linux-kernel - Re: [PATCH 1/2] uio: add ioctl to uio

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [day] [month] [year] [list]

Message-ID: <Yg4/2rZsCmOazYgb@kroah.com>
Date:   Thu, 17 Feb 2022 13:30:18 +0100
From:   Greg KH <gregkh@...uxfoundation.org>
To:     Xiaoguang Wang <xiaoguang.wang@...ux.alibaba.com>
Cc:     Guixin Liu <kanie@...ux.alibaba.com>, bostroesser@...il.com,
        martin.petersen@...cle.com, linux-scsi@...r.kernel.org,
        target-devel@...r.kernel.org, linux-kernel@...r.kernel.org,
        xlpang@...ux.alibaba.com
Subject: Re: [PATCH 1/2] uio: add ioctl to uio

On Thu, Feb 17, 2022 at 08:15:29PM +0800, Xiaoguang Wang wrote:
> hi,
> 
> > On Thu, Feb 17, 2022 at 10:29:21AM +0800, Guixin Liu wrote:
> > > In TCMU, if backstore holds its own userspace buffer, for read cmd, the
> > > data needs to be copied from userspace buffer to tcmu data area first,
> > > and then needs to be copied from tcmu data area to scsi sgl pages again.
> > > 
> > > To solve this problem, add ioctl to uio to let userspace backstore can
> > > copy data between scsi sgl pages and its own buffer directly.
> > > 
> > > Reviewed-by: Xiaoguang Wang<xiaoguang.wang@...ux.alibaba.com>
> > > Signed-off-by: Guixin Liu<kanie@...ux.alibaba.com>
> > > ---
> > >   drivers/uio/uio.c          | 22 ++++++++++++++++++++++
> > >   include/linux/uio_driver.h |  1 +
> > No, sorry, thie uio driver will not be adding ioctls to them.  If you
> > need an ioctl, then you should not be using the UIO api but rather use a
> > custom character driver instead.
> 
> I found that early in 2015, there was developer trying to add ioctl interface
> to uio framework:https://lore.kernel.org/lkml/20151005080149.GB1747@kroah.com/

See, I rejected your ioctl for the very same reasons :)

It's good that I was consistent...

> Some of my customers use tcm_loop & tcmu to simulate block devices, it's tcmu
> driver that uses uio framework. There maybe more extra work if we tries to replace
> uio with a new character driver.

Why is tcmu using uio at all?

> Currently tcmu has performance bottleneck, Guixin's patch uses ioctl interface
> to bypass tcmu data area. I also have implemented a tcmu zero-copy feature that
> allows tcmu driver map io request sgl's pages to user space, which uses ioctl
> to do this mapping work, similar to network getsockopt(TCP_ZEROCOPY_RECEIVE).
> 
> I also understand your concerns about ioctl interface. Except that replacing
> uio with a new character driver in tcmu, are there any less complicated methods
> to complete our needs? Thanks.

I do not know what your needs are, nor what tcmu is, nor why you would
want to simulate a block device to userspace like this through the uio
api, sorry.

good luck!

greg k-h