The Samba-Bugzilla – Bug 370
Garbage in locking database after 7..10 days of uptime
Last modified: 2005-11-14 09:25:03 UTC
Linux Debian 3.01, Samba 2.2.8a self-compiled, kernel 2.4.21 self-compiled.
After 1..2 weeks of normal operation the first symptoma of upcoming crash are
bogus records in "smbstatus" output: non-existent (actually, already finished
without error) smbd process keeps a file with random unreadable 8-bit "name". In
1..2 days the anmount of such lines increase. Then, regular client application
failures. Samba restart with locking.tdb deletion does not help; problem returns
in 1 day. Linux restart helps, for same 1..2 weeks.
Maximal number of (true) smbstatus records during normal operation is like 800
total, 300 per client. Just Clipper and MS-Access-driven accounting.
Replaced: all hardware (motherboard, power, memory, HDD for /var), kernel 2.4.19
to 2.4.21, kernel SMP to single, Sambe 2.2.7 to 2.2.8a. The problem persists.
Non-standard features: ACL kernel patch, ACL in Samba, and ACL attributes at
Seems to help for the price of significant databases performance downgrade: "use
mmap = no". Not sure, because of insufficient test duration (3 weeks). maybe, it
just delays the crash.
Sorry, but the 2.2 is not under development any longer.
If you can reproduce this bug against the latest 3.0 release,
please reopen this bug and change the version in the report.