Re: [PATCH v4 3/3] FSD: Fix NFS server hang when there are multiple layout conflicts

[Christoph Hellwig <hch@lst.de>]

On Wed, Nov 19, 2025 at 08:52:25AM -0800, Dai Ngo wrote:
>
> On 11/19/25 2:08 AM, Christoph Hellwig wrote:
>> On Mon, Nov 17, 2025 at 02:00:07PM -0800, Dai Ngo wrote:
>>> Perhaps I overstated the severity of the risk. The real issue is, in the
>>> current state, SCSI layout recall has no timeout and if there are enough
>>> activities on the server that results in lots of layout conflicts then the
>>> server can hang.
>> All this is really caused by the synchronous waiting.  I'm not against
>> the workaround here, but I think we need to address that.  There's
>> really no reason to consumer threads for this waiting activity and
>> we'll need to stop doing it.
>
> Yes, I think __break_lease needs to work asynchronously. But I'm concerned
> about touching a bunch of other callers.

You really don't need to touch other callers, keep the existing semantics
for them.  Just like we don't force all file I/O to be asynchronous,
just because we also offer that.