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@

1 files changed, 1 insertions, 372 deletions

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
  */