X-Git-Url: https://git.librecmc.org/?a=blobdiff_plain;f=util-linux%2Fmkfs_ext2.c;h=6dccd3a5d8e7b76a73b891e501c56353e460535f;hb=38292b68c962b9d470fa4e577020749c8c69226d;hp=8df9f1a66c80713e93952df7b78def1c0dceafed;hpb=7673e7fda42a2665f8528d531e9596f264643149;p=oweals%2Fbusybox.git diff --git a/util-linux/mkfs_ext2.c b/util-linux/mkfs_ext2.c index 8df9f1a66..6dccd3a5d 100644 --- a/util-linux/mkfs_ext2.c +++ b/util-linux/mkfs_ext2.c @@ -5,41 +5,41 @@ * * Busybox'ed (2009) by Vladimir Dronnikov * - * Licensed under GPLv2, see file LICENSE in this tarball for details. + * Licensed under GPLv2, see file LICENSE in this source tree. */ #include "libbb.h" #include #include -#include "volume_id/volume_id_internal.h" #define ENABLE_FEATURE_MKFS_EXT2_RESERVED_GDT 0 -#define ENABLE_FEATURE_MKFS_EXT2_DIR_INDEX 1 +#define ENABLE_FEATURE_MKFS_EXT2_DIR_INDEX 1 // from e2fsprogs #define s_reserved_gdt_blocks s_padding1 #define s_mkfs_time s_reserved[0] #define s_flags s_reserved[22] -#define EXT2_HASH_HALF_MD4 1 -#define EXT2_FLAGS_SIGNED_HASH 0x0001 -// whiteout: for writable overlays -//#define LINUX_S_IFWHT 0160000 -//#define EXT2_FEATURE_INCOMPAT_WHITEOUT 0x0020 +#define EXT2_HASH_HALF_MD4 1 +#define EXT2_FLAGS_SIGNED_HASH 0x0001 +#define EXT2_FLAGS_UNSIGNED_HASH 0x0002 -// storage helper -void BUG_unsupported_field_size(void); +// storage helpers +char BUG_wrong_field_size(void); #define STORE_LE(field, value) \ do { \ if (sizeof(field) == 4) \ - field = cpu_to_le32(value); \ + field = SWAP_LE32(value); \ else if (sizeof(field) == 2) \ - field = cpu_to_le16(value); \ + field = SWAP_LE16(value); \ else if (sizeof(field) == 1) \ field = (value); \ else \ - BUG_unsupported_field_size(); \ + BUG_wrong_field_size(); \ } while (0) +#define FETCH_LE32(field) \ + (sizeof(field) == 4 ? SWAP_LE32(field) : BUG_wrong_field_size()) + // All fields are little-endian struct ext2_dir { uint32_t inode1; @@ -68,42 +68,33 @@ static unsigned int_log2(unsigned arg) } // taken from mkfs_minix.c. libbb candidate? -static unsigned div_roundup(uint32_t size, uint32_t n) +// "uint32_t size", since we never use it for anything >32 bits +static uint32_t div_roundup(uint32_t size, uint32_t n) { - return (size + n-1) / n; + // Overflow-resistant + uint32_t res = size / n; + if (res * n != size) + res++; + return res; } static void allocate(uint8_t *bitmap, uint32_t blocksize, uint32_t start, uint32_t end) { uint32_t i; + +//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))); memset(bitmap, 0, blocksize); - i = start/8; + i = start / 8; memset(bitmap, 0xFF, i); - bitmap[i] = 0xFF >> (8-(start&7)); -//bb_info_msg("ALLOC: [%u][%u][%u]: [%u]:=[%x]", blocksize, start, end, blocksize - end/8 - 1, (uint8_t)(0xFF << (8-(end&7)))); - i = end/8; - bitmap[blocksize - i - 1] = 0xFF << (8-(end&7)); + bitmap[i] = (1 << (start & 7)) - 1; //0..7 => 00000000..01111111 + i = end / 8; + bitmap[blocksize - i - 1] |= 0x7F00 >> (end & 7); //0..7 => 00000000..11111110 memset(bitmap + blocksize - i, 0xFF, i); // N.B. no overflow here! } -#if 0 -// TODO: get rid of FPU -static bool is_power_of(uint32_t x, uint16_t n) -{ -// return (!(x % n) && is_power_of(x / n, n)); - double z = logf(x)/logf(n); - return (z == (int)z); -} - -static uint32_t has_super(uint32_t x) -{ - return (0 == x || 1 == x || is_power_of(x, 3) || is_power_of(x, 5) || is_power_of(x, 7)); -} - -#else - static uint32_t has_super(uint32_t x) { + // 0, 1 and powers of 3, 5, 7 up to 2^32 limit static const uint32_t supers[] = { 0, 1, 3, 5, 7, 9, 25, 27, 49, 81, 125, 243, 343, 625, 729, 2187, 2401, 3125, 6561, 15625, 16807, 19683, 59049, 78125, @@ -112,34 +103,62 @@ static uint32_t has_super(uint32_t x) 48828125, 129140163, 244140625, 282475249, 387420489, 1162261467, 1220703125, 1977326743, 3486784401/* >2^31 */, }; - const uint32_t *sp = supers + ARRAY_SIZE(supers)-1; + const uint32_t *sp = supers + ARRAY_SIZE(supers); while (1) { + sp--; if (x == *sp) return 1; - if (0 == *sp) + if (x > *sp) return 0; - sp--; } } -#endif - -/* Standard mke2fs 1.41.9: - * Usage: mke2fs [-c|-l filename] [-b block-size] [-f fragment-size] - * [-i bytes-per-inode] [-I inode-size] [-J journal-options] - * [-G meta group size] [-N number-of-inodes] - * [-m reserved-blocks-percentage] [-o creator-os] - * [-g blocks-per-group] [-L volume-label] [-M last-mounted-directory] - * [-O feature[,...]] [-r fs-revision] [-E extended-option[,...]] - * [-T fs-type] [-U UUID] [-jnqvFSV] device [blocks-count] -*/ -// N.B. not commented below options are taken and silently ignored + +#define fd 3 /* predefined output descriptor */ + +static void PUT(uint64_t off, void *buf, uint32_t size) +{ +// bb_info_msg("PUT[%llu]:[%u]", off, size); + xlseek(fd, off, SEEK_SET); + xwrite(fd, buf, size); +} + +// 128 and 256-byte inodes: +// 128-byte inode is described by struct ext2_inode. +// 256-byte one just has these fields appended: +// __u16 i_extra_isize; +// __u16 i_pad1; +// __u32 i_ctime_extra; /* extra Change time (nsec << 2 | epoch) */ +// __u32 i_mtime_extra; /* extra Modification time (nsec << 2 | epoch) */ +// __u32 i_atime_extra; /* extra Access time (nsec << 2 | epoch) */ +// __u32 i_crtime; /* File creation time */ +// __u32 i_crtime_extra; /* extra File creation time (nsec << 2 | epoch)*/ +// __u32 i_version_hi; /* high 32 bits for 64-bit version */ +// the rest is padding. +// +// linux/ext2_fs.h has "#define i_size_high i_dir_acl" which suggests that even +// 128-byte inode is capable of describing large files (i_dir_acl is meaningful +// only for directories, which never need i_size_high). +// +// Standard mke2fs creates a filesystem with 256-byte inodes if it is +// bigger than 0.5GB. + +// Standard mke2fs 1.41.9: +// Usage: mke2fs [-c|-l filename] [-b block-size] [-f fragment-size] +// [-i bytes-per-inode] [-I inode-size] [-J journal-options] +// [-G meta group size] [-N number-of-inodes] +// [-m reserved-blocks-percentage] [-o creator-os] +// [-g blocks-per-group] [-L volume-label] [-M last-mounted-directory] +// [-O feature[,...]] [-r fs-revision] [-E extended-option[,...]] +// [-T fs-type] [-U UUID] [-jnqvFSV] device [blocks-count] +// +// Options not commented below are taken but silently ignored: enum { OPT_c = 1 << 0, OPT_l = 1 << 1, OPT_b = 1 << 2, // block size, in bytes OPT_f = 1 << 3, OPT_i = 1 << 4, // bytes per inode - OPT_I = 1 << 5, + OPT_I = 1 << 5, // custom inode size, in bytes OPT_J = 1 << 6, OPT_G = 1 << 7, OPT_N = 1 << 8, @@ -154,7 +173,7 @@ enum { OPT_T = 1 << 17, OPT_U = 1 << 18, OPT_j = 1 << 19, - OPT_n = 1 << 20, + OPT_n = 1 << 20, // dry run: do not write anything OPT_q = 1 << 21, OPT_v = 1 << 22, OPT_F = 1 << 23, @@ -162,33 +181,26 @@ enum { //OPT_V = 1 << 25, // -V version. bbox applets don't support that }; -#define fd 3 /* predefined output descriptor */ - -static void PUT(uint64_t off, void *buf, uint32_t size) -{ - if (!(option_mask32 & OPT_n)) { - xlseek(fd, off, SEEK_SET); - xwrite(fd, buf, size); - } -} - 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; - unsigned nreserved = 5; - uint32_t nblocks; + unsigned bs, bpi; + unsigned blocksize, blocksize_log2; + unsigned inodesize, user_inodesize; + unsigned reserved_percent = 5; + unsigned long long kilobytes; + uint32_t nblocks, nblocks_full; + uint32_t nreserved; uint32_t ngroups; - unsigned bytes_per_inode; - uint32_t nblocks_per_group; - uint32_t first_data_block; - uint32_t ninodes; - uint32_t ninodes_per_group; - uint32_t gdtsz, rgdtsz, itsz; + uint32_t bytes_per_inode; + uint32_t first_block; + uint32_t inodes_per_group; + uint32_t group_desc_blocks; + uint32_t inode_table_blocks; + uint32_t lost_and_found_blocks; time_t timestamp; - unsigned opts; - const char *label; + const char *label = ""; struct stat st; struct ext2_super_block *sb; // superblock struct ext2_group_desc *gd; // group descriptors @@ -196,257 +208,420 @@ int mkfs_ext2_main(int argc UNUSED_PARAM, char **argv) struct ext2_dir *dir; uint8_t *buf; - bs = EXT2_MIN_BLOCK_SIZE; - 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, - &nreserved, NULL, NULL, &label, NULL, NULL, NULL, NULL, NULL, NULL); + // using global "option_mask32" instead of local "opts": + // we are register starved here + opt_complementary = "-1:b+:i+:I+:m+"; + /*opts =*/ getopt32(argv, "cl:b:f:i:I:J:G:N:m:o:g:L:M:O:r:E:T:U:jnqvFS", + /*lbfi:*/ NULL, &bs, NULL, &bpi, + /*IJGN:*/ &user_inodesize, NULL, NULL, NULL, + /*mogL:*/ &reserved_percent, NULL, NULL, &label, + /*MOrE:*/ NULL, NULL, NULL, NULL, + /*TU:*/ NULL, NULL); argv += optind; // argv[0] -- device - // block size minimax, block size is a multiple of minimum - blocksize = bs; + // open the device, check the device is a block device + xmove_fd(xopen(argv[0], O_WRONLY), fd); + xfstat(fd, &st, argv[0]); + if (!S_ISBLK(st.st_mode) && !(option_mask32 & OPT_F)) + bb_error_msg_and_die("%s: not a block device", argv[0]); + + // 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"); + + // get size in kbytes + kilobytes = get_volume_size_in_bytes(fd, argv[1], 1024, /*extend:*/ !(option_mask32 & OPT_n)) / 1024; + + bytes_per_inode = 16384; + if (kilobytes < 512*1024) + bytes_per_inode = 4096; + if (kilobytes < 3*1024) + bytes_per_inode = 8192; + if (option_mask32 & OPT_i) + bytes_per_inode = bpi; + + // Determine block size and inode size + // block size is a multiple of 1024 + // inode size is a multiple of 128 + blocksize = 1024; + inodesize = sizeof(struct ext2_inode); // 128 + if (kilobytes >= 512*1024) { // mke2fs 1.41.9 compat + blocksize = 4096; + inodesize = 256; + } + if (EXT2_MAX_BLOCK_SIZE > 4096) { + // kilobytes >> 22 == size in 4gigabyte chunks. + // if size >= 16k gigs, blocksize must be increased. + // Try "mke2fs -F image $((16 * 1024*1024*1024))" + while ((kilobytes >> 22) >= blocksize) + blocksize *= 2; + } + if (option_mask32 & OPT_b) + blocksize = bs; if (blocksize < EXT2_MIN_BLOCK_SIZE || blocksize > EXT2_MAX_BLOCK_SIZE || (blocksize & (blocksize - 1)) // not power of 2 ) { - bb_error_msg_and_die("-%c is bad", 'b'); + bb_error_msg_and_die("blocksize %u is bad", blocksize); + } + // Do we have custom inode size? + if (option_mask32 & OPT_I) { + if (user_inodesize < sizeof(*inode) + || user_inodesize > blocksize + || (user_inodesize & (user_inodesize - 1)) // not power of 2 + ) { + bb_error_msg("-%c is bad", 'I'); + } else { + inodesize = user_inodesize; + } } - // reserved blocks count - if (nreserved > 50) + if ((int32_t)bytes_per_inode < blocksize) + bb_error_msg_and_die("-%c is bad", 'i'); + // number of bits in one block, i.e. 8*blocksize +#define blocks_per_group (8 * blocksize) + first_block = (EXT2_MIN_BLOCK_SIZE == blocksize); + blocksize_log2 = int_log2(blocksize); + + // Determine number of blocks + kilobytes >>= (blocksize_log2 - EXT2_MIN_BLOCK_LOG_SIZE); + nblocks = kilobytes; + if (nblocks != kilobytes) + bb_error_msg_and_die("block count doesn't fit in 32 bits"); +#define kilobytes kilobytes_unused_after_this + // Experimentally, standard mke2fs won't work on images smaller than 60k + if (nblocks < 60) + bb_error_msg_and_die("need >= 60 blocks"); + + // How many reserved blocks? + if (reserved_percent > 50) bb_error_msg_and_die("-%c is bad", 'm'); + nreserved = (uint64_t)nblocks * reserved_percent / 100; + + // N.B. killing e2fsprogs feature! Unused blocks don't account in calculations + nblocks_full = nblocks; + + // If last block group is too small, nblocks may be decreased in order + // to discard it, and control returns here to recalculate some + // parameters. + // Note: blocksize and bytes_per_inode are never recalculated. + retry: + // N.B. a block group can have no more than blocks_per_group blocks + ngroups = div_roundup(nblocks - first_block, blocks_per_group); + + group_desc_blocks = div_roundup(ngroups, blocksize / sizeof(*gd)); + // TODO: reserved blocks must be marked as such in the bitmaps, + // or resulting filesystem is corrupt + if (ENABLE_FEATURE_MKFS_EXT2_RESERVED_GDT) { + /* + * From e2fsprogs: Calculate the number of GDT blocks to reserve for online + * filesystem growth. + * The absolute maximum number of GDT blocks we can reserve is determined by + * the number of block pointers that can fit into a single block. + * We set it at 1024x the current filesystem size, or + * the upper block count limit (2^32), whichever is lower. + */ + uint32_t reserved_group_desc_blocks = 0xFFFFFFFF; // maximum block number + if (nblocks < reserved_group_desc_blocks / 1024) + reserved_group_desc_blocks = nblocks * 1024; + reserved_group_desc_blocks = div_roundup(reserved_group_desc_blocks - first_block, blocks_per_group); + reserved_group_desc_blocks = div_roundup(reserved_group_desc_blocks, blocksize / sizeof(*gd)) - group_desc_blocks; + if (reserved_group_desc_blocks > blocksize / sizeof(uint32_t)) + reserved_group_desc_blocks = blocksize / sizeof(uint32_t); + //TODO: STORE_LE(sb->s_reserved_gdt_blocks, reserved_group_desc_blocks); + group_desc_blocks += reserved_group_desc_blocks; + } + + { + // N.B. e2fsprogs does as follows! + uint32_t overhead, remainder; + // ninodes is the max number of inodes in this filesystem + uint32_t ninodes = ((uint64_t) nblocks_full * blocksize) / bytes_per_inode; + if (ninodes < EXT2_GOOD_OLD_FIRST_INO+1) + ninodes = EXT2_GOOD_OLD_FIRST_INO+1; + inodes_per_group = div_roundup(ninodes, ngroups); + // minimum number because the first EXT2_GOOD_OLD_FIRST_INO-1 are reserved + if (inodes_per_group < 16) + inodes_per_group = 16; + // a block group can't have more inodes than blocks + if (inodes_per_group > blocks_per_group) + inodes_per_group = blocks_per_group; + // adjust inodes per group so they completely fill the inode table blocks in the descriptor + inodes_per_group = (div_roundup(inodes_per_group * inodesize, blocksize) * blocksize) / inodesize; + // make sure the number of inodes per group is a multiple of 8 + inodes_per_group &= ~7; + inode_table_blocks = div_roundup(inodes_per_group * inodesize, blocksize); + + // to be useful, lost+found should occupy at least 2 blocks (but not exceeding 16*1024 bytes), + // and at most EXT2_NDIR_BLOCKS. So reserve these blocks right now + /* Or e2fsprogs comment verbatim (what does it mean?): + * Ensure that lost+found is at least 2 blocks, so we always + * test large empty blocks for big-block filesystems. */ + lost_and_found_blocks = MIN(EXT2_NDIR_BLOCKS, 16 >> (blocksize_log2 - EXT2_MIN_BLOCK_LOG_SIZE)); + + // the last group needs more attention: isn't it too small for possible overhead? + overhead = (has_super(ngroups - 1) ? (1/*sb*/ + group_desc_blocks) : 0) + 1/*bbmp*/ + 1/*ibmp*/ + inode_table_blocks; + remainder = (nblocks - first_block) % blocks_per_group; + ////can't happen, nblocks >= 60 guarantees this + ////if ((1 == ngroups) + //// && remainder + //// && (remainder < overhead + 1/* "/" */ + lost_and_found_blocks) + ////) { + //// bb_error_msg_and_die("way small device"); + ////} + + // Standard mke2fs uses 50. Looks like a bug in our calculation + // of "remainder" or "overhead" - we don't match standard mke2fs + // when we transition from one group to two groups + // (a bit after 8M image size), but it works for two->three groups + // transition (at 16M). + if (remainder && (remainder < overhead + 50)) { +//bb_info_msg("CHOP[%u]", remainder); + nblocks -= remainder; + goto retry; + } + } - // check the device is a block device - xstat(argv[0], &st); - if (!S_ISBLK(st.st_mode) && !(opts & OPT_F)) - bb_error_msg_and_die("not a block device"); + if (nblocks_full - nblocks) + printf("warning: %u blocks unused\n\n", nblocks_full - nblocks); + printf( + "Filesystem label=%s\n" + "OS type: Linux\n" + "Block size=%u (log=%u)\n" + "Fragment size=%u (log=%u)\n" + "%u inodes, %u blocks\n" + "%u blocks (%u%%) reserved for the super user\n" + "First data block=%u\n" + "Maximum filesystem blocks=%u\n" + "%u block groups\n" + "%u blocks per group, %u fragments per group\n" + "%u inodes per group" + , label + , blocksize, blocksize_log2 - EXT2_MIN_BLOCK_LOG_SIZE + , blocksize, blocksize_log2 - EXT2_MIN_BLOCK_LOG_SIZE + , inodes_per_group * ngroups, nblocks + , nreserved, reserved_percent + , first_block + , group_desc_blocks * (blocksize / (unsigned)sizeof(*gd)) * blocks_per_group + , ngroups + , blocks_per_group, blocks_per_group + , inodes_per_group + ); + { + const char *fmt = "\nSuperblock backups stored on blocks:\n" + "\t%u"; + pos = first_block; + for (i = 1; i < ngroups; i++) { + pos += blocks_per_group; + if (has_super(i)) { + printf(fmt, (unsigned)pos); + fmt = ", %u"; + } + } + } + bb_putchar('\n'); - // 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 mounted! - if (find_mount_point(argv[0], 0)) - bb_error_msg_and_die("can't format mounted filesystem"); + if (option_mask32 & OPT_n) { + if (ENABLE_FEATURE_CLEAN_UP) + close(fd); + return EXIT_SUCCESS; + } - // TODO: 5?/5 WE MUST NOT DEPEND ON WHETHER DEVICE IS /dev/zero 'ed OR NOT // TODO: 3/5 refuse if mounted // TODO: 4/5 compat options // TODO: 1/5 sanity checks // TODO: 0/5 more verbose error messages - // TODO: 0/5 info printing - // TODO: 2/5 bigendianness! Spot where it comes to play! sb->, gd-> + // TODO: 4/5 bigendianness: recheck, wait for ARM reporters // TODO: 2/5 reserved GDT: how to mark but not allocate? - // TODO: 0/5 dir_index? + // TODO: 3/5 dir_index? // fill the superblock - sb = xzalloc(blocksize); - sb->s_rev_level = 1; // revision 1 filesystem - sb->s_magic = EXT2_SUPER_MAGIC; - sb->s_inode_size = sizeof(*inode); - sb->s_first_ino = EXT2_GOOD_OLD_FIRST_INO; - sb->s_log_block_size = sb->s_log_frag_size = int_log2(blocksize >> EXT2_MIN_BLOCK_LOG_SIZE); + sb = xzalloc(1024); + STORE_LE(sb->s_rev_level, EXT2_DYNAMIC_REV); // revision 1 filesystem + STORE_LE(sb->s_magic, EXT2_SUPER_MAGIC); + STORE_LE(sb->s_inode_size, inodesize); + // set "Required extra isize" and "Desired extra isize" fields to 28 + if (inodesize != sizeof(*inode)) + STORE_LE(sb->s_reserved[21], 0x001C001C); + STORE_LE(sb->s_first_ino, EXT2_GOOD_OLD_FIRST_INO); + STORE_LE(sb->s_log_block_size, blocksize_log2 - EXT2_MIN_BLOCK_LOG_SIZE); + STORE_LE(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); + // the first block is 1, otherwise 0 + STORE_LE(sb->s_first_data_block, first_block); // 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; + STORE_LE(sb->s_blocks_per_group, blocks_per_group); + STORE_LE(sb->s_frags_per_group, blocks_per_group); + // blocks + STORE_LE(sb->s_blocks_count, nblocks); + // reserve blocks for superuser + STORE_LE(sb->s_r_blocks_count, nreserved); + // ninodes + STORE_LE(sb->s_inodes_per_group, inodes_per_group); + STORE_LE(sb->s_inodes_count, inodes_per_group * ngroups); + STORE_LE(sb->s_free_inodes_count, inodes_per_group * ngroups - EXT2_GOOD_OLD_FIRST_INO); + // timestamps 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 + STORE_LE(sb->s_mkfs_time, timestamp); + STORE_LE(sb->s_wtime, timestamp); + STORE_LE(sb->s_lastcheck, timestamp); + // misc. Values are chosen to match mke2fs 1.41.9 + STORE_LE(sb->s_state, 1); // TODO: what's 1? + STORE_LE(sb->s_creator_os, EXT2_OS_LINUX); + STORE_LE(sb->s_checkinterval, 24*60*60 * 180); // 180 days + STORE_LE(sb->s_errors, EXT2_ERRORS_DEFAULT); + // mke2fs 1.41.9 also sets EXT3_FEATURE_COMPAT_RESIZE_INODE + // and if >= 0.5GB, EXT3_FEATURE_RO_COMPAT_LARGE_FILE. + // we use values which match "mke2fs -O ^resize_inode": + // in this case 1.41.9 never sets EXT3_FEATURE_RO_COMPAT_LARGE_FILE. + STORE_LE(sb->s_feature_compat, EXT2_FEATURE_COMPAT_SUPP | (EXT2_FEATURE_COMPAT_RESIZE_INO * ENABLE_FEATURE_MKFS_EXT2_RESERVED_GDT) | (EXT2_FEATURE_COMPAT_DIR_INDEX * ENABLE_FEATURE_MKFS_EXT2_DIR_INDEX) - ; - // e2fsprogs-1.41.9 doesn't like EXT2_FEATURE_INCOMPAT_WHITEOUT - sb->s_feature_incompat = EXT2_FEATURE_INCOMPAT_FILETYPE;// | EXT2_FEATURE_INCOMPAT_WHITEOUT; - sb->s_feature_ro_compat = EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER; - sb->s_flags = EXT2_FLAGS_SIGNED_HASH * ENABLE_FEATURE_MKFS_EXT2_DIR_INDEX; + ); + STORE_LE(sb->s_feature_incompat, EXT2_FEATURE_INCOMPAT_FILETYPE); + STORE_LE(sb->s_feature_ro_compat, EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER); + STORE_LE(sb->s_flags, EXT2_FLAGS_UNSIGNED_HASH * ENABLE_FEATURE_MKFS_EXT2_DIR_INDEX); generate_uuid(sb->s_uuid); -#if ENABLE_FEATURE_MKFS_EXT2_DIR_INDEX - sb->s_def_hash_version = EXT2_HASH_HALF_MD4; - generate_uuid((uint8_t *)sb->s_hash_seed); -#endif + if (ENABLE_FEATURE_MKFS_EXT2_DIR_INDEX) { + STORE_LE(sb->s_def_hash_version, EXT2_HASH_HALF_MD4); + generate_uuid((uint8_t *)sb->s_hash_seed); + } /* * From e2fsprogs: add "jitter" to the superblock's check interval so that we * 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; - - // open the device, get number of blocks - xmove_fd(xopen3(argv[0], O_WRONLY | O_CREAT, 0666), fd); - if (argv[1]) { - nblocks = xatou(argv[1]); - } else { - nblocks = ((uoff_t)xlseek(fd, 0, SEEK_END)) / blocksize; - } - sb->s_blocks_count = nblocks; + STORE_LE(sb->s_max_mnt_count, + EXT2_DFL_MAX_MNT_COUNT + + (sb->s_uuid[ARRAY_SIZE(sb->s_uuid)-1] % EXT2_DFL_MAX_MNT_COUNT)); - // nblocks is the total number of blocks in the filesystem - if (nblocks < 8) - bb_error_msg_and_die("nblocks"); - // reserve blocks for superuser - sb->s_r_blocks_count = ((uint64_t) nblocks * nreserved) / 100; + // write the label + safe_strncpy((char *)sb->s_volume_name, label, sizeof(sb->s_volume_name)); - // 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 - * filesystem growth. - * The absolute maximum number of GDT blocks we can reserve is determined by - * the number of block pointers that can fit into a single block. - */ - /* 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; - } - ninodes = nblocks / (bytes_per_inode / blocksize); - 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 10 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; - // 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) / 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; - - itsz = ninodes_per_group * sb->s_inode_size / blocksize; - sb->s_free_inodes_count = sb->s_inodes_count - EXT2_GOOD_OLD_FIRST_INO; - - // write the label, if any - if (opts & OPT_L) - safe_strncpy((char *)sb->s_volume_name, label, sizeof(sb->s_volume_name)); - - // fill group descriptors - gd = xzalloc((gdtsz + rgdtsz) * blocksize); + // calculate filesystem skeleton structures + gd = xzalloc(group_desc_blocks * blocksize); + buf = xmalloc(blocksize); sb->s_free_blocks_count = 0; - for (i = 0, pos = first_data_block, n = nblocks; + for (i = 0, pos = first_block, n = nblocks - first_block; 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); - gd[i].bg_block_bitmap = overhead + 0; - gd[i].bg_inode_bitmap = overhead + 1; - gd[i].bg_inode_table = overhead + 2; - overhead = overhead - pos + 1/*bbmp*/ + 1/*ibmp*/ + itsz; - gd[i].bg_free_inodes_count = ninodes_per_group; - // N.B. both root and lost+found dirs are within the first block group, thus +2 - //gd[i].bg_used_dirs_count = 0; + uint32_t overhead = pos + (has_super(i) ? (1/*sb*/ + group_desc_blocks) : 0); + uint32_t free_blocks; + // fill group descriptors + STORE_LE(gd[i].bg_block_bitmap, overhead + 0); + STORE_LE(gd[i].bg_inode_bitmap, overhead + 1); + STORE_LE(gd[i].bg_inode_table, overhead + 2); + overhead = overhead - pos + 1/*bbmp*/ + 1/*ibmp*/ + inode_table_blocks; + gd[i].bg_free_inodes_count = inodes_per_group; + //STORE_LE(gd[i].bg_used_dirs_count, 0); + // N.B. both "/" and "/lost+found" are within the first block group + // "/" occupies 1 block, "/lost+found" occupies lost_and_found_blocks... if (0 == i) { - overhead += 2; - gd[i].bg_used_dirs_count = 2; + // ... thus increased overhead for the first block group ... + overhead += 1 + lost_and_found_blocks; + // ... and 2 used directories + STORE_LE(gd[i].bg_used_dirs_count, 2); + // well known reserved inodes belong to the first block too gd[i].bg_free_inodes_count -= EXT2_GOOD_OLD_FIRST_INO; } - // N.B. the following is pure heuristics! - // Likely to cope with 1024-byte blocks, when first block is for boot sectors - if (ngroups-1 == i) { - overhead += first_data_block; - } - gd[i].bg_free_blocks_count = (n < nblocks_per_group ? n : nblocks_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); - - // dump filesystem skeleton structures - 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; - 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 -//bb_info_msg("SUPER@[%d]", pos); - // N.B. 1024 byte blocks are special - PUT(blocksize * pos + ((0 == i && 0 == first_data_block) ? 1024 : 0), sb, blocksize); - PUT(blocksize * pos + blocksize, gd, (gdtsz + rgdtsz) * blocksize); - } + // cache free block count of the group + free_blocks = (n < blocks_per_group ? n : blocks_per_group) - overhead; - 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); // mark preallocated blocks as allocated - allocate(buf, blocksize, start, end); +//bb_info_msg("ALLOC: [%u][%u][%u]", blocksize, overhead, blocks_per_group - (free_blocks + overhead)); + allocate(buf, blocksize, + // reserve "overhead" blocks + overhead, + // mark unused trailing blocks + blocks_per_group - (free_blocks + overhead) + ); // dump block bitmap - PUT((pos + overhead) * blocksize, buf, blocksize); + PUT((uint64_t)(FETCH_LE32(gd[i].bg_block_bitmap)) * blocksize, buf, blocksize); + STORE_LE(gd[i].bg_free_blocks_count, free_blocks); // mark preallocated inodes as allocated allocate(buf, blocksize, - ninodes_per_group - gd[i].bg_free_inodes_count, - 8*blocksize - ninodes_per_group + // mark reserved inodes + inodes_per_group - gd[i].bg_free_inodes_count, + // mark unused trailing inodes + blocks_per_group - inodes_per_group ); // dump inode bitmap - PUT((pos + overhead + 1) * blocksize, buf, blocksize); + //PUT((uint64_t)(FETCH_LE32(gd[i].bg_block_bitmap)) * blocksize, buf, blocksize); + //but it's right after block bitmap, so we can just: + xwrite(fd, buf, blocksize); + STORE_LE(gd[i].bg_free_inodes_count, gd[i].bg_free_inodes_count); + + // count overall free blocks + sb->s_free_blocks_count += free_blocks; + } + STORE_LE(sb->s_free_blocks_count, sb->s_free_blocks_count); + + // dump filesystem skeleton structures +// printf("Writing superblocks and filesystem accounting information: "); + for (i = 0, pos = first_block; i < ngroups; i++, pos += blocks_per_group) { + // dump superblock and group descriptors and their backups + if (has_super(i)) { + // N.B. 1024 byte blocks are special + PUT(((uint64_t)pos * blocksize) + ((0 == i && 1024 != blocksize) ? 1024 : 0), + sb, 1024); + PUT(((uint64_t)pos * blocksize) + blocksize, + gd, group_desc_blocks * blocksize); + } } // zero boot sectors memset(buf, 0, blocksize); - PUT(0, buf, 1024); // N.B. 1024 <= blocksize + PUT(0, buf, 1024); // N.B. 1024 <= blocksize, so buf[0..1023] contains zeros // zero inode tables for (i = 0; i < ngroups; ++i) - for (n = 0; n < itsz; ++n) - PUT((gd[i].bg_inode_table + n) * blocksize, buf, blocksize); + for (n = 0; n < inode_table_blocks; ++n) + PUT((uint64_t)(FETCH_LE32(gd[i].bg_inode_table) + n) * blocksize, + buf, blocksize); // prepare directory inode inode = (struct ext2_inode *)buf; STORE_LE(inode->i_mode, S_IFDIR | S_IRWXU | S_IRGRP | S_IROTH | S_IXGRP | S_IXOTH); - inode->i_mtime = inode->i_atime = timestamp; + STORE_LE(inode->i_mtime, timestamp); + STORE_LE(inode->i_atime, timestamp); STORE_LE(inode->i_ctime, timestamp); STORE_LE(inode->i_size, blocksize); - // N.B. inode->i_blocks stores the number of 512 byte data blocks. Why on Earth?! + // inode->i_blocks stores the number of 512 byte data blocks + // (512, because it goes directly to struct stat without scaling) STORE_LE(inode->i_blocks, blocksize / 512); // dump root dir inode STORE_LE(inode->i_links_count, 3); // "/.", "/..", "/lost+found/.." point to this inode - STORE_LE(inode->i_block[0], gd[0].bg_inode_table + itsz); - PUT(gd[0].bg_inode_table * blocksize + (EXT2_ROOT_INO-1) * sizeof(*inode), buf, sizeof(*inode)); + STORE_LE(inode->i_block[0], FETCH_LE32(gd[0].bg_inode_table) + inode_table_blocks); + PUT(((uint64_t)FETCH_LE32(gd[0].bg_inode_table) * blocksize) + (EXT2_ROOT_INO-1) * inodesize, + buf, inodesize); // dump lost+found dir inode STORE_LE(inode->i_links_count, 2); // both "/lost+found" and "/lost+found/." point to this inode - STORE_LE(inode->i_block[0], inode->i_block[0]+1); // use next block //= gd[0].bg_inode_table + itsz + 1; - PUT(gd[0].bg_inode_table * blocksize + (EXT2_GOOD_OLD_FIRST_INO-1) * sizeof(*inode), buf, sizeof(*inode)); + STORE_LE(inode->i_size, lost_and_found_blocks * blocksize); + STORE_LE(inode->i_blocks, (lost_and_found_blocks * blocksize) / 512); + n = FETCH_LE32(inode->i_block[0]) + 1; + for (i = 0; i < lost_and_found_blocks; ++i) + STORE_LE(inode->i_block[i], i + n); // use next block +//bb_info_msg("LAST BLOCK USED[%u]", i + n); + PUT(((uint64_t)FETCH_LE32(gd[0].bg_inode_table) * blocksize) + (EXT2_GOOD_OLD_FIRST_INO-1) * inodesize, + buf, inodesize); // dump directories memset(buf, 0, blocksize); dir = (struct ext2_dir *)buf; - // dump lost+found dir block + // dump 2nd+ blocks of "/lost+found" + STORE_LE(dir->rec_len1, blocksize); // e2fsck 1.41.4 compat (1.41.9 does not need this) + for (i = 1; i < lost_and_found_blocks; ++i) + PUT((uint64_t)(FETCH_LE32(gd[0].bg_inode_table) + inode_table_blocks + 1+i) * blocksize, + buf, blocksize); + + // dump 1st block of "/lost+found" STORE_LE(dir->inode1, EXT2_GOOD_OLD_FIRST_INO); STORE_LE(dir->rec_len1, 12); STORE_LE(dir->name_len1, 1); @@ -457,7 +632,7 @@ int mkfs_ext2_main(int argc UNUSED_PARAM, char **argv) STORE_LE(dir->name_len2, 2); STORE_LE(dir->file_type2, EXT2_FT_DIR); dir->name2[0] = '.'; dir->name2[1] = '.'; - PUT((gd[0].bg_inode_table + itsz + 1) * blocksize, buf, blocksize); + PUT((uint64_t)(FETCH_LE32(gd[0].bg_inode_table) + inode_table_blocks + 1) * blocksize, buf, blocksize); // dump root dir block STORE_LE(dir->inode1, EXT2_ROOT_INO); @@ -467,15 +642,15 @@ int mkfs_ext2_main(int argc UNUSED_PARAM, char **argv) STORE_LE(dir->name_len3, 10); STORE_LE(dir->file_type3, EXT2_FT_DIR); strcpy(dir->name3, "lost+found"); - PUT((gd[0].bg_inode_table + itsz + 0) * blocksize, buf, blocksize); + PUT((uint64_t)(FETCH_LE32(gd[0].bg_inode_table) + inode_table_blocks + 0) * blocksize, buf, blocksize); // cleanup if (ENABLE_FEATURE_CLEAN_UP) { free(buf); free(gd); free(sb); - close(fd); } + xclose(fd); return EXIT_SUCCESS; }