[PATCH] nvme: Fix zns drives without append support to export correct permissions

[Pankaj Raghav]

On 2022-03-14 09:44, Christoph Hellwig wrote:
> On Mon, Mar 14, 2022 at 09:36:35AM +0100, Pankaj Raghav wrote:
>> diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
>> index 51c08f206cbf..cde33f2a3a5a 100644
>> --- a/drivers/nvme/host/core.c
>> +++ b/drivers/nvme/host/core.c
>> @@ -1855,8 +1855,7 @@ static void nvme_update_disk_info(struct gendisk
>> *disk,
>>  	blk_queue_max_write_zeroes_sectors(disk->queue,
>>  					   ns->ctrl->max_zeroes_sectors);
>>
>> -	set_disk_ro(disk, (id->nsattr & NVME_NS_ATTR_RO) ||
>> -		test_bit(NVME_NS_FORCE_RO, &ns->flags));
>> +	set_disk_ro(disk, (id->nsattr & NVME_NS_ATTR_RO));
>>  }
> 
> This creates a race window during revalidation where the zone will be
> set writable here only to be set read-only again a little later.

We call the blk_revalidate_zone after nvme_update_zone_info. So before
we revalidate, the correct permission is already set.

	// sets the disk to ro based on nsattr
	nvme_update_disk_info(ns->disk, ns, id);

	if (ns->head->ids.csi == NVME_CSI_ZNS) {
		// Sets the disk to ro based on append
		ret = nvme_update_zone_info(ns, lbaf);
		if (ret)
			goto out_unfreeze;
	}

	set_bit(NVME_NS_READY, &ns->flags);
	blk_mq_unfreeze_queue(ns->disk->queue);

	if (blk_queue_is_zoned(ns->queue)) {
		// Revalidate is called after the permission is set
		ret = nvme_revalidate_zones(ns);
		if (ret && !nvme_first_scan(ns->disk))
			return ret;
	}

Am I missing something?

-- 
Regards,
Pankaj