Hi!
On Fri, 2026-11-13 at 15:00:17 +0100, Guillem Jover wrote:
> On Fri, 2026-11-13 at 14:45:52 +0100, Guillem Jover wrote:
> > Package: release.debian.org
> > Severity: normal
> > Tags: trixie
> > X-Debbugs-Cc: [email protected]
> > Control: affects -1 + src:dpkg
> > User: [email protected]
> > Usertags: pu
>
> > [ Reason ]
> >
> > This update includes several (low severity) security fixes, several
> > memory leak fixes, and a regression in the Perl API.
> >
> > [ Impact ]
> >
> > The security issues are either a potential DoS vector after crashing
> > dpkg-deb due to bogus data parsing and leaving temporary files behind
> > (and fill up disk on an automated system for example), or directory
> > traversal bugs when handling either .deb or .dsc untrusted packages
> > (the latter could be used to exfiltrate secrets from automated CI or
> > testing systems for example).
> >
> > The memory leaks, well leak. :)
> >
> > The Perl API breakage is a regression that changed the semantics of the
> > comparison functions, which then started giving wrong results.
> >
> > [ Tests ]
> >
> > The changes that are only in the latest dpkg in sid, are either trivial
> > (but still tested manually), or have tests covering their behavior.
> >
> > [ Risks ]
> >
> > Most of the included fixes have been in sid for a long while now, with
> > no reported regressions. The new changes that are only in sid, as the
> > potentially riskier ones, but as mentioned above should either be
> > trivial or be covered by tests.
>
> Ah, had forgotten. For the dpkg-deb symlink directory traversal issue, I
> had initially implemented a fix to completely ban symlinks in control.tar
> .deb members, but after pondering a bit, it looked too risky for a stable
> update as a potential source of regressions on out of archive packages.
> So I've gone instead with disallowing directory traversal for now, and
> will consider banning them with a deprecation transition starting in sid,
> as mentioned in the relevant commit message (in the ChangeLog).
>
> > Also I'm sorry this came this late, as this means the newest changes
> > are not yet in testing, so I would understand if you do not feel
> > comfortable accepting this. The reason is a combination of the
> > changes included not fitting entirely in a security update, me not
> > having found the needed blocks of time, and when deciding it would
> > be better to handle this via a stable update, not wanting to publish
> > this too much in advance (even if the security implications are not
> > high), and then getting a new security report at the beginning of this
> > week, and then the Perl transition starting and not being able to upload
> > dpkg to sid at that time either. :)
> >
> > [ Checklist ]
> >
> > [√] *all* changes are documented in the d/changelog
> > [√] I reviewed all changes and I approve them
> > [√] attach debdiff against the package in (old)stable
> > [√] the issue is verified as fixed in unstable
> >
> > [ Changes ]
> >
> > The detailed explanation of all the changes is included in the ChangeLog
> > in the debdiff.
> >
> > [ Other info ]
> >
> > As usual, I've included the full debdiff, and the following can be
> > used to filter all autogenerated stuff from it:
> >
> > ,---
> > xzcat dpkg-1.22.22-1.22.23.debdiff.xz \
> > | filterdiff -x '*.po' -x '*.pot' -x '*.in' -x '*/man/*/*.pod' \
> > -x '*/configure' -x '*/build-aux/*' -x '*/src/at/*' \
> > | less
> > `---
Nilesh mentioned on #debian-qa that the dpkg upload to sid broke
lintian's test suite, due to the «erasedir("$dir/.pc");» security fix.
So I'm assuming a lintian pu would also be needed to fix its test suite
otherwise it would start to FTBFS in stable. Lintian maintainers CCed.
Thanks,
Guillem