aa-logprof crash if log contains (encoded) non-utf8 filename

[Christian Boltz]

aa-logprof crashes if the log contains an (encoded) filename (for example, for a denied read access) and that filename contains a non-utf8 special character.

# cat audit.log-bärendienst
type=AVC msg=audit(1411473165.591:504): apparmor="ALLOWED" operation="open" profile="/sbin/klogd" name=2F686F6D652F7379732D746D702F62E472656E6469656E7374 pid=13468 comm="klogd" requested_mask="wc" denied_mask="wc" fsuid=1000 ouid=1000

(the original event/log was for a test profile - I changed it to klogd as an easy reproducer)

# aa-logprof -f audit.log-bärendienst
Reading log entries from audit.log-bärendienst.
Aktualisiere AppArmor-Profile in /etc/apparmor.d.
Änderungen im complain-Modus
Traceback (most recent call last):
  File "aa-logprof", line 52, in <module>
    apparmor.do_logprof_pass(logmark)
  File "/home/cb/apparmor/HEAD-CLEAN/utils/apparmor/aa.py", line 2276, in do_logprof_pass
    ask_the_questions()
  File "/home/cb/apparmor/HEAD-CLEAN/utils/apparmor/aa.py", line 1826, in ask_the_questions
    ans, selected = q.promptUser()
  File "/home/cb/apparmor/HEAD-CLEAN/utils/apparmor/ui.py", line 304, in promptUser
    cmd, arg = self.Text_PromptUser()
  File "/home/cb/apparmor/HEAD-CLEAN/utils/apparmor/ui.py", line 408, in Text_PromptUser
    sys.stdout.write(prompt + '\n')
  File "/usr/lib64/python3.4/encodings/iso8859_15.py", line 19, in encode
    return codecs.charmap_encode(input,self.errors,encoding_table)[0]
UnicodeEncodeError: 'charmap' codec can't encode character '\udce4' in position 54: character maps to <undefined>

Bonus content (note: I'm using de_DE@euro (aka iso-8859-15) as locale, and the filename is also iso-8859-15)

# LANG=C aa-logprof -f audit.log-bärendienst
Traceback (most recent call last):
  File "aa-logprof", line 52, in <module>
    apparmor.do_logprof_pass(logmark)
  File "/home/cb/apparmor/HEAD-CLEAN/utils/apparmor/aa.py", line 2248, in do_logprof_pass
    aaui.UI_Info(_('Reading log entries from %s.') % filename)
  File "/home/cb/apparmor/HEAD-CLEAN/utils/apparmor/ui.py", line 46, in UI_Info
    sys.stdout.write(text + '\n')
UnicodeEncodeError: 'ascii' codec can't encode character '\udce4' in position 36: ordinal not in range(128)

[Paweł Krawczyk (pawel-krawczyk)]

This is easy to fix - just add errors='ignore' in open_file_read() in /usr/lib/python3/dist-packages/apparmor/common.py

Nice if this could be added to the release as this is making logprof unusable if there's a single non-ASCII character in the logs.

[Christian Boltz (cboltz)]

I'm afraid it's not _that_ easy ;-)

We already use errors='surrogateescape' (or 'replace for python 2.x), which means invalid characters (non-ascii, non-utf8) in the logs shouldn't cause any problem.

However this bug is about _encoded_ invalid chars - which means we have hex string (thus ASCII) in the log which then gets decoded by libapparmor - and if the decoded string isn't valid utf8, you'll hit this bug.

If you have an idea or even a patch to fix this, it is more than welcome ;-)