Re: [PATCH v4 2/3] locks: Threads with layout conflict must wait until client was fenced.

[Christoph Hellwig <hch@lst.de>]

On Mon, Nov 17, 2025 at 01:21:32PM -0500, Jeff Layton wrote:
> > +	if (type == FL_LAYOUT && !ctx->flc_conflict) {
> > +		ctx->flc_conflict = true;
> > +		ctx->flc_wait_for_dispose = false;
> > +	}
> 
> I don't like special casing this for FL_LAYOUT leases. It seems like we
> ought to be able to set up a lm_breaker_timedout operation on any sort
> of lease.

Yes, this should check for a lm_breaker_timedout operation instead.

> > +
> > +	/* for FL_LAYOUT */
> > +	bool			flc_conflict;
> > +	bool			flc_wait_for_dispose;
> 
> I'm also not a fan of this particular bool. Waiting for any
> lm_breaker_timeout operations to complete seems like something we ought
> to just always do.

Yes.

> In the trivial case where we have no special fencing
> to do, that should just return quickly anyway.

Exactly.