Linux下omfs文件系统的硬盘布局-服务器评测

文件系统是linux内核的重要组成部分，涉及到vfs、块IO层的调度机制，块设备驱动以及具体文件系统所采用的数据结构。所使用linux内核版本是2.6.34.1。

以fs/omfs为例，主要学习vfs的实现，omfs的硬盘布局，它所采用的数据结构为何能够优化MPEG文件系统。Omfs文件系统的具体文件读写又是如何实现的。

OMFS：Optimized MPEG Filesystem

OMFS是由SonicBlue公司创建的用于ReplayTV DVR和MP3 player的文件系统。该文件系统是基于extent的（现代很多文件系统都采用extent替代block来管理磁盘。Extent就是一些连续的block，可以有效减少元数据开销。），可用的block大小在2k到8k之间，目录结构是基于hash的。

该文件系统在特定的流媒体设备中性能很好，但对于一般的应用，linux主流的文件系统应该性能更优。Omfs是如何针对MPEG做性能优化，这点还在探索中。

相关阅读：

Linux文件系统omfs的目录创建和删除 http://www.linuxidc.com/Linux/2012-02/54025.htm

Linux文件系统omfs的普通文件创建和删除 http://www.linuxidc.com/Linux/2012-02/54026.htm

硬盘布局格式：

Omfs区分sysblocks和一般的数据blocks。Sysblock group由superblock信息、文件的metadata元数据、目录结构和extents构成。每一个sysblock都有一个包含CRC校验的头，而且可以在硬盘上备份。Sysblock大小比一个数据block小，但是它们都用64位的块号寻址。

Sysblock 头信息：

struct omfs_header { __be64 h_self; /* FS block where this is located */ __be32 h_body_size; /* size of useful data after header */ __be16 h_crc; /* crc-ccitt of body_size bytes */

char h_fill1[2];

u8 h_version; /* version, always 1 */

char h_type; /* OMFS_INODE_X */

u8 h_magic; /* OMFS_IMAGIC */

u8 h_check_xor; /* XOR of header bytes before this */ __be32 h_fill2; };

文件和目录都由omfs_inode表示：

struct omfs_inode { struct omfs_header i_head; /* header */

__be64 i_parent; /* parent containing this inode */ __be64 i_sibling; /* next inode in hash bucket */

__be64 i_ctime; /* ctime, in milliseconds */ char i_fill1[35];

char i_type; /* OMFS_[DIR,FILE] */

__be32 i_fill2;

char i_fill3[64];

char i_name[OMFS_NAMELEN]; /* filename */

__be64 i_size; /* size of file, in bytes */ };

OMFS中的目录是一个大的hash表。文件名经过hash计算，然后放到以OMFS_DIR_START开始的桶中。查找的时候需要hash文件名，然后通过i_sibling指针查找到匹配的i_name。

文件以omfs_inode结构体开头，后面跟着在OMFS_EXTENT_START开始的extent table。

struct omfs_extent_entry {

__be64 e_cluster; /* start location of a set of blocks */

__be64 e_blocks; /* number of blocks after e_cluster */

};

struct omfs_extent {

__be64 e_next; /* next extent table location */

__be32 e_extent_count; /* total # extents in this table */

__be32 e_fill;

struct omfs_extent_entry e_entry; /* start of extent entries */

};

通过Mkomfs.c可以了解omfs的基本布局：

在这里我们不使用实际的硬盘，用loop设备模拟一下：

生成一个文件：dd if=/dev/zero of=file.img bs=512 count=10000

10000+0 records in

10000+0 records out

5120000 bytes (5.1 MB) copied, 0.102142 s, 50.1 MB/s

用losetup /dev/loop0 file.img将loop设备和file.img关联上。

再用mkomfs /dev/loop0来布局：

Mkomfs的默认配置参数是：

fs_config_t config = {

.block_size = 8192,

.cluster_size = 8,

.clear_dev = 0

};

获取设备的大小：size=bs×count=512×10000=5120000

create_fs(fp, size/512, &config);中

block_size = 8192

blocks_per_sector = block_size / SECTOR_SIZE = 8192/512 = 16

blocks = sectors / blocks_per_sector = 10000/16 =625=0x271

初始化omfs_super_block结构体：

struct omfs_super_block {

char s_fill1[192];

char s_name[OMFS_SUPER_NAMELEN]; :”omfs”

__be64 s_root_block; /* block number of omfs_root_block */ :ROOT_BLK=1

__be64 s_num_blocks; /* total number of FS blocks */ :blocks=625=0x271

__be32 s_magic; /* OMFS_MAGIC */ : OMFS_MAGIC 0xC2993D87

__be32 s_blocksize; /* size of a block */ :block_size=8192=0x2000

__be32 s_mirrors; /* # of mirrors of system blocks */ :2

__be32 s_sys_blocksize; /* size of non-data blocks */ :2048=0x800

}; 288个字节

hexdump -C -s 0 -n 512 /dev/loop0

00000000 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |…………….|

000000c0 6f 6d 66 73 00 00 00 00 00 00 00 00 00 00 00 00 |omfs…………|

000000d0 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |…………….|

00000100 00 00 00 00 00 00 00 01 00 00 00 00 00 00 02 71 |……………q|

00000110 c2 99 3d 87 00 00 20 00 00 00 00 02 00 00 08 00 |..=… ………|

00000120 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |…………….|

00000200

将superblock的288个字节写在最开头，

初始化omfs_root_block结构体：

struct omfs_root_block {

struct omfs_header r_head; /* header */

__be64 r_fill1;

__be64 r_num_blocks; /* total number of FS blocks */ :blocks=625

__be64 r_root_dir; /* block # of root directory */ : ROOT_DIR_BLK=3

__be64 r_bitmap; /* block # of free space bitmap */ :BITMAP_BLK 5

__be32 r_blocksize; /* size of a block */ :block_size=8192

__be32 r_clustersize; /* size allocated for data blocks */ : cluster_size = 8

__be64 r_mirrors; /* # of mirrors of system blocks */ :2

char r_name[OMFS_NAMELEN]; /* partition label */ :”omfs”

__be64 r_fill2;

}; 336个字节

struct omfs_header {

__be64 h_self; /* FS block where this is located */ :ROOT_BLK=1

__be32 h_body_size; /* size of useful data after header */ :336-24=312=0x138

__be16 h_crc; /* crc-ccitt of body_size bytes */

char h_fill1[2];

u8 h_version; /* version, always 1 */ :1

char h_type; /* OMFS_INODE_X */ : OMFS_INODE_SYSTEM=’s’

u8 h_magic; /* OMFS_IMAGIC */ : OMFS_IMAGIC 0xD2

u8 h_check_xor; /* XOR of header bytes before this */

__be32 h_fill2;

}; 24个字节

然后移动到8192个字节，0x2000处，写入rootblock；再移动到0x4000处，再写一遍rootblock。

hexdump -C -s 0x2000 -n 512 /dev/loop0

00002000 00 00 00 00 00 00 00 01 00 00 01 38 00 00 00 00 |………..8….|

00002010 01 73 d2 00 00 00 00 00 00 00 00 00 00 00 00 00 |.s…………..|

00002020 00 00 00 00 00 00 02 71 00 00 00 00 00 00 00 03 |…….q……..|

00002030 00 00 00 00 00 00 00 05 00 00 20 00 00 00 00 08 |………. …..|

00002040 00 00 00 00 02 00 00 00 6f 6d 66 73 00 00 00 00 |……..omfs….|

00002050 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |…………….|

00002200

hexdump -C -s 0x4000 -n 512 /dev/loop0

00004000 00 00 00 00 00 00 00 01 00 00 01 38 00 00 00 00 |………..8….|

00004010 01 73 d2 00 00 00 00 00 00 00 00 00 00 00 00 00 |.s…………..|

00004020 00 00 00 00 00 00 02 71 00 00 00 00 00 00 00 03 |…….q……..|

00004030 00 00 00 00 00 00 00 05 00 00 20 00 00 00 00 08 |………. …..|

00004040 00 00 00 00 02 00 00 00 6f 6d 66 73 00 00 00 00 |……..omfs….|

00004050 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |…………….|

00004200

Root directory inode信息：

struct omfs_inode {

struct omfs_header i_head; /* header */

__be64 i_parent; /* parent containing this inode */ :~0

__be64 i_sibling; /* next inode in hash bucket */ :~0

__be64 i_ctime; /* ctime, in milliseconds */

char i_fill1[35];

char i_type; /* OMFS_[DIR,FILE] */ :’D’

__be32 i_fill2; :1

char i_fill3[64];

char i_name[OMFS_NAMELEN]; /* filename */

__be64 i_size; /* size of file, in bytes */ :2048

};

struct omfs_header {

__be64 h_self; /* FS block where this is located */ : ROOT_DIR_BLK 3

__be32 h_body_size; /* size of useful data after header */ :2048-24=2024

__be16 h_crc; /* crc-ccitt of body_size bytes */

char h_fill1[2];

u8 h_version; /* version, always 1 */ :1

char h_type; /* OMFS_INODE_X */ : OMFS_INODE_NORMAL ‘e’

u8 h_magic; /* OMFS_IMAGIC */ :OMFS_IMAGIC 0xD2

u8 h_check_xor; /* XOR of header bytes before this */

__be32 h_fill2;

};

申请一块大小为2048字节的内存，前面416=0x1a0个字节放root inode，

从OMFS_DIR_START 0x1b8到2048=0x800全部设置为0xff。

将这2048个字节写到0x6000位置，再重复将这2048个字节写到0x8000位置。

hexdump -C -s 0x6000 -n 512 /dev/loop0

00006000 00 00 00 00 00 00 00 03 00 00 07 e8 1c 5c 00 00 |………….\..|

00006010 01 65 d2 1a 00 00 00 00 ff ff ff ff ff ff ff ff |.e…………..|

00006020 ff ff ff ff ff ff ff ff 00 00 01 35 5a 8e 42 bb |………..5Z.B.|

00006030 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |…………….|

00006050 00 00 00 44 00 00 00 01 00 00 00 00 00 00 00 00 |…D…………|

00006060 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |…………….|

00006190 00 00 00 00 00 00 00 00 00 00 00 00 00 00 08 00 |…………….|

000061a0 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |…………….|

000061b0 00 00 00 00 00 00 00 00 ff ff ff ff ff ff ff ff |…………….|

000061c0 ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff |…………….|

00006200

hexdump -C -s 0x8000 -n 512 /dev/loop0

00008000 00 00 00 00 00 00 00 03 00 00 07 e8 1c 5c 00 00 |………….\..|

00008010 01 65 d2 1a 00 00 00 00 ff ff ff ff ff ff ff ff |.e…………..|

00008020 ff ff ff ff ff ff ff ff 00 00 01 35 5a 8e 42 bb |………..5Z.B.|

00008030 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |…………….|

00008050 00 00 00 44 00 00 00 01 00 00 00 00 00 00 00 00 |…D…………|

00008060 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |…………….|

00008190 00 00 00 00 00 00 00 00 00 00 00 00 00 00 08 00 |…………….|

000081a0 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |…………….|

000081b0 00 00 00 00 00 00 00 00 ff ff ff ff ff ff ff ff |…………….|

000081c0 ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff |…………….|

00008200

最后，就是free space bitmap了。

bitmap_size = (swap_be64(super.s_num_blocks) + 7)/8; =（625+7）/8=79

dirty_size = (bitmap_size + 7)/8; =（79+7）/8=10

first_blk = BITMAP_BLK + (bitmap_size +

swap_be32(super.s_blocksize)-1) / swap_be32(super.s_blocksize); =5+（79+8192-1）/8192=6

因为bitmap的每一位代表一个block，所以可以计算出bitmap的字节数。

Bitmap.bmap申请bitmap_size=79个字节的内存大小

for (i=0; i<first_blk; i++)

{

bitmap.bmap[i/8] |= 1<<(i & 7);

}

Bitmap.bmap[0] = 0011 1111 2进制 0x3f

因为blocks 0-5被用了，所以bitmap的相应位都置上1。

Bitmap.dirty申请dirty_size=10个字节的大小，每个字节都置为0xff

将bitmap写入0xa000。（第6个block,block是从block 0开始的，block 0放的是super block）。

hexdump -C -s 0xa000 -n 512 /dev/loop0

0000a000 3f 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |?……………|

0000a010 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |…………….|

0000a050 00 00 00 00 11 00 00 00 ff ff ff ff ff ff ff ff |…………….|

0000a060 ff ff 00 00 c9 0c 02 00 00 00 00 00 00 00 00 00 |…………….|

0000a070 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |…………….|

0000a190 00 00 00 00 00 00 00 00 00 00 00 00 00 00 08 00 |…………….|

0000a1a0 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |…………….|

0000a1b0 00 00 00 00 00 00 00 00 ff ff ff ff ff ff ff ff |…………….|

0000a1c0 ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff |…………….|

0000a200

到此，omfs文件系统格式化ok了，mount之后就可以进行常见的文件操作了。

Linux下omfs文件系统的硬盘布局

相关推荐

分类

听说打赏我的人，都进福布斯排行榜啦！

支付宝扫一扫打赏

微信扫一扫打赏

QQ咨询

回顶部