[lkml]   [2006]   [Oct]   [2]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
Patch in this message
SubjectRe: [NFS] [PATCH 008 of 11] knfsd: Prepare knfsd for support of rsize/wsize of up to 1MB, over TCP.
On Monday September 25, wrote:
> We're reporting svc_max_payload(rqstp) as the server's maximum
> read/write block size:

Yes. So I'm going to change the number returned by
svc_max_payload(rqstp) to mean the maximum read/write block size.
i.e. when a service is created, the number passed isn't the maximum
packet size, but is the maximum payload size.
The assumption is that all of the request that is not payload will fit
into one page, and all of the reply that is not payload will also fit
into one page (though a different page).

It means that RPC services that have lots of non-payload data combined
with payload data won't work, but making sunrpc code completely
general when there are only two users is just too painful.

The only real problem is that NFSv4 can have arbitrarily large
non-payload data, and arbitrarily many payloads. But I guess any
client that trying to send two full-sized payloads in the one request
is asking for trouble (I don't suppose the RPC spells this out at

> > +/*
> > + * Largest number of bytes we need to allocate for an NFS
> > + * call or reply. Used to control buffer sizes. We use
> > + * the length of v3 WRITE, READDIR and READDIR replies
> > + * which are an RPC header, up to 26 XDR units of reply
> > + * data, and some page data.
> > + *
> > + * Note that accuracy here doesn't matter too much as the
> > + * size is rounded up to a page size when allocating space.
> > + */
> Is the rounding up *always* going to increase the size? And if not,
> then why doesn't accuracy matter?
> I think this results in 80 less bytes less than before, I think.
> No doubt we have lots of wiggle room here, but I'd rather we didn't
> decrease that size without seeing a careful analysis.

The above change makes this loss in bytes irrelevant. NFSD_BUFSIZE
will now only be used once - near the end of nfs4proc.c and there if
it is wrong you just get a warning.

And the fact that the code change to effect this is so tiny seems to
imply that most of the code was already assuming that sv_bufsz was
really the payload size rather than the packet size.

So this is my proposed 'fix' for


Make sv_bufsiz really be the payload size for rpc requests.

svc.c already allocated 2 extra pages for the request and the reply,
so it is perfectly consistent to assume that the size passed to
svc_create_pooled is the size of the payload. This means that
the number returned by svc_max_payload - and thus returned to the client
as the maxiumu IO size - is exactly the chosen max block size.

Signed-off-by: Neil Brown <>

### Diffstat output
./fs/nfsd/nfssvc.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff .prev/fs/nfsd/nfssvc.c ./fs/nfsd/nfssvc.c
--- .prev/fs/nfsd/nfssvc.c 2006-09-29 11:57:27.000000000 +1000
+++ ./fs/nfsd/nfssvc.c 2006-10-03 11:23:11.000000000 +1000
@@ -217,7 +217,7 @@ int nfsd_create_serv(void)

atomic_set(&nfsd_busy, 0);
nfsd_serv = svc_create_pooled(&nfsd_program,
- NFSD_BUFSIZE - NFSSVC_MAXBLKSIZE + nfsd_max_blksize,
+ nfsd_max_blksize,
if (nfsd_serv == NULL)
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at
 \ /
  Last update: 2006-10-03 03:39    [W:0.097 / U:0.420 seconds]
©2003-2017 Jasper Spaans. hosted at Digital OceanAdvertise on this site