Software 45238 Published by

MariaDB released maintenance updates for all four of its supported LTS series, with 12.3.4, 11.8.10, 11.4.14, and 10.11.20 all receiving the exact same set of thirteen security patches. The highest-rated CVE scores 7.5, and the fixes target classic remote-attack-surface problems like buffer overflows, use-after-free bugs, and privilege-validation gaps, making upgrades a time-sensitive rather than optional task. The standout wrinkle is that the oldest maintained branch, 10.11.20, carries those identical fixes despite having just under a year of community support left before its February 2028 end of life. Tucked into the release is MariaDB Foundry, a new way to build external plugins without recompiling the whole server, signaling a broader push toward a developer-facing plugin ecosystem.



MariaDB ships four LTS maintenance releases, patching the same 13 CVEs across every active series

Every supported branch got identical security fixes with the oldest one is running out of clock.

Four of MariaDB's supported long-term support branches got maintenance updates on 16 November 2026. All four received the same 13 security patches. That last part is the one worth sitting with, because the oldest branch in that group has just under a year of community support left.

Mariadb

The versions are 12.3.4, 11.8.10, 11.4.14, and 10.11.20. The announcements landed on 18 November, split between a Foundation blog post by Frédéric Descamps (lefred) and a companion release post by Daniel Bartholomew.

Maintenance releases are the unglamorous end of the calendar. You won't find new SQL keywords or benchmark flexing here. For anyone running MariaDB in production, though, these point updates are often the software that keeps you up at night for the right reasons. They close exploitable holes, fix data-corruption bugs that only surface after a crash, and correct the subtle query errors that make wrong results look completely plausible.

The four series, briefly

12.3 is the current flagship LTS, and MariaDB is clearly pitching it as the AI-era database. Vector search optimized for OpenAI's matryoshka embeddings, a reworked binary-logging path that claims up to 4x write throughput, and a full optimizer-hint system round out the story.

11.8.10 is the previous LTS, still carrying native vector indexing, a default-character-set flip from latin1 to utf8mb4, and a TIMESTAMP range pushed out to the year 2106.

11.4.14 is the workhorse, still widely deployed and supported until May 2029. Its release notes are leaner by design, since LTS branches stabilize instead of inheriting every new feature.

10.11.20 is the oldest maintained branch, with community support expiring in February 2028. It receives the identical 13-CVE patch set anyway. MariaDB doesn't hand out a cut-down security fix for older releases, which is a fair bit of respect for operators who haven't migrated yet.

The security story

Thirteen CVEs. Highest CVSS score sits at 7.5.

The release notes don't spell out each vulnerability, but the surrounding fix categories make the threat model obvious. Buffer overflows. Out-of-bounds reads and writes. Use-after-free conditions. Format-string crashes. Privilege-validation gaps. Several of the fixes touch the CONNECT storage engine, JSON handling, the MySQL client-server protocol, authentication, and SSL/TLS certificate matching.

Since these are memory-safety and privilege issues reachable through a normal query or connection path, patching here is time-sensitive. Not optional. The fact that 10.11.20 gets the same CVEs as 12.3.4 is a pretty pointed reminder to finish that migration.

What actually got fixed

Beyond the CVEs, four areas absorbed the heaviest lifting, and these are the kinds of changes that prevent production incidents rather than headline them.

InnoDB crash recovery got the biggest batch. The engine previously failed to recover when killed mid-DDL, wouldn't start up when recovery involved multiple RENAME operations, and could even overwrite a renamed table with a freshly created one using the old name. The LRU eviction algorithm also flipped to a sweeping approach to dodge buffer-pool contention.

Then there were the correctness fixes that quietly ruin your day. Repeated references to a common table expression were reparsing exponentially in time and memory. Window functions over a frame containing NULL returned wrong results. A crafted JSON_CONTAINS_PATH input could write beyond an internal array.

Wrong results are worse than crashes, honestly. A database that returns plausible-looking garbage is far harder to catch than one that just goes down.

Head here to the release announcement.