读写文件（3）

2、读入操作。完成了上面的准备工作，下一步就是执行读操作的核心函数do_generic_mapping_read，这是一个比较复杂的函数，里面有大量的goto跳转，但还是比较清晰的。

       它工作过程可以描述如下：

a.       如果所要读取的文件在页面缓存中，则跳转到步骤d。

b.       文件还没有被缓冲，所以要从设备中去读取，首先分配一个页面，并将这个页面链入到相应的address_space中去

c.       然后调用address_space中的readpage()函数，去从设备中读出一个页面大小的数据到这个页面缓存中。

d.       检查PageUptodate(page)

e.       调用由参数传入的actor函数指针，在此为file_read_actor()，将数据中页面缓存中拷贝到用户缓冲区。

f.        如果请求读取的数据长度已完成，则函数返回，否则跳转到步骤a重复执行。

先看看file_read_actor()：

int file_read_actor(read_descriptor_t *desc, struct page *page,

                     unsigned long offset, unsigned long size)

{

       char *kaddr;

       unsigned long left, count = desc->count;

       if (size > count)

              size = count;

……

       /* Do it the slow way */

       kaddr = kmap(page);

       left = __copy_to_user(desc->arg.buf, kaddr + offset, size); //将数据拷贝到用户空间

       kunmap(page);

       if (left) {

              size -= left;

              desc->error = -EFAULT;

       }

success:

       desc->count = count - size;

       desc->written += size;

       desc->arg.buf += size;

       return size;

}

/**

* This is a generic file read routine, and uses the

* mapping->a_ops->readpage() function for the actual low-level stuff.

*/

void do_generic_mapping_read(struct address_space *mapping,

                          struct file_ra_state *_ra,

                          struct file *filp,

                          loff_t *ppos,

                          read_descriptor_t *desc,

                          read_actor_t actor)

{

       struct inode *inode = mapping->host;

       unsigned long index;

       unsigned long end_index;

       unsigned long offset;

       unsigned long last_index;

       unsigned long next_index;

       unsigned long prev_index;

       loff_t isize;

       struct page *cached_page;

       int error;

       struct file_ra_state ra = *_ra;

       cached_page = NULL;

       index = *ppos >> PAGE_CACHE_SHIFT;

       next_index = index;

       prev_index = ra.prev_page;

       last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;

       offset = *ppos & ~PAGE_CACHE_MASK;

       isize = i_size_read(inode);

       if (!isize)

              goto out;

       end_index = (isize - 1) >> PAGE_CACHE_SHIFT;

       for (;;) {

              struct page *page;

              unsigned long nr, ret;

              /* nr is the maximum number of bytes to copy from this page */

              nr = PAGE_CACHE_SIZE;

              if (index >= end_index) {

                     if (index > end_index)

                            goto out;

                     nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;

                     if (nr <= offset) {

                            goto out;

                     }

              }

              nr = nr - offset;

              cond_resched();

              if (index == next_index)

                     next_index = page_cache_readahead(mapping, &ra, filp,

                                   index, last_index - index);

find_page:

              page = find_get_page(mapping, index); //在缓存中查找

              if (unlikely(page == NULL)) {

                     handle_ra_miss(mapping, &ra, index);

                     goto no_cached_page; //没有找到

              }

              if (!PageUptodate(page)) //Uptodate

                     goto page_not_up_to_date;

page_ok: //找到了相关缓存页面

              ret = actor(desc, page, offset, nr); //拷贝数据到用户缓冲区

              //更新一些变量值

              offset += ret;

              index += offset >> PAGE_CACHE_SHIFT;

              offset &= ~PAGE_CACHE_MASK;

              page_cache_release(page);

              if (ret == nr && desc->count)

                     continue; //未完成，进入下一次循环

              goto out; //完成

page_not_up_to_date:

              /* Get exclusive access to the page ... */

              lock_page(page);

              /* Did it get truncated before we got the lock? */

              if (!page->mapping) {

                     unlock_page(page);

                     page_cache_release(page);

                     continue;

              }

              /* Did somebody else fill it already? */

              if (PageUptodate(page)) {

                     unlock_page(page);

                     goto page_ok;

              }

readpage: //读操作

              /* Start the actual read. The read will unlock the page. */

              error = mapping->a_ops->readpage(filp, page); //真正的读操作

              ……             

              /* nr is the maximum number of bytes to copy from this page */

              nr = PAGE_CACHE_SIZE;

              if (index == end_index) {

                     nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;

                     if (nr <= offset) {

                            page_cache_release(page);

                            goto out;

                     }

              }

              nr = nr - offset;

              goto page_ok;

readpage_error:

              /* UHHUH! A synchronous read error occurred. Report it */

              desc->error = error;

              page_cache_release(page);

              goto out;

no_cached_page: //分配一个新的页面，比将它链入缓存树中。

              /*

              * Ok, it wasn't cached, so we need to create a new

              * page..

              */

              if (!cached_page) {

                     cached_page = page_cache_alloc_cold(mapping);

                     if (!cached_page) {

                            desc->error = -ENOMEM;

                            goto out;

                     }

              }

              error = add_to_page_cache_lru(cached_page, mapping,

                                          index, GFP_KERNEL);

              page = cached_page;

              cached_page = NULL;

              goto readpage;

       }

out:

       *_ra = ra;

       *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;

       if (cached_page)

              page_cache_release(cached_page);

       if (filp)

              file_accessed(filp);

}

3、从设备读取

对于不同的文件系统有不同的address_space，而且有不同的address_space_operations，对于ext2文件系统来说，这个是如下一个结构：

const struct address_space_operations ext2_aops = {

       .readpage               = ext2_readpage,

       .readpages             = ext2_readpages,

       .writepage             = ext2_writepage,

       .sync_page            = block_sync_page,

       .prepare_write        = ext2_prepare_write,

       .commit_write              = generic_commit_write,

       .bmap                   = ext2_bmap,

       .direct_IO              = ext2_direct_IO,

       .writepages            = ext2_writepages,

       .migratepage          = buffer_migrate_page,

};

可见，这个readpage()便是ext2_readpage()，它负责从设备中读取一个页面。

static int ext2_readpage(struct file *file, struct page *page)

{

       return mpage_readpage(page, ext2_get_block);

}

/*

* This isn't called much at all

*/

int mpage_readpage(struct page *page, get_block_t get_block)

{

       struct bio *bio = NULL;

       sector_t last_block_in_bio = 0;

       struct buffer_head map_bh;

       unsigned long first_logical_block = 0;

       clear_buffer_mapped(&map_bh);

       bio = do_mpage_readpage(bio, page, 1, &last_block_in_bio,

                     &map_bh, &first_logical_block, get_block);

       if (bio)

              mpage_bio_submit(READ, bio);

       return 0;

}

这个函数最终将读请求转成submit_bio()，之后就是通用块层的事情了。

原文链接：http://hi.baidu.com/rwen2012/blog/item/302997a2c1dfd4accbefd0a8.html