diff options
author | David Oberhollenzer <david.oberhollenzer@sigma-star.at> | 2019-04-30 13:06:16 +0200 |
---|---|---|
committer | David Oberhollenzer <david.oberhollenzer@sigma-star.at> | 2019-05-02 12:40:06 +0200 |
commit | e711928cb0931c1d14ac14cfad709c2a453d6c31 (patch) | |
tree | fca3ce13a04c37f9266f749cd1a094a9ae7979b2 /mkfs/super.c | |
parent | 5ffeac25e3ae614e9c6b7d274962c8e0084907fd (diff) |
Store globally interesting data in sqfs_info_t structure
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
Diffstat (limited to 'mkfs/super.c')
-rw-r--r-- | mkfs/super.c | 98 |
1 files changed, 49 insertions, 49 deletions
diff --git a/mkfs/super.c b/mkfs/super.c index b3755d1..c924b6c 100644 --- a/mkfs/super.c +++ b/mkfs/super.c @@ -4,52 +4,52 @@ #include <endian.h> -int sqfs_super_init(sqfs_super_t *s, const options_t *opt) +int sqfs_super_init(sqfs_info_t *info) { unsigned int i; - if (opt->blksz & (opt->blksz - 1)) { + if (info->opt.blksz & (info->opt.blksz - 1)) { fputs("Block size must be a power of 2!\n", stderr); return -1; } - if (opt->blksz < 8192 || opt->blksz >= (1 << 24)) { + if (info->opt.blksz < 8192 || info->opt.blksz >= (1 << 24)) { fputs("Block size must be between 8k and 1M\n", stderr); return -1; } - memset(s, 0, sizeof(*s)); - s->magic = SQFS_MAGIC; - s->modification_time = opt->def_mtime; - s->block_size = opt->blksz; - s->compression_id = opt->compressor; - s->flags = SQFS_FLAG_NO_FRAGMENTS | SQFS_FLAG_NO_XATTRS; - s->version_major = SQFS_VERSION_MAJOR; - s->version_minor = SQFS_VERSION_MINOR; - s->bytes_used = sizeof(*s); - s->id_table_start = 0xFFFFFFFFFFFFFFFFUL; - s->xattr_id_table_start = 0xFFFFFFFFFFFFFFFFUL; - s->inode_table_start = 0xFFFFFFFFFFFFFFFFUL; - s->directory_table_start = 0xFFFFFFFFFFFFFFFFUL; - s->fragment_table_start = 0xFFFFFFFFFFFFFFFFUL; - s->export_table_start = 0xFFFFFFFFFFFFFFFFUL; - - for (i = opt->blksz; i != 0x01; i >>= 1) - s->block_log += 1; + memset(&info->super, 0, sizeof(info->super)); + info->super.magic = SQFS_MAGIC; + info->super.modification_time = info->opt.def_mtime; + info->super.block_size = info->opt.blksz; + info->super.compression_id = info->opt.compressor; + info->super.flags = SQFS_FLAG_NO_FRAGMENTS | SQFS_FLAG_NO_XATTRS; + info->super.version_major = SQFS_VERSION_MAJOR; + info->super.version_minor = SQFS_VERSION_MINOR; + info->super.bytes_used = sizeof(info->super); + info->super.id_table_start = 0xFFFFFFFFFFFFFFFFUL; + info->super.xattr_id_table_start = 0xFFFFFFFFFFFFFFFFUL; + info->super.inode_table_start = 0xFFFFFFFFFFFFFFFFUL; + info->super.directory_table_start = 0xFFFFFFFFFFFFFFFFUL; + info->super.fragment_table_start = 0xFFFFFFFFFFFFFFFFUL; + info->super.export_table_start = 0xFFFFFFFFFFFFFFFFUL; + + for (i = info->opt.blksz; i != 0x01; i >>= 1) + info->super.block_log += 1; return 0; } -int sqfs_padd_file(sqfs_super_t *s, const options_t *opt, int outfd) +int sqfs_padd_file(sqfs_info_t *info) { - size_t padd_sz = s->bytes_used % opt->devblksz; + size_t padd_sz = info->super.bytes_used % info->opt.devblksz; uint8_t *buffer; ssize_t ret; if (padd_sz == 0) return 0; - padd_sz = opt->devblksz - padd_sz; + padd_sz = info->opt.devblksz - padd_sz; buffer = calloc(1, padd_sz); if (buffer == NULL) { @@ -57,7 +57,7 @@ int sqfs_padd_file(sqfs_super_t *s, const options_t *opt, int outfd) return -1; } - ret = write_retry(outfd, buffer, padd_sz); + ret = write_retry(info->outfd, buffer, padd_sz); if (ret < 0) { perror("Error padding squashfs image to page size"); @@ -75,35 +75,35 @@ int sqfs_padd_file(sqfs_super_t *s, const options_t *opt, int outfd) return 0; } -int sqfs_super_write(const sqfs_super_t *super, int outfd) +int sqfs_super_write(sqfs_info_t *info) { sqfs_super_t copy; ssize_t ret; - copy.magic = htole32(super->magic); - copy.inode_count = htole32(super->inode_count); - copy.modification_time = htole32(super->modification_time); - copy.block_size = htole32(super->block_size); - copy.fragment_entry_count = htole32(super->fragment_entry_count); - copy.compression_id = htole16(super->compression_id); - copy.block_log = htole16(super->block_log); - copy.flags = htole16(super->flags); - copy.id_count = htole16(super->id_count); - copy.version_major = htole16(super->version_major); - copy.version_minor = htole16(super->version_minor); - copy.root_inode_ref = htole64(super->root_inode_ref); - copy.bytes_used = htole64(super->bytes_used); - copy.id_table_start = htole64(super->id_table_start); - copy.xattr_id_table_start = htole64(super->xattr_id_table_start); - copy.inode_table_start = htole64(super->inode_table_start); - copy.directory_table_start = htole64(super->directory_table_start); - copy.fragment_table_start = htole64(super->fragment_table_start); - copy.export_table_start = htole64(super->export_table_start); - - if (lseek(outfd, 0, SEEK_SET) == (off_t)-1) + copy.magic = htole32(info->super.magic); + copy.inode_count = htole32(info->super.inode_count); + copy.modification_time = htole32(info->super.modification_time); + copy.block_size = htole32(info->super.block_size); + copy.fragment_entry_count = htole32(info->super.fragment_entry_count); + copy.compression_id = htole16(info->super.compression_id); + copy.block_log = htole16(info->super.block_log); + copy.flags = htole16(info->super.flags); + copy.id_count = htole16(info->super.id_count); + copy.version_major = htole16(info->super.version_major); + copy.version_minor = htole16(info->super.version_minor); + copy.root_inode_ref = htole64(info->super.root_inode_ref); + copy.bytes_used = htole64(info->super.bytes_used); + copy.id_table_start = htole64(info->super.id_table_start); + copy.xattr_id_table_start = htole64(info->super.xattr_id_table_start); + copy.inode_table_start = htole64(info->super.inode_table_start); + copy.directory_table_start = htole64(info->super.directory_table_start); + copy.fragment_table_start = htole64(info->super.fragment_table_start); + copy.export_table_start = htole64(info->super.export_table_start); + + if (lseek(info->outfd, 0, SEEK_SET) == (off_t)-1) goto fail_seek; - ret = write_retry(outfd, ©, sizeof(copy)); + ret = write_retry(info->outfd, ©, sizeof(copy)); if (ret < 0) { perror("squashfs writing super block"); @@ -116,7 +116,7 @@ int sqfs_super_write(const sqfs_super_t *super, int outfd) return -1; } - if (lseek(outfd, 0, SEEK_END) == (off_t)-1) + if (lseek(info->outfd, 0, SEEK_END) == (off_t)-1) goto fail_seek; return 0; |