[dm-devel] [PATCH v1 0/5] dm: dm-user: New target that proxies BIOs to userspace

[Desnoyers, Peter]

I seemed to have missed this during the OSDI and end-of-term rush. I've implemented a couple of user-space block device components for research projects, so I thought I'd weigh in.

In particular, does this offer significant advantages over TCMU? It uses the UIO framework for a ring of SCSI requests to user space, and seems to be quite efficient. The differences seem to be:

  *   dm-user is at bio level, while TCMU is a SCSI device, so like NBD you get request merging. Depending on your purposes, that either does no harm or is really useful.
  *   TCMU requires either accepting the entire tcmu-runner architecture and philosophy, which may be a no-go, or writing your own SCSI implementation

(I'm actually really bummed that I didn't discover TCMU until recently, as I think we would have used it on a few research projects in the past.)

I have no idea whether you can hand things off to another process - I don't think the UIO framework would have any problems with that, so it probably comes down to whether the target_core_user module would allow its memory region to be mapped by multiple processes.


I've got the world's quickest and dirtiest SCSI implementation that I've hacked up to use with TCMU, although the only real experimentation I've done is by hacking tcmu-runner. I'd definitely be interested in talking to any people who are interested in this, to develop a proper SCSI library that would allow you to use TCMU outside of tcmu-runner.

________________________________
From: Bart Van Assche <bvanassche at acm.org>
Sent: Wednesday, December 9, 2020 10:38 PM
To: Palmer Dabbelt <palmer at dabbelt.com>; Christoph Hellwig <hch at infradead.org>
Cc: snitzer at redhat.com <snitzer at redhat.com>; linux-doc at vger.kernel.org <linux-doc at vger.kernel.org>; shuah at kernel.org <shuah at kernel.org>; corbet at lwn.net <corbet at lwn.net>; linux-kernel at vger.kernel.org <linux-kernel at vger.kernel.org>; Josef Bacik <josef at toxicpanda.com>; linux-raid at vger.kernel.org <linux-raid at vger.kernel.org>; song at kernel.org <song at kernel.org>; dm-devel at redhat.com <dm-devel at redhat.com>; Mike Christie <michael.christie at oracle.com>; linux-kselftest at vger.kernel.org <linux-kselftest at vger.kernel.org>; kernel-team at android.com <kernel-team at android.com>; agk at redhat.com <agk at redhat.com>
Subject: Re: [dm-devel] [PATCH v1 0/5] dm: dm-user: New target that proxies BIOs to userspace

On 12/7/20 10:55 AM, Palmer Dabbelt wrote:
> All in all, I've found it a bit hard to figure out what sort of interest
> people
> have in dm-user: when I bring this up I seem to run into people who've done
> similar things before and are vaguely interested, but certainly nobody is
> chomping at the bit.  I'm sending it out in this early state to try and
> figure
> out if it's interesting enough to keep going.

Cc-ing Josef and Mike since their nbd contributions make me wonder
whether this new driver could be useful to their use cases?

Thanks,

Bart.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/dm-devel/attachments/20210105/8ffdb5f4/attachment.htm>