Linux缓存机制之块缓存

作者：bullbat

在Linux内核中，并非总使用基于页的方法来承担缓存的任务。内核的早期版本只包含了块缓存，来加速文件操作和提高系统性能。这是来自于其他具有相同结构的类UNIX操作系统的遗产。来自于底层块设备的块缓存在内存的缓冲区中，可以加速读写操作。

与内存页相比，块不仅比较小（大多数情况下），而且长度可变的，依赖于使用的块设备（或文件系统）。随着日渐倾向于使用基于页操作实现的通用文件存取方法，块缓存作为中枢系统缓存的重要性已经逐渐失去。主要的缓存任务现在由页缓存承担。另外，基于块的I/O的标准数据结构，现在已经不再是缓冲区，而是struct bio结构。

缓冲区用作小型的数据传输，一般设计的数据量是与块长度可比拟的。文件系统在处理元数据时，通常会使用此类方法。而裸数据的传输则按页进行，而缓冲区的实现也基于也缓存。

块缓存在结构上由两个部分组成：

1） 缓冲头（buffer head）包含了与缓冲区状态相关的所有管理数据，包括快号、块长度、访问计数器等。这些数据不是直接存储在缓冲头之后，而是存储在物理内存的一个独立区域中，由缓冲头结构中的一个对应的指针表示。

2） 有用数据保存在专门分配的页中，这些页也可能同时存在于页缓存中。这进一步细分了页缓存，如下图所示，在我们的例子中，页划分为4个长度相同的部分，每一部分由其自身的缓冲头描述。缓冲头存储的内存区域与有用数据存储的区域是有关的。



这使得页面可以细分为更小的部分，各顾各部分之间完全连续的（因为缓冲区数据和缓冲头数据是分离的）。因为一个缓冲区由至少512字节组成，每页最多可包括MAX_BUF_PER_PAGE个缓冲区。该常数定义为页面长度的函数。

如果修改了某个缓冲区，则会立即印象到页面的内容（反之也是），因而两个缓存不需要显示同步，毕竟二者的数据是共享的。

当然，有些应用程序在访问块设备时，使用的是块而不是页面，读取文件系统的操作几块，就是一个例子。一个独立的块缓存用于加速此类访问。该块缓存的运作独立于页面缓存，而不是在其上建立的。为此，缓冲头数据结构（对于块缓存和页面缓存是相同的）群聚在一个长度恒定的数组中，各个数组项按LUR方式管理。在一个三个数组项用过之后，将其置于索引位置0，其他数组项相应下移。这意味这最常使用的数组项位于数组的开头，而不常用的数组项将被后退，如果很长时间不使用，则会“掉出”数组。

因为数组的长度，或者说LUR列表中的项数，是一个固定值，在内核运行期间不改变，内核无需运行独立的线程来将缓存长度修正为合理值。相反，内核只需要在一项“掉出”数组时，将相关的缓冲区从缓存删除，以释放内存，用于其他目地。

块缓存实现

块患处不仅仅用作页面缓存的附加功能，对以块而不是页面进行处理的对象来说，块缓存是一个独立的缓存。

数据结构

块缓冲区头

struct buffer_head {
	unsigned long b_state;		/* buffer state bitmap (see above) */
	struct buffer_head *b_this_page;/* circular list of page's buffers */
	struct page *b_page;		/* the page this bh is mapped to */

	sector_t b_blocknr;		/* start block number */
	size_t b_size;			/* size of mapping */
	char *b_data;			/* pointer to data within the page */

	struct block_device *b_bdev;
	bh_end_io_t *b_end_io;		/* I/O completion */
 	void *b_private;		/* reserved for b_end_io */
	struct list_head b_assoc_buffers; /* associated with another mapping */
	struct address_space *b_assoc_map;	/* mapping this buffer is
						   associated with */
	atomic_t b_count;		/* users using this buffer_head */
};

操作

内核必须提供一组操作，使得其余代码能够轻松有效地利用缓冲区的功能。切记：这些机制对内存中实际缓存的数据没有贡献。

在使用缓冲区之前，内核首先必须创建一个buffer_head结构实例，而其余的函数则对该结构进行操作。因为创建新缓冲头是一个频繁重现的任务，他应该尽快执行。这是一种很经典的情形，可使用slab缓存解决。

切记：内核源代码确实提供了一些函数，可用作前端，来创建和销毁缓冲头。alloc_buffer_head生成一个新的缓冲头，而free_buffer_head销毁一个显存的缓冲头。

/*分配buffer_head*/
struct buffer_head *alloc_buffer_head(gfp_t gfp_flags)
{
	/*从slab中分配空间*/
	struct buffer_head *ret = kmem_cache_alloc(bh_cachep, gfp_flags);
	if (ret) {
		/*初始化*/
		INIT_LIST_HEAD(&ret->b_assoc_buffers);
		get_cpu_var(bh_accounting).nr++;
		recalc_bh_state();
		put_cpu_var(bh_accounting);
	}
	return ret;
}

页缓存和块缓存的交互

一页划分为几个数据单元，但缓冲头保存在独立的内存区中，与实际数据无关。与缓冲区的交互没有改变的页的内容，缓冲区只不过为页的数据提供了一个新的视图。

为支持页与缓冲区的交互，需要使用struct page的private成员。其类型为unsigned long,可用作指向虚拟地址空间中任何位置的指针。

Private成员还可以用作其他用途，根据页的具体用途，可能与缓冲头完全无关。但其主要的用途是关联缓冲区和页。这样的话，private指向将页划分为更小单位的第一个缓冲头。各个缓冲头通过b_this_page链接为一个环形链表。在该链表中每个缓冲头的b_this_page成员指向下一个缓冲头，而最后一个缓冲头的b_this_page成员指向第一个缓冲头。这使得内核从page结构开始，可以轻易地扫描与页关联的所有buffer_head实例。

内核提供cteate_empty_buffers函数关联page和buffer_head结构之间的关联：

/*
 * We attach and possibly dirty the buffers atomically wrt
 * __set_page_dirty_buffers() via private_lock.  try_to_free_buffers
 * is already excluded via the page lock.
 */
void create_empty_buffers(struct page *page,
			unsigned long blocksize, unsigned long b_state)
{
	struct buffer_head *bh, *head, *tail;

	head = alloc_page_buffers(page, blocksize, 1);
	bh = head;
	/*遍历所有缓冲头，设置其状态，并建立一个环形链表*/
	do {
		bh->b_state |= b_state;
		tail = bh;
		bh = bh->b_this_page;
	} while (bh);
	tail->b_this_page = head;

	spin_lock(&page->mapping->private_lock);
	/*缓冲区的状态依赖于内存页面中数据的状态*/
	if (PageUptodate(page) || PageDirty(page)) {
		bh = head;
		do {/*设置相关标志*/
			if (PageDirty(page))
				set_buffer_dirty(bh);
			if (PageUptodate(page))
				set_buffer_uptodate(bh);
			bh = bh->b_this_page;
		} while (bh != head);
	}
	/*将缓冲区关联到页面*/
	attach_page_buffers(page, head);
	spin_unlock(&page->mapping->private_lock);
}

static inline void attach_page_buffers(struct page *page,
		struct buffer_head *head)
{
	page_cache_get(page);/*递增引用计数*/
	/*设置PG_private标志，通知内核其他部分，page实例的private成员正在使用中*/
	SetPagePrivate(page);
	/*将页的private成员设置为一个指向环形链表中第一个缓冲头的指针*/
	set_page_private(page, (unsigned long)head);
}

交互

如果对内核的其他部分无益，那么在页和缓冲区之间建立关联就没起作用。一些与块设备之间的传输操作，传输单位的长度依赖于底层设备的块长度，而内核的许多部分更喜欢按页的粒度来执行I/O操作，因为这使得其他事情更容易处理，特别是内存管理方面。在这种场景下，缓冲头区充当了双方的中介。

从缓冲区中读取整页

首先考察内核在从块设备读取整页时采用的方法，以block_read_full_page为例。我们讨论缓冲区实现所关注的部分。

/*
 * Generic "read page" function for block devices that have the normal
 * get_block functionality. This is most of the block device filesystems.
 * Reads the page asynchronously --- the unlock_buffer() and
 * set/clear_buffer_uptodate() functions propagate buffer state into the
 * page struct once IO has completed.
 */
int block_read_full_page(struct page *page, get_block_t *get_block)
{
	struct inode *inode = page->mapping->host;
	sector_t iblock, lblock;
	struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
	unsigned int blocksize;
	int nr, i;
	int fully_mapped = 1;

	BUG_ON(!PageLocked(page));
	blocksize = 1 << inode->i_blkbits;
	/*检查页是否有相关联的缓冲区，如果没有，则创建他*/
	if (!page_has_buffers(page))
		create_empty_buffers(page, blocksize, 0);
	/*获得这些缓冲区，无论是新建的还是已经存在的
	只是将page的private成员转换为buffer_head指针，因为按照
	惯例，private指向与page关联的第一个缓冲头*/
	head = page_buffers(page);

	iblock = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
	lblock = (i_size_read(inode)+blocksize-1) >> inode->i_blkbits;
	bh = head;
	nr = 0;
	i = 0;
	/*内核遍历与页面关联的所有缓冲区*/
	do {
		/*如果缓冲区内容是最新的，内核继续处理下一个
		缓冲区。在这种情况下，页面缓冲区中的数据与块
		设备匹配，无需额外的读操作*/
		if (buffer_uptodate(bh))
			continue;
		/*如果没有映射*/
		if (!buffer_mapped(bh)) {
			int err = 0;

			fully_mapped = 0;
			if (iblock < lblock) {
				WARN_ON(bh->b_size != blocksize);
				/*确定块在块设备上的位置*/
				err = get_block(inode, iblock, bh, 0);
				if (err)
					SetPageError(page);
			}
			if (!buffer_mapped(bh)) {
				zero_user(page, i * blocksize, blocksize);
				if (!err)
					set_buffer_uptodate(bh);
				continue;
			}
			/*
			 * get_block() might have updated the buffer
			 * synchronously
			 */
			if (buffer_uptodate(bh))
				continue;
		}
		/*如果缓冲区已经建立了与块的映射，但是其内容不是最新
		的则将缓冲区放置到一个临时的数组中*/
		arr[nr++] = bh;
	} while (i++, iblock++, (bh = bh->b_this_page) != head);

	if (fully_mapped)
		SetPageMappedToDisk(page);

	if (!nr) {
		/*
		 * All buffers are uptodate - we can set the page uptodate
		 * as well. But not if get_block() returned an error.
		 */
		if (!PageError(page))
			SetPageUptodate(page);
		unlock_page(page);
		return 0;
	}

	/* Stage two: lock the buffers */
	for (i = 0; i < nr; i++) {
		bh = arr[i];
		lock_buffer(bh);
		/*将b_end_io设置为end_buffer_async_read，该函数将在数据传输结构时
		调用*/
		mark_buffer_async_read(bh);
	}

	/*
	 * Stage 3: start the IO.  Check for uptodateness
	 * inside the buffer lock in case another process reading
	 * the underlying blockdev brought it uptodate (the sct fix).
	 */
	for (i = 0; i < nr; i++) {
		bh = arr[i];
		if (buffer_uptodate(bh))
			end_buffer_async_read(bh, 1);
		else
			/*将所有需要读取的缓冲区转交给块层
			也就是BIO层，在其中开始读操作*/
			submit_bh(READ, bh);
	}
	return 0;
}

将整页写入到缓冲区

除了读操作之外，页面的写操作也可以划分为更小的单位。只有页中实际修改的内容需要回写，而不用回写整页的内容。遗憾的是，从缓冲区的角度来看，写操作的实现比上述的读操作复杂的多。

__block_wirte_full_page函数中回写脏页面设计的缓冲区相关操作。

/*
 * NOTE! All mapped/uptodate combinations are valid:
 *
 *	Mapped	Uptodate	Meaning
 *
 *	No	No		"unknown" - must do get_block()
 *	No	Yes		"hole" - zero-filled
 *	Yes	No		"allocated" - allocated on disk, not read in
 *	Yes	Yes		"valid" - allocated and up-to-date in memory.
 *
 * "Dirty" is valid only with the last case (mapped+uptodate).
 */

/*
 * While block_write_full_page is writing back the dirty buffers under
 * the page lock, whoever dirtied the buffers may decide to clean them
 * again at any time.  We handle that by only looking at the buffer
 * state inside lock_buffer().
 *
 * If block_write_full_page() is called for regular writeback
 * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
 * locked buffer.   This only can happen if someone has written the buffer
 * directly, with submit_bh().  At the address_space level PageWriteback
 * prevents this contention from occurring.
 *
 * If block_write_full_page() is called with wbc->sync_mode ==
 * WB_SYNC_ALL, the writes are posted using WRITE_SYNC_PLUG; this
 * causes the writes to be flagged as synchronous writes, but the
 * block device queue will NOT be unplugged, since usually many pages
 * will be pushed to the out before the higher-level caller actually
 * waits for the writes to be completed.  The various wait functions,
 * such as wait_on_writeback_range() will ultimately call sync_page()
 * which will ultimately call blk_run_backing_dev(), which will end up
 * unplugging the device queue.
 */
static int __block_write_full_page(struct inode *inode, struct page *page,
			get_block_t *get_block, struct writeback_control *wbc,
			bh_end_io_t *handler)
{
	int err;
	sector_t block;
	sector_t last_block;
	struct buffer_head *bh, *head;
	const unsigned blocksize = 1 << inode->i_blkbits;
	int nr_underway = 0;
	int write_op = (wbc->sync_mode == WB_SYNC_ALL ?
			WRITE_SYNC_PLUG : WRITE);

	BUG_ON(!PageLocked(page));

	last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
	/*页面是否有关联缓冲区，如果没有创建他*/
	if (!page_has_buffers(page)) {
		create_empty_buffers(page, blocksize,
					(1 << BH_Dirty)|(1 << BH_Uptodate));
	}

	/*
	 * Be very careful.  We have no exclusion from __set_page_dirty_buffers
	 * here, and the (potentially unmapped) buffers may become dirty at
	 * any time.  If a buffer becomes dirty here after we've inspected it
	 * then we just miss that fact, and the page stays dirty.
	 *
	 * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
	 * handle that here by just cleaning them.
	 */

	block = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
	head = page_buffers(page);
	bh = head;

	/*
	 * Get all the dirty buffers mapped to disk addresses and
	 * handle any aliases from the underlying blockdev's mapping.
	 */
	 /*对所有未映射的脏缓冲区，在缓冲区和块设备
	之间建立映射*/
	do {
		if (block > last_block) {
			/*
			 * mapped buffers outside i_size will occur, because
			 * this page can be outside i_size when there is a
			 * truncate in progress.
			 */
			/*
			 * The buffer was zeroed by block_write_full_page()
			 */
			clear_buffer_dirty(bh);
			set_buffer_uptodate(bh);
		} else if ((!buffer_mapped(bh) || buffer_delay(bh)) &&
			   buffer_dirty(bh)) {
			WARN_ON(bh->b_size != blocksize);
			/*查找块设备上与缓冲区项匹配的块*/
			err = get_block(inode, block, bh, 1);
			if (err)
				goto recover;
			clear_buffer_delay(bh);
			if (buffer_new(bh)) {
				/* blockdev mappings never come here */
				clear_buffer_new(bh);
				unmap_underlying_metadata(bh->b_bdev,
							bh->b_blocknr);
			}
		}
		bh = bh->b_this_page;
		block++;
	} while (bh != head);
	/*第二遍遍历，将滤出所有的脏缓冲区*/
	do {
		if (!buffer_mapped(bh))
			continue;
		/*
		 * If it's a fully non-blocking write attempt and we cannot
		 * lock the buffer then redirty the page.  Note that this can
		 * potentially cause a busy-wait loop from writeback threads
		 * and kswapd activity, but those code paths have their own
		 * higher-level throttling.
		 */
		if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
			lock_buffer(bh);
		} else if (!trylock_buffer(bh)) {
			redirty_page_for_writepage(wbc, page);
			continue;
		}
		/*如果设置了脏页标志，则会在调用该函数时清除
		因为缓冲区的内容将立即回写*/
		if (test_clear_buffer_dirty(bh)) {
			/*设置BH_Async_Write状态位，并将end_buffer_async_write
			指定为BIO完成处理程序即b_end_io*/
			mark_buffer_async_write_endio(bh, handler);
		} else {
			unlock_buffer(bh);
		}
	} while ((bh = bh->b_this_page) != head);

	/*
	 * The page and its buffers are protected by PageWriteback(), so we can
	 * drop the bh refcounts early.
	 */
	BUG_ON(PageWriteback(page));
	set_page_writeback(page);
	/*最后一次遍历*/
	do {
		struct buffer_head *next = bh->b_this_page;
		if (buffer_async_write(bh)) {
			/*将前一次遍历中标记为BH_Async_Write的所有缓冲区
			转交给块层执行实际的写操作，该函数向块层提交
			了对应的请求*/
			submit_bh(write_op, bh);
			nr_underway++;
		}
		bh = next;
	} while (bh != head);
	unlock_page(page);

	err = 0;
done:
	if (nr_underway == 0) {
		/*
		 * The page was marked dirty, but the buffers were
		 * clean.  Someone wrote them back by hand with
		 * ll_rw_block/submit_bh.  A rare case.
		 */
		end_page_writeback(page);

		/*
		 * The page and buffer_heads can be released at any time from
		 * here on.
		 */
	}
	return err;

recover:
	/*
	 * ENOSPC, or some other error.  We may already have added some
	 * blocks to the file, so we need to write these out to avoid
	 * exposing stale data.
	 * The page is currently locked and not marked for writeback
	 */
	bh = head;
	/* Recovery: lock and submit the mapped buffers */
	do {
		if (buffer_mapped(bh) && buffer_dirty(bh) &&
		    !buffer_delay(bh)) {
			lock_buffer(bh);
			mark_buffer_async_write_endio(bh, handler);
		} else {
			/*
			 * The buffer may have been set dirty during
			 * attachment to a dirty page.
			 */
			clear_buffer_dirty(bh);
		}
	} while ((bh = bh->b_this_page) != head);
	SetPageError(page);
	BUG_ON(PageWriteback(page));
	mapping_set_error(page->mapping, err);
	set_page_writeback(page);
	do {
		struct buffer_head *next = bh->b_this_page;
		if (buffer_async_write(bh)) {
			clear_buffer_dirty(bh);
			submit_bh(write_op, bh);
			nr_underway++;
		}
		bh = next;
	} while (bh != head);
	unlock_page(page);
	goto done;
}