Submitted by: @pavel-zotov

Attachments:
core_firebird_log_+_LT_rawdump.rar

Scenario:

1. Database with table TLOCK(id int, f01 int) and single record in it (e.g ID=1, F01=100).

2. Machine #⁠1.
Connect to this database via TCP and do:
SELECT * FROM TLOCK WHERE ID=1;
UPDATE TLOCK SET F01=RAND()*10000 WHERE ID=1;
SELECT * FROM TLOCK WHERE ID=1;

3. Machine #⁠2.
Run batch (or shell) that:
3.1. Opens simultaneously N test ISQLs (I've reproduced result on at least N=10) which all should just stay quiet in "SQL>" prompt after being loaded;
3.2. Waits ~10 seconds to insure that all ISQLs do establishe connect to the database;
3.3. Runs the utility from SysInternals package to kill all ISQL processes: PSKILL.EXE ISQL;
3.4. Waits again ~10 seconds to insure that all test windows and processes are killed;
3.5. Loop to step 3.1

After some (short) time the Firebird daemon will craches with producing core file, LockTable dump and the following message in it's log:
---------------
vmoel63.local Tue Aug 20 13:37:35 2013
Fatal lock manager error: invalid lock id (143960), errno: 0

vmoel63.local Tue Aug 20 13:37:59 2013
/opt/fb30cs/bin/fbguard: /opt/fb30cs/bin/firebird terminated abnormally (-1)

vmoel63.local Tue Aug 20 13:37:59 2013
/opt/fb30cs/bin/fbguard: guardian starting /opt/fb30cs/bin/firebird
--------------

Commits: edd2f69 40bacdd d2b7575 ee0be48 FirebirdSQL/fbt-repository@ee0fee1 FirebirdSQL/fbt-repository@90f65e7 FirebirdSQL/fbt-repository@59d0569 FirebirdSQL/fbt-repository@5c72735