Re: Help wanted: ENOCLK returned during lock test#2 in connectathon's test

["J. Bruce Fields" <bfields@fieldses.org>]

On Tue, Dec 06, 2011 at 01:11:05PM +0100, DENIEL Philippe wrote:
> Hi Trond,
> 
> many thanks for your reply.
> In fact, the "rflag" in OP4_OPEN's reply is set to 6 = 4|2 =
> OPEN4_RESULT_LOCKTYPE_POSIX|OPEN4_RESULT_CONFIRM
> For some reason I do not understand, wireshark see
> OPEN4_RESULT_LOCKTYPE_POSIX as an 'unknown' flag and do not print
> it.  Bug actually it seems like OPEN4_RESULT_LOCKTYPE_POSIX is set.
> 
> Your mail made me have a closer look to my implementation of
> OP4_OPEN and OP4_OPEN_CONFIRM in NFSv4.0 . Since the beginning
> (since I met this bug), I suspect something related to seqids : it
> does not occur in NFSv4.1 where seqids 's management is made in
> OP4_SEQUENCE, at the beginning of the request. So I ran lock test#2
> on a kernel nfsd, capture the result and compared to what ganesha
> produces. I saw a difference:
> - when OP4_OPEN is invoked, the nfsd replies with a stateid
> containing seqid=0. This seqid is passed to OP4_OPEN_CONFIRM which
> confirms it and (if OK) replies with an updated stateid (seqid is
> now 1)
> - when ganesha does the same OP4_OPEN return a (unconfirmed) stateid
> whose seqid is equal to 1, then OP4_OPEN_CONFIRM set this seqid to 2
> when confirming the stateid.

Sounds like you're talking about the seqid field that's contained in the
stateid itself--I'd be suprised if the client cares about it.  The spec
does allow the client to inspect that field to decide what order opens
were done in, but other than that a client normally treats the whole
stateid as opaque.

--b.

> 
> From your point of view, could this mess in seqid's management
> produce the bug that I see when running lock test#2 ?
> 
>    Regards
> 
>       Philippe
> 
> Trond Myklebust a écrit :
> >On Mon, 2011-12-05 at 14:52 +0100, DENIEL Philippe wrote:
> >>Hi,
> >>
> >>as you may know (we may have met at Bake-A-Thon), I am working
> >>on NFS-Ganesha, a NFS server running in userspace. I currently
> >>face an issue when running cthon04 test suite, during the "lock
> >>step".
> >>Client is linux 3.1.0-rc4, server is nfs-ganesha compiled with
> >>FSAL_VFS support. Server is mounted via command "mount
> >>-overs=4.minorversion=1,lock <server>:<path> /mnt"
> >>
> >>During the test#2 in "lock" tests, I got the following error:
> >>
> >>    Creating parent/child synchronization pipes.
> >>
> >>    Test #2 - Try to lock the whole file.
> >>            Parent: 2.0  - F_TLOCK [               0,          ENDING]
> >>    FAILED!
> >>            Parent: **** Expected success, returned errno=37...
> >>            Parent: **** Probably implementation error.
> >>
> >>    ** PARENT pass 1 results: 0/0 pass, 0/0 warn, 1/1 fail (pass/total).
> >>
> >>    **  CHILD pass 1 results: 0/0 pass, 0/0 warn, 0/0 fail (pass/total).
> >>
> >>
> >>I made a wireshark capture of the packet (see attachement).
> >>Apparently, the client does 2 compounds, one for OP4_OPEN and a
> >>second to call OP4_OPEN_CONFIRM.
> >
> >Hi Philippe,
> >
> >As far as I can see from the pcap file, your server isn't setting the
> >OPEN4_RESULT_LOCKTYPE_POSIX flag in the OPEN reply, and so the client
> >can't support posix locking semantics. In that case, it will return
> >ENOLCK to all fcntl locking requests.
> >
> >Cheers
> >  Trond
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html