When fsync'ing a file on a filesystem using soft updates, we first try

to write all the dirty blocks. If some of those blocks have dependencies,
they will be remarked dirty when the I/O completes. On systems with
really fast I/O systems, it is possible to get in an infinite loop trying
to flush the buffers, because the I/O finishes before we can get all the
dirty buffers off the v_dirtyblkhd list and into the I/O queue. (The
previous algorithm looped over the v_dirtyblkhd list writing out buffers
until the list emptied.) So, now we mark each buffer that we try to
write so that we can distinguish the ones that are being remarked dirty
from those that we have not yet tried to flush. Once we have tried to
push every buffer once, we then push any associated metadata that is
causing the remaining buffers to be redirtied.

Submitted by:	Matthew Dillon <dillon@apollo.backplane.com>

1 files changed, 18 insertions, 10 deletions

diff --git a/sys/ufs/ffs/ffs_vnops.c b/sys/ufs/ffs/ffs_vnops.c
index 1823c0f..c44b133 100644
--- a/sys/ufs/ffs/ffs_vnops.c
+++ b/sys/ufs/ffs/ffs_vnops.c
@@ -31,7 +31,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)ffs_vnops.c	8.15 (Berkeley) 5/14/95
- * $Id: ffs_vnops.c,v 1.53 1998/10/31 15:31:27 peter Exp $
+ * $Id: ffs_vnops.c,v 1.54 1999/01/07 16:14:17 bde Exp $
  */
 
 #include <sys/param.h>
@@ -142,9 +142,11 @@ ffs_fsync(ap)
 	skipmeta = 0;
 	if (ap->a_waitfor == MNT_WAIT)
 		skipmeta = 1;
-loop:
 	s = splbio();
-loop2:
+loop:
+	for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp;
+	     bp = TAILQ_NEXT(bp, b_vnbufs))
+		bp->b_flags &= ~B_SCANNED;
 	for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
 		nbp = TAILQ_NEXT(bp, b_vnbufs);
 		/* 
@@ -152,7 +154,7 @@ loop2:
 		 * or if it's already scheduled, skip to the next 
 		 * buffer
 		 */
-		if ((bp->b_flags & B_BUSY) ||
+		if ((bp->b_flags & (B_BUSY | B_SCANNED)) ||
 		    ((skipmeta == 1) && (bp->b_lblkno < 0)))
 			continue;
 		if ((bp->b_flags & B_DELWRI) == 0)
@@ -162,6 +164,7 @@ loop2:
 		 * asked to wait for everything, or it's not a file or BDEV,
 		 * start the IO on this buffer immediatly.
 		 */
+		bp->b_flags |= B_SCANNED;
 		if (((bp->b_vp != vp) || (ap->a_waitfor == MNT_WAIT)) ||
 		    ((vp->v_type != VREG) && (vp->v_type != VBLK))) {
 
@@ -174,18 +177,19 @@ loop2:
 				if ((bp->b_flags & B_CLUSTEROK) &&
 				    ap->a_waitfor != MNT_WAIT) {
 					(void) vfs_bio_awrite(bp);
-					splx(s);
 				} else {
 					bremfree(bp);
 					bp->b_flags |= B_BUSY;
 					splx(s);
 					(void) bawrite(bp);
+					s = splbio();
 				}
 			} else {
 				bremfree(bp);
 				bp->b_flags |= B_BUSY;
 				splx(s);
 				(void) bwrite(bp);
+				s = splbio();
 			}
 		} else if ((vp->v_type == VREG) && (bp->b_lblkno >= lbn)) {
 			/* 
@@ -194,13 +198,17 @@ loop2:
 			 */
 			bremfree(bp);
 			bp->b_flags |= B_BUSY | B_INVAL | B_NOCACHE;
-			brelse(bp);
 			splx(s);
+			brelse(bp);
+			s = splbio();
 		} else {
 			vfs_bio_awrite(bp);
-			splx(s);
 		}
-		goto loop;
+		/*
+		 * Since we may have slept during the I/O, we need 
+		 * to start from a known point.
+		 */
+		nbp = TAILQ_FIRST(&vp->v_dirtyblkhd);
 	}
 	/*
 	 * If we were asked to do this synchronously, then go back for
@@ -208,7 +216,7 @@ loop2:
 	 */
 	if (skipmeta) {
 		skipmeta = 0;
-		goto loop2; /* stay within the splbio() */
+		goto loop;
 	}
 
 	if (ap->a_waitfor == MNT_WAIT) {
@@ -238,7 +246,7 @@ loop2:
 			 */
 			if (passes > 0) {
 				passes -= 1;
-				goto loop2;
+				goto loop;
 			}
 #ifdef DIAGNOSTIC
 			if (vp->v_type != VBLK)