[PATCH 0/2] nvme-cli: add support for an abort command with documentation

[minwoo.im.dev@gmail.com (Minwoo Im)]

On 01/24/2018 10:03 PM, Johannes Thumshirn wrote:
> On Wed, Jan 24, 2018@09:55:20PM +0900, Minwoo Im wrote:
>> Hello Johannes,
>>
>> On 01/24/2018 09:31 PM, Johannes Thumshirn wrote:
>>> On Wed, Jan 24, 2018@09:24:42PM +0900, Minwoo Im wrote:
>>>> It would be convenient to use abort command by a subcommand.
>>>> This command will strongly used by device developers to test device.
>>>
>>> Ahm... please no.
>>>
>>> If you need it for development just hack a small tool to create the
>>> passthrough ioctl() call but don't hand out loaded guns to children^Wusers.
>>>
>>
>> Thanks for your kindly comment.
>>
>> I thought nvme-cli tool would be used mostly by developers, not children
>> users. Also thought that it will not need any helps from kernel to abort a
>> specified command not just like delete I/O SQ/CQ so that nvme-cli could
>> handle this kind of thing.
> 
> nvme-cli will also be used by administrators intending to setup nvme over
> fabrics connections or debug nvme multipathing as well. It is certainly not a
> developer only tool.

Got it, Thanks for your kindly comment. :)

>> But, actually there is no limit for anybody to use this interface tool.
> 
> Yes, actually thinking more of it, the whole passthrough interface is
> dangerous. What happens if a user sends a delete CQ/SQ command via this
> interface and the kernel still tries to do I/O to it

Yes, That situation is one of the reason why I tried to add abort
command to nvme-cli because it does not need any helps from kernel.
As you said, delete CQ/SQ obviously needs helps from kernel to free
resources about queues.

Or.. Is there any cases that breaks behavior of kernel by an abort
command from user space?

Thanks,