mkfs_ext2: more work. added test script

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

2 files changed, 153 insertions, 75 deletions

diff --git a/util-linux/mkfs_ext2.c b/util-linux/mkfs_ext2.c
index 56203c7..38654ef 100644
--- a/util-linux/mkfs_ext2.c
+++ b/util-linux/mkfs_ext2.c
@@ -163,17 +163,16 @@ int mkfs_ext2_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
 int mkfs_ext2_main(int argc UNUSED_PARAM, char **argv)
 {
 	unsigned i, pos, n;
-	unsigned bs, blocksize, blocksize_log2;
+	unsigned bs, bpi;
+	unsigned blocksize, blocksize_log2;
 	unsigned nreserved = 5;
-	unsigned long long nblocks_ull;
+	unsigned long long kilobytes;
 	uint32_t nblocks;
 	uint32_t ngroups;
-	unsigned bytes_per_inode;
-	uint32_t nblocks_per_group;
+	uint32_t bytes_per_inode;
 	uint32_t first_data_block;
-	uint32_t ninodes;
 	uint32_t ninodes_per_group;
-	uint32_t gdtsz, rgdtsz, itsz;
+	uint32_t gdtsz, itsz;
 	time_t timestamp;
 	unsigned opts;
 	const char *label;
@@ -186,7 +185,7 @@ int mkfs_ext2_main(int argc UNUSED_PARAM, char **argv)
 
 	opt_complementary = "-1:b+:m+:i+";
 	opts = getopt32(argv, "cl:b:f:i:I:J:G:N:m:o:g:L:M:O:r:E:T:U:jnqvFS",
-		NULL, &bs, NULL, &bytes_per_inode, NULL, NULL, NULL, NULL,
+		NULL, &bs, NULL, &bpi, NULL, NULL, NULL, NULL,
 		&nreserved, NULL, NULL, &label, NULL, NULL, NULL, NULL, NULL, NULL);
 	argv += optind; // argv[0] -- device
 
@@ -195,7 +194,8 @@ int mkfs_ext2_main(int argc UNUSED_PARAM, char **argv)
 		bb_error_msg_and_die("-%c is bad", 'm');
 
 	// check the device is a block device
-	xstat(argv[0], &st);
+	xmove_fd(xopen(argv[0], O_WRONLY), fd);
+	fstat(fd, &st);
 	if (!S_ISBLK(st.st_mode) && !(opts & OPT_F))
 		bb_error_msg_and_die("not a block device");
 
@@ -206,22 +206,29 @@ int mkfs_ext2_main(int argc UNUSED_PARAM, char **argv)
 		bb_error_msg_and_die("can't format mounted filesystem");
 
 	// open the device, get size in kbytes
-	xmove_fd(xopen3(argv[0], O_WRONLY | O_CREAT, 0666), fd);
 	if (argv[1]) {
-		nblocks_ull = xatoull(argv[1]);
+		kilobytes = xatoull(argv[1]);
 	} else {
-		nblocks_ull = (uoff_t)xlseek(fd, 0, SEEK_END) / 1024;
+		kilobytes = (uoff_t)xlseek(fd, 0, SEEK_END) / 1024;
 	}
 
+	bytes_per_inode = 16384;
+	if (kilobytes < 512*1024)
+		bytes_per_inode = 4096;
+	if (kilobytes < 3*1024)
+		bytes_per_inode = 8192;
+	if (opts & OPT_i)
+		bytes_per_inode = bpi;
+
 	// block size is a multiple of 1024
 	blocksize = 1024;
-	if (nblocks_ull >= 512*1024) // mke2fs 1.41.9 compat
+	if (kilobytes >= 512*1024) // mke2fs 1.41.9 compat
 		blocksize = 4096;
 	if (EXT2_MAX_BLOCK_SIZE > 4096) {
-		// nblocks_ull >> 22 == size in 4gigabyte chunks.
+		// kilobytes >> 22 == size in 4gigabyte chunks.
 		// if it is >= 16k gigs, blocksize must be increased.
 		// Try "mke2fs -F image_std $((16 * 1024*1024*1024))"
-		while ((nblocks_ull >> 22) >= blocksize)
+		while ((kilobytes >> 22) >= blocksize)
 			blocksize *= 2;
 	}
 	if (opts & OPT_b)
@@ -233,13 +240,49 @@ int mkfs_ext2_main(int argc UNUSED_PARAM, char **argv)
 		bb_error_msg_and_die("blocksize %u is bad", blocksize);
 	}
 	blocksize_log2 = int_log2(blocksize);
-	nblocks_ull >>= (blocksize_log2 - EXT2_MIN_BLOCK_LOG_SIZE);
+	kilobytes >>= (blocksize_log2 - EXT2_MIN_BLOCK_LOG_SIZE);
 	// nblocks: the total number of blocks in the filesystem
-	nblocks = nblocks_ull;
-	if (nblocks != nblocks_ull)
+	nblocks = kilobytes;
+	if (nblocks != kilobytes)
 		bb_error_msg_and_die("block count doesn't fit in 32 bits");
 	if (nblocks < 8)
 		bb_error_msg_and_die("need >= 8 blocks");
+#define kilobytes kilobytes_unused_after_this
+
+	// number of bits in one block, i.e. 8*blocksize
+#define	blocks_per_group (8 * blocksize)
+
+/* e2fsprogs-1.41.9
+	overhead = 2 + fs->inode_blocks_per_group;
+	if (has_super(fs->group_desc_count - 1))
+		overhead += 1 + fs->desc_blocks + super->s_reserved_gdt_blocks;
+	rem = (nblocks - first_data_block) % blocks_per_group);
+	if ((fs->group_desc_count == 1) && rem && (rem < overhead)) {
+		retval = EXT2_ET_TOOSMALL;
+		goto cleanup;
+	}
+	if (rem && (rem < overhead+50)) {
+		nblocks -= rem;
+		goto retry;
+	}
+*/
+
+	// N.B. a block group can have no more than blocks_per_group blocks
+	first_data_block = (EXT2_MIN_BLOCK_SIZE == blocksize);
+	ngroups = div_roundup(nblocks - first_data_block, blocks_per_group);
+	if (0 == ngroups)
+		bb_error_msg_and_die("ngroups");
+
+	{
+		// ninodes is the total number of inodes (files) in the file system
+		uint32_t ninodes = nblocks / (bytes_per_inode >> blocksize_log2);
+		if (ninodes < EXT2_GOOD_OLD_FIRST_INO+1)
+			ninodes = EXT2_GOOD_OLD_FIRST_INO+1;
+		ninodes_per_group = div_roundup(ninodes, ngroups);
+		// minimum number because the first EXT2_GOOD_OLD_FIRST_INO-1 are reserved
+		if (ninodes_per_group < 16)
+			ninodes_per_group = 16;
+	}
 
 	// TODO: 5?/5 WE MUST NOT DEPEND ON WHETHER DEVICE IS /dev/zero 'ed OR NOT
 	// TODO: 3/5 refuse if mounted
@@ -260,15 +303,13 @@ int mkfs_ext2_main(int argc UNUSED_PARAM, char **argv)
 	sb->s_log_block_size = sb->s_log_frag_size = blocksize_log2 - EXT2_MIN_BLOCK_LOG_SIZE;
 	// first 1024 bytes of the device are for boot record. If block size is 1024 bytes, then
 	// the first block available for data is 1, otherwise 0
-	first_data_block = sb->s_first_data_block = (EXT2_MIN_BLOCK_SIZE == blocksize);
+	sb->s_first_data_block = first_data_block; // 0 or 1
 	// block and inode bitmaps occupy no more than one block, so maximum number of blocks is
-	// number of bits in one block, i.e. 8*blocksize
-	nblocks_per_group = sb->s_blocks_per_group = sb->s_frags_per_group = sb->s_inodes_per_group = 8 * blocksize;
+	sb->s_blocks_per_group = sb->s_frags_per_group = blocks_per_group;
 	timestamp = time(NULL);
 	sb->s_mkfs_time = sb->s_wtime = sb->s_lastcheck = timestamp;
 	sb->s_state = 1;
 	sb->s_creator_os = EXT2_OS_LINUX;
-	sb->s_max_mnt_count = EXT2_DFL_MAX_MNT_COUNT;
 	sb->s_checkinterval = 24*60*60 * 180; // 180 days
 	sb->s_errors = EXT2_ERRORS_DEFAULT;
 	sb->s_feature_compat = EXT2_FEATURE_COMPAT_SUPP
@@ -289,17 +330,14 @@ int mkfs_ext2_main(int argc UNUSED_PARAM, char **argv)
 	 * don't check all the filesystems at the same time.  We use a
 	 * kludgy hack of using the UUID to derive a random jitter value.
 	 */
-	sb->s_max_mnt_count += sb->s_uuid[ARRAY_SIZE(sb->s_uuid)-1] % EXT2_DFL_MAX_MNT_COUNT;
+	sb->s_max_mnt_count = EXT2_DFL_MAX_MNT_COUNT
+		+ (sb->s_uuid[ARRAY_SIZE(sb->s_uuid)-1] % EXT2_DFL_MAX_MNT_COUNT);
 
 	sb->s_blocks_count = nblocks;
 
 	// reserve blocks for superuser
 	sb->s_r_blocks_count = ((uint64_t) nblocks * nreserved) / 100;
 
-	// N.B. a block group can have no more than nblocks_per_group blocks
-	ngroups = div_roundup(nblocks - first_data_block, nblocks_per_group);
-	if (0 == ngroups)
-		bb_error_msg_and_die("ngroups");
 	gdtsz = div_roundup(ngroups, EXT2_DESC_PER_BLOCK(sb));
 	/*
 	 * From e2fsprogs: Calculate the number of GDT blocks to reserve for online
@@ -310,44 +348,28 @@ int mkfs_ext2_main(int argc UNUSED_PARAM, char **argv)
 	/* We set it at 1024x the current filesystem size, or
 	 * the upper block count limit (2^32), whichever is lower.
 	 */
-#if ENABLE_FEATURE_MKFS_EXT2_RESERVED_GDT
-	rgdtsz = 0xFFFFFFFF; // maximum block number
-	if (nblocks < rgdtsz / 1024)
-		rgdtsz = nblocks * 1024;
-	rgdtsz = div_roundup(rgdtsz - first_data_block, nblocks_per_group);
-	rgdtsz = div_roundup(rgdtsz, EXT2_DESC_PER_BLOCK(sb)) - gdtsz;
-	if (rgdtsz > EXT2_ADDR_PER_BLOCK(sb))
-		rgdtsz = EXT2_ADDR_PER_BLOCK(sb);
-	sb->s_reserved_gdt_blocks = rgdtsz;
-	//bb_info_msg("RSRVD[%u]", n);
-#else
-	rgdtsz = 0;
-#endif
-
-	// ninodes is the total number of inodes (files) in the file system
-	if (!(opts & OPT_i)) {
-		bytes_per_inode = 16384;
-		if (nblocks < 512*1024)
-			bytes_per_inode = 4096;
-		if (nblocks < 3*1024)
-			bytes_per_inode = 8192;
+	if (ENABLE_FEATURE_MKFS_EXT2_RESERVED_GDT) {
+		uint32_t rgdtsz = 0xFFFFFFFF; // maximum block number
+		if (nblocks < rgdtsz / 1024)
+			rgdtsz = nblocks * 1024;
+		rgdtsz = div_roundup(rgdtsz - first_data_block, blocks_per_group);
+		rgdtsz = div_roundup(rgdtsz, EXT2_DESC_PER_BLOCK(sb)) - gdtsz;
+		if (rgdtsz > EXT2_ADDR_PER_BLOCK(sb))
+			rgdtsz = EXT2_ADDR_PER_BLOCK(sb);
+		STORE_LE(sb->s_reserved_gdt_blocks, rgdtsz);
+		gdtsz += rgdtsz;
 	}
-	ninodes = nblocks / (bytes_per_inode >> blocksize_log2);
-	if (ninodes < EXT2_GOOD_OLD_FIRST_INO+1)
-		ninodes = EXT2_GOOD_OLD_FIRST_INO+1;
-	ninodes_per_group = div_roundup(ninodes, ngroups);
-	if (ninodes_per_group < 16)
-		ninodes_per_group = 16; // minimum number because the first EXT2_GOOD_OLD_FIRST_INO-1 are reserved
-	// N.B. a block group can have no more than 8*blocksize = sb->s_inodes_per_group inodes
-	if (ninodes_per_group > sb->s_inodes_per_group)
-		ninodes_per_group = sb->s_inodes_per_group;
+
+	// N.B. a block group can have no more than 8*blocksize inodes
+	if (ninodes_per_group > blocks_per_group)
+		ninodes_per_group = blocks_per_group;
 	// adjust inodes per group so they completely fill the inode table blocks in the descriptor
 	ninodes_per_group = (div_roundup(ninodes_per_group * EXT2_INODE_SIZE(sb), blocksize) << blocksize_log2) / EXT2_INODE_SIZE(sb);
 	// make sure the number of inodes per group is a multiple of 8
 	ninodes_per_group &= ~7;
 	sb->s_inodes_per_group = ninodes_per_group;// = div_roundup(ninodes_per_group * sb->s_inode_size, blocksize);
 	// total ninodes
-	ninodes = sb->s_inodes_count = ninodes_per_group * ngroups;
+	sb->s_inodes_count = ninodes_per_group * ngroups;
 
 	itsz = ninodes_per_group * sb->s_inode_size / blocksize;
 	sb->s_free_inodes_count = sb->s_inodes_count - EXT2_GOOD_OLD_FIRST_INO;
@@ -370,7 +392,7 @@ int mkfs_ext2_main(int argc UNUSED_PARAM, char **argv)
 		"%u inodes, %u blocks\n"
 		"%u blocks (%u%%) reserved for the super user\n"
 		"First data block=%u\n"
-//		"Maximum filesystem blocks=%lu\n"
+		"Maximum filesystem blocks=%u\n"
 		"%u block groups\n"
 		"%u blocks per group, %u fragments per group\n"
 		"%u inodes per group"
@@ -380,16 +402,17 @@ int mkfs_ext2_main(int argc UNUSED_PARAM, char **argv)
 		, sb->s_inodes_count, sb->s_blocks_count
 		, sb->s_r_blocks_count, nreserved
 		, first_data_block
-//		, (rgdtsz + gdtsz) * EXT2_DESC_PER_BLOCK(sb) * nblocks_per_group
+		, gdtsz * EXT2_DESC_PER_BLOCK(sb) * blocks_per_group
 		, ngroups
-		, nblocks_per_group, nblocks_per_group
+		, blocks_per_group, blocks_per_group
 		, ninodes_per_group
 	);
 	{
-		const char *fmt = "\nSuperblock backups stored on blocks: %u";
+		const char *fmt = "\nSuperblock backups stored on blocks:\n"
+			"\t%u";
 		pos = first_data_block;
 		for (i = 1; i < ngroups; i++) {
-			pos += nblocks_per_group;
+			pos += blocks_per_group;
 			if (has_super(i)) {
 				printf(fmt, (unsigned)pos);
 				fmt = ", %u";
@@ -400,16 +423,16 @@ int mkfs_ext2_main(int argc UNUSED_PARAM, char **argv)
 #endif
 
 	if (opts & OPT_n)
-		return EXIT_SUCCESS;
+		goto done;
 
 	// fill group descriptors
-	gd = xzalloc((gdtsz + rgdtsz) * blocksize);
+	gd = xzalloc(gdtsz * blocksize);
 	sb->s_free_blocks_count = 0;
 	for (i = 0, pos = first_data_block, n = nblocks;
 		i < ngroups;
-		i++, pos += nblocks_per_group, n -= nblocks_per_group
+		i++, pos += blocks_per_group, n -= blocks_per_group
 	) {
-		uint32_t overhead = pos + (has_super(i) ? (1/*sb*/ + gdtsz + rgdtsz) : 0);
+		uint32_t overhead = pos + (has_super(i) ? (1/*sb*/ + gdtsz) : 0);
 		gd[i].bg_block_bitmap = overhead + 0;
 		gd[i].bg_inode_bitmap = overhead + 1;
 		gd[i].bg_inode_table  = overhead + 2;
@@ -427,7 +450,7 @@ int mkfs_ext2_main(int argc UNUSED_PARAM, char **argv)
 		if (ngroups-1 == i) {
 			overhead += first_data_block;
 		}
-		gd[i].bg_free_blocks_count = (n < nblocks_per_group ? n : nblocks_per_group) - overhead;
+		gd[i].bg_free_blocks_count = (n < blocks_per_group ? n : blocks_per_group) - overhead;
 		sb->s_free_blocks_count += gd[i].bg_free_blocks_count;
 	}
 	STORE_LE(sb->s_free_blocks_count, sb->s_free_blocks_count);
@@ -435,22 +458,22 @@ int mkfs_ext2_main(int argc UNUSED_PARAM, char **argv)
 	// dump filesystem skeleton structures
 //	printf("Writing superblocks and filesystem accounting information: ");
 	buf = xmalloc(blocksize);
-	for (i = 0, pos = first_data_block; i < ngroups; i++, pos += nblocks_per_group) {
-		uint32_t overhead = has_super(i) ? (1/*sb*/ + gdtsz + rgdtsz) : 0;
-		uint32_t start;// = has_super(i) ? (1/*sb*/ + gdtsz + rgdtsz) : 0;
+	for (i = 0, pos = first_data_block; i < ngroups; i++, pos += blocks_per_group) {
+		uint32_t overhead = has_super(i) ? (1/*sb*/ + gdtsz) : 0;
+		uint32_t start;
 		uint32_t end;
 
 		// dump superblock and group descriptors and their backups
 		if (overhead) { // N.B. in fact, we want (has_super(i)) condition, but it is equal to (overhead != 0) and is cheaper
 			// N.B. 1024 byte blocks are special
 			PUT(((uint64_t)pos << blocksize_log2) + ((0 == i && 0 == first_data_block) ? 1024 : 0), sb, 1024);//blocksize);
-			PUT(((uint64_t)pos << blocksize_log2) + blocksize, gd, (gdtsz + rgdtsz) * blocksize);
+			PUT(((uint64_t)pos << blocksize_log2) + blocksize, gd, gdtsz * blocksize);
 		}
 
 		start = overhead + 1/*bbmp*/ + 1/*ibmp*/ + itsz;
 		if (i == 0)
 			start += 2; // for "/" and "/lost+found"
-		end = nblocks_per_group - (start + gd[i].bg_free_blocks_count);
+		end = blocks_per_group - (start + gd[i].bg_free_blocks_count);
 
 		// mark preallocated blocks as allocated
 		allocate(buf, blocksize, start, end);
@@ -460,7 +483,7 @@ int mkfs_ext2_main(int argc UNUSED_PARAM, char **argv)
 		// mark preallocated inodes as allocated
 		allocate(buf, blocksize,
 			ninodes_per_group - gd[i].bg_free_inodes_count,
-			8 * blocksize - ninodes_per_group
+			blocks_per_group - ninodes_per_group
 		);
 		// dump inode bitmap
 		//PUT((uint64_t)(pos + overhead + 1) * blocksize, buf, blocksize);
@@ -525,10 +548,11 @@ int mkfs_ext2_main(int argc UNUSED_PARAM, char **argv)
 
 //	bb_info_msg("done\n"
 //		"This filesystem will be automatically checked every %u mounts or\n"
-//		"%u days, whichever comes first. Use tune2fs -c or -i to override.",
-//		sb->s_max_mnt_count, sb->s_checkinterval / (3600 * 24)
+//		"180 days, whichever comes first. Use tune2fs -c or -i to override.",
+//		sb->s_max_mnt_count
 //	);
 
+ done:
 	// cleanup
 	if (ENABLE_FEATURE_CLEAN_UP) {
 		free(buf);
diff --git a/util-linux/mkfs_ext2_test.sh b/util-linux/mkfs_ext2_test.sh
new file mode 100755
index 0000000..fe6291f
--- /dev/null
+++ b/util-linux/mkfs_ext2_test.sh
@@ -0,0 +1,54 @@
+#!/bin/sh
+
+test_mke2fs() {
+    echo Testing $kilobytes
+
+    >image_std
+    dd seek=$((kilobytes-1)) bs=1K count=1 </dev/zero of=image_std >/dev/null 2>&1 || exit 1
+    /usr/bin/mke2fs -F image_std $kilobytes >image_std.raw_out 2>&1 || return 1
+    cat image_std.raw_out \
+    | grep -v '^mke2fs ' \
+    | grep -v '^Maximum filesystem' \
+    | grep -v '^warning: .* blocks unused' \
+    | grep -v '^Writing inode tables' \
+    | grep -v '^Writing superblocks and filesystem accounting information' \
+    | grep -v '^This filesystem will be automatically checked every' \
+    | grep -v '^180 days, whichever comes first' \
+    | sed 's/block groups/block group/' \
+    | sed 's/ *$//' \
+    | sed 's/blocks (.*%) reserved/blocks reserved/' \
+    | grep -v '^$' \
+    >image_std.out
+
+    >image_bb
+    dd seek=$((kilobytes-1)) bs=1K count=1 </dev/zero of=image_bb >/dev/null 2>&1 || exit 1
+    ./busybox mke2fs -F image_bb $kilobytes >image_bb.raw_out 2>&1 || return 1
+    cat image_bb.raw_out \
+    | grep -v '^mke2fs ' \
+    | grep -v '^Maximum filesystem' \
+    | grep -v '^warning: .* blocks unused' \
+    | grep -v '^Writing inode tables' \
+    | grep -v '^Writing superblocks and filesystem accounting information' \
+    | grep -v '^This filesystem will be automatically checked every' \
+    | grep -v '^180 days, whichever comes first' \
+    | sed 's/block groups/block group/' \
+    | sed 's/ *$//' \
+    | sed 's/blocks (.*%) reserved/blocks reserved/' \
+    | grep -v '^$' \
+    >image_bb.out
+
+    diff -ua image_bb.out image_std.out >image.out.diff || {
+	cat image.out.diff
+	return 1
+    }
+}
+
+kilobytes=24908 test_mke2fs
+kilobytes=81940 test_mke2fs
+kilobytes=98392 test_mke2fs
+exit
+
+while true; do
+    kilobytes=$(( (RANDOM*RANDOM) % 100000 + 100))
+    test_mke2fs || exit 1
+done