Is there something wrong with my question?  I asked it 1.5 months ago...

Meanwhile, we're using this patch on our database server since
Aug-2011, and it appears to work correctly - direct and buffered
I/O works together without surprizes.  Without this patch, I see
unexpected results.

Thanks,

/mjt

On 01.12.2011 00:38, Michael Tokarev wrote:

Hello.

Back in August 2011, a commit has been tagged to be included
into stable, this one:

commit dccaf33fa37a1bc5d651baeb3bfeb6becb86597b
Author: Jiaying Zhang [off-list ref]
Date:   Fri Aug 19 19:13:32 2011 -0400

    ext4: flush any pending end_io requests before DIO reads w/dioread_nolock

    There is a race between ext4 buffer write and direct_IO read with
    dioread_nolock mount option enabled. The problem is that we clear
    PageWriteback flag during end_io time but will do
    uninitialized-to-initialized extent conversion later with dioread_nolock.
    If an O_direct read request comes in during this period, ext4 will return
    zero instead of the recently written data.

    This patch checks whether there are any pending uninitialized-to-initialized
    extent conversion requests before doing O_direct read to close the race.
    Note that this is just a bandaid fix. The fundamental issue is that we
    clear PageWriteback flag before we really complete an IO, which is
    problem-prone. To fix the fundamental issue, we may need to implement an
    extent tree cache that we can use to look up pending to-be-converted extents.

    Signed-off-by: Jiaying Zhang [off-list ref]
    Signed-off-by: "Theodore Ts'o" [off-list ref]
    Cc: stable@kernel.org


There was one more ext4 commit at that time, which made its way into
stable but this one did not.

I wonder if the reason for that was the fact that it needed a small
"backport" for 3.0, since in 3.1+ the code has been moved into another
file, and the context is slightly different.  In that case, attached
is the "backport" which we use with 3.0.x since that time.

Thanks!

/mjt