Re: [PATCH v2 2/2] dax,pmem: Implement pmem based dax data recovery

[Jane Chu]

On 11/5/2021 7:04 PM, Darrick J. Wong wrote:
<snip>
>>
>> diff --git a/drivers/md/dm.c b/drivers/md/dm.c
>> index dc354db22ef9..9b3dac916f22 100644
>> --- a/drivers/md/dm.c
>> +++ b/drivers/md/dm.c
>> @@ -1043,6 +1043,7 @@ static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
>>   	if (!ti)
>>   		goto out;
>>   	if (!ti->type->dax_copy_from_iter) {
>> +		WARN_ON(mode == DAX_OP_RECOVERY);
>>   		ret = copy_from_iter(addr, bytes, i);
>>   		goto out;
>>   	}
>> @@ -1067,6 +1068,7 @@ static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
>>   	if (!ti)
>>   		goto out;
>>   	if (!ti->type->dax_copy_to_iter) {
>> +		WARN_ON(mode == DAX_OP_RECOVERY);
> 
> Maybe just return -EOPNOTSUPP here?
> 
> Warnings are kinda loud.
> 

Indeed.  Looks like the
   "if (!ti->type->dax_copy_to_iter) {"
clause was to allow mixed dax targets in dm, such as dcss, fuse and
virtio_fs targets. These targets either don't export
.dax_copy_from/to_iter, or don't need to.
And their .dax_direct_access don't check poison, and can't repair
poison anyway.

I think these targets may safely ignore the flag.  However, returning
-EOPNOTSUPP is helpful to catch future bug, such as someone add a
method to detect poison, but didn't add a method to clear poison, in
that case, we fail the call.

Dan, do you have a preference?

thanks!
-jane