Re: [PATCH] btrfs: always wait for ordered extents to avoid OE races

在 2026/6/26 18:45, Filipe Manana 写道:
> On Thu, Jun 25, 2026 at 11:39 PM Qu Wenruo <[email protected]> wrote:
[...]
>> If acceptable, I'll add the microbenchmark into the commit message in
>> the next update.
> 
> Instead of this approach, can we wait for the ordered extent(s) in DIO
> path that reverts i_size after the short write?

OK, I think that's possible.
I didn't see any special context requirement for iomap_end() callback.

But doing btrfs_start_ordered_extent() at the isize revert timing has 
one extra problem, it will break nowait.

As btrfs_start_ordered_extent() will call wait_event().

At least the current one we won't directly sleep (the wait is happening 
on the next buffered write).

Thanks,
Qu