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 <sys/user.h> /* PAGE_SIZE */
15 # define PAGE_SIZE 4096
17 #include "volume_id/volume_id_internal.h"
19 #define ENABLE_FEATURE_MKFS_EXT2_RESERVED_GDT 0
20 #define ENABLE_FEATURE_MKFS_EXT2_DIR_INDEX 1
23 #define s_reserved_gdt_blocks s_padding1
24 #define s_mkfs_time s_reserved[0]
25 #define s_flags s_reserved[22]
26 #define EXT2_HASH_HALF_MD4 1
27 #define EXT2_FLAGS_SIGNED_HASH 0x0001
29 // whiteout: for writable overlays
30 //#define LINUX_S_IFWHT 0160000
31 //#define EXT2_FEATURE_INCOMPAT_WHITEOUT 0x0020
34 char BUG_wrong_field_size(void);
35 #define STORE_LE(field, value) \
37 if (sizeof(field) == 4) \
38 field = cpu_to_le32(value); \
39 else if (sizeof(field) == 2) \
40 field = cpu_to_le16(value); \
41 else if (sizeof(field) == 1) \
44 BUG_wrong_field_size(); \
47 #define FETCH_LE32(field) \
48 (sizeof(field) == 4 ? cpu_to_le32(field) : BUG_wrong_field_size())
50 // All fields are little-endian
69 static unsigned int_log2(unsigned arg)
72 while ((arg >>= 1) != 0)
77 // taken from mkfs_minix.c. libbb candidate?
78 // why "uint64_t size"? we never use it for anything >32 bits
79 static uint32_t div_roundup(uint64_t size, uint32_t n)
82 uint32_t res = size / n;
88 static void allocate(uint8_t *bitmap, uint32_t blocksize, uint32_t start, uint32_t end)
91 memset(bitmap, 0, blocksize);
93 memset(bitmap, 0xFF, i);
94 bitmap[i] = 0xFF >> (8 - (start & 7));
95 //bb_info_msg("ALLOC: [%u][%u][%u]: [%u-%u]:=[%x],[%x]", blocksize, start, end, start/8, blocksize - end/8 - 1, 0xFF >> (8 - (start & 7)), (uint8_t)(0xFF << (8-(end&7))));
97 bitmap[blocksize - i - 1] = 0xFF << (8 - (end & 7));
98 memset(bitmap + blocksize - i, 0xFF, i); // N.B. no overflow here!
101 static uint32_t has_super(uint32_t x)
103 // 0, 1 and powers of 3, 5, 7 up to 2^32 limit
104 static const uint32_t supers[] = {
105 0, 1, 3, 5, 7, 9, 25, 27, 49, 81, 125, 243, 343, 625, 729,
106 2187, 2401, 3125, 6561, 15625, 16807, 19683, 59049, 78125,
107 117649, 177147, 390625, 531441, 823543, 1594323, 1953125,
108 4782969, 5764801, 9765625, 14348907, 40353607, 43046721,
109 48828125, 129140163, 244140625, 282475249, 387420489,
110 1162261467, 1220703125, 1977326743, 3486784401/* >2^31 */,
112 const uint32_t *sp = supers + ARRAY_SIZE(supers);
122 /* Standard mke2fs 1.41.9:
123 * Usage: mke2fs [-c|-l filename] [-b block-size] [-f fragment-size]
124 * [-i bytes-per-inode] [-I inode-size] [-J journal-options]
125 * [-G meta group size] [-N number-of-inodes]
126 * [-m reserved-blocks-percentage] [-o creator-os]
127 * [-g blocks-per-group] [-L volume-label] [-M last-mounted-directory]
128 * [-O feature[,...]] [-r fs-revision] [-E extended-option[,...]]
129 * [-T fs-type] [-U UUID] [-jnqvFSV] device [blocks-count]
131 // N.B. not commented below options are taken and silently ignored
135 OPT_b = 1 << 2, // block size, in bytes
137 OPT_i = 1 << 4, // bytes per inode
142 OPT_m = 1 << 9, // percentage of blocks reserved for superuser
145 OPT_L = 1 << 12, // label
153 OPT_n = 1 << 20, // dry run: do not write anything
158 //OPT_V = 1 << 25, // -V version. bbox applets don't support that
161 #define fd 3 /* predefined output descriptor */
163 static void PUT(uint64_t off, void *buf, uint32_t size)
165 // bb_info_msg("PUT[%llu]:[%u]", off, size);
166 xlseek(fd, off, SEEK_SET);
167 xwrite(fd, buf, size);
170 int mkfs_ext2_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
171 int mkfs_ext2_main(int argc UNUSED_PARAM, char **argv)
175 unsigned blocksize, blocksize_log2;
176 unsigned reserved_percent = 5;
177 unsigned long long kilobytes;
178 uint32_t nblocks, nblocks_full, nreserved;
180 uint32_t bytes_per_inode;
181 uint32_t first_data_block;
182 uint32_t inodes_per_group;
183 uint32_t gdtsz, itsz;
186 const char *label = "";
188 struct ext2_super_block *sb; // superblock
189 struct ext2_group_desc *gd; // group descriptors
190 struct ext2_inode *inode;
191 struct ext2_dir *dir;
194 opt_complementary = "-1:b+:m+:i+";
195 opts = getopt32(argv, "cl:b:f:i:I:J:G:N:m:o:g:L:M:O:r:E:T:U:jnqvFS",
196 NULL, &bs, NULL, &bpi, NULL, NULL, NULL, NULL,
197 &reserved_percent, NULL, NULL, &label, NULL, NULL, NULL, NULL, NULL, NULL);
198 argv += optind; // argv[0] -- device
200 // check the device is a block device
201 xmove_fd(xopen(argv[0], O_WRONLY), fd);
203 if (!S_ISBLK(st.st_mode) && !(opts & OPT_F))
204 bb_error_msg_and_die("not a block device");
206 // check if it is mounted
207 // N.B. what if we format a file? find_mount_point will return false negative since
208 // it is loop block device which mounted!
209 if (find_mount_point(argv[0], 0))
210 bb_error_msg_and_die("can't format mounted filesystem");
212 // open the device, get size in kbytes
214 kilobytes = xatoull(argv[1]);
216 kilobytes = (uoff_t)xlseek(fd, 0, SEEK_END) / 1024;
219 bytes_per_inode = 16384;
220 if (kilobytes < 512*1024)
221 bytes_per_inode = 4096;
222 if (kilobytes < 3*1024)
223 bytes_per_inode = 8192;
225 bytes_per_inode = bpi;
227 // Determine block size
228 // block size is a multiple of 1024
230 if (kilobytes >= 512*1024) // mke2fs 1.41.9 compat
232 if (EXT2_MAX_BLOCK_SIZE > 4096) {
233 // kilobytes >> 22 == size in 4gigabyte chunks.
234 // if size >= 16k gigs, blocksize must be increased.
235 // Try "mke2fs -F image $((16 * 1024*1024*1024))"
236 while ((kilobytes >> 22) >= blocksize)
241 if (blocksize < EXT2_MIN_BLOCK_SIZE
242 || blocksize > EXT2_MAX_BLOCK_SIZE
243 || (blocksize & (blocksize - 1)) // not power of 2
245 bb_error_msg_and_die("blocksize %u is bad", blocksize);
247 // number of bits in one block, i.e. 8*blocksize
248 #define blocks_per_group (8 * blocksize)
249 first_data_block = (EXT2_MIN_BLOCK_SIZE == blocksize);
250 blocksize_log2 = int_log2(blocksize);
252 // Determine number of blocks
253 kilobytes >>= (blocksize_log2 - EXT2_MIN_BLOCK_LOG_SIZE);
255 if (nblocks != kilobytes)
256 bb_error_msg_and_die("block count doesn't fit in 32 bits");
257 #define kilobytes kilobytes_unused_after_this
258 if (blocksize < PAGE_SIZE)
259 nblocks &= ~((PAGE_SIZE >> blocksize_log2)-1);
260 // Experimentally, standard mke2fs won't work on images smaller than 60k
262 bb_error_msg_and_die("need >= 60 blocks");
264 // How many reserved blocks?
265 if (reserved_percent > 50)
266 bb_error_msg_and_die("-%c is bad", 'm');
267 //nreserved = div_roundup((uint64_t) nblocks * reserved_percent, 100);
268 nreserved = (uint64_t)nblocks * reserved_percent / 100;
270 // N.B. killing e2fsprogs feature! Unused blocks don't account in calculations
271 nblocks_full = nblocks;
273 // If last block group is too small, nblocks may be decreased in order
274 // to discard it, and control returns here to recalculate some
276 // Note: blocksize and bytes_per_inode are never recalculated.
278 // N.B. a block group can have no more than blocks_per_group blocks
279 ngroups = div_roundup(nblocks - first_data_block, blocks_per_group);
281 bb_error_msg_and_die("ngroups");
283 gdtsz = div_roundup(ngroups, blocksize / sizeof(*gd));
284 // TODO: reserved blocks must be marked as such in the bitmaps,
285 // or resulting filesystem is corrupt
286 if (ENABLE_FEATURE_MKFS_EXT2_RESERVED_GDT) {
288 * From e2fsprogs: Calculate the number of GDT blocks to reserve for online
290 * The absolute maximum number of GDT blocks we can reserve is determined by
291 * the number of block pointers that can fit into a single block.
292 * We set it at 1024x the current filesystem size, or
293 * the upper block count limit (2^32), whichever is lower.
295 uint32_t rgdtsz = 0xFFFFFFFF; // maximum block number
296 if (nblocks < rgdtsz / 1024)
297 rgdtsz = nblocks * 1024;
298 rgdtsz = div_roundup(rgdtsz - first_data_block, blocks_per_group);
299 rgdtsz = div_roundup(rgdtsz, blocksize / sizeof(*gd)) - gdtsz;
300 if (rgdtsz > blocksize / sizeof(uint32_t))
301 rgdtsz = blocksize / sizeof(uint32_t);
302 //TODO: STORE_LE(sb->s_reserved_gdt_blocks, rgdtsz);
307 // N.B. e2fsprogs does as follows!
308 // ninodes is the total number of inodes (files) in the file system
309 uint32_t ninodes = nblocks_full / (blocksize >= 4096 ? 1 : 4096 / blocksize);
310 uint32_t overhead, remainder;
311 if (ninodes < EXT2_GOOD_OLD_FIRST_INO+1)
312 ninodes = EXT2_GOOD_OLD_FIRST_INO+1;
313 inodes_per_group = div_roundup(ninodes, ngroups);
314 // minimum number because the first EXT2_GOOD_OLD_FIRST_INO-1 are reserved
315 if (inodes_per_group < 16)
316 inodes_per_group = 16;
318 // a block group can have no more than 8*blocksize inodes
319 if (inodes_per_group > blocks_per_group)
320 inodes_per_group = blocks_per_group;
321 // adjust inodes per group so they completely fill the inode table blocks in the descriptor
322 inodes_per_group = (div_roundup(inodes_per_group * sizeof(*inode), blocksize) * blocksize) / sizeof(*inode);
323 // make sure the number of inodes per group is a multiple of 8
324 inodes_per_group &= ~7;
325 itsz = div_roundup(inodes_per_group * sizeof(*inode), blocksize);
327 // the last group needs more attention: isn't it too small for possible overhead?
328 overhead = (has_super(ngroups - 1) ? (1/*sb*/ + gdtsz) : 0) + 1/*bbmp*/ + 1/*ibmp*/ + itsz;
329 remainder = (nblocks - first_data_block) % blocks_per_group;
330 if ((1 == ngroups) && remainder && (remainder < overhead))
331 bb_error_msg_and_die("way small device");
332 if (remainder && (remainder < overhead + 50/* e2fsprogs hardcoded */)) {
333 //bb_info_msg("CHOP[%u]", remainder);
334 nblocks -= remainder;
340 if (nblocks_full - nblocks)
341 printf("warning: %u blocks unused\n\n", nblocks_full - nblocks);
343 "Filesystem label=%s\n"
345 "Block size=%u (log=%u)\n"
346 "Fragment size=%u (log=%u)\n"
347 "%u inodes, %u blocks\n"
348 "%u blocks (%u%%) reserved for the super user\n"
349 "First data block=%u\n"
350 "Maximum filesystem blocks=%u\n"
352 "%u blocks per group, %u fragments per group\n"
353 "%u inodes per group"
355 , blocksize, blocksize_log2 - EXT2_MIN_BLOCK_LOG_SIZE
356 , blocksize, blocksize_log2 - EXT2_MIN_BLOCK_LOG_SIZE
357 , inodes_per_group * ngroups, nblocks
358 , nreserved, reserved_percent
360 , gdtsz * (blocksize / sizeof(*gd)) * blocks_per_group
362 , blocks_per_group, blocks_per_group
366 const char *fmt = "\nSuperblock backups stored on blocks:\n"
368 pos = first_data_block;
369 for (i = 1; i < ngroups; i++) {
370 pos += blocks_per_group;
372 printf(fmt, (unsigned)pos);
379 // dry run? -> we are done
383 // TODO: 3/5 refuse if mounted
384 // TODO: 4/5 compat options
385 // TODO: 1/5 sanity checks
386 // TODO: 0/5 more verbose error messages
387 // TODO: 4/5 bigendianness: recheck, wait for ARM reporters
388 // TODO: 2/5 reserved GDT: how to mark but not allocate?
389 // TODO: 3/5 dir_index?
391 // fill the superblock
392 sb = xzalloc(blocksize);
393 STORE_LE(sb->s_rev_level, 1); // revision 1 filesystem
394 STORE_LE(sb->s_magic, EXT2_SUPER_MAGIC);
395 STORE_LE(sb->s_inode_size, sizeof(*inode));
396 STORE_LE(sb->s_first_ino, EXT2_GOOD_OLD_FIRST_INO);
397 STORE_LE(sb->s_log_block_size, blocksize_log2 - EXT2_MIN_BLOCK_LOG_SIZE);
398 STORE_LE(sb->s_log_frag_size, blocksize_log2 - EXT2_MIN_BLOCK_LOG_SIZE);
399 // first 1024 bytes of the device are for boot record. If block size is 1024 bytes, then
400 // the first block available for data is 1, otherwise 0
401 STORE_LE(sb->s_first_data_block, first_data_block); // 0 or 1
402 // block and inode bitmaps occupy no more than one block, so maximum number of blocks is
403 STORE_LE(sb->s_blocks_per_group, blocks_per_group);
404 STORE_LE(sb->s_frags_per_group, blocks_per_group);
406 STORE_LE(sb->s_blocks_count, nblocks);
407 // reserve blocks for superuser
408 STORE_LE(sb->s_r_blocks_count, nreserved);
410 STORE_LE(sb->s_inodes_per_group, inodes_per_group);
411 STORE_LE(sb->s_inodes_count, inodes_per_group * ngroups);
412 STORE_LE(sb->s_free_inodes_count, inodes_per_group * ngroups - EXT2_GOOD_OLD_FIRST_INO);
414 timestamp = time(NULL);
415 STORE_LE(sb->s_mkfs_time, timestamp);
416 STORE_LE(sb->s_wtime, timestamp);
417 STORE_LE(sb->s_lastcheck, timestamp);
419 STORE_LE(sb->s_state, 1); // TODO: what's 1?
420 STORE_LE(sb->s_creator_os, EXT2_OS_LINUX);
421 STORE_LE(sb->s_checkinterval, 24*60*60 * 180); // 180 days
422 STORE_LE(sb->s_errors, EXT2_ERRORS_DEFAULT);
423 STORE_LE(sb->s_feature_compat, EXT2_FEATURE_COMPAT_SUPP
424 | (EXT2_FEATURE_COMPAT_RESIZE_INO * ENABLE_FEATURE_MKFS_EXT2_RESERVED_GDT)
425 | (EXT2_FEATURE_COMPAT_DIR_INDEX * ENABLE_FEATURE_MKFS_EXT2_DIR_INDEX)
427 // e2fsck from 1.41.9 doesn't like EXT2_FEATURE_INCOMPAT_WHITEOUT
428 STORE_LE(sb->s_feature_incompat, EXT2_FEATURE_INCOMPAT_FILETYPE);// | EXT2_FEATURE_INCOMPAT_WHITEOUT;
429 STORE_LE(sb->s_feature_ro_compat, EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER);
430 STORE_LE(sb->s_flags, EXT2_FLAGS_SIGNED_HASH * ENABLE_FEATURE_MKFS_EXT2_DIR_INDEX);
431 generate_uuid(sb->s_uuid);
432 if (ENABLE_FEATURE_MKFS_EXT2_DIR_INDEX) {
433 STORE_LE(sb->s_def_hash_version, EXT2_HASH_HALF_MD4);
434 generate_uuid((uint8_t *)sb->s_hash_seed);
437 * From e2fsprogs: add "jitter" to the superblock's check interval so that we
438 * don't check all the filesystems at the same time. We use a
439 * kludgy hack of using the UUID to derive a random jitter value.
441 STORE_LE(sb->s_max_mnt_count, EXT2_DFL_MAX_MNT_COUNT
442 + (sb->s_uuid[ARRAY_SIZE(sb->s_uuid)-1] % EXT2_DFL_MAX_MNT_COUNT));
445 safe_strncpy((char *)sb->s_volume_name, label, sizeof(sb->s_volume_name));
447 // fill group descriptors
448 gd = xzalloc(gdtsz * blocksize);
449 buf = xmalloc(blocksize);
450 sb->s_free_blocks_count = 0;
451 for (i = 0, pos = first_data_block, n = nblocks - first_data_block;
453 i++, pos += blocks_per_group, n -= blocks_per_group
455 uint32_t overhead = pos + (has_super(i) ? (1/*sb*/ + gdtsz) : 0);
457 STORE_LE(gd[i].bg_block_bitmap, overhead + 0);
458 STORE_LE(gd[i].bg_inode_bitmap, overhead + 1);
459 STORE_LE(gd[i].bg_inode_table, overhead + 2);
460 overhead = overhead - pos + 1/*bbmp*/ + 1/*ibmp*/ + itsz;
461 gd[i].bg_free_inodes_count = inodes_per_group;
462 //STORE_LE(gd[i].bg_used_dirs_count, 0);
463 // N.B. both root and lost+found dirs are within the first block group, thus +2
466 STORE_LE(gd[i].bg_used_dirs_count, 2);
467 gd[i].bg_free_inodes_count -= EXT2_GOOD_OLD_FIRST_INO;
470 // mark preallocated blocks as allocated
471 fb = (n < blocks_per_group ? n : blocks_per_group) - overhead;
472 //bb_info_msg("ALLOC: [%u][%u][%u]", blocksize, overhead, blocks_per_group - (fb + overhead));
473 allocate(buf, blocksize,
475 blocks_per_group - (fb + overhead)
478 PUT((uint64_t)(FETCH_LE32(gd[i].bg_block_bitmap)) * blocksize, buf, blocksize);
479 STORE_LE(gd[i].bg_free_blocks_count, fb);
481 // mark preallocated inodes as allocated
482 allocate(buf, blocksize,
483 inodes_per_group - gd[i].bg_free_inodes_count,
484 blocks_per_group - inodes_per_group
487 //PUT((uint64_t)(FETCH_LE32(gd[i].bg_block_bitmap)) * blocksize, buf, blocksize);
488 //but it's right after block bitmap, so we can just:
489 xwrite(fd, buf, blocksize);
490 STORE_LE(gd[i].bg_free_inodes_count, gd[i].bg_free_inodes_count);
492 // count overall free blocks
493 sb->s_free_blocks_count += fb;
495 STORE_LE(sb->s_free_blocks_count, sb->s_free_blocks_count);
497 // dump filesystem skeleton structures
498 // printf("Writing superblocks and filesystem accounting information: ");
499 for (i = 0, pos = first_data_block; i < ngroups; i++, pos += blocks_per_group) {
500 // dump superblock and group descriptors and their backups
502 // N.B. 1024 byte blocks are special
503 PUT(((uint64_t)pos * blocksize) + ((0 == i && 0 == first_data_block) ? 1024 : 0), sb, 1024);//blocksize);
504 PUT(((uint64_t)pos * blocksize) + blocksize, gd, gdtsz * blocksize);
509 memset(buf, 0, blocksize);
510 PUT(0, buf, 1024); // N.B. 1024 <= blocksize, so buf[0..1023] contains zeros
512 for (i = 0; i < ngroups; ++i)
513 for (n = 0; n < itsz; ++n)
514 PUT((uint64_t)(FETCH_LE32(gd[i].bg_inode_table) + n) * blocksize, buf, blocksize);
516 // prepare directory inode
517 inode = (struct ext2_inode *)buf;
518 STORE_LE(inode->i_mode, S_IFDIR | S_IRWXU | S_IRGRP | S_IROTH | S_IXGRP | S_IXOTH);
519 STORE_LE(inode->i_mtime, timestamp);
520 STORE_LE(inode->i_atime, timestamp);
521 STORE_LE(inode->i_ctime, timestamp);
522 STORE_LE(inode->i_size, blocksize);
523 // N.B. inode->i_blocks stores the number of 512 byte data blocks. Why on Earth?!
524 STORE_LE(inode->i_blocks, blocksize / 512);
526 // dump root dir inode
527 STORE_LE(inode->i_links_count, 3); // "/.", "/..", "/lost+found/.." point to this inode
528 STORE_LE(inode->i_block[0], FETCH_LE32(gd[0].bg_inode_table) + itsz);
529 PUT(((uint64_t)FETCH_LE32(gd[0].bg_inode_table) * blocksize) + (EXT2_ROOT_INO-1) * sizeof(*inode), buf, sizeof(*inode));
531 // dump lost+found dir inode
532 STORE_LE(inode->i_links_count, 2); // both "/lost+found" and "/lost+found/." point to this inode
533 STORE_LE(inode->i_block[0], inode->i_block[0]+1); // use next block
534 PUT(((uint64_t)FETCH_LE32(gd[0].bg_inode_table) * blocksize) + (EXT2_GOOD_OLD_FIRST_INO-1) * sizeof(*inode), buf, sizeof(*inode));
537 memset(buf, 0, blocksize);
538 dir = (struct ext2_dir *)buf;
540 // dump lost+found dir block
541 STORE_LE(dir->inode1, EXT2_GOOD_OLD_FIRST_INO);
542 STORE_LE(dir->rec_len1, 12);
543 STORE_LE(dir->name_len1, 1);
544 STORE_LE(dir->file_type1, EXT2_FT_DIR);
546 STORE_LE(dir->inode2, EXT2_ROOT_INO);
547 STORE_LE(dir->rec_len2, blocksize - 12);
548 STORE_LE(dir->name_len2, 2);
549 STORE_LE(dir->file_type2, EXT2_FT_DIR);
550 dir->name2[0] = '.'; dir->name2[1] = '.';
551 PUT((uint64_t)(FETCH_LE32(gd[0].bg_inode_table) + itsz + 1) * blocksize, buf, blocksize);
553 // dump root dir block
554 STORE_LE(dir->inode1, EXT2_ROOT_INO);
555 STORE_LE(dir->rec_len2, 12);
556 STORE_LE(dir->inode3, EXT2_GOOD_OLD_FIRST_INO);
557 STORE_LE(dir->rec_len3, blocksize - 12 - 12);
558 STORE_LE(dir->name_len3, 10);
559 STORE_LE(dir->file_type3, EXT2_FT_DIR);
560 strcpy(dir->name3, "lost+found");
561 PUT((uint64_t)(FETCH_LE32(gd[0].bg_inode_table) + itsz + 0) * blocksize, buf, blocksize);
565 if (ENABLE_FEATURE_CLEAN_UP) {