bug reporting and file descriptors...

[Reinhard Tartler]

On Thu, Jul 22, 2010 at 23:53:15 (CEST), Bruno Girin wrote:

> On Thu, 2010-07-22 at 21:22 +0200, Reinhard Tartler wrote:
>> On Thu, Jul 22, 2010 at 14:10:23 (CEST), Martin Pitt wrote:
>> 
>> > Reinhard Tartler [2010-07-22 13:00 +0200]:
>> >> > 7 -> /home/daniel/test.mp3
>> >> >
>> >> > Would it make sense to have this information, or have I overlooked something?
>> >> 
>> >> I think it would be incredibly helpful.
>> >
>> > Note that we got quite a lot of complaints when hooks exposed
>> > arbitrary paths to files or even directories, due to privacy reasons.
>> > If we are going to do this, we need to do some anonymization there.
>> 
>> In this case, I think we can:
>> 
>>  - interactively ask the user if he/she is OK with that
>
> That's always good :-)
>
>>  - only attach the first few megabytes of the file
>
> Not sure about that: if the file is a binary file, like the mp3 above,
> you will need to attach all or none of it as a partial file is not much
> use. If it is a text file, the interesting bit may be at the end
> (especially in the case of a log file). You would also need a blacklist
> for some files so that they are never uploaded (such as /etc/shadow).

We are talking here about bugs in gstreamer having problems with certain
media files. Other use cases might have different requirements. Please
see http://ffmpeg.org/bugreports.html for FFmpeg's requirements for
upstream bug reports for reference.

MP3 files are very likely to be less than 10MB, so they will probably be
uploaded uncut. For other, larger media files it's usually not necessary
to upload the whole file, the first few packets are usually enough.

If someone wants to invest more thinking about what's the minimal amount
of the offending file that needs to be transmitted, I'd suggest to have
a look at the ffprobe tool, which may or may not be helpful here to
determine such information on a given sample.

>
>>  - randomize the filename
>
> You could use the name of the soft link, which is a sequence number and
> keep the extension, e.g. 7.mp3 (or fd7.mp3) for the mp3 file above. It
> may also be good to selectively randomize, e.g. only randomize what's
> under /home, the reason being that knowing that one of the files
> descriptors points to a well known file (like anything in /etc) may help
> analyse the cause of the bug. Having said that, knowing that one of the
> file descriptors points to /home/user/.gvfs/something-or-other may be
> useful too. Maybe replace /home/user with ~?

interesting idea.

-- 
Gruesse/greetings,
Reinhard Tartler, KeyID 945348A4