A new implementation of the loader block cache

The block cache implementation in loader has proven to be almost useless, and in worst case even slowing down the disk reads due to insufficient cache size and extra memory copy.
Also the current cache implementation does not cache reads from CDs, or work with zfs built on top of multiple disks.
Instead of an LRU, this code uses a simple hash (O(1) read from cache), and instead of a single global cache, a separate cache per block device.
The cache also implements limited read-ahead to increase performance.
To simplify read ahead management, the read ahead will not wrap over bcache end, so in worst case, single block physical read will be performed to fill the last block in bcache.

Booting from a virtual CD over IPMI:
0ms latency, before: 27 second, after: 7 seconds
60ms latency, before: over 12 minutes, after: under 5 minutes.

Submitted by:	Toomas Soome <tsoome@me.com>
Reviewed by:	delphij (previous version), emaste (previous version)
Relnotes:	yes
Differential Revision:	https://reviews.freebsd.org/D4713

38 files changed, 728 insertions, 382 deletions

diff --git a/lib/libstand/cd9660.c b/lib/libstand/cd9660.c
index a730a8e..5d9b57b 100644
--- a/lib/libstand/cd9660.c
+++ b/lib/libstand/cd9660.c
@@ -143,7 +143,7 @@ susp_lookup_record(struct open_file *f, const char *identifier,
 		if (bcmp(sh->type, SUSP_CONTINUATION, 2) == 0) {
 			shc = (ISO_RRIP_CONT *)sh;
 			error = f->f_dev->dv_strategy(f->f_devdata, F_READ,
-			    cdb2devb(isonum_733(shc->location)),
+			    cdb2devb(isonum_733(shc->location)), 0,
 			    ISO_DEFAULT_BLOCK_SIZE, susp_buffer, &read);
 
 			/* Bail if it fails. */
@@ -288,7 +288,7 @@ cd9660_open(const char *path, struct open_file *f)
 	for (bno = 16;; bno++) {
 		twiddle(1);
 		rc = f->f_dev->dv_strategy(f->f_devdata, F_READ, cdb2devb(bno),
-					   ISO_DEFAULT_BLOCK_SIZE, buf, &read);
+					0, ISO_DEFAULT_BLOCK_SIZE, buf, &read);
 		if (rc)
 			goto out;
 		if (read != ISO_DEFAULT_BLOCK_SIZE) {
@@ -322,7 +322,7 @@ cd9660_open(const char *path, struct open_file *f)
 				twiddle(1);
 				rc = f->f_dev->dv_strategy
 					(f->f_devdata, F_READ,
-					 cdb2devb(bno + boff),
+					 cdb2devb(bno + boff), 0,
 					 ISO_DEFAULT_BLOCK_SIZE,
 					 buf, &read);
 				if (rc)
@@ -381,7 +381,7 @@ cd9660_open(const char *path, struct open_file *f)
 		bno = isonum_733(rec.extent) + isonum_711(rec.ext_attr_length);
 		twiddle(1);
 		rc = f->f_dev->dv_strategy(f->f_devdata, F_READ, cdb2devb(bno),
-		    ISO_DEFAULT_BLOCK_SIZE, buf, &read);
+		    0, ISO_DEFAULT_BLOCK_SIZE, buf, &read);
 		if (rc)
 			goto out;
 		if (read != ISO_DEFAULT_BLOCK_SIZE) {
@@ -438,7 +438,8 @@ buf_read_file(struct open_file *f, char **buf_p, size_t *size_p)
 
 		twiddle(16);
 		rc = f->f_dev->dv_strategy(f->f_devdata, F_READ,
-		    cdb2devb(blkno), ISO_DEFAULT_BLOCK_SIZE, fp->f_buf, &read);
+		    cdb2devb(blkno), 0, ISO_DEFAULT_BLOCK_SIZE,
+		    fp->f_buf, &read);
 		if (rc)
 			return (rc);
 		if (read != ISO_DEFAULT_BLOCK_SIZE)
diff --git a/lib/libstand/dosfs.c b/lib/libstand/dosfs.c
index b75d752..0ab5b97 100644
--- a/lib/libstand/dosfs.c
+++ b/lib/libstand/dosfs.c
@@ -131,7 +131,18 @@ static DOS_DE dot[2] = {
 #define stclus(sz, de)  ((sz) != 32 ? cv2((de)->clus) :          \
                          ((u_int)cv2((de)->dex.h_clus) << 16) |  \
 			 cv2((de)->clus))
-    
+
+/*
+ * fat cache metadata
+ */
+struct fatcache {
+	int unit;	/* disk unit number */
+	int size;	/* buffer (and fat) size in sectors */
+	u_char *buf;
+};
+
+static struct fatcache fat;
+
 static int dosunmount(DOS_FS *);
 static int parsebs(DOS_FS *, DOS_BS *);
 static int namede(DOS_FS *, const char *, DOS_DE **);
@@ -143,8 +154,36 @@ static int fatcnt(DOS_FS *, u_int);
 static int fatget(DOS_FS *, u_int *);
 static int fatend(u_int, u_int);
 static int ioread(DOS_FS *, u_int, void *, u_int);
-static int iobuf(DOS_FS *, u_int);
-static int ioget(struct open_file *, u_int, void *, u_int);
+static int ioget(struct open_file *, daddr_t, size_t, void *, u_int);
+
+static void
+dos_read_fat(DOS_FS *fs, struct open_file *fd)
+{
+    struct devdesc *dd = fd->f_devdata;
+
+    if (fat.buf != NULL) {		/* can we reuse old buffer? */
+	if (fat.size != fs->spf) {
+	    free(fat.buf);		/* no, free old buffer */
+	    fat.buf = NULL;
+	}
+    }
+
+    if (fat.buf == NULL)
+	fat.buf = malloc(secbyt(fs->spf));
+
+    if (fat.buf != NULL) {
+	if (ioget(fd, fs->lsnfat, 0, fat.buf, secbyt(fs->spf)) == 0) {
+	    fat.size = fs->spf;
+	    fat.unit = dd->d_unit;
+	    return;
+	}
+    }
+    if (fat.buf != NULL)	/* got IO error */
+	free(fat.buf);
+    fat.buf = NULL;
+    fat.unit = -1;	/* impossible unit */
+    fat.size = 0;
+}
 
 /*
  * Mount DOS filesystem
@@ -153,15 +192,25 @@ static int
 dos_mount(DOS_FS *fs, struct open_file *fd)
 {
     int err;
+    struct devdesc *dd = fd->f_devdata;
+    u_char *buf;
 
     bzero(fs, sizeof(DOS_FS));
     fs->fd = fd;
-    if ((err = !(fs->buf = malloc(SECSIZ)) ? errno : 0) ||
-        (err = ioget(fs->fd, 0, fs->buf, 1)) ||
-        (err = parsebs(fs, (DOS_BS *)fs->buf))) {
+
+    if ((err = !(buf = malloc(secbyt(1))) ? errno : 0) ||
+        (err = ioget(fs->fd, 0, 0, buf, secbyt(1))) ||
+        (err = parsebs(fs, (DOS_BS *)buf))) {
+	if (buf != NULL)
+	    free(buf);
         (void)dosunmount(fs);
         return(err);
     }
+    free(buf);
+
+    if (fat.buf == NULL || fat.unit != dd->d_unit)
+	dos_read_fat(fs, fd);
+
     fs->root = dot[0];
     fs->root.name[0] = ' ';
     if (fs->fatsz == 32) {
@@ -194,8 +243,6 @@ dos_unmount(DOS_FS *fs)
 static int
 dosunmount(DOS_FS *fs)
 {
-    if (fs->buf)
-        free(fs->buf);
     free(fs);
     return(0);
 }
@@ -252,42 +299,47 @@ dos_read(struct open_file *fd, void *buf, size_t nbyte, size_t *resid)
     DOS_FILE *f = (DOS_FILE *)fd->f_fsdata;
     int err = 0;
 
+    /*
+     * as ioget() can be called *a lot*, use twiddle here.
+     * also 4 seems to be good value not to slow loading down too much:
+     * with 270MB file (~540k ioget() calls, twiddle can easily waste 4-5sec.
+     */
+    twiddle(4);
     nb = (u_int)nbyte;
     if ((size = fsize(f->fs, &f->de)) == -1)
 	return EINVAL;
     if (nb > (n = size - f->offset))
-        nb = n;
+	nb = n;
     off = f->offset;
     if ((clus = stclus(f->fs->fatsz, &f->de)))
-        off &= f->fs->bsize - 1;
+	off &= f->fs->bsize - 1;
     c = f->c;
     cnt = nb;
     while (cnt) {
-        n = 0;
-        if (!c) {
-            if ((c = clus))
-                n = bytblk(f->fs, f->offset);
-        } else if (!off)
-            n++;
-        while (n--) {
-            if ((err = fatget(f->fs, &c)))
+	n = 0;
+	if (!c) {
+	    if ((c = clus))
+		n = bytblk(f->fs, f->offset);
+	} else if (!off)
+	    n++;
+	while (n--) {
+	    if ((err = fatget(f->fs, &c)))
 		goto out;
-            if (!okclus(f->fs, c)) {
+	    if (!okclus(f->fs, c)) {
 		err = EINVAL;
 		goto out;
 	    }
-        }
-        if (!clus || (n = f->fs->bsize - off) > cnt)
-            n = cnt;
-        if ((err = ioread(f->fs, (c ? blkoff(f->fs, c) :
-				      secbyt(f->fs->lsndir)) + off,
-			  buf, n)))
+	}
+	if (!clus || (n = f->fs->bsize - off) > cnt)
+	    n = cnt;
+	if ((err = ioread(f->fs, (c ? blkoff(f->fs, c) :
+				      secbyt(f->fs->lsndir)) + off, buf, n)))
 	    goto out;
-        f->offset += n;
-        f->c = c;
-        off = 0;
-        buf = (char *)buf + n;
-        cnt -= n;
+	f->offset += n;
+	f->c = c;
+	off = 0;
+	buf = (char *)buf + n;
+	cnt -= n;
     }
  out:
     if (resid)
@@ -364,6 +416,23 @@ dos_stat(struct open_file *fd, struct stat *sb)
 }
 
 static int
+dos_checksum(char *name, char *ext)
+{
+    int x, i;
+    char buf[11];
+
+    bcopy(name, buf, 8);
+    bcopy(ext, buf+8, 3);
+    x = 0;
+    for (i = 0; i < 11; i++) {
+	x = ((x & 1) << 7) | (x >> 1);
+	x += buf[i];
+	x &= 0xff;
+    }
+    return (x);
+}
+
+static int
 dos_readdir(struct open_file *fd, struct dirent *d)
 {
     /* DOS_FILE *f = (DOS_FILE *)fd->f_fsdata; */
@@ -417,12 +486,7 @@ dos_readdir(struct open_file *fd, struct dirent *d)
 	    }
 	} else {
 	    if (xdn == 1) {
-		x = 0;
-		for (i = 0; i < 11; i++) {
-		    x = ((x & 1) << 7) | (x >> 1);
-		    x += dd.de.name[i];
-		    x &= 0xff;
-		}
+		x = dos_checksum(dd.de.name, dd.de.ext);
 		if (x == chk)
 		    break;
 	    } else {
@@ -555,7 +619,7 @@ lookup(DOS_FS *fs, u_int clus, const char *name, DOS_DE **dep)
         else
             return EINVAL;
         for (sec = 0; sec < nsec; sec++) {
-            if ((err = ioget(fs->fd, lsec + sec, dir, 1)))
+            if ((err = ioget(fs->fd, lsec + sec, 0, dir, secbyt(1))))
                 return err;
             for (ent = 0; ent < DEPSEC; ent++) {
                 if (!*dir[ent].de.name)
@@ -577,9 +641,7 @@ lookup(DOS_FS *fs, u_int clus, const char *name, DOS_DE **dep)
                         }
                     } else if (!(dir[ent].de.attr & FA_LABEL)) {
                         if ((ok = xdn == 1)) {
-                            for (x = 0, i = 0; i < 11; i++)
-                                x = ((((x & 1) << 7) | (x >> 1)) +
-                                     dir[ent].de.name[i]) & 0xff;
+			    x = dos_checksum(dir[ent].de.name, dir[ent].de.ext);
                             ok = chk == x &&
                                 !strcasecmp(name, (const char *)lfn);
                         }
@@ -699,22 +761,52 @@ fatcnt(DOS_FS *fs, u_int c)
 }
 
 /*
- * Get next cluster in cluster chain
+ * Get next cluster in cluster chain. Use in core fat cache unless another
+ * device replaced it.
  */
 static int
 fatget(DOS_FS *fs, u_int *c)
 {
     u_char buf[4];
-    u_int x;
-    int err;
+    u_char *s;
+    u_int x, offset, off, n, nbyte, lsec;
+    struct devdesc *dd = fs->fd->f_devdata;
+    int err = 0;
+
+    if (fat.unit != dd->d_unit) {
+	/* fat cache was changed to another device, dont use it */
+	err = ioread(fs, secbyt(fs->lsnfat) + fatoff(fs->fatsz, *c), buf,
+	    fs->fatsz != 32 ? 2 : 4);
+	if (err)
+	    return err;
+    } else {
+	offset = fatoff(fs->fatsz, *c);
+	nbyte = fs->fatsz != 32 ? 2 : 4;
+
+	s = buf;
+	if ((off = offset & (SECSIZ - 1))) {
+	    offset -= off;
+	    lsec = bytsec(offset);
+	    offset += SECSIZ;
+	    if ((n = SECSIZ - off) > nbyte)
+		n = nbyte;
+	    memcpy(s, fat.buf + secbyt(lsec) + off, n);
+	    s += n;
+	    nbyte -= n;
+	}
+	n = nbyte & (SECSIZ - 1);
+	if (nbyte -= n) {
+	    memcpy(s, fat.buf + secbyt(bytsec(offset)), nbyte);
+	    offset += nbyte;
+	    s += nbyte;
+	}
+	if (n)
+	    memcpy(s, fat.buf + secbyt(bytsec(offset)), n);
+    }
 
-    err = ioread(fs, secbyt(fs->lsnfat) + fatoff(fs->fatsz, *c), buf,
-                 fs->fatsz != 32 ? 2 : 4);
-    if (err)
-        return err;
     x = fs->fatsz != 32 ? cv2(buf) : cv4(buf);
     *c = fs->fatsz == 12 ? *c & 1 ? x >> 4 : x & 0xfff : x;
-    return 0;
+    return (0);
 }
 
 /*
@@ -739,42 +831,24 @@ ioread(DOS_FS *fs, u_int offset, void *buf, u_int nbyte)
     s = buf;
     if ((off = offset & (SECSIZ - 1))) {
         offset -= off;
-        if ((err = iobuf(fs, bytsec(offset))))
-            return err;
-        offset += SECSIZ;
         if ((n = SECSIZ - off) > nbyte)
             n = nbyte;
-        memcpy(s, fs->buf + off, n);
+        if ((err = ioget(fs->fd, bytsec(offset), off, s, n)))
+            return err;
+        offset += SECSIZ;
         s += n;
         nbyte -= n;
     }
     n = nbyte & (SECSIZ - 1);
     if (nbyte -= n) {
-        if ((err = ioget(fs->fd, bytsec(offset), s, bytsec(nbyte))))
+        if ((err = ioget(fs->fd, bytsec(offset), 0, s, nbyte)))
             return err;
         offset += nbyte;
         s += nbyte;
     }
     if (n) {
-        if ((err = iobuf(fs, bytsec(offset))))
-            return err;
-        memcpy(s, fs->buf, n);
-    }
-    return 0;
-}
-
-/*
- * Buffered sector-based I/O primitive
- */
-static int
-iobuf(DOS_FS *fs, u_int lsec)
-{
-    int err;
-
-    if (fs->bufsec != lsec) {
-        if ((err = ioget(fs->fd, lsec, fs->buf, 1)))
+        if ((err = ioget(fs->fd, bytsec(offset), 0, s, n)))
             return err;
-        fs->bufsec = lsec;
     }
     return 0;
 }
@@ -783,13 +857,8 @@ iobuf(DOS_FS *fs, u_int lsec)
  * Sector-based I/O primitive
  */
 static int
-ioget(struct open_file *fd, u_int lsec, void *buf, u_int nsec)
+ioget(struct open_file *fd, daddr_t lsec, size_t offset, void *buf, u_int size)
 {
-    int	err;
-
-    twiddle(1);
-    if ((err = (fd->f_dev->dv_strategy)(fd->f_devdata, F_READ, lsec, 
-					secbyt(nsec), buf, NULL)))
-	return(err);
-    return(0);
+    return ((fd->f_dev->dv_strategy)(fd->f_devdata, F_READ, lsec, offset,
+	size, buf, NULL));
 }
diff --git a/lib/libstand/dosfs.h b/lib/libstand/dosfs.h
index 9e5744d..f2370ee 100644
--- a/lib/libstand/dosfs.h
+++ b/lib/libstand/dosfs.h
@@ -96,8 +96,6 @@ typedef union {
 
 typedef struct {
     struct open_file *fd;       /* file descriptor */
-    u_char *buf;                /* buffer */
-    u_int bufsec;               /* buffered sector */
     u_int links;                /* active links to structure */
     u_int spc;                  /* sectors per cluster */
     u_int bsize;                /* cluster size in bytes */
diff --git a/lib/libstand/ext2fs.c b/lib/libstand/ext2fs.c
index d0b91e0..bbc3be4 100644
--- a/lib/libstand/ext2fs.c
+++ b/lib/libstand/ext2fs.c
@@ -355,7 +355,7 @@ ext2fs_open(const char *upath, struct open_file *f)
 	fp->f_fs = fs;
 	twiddle(1);
 	error = (f->f_dev->dv_strategy)(f->f_devdata, F_READ,
-	    EXT2_SBLOCK, EXT2_SBSIZE, (char *)fs, &buf_size);
+	    EXT2_SBLOCK, 0, EXT2_SBSIZE, (char *)fs, &buf_size);
 	if (error)
 		goto out;
 
@@ -397,7 +397,7 @@ ext2fs_open(const char *upath, struct open_file *f)
 	fp->f_bg = malloc(len);
 	twiddle(1);
 	error = (f->f_dev->dv_strategy)(f->f_devdata, F_READ,
-	    EXT2_SBLOCK + EXT2_SBSIZE / DEV_BSIZE, len,
+	    EXT2_SBLOCK + EXT2_SBSIZE / DEV_BSIZE, 0, len,
 	    (char *)fp->f_bg, &buf_size);
 	if (error)
 		goto out;
@@ -509,7 +509,7 @@ ext2fs_open(const char *upath, struct open_file *f)
 				
 				twiddle(1);
 				error = (f->f_dev->dv_strategy)(f->f_devdata,
-				    F_READ, fsb_to_db(fs, disk_block),
+				    F_READ, fsb_to_db(fs, disk_block), 0,
 				    fs->fs_bsize, buf, &buf_size);
 				if (error)
 					goto out;
@@ -570,7 +570,7 @@ read_inode(ino_t inumber, struct open_file *f)
 	buf = malloc(fs->fs_bsize);
 	twiddle(1);
 	error = (f->f_dev->dv_strategy)(f->f_devdata, F_READ,
-	    ino_to_db(fs, fp->f_bg, inumber), fs->fs_bsize, buf, &rsize);
+	    ino_to_db(fs, fp->f_bg, inumber), 0, fs->fs_bsize, buf, &rsize);
 	if (error)
 		goto out;
 	if (rsize != fs->fs_bsize) {
@@ -667,7 +667,7 @@ block_map(struct open_file *f, daddr_t file_block, daddr_t *disk_block_p)
 					malloc(fs->fs_bsize);
 			twiddle(1);
 			error = (f->f_dev->dv_strategy)(f->f_devdata, F_READ,
-			    fsb_to_db(fp->f_fs, ind_block_num), fs->fs_bsize,
+			    fsb_to_db(fp->f_fs, ind_block_num), 0, fs->fs_bsize,
 			    fp->f_blk[level], &fp->f_blksize[level]);
 			if (error)
 				return (error);
@@ -725,7 +725,7 @@ buf_read_file(struct open_file *f, char **buf_p, size_t *size_p)
 		} else {
 			twiddle(4);
 			error = (f->f_dev->dv_strategy)(f->f_devdata, F_READ,
-			    fsb_to_db(fs, disk_block), block_size,
+			    fsb_to_db(fs, disk_block), 0, block_size,
 			    fp->f_buf, &fp->f_buf_size);
 			if (error)
 				goto done;
diff --git a/lib/libstand/read.c b/lib/libstand/read.c
index a984dbe..0e47fc2 100644
--- a/lib/libstand/read.c
+++ b/lib/libstand/read.c
@@ -79,7 +79,7 @@ read(int fd, void *dest, size_t bcount)
     if (f->f_flags & F_RAW) {
 	twiddle(4);
 	errno = (f->f_dev->dv_strategy)(f->f_devdata, F_READ,
-					btodb(f->f_offset), bcount, dest, &resid);
+				btodb(f->f_offset), 0, bcount, dest, &resid);
 	if (errno)
 	    return (-1);
 	f->f_offset += resid;
diff --git a/lib/libstand/stand.h b/lib/libstand/stand.h
index a841523..238f9ed 100644
--- a/lib/libstand/stand.h
+++ b/lib/libstand/stand.h
@@ -138,8 +138,8 @@ struct devsw {
     const char	dv_name[8];
     int		dv_type;		/* opaque type constant, arch-dependant */
     int		(*dv_init)(void);	/* early probe call */
-    int		(*dv_strategy)(void *devdata, int rw, daddr_t blk, size_t size,
-			       char *buf, size_t *rsize);
+    int		(*dv_strategy)(void *devdata, int rw, daddr_t blk,
+			size_t offset, size_t size, char *buf, size_t *rsize);
     int		(*dv_open)(struct open_file *f, ...);
     int		(*dv_close)(struct open_file *f);
     int		(*dv_ioctl)(struct open_file *f, u_long cmd, void *data);
@@ -154,6 +154,24 @@ extern struct devsw netdev;
 
 extern int errno;
 
+/*
+ * Generic device specifier; architecture-dependent
+ * versions may be larger, but should be allowed to
+ * overlap.
+ */
+struct devdesc
+{
+    struct devsw	*d_dev;
+    int			d_type;
+#define DEVT_NONE	0
+#define DEVT_DISK	1
+#define DEVT_NET	2
+#define DEVT_CD		3
+#define DEVT_ZFS	4
+    int			d_unit;
+    void		*d_opendata;
+};
+
 struct open_file {
     int			f_flags;	/* see F_* below */
     struct devsw	*f_dev;		/* pointer to device operations */
diff --git a/lib/libstand/ufs.c b/lib/libstand/ufs.c
index 928a1d1..21e341f 100644
--- a/lib/libstand/ufs.c
+++ b/lib/libstand/ufs.c
@@ -157,7 +157,7 @@ read_inode(inumber, f)
 	buf = malloc(fs->fs_bsize);
 	twiddle(1);
 	rc = (f->f_dev->dv_strategy)(f->f_devdata, F_READ,
-		fsbtodb(fs, ino_to_fsba(fs, inumber)), fs->fs_bsize,
+		fsbtodb(fs, ino_to_fsba(fs, inumber)), 0, fs->fs_bsize,
 		buf, &rsize);
 	if (rc)
 		goto out;
@@ -267,7 +267,7 @@ block_map(f, file_block, disk_block_p)
 					malloc(fs->fs_bsize);
 			twiddle(1);
 			rc = (f->f_dev->dv_strategy)(f->f_devdata, F_READ,
-				fsbtodb(fp->f_fs, ind_block_num),
+				fsbtodb(fp->f_fs, ind_block_num), 0,
 				fs->fs_bsize,
 				fp->f_blk[level],
 				&fp->f_blksize[level]);
@@ -348,7 +348,7 @@ buf_write_file(f, buf_p, size_p)
 
 		twiddle(4);
 		rc = (f->f_dev->dv_strategy)(f->f_devdata, F_READ,
-			fsbtodb(fs, disk_block),
+			fsbtodb(fs, disk_block), 0,
 			block_size, fp->f_buf, &fp->f_buf_size);
 		if (rc)
 			return (rc);
@@ -367,7 +367,7 @@ buf_write_file(f, buf_p, size_p)
 
 	twiddle(4);
 	rc = (f->f_dev->dv_strategy)(f->f_devdata, F_WRITE,
-		fsbtodb(fs, disk_block),
+		fsbtodb(fs, disk_block), 0,
 		block_size, fp->f_buf, &fp->f_buf_size);
 	return (rc);
 }
@@ -408,7 +408,7 @@ buf_read_file(f, buf_p, size_p)
 		} else {
 			twiddle(4);
 			rc = (f->f_dev->dv_strategy)(f->f_devdata, F_READ,
-				fsbtodb(fs, disk_block),
+				fsbtodb(fs, disk_block), 0,
 				block_size, fp->f_buf, &fp->f_buf_size);
 			if (rc)
 				return (rc);
@@ -521,7 +521,7 @@ ufs_open(upath, f)
 	 */
 	for (i = 0; sblock_try[i] != -1; i++) {
 		rc = (f->f_dev->dv_strategy)(f->f_devdata, F_READ,
-		    sblock_try[i] / DEV_BSIZE, SBLOCKSIZE,
+		    sblock_try[i] / DEV_BSIZE, 0, SBLOCKSIZE,
 		    (char *)fs, &buf_size);
 		if (rc)
 			goto out;
@@ -651,7 +651,7 @@ ufs_open(upath, f)
 				
 				twiddle(1);
 				rc = (f->f_dev->dv_strategy)(f->f_devdata,
-					F_READ, fsbtodb(fs, disk_block),
+					F_READ, fsbtodb(fs, disk_block), 0,
 					fs->fs_bsize, buf, &buf_size);
 				if (rc)
 					goto out;
diff --git a/lib/libstand/write.c b/lib/libstand/write.c
index 9e02f08..daf33cf 100644
--- a/lib/libstand/write.c
+++ b/lib/libstand/write.c
@@ -82,7 +82,7 @@ write(fd, dest, bcount)
 	if (f->f_flags & F_RAW) {
 		twiddle(4);
 		errno = (f->f_dev->dv_strategy)(f->f_devdata, F_WRITE,
-			btodb(f->f_offset), bcount, dest, &resid);
+			btodb(f->f_offset), 0, bcount, dest, &resid);
 		if (errno)
 			return (-1);
 		f->f_offset += resid;
diff --git a/sys/boot/common/bcache.c b/sys/boot/common/bcache.c
index c88adca..e5cf75b 100644
--- a/sys/boot/common/bcache.c
+++ b/sys/boot/common/bcache.c
@@ -1,5 +1,6 @@
 /*-
  * Copyright (c) 1998 Michael Smith <msmith@freebsd.org>
+ * Copyright 2015 Toomas Soome <tsoome@me.com>
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -25,99 +26,155 @@
  */
 
 #include <sys/cdefs.h>
+#include <sys/param.h>
 __FBSDID("$FreeBSD$");
 
 /*
- * Simple LRU block cache
+ * Simple hashed block cache
  */
 
 #include <sys/stdint.h>
 
 #include <stand.h>
 #include <string.h>
-#include <bitstring.h>
+#include <strings.h>
 
 #include "bootstrap.h"
 
 /* #define BCACHE_DEBUG */
 
 #ifdef BCACHE_DEBUG
-#define BCACHE_TIMEOUT	10
 # define DEBUG(fmt, args...)	printf("%s: " fmt "\n" , __func__ , ## args)
 #else
-#define BCACHE_TIMEOUT	2
 # define DEBUG(fmt, args...)
 #endif
 
-
 struct bcachectl
 {
     daddr_t	bc_blkno;
-    time_t	bc_stamp;
     int		bc_count;
 };
 
-static struct bcachectl	*bcache_ctl;
-static caddr_t		bcache_data;
-static bitstr_t		*bcache_miss;
-static u_int		bcache_nblks;
-static u_int		bcache_blksize;
-static u_int		bcache_hits, bcache_misses, bcache_ops, bcache_bypasses;
-static u_int		bcache_flushes;
-static u_int		bcache_bcount;
+/*
+ * bcache per device node. cache is allocated on device first open and freed
+ * on last close, to save memory. The issue there is the size; biosdisk
+ * supports up to 31 (0x1f) devices. Classic setup would use single disk
+ * to boot from, but this has changed with zfs.
+ */
+struct bcache {
+    struct bcachectl	*bcache_ctl;
+    caddr_t		bcache_data;
+    u_int		bcache_nblks;
+    size_t		ra;
+};
 
-static void	bcache_invalidate(daddr_t blkno);
-static void	bcache_insert(caddr_t buf, daddr_t blkno);
-static int	bcache_lookup(caddr_t buf, daddr_t blkno);
+static u_int bcache_total_nblks;	/* set by bcache_init */
+static u_int bcache_blksize;		/* set by bcache_init */
+static u_int bcache_numdev;		/* set by bcache_add_dev */
+/* statistics */
+static u_int bcache_units;	/* number of devices with cache */
+static u_int bcache_unit_nblks;	/* nblocks per unit */
+static u_int bcache_hits;
+static u_int bcache_misses;
+static u_int bcache_ops;
+static u_int bcache_bypasses;
+static u_int bcache_bcount;
+static u_int bcache_rablks;
+
+#define	BHASH(bc, blkno)	((blkno) & ((bc)->bcache_nblks - 1))
+#define	BCACHE_LOOKUP(bc, blkno)	\
+	((bc)->bcache_ctl[BHASH((bc), (blkno))].bc_blkno != (blkno))
+#define	BCACHE_READAHEAD	256
+#define	BCACHE_MINREADAHEAD	32
+
+static void	bcache_invalidate(struct bcache *bc, daddr_t blkno);
+static void	bcache_insert(struct bcache *bc, daddr_t blkno);
+static void	bcache_free_instance(struct bcache *bc);
 
 /*
  * Initialise the cache for (nblks) of (bsize).
  */
-int
+void
 bcache_init(u_int nblks, size_t bsize)
 {
-    /* discard any old contents */
-    if (bcache_data != NULL) {
-	free(bcache_data);
-	bcache_data = NULL;
-	free(bcache_ctl);
-    }
-
-    /* Allocate control structures */
-    bcache_nblks = nblks;
+    /* set up control data */
+    bcache_total_nblks = nblks;
     bcache_blksize = bsize;
-    bcache_data = malloc(bcache_nblks * bcache_blksize);
-    bcache_ctl = (struct bcachectl *)malloc(bcache_nblks * sizeof(struct bcachectl));
-    bcache_miss = bit_alloc((bcache_nblks + 1) / 2);
-    if ((bcache_data == NULL) || (bcache_ctl == NULL) || (bcache_miss == NULL)) {
-	if (bcache_miss)
-	    free(bcache_miss);
-	if (bcache_ctl)
-	    free(bcache_ctl);
-	if (bcache_data)
-	    free(bcache_data);
-	bcache_data = NULL;
-	return(ENOMEM);
-    }
-
-    return(0);
 }
 
 /*
- * Flush the cache
+ * add number of devices to bcache. we have to divide cache space
+ * between the devices, so bcache_add_dev() can be used to set up the
+ * number. The issue is, we need to get the number before actual allocations.
+ * bcache_add_dev() is supposed to be called from device init() call, so the
+ * assumption is, devsw dv_init is called for plain devices first, and
+ * for zfs, last.
  */
 void
-bcache_flush(void)
+bcache_add_dev(int devices)
 {
-    u_int	i;
+    bcache_numdev += devices;
+}
+
+void *
+bcache_allocate(void)
+{
+    u_int i;
+    struct bcache *bc = malloc(sizeof (struct bcache));
+    int disks = bcache_numdev;
+
+    if (disks == 0)
+	disks = 1;	/* safe guard */
+
+    if (bc == NULL) {
+	errno = ENOMEM;
+	return (bc);
+    }
+
+    /*
+     * the bcache block count must be power of 2 for hash function
+     */
+    i = fls(disks) - 1;		/* highbit - 1 */
+    if (disks > (1 << i))	/* next power of 2 */
+	i++;
+
+    bc->bcache_nblks = bcache_total_nblks >> i;
+    bcache_unit_nblks = bc->bcache_nblks;
+    bc->bcache_data = malloc(bc->bcache_nblks * bcache_blksize);
+    if (bc->bcache_data == NULL) {
+	/* dont error out yet. fall back to 32 blocks and try again */
+	bc->bcache_nblks = 32;
+	bc->bcache_data = malloc(bc->bcache_nblks * bcache_blksize);
+    }
+
+    bc->bcache_ctl = malloc(bc->bcache_nblks * sizeof(struct bcachectl));
 
-    bcache_flushes++;
+    if ((bc->bcache_data == NULL) || (bc->bcache_ctl == NULL)) {
+	bcache_free_instance(bc);
+	errno = ENOMEM;
+	return(NULL);
+    }
 
     /* Flush the cache */
-    for (i = 0; i < bcache_nblks; i++) {
-	bcache_ctl[i].bc_count = -1;
-	bcache_ctl[i].bc_blkno = -1;
+    for (i = 0; i < bc->bcache_nblks; i++) {
+	bc->bcache_ctl[i].bc_count = -1;
+	bc->bcache_ctl[i].bc_blkno = -1;
     }
+    bcache_units++;
+    bc->ra = BCACHE_READAHEAD;	/* optimistic read ahead */
+    return (bc);
+}
+
+void
+bcache_free(void *cache)
+{
+    struct bcache *bc = cache;
+
+    if (bc == NULL)
+	return;
+
+    bcache_free_instance(bc);
+    bcache_units--;
 }
 
 /*
@@ -125,31 +182,22 @@ bcache_flush(void)
  * cache with the new values.
  */
 static int
-write_strategy(void *devdata, int unit, int rw, daddr_t blk, size_t size,
-		char *buf, size_t *rsize)
+write_strategy(void *devdata, int rw, daddr_t blk, size_t offset,
+    size_t size, char *buf, size_t *rsize)
 {
     struct bcache_devdata	*dd = (struct bcache_devdata *)devdata;
+    struct bcache		*bc = dd->dv_cache;
     daddr_t			i, nblk;
-    int				err;
 
     nblk = size / bcache_blksize;
 
     /* Invalidate the blocks being written */
     for (i = 0; i < nblk; i++) {
-	bcache_invalidate(blk + i);
+	bcache_invalidate(bc, blk + i);
     }
 
     /* Write the blocks */
-    err = dd->dv_strategy(dd->dv_devdata, rw, blk, size, buf, rsize);
-
-    /* Populate the block cache with the new data */
-    if (err == 0) {
-    	for (i = 0; i < nblk; i++) {
-	    bcache_insert(buf + (i * bcache_blksize),blk + i);
-    	}
-    }
-
-    return err;
+    return (dd->dv_strategy(dd->dv_devdata, rw, blk, offset, size, buf, rsize));
 }
 
 /*
@@ -158,61 +206,87 @@ write_strategy(void *devdata, int unit, int rw, daddr_t blk, size_t size,
  * device I/O and then use the I/O results to populate the cache. 
  */
 static int
-read_strategy(void *devdata, int unit, int rw, daddr_t blk, size_t size,
-		char *buf, size_t *rsize)
+read_strategy(void *devdata, int rw, daddr_t blk, size_t offset,
+    size_t size, char *buf, size_t *rsize)
 {
     struct bcache_devdata	*dd = (struct bcache_devdata *)devdata;
-    int				p_size, result;
-    daddr_t			p_blk, i, j, nblk;
+    struct bcache		*bc = dd->dv_cache;
+    size_t			i, nblk, p_size, r_size, complete, ra;
+    int				result;
+    daddr_t			p_blk;
     caddr_t			p_buf;
 
+    if (bc == NULL) {
+	errno = ENODEV;
+	return (-1);
+    }
+
+    if (rsize != NULL)
+	*rsize = 0;
+
     nblk = size / bcache_blksize;
+    if ((nblk == 0 && size != 0) || offset != 0)
+	nblk++;
     result = 0;
+    complete = 1;
 
-    /* Satisfy any cache hits up front */
+    /* Satisfy any cache hits up front, break on first miss */
     for (i = 0; i < nblk; i++) {
-	if (bcache_lookup(buf + (bcache_blksize * i), blk + i)) {
-	    bit_set(bcache_miss, i);	/* cache miss */
-	    bcache_misses++;
+	if (BCACHE_LOOKUP(bc, (daddr_t)(blk + i))) {
+	    bcache_misses += (nblk - i);
+	    complete = 0;
+	    if (nblk - i > BCACHE_MINREADAHEAD && bc->ra > BCACHE_MINREADAHEAD)
+		bc->ra >>= 1;	/* reduce read ahead */
+	    break;
 	} else {
-	    bit_clear(bcache_miss, i);	/* cache hit */
 	    bcache_hits++;
 	}
     }
 
-    /* Go back and fill in any misses  XXX optimise */
-    p_blk = -1;
-    p_buf = NULL;
-    p_size = 0;
-    for (i = 0; i < nblk; i++) {
-	if (bit_test(bcache_miss, i)) {
-	    /* miss, add to pending transfer */
-	    if (p_blk == -1) {
-		p_blk = blk + i;
-		p_buf = buf + (bcache_blksize * i);
-		p_size = 1;
-	    } else {
-		p_size++;
-	    }
-	} else if (p_blk != -1) {
-	    /* hit, complete pending transfer */
-	    result = dd->dv_strategy(dd->dv_devdata, rw, p_blk, p_size * bcache_blksize, p_buf, NULL);
-	    if (result != 0)
-		goto done;
-	    for (j = 0; j < p_size; j++)
-		bcache_insert(p_buf + (j * bcache_blksize), p_blk + j);
-	    p_blk = -1;
-	}
+   if (complete) {	/* whole set was in cache, return it */
+	if (bc->ra < BCACHE_READAHEAD)
+		bc->ra <<= 1;	/* increase read ahead */
+	bcopy(bc->bcache_data + (bcache_blksize * BHASH(bc, blk)) + offset,
+	    buf, size);
+	goto done;
+   }
+
+    /*
+     * Fill in any misses. From check we have i pointing to first missing
+     * block, read in all remaining blocks + readahead.
+     * We have space at least for nblk - i before bcache wraps.
+     */
+    p_blk = blk + i;
+    p_buf = bc->bcache_data + (bcache_blksize * BHASH(bc, p_blk));
+    r_size = bc->bcache_nblks - BHASH(bc, p_blk); /* remaining blocks */
+
+    p_size = MIN(r_size, nblk - i);	/* read at least those blocks */
+
+    ra = bc->bcache_nblks - BHASH(bc, p_blk + p_size);
+    if (ra != bc->bcache_nblks) { /* do we have RA space? */
+	ra = MIN(bc->ra, ra);
+	p_size += ra;
     }
-    if (p_blk != -1) {
-	/* pending transfer left */
-	result = dd->dv_strategy(dd->dv_devdata, rw, p_blk, p_size * bcache_blksize, p_buf, NULL);
-	if (result != 0)
-	    goto done;
-	for (j = 0; j < p_size; j++)
-	    bcache_insert(p_buf + (j * bcache_blksize), p_blk + j);
+
+    /* invalidate bcache */
+    for (i = 0; i < p_size; i++) {
+	bcache_invalidate(bc, p_blk + i);
     }
-    
+    r_size = 0;
+    result = dd->dv_strategy(dd->dv_devdata, rw, p_blk, 0,
+	p_size * bcache_blksize, p_buf, &r_size);
+
+    if (result)
+	goto done;
+
+    r_size /= bcache_blksize;
+    for (i = 0; i < r_size; i++)
+	bcache_insert(bc, p_blk + i);
+
+    bcache_rablks += ra;
+    bcopy(bc->bcache_data + (bcache_blksize * BHASH(bc, blk)) + offset, buf,
+	size);
+
  done:
     if ((result == 0) && (rsize != NULL))
 	*rsize = size;
@@ -220,130 +294,144 @@ read_strategy(void *devdata, int unit, int rw, daddr_t blk, size_t size,
 }
 
 /* 
- * Requests larger than 1/2 the cache size will be bypassed and go
+ * Requests larger than 1/2 cache size will be bypassed and go
  * directly to the disk.  XXX tune this.
  */
 int
-bcache_strategy(void *devdata, int unit, int rw, daddr_t blk, size_t size,
-		char *buf, size_t *rsize)
+bcache_strategy(void *devdata, int rw, daddr_t blk, size_t offset,
+    size_t size, char *buf, size_t *rsize)
 {
-    static int			bcache_unit = -1;
     struct bcache_devdata	*dd = (struct bcache_devdata *)devdata;
+    struct bcache		*bc = dd->dv_cache;
+    u_int bcache_nblks = 0;
+    int nblk, cblk, ret;
+    size_t csize, isize, total;
 
     bcache_ops++;
 
-    if(bcache_unit != unit) {
-	bcache_flush();
-	bcache_unit = unit;
-    }
+    if (bc != NULL)
+	bcache_nblks = bc->bcache_nblks;
 
     /* bypass large requests, or when the cache is inactive */
-    if ((bcache_data == NULL) || ((size * 2 / bcache_blksize) > bcache_nblks)) {
+    if (bc == NULL ||
+	(offset == 0 && ((size * 2 / bcache_blksize) > bcache_nblks))) {
 	DEBUG("bypass %d from %d", size / bcache_blksize, blk);
 	bcache_bypasses++;
-	return(dd->dv_strategy(dd->dv_devdata, rw, blk, size, buf, rsize));
+	return (dd->dv_strategy(dd->dv_devdata, rw, blk, offset, size, buf,
+	    rsize));
+    }
+
+    /* normalize offset */
+    while (offset >= bcache_blksize) {
+	blk++;
+	offset -= bcache_blksize;
     }
 
     switch (rw) {
     case F_READ:
-	return read_strategy(devdata, unit, rw, blk, size, buf, rsize);
+	nblk = size / bcache_blksize;
+	if (offset || (size != 0 && nblk == 0))
+	    nblk++;	/* read at least one block */
+
+	ret = 0;
+	total = 0;
+	while(size) {
+	    cblk = bcache_nblks - BHASH(bc, blk); /* # of blocks left */
+	    cblk = MIN(cblk, nblk);
+
+	    if (size <= bcache_blksize)
+		csize = size;
+	    else {
+		csize = cblk * bcache_blksize;
+		if (offset)
+		    csize -= (bcache_blksize - offset);
+	    }
+
+	    ret = read_strategy(devdata, rw, blk, offset,
+		csize, buf+total, &isize);
+	    if (ret != 0)
+		return (ret);
+	    blk += (offset+isize) / bcache_blksize;
+	    offset = 0;
+	    total += isize;
+	    size -= isize;
+	    nblk = size / bcache_blksize;
+	}
+
+	if (rsize)
+	    *rsize = total;
+
+	return (ret);
     case F_WRITE:
-	return write_strategy(devdata, unit, rw, blk, size, buf, rsize);
+	return write_strategy(devdata, rw, blk, offset, size, buf, rsize);
     }
     return -1;
 }
 
-
 /*
- * Insert a block into the cache.  Retire the oldest block to do so, if required.
- *
- * XXX the LRU algorithm will fail after 2^31 blocks have been transferred.
+ * Free allocated bcache instance
  */
 static void
-bcache_insert(caddr_t buf, daddr_t blkno) 
+bcache_free_instance(struct bcache *bc)
 {
-    time_t	now;
-    int		cand, ocount;
-    u_int	i;
-    
-    time(&now);
-    cand = 0;				/* assume the first block */
-    ocount = bcache_ctl[0].bc_count;
-
-    /* find the oldest block */
-    for (i = 1; i < bcache_nblks; i++) {
-	if (bcache_ctl[i].bc_blkno == blkno) {
-	    /* reuse old entry */
-	    cand = i;
-	    break;
-	}
-	if (bcache_ctl[i].bc_count < ocount) {
-	    ocount = bcache_ctl[i].bc_count;
-	    cand = i;
-	}
+    if (bc != NULL) {
+	if (bc->bcache_ctl)
+	    free(bc->bcache_ctl);
+	if (bc->bcache_data)
+	    free(bc->bcache_data);
+	free(bc);
     }
-    
-    DEBUG("insert blk %d -> %d @ %d # %d", blkno, cand, now, bcache_bcount);
-    bcopy(buf, bcache_data + (bcache_blksize * cand), bcache_blksize);
-    bcache_ctl[cand].bc_blkno = blkno;
-    bcache_ctl[cand].bc_stamp = now;
-    bcache_ctl[cand].bc_count = bcache_bcount++;
 }
 
 /*
- * Look for a block in the cache.  Blocks more than BCACHE_TIMEOUT seconds old
- * may be stale (removable media) and thus are discarded.  Copy the block out 
- * if successful and return zero, or return nonzero on failure.
+ * Insert a block into the cache.
  */
-static int
-bcache_lookup(caddr_t buf, daddr_t blkno)
+static void
+bcache_insert(struct bcache *bc, daddr_t blkno)
 {
-    time_t	now;
-    u_int	i;
+    u_int	cand;
     
-    time(&now);
-
-    for (i = 0; i < bcache_nblks; i++)
-	/* cache hit? */
-	if ((bcache_ctl[i].bc_blkno == blkno) && ((bcache_ctl[i].bc_stamp + BCACHE_TIMEOUT) >= now)) {
-	    bcopy(bcache_data + (bcache_blksize * i), buf, bcache_blksize);
-	    DEBUG("hit blk %d <- %d (now %d then %d)", blkno, i, now, bcache_ctl[i].bc_stamp);
-	    return(0);
-	}
-    return(ENOENT);
+    cand = BHASH(bc, blkno);
+
+    DEBUG("insert blk %llu -> %u # %d", blkno, cand, bcache_bcount);
+    bc->bcache_ctl[cand].bc_blkno = blkno;
+    bc->bcache_ctl[cand].bc_count = bcache_bcount++;
 }
 
 /*
  * Invalidate a block from the cache.
  */
 static void
-bcache_invalidate(daddr_t blkno)
+bcache_invalidate(struct bcache *bc, daddr_t blkno)
 {
     u_int	i;
     
-    for (i = 0; i < bcache_nblks; i++) {
-	if (bcache_ctl[i].bc_blkno == blkno) {
-	    bcache_ctl[i].bc_count = -1;
-	    bcache_ctl[i].bc_blkno = -1;
-	    DEBUG("invalidate blk %d", blkno);
-	    break;
-	}
+    i = BHASH(bc, blkno);
+    if (bc->bcache_ctl[i].bc_blkno == blkno) {
+	bc->bcache_ctl[i].bc_count = -1;
+	bc->bcache_ctl[i].bc_blkno = -1;
+	DEBUG("invalidate blk %llu", blkno);
     }
 }
 
+#ifndef BOOT2
 COMMAND_SET(bcachestat, "bcachestat", "get disk block cache stats", command_bcache);
 
 static int
 command_bcache(int argc, char *argv[])
 {
-    u_int	i;
-    
-    for (i = 0; i < bcache_nblks; i++) {
-	printf("%08jx %04x %04x|", (uintmax_t)bcache_ctl[i].bc_blkno, (unsigned int)bcache_ctl[i].bc_stamp & 0xffff, bcache_ctl[i].bc_count & 0xffff);
-	if (((i + 1) % 4) == 0)
-	    printf("\n");
+    if (argc != 1) {
+	command_errmsg = "wrong number of arguments";
+	return(CMD_ERROR);
     }
-    printf("\n%d ops  %d bypasses  %d hits  %d misses  %d flushes\n", bcache_ops, bcache_bypasses, bcache_hits, bcache_misses, bcache_flushes);
+
+    printf("\ncache blocks: %d\n", bcache_total_nblks);
+    printf("cache blocksz: %d\n", bcache_blksize);
+    printf("cache readahead: %d\n", bcache_rablks);
+    printf("unit cache blocks: %d\n", bcache_unit_nblks);
+    printf("cached units: %d\n", bcache_units);
+    printf("%d ops  %d bypasses  %d hits  %d misses\n", bcache_ops,
+	bcache_bypasses, bcache_hits, bcache_misses);
     return(CMD_OK);
 }
-
+#endif
diff --git a/sys/boot/common/bootstrap.h b/sys/boot/common/bootstrap.h
index 7efc2c2..cbfc6f0 100644
--- a/sys/boot/common/bootstrap.h
+++ b/sys/boot/common/bootstrap.h
@@ -33,24 +33,6 @@
 #include <sys/queue.h>
 #include <sys/linker_set.h>
 
-/*
- * Generic device specifier; architecture-dependant 
- * versions may be larger, but should be allowed to
- * overlap.
- */
-struct devdesc 
-{
-    struct devsw	*d_dev;
-    int			d_type;
-#define DEVT_NONE	0
-#define DEVT_DISK	1
-#define DEVT_NET	2
-#define	DEVT_CD		3
-#define DEVT_ZFS	4
-    int			d_unit;
-    void		*d_opendata;
-};
-
 /* Commands and return values; nonzero return sets command_errmsg != NULL */
 typedef int	(bootblk_cmd_t)(int argc, char *argv[]);
 extern char	*command_errmsg;	
@@ -90,9 +72,11 @@ int	kern_pread(int fd, vm_offset_t dest, size_t len, off_t off);
 void	*alloc_pread(int fd, off_t off, size_t len);
 
 /* bcache.c */
-int	bcache_init(u_int nblks, size_t bsize);
-void	bcache_flush(void);
-int	bcache_strategy(void *devdata, int unit, int rw, daddr_t blk,
+void	bcache_init(u_int nblks, size_t bsize);
+void	bcache_add_dev(int);
+void	*bcache_allocate(void);
+void	bcache_free(void *);
+int	bcache_strategy(void *devdata, int rw, daddr_t blk, size_t offset,
 			size_t size, char *buf, size_t *rsize);
 
 /*
@@ -100,8 +84,10 @@ int	bcache_strategy(void *devdata, int unit, int rw, daddr_t blk,
  */
 struct bcache_devdata
 {
-    int         (*dv_strategy)(void *devdata, int rw, daddr_t blk, size_t size, char *buf, size_t *rsize);
+    int         (*dv_strategy)(void *devdata, int rw, daddr_t blk,
+			size_t offset, size_t size, char *buf, size_t *rsize);
     void	*dv_devdata;
+    void	*dv_cache;
 };
 
 /*
diff --git a/sys/boot/common/disk.c b/sys/boot/common/disk.c
index c862d30..804fb6b 100644
--- a/sys/boot/common/disk.c
+++ b/sys/boot/common/disk.c
@@ -178,7 +178,7 @@ ptblread(void *d, void *buf, size_t blocks, off_t offset)
 
 	dev = (struct disk_devdesc *)d;
 	od = (struct open_disk *)dev->d_opendata;
-	return (dev->d_dev->dv_strategy(dev, F_READ, offset,
+	return (dev->d_dev->dv_strategy(dev, F_READ, offset, 0,
 	    blocks * od->sectorsize, (char *)buf, NULL));
 }
 
@@ -239,7 +239,7 @@ disk_read(struct disk_devdesc *dev, void *buf, off_t offset, u_int blocks)
 	int ret;
 
 	od = (struct open_disk *)dev->d_opendata;
-	ret = dev->d_dev->dv_strategy(dev, F_READ, dev->d_offset + offset,
+	ret = dev->d_dev->dv_strategy(dev, F_READ, dev->d_offset + offset, 0,
 	    blocks * od->sectorsize, buf, NULL);
 
 	return (ret);
@@ -252,7 +252,7 @@ disk_write(struct disk_devdesc *dev, void *buf, off_t offset, u_int blocks)
 	int ret;
 
 	od = (struct open_disk *)dev->d_opendata;
-	ret = dev->d_dev->dv_strategy(dev, F_WRITE, dev->d_offset + offset,
+	ret = dev->d_dev->dv_strategy(dev, F_WRITE, dev->d_offset + offset, 0,
 	    blocks * od->sectorsize, buf, NULL);
 
 	return (ret);
diff --git a/sys/boot/common/md.c b/sys/boot/common/md.c
index a8f092b9..e5e8a48 100644
--- a/sys/boot/common/md.c
+++ b/sys/boot/common/md.c
@@ -60,7 +60,7 @@ static struct {
 
 /* devsw I/F */
 static int md_init(void);
-static int md_strategy(void *, int, daddr_t, size_t, char *, size_t *);
+static int md_strategy(void *, int, daddr_t, size_t, size_t, char *, size_t *);
 static int md_open(struct open_file *, ...);
 static int md_close(struct open_file *);
 static void md_print(int);
@@ -84,8 +84,8 @@ md_init(void)
 }
 
 static int
-md_strategy(void *devdata, int rw, daddr_t blk, size_t size, char *buf,
-    size_t *rsize)
+md_strategy(void *devdata, int rw, daddr_t blk, size_t offset, size_t size,
+    char *buf, size_t *rsize)
 {
 	struct devdesc *dev = (struct devdesc *)devdata;
 	size_t ofs;
diff --git a/sys/boot/common/module.c b/sys/boot/common/module.c
index 923da5b..76fdf2c 100644
--- a/sys/boot/common/module.c
+++ b/sys/boot/common/module.c
@@ -769,7 +769,7 @@ mod_search_hints(struct moduledir *mdp, const char *modname,
 	intp = (int*)recptr;
 	reclen = *intp++;
 	ival = *intp++;
-	cp = (char*)intp;
+	cp = (u_char*)intp;
 	switch (ival) {
 	case MDT_VERSION:
 	    clen = *cp++;
@@ -801,9 +801,9 @@ mod_search_hints(struct moduledir *mdp, const char *modname,
      * Finally check if KLD is in the place
      */
     if (found)
-	result = file_lookup(mdp->d_path, cp, clen, NULL);
+	result = file_lookup(mdp->d_path, (const char *)cp, clen, NULL);
     else if (best)
-	result = file_lookup(mdp->d_path, best, blen, NULL);
+	result = file_lookup(mdp->d_path, (const char *)best, blen, NULL);
 bad:
     /*
      * If nothing found or hints is absent - fallback to the old way
diff --git a/sys/boot/efi/libefi/efipart.c b/sys/boot/efi/libefi/efipart.c
index 757d64f..2cf009a 100644
--- a/sys/boot/efi/libefi/efipart.c
+++ b/sys/boot/efi/libefi/efipart.c
@@ -42,7 +42,10 @@ static EFI_GUID blkio_guid = BLOCK_IO_PROTOCOL;
 static EFI_GUID devpath_guid = DEVICE_PATH_PROTOCOL;
 
 static int efipart_init(void);
-static int efipart_strategy(void *, int, daddr_t, size_t, char *, size_t *);
+static int efipart_strategy(void *, int, daddr_t, size_t, size_t, char *,
+    size_t *);
+static int efipart_realstrategy(void *, int, daddr_t, size_t, size_t, char *,
+    size_t *);
 static int efipart_open(struct open_file *, ...);
 static int efipart_close(struct open_file *);
 static void efipart_print(int);
@@ -59,6 +62,21 @@ struct devsw efipart_dev = {
 	.dv_cleanup = NULL
 };
 
+/*
+ * info structure to support bcache
+ */
+#define	MAXPDDEV	31	/* see MAXDEV in libi386.h */
+
+static struct pdinfo
+{
+	int	pd_unit;	/* unit number */
+	int	pd_open;	/* reference counter */
+	void	*pd_bcache;	/* buffer cache data */
+} pdinfo [MAXPDDEV];
+static int npdinfo = 0;
+
+#define PD(dev)         (pdinfo[(dev)->d_unit])
+
 static int
 efipart_init(void) 
 {
@@ -140,8 +158,13 @@ efipart_init(void)
 		} else
 			hout[nout] = hin[n];
 		nout++;
+		pdinfo[npdinfo].pd_open = 0;
+		pdinfo[npdinfo].pd_bcache = NULL;
+		pdinfo[npdinfo].pd_unit = npdinfo;
+		npdinfo++;
 	}
 
+	bcache_add_dev(npdinfo);
 	err = efi_register_handles(&efipart_dev, hout, aliases, nout);
 	free(hin);
 	return (err);
@@ -173,7 +196,7 @@ efipart_print(int verbose)
 	}
 }
 
-static int 
+static int
 efipart_open(struct open_file *f, ...)
 {
 	va_list args;
@@ -198,10 +221,13 @@ efipart_open(struct open_file *f, ...)
 		return (EAGAIN);
 
 	dev->d_opendata = blkio;
+	PD(dev).pd_open++;
+	if (PD(dev).pd_bcache == NULL)
+		PD(dev).pd_bcache = bcache_allocate();
 	return (0);
 }
 
-static int 
+static int
 efipart_close(struct open_file *f)
 {
 	struct devdesc *dev;
@@ -211,6 +237,11 @@ efipart_close(struct open_file *f)
 		return (EINVAL);
 
 	dev->d_opendata = NULL;
+	PD(dev).pd_open--;
+	if (PD(dev).pd_open == 0) {
+		bcache_free(PD(dev).pd_bcache);
+		PD(dev).pd_bcache = NULL;
+	}
 	return (0);
 }
 
@@ -254,9 +285,24 @@ efipart_readwrite(EFI_BLOCK_IO *blkio, int rw, daddr_t blk, daddr_t nblks,
 	return (efi_status_to_errno(status));
 }
 
-static int 
-efipart_strategy(void *devdata, int rw, daddr_t blk, size_t size, char *buf,
-    size_t *rsize)
+static int
+efipart_strategy(void *devdata, int rw, daddr_t blk, size_t offset,
+    size_t size, char *buf, size_t *rsize)
+{
+	struct bcache_devdata bcd;
+	struct devdesc *dev;
+
+	dev = (struct devdesc *)devdata;
+	bcd.dv_strategy = efipart_realstrategy;
+	bcd.dv_devdata = devdata;
+	bcd.dv_cache = PD(dev).pd_bcache;
+	return (bcache_strategy(&bcd, rw, blk, offset, size,
+	    buf, rsize));
+}
+
+static int
+efipart_realstrategy(void *devdata, int rw, daddr_t blk, size_t offset,
+    size_t size, char *buf, size_t *rsize)
 {
 	struct devdesc *dev = (struct devdesc *)devdata;
 	EFI_BLOCK_IO *blkio;
diff --git a/sys/boot/efi/libefi/libefi.c b/sys/boot/efi/libefi/libefi.c
index c76bd6c..0807eb5 100644
--- a/sys/boot/efi/libefi/libefi.c
+++ b/sys/boot/efi/libefi/libefi.c
@@ -102,7 +102,7 @@ efi_main(EFI_HANDLE image_handle, EFI_SYSTEM_TABLE *system_table)
 		(void)console_control->SetMode(console_control,
 		    EfiConsoleControlScreenText);
 
-	heapsize = 3 * 1024 * 1024;
+	heapsize = 64 * 1024 * 1024;
 	status = BS->AllocatePages(AllocateAnyPages, EfiLoaderData,
 	    EFI_SIZE_TO_PAGES(heapsize), &heap);
 	if (status != EFI_SUCCESS)
diff --git a/sys/boot/efi/loader/main.c b/sys/boot/efi/loader/main.c
index 361d3bb..91b39b1 100644
--- a/sys/boot/efi/loader/main.c
+++ b/sys/boot/efi/loader/main.c
@@ -210,6 +210,11 @@ main(int argc, CHAR16 *argv[])
 	cons_probe();
 
 	/*
+	 * Initialise the block cache. Set the upper limit.
+	 */
+	bcache_init(32768, 512);
+
+	/*
 	 * Parse the args to set the console settings, etc
 	 * boot1.efi passes these in, if it can read /boot.config or /boot/config
 	 * or iPXE may be setup to pass these in.
diff --git a/sys/boot/i386/libfirewire/firewire.c b/sys/boot/i386/libfirewire/firewire.c
index 8191a63..2c7ee32 100644
--- a/sys/boot/i386/libfirewire/firewire.c
+++ b/sys/boot/i386/libfirewire/firewire.c
@@ -66,7 +66,7 @@ struct crom_src_buf {
 
 static int	fw_init(void);
 static int	fw_strategy(void *devdata, int flag, daddr_t dblk,
-		    size_t size, char *buf, size_t *rsize);
+		    size_t offset, size_t size, char *buf, size_t *rsize);
 static int	fw_open(struct open_file *f, ...);
 static int	fw_close(struct open_file *f);
 static void	fw_print(int verbose);
@@ -201,7 +201,8 @@ fw_cleanup()
 }
 
 static int 
-fw_strategy(void *devdata, int rw, daddr_t dblk, size_t size, char *buf, size_t *rsize)
+fw_strategy(void *devdata, int rw, daddr_t dblk, size_t offset, size_t size,
+    char *buf, size_t *rsize)
 {
 	return (EIO);
 }
diff --git a/sys/boot/i386/libi386/bioscd.c b/sys/boot/i386/libi386/bioscd.c
index 9babcdf..c387513 100644
--- a/sys/boot/i386/libi386/bioscd.c
+++ b/sys/boot/i386/libi386/bioscd.c
@@ -85,13 +85,19 @@ struct specification_packet {
 static struct bcinfo {
 	int	bc_unit;		/* BIOS unit number */
 	struct specification_packet bc_sp;
+	int	bc_open;		/* reference counter */
+	void	*bc_bcache;		/* buffer cache data */
 } bcinfo [MAXBCDEV];
 static int nbcinfo = 0;
 
+#define	BC(dev)	(bcinfo[(dev)->d_unit])
+
 static int	bc_read(int unit, daddr_t dblk, int blks, caddr_t dest);
 static int	bc_init(void);
 static int	bc_strategy(void *devdata, int flag, daddr_t dblk,
-		    size_t size, char *buf, size_t *rsize);
+    size_t offset, size_t size, char *buf, size_t *rsize);
+static int	bc_realstrategy(void *devdata, int flag, daddr_t dblk,
+    size_t offset, size_t size, char *buf, size_t *rsize);
 static int	bc_open(struct open_file *f, ...);
 static int	bc_close(struct open_file *f);
 static void	bc_print(int verbose);
@@ -164,6 +170,7 @@ bc_add(int biosdev)
 
 	printf("BIOS CD is cd%d\n", nbcinfo);
 	nbcinfo++;
+	bcache_add_dev(nbcinfo);	/* register cd device in bcache */
 	return(0);
 }
 
@@ -200,19 +207,44 @@ bc_open(struct open_file *f, ...)
 		return(ENXIO);
 	}
 
+	BC(dev).bc_open++;
+	if (BC(dev).bc_bcache == NULL)
+		BC(dev).bc_bcache = bcache_allocate();
 	return(0);
 }
  
 static int 
 bc_close(struct open_file *f)
 {
+	struct i386_devdesc *dev;
 
+	dev = (struct i386_devdesc *)f->f_devdata;
+	BC(dev).bc_open--;
+	if (BC(dev).bc_open == 0) {
+		bcache_free(BC(dev).bc_bcache);
+		BC(dev).bc_bcache = NULL;
+	}
 	return(0);
 }
 
+static int
+bc_strategy(void *devdata, int rw, daddr_t dblk, size_t offset, size_t size,
+    char *buf, size_t *rsize)
+{
+	struct bcache_devdata bcd;
+	struct i386_devdesc *dev;
+
+	dev = (struct i386_devdesc *)devdata;
+	bcd.dv_strategy = bc_realstrategy;
+	bcd.dv_devdata = devdata;
+	bcd.dv_cache = BC(dev).bc_bcache;
+
+	return (bcache_strategy(&bcd, rw, dblk, offset, size, buf, rsize));
+}
+
 static int 
-bc_strategy(void *devdata, int rw, daddr_t dblk, size_t size, char *buf,
-    size_t *rsize)
+bc_realstrategy(void *devdata, int rw, daddr_t dblk, size_t offset, size_t size,
+    char *buf, size_t *rsize)
 {
 	struct i386_devdesc *dev;
 	int unit;
diff --git a/sys/boot/i386/libi386/biosdisk.c b/sys/boot/i386/libi386/biosdisk.c
index 38ca85d..4dfb6f8 100644
--- a/sys/boot/i386/libi386/biosdisk.c
+++ b/sys/boot/i386/libi386/biosdisk.c
@@ -114,6 +114,8 @@ static struct bdinfo
 	int		bd_type;	/* BIOS 'drive type' (floppy only) */
 	uint16_t	bd_sectorsize;	/* Sector size */
 	uint64_t	bd_sectors;	/* Disk size */
+	int		bd_open;	/* reference counter */
+	void		*bd_bcache;	/* buffer cache data */
 } bdinfo [MAXBDDEV];
 static int nbdinfo = 0;
 
@@ -126,9 +128,9 @@ static int bd_write(struct disk_devdesc *dev, daddr_t dblk, int blks,
 static int bd_int13probe(struct bdinfo *bd);
 
 static int bd_init(void);
-static int bd_strategy(void *devdata, int flag, daddr_t dblk, size_t size,
-    char *buf, size_t *rsize);
-static int bd_realstrategy(void *devdata, int flag, daddr_t dblk,
+static int bd_strategy(void *devdata, int flag, daddr_t dblk, size_t offset,
+    size_t size, char *buf, size_t *rsize);
+static int bd_realstrategy(void *devdata, int flag, daddr_t dblk, size_t offset,
     size_t size, char *buf, size_t *rsize);
 static int bd_open(struct open_file *f, ...);
 static int bd_close(struct open_file *f);
@@ -209,6 +211,8 @@ bd_init(void)
 			    (nfd >= *(unsigned char *)PTOV(BIOS_NUMDRIVES)))
 				break;
 #endif
+			bdinfo[nbdinfo].bd_open = 0;
+			bdinfo[nbdinfo].bd_bcache = NULL;
 			bdinfo[nbdinfo].bd_unit = unit;
 			bdinfo[nbdinfo].bd_flags = unit < 0x80 ? BD_FLOPPY: 0;
 			if (!bd_int13probe(&bdinfo[nbdinfo]))
@@ -222,6 +226,7 @@ bd_init(void)
 				nfd++;
 		}
 	}
+	bcache_add_dev(nbdinfo);
 	return(0);
 }
 
@@ -352,7 +357,9 @@ bd_open(struct open_file *f, ...)
 
 	if (dev->d_unit < 0 || dev->d_unit >= nbdinfo)
 		return (EIO);
-
+	BD(dev).bd_open++;
+	if (BD(dev).bd_bcache == NULL)
+	    BD(dev).bd_bcache = bcache_allocate();
 	err = disk_open(dev, BD(dev).bd_sectors * BD(dev).bd_sectorsize,
 	    BD(dev).bd_sectorsize, (BD(dev).bd_flags & BD_FLOPPY) ?
 	    DISK_F_NOCACHE: 0);
@@ -438,6 +445,11 @@ bd_close(struct open_file *f)
 	struct disk_devdesc *dev;
 
 	dev = (struct disk_devdesc *)f->f_devdata;
+	BD(dev).bd_open--;
+	if (BD(dev).bd_open == 0) {
+	    bcache_free(BD(dev).bd_bcache);
+	    BD(dev).bd_bcache = NULL;
+	}
 	return (disk_close(dev));
 }
 
@@ -461,8 +473,8 @@ bd_ioctl(struct open_file *f, u_long cmd, void *data)
 }
 
 static int
-bd_strategy(void *devdata, int rw, daddr_t dblk, size_t size, char *buf,
-    size_t *rsize)
+bd_strategy(void *devdata, int rw, daddr_t dblk, size_t offset, size_t size,
+    char *buf, size_t *rsize)
 {
 	struct bcache_devdata bcd;
 	struct disk_devdesc *dev;
@@ -470,13 +482,14 @@ bd_strategy(void *devdata, int rw, daddr_t dblk, size_t size, char *buf,
 	dev = (struct disk_devdesc *)devdata;
 	bcd.dv_strategy = bd_realstrategy;
 	bcd.dv_devdata = devdata;
-	return (bcache_strategy(&bcd, BD(dev).bd_unit, rw, dblk + dev->d_offset,
+	bcd.dv_cache = BD(dev).bd_bcache;
+	return (bcache_strategy(&bcd, rw, dblk + dev->d_offset, offset,
 	    size, buf, rsize));
 }
 
 static int
-bd_realstrategy(void *devdata, int rw, daddr_t dblk, size_t size, char *buf,
-    size_t *rsize)
+bd_realstrategy(void *devdata, int rw, daddr_t dblk, size_t offset, size_t size,
+    char *buf, size_t *rsize)
 {
     struct disk_devdesc *dev = (struct disk_devdesc *)devdata;
     int			blks;
diff --git a/sys/boot/i386/libi386/biosmem.c b/sys/boot/i386/libi386/biosmem.c
index 399a208..abcbddb 100644
--- a/sys/boot/i386/libi386/biosmem.c
+++ b/sys/boot/i386/libi386/biosmem.c
@@ -56,7 +56,7 @@ static uint8_t b_bios_probed;
 /*
  * The minimum amount of memory to reserve in bios_extmem for the heap.
  */
-#define	HEAP_MIN	(3 * 1024 * 1024)
+#define	HEAP_MIN	(64 * 1024 * 1024)
 
 /*
  * Products in this list need quirks to detect
diff --git a/sys/boot/i386/libi386/pxe.c b/sys/boot/i386/libi386/pxe.c
index d67665e..096ebc4 100644
--- a/sys/boot/i386/libi386/pxe.c
+++ b/sys/boot/i386/libi386/pxe.c
@@ -72,7 +72,7 @@ static void	bangpxe_call(int func);
 
 static int	pxe_init(void);
 static int	pxe_strategy(void *devdata, int flag, daddr_t dblk,
-			     size_t size, char *buf, size_t *rsize);
+			     size_t offset, size_t size, char *buf, size_t *rsize);
 static int	pxe_open(struct open_file *f, ...);
 static int	pxe_close(struct open_file *f);
 static void	pxe_print(int verbose);
@@ -247,7 +247,7 @@ pxe_init(void)
 
 
 static int
-pxe_strategy(void *devdata, int flag, daddr_t dblk, size_t size,
+pxe_strategy(void *devdata, int flag, daddr_t dblk, size_t offset, size_t size,
 		char *buf, size_t *rsize)
 {
 	return (EIO);
diff --git a/sys/boot/i386/loader/main.c b/sys/boot/i386/loader/main.c
index b3e7b41..bc32593 100644
--- a/sys/boot/i386/loader/main.c
+++ b/sys/boot/i386/loader/main.c
@@ -141,9 +141,9 @@ main(void)
     cons_probe();
 
     /*
-     * Initialise the block cache
+     * Initialise the block cache. Set the upper limit.
      */
-    bcache_init(32, 512);	/* 16k cache XXX tune this */
+    bcache_init(32768, 512);
 
     /*
      * Special handling for PXE and CD booting.
diff --git a/sys/boot/mips/beri/loader/beri_disk_cfi.c b/sys/boot/mips/beri/loader/beri_disk_cfi.c
index a6776b9..b2b6d00 100644
--- a/sys/boot/mips/beri/loader/beri_disk_cfi.c
+++ b/sys/boot/mips/beri/loader/beri_disk_cfi.c
@@ -45,8 +45,8 @@ static int	beri_cfi_disk_init(void);
 static int	beri_cfi_disk_open(struct open_file *, ...);
 static int	beri_cfi_disk_close(struct open_file *);
 static void	beri_cfi_disk_cleanup(void);
-static int	beri_cfi_disk_strategy(void *, int, daddr_t, size_t, char *,
-		    size_t *);
+static int	beri_cfi_disk_strategy(void *, int, daddr_t, size_t, size_t,
+		    char *, size_t *);
 static void	beri_cfi_disk_print(int);
 
 struct devsw beri_cfi_disk = {
@@ -69,8 +69,8 @@ beri_cfi_disk_init(void)
 }
 
 static int
-beri_cfi_disk_strategy(void *devdata, int flag, daddr_t dblk, size_t size,
-    char *buf, size_t *rsizep)
+beri_cfi_disk_strategy(void *devdata, int flag, daddr_t dblk, size_t offset,
+    size_t size, char *buf, size_t *rsizep)
 {
 	int error;
 
diff --git a/sys/boot/mips/beri/loader/beri_disk_sdcard.c b/sys/boot/mips/beri/loader/beri_disk_sdcard.c
index f47e067..2577e14 100644
--- a/sys/boot/mips/beri/loader/beri_disk_sdcard.c
+++ b/sys/boot/mips/beri/loader/beri_disk_sdcard.c
@@ -45,8 +45,8 @@ static int	beri_sdcard_disk_init(void);
 static int	beri_sdcard_disk_open(struct open_file *, ...);
 static int	beri_sdcard_disk_close(struct open_file *);
 static void	beri_sdcard_disk_cleanup(void);
-static int	beri_sdcard_disk_strategy(void *, int, daddr_t, size_t, char *,
-		    size_t *);
+static int	beri_sdcard_disk_strategy(void *, int, daddr_t, size_t, size_t,
+		    char *, size_t *);
 static void	beri_sdcard_disk_print(int);
 
 struct devsw beri_sdcard_disk = {
@@ -69,8 +69,8 @@ beri_sdcard_disk_init(void)
 }
 
 static int
-beri_sdcard_disk_strategy(void *devdata, int flag, daddr_t dblk, size_t size,
-    char *buf, size_t *rsizep)
+beri_sdcard_disk_strategy(void *devdata, int flag, daddr_t dblk, size_t offset,
+    size_t size, char *buf, size_t *rsizep)
 {
 	int error;
 
diff --git a/sys/boot/ofw/libofw/ofw_disk.c b/sys/boot/ofw/libofw/ofw_disk.c
index 9b2e11c..9c46ccc 100644
--- a/sys/boot/ofw/libofw/ofw_disk.c
+++ b/sys/boot/ofw/libofw/ofw_disk.c
@@ -43,7 +43,7 @@ __FBSDID("$FreeBSD$");
 
 static int	ofwd_init(void);
 static int	ofwd_strategy(void *devdata, int flag, daddr_t dblk,
-		    size_t size, char *buf, size_t *rsize);
+		    size_t offset, size_t size, char *buf, size_t *rsize);
 static int	ofwd_open(struct open_file *f, ...);
 static int	ofwd_close(struct open_file *f);
 static int	ofwd_ioctl(struct open_file *f, u_long cmd, void *data);
@@ -83,8 +83,8 @@ ofwd_init(void)
 }
 
 static int
-ofwd_strategy(void *devdata, int flag __unused, daddr_t dblk, size_t size,
-    char *buf, size_t *rsize)
+ofwd_strategy(void *devdata, int flag __unused, daddr_t dblk, size_t offset,
+    size_t size, char *buf, size_t *rsize)
 {
 	struct ofw_devdesc *dp = (struct ofw_devdesc *)devdata;
 	daddr_t pos;
diff --git a/sys/boot/pc98/libpc98/bioscd.c b/sys/boot/pc98/libpc98/bioscd.c
index e3df391..d1d1ca1 100644
--- a/sys/boot/pc98/libpc98/bioscd.c
+++ b/sys/boot/pc98/libpc98/bioscd.c
@@ -84,13 +84,19 @@ struct specification_packet {
 static struct bcinfo {
 	int	bc_unit;		/* BIOS unit number */
 	struct specification_packet bc_sp;
+	int	bc_open;		/* reference counter */
+	void	*bc_bcache;		/* buffer cache data */
 } bcinfo [MAXBCDEV];
 static int nbcinfo = 0;
 
+#define	BC(dev)	(bcinfo[(dev)->d_unit])
+
 static int	bc_read(int unit, daddr_t dblk, int blks, caddr_t dest);
 static int	bc_init(void);
 static int	bc_strategy(void *devdata, int flag, daddr_t dblk,
-		    size_t size, char *buf, size_t *rsize);
+		    size_t offset, size_t size, char *buf, size_t *rsize);
+static int	bc_realstrategy(void *devdata, int flag, daddr_t dblk,
+		    size_t offset, size_t size, char *buf, size_t *rsize);
 static int	bc_open(struct open_file *f, ...);
 static int	bc_close(struct open_file *f);
 static void	bc_print(int verbose);
@@ -160,6 +166,7 @@ bc_add(int biosdev)
 
 	printf("BIOS CD is cd%d\n", nbcinfo);
 	nbcinfo++;
+	bcache_add_dev(nbcinfo);	/* register cd device in bcache */
 	return(0);
 }
 
@@ -196,19 +203,44 @@ bc_open(struct open_file *f, ...)
 		return(ENXIO);
 	}
 
+	BC(dev).bc_open++;
+	if (BC(dev).bc_bcache == NULL)
+		BC(dev).bc_bcache = bcache_allocate();
 	return(0);
 }
  
 static int 
 bc_close(struct open_file *f)
 {
+	struct i386_devdesc *dev;
 
+	dev = (struct i386_devdesc *)f->f_devdata;
+	BC(dev).bc_open--;
+	if (BC(dev).bc_open == 0) {
+		bcache_free(BC(dev).bc_bcache);
+		BC(dev).bc_bcache = NULL;
+	}
 	return(0);
 }
 
+static int
+bc_strategy(void *devdata, int rw, daddr_t dblk, size_t offset, size_t size,
+    char *buf, size_t *rsize)
+{
+	struct bcache_devdata bcd;
+	struct i386_devdesc *dev;
+
+	dev = (struct i386_devdesc *)devdata;
+	bcd.dv_strategy = bc_realstrategy;
+	bcd.dv_devdata = devdata;
+	bcd.dv_cache = BC(dev).bc_bcache;
+
+	return (bcache_strategy(&bcd, rw, dblk, offset, size, buf, rsize));
+}
+
 static int 
-bc_strategy(void *devdata, int rw, daddr_t dblk, size_t size, char *buf,
-    size_t *rsize)
+bc_realstrategy(void *devdata, int rw, daddr_t dblk, size_t offset, size_t size,
+    char *buf, size_t *rsize)
 {
 	struct i386_devdesc *dev;
 	int unit;
diff --git a/sys/boot/pc98/libpc98/biosdisk.c b/sys/boot/pc98/libpc98/biosdisk.c
index b777d25..d55722d 100644
--- a/sys/boot/pc98/libpc98/biosdisk.c
+++ b/sys/boot/pc98/libpc98/biosdisk.c
@@ -96,9 +96,13 @@ static struct bdinfo
     int		bd_flags;
     int		bd_type;		/* BIOS 'drive type' (floppy only) */
     int		bd_da_unit;		/* kernel unit number for da */
+    int		bd_open;		/* reference counter */
+    void	*bd_bcache;		/* buffer cache data */
 } bdinfo [MAXBDDEV];
 static int nbdinfo = 0;
 
+#define	BD(dev)	(bdinfo[(dev)->d_unit])
+
 static int	bd_getgeom(struct open_disk *od);
 static int	bd_read(struct open_disk *od, daddr_t dblk, int blks,
 		    caddr_t dest);
@@ -176,6 +180,8 @@ bd_init(void)
     /* sequence 0x90, 0x80, 0xa0 */
     for (base = 0x90; base <= 0xa0; base += n, n += 0x30) {
 	for (unit = base; (nbdinfo < MAXBDDEV) || ((unit & 0x0f) < 4); unit++) {
+	    bdinfo[nbdinfo].bd_open = 0;
+	    bdinfo[nbdinfo].bd_bcache = NULL;
 	    bdinfo[nbdinfo].bd_unit = unit;
 	    bdinfo[nbdinfo].bd_flags = (unit & 0xf0) == 0x90 ? BD_FLOPPY : 0;
 
@@ -205,6 +211,7 @@ bd_init(void)
 	    nbdinfo++;
 	}
     }
+    bcache_add_dev(nbdinfo);
     return(0);
 }
 
@@ -427,6 +434,10 @@ bd_open(struct open_file *f, ...)
     if ((error = bd_opendisk(&od, dev)))
 	return(error);
     
+    BD(dev).bd_open++;
+    if (BD(dev).bd_bcache == NULL)
+	BD(dev).bd_bcache = bcache_allocate();
+
     /*
      * Save our context
      */
@@ -696,7 +707,14 @@ bd_bestslice(struct open_disk *od)
 static int 
 bd_close(struct open_file *f)
 {
-    struct open_disk	*od = (struct open_disk *)(((struct i386_devdesc *)(f->f_devdata))->d_kind.biosdisk.data);
+    struct i386_devdesc		*dev = f->f_devdata;
+    struct open_disk	*od = (struct open_disk *)(dev->d_kind.biosdisk.data);
+
+    BD(dev).bd_open--;
+    if (BD(dev).bd_open == 0) {
+	bcache_free(BD(dev).bd_bcache);
+	BD(dev).bd_bcache = NULL;
+    }
 
     bd_closedisk(od);
     return(0);
@@ -715,18 +733,23 @@ bd_closedisk(struct open_disk *od)
 }
 
 static int 
-bd_strategy(void *devdata, int rw, daddr_t dblk, size_t size, char *buf, size_t *rsize)
+bd_strategy(void *devdata, int rw, daddr_t dblk, size_t offset, size_t size,
+    char *buf, size_t *rsize)
 {
     struct bcache_devdata	bcd;
-    struct open_disk	*od = (struct open_disk *)(((struct i386_devdesc *)devdata)->d_kind.biosdisk.data);
+    struct i386_devdesc		*dev = f->f_devdata;
+    struct open_disk	*od = (struct open_disk *)(dev->d_kind.biosdisk.data);
 
     bcd.dv_strategy = bd_realstrategy;
     bcd.dv_devdata = devdata;
-    return(bcache_strategy(&bcd, od->od_unit, rw, dblk+od->od_boff, size, buf, rsize));
+    bcd.dv_cache = BD(dev).bd_bcache;
+    return(bcache_strategy(&bcd, od->od_unit, rw, dblk+od->od_boff, offset,
+	size, buf, rsize));
 }
 
 static int 
-bd_realstrategy(void *devdata, int rw, daddr_t dblk, size_t size, char *buf, size_t *rsize)
+bd_realstrategy(void *devdata, int rw, daddr_t dblk, size_t offset,
+    size_t size, char *buf, size_t *rsize)
 {
     struct open_disk	*od = (struct open_disk *)(((struct i386_devdesc *)devdata)->d_kind.biosdisk.data);
     int			blks;
diff --git a/sys/boot/pc98/libpc98/biosmem.c b/sys/boot/pc98/libpc98/biosmem.c
index a25d8c2..c5a9b30 100644
--- a/sys/boot/pc98/libpc98/biosmem.c
+++ b/sys/boot/pc98/libpc98/biosmem.c
@@ -40,7 +40,7 @@ uint32_t	bios_basemem, bios_extmem, high_heap_size;
 /*
  * The minimum amount of memory to reserve in bios_extmem for the heap.
  */
-#define	HEAP_MIN	(3 * 1024 * 1024)
+#define	HEAP_MIN	(64 * 1024 * 1024)
 
 void
 bios_getmem(void)
diff --git a/sys/boot/pc98/loader/main.c b/sys/boot/pc98/loader/main.c
index da9bca5..39444ad 100644
--- a/sys/boot/pc98/loader/main.c
+++ b/sys/boot/pc98/loader/main.c
@@ -147,9 +147,9 @@ main(void)
     cons_probe();
 
     /*
-     * Initialise the block cache
+     * Initialise the block cache. Set the upper limit.
      */
-    bcache_init(32, 512);	/* 16k cache XXX tune this */
+    bcache_init(32768, 512);
 
     /*
      * Special handling for PXE and CD booting.
diff --git a/sys/boot/powerpc/kboot/hostdisk.c b/sys/boot/powerpc/kboot/hostdisk.c
index 2deb956..ac4534c 100644
--- a/sys/boot/powerpc/kboot/hostdisk.c
+++ b/sys/boot/powerpc/kboot/hostdisk.c
@@ -33,7 +33,7 @@ __FBSDID("$FreeBSD$");
 
 static int hostdisk_init(void);
 static int hostdisk_strategy(void *devdata, int flag, daddr_t dblk,
-    size_t size, char *buf, size_t *rsize);
+    size_t offset, size_t size, char *buf, size_t *rsize);
 static int hostdisk_open(struct open_file *f, ...);
 static int hostdisk_close(struct open_file *f);
 static int hostdisk_ioctl(struct open_file *f, u_long cmd, void *data);
@@ -58,8 +58,8 @@ hostdisk_init(void)
 }
 
 static int
-hostdisk_strategy(void *devdata, int flag, daddr_t dblk, size_t size,
-    char *buf, size_t *rsize)
+hostdisk_strategy(void *devdata, int flag, daddr_t dblk, size_t offset,
+    size_t size, char *buf, size_t *rsize)
 {
 	struct devdesc *desc = devdata;
 	daddr_t pos;
diff --git a/sys/boot/powerpc/ps3/ps3cdrom.c b/sys/boot/powerpc/ps3/ps3cdrom.c
index 843ecd5..c5019e0 100644
--- a/sys/boot/powerpc/ps3/ps3cdrom.c
+++ b/sys/boot/powerpc/ps3/ps3cdrom.c
@@ -46,7 +46,7 @@
 
 static int ps3cdrom_init(void);
 static int ps3cdrom_strategy(void *devdata, int flag, daddr_t dblk,
-	size_t size, char *buf, size_t *rsize);
+	size_t offset, size_t size, char *buf, size_t *rsize);
 static int ps3cdrom_open(struct open_file *f, ...);
 static int ps3cdrom_close(struct open_file *f);
 static void ps3cdrom_print(int verbose);
@@ -76,7 +76,7 @@ static int ps3cdrom_init(void)
 }
 
 static int ps3cdrom_strategy(void *devdata, int flag, daddr_t dblk,
-	size_t size, char *buf, size_t *rsize)
+	size_t offset, size_t size, char *buf, size_t *rsize)
 {
 	struct ps3_devdesc *dev = (struct ps3_devdesc *) devdata;
 	int err;
diff --git a/sys/boot/powerpc/ps3/ps3disk.c b/sys/boot/powerpc/ps3/ps3disk.c
index 5c5195b..52a43f1 100644
--- a/sys/boot/powerpc/ps3/ps3disk.c
+++ b/sys/boot/powerpc/ps3/ps3disk.c
@@ -58,7 +58,7 @@ static void ps3disk_uuid_letoh(uuid_t *uuid);
 
 static int ps3disk_init(void);
 static int ps3disk_strategy(void *devdata, int flag, daddr_t dblk,
-	size_t size, char *buf, size_t *rsize);
+	size_t offset, size_t size, char *buf, size_t *rsize);
 static int ps3disk_open(struct open_file *f, ...);
 static int ps3disk_close(struct open_file *f);
 static void ps3disk_print(int verbose);
@@ -109,7 +109,7 @@ static int ps3disk_init(void)
 }
 
 static int ps3disk_strategy(void *devdata, int flag, daddr_t dblk,
-	size_t size, char *buf, size_t *rsize)
+    size_t offset, size_t size, char *buf, size_t *rsize)
 {
 	struct ps3_devdesc *dev = (struct ps3_devdesc *) devdata;
 	struct open_dev *od = (struct open_dev *) dev->d_disk.data;
diff --git a/sys/boot/uboot/lib/disk.c b/sys/boot/uboot/lib/disk.c
index a8b7853..018dfed 100644
--- a/sys/boot/uboot/lib/disk.c
+++ b/sys/boot/uboot/lib/disk.c
@@ -73,7 +73,8 @@ static int stor_readdev(struct disk_devdesc *, daddr_t, size_t, char *);
 
 /* devsw I/F */
 static int stor_init(void);
-static int stor_strategy(void *, int, daddr_t, size_t, char *, size_t *);
+static int stor_strategy(void *, int, daddr_t, size_t, size_t, char *,
+    size_t *);
 static int stor_open(struct open_file *, ...);
 static int stor_close(struct open_file *);
 static int stor_ioctl(struct open_file *f, u_long cmd, void *data);
@@ -143,8 +144,8 @@ stor_cleanup(void)
 }
 
 static int
-stor_strategy(void *devdata, int rw, daddr_t blk, size_t size, char *buf,
-    size_t *rsize)
+stor_strategy(void *devdata, int rw, daddr_t blk, size_t offset, size_t size,
+    char *buf, size_t *rsize)
 {
 	struct disk_devdesc *dev = (struct disk_devdesc *)devdata;
 	daddr_t bcount;
diff --git a/sys/boot/usb/storage/umass_loader.c b/sys/boot/usb/storage/umass_loader.c
index 4311b2b..85b89e8 100644
--- a/sys/boot/usb/storage/umass_loader.c
+++ b/sys/boot/usb/storage/umass_loader.c
@@ -48,7 +48,8 @@ static int umass_disk_open(struct open_file *,...);
 static int umass_disk_close(struct open_file *);
 static void umass_disk_cleanup(void);
 static int umass_disk_ioctl(struct open_file *, u_long, void *);
-static int umass_disk_strategy(void *, int, daddr_t, size_t, char *, size_t *);
+static int umass_disk_strategy(void *, int, daddr_t, size_t, size_t, char *,
+    size_t *);
 static void umass_disk_print(int);
 
 struct devsw umass_disk = {
@@ -84,8 +85,8 @@ umass_disk_init(void)
 }
 
 static int
-umass_disk_strategy(void *devdata, int flag, daddr_t dblk, size_t size,
-    char *buf, size_t *rsizep)
+umass_disk_strategy(void *devdata, int flag, daddr_t dblk, size_t offset,
+    size_t size, char *buf, size_t *rsizep)
 {
 	if (umass_uaa.device == NULL)
 		return (ENXIO);
diff --git a/sys/boot/userboot/userboot/host.c b/sys/boot/userboot/userboot/host.c
index 81858a9..3320a7f 100644
--- a/sys/boot/userboot/userboot/host.c
+++ b/sys/boot/userboot/userboot/host.c
@@ -167,8 +167,8 @@ host_dev_close(struct open_file *f)
 }
 
 static int
-host_dev_strategy(void *devdata, int rw, daddr_t dblk, size_t size,
-    char *buf, size_t *rsize)
+host_dev_strategy(void *devdata, int rw, daddr_t dblk, size_t offset,
+    size_t size, char *buf, size_t *rsize)
 {
 
 	return (ENOSYS);
diff --git a/sys/boot/userboot/userboot/main.c b/sys/boot/userboot/userboot/main.c
index 3a7a7fb..4c50400 100644
--- a/sys/boot/userboot/userboot/main.c
+++ b/sys/boot/userboot/userboot/main.c
@@ -131,6 +131,10 @@ loader_main(struct loader_callbacks *cb, void *arg, int version, int ndisks)
 #endif
 
 	/*
+	 * Initialise the block cache. Set the upper limit.
+	 */
+	bcache_init(32768, 512);
+	/*
 	 * March through the device switch probing for things.
 	 */
 	for (i = 0; devsw[i] != NULL; i++)
diff --git a/sys/boot/userboot/userboot/userboot_disk.c b/sys/boot/userboot/userboot/userboot_disk.c
index bbd9efc..56fe7e3 100644
--- a/sys/boot/userboot/userboot/userboot_disk.c
+++ b/sys/boot/userboot/userboot/userboot_disk.c
@@ -42,6 +42,8 @@ __FBSDID("$FreeBSD$");
 struct userdisk_info {
 	uint64_t	mediasize;
 	uint16_t	sectorsize;
+	int		ud_open;	/* reference counter */
+	void		*ud_bcache;	/* buffer cache data */
 };
 
 int userboot_disk_maxunit = 0;
@@ -52,7 +54,9 @@ static struct userdisk_info	*ud_info;
 static int	userdisk_init(void);
 static void	userdisk_cleanup(void);
 static int	userdisk_strategy(void *devdata, int flag, daddr_t dblk,
-		    size_t size, char *buf, size_t *rsize);
+		    size_t offset, size_t size, char *buf, size_t *rsize);
+static int	userdisk_realstrategy(void *devdata, int flag, daddr_t dblk,
+		    size_t offset, size_t size, char *buf, size_t *rsize);
 static int	userdisk_open(struct open_file *f, ...);
 static int	userdisk_close(struct open_file *f);
 static int	userdisk_ioctl(struct open_file *f, u_long cmd, void *data);
@@ -92,9 +96,11 @@ userdisk_init(void)
 				return (ENXIO);
 			ud_info[i].mediasize = mediasize;
 			ud_info[i].sectorsize = sectorsize;
+			ud_info[i].ud_open = 0;
+			ud_info[i].ud_bcache = NULL;
 		}
 	}
-
+	bcache_add_dev(userdisk_maxunit);
 	return(0);
 }
 
@@ -148,7 +154,9 @@ userdisk_open(struct open_file *f, ...)
 
 	if (dev->d_unit < 0 || dev->d_unit >= userdisk_maxunit)
 		return (EIO);
-
+	ud_info[dev->d_unit].ud_open++;
+	if (ud_info[dev->d_unit].ud_bcache == NULL)
+		ud_info[dev->d_unit].ud_bcache = bcache_allocate();
 	return (disk_open(dev, ud_info[dev->d_unit].mediasize,
 	    ud_info[dev->d_unit].sectorsize, 0));
 }
@@ -159,12 +167,32 @@ userdisk_close(struct open_file *f)
 	struct disk_devdesc *dev;
 
 	dev = (struct disk_devdesc *)f->f_devdata;
+	ud_info[dev->d_unit].ud_open--;
+	if (ud_info[dev->d_unit].ud_open == 0) {
+		bcache_free(ud_info[dev->d_unit].ud_bcache);
+		ud_info[dev->d_unit].ud_bcache = NULL;
+	}
 	return (disk_close(dev));
 }
 
 static int
-userdisk_strategy(void *devdata, int rw, daddr_t dblk, size_t size,
-    char *buf, size_t *rsize)
+userdisk_strategy(void *devdata, int rw, daddr_t dblk, size_t offset,
+    size_t size, char *buf, size_t *rsize)
+{
+	struct bcache_devdata bcd;
+	struct disk_devdesc *dev;
+
+	dev = (struct disk_devdesc *)devdata;
+	bcd.dv_strategy = userdisk_realstrategy;
+	bcd.dv_devdata = devdata;
+	bcd.dv_cache = ud_info[dev->d_unit].ud_bcache;
+	return (bcache_strategy(&bcd, rw, dblk + dev->d_offset, offset,
+	    size, buf, rsize));
+}
+
+static int
+userdisk_realstrategy(void *devdata, int rw, daddr_t dblk, size_t offset,
+    size_t size, char *buf, size_t *rsize)
 {
 	struct disk_devdesc *dev = devdata;
 	uint64_t	off;
@@ -177,7 +205,7 @@ userdisk_strategy(void *devdata, int rw, daddr_t dblk, size_t size,
 		return (EINVAL);
 	if (rsize)
 		*rsize = 0;
-	off = (dblk + dev->d_offset) * ud_info[dev->d_unit].sectorsize;
+	off = dblk * ud_info[dev->d_unit].sectorsize;
 	rc = CALLBACK(diskread, dev->d_unit, off, buf, size, &resid);
 	if (rc)
 		return (rc);
diff --git a/sys/boot/zfs/zfs.c b/sys/boot/zfs/zfs.c
index bf2fc6f8..059e80a 100644
--- a/sys/boot/zfs/zfs.c
+++ b/sys/boot/zfs/zfs.c
@@ -578,7 +578,7 @@ zfs_dev_close(struct open_file *f)
 }
 
 static int
-zfs_dev_strategy(void *devdata, int rw, daddr_t dblk, size_t size, char *buf, size_t *rsize)
+zfs_dev_strategy(void *devdata, int rw, daddr_t dblk, size_t offset, size_t size, char *buf, size_t *rsize)
 {
 
 	return (ENOSYS);