Change the comment that describes the kern.ipc.maxsockbuf
suggested value, in order to clarify what the calculation is.

Yes, since kern.maxsockbuf can be set in sysctl.conf, I wanted
vfs.nfsd.srvmaxio to be set that way as well.
It can also be changed without rebooting the server, by stopping
and restarting the nfsd.

Just fyi Dave, the change that makes nfs_bufpackets global was
just MFC'd to stable/13. It was in a commit already applied
to "main".

Made inline comment on calculation.

This looks ok to me. It does assume that "mnt_stat" is initialized '0', but that is the case.

Add locking using NFSD_LOCK()/NFSD_UNLOCK() to
the sysctl function to ensure that the nfsd threads are
not running when vfs.nfsd.srvmaxio is updated.

I had used NFSLOCKMNT(nmp) instead of atomic..() because I wanted
to maintain nm_nextaconn at < nm_aconnect and couldn't do "%" via
an atomic.

As suggested by markj@, this patch redefines the fields in
struct nfsmount as additional connections, to somewhat simplify
the code.
nm_nconnect renamed to nm_aconnect
nm_nextnconn renamed to nm_nextaconn

If the last catch was good, this one is a great catch.
I'm a little embarrassed that I didn't think to make
manipulation of nm_nextnconn MP-safe.

Add NFSPROC_READDIRPLUS to the list of RPCs
with large messages as suggested by markj@.

This patch depends on the property that the NFS
server sends the RPC reply on the same TCP connection
as the one the request was received on. Although the
FreeBSD server always does this, it is only required
for NFSv4.1/4.2 (RFC 5661)

This comment was imbedded in a recent email message on linu-nfs@vger.kernel.org.
Although other posters felt more evidence was needed to determine this,
it at least suggests that separating the large RPC messages from the small ones may
improve performance under certain circumstances.

The original issue described was how a high read/write process on the
client could slow another process trying to do heavy metadata
operations (like walking the filesystem). Using a different mount to
the same multi-homed server seems to help a lot (probably because of
the independent slot table).

Fixed nfsstat to be 1 instead of 8.

Sorry, I don't understand your comment. mount_nfs.8.sav is
simply the old (what is in "main") version that is being patched.

Add printf()s indicating why setting vfs.nfsd.srvmaxio failed.

Update comments to clarify what xp_p2 points to
on both the server (client end for callbacks) and
client (server end for callbacks).

I just did a little experiment where I had the server file systems
mounted in a shallow tree (side by side under the exported root)
and where all these server file systems (other than the root) were
assigned the same va_fsid.

Everything in your comment is correct.