[PATCH] nvme: Remove namespace when nvme_identify_ns_descs() failed

[Hannes Reinecke]

On 1/15/25 09:18, Nilay Shroff wrote:
> 
> 
> On 1/15/25 1:32 PM, Hannes Reinecke wrote:
>> On 1/15/25 08:48, Nilay Shroff wrote:
>>>
>>>
>>> On 1/13/25 7:59 PM, Hannes Reinecke wrote:
>>>> On 1/13/25 15:12, Nilay Shroff wrote:
>>>>>
>>>>>
>>>>> On 1/13/25 1:13 PM, Hannes Reinecke wrote:
>>>>>> On 1/11/25 15:01, Nilay Shroff wrote:
>>>>>>>
>>>>>>>
>>>> [ .. ]
>>>>>> So my argument is that in this specific case the 'ANA inaccessible' nvme
>>>>>> state should _not_ be retried, but should be treated as identical to
>>>>>> 'invalid namespace' errors.
>>>>>>
>>>>> I think I got what you're trying to propose. So when this issue manifests, on host, if we
>>>>> could possibly differentiate between nvme_identify_ns_descs() failed reasons : is it failed
>>>>> because the nsid has been removed/un-mapped on the target or is it failed due to "ANA inaccessible"
>>>>> state? IMO, for "ANA inaccessible" status, we may not want to immediately remove the ns from
>>>>> the host (due to reason I mentioned earlier per NVMe spec section 8.1.3.3), however for the
>>>>> other error case we may remove the ns from the host.
>>>>> I think issuing ns descriptor list command to target for a nsid which doesn't exist on the
>>>>> target would return buffer filled with all zeros. So that might be an indication that ns has
>>>>> been removed from the target.
>>>>>     
>>>> But only if the NSID has not been remapped in the meantime.
>>>> If it has (as in my case) the ns descriptor list will be valid, it just
>>>> refers to another namespace.
>>>>
>>> If NSID has been unmapped and then remapped on the targer then in that case,
>>> host would hit the mismatch uuid case (under nvme_validate_ns()) and so host
>>> would then remove the namespace.
>>>
>>> I think there are two cases,
>>> Case1:
>>> 1. AEN triggers rescan
>>> 2. List of active nsid is retrieved
>>> -> NSID A is removed on the target
>>> 3. Scanning of NSID A fails (i.e. nvme_identify_ns_descs() returns buffer filled with all zeros)
>>> -> host removes the respective namespace
>>>
>>> Case2:
>>> 1. AEN triggers rescan
>>> 2. List of active nsid is retrieved
>>> -> NSID A is unmapped and remapped (possibly with different uuid) on target
>>> 3. Scanning of NSID A succeed
>>> 4. host finds the mismatch uuid for NSID A (i.e. nvme_validate_ns() fails)
>>> -> host removes the respective namespace
>>>    
>> Entirely correct.
>> But Case2 results in the new namespace never to be scanned, and not visible to the OS. Which is the error I'm fighting with.
>>
> Ok but then it seems that your proposed patch doesn't address Case2, isn't it?
> It appears to me that the patch tries to address Case1 but with error code
> of "ANA inaccessible and DNR" set. IMO for case2, we may want to schedule
> queue scan again if nvme_validate_ns() fails due to the mismatch uuid.
> 
That would be possible, too. I'll see to whip up a patch.

Cheers,

Hannes
-- 
Dr. Hannes Reinecke                  Kernel Storage Architect
hare at suse.de                                +49 911 74053 688
SUSE Software Solutions GmbH, Frankenstr. 146, 90461 Nürnberg
HRB 36809 (AG Nürnberg), GF: I. Totev, A. McDonald, W. Knoblich