Changes to centralise the default blocksize behaviour.

More likely to follow.

Submitted by: phk@freebsd.org

20 files changed, 64 insertions, 63 deletions

diff --git a/sys/dev/ccd/ccd.c b/sys/dev/ccd/ccd.c
index 7badacb..a889ca3 100644
--- a/sys/dev/ccd/ccd.c
+++ b/sys/dev/ccd/ccd.c
@@ -633,10 +633,6 @@ ccdopen(dev, flags, fmt, p)
 	part = ccdpart(dev);
 	pmask = (1 << part);
 
-	dev->si_bsize_phys = DEV_BSIZE;
-	dev->si_bsize_best = BLKDEV_IOSIZE;
-	dev->si_bsize_max = MAXBSIZE;
-
 	/*
 	 * If we're initialized, check to see if there are any other
 	 * open partitions.  If not, then it's safe to update
diff --git a/sys/dev/fdc/fdc.c b/sys/dev/fdc/fdc.c
index ee74217..7e6b0b3 100644
--- a/sys/dev/fdc/fdc.c
+++ b/sys/dev/fdc/fdc.c
@@ -1279,9 +1279,6 @@ Fdopen(dev_t dev, int flags, int mode, struct proc *p)
 	fd_p	fd;
 	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)
 		return (ENXIO);
diff --git a/sys/dev/mcd/mcd.c b/sys/dev/mcd/mcd.c
index 37bd508..19f35e6 100644
--- a/sys/dev/mcd/mcd.c
+++ b/sys/dev/mcd/mcd.c
@@ -296,8 +296,6 @@ int mcdopen(dev_t dev, int flags, int fmt, struct proc *p)
 	if (!(cd->flags & MCDVALID) && cd->openflags)
 		return ENXIO;
 
-	dev->si_bsize_phys = 2048;
-	dev->si_bsize_max = MAXBSIZE;
 	if (mcd_getstat(unit,1) == -1)
 		return EIO;
 
@@ -352,6 +350,8 @@ int mcdopen(dev_t dev, int flags, int fmt, struct proc *p)
 MCD_TRACE("open: partition=%d, disksize = %ld, blksize=%d\n",
 	part, cd->disksize, cd->blksize);
 
+	dev->si_bsize_phys = cd->blksize;
+
 	if (part == RAW_PART ||
 		(part < cd->dlabel.d_npartitions &&
 		cd->dlabel.d_partitions[part].p_fstype != FS_UNUSED)) {
diff --git a/sys/dev/scd/scd.c b/sys/dev/scd/scd.c
index 1742fdd..719c86b 100644
--- a/sys/dev/scd/scd.c
+++ b/sys/dev/scd/scd.c
@@ -258,8 +258,6 @@ scdopen(dev_t dev, int flags, int fmt, struct proc *p)
 
 	XDEBUG(1,("scd%d: DEBUG: status = 0x%x\n", unit, inb(cd->iobase+IREG_STATUS)));
 
-	dev->si_bsize_phys = 2048;
-	dev->si_bsize_max = MAXBSIZE;
 	if ((rc = spin_up(unit)) != 0) {
 		print_error(unit, rc);
 		return EIO;
@@ -281,6 +279,8 @@ scdopen(dev_t dev, int flags, int fmt, struct proc *p)
 		}
 	}
 
+	dev->si_bsize_phys = cd->blksize;
+
 	cd->openflag = 1;
 	cd->flags |= SCDVALID;
 
diff --git a/sys/dev/vinum/vinum.c b/sys/dev/vinum/vinum.c
index 2482185..c70b1c7 100644
--- a/sys/dev/vinum/vinum.c
+++ b/sys/dev/vinum/vinum.c
@@ -268,9 +268,6 @@ vinumopen(dev_t dev,
     int devminor;					    /* minor number */
 
     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 */
     switch (DEVTYPE(dev)) {
diff --git a/sys/dev/vn/vn.c b/sys/dev/vn/vn.c
index f51efca..199c804 100644
--- a/sys/dev/vn/vn.c
+++ b/sys/dev/vn/vn.c
@@ -510,10 +510,7 @@ vniocattach_file(vn, vio, dev, flag, p)
 		(void) vn_close(nd.ni_vp, flags, p->p_ucred, p);
 		return(error);
 	}
-	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;
+	dev->si_bsize_phys = 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 9d85bfe..7b2d712 100644
--- a/sys/fs/specfs/spec_vnops.c
+++ b/sys/fs/specfs/spec_vnops.c
@@ -234,8 +234,6 @@ spec_open(ap)
 	if (vn_isdisk(vp)) {
 		if (!dev->si_bsize_phys)
 			dev->si_bsize_phys = DEV_BSIZE;
-		if (dev->si_bsize_best < dev->si_bsize_phys)
-			dev->si_bsize_best = BLKDEV_IOSIZE;
 		if (!dev->si_bsize_max)
 			dev->si_bsize_max = MAXBSIZE;
 	}
@@ -296,6 +294,10 @@ spec_read(ap)
 		 */
 
 		bsize = vp->v_rdev->si_bsize_best;
+		if (bsize < vp->v_rdev->si_bsize_phys)
+			bsize = vp->v_rdev->si_bsize_phys;
+		if (bsize < BLKDEV_IOSIZE)
+			bsize = BLKDEV_IOSIZE;
 
 		if ((ioctl = devsw(dev)->d_ioctl) != NULL &&
 		    (*ioctl)(dev, DIOCGPART, (caddr_t)&dpart, FREAD, p) == 0 &&
@@ -380,6 +382,10 @@ spec_write(ap)
 		 * be larger then the physical minimum.
 		 */
 		bsize = vp->v_rdev->si_bsize_best;
+		if (bsize < vp->v_rdev->si_bsize_phys)
+			bsize = vp->v_rdev->si_bsize_phys;
+		if (bsize < BLKDEV_IOSIZE)
+			bsize = BLKDEV_IOSIZE;
 
 		if ((*devsw(vp->v_rdev)->d_ioctl)(vp->v_rdev, DIOCGPART,
 		    (caddr_t)&dpart, FREAD, p) == 0) {
diff --git a/sys/geom/geom_ccd.c b/sys/geom/geom_ccd.c
index 7badacb..a889ca3 100644
--- a/sys/geom/geom_ccd.c
+++ b/sys/geom/geom_ccd.c
@@ -633,10 +633,6 @@ ccdopen(dev, flags, fmt, p)
 	part = ccdpart(dev);
 	pmask = (1 << part);
 
-	dev->si_bsize_phys = DEV_BSIZE;
-	dev->si_bsize_best = BLKDEV_IOSIZE;
-	dev->si_bsize_max = MAXBSIZE;
-
 	/*
 	 * If we're initialized, check to see if there are any other
 	 * open partitions.  If not, then it's safe to update
diff --git a/sys/i386/isa/atapi-cd.c b/sys/i386/isa/atapi-cd.c
index a8adbab..678603f 100644
--- a/sys/i386/isa/atapi-cd.c
+++ b/sys/i386/isa/atapi-cd.c
@@ -399,8 +399,6 @@ acdopen(dev_t dev, int flags, int fmt, struct proc *p)
         return ENXIO;
     cdp = acdtab[lun];
 
-    dev->si_bsize_phys = 2048;
-    dev->si_bsize_max = MAXBSIZE;
     if (!(cdp->flags & F_BOPEN) && !cdp->refcnt) {
         /* Prevent user eject */
         acd_request_wait(cdp, ATAPI_PREVENT_ALLOW,
@@ -411,7 +409,7 @@ acdopen(dev_t dev, int flags, int fmt, struct proc *p)
         cdp->flags |= F_BOPEN;
     else
         ++cdp->refcnt;
-
+    dev->si_bsize_phys = cdp->block_size;
     if (!(flags & O_NONBLOCK) && acd_read_toc(cdp) && !(flags & FWRITE))
         printf("acd%d: read_toc failed\n", lun);
     return 0;
diff --git a/sys/i386/isa/matcd/matcd.c b/sys/i386/isa/matcd/matcd.c
index 78e7ce0..bef362d 100644
--- a/sys/i386/isa/matcd/matcd.c
+++ b/sys/i386/isa/matcd/matcd.c
@@ -623,9 +623,6 @@ int	matcdopen(dev_t dev, int flags, int fmt,
 
 	if (ldrive >= TOTALDRIVES) return(ENXIO);
 
-	dev->si_bsize_phys = 2048;
-	dev->si_bsize_max = MAXBSIZE;
-
 #ifdef DEBUGOPEN
 	printf("matcd%d: Open: dev %x partition %x controller %x flags %x cdrive %x\n",
 	       ldrive,(int)dev,partition,controller,cd->flags,
@@ -696,7 +693,6 @@ int	matcdopen(dev_t dev, int flags, int fmt,
 	if ((cd->flags & MATCDLABEL)==0) {
 		bzero(&cd->dlabel,sizeof(struct disklabel));
 
-
 /*	Now we query the drive for the actual size of the media.
 	This is where we find out of there is any media or if the
 	media isn't a Mode 1 or Mode 2/XA disc.
@@ -749,6 +745,8 @@ int	matcdopen(dev_t dev, int flags, int fmt,
 		cd->flags |= MATCDLABEL;	/*Mark drive as having TOC*/
 	}
 
+	dev->si_bsize_phys = cd->blksize;
+
 #ifdef DEBUGOPEN
 	printf("matcd%d open2: partition=%d disksize=%d blksize=%x flags=%x\n",
 	       ldrive,partition,(int)cd->disksize,cd->blksize,cd->flags);
diff --git a/sys/i386/isa/mcd.c b/sys/i386/isa/mcd.c
index 37bd508..19f35e6 100644
--- a/sys/i386/isa/mcd.c
+++ b/sys/i386/isa/mcd.c
@@ -296,8 +296,6 @@ int mcdopen(dev_t dev, int flags, int fmt, struct proc *p)
 	if (!(cd->flags & MCDVALID) && cd->openflags)
 		return ENXIO;
 
-	dev->si_bsize_phys = 2048;
-	dev->si_bsize_max = MAXBSIZE;
 	if (mcd_getstat(unit,1) == -1)
 		return EIO;
 
@@ -352,6 +350,8 @@ int mcdopen(dev_t dev, int flags, int fmt, struct proc *p)
 MCD_TRACE("open: partition=%d, disksize = %ld, blksize=%d\n",
 	part, cd->disksize, cd->blksize);
 
+	dev->si_bsize_phys = cd->blksize;
+
 	if (part == RAW_PART ||
 		(part < cd->dlabel.d_npartitions &&
 		cd->dlabel.d_partitions[part].p_fstype != FS_UNUSED)) {
diff --git a/sys/i386/isa/scd.c b/sys/i386/isa/scd.c
index 1742fdd..719c86b 100644
--- a/sys/i386/isa/scd.c
+++ b/sys/i386/isa/scd.c
@@ -258,8 +258,6 @@ scdopen(dev_t dev, int flags, int fmt, struct proc *p)
 
 	XDEBUG(1,("scd%d: DEBUG: status = 0x%x\n", unit, inb(cd->iobase+IREG_STATUS)));
 
-	dev->si_bsize_phys = 2048;
-	dev->si_bsize_max = MAXBSIZE;
 	if ((rc = spin_up(unit)) != 0) {
 		print_error(unit, rc);
 		return EIO;
@@ -281,6 +279,8 @@ scdopen(dev_t dev, int flags, int fmt, struct proc *p)
 		}
 	}
 
+	dev->si_bsize_phys = cd->blksize;
+
 	cd->openflag = 1;
 	cd->flags |= SCDVALID;
 
diff --git a/sys/i386/isa/wd_cd.c b/sys/i386/isa/wd_cd.c
index a8adbab..678603f 100644
--- a/sys/i386/isa/wd_cd.c
+++ b/sys/i386/isa/wd_cd.c
@@ -399,8 +399,6 @@ acdopen(dev_t dev, int flags, int fmt, struct proc *p)
         return ENXIO;
     cdp = acdtab[lun];
 
-    dev->si_bsize_phys = 2048;
-    dev->si_bsize_max = MAXBSIZE;
     if (!(cdp->flags & F_BOPEN) && !cdp->refcnt) {
         /* Prevent user eject */
         acd_request_wait(cdp, ATAPI_PREVENT_ALLOW,
@@ -411,7 +409,7 @@ acdopen(dev_t dev, int flags, int fmt, struct proc *p)
         cdp->flags |= F_BOPEN;
     else
         ++cdp->refcnt;
-
+    dev->si_bsize_phys = cdp->block_size;
     if (!(flags & O_NONBLOCK) && acd_read_toc(cdp) && !(flags & FWRITE))
         printf("acd%d: read_toc failed\n", lun);
     return 0;
diff --git a/sys/isa/fd.c b/sys/isa/fd.c
index ee74217..7e6b0b3 100644
--- a/sys/isa/fd.c
+++ b/sys/isa/fd.c
@@ -1279,9 +1279,6 @@ Fdopen(dev_t dev, int flags, int mode, struct proc *p)
 	fd_p	fd;
 	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)
 		return (ENXIO);
diff --git a/sys/kern/subr_disk.c b/sys/kern/subr_disk.c
index dbf9c14..cbdb2af 100644
--- a/sys/kern/subr_disk.c
+++ b/sys/kern/subr_disk.c
@@ -13,6 +13,7 @@
 #include <sys/param.h>
 #include <sys/systm.h>
 #include <sys/kernel.h>
+#include <sys/sysctl.h>
 #include <sys/buf.h>
 #include <sys/conf.h>
 #include <sys/disk.h>
@@ -235,3 +236,14 @@ diskpsize(dev_t dev)
 	}
 	return (dssize(dev, &dp->d_slice));
 }
+
+SYSCTL_DECL(_debug_sizeof);
+
+SYSCTL_INT(_debug_sizeof, OID_AUTO, disklabel, CTLFLAG_RD, 
+    0, sizeof(struct disklabel), "sizeof(struct disklabel)");
+
+SYSCTL_INT(_debug_sizeof, OID_AUTO, diskslices, CTLFLAG_RD, 
+    0, sizeof(struct diskslices), "sizeof(struct diskslices)");
+
+SYSCTL_INT(_debug_sizeof, OID_AUTO, disk, CTLFLAG_RD, 
+    0, sizeof(struct disk), "sizeof(struct disk)");
diff --git a/sys/kern/subr_diskslice.c b/sys/kern/subr_diskslice.c
index 49e86f6..6a31e11 100644
--- a/sys/kern/subr_diskslice.c
+++ b/sys/kern/subr_diskslice.c
@@ -718,8 +718,6 @@ 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);
 	if (lp->d_secsize % DEV_BSIZE) {
diff --git a/sys/kern/vfs_vnops.c b/sys/kern/vfs_vnops.c
index 0a08938..c42ebf6 100644
--- a/sys/kern/vfs_vnops.c
+++ b/sys/kern/vfs_vnops.c
@@ -429,22 +429,29 @@ vn_stat(vp, sb, p)
 	sb->st_atimespec = vap->va_atime;
 	sb->st_mtimespec = vap->va_mtime;
 	sb->st_ctimespec = vap->va_ctime;
-	/*
-	 * 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
+
+        /*
+	 * 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.
 	 */
-	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:
+
+	if (vap->va_type == VREG) {
 		sb->st_blksize = vap->va_blocksize;
-		break;
+	} else if ((vp->v_type == VBLK || vp->v_type == VCHR) &&
+	    devsw(vp->v_rdev) && (devsw(vp->v_rdev)->d_flags & D_DISK)) {
+		/* XXX use vn_isdisk() above once VCHR is also disk */
+		sb->st_blksize = vp->v_rdev->si_bsize_best;
+		if (sb->st_blksize < vp->v_rdev->si_bsize_phys)
+			sb->st_blksize = vp->v_rdev->si_bsize_phys;
+		if (sb->st_blksize < BLKDEV_IOSIZE)
+			sb->st_blksize = BLKDEV_IOSIZE;
+	} else {
+		sb->st_blksize = 0;
 	}
+	
 	sb->st_flags = vap->va_flags;
 	if (suser_xxx(p->p_ucred, 0, 0))
 		sb->st_gen = 0;
diff --git a/sys/miscfs/specfs/spec_vnops.c b/sys/miscfs/specfs/spec_vnops.c
index 9d85bfe..7b2d712 100644
--- a/sys/miscfs/specfs/spec_vnops.c
+++ b/sys/miscfs/specfs/spec_vnops.c
@@ -234,8 +234,6 @@ spec_open(ap)
 	if (vn_isdisk(vp)) {
 		if (!dev->si_bsize_phys)
 			dev->si_bsize_phys = DEV_BSIZE;
-		if (dev->si_bsize_best < dev->si_bsize_phys)
-			dev->si_bsize_best = BLKDEV_IOSIZE;
 		if (!dev->si_bsize_max)
 			dev->si_bsize_max = MAXBSIZE;
 	}
@@ -296,6 +294,10 @@ spec_read(ap)
 		 */
 
 		bsize = vp->v_rdev->si_bsize_best;
+		if (bsize < vp->v_rdev->si_bsize_phys)
+			bsize = vp->v_rdev->si_bsize_phys;
+		if (bsize < BLKDEV_IOSIZE)
+			bsize = BLKDEV_IOSIZE;
 
 		if ((ioctl = devsw(dev)->d_ioctl) != NULL &&
 		    (*ioctl)(dev, DIOCGPART, (caddr_t)&dpart, FREAD, p) == 0 &&
@@ -380,6 +382,10 @@ spec_write(ap)
 		 * be larger then the physical minimum.
 		 */
 		bsize = vp->v_rdev->si_bsize_best;
+		if (bsize < vp->v_rdev->si_bsize_phys)
+			bsize = vp->v_rdev->si_bsize_phys;
+		if (bsize < BLKDEV_IOSIZE)
+			bsize = BLKDEV_IOSIZE;
 
 		if ((*devsw(vp->v_rdev)->d_ioctl)(vp->v_rdev, DIOCGPART,
 		    (caddr_t)&dpart, FREAD, p) == 0) {
diff --git a/sys/pc98/pc98/wd_cd.c b/sys/pc98/pc98/wd_cd.c
index a8adbab..678603f 100644
--- a/sys/pc98/pc98/wd_cd.c
+++ b/sys/pc98/pc98/wd_cd.c
@@ -399,8 +399,6 @@ acdopen(dev_t dev, int flags, int fmt, struct proc *p)
         return ENXIO;
     cdp = acdtab[lun];
 
-    dev->si_bsize_phys = 2048;
-    dev->si_bsize_max = MAXBSIZE;
     if (!(cdp->flags & F_BOPEN) && !cdp->refcnt) {
         /* Prevent user eject */
         acd_request_wait(cdp, ATAPI_PREVENT_ALLOW,
@@ -411,7 +409,7 @@ acdopen(dev_t dev, int flags, int fmt, struct proc *p)
         cdp->flags |= F_BOPEN;
     else
         ++cdp->refcnt;
-
+    dev->si_bsize_phys = cdp->block_size;
     if (!(flags & O_NONBLOCK) && acd_read_toc(cdp) && !(flags & FWRITE))
         printf("acd%d: read_toc failed\n", lun);
     return 0;
diff --git a/sys/ufs/mfs/mfs_vfsops.c b/sys/ufs/mfs/mfs_vfsops.c
index 044bf63..1ac1dc8 100644
--- a/sys/ufs/mfs/mfs_vfsops.c
+++ b/sys/ufs/mfs/mfs_vfsops.c
@@ -332,8 +332,8 @@ 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);
+	/* It is not clear that these will get initialized otherwise */
 	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;
@@ -494,8 +494,8 @@ mfs_init(vfsp)
 		mountrootfsname = "mfs";
 		rootdev = make_dev(&mfs_cdevsw, mfs_minor, 
 		    0, 0, 0, "MFS%d", mfs_minor);
+		/* It is not clear that these will get initialized otherwise */
 		rootdev->si_bsize_phys = DEV_BSIZE;
-		rootdev->si_bsize_best = BLKDEV_IOSIZE;
 		rootdev->si_bsize_max = MAXBSIZE;
 		mfs_minor++;
 	} else if (bootverbose)