Re: [PATCH] Avoid infinite loop in malformed packfiles

From: Jeff King <peff@peff.net>
To: Junio C Hamano <gitster@pobox.com>
Cc: "René Scharfe" <l.s.r@web.de>,
	"Ori Bernstein" <ori@eigenstate.org>,
	git@vger.kernel.org
Subject: Re: [PATCH] Avoid infinite loop in malformed packfiles
Date: Mon, 24 Aug 2020 16:52:31 -0400	[thread overview]
Message-ID: <20200824205231.GA787628@coredump.intra.peff.net> (raw)
In-Reply-To: <xmqq5z974w50.fsf@gitster.c.googlers.com>

On Mon, Aug 24, 2020 at 01:38:35PM -0700, Junio C Hamano wrote:

> Jeff King <peff@peff.net> writes:
> 
> > I think it may be worth making this a configurable value
> > (core.maxDeltaDepth or something). Nobody would generally need to tweak
> > it, but it would give an escape hatch for getting people out of a broken
> > situation ("git -c core.maxDeltaDepth=50000 repack" or similar).
> 
> ... meaning "the pack I have has overlong delta chains to read, and
> I am running repack to cut these chains down to more manageable
> level"?  Makes sense.

Exactly.

> As it may be a bit tricky to figure out where we should read such a
> configuration for those who are new to our codebase, here is an
> illustration to give a starting point.  Docs and tests are probably
> needed, too.

It may be hard to test, as I suspect modern versions of Git are not
happy to create such a deep chain. We could test with a lowered value of
the config option, though.

It may also be worth introducing a true cycle using non-git commands.
There's some coverage there in t/t5309-pack-delta-cycles.sh. I think we
were mainly concerned there with how index-pack treats them, and it
would be nice to see how other commands react. Though I guess that
creates another testing difficulty: those other commands would need a
pack index, and we'd refuse to create one. :) So I think it would
require adding code to manually create a bogus idx file (or I guess
shipping one as a fixture).

-Peff