diff --git a/bcache-test.c b/bcache-test.c
index 33218e53..69a3c48b 100644
--- a/bcache-test.c
+++ b/bcache-test.c
@@ -71,8 +71,9 @@ long getblocks(int fd)
 int main(int argc, char **argv)
 {
 	bool walk = false, randsize = false, verbose = false;
-	int fd1, fd2, direct = 0;
-	long size, i;
+	int fd1, fd2, direct = 0, nbytes = 4096, j;
+	unsigned long size, i = 0, offset = 0;
+	void *buf1 = NULL, *buf2 = NULL;
 
 	if (argc < 3) {
 		printf("Please enter a cache device and raw device\n");
@@ -82,12 +83,16 @@ int main(int argc, char **argv)
 	for (i = 3; i < argc; i++) {
 		if (strcmp(argv[i], "direct") == 0)
 			direct = O_DIRECT;
-		if (strcmp(argv[i], "walk") == 0)
+		else if (strcmp(argv[i], "walk") == 0)
 			walk = true;
-		if (strcmp(argv[i], "verbose") == 0)
+		else if (strcmp(argv[i], "verbose") == 0)
 			verbose = true;
-		if (strcmp(argv[i], "size") == 0)
+		else if (strcmp(argv[i], "size") == 0)
 			randsize = true;
+		else {
+			printf("Uknown argument %s\n", argv[i]);
+			exit(EXIT_FAILURE);
+		}
 	}
 
 	fd1 = open(argv[1], O_RDONLY|direct);
@@ -97,29 +102,42 @@ int main(int argc, char **argv)
 		exit(EXIT_FAILURE);
 	}
 
-	size = MIN(getblocks(fd1), getblocks(fd2)) / 8;
+	size = MIN(getblocks(fd1), getblocks(fd2)) / 8 - 16;
 	printf("size %li\n", size);
 
-	for (i = 0;; i++) {
-		char buf1[4096 * 16], buf2[4096 * 16];
-		long offset;
-		int pages = randsize ? MAX(MIN(abs(normal()) * 4, 16), 1) : 1;
+	if (posix_memalign(&buf1, 4096, 4096 * 16) ||
+	    posix_memalign(&buf2, 4096, 4096 * 16)) {
+		printf("Could not allocate buffers\n");
+		exit(EXIT_FAILURE);
+	}
 
-		offset = walk ? offset * normal() * 2 : random();
+	while (1) {
+		if (randsize)
+			nbytes = 4096 * (int) (drand48() * 16 + 1);
+
+		offset += walk ? normal() * 60 : random();
 		offset %= size;
+		assert(offset < size);
 
-		if (verbose)
-			printf("Loop %li offset %li\n", i, offset);
-		else if (!(i % 100))
-			printf("Loop %li\n", i);
+		do {
+			if (verbose)
+				printf("Loop %li offset %li sectors %i\n",
+				       i, offset << 3, nbytes >> 9);
+			else if (!(i % 100))
+				printf("Loop %li\n", i);
 
-		Pread(fd1, buf1, 4096 * pages, offset << 12);
-		Pread(fd2, buf2, 4096 * pages, offset << 12);
+			Pread(fd1, buf1, nbytes, offset << 12);
+			Pread(fd2, buf2, nbytes, offset << 12);
 
-		if (memcmp(buf1, buf2, 4096 * pages)) {
-			printf("Bad read! offset %li", offset << 12);
-			exit(EXIT_FAILURE);
-		}
+			for (j = 0; j < nbytes; j += 512)
+				if (memcmp(buf1 + j,
+					   buf2 + j,
+					   512)) {
+					printf("Bad read! offset %li sectors %i, sector %i\n",
+					       offset << 3, nbytes >> 9, j >> 9);
+					exit(EXIT_FAILURE);
+				}
+		} while (!(i++ & 1));
 	}
 err:
 	perror("Read error");
diff --git a/make-bcache.c b/make-bcache.c
index 54828bbc..38a4d436 100644
--- a/make-bcache.c
+++ b/make-bcache.c
@@ -1,3 +1,5 @@
+#define _FILE_OFFSET_BITS	64
+#define __USE_FILE_OFFSET64
 #define _XOPEN_SOURCE 500
 
 #include <fcntl.h>
@@ -10,6 +12,7 @@
 #include <sys/ioctl.h>
 #include <sys/types.h>
 #include <sys/stat.h>
+#include <unistd.h>
 
 static const char bcache_magic[] = {
 	0xc6, 0x85, 0x73, 0xf6, 0x4e, 0x1a, 0x45, 0xca,
@@ -34,47 +37,86 @@ struct bucket_disk {
 
 char zero[4096];
 
-int main(int argc, char **argv)
+long getblocks(int fd)
 {
-	long n;
-	int fd, i;
+	long ret;
 	struct stat statbuf;
-	struct cache_sb sb;
-
-	if (argc < 2) {
-		printf("Please supply a device\n");
-		exit(EXIT_FAILURE);
-	}
-
-	fd = open(argv[1], O_RDWR);
-	if (!fd) {
-		perror("Can't open dev\n");
-		exit(EXIT_FAILURE);
-	}
-
 	if (fstat(fd, &statbuf)) {
 		perror("stat error\n");
 		exit(EXIT_FAILURE);
 	}
-	if (!S_ISBLK(statbuf.st_mode))
-		n = statbuf.st_blocks;
-	else
-		if (ioctl(fd, BLKGETSIZE, &n)) {
+	ret = statbuf.st_blocks;
+	if (S_ISBLK(statbuf.st_mode))
+		if (ioctl(fd, BLKGETSIZE, &ret)) {
 			perror("ioctl error");
 			exit(EXIT_FAILURE);
 		}
+	return ret;
+}
+
+long hatoi(const char *s)
+{
+	char *e;
+	long long i = strtol(s, &e, 10);
+	switch (*e) {
+		case 't':
+		case 'T':
+			i *= 1024;
+		case 'g':
+		case 'G':
+			i *= 1024;
+		case 'm':
+		case 'M':
+			i *= 1024;
+		case 'k':
+		case 'K':
+			i *= 1024;
+	}
+	return i;
+}
+
+int main(int argc, char **argv)
+{
+	int64_t nblocks, bucketsize = 1024, blocksize = 8;
+	int fd, i, c;
+	struct cache_sb sb;
+
+	while ((c = getopt(argc, argv, "b:")) != -1) {
+		switch (c) {
+		case 'b':
+			bucketsize = hatoi(optarg) / 512;
+			break;
+		}
+	}
+
+	if (argc <= optind) {
+		printf("Please supply a device\n");
+		exit(EXIT_FAILURE);
+	}
+
+	fd = open(argv[optind], O_RDWR);
+	if (fd == -1) {
+		perror("Can't open dev\n");
+		exit(EXIT_FAILURE);
+	}
+	nblocks = getblocks(fd);
+
+	if (bucketsize < blocksize ||
+	    bucketsize > nblocks / 8) {
+		printf("Bad bucket size %li\n", bucketsize);
+		exit(EXIT_FAILURE);
+	}
 
 	memcpy(sb.magic, bcache_magic, 16);
 	sb.version = 0;
-	sb.block_size = 8;
-	sb.bucket_size = 32;
-	sb.nbuckets = n / sb.bucket_size;
+	sb.block_size = blocksize;
+	sb.bucket_size = bucketsize;
+	sb.nbuckets = nblocks / sb.bucket_size;
 
 	do
 		sb.first_bucket = ((--sb.nbuckets * sizeof(struct bucket_disk))
 				   + 4096 * 3) / (sb.bucket_size * 512) + 1;
-	while ((sb.nbuckets + sb.first_bucket) * sb.bucket_size * 512
-	       > statbuf.st_size);
+	while ((sb.nbuckets + sb.first_bucket) * sb.bucket_size > nblocks);
 
 	sb.journal_start = sb.first_bucket;