From owner-freebsd-current  Wed Dec  6 16:35:13 1995
Return-Path: owner-current
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.3/8.7.3) id QAA12108
          for current-outgoing; Wed, 6 Dec 1995 16:35:13 -0800 (PST)
Received: from who.cdrom.com (who.cdrom.com [192.216.222.3])
          by freefall.freebsd.org (8.7.3/8.7.3) with SMTP id QAA12096
          for <current@freebsd.org>; Wed, 6 Dec 1995 16:35:07 -0800 (PST)
Received: from phaeton.artisoft.com (phaeton.Artisoft.COM [198.17.250.211])
          by who.cdrom.com (8.6.12/8.6.11) with ESMTP id QAA04831
          for <current@freebsd.org>; Wed, 6 Dec 1995 16:34:58 -0800
Received: (from terry@localhost) by phaeton.artisoft.com (8.6.11/8.6.9) id RAA02483 for current@freebsd.org; Wed, 6 Dec 1995 17:32:36 -0700
From: Terry Lambert <terry@lambert.org>
Message-Id: <199512070032.RAA02483@phaeton.artisoft.com>
Subject: Reconsidering leases
To: current@freebsd.org
Date: Wed, 6 Dec 1995 17:32:36 -0700 (MST)
X-Mailer: ELM [version 2.4 PL24]
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Sender: owner-current@freebsd.org
Precedence: bulk


Really, this is a posting about "low-hanging-fruit" optimization.

Leases are a facility for supporting client cache coherency in NFS
clients.


In vnode.h, the following is defined:


extern void	(*lease_check) __P((struct vnode *vp, struct proc *p,
				    struct ucred *ucred, int flag));
extern void	(*lease_updatetime) __P((int deltat));

#ifdef NFS
#ifdef NQNFS
#define	LEASE_CHECK(vp, p, cred, flag)	lease_check((vp), (p), (cred), (flag))
#define	LEASE_UPDATETIME(dt)		lease_updatetime(dt)
#else
#define LEASE_CHECK(vp, p, cred, flag)
#define LEASE_UPDATETIME(dt)
#endif /* NQNFS */
#else
#define LEASE_CHECK(vp, p, cred, flag) \
	do { if(lease_check) lease_check((vp), (p), (cred), (flag)); } while(0)
#define LEASE_UPDATETIME(dt) \
	do { if(lease_updatetime) lease_updatetime(dt); } while(0)
#endif /* NFS */
#endif /* KERNEL */


Basically in the non-NFS case, extra code is generated to support the
loading of NFS as an LKM to support lease checking.  Not only that,
the VOP_LOCK'ing is delayed because LEASE_CHECK wants to be called
with the vp unlocked.

In point of fact, there is now now way to disable lease checking without:

# Note: 4.4BSD NQNFS lease checking has relatively high cost for
# _local_ I/O as well as remote I/O. Don't use it unless you will
# using NQNFS.
options		NFS			#Network File System
#options	NQNFS			#Enable NQNFS lease checking

since if NQFS is defined or NFS *isn't* defined, the compare for
the lease-cheking goes in.

So really, what's being supported is loading of NFS with NQNFS
enabled.


I'd like to see a seperate definition of

options		FS_LEASES		#Enable lease checking

Since I don't believe the client caching abstraction is limited to
NQNFS use of the interface.

What that means is that NQNFS leasing would be dependent on whether
FS_LEASES was defined in the kernel where the module is loaded.  Maybe
something like

#ifdef FS_LEASES
int	fs_has_lease_checking = 0;
#else
int	fs_has_lease_checking = 1;
#endif


Then in the VOP_LOCK delays to allow the LEASE_CHECK on an unlocked vp
can be conditionalized on compilation.

Small optimization, I know, but every 3-5% helps.  8-).


There's also some short read fixes that can go in using a generation count
that should speed small sequential reads significantly if there are 2 or
more of the reads possible in a single logical FS block.  More on that
one later.


					Terry Lambert
					terry@lambert.org
---
Any opinions in this posting are my own and not those of my present
or previous employers.