[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

Re: [PATCH] reject SHA1 collisions (was: Re: Progress on SHA-1 fixes in patch releases?)

From: Johan Corveleyn <jcorvel_at_gmail.com>
Date: Tue, 9 May 2017 18:48:03 +0200

On Tue, May 9, 2017 at 6:27 PM, Mark Phippard <markphip_at_gmail.com> wrote:
> On Tue, May 9, 2017 at 12:08 PM, Stefan Sperling <stsp_at_elego.de> wrote:
>>
>> On Tue, May 09, 2017 at 03:44:22PM +0000, Daniel Shahaf wrote:
>> > Stefan Sperling wrote on Tue, May 09, 2017 at 15:25:23 +0200:
>> > > This could be further extended by the config knob to give users a
>> > > choice.
>> > > I don't see a good way of adding such a knob in a patch release,
>> > > though.
>> >
>> > Just give the knob a name that indicates it's not forward compatible?
>> >
>> > For illustration, if the knob in 1.10 will be called "foo", then the
>> > knob in 1.9.6 could be named "SVN_NFC_foo", where the prefix stands for
>> > "svn not forward compatible".
>>
>> For a patch release I would generally prefer a simple solution that
>> does not add knobs. A fix that people can install and forget about
>> is going to be appreciated the most after all the hype and worrying
>> this problem has caused.
>>
>> And I wonder who would really want to tweak such a knob in the first
>> place.
>> People who really wish to store SHA1 collisions in their FSFS repository
>> could just disable rep-sharing, couldn't they?
>
>
>
> From the best I can tell we have no plan on how or when we could support
> this in the working copy. I have also seen a lot of people express interest
> in the hook scripts to block the sha1 collisions and not any real
> conversation about wanting to fully support storing collisions. With that
> in mind, why not just simplify this and say we have no plans to provide
> support for supporting two files with the same sha1 and put the code in
> place in 1.9.x to block this from happening. This removes the need for
> people to add hooks and it solves the problem.
>
> If and when someone can state enough of a need for storing files with the
> same sha1 let them step forward with a proposal and code for how to do this.
> Until then it seems like just flat out blocking this from happening is
> better then a half hearted attempt to support it.
>
> This approach does not need any knobs or hooks.

Yeah, I was pushing on IRC for making it configurable (with some hope
that a future client release might support it, and then a 1.10
repository could be "opened up" if the svn admin wanted to allow
collisions). But I'm reconsidering. I guess you guys are right ...
nobody is really asking today for full support of sha1 collisions. So
better to keep it simple now, and flat out block them (and leave "full
sha1 collision support" for some distant future when someone really
wants / needs it.

And like Stefan said: one can always disable rep-sharing. Speaking of
which: we must make it really clear in docs that this collision
rejection will only work with if rep-sharing is on. If needed, admins
can (re-)enable rep-sharing for an existing repository (as long as a
collision hasn't been committed yet), right?

-- 
Johan
Received on 2017-05-09 18:48:27 CEST

This is an archived mail posted to the Subversion Dev mailing list.

This site is subject to the Apache Privacy Policy and the Apache Public Forum Archive Policy.