Revert a bunch of contraversial changes by PHK. After

a quick think and discussion among various people some form of some of
these changes will probably be recommitted.

The reversion requested was requested by dg while discussions proceed.
PHK has indicated that he can live with this, and it has been agreed
that some form of some of these changes may return shortly after further
discussion.

17 files changed, 95 insertions, 122 deletions

diff --git a/sys/conf/NOTES b/sys/conf/NOTES
index 7868bd8..544b2f4 100644
--- a/sys/conf/NOTES
+++ b/sys/conf/NOTES
@@ -625,9 +625,6 @@ pseudo-device	vcoda	4		#coda minicache <-> venus comm.
 #
 options 	EXT2FS
 
-#
-# Only set this if you positively know why you should never do that.
-options		ALLOW_BDEV_ACCESS	# enable bdev access
 
 
 #####################################################################
diff --git a/sys/conf/options b/sys/conf/options
index 2146ffe..ba04359 100644
--- a/sys/conf/options
+++ b/sys/conf/options
@@ -126,9 +126,6 @@ CD9660_ROOTDELAY	opt_cd9660.h
 # hidden yet.
 UNION
 
-# Options for all filesystems
-ALLOW_BDEV_ACCESS	opt_fs.h
-
 # Options used only in param.c.
 HZ		opt_param.h
 MAXFILES	opt_param.h
diff --git a/sys/dev/ccd/ccd.c b/sys/dev/ccd/ccd.c
index c4a0a8e..7badacb 100644
--- a/sys/dev/ccd/ccd.c
+++ b/sys/dev/ccd/ccd.c
@@ -634,6 +634,7 @@ ccdopen(dev, flags, fmt, p)
 	pmask = (1 << part);
 
 	dev->si_bsize_phys = DEV_BSIZE;
+	dev->si_bsize_best = BLKDEV_IOSIZE;
 	dev->si_bsize_max = MAXBSIZE;
 
 	/*
diff --git a/sys/dev/fdc/fdc.c b/sys/dev/fdc/fdc.c
index 539569d..7704cca 100644
--- a/sys/dev/fdc/fdc.c
+++ b/sys/dev/fdc/fdc.c
@@ -1227,6 +1227,7 @@ Fdopen(dev_t dev, int flags, int mode, struct proc *p)
 	fdc_p	fdc;
 
 	dev->si_bsize_phys = DEV_BSIZE;
+	dev->si_bsize_best = BLKDEV_IOSIZE;
 	dev->si_bsize_max = MAXBSIZE;
 	/* check bounds */
 	if ((fd = devclass_get_softc(fd_devclass, fdu)) == 0)
diff --git a/sys/dev/vinum/vinum.c b/sys/dev/vinum/vinum.c
index 46ff86a..2482185 100644
--- a/sys/dev/vinum/vinum.c
+++ b/sys/dev/vinum/vinum.c
@@ -269,6 +269,7 @@ vinumopen(dev_t dev,
 
     devminor = minor(dev);
     dev->si_bsize_phys = DEV_BSIZE;
+    dev->si_bsize_best = VINUM_BSIZE_BEST;		    /* kludge until we track drive block sizes */
     dev->si_bsize_max = MAXBSIZE;
     error = 0;
     /* First, decide what we're looking at */
diff --git a/sys/dev/vn/vn.c b/sys/dev/vn/vn.c
index 199c804..f51efca 100644
--- a/sys/dev/vn/vn.c
+++ b/sys/dev/vn/vn.c
@@ -510,7 +510,10 @@ vniocattach_file(vn, vio, dev, flag, p)
 		(void) vn_close(nd.ni_vp, flags, p->p_ucred, p);
 		return(error);
 	}
-	dev->si_bsize_phys = vn->sc_secsize;
+	if (dev->si_bsize_phys < vn->sc_secsize)
+		dev->si_bsize_phys = vn->sc_secsize;
+	if (dev->si_bsize_best < vn->sc_secsize)
+		dev->si_bsize_best = vn->sc_secsize;
 	vn->sc_flags |= VNF_INITED;
 	if (flags == FREAD)
 		vn->sc_flags |= VNF_READONLY;
diff --git a/sys/fs/specfs/spec_vnops.c b/sys/fs/specfs/spec_vnops.c
index 8fd4539..1afc400 100644
--- a/sys/fs/specfs/spec_vnops.c
+++ b/sys/fs/specfs/spec_vnops.c
@@ -34,7 +34,6 @@
  * $FreeBSD$
  */
 
-#include "opt_fs.h"
 #include <sys/param.h>
 #include <sys/proc.h>
 #include <sys/systm.h>
@@ -47,7 +46,6 @@
 #include <sys/fcntl.h>
 #include <sys/disklabel.h>
 #include <sys/vmmeter.h>
-#include <sys/sysctl.h>
 
 #include <vm/vm.h>
 #include <vm/vm_prot.h>
@@ -235,15 +233,6 @@ spec_open(ap)
 	return (0);
 }
 
-#ifdef ALLOW_BDEV_ACCESS
-static int bdev_access = 1;
-#else
-static int bdev_access;
-#endif
-
-SYSCTL_INT(_vfs, OID_AUTO, bdev_access, CTLFLAG_RW, &bdev_access, 0,
-	"allow block device access");
-
 /*
  * Vnode op for read
  */
@@ -257,8 +246,8 @@ spec_read(ap)
 		struct ucred *a_cred;
 	} */ *ap;
 {
-	struct vnode *vp = ap->a_vp;
-	struct uio *uio = ap->a_uio;
+	register struct vnode *vp = ap->a_vp;
+	register struct uio *uio = ap->a_uio;
  	struct proc *p = uio->uio_procp;
 	struct buf *bp;
 	daddr_t bn, nextbn;
@@ -269,8 +258,6 @@ spec_read(ap)
 	int error = 0;
 	dev_t dev;
 
-	dev = vp->v_rdev;
-
 #ifdef DIAGNOSTIC
 	if (uio->uio_rw != UIO_READ)
 		panic("spec_read mode");
@@ -282,14 +269,24 @@ spec_read(ap)
 
 	switch (vp->v_type) {
 
+	case VCHR:
+		VOP_UNLOCK(vp, 0, p);
+		error = (*devsw(vp->v_rdev)->d_read)
+			(vp->v_rdev, uio, ap->a_ioflag);
+		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
+		return (error);
+
 	case VBLK:
-if (bdev_access) {
 		if (uio->uio_offset < 0)
 			return (EINVAL);
+		dev = vp->v_rdev;
 
-		bsize = dev->si_bsize_phys;
-		if (bsize < BLKDEV_IOSIZE)
-			bsize = BLKDEV_IOSIZE;
+		/*
+		 * Calculate block size for block device.  The block size must
+		 * be larger then the physical minimum.
+		 */
+
+		bsize = vp->v_rdev->si_bsize_best;
 
 		if ((ioctl = devsw(dev)->d_ioctl) != NULL &&
 		    (*ioctl)(dev, DIOCGPART, (caddr_t)&dpart, FREAD, p) == 0 &&
@@ -317,12 +314,6 @@ if (bdev_access) {
 			brelse(bp);
 		} while (error == 0 && uio->uio_resid > 0 && n != 0);
 		return (error);
-}
-	case VCHR:
-		VOP_UNLOCK(vp, 0, p);
-		error = (*devsw(dev)->d_read) (dev, uio, ap->a_ioflag);
-		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
-		return (error);
 
 	default:
 		panic("spec_read type");
@@ -343,8 +334,8 @@ spec_write(ap)
 		struct ucred *a_cred;
 	} */ *ap;
 {
-	struct vnode *vp = ap->a_vp;
-	struct uio *uio = ap->a_uio;
+	register struct vnode *vp = ap->a_vp;
+	register struct uio *uio = ap->a_uio;
 	struct proc *p = uio->uio_procp;
 	struct buf *bp;
 	daddr_t bn;
@@ -352,30 +343,36 @@ spec_write(ap)
 	struct partinfo dpart;
 	register int n, on;
 	int error = 0;
-	dev_t dev;
 
-	dev = vp->v_rdev;
 #ifdef DIAGNOSTIC
 	if (uio->uio_rw != UIO_WRITE)
 		panic("spec_write mode");
 	if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc)
 		panic("spec_write proc");
 #endif
-	if (uio->uio_resid == 0)
-		return (0);
 
 	switch (vp->v_type) {
 
+	case VCHR:
+		VOP_UNLOCK(vp, 0, p);
+		error = (*devsw(vp->v_rdev)->d_write)
+			(vp->v_rdev, uio, ap->a_ioflag);
+		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
+		return (error);
+
 	case VBLK:
-if (bdev_access) {
+		if (uio->uio_resid == 0)
+			return (0);
 		if (uio->uio_offset < 0)
 			return (EINVAL);
 
-		bsize = dev->si_bsize_phys;
-		if (bsize < BLKDEV_IOSIZE)
-			bsize = BLKDEV_IOSIZE;
+		/*
+		 * Calculate block size for block device.  The block size must
+		 * be larger then the physical minimum.
+		 */
+		bsize = vp->v_rdev->si_bsize_best;
 
-		if ((*devsw(dev)->d_ioctl)(dev, DIOCGPART,
+		if ((*devsw(vp->v_rdev)->d_ioctl)(vp->v_rdev, DIOCGPART,
 		    (caddr_t)&dpart, FREAD, p) == 0) {
 			if (dpart.part->p_fstype == FS_BSDFFS &&
 			    dpart.part->p_frag != 0 && dpart.part->p_fsize != 0)
@@ -403,14 +400,6 @@ if (bdev_access) {
 				bdwrite(bp);
 		} while (error == 0 && uio->uio_resid > 0 && n != 0);
 		return (error);
-}
-
-	case VCHR:
-		VOP_UNLOCK(vp, 0, p);
-		error = (*devsw(dev)->d_write)
-			(dev, uio, ap->a_ioflag);
-		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
-		return (error);
 
 	default:
 		panic("spec_write type");
diff --git a/sys/geom/geom_ccd.c b/sys/geom/geom_ccd.c
index c4a0a8e..7badacb 100644
--- a/sys/geom/geom_ccd.c
+++ b/sys/geom/geom_ccd.c
@@ -634,6 +634,7 @@ ccdopen(dev, flags, fmt, p)
 	pmask = (1 << part);
 
 	dev->si_bsize_phys = DEV_BSIZE;
+	dev->si_bsize_best = BLKDEV_IOSIZE;
 	dev->si_bsize_max = MAXBSIZE;
 
 	/*
diff --git a/sys/i386/conf/LINT b/sys/i386/conf/LINT
index 7868bd8..544b2f4 100644
--- a/sys/i386/conf/LINT
+++ b/sys/i386/conf/LINT
@@ -625,9 +625,6 @@ pseudo-device	vcoda	4		#coda minicache <-> venus comm.
 #
 options 	EXT2FS
 
-#
-# Only set this if you positively know why you should never do that.
-options		ALLOW_BDEV_ACCESS	# enable bdev access
 
 
 #####################################################################
diff --git a/sys/i386/conf/NOTES b/sys/i386/conf/NOTES
index 7868bd8..544b2f4 100644
--- a/sys/i386/conf/NOTES
+++ b/sys/i386/conf/NOTES
@@ -625,9 +625,6 @@ pseudo-device	vcoda	4		#coda minicache <-> venus comm.
 #
 options 	EXT2FS
 
-#
-# Only set this if you positively know why you should never do that.
-options		ALLOW_BDEV_ACCESS	# enable bdev access
 
 
 #####################################################################
diff --git a/sys/isa/fd.c b/sys/isa/fd.c
index 539569d..7704cca 100644
--- a/sys/isa/fd.c
+++ b/sys/isa/fd.c
@@ -1227,6 +1227,7 @@ Fdopen(dev_t dev, int flags, int mode, struct proc *p)
 	fdc_p	fdc;
 
 	dev->si_bsize_phys = DEV_BSIZE;
+	dev->si_bsize_best = BLKDEV_IOSIZE;
 	dev->si_bsize_max = MAXBSIZE;
 	/* check bounds */
 	if ((fd = devclass_get_softc(fd_devclass, fdu)) == 0)
diff --git a/sys/kern/subr_diskslice.c b/sys/kern/subr_diskslice.c
index d8dda72..df187ef 100644
--- a/sys/kern/subr_diskslice.c
+++ b/sys/kern/subr_diskslice.c
@@ -716,6 +716,7 @@ dsopen(dev, mode, flags, sspp, lp)
 	int	unit;
 
 	dev->si_bsize_phys = lp->d_secsize;
+	dev->si_bsize_best = BLKDEV_IOSIZE;
 	dev->si_bsize_max = MAXBSIZE;
 
 	unit = dkunit(dev);
diff --git a/sys/kern/vfs_vnops.c b/sys/kern/vfs_vnops.c
index c64a77e..0a08938 100644
--- a/sys/kern/vfs_vnops.c
+++ b/sys/kern/vfs_vnops.c
@@ -429,31 +429,23 @@ vn_stat(vp, sb, p)
 	sb->st_atimespec = vap->va_atime;
 	sb->st_mtimespec = vap->va_mtime;
 	sb->st_ctimespec = vap->va_ctime;
-
 	/*
-	 * According to www.opengroup.org, the meaning of st_blksize is 
-	 *   "a filesystem-specific preferred I/O block size for this 
-	 *    object.  In some filesystem types, this may vary from file
-	 *    to file"
-	 * Default to zero to catch bogus uses of this field.
+	 * For block and char device nodes we don't really care
+	 * about what the filesystem told us, we want to know
+	 * what the device told us
 	 */
-	sb->st_blksize = 0;
-
-	if (vp->v_type == VREG)
-		sb->st_blksize = vap->va_blocksize;
-
-	/*
-	 * For disks we can say something sensible.  We report the max
-	 * size because we prefer few big transfers than many small.
-	 * XXX: Only reliable if the disk is opened.
-	 * XXX: Use vn_isdisk when it allows VCHR too 
-	 */
-	if ((vp->v_type == VBLK || vp->v_type == VCHR) &&
-	    devsw(vp->v_rdev) && (devsw(vp->v_rdev)->d_flags & D_DISK))
+	switch (vap->va_type) {
+	case VBLK:
+		sb->st_blksize = vp->v_rdev->si_bsize_best;
+		break;
+	case VCHR:
 		sb->st_blksize = vp->v_rdev->si_bsize_max;
-
+		break;
+	default:
+		sb->st_blksize = vap->va_blocksize;
+		break;
+	}
 	sb->st_flags = vap->va_flags;
-
 	if (suser_xxx(p->p_ucred, 0, 0))
 		sb->st_gen = 0;
 	else
diff --git a/sys/miscfs/specfs/spec_vnops.c b/sys/miscfs/specfs/spec_vnops.c
index 8fd4539..1afc400 100644
--- a/sys/miscfs/specfs/spec_vnops.c
+++ b/sys/miscfs/specfs/spec_vnops.c
@@ -34,7 +34,6 @@
  * $FreeBSD$
  */
 
-#include "opt_fs.h"
 #include <sys/param.h>
 #include <sys/proc.h>
 #include <sys/systm.h>
@@ -47,7 +46,6 @@
 #include <sys/fcntl.h>
 #include <sys/disklabel.h>
 #include <sys/vmmeter.h>
-#include <sys/sysctl.h>
 
 #include <vm/vm.h>
 #include <vm/vm_prot.h>
@@ -235,15 +233,6 @@ spec_open(ap)
 	return (0);
 }
 
-#ifdef ALLOW_BDEV_ACCESS
-static int bdev_access = 1;
-#else
-static int bdev_access;
-#endif
-
-SYSCTL_INT(_vfs, OID_AUTO, bdev_access, CTLFLAG_RW, &bdev_access, 0,
-	"allow block device access");
-
 /*
  * Vnode op for read
  */
@@ -257,8 +246,8 @@ spec_read(ap)
 		struct ucred *a_cred;
 	} */ *ap;
 {
-	struct vnode *vp = ap->a_vp;
-	struct uio *uio = ap->a_uio;
+	register struct vnode *vp = ap->a_vp;
+	register struct uio *uio = ap->a_uio;
  	struct proc *p = uio->uio_procp;
 	struct buf *bp;
 	daddr_t bn, nextbn;
@@ -269,8 +258,6 @@ spec_read(ap)
 	int error = 0;
 	dev_t dev;
 
-	dev = vp->v_rdev;
-
 #ifdef DIAGNOSTIC
 	if (uio->uio_rw != UIO_READ)
 		panic("spec_read mode");
@@ -282,14 +269,24 @@ spec_read(ap)
 
 	switch (vp->v_type) {
 
+	case VCHR:
+		VOP_UNLOCK(vp, 0, p);
+		error = (*devsw(vp->v_rdev)->d_read)
+			(vp->v_rdev, uio, ap->a_ioflag);
+		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
+		return (error);
+
 	case VBLK:
-if (bdev_access) {
 		if (uio->uio_offset < 0)
 			return (EINVAL);
+		dev = vp->v_rdev;
 
-		bsize = dev->si_bsize_phys;
-		if (bsize < BLKDEV_IOSIZE)
-			bsize = BLKDEV_IOSIZE;
+		/*
+		 * Calculate block size for block device.  The block size must
+		 * be larger then the physical minimum.
+		 */
+
+		bsize = vp->v_rdev->si_bsize_best;
 
 		if ((ioctl = devsw(dev)->d_ioctl) != NULL &&
 		    (*ioctl)(dev, DIOCGPART, (caddr_t)&dpart, FREAD, p) == 0 &&
@@ -317,12 +314,6 @@ if (bdev_access) {
 			brelse(bp);
 		} while (error == 0 && uio->uio_resid > 0 && n != 0);
 		return (error);
-}
-	case VCHR:
-		VOP_UNLOCK(vp, 0, p);
-		error = (*devsw(dev)->d_read) (dev, uio, ap->a_ioflag);
-		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
-		return (error);
 
 	default:
 		panic("spec_read type");
@@ -343,8 +334,8 @@ spec_write(ap)
 		struct ucred *a_cred;
 	} */ *ap;
 {
-	struct vnode *vp = ap->a_vp;
-	struct uio *uio = ap->a_uio;
+	register struct vnode *vp = ap->a_vp;
+	register struct uio *uio = ap->a_uio;
 	struct proc *p = uio->uio_procp;
 	struct buf *bp;
 	daddr_t bn;
@@ -352,30 +343,36 @@ spec_write(ap)
 	struct partinfo dpart;
 	register int n, on;
 	int error = 0;
-	dev_t dev;
 
-	dev = vp->v_rdev;
 #ifdef DIAGNOSTIC
 	if (uio->uio_rw != UIO_WRITE)
 		panic("spec_write mode");
 	if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc)
 		panic("spec_write proc");
 #endif
-	if (uio->uio_resid == 0)
-		return (0);
 
 	switch (vp->v_type) {
 
+	case VCHR:
+		VOP_UNLOCK(vp, 0, p);
+		error = (*devsw(vp->v_rdev)->d_write)
+			(vp->v_rdev, uio, ap->a_ioflag);
+		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
+		return (error);
+
 	case VBLK:
-if (bdev_access) {
+		if (uio->uio_resid == 0)
+			return (0);
 		if (uio->uio_offset < 0)
 			return (EINVAL);
 
-		bsize = dev->si_bsize_phys;
-		if (bsize < BLKDEV_IOSIZE)
-			bsize = BLKDEV_IOSIZE;
+		/*
+		 * Calculate block size for block device.  The block size must
+		 * be larger then the physical minimum.
+		 */
+		bsize = vp->v_rdev->si_bsize_best;
 
-		if ((*devsw(dev)->d_ioctl)(dev, DIOCGPART,
+		if ((*devsw(vp->v_rdev)->d_ioctl)(vp->v_rdev, DIOCGPART,
 		    (caddr_t)&dpart, FREAD, p) == 0) {
 			if (dpart.part->p_fstype == FS_BSDFFS &&
 			    dpart.part->p_frag != 0 && dpart.part->p_fsize != 0)
@@ -403,14 +400,6 @@ if (bdev_access) {
 				bdwrite(bp);
 		} while (error == 0 && uio->uio_resid > 0 && n != 0);
 		return (error);
-}
-
-	case VCHR:
-		VOP_UNLOCK(vp, 0, p);
-		error = (*devsw(dev)->d_write)
-			(dev, uio, ap->a_ioflag);
-		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
-		return (error);
 
 	default:
 		panic("spec_write type");
diff --git a/sys/sys/conf.h b/sys/sys/conf.h
index 7232781..ddf4c68 100644
--- a/sys/sys/conf.h
+++ b/sys/sys/conf.h
@@ -69,6 +69,7 @@ struct specinfo {
 			struct disk *__sid_disk;
 			struct mount *__sid_mountpoint;
 			int __sid_bsize_phys; /* min physical block size */
+			int __sid_bsize_best; /* optimal block size */
 			int __sid_bsize_max;  /* maximum block size */
 		} __si_disk;
 	} __si_u;
@@ -78,6 +79,7 @@ struct specinfo {
 #define si_disk		__si_u.__si_disk.__sid_disk
 #define si_mountpoint	__si_u.__si_disk.__sid_mountpoint
 #define si_bsize_phys	__si_u.__si_disk.__sid_bsize_phys
+#define si_bsize_best	__si_u.__si_disk.__sid_bsize_best
 #define si_bsize_max	__si_u.__si_disk.__sid_bsize_max
 
 /*
diff --git a/sys/sys/linedisc.h b/sys/sys/linedisc.h
index 7232781..ddf4c68 100644
--- a/sys/sys/linedisc.h
+++ b/sys/sys/linedisc.h
@@ -69,6 +69,7 @@ struct specinfo {
 			struct disk *__sid_disk;
 			struct mount *__sid_mountpoint;
 			int __sid_bsize_phys; /* min physical block size */
+			int __sid_bsize_best; /* optimal block size */
 			int __sid_bsize_max;  /* maximum block size */
 		} __si_disk;
 	} __si_u;
@@ -78,6 +79,7 @@ struct specinfo {
 #define si_disk		__si_u.__si_disk.__sid_disk
 #define si_mountpoint	__si_u.__si_disk.__sid_mountpoint
 #define si_bsize_phys	__si_u.__si_disk.__sid_bsize_phys
+#define si_bsize_best	__si_u.__si_disk.__sid_bsize_best
 #define si_bsize_max	__si_u.__si_disk.__sid_bsize_max
 
 /*
diff --git a/sys/ufs/mfs/mfs_vfsops.c b/sys/ufs/mfs/mfs_vfsops.c
index 4f8c9f6..044bf63 100644
--- a/sys/ufs/mfs/mfs_vfsops.c
+++ b/sys/ufs/mfs/mfs_vfsops.c
@@ -333,6 +333,7 @@ mfs_mount(mp, path, data, ndp, p)
 	devvp->v_type = VBLK;
 	dev = make_dev(&mfs_cdevsw, mfs_minor, 0, 0, 0, "MFS%d", mfs_minor);
 	dev->si_bsize_phys = DEV_BSIZE;
+	dev->si_bsize_best = BLKDEV_IOSIZE;
 	dev->si_bsize_max = MAXBSIZE;
 	addaliasu(devvp, makeudev(253, mfs_minor++));
 	devvp->v_data = mfsp;
@@ -494,6 +495,7 @@ mfs_init(vfsp)
 		rootdev = make_dev(&mfs_cdevsw, mfs_minor, 
 		    0, 0, 0, "MFS%d", mfs_minor);
 		rootdev->si_bsize_phys = DEV_BSIZE;
+		rootdev->si_bsize_best = BLKDEV_IOSIZE;
 		rootdev->si_bsize_max = MAXBSIZE;
 		mfs_minor++;
 	} else if (bootverbose)