RPM file format questions

[Panu Matilainen]

On 02/19/2012 09:23 AM, Jordan Sissel wrote:
> Howdy!
>
> I am implementing a library for reading and writing RPM files. I've used
> the rpm source and from max-rpm[1] as documentation on the file format.
>
> [1] http://www.rpm.org/max-rpm/s1-rpm-file-format-rpm-file-format.html
>
> I'm thus far able to happily read older RPMs from,CentOS5 and Fedora 10
> (and earlier). I am not able to correctly read RPMs from CentOS 6 and
> Fedora 11 (and beyond)
>
> The problem seems to be around the length of each header section
> (signature and regular header). With older RPMs, my calculation
> correctly computes the length, but for newer rpms (Fedora 11+ and
> CentOS6+) I am usually short by 4 bytes.
>
> This causes me a problem because I'll read the Signature header and fail
> when reading the next header because I'm 4 bytes behind the actual start
> of the next header.
>
> I have wrangled it down to a small example script that can read older
> rpms but fails on newer ones as described. I have also included some
> data and output as well in a gist found here:
>
> https://gist.github.com/1862256
>
> If you have any pointers on how I'm reading these newer RPM files
> incorrectly, I'd love to know :)

The fundamental package file format hasn't changed. There's a padding 
between signature and regular header which your script doesn't appear to 
take signature header padding into account, so you've basically just 
been lucky if you haven't been hit by that with older rpms.

See eg 
http://rpm.org/gitweb?p=rpm.git;a=blob;f=lib/signature.c;h=63e59c00f255a538e48cbc8b0cf3b9bd4a4dbd56;hb=HEAD#l203 
how to calculate the pad, or alternatively if you prefer reading obscure 
shell-script to C:
http://rpm.org/gitweb?p=rpm.git;a=blob;f=scripts/rpm2cpio.sh;h=b03ee361f7f6948c979195eefc50295aa8fa0cf7;hb=HEAD

	- Panu -