lkml.org 
[lkml]   [2010]   [Feb]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH 01/17] MTD: create lockless versions of {get,put}_mtd_device This will be used to resolve deadlock in block translation layer.
From
Date
On Tue, 2010-02-09 at 19:23 +0200, Maxim Levitsky wrote: 
> On Tue, 2010-02-09 at 18:14 +0100, Peter Zijlstra wrote:
> > On Tue, 2010-02-09 at 18:57 +0200, Maxim Levitsky wrote:
> > > These functions can be used as long as we don't need access to global mtd table, but have
> > > a pointer to the mtd device.
> > >
> > > Signed-off-by: Maxim Levitsky <maximlevitsky@gmail.com>
> > > ---
> > > drivers/mtd/mtdcore.c | 60 ++++++++++++++++++++++++++++++----------------
> > > include/linux/mtd/mtd.h | 3 +-
> > > 2 files changed, 41 insertions(+), 22 deletions(-)
> >
> > > +int __get_mtd_device(struct mtd_info *mtd)
> > > +{
> > > + int err;
> > > +
> > > + if (!try_module_get(mtd->owner))
> > > + return -ENODEV;
> > > +
> > > + if (mtd->get_device) {
> > > +
> > > + err = mtd->get_device(mtd);
> > > +
> > > + if (err) {
> > > + module_put(mtd->owner);
> > > + return err;
> > > + }
> > > + }
> > > + mtd->usecount++;
> > > + return 0;
> > > }
> >
> > > +void __put_mtd_device(struct mtd_info *mtd)
> > > +{
> > > + --mtd->usecount;
> > > + BUG_ON(mtd->usecount < 0);
> > > +
> > > if (mtd->put_device)
> > > mtd->put_device(mtd);
> > >
> > > module_put(mtd->owner);
> > > }
> >
> > That's racy, use kref.
> >
> Couldn't agree with you more.
>
> However, these functions aren't intended for general use, and probably
> will be used by mtd translation layer only. I do have a lock that
> protects concurrent use of these functions.
>
> Thus, I better add a comment about this?
However on second thought, there is still a race if two FTLs access same
mtd device.
While this might seem impossible, and I say it is quite dangerous. I can
imagine using both some FTL and mtdblock for testing.

Just using kref (nothing against it) won't help here.
The mtd->get_device/put_device aren't expecting to be called
concurrently ether...

I can add per mtd lock, but it is a bit ugly...

What do you think?

Best regards,
Maxim Levitsky





\
 
 \ /
  Last update: 2010-02-09 18:49    [W:0.363 / U:0.056 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site