Details
-
Type:
Bug
-
Status: In Progress
-
Priority:
Critical
-
Resolution: Unresolved
-
Affects Version/s: N/A
-
Fix Version/s: 10.0
-
Component/s: Plugin - Audit
-
Labels:
Description
if an external program will move or remove the audit log, a segmentation fault will occur. this error will not happen, if the audit log file exist while mysql is started. if flush logs is executed after file has been moved but before stop+start of audit is triggered (via setting the file path or setting audit=OFF and audit=ON, see logrotate config), the probability of getting a segfault will be reduced.
problem will occur with audit plugin (file rotation disabled) + mysql 5.6 + ubuntu 14.04 on git commit cc84ac3be41d9d6ac480d55449d5bf4e324cca10
gdb output: http://pastebin.com/ccs2Fug7
logrotate script: http://pastebin.com/CnLANyEm
mysql error log: http://pastebin.com/P51NsFxs
my naive solution would be to call stop_logging/start_logging on rotate() if file rotations are set to 0. this will trigger the needed reopen of the logfile
Gliffy Diagrams
Attachments
Activity
- All
- Comments
- Work Log
- History
- Activity
- Transitions
i added the versions because there is no mysql/audit version available and i'm using the current trunk version of audit plugin. i'm using ubuntu 14.04 LTS and my mysql version is 5.6.19. could you find the reason for 5.6.23?
removed affected versions
Apparently, it was this commit on MySQL 5.6 after which the problem disappeared:
commit 38dcb9fb00d5cb1a398977edc56fa0f454037498
Author: Kristofer Pettersson <kristofer.pettersson@oracle.com>
Date: Tue Mar 17 13:59:31 2015 +0100
Bug #20536590: BACKPORT BUG #12368204 AND BUG #16953758 TO MYSQL-5.6
Backported the fixes.
Partially backported the fix for bug #19463877 too :
the part about different audit events count for the different
protocols.
Changed the strategy of the original fix to account for the lack
of the changes done in WL#6613 as follows:
Instead of mixing the calls to the audit log API by calling them
both outside of the general log hooks and inside some
moved all audit log API calls inside the general log hooks but
before any checks to the general log options. Made sure that the
general log hooks are called regardless of the value of the
general log option and moved all checking of it inside the
hooks.
Changed the binary log calles used by the prepared statement
code to use the convenience hooks instead of directly calling the
LOGGER functions. Ensured that prepared statements pass down the
query text.
Getting it from the THD is not correct for prepared statements.
** Added an extra test file that tests firewall with general log off.
Still, Alexey Botchkov, could you please double-check that there is nothing to do/fix on the audit plugin side?
i can confirm that this commit will fix the issue. i created a backport patch for current mysql-server-5.6 version on trusty: https://bugs.launchpad.net/ubuntu/+source/mysql-5.6/+bug/1444904
i was able to fix this problem for mysql on plugin level too. could you please take a look at the attached patch?
created pull request https://github.com/MariaDB/server/pull/46
any progress on this issue?
Hi,
Are you really also getting the crash on MariaDB 5.5, 10.0, 10.1.1 as the 'Affects Version/s' field states?
And which exact version of MySQL 5.6 are you using?
I can reproduce it on MySQL 5.6.23, but so far only there. Couldn't reproduce on MariaDB, or on MySQL 5.5, or on MySQL 5.6.24.
It's possible though it's just the matter of "luck"...