-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA512
Format: 1.8
Date: Tue, 11 Aug 2026 22:58:48 +0200
Source: postgresql-18
Architecture: source
Version: 18.6-1
Distribution: unstable
Urgency: medium
Maintainer: Debian PostgreSQL Maintainers <team+postgresql@tracker.debian.org>
Changed-By: Christoph Berg <myon@debian.org>
Changes:
postgresql-18 (18.6-1) unstable; urgency=medium
.
* New upstream version 18.6. (18.5 was never released.)
.
+ Restrict logical decoding output plugins to the set specified by
a new server parameter `output_plugin_libraries` (Jacob
Champion)
.
Previously, a replication user could select any loadable library
for logical decoding, allowing exploits of various sorts. To
allow locking this down without breaking setups that worked
before, introduce a whitelist of allowed output plugins.
.
By default, only the output plugins shipped as part of
PostgreSQL (`pgoutput` and `test_decoding`) are included in
`output_plugin_libraries`. Installations that rely on other
output plugins must add them after updating the server, for
example
.
output_plugin_libraries = 'pgoutput, test_decoding, my_trusted_decoder'
.
Additionally, pg_upgrade --check will fail if the
`output_plugin_libraries` parameter on the new cluster does not
permit the plugins of logical replication slots on the old
cluster, when migrating from versions 17 and later. Make
necessary additions to the new cluster's setting before
performing pg_upgrade.
.
The PostgreSQL Project thanks Vladimir Tokarev and Yu Kunpeng
for reporting this problem. (CVE-2026-6471)
.
+ Fix contrib/pgcrypto's PGP encryption to detect unsupported
ciphers (Daniel Gustafsson)
.
Previously, if OpenSSL rejected the requested cipher (for
example, because it is running in FIPS mode, or the legacy
provider hasn't been loaded), pgcrypto failed to notice the
failure and simply XOR'd the non-encrypted block with the
plaintext, rendering the "encryption" trivially breakable. This
will typically occur with deprecated or non-FIPS cipher
algorithms (cipher-algo=blowfish/bf, twofish, cast5, or 3des).
.
By default, pgcrypto will now fail to decrypt any messages that
were affected in this way. To allow retrieval of such data, a
new option `ignore-cipher-failure` has been added to
pgp_pub_decrypt() and pgp_sym_decrypt(). Setting
`ignore-cipher-failure=1` will restore their previous behavior,
allowing the faulty encryption wrapper to be stripped off:
.
pgp_sym_decrypt(encrypted_column, any key, 'ignore-cipher-failure=1')
.
Once the affected messages are identified and stripped of their
wrappers, they can then be re-encrypted with a modern algorithm.
It is important however that the behavior of OpenSSL be the same
as it was when the faulty messages were created: if the set of
unsupported algorithms is not the same, this approach will not
work. See the documentation for `ignore-cipher-failure`.
.
The PostgreSQL Project thanks Shishir Sharma for reporting this
problem. (CVE-2026-14663)
.
+ Fix psql to skip in-line data following a scripted COPY ... FROM
STDIN command, even if the COPY fails before sending
`PGRES_COPY_IN` (Tom Lane)
.
Previously, if a `COPY` command failed at startup (for instance,
because the target table doesn't exist) psql would not realize
that and would proceed to read the following in-line data as SQL
commands. In the best case that's wrong and in the worst case
it's a SQL-injection hazard. Teach psql to recognize
syntactically-valid COPY ... FROM STDIN commands and to skip
data on its own authority if the server doesn't respond with
`PGRES_COPY_IN`.
.
While this fix is unlikely to affect any production SQL scripts,
test scripts might intentionally exercise failing COPY ... FROM
STDIN commands. Those will need to gain a `\.` data terminator
line after each such command.
.
The PostgreSQL Project thanks Alexander Lakhin for reporting
this problem. (CVE-2026-6464)
.
+ Cross-check the output row type of a portal running EXECUTE or
FETCH (Robert Haas)
.
EXECUTE and FETCH use two portals: an outer one for the
statement itself, and an inner one running the query being
executed on its behalf. It was previously possible to make the
declared row types of the two portals diverge, leading to server
memory disclosure and arbitrary code execution.
.
The PostgreSQL Project thanks Ben Morris (in collaboration with
Claude and Anthropic Research) and Peter Geoghegan for reporting
this problem. (CVE-2026-16239)
.
+ Fix buffer overrun with long time zone abbreviation in to_char()
(Tom Lane)
.
This can easily crash the server, and exploits leading to
arbitrary code execution have been reported.
.
The PostgreSQL Project thanks Hcamael, Amjad Shahzad, Tan Zhen
of AntAISecurityLab, Tomer Fichman, Zheng Yu, Amy Burnett
(OpenAI Codex Security), Rick de Jager, Heewon Song, Sylvie
Mayer, Aleksander Alekseev, and Hillai Ben Sasson for reporting
this problem. (CVE-2026-14669)
.
+ Fix buffer overrun in regexp match/split functions (Masahiko
Sawada)
.
If passed invalidly-encoded data, these functions could write
past the end of their conversion buffer.
.
The PostgreSQL Project thanks Francesco Verardi for reporting
this problem. (CVE-2026-14664)
.
+ Harden the ascii() function against invalid input (Michael
Paquier)
.
By supplying invalidly-encoded input, this function could be
coaxed to read and return a few bytes of data that it shouldn't.
In assert-enabled builds, its assertions could be triggered too.
.
The PostgreSQL Project thanks Hcamael for reporting this
problem. (CVE-2026-18024)
.
+ Fix multirange type handling in pg_restore_attribute_stats()
(OpenAI Security Research Team)
.
pg_restore_attribute_stats() treated multirange types just like
their underlying range type. This works correctly for the bounds
histogram, but it was wrong for all the other statistics kinds.
.
The PostgreSQL Project thanks Amy Burnett (OpenAI Codex
Security) for reporting this problem. (CVE-2026-16238)
.
+ Make scalarineqsel() check that a constant it expects to be of
type tid actually is (Tom Lane)
.
This expectation will hold for all the built-in operators that
use this estimator, but a maliciously-constructed operator could
violate it, leading to a crash or server memory disclosure.
.
The PostgreSQL Project thanks Hcamael for reporting this
problem. (CVE-2026-14668)
.
+ Harden tsvector and tsquery code against overly long values
(both individual lexemes and total vector/query length) (Tom
Lane)
.
The documented limits were not enforced in all code paths.
.
The PostgreSQL Project thanks Yuhang Wu, Zhenpeng Lin, Zheng Yu,
and Hcamael for reporting these problems. (CVE-2026-14662)
.
+ Fix various places that mistakenly assumed they would not have
to deal with more than `FUNC_MAX_ARGS` function arguments (Tom
Lane)
.
Notably, the server's actual limit on the number of arguments to
an aggregate function is `FUNC_MAX_ARGS - 1`, but the parser
failed to enforce that, creating hazards downstream.
.
The PostgreSQL Project thanks Zheng Yu, ylwangtju, and Masahiko
Sawada for reporting these problems. (CVE-2026-14679)
.
+ Reject calls from SQL to functions that take or return type
internal (Tom Lane)
.
The existing defenses against doing this have been shown to be
insufficient, so add more explicit checks.
.
The PostgreSQL Project thanks Amy Burnett (OpenAI Codex
Security) for reporting this problem. (CVE-2026-14680)
.
+ Preserve the ownership of extended statistics objects when they
are rebuilt by ALTER TABLE (Masahiko Sawada)
.
Previously, the role running ALTER TABLE gained ownership of
such objects, but that seems inappropriate.
.
The PostgreSQL Project thanks Noah Misch for reporting this
problem. (CVE-2026-6469)
.
+ When deparsing an EXTRACT() function call, quote the field name
if needed (Nathan Bossart)
.
The parser accepts any string literal as a field name in
EXTRACT(), deferring validation to execution. If the call is
stored and deparsed (for example during pg_dump), the string
body was regurgitated verbatim, allowing SQL injection.
.
The PostgreSQL Project thanks Ben Morris (in collaboration with
Claude and Anthropic Research) for reporting this problem.
(CVE-2026-15741)
.
+ Check for `USAGE` privilege on data types in places that
formerly failed to check that (Nathan Bossart)
.
CREATE TYPE AS RANGE did not check, nor did ALTER TABLE OF, nor
did commands that create stored expressions. These omissions
allowed roles without `USAGE` privilege to nonetheless create
objects depending on the type, possibly blocking the type's
owner from changing the type later.
.
The PostgreSQL Project thanks Jingzhou Fu for reporting this
problem. (CVE-2026-6470)
.
+ Invalidate role-dependent cached plans after role changes (Ilya
Staroverov, Shinya Kato, Nathan Bossart)
.
Role membership, role attribute, and database ownership changes
may impact the expected behavior of row-level security policies,
but previously we'd continue to use cached plans that were made
according to the old state of affairs.
.
The PostgreSQL Project thanks Ilya Staroverov and Shinya Kato
for reporting this problem. (CVE-2026-14666)
.
+ Reject GSSEncRequest after direct SSL connection (Michael
Paquier)
.
After establishing a TLS-encrypted connection, the server would
still accept a request for GSSAPI encryption. If that succeeded,
the connection would proceed using TLS encryption, but it would
look like a GSS connection to the pg_hba rules. Thus, a pg_hba
policy intending to disallow TLS would not be enforced
correctly.
.
The PostgreSQL Project thanks p4p3r for reporting this problem.
(CVE-2026-14681)
.
+ Make mock SCRAM authentication secrets more plausible (Nathan
Bossart)
.
If a SCRAM login is attempted against a role that doesn't exist
or doesn't have a SCRAM secret, we generate a mock secret and
carry out the authentication handshake anyway, to avoid
revealing these facts to an attacker. But the mock secret was
made with a fixed iteration count, which in itself can be an
observable response discrepancy. Use the configuration setting
`scram_iterations` instead, to make the mock secret look more
like the installation's real secrets.
.
The PostgreSQL Project thanks Radim Marek for reporting this
problem. (CVE-2026-14672)
.
+ Fix out-of-bounds writes in ecpg applications caused by invalid
bytea data received from the server (Michael Paquier)
.
ecpg assumed without checking that any bytea value must begin
with `\x`. A broken or malicious server might send a string
shorter than 2 bytes, resulting in memory clobber in the
application.
.
The PostgreSQL Project thanks ylwangtju for reporting this
problem. (CVE-2026-16241)
.
+ Do not do backquote expansion on the argument of psql's
\unrestrict command (Nathan Bossart)
.
This oversight in the fix for CVE-2025-8714 allows a malicious
server to inject shell commands into plain-text dump output that
will be run at restore time on the machine running psql, the
exact scenario that CVE-2025-8714 intended to prevent.
.
The PostgreSQL Project thanks Lucas Velgus, Filip Janus, and
Daniel Bakker for reporting this problem. (CVE-2026-18408)
.
+ Remove pg_dump's assumption that pg_proc.protrftypes cannot have
more than `FUNC_MAX_ARGS` entries (Tom Lane)
.
Since there could be entries for both input and output
arguments, it's feasible for this array's length to exceed
`FUNC_MAX_ARGS` (which constrains only input arguments). Even if
that were not so, pg_dump cannot assume that the server was
built with the same value of `FUNC_MAX_ARGS` that it has. An
overrun would lead to a memory clobber inside pg_dump.
.
The PostgreSQL Project thanks Masahiko Sawada for reporting this
problem. (CVE-2026-19385)
.
+ Harden PL/Perl against "tied" Perl arrays and hashes (Tom Lane)
.
A tied object that doesn't behave like a regular one could lead
to memory overwrite, or to constructing a corrupt result array
(which would likely cause problems later).
.
The PostgreSQL Project thanks Hcamael for reporting this
problem. (CVE-2026-14670)
.
+ Fix integer overflows in memory-allocation calculations in
PL/Perl and PL/Tcl (Heikki Linnakangas)
.
This is the same type of problem as CVE-2026-6473, just in a
different part of the code, and is fixed in the same way.
.
The PostgreSQL Project thanks the Tulya Project (Team Dhiutsa,
Bitecope Technologies Private Ltd) for reporting this problem.
(CVE-2026-14677)
.
+ Ensure that contrib/amcheck functions restrict `search_path`
before executing index expressions (Noah Misch)
.
Because amcheck will run such index expressions as the owner of
their tables, a caller could potentially hijack
`search_path`-dependent functions to run arbitrary code as the
table owner. By default this is not a vulnerability because only
superusers are allowed to call amcheck functions; but if that
privilege was granted out, it created a larger hazard than the
documentation suggests.
.
The PostgreSQL Project thanks Yuelin Wang and Jacob Brazeal for
reporting this problem. (CVE-2026-14673)
.
+ Fix integer overflows in contrib/fuzzystrmatch's levenshtein()
and levenshtein_less_equal() functions (Nathan Bossart)
.
Passing large cost values to these functions could cause integer
overflows, thereby producing nonsensical results, and even
causing out-of-bounds writes in some cases.
.
The PostgreSQL Project thanks Ben Morris (in collaboration with
Claude and Anthropic Research) for reporting this problem.
(CVE-2026-15742)
.
+ Fix buffer overrun in contrib/pg_stat_statements (Álvaro
Herrera)
.
Query normalization didn't accurately account for the amount of
space the normalized string would require.
.
The PostgreSQL Project thanks Sajeeb Lohani (with TrendAI Zero
Day Initiative) and Yuelin Wang for reporting this problem.
(CVE-2026-14676)
.
+ Fix datatype error in contrib/pg_trgm's GiST picksplit function
(Heikki Linnakangas)
.
This mistake resulted in reading past the end of the buffer,
typically causing bad split decisions; but a crash could ensue
if you're very unlucky.
.
The PostgreSQL Project thanks Mehmet D. Ince for reporting this
problem. (CVE-2026-14678)
.
+ Remove the plan cache in contrib/refint (Ayush Tiwari)
.
This caching behavior has several serious bugs, notably that
check_foreign_key() embeds the new key values in its
cascade-UPDATE queries, so a cached plan reuses the
originally-needed values rather than the key values that should
be used. The simplest solution is to remove it.
.
The PostgreSQL Project thanks Hcamael for reporting this
problem. (CVE-2026-14671)
.
+ Ensure that parallel GIN index builds update the table's
pg_class.reltuples value correctly (Jan Nidzwetzki, Tomas
Vondra)
.
A parallel worker could report an uninitialized value for the
number of rows it processed, leading to a bogus value for
reltuples, even Infinity or NaN. Such values could lead to
subsequent autovacuum and autoanalyze operations never deciding
that the table needs to be processed. If so, the situation will
not self-heal. A manual ANALYZE command, or creation of another
index, will be needed to reset reltuples to the correct value.
If you have any tables with GIN indexes, it's recommended to
check to see if their reltuples entries look sane. A query such
as this may be helpful:
.
SELECT DISTINCT t.oid::regclass, t.reltuples
FROM pg_class t
JOIN pg_index i ON t.oid = i.indrelid
JOIN pg_class ic ON i.indexrelid = ic.oid
WHERE t.relhasindex AND ic.relam = 2742;
.
* Fix psql -c 'truncate/create table; copy from stdin'.
Checksums-Sha1:
a85790423b6d52bc11dbdbcb76848475f2f79f66 4752 postgresql-18_18.6-1.dsc
a0b624ab95fd6fae676bc929d0e6fbcbc777c327 22632314 postgresql-18_18.6.orig.tar.bz2
65d427a499155e6d53ed332e9c92c9b26f95de06 34340 postgresql-18_18.6-1.debian.tar.xz
Checksums-Sha256:
023d59351ad6f78714a2b111814736976948358d89fdf4782a85ac3572436240 4752 postgresql-18_18.6-1.dsc
555610c24d53e4316da5b7d3fc25c279d96856d5e0e23ee308c328c5fa881d9f 22632314 postgresql-18_18.6.orig.tar.bz2
83e02f228f8fb453a50105fc6b5ee48e4c15e50ce351665b8d12eb280b82d33f 34340 postgresql-18_18.6-1.debian.tar.xz
Files:
08b022014e622a79c8294eccacd6076d 4752 database optional postgresql-18_18.6-1.dsc
6143c18ab91bdfcaa056a61861814cde 22632314 database optional postgresql-18_18.6.orig.tar.bz2
148628ba342e6812ec4ef775c69f4636 34340 database optional postgresql-18_18.6-1.debian.tar.xz
-----BEGIN PGP SIGNATURE-----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=6glP
-----END PGP SIGNATURE-----
