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)
92 //bb_info_msg("ALLOC: [%u][%u][%u]: [%u-%u]:=[%x],[%x]", blocksize, start, end, start/8, blocksize - end/8 - 1, (1 << (start & 7)) - 1, (uint8_t)(0xFF00 >> (end & 7)));
93 memset(bitmap, 0, blocksize);
95 memset(bitmap, 0xFF, i);
96 bitmap[i] = (1 << (start & 7)) - 1; //0..7 => 00000000..01111111
98 bitmap[blocksize - i - 1] |= 0x7F00 >> (end & 7); //0..7 => 00000000..11111110
99 memset(bitmap + blocksize - i, 0xFF, i); // N.B. no overflow here!
102 static uint32_t has_super(uint32_t x)
104 // 0, 1 and powers of 3, 5, 7 up to 2^32 limit
105 static const uint32_t supers[] = {
106 0, 1, 3, 5, 7, 9, 25, 27, 49, 81, 125, 243, 343, 625, 729,
107 2187, 2401, 3125, 6561, 15625, 16807, 19683, 59049, 78125,
108 117649, 177147, 390625, 531441, 823543, 1594323, 1953125,
109 4782969, 5764801, 9765625, 14348907, 40353607, 43046721,
110 48828125, 129140163, 244140625, 282475249, 387420489,
111 1162261467, 1220703125, 1977326743, 3486784401/* >2^31 */,
113 const uint32_t *sp = supers + ARRAY_SIZE(supers);
123 /* Standard mke2fs 1.41.9:
124 * Usage: mke2fs [-c|-l filename] [-b block-size] [-f fragment-size]
125 * [-i bytes-per-inode] [-I inode-size] [-J journal-options]
126 * [-G meta group size] [-N number-of-inodes]
127 * [-m reserved-blocks-percentage] [-o creator-os]
128 * [-g blocks-per-group] [-L volume-label] [-M last-mounted-directory]
129 * [-O feature[,...]] [-r fs-revision] [-E extended-option[,...]]
130 * [-T fs-type] [-U UUID] [-jnqvFSV] device [blocks-count]
132 // N.B. not commented below options are taken and silently ignored
136 OPT_b = 1 << 2, // block size, in bytes
138 OPT_i = 1 << 4, // bytes per inode
143 OPT_m = 1 << 9, // percentage of blocks reserved for superuser
146 OPT_L = 1 << 12, // label
154 OPT_n = 1 << 20, // dry run: do not write anything
159 //OPT_V = 1 << 25, // -V version. bbox applets don't support that
162 #define fd 3 /* predefined output descriptor */
164 static void PUT(uint64_t off, void *buf, uint32_t size)
166 // bb_info_msg("PUT[%llu]:[%u]", off, size);
167 xlseek(fd, off, SEEK_SET);
168 xwrite(fd, buf, size);
171 int mkfs_ext2_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
172 int mkfs_ext2_main(int argc UNUSED_PARAM, char **argv)
176 unsigned blocksize, blocksize_log2;
177 unsigned reserved_percent = 5;
178 unsigned long long kilobytes;
179 uint32_t nblocks, nblocks_full, nreserved;
181 uint32_t bytes_per_inode;
182 uint32_t first_block;
183 uint32_t inodes_per_group;
184 uint32_t gdtsz, itsz;
187 const char *label = "";
189 struct ext2_super_block *sb; // superblock
190 struct ext2_group_desc *gd; // group descriptors
191 struct ext2_inode *inode;
192 struct ext2_dir *dir;
195 opt_complementary = "-1:b+:m+:i+";
196 opts = getopt32(argv, "cl:b:f:i:I:J:G:N:m:o:g:L:M:O:r:E:T:U:jnqvFS",
197 NULL, &bs, NULL, &bpi, NULL, NULL, NULL, NULL,
198 &reserved_percent, NULL, NULL, &label, NULL, NULL, NULL, NULL, NULL, NULL);
199 argv += optind; // argv[0] -- device
201 // check the device is a block device
202 xmove_fd(xopen(argv[0], O_WRONLY), fd);
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 // open the device, get size in kbytes
215 kilobytes = xatoull(argv[1]);
217 kilobytes = (uoff_t)xlseek(fd, 0, SEEK_END) / 1024;
220 bytes_per_inode = 16384;
221 if (kilobytes < 512*1024)
222 bytes_per_inode = 4096;
223 if (kilobytes < 3*1024)
224 bytes_per_inode = 8192;
226 bytes_per_inode = bpi;
228 // Determine block size
229 // block size is a multiple of 1024
231 if (kilobytes >= 512*1024) // mke2fs 1.41.9 compat
233 if (EXT2_MAX_BLOCK_SIZE > 4096) {
234 // kilobytes >> 22 == size in 4gigabyte chunks.
235 // if size >= 16k gigs, blocksize must be increased.
236 // Try "mke2fs -F image $((16 * 1024*1024*1024))"
237 while ((kilobytes >> 22) >= blocksize)
242 if (blocksize < EXT2_MIN_BLOCK_SIZE
243 || blocksize > EXT2_MAX_BLOCK_SIZE
244 || (blocksize & (blocksize - 1)) // not power of 2
246 bb_error_msg_and_die("blocksize %u is bad", blocksize);
248 // number of bits in one block, i.e. 8*blocksize
249 #define blocks_per_group (8 * blocksize)
250 first_block = (EXT2_MIN_BLOCK_SIZE == blocksize);
251 blocksize_log2 = int_log2(blocksize);
253 // Determine number of blocks
254 kilobytes >>= (blocksize_log2 - EXT2_MIN_BLOCK_LOG_SIZE);
256 if (nblocks != kilobytes)
257 bb_error_msg_and_die("block count doesn't fit in 32 bits");
258 #define kilobytes kilobytes_unused_after_this
260 // if (blocksize < PAGE_SIZE)
261 // nblocks &= ~((PAGE_SIZE >> blocksize_log2)-1);
262 // Experimentally, standard mke2fs won't work on images smaller than 60k
264 bb_error_msg_and_die("need >= 60 blocks");
266 // How many reserved blocks?
267 if (reserved_percent > 50)
268 bb_error_msg_and_die("-%c is bad", 'm');
269 //nreserved = div_roundup((uint64_t) nblocks * reserved_percent, 100);
270 nreserved = (uint64_t)nblocks * reserved_percent / 100;
272 // N.B. killing e2fsprogs feature! Unused blocks don't account in calculations
273 nblocks_full = nblocks;
275 // If last block group is too small, nblocks may be decreased in order
276 // to discard it, and control returns here to recalculate some
278 // Note: blocksize and bytes_per_inode are never recalculated.
280 // N.B. a block group can have no more than blocks_per_group blocks
281 ngroups = div_roundup(nblocks - first_block, blocks_per_group);
283 bb_error_msg_and_die("ngroups");
285 gdtsz = div_roundup(ngroups, blocksize / sizeof(*gd));
286 // TODO: reserved blocks must be marked as such in the bitmaps,
287 // or resulting filesystem is corrupt
288 if (ENABLE_FEATURE_MKFS_EXT2_RESERVED_GDT) {
290 * From e2fsprogs: Calculate the number of GDT blocks to reserve for online
292 * The absolute maximum number of GDT blocks we can reserve is determined by
293 * the number of block pointers that can fit into a single block.
294 * We set it at 1024x the current filesystem size, or
295 * the upper block count limit (2^32), whichever is lower.
297 uint32_t rgdtsz = 0xFFFFFFFF; // maximum block number
298 if (nblocks < rgdtsz / 1024)
299 rgdtsz = nblocks * 1024;
300 rgdtsz = div_roundup(rgdtsz - first_block, blocks_per_group);
301 rgdtsz = div_roundup(rgdtsz, blocksize / sizeof(*gd)) - gdtsz;
302 if (rgdtsz > blocksize / sizeof(uint32_t))
303 rgdtsz = blocksize / sizeof(uint32_t);
304 //TODO: STORE_LE(sb->s_reserved_gdt_blocks, rgdtsz);
309 // N.B. e2fsprogs does as follows!
310 // ninodes is the total number of inodes (files) in the file system
311 uint32_t ninodes = ((uint64_t) nblocks_full * blocksize) / bytes_per_inode;
312 uint32_t overhead, remainder;
313 if (ninodes < EXT2_GOOD_OLD_FIRST_INO+1)
314 ninodes = EXT2_GOOD_OLD_FIRST_INO+1;
315 inodes_per_group = div_roundup(ninodes, ngroups);
316 // minimum number because the first EXT2_GOOD_OLD_FIRST_INO-1 are reserved
317 if (inodes_per_group < 16)
318 inodes_per_group = 16;
320 // a block group can have no more than 8*blocksize inodes
321 if (inodes_per_group > blocks_per_group)
322 inodes_per_group = blocks_per_group;
323 // adjust inodes per group so they completely fill the inode table blocks in the descriptor
324 inodes_per_group = (div_roundup(inodes_per_group * sizeof(*inode), blocksize) * blocksize) / sizeof(*inode);
325 // make sure the number of inodes per group is a multiple of 8
326 inodes_per_group &= ~7;
327 itsz = div_roundup(inodes_per_group * sizeof(*inode), blocksize);
329 // the last group needs more attention: isn't it too small for possible overhead?
330 overhead = (has_super(ngroups - 1) ? (1/*sb*/ + gdtsz) : 0) + 1/*bbmp*/ + 1/*ibmp*/ + itsz;
331 remainder = (nblocks - first_block) % blocks_per_group;
332 if ((1 == ngroups) && remainder && (remainder < overhead))
333 bb_error_msg_and_die("way small device");
334 // Standard mke2fs uses 50. Looks like a bug in our calculation
335 // of "remainder" or "overhead" - we don't match standard mke2fs
336 // when we transition from one group to two groups
337 // (a bit after 8M image size), but it works for two->three groups
338 // transition (at 16M).
339 if (remainder && (remainder < overhead + 50)) {
340 //bb_info_msg("CHOP[%u]", remainder);
341 nblocks -= remainder;
346 if (nblocks_full - nblocks)
347 printf("warning: %u blocks unused\n\n", nblocks_full - nblocks);
349 "Filesystem label=%s\n"
351 "Block size=%u (log=%u)\n"
352 "Fragment size=%u (log=%u)\n"
353 "%u inodes, %u blocks\n"
354 "%u blocks (%u%%) reserved for the super user\n"
355 "First data block=%u\n"
356 "Maximum filesystem blocks=%u\n"
358 "%u blocks per group, %u fragments per group\n"
359 "%u inodes per group"
361 , blocksize, blocksize_log2 - EXT2_MIN_BLOCK_LOG_SIZE
362 , blocksize, blocksize_log2 - EXT2_MIN_BLOCK_LOG_SIZE
363 , inodes_per_group * ngroups, nblocks
364 , nreserved, reserved_percent
366 , gdtsz * (blocksize / sizeof(*gd)) * blocks_per_group
368 , blocks_per_group, blocks_per_group
372 const char *fmt = "\nSuperblock backups stored on blocks:\n"
375 for (i = 1; i < ngroups; i++) {
376 pos += blocks_per_group;
378 printf(fmt, (unsigned)pos);
385 // dry run? -> we are done
389 // TODO: 3/5 refuse if mounted
390 // TODO: 4/5 compat options
391 // TODO: 1/5 sanity checks
392 // TODO: 0/5 more verbose error messages
393 // TODO: 4/5 bigendianness: recheck, wait for ARM reporters
394 // TODO: 2/5 reserved GDT: how to mark but not allocate?
395 // TODO: 3/5 dir_index?
397 // fill the superblock
398 sb = xzalloc(blocksize);
399 STORE_LE(sb->s_rev_level, 1); // revision 1 filesystem
400 STORE_LE(sb->s_magic, EXT2_SUPER_MAGIC);
401 STORE_LE(sb->s_inode_size, sizeof(*inode));
402 STORE_LE(sb->s_first_ino, EXT2_GOOD_OLD_FIRST_INO);
403 STORE_LE(sb->s_log_block_size, blocksize_log2 - EXT2_MIN_BLOCK_LOG_SIZE);
404 STORE_LE(sb->s_log_frag_size, blocksize_log2 - EXT2_MIN_BLOCK_LOG_SIZE);
405 // first 1024 bytes of the device are for boot record. If block size is 1024 bytes, then
406 // the first block is 1, otherwise 0
407 STORE_LE(sb->s_first_data_block, first_block);
408 // block and inode bitmaps occupy no more than one block, so maximum number of blocks is
409 STORE_LE(sb->s_blocks_per_group, blocks_per_group);
410 STORE_LE(sb->s_frags_per_group, blocks_per_group);
412 STORE_LE(sb->s_blocks_count, nblocks);
413 // reserve blocks for superuser
414 STORE_LE(sb->s_r_blocks_count, nreserved);
416 STORE_LE(sb->s_inodes_per_group, inodes_per_group);
417 STORE_LE(sb->s_inodes_count, inodes_per_group * ngroups);
418 STORE_LE(sb->s_free_inodes_count, inodes_per_group * ngroups - EXT2_GOOD_OLD_FIRST_INO);
420 timestamp = time(NULL);
421 STORE_LE(sb->s_mkfs_time, timestamp);
422 STORE_LE(sb->s_wtime, timestamp);
423 STORE_LE(sb->s_lastcheck, timestamp);
425 STORE_LE(sb->s_state, 1); // TODO: what's 1?
426 STORE_LE(sb->s_creator_os, EXT2_OS_LINUX);
427 STORE_LE(sb->s_checkinterval, 24*60*60 * 180); // 180 days
428 STORE_LE(sb->s_errors, EXT2_ERRORS_DEFAULT);
429 STORE_LE(sb->s_feature_compat, EXT2_FEATURE_COMPAT_SUPP
430 | (EXT2_FEATURE_COMPAT_RESIZE_INO * ENABLE_FEATURE_MKFS_EXT2_RESERVED_GDT)
431 | (EXT2_FEATURE_COMPAT_DIR_INDEX * ENABLE_FEATURE_MKFS_EXT2_DIR_INDEX)
433 // e2fsck from 1.41.9 doesn't like EXT2_FEATURE_INCOMPAT_WHITEOUT
434 STORE_LE(sb->s_feature_incompat, EXT2_FEATURE_INCOMPAT_FILETYPE);// | EXT2_FEATURE_INCOMPAT_WHITEOUT;
435 STORE_LE(sb->s_feature_ro_compat, EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER);
436 STORE_LE(sb->s_flags, EXT2_FLAGS_SIGNED_HASH * ENABLE_FEATURE_MKFS_EXT2_DIR_INDEX);
437 generate_uuid(sb->s_uuid);
438 if (ENABLE_FEATURE_MKFS_EXT2_DIR_INDEX) {
439 STORE_LE(sb->s_def_hash_version, EXT2_HASH_HALF_MD4);
440 generate_uuid((uint8_t *)sb->s_hash_seed);
443 * From e2fsprogs: add "jitter" to the superblock's check interval so that we
444 * don't check all the filesystems at the same time. We use a
445 * kludgy hack of using the UUID to derive a random jitter value.
447 STORE_LE(sb->s_max_mnt_count, EXT2_DFL_MAX_MNT_COUNT
448 + (sb->s_uuid[ARRAY_SIZE(sb->s_uuid)-1] % EXT2_DFL_MAX_MNT_COUNT));
451 safe_strncpy((char *)sb->s_volume_name, label, sizeof(sb->s_volume_name));
453 // fill group descriptors
454 gd = xzalloc(gdtsz * blocksize);
455 buf = xmalloc(blocksize);
456 sb->s_free_blocks_count = 0;
457 for (i = 0, pos = first_block, n = nblocks - first_block;
459 i++, pos += blocks_per_group, n -= blocks_per_group
461 uint32_t overhead = pos + (has_super(i) ? (1/*sb*/ + gdtsz) : 0);
463 STORE_LE(gd[i].bg_block_bitmap, overhead + 0);
464 STORE_LE(gd[i].bg_inode_bitmap, overhead + 1);
465 STORE_LE(gd[i].bg_inode_table, overhead + 2);
466 overhead = overhead - pos + 1/*bbmp*/ + 1/*ibmp*/ + itsz;
467 gd[i].bg_free_inodes_count = inodes_per_group;
468 //STORE_LE(gd[i].bg_used_dirs_count, 0);
469 // N.B. both root and lost+found dirs are within the first block group, thus +2
472 STORE_LE(gd[i].bg_used_dirs_count, 2);
473 gd[i].bg_free_inodes_count -= EXT2_GOOD_OLD_FIRST_INO;
476 // mark preallocated blocks as allocated
477 fb = (n < blocks_per_group ? n : blocks_per_group) - overhead;
478 //bb_info_msg("ALLOC: [%u][%u][%u]", blocksize, overhead, blocks_per_group - (fb + overhead));
479 allocate(buf, blocksize,
481 blocks_per_group - (fb + overhead)
484 PUT((uint64_t)(FETCH_LE32(gd[i].bg_block_bitmap)) * blocksize, buf, blocksize);
485 STORE_LE(gd[i].bg_free_blocks_count, fb);
487 // mark preallocated inodes as allocated
488 allocate(buf, blocksize,
489 inodes_per_group - gd[i].bg_free_inodes_count,
490 blocks_per_group - inodes_per_group
493 //PUT((uint64_t)(FETCH_LE32(gd[i].bg_block_bitmap)) * blocksize, buf, blocksize);
494 //but it's right after block bitmap, so we can just:
495 xwrite(fd, buf, blocksize);
496 STORE_LE(gd[i].bg_free_inodes_count, gd[i].bg_free_inodes_count);
498 // count overall free blocks
499 sb->s_free_blocks_count += fb;
501 STORE_LE(sb->s_free_blocks_count, sb->s_free_blocks_count);
503 // dump filesystem skeleton structures
504 // printf("Writing superblocks and filesystem accounting information: ");
505 for (i = 0, pos = first_block; i < ngroups; i++, pos += blocks_per_group) {
506 // dump superblock and group descriptors and their backups
508 // N.B. 1024 byte blocks are special
509 PUT(((uint64_t)pos * blocksize) + ((0 == i && 0 == first_block) ? 1024 : 0), sb, 1024);//blocksize);
510 PUT(((uint64_t)pos * blocksize) + blocksize, gd, gdtsz * blocksize);
515 memset(buf, 0, blocksize);
516 PUT(0, buf, 1024); // N.B. 1024 <= blocksize, so buf[0..1023] contains zeros
518 for (i = 0; i < ngroups; ++i)
519 for (n = 0; n < itsz; ++n)
520 PUT((uint64_t)(FETCH_LE32(gd[i].bg_inode_table) + n) * blocksize, buf, blocksize);
522 // prepare directory inode
523 inode = (struct ext2_inode *)buf;
524 STORE_LE(inode->i_mode, S_IFDIR | S_IRWXU | S_IRGRP | S_IROTH | S_IXGRP | S_IXOTH);
525 STORE_LE(inode->i_mtime, timestamp);
526 STORE_LE(inode->i_atime, timestamp);
527 STORE_LE(inode->i_ctime, timestamp);
528 STORE_LE(inode->i_size, blocksize);
529 // N.B. inode->i_blocks stores the number of 512 byte data blocks. Why on Earth?!
530 STORE_LE(inode->i_blocks, blocksize / 512);
532 // dump root dir inode
533 STORE_LE(inode->i_links_count, 3); // "/.", "/..", "/lost+found/.." point to this inode
534 STORE_LE(inode->i_block[0], FETCH_LE32(gd[0].bg_inode_table) + itsz);
535 PUT(((uint64_t)FETCH_LE32(gd[0].bg_inode_table) * blocksize) + (EXT2_ROOT_INO-1) * sizeof(*inode), buf, sizeof(*inode));
537 // dump lost+found dir inode
538 STORE_LE(inode->i_links_count, 2); // both "/lost+found" and "/lost+found/." point to this inode
539 STORE_LE(inode->i_block[0], inode->i_block[0]+1); // use next block
540 PUT(((uint64_t)FETCH_LE32(gd[0].bg_inode_table) * blocksize) + (EXT2_GOOD_OLD_FIRST_INO-1) * sizeof(*inode), buf, sizeof(*inode));
543 memset(buf, 0, blocksize);
544 dir = (struct ext2_dir *)buf;
546 // dump lost+found dir block
547 STORE_LE(dir->inode1, EXT2_GOOD_OLD_FIRST_INO);
548 STORE_LE(dir->rec_len1, 12);
549 STORE_LE(dir->name_len1, 1);
550 STORE_LE(dir->file_type1, EXT2_FT_DIR);
552 STORE_LE(dir->inode2, EXT2_ROOT_INO);
553 STORE_LE(dir->rec_len2, blocksize - 12);
554 STORE_LE(dir->name_len2, 2);
555 STORE_LE(dir->file_type2, EXT2_FT_DIR);
556 dir->name2[0] = '.'; dir->name2[1] = '.';
557 PUT((uint64_t)(FETCH_LE32(gd[0].bg_inode_table) + itsz + 1) * blocksize, buf, blocksize);
559 // dump root dir block
560 STORE_LE(dir->inode1, EXT2_ROOT_INO);
561 STORE_LE(dir->rec_len2, 12);
562 STORE_LE(dir->inode3, EXT2_GOOD_OLD_FIRST_INO);
563 STORE_LE(dir->rec_len3, blocksize - 12 - 12);
564 STORE_LE(dir->name_len3, 10);
565 STORE_LE(dir->file_type3, EXT2_FT_DIR);
566 strcpy(dir->name3, "lost+found");
567 PUT((uint64_t)(FETCH_LE32(gd[0].bg_inode_table) + itsz + 0) * blocksize, buf, blocksize);
571 if (ENABLE_FEATURE_CLEAN_UP) {