1 /* vi: set sw=4 ts=4: */
3 * mkfs_ext2: utility to create EXT2 filesystem
4 * inspired by genext2fs
6 * Busybox'ed (2009) by Vladimir Dronnikov <dronnikov@gmail.com>
8 * Licensed under GPLv2, see file LICENSE in this tarball for details.
12 #include <linux/ext2_fs.h>
13 #include "volume_id/volume_id_internal.h"
15 #define ENABLE_FEATURE_MKFS_EXT2_RESERVED_GDT 0
16 #define ENABLE_FEATURE_MKFS_EXT2_DIR_INDEX 1
19 #define s_reserved_gdt_blocks s_padding1
20 #define s_mkfs_time s_reserved[0]
21 #define s_flags s_reserved[22]
22 #define EXT2_HASH_HALF_MD4 1
23 #define EXT2_FLAGS_SIGNED_HASH 0x0001
25 // whiteout: for writable overlays
26 //#define LINUX_S_IFWHT 0160000
27 //#define EXT2_FEATURE_INCOMPAT_WHITEOUT 0x0020
30 void BUG_unsupported_field_size(void);
31 #define STORE_LE(field, value) \
33 if (sizeof(field) == 4) \
34 field = cpu_to_le32(value); \
35 else if (sizeof(field) == 2) \
36 field = cpu_to_le16(value); \
37 else if (sizeof(field) == 1) \
40 BUG_unsupported_field_size(); \
43 // All fields are little-endian
62 static unsigned int_log2(unsigned arg)
65 while ((arg >>= 1) != 0)
70 // taken from mkfs_minix.c. libbb candidate?
71 static unsigned div_roundup(uint32_t size, uint32_t n)
73 return (size + n-1) / n;
76 static void allocate(uint8_t *bitmap, uint32_t blocksize, uint32_t start, uint32_t end)
79 memset(bitmap, 0, blocksize);
81 memset(bitmap, 0xFF, i);
82 bitmap[i] = 0xFF >> (8-(start&7));
83 //bb_info_msg("ALLOC: [%u][%u][%u]: [%u]:=[%x]", blocksize, start, end, blocksize - end/8 - 1, (uint8_t)(0xFF << (8-(end&7))));
85 bitmap[blocksize - i - 1] = 0xFF << (8 - (end & 7));
86 memset(bitmap + blocksize - i, 0xFF, i); // N.B. no overflow here!
89 static uint32_t has_super(uint32_t x)
91 // 0, 1 and powers of 3, 5, 7 up to 2^32 limit
92 static const uint32_t supers[] = {
93 0, 1, 3, 5, 7, 9, 25, 27, 49, 81, 125, 243, 343, 625, 729,
94 2187, 2401, 3125, 6561, 15625, 16807, 19683, 59049, 78125,
95 117649, 177147, 390625, 531441, 823543, 1594323, 1953125,
96 4782969, 5764801, 9765625, 14348907, 40353607, 43046721,
97 48828125, 129140163, 244140625, 282475249, 387420489,
98 1162261467, 1220703125, 1977326743, 3486784401/* >2^31 */,
100 const uint32_t *sp = supers + ARRAY_SIZE(supers);
110 /* Standard mke2fs 1.41.9:
111 * Usage: mke2fs [-c|-l filename] [-b block-size] [-f fragment-size]
112 * [-i bytes-per-inode] [-I inode-size] [-J journal-options]
113 * [-G meta group size] [-N number-of-inodes]
114 * [-m reserved-blocks-percentage] [-o creator-os]
115 * [-g blocks-per-group] [-L volume-label] [-M last-mounted-directory]
116 * [-O feature[,...]] [-r fs-revision] [-E extended-option[,...]]
117 * [-T fs-type] [-U UUID] [-jnqvFSV] device [blocks-count]
119 // N.B. not commented below options are taken and silently ignored
123 OPT_b = 1 << 2, // block size, in bytes
125 OPT_i = 1 << 4, // bytes per inode
130 OPT_m = 1 << 9, // percentage of blocks reserved for superuser
133 OPT_L = 1 << 12, // label
141 OPT_n = 1 << 20, // dry run: do not write anything
146 //OPT_V = 1 << 25, // -V version. bbox applets don't support that
149 #define fd 3 /* predefined output descriptor */
151 static void PUT(uint64_t off, void *buf, uint32_t size)
153 // bb_info_msg("PUT[%llu]:[%u]", off, size);
154 xlseek(fd, off, SEEK_SET);
155 xwrite(fd, buf, size);
158 int mkfs_ext2_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
159 int mkfs_ext2_main(int argc UNUSED_PARAM, char **argv)
162 unsigned bs, blocksize, blocksize_log2;
163 unsigned nreserved = 5;
166 unsigned bytes_per_inode;
167 uint32_t nblocks_per_group;
168 uint32_t first_data_block;
170 uint32_t ninodes_per_group;
171 uint32_t gdtsz, rgdtsz, itsz;
176 struct ext2_super_block *sb; // superblock
177 struct ext2_group_desc *gd; // group descriptors
178 struct ext2_inode *inode;
179 struct ext2_dir *dir;
182 bs = EXT2_MIN_BLOCK_SIZE;
183 opt_complementary = "-1:b+:m+:i+";
184 opts = getopt32(argv, "cl:b:f:i:I:J:G:N:m:o:g:L:M:O:r:E:T:U:jnqvFS",
185 NULL, &bs, NULL, &bytes_per_inode, NULL, NULL, NULL, NULL,
186 &nreserved, NULL, NULL, &label, NULL, NULL, NULL, NULL, NULL, NULL);
187 argv += optind; // argv[0] -- device
189 // block size minimax, block size is a multiple of minimum
191 if (blocksize < EXT2_MIN_BLOCK_SIZE
192 || blocksize > EXT2_MAX_BLOCK_SIZE
193 || (blocksize & (blocksize - 1)) // not power of 2
195 bb_error_msg_and_die("-%c is bad", 'b');
198 // reserved blocks count
200 bb_error_msg_and_die("-%c is bad", 'm');
202 // check the device is a block device
204 if (!S_ISBLK(st.st_mode) && !(opts & OPT_F))
205 bb_error_msg_and_die("not a block device");
207 // check if it is mounted
208 // N.B. what if we format a file? find_mount_point will return false negative since
209 // it is loop block device which mounted!
210 if (find_mount_point(argv[0], 0))
211 bb_error_msg_and_die("can't format mounted filesystem");
213 // TODO: 5?/5 WE MUST NOT DEPEND ON WHETHER DEVICE IS /dev/zero 'ed OR NOT
214 // TODO: 3/5 refuse if mounted
215 // TODO: 4/5 compat options
216 // TODO: 1/5 sanity checks
217 // TODO: 0/5 more verbose error messages
218 // TODO: 0/5 info printing
219 // TODO: 2/5 bigendianness! Spot where it comes to play! sb->, gd->
220 // TODO: 2/5 reserved GDT: how to mark but not allocate?
221 // TODO: 3/5 dir_index?
223 // fill the superblock
224 sb = xzalloc(blocksize);
225 sb->s_rev_level = 1; // revision 1 filesystem
226 sb->s_magic = EXT2_SUPER_MAGIC;
227 sb->s_inode_size = sizeof(*inode);
228 sb->s_first_ino = EXT2_GOOD_OLD_FIRST_INO;
229 blocksize_log2 = int_log2(blocksize);
230 sb->s_log_block_size = sb->s_log_frag_size = blocksize_log2 - EXT2_MIN_BLOCK_LOG_SIZE;
231 // first 1024 bytes of the device are for boot record. If block size is 1024 bytes, then
232 // the first block available for data is 1, otherwise 0
233 first_data_block = sb->s_first_data_block = (EXT2_MIN_BLOCK_SIZE == blocksize);
234 // block and inode bitmaps occupy no more than one block, so maximum number of blocks is
235 // number of bits in one block, i.e. 8*blocksize
236 nblocks_per_group = sb->s_blocks_per_group = sb->s_frags_per_group = sb->s_inodes_per_group = 8 * blocksize;
237 timestamp = time(NULL);
238 sb->s_mkfs_time = sb->s_wtime = sb->s_lastcheck = timestamp;
240 sb->s_creator_os = EXT2_OS_LINUX;
241 sb->s_max_mnt_count = EXT2_DFL_MAX_MNT_COUNT;
242 sb->s_checkinterval = 24*60*60 * 180; // 180 days
243 sb->s_errors = EXT2_ERRORS_DEFAULT;
244 sb->s_feature_compat = EXT2_FEATURE_COMPAT_SUPP
245 | (EXT2_FEATURE_COMPAT_RESIZE_INO * ENABLE_FEATURE_MKFS_EXT2_RESERVED_GDT)
246 | (EXT2_FEATURE_COMPAT_DIR_INDEX * ENABLE_FEATURE_MKFS_EXT2_DIR_INDEX)
248 // e2fsprogs-1.41.9 doesn't like EXT2_FEATURE_INCOMPAT_WHITEOUT
249 sb->s_feature_incompat = EXT2_FEATURE_INCOMPAT_FILETYPE;// | EXT2_FEATURE_INCOMPAT_WHITEOUT;
250 sb->s_feature_ro_compat = EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER;
251 sb->s_flags = EXT2_FLAGS_SIGNED_HASH * ENABLE_FEATURE_MKFS_EXT2_DIR_INDEX;
252 generate_uuid(sb->s_uuid);
253 if (ENABLE_FEATURE_MKFS_EXT2_DIR_INDEX) {
254 sb->s_def_hash_version = EXT2_HASH_HALF_MD4;
255 generate_uuid((uint8_t *)sb->s_hash_seed);
258 * From e2fsprogs: add "jitter" to the superblock's check interval so that we
259 * don't check all the filesystems at the same time. We use a
260 * kludgy hack of using the UUID to derive a random jitter value.
262 sb->s_max_mnt_count += sb->s_uuid[ARRAY_SIZE(sb->s_uuid)-1] % EXT2_DFL_MAX_MNT_COUNT;
264 // open the device, get number of blocks
265 xmove_fd(xopen3(argv[0], O_WRONLY | O_CREAT, 0666), fd);
267 nblocks = xatou(argv[1]);
269 nblocks = (uoff_t)xlseek(fd, 0, SEEK_END) >> blocksize_log2;
271 sb->s_blocks_count = nblocks;
273 // nblocks is the total number of blocks in the filesystem
275 bb_error_msg_and_die("nblocks");
276 // reserve blocks for superuser
277 sb->s_r_blocks_count = ((uint64_t) nblocks * nreserved) / 100;
279 // N.B. a block group can have no more than nblocks_per_group blocks
280 ngroups = div_roundup(nblocks - first_data_block, nblocks_per_group);
282 bb_error_msg_and_die("ngroups");
283 gdtsz = div_roundup(ngroups, EXT2_DESC_PER_BLOCK(sb));
285 * From e2fsprogs: Calculate the number of GDT blocks to reserve for online
287 * The absolute maximum number of GDT blocks we can reserve is determined by
288 * the number of block pointers that can fit into a single block.
290 /* We set it at 1024x the current filesystem size, or
291 * the upper block count limit (2^32), whichever is lower.
293 #if ENABLE_FEATURE_MKFS_EXT2_RESERVED_GDT
294 rgdtsz = 0xFFFFFFFF; // maximum block number
295 if (nblocks < rgdtsz / 1024)
296 rgdtsz = nblocks * 1024;
297 rgdtsz = div_roundup(rgdtsz - first_data_block, nblocks_per_group);
298 rgdtsz = div_roundup(rgdtsz, EXT2_DESC_PER_BLOCK(sb)) - gdtsz;
299 if (rgdtsz > EXT2_ADDR_PER_BLOCK(sb))
300 rgdtsz = EXT2_ADDR_PER_BLOCK(sb);
301 sb->s_reserved_gdt_blocks = rgdtsz;
302 //bb_info_msg("RSRVD[%u]", n);
307 // ninodes is the total number of inodes (files) in the file system
308 if (!(opts & OPT_i)) {
309 bytes_per_inode = 16384;
310 if (nblocks < 512*1024)
311 bytes_per_inode = 4096;
312 if (nblocks < 3*1024)
313 bytes_per_inode = 8192;
315 ninodes = nblocks / (bytes_per_inode >> blocksize_log2);
316 if (ninodes < EXT2_GOOD_OLD_FIRST_INO+1)
317 ninodes = EXT2_GOOD_OLD_FIRST_INO+1;
318 ninodes_per_group = div_roundup(ninodes, ngroups);
319 if (ninodes_per_group < 16)
320 ninodes_per_group = 16; // minimum number because the first EXT2_GOOD_OLD_FIRST_INO-1 are reserved
321 // N.B. a block group can have no more than 8*blocksize = sb->s_inodes_per_group inodes
322 if (ninodes_per_group > sb->s_inodes_per_group)
323 ninodes_per_group = sb->s_inodes_per_group;
324 // adjust inodes per group so they completely fill the inode table blocks in the descriptor
325 ninodes_per_group = (div_roundup(ninodes_per_group * EXT2_INODE_SIZE(sb), blocksize) << blocksize_log2) / EXT2_INODE_SIZE(sb);
326 // make sure the number of inodes per group is a multiple of 8
327 ninodes_per_group &= ~7;
328 sb->s_inodes_per_group = ninodes_per_group;// = div_roundup(ninodes_per_group * sb->s_inode_size, blocksize);
330 ninodes = sb->s_inodes_count = ninodes_per_group * ngroups;
332 itsz = ninodes_per_group * sb->s_inode_size / blocksize;
333 sb->s_free_inodes_count = sb->s_inodes_count - EXT2_GOOD_OLD_FIRST_INO;
335 // write the label, if any
337 safe_strncpy((char *)sb->s_volume_name, label, sizeof(sb->s_volume_name));
340 /* if (fs_param.s_blocks_count != s->s_blocks_count)
341 fprintf(stderr, _("warning: %u blocks unused.\n\n"),
342 fs_param.s_blocks_count - s->s_blocks_count);
346 "Filesystem label=%s\n"
348 "Block size=%u (log=%u)\n"
349 "Fragment size=%u (log=%u)\n"
350 "%u inodes, %u blocks\n"
351 "%u blocks (%u%%) reserved for the super user\n"
352 "First data block=%u\n"
353 // "Maximum filesystem blocks=%lu\n"
355 "%u blocks per group, %u fragments per group\n"
356 "%u inodes per group"
357 , (char *)sb->s_volume_name
358 , blocksize, sb->s_log_block_size
359 , blocksize, sb->s_log_block_size
360 , sb->s_inodes_count, sb->s_blocks_count
361 , sb->s_r_blocks_count, nreserved
363 // , (rgdtsz + gdtsz) * EXT2_DESC_PER_BLOCK(sb) * nblocks_per_group
365 , nblocks_per_group, nblocks_per_group
369 const char *fmt = "\nSuperblock backups stored on blocks: %u";
370 pos = first_data_block;
371 for (i = 1; i < ngroups; i++) {
372 pos += nblocks_per_group;
374 printf(fmt, (unsigned)pos);
385 // fill group descriptors
386 gd = xzalloc((gdtsz + rgdtsz) * blocksize);
387 sb->s_free_blocks_count = 0;
388 for (i = 0, pos = first_data_block, n = nblocks;
390 i++, pos += nblocks_per_group, n -= nblocks_per_group
392 uint32_t overhead = pos + (has_super(i) ? (1/*sb*/ + gdtsz + rgdtsz) : 0);
393 gd[i].bg_block_bitmap = overhead + 0;
394 gd[i].bg_inode_bitmap = overhead + 1;
395 gd[i].bg_inode_table = overhead + 2;
396 overhead = overhead - pos + 1/*bbmp*/ + 1/*ibmp*/ + itsz;
397 gd[i].bg_free_inodes_count = ninodes_per_group;
398 //gd[i].bg_used_dirs_count = 0;
399 // N.B. both root and lost+found dirs are within the first block group, thus +2
402 gd[i].bg_used_dirs_count = 2;
403 gd[i].bg_free_inodes_count -= EXT2_GOOD_OLD_FIRST_INO;
405 // N.B. the following is pure heuristics!
406 // Likely to cope with 1024-byte blocks, when first block is for boot sectors
407 if (ngroups-1 == i) {
408 overhead += first_data_block;
410 gd[i].bg_free_blocks_count = (n < nblocks_per_group ? n : nblocks_per_group) - overhead;
411 sb->s_free_blocks_count += gd[i].bg_free_blocks_count;
413 STORE_LE(sb->s_free_blocks_count, sb->s_free_blocks_count);
415 // dump filesystem skeleton structures
416 // printf("Writing superblocks and filesystem accounting information: ");
417 buf = xmalloc(blocksize);
418 for (i = 0, pos = first_data_block; i < ngroups; i++, pos += nblocks_per_group) {
419 uint32_t overhead = has_super(i) ? (1/*sb*/ + gdtsz + rgdtsz) : 0;
420 uint32_t start;// = has_super(i) ? (1/*sb*/ + gdtsz + rgdtsz) : 0;
423 // dump superblock and group descriptors and their backups
424 if (overhead) { // N.B. in fact, we want (has_super(i)) condition, but it is equal to (overhead != 0) and is cheaper
425 // N.B. 1024 byte blocks are special
426 PUT(((uint64_t)pos << blocksize_log2) + ((0 == i && 0 == first_data_block) ? 1024 : 0), sb, 1024);//blocksize);
427 PUT(((uint64_t)pos << blocksize_log2) + blocksize, gd, (gdtsz + rgdtsz) * blocksize);
430 start = overhead + 1/*bbmp*/ + 1/*ibmp*/ + itsz;
432 start += 2; // for "/" and "/lost+found"
433 end = nblocks_per_group - (start + gd[i].bg_free_blocks_count);
435 // mark preallocated blocks as allocated
436 allocate(buf, blocksize, start, end);
438 PUT((uint64_t)(pos + overhead) * blocksize, buf, blocksize);
440 // mark preallocated inodes as allocated
441 allocate(buf, blocksize,
442 ninodes_per_group - gd[i].bg_free_inodes_count,
443 8 * blocksize - ninodes_per_group
446 //PUT((uint64_t)(pos + overhead + 1) * blocksize, buf, blocksize);
447 //but it's right after block bitmap, so we can just:
448 xwrite(fd, buf, blocksize);
452 memset(buf, 0, blocksize);
453 PUT(0, buf, 1024); // N.B. 1024 <= blocksize
455 for (i = 0; i < ngroups; ++i)
456 for (n = 0; n < itsz; ++n)
457 PUT((uint64_t)(gd[i].bg_inode_table + n) * blocksize, buf, blocksize);
459 // prepare directory inode
460 inode = (struct ext2_inode *)buf;
461 STORE_LE(inode->i_mode, S_IFDIR | S_IRWXU | S_IRGRP | S_IROTH | S_IXGRP | S_IXOTH);
462 STORE_LE(inode->i_mtime, timestamp);
463 STORE_LE(inode->i_atime, timestamp);
464 STORE_LE(inode->i_ctime, timestamp);
465 STORE_LE(inode->i_size, blocksize);
466 // N.B. inode->i_blocks stores the number of 512 byte data blocks. Why on Earth?!
467 STORE_LE(inode->i_blocks, blocksize / 512);
469 // dump root dir inode
470 STORE_LE(inode->i_links_count, 3); // "/.", "/..", "/lost+found/.." point to this inode
471 STORE_LE(inode->i_block[0], gd[0].bg_inode_table + itsz);
472 PUT(((uint64_t)gd[0].bg_inode_table << blocksize_log2) + (EXT2_ROOT_INO-1) * sizeof(*inode), buf, sizeof(*inode));
474 // dump lost+found dir inode
475 STORE_LE(inode->i_links_count, 2); // both "/lost+found" and "/lost+found/." point to this inode
476 STORE_LE(inode->i_block[0], inode->i_block[0]+1); // use next block //= gd[0].bg_inode_table + itsz + 1;
477 PUT(((uint64_t)gd[0].bg_inode_table << blocksize_log2) + (EXT2_GOOD_OLD_FIRST_INO-1) * sizeof(*inode), buf, sizeof(*inode));
480 memset(buf, 0, blocksize);
481 dir = (struct ext2_dir *)buf;
483 // dump lost+found dir block
484 STORE_LE(dir->inode1, EXT2_GOOD_OLD_FIRST_INO);
485 STORE_LE(dir->rec_len1, 12);
486 STORE_LE(dir->name_len1, 1);
487 STORE_LE(dir->file_type1, EXT2_FT_DIR);
489 STORE_LE(dir->inode2, EXT2_ROOT_INO);
490 STORE_LE(dir->rec_len2, blocksize - 12);
491 STORE_LE(dir->name_len2, 2);
492 STORE_LE(dir->file_type2, EXT2_FT_DIR);
493 dir->name2[0] = '.'; dir->name2[1] = '.';
494 PUT((uint64_t)(gd[0].bg_inode_table + itsz + 1) * blocksize, buf, blocksize);
496 // dump root dir block
497 STORE_LE(dir->inode1, EXT2_ROOT_INO);
498 STORE_LE(dir->rec_len2, 12);
499 STORE_LE(dir->inode3, EXT2_GOOD_OLD_FIRST_INO);
500 STORE_LE(dir->rec_len3, blocksize - 12 - 12);
501 STORE_LE(dir->name_len3, 10);
502 STORE_LE(dir->file_type3, EXT2_FT_DIR);
503 strcpy(dir->name3, "lost+found");
504 PUT((uint64_t)(gd[0].bg_inode_table + itsz + 0) * blocksize, buf, blocksize);
506 // bb_info_msg("done\n"
507 // "This filesystem will be automatically checked every %u mounts or\n"
508 // "%u days, whichever comes first. Use tune2fs -c or -i to override.",
509 // sb->s_max_mnt_count, sb->s_checkinterval / (3600 * 24)
513 if (ENABLE_FEATURE_CLEAN_UP) {