On 10/07/07, Johnny Bufu <[EMAIL PROTECTED]> wrote: > > On 6-Jul-07, at 3:54 AM, James Henstridge wrote: > >> Not entirely; the OP MUST NOT honor check_authentication requests for > >> shared associations (this would allow a type of attack). > > > > Okay. In that case it sounds like it would be best practice to > > generate a private association handle for each unsolicited response > > since there is no guarantee that the RP has kept hold of its > > association handle, so may not be able to verify the update if a > > pre-existing handle is used. > > The OPs knows the expiration time for each handle, so if they link > them with the update URLs they can determine their validity. > > But most of the time the updates will probably happen at much longer > intervals than association expiry time, so using private association > would be required.
The expiration time on an association handle only indicates the maximum time an RP can keep on using a particular association. The authentication specs don't mandate that the RP keeps hold of them that long (if the RP stores associations in memory, it could easily lose associations when the app restarts). The only real constraint the authentication spec places on the RP is that it maintain the association for the duration of an OpenID authentication request. With unsolicited response though, there is no prior request that tells us the RP is holding a particular association _right now_. > > Would that be appropriate to include in the spec or some best > > practices document? > > I see this as a pure OpenID (core) issue and don't feel the need to > touch it in the AX spec. That would be appropriate if the OpenID authentication spec covered the idea of unsolicited OpenID responses. Given that it does not cover unsolicited responses and the AX spec uses them, it would seem sensible for the AX spec to describe in detail how they are meant to work. > >> I don't think it's implied anywhere (or a good design) to keep state > >> between the original request and subsequent updates. So the RP cannot > >> infer the 'removed' statement just because an update did not contain > >> an attribute that was part of the original exchange. > >> > >> The update message is a fetch response, so I think it should be > >> interpreted as such by the RP: "the user has a new phone number". > >> Then the RP can decide what it wants to do with the new value, as if > >> it had requested the same attributes again. > > > > Thank you for the clarification. It seems that an OP will get the > > most consistent results if it always sends all attributes in an update > > then, so it doesn't need to track whether intermediate updates failed > > (or track exactly which attributes were changed). > > Sending all of the originally requested attributes would also require > the OP to keep an "original request" data structure for each Fetch > Request with an update_url in it, with the possibility of > conflicting / overlapping requests. > > A cleaner way would be to attach a list of update URLs to each > attribute in the user's profile, and when that attribute's value > changes to post an update to the RP (after prompting the user etc.). The issue I was thinking of was how to handle a "lost update". In cases where two attributes are related (like two components of a postal address), I'd want to make sure the RP has matching versions of those attributes. An update could be lost due to temporary network failures, or possibly if the RP could not verify the update (e.g. if an association handle is used that the RP does not have). > >> To indicate that the user has deleted an attribute, the count=0 > >> mechanism can be used: > >> > >> > An "openid.ax.count.<alias>" with a value of "0" together with its > >> > corresponding "openid.ax.type.<alias>" field MAY be included to > >> > explicitly state that no values are provided for an attribute. > > > > It might be worth demonstrating this in the example from section 5.2 > > then. Currently it reads: > > openid.ax.type.gender=http://example.com/schema/gender > > ... > > openid.ax.value.gender= > > > > If this is a case where the user has not given their gender it should > > perhaps use "openid.ax.count.gender=0" instead. > > You are right - thanks for catching this one as well! Previous drafts > required that empty values are sent if the user did not send a > meaningful value (which led to confusions and were clarified with the > count param). That looks better. James. _______________________________________________ specs mailing list specs@openid.net http://openid.net/mailman/listinfo/specs
