Remove the v2 writegather code. It did help alot back in the 80s

but extensive performance benchmarking done by myself and jasper@
has shown that it doesn't help, at all - even on vaxens and in some
cases it makes things significantly slower.

"this excites me sexually" jetpack@
Tested by jasper@.
OK blambert@

6 files changed, 136 insertions, 658 deletions

diff --git a/sys/nfs/nfs.h b/sys/nfs/nfs.h
index 0e1b483fd81..187112bf705 100644
--- a/sys/nfs/nfs.h
+++ b/sys/nfs/nfs.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: nfs.h,v 1.45 2009/07/14 16:40:29 thib Exp $	*/
+/*	$OpenBSD: nfs.h,v 1.46 2009/08/10 09:38:44 thib Exp $	*/
 /*	$NetBSD: nfs.h,v 1.10.4.1 1996/05/27 11:23:56 fvdl Exp $	*/
 
 /*
@@ -56,8 +56,6 @@
 #define	NFS_DEFRAHEAD	1		/* Def. read ahead # blocks */
 #define	NFS_MAXRAHEAD	4		/* Max. read ahead # blocks */
 #define	NFS_MAXASYNCDAEMON 	20	/* Max. number async_daemons runable */
-#define NFS_MAXGATHERDELAY	100	/* Max. write gather delay (msec) */
-#define NFS_GATHERDELAY		10	/* Default write gather delay (msec) */
 
 /*
  * Ideally, NFS_DIRBLKSIZ should be bigger, but I've seen servers with
@@ -265,15 +263,6 @@ enum nfs_rto_timers {
 
 #define NFS_INITRTT	(NFS_HZ << 3)
 
-/*
- * A list of nfssvc_sock structures is maintained with all the sockets
- * that require service by the nfsd.
- */
-#ifndef NFS_WDELAYHASHSIZ
-#define	NFS_WDELAYHASHSIZ 16	/* and with this */
-#endif
-#define	NWDELAYHASH(sock, f) \
-	(&(sock)->ns_wdelayhashtbl[(*((u_int32_t *)(f))) % NFS_WDELAYHASHSIZ])
 #define	NFSNOHASH(fhsum) \
 	(&nfsnodehashtbl[(fhsum) & nfsnodehash])
 
@@ -300,9 +289,6 @@ struct nfssvc_sock {
 	int		ns_cc;		/* actual chars queued */
 	int		ns_reclen;	/* length of first queued record */
 	u_int32_t	ns_sref;	/* # of refs to this struct */
-	LIST_HEAD(, nfsrv_descript) ns_tq; /* Write gather lists */
-	LIST_HEAD(nfsrvw_delayhash, nfsrv_descript)
-			ns_wdelayhashtbl[NFS_WDELAYHASHSIZ];
 };
 
 /* Bits for "ns_flag" */
@@ -338,53 +324,27 @@ struct nfsd {
 
 /*
  * This structure is used by the server for describing each request.
- * Some fields are used only when write request gathering is performed.
  */
 struct nfsrv_descript {
-	struct timeval		nd_time;	/* Write deadline */
-	off_t			nd_off;		/* Start byte offset */
-	off_t			nd_eoff;	/* and end byte offset */
-	LIST_ENTRY(nfsrv_descript) nd_hash;	/* Hash list */
-	LIST_ENTRY(nfsrv_descript) nd_tq;	/* and timer list */
-	LIST_HEAD(,nfsrv_descript) nd_coalesce;	/* coalesced writes */
 	struct mbuf		*nd_mrep;	/* Request mbuf list */
 	struct mbuf		*nd_md;		/* Current dissect mbuf */
-	struct mbuf		*nd_mreq;	/* Reply mbuf list */
 	struct mbuf		*nd_nam;	/* and socket addr */
 	struct mbuf		*nd_nam2;	/* return socket addr */
 	caddr_t			nd_dpos;	/* Current dissect pos */
 	unsigned int		nd_procnum;	/* RPC # */
-	int			nd_stable;	/* storage type */
 	int			nd_flag;	/* nd_flag */
-	int			nd_len;		/* Length of this write */
 	int			nd_repstat;	/* Reply status */
 	u_int32_t		nd_retxid;	/* Reply xid */
-	fhandle_t		nd_fh;		/* File handle */
 	struct ucred		nd_cr;		/* Credentials */
 };
 
 /* Bits for "nd_flag" */
 #define ND_NFSV3	0x08
-#define ND_KERBNICK	0x20
-#define ND_KERBFULL	0x40
-#define ND_KERBAUTH	(ND_KERBNICK | ND_KERBFULL)
 
 extern struct pool nfsreqpl;
 extern TAILQ_HEAD(nfsdhead, nfsd) nfsd_head;
 extern int nfsd_head_flag;
 #define	NFSD_CHECKSLP	0x01
 
-/*
- * These macros compare nfsrv_descript structures.
- */
-#define NFSW_CONTIG(o, n) \
-		((o)->nd_eoff >= (n)->nd_off && \
-		 !bcmp((caddr_t)&(o)->nd_fh, (caddr_t)&(n)->nd_fh, NFSX_V3FH))
-
-#define NFSW_SAMECRED(o, n) \
-	(((o)->nd_flag & ND_KERBAUTH) == ((n)->nd_flag & ND_KERBAUTH) && \
- 	 !bcmp((caddr_t)&(o)->nd_cr, (caddr_t)&(n)->nd_cr, \
-		sizeof (struct ucred)))
-
 #endif	/* _KERNEL */
 #endif /* _NFS_NFS_H */
diff --git a/sys/nfs/nfs_serv.c b/sys/nfs/nfs_serv.c
index cffb7abadb4..e689148102d 100644
--- a/sys/nfs/nfs_serv.c
+++ b/sys/nfs/nfs_serv.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: nfs_serv.c,v 1.81 2009/08/09 17:35:31 blambert Exp $	*/
+/*	$OpenBSD: nfs_serv.c,v 1.82 2009/08/10 09:38:44 thib Exp $	*/
 /*     $NetBSD: nfs_serv.c,v 1.34 1997/05/12 23:37:12 fvdl Exp $       */
 
 /*
@@ -88,11 +88,6 @@ extern enum vtype nv3tov_type[8];
 extern struct nfsstats nfsstats;
 extern nfstype nfsv2_type[9];
 extern nfstype nfsv3_type[9];
-int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000;
-struct timeval nfsrvw_procrastinate_tv = {
-	(NFS_GATHERDELAY * 1000) / 1000000,	/* tv_sec */
-	(NFS_GATHERDELAY * 1000) % 1000000	/* tv_usec */
-};
 
 /*
  * nfs v3 access service
@@ -867,372 +862,6 @@ bad:
 }
 
 /*
- * NFS write service with write gathering support. Called when
- * nfsrvw_procrastinate > 0.
- * See: Chet Juszczak, "Improving the Write Performance of an NFS Server",
- * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco,
- * Jan. 1994.
- */
-int
-nfsrv_writegather(ndp, slp, procp, mrq)
-	struct nfsrv_descript **ndp;
-	struct nfssvc_sock *slp;
-	struct proc *procp;
-	struct mbuf **mrq;
-{
-	struct iovec *ivp;
-	struct mbuf *mp;
-	struct nfsrv_descript *wp, *nfsd, *nnfsd, *owp, *swp;
-	struct nfs_fattr *fp;
-	struct nfsm_info	info;
-	int i = 0;
-	struct iovec *iov;
-	struct nfsrvw_delayhash *wpp;
-	struct ucred *cred;
-	struct vattr va, forat;
-	u_int32_t *tl;
-	int32_t t1;
-	int error = 0, rdonly, len = 0, forat_ret = 1;
-	int ioflags, aftat_ret = 1, s, adjust, zeroing;
-	char *cp2;
-	struct vnode *vp;
-	struct uio io, *uiop = &io;
-	struct timeval tv;
-
-	*mrq = NULL;
-	if (*ndp) {
-	    nfsd = *ndp;
-	    *ndp = NULL;
-	    info.nmi_mreq = NULL;
-	    info.nmi_mrep = nfsd->nd_mrep;
-	    info.nmi_md = nfsd->nd_md;
-	    info.nmi_dpos = nfsd->nd_dpos;
-	    cred = &nfsd->nd_cr;
-	    info.nmi_v3 = (nfsd->nd_flag & ND_NFSV3);
-	    LIST_INIT(&nfsd->nd_coalesce);
-	    nfsd->nd_mreq = NULL;
-	    nfsd->nd_stable = NFSV3WRITE_FILESYNC;
-	    getmicrotime(&tv);
-	    timeradd(&tv, &nfsrvw_procrastinate_tv, &nfsd->nd_time);
-    
-	    /*
-	     * Now, get the write header..
-	     */
-	    nfsm_srvmtofh(&nfsd->nd_fh);
-	    if (info.nmi_v3) {
-		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
-		nfsd->nd_off = fxdr_hyper(tl);
-		tl += 3;
-		nfsd->nd_stable = fxdr_unsigned(int, *tl++);
-	    } else {
-		nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
-		nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
-		tl += 2;
-	    }
-	    len = fxdr_unsigned(int32_t, *tl);
-	    nfsd->nd_len = len;
-	    nfsd->nd_eoff = nfsd->nd_off + len;
-    
-	    /*
-	     * Trim the header out of the mbuf list and trim off any trailing
-	     * junk so that the mbuf list has only the write data.
-	     */
-	    zeroing = 1;
-	    i = 0;
-	    mp = info.nmi_mrep;
-	    while (mp) {
-		if (mp == info.nmi_md) {
-		    zeroing = 0;
-		    adjust = info.nmi_dpos - mtod(mp, caddr_t);
-		    mp->m_len -= adjust;
-		    if (mp->m_len > 0 && adjust > 0)
-		    	mp->m_data += adjust;
-		}
-		if (zeroing)
-		    mp->m_len = 0;
-		else {
-		    i += mp->m_len;
-		    if (i > len) {
-			mp->m_len -= (i - len);
-			zeroing = 1;
-		    }
-		}
-		mp = mp->m_next;
-	    }
-	    if (len > NFS_MAXDATA || len < 0  || i < len) {
-		m_freem(info.nmi_mrep);
-nfsmout:
-		error = EIO;
-		nfsm_writereply(2 * NFSX_UNSIGNED, info.nmi_v3);
-		if (info.nmi_v3)
-		    nfsm_srvwcc(nfsd, forat_ret, &forat, aftat_ret, &va,
-		        &info.nmi_mb);
-		nfsd->nd_mreq = info.nmi_mreq;
-		nfsd->nd_mrep = NULL;
-		timerclear(&nfsd->nd_time);
-	    }
-    
-	    /*
-	     * Add this entry to the hash and time queues.
-	     */
-	    s = splsoftclock();
-	    owp = NULL;
-	    wp = LIST_FIRST(&slp->ns_tq);
-	    while (wp && timercmp(&wp->nd_time, &nfsd->nd_time, <)) {
-		owp = wp;
-		wp = LIST_NEXT(wp, nd_tq);
-	    }
-	    if (owp) {
-		LIST_INSERT_AFTER(owp, nfsd, nd_tq);
-	    } else {
-		LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
-	    }
-	    if (nfsd->nd_mrep) {
-		wpp = NWDELAYHASH(slp, nfsd->nd_fh.fh_fid.fid_data);
-		owp = NULL;
-		wp = LIST_FIRST(wpp);
-		while (wp &&
-		    bcmp((caddr_t)&nfsd->nd_fh,(caddr_t)&wp->nd_fh,NFSX_V3FH)) {
-		    owp = wp;
-		    wp = LIST_NEXT(wp, nd_hash);
-		}
-		while (wp && wp->nd_off < nfsd->nd_off &&
-		   !bcmp((caddr_t)&nfsd->nd_fh,(caddr_t)&wp->nd_fh,NFSX_V3FH)) {
-		    owp = wp;
-		    wp = LIST_NEXT(wp, nd_hash);
-		}
-		if (owp) {
-		    LIST_INSERT_AFTER(owp, nfsd, nd_hash);
-
-		    /*
-		     * Search the hash list for overlapping entries and
-		     * coalesce.
-		     */
-		    for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) {
-			wp = LIST_NEXT(nfsd, nd_hash);
-			if (NFSW_SAMECRED(owp, nfsd))
-			    nfsrvw_coalesce(owp, nfsd);
-		    }
-		} else {
-		    LIST_INSERT_HEAD(wpp, nfsd, nd_hash);
-		}
-	    }
-	    splx(s);
-	}
-    
-	/*
-	 * Now, do VOP_WRITE()s for any one(s) that need to be done now
-	 * and generate the associated reply mbuf list(s).
-	 */
-loop1:
-	getmicrotime(&tv);
-	s = splsoftclock();
-	for (nfsd = LIST_FIRST(&slp->ns_tq); nfsd != NULL; nfsd = owp) {
-		owp = LIST_NEXT(nfsd, nd_tq);
-		if (timercmp(&nfsd->nd_time, &tv, >))
-		    break;
-		if (nfsd->nd_mreq)
-		    continue;
-		LIST_REMOVE(nfsd, nd_tq);
-		LIST_REMOVE(nfsd, nd_hash);
-		splx(s);
-		info.nmi_mrep = nfsd->nd_mrep;
-		nfsd->nd_mrep = NULL;
-		cred = &nfsd->nd_cr;
-		info.nmi_v3 = (nfsd->nd_flag & ND_NFSV3);
-		forat_ret = aftat_ret = 1;
-		error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp,
-		    nfsd->nd_nam, &rdonly);
-		if (!error) {
-		    if (info.nmi_v3)
-			forat_ret = VOP_GETATTR(vp, &forat, cred, procp);
-		    if (vp->v_type != VREG) {
-			if (info.nmi_v3)
-			    error = EINVAL;
-			else
-			    error = (vp->v_type == VDIR) ? EISDIR : EACCES;
-		    }
-		} else
-		    vp = NULL;
-		if (!error) {
-		    error = nfsrv_access(vp, VWRITE, cred, rdonly, procp, 1);
-		}
-    
-		if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE)
-		    ioflags = IO_NODELOCKED;
-		else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC)
-		    ioflags = (IO_SYNC | IO_NODELOCKED);
-		else
-		    ioflags = (IO_SYNC | IO_NODELOCKED);
-		uiop->uio_rw = UIO_WRITE;
-		uiop->uio_segflg = UIO_SYSSPACE;
-		uiop->uio_procp = NULL;
-		uiop->uio_offset = nfsd->nd_off;
-		uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off;
-		if (uiop->uio_resid > 0) {
-		    mp = info.nmi_mrep;
-		    i = 0;
-		    while (mp) {
-			if (mp->m_len > 0)
-			    i++;
-			mp = mp->m_next;
-		    }
-		    uiop->uio_iovcnt = i;
-		    iov = malloc(i * sizeof(struct iovec), M_TEMP, M_WAITOK);
-		    uiop->uio_iov = ivp = iov;
-		    mp = info.nmi_mrep;
-		    while (mp) {
-			if (mp->m_len > 0) {
-			    ivp->iov_base = mtod(mp, caddr_t);
-			    ivp->iov_len = mp->m_len;
-			    ivp++;
-			}
-			mp = mp->m_next;
-		    }
-		    if (!error) {
-			error = VOP_WRITE(vp, uiop, ioflags, cred);
-			nfsstats.srvvop_writes++;
-		    }
-		    free(iov, M_TEMP);
-		}
-		m_freem(info.nmi_mrep);
-		if (vp) {
-		    aftat_ret = VOP_GETATTR(vp, &va, cred, procp);
-		    vput(vp);
-		}
-
-		/*
-		 * Loop around generating replies for all write rpcs that have
-		 * now been completed.
-		 */
-		swp = nfsd;
-		do {
-		    if (error) {
-			nfsm_writereply(NFSX_WCCDATA(info.nmi_v3), info.nmi_v3);
-			if (info.nmi_v3) {
-			    nfsm_srvwcc(nfsd, forat_ret, &forat, aftat_ret,
-			        &va, &info.nmi_mb);
-			}
-		    } else {
-			nfsm_writereply(NFSX_PREOPATTR(info.nmi_v3) +
-			    NFSX_POSTOPORFATTR(info.nmi_v3) + 2 * NFSX_UNSIGNED +
-			    NFSX_WRITEVERF(info.nmi_v3), info.nmi_v3);
-			if (info.nmi_v3) {
-			    nfsm_srvwcc(nfsd, forat_ret, &forat, aftat_ret,
-			        &va, &info.nmi_mb);
-			    tl = nfsm_build(&info.nmi_mb, 4 * NFSX_UNSIGNED);
-			    *tl++ = txdr_unsigned(nfsd->nd_len);
-			    *tl++ = txdr_unsigned(swp->nd_stable);
-			    /*
-			     * Actually, there is no need to txdr these fields,
-			     * but it may make the values more human readable,
-			     * for debugging purposes.
-			     */
-			    *tl++ = txdr_unsigned(boottime.tv_sec);
-			    *tl = txdr_unsigned(boottime.tv_usec);
-			} else {
-			    fp = nfsm_build(&info.nmi_mb, NFSX_V2FATTR);
-			    nfsm_srvfattr(nfsd, &va, fp);
-			}
-		    }
-		    nfsd->nd_mreq = info.nmi_mreq;
-		    if (nfsd->nd_mrep)
-			panic("nfsrv_write: nd_mrep not free");
-
-		    /*
-		     * Done. Put it at the head of the timer queue so that
-		     * the final phase can return the reply.
-		     */
-		    s = splsoftclock();
-		    if (nfsd != swp) {
-			timerclear(&nfsd->nd_time);
-			LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
-		    }
-		    nfsd = LIST_FIRST(&swp->nd_coalesce);
-		    if (nfsd) {
-			LIST_REMOVE(nfsd, nd_tq);
-		    }
-		    splx(s);
-		} while (nfsd);
-		s = splsoftclock();
-		timerclear(&swp->nd_time);
-		LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq);
-		splx(s);
-		goto loop1;
-	}
-	splx(s);
-
-	/*
-	 * Search for a reply to return.
-	 */
-	s = splsoftclock();
-	for (nfsd = LIST_FIRST(&slp->ns_tq); nfsd != NULL; nfsd = nnfsd) {
-		nnfsd = LIST_NEXT(nfsd, nd_tq);
-		if (nfsd->nd_mreq) {
-		    LIST_REMOVE(nfsd, nd_tq);
-		    *mrq = nfsd->nd_mreq;
-		    *ndp = nfsd;
-		    break;
-		}
-	}
-	splx(s);
-	return (0);
-}
-
-/*
- * Coalesce the write request nfsd into owp. To do this we must:
- * - remove nfsd from the queues
- * - merge nfsd->nd_mrep into owp->nd_mrep
- * - update the nd_eoff and nd_stable for owp
- * - put nfsd on owp's nd_coalesce list
- * NB: Must be called at splsoftclock().
- */
-void
-nfsrvw_coalesce(struct nfsrv_descript *owp, struct nfsrv_descript *nfsd)
-{
-        int overlap;
-        struct mbuf *mp;
-
-	splsoftassert(IPL_SOFTCLOCK);
-
-        LIST_REMOVE(nfsd, nd_hash);
-        LIST_REMOVE(nfsd, nd_tq);
-        if (owp->nd_eoff < nfsd->nd_eoff) {
-            overlap = owp->nd_eoff - nfsd->nd_off;
-            if (overlap < 0)
-                panic("nfsrv_coalesce: bad off");
-            if (overlap > 0)
-                m_adj(nfsd->nd_mrep, overlap);
-            mp = owp->nd_mrep;
-            while (mp->m_next)
-                mp = mp->m_next;
-            mp->m_next = nfsd->nd_mrep;
-            owp->nd_eoff = nfsd->nd_eoff;
-        } else
-            m_freem(nfsd->nd_mrep);
-        nfsd->nd_mrep = NULL;
-        if (nfsd->nd_stable == NFSV3WRITE_FILESYNC)
-            owp->nd_stable = NFSV3WRITE_FILESYNC;
-        else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC &&
-            owp->nd_stable == NFSV3WRITE_UNSTABLE)
-            owp->nd_stable = NFSV3WRITE_DATASYNC;
-        LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq);
-
-	/*
-	 * nfsd might hold coalesce elements! Move them to owp.
-	 * Otherwise, requests may be lost and clients will be stuck.
-	 */
-	if (LIST_FIRST(&nfsd->nd_coalesce) != NULL) {
-		struct nfsrv_descript *m;
-
-		while ((m = LIST_FIRST(&nfsd->nd_coalesce)) != NULL) {
-			LIST_REMOVE(m, nd_tq);
-			LIST_INSERT_HEAD(&owp->nd_coalesce, m, nd_tq);
-		}
-	}
-}
-
-/*
  * nfs create service
  * now does a truncate to 0 length via. setattr if it already exists
  */
diff --git a/sys/nfs/nfs_socket.c b/sys/nfs/nfs_socket.c
index 85001370b0e..63f4c2c604d 100644
--- a/sys/nfs/nfs_socket.c
+++ b/sys/nfs/nfs_socket.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: nfs_socket.c,v 1.93 2009/08/04 17:12:39 thib Exp $	*/
+/*	$OpenBSD: nfs_socket.c,v 1.94 2009/08/10 09:38:44 thib Exp $	*/
 /*	$NetBSD: nfs_socket.c,v 1.27 1996/04/15 20:20:00 thorpej Exp $	*/
 
 /*
@@ -1151,10 +1151,6 @@ nfs_timer(arg)
 	struct nfsmount *nmp;
 	int timeo;
 	int s, error;
-#ifdef NFSSERVER
-	struct nfssvc_sock *slp;
-	struct timeval tv;
-#endif
 
 	s = splsoftnet();
 	TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
@@ -1240,19 +1236,6 @@ nfs_timer(arg)
 			}
 		}
 	}
-
-#ifdef NFSSERVER
-	/*
-	 * Scan the write gathering queues for writes that need to be
-	 * completed now.
-	 */
-	getmicrotime(&tv);
-	TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
-		if (LIST_FIRST(&slp->ns_tq) &&
-		    timercmp(&LIST_FIRST(&slp->ns_tq)->nd_time, &tv, <=))
-			nfsrv_wakenfsd(slp);
-	}
-#endif /* NFSSERVER */
 	splx(s);
 	timeout_add(to, nfs_ticks);
 }
@@ -1565,8 +1548,6 @@ nfs_getreq(nd, nfsd, has_header)
 		    else
 			tl++;
 		nd->nd_cr.cr_ngroups = (len > NGROUPS) ? NGROUPS : len;
-		if (nd->nd_cr.cr_ngroups > 1)
-		    nfsrvw_sort(nd->nd_cr.cr_groups, nd->nd_cr.cr_ngroups);
 		len = fxdr_unsigned(int, *++tl);
 		if (len < 0 || len > RPCAUTH_MAXSIZ) {
 			m_freem(info.nmi_mrep);
diff --git a/sys/nfs/nfs_subs.c b/sys/nfs/nfs_subs.c
index f7384ad0334..4446cf1e3c6 100644
--- a/sys/nfs/nfs_subs.c
+++ b/sys/nfs/nfs_subs.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: nfs_subs.c,v 1.99 2009/08/04 17:12:39 thib Exp $	*/
+/*	$OpenBSD: nfs_subs.c,v 1.100 2009/08/10 09:38:44 thib Exp $	*/
 /*	$NetBSD: nfs_subs.c,v 1.27.4.3 1996/07/08 20:34:24 jtc Exp $	*/
 
 /*
@@ -1790,48 +1790,6 @@ nfsrv_errmap(nd, err)
 }
 
 /*
- * Sort the group list in increasing numerical order.
- * (Insertion sort by Chris Torek, who was grossed out by the bubble sort
- *  that used to be here.)
- */
-void
-nfsrvw_sort(list, num)
-        gid_t *list;
-        int num;
-{
-	int i, j;
-	gid_t v;
-
-	/* Insertion sort. */
-	for (i = 1; i < num; i++) {
-		v = list[i];
-		/* find correct slot for value v, moving others up */
-		for (j = i; --j >= 0 && v < list[j];)
-			list[j + 1] = list[j];
-		list[j + 1] = v;
-	}
-}
-
-/*
- * copy credentials making sure that the result can be compared with bcmp().
- */
-void
-nfsrv_setcred(incred, outcred)
-	struct ucred *incred, *outcred;
-{
-	int i;
-
-	bzero((caddr_t)outcred, sizeof (struct ucred));
-	outcred->cr_ref = 1;
-	outcred->cr_uid = incred->cr_uid;
-	outcred->cr_gid = incred->cr_gid;
-	outcred->cr_ngroups = incred->cr_ngroups;
-	for (i = 0; i < incred->cr_ngroups; i++)
-		outcred->cr_groups[i] = incred->cr_groups[i];
-	nfsrvw_sort(outcred->cr_groups, outcred->cr_ngroups);
-}
-
-/*
  * If full is non zero, set all fields, otherwise just set mode and time fields
  */
 void
diff --git a/sys/nfs/nfs_syscalls.c b/sys/nfs/nfs_syscalls.c
index 966c0ce2af1..dc0835a1876 100644
--- a/sys/nfs/nfs_syscalls.c
+++ b/sys/nfs/nfs_syscalls.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: nfs_syscalls.c,v 1.84 2009/07/20 16:49:40 thib Exp $	*/
+/*	$OpenBSD: nfs_syscalls.c,v 1.85 2009/08/10 09:38:44 thib Exp $	*/
 /*	$NetBSD: nfs_syscalls.c,v 1.19 1996/02/18 11:53:52 fvdl Exp $	*/
 
 /*
@@ -75,8 +75,6 @@
 /* Global defs. */
 extern int nfs_numasync;
 extern struct nfsstats nfsstats;
-extern int nfsrvw_procrastinate;
-extern struct timeval nfsrvw_procrastinate_tv;
 struct nfssvc_sock *nfs_udpsock;
 int nfsd_waiting = 0;
 
@@ -305,190 +303,157 @@ nfssvc_nfsd(struct nfsd *nfsd)
 	int *solockp;
 	struct nfsrv_descript *nd = NULL;
 	struct mbuf *mreq;
-	int error = 0, cacherep, s, sotype, writes_todo;
-	struct timeval tv;
+	int error = 0, cacherep, s, sotype;
 
 	cacherep = RC_DOIT;
-	writes_todo = 0;
 
 	s = splsoftnet();
 	TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain);
 	nfs_numnfsd++;
-	/*
-	 * Loop getting rpc requests until SIGKILL.
-	 */
-	for (;;) {
-		if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) {
-
-			/* attach an nfssvc_sock to nfsd */
-			error = nfsrv_getslp(nfsd);
-			if (error)
-				goto done;
-			slp = nfsd->nfsd_slp;
-
-			if (slp->ns_flag & SLP_VALID) {
-				if (slp->ns_flag & SLP_DISCONN)
-					nfsrv_zapsock(slp);
-				else if (slp->ns_flag & SLP_NEEDQ) {
-					slp->ns_flag &= ~SLP_NEEDQ;
-					(void) nfs_sndlock(&slp->ns_solock,
-						NULL);
-					nfsrv_rcv(slp->ns_so, (caddr_t)slp,
-						M_WAIT);
-					nfs_sndunlock(&slp->ns_solock);
-				}
-				error = nfsrv_dorec(slp, nfsd, &nd);
-				getmicrotime(&tv);
-				if (error && LIST_FIRST(&slp->ns_tq) &&
-				    timercmp(&LIST_FIRST(&slp->ns_tq)->nd_time,
-				    &tv, <=)) {
-					error = 0;
-					cacherep = RC_DOIT;
-					writes_todo = 1;
-				} else
-					writes_todo = 0;
-				nfsd->nfsd_flag |= NFSD_REQINPROG;
+
+	/* Loop getting rpc requests until SIGKILL. */
+loop:
+	if (!ISSET(nfsd->nfsd_flag, NFSD_REQINPROG)) {
+
+		/* attach an nfssvc_sock to nfsd */
+		error = nfsrv_getslp(nfsd);
+		if (error)
+			goto done;
+
+		slp = nfsd->nfsd_slp;
+
+		if (ISSET(slp->ns_flag, SLP_VALID)) {
+			if (ISSET(slp->ns_flag, SLP_DISCONN)) {
+				nfsrv_zapsock(slp);
+			} else if (ISSET(slp->ns_flag, SLP_NEEDQ)) {
+				CLR(slp->ns_flag, SLP_NEEDQ);
+				nfs_sndlock(&slp->ns_solock, NULL);
+				nfsrv_rcv(slp->ns_so, (caddr_t)slp, M_WAIT);
+				nfs_sndunlock(&slp->ns_solock);
 			}
-		} else {
-			error = 0;
-			slp = nfsd->nfsd_slp;
+
+			error = nfsrv_dorec(slp, nfsd, &nd);
+			SET(nfsd->nfsd_flag, NFSD_REQINPROG);
 		}
-		if (error || (slp->ns_flag & SLP_VALID) == 0) {
-			if (nd) {
-				pool_put(&nfsrv_descript_pl, nd);
-				nd = NULL;
-			}
-			nfsd->nfsd_slp = NULL;
-			nfsd->nfsd_flag &= ~NFSD_REQINPROG;
-			nfsrv_slpderef(slp);
-			continue;
+	} else {
+		error = 0;
+		slp = nfsd->nfsd_slp;
+	}
+
+	if (error || !ISSET(slp->ns_flag, SLP_VALID)) {
+		if (nd != NULL) {
+			pool_put(&nfsrv_descript_pl, nd);
+			nd = NULL;
 		}
-		splx(s);
-		so = slp->ns_so;
-		sotype = so->so_type;
-		if (so->so_proto->pr_flags & PR_CONNREQUIRED)
-			solockp = &slp->ns_solock;
-		else
-			solockp = NULL;
-		if (nd) {
-		    if (nd->nd_nam2)
+		nfsd->nfsd_slp = NULL;
+		CLR(nfsd->nfsd_flag, NFSD_REQINPROG);
+		nfsrv_slpderef(slp);
+		goto loop;
+	}
+
+	splx(s);
+
+	so = slp->ns_so;
+	sotype = so->so_type;
+	if (ISSET(so->so_proto->pr_flags, PR_CONNREQUIRED))
+		solockp = &slp->ns_solock;
+	else
+		solockp = NULL;
+
+	if (nd) {
+		if (nd->nd_nam2)
 			nd->nd_nam = nd->nd_nam2;
-		    else
+		else
 			nd->nd_nam = slp->ns_nam;
+	}
 
-		    cacherep = nfsrv_getcache(nd, slp, &mreq);
-		}
-
-		/*
-		 * Loop to get all the write rpc relies that have been
-		 * gathered together.
-		 */
-		do {
-		    switch (cacherep) {
-		    case RC_DOIT:
-			if (writes_todo || (!(nd->nd_flag & ND_NFSV3) &&
-			    nd->nd_procnum == NFSPROC_WRITE &&
-			    nfsrvw_procrastinate > 0))
-				error = nfsrv_writegather(&nd, slp,
-				    nfsd->nfsd_procp, &mreq);
-			else
-				error = (*(nfsrv3_procs[nd->nd_procnum]))(nd,
-				    slp, nfsd->nfsd_procp, &mreq);
-			if (mreq == NULL) {
-				if (nd != NULL) {
-					m_freem(nd->nd_nam2);
-					m_freem(nd->nd_mrep);
-				}
-				break;
-			}
-			if (error) {
-				nfsstats.srv_errs++;
-				nfsrv_updatecache(nd, 0, mreq);
-				if (nd->nd_nam2)
-					m_freem(nd->nd_nam2);
-				break;
-			}
-			nfsstats.srvrpccnt[nd->nd_procnum]++;
-			nfsrv_updatecache(nd, 1, mreq);
-			nd->nd_mrep = NULL;
-
-			/* FALLTHROUGH */
-		    case RC_REPLY:
-			m = mreq;
-			siz = 0;
-			while (m) {
-				siz += m->m_len;
-				m = m->m_next;
-			}
-			if (siz <= 0 || siz > NFS_MAXPACKET) {
-				printf("mbuf siz=%d\n",siz);
-				panic("Bad nfs svc reply");
-			}
-			m = mreq;
-			m->m_pkthdr.len = siz;
-			m->m_pkthdr.rcvif = NULL;
-			/*
-			 * For stream protocols, prepend a Sun RPC
-			 * Record Mark.
-			 */
-			if (sotype == SOCK_STREAM) {
-				M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
-				*mtod(m, u_int32_t *) = htonl(0x80000000 | siz);
-			}
-			if (solockp)
-				(void) nfs_sndlock(solockp, NULL);
-			if (slp->ns_flag & SLP_VALID)
-			    error = nfs_send(so, nd->nd_nam2, m, NULL);
-			else {
-			    error = EPIPE;
-			    m_freem(m);
-			}
-			if (nd->nd_nam2)
+	cacherep = nfsrv_getcache(nd, slp, &mreq);
+	switch (cacherep) {
+	case RC_DOIT:
+		error = (*(nfsrv3_procs[nd->nd_procnum]))(nd, slp, nfsd->nfsd_procp, &mreq);
+		if (mreq == NULL) {
+			if (nd != NULL) {
 				m_freem(nd->nd_nam2);
-			if (nd->nd_mrep)
 				m_freem(nd->nd_mrep);
-			if (error == EPIPE)
-				nfsrv_zapsock(slp);
-			if (solockp)
-				nfs_sndunlock(solockp);
-			if (error == EINTR || error == ERESTART) {
-		    		pool_put(&nfsrv_descript_pl, nd);
-				nfsrv_slpderef(slp);
-				s = splsoftnet();
-				goto done;
 			}
 			break;
-		    case RC_DROPIT:
-			m_freem(nd->nd_mrep);
-			m_freem(nd->nd_nam2);
+		}
+		if (error) {
+			nfsstats.srv_errs++;
+			nfsrv_updatecache(nd, 0, mreq);
+			if (nd->nd_nam2)
+				m_freem(nd->nd_nam2);
 			break;
-		    };
-		    if (nd) {
-		    	pool_put(&nfsrv_descript_pl, nd);
-			nd = NULL;
-		    }
+		}
+		nfsstats.srvrpccnt[nd->nd_procnum]++;
+		nfsrv_updatecache(nd, 1, mreq);
+		nd->nd_mrep = NULL;
+
+		/* FALLTHROUGH */
+	case RC_REPLY:
+		m = mreq;
+		siz = 0;
+		while (m) {
+			siz += m->m_len;
+			m = m->m_next;
+		}
 
-		    /*
-		     * Check to see if there are outstanding writes that
-		     * need to be serviced.
-		     */
-		    getmicrotime(&tv);
-		    s = splsoftclock();
-		    if (LIST_FIRST(&slp->ns_tq) &&
-			timercmp(&LIST_FIRST(&slp->ns_tq)->nd_time, &tv, <=)) {
-			cacherep = RC_DOIT;
-			writes_todo = 1;
-		    } else
-			writes_todo = 0;
-		    splx(s);
-		} while (writes_todo);
-		s = splsoftnet();
-		if (nfsrv_dorec(slp, nfsd, &nd)) {
-			nfsd->nfsd_flag &= ~NFSD_REQINPROG;
-			nfsd->nfsd_slp = NULL;
+		if (siz <= 0 || siz > NFS_MAXPACKET)
+			panic("bad nfs svc reply, siz = %i\n", siz);
+
+		m = mreq;
+		m->m_pkthdr.len = siz;
+		m->m_pkthdr.rcvif = NULL;
+
+		/* For stream protocols, prepend a Sun RPC Record Mark. */
+		if (sotype == SOCK_STREAM) {
+			M_PREPEND(m, NFSX_UNSIGNED, M_WAIT);
+			*mtod(m, u_int32_t *) = htonl(0x80000000 | siz);
+		}
+
+		if (solockp)
+			nfs_sndlock(solockp, NULL);
+
+		if (ISSET(slp->ns_flag, SLP_VALID))
+		    error = nfs_send(so, nd->nd_nam2, m, NULL);
+		else {
+		    error = EPIPE;
+		    m_freem(m);
+		}
+		if (nd->nd_nam2)
+			m_freem(nd->nd_nam2);
+		if (nd->nd_mrep)
+			m_freem(nd->nd_mrep);
+		if (error == EPIPE)
+			nfsrv_zapsock(slp);
+		if (solockp)
+			nfs_sndunlock(solockp);
+		if (error == EINTR || error == ERESTART) {
+			pool_put(&nfsrv_descript_pl, nd);
 			nfsrv_slpderef(slp);
+			s = splsoftnet();
+			goto done;
 		}
+		break;
+	case RC_DROPIT:
+		m_freem(nd->nd_mrep);
+		m_freem(nd->nd_nam2);
+		break;
+	};
+
+	if (nd) {
+		pool_put(&nfsrv_descript_pl, nd);
+		nd = NULL;
 	}
+
+	s = splsoftnet();
+	if (nfsrv_dorec(slp, nfsd, &nd)) {
+		nfsd->nfsd_flag &= ~NFSD_REQINPROG;
+		nfsd->nfsd_slp = NULL;
+		nfsrv_slpderef(slp);
+	}
+	goto loop;
+
 done:
 	TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain);
 	splx(s);
@@ -509,11 +474,9 @@ void
 nfsrv_zapsock(slp)
 	struct nfssvc_sock *slp;
 {
-	struct nfsrv_descript *nwp, *nnwp;
 	struct socket *so;
 	struct file *fp;
 	struct mbuf *m, *n;
-	int s;
 
 	slp->ns_flag &= ~SLP_ALLFLAGS;
 	fp = slp->ns_fp;
@@ -533,14 +496,6 @@ nfsrv_zapsock(slp)
 			m_freem(m);
 			m = n;
 		}
-		s = splsoftclock();
-		for (nwp = LIST_FIRST(&slp->ns_tq); nwp != NULL; nwp = nnwp) {
-			nnwp = LIST_NEXT(nwp, nd_tq);
-			LIST_REMOVE(nwp, nd_tq);
-			pool_put(&nfsrv_descript_pl, nwp);
-		}
-		LIST_INIT(&slp->ns_tq);
-		splx(s);
 	}
 }
 
diff --git a/sys/nfs/nfs_var.h b/sys/nfs/nfs_var.h
index 71d9e46ac2f..3b050439731 100644
--- a/sys/nfs/nfs_var.h
+++ b/sys/nfs/nfs_var.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: nfs_var.h,v 1.52 2009/08/04 17:12:39 thib Exp $	*/
+/*	$OpenBSD: nfs_var.h,v 1.53 2009/08/10 09:38:44 thib Exp $	*/
 /*	$NetBSD: nfs_var.h,v 1.3 1996/02/18 11:53:54 fvdl Exp $	*/
 
 /*
@@ -139,9 +139,6 @@ int nfsrv_read(struct nfsrv_descript *, struct nfssvc_sock *,
 		    struct proc *, struct mbuf **);
 int nfsrv_write(struct nfsrv_descript *, struct nfssvc_sock *,
 		     struct proc *, struct mbuf **);
-int nfsrv_writegather(struct nfsrv_descript **, struct nfssvc_sock *,
-			   struct proc *, struct mbuf **);
-void nfsrvw_coalesce(struct nfsrv_descript *, struct nfsrv_descript *);
 int nfsrv_create(struct nfsrv_descript *, struct nfssvc_sock *,
 		      struct proc *, struct mbuf **);
 int nfsrv_mknod(struct nfsrv_descript *, struct nfssvc_sock *,
@@ -249,8 +246,6 @@ void nfs_add_tobecommitted_range(struct vnode *, struct buf *);
 void nfs_del_tobecommitted_range(struct vnode *, struct buf *);
 void nfs_merge_commit_ranges(struct vnode *);
 int nfsrv_errmap(struct nfsrv_descript *, int);
-void nfsrvw_sort(gid_t *, int);
-void nfsrv_setcred(struct ucred *, struct ucred *);
 int nfsm_srvsattr(struct mbuf **, struct vattr *, struct mbuf *, caddr_t *);
 void nfsm_fhtom(struct mbuf **, struct vnode *, int);
 void nfsm_srvfhtom(struct mbuf **, fhandle_t *, int);