Re: [dispatch] Work for IETF114

[Mark Nottingham <[e-mail]>]

Tim's approach is the direction I'd encourage you to go in -- being able to reusing an existing scheme has a lot of merit.

Cheers,


> On 11 Jun 2022, at 2:06 am, Tim Bray <[e-mail]> wrote:
> 
> You know, you wouldn't have to invent *that* much new. When you retrieve https://example.com/foo#bar, the "#bar" part isn't sent to the server, and the rules say it is interpreted with respect to the content-type of whatever you get.  It's called the Fragment Identifier (https://datatracker.ietf.org/doc/html/rfc3986#section-3.5) and normally used to identify a fragment within the resource representation, but let's ignore that.  So suppose you invent a new media type, say app/encrypted-file.  Then you invent a tiny little language to go in the fragment identifier. E.g.
> 
> https://example.com/#KeyID=48954289574239
> 
> or possibly even
> 
> https://example.com/#KeyHost=tbray.org&KeyID=48954289574239
> 
> When this thing is retrieved (without the #fragment) you get something with Content-type: app/encrypted-file
> 
> Now, there'd need to be serious thinking about chain of trust.  Obviously you'd require trust and encryption for the channel over which the URI-with-fragment is transmitted.  Then there's the matter of where the key comes from. In the second form, one assumes the key is in example.org//.well-known/something  and the receiver only really needs to decide to trust example.org.  The first form, where there's a pre-existing out-of-band arrangement about where to get keys from, is attractive.  I think you still need the KeyID thing.
> 
> Wouldn't be that hard to implement.
> 
> 
> On Fri, Jun 10, 2022 at 6:22 AM Bron Gondwana <[e-mail]> wrote:
> Yep, me again with something new that's bubbling up out of the "large files over email" problem.
> 
> I'm not quite sure the shape of this, but I'd like some time to talk about something which might be a media type, might be a URI scheme, might be... I dunno.  I'm sure the group will have zillions of opinions about both what the work should be, and what it should look like.  I can give you the user experience and use-case that I want anyway.
> 
> ...
> 
> Use case: user wants to store a file encrypted on a server, and provide a link such that their recipient can get the decrypted file, without the server ever seeing the plaintext content.
> 
> Which basically means - the URL/URI that the sender provides to the recipient contains the key data to decrypt the content, encoded into the URI in such a way that it's not provided to the server.  There are pros and cons of different approaches, but ideally the operating system / browser / library transparently does the decryption as part of the fetching such that a piece of client software doesn't care if the data is stored in plaintext and transferred over HTTPS (or whatever), or stored encrypted on the server with zero knowledge - either way the client program / library gets the raw data plaintext and doesn't have to care what the sender intended.
> 
> Obviously, this only works if the recipient supports it, so bootstrapping is going to be a pain.  But I haven't seen something like this existing, and it would be SOOOOOO much nicer to have this transparently able to be embedded in the URI and handled by lower levels rather than having to design and specify an optional encryption layer in the email large file handling stuff.
> 
> ...
> 
> I'll put together a better proposal in the next couple of weeks, but just requesting the agenda time now :)
> 
> And I would LOVE it if someone already knows that such a thing exists and can point me to it instead of having to reinvent the wheel.
> 
> Cheers,
> 
> Bron.
> 
> --
>   Bron Gondwana, CEO, Fastmail Pty Ltd
>   [e-mail]
> 
> 
> _______________________________________________
> dispatch mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/dispatch
> _______________________________________________
> dispatch mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/dispatch

--
Mark Nottingham   https://www.mnot.net/