[PATCH 2/3] nvme-multipath: cannot disconnect controller on stuck partition scan

[Sagi Grimberg]

On 15/10/2024 18:10, Keith Busch wrote:
> On Tue, Oct 15, 2024 at 04:56:08PM +0200, Hannes Reinecke wrote:
>> On 10/15/24 16:33, Keith Busch wrote:
>>> Is that actually happening for you? I don't think it's supposed to
>>> happen because mpath_shutdwon_disk only happens if the head is detached
>>> from the subsystem, so no other thread should be able to possibly
>>> attempt to access that head's disk for a different controller path. That
>>> new controller should have to allocate an entirely new head.
>> Na, forget it. Not only doesn't it happen, but we need the
>> 'test_and_set_bit()' at the start otherwise we can't abort I/O.
>> So all's fine with your patch, and it even survives my testcase.
>>
>> (The first to do so even after several runs, I might add :-)
>>
>> Care to send an official patch?
> Should be in your mailbox. Here's the link:
>
>    https://lore.kernel.org/linux-nvme/20241015143136.810779-1-kbusch@meta.com/T/#u

I'm glad we didn't take the original version of this patch :)