Previous code would store md5sum of an empty line as the one
identifying the monitored file. That file then was read and possibly
failures were found. Upon next "container.open()", md5 digest of now
present first line was compared against previous digest of an empty
line, which was different, thus file was assumed to be rotated and all
the log lines were read once again.
The History
-----------
In rare cases various tests failed somewhat consistently. Below you
can find one case in test_move_file where such failure really made no
sense -- we should have not had 4 failures by that point.
Fail2ban 0.8.10.dev test suite. Python 2.4.6 (#2, Sep 25 2009, 22:22:06) [GCC 4.3.4]. Please wait...
I: Skipping gamin backend testing. Got exception 'No module named gamin'
I: Skipping pyinotify backend testing. Got exception 'No module named pyinotify'
D: fn=/home/yoh/test/monitorfailures_FilterPoll9nUKoCfail2ban-0 hashes=d41d8cd98f00b204e9800998ecf8427e/d41d8cd98f00b204e9800998ecf8427e inos=5398862/5398862 pos=0 rotate=False
D: Adding a ticket for ('193.168.0.128', 1124013599.0, ['Aug 14 11:59:59 [sshd] error: PAM: Authentication failure for kevin from 193.168.0.128\n'])
D: Adding a ticket for ('193.168.0.128', 1124013599.0, ['Aug 14 11:59:59 [sshd] error: PAM: Authentication failure for kevin from 193.168.0.128\n'])
D: Closed <closed file '/home/yoh/test/monitorfailures_FilterPoll9nUKoCfail2ban-0', mode 'r' at 0x7f472b2efdc8> with pos 1231
D: fn=/home/yoh/test/monitorfailures_FilterPoll9nUKoCfail2ban-0 hashes=d41d8cd98f00b204e9800998ecf8427e/aeb4e73e6922a746d027eb365ece2149 inos=5398862/5398862 pos=1231 rotate=True
D: Adding a ticket for ('193.168.0.128', 1124013599.0, ['Aug 14 11:59:59 [sshd] error: PAM: Authentication failure for kevin from 193.168.0.128\n'])
D: Adding a ticket for ('193.168.0.128', 1124013599.0, ['Aug 14 11:59:59 [sshd] error: PAM: Authentication failure for kevin from 193.168.0.128\n'])
D: Closed <closed file '/home/yoh/test/monitorfailures_FilterPoll9nUKoCfail2ban-0', mode 'r' at 0x7f472b2ef558> with pos 1231
F
======================================================================
FAIL: test_move_file (testcases.filtertestcase.MonitorFailures<FilterPoll>(/home/yoh/test/monitorfailures_FilterPoll9nUKoCfail2ban))
----------------------------------------------------------------------
Traceback (most recent call last):
File "/home/yoh/deb/gits/fail2ban/testcases/filtertestcase.py", line 451, in test_move_file
"Queue must be empty but it is not: %s."
AssertionError: Queue must be empty but it is not: server.ticket.FailTicket: ip=193.168.0.128 time=1124013599.0 #attempts=4.
----------------------------------------------------------------------
N.B.1 I preserved here and in the code corresponding additional debug
print statements, which are commented out by default. sensible md5
digest was generated by using hexdigest() instead of current a bit
faster digest(). Running tests with all the debug output simply
breaks the race loose and failure doesn't trigger.
N.B.2 d41d8cd98f00b204e9800998ecf8427e is an md5sum of an empty
string, and aeb4e73e6922a746d027eb365ece2149 of the first line in that
file.
mtime alone is a poor measure here as many of our tests shown -- on older Pythons and some file systems
mtime might be reported only with precision up to a second. If file gets rotated fast, or there are new
modifications within the same second, fail2ban might miss them with Polling backend if judging only by
mtime. With this modification we will track also inode and size which are all indicative of a file
change.
* 'dovecot' of https://github.com/grooverdan/fail2ban:
ENH: remove non-capturing groups for readibility
BF: fix dovecot filter for when no TLS is enabled on pop/imap
Conflicts:
ChangeLog -- changelog entries. Also untabified few other spots
* 'exim' of https://github.com/grooverdan/fail2ban:
DOC: credits/blame to me for changes to exim
ENH: new exim filter regexs. Also note a begining PID in this format. Thanks to ftoppi for the log entries
TST/ENH: Improve regex around exim
otherwise it
1. 'interferes' with meaninful output of the client
2. if ERROR is logged it better go to stderr and separating ERROR from other levels is not that transparent with python's logging
Yaroslav Halchenko