Comment # 7 on bug 1020028 from Jonathan Kang

(In reply to Michael Andres from comment #6)
> This bug, as well as bug#955557, suffer from a buffer overflows in
> backend_get_updates_thread. This one inside libzypp the other one outside. 
> 
> OTOH, libzypp (and also it's PathInfo class) is heavily used by zypper and
> yast. All their tests succeed on i586. Also no similar bug reports from i586
> systems about zypper/yast.
> 
> Also no problems in other backend jobs; just in backend_get_updates_thread.
> 
> Franky, I doubt that we'll find the problem inside the libzypp code.
> 
> 
> (In reply to Jonathan Kang from comment #3)
> > We can see the stack smashing happened in
> > libzypp-16.4.3/zypp/PathInfo.cc:172.
> > 
> > I looked the implementation of PathInfo class, and found it calls xstat64()
> > somehow in this 32-bit machine, which is quite odd.
> 
> (https://www.tcm.phy.cam.ac.uk/sw/inodes64.html) The 64 in stat64 refers to
> the size of the filesystem inodes, not tho the systems architecture. The
> glibc stat() wrapper function hides these details from applications and
> under the hood calls whatever is appropriate.

Replacing
> if (PathInfo("/etc/PackageKit/ZYpp.conf").isExist())
with
> if (g_file_test ("/etc/PackageKit/ZYpp.conf", G_FILE_TEST_EXISTS))
to check if "/etc/PackageKit/ZYpp.conf" fixes the problem.

So I highly doubt that there is something wrong with PathInfo.