Part 2 in the series, please see the first part of the JIL BONDI comparison.
Widgets need to be signed, because in many cases they take advantage of new experimental Device APIs that must be authorised (or checked over, like a MOT), else they might harm your privacy and security. You should not use a potentially harmful service unless you know who accounts for it, so you have some sort of recourse if things go bad. If Widgets go bad, the idea is that the signature can be revoked and the harmful widget can be disabled.
Assuming JIL, has one operator root authority your widget can be in one of three security domains:
- Unidentified (~not signed)
- Identified (~author signed, though with JIL as the Certificate Authority most likely)
- Operator (~signed and approved explicitly by a JIL operator)
This is documented in section four "API Support Levels" of their JIL Widget System API Specification - Handset API v1 1 Revision 4.pdf from http://www.jil.org/#DOCUMENTS.
JIL API policy outcomes (security profiles) are actually the same as BONDI's:
- Unrestricted (allowed)
- One shot (prompt everytime the API is used)
- Session (prompt once in the session)
- Blanket (saved preference)
- Disallowed (deny in BONDI)
The difference between JIL and BONDI, is that JIL's "security profiles" are simply fixed to each handset API under each security domain.
For example, Geolocation or JIL's version of it DeviceStateInfo.requestPosistionInfo is "Allowed" if you are an "Operator" signed widget. "Session" if your widget is only author signed and "Oneshot" if not signed/identified. Now ask yourself if you trust your operator enough to make decisions whether applications know your location or not?
JIL's API support levels are too rigid and ultimately makes the operator have all the power. This makes the same mistakes as the Java security model does, as we all know developers have had problems getting their widgets signed by one particular operator.
BONDI has adopted Widgets 1.0: Digital Signatures and contributed to the specification rnc. The specification allows for one author signature and several distributor signatures. Therefore a BONDI widget can end up in many several arbitrary domains. You could for example trust a certain distributor or author and certain groupings of them.
BONDI's security model can implement the JIL model, however BONDI allows the flexibility to trust different signers in a decentralised trust model. This helps make Widgets work seamlessly across networks. With BONDI widgets you might begin to explicitly trust a certain author's signature and if it is not signed by the distributor, it doesn't matter. You'll be able to finally enjoy a seamless install and update process, without completely losing trust, verification and API functionality if a certain distributor is slow or reluctant to sign a new widget.
Perhaps you're a high security government institution. You can verify and start using a widget that has had it's entire chain verified potentially with many several signatures of concerned parties. BONDI's security model extends to allow this use case, whilst JIL's simpler one does not.
JIL does not have a standard policy format. So for all the stored "Blanket" decisions an end user has made about his/her security/privacy settings with certain APIs, these preferences cannot be moved between devices or widget runtimes like BONDI's compliant widget runtimes can.
BONDI allows the user to have full control of his/her privacy if the user wants to. You could delegate the decision to a trusted party (your operator) to decide whether or not an application should know your location or not (like JIL). However in some cases it's a privacy requirement that you alone want to decide this preference / security decision based on a whole host of factors. Once your decision has been made, BONDI's security model can record your particular preferences as a standardised XML policy fragment. And you can take these security settings around with you and customise them over time.
Unfortunately privacy and security considerations are complex and differ from person to person and API to API. BONDI's security model can accommodate JIL's simpler security model, though in the reality you will need BONDI's flexible fine grained approach to security to accommodate users changing individual needs.