[Ocfs2-tools-devel] Re: [PATCH 1/2] The io_cache implementation

[Joel Becker]

On Tue, Mar 27, 2007 at 02:48:45PM +0800, tao.ma wrote:
> >+		ret = io_cache_read_one_block(channel, blkno, data);
> >  
> Should data change accordingly? data + i * channel->io_blksize?
> The same error happens in io_cache_write_block.

	You know, I could swear I did that.  Let me peruse, because I
think you are right.

> >+	memcpy(icb->icb_buf, data, channel->io_blksize);
> >+	io_cache_seen(ic, icb);
> >+
> >+	ret = unix_io_write_block(channel, blkno, 1, icb->icb_buf);
> >+	return ret;
> >+}
> >  
> I think the "memcpy" should be moved after the success of 
> "unix_io_write_block" so that we can keep consistent of the buffer and 
> the real content on the disk.

	Hmm, while I liked the idea of "writing out of the cache", you
make a valid point that we want the cache to reflect the disk.  On the
other hand, we have no idea what the disk looks like when a write fails
- it could be an EIO that actually modified the platter.
	Perhaps just remove the block from the cache on failure to force
a real re-read from disk?

Joel

-- 

Life's Little Instruction Book #198

	"Feed a stranger's expired parking meter."

Joel Becker
Principal Software Developer
Oracle
E-mail: joel.becker at oracle.com
Phone: (650) 506-8127