Correct cases of mishandling of pending reads and writes to prevent

them going negative - this consists of identifying a number of cases of
IO not going through the buffer cache and marking those buffers with
B_RAW - as well as fixing nfs_bio to show pending writes and reads through
the buffer cache via NFS

still has a problem with mishandling the counters I believe in the
async/sync fallback case where counters stay positive which will be
addressed seperately.

ok tedu@ deraadt@

8 files changed, 29 insertions, 16 deletions

diff --git a/share/man/man9/buffercache.9 b/share/man/man9/buffercache.9
index dc6d02ec441..6b855defed8 100644
--- a/share/man/man9/buffercache.9
+++ b/share/man/man9/buffercache.9
@@ -1,4 +1,4 @@
-.\"	$OpenBSD: buffercache.9,v 1.5 2007/05/31 19:20:00 jmc Exp $
+.\"	$OpenBSD: buffercache.9,v 1.6 2008/07/23 16:24:43 beck Exp $
 .\"     $NetBSD: buffercache.9,v 1.13 2004/06/25 15:31:37 wiz Exp $
 .\"
 .\" Copyright (c)2003 YAMAMOTO Takashi,
@@ -102,7 +102,7 @@
 .\"
 .\"
 .\" ------------------------------------------------------------
-.Dd $Mdocdate: May 31 2007 $
+.Dd $Mdocdate: July 23 2008 $
 .Dt BUFFERCACHE 9
 .Os
 .Sh NAME
@@ -327,7 +327,12 @@ Otherwise, wake up the waiting processes.
 .\" - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
 .It Fn biowait "bp"
 Wait for operations on the buffer to complete.
-When they do, extract and return the I/O's error value.
+When they do, extract and return the I/O's error value. 
+If the operation on the buffer is being done via a direct call to a 
+.Fn strategy 
+type function, then the buffer must be previously initialized with the 
+.Dv B_RAW
+flag.
 .El
 .\" ------------------------------------------------------------
 .Sh CODE REFERENCES
diff --git a/sys/arch/aviion/aviion/disksubr.c b/sys/arch/aviion/aviion/disksubr.c
index 575ae057f1c..e9cc38389f5 100644
--- a/sys/arch/aviion/aviion/disksubr.c
+++ b/sys/arch/aviion/aviion/disksubr.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: disksubr.c,v 1.43 2008/06/12 06:58:33 deraadt Exp $	*/
+/*	$OpenBSD: disksubr.c,v 1.44 2008/07/23 16:24:42 beck Exp $	*/
 /*	$NetBSD: disksubr.c,v 1.21 1996/05/03 19:42:03 christos Exp $	*/
 
 /*
@@ -114,7 +114,7 @@ writedisklabel(dev_t dev, void (*strat)(struct buf *), struct disklabel *lp)
 	/* Read it in, slap the new label in, and write it back out */
 	bp->b_blkno = partoff + LABELSECTOR;
 	bp->b_bcount = lp->d_secsize;
-	bp->b_flags = B_BUSY | B_READ;
+	bp->b_flags = B_BUSY | B_READ | B_RAW;
 	(*strat)(bp);
 	if ((error = biowait(bp)) != 0)
 		goto done;
diff --git a/sys/dev/raidframe/rf_openbsdkintf.c b/sys/dev/raidframe/rf_openbsdkintf.c
index 46f9f9b8054..cac048abce7 100644
--- a/sys/dev/raidframe/rf_openbsdkintf.c
+++ b/sys/dev/raidframe/rf_openbsdkintf.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: rf_openbsdkintf.c,v 1.49 2008/06/26 05:42:17 ray Exp $	*/
+/* $OpenBSD: rf_openbsdkintf.c,v 1.50 2008/07/23 16:24:43 beck Exp $	*/
 /* $NetBSD: rf_netbsdkintf.c,v 1.109 2001/07/27 03:30:07 oster Exp $	*/
 
 /*-
@@ -2285,7 +2285,7 @@ raidread_component_label(dev_t dev, struct vnode *b_vp,
 	/* Get our ducks in a row for the read. */
 	bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE;
 	bp->b_bcount = RF_COMPONENT_INFO_SIZE;
-	bp->b_flags |= B_READ;
+	bp->b_flags |= (B_READ | B_RAW);
  	bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE;
 
 	(*bdevsw[major(bp->b_dev)].d_strategy)(bp);
@@ -2320,7 +2320,7 @@ raidwrite_component_label(dev_t dev, struct vnode *b_vp,
 	/* Get our ducks in a row for the write. */
 	bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE;
 	bp->b_bcount = RF_COMPONENT_INFO_SIZE;
-	bp->b_flags |= B_WRITE;
+	bp->b_flags |= (B_WRITE | B_RAW);
  	bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE;
 
 	memset(bp->b_data, 0, RF_COMPONENT_INFO_SIZE );
diff --git a/sys/dev/vnd.c b/sys/dev/vnd.c
index 85be177912d..602f56f2006 100644
--- a/sys/dev/vnd.c
+++ b/sys/dev/vnd.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vnd.c,v 1.87 2008/07/20 01:53:43 krw Exp $	*/
+/*	$OpenBSD: vnd.c,v 1.88 2008/07/23 16:24:43 beck Exp $	*/
 /*	$NetBSD: vnd.c,v 1.26 1996/03/30 23:06:11 christos Exp $	*/
 
 /*
@@ -652,6 +652,8 @@ vndiodone(struct buf *bp)
 		    vbp->vb_buf.b_error);
 
 		pbp->b_flags |= B_ERROR;
+		/* XXX does this matter here? */
+		(&vbp->vb_buf)->b_flags |= B_RAW;
 		pbp->b_error = biowait(&vbp->vb_buf);
 	}
 	pbp->b_resid -= vbp->vb_buf.b_bcount;
diff --git a/sys/isofs/cd9660/cd9660_vfsops.c b/sys/isofs/cd9660/cd9660_vfsops.c
index 7a162c87f6e..8e9b64f36e8 100644
--- a/sys/isofs/cd9660/cd9660_vfsops.c
+++ b/sys/isofs/cd9660/cd9660_vfsops.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cd9660_vfsops.c,v 1.47 2007/10/03 10:52:11 krw Exp $	*/
+/*	$OpenBSD: cd9660_vfsops.c,v 1.48 2008/07/23 16:24:43 beck Exp $	*/
 /*	$NetBSD: cd9660_vfsops.c,v 1.26 1997/06/13 15:38:58 pk Exp $	*/
 
 /*-
@@ -476,7 +476,7 @@ iso_disklabelspoof(dev, strat, lp)
 	for (iso_blknum = 16; iso_blknum < 100; iso_blknum++) {
 		bp->b_blkno = iso_blknum * btodb(ISO_DEFAULT_BLOCK_SIZE);
 		bp->b_bcount = ISO_DEFAULT_BLOCK_SIZE;
-		bp->b_flags = B_BUSY | B_READ;
+		bp->b_flags = B_BUSY | B_READ | B_RAW;
 		bp->b_cylinder = bp->b_blkno / lp->d_secpercyl;
 
 		/*printf("d_secsize %d iso_blknum %d b_blkno %d bcount %d\n",
diff --git a/sys/isofs/udf/udf_subr.c b/sys/isofs/udf/udf_subr.c
index 3ffa57e28eb..794d69c4f26 100644
--- a/sys/isofs/udf/udf_subr.c
+++ b/sys/isofs/udf/udf_subr.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: udf_subr.c,v 1.17 2008/06/12 06:58:39 deraadt Exp $	*/
+/*	$OpenBSD: udf_subr.c,v 1.18 2008/07/23 16:24:43 beck Exp $	*/
 
 /*
  * Copyright (c) 2006, Miodrag Vallat
@@ -110,7 +110,7 @@ udf_disklabelspoof(dev_t dev, void (*strat)(struct buf *),
 	 */
 	bp->b_blkno = sector * btodb(bsize);
 	bp->b_bcount = bsize;
-	bp->b_flags |= B_READ;
+	bp->b_flags |= (B_READ | B_RAW);
 	bp->b_resid = bp->b_blkno / lp->d_secpercyl;
 
 	(*strat)(bp);
@@ -130,7 +130,7 @@ udf_disklabelspoof(dev_t dev, void (*strat)(struct buf *),
 	for (sector = mvds_start; sector < mvds_end; sector++) {
 		bp->b_blkno = sector * btodb(bsize);
 		bp->b_bcount = bsize;
-		bp->b_flags |= B_READ;
+		bp->b_flags |= (B_READ | B_RAW);
 		bp->b_resid = bp->b_blkno / lp->d_secpercyl;
 
 		(*strat)(bp);
diff --git a/sys/nfs/nfs_bio.c b/sys/nfs/nfs_bio.c
index 929152321e0..70e6e3fc398 100644
--- a/sys/nfs/nfs_bio.c
+++ b/sys/nfs/nfs_bio.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: nfs_bio.c,v 1.51 2008/06/14 19:33:58 beck Exp $	*/
+/*	$OpenBSD: nfs_bio.c,v 1.52 2008/07/23 16:24:43 beck Exp $	*/
 /*	$NetBSD: nfs_bio.c,v 1.25.4.2 1996/07/08 20:47:04 jtc Exp $	*/
 
 /*
@@ -607,11 +607,13 @@ nfs_doio(bp, p)
 	    if (bp->b_flags & B_READ) {
 		uiop->uio_rw = UIO_READ;
 		nfsstats.read_physios++;
+		bcstats.pendingreads++; /* XXX */
 		error = nfs_readrpc(vp, uiop);
 	    } else {
 		iomode = NFSV3WRITE_DATASYNC;
 		uiop->uio_rw = UIO_WRITE;
 		nfsstats.write_physios++;
+		bcstats.pendingwrites++; /* XXX */
 		error = nfs_writerpc(vp, uiop, &iomode, &must_commit);
 	    }
 	    if (error) {
@@ -626,6 +628,7 @@ nfs_doio(bp, p)
 	    case VREG:
 		uiop->uio_offset = ((off_t)bp->b_blkno) << DEV_BSHIFT;
 		nfsstats.read_bios++;
+		bcstats.pendingreads++;
 		error = nfs_readrpc(vp, uiop);
 		if (!error) {
 		    bp->b_validoff = 0;
@@ -657,6 +660,7 @@ nfs_doio(bp, p)
 	    case VLNK:
 		uiop->uio_offset = (off_t)0;
 		nfsstats.readlink_bios++;
+		bcstats.pendingreads++; /* XXX */
 		error = nfs_readlinkrpc(vp, uiop, curproc->p_ucred);
 		break;
 	    default:
@@ -675,6 +679,7 @@ nfs_doio(bp, p)
 	    io.iov_base = (char *)bp->b_data + bp->b_dirtyoff;
 	    uiop->uio_rw = UIO_WRITE;
 	    nfsstats.write_bios++;
+	    bcstats.pendingwrites++; /* XXX */
 	    if ((bp->b_flags & (B_ASYNC | B_NEEDCOMMIT | B_NOCACHE)) == B_ASYNC)
 		iomode = NFSV3WRITE_UNSTABLE;
 	    else
diff --git a/sys/nfs/nfs_vnops.c b/sys/nfs/nfs_vnops.c
index 2741ca34473..bc1e604c86a 100644
--- a/sys/nfs/nfs_vnops.c
+++ b/sys/nfs/nfs_vnops.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: nfs_vnops.c,v 1.95 2008/07/23 11:53:07 art Exp $	*/
+/*	$OpenBSD: nfs_vnops.c,v 1.96 2008/07/23 16:24:43 beck Exp $	*/
 /*	$NetBSD: nfs_vnops.c,v 1.62.4.1 1996/07/08 20:26:52 jtc Exp $	*/
 
 /*
@@ -2997,6 +2997,7 @@ nfs_writebp(bp, force)
 	}
 
 	if( (oldflags & B_ASYNC) == 0) {
+		bp->b_flags |= B_RAW;
 		int rtval = biowait(bp);
 		if (!(oldflags & B_DELWRI) && p) {
 			++p->p_stats->p_ru.ru_oublock;