On 17/03/2021 14:20, Johannes Thumshirn wrote:
> On 17/03/2021 11:54, David Sterba wrote:
>> On Wed, Mar 17, 2021 at 05:57:31PM +0900, Johannes Thumshirn wrote:
>>> In btrfs_submit_direct() there's a WAN_ON_ONCE() that will trigger if
>>> we're submitting a DIO write on a zoned filesystem but are not using
>>> REQ_OP_ZONE_APPEND to submit the IO to the block device.
>>>
>>> This is a left over from a previous version where btrfs_dio_iomap_begin()
>>> didn't use btrfs_use_zone_append() to check for sequential write only
>>> zones.
>>
>> I can't identify the patch where this got changed. I've landed on
>> 544d24f9de73 ("btrfs: zoned: enable zone append writing for direct IO")
>> but that adds the btrfs_use_zone_append, the append flag and also the
>> warning.
>>
>
> It is an oversight from the development phase. In v11 (I think) I've added
> 08f455593fff ("btrfs: zoned: cache if block group is on a sequential zone")
> and forgot to remove the WARN_ON_ONCE() for 544d24f9de73 ("btrfs: zoned:
> enable zone append writing for direct IO").
>
> When developing auto relocation I got hit by the WARN as a block groups
> where relocated to conventional zone and the dio code calls
> btrfs_use_zone_append() introduced by 08f455593fff to check if it can use
> zone append (a.k.a. if it's a sequential zone) or not and sets the appropriate
> flags for iomap.
>
> I've never hit it in testing before, as I was relying on emulation to test
> the conventional zones code but this one case wasn't hit, because on emulation
> fs_info->max_zone_append_size is 0 and the WARN doesn't trigger either.
>
I just realized that explanation should have gone into the commit message, do
you
want me to resend with a more elaborate commit message?
