[ext4]12分配机制-关键的数据结构

11-27

在块分配机制中,涉及到几个主要的数据结构。

通过ext4_allocation_request描述块请求,然后基于块查找结果即上层需求来决定是否执行块分配操作。

在分配过程中,为了更好执行分配,记录一些信息,需要对分配行为进行描述,就有结构体ext4_allocation_contex。

在搜寻可用空间过程中,是有可能使用预分配空间的,因此还需要有能够描述预分配空间大小等属性的描述符ext4_prealloc_space。

下面,对各个关键结构体进行详细的分析。

1. 块请求描述符ext4_allocation_request

块分配请求属性,有请求描述符ext4_allocation_request来描述:

structext4_allocation_request {

/* target inode for block we'reallocating */

struct inode *inode;

/* how many blocks we want to allocate*/

unsigned int len;

/* logical block in target inode */

ext4_lblk_t logical;

/* the closest logical allocated blockto the left */

ext4_lblk_t lleft;

/* the closest logical allocated blockto the right */

ext4_lblk_t lright;

/* phys. target (a hint) */

ext4_fsblk_t goal;

/* phys. block for the closest logicalallocated block to the left */

ext4_fsblk_t pleft;

/* phys. block for the closest logicalallocated block to the right */

ext4_fsblk_t pright;

/* flags. see above EXT4_MB_HINT_* */

unsigned int flags;

};

这个请求描述符结构体在ext4_ext_map_blocks()中初始化(注:ext4_ext_map_blocks()的作用是查找或分配指定的block块,并完成与缓存空间的映射)。

具体上述信息也就一个成员变量goal值的我们分析一下,goal记录是物理块号,其隐含含义比较重要:goal虽然只是记录物理块号,但是这个物理块号的选择可以很大程度的是文件保证locality特性及其物理地址连续性。

goal是由函数ext4_ext_find_goal()来定义:

static ext4_fsblk_t ext4_ext_find_goal(struct inode*inode,

struct ext4_ext_path *path,

ext4_lblk_t block)

{

if(path) {

intdepth = path->p_depth;

structext4_extent *ex;

/*

* Try to predict block placement assuming thatwe are

* filling in a file which will eventually be

* non-sparse --- i.e., in the case of libbfdwriting

* an ELF object sections out-of-order but in away

* the eventually results in a contiguousobject or

* executable file, or some database extendinga table

* space file. However, this is actually somewhat

* non-ideal if we are writing a sparse filesuch as

* qemu or KVM writing a raw image file that isgoing

* to stay fairly sparse, since it will end up

* fragmenting the file system's free space. Maybe we

* should have some hueristics or some way toallow

* userspace to pass a hint to file system,

* especially if the latter case turns out tobe

* common.

*/

ex= path[depth].p_ext;

if(ex) {

ext4_fsblk_text_pblk = ext4_ext_pblock(ex);

ext4_lblk_text_block = le32_to_cpu(ex->ee_block);

if(block > ext_block)

returnext_pblk + (block - ext_block);

else

returnext_pblk - (ext_block - block);

}

/*it looks like index is empty;

* try to find starting block from index itself*/

if(path[depth].p_bh)

returnpath[depth].p_bh->b_blocknr;

}

/*OK. use inode's group */

returnext4_inode_to_goal_block(inode);

}

细细分析这段代码,如果从根目录到指定逻辑块的path存在,那么就需要根据path来计算目标物理块的地址。

(1) Path的终点若是dataextent,则说明该path是从根到叶子的。当请求block号大于path叶子extent的起始逻辑块号ext_block (对应物理块号为pblk),其逻辑块的距离为(block-ext_block),为在最可能上保证对应物理地址的连续性;只需返回与pblk+(block-ext_block)物理块号最接近的空闲物理块即可;而对于请求block号小于extent的起始逻辑块号ext_block的情况,只需尽最可能以pblk-( ext_block -block)物理块号为目标寻找与其物理地址最接近的空闲物理块即可。因此,我们指定goal分别为pblk+(block-ext_block)和pblk-(block-ext_block)。

(2) 而如果path存在,却没有叶子,那则么办,很简单,我们只需要将goal物理块号指定为最后一个的extent block对应的物理块号既可。

(3) 还有一种情况,没有给出path。个人认为,这种场景即inode刚create的情况。有专门的ext4_inode_to_goal_block()来实现:

ext4_fsblk_t ext4_inode_to_goal_block(struct inode*inode)

{

structext4_inode_info *ei = EXT4_I(inode);

ext4_group_tblock_group;

ext4_grpblk_tcolour;

intflex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));

ext4_fsblk_tbg_start;

ext4_fsblk_tlast_block;

block_group= ei->i_block_group;

if(flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {

/*

* If there are at leastEXT4_FLEX_SIZE_DIR_ALLOC_SCHEME

* block groups per flexgroup, reserve thefirst block

* group for directories and special files. Regular

* files will start at the second blockgroup. This

* tends to speed up directory access andimproves

* fsck times.

*/

block_group&= ~(flex_size-1);

if(S_ISREG(inode->i_mode))

block_group++;

}

bg_start= ext4_group_first_block_no(inode->i_sb, block_group);

last_block= ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;

/*

* If we are doing delayed allocation, we don'tneed take

* colour into account.

*/

if(test_opt(inode->i_sb, DELALLOC))

returnbg_start;

if(bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)

colour= (current->pid % 16) *

(EXT4_BLOCKS_PER_GROUP(inode->i_sb)/ 16);

else

colour= (current->pid % 16) * ((last_block - bg_start) / 16);

returnbg_start + colour;

}

其思想是:如果flex_size至少有EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME个block groups,则定义inode所在flex_group的第二个block group的首个可用block为起始物理块号bg_block。

当然,如果该flex_group的所有文件都以bg_block为goal的,肯定会产生竞争,所以增加color的作用,目的就是加入一个随机值,降低可能带来的竞争。

因此,最后这种情况的goal会选择inode所在flex_group中某个随机值。

【说明:如果flex_size只有不小于EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME,则才有可能将flex_group中第一个group分离出来,用于专门存放directories和一些特殊文件,普通文件从第二个group中分配,该特可以加速directory的访问及fsync效率。】

2. 分配行为描述符ext4_allocation_contex

在分配过程中,为了更好执行分配,记录一些信息,需要对分配行为进行描述,就有结构体ext4_allocation_contex:

struct ext4_allocation_context{

struct inode *ac_inode;

struct super_block *ac_sb;

/* original request */

struct ext4_free_extent ac_o_ex;

/* goal request (normalized ac_o_ex) */

struct ext4_free_extent ac_g_ex;

/* the best found extent */

struct ext4_free_extent ac_b_ex;

/* copy of the best found extent takenbefore preallocation efforts */

struct ext4_free_extent ac_f_ex;

__u16 ac_groups_scanned;

__u16 ac_found;

__u16 ac_tail;

__u16 ac_buddy;

__u16 ac_flags; /* allocation hints */

__u8 ac_status;

__u8 ac_criteria;

__u8 ac_2order; /* if request is to allocate 2^N blocks and

* N > 0, the field stores N, otherwise 0 */

__u8 ac_op; /* operation, for history only */

struct page *ac_bitmap_page;

struct page *ac_buddy_page;

struct ext4_prealloc_space *ac_pa;

struct ext4_locality_group *ac_lg;

};

这个数据结构用来描述分配上下文的属性。基于结构体ext4_allocation_request,由函数ext4_mb_initialize_context()进行初始化。

ext4_mb_initialize_context()主要工作: 利用请求描述符的信息初始化ac->ac_o_ex:申请的逻辑块号fe_logical、goal所在的group,goal的cluster号(暂时理解为物理块号);然后将ac_g_ex 赋值为ac_o_ex。

ext4_mb_normalize_request()会对ext4_allocation_contex结构体进行normalization:

1.计算file的大小size应该是i_size_read(ac->ac_inode)和(offset+请求长度)中的大值,其中offset是有指定block转化而来。

2. 根据已定的算法估算文件可能的大小;

#define NRL_CHECK_SIZE(req, size, max, chunk_size)

(req<= (size) || max <= (chunk_size))

/*first, try to predict filesize */

/*XXX: should this table be tunable? */

start_off= 0;

if(size <= 16 * 1024) {

size= 16 * 1024;

}else if (size <= 32 * 1024) {

size= 32 * 1024;

}else if (size <= 64 * 1024) {

size= 64 * 1024;

}else if (size <= 128 * 1024) {

size= 128 * 1024;

}else if (size <= 256 * 1024) {

size= 256 * 1024;

}else if (size <= 512 * 1024) {

size= 512 * 1024;

}else if (size <= 1024 * 1024) {

size= 1024 * 1024;

}else if (NRL_CHECK_SIZE(size, 4 * 1024 * 1024, max, 2 * 1024)) {

start_off= ((loff_t)ac->ac_o_ex.fe_logical >>

(21- bsbits)) << 21;

size= 2 * 1024 * 1024;

}else if (NRL_CHECK_SIZE(size, 8 * 1024 * 1024, max, 4 * 1024)) {

start_off= ((loff_t)ac->ac_o_ex.fe_logical >>

(22- bsbits)) << 22;

size= 4 * 1024 * 1024;

}else if (NRL_CHECK_SIZE(ac->ac_o_ex.fe_len,

(8<<20)>>bsbits,max, 8 * 1024)) {

start_off= ((loff_t)ac->ac_o_ex.fe_logical >>

(23- bsbits)) << 23;

size= 8 * 1024 * 1024;

}else {

start_off= (loff_t)ac->ac_o_ex.fe_logical << bsbits;

size =ac->ac_o_ex.fe_len << bsbits;

}

size= size >> bsbits;

start= start_off >> bsbits;

由此可见,预估文件大小之后得到的size和start肯定比原来的要大一些。

3. check一下,是否覆盖了已有的prealloc空间。(如果覆盖,那就BUG);

4. 更新ac_g_ex:根据(2)中size和start更新ac_g_ex;

ac->ac_g_ex.fe_logical= start;

ac->ac_g_ex.fe_len= EXT4_NUM_B2C(sbi, size);

由上可见,通过ext4_mb_normalize_request()函数主要更新了ac->ac_g_ex成员。

而ac->ac_b_ex是在ext4_mb_regular_allocator()函数初始化的,其表示可以分配的最佳的extent;隐含意思,就是就按这么分配。

而ac-> ac_f_ex是在prealloc空间初始化之前保留ac_b_ex的副本,在ext4_mb_new_inode_pa()或ext4_mb_new_group_pa()中定义。

3. 预分配空间描述符ext4_allocation_contex

描述预分配空间大小等属性的描述符ext4_prealloc_space:

structext4_prealloc_space {

struct list_head pa_inode_list;

struct list_head pa_group_list;

union {

struct list_head pa_tmp_list;

struct rcu_head pa_rcu;

} u;

spinlock_t pa_lock;

atomic_t pa_count;

unsigned pa_deleted;

ext4_fsblk_t pa_pstart; /*phys. block */

ext4_lblk_t pa_lstart; /*log. block */

ext4_grpblk_t pa_len; /*len of preallocated chunk */

ext4_grpblk_t pa_free; /* howmany blocks are free */

unsigned short pa_type; /* pa type.inode or group */

spinlock_t *pa_obj_lock;

struct inode *pa_inode; /*hack, for history only */

};

其中有四个结构体非常重要:

pa_lstart -> prealloc 空间的起始逻辑地址(对文件而言);

pa_pstart -> prealloc 空间的起始物理地址;

pa_len -> prealloc 空间的长度;

pa_free -> prealloc 空间的可用长度;

这个结构体是在函数ext4_mb_new_inode_pa()或ext4_mb_new_group_pa()中初始化。

暂时就分析这么几个结构体吧。

作者:Younger Liu,

本作品采用知识共享署名-非商业性使用-相同方式共享 3.0 未本地化版本许可协议进行许可。