No i386 images in rawhide,

[Panu Matilainen]

On Wed, 8 Apr 2009, Peter Robinson wrote:

>>>>>>>>> Its just plain rpm. An example doing a simple package query gives me
>>>>>>>>> the following
>>>>>>>>>
>>>>>>>>> [root at cypher ~]# rpm -q glibc
>>>>>>>>> error: Unterminated {: {_%{_keyringpath}/*.k
>>>>>>>>>  0< /
>>>>>>>>> error: /: reading of public key failed.
>>>>>>>>> error: cannot open Packages index using db3 - No such file or
>>>>>>>>> directory
>>>>>>>>> (2)
>>>>>>>>> error: Unterminated {: {_%{_dbpat???
>>>>>>>>>  0< /
>>>>>>>>> error: cannot open Packages database in /
>>>>>>>>> package glibc is not installed
>>>>>>>>> [root at cypher ~]#
>>>>>>>>>
>>>>>>>>> Let me know if you want me to do further testing.
>>>>>>>>
>>>>>>>> Smells strongly of memory corruption... Can you run the same with
>>>>>>>> valgrind?
>>>>>>>>
>>>>>>>> # valgrind --leak-check=yes rpm -q glibc
>>>>>>>
>>>>>>> Hmm. don't have valgrind installed and rpm is screwed so I can't.....
>>>>>>> suggestions?
>>>>>>
>>>>>> rpm2cpio should work even if rpm is otherwise screwed (and if even that
>>>>>> fails, there's a script version in /usr/lib/rpm/rpm2cpio.sh). Also just
>>>>>> a
>>>>>> strace might provide some clues, if you have that installed.
>>>>>
>>>>> Any idea what provides the memcheck tool?
>>>>>
>>>>> [root at cypher bin]# ./valgrind --leak-check=yes rpm -q glibc
>>>>> valgrind: failed to start tool 'memcheck' for platform 'x86-linux': No
>>>>> such file or directory
>>>>> [root at cypher bin]#
>>>>
>>>> valgrind expects to find the bits and pieces in certain places - you'll
>>>> need
>>>> to do the rpm2cpio unpacking from / to get the paths right.
>>>
>>> Output can be found here. Let me know if you want any other things run.
>>>
>>> http://pbrobinson.fedorapeople.org/rpm-valgrind.txt
>>
>> Hmm, this appears to be F10 valgrind (==3619== Using valgrind-3.3.0), which
>> is fairly broken at least as far as rpm goes (it doesn't handle pipe2()
>> syscall which prevents macros from getting properly processed), try grabbing
>> rawhide valgrind instead and rerun.
>>
>> Some related items: despite macros being somewhat busted, it actually
>> manages to return the glibc version: glibc-2.9.90-12.i586. That is rawhide
>> alright, but then:
>>
>> ==3623==    by 0xBAB7C3: rpmInitCrypto (in /usr/lib/librpmio-4.6.so)
>>
>> ...it still has rpm 4.6.0, whereas rawhide has 4.7.0-beta1. That this
>> happens with rpm 4.6.0 too is probably a useful piece of information.
>> Since rpm seems to limp on somehow under valgrind (which is might be a clue
>> in itself), try grabbing the entire package list with 'valgrind rpm -qa',
>> maybe there are some further clues there. (and yes this is pretty much
>> shooting in the dark, trying to hunt for some clues)
>
> Basically the only thing on the box that's rawhide is glibc which is
> about where it got to before it barfed.

Ahhah, this might be an all-important clue:
* Wed Apr  1 2009 Jakub Jelinek <jakub at redhat.com> 2.9.90-12
- update from trunk
- configure with --enable-experimental-malloc

Note "experimental malloc", Apr 1st is the first known occurrence 
of what appears to be memory corruption, and except for glibc, you have 
essentially F10 system exhibiting this issue. It's at least a rather 
promising lead if nothing else.

> If you contact me off list I can give you a login to the device to 
> investigate if you like.

Thanks, I might contact you later depending on where the trail leads :)

 	- Panu -