- From: Ben Laurie <benl@google.com>
- Date: Thu, 27 Sep 2012 09:15:51 +0100
- To: Melvin Carvalho <melvincarvalho@gmail.com>
- Cc: Kingsley Idehen <kidehen@openlinksw.com>, Henry Story <henry.story@bblfish.net>, "public-webid@w3.org" <public-webid@w3.org>, Andrei Sambra <andrei@fcns.eu>
On 27 September 2012 08:56, Melvin Carvalho <melvincarvalho@gmail.com> wrote: > > > On 27 September 2012 09:49, Ben Laurie <benl@google.com> wrote: >> >> On 26 September 2012 23:43, Kingsley Idehen <kidehen@openlinksw.com> >> wrote: >> > On 9/26/12 5:18 PM, Ben Laurie wrote: >> >> >> >> On 26 September 2012 19:02, Henry Story <henry.story@bblfish.net> >> >> wrote: >> >>> >> >>> On 26 Sep 2012, at 19:10, Kingsley Idehen <kidehen@openlinksw.com> >> >>> wrote: >> >>> >> >>>> On 9/26/12 11:48 AM, Ben Laurie wrote: >> >>>>> >> >>>>> No, the point you are missing is that in capabilities the_only_ >> >>>>> authority I need to access a resource is the name of that resource - >> >>>>> the URI in your case. >> >>>> >> >>>> You can seriously believe I am missing that point while also >> >>>> espousing >> >>>> the virtues of hyperlinks as denotation mechanisms for a global web >> >>>> of >> >>>> linked data. That doesn't compute. That's a contradiction. >> >>>> >> >>>>> Security derives from the unforgeability of the >> >>>>> URI, rather than an independent system that decides if some >> >>>>> principal >> >>>>> has permission. >> >>>> >> >>>> Security is not derived from the persistence of a URI, its derived >> >>>> from >> >>>> the values exposed directly or indirectly via URI which logic >> >>>> handling >> >>>> routing. I can have many identifiers, but relationship semantics >> >>>> ultimately >> >>>> determine if I can access a resource at an address, directly or >> >>>> indirectly >> >>>> (i.e., name based indirection). >> >>> >> >>> +1 >> >>> >> >>> the idea of an unforgeable URI seems gobbledegook to me, frankly. When >> >>> people spoke of unforgeable things they spoke of things like diamonds >> >>> that >> >>> could not be copied, swords that were made to such perfection that >> >>> never >> >>> could there be two identical versions of them, etc... A URI is by >> >>> definition >> >>> something that can be copied. In fact there is no way of telling of >> >>> one URI >> >>> is an original or another a copy! >> >> >> >> This is true, and is one reason it is hard to simulate capabilities >> >> using >> >> URIs. >> > >> > >> > >> > How do you arrive at these conclusions? If you don't mind, have you >> > taken a >> > look at the Linked Data meme and what its about? How URIs resolve to >> > structured data etc. Basically, you can denote anything entity using a >> > URI. >> > Even better is you take advantage of an HTTP URI since you have the >> > power of >> > indirection in play such that a denotation resolves to representation. A >> > URI >> > becomes an extremely powerful data source name that enables you work >> > wonders >> > with structured data. You can simulate all kinds of capabilities via >> > URIs, >> > the only limits are: >> > >> > 1. our imagination >> > 2. our willingness to look at what Linked Data is really about -- forget >> > any >> > distracting political wars from the past, it's 2012, let start afresh >> > re. >> > AWWW and its capabilities. >> >> Now we're overloading "capabilities" :-) >> >> So, the point is this: object capabilities are a security mechanism, >> like ACLs. Their purpose is to restrict access to resources to only >> the intended accessors. >> >> With URIs, there are two obvious ways to implement this: >> >> 1. Make the URIs unguessable - so, I only get access to the resource >> if someone tells me the URI. >> >> 2. Link the URI to a public key - so, I only get access to the >> resource if I can prove I have the corresponding private key. >> >> The problem with 1 is that the nature of URIs makes it hard to keep >> them secret. People like to send them in emails and IMs and they leak >> quite easily. >> >> The problem with 2 (which, it should be obvious, fits quite neatly >> with WebIDs) is that it requires changes to clients and servers to do >> the key proof. Or maybe not, actually ... I guess it could be done at >> the back end, wherever you do ACL checks, by instead correlating the >> URI and the key presented in the WebID cert. >> >> So ... maybe not such a red herring after all. >> >> Not that this fixes any of the problems I've talked about :-) > > > Spot on. > > I think Tim's intention with ACLs is to bring an analogy, of the UNIX file > system, to the Web. My point is that ACLs don't work well for us in UNIX and there's really no reason to suppose they will work any better on the Web. > In a sense, PKI is only one class of solution to this kind of authentication > and authorization. A cookie, for example, could work too. Capabilities don't really have anything to do with PKI - at least, not the I part. There is no need for infrastructure - in fact, it is likely to reduce effectiveness. A cookie actually is a capability, at least as often used - i.e. an unguessable string that gives the holder access to certain resources. > >> >> >> > >> > >> > >> >> >> >>> The idea of unforgeable URIs, the idea of a web that cannot be linked, >> >>> all of these ideas seem to be like weird beasts from a netherworld >> >>> that >> >>> nobody has ever heard of, a Medusa that turns all that look at her >> >>> into >> >>> stone. >> >> >> >> Not that cannot be linked, but that can only be linked by those you >> >> choose to allow to link. The same goal as ACLs, of course. >> >> >> >> But I should not have introduced the idea, it raises as many questions >> >> as it answers. Definitely a red herring. Apologies. >> >> >> >> >> >> >> > >> > >> > -- >> > >> > Regards, >> > >> > Kingsley Idehen >> > Founder & CEO >> > OpenLink Software >> > Company Web: http://www.openlinksw.com >> > Personal Weblog: http://www.openlinksw.com/blog/~kidehen >> > Twitter/Identi.ca handle: @kidehen >> > Google+ Profile: https://plus.google.com/112399767740508618350/about >> > LinkedIn Profile: http://www.linkedin.com/in/kidehen >> > >> > >> > >> > >> > >> >
Received on Thursday, 27 September 2012 08:16:24 UTC