Re: [PATCH v7 09/12] sysfs: fix deadlock race with module removal

[Luis Chamberlain]

On Mon, Sep 20, 2021 at 07:53:16AM -1000, Tejun Heo wrote:
> Hello,
> 
> On Fri, Sep 17, 2021 at 10:04:27PM -0700, Luis Chamberlain wrote:
> > If try_module_get() fails we fail the operation on the kernfs node.
> > 
> > We use a try method as a full lock means we'd then make our sysfs
> > attributes busy us out from possible module removal, and so userspace
> > could force denying module removal, a silly form of "DOS" against module
> > removal. A try lock on the module removal ensures we give priority to
> > module removal and interacting with sysfs attributes only comes second.
> > Using a full lock could mean for instance that if you don't stop poking
> > at sysfs files you cannot remove a module.
> 
> I find this explanation odd because there's no real equivalent to locking
> the module (as opposed to try locking) 

Actually there is, __module_get() but I suspect some of these users are
probably incorrect and should be be moved to try. The documentation
about "rmmod --wait" for __module_get() is also outdated as that option
is no longer supported. I'll send an update for that later.

> because you can't wait for the
> removal to finish and then grant the lock, so any operation which increases
> the reference *has* to be a try method unless the caller already holds a
> reference to the same module and thus knows that the module is already
> pinned.

Right, the reason I mention the alternative is that we technically don't
need to use try in this case since during a kernfs op it is implied the
module will be pinned, but we have further motivations to use a try
method here: to avoid a possible DOS from module removal by userspace
mucking with ops.

> The code isn't wrong, so maybe just drop the related paragraphs in
> the commit message?

Does it make sense to clarify the above a bit more somehow? Or do think
its not needed?

> >  static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
> >  					     struct kernfs_node *parent,
> >  					     const char *name, umode_t mode,
> > +					     struct module *owner,
> >  					     kuid_t uid, kgid_t gid,
> >  					     unsigned flags)
> 
> Is there a particular reason why @owner is added between @mode and @uid?
> Sitting between two fs attributes seems a bit awkward. Maybe it can just be
> the last one?

No, I just picked an arbitrary place. Sure I'll move it to the end.

  Luis