Re: [PATCH 1/1] fs/mpage.c: forgotten WRITE_SYNC in case of data integrity write

[Tejun Heo <tj@kernel.org>]

Hello, Andrew.

On Thu, Mar 13, 2014 at 02:34:56PM -0700, Andrew Morton wrote:
> Jens isn't talking to us.  Tejun, are you able explain REQ_SYNC?

It has nothing to do with data integrity.  It's just a hint telling
the block layer that someone is waiting for the IO so it'd be a good
idea to prioritize it.  For example, nothing visible to userland
really waits for periodic writebacks, so we can delay their processing
to prioritize, for example, READs triggered from a page fault, which
is obviously causing userland visible latency.

Block layer treats all READs as REQ_SYNC and also allows upper layers
to mark some writes REQ_SYNC for cases where somebody is waiting for
the write to complete for cases like flush(2).

> From: Roman Pen <r.peniaev@gmail.com>
> Subject: fs/mpage.c: forgotten WRITE_SYNC in case of data integrity write
> 
> In case of wbc->sync_mode == WB_SYNC_ALL we need to do data integrity
> write, thus mark request as WRITE_SYNC.

So, at least this patch description is very misleading.  WRITE_SYNC
has *NOTHING* to do with data integrity.  The only thing matters is
whether somebody is waiting for its completion or not.

Thanks.

-- 
tejun