[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Bug#673065: rred segfaults during apt-get update



On Wed, May 16, 2012 at 12:30 PM, Daniel Hartwig <mandyke@gmail.com> wrote:
> On 16 May 2012 04:09, Paul Menzel <pm.debian@googlemail.com> wrote:
>
>> on Saturday,  APT was upgraded from 0.9.2 to 0.9.3. But I saw a segfault
>> from rred just today.

Just to be sure (you know, clutching at a straw):
You upgraded also libapt-pkg4.12, right? (or preferable your complete system)
as the supposed fix for this is not in the apt package, even through
the failing program (rred) is in it.
(It was and presumable still is FileFd::Close failing)


> So these are the same symptoms you reported in #656865.  In that
> report you mention that 'apt-get update' fails "most of the time",
> would you say this is still true for 0.9.3?

I haven't seen it that often (in fact only once) with 0.9.2 and this segfault
was fixed with 0.9.3, but these memory bugs tend to be really nasty as
just reordering the involved code or something as simple as enabling
debug (or adding output by hand) "fixed" the bug.

So it's possible i only fixed one of possibly many or none of them,
just another which happened to "order" everything "correctly".

A bit mysteries is that valgrind doesn't talk about accessing invalid
memory, it only happens if rred is called from apt - if i call it by hand
with the needed files everything works as expected, as it does if you rerun
apt - and it only happens in rred even through the code which segfaults is
the closing of a (gzip compressed) file, so that the gzip method should fail,
too (but it is not used that much anymore, so we might just not see it), and
given that the bzip2 code isn't that different it should be failing, too
(and bz2 is run all the time as it is the preferred compression type).

(From the provided traces aswell as my own tests with reordered code it
 not only happens with gzip files, it does also with uncompressed, but this
 makes even less sense as we open files in a lot of places and not just rred,
 so i presume it first gets it wrong with gzip but survives deathly wounded
 finally dying "at random" files)


>> Unfortunately I do not know how to reproduce that problem as it only
>> happens “randomly”.
>>
>
> Not sure if this would be helpful: next time it happens and before you
> run another apt command could you take a copy of the list files:
>
> $ tar czf  /var/lib/apt/lists/

It should be, at least that is what i did and do now again in the hope that
i will hit it again to have a testcase for it… I love it™


Best regards

David Kalnischkies



Reply to: