[PATCH 2.6.24] block2mtd: removing a device and typo fixes

[Stephane Chazelas]

2008-02-12 16:21:24 +0100, Jörn Engel:
> On Tue, 12 February 2008 13:47:51 +0000, Stephane Chazelas wrote:
> > 
> > this patch addresses a number of small issues mainly regarding
> > the output made by this driver to dmesg:
> >   - Some of the "blkmtd"'s had not been changed to "block2mtd"
> >     which caused display problem
> >   - the parse_err() macro was displaying "block2mtd: " twice
> 
> Fairly obvious fixes.
> 
> > Also, one can add a block2mtd mtd device with things like:
> > 
> > echo /dev/loop3,$((256*1024)) |
> >   sudo tee /sys/module/block2mtd/parameters/block2mtd
> > 
> > but individual mtds cannot be removed. You can only do a
> > modprobe -r block2mtd to remove *all* the block2mtd mtds.
> > 
> > This patch proposes to add the cabability with:
> > 
> > echo /dev/loop3,remove |
> >   sudo tee /sys/module/block2mtd/parameters/block2mtd
> 
> Sounds sane enough.  But I do have some reservations about the
> implementation.  It would be best if you split the patch in two.  One
> with the obvious stuff above and one for this.

Hi Jörn,

OK, I can do that. Would the "simple" fix go to the
trivial at kernel.org Trivial Patch Monkey?

> The core of remove_device_by_name() is shared with block2mtd_exit(),
> so a common helper would be good.  Your error handling is better, so
> let's keep that version.

Well, yes that raised a concern to me, the "exit" function
returns "void". If the del_mtd_device fails with EBUSY at the
moment (such as when a /dev/mtdblock<x> is mounted), we ignore
it and go on with freeing everything leaving a dangling mtd.

Is there a way around that?

Another problem is that it's not easy to check whether a mtd
creation was successful or not. Basically, you have to write to
a /sys file and then parse /proc/mtd to get the result.

What about having a /dev/block2mtd (with owner/permissions that
could allow non-root users to use it), with 2 ioctls:

- one to "link" a block dev to a mtd that would take as
  parameter a fd to an open block dev (again allowing for
  flexible permissions) and would return the number of the
  allocated mtd and success/failure in errno. Upon sucess it
  would increase the refcnt of block2mtd.

- and one to "release" the link. That would fail if the mtd is
  in use and decrease block2mtd's refcnt upon success.

A bit like the loop devices (or /dev/ptmx) actually. What do you
think?

(also, with the /sys interface, isn't there a potential problem
with chroots wrt the path given to the /sys file?)

> And independently of your patch a mutex protecting the device list from
> simultaneous modifications would be good to have.
> 
> Side note: I may not have internet access until 19th or so.
[...]

I'll dig a little deeper, but I think I'll need some advice/help
at some point, I'm not a Linux kernel expert.

Regards,
Stephane