transformation in "contrib/pgcrypto"'s DES
      crypt() function.
      If a password string contained the byte value 0x80, the remainder
      of the password was ignored, causing the password to be much weaker
      than it appeared. With this fix, the rest of the string is properly
      included in the DES hash. Any stored password values that are
      affected by this bug will thus no longer match, so the stored
      values may need to be updated. (CVE-2012-2143)
    - Ignore SECURITY DEFINER and SET attributes for a procedural
      language's call handler. Applying such attributes to a call handler
      could crash the server. (CVE-2012-2655)
    - Make "contrib/citext"'s upgrade script fix collations of citext
      arrays and domains over citext.
      Release 9.1.2 provided a fix for collations of citext columns and
      indexes in databases upgraded or reloaded from pre-9.1
      installations, but that fix was incomplete: it neglected to handle
      arrays and domains over citext. This release extends the module's
      upgrade script to handle these cases. As before, if you have
      already run the upgrade script, you'll need to run the collation
      update commands by hand instead. See the 9.1.2 release notes for
      more information about doing this.
    - Allow numeric timezone offsets in timestamp input to be up to 16
      hours away from UTC. Some historical time zones have offsets larger than
      15 hours, the previous limit. This could result in dumped data values
      being rejected during reload.
    - Fix timestamp conversion to cope when the given time is exactly the
      last DST transition time for the current timezone.
      This oversight has been there a long time, but was not noticed
      previously because most DST-using zones are presumed to have an
      indefinite sequence of future DST transitions.
    - Fix text to name and char to name casts to perform string
      truncation correctly in multibyte encodings.
    - Fix memory copying bug in to_tsquery().
    - Ensure txid_current() reports the correct epoch when executed in
      hot standby.
    - Fix planner's handling of outer PlaceHolderVars within subqueries.
      This bug concerns sub-SELECTs that reference variables coming from
      the nullable side of an outer join of the surrounding query. In
      9.1, queries affected by this bug would fail with "ERROR:
      Upper-level PlaceHolderVar found where not expected". But in 9.0
      and 8.4, you'd silently get possibly-wrong answers, since the value
      transmitted into the subquery wouldn't go to null when it should.
    - Fix planning of UNION ALL subqueries with output columns that are
      not simple variables.
      Planning of such cases got noticeably worse in 9.1 as a result of a
      misguided fix for "MergeAppend child's targetlist doesn't match
      MergeAppend" errors. Revert that fix and do it another way.
    - Fix slow session startup when pg_attribute is very large.
      If pg_attribute exceeds one-fourth of shared_buffers, cache
      rebuilding code that is sometimes needed during session start would
      trigger the synchronized-scan logic, causing it to take many times
      longer than normal. The problem was particularly acute if many new
      sessions were starting at once.
    - Ensure sequential scans check for query cancel reasonably often.
      A scan encountering many consecutive pages that contain no live
      tuples would not respond to interrupts meanwhile.
    - Ensure the Windows implementation of PGSemaphoreLock() clears
      ImmediateInterruptOK before returning.
      This oversight meant that a query-cancel interrupt received later
      in the same query could be accepted at an unsafe time, with
      unpredictable but not good consequences.
    - Show whole-row variables safely when printing views or rules.
      Corner cases involving ambiguous names (that is, the name could be
      either a table or column name of the query) were printed in an
      ambiguous way, risking that the view or rule would be interpreted
      differently after dump and reload. Avoid the ambiguous case by
      attaching a no-op cast.
    - Fix "COPY FROM" to properly handle null marker strings that
      correspond to invalid encoding.
      A null marker string such as E'\\0' should work, and did work in
      the past, but the case got broken in 8.4.
    - Fix "EXPLAIN VERBOSE" for writable CTEs containing RETURNING
      clauses.
    - Fix "PREPARE TRANSACTION" to work correctly in the presence of
      advisory locks.
      Historically, "PREPARE TRANSACTION" has simply ignored any
      session-level advisory locks the session holds, but this case was
      accidentally broken in 9.1.
    - Fix truncation of unlogged tables.
    - Ignore missing schemas during non-interactive assignments of
      search_path.
      This re-aligns 9.1's behavior with that of older branches.
      Previously 9.1 would throw an error for nonexistent schemas
      mentioned in search_path settings obtained from places such as
      "ALTER DATABASE SET".
    - Fix bugs with temporary or transient tables used in extension
      scripts.
      This includes cases such as a rewriting "ALTER TABLE" within an
      extension update script, since that uses a transient table behind
      the scenes.
    - Ensure autovacuum worker processes perform stack depth checking
      properly.
      Previously, infinite recursion in a function invoked by
      auto-"ANALYZE" could crash worker processes.
    - Fix logging collector to not lose log coherency under high load.
      The collector previously could fail to reassemble large messages if
      it got too busy.
    - Fix logging collector to ensure it will restart file rotation after
      receiving SIGHUP.
    - Fix "too many LWLocks taken" failure in GiST indexes.
    - Fix WAL replay logic for GIN indexes to not fail if the index was
      subsequently dropped.
    - Correctly detect SSI conflicts of prepared transactions after a
      crash.
    - Avoid synchronous replication delay when committing a transaction
      that only modified temporary tables.
      In such a case the transaction's commit record need not be flushed
      to standby servers, but some of the code didn't know that and
      waited for it to happen anyway.
    - Fix error handling in pg_basebackup.
    - Fix walsender to not go into a busy loop if connection is
      terminated.
    - Fix memory leak in PL/pgSQL's "RETURN NEXT" command.
    - Fix PL/pgSQL's "GET DIAGNOSTICS" command when the target is the
      function's first variable.
    - Ensure that PL/Perl package-qualifies the _TD variable.
      This bug caused trigger invocations to fail when they are nested
      within a function invocation that changes the current package.
    - Fix PL/Python functions returning composite types to accept a
      string for their result value.
      This case was accidentally broken by the 9.1 additions to allow a
      composite result value to be supplied in other formats, such as
      dictionaries.
    - Fix potential access off the end of memory in psql's expanded
      display ("\x") mode.
    - Fix several performance problems in pg_dump when the database
      contains many objects.
      pg_dump could get very slow if the database contained many schemas,
      or if many objects are in dependency loops, or if there are many
      owned sequences.
    - Fix memory and file descriptor leaks in pg_restore when reading a
      directory-format archive.
    - Fix pg_upgrade for the case that a database stored in a non-default
      tablespace contains a table in the cluster's default tablespace.
    - In ecpg, fix rare memory leaks and possible overwrite of one byte
      after the sqlca_t structure.
    - Fix "contrib/dblink"'s dblink_exec() to not leak temporary database
      connections upon error.
    - Fix "contrib/dblink" to report the correct connection name in error
      messages.
    - Fix "contrib/vacuumlo" to use multiple transactions when dropping
      many large objects.
      This change avoids exceeding max_locks_per_transaction when many
      objects need to be dropped. The behavior can be adjusted with the
      new -l (limit) option.

441ebd7... by Martin Pitt on 2012-02-27

Import patches-unapplied version 9.1.3-0ubuntu0.11.10 to ubuntu/oneiric-security

Imported using git-ubuntu import.

Changelog parent: f6f2a9092b4b820b76804da606ca464b05cafbd5

New changelog entries:
  * New upstream security/bug fix release: (LP: #941912)
    - Require execute permission on the trigger function for "CREATE
      TRIGGER".
      This missing check could allow another user to execute a trigger
      function with forged input data, by installing it on a table he
      owns. This is only of significance for trigger functions marked
      SECURITY DEFINER, since otherwise trigger functions run as the
      table owner anyway. (CVE-2012-0866)
    - Remove arbitrary limitation on length of common name in SSL
      certificates.
      Both libpq and the server truncated the common name extracted from
      an SSL certificate at 32 bytes. Normally this would cause nothing
      worse than an unexpected verification failure, but there are some
      rather-implausible scenarios in which it might allow one
      certificate holder to impersonate another. The victim would have to
      have a common name exactly 32 bytes long, and the attacker would
      have to persuade a trusted CA to issue a certificate in which the
      common name has that string as a prefix. Impersonating a server
      would also require some additional exploit to redirect client
      connections. (CVE-2012-0867)
    - Convert newlines to spaces in names written in pg_dump comments.
      pg_dump was incautious about sanitizing object names that are
      emitted within SQL comments in its output script. A name containing
      a newline would at least render the script syntactically incorrect.
      Maliciously crafted object names could present a SQL injection risk
      when the script is reloaded. (CVE-2012-0868)
    - Fix btree index corruption from insertions concurrent with
      vacuuming.
      An index page split caused by an insertion could sometimes cause a
      concurrently-running "VACUUM" to miss removing index entries that
      it should remove. After the corresponding table rows are removed,
      the dangling index entries would cause errors (such as "could not
      read block N in file ...") or worse, silently wrong query results
      after unrelated rows are re-inserted at the now-free table
      locations. This bug has been present since release 8.2, but occurs
      so infrequently that it was not diagnosed until now. If you have
      reason to suspect that it has happened in your database, reindexing
      the affected index will fix things.
    - Fix transient zeroing of shared buffers during WAL replay.
      The replay logic would sometimes zero and refill a shared buffer,
      so that the contents were transiently invalid. In hot standby mode
      this can result in a query that's executing in parallel seeing
      garbage data. Various symptoms could result from that, but the most
      common one seems to be "invalid memory alloc request size".
    - Fix handling of data-modifying WITH subplans in READ COMMITTED
      rechecking.
      A WITH clause containing "INSERT"/"UPDATE"/"DELETE" would crash if
      the parent "UPDATE" or "DELETE" command needed to be re-evaluated
      at one or more rows due to concurrent updates in READ COMMITTED
      mode.
    - Fix corner case in SSI transaction cleanup.
      When finishing up a read-write serializable transaction, a crash
      could occur if all remaining active serializable transactions are
      read-only.
    - Fix postmaster to attempt restart after a hot-standby crash.
      A logic error caused the postmaster to terminate, rather than
      attempt to restart the cluster, if any backend process crashed
      while operating in hot standby mode.
    - Fix "CLUSTER"/"VACUUM FULL" handling of toast values owned by
      recently-updated rows.
      This oversight could lead to "duplicate key value violates unique
      constraint" errors being reported against the toast table's index
      during one of these commands.
    - Update per-column permissions, not only per-table permissions, when
      changing table owner.
      Failure to do this meant that any previously granted column
      permissions were still shown as having been granted by the old
      owner. This meant that neither the new owner nor a superuser could
      revoke the now-untraceable-to-table-owner permissions.
    - Support foreign data wrappers and foreign servers in "REASSIGN
      OWNED".
      This command failed with "unexpected classid" errors if it needed
      to change the ownership of any such objects.
    - Allow non-existent values for some settings in "ALTER USER/DATABASE
      SET".
      Allow default_text_search_config, default_tablespace, and
      temp_tablespaces to be set to names that are not known. This is
      because they might be known in another database where the setting
      is intended to be used, or for the tablespace cases because the
      tablespace might not be created yet. The same issue was previously
      recognized for search_path, and these settings now act like that
      one.
    - Fix "unsupported node type" error caused by COLLATE in an "INSERT"
      expression.
    - Avoid crashing when we have problems deleting table files
      post-commit.
      Dropping a table should lead to deleting the underlying disk files
      only after the transaction commits. In event of failure then (for
      instance, because of wrong file permissions) the code is supposed
      to just emit a warning message and go on, since it's too late to
      abort the transaction. This logic got broken as of release 8.4,
      causing such situations to result in a PANIC and an unrestartable
      database.
    - Recover from errors occurring during WAL replay of "DROP
      TABLESPACE".
      Replay will attempt to remove the tablespace's directories, but
      there are various reasons why this might fail (for example,
      incorrect ownership or permissions on those directories). Formerly
      the replay code would panic, rendering the database unrestartable
      without manual intervention. It seems better to log the problem and
      continue, since the only consequence of failure to remove the
      directories is some wasted disk space.
    - Fix race condition in logging AccessExclusiveLocks for hot standby.
      Sometimes a lock would be logged as being held by "transaction
      zero". This is at least known to produce assertion failures on
      slave servers, and might be the cause of more serious problems.
    - Track the OID counter correctly during WAL replay, even when it
      wraps around.
    - Prevent emitting misleading "consistent recovery state reached" log
      message at the beginning of crash recovery.
    - Fix initial value of pg_stat_replication.replay_location.
    - Fix regular expression back-references with - attached.
      Rather than enforcing an exact string match, the code would
      effectively accept any string that satisfies the pattern
      sub-expression referenced by the back-reference symbol.
      A similar problem still afflicts back-references that are embedded
      in a larger quantified expression, rather than being the immediate
      subject of the quantifier. This will be addressed in a future
      PostgreSQL release.
    - Fix recently-introduced memory leak in processing of inet/cidr
      values.
    - Fix planner's ability to push down index-expression restrictions
      through UNION ALL.
    - Fix planning of WITH clauses referenced in "UPDATE"/"DELETE" on an
      inherited table.
      This bug led to "could not find plan for CTE" failures.
    - Fix GIN cost estimation to handle column IN (...) index conditions.
      This oversight would usually lead to crashes if such a condition
      could be used with a GIN index.
    - Fix dangling pointer after "CREATE TABLE AS"/"SELECT INTO" in a
      SQL-language function.
      In most cases this only led to an assertion failure in
      assert-enabled builds, but worse consequences seem possible.
    - Fix I/O-conversion-related memory leaks in plpgsql.
    - Work around bug in perl's SvPVutf8() function.
      This function crashes when handed a typeglob or certain read-only
      objects such as $^V. Make plperl avoid passing those to it.
    - In pg_dump, don't dump contents of an extension's configuration
      tables if the extension itself is not being dumped.
    - Improve pg_dump's handling of inherited table columns.
      pg_dump mishandled situations where a child column has a different
      default expression than its parent column. If the default is
      textually identical to the parent's default, but not actually the
      same (for instance, because of schema search path differences) it
      would not be recognized as different, so that after dump and
      restore the child would be allowed to inherit the parent's default.
      Child columns that are NOT NULL where their parent is not could
      also be restored subtly incorrectly.
    - Fix pg_restore's direct-to-database mode for INSERT-style table data.
      Direct-to-database restores from archive files made with
      "--inserts" or "--column-inserts" options fail when using
      pg_restore from a release dated September or December 2011, as a
      result of an oversight in a fix for another problem. The archive
      file itself is not at fault, and text-mode output is okay.
    - Teach pg_upgrade to handle renaming of plpython's shared library.
      Upgrading a pre-9.1 database that included plpython would fail
      because of this oversight.
    - Allow pg_upgrade to process tables containing regclass columns.
      Since pg_upgrade now takes care to preserve pg_class OIDs, there
      was no longer any reason for this restriction.
    - Make libpq ignore ENOTDIR errors when looking for an SSL client
      certificate file.
      This allows SSL connections to be established, though without a
      certificate, even when the user's home directory is set to
      something like /dev/null.
    - Fix some more field alignment issues in ecpg's SQLDA area.
    - Allow AT option in ecpg DEALLOCATE statements.
      The infrastructure to support this has been there for awhile, but
      through an oversight there was still an error check rejecting the
      case.
    - Do not use the variable name when defining a varchar structure in
      ecpg.
    - Fix "contrib/auto_explain"'s JSON output mode to produce valid JSON.
    - Fix error in "contrib/intarray"'s int[] & int[] operator.
      If the smallest integer the two input arrays have in common is 1,
      and there are smaller values in either array, then 1 would be
      incorrectly omitted from the result.
    - Fix error detection in "contrib/pgcrypto"'s encrypt_iv() and
      decrypt_iv().
      These functions failed to report certain types of invalid-input
      errors, and would instead return random garbage values for
      incorrect input.
    - Fix one-byte buffer overrun in "contrib/test_parser".
      The code would try to read one more byte than it should, which
      would crash in corner cases. Since "contrib/test_parser" is only
      example code, this is not a security issue in itself, but bad
      example code is still bad.
    - Use __sync_lock_test_and_set() for spinlocks on ARM, if available.
      This function replaces our previous use of the SWPB instruction,
      which is deprecated and not available on ARMv6 and later. Reports
      suggest that the old code doesn't fail in an obvious way on recent
      ARM boards, but simply doesn't interlock concurrent accesses,
      leading to bizarre failures in multiprocess operation.
    - Use "-fexcess-precision=standard" option when building with gcc
      versions that accept it.
      This prevents assorted scenarios wherein recent versions of gcc
      will produce creative results.
    - Allow use of threaded Python on FreeBSD (Chris Rees)
      Our configure script previously believed that this combination
      wouldn't work; but FreeBSD fixed the problem, so remove that error
      check.
  * Drop 00git_inet_cidr_unpack.patch, 01-armel-tas.patch: Applied upstream.

f6f2a90... by Martin Pitt on 2012-01-05

Import patches-unapplied version 9.1.2-0ubuntu0.11.10.2 to ubuntu/oneiric-proposed

Imported using git-ubuntu import.

Changelog parent: 4b842cd02a17ae9f3693df5b5f67694e5a2a46b3

New changelog entries:
  * Add 00git_inet_cidr_unpack.patch: Revert the behavior of inet/cidr
    functions to not unpack the arguments. This fixes the memory leak when
    sorting inet values. Patch taken from upstream git HEAD. Spotted during
    testing in LP #904631.

4b842cd... by Martin Pitt on 2011-12-16

Import patches-unapplied version 9.1.2-0ubuntu0.11.10.1 to ubuntu/oneiric-proposed

Imported using git-ubuntu import.

Changelog parent: c100644a05abf9c1d93213b9a5a1f52288c00988

New changelog entries:
  * 01-armel-tas.patch: Turn slock_t datatype into an int, and define
    S_UNLOCK() to call __sync_lock_release() instead of using the default
    implementation. This complies to the gcc built-in atomic operations
    specifiction more strictly and now also works on the Panda boards.
    (LP: #904828)

c100644... by Martin Pitt on 2011-12-15

Import patches-unapplied version 9.1.2-0ubuntu0.11.10 to ubuntu/oneiric-proposed

Imported using git-ubuntu import.

Changelog parent: d05b6d368d91e14b3066235f179978dee8d3e770

New changelog entries:
  * New upstream release (LP: #904631):
    - Fix bugs in information_schema.referential_constraints view.
      This view was being insufficiently careful about matching the
      foreign-key constraint to the depended-on primary or unique key
      constraint. That could result in failure to show a foreign key
      constraint at all, or showing it multiple times, or claiming that
      it depends on a different constraint than the one it really does.
      Since the view definition is installed by initdb, merely upgrading
      will not fix the problem. If you need to fix this in an existing
      installation, you can (as a superuser) drop the information_schema
      schema then re-create it by sourcing
      "SHAREDIR/information_schema.sql". (Run pg_config --sharedir if
      you're uncertain where "SHAREDIR" is.) This must be repeated in
      each database to be fixed.
    - Make "contrib/citext"'s upgrade script fix collations of citext
      columns and indexes.
      Existing citext columns and indexes aren't correctly marked as
      being of a collatable data type during pg_upgrade from a pre-9.1
      server. That leads to operations on them failing with errors such
      as "could not determine which collation to use for string
      comparison". This change allows them to be fixed by the same script
      that upgrades the citext module into a proper 9.1 extension during
      CREATE EXTENSION citext FROM unpackaged.
      If you have a previously-upgraded database that is suffering from
      this problem, and you already ran the "CREATE EXTENSION" command,
      you can manually run (as superuser) the "UPDATE" commands found at
      the end of "SHAREDIR/extension/citext--unpackaged--1.0.sql". (Run
      pg_config --sharedir if you're uncertain where "SHAREDIR" is.)
    - Fix possible crash during "UPDATE" or "DELETE" that joins to the
      output of a scalar-returning function.
    - Fix incorrect replay of WAL records for GIN index updates.
    - Fix TOAST-related data corruption during CREATE TABLE dest AS
      SELECT - FROM src or INSERT INTO dest SELECT * FROM src.
    - Fix possible failures during hot standby startup.
    - Start hot standby faster when initial snapshot is incomplete.
    - Fix race condition during toast table access from stale syscache
      entries. The typical symptom was transient errors like "missing chunk
      number 0 for toast value NNNNN in pg_toast_2619", where the cited toast
      table would always belong to a system catalog.
    - Track dependencies of functions on items used in parameter default
      expressions. Previously, a referenced object could be dropped without
      having dropped or modified the function, leading to misbehavior when
      the function was used. Note that merely installing this update will not
      fix the missing dependency entries; to do that, you'd need to
      "CREATE OR REPLACE" each such function afterwards. If you have
      functions whose defaults depend on non-built-in objects, doing so
      is recommended.
    - Fix incorrect management of placeholder variables in nestloop joins.
      This bug is known to lead to "variable not found in subplan target
      list" planner errors, and could possibly result in wrong query
      output when outer joins are involved.
    - Fix window functions that sort by expressions involving aggregates.
    - Fix "MergeAppend child's targetlist doesn't match MergeAppend"
      planner errors.
    - Fix index matching for operators with both collatable and
      noncollatable inputs. In 9.1.0, an indexable operator that has a
      non-collatable left-hand input type and a collatable right-hand input
      type would not be recognized as matching the left-hand column's index.
      An example is the hstore ? text operator.
    - Allow inlining of set-returning SQL functions with multiple OUT
      parameters.
    - Don't trust deferred-unique indexes for join removal.
    - Make DatumGetInetP() unpack inet datums that have a 1-byte header,
      and add a new macro, DatumGetInetPP(), that does not.
    - Improve locale support in money type's input and output.
      Aside from not supporting all standard lc_monetary formatting
      options, the input and output functions were inconsistent, meaning
      there were locales in which dumped money values could not be
      re-read.
    - Don't let transform_null_equals affect CASE foo WHEN NULL ...
      constructs. transform_null_equals is only supposed to affect foo =
      NULL expressions written directly by the user, not equality checks
      generated internally by this form of CASE.
    - Change foreign-key trigger creation order to better support
      self-referential foreign keys.
    - Fix IF EXISTS to work correctly in "DROP OPERATOR FAMILY".
    - Disallow dropping of an extension from within its own script.
    - Don't mark auto-generated types as extension members.
    - Cope with invalid pre-existing search_path settings during "CREATE
      EXTENSION".
    - Avoid floating-point underflow while tracking buffer allocation
      rate.
    - Prevent autovacuum transactions from running in serializable mode.
      Autovacuum formerly used the cluster-wide default transaction
      isolation level, but there is no need for it to use anything higher
      than READ COMMITTED, and using SERIALIZABLE could result in
      unnecessary delays for other processes.
    - Ensure walsender processes respond promptly to SIGTERM.
    - Exclude "postmaster.opts" from base backups.
    - Fix incorrect field alignment in ecpg's SQLDA area.
    - Preserve blank lines within commands in psql's command history.
      The former behavior could cause problems if an empty line was
      removed from within a string literal, for example.
    - Avoid platform-specific infinite loop in pg_dump.
    - Fix compression of plain-text output format in pg_dump.
      pg_dump has historically understood -Z with no -F switch to mean
      that it should emit a gzip-compressed version of its plain text
      output. Restore that behavior.
    - Fix pg_dump to dump user-defined casts between auto-generated
      types, such as table rowtypes.
    - Fix missed quoting of foreign server names in pg_dump.
    - Assorted fixes for pg_upgrade. Handle exclusion constraints correctly,
      avoid failures on Windows, don't complain about mismatched toast table
      names in 8.4 databases.
    - In PL/pgSQL, allow foreign tables to define row types.
    - Fix up conversions of PL/Perl functions' results.
      Restore the pre-9.1 behavior that PL/Perl functions returning void
      ignore the result value of their last Perl statement; 9.1.0 would
      throw an error if that statement returned a reference. Also, make
      sure it works to return a string value for a composite type, so
      long as the string meets the type's input format. In addition,
      throw errors for attempts to return Perl arrays or hashes when the
      function's declared result type is not an array or composite type,
      respectively. (Pre-9.1 versions rather uselessly returned strings
      like ARRAY(0x221a9a0) or HASH(0x221aa90) in such cases.)
    - Ensure PL/Perl strings are always correctly UTF8-encoded.
    - Use the preferred version of xsubpp to build PL/Perl, not
      necessarily the operating system's main copy.
    - Correctly propagate SQLSTATE in PL/Python exceptions.
    - Do not install PL/Python extension files for Python major versions
      other than the one built against.
    - Change all the "contrib" extension script files to report a useful
      error message if they are fed to psql. This should help teach people
      about the new method of using "CREATE EXTENSION" to load these files.
      In most cases, sourcing the scripts directly would fail anyway, but
      with harder-to-interpret messages.
    - Fix incorrect coding in "contrib/dict_int" and "contrib/dict_xsyn".
    - Remove "contrib/sepgsql" tests from the regular regression test
      mechanism. Since these tests require root privileges for setup, they're
      impractical to run automatically. Switch over to a manual approach
      instead, and provide a testing script to help with that.
    - Fix assorted errors in "contrib/unaccent"'s configuration file
      parsing.
    - Honor query cancel interrupts promptly in pgstatindex().
    - Revert unintentional enabling of WAL_DEBUG. Fortunately, as debugging
      tools go, this one is pretty cheap; but it's not intended to be enabled
      by default, so revert.
    - Ensure VPATH builds properly install all server header files.
    - Shorten file names reported in verbose error messages.
      Regular builds have always reported just the name of the C file
      containing the error message call, but VPATH builds formerly
      reported an absolute path name.