Linux文件系统omfs文件的读写-服务器评测

为了研究omfs文件系统是如何进行文件的write操作，我在应用层写了个小程序：

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

#include <stdio.h>

#include <stdlib.h>

#include <sys/types.h>

#include <sys/stat.h>

#include <unistd.h>

#include <fcntl.h>

#include <errno.h>

Void main()

{

Int I;

Char w[16384];

Int fd;

Int count;

Fd = open(“/mnt/point1/cccc”,O_RDWR);

Perror(“open”);

For( I = 0; I < 16384; i++)

{

W[i] = 0x99;

}

Lseek(fd,0,SEEK_END);

Count = write(fd,w,16384);

Perror(“program”);

Printf(“count: %d\n”, count);

}

这个程序的目的主要在于写入/mnt/point1/目录下的某个文件，而我们的omfs文件系统就挂载在该目录下。

目前该目录下的情况如下：

ls –ali

接下来我们对cccc这个文件写入16384个字节，每个字节都是0x99.

在omfs_get_block函数下断点：

函数堆栈如下：

#0 omfs_get_block (inode=0xddf2c870, block=0, bh_result=0xddfd6818, create=1)

at fs/omfs/file.c:230

#1 0xc02c1c5f in __block_prepare_write (inode=0xddf2c870, page=0xc10d1020,

from=0, to=4096, get_block=0xe278d2ca <omfs_get_block>) at fs/buffer.c:1880

#2 0xc02c1f9d in block_write_begin (file=0xdd046480, mapping=0xddf2c920,

pos=0, len=4096, flags=0, pagep=0xddbb7d80, fsdata=0xddbb7d6c,

get_block=0xe278d2ca <omfs_get_block>) at fs/buffer.c:1995

#3 0xe278d642 in omfs_write_begin (file=0xdd046480, mapping=0xddf2c920,

pos=0, len=4096, flags=0, pagep=0xddbb7d80, fsdata=0xddbb7d6c)

at fs/omfs/file.c:316

#4 0xc0231a32 in generic_perform_write (file=0xdd046480, i=0xddbb7dc0, pos=0)

at mm/filemap.c:2236

#5 0xc0231bd9 in generic_file_buffered_write (iocb=0xddbb7ec4,

iov=0xddbb7f34, nr_segs=1, pos=0, ppos=0xddbb7ef8, count=16384, written=0)

at mm/filemap.c:2292

#6 0xc0231f99 in __generic_file_aio_write (iocb=0xddbb7ec4, iov=0xddbb7f34,

nr_segs=1, ppos=0xddbb7ef8) at mm/filemap.c:2410

#7 0xc023204a in generic_file_aio_write (iocb=0xddbb7ec4, iov=0xddbb7f34,

nr_segs=1, pos=0) at mm/filemap.c:2440

#8 0xc02902cf in do_sync_write (filp=0xdd046480,

\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231″…, len=16384, ppos=0xddbb7f94) at fs/read_write.c:320

#9 0xc02903fc in vfs_write (file=0xdd046480,

buf=0xbf90ab7c “\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231″…, count=16384, pos=0xddbb7f94) at fs/read_write.c:349

#10 0xc0290557 in sys_write (fd=3,

\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231\231″…, count=16384) at fs/read_write.c:401

#11 0xc0104657 in ?? () at arch/x86/kernel/entry_32.S:457

(gdb) p max_blocks

$61 = 1

(gdb) p bh_result->b_size

$62 = 2048

(gdb) p inode->i_blkbits

$63 = 11

因为cccc这个inode的i_no是18，所以sb_bread读的block是72=18×4

获取oe oe = (struct omfs_extent *)(&bh->b_data[OMFS_EXTENT_START]);

这个oe还是当初omfs_create的最初状态

max_extents = omfs_max_extents(sbi, OMFS_EXTENT_START);

得到的max_extents值为98

(sbi->s_sys_blocksize – offset –

sizeof(struct omfs_extent)) /

sizeof(struct omfs_extent_entry) + 1;

(2048-0x1d0-32)/16+1 = 98

从这个数据我们可以算出最大文件的大小是：98×8192×8 = 6422528字节=6.125M

next = inode->i_ino; next=18

接下来是一个循环：

for (;;) {

if (omfs_is_bad(sbi, (struct omfs_header *) bh->b_data, next)) //做一个检查

goto out_brelse;

extent_count = be32_to_cpu(oe->e_extent_count); //值为1 omfs_create时设定的

next = be64_to_cpu(oe->e_next); //值为0xffff ffff 同样是omfs_create时设定的

entry = &oe->e_entry;

if (extent_count > max_extents)

goto out_brelse;

offset = find_block(inode, entry, block, extent_count, &remain);

if (offset > 0) {

ret = 0;

map_bh(bh_result, inode->i_sb, offset);

if (remain > max_blocks)

remain = max_blocks;

bh_result->b_size = (remain << inode->i_blkbits);

goto out_brelse;

}

if (next == ~0)

break;

brelse(bh);

bh = sb_bread(inode->i_sb, clus_to_blk(sbi, next));

if (!bh)

goto out;

oe = (struct omfs_extent *) (&bh->b_data[OMFS_EXTENT_CONT]);

max_extents = omfs_max_extents(sbi, OMFS_EXTENT_CONT);

}

这个循环到底做了什么：

先看看find_block干了什么，从字面来看这个函数应该是查找空闲的block

因为count是1，所以这个find_block中的循环不执行，直接返回0.

接下来继续看omfs_get_block，

因为next == ～0，所以跳出循环。

if (create) {

ret = omfs_grow_extent(inode, oe, &new_block);

if (ret == 0) {

mark_buffer_dirty(bh);

mark_inode_dirty(inode);

map_bh(bh_result, inode->i_sb,

clus_to_blk(sbi, new_block));

}

omfs_grow_extent是个关键的函数，继续往里面看：

这个函数的注释是：

* Add new blocks to the current extent, or create new entries/continuations

* as necessary.

ret = omfs_allocate_range(inode->i_sb, 1, sbi->s_clustersize,

&new_block, &new_count);

New_block=28 new_count=8

entry->e_cluster = cpu_to_be64(new_block);

entry->e_blocks = cpu_to_be64((u64) new_count);

Breakpoint 5, omfs_get_block (inode=0xddf2c870, block=1, bh_result=0xddfd67e0,