Comment # 10
on bug 13202
from Parav Pandit
(In reply to Chuck Lever from comment #7)
> Parav:
>
> I drew similar conclusions while trying to address a problem where the RPC
> dissector could not match Replies to Calls on RPC-over-RDMA. I had proposed
> change 19033 to the RPC dissector to ignore the source port when building
> conversations, but Michael suggested I abandon it in favor of your work. I
> will investigate change 19107 to see if that addresses this issue.
>
> However, for other reasons, I want to remove the existing conversation
> infrastructure from packet-rpcrdma.c. See change 19032.
>
> At some later point, some kind of conversation flow tracking will be needed
> in order to match RDMA Read/Write requests to RPC-over-RDMA transport
> headers. Perhaps you and I can work out a scheme to do this.
Yes. Its already in todo list once some of this basic things sorted out. So
current design is to to pass read/write context from infiniband.c to ulp
dissectors on read/write requests and responses. ULP will have ULP level IO
context in which they can match up the RDMA keys. RDMA key level context come
from infiniband.c so that any transport level errors that occurs transport can
flag.
You are receiving this mail because:
- You are watching all bug changes.