Re: [PATCHv4 RFC 1/1] nvme-multipath: Add sysfs attributes for showing multipath info

[Nilay Shroff <nilay@linux.ibm.com>]

On 10/16/24 12:22, Hannes Reinecke wrote:
> On 10/16/24 05:19, Nilay Shroff wrote:
>> Hi Hannes,
>>
>> On 10/7/24 21:03, Nilay Shroff wrote:
>>>
>>>
>>> On 10/7/24 19:34, Hannes Reinecke wrote:
>>>> On 10/7/24 15:47, Nilay Shroff wrote:
>>>>>
>>>>>
>>>>> On 10/7/24 15:44, Hannes Reinecke wrote:
>>>>>> On 9/11/24 08:26, Nilay Shroff wrote:
>>>>>>> NVMe native multipath supports different IO policies for selecting I/O
>>>>>>> path, however we don't have any visibility about which path is being
>>>>>>> selected by multipath code for forwarding I/O.
>>>>>>> This patch helps add that visibility by adding new sysfs attribute files
>>>>>>> named "numa_nodes" and "queue_depth" under each namespace block device
>>>>>>> path /sys/block/nvmeXcYnZ/. We also create a "multipath" sysfs directory
>>>>>>> under head disk node and then from this directory add a link to each
>>>>>>> namespace path device this head disk node points to.
>>>>>>>
>>>>>>> For instance, /sys/block/nvmeXnY/multipath/ would create a soft link to
>>>>>>> each path the head disk node <nvmeXnY> points to:
>>>>>>>
>>>>>>> $ ls -1 /sys/block/nvme1n1/
>>>>>>> nvme1c1n1 -> ../../../../../pci052e:78/052e:78:00.0/nvme/nvme1/nvme1c1n1
>>>>>>> nvme1c3n1 -> ../../../../../pci058e:78/058e:78:00.0/nvme/nvme3/nvme1c3n1
>>>>>>>
>>>>>>> For round-robin I/O policy, we could easily infer from the above output
>>>>>>> that I/O workload targeted to nvme3n1 would toggle across paths nvme1c1n1
>>>>>>> and nvme1c3n1.
>>>>>>>
>>>>>>> For numa I/O policy, the "numa_nodes" attribute file shows the numa nodes
>>>>>>> being preferred by the respective block device path. The numa nodes value
>>>>>>> is comma delimited list of nodes or A-B range of nodes.
>>>>>>>
>>>>>>> For queue-depth I/O policy, the "queue_depth" attribute file shows the
>>>>>>> number of active/in-flight I/O requests currently queued for each path.
>>>>>>>
>>>>>>> Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
>>>>>>> ---
>>>>>>>     drivers/nvme/host/core.c      |  3 ++
>>>>>>>     drivers/nvme/host/multipath.c | 71 +++++++++++++++++++++++++++++++++++
>>>>>>>     drivers/nvme/host/nvme.h      | 20 ++++++++--
>>>>>>>     drivers/nvme/host/sysfs.c     | 20 ++++++++++
>>>>>>>     4 files changed, 110 insertions(+), 4 deletions(-)
>>>>>>>
>>>>>>> diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
>>>>>>> index 983909a600ad..6be29fd64236 100644
>>>>>>> --- a/drivers/nvme/host/core.c
>>>>>>> +++ b/drivers/nvme/host/core.c
>>>>>>> @@ -3951,6 +3951,9 @@ static void nvme_ns_remove(struct nvme_ns *ns)
>>>>>>>           if (!nvme_ns_head_multipath(ns->head))
>>>>>>>             nvme_cdev_del(&ns->cdev, &ns->cdev_device);
>>>>>>> +
>>>>>>> +    nvme_mpath_remove_sysfs_link(ns);
>>>>>>> +
>>>>>>>         del_gendisk(ns->disk);
>>>>>>>           mutex_lock(&ns->ctrl->namespaces_lock);
>>>>>>> diff --git a/drivers/nvme/host/multipath.c b/drivers/nvme/host/multipath.c
>>>>>>> index 518e22dd4f9b..7d9c36a7a261 100644
>>>>>>> --- a/drivers/nvme/host/multipath.c
>>>>>>> +++ b/drivers/nvme/host/multipath.c
>>>>>>> @@ -654,6 +654,8 @@ static void nvme_mpath_set_live(struct nvme_ns *ns)
>>>>>>>             nvme_add_ns_head_cdev(head);
>>>>>>>         }
>>>>>>>     +    nvme_mpath_add_sysfs_link(ns);
>>>>>>> +
>>>>>>>         mutex_lock(&head->lock);
>>>>>>>         if (nvme_path_is_optimized(ns)) {
>>>>>>>             int node, srcu_idx;
>>>>>> Nearly there.
>>>>> Thank you for your review comments!
>>>>>
>>>>>>
>>>>>> You can only call 'nvme_mpath_add_sysfs_link()' if the gendisk on the head had been created.
>>>>>>
>>>>>> And there is one branch in nvme_mpath_add_disk():
>>>>>>
>>>>>>                   if (desc.state) {
>>>>>>                           /* found the group desc: update */
>>>>>>                           nvme_update_ns_ana_state(&desc, ns);
>>>>>>
>>>>>> which does not go via nvme_mpath_set_live(), yet a device link would need to be create here, too.
>>>>>> But you can't call nvme_mpath_add_sysfs_link() from nvme_mpath_add_disk(), as the actual gendisk
>>>>>> might only be created later on during ANA log parsing.>>
>>>>>> It is a tangle, and I haven't found a good way out of this.
>>>>>> But I am _very much_ in favour of having these links, so please
>>>>>> update your patch.
>>>>>> In case disk supports ANA group then yes it would go through nvme_mpath_add_disk()->nvme_update_ns_ana_state();
>>>>>> and later nvme_update_ns_ana_state() would also fall through function nvme_mpath_set_live where we call
>>>>>> nvme_mpath_add_sysfs_link().
>>>>>
>>>>> So I think that in any case while multipath namespace is being created it has to go through
>>>>> nvme_mpath_set_live function. And as we see in nvme_mpath_set_live function, we only create
>>>>> sysfs link after the gendisk on the head is created. Do you agree with this? Or please let
>>>>> me know if you have any further question.
>>>>>
>>>> But it doesn't:
>>>>
>>>>          if (nvme_state_is_live(ns->ana_state) &&
>>>>              nvme_ctrl_state(ns->ctrl) == NVME_CTRL_LIVE)
>>>>                  nvme_mpath_set_live(ns);
>>>>
>>>> so if a namespace is being created for which nvme_state_is_live() returns 'false' nvme_mpath_set_live() is not called, and no links are being created.
>>>> This can happen eg. if the first namespace to be encountered is in any other state than 'optimized' or 'non-optimized'.
>>>>
>>>
>>> OK I got what you're suggesting here. So in this particular case when ANA state of a shared namespace
>>> is neither "optimized" nor "non-optimized", we would have gendisk for shared namepspace (i.e.
>>> nvmeXcYnZ) created but we don't have yet gendisk for corresponding head node (i.e. nvmeXnY) created.
>>> So without the gendisk for head node created, how could we create a link from it to the namespace node?
>>>
>>> The link from gendisk head node would be eventually created when the ANA state of the namespace transition
>>> to "optimized" or "non-optimized" state. I think, it's not anyways possible to have multipathing function
>>> enabled until the gendisk for the head node created, isn't it? So I don't yet understand why do we really
>>> need device link created if we don't have gendisk for head not ready? Am I missing anything here?
>>>
> You are right, we will create 'the' link for the gendisk node once 'nvme_mpath_set_live()' is called. But:
> Then we would have _one_ link (for the active path) only.
> All other paths (ie the non-active ones) might already been present, but won't get a link.
> 
> I'd love to have links to all paths present, active or not.
> 
OK, I think I got what you're asking here. I think we can make it the way you want.
I will work it out in the next patch.

Thanks,
--Nilay