Re: [PATCH v13 0/9] nvmet: add target passthru commands support

[Chaitanya Kulkarni]

On 5/14/20 10:23 AM, Logan Gunthorpe wrote:
> This is v13 of the passthru patchset which is mostly a resend of v12
> with Sagi's reviewed-by tags collected.
> 
> Below, I'll reiterrate some points I made previously that haven't been
> responded to:
> 
> I don't think cloning the ctrl_id or the subsysnqn is a good idea.
> I sent an email trying to explain why here[1] but there was no response.
> In short, I think cloning the ctrl_id will break multipathing over
> fabrics and copying the subsysnqn only has the effect of breaking
> loopback; the user can always copy the underlying subsysnqn if it
> makes sense for their overall system.
> 
> I maintain overriding the CMIC bit in the ctrl id is necessary to
> allow multipath over fabrics even if the underlying device did
> not support multipath.
> 
> I also think the black list for admin commands is appropriate, and I
> added it based on Sagi's feedback[2]. There are plenty of commands that
> may be dangerous like firmware update and format NVM commands, and NS
> attach commands won't work out of the box because we don't copy the
> ctrl_id. It seems like there's more commands to be careful of than ones
> that are that are obviously acceptable. So, I think the prudent course
> is blacklisting by default until someone has a usecase and can show
> the command is safe seems and makes sense. For our present use cases,
> the identify, log page and vendor specific commands are all that we
> care about.
> 
> A git branch is available here and is based on v5.7-rc5:
> 
> https://github.com/sbates130272/linux-p2pmem  nvmet_passthru_v13
> 
> [1]https://lore.kernel.org/linux-block/247eca47-c3bc-6452-fb19-f7aa27b05a60@xxxxxxxxxxxx/
> [2]https://lore.kernel.org/linux-block/e4430207-7def-8776-0289-0d58689dc0cd@xxxxxxxxxxx/
> 
> --
> 
> v13 Changes:
>    1. Rebased onto v5.7-rc5
>    2. Collected Sagi's Reviewed-by tags

Are you planning to send V14 based on nvme-5.9 branch ?