Fwd: Corrected diagram and description for SMB-Direct (RDMA) support in Samba

Ira Cooper ira at samba.org
Wed May 21 13:20:32 MDT 2014


On Wed, May 21, 2014 at 2:48 PM, Michael Adam <obnox at samba.org> wrote:

> On 2014-05-21 at 14:23 -0400, Ira Cooper wrote:
> > On Wed, May 21, 2014 at 2:11 PM, Stefan (metze) Metzmacher
> > <metze at samba.org>wrote:
> > > >> The session bind using the sessionID will still happen.
> > > >> The fd-passing would just have happened before.
> > > >> The fd-passing is something we plan to do under the hood
> > > >> and is not necessarily tied to the session-bind.
> > > >>
> > > >
> > > > What about VFS modules that don't have file descriptors?
> > >
> > > FD-passing will be only used for the tcp connection
> > > not for open file handles.
> > >
> > > So VFS modules will just work fine.
> > >
> >
> > So what actually creates the RDMA queue, and does the processing in this
> > model?
> >
> > Can someone show me the event trace for read/write?
>
> I am not sure that this is what you are asking:
> The chain of events for bringing up a smb-direct-daemon
> and establishing a proxy channel and mmap area between
> smbd-process for a client and this (one) smbd-direct-daemon
> is skeched in the earlier mails in this thread and the
> thread "SMB3 and RDMA".
>
> then the one smb-direct-daemon does all the RDMA connections
> and speaks smb-direct. The rdma read and write calls are then
> using addresses from the mmap-ed area from the corresponding
> smbd child process.
>
> I need to write all this up in greater detail in the wiki,
> but please read up on these two recent threads.
>

You explained it just fine.  Thank you.

My apologies.

-Ira


More information about the samba-technical mailing list