From b07b491ee68d156b2162c24f16fd4667d1c6700f Mon Sep 17 00:00:00 2001
From: dfr <dfr@FreeBSD.org>
Date: Mon, 19 May 1997 14:36:56 +0000
Subject: [PATCH] Fix a few bugs with NFS and mmap caused by NFS' use of
 b_validoff and b_validend.  The changes to vfs_bio.c are a bit ugly but
 hopefully can be tidied up later by a slight redesign.

PR:		kern/2573, kern/2754, kern/3046 (possibly)
Reviewed by:	dyson
---
 sys/kern/vfs_bio.c         | 152 +++++++++++++++++++++++++------------
 sys/nfs/nfs.h              |   4 +-
 sys/nfs/nfs_bio.c          |  85 ++++++++++++++++++++-
 sys/nfs/nfs_vnops.c        |   9 ++-
 sys/nfs/nfsnode.h          |   3 +-
 sys/nfsclient/nfs.h        |   4 +-
 sys/nfsclient/nfs_bio.c    |  85 ++++++++++++++++++++-
 sys/nfsclient/nfs_vnops.c  |   9 ++-
 sys/nfsclient/nfsargs.h    |   4 +-
 sys/nfsclient/nfsnode.h    |   3 +-
 sys/nfsclient/nfsstats.h   |   4 +-
 sys/nfsserver/nfs.h        |   4 +-
 sys/nfsserver/nfsrvstats.h |   4 +-
 sys/vm/vm_fault.c          |   4 +-
 sys/vm/vnode_pager.c       |   7 +-
 15 files changed, 302 insertions(+), 79 deletions(-)

diff --git a/sys/kern/vfs_bio.c b/sys/kern/vfs_bio.c
index d5f6a9d047b4..0023dcad97c9 100644
--- a/sys/kern/vfs_bio.c
+++ b/sys/kern/vfs_bio.c
@@ -18,7 +18,7 @@
  * 5. Modifications may be freely made to this file if the above conditions
  *    are met.
  *
- * $Id: vfs_bio.c,v 1.114 1997/04/13 03:33:25 dyson Exp $
+ * $Id: vfs_bio.c,v 1.115 1997/05/10 09:09:42 joerg Exp $
  */
 
 /*
@@ -77,6 +77,10 @@ static void vm_hold_free_pages(struct buf * bp, vm_offset_t from,
 		vm_offset_t to);
 static void vm_hold_load_pages(struct buf * bp, vm_offset_t from,
 		vm_offset_t to);
+static void vfs_buf_set_valid(struct buf *bp, vm_ooffset_t foff,
+			      vm_offset_t off, vm_offset_t size,
+			      vm_page_t m);
+static void vfs_page_set_valid(struct buf *bp, vm_offset_t off, vm_page_t m);
 static void vfs_clean_pages(struct buf * bp);
 static void vfs_setdirty(struct buf *bp);
 static void vfs_vmio_release(struct buf *bp);
@@ -501,8 +505,19 @@ brelse(struct buf * bp)
 	 * constituted, so the B_INVAL flag is used to *invalidate* the buffer,
 	 * but the VM object is kept around.  The B_NOCACHE flag is used to
 	 * invalidate the pages in the VM object.
+	 *
+	 * If the buffer is a partially filled NFS buffer, keep it
+	 * since invalidating it now will lose informatio.  The valid
+	 * flags in the vm_pages have only DEV_BSIZE resolution but
+	 * the b_validoff, b_validend fields have byte resolution.
+	 * This can avoid unnecessary re-reads of the buffer.
 	 */
-	if (bp->b_flags & B_VMIO) {
+	if ((bp->b_flags & B_VMIO)
+	    && (bp->b_vp->v_tag != VT_NFS
+		|| (bp->b_flags & (B_NOCACHE | B_INVAL | B_ERROR))
+		|| bp->b_validend == 0
+		|| (bp->b_validoff == 0
+		    && bp->b_validend == bp->b_bufsize))) {
 		vm_ooffset_t foff;
 		vm_object_t obj;
 		int i, resid;
@@ -1418,6 +1433,7 @@ allocbuf(struct buf * bp, int size)
 				curbpnpages = bp->b_npages;
 		doretry:
 				bp->b_flags |= B_CACHE;
+				bp->b_validoff = bp->b_validend = 0;
 				for (toff = 0; toff < newbsize; toff += tinc) {
 					int bytesinpage;
 
@@ -1433,12 +1449,8 @@ allocbuf(struct buf * bp, int size)
 						bytesinpage = tinc;
 						if (tinc > (newbsize - toff))
 							bytesinpage = newbsize - toff;
-						if ((bp->b_flags & B_CACHE) &&
-							!vm_page_is_valid(m,
-							(vm_offset_t) ((toff + off) & PAGE_MASK),
-							bytesinpage)) {
-							bp->b_flags &= ~B_CACHE;
-						}
+						if (bp->b_flags & B_CACHE)
+							vfs_buf_set_valid(bp, off, toff, bytesinpage, m);
 						continue;
 					}
 					m = vm_page_lookup(obj, objoff);
@@ -1474,17 +1486,15 @@ allocbuf(struct buf * bp, int size)
 						bytesinpage = tinc;
 						if (tinc > (newbsize - toff))
 							bytesinpage = newbsize - toff;
-						if ((bp->b_flags & B_CACHE) &&
-							!vm_page_is_valid(m,
-							(vm_offset_t) ((toff + off) & PAGE_MASK),
-							bytesinpage)) {
-							bp->b_flags &= ~B_CACHE;
-						}
+						if (bp->b_flags & B_CACHE)
+							vfs_buf_set_valid(bp, off, toff, bytesinpage, m);
 						vm_page_wire(m);
 					}
 					bp->b_pages[pageindex] = m;
 					curbpnpages = pageindex + 1;
 				}
+				if (vp->v_tag == VT_NFS && bp->b_validend == 0)
+					bp->b_flags &= ~B_CACHE;
 				bp->b_data = (caddr_t) trunc_page(bp->b_data);
 				bp->b_npages = curbpnpages;
 				pmap_qenter((vm_offset_t) bp->b_data,
@@ -1562,7 +1572,7 @@ biodone(register struct buf * bp)
 	}
 	if (bp->b_flags & B_VMIO) {
 		int i, resid;
-		vm_ooffset_t foff;
+		vm_ooffset_t foff, bfoff;
 		vm_page_t m;
 		vm_object_t obj;
 		int iosize;
@@ -1572,6 +1582,7 @@ biodone(register struct buf * bp)
 			foff = (vm_ooffset_t) DEV_BSIZE * bp->b_lblkno;
 		else
 			foff = (vm_ooffset_t) vp->v_mount->mnt_stat.f_iosize * bp->b_lblkno;
+		bfoff = foff;
 		obj = vp->v_object;
 		if (!obj) {
 			panic("biodone: no object");
@@ -1613,8 +1624,7 @@ biodone(register struct buf * bp)
 			 * here in the read case.
 			 */
 			if ((bp->b_flags & B_READ) && !bogusflag && resid > 0) {
-				vm_page_set_validclean(m,
-					(vm_offset_t) (foff & PAGE_MASK), resid);
+				vfs_page_set_valid(bp, foff - bfoff, m);
 			}
 
 			/*
@@ -1760,6 +1770,73 @@ vfs_unbusy_pages(struct buf * bp)
 	}
 }
 
+/*
+ * Set NFS' b_validoff and b_validend fields from the valid bits
+ * of a page.  If the consumer is not NFS, and the page is not
+ * valid for the entire range, clear the B_CACHE flag to force
+ * the consumer to re-read the page.
+ */
+static void
+vfs_buf_set_valid(struct buf *bp,
+		  vm_ooffset_t foff, vm_offset_t off, vm_offset_t size,
+		  vm_page_t m)
+{
+	if (bp->b_vp->v_tag == VT_NFS) {
+		vm_offset_t svalid, evalid;
+		int validbits = m->valid;
+
+		/*
+		 * This only bothers with the first valid range in the
+		 * page.
+		 */
+		svalid = off;
+		while (validbits && !(validbits & 1)) {
+			svalid += DEV_BSIZE;
+			validbits >>= 1;
+		}
+		evalid = svalid;
+		while (validbits & 1) {
+			evalid += DEV_BSIZE;
+			validbits >>= 1;
+		}
+		/*
+		 * Make sure this range is contiguous with the range
+		 * built up from previous pages.  If not, then we will
+		 * just use the range from the previous pages.
+		 */
+		if (svalid == bp->b_validend) {
+			bp->b_validoff = min(bp->b_validoff, svalid);
+			bp->b_validend = max(bp->b_validend, evalid);
+		}
+	} else if (!vm_page_is_valid(m,
+				     (vm_offset_t) ((foff + off) & PAGE_MASK),
+				     size)) {
+		bp->b_flags &= ~B_CACHE;
+	}
+}
+
+/*
+ * Set the valid bits in a page, taking care of the b_validoff,
+ * b_validend fields which NFS uses to optimise small reads.  Off is
+ * the offset of the page within the buf.
+ */
+static void
+vfs_page_set_valid(struct buf *bp, vm_offset_t off, vm_page_t m)
+{
+	struct vnode *vp = bp->b_vp;
+	vm_offset_t soff, eoff;
+
+	soff = off;
+	eoff = min(off + PAGE_SIZE, bp->b_bufsize);
+	if (vp->v_tag == VT_NFS) {
+		soff = max((bp->b_validoff + DEV_BSIZE - 1) & -DEV_BSIZE, soff);
+		eoff = min(bp->b_validend & -DEV_BSIZE, eoff);
+	}
+	vm_page_set_invalid(m, 0, PAGE_SIZE);
+	if (eoff > soff)
+		vm_page_set_validclean(m, soff, eoff - soff);
+}
+
 /*
  * This routine is called before a device strategy routine.
  * It is used to tell the VM system that paging I/O is in
@@ -1775,36 +1852,25 @@ vfs_busy_pages(struct buf * bp, int clear_modify)
 
 	if (bp->b_flags & B_VMIO) {
 		vm_object_t obj = bp->b_vp->v_object;
-		vm_ooffset_t foff;
-		int iocount = bp->b_bufsize;
+		vm_offset_t off;
 
-		if (bp->b_vp->v_type == VBLK)
-			foff = (vm_ooffset_t) DEV_BSIZE * bp->b_lblkno;
-		else
-			foff = (vm_ooffset_t) bp->b_vp->v_mount->mnt_stat.f_iosize * bp->b_lblkno;
 		vfs_setdirty(bp);
-		for (i = 0; i < bp->b_npages; i++) {
+		for (i = 0, off = 0; i < bp->b_npages; i++, off += PAGE_SIZE) {
 			vm_page_t m = bp->b_pages[i];
-			int resid = IDX_TO_OFF(m->pindex + 1) - foff;
 
-			if (resid > iocount)
-				resid = iocount;
 			if ((bp->b_flags & B_CLUSTER) == 0) {
 				obj->paging_in_progress++;
 				m->busy++;
 			}
 			vm_page_protect(m, VM_PROT_NONE);
-			if (clear_modify) {
-				vm_page_set_validclean(m,
-					(vm_offset_t) (foff & PAGE_MASK), resid);
-			} else if (bp->b_bcount >= PAGE_SIZE) {
+			if (clear_modify)
+				vfs_page_set_valid(bp, off, m);
+			else if (bp->b_bcount >= PAGE_SIZE) {
 				if (m->valid && (bp->b_flags & B_CACHE) == 0) {
 					bp->b_pages[i] = bogus_page;
 					pmap_qenter(trunc_page(bp->b_data), bp->b_pages, bp->b_npages);
 				}
 			}
-			foff += resid;
-			iocount -= resid;
 		}
 	}
 }
@@ -1820,26 +1886,12 @@ vfs_clean_pages(struct buf * bp)
 	int i;
 
 	if (bp->b_flags & B_VMIO) {
-		vm_ooffset_t foff;
-		int iocount = bp->b_bufsize;
+		vm_offset_t off;
 
-		if (bp->b_vp->v_type == VBLK)
-			foff = (vm_ooffset_t) DEV_BSIZE * bp->b_lblkno;
-		else
-			foff = (vm_ooffset_t) bp->b_vp->v_mount->mnt_stat.f_iosize * bp->b_lblkno;
-
-		for (i = 0; i < bp->b_npages; i++) {
+		for (i = 0, off = 0; i < bp->b_npages; i++, off += PAGE_SIZE) {
 			vm_page_t m = bp->b_pages[i];
-			int resid = IDX_TO_OFF(m->pindex + 1) - foff;
 
-			if (resid > iocount)
-				resid = iocount;
-			if (resid > 0) {
-				vm_page_set_validclean(m,
-					((vm_offset_t) foff & PAGE_MASK), resid);
-			}
-			foff += resid;
-			iocount -= resid;
+			vfs_page_set_valid(bp, off, m);
 		}
 	}
 }
diff --git a/sys/nfs/nfs.h b/sys/nfs/nfs.h
index 406024ef0875..2dc19c176136 100644
--- a/sys/nfs/nfs.h
+++ b/sys/nfs/nfs.h
@@ -34,7 +34,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)nfs.h	8.4 (Berkeley) 5/1/95
- * $Id: nfs.h,v 1.25 1997/05/10 16:12:02 dfr Exp $
+ * $Id: nfs.h,v 1.26 1997/05/13 17:25:44 dfr Exp $
  */
 
 #ifndef _NFS_NFS_H_
@@ -583,7 +583,7 @@ int	nfs_connect __P((struct nfsmount *,struct nfsreq *));
 void	nfs_disconnect __P((struct nfsmount *));
 int	nfs_getattrcache __P((struct vnode *,struct vattr *));
 int	nfsm_strtmbuf __P((struct mbuf **,char **,char *,long));
-int	nfs_bioread __P((struct vnode *,struct uio *,int,struct ucred *));
+int	nfs_bioread __P((struct vnode *,struct uio *,int,struct ucred *, int getpages));
 int	nfsm_uiotombuf __P((struct uio *,struct mbuf **,int,caddr_t *));
 void	nfsrv_init __P((int));
 void	nfs_clearcommit __P((struct mount *));
diff --git a/sys/nfs/nfs_bio.c b/sys/nfs/nfs_bio.c
index bc550fc071fe..2a02371a7ac9 100644
--- a/sys/nfs/nfs_bio.c
+++ b/sys/nfs/nfs_bio.c
@@ -34,7 +34,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)nfs_bio.c	8.9 (Berkeley) 3/30/95
- * $Id: nfs_bio.c,v 1.36 1997/04/19 14:28:36 dfr Exp $
+ * $Id: nfs_bio.c,v 1.37 1997/05/13 19:41:32 dfr Exp $
  */
 
 
@@ -52,6 +52,11 @@
 #include <vm/vm.h>
 #include <vm/vm_param.h>
 #include <vm/vm_extern.h>
+#include <vm/vm_prot.h>
+#include <vm/vm_page.h>
+#include <vm/vm_object.h>
+#include <vm/vm_pager.h>
+#include <vm/vnode_pager.h>
 
 #include <nfs/rpcv2.h>
 #include <nfs/nfsproto.h>
@@ -66,16 +71,71 @@ static struct buf *nfs_getcacheblk __P((struct vnode *vp, daddr_t bn, int size,
 extern int nfs_numasync;
 extern struct nfsstats nfsstats;
 
+/*
+ * Vnode op for VM getpages.
+ */
+int
+nfs_getpages(ap)
+	struct vop_getpages_args *ap;
+{
+	int i, bsize;
+	vm_object_t obj;
+	int pcount;
+	struct uio auio;
+	struct iovec aiov;
+	int error;
+	vm_page_t m;
+
+	if (!(ap->a_vp->v_flag & VVMIO)) {
+		printf("nfs_getpages: called with non-VMIO vnode??\n");
+		return EOPNOTSUPP;
+	}
+
+	pcount = round_page(ap->a_count) / PAGE_SIZE;
+
+	obj = ap->a_m[ap->a_reqpage]->object;
+	bsize = ap->a_vp->v_mount->mnt_stat.f_iosize;
+
+	for (i = 0; i < pcount; i++) {
+		if (i != ap->a_reqpage) {
+			vnode_pager_freepage(ap->a_m[i]);
+		}
+	}
+	m = ap->a_m[ap->a_reqpage];
+
+	m->busy++;
+	m->flags &= ~PG_BUSY;
+
+	auio.uio_iov = &aiov;
+	auio.uio_iovcnt = 1;
+	aiov.iov_base = 0;
+	aiov.iov_len = MAXBSIZE;
+	auio.uio_resid = MAXBSIZE;
+	auio.uio_offset = IDX_TO_OFF(m->pindex);
+	auio.uio_segflg = UIO_NOCOPY;
+	auio.uio_rw = UIO_READ;
+	auio.uio_procp = curproc;
+	error = nfs_bioread(ap->a_vp, &auio, IO_NODELOCKED, curproc->p_ucred, 1);
+
+	m->flags |= PG_BUSY;
+	m->busy--;
+
+	if (error && (auio.uio_resid == MAXBSIZE))
+		return VM_PAGER_ERROR;
+	return 0;
+}
+
 /*
  * Vnode op for read using bio
  * Any similarity to readip() is purely coincidental
  */
 int
-nfs_bioread(vp, uio, ioflag, cred)
+nfs_bioread(vp, uio, ioflag, cred, getpages)
 	register struct vnode *vp;
 	register struct uio *uio;
 	int ioflag;
 	struct ucred *cred;
+	int getpages;
 {
 	register struct nfsnode *np = VTONFS(vp);
 	register int biosize, diff, i;
@@ -235,6 +295,27 @@ nfs_bioread(vp, uio, ioflag, cred)
 		bp = nfs_getcacheblk(vp, lbn, bufsize, p);
 		if (!bp)
 			return (EINTR);
+		/*
+		 * If we are being called from nfs_getpages, we must
+		 * make sure the buffer is a vmio buffer.  The vp will
+		 * already be setup for vmio but there may be some old
+		 * non-vmio buffers attached to it.
+		 */
+		if (getpages && !(bp->b_flags & B_VMIO)) {
+#ifdef DIAGNOSTIC
+			printf("nfs_bioread: non vmio buf found, discarding\n");
+#endif
+			bp->b_flags |= B_NOCACHE;
+			bp->b_flags |= B_INVAFTERWRITE;
+			if (bp->b_dirtyend > 0) {
+				if ((bp->b_flags & B_DELWRI) == 0)
+					panic("nfsbioread");
+				if (VOP_BWRITE(bp) == EINTR)
+					return (EINTR);
+			} else
+				brelse(bp);
+			goto again;
+		}
 		if ((bp->b_flags & B_CACHE) == 0) {
 			bp->b_flags |= B_READ;
 			bp->b_flags &= ~(B_DONE | B_ERROR | B_INVAL);
diff --git a/sys/nfs/nfs_vnops.c b/sys/nfs/nfs_vnops.c
index d3ec002b6551..ff5ad943e67e 100644
--- a/sys/nfs/nfs_vnops.c
+++ b/sys/nfs/nfs_vnops.c
@@ -34,7 +34,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)nfs_vnops.c	8.16 (Berkeley) 5/27/95
- * $Id: nfs_vnops.c,v 1.48 1997/05/09 13:18:42 dfr Exp $
+ * $Id: nfs_vnops.c,v 1.49 1997/05/17 18:32:52 phk Exp $
  */
 
 
@@ -184,6 +184,7 @@ static struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
 	{ &vop_truncate_desc, (vop_t *)nfs_truncate },	/* truncate */
 	{ &vop_update_desc, (vop_t *)nfs_update },	/* update */
 	{ &vop_bwrite_desc, (vop_t *)nfs_bwrite },	/* bwrite */
+	{ &vop_getpages_desc, (vop_t *)nfs_getpages },	/* getpages */
 	{ NULL, NULL }
 };
 static struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
@@ -1035,7 +1036,7 @@ nfs_read(ap)
 
 	if (vp->v_type != VREG)
 		return (EPERM);
-	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
+	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
 }
 
 /*
@@ -1053,7 +1054,7 @@ nfs_readlink(ap)
 
 	if (vp->v_type != VLNK)
 		return (EPERM);
-	return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
+	return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
 }
 
 /*
@@ -2055,7 +2056,7 @@ nfs_readdir(ap)
 	 * Call nfs_bioread() to do the real work.
 	 */
 	tresid = uio->uio_resid;
-	error = nfs_bioread(vp, uio, 0, ap->a_cred);
+	error = nfs_bioread(vp, uio, 0, ap->a_cred, 0);
 
 	if (!error && uio->uio_resid == tresid)
 		nfsstats.direofcache_misses++;
diff --git a/sys/nfs/nfsnode.h b/sys/nfs/nfsnode.h
index 540d80387863..12553ef0c555 100644
--- a/sys/nfs/nfsnode.h
+++ b/sys/nfs/nfsnode.h
@@ -34,7 +34,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)nfsnode.h	8.9 (Berkeley) 5/14/95
- * $Id: nfsnode.h,v 1.19 1997/02/22 09:42:49 peter Exp $
+ * $Id: nfsnode.h,v 1.20 1997/04/04 17:49:35 dfr Exp $
  */
 
 
@@ -161,6 +161,7 @@ extern	vop_t	**spec_nfsv2nodeop_p;
 /*
  * Prototypes for NFS vnode operations
  */
+int	nfs_getpages __P((struct vop_getpages_args *));
 int	nfs_write __P((struct vop_write_args *));
 #define	nfs_lease_check ((int (*) __P((struct  vop_lease_args *)))nullop)
 int	nqnfs_vop_lease_check __P((struct vop_lease_args *));
diff --git a/sys/nfsclient/nfs.h b/sys/nfsclient/nfs.h
index 406024ef0875..2dc19c176136 100644
--- a/sys/nfsclient/nfs.h
+++ b/sys/nfsclient/nfs.h
@@ -34,7 +34,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)nfs.h	8.4 (Berkeley) 5/1/95
- * $Id: nfs.h,v 1.25 1997/05/10 16:12:02 dfr Exp $
+ * $Id: nfs.h,v 1.26 1997/05/13 17:25:44 dfr Exp $
  */
 
 #ifndef _NFS_NFS_H_
@@ -583,7 +583,7 @@ int	nfs_connect __P((struct nfsmount *,struct nfsreq *));
 void	nfs_disconnect __P((struct nfsmount *));
 int	nfs_getattrcache __P((struct vnode *,struct vattr *));
 int	nfsm_strtmbuf __P((struct mbuf **,char **,char *,long));
-int	nfs_bioread __P((struct vnode *,struct uio *,int,struct ucred *));
+int	nfs_bioread __P((struct vnode *,struct uio *,int,struct ucred *, int getpages));
 int	nfsm_uiotombuf __P((struct uio *,struct mbuf **,int,caddr_t *));
 void	nfsrv_init __P((int));
 void	nfs_clearcommit __P((struct mount *));
diff --git a/sys/nfsclient/nfs_bio.c b/sys/nfsclient/nfs_bio.c
index bc550fc071fe..2a02371a7ac9 100644
--- a/sys/nfsclient/nfs_bio.c
+++ b/sys/nfsclient/nfs_bio.c
@@ -34,7 +34,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)nfs_bio.c	8.9 (Berkeley) 3/30/95
- * $Id: nfs_bio.c,v 1.36 1997/04/19 14:28:36 dfr Exp $
+ * $Id: nfs_bio.c,v 1.37 1997/05/13 19:41:32 dfr Exp $
  */
 
 
@@ -52,6 +52,11 @@
 #include <vm/vm.h>
 #include <vm/vm_param.h>
 #include <vm/vm_extern.h>
+#include <vm/vm_prot.h>
+#include <vm/vm_page.h>
+#include <vm/vm_object.h>
+#include <vm/vm_pager.h>
+#include <vm/vnode_pager.h>
 
 #include <nfs/rpcv2.h>
 #include <nfs/nfsproto.h>
@@ -66,16 +71,71 @@ static struct buf *nfs_getcacheblk __P((struct vnode *vp, daddr_t bn, int size,
 extern int nfs_numasync;
 extern struct nfsstats nfsstats;
 
+/*
+ * Vnode op for VM getpages.
+ */
+int
+nfs_getpages(ap)
+	struct vop_getpages_args *ap;
+{
+	int i, bsize;
+	vm_object_t obj;
+	int pcount;
+	struct uio auio;
+	struct iovec aiov;
+	int error;
+	vm_page_t m;
+
+	if (!(ap->a_vp->v_flag & VVMIO)) {
+		printf("nfs_getpages: called with non-VMIO vnode??\n");
+		return EOPNOTSUPP;
+	}
+
+	pcount = round_page(ap->a_count) / PAGE_SIZE;
+
+	obj = ap->a_m[ap->a_reqpage]->object;
+	bsize = ap->a_vp->v_mount->mnt_stat.f_iosize;
+
+	for (i = 0; i < pcount; i++) {
+		if (i != ap->a_reqpage) {
+			vnode_pager_freepage(ap->a_m[i]);
+		}
+	}
+	m = ap->a_m[ap->a_reqpage];
+
+	m->busy++;
+	m->flags &= ~PG_BUSY;
+
+	auio.uio_iov = &aiov;
+	auio.uio_iovcnt = 1;
+	aiov.iov_base = 0;
+	aiov.iov_len = MAXBSIZE;
+	auio.uio_resid = MAXBSIZE;
+	auio.uio_offset = IDX_TO_OFF(m->pindex);
+	auio.uio_segflg = UIO_NOCOPY;
+	auio.uio_rw = UIO_READ;
+	auio.uio_procp = curproc;
+	error = nfs_bioread(ap->a_vp, &auio, IO_NODELOCKED, curproc->p_ucred, 1);
+
+	m->flags |= PG_BUSY;
+	m->busy--;
+
+	if (error && (auio.uio_resid == MAXBSIZE))
+		return VM_PAGER_ERROR;
+	return 0;
+}
+
 /*
  * Vnode op for read using bio
  * Any similarity to readip() is purely coincidental
  */
 int
-nfs_bioread(vp, uio, ioflag, cred)
+nfs_bioread(vp, uio, ioflag, cred, getpages)
 	register struct vnode *vp;
 	register struct uio *uio;
 	int ioflag;
 	struct ucred *cred;
+	int getpages;
 {
 	register struct nfsnode *np = VTONFS(vp);
 	register int biosize, diff, i;
@@ -235,6 +295,27 @@ nfs_bioread(vp, uio, ioflag, cred)
 		bp = nfs_getcacheblk(vp, lbn, bufsize, p);
 		if (!bp)
 			return (EINTR);
+		/*
+		 * If we are being called from nfs_getpages, we must
+		 * make sure the buffer is a vmio buffer.  The vp will
+		 * already be setup for vmio but there may be some old
+		 * non-vmio buffers attached to it.
+		 */
+		if (getpages && !(bp->b_flags & B_VMIO)) {
+#ifdef DIAGNOSTIC
+			printf("nfs_bioread: non vmio buf found, discarding\n");
+#endif
+			bp->b_flags |= B_NOCACHE;
+			bp->b_flags |= B_INVAFTERWRITE;
+			if (bp->b_dirtyend > 0) {
+				if ((bp->b_flags & B_DELWRI) == 0)
+					panic("nfsbioread");
+				if (VOP_BWRITE(bp) == EINTR)
+					return (EINTR);
+			} else
+				brelse(bp);
+			goto again;
+		}
 		if ((bp->b_flags & B_CACHE) == 0) {
 			bp->b_flags |= B_READ;
 			bp->b_flags &= ~(B_DONE | B_ERROR | B_INVAL);
diff --git a/sys/nfsclient/nfs_vnops.c b/sys/nfsclient/nfs_vnops.c
index d3ec002b6551..ff5ad943e67e 100644
--- a/sys/nfsclient/nfs_vnops.c
+++ b/sys/nfsclient/nfs_vnops.c
@@ -34,7 +34,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)nfs_vnops.c	8.16 (Berkeley) 5/27/95
- * $Id: nfs_vnops.c,v 1.48 1997/05/09 13:18:42 dfr Exp $
+ * $Id: nfs_vnops.c,v 1.49 1997/05/17 18:32:52 phk Exp $
  */
 
 
@@ -184,6 +184,7 @@ static struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
 	{ &vop_truncate_desc, (vop_t *)nfs_truncate },	/* truncate */
 	{ &vop_update_desc, (vop_t *)nfs_update },	/* update */
 	{ &vop_bwrite_desc, (vop_t *)nfs_bwrite },	/* bwrite */
+	{ &vop_getpages_desc, (vop_t *)nfs_getpages },	/* getpages */
 	{ NULL, NULL }
 };
 static struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
@@ -1035,7 +1036,7 @@ nfs_read(ap)
 
 	if (vp->v_type != VREG)
 		return (EPERM);
-	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
+	return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
 }
 
 /*
@@ -1053,7 +1054,7 @@ nfs_readlink(ap)
 
 	if (vp->v_type != VLNK)
 		return (EPERM);
-	return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
+	return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
 }
 
 /*
@@ -2055,7 +2056,7 @@ nfs_readdir(ap)
 	 * Call nfs_bioread() to do the real work.
 	 */
 	tresid = uio->uio_resid;
-	error = nfs_bioread(vp, uio, 0, ap->a_cred);
+	error = nfs_bioread(vp, uio, 0, ap->a_cred, 0);
 
 	if (!error && uio->uio_resid == tresid)
 		nfsstats.direofcache_misses++;
diff --git a/sys/nfsclient/nfsargs.h b/sys/nfsclient/nfsargs.h
index 406024ef0875..2dc19c176136 100644
--- a/sys/nfsclient/nfsargs.h
+++ b/sys/nfsclient/nfsargs.h
@@ -34,7 +34,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)nfs.h	8.4 (Berkeley) 5/1/95
- * $Id: nfs.h,v 1.25 1997/05/10 16:12:02 dfr Exp $
+ * $Id: nfs.h,v 1.26 1997/05/13 17:25:44 dfr Exp $
  */
 
 #ifndef _NFS_NFS_H_
@@ -583,7 +583,7 @@ int	nfs_connect __P((struct nfsmount *,struct nfsreq *));
 void	nfs_disconnect __P((struct nfsmount *));
 int	nfs_getattrcache __P((struct vnode *,struct vattr *));
 int	nfsm_strtmbuf __P((struct mbuf **,char **,char *,long));
-int	nfs_bioread __P((struct vnode *,struct uio *,int,struct ucred *));
+int	nfs_bioread __P((struct vnode *,struct uio *,int,struct ucred *, int getpages));
 int	nfsm_uiotombuf __P((struct uio *,struct mbuf **,int,caddr_t *));
 void	nfsrv_init __P((int));
 void	nfs_clearcommit __P((struct mount *));
diff --git a/sys/nfsclient/nfsnode.h b/sys/nfsclient/nfsnode.h
index 540d80387863..12553ef0c555 100644
--- a/sys/nfsclient/nfsnode.h
+++ b/sys/nfsclient/nfsnode.h
@@ -34,7 +34,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)nfsnode.h	8.9 (Berkeley) 5/14/95
- * $Id: nfsnode.h,v 1.19 1997/02/22 09:42:49 peter Exp $
+ * $Id: nfsnode.h,v 1.20 1997/04/04 17:49:35 dfr Exp $
  */
 
 
@@ -161,6 +161,7 @@ extern	vop_t	**spec_nfsv2nodeop_p;
 /*
  * Prototypes for NFS vnode operations
  */
+int	nfs_getpages __P((struct vop_getpages_args *));
 int	nfs_write __P((struct vop_write_args *));
 #define	nfs_lease_check ((int (*) __P((struct  vop_lease_args *)))nullop)
 int	nqnfs_vop_lease_check __P((struct vop_lease_args *));
diff --git a/sys/nfsclient/nfsstats.h b/sys/nfsclient/nfsstats.h
index 406024ef0875..2dc19c176136 100644
--- a/sys/nfsclient/nfsstats.h
+++ b/sys/nfsclient/nfsstats.h
@@ -34,7 +34,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)nfs.h	8.4 (Berkeley) 5/1/95
- * $Id: nfs.h,v 1.25 1997/05/10 16:12:02 dfr Exp $
+ * $Id: nfs.h,v 1.26 1997/05/13 17:25:44 dfr Exp $
  */
 
 #ifndef _NFS_NFS_H_
@@ -583,7 +583,7 @@ int	nfs_connect __P((struct nfsmount *,struct nfsreq *));
 void	nfs_disconnect __P((struct nfsmount *));
 int	nfs_getattrcache __P((struct vnode *,struct vattr *));
 int	nfsm_strtmbuf __P((struct mbuf **,char **,char *,long));
-int	nfs_bioread __P((struct vnode *,struct uio *,int,struct ucred *));
+int	nfs_bioread __P((struct vnode *,struct uio *,int,struct ucred *, int getpages));
 int	nfsm_uiotombuf __P((struct uio *,struct mbuf **,int,caddr_t *));
 void	nfsrv_init __P((int));
 void	nfs_clearcommit __P((struct mount *));
diff --git a/sys/nfsserver/nfs.h b/sys/nfsserver/nfs.h
index 406024ef0875..2dc19c176136 100644
--- a/sys/nfsserver/nfs.h
+++ b/sys/nfsserver/nfs.h
@@ -34,7 +34,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)nfs.h	8.4 (Berkeley) 5/1/95
- * $Id: nfs.h,v 1.25 1997/05/10 16:12:02 dfr Exp $
+ * $Id: nfs.h,v 1.26 1997/05/13 17:25:44 dfr Exp $
  */
 
 #ifndef _NFS_NFS_H_
@@ -583,7 +583,7 @@ int	nfs_connect __P((struct nfsmount *,struct nfsreq *));
 void	nfs_disconnect __P((struct nfsmount *));
 int	nfs_getattrcache __P((struct vnode *,struct vattr *));
 int	nfsm_strtmbuf __P((struct mbuf **,char **,char *,long));
-int	nfs_bioread __P((struct vnode *,struct uio *,int,struct ucred *));
+int	nfs_bioread __P((struct vnode *,struct uio *,int,struct ucred *, int getpages));
 int	nfsm_uiotombuf __P((struct uio *,struct mbuf **,int,caddr_t *));
 void	nfsrv_init __P((int));
 void	nfs_clearcommit __P((struct mount *));
diff --git a/sys/nfsserver/nfsrvstats.h b/sys/nfsserver/nfsrvstats.h
index 406024ef0875..2dc19c176136 100644
--- a/sys/nfsserver/nfsrvstats.h
+++ b/sys/nfsserver/nfsrvstats.h
@@ -34,7 +34,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)nfs.h	8.4 (Berkeley) 5/1/95
- * $Id: nfs.h,v 1.25 1997/05/10 16:12:02 dfr Exp $
+ * $Id: nfs.h,v 1.26 1997/05/13 17:25:44 dfr Exp $
  */
 
 #ifndef _NFS_NFS_H_
@@ -583,7 +583,7 @@ int	nfs_connect __P((struct nfsmount *,struct nfsreq *));
 void	nfs_disconnect __P((struct nfsmount *));
 int	nfs_getattrcache __P((struct vnode *,struct vattr *));
 int	nfsm_strtmbuf __P((struct mbuf **,char **,char *,long));
-int	nfs_bioread __P((struct vnode *,struct uio *,int,struct ucred *));
+int	nfs_bioread __P((struct vnode *,struct uio *,int,struct ucred *, int getpages));
 int	nfsm_uiotombuf __P((struct uio *,struct mbuf **,int,caddr_t *));
 void	nfsrv_init __P((int));
 void	nfs_clearcommit __P((struct mount *));
diff --git a/sys/vm/vm_fault.c b/sys/vm/vm_fault.c
index c96c42a4db85..c138ff271c80 100644
--- a/sys/vm/vm_fault.c
+++ b/sys/vm/vm_fault.c
@@ -66,7 +66,7 @@
  * any improvements or extensions that they make and grant Carnegie the
  * rights to redistribute these changes.
  *
- * $Id: vm_fault.c,v 1.67 1997/04/06 02:29:41 dyson Exp $
+ * $Id: vm_fault.c,v 1.68 1997/04/06 16:16:11 peter Exp $
  */
 
 /*
@@ -326,7 +326,7 @@ RetryFault:;
 
 			m->flags |= PG_BUSY;
 
-			if (m->valid &&
+			if (/*m->valid && */
 				((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) &&
 				m->object != kernel_object && m->object != kmem_object) {
 				goto readrest;
diff --git a/sys/vm/vnode_pager.c b/sys/vm/vnode_pager.c
index e4d7404888a0..d2583513144a 100644
--- a/sys/vm/vnode_pager.c
+++ b/sys/vm/vnode_pager.c
@@ -38,7 +38,7 @@
  * SUCH DAMAGE.
  *
  *	from: @(#)vnode_pager.c	7.5 (Berkeley) 4/20/91
- *	$Id: vnode_pager.c,v 1.69 1997/02/22 09:48:42 peter Exp $
+ *	$Id: vnode_pager.c,v 1.70 1997/03/08 04:33:47 dyson Exp $
  */
 
 /*
@@ -692,7 +692,12 @@ vnode_pager_leaf_getpages(object, m, count, reqpage)
 	/*
 	 * if ANY DEV_BSIZE blocks are valid on a large filesystem block
 	 * then, the entire page is valid --
+	 * XXX no it isn't
 	 */
+
+	if (m[reqpage]->valid != VM_PAGE_BITS_ALL)
+	    m[reqpage]->valid = 0;
+
 	if (m[reqpage]->valid) {
 		m[reqpage]->valid = VM_PAGE_BITS_ALL;
 		for (i = 0; i < count; i++) {