Re: Possible kernel fs block code regression in 6.2.3 umounting usb drives

[Jens Axboe]

n 3/11/23 2:32?AM, Greg Kroah-Hartman wrote:
> On Fri, Mar 10, 2023 at 02:53:19PM -0800, Eric Biggers wrote:
>> On Fri, Mar 10, 2023 at 04:08:21PM -0500, Genes Lists wrote:
>>> On 3/10/23 15:23, Jens Axboe wrote:
>>>> On 3/10/23 1:16?PM, Eric Biggers wrote:
>>> ...
>>>> But I would revert:
>>>>
>>>> bfe46d2efe46c5c952f982e2ca94fe2ec5e58e2a
>>>> 57a425badc05c2e87e9f25713e5c3c0298e4202c
>>>>
>>>> in that order from 6.2.3 and see if that helps. Adding Yu.
>>>>
>>> Confirm the 2 Reverts fixed in my tests as well (nvme + sata drives).
>>> Nasty crash - some needed to be power cycled as they hung on shutdown.
>>>
>>> Thank you!
>>>
>>> gene
>>>
>>>
>>
>> Great, thanks.  BTW, 6.1 is also affected.  A simple reproducer is to run:
>>
>> 	dmsetup create dev --table "0 128 zero"
>> 	dmsetup remove dev
>>
>> The following kconfigs are needed for the bug to be hit:
>>
>> 	CONFIG_BLK_CGROUP=y
>> 	CONFIG_BLK_DEV_THROTTLING=y
>> 	CONFIG_BLK_DEV_THROTTLING_LOW=y
>>
>> Sasha or Greg, can you please revert the indicated commits from 6.1 and 6.2?
> 
> Yes, will go do that right now, thanks for debugging this so quickly!

The issue here is that parts of a series was auto-selected. That seems
like a bad idea to do for stable. Just because something applies without
other parts of the series doesn't mean it's sane to backport by itself.

How do we prevent that from happening? Maybe we just need to default
to "if whole series doesn't pick cleanly, don't grab any parts of it
in auto-selection"? Exception being if it's explicitly marked for stable,
not uncommon to have a series that starts with a fix or two which should
go to stable, then the feature bits.

-- 
Jens Axboe