Make flushing dirty pages work correctly on filesystems that

unexpectedly do not complete writes even with sync I/O requests.
This should help the behavior of mmaped files when using
softupdates (and perhaps in other circumstances also.)

7 files changed, 39 insertions, 25 deletions

diff --git a/sys/amd64/amd64/pmap.c b/sys/amd64/amd64/pmap.c
index 5820a4e..33f9a9e 100644
--- a/sys/amd64/amd64/pmap.c
+++ b/sys/amd64/amd64/pmap.c
@@ -39,7 +39,7 @@
  * SUCH DAMAGE.
  *
  *	from:	@(#)pmap.c	7.7 (Berkeley)	5/12/91
- *	$Id: pmap.c,v 1.200 1998/05/19 00:06:35 tegge Exp $
+ *	$Id: pmap.c,v 1.201 1998/05/19 08:58:48 phk Exp $
  */
 
 /*
@@ -2496,6 +2496,7 @@ retry:
 				continue;
 			}
 			if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
+				(p->busy == 0) &&
 			    (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
 				if ((p->queue - p->pc) == PQ_CACHE)
 					vm_page_deactivate(p);
@@ -2516,6 +2517,7 @@ retry:
 			p = vm_page_lookup(object, tmpidx + pindex);
 			if (p &&
 			    ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
+				(p->busy == 0) &&
 			    (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
 				if ((p->queue - p->pc) == PQ_CACHE)
 					vm_page_deactivate(p);
@@ -2610,6 +2612,7 @@ pmap_prefault(pmap, addra, entry)
 			break;
 
 		if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
+			(m->busy == 0) &&
 		    (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
 
 			if ((m->queue - m->pc) == PQ_CACHE) {
diff --git a/sys/fs/specfs/spec_vnops.c b/sys/fs/specfs/spec_vnops.c
index 82f001d..a168bc8 100644
--- a/sys/fs/specfs/spec_vnops.c
+++ b/sys/fs/specfs/spec_vnops.c
@@ -31,7 +31,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)spec_vnops.c	8.14 (Berkeley) 5/21/95
- * $Id: spec_vnops.c,v 1.60 1998/03/08 09:57:36 julian Exp $
+ * $Id: spec_vnops.c,v 1.61 1998/04/19 23:32:26 julian Exp $
  */
 
 #include <sys/param.h>
@@ -551,7 +551,8 @@ spec_strategy(ap)
 	struct buf *bp;
 
 	bp = ap->a_bp;
-	if ((LIST_FIRST(&bp->b_dep)) != NULL && bioops.io_start)
+	if (((bp->b_flags & B_READ) == 0) &&
+		(LIST_FIRST(&bp->b_dep)) != NULL && bioops.io_start)
 		(*bioops.io_start)(bp);
 	(*bdevsw[major(bp->b_dev)]->d_strategy)(bp);
 	return (0);
diff --git a/sys/i386/i386/pmap.c b/sys/i386/i386/pmap.c
index 5820a4e..33f9a9e 100644
--- a/sys/i386/i386/pmap.c
+++ b/sys/i386/i386/pmap.c
@@ -39,7 +39,7 @@
  * SUCH DAMAGE.
  *
  *	from:	@(#)pmap.c	7.7 (Berkeley)	5/12/91
- *	$Id: pmap.c,v 1.200 1998/05/19 00:06:35 tegge Exp $
+ *	$Id: pmap.c,v 1.201 1998/05/19 08:58:48 phk Exp $
  */
 
 /*
@@ -2496,6 +2496,7 @@ retry:
 				continue;
 			}
 			if (((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
+				(p->busy == 0) &&
 			    (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
 				if ((p->queue - p->pc) == PQ_CACHE)
 					vm_page_deactivate(p);
@@ -2516,6 +2517,7 @@ retry:
 			p = vm_page_lookup(object, tmpidx + pindex);
 			if (p &&
 			    ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
+				(p->busy == 0) &&
 			    (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
 				if ((p->queue - p->pc) == PQ_CACHE)
 					vm_page_deactivate(p);
@@ -2610,6 +2612,7 @@ pmap_prefault(pmap, addra, entry)
 			break;
 
 		if (((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) &&
+			(m->busy == 0) &&
 		    (m->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
 
 			if ((m->queue - m->pc) == PQ_CACHE) {
diff --git a/sys/kern/vfs_cluster.c b/sys/kern/vfs_cluster.c
index 128940c..079c4da 100644
--- a/sys/kern/vfs_cluster.c
+++ b/sys/kern/vfs_cluster.c
@@ -33,7 +33,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)vfs_cluster.c	8.7 (Berkeley) 2/13/94
- * $Id: vfs_cluster.c,v 1.60 1998/03/19 22:48:13 dyson Exp $
+ * $Id: vfs_cluster.c,v 1.61 1998/05/01 16:29:27 bde Exp $
  */
 
 #include "opt_debug_cluster.h"
@@ -243,19 +243,15 @@ single_block_read:
 	 * handle the synchronous read
 	 */
 	if (bp) {
-		if (bp->b_flags & (B_DONE | B_DELWRI)) {
-			panic("cluster_read: DONE bp");
-		} else {
 #if defined(CLUSTERDEBUG)
-			if (rcluster)
-				printf("S(%d,%d,%d) ",
-					bp->b_lblkno, bp->b_bcount, seqcount);
+		if (rcluster)
+			printf("S(%d,%d,%d) ",
+				bp->b_lblkno, bp->b_bcount, seqcount);
 #endif
-			if ((bp->b_flags & B_CLUSTER) == 0)
-				vfs_busy_pages(bp, 0);
-			error = VOP_STRATEGY(bp);
-			curproc->p_stats->p_ru.ru_inblock++;
-		}
+		if ((bp->b_flags & B_CLUSTER) == 0)
+			vfs_busy_pages(bp, 0);
+		error = VOP_STRATEGY(bp);
+		curproc->p_stats->p_ru.ru_inblock++;
 	}
 
 	/*
diff --git a/sys/miscfs/specfs/spec_vnops.c b/sys/miscfs/specfs/spec_vnops.c
index 82f001d..a168bc8 100644
--- a/sys/miscfs/specfs/spec_vnops.c
+++ b/sys/miscfs/specfs/spec_vnops.c
@@ -31,7 +31,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)spec_vnops.c	8.14 (Berkeley) 5/21/95
- * $Id: spec_vnops.c,v 1.60 1998/03/08 09:57:36 julian Exp $
+ * $Id: spec_vnops.c,v 1.61 1998/04/19 23:32:26 julian Exp $
  */
 
 #include <sys/param.h>
@@ -551,7 +551,8 @@ spec_strategy(ap)
 	struct buf *bp;
 
 	bp = ap->a_bp;
-	if ((LIST_FIRST(&bp->b_dep)) != NULL && bioops.io_start)
+	if (((bp->b_flags & B_READ) == 0) &&
+		(LIST_FIRST(&bp->b_dep)) != NULL && bioops.io_start)
 		(*bioops.io_start)(bp);
 	(*bdevsw[major(bp->b_dev)]->d_strategy)(bp);
 	return (0);
diff --git a/sys/vm/vm_map.c b/sys/vm/vm_map.c
index d17f6ea..24c028a 100644
--- a/sys/vm/vm_map.c
+++ b/sys/vm/vm_map.c
@@ -61,7 +61,7 @@
  * any improvements or extensions that they make and grant Carnegie the
  * rights to redistribute these changes.
  *
- * $Id: vm_map.c,v 1.125 1998/05/04 17:12:52 dyson Exp $
+ * $Id: vm_map.c,v 1.126 1998/05/16 23:03:20 dyson Exp $
  */
 
 /*
@@ -76,6 +76,7 @@
 #include <sys/mman.h>
 #include <sys/buf.h>
 #include <sys/vnode.h>
+#include <sys/mount.h>
 
 #include <vm/vm.h>
 #include <vm/vm_param.h>
@@ -1631,6 +1632,8 @@ vm_map_clean(map, start, end, syncio, invalidate)
 		}
 	}
 
+	if (invalidate)
+		pmap_remove(vm_map_pmap(map), start, end);
 	/*
 	 * Make a second pass, cleaning/uncaching pages from the indicated
 	 * objects as we go.
@@ -1668,9 +1671,6 @@ vm_map_clean(map, start, end, syncio, invalidate)
 			if (object->size < OFF_TO_IDX( offset + size))
 				size = IDX_TO_OFF(object->size) - offset;
 		}
-		if (invalidate)
-			pmap_remove(vm_map_pmap(map), current->start,
-				current->start + size);
 		if (object && (object->type == OBJT_VNODE)) {
 			/*
 			 * Flush pages if writing is allowed. XXX should we continue
@@ -1690,11 +1690,13 @@ vm_map_clean(map, start, end, syncio, invalidate)
 					OFF_TO_IDX(offset),
 					OFF_TO_IDX(offset + size + PAGE_MASK),
 					flags);
-				if (invalidate)
+				if (invalidate) {
+					vm_object_pip_wait(object, "objmcl");
 					vm_object_page_remove(object,
 						OFF_TO_IDX(offset),
 						OFF_TO_IDX(offset + size + PAGE_MASK),
 						FALSE);
+				}
 				if (object->type == OBJT_VNODE)
 					VOP_UNLOCK(object->handle, 0, curproc);
 			}
diff --git a/sys/vm/vm_object.c b/sys/vm/vm_object.c
index f1abad0..df91aa0 100644
--- a/sys/vm/vm_object.c
+++ b/sys/vm/vm_object.c
@@ -61,7 +61,7 @@
  * any improvements or extensions that they make and grant Carnegie the
  * rights to redistribute these changes.
  *
- * $Id: vm_object.c,v 1.119 1998/03/16 01:55:52 dyson Exp $
+ * $Id: vm_object.c,v 1.120 1998/04/29 04:28:09 dyson Exp $
  */
 
 /*
@@ -74,6 +74,7 @@
 #include <sys/vnode.h>
 #include <sys/vmmeter.h>
 #include <sys/mman.h>
+#include <sys/mount.h>
 
 #include <vm/vm.h>
 #include <vm/vm_param.h>
@@ -651,13 +652,20 @@ rescan:
 			ma[index]->flags &= ~PG_CLEANCHK;
 		}
 		runlen = maxb + maxf + 1;
+
 		splx(s);
 		vm_pageout_flush(ma, runlen, pagerflags);
+		for (i = 0; i<runlen; i++) {
+			if (ma[i]->valid & ma[i]->dirty) {
+				vm_page_protect(ma[i], VM_PROT_READ);
+				ma[i]->flags |= PG_CLEANCHK;
+			}
+		}
 		if (object->generation != curgeneration)
 			goto rescan;
 	}
 
-	VOP_FSYNC(vp, NULL, (pagerflags & VM_PAGER_PUT_SYNC)?1:0, curproc);
+	VOP_FSYNC(vp, NULL, (pagerflags & VM_PAGER_PUT_SYNC)?MNT_WAIT:0, curproc);
 
 	object->flags &= ~OBJ_CLEANING;
 	return;