On Mon, Dec 5, 2011 at 10:15 AM, Sergey Beryozkin <sberyoz...@gmail.com> wrote: > On 05/12/11 13:23, Benson Margulies wrote: >> >> On Mon, Dec 5, 2011 at 7:15 AM, Sergey Beryozkin<sberyoz...@gmail.com> >> wrote: >>> >>> Hi Benson, all >>> >>> At the moment the in CORS filter returns 'null' during a preflight check, >>> whenever some check fails, which means that most likely an HTTP status >>> code >>> will be returned to do with failure at the selection algorithm stage, but >>> that status code may not necessarily to be the one expected by the CORS >>> client ? I'm wondering of we should return some more specific HTTP status >>> code instead of depending on the runtime to eventually fail this >>> preflight >>> request. >> >> >> Maybe I don't understand filters. >> >> The cors spec never, ever, calls for failing the overall HTTP request. >> It calls for adding extra headers if the request is good, and not >> adding them if it is bad, and otherwise leaving it alone. > > Are you referring to the actual request which follows a preflight request ? > > I'm looking at [1] and I'm not sure how does the client (browser ?) can > decide that a preflight request was not successful. > > The filter returns Response.ok().build() in the end of the preflight check, > which indeed will let the out CORS filter to finalize the preflight response > but in cases where the preflight check was not good then I believe a random > HTTP error status will be returned depending on where the selection > algorithm fails afterwards (may be it is a path mismatch or unexpected > verb/content-type/accept-type).
Yes, I see the problem here, but I don't quite know what to do. Preflight seems to be carefully defined to get along with any existing OPTIONS handling that is going on. So, if the programmer has an OPTIONS method that matches, the situation is supposed to be the same as the situation with simple requests and, say, GET handlers. Will the JAX-RS code ever dispatch OPTIONS to a function that doesn't have an @OPTIONS? If so, I think that the problem here is more serious. If not, I'm not sure we have a problem. I'm also not sure that the CORS spec exactly makes sense, and I'm going to send them some email. > > Cheers, Sergey > > > >> >> Now, we could design a unified JAX-RS security feature that >> incorporated CORS as part of its job. It could, if asked, fail >> requests if they failed to meet the requirements. >> > > [1] http://www.w3.org/TR/cors/ > > >>> >>> The other question which we've discussed with Benson is what to do in the >>> case like this: >>> >>> @Path("/somepath") >>> public class Resource { >>> @GET >>> @Produces("application/xml") >>> public Book getXML() {} >>> >>> @GET >>> @Produces("application/json") >>> public Book getXML() {} >>> } >>> >>> The info CORS provides is sufficient enough to select either of the the >>> above 2 methods thus the question is what to do at the preflight check. >>> In this case we thought we can expect a CrossResourceSharingAnnotation >>> being >>> added to the 'good' method, or even to the all of them, possibly uing a >>> class-level annotation: >>> >>> @Path("/somepath") >>> @CrossResourceSharingAnnotation(...) >>> public class Resource { >>> @GET >>> @Produces("application/xml") >>> public Book getXML() {} >>> >>> @GET >>> @Produces("application/json") >>> public Book getXML() {} >>> } >>> >>> or in case of POST: >>> >>> @Path("/somepath") >>> public class Resource { >>> @POST >>> @Consumes("application/xml") >>> @CrossResourceSharingAnnotation(...) >>> public void addXML(Book) {} >>> >>> @POST >>> @Consumes("multipart/form-data") >>> public void upload(MultipartBody) {} >>> } >>> >>> We can also think of some configuration tricks. >>> Ex, if the consumer does know that only an upload POST method is 'valid' >>> then we can configure a CORS filter with the acceptType value which will >>> be >>> passed on to the JAXRS runtime to confirm that such a method actually >>> exists >>> >>> For the record, as agreed with Benson, I updated the filter to delegate >>> to >>> the runtime to find a valid matching method during a preflight check >>> which >>> is more secure than depending on the custom annotation >>> >>> Cheers, Sergey >>> >>> -- >>> Sergey Beryozkin >>> >>> Talend Community Coders >>> http://coders.talend.com/ >>> >>> Blog: http://sberyozkin.blogspot.com > >
