KVM虚拟机IO处理过程(一) ----Guest VM I/O 处理过程

    虚拟化技术主要包含三部分内容:CPU虚拟化,内存虚拟化,设备虚拟化.本系列文章主要描述磁盘设备的虚拟化过程,包含了一个读操作的I/O请求如何从Guest Vm到其最终被处理的整个过程.本系列文章中引用到的linux内核代码版本为3.7.10,使用的虚拟化平台是KVM,qemu的版本是1.6.1.

    用户程序想要访问IO设备需要调用操作系统提供的接口,即系统调用.当在用户程序中调用一个read操作时,系统先保存好read操作的参数,然后调用int 80命令(也可能是sysenter)进入内核空间,在内核空间中,读操作的逻辑由sys_read函数实现.

    在讲sys_read的实现过程之前,我们先来看看read操作在内核空间需要经历的层次结构.从图中可以看出,read操作首先经过虚拟文件系统曾(vfs), 接下来是具体的文件系统层,Page cache层,通用块层(generic block layer),I/O调度层(I/O scheduler layer),块设备驱动层(block device driver layer),最后是块物理设备层(block device layer).

• 虚拟文件系统层:该层屏蔽了下层的具体操作,为上层提供统一的接口,如vfs_read,vfs_write等.vfs_read,vfs_write通过调用下层具体文件系统的接口来实现相应的功能.
• 具体文件系统层:该层针对每一类文件系统都有相应的操作和实现了,包含了具体文件系统的处理逻辑.
• page cache层:该层缓存了从块设备中获取的数据.引入该层的目的是避免频繁的块设备访问,如果在page cache中已经缓存了I/O请求的数据,则可以将数据直接返回,无需访问块设备.
• 通过块层:接收上层的I/O请求,并最终发出I/O请求.该层向上层屏蔽了下层设备的特性.
• I/O调度层:   接收通用块层发出的 IO 请求，缓存请求并试图合并相邻的请求（如果这两个请求的数据在磁盘上是相邻的）。并根据设置好的调度算法，回调驱动层提供的请求处理函数，以处理具体的 IO 请求
• 块设备驱动层:从上层取出请求,并根据参数,操作具体的设备.
• 块设备层:真正的物理设备.

    了解了内核层次的结构,让我们来看一下read操作的代码实现.

     sys_read函数声明在include/linux/syscalls.h文件中,

asmlinkage long sys_read(unsigned int fd, char __user *buf, size_t count);

     

      其函数实现在fs/read_write.c文件中:

SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
{
	struct fd f = fdget(fd);
	ssize_t ret = -EBADF;

	if (f.file) {
		loff_t pos = file_pos_read(f.file);
		ret = vfs_read(f.file, buf, count, &pos); //调用vfs layer中的read操作
		file_pos_write(f.file, pos);//设置当前文件的位置
		fdput(f);
	}
	return ret;
}

    vfs_read函数属于vfs layer,定义在fs/read_write.c, 其主要功能是调用具体文件系统中对应的read操作,如果具体文件系统没有提供read操作,则使用默认的do_sync_read函数.

ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
{
	ssize_t ret;

	if (!(file->f_mode & FMODE_READ))
		return -EBADF;
	if (!file->f_op || (!file->f_op->read && !file->f_op->aio_read))
		return -EINVAL;
	if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
		return -EFAULT;

	ret = rw_verify_area(READ, file, pos, count);
	if (ret >= 0) {
		count = ret;
		if (file->f_op->read) {
			ret = file->f_op->read(file, buf, count, pos); //该函数由具体的文件系统指定
		} else
			ret = do_sync_read(file, buf, count, pos);  //内核默认的读文件操作
		if (ret > 0) {
			fsnotify_access(file);
			add_rchar(current, ret);
		}
		inc_syscr(current);
	}

	return ret;
}

    file->f_op的类型为struct file_operations, 该类型定义了一系列涉及文件操作的函数指针,针对不同的文件系统,这些函数指针指向不同的实现.以ext4 文件系统为例子,该数据结构的初始化在fs/ext4/file.c,从该初始化可以知道,ext4的read操作调用了内核自带的do_sync_read()函数

const struct file_operations ext4_file_operations = {
	.llseek		= ext4_llseek,
	.read		= do_sync_read,
	.write		= do_sync_write,
	.aio_read	= generic_file_aio_read,
	.aio_write	= ext4_file_write,
	.unlocked_ioctl = ext4_ioctl,
#ifdef CONFIG_COMPAT
	.compat_ioctl	= ext4_compat_ioctl,
#endif
	.mmap		= ext4_file_mmap,
	.open		= ext4_file_open,
	.release	= ext4_release_file,
	.fsync		= ext4_sync_file,
	.splice_read	= generic_file_splice_read,
	.splice_write	= generic_file_splice_write,
	.fallocate	= ext4_fallocate,
};

    do_sync_read()函数定义fs/read_write.c中,

ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
{
	struct iovec iov = { .iov_base = buf, .iov_len = len };
	struct kiocb kiocb;
	ssize_t ret;

	init_sync_kiocb(&kiocb, filp);//初始化kiocp,描述符kiocb是用来记录I/O操作的完成状态
	kiocb.ki_pos = *ppos;
	kiocb.ki_left = len;
	kiocb.ki_nbytes = len;

	for (;;) {
		ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);//调用真正做读操作的函数,ext4文件系统在fs/ext4/file.c中配置
		if (ret != -EIOCBRETRY)
			break;
		wait_on_retry_sync_kiocb(&kiocb);
	}

	if (-EIOCBQUEUED == ret)
		ret = wait_on_sync_kiocb(&kiocb);
	*ppos = kiocb.ki_pos;
	return ret;
}

    在ext4文件系统中filp->f_op->aio_read函数指针只想generic_file_aio_read, 该函数定义于mm/filemap.c文件中,该函数有两个执行路径,如果是以O_DIRECT方式打开文件,则读操作跳过page cache直接去读取磁盘,否则调用do_generic_sync_read函数尝试从page cache中获取所需的数据.

ssize_t
generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
		unsigned long nr_segs, loff_t pos)
{
	struct file *filp = iocb->ki_filp;
	ssize_t retval;
	unsigned long seg = 0;
	size_t count;
	loff_t *ppos = &iocb->ki_pos;

	count = 0;
	retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
	if (retval)
		return retval;

	/* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
	if (filp->f_flags & O_DIRECT) {
		loff_t size;
		struct address_space *mapping;
		struct inode *inode;

		struct timex txc;
		do_gettimeofday(&(txc.time));

		mapping = filp->f_mapping;
		inode = mapping->host;
		if (!count)
			goto out; /* skip atime */
		size = i_size_read(inode);
		if (pos < size) {
			retval = filemap_write_and_wait_range(mapping, pos,
					pos + iov_length(iov, nr_segs) - 1);
			if (!retval) {
				retval = mapping->a_ops->direct_IO(READ, iocb,
							iov, pos, nr_segs);
			}
			if (retval > 0) {
				*ppos = pos + retval;
				count -= retval;
			}

			/*
			 * Btrfs can have a short DIO read if we encounter
			 * compressed extents, so if there was an error, or if
			 * we've already read everything we wanted to, or if
			 * there was a short read because we hit EOF, go ahead
			 * and return.  Otherwise fallthrough to buffered io for
			 * the rest of the read.
			 */
			if (retval < 0 || !count || *ppos >= size) {
				file_accessed(filp);
				goto out;
			}
		}
	}

	count = retval;
	for (seg = 0; seg < nr_segs; seg++) {
		read_descriptor_t desc;
		loff_t offset = 0;

		/*
		 * If we did a short DIO read we need to skip the section of the
		 * iov that we've already read data into.
		 */
		if (count) {
			if (count > iov[seg].iov_len) {
				count -= iov[seg].iov_len;
				continue;
			}
			offset = count;
			count = 0;
		}

		desc.written = 0;
		desc.arg.buf = iov[seg].iov_base + offset;
		desc.count = iov[seg].iov_len - offset;
		if (desc.count == 0)
			continue;
		desc.error = 0;
		do_generic_file_read(filp, ppos, &desc, file_read_actor);
		retval += desc.written;
		if (desc.error) {
			retval = retval ?: desc.error;
			break;
		}
		if (desc.count > 0)
			break;
	}
out:
	return retval;
}

    do_generic_file_read定义在mm/filemap.c文件中,该函数调用page cache层中相关的函数.如果所需数据存在与page cache中,并且数据不是dirty的,则从page cache中直接获取数据返回.如果数据在page cache中不存在,或者数据是dirty的,则page cache会引发读磁盘的操作.该函数的读磁盘并不是简单的只读取所需数据的所在的block,而是会有一定的预读机制来提高cache的命中率,减少磁盘访问的次数. 

    page cache层中真正读磁盘的操作为readpage系列,readpage系列函数具体指向的函数实现在fs/ext4/inode.c文件中定义,该文件中有很多个struct address_space_operation对象来对应与不同日志机制,我们选择linux默认的ordered模式的日志机制来描述I/O的整个流程, ordered模式对应的readpage系列函数如下所示.

static const struct address_space_operations ext4_ordered_aops = {
	.readpage		= ext4_readpage,
	.readpages		= ext4_readpages,
	.writepage		= ext4_writepage,
	.write_begin		= ext4_write_begin,
	.write_end		= ext4_ordered_write_end,
	.bmap			= ext4_bmap,
	.invalidatepage		= ext4_invalidatepage,
	.releasepage		= ext4_releasepage,
	.direct_IO		= ext4_direct_IO,
	.migratepage		= buffer_migrate_page,
	.is_partially_uptodate  = block_is_partially_uptodate,
	.error_remove_page	= generic_error_remove_page,
};

    为简化流程,我们选取最简单的ext4_readpage函数来说明,该函数实现位于fs/ext4/inode.c中,函数很简单,只是调用了mpage_readpage函数.mpage_readpage位于fs/mpage.c文件中,该函数生成一个IO请求,并提交给Generic block layer.

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;

	map_bh.b_state = 0;
	map_bh.b_size = 0;
	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;
}

    Generic block layer会将该请求分发到具体设备的IO队列中,由I/O Scheduler去调用具体的driver接口获取所需的数据.

    至此,在Guest vm中整个I/O的流程已经介绍完了,后续的文章会介绍I/O操作如何从Guest vm跳转到kvm及如何在qemu中模拟I/O设备.

参考资料:

1. read系统调用剖析: http://www.ibm.com/developerworks/cn/linux/l-cn-read/