123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367 |
- /* vi: set sw=4 ts=4: */
- /*
- * mkfs_reiser: utility to create ReiserFS filesystem
- *
- * Busybox'ed (2009) by Vladimir Dronnikov <dronnikov@gmail.com>
- *
- * Licensed under GPLv2, see file LICENSE in this tarball for details.
- */
- #include "libbb.h"
- #include <linux/fs.h>
- char BUG_wrong_field_size(void);
- #define STORE_LE(field, value) \
- do { \
- if (sizeof(field) == 4) \
- field = SWAP_LE32(value); \
- else if (sizeof(field) == 2) \
- field = SWAP_LE16(value); \
- else if (sizeof(field) == 1) \
- field = (value); \
- else \
- BUG_wrong_field_size(); \
- } while (0)
- #define FETCH_LE32(field) \
- (sizeof(field) == 4 ? SWAP_LE32(field) : BUG_wrong_field_size())
- struct journal_params {
- uint32_t jp_journal_1st_block; /* where does journal start from on its device */
- uint32_t jp_journal_dev; /* journal device st_rdev */
- uint32_t jp_journal_size; /* size of the journal on FS creation. used to make sure they don't overflow it */
- uint32_t jp_journal_trans_max; /* max number of blocks in a transaction. */
- uint32_t jp_journal_magic; /* random value made on fs creation (this was sb_journal_block_count) */
- uint32_t jp_journal_max_batch; /* max number of blocks to batch into a trans */
- uint32_t jp_journal_max_commit_age; /* in seconds, how old can an async commit be */
- uint32_t jp_journal_max_trans_age; /* in seconds, how old can a transaction be */
- };
- struct reiserfs_journal_header {
- uint32_t jh_last_flush_trans_id; /* id of last fully flushed transaction */
- uint32_t jh_first_unflushed_offset; /* offset in the log of where to start replay after a crash */
- uint32_t jh_mount_id;
- struct journal_params jh_journal;
- uint32_t jh_last_check_mount_id; /* the mount id of the fs during the last reiserfsck --check. */
- };
- struct reiserfs_super_block {
- uint32_t sb_block_count; /* 0 number of block on data device */
- uint32_t sb_free_blocks; /* 4 free blocks count */
- uint32_t sb_root_block; /* 8 root of the tree */
- struct journal_params sb_journal; /* 12 */
- uint16_t sb_blocksize; /* 44 */
- uint16_t sb_oid_maxsize; /* 46 max size of object id array, see get_objectid() commentary */
- uint16_t sb_oid_cursize; /* 48 current size of object id array */
- uint16_t sb_umount_state; /* 50 this is set to 1 when filesystem was umounted, to 2 - when not */
- char s_magic[10]; /* 52 "ReIsErFs" or "ReIsEr2Fs" or "ReIsEr3Fs" */
- uint16_t sb_fs_state; /* 62 it is set to used by fsck to mark which phase of rebuilding is done (used for fsck debugging) */
- uint32_t sb_hash_function_code; /* 64 code of fuction which was/is/will be used to sort names in a directory. See codes in above */
- uint16_t sb_tree_height; /* 68 height of filesytem tree. Tree consisting of only one root block has 2 here */
- uint16_t sb_bmap_nr; /* 70 amount of bitmap blocks needed to address each block of file system */
- uint16_t sb_version; /* 72 this field is only reliable on filesystem with non-standard journal */
- uint16_t sb_reserved_for_journal; /* 74 size in blocks of journal area on main device, we need to keep after non-standard journal relocation */
- uint32_t sb_inode_generation; /* 76 */
- uint32_t sb_flags; /* 80 Right now used only by inode-attributes, if enabled */
- unsigned char s_uuid[16]; /* 84 filesystem unique identifier */
- unsigned char s_label[16]; /* 100 filesystem volume label */
- uint16_t sb_mnt_count; /* 116 */
- uint16_t sb_max_mnt_count; /* 118 */
- uint32_t sb_lastcheck; /* 120 */
- uint32_t sb_check_interval; /* 124 */
- /* zero filled by mkreiserfs and reiserfs_convert_objectid_map_v1() so any additions must be updated there as well. */
- char s_unused[76]; /* 128 */
- /* 204 */
- };
- /* Header of a disk block. More precisely, header of a formatted leaf
- or internal node, and not the header of an unformatted node. */
- struct block_head {
- uint16_t blk2_level; /* Level of a block in the tree. */
- uint16_t blk2_nr_item; /* Number of keys/items in a block. */
- uint16_t blk2_free_space; /* Block free space in bytes. */
- uint16_t blk_reserved;
- uint32_t reserved[4];
- };
- #define REISERFS_DISK_OFFSET_IN_BYTES (64 * 1024)
- #define REISERFS_3_6_SUPER_MAGIC_STRING "ReIsEr2Fs"
- #define REISERFS_FORMAT_3_6 2
- #define DEFAULT_MAX_MNT_COUNT 30 /* 30 mounts */
- #define DEFAULT_CHECK_INTERVAL (180 * 60 * 60 * 24) /* 180 days */
- #define FS_CLEANLY_UMOUNTED 1 /* this was REISERFS_VALID_FS */
- #define JOURNAL_MIN_SIZE 512
- /* biggest possible single transaction, don't change for now (8/3/99) */
- #define JOURNAL_TRANS_MAX 1024
- #define JOURNAL_TRANS_MIN 256 /* need to check whether it works */
- #define JOURNAL_DEFAULT_RATIO 8 /* default journal size / max trans length */
- #define JOURNAL_MIN_RATIO 2
- /* max blocks to batch into one transaction, don't make this any bigger than 900 */
- #define JOURNAL_MAX_BATCH 900
- #define JOURNAL_MAX_COMMIT_AGE 30
- // Standard mkreiserfs 3.6.21:
- // -b | --block-size N size of file-system block, in bytes
- // -j | --journal-device FILE path to separate device to hold journal
- // -s | --journal-size N size of the journal in blocks
- // -o | --journal-offset N offset of the journal from the start of
- // the separate device, in blocks
- // -t | --transaction-max-size N maximal size of transaction, in blocks
- // -B | --badblocks file store all bad blocks given in file on the fs
- // -h | --hash rupasov|tea|r5 hash function to use by default
- // -u | --uuid UUID store UUID in the superblock
- // -l | --label LABEL store LABEL in the superblock
- // --format 3.5|3.6 old 3.5 format or newer 3.6
- // -f | --force specified once, make mkreiserfs the whole
- // disk, not block device or mounted partition;
- // specified twice, do not ask for confirmation
- // -q | --quiet quiet work without messages, progress and
- // questions. Useful if run in a script. For use
- // by end users only.
- // -d | --debug print debugging information during mkreiser
- // -V print version and exit
- // Options not commented below are taken but silently ignored:
- enum {
- OPT_b = 1 << 0,
- OPT_j = 1 << 1,
- OPT_s = 1 << 2,
- OPT_o = 1 << 3,
- OPT_t = 1 << 4,
- OPT_B = 1 << 5,
- OPT_h = 1 << 6,
- OPT_u = 1 << 7,
- OPT_l = 1 << 8, // label
- OPT_f = 1 << 9, // ask no questions
- OPT_q = 1 << 10,
- OPT_d = 1 << 11,
- //OPT_V = 1 << 12, // -V version. bbox applets don't support that
- };
- int mkfs_reiser_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
- int mkfs_reiser_main(int argc UNUSED_PARAM, char **argv)
- {
- unsigned blocksize = 4096;
- unsigned journal_blocks = 8192;
- unsigned blocks, bitmap_blocks, i, block;
- time_t timestamp;
- const char *label = "";
- struct stat st;
- int fd;
- uint8_t *buf;
- struct reiserfs_super_block *sb;
- struct journal_params *jp;
- struct block_head *root;
- // using global "option_mask32" instead of local "opts":
- // we are register starved here
- opt_complementary = "-1:b+";
- /*opts =*/ getopt32(argv, "b:j:s:o:t:B:h:u:l:fqd",
- NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, &label);
- argv += optind; // argv[0] -- device
- // check the device is a block device
- fd = xopen(argv[0], O_WRONLY | O_EXCL);
- fstat(fd, &st);
- if (!S_ISBLK(st.st_mode) && !(option_mask32 & OPT_f))
- bb_error_msg_and_die("not a block device");
- // check if it is mounted
- // N.B. what if we format a file? find_mount_point will return false negative since
- // it is loop block device which is mounted!
- if (find_mount_point(argv[0], 0))
- bb_error_msg_and_die("can't format mounted filesystem");
- // open the device, get size in blocks
- blocks = get_volume_size_in_bytes(fd, argv[1], blocksize, /*extend:*/ 1) / blocksize;
- // block number sanity check
- // we have a limit: skipped area, super block, journal and root block
- // all have to be addressed by one first bitmap
- block = REISERFS_DISK_OFFSET_IN_BYTES / blocksize // boot area
- + 1 // sb
- + 1 // bitmap#0
- + journal_blocks+1 // journal
- ;
- // count overhead
- bitmap_blocks = (blocks - 1) / (blocksize * 8) + 1;
- i = block + bitmap_blocks;
- // check overhead
- if (MIN(blocksize * 8, blocks) < i)
- bb_error_msg_and_die("need >= %u blocks", i);
- // ask confirmation?
- // TODO: ???
- // wipe out first REISERFS_DISK_OFFSET_IN_BYTES of device
- // TODO: do we really need to wipe?!
- xlseek(fd, REISERFS_DISK_OFFSET_IN_BYTES, SEEK_SET);
- // fill superblock
- sb = (struct reiserfs_super_block *)xzalloc(blocksize);
- // block count
- STORE_LE(sb->sb_block_count, blocks);
- STORE_LE(sb->sb_free_blocks, blocks - i);
- // TODO: decypher!
- STORE_LE(sb->sb_root_block, block);
- // fill journal related fields
- jp = &sb->sb_journal;
- STORE_LE(jp->jp_journal_1st_block, REISERFS_DISK_OFFSET_IN_BYTES / blocksize + 1/*sb*/ + 1/*bmp#0*/);
- timestamp = time(NULL);
- srandom(timestamp);
- STORE_LE(jp->jp_journal_magic, random());
- STORE_LE(jp->jp_journal_size, journal_blocks);
- STORE_LE(jp->jp_journal_trans_max, JOURNAL_TRANS_MAX);
- STORE_LE(jp->jp_journal_max_batch, JOURNAL_MAX_BATCH);
- STORE_LE(jp->jp_journal_max_commit_age, JOURNAL_MAX_COMMIT_AGE);
- // sizes
- STORE_LE(sb->sb_blocksize, blocksize);
- STORE_LE(sb->sb_oid_maxsize, (blocksize - sizeof(*sb)) / sizeof(uint32_t) / 2 * 2);
- STORE_LE(sb->sb_oid_cursize, 2); // "." and ".."
- strcpy(sb->s_magic, REISERFS_3_6_SUPER_MAGIC_STRING);
- STORE_LE(sb->sb_bmap_nr, (bitmap_blocks > ((1LL << 16) - 1)) ? 0 : bitmap_blocks);
- // misc
- STORE_LE(sb->sb_version, REISERFS_FORMAT_3_6);
- STORE_LE(sb->sb_lastcheck, timestamp);
- STORE_LE(sb->sb_check_interval, DEFAULT_CHECK_INTERVAL);
- STORE_LE(sb->sb_mnt_count, 1);
- STORE_LE(sb->sb_max_mnt_count, DEFAULT_MAX_MNT_COUNT);
- STORE_LE(sb->sb_umount_state, FS_CLEANLY_UMOUNTED);
- STORE_LE(sb->sb_tree_height, 2);
- STORE_LE(sb->sb_hash_function_code, 3); // R5_HASH
- STORE_LE(sb->sb_flags, 1);
- //STORE_LE(sb->sb_reserved_for_journal, 0);
- // create UUID
- generate_uuid(sb->s_uuid);
- // write the label
- safe_strncpy((char *)sb->s_label, label, sizeof(sb->s_label));
- // TODO: EMPIRIC! ENDIANNESS!
- // superblock has only 204 bytes. What are these?
- buf = (uint8_t *)sb;
- buf[205] = 1;
- buf[209] = 3;
- // put superblock
- xwrite(fd, sb, blocksize);
- // create bitmaps
- buf = xzalloc(blocksize);
- // bitmap #0 uses initial "block"+1 blocks
- i = block + 1;
- memset(buf, 0xFF, i / 8);
- buf[i / 8] = (1 << (i & 7)) - 1; //0..7 => 00000000..01111111
- // mark trailing absent blocks, if any
- if (blocks < 8*blocksize) {
- unsigned n = 8*blocksize - blocks;
- i = n / 8;
- buf[blocksize - i - 1] |= 0x7F00 >> (n & 7); //0..7 => 00000000..11111110
- memset(buf + blocksize - i, 0xFF, i); // N.B. no overflow here!
- }
- // put bitmap #0
- xwrite(fd, buf, blocksize);
- // now go journal blocks
- memset(buf, 0, blocksize);
- for (i = 0; i < journal_blocks; i++)
- xwrite(fd, buf, blocksize);
- // dump journal control block
- memcpy(&((struct reiserfs_journal_header *)buf)->jh_journal, &sb->sb_journal, sizeof(sb->sb_journal));
- xwrite(fd, buf, blocksize);
- // other bitmaps are in every (8*blocksize)-th block
- // N.B. they use the only block -- namely bitmap itself!
- buf[0] = 0x01;
- // put bitmaps
- for (i = 1; i < bitmap_blocks; i++) {
- xlseek(fd, i*8*blocksize * blocksize, SEEK_SET);
- // mark trailing absent blocks, if any
- if (i == bitmap_blocks - 1 && (blocks % (8*blocksize))) {
- unsigned n = 8*blocksize - blocks % (8*blocksize);
- unsigned j = n / 8;
- buf[blocksize - j - 1] |= 0x7F00 >> (n & 7); //0..7 => 00000000..11111110
- memset(buf + blocksize - j, 0xFF, j); // N.B. no overflow here!
- }
- xwrite(fd, buf, blocksize);
- }
- // fill root block
- // block head
- memset(buf, 0, blocksize);
- root = (struct block_head *)buf;
- STORE_LE(root->blk2_level, 1); // leaf node
- STORE_LE(root->blk2_nr_item, 2); // "." and ".."
- STORE_LE(root->blk2_free_space, blocksize - sizeof(struct block_head));
- // item head
- // root directory
- // TODO: EMPIRIC! ENDIANNESS!
- // TODO: indented assignments seem to be timestamps
- buf[4] = 0134;
- buf[24] = 01;
- buf[28] = 02;
- buf[42] = 054;
- buf[44] = 0324;
- buf[45] = 017;
- buf[46] = 01;
- buf[48] = 01;
- buf[52] = 02;
- buf[56] = 01;
- buf[60] = 0364;
- buf[61] = 01;
- buf[64] = 02;
- buf[66] = 060;
- buf[68] = 0244;
- buf[69] = 017;
- buf[4004] = 01;
- buf[4008] = 01;
- buf[4012] = 02;
- buf[4016] = 050;
- buf[4018] = 04;
- buf[4020] = 02;
- buf[4028] = 01;
- buf[4032] = 040;
- buf[4034] = 04;
- buf[4036] = 056; buf[4037] = 056; // ".."
- buf[4044] = 056; // "."
- buf[4052] = 0355;
- buf[4053] = 0101;
- buf[4056] = 03;
- buf[4060] = 060;
- buf[4076] = 0173;
- buf[4077] = 0240;
- buf[4078] = 0344;
- buf[4079] = 0112;
- buf[4080] = 0173;
- buf[4081] = 0240;
- buf[4082] = 0344;
- buf[4083] = 0112;
- buf[4084] = 0173;
- buf[4085] = 0240;
- buf[4086] = 0344;
- buf[4087] = 0112;
- buf[4088] = 01;
- // put root block
- xlseek(fd, block * blocksize, SEEK_SET);
- xwrite(fd, buf, blocksize);
- // cleanup
- if (ENABLE_FEATURE_CLEAN_UP) {
- free(buf);
- free(sb);
- }
- xclose(fd);
- return EXIT_SUCCESS;
- }
|