On Sat, Jun 20, 2009 at 05:44:08PM -0400, Trond Myklebust wrote: > On Fri, 2009-06-19 at 18:23 -0400, J. Bruce Fields wrote: > > On Mon, May 18, 2009 at 05:47:56PM -0400, Trond Myklebust wrote: > > > Ensure that deferred requests are accounted for correctly by the write > > > space reservation mechanism. In order to avoid double counting, remove the > > > reservation when we defer the request, and save any calculated value, so > > > that we can restore it when the request is requeued. > > > > I like that it does the addition to xpt_reserved in just one place > > instead of two, and carrying over the reserved_space is nice, but I > > don't understand the "double accounting" comment--where exactly is > > something counted twice? > > Correct me if I'm wrong, but as far as I can see, the current code bumps > xpt_reserved every time you call svc_xprt_enqueue() irrespective of > whether or not you are going to process a new RPC call, or whether you > are processing a deferred call. > > In the latter case, there doesn't appear to be any code that subtracts > from xpt_reserved prior to the re-enqueue process, hence my belief that > we are double counting those events... Sorry for the slow response! The original request is dropped--cache_check() returns -EAGAIN on queueing up the deferred request, the caller returns back to svc_process, and the reservation is removed from xpt_reserved by svc_drop(). And a new reservation isn't made until the deferred request is revisited, at which point it's treated as any new request received from a socket would be. Your patch would instead have the server keep that reservation while the request is deferred. I'm not really sure how to decide which is better. --b. -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
