Re: [PATCH 1/1] NFSD: Enforce recall timeout for layout conflict

[Christoph Hellwig <hch@lst.de>]

On Tue, Jan 20, 2026 at 10:54:39AM -0800, Dai Ngo wrote:
> Thank you Christoph! I have a couple of questions regarding to
> xfs_break_layouts and xfs_break_leased_layouts.
>
> . Should we break out of the while loop in xfs_break_leased_layouts
> if the 2nd call to break_layout, inside the while loop, returns error
> other than -EWOULDBLOCK?

Good question.

> . In xfs_break_leased_layouts, the return value of the 2nd call to
> break_layout was rightly ignored since the call was made without
> holding the xfs_inode lock so there could be a race condition where
> a new layout was handled out to another client.

I have to admin that I'm not sure what other errors we could
have.  Looking through the code I see:

 o -EINVAL for incorrect flags.
 o the error from lease_alloc, which could be -ENOMEM, -EINVAL
   again for a wrong type
 o -EINTR or similar from wait_event_interruptible_timeout

The -EINVAL cases can't happen, for code hygiene they probably should be
handled.  -EINTR means the caller gave up, so it should be handled.
-ENOMEM for the tiny structure is basically impossible to hit, but there
is no point in not giving up, so it should be handled as well.

So yeah, I think we should break out of the loop on error.