Samba 4.24.6 stabilizes CephFS and CTDB clusters, rounds out the 4.24 line
The new release follows up on the CVE-heavy 4.24.5 security update, targeting a smbd panic on Ceph snapshots and race conditions in the threading pool.
Samba 4.24.6 has landed. Björn Jacke announced the release today on behalf of the Samba Team, and it's a stable bug-fix update sitting on top of the 4.24.x branch. The package contains 10 bug fixes contributed by five developers.
It's not a security release. You can breathe a sigh of relief if you survived 4.24.5. That previous update was a heavy hitter, patching six CVEs that covered everything from authenticated domain takeovers to KDC crashes via malformed ASN.1. If you missed that one, you should have applied it. This release is different. It's focused on stability for the edge cases that tend to keep sysadmins up at night.
Ceph and Clustering Take the Stage
The biggest wins in this round come from the storage and clustering camps. Samba's integration with CephFS remains a high-effort area, and two critical crash bugs get patched here.
Croit GmbH's Matthias Grandl fixed a panic in vfs_ceph_snapshots where smbd would crash when accessing snapshots on a CephFS root share. smbd panics are essentially server crashes. If you're using Ceph snapshots with a root-mounted share, you're vulnerable to this until you update. Next, Anoop C S from the Samba team squashed a crash in vfs_ceph_new when a server mixes proxy and non-proxy CephFS shares. Mixed-mode deployments were completely broken, and this fix unblocks that topology.
The density of crash bugs in the Ceph VFS modules suggests this is still a volatile area. However, at the same time, the fact that Croit and DDN are handing upstream the fixes their customers actually hit is the kind of ecosystem health that makes open source work.
DDN's Martin Schwenke contributed two protocol-level fixes for CTDB, Samba's clustering layer. One addresses missing "tickle acknowledgements" during IP takeover, and the other prevents CTDB from nesting elections in rare circumstances. Both are the kind of subtle state bugs that can trigger split-brain behavior or unnecessary failovers in production HA setups.
Internal Hardening and AD Fixes
The remaining fixes target memory leaks, race conditions, and encoding inconsistencies.
Ralph Boehme and Stefan Metzmacher tackled a memory leak in the Directory Replication Service (DRS) that occurs when replication fails. Long-running Active Directory domains with heavy topology churn could see performance degrade over time without this patch.
Volker Lendecke fixed a race condition in the pthreadpool implementation when smbd forks child processes. Race conditions are hard to pin down, but this one affects the core threading infrastructure. Metzmacher also cleaned up a read of uninitialized memory in the S3 DFS server, which could potentially expose stale data. On top of that, he resolved an inconsistency in cracknames.c that might cause DN parsing errors for accounts with binary-encoded attributes, and untangled a redundant double-encode in the NDR time value functions.
Keep in mind that one fix addresses dsgetdcname() failing to locate domain controllers when neither nmbd nor the NBT service is running. As NetBIOS continues to fade in modern Windows environments, this matters more than you might think. Clients handed a NetBIOS name could be unable to find a DC in a zero-NetBIOS setup.
Moving Forward
It's a dual-release day. Alongside 4.24.6, the team announced Samba 4.25.0rc1, the first release candidate for the next major series. The 4.25 branch is pushing experimental SMB3 Persistent Handles and introduces cluster-wide rate limiting in vfs_aio_ratelimit.
Head here to the full release notes for the complete diff. You can grab Samba 4.24.6 from here. The source tarballs come with GPG signatures, and you can verify them against key ID AA99442FB680B620.
