Re: [PATCH] nvme: zns: limit max_zone_append by max_segments

[Damien Le Moal <dlemoal@kernel.org>]

On 7/31/23 23:06, Christoph Hellwig wrote:
> On Mon, Jul 31, 2023 at 11:01:21PM +0900, Damien Le Moal wrote:
>> We cannot do that in zonefs because there is no metadata to handle a possible
>> reordering of the fragments of a split large zone append. Hence zonefs limits
>> writes size to max zone append on entry and never tries to do larger writes. But
>> the ZNS limit bug resulted in the split. At least for zonefs, I think there is
>> no need to use the special bio add page since with a proper limit, we should
>> never see a split.
> 
> Then we need to bring back something like the code removed in
> 8e81aa16a42169faae1ba15cd648cc8bb83eaa48, although I'd kinda hate that,
> and I don't really say how that would protect against reordering either.

For zonefs, their is never any reordering because their is no split, as long as
max zone append sectors limit indicates the maximum size of a zone append BIO
that can be built without ever needing splitting.

So if we fix the zns limit, zonefs does not really need a fix, eventhough the
code would be a little weird as-is.

-- 
Damien Le Moal
Western Digital Research