Re: [PATCH] vfs: Tighten up linkat(..., AT_EMPTY_PATH)

[Oleg Nesterov <oleg@redhat.com>]

On 08/21, Andy Lutomirski wrote:
>
> On Wed, Aug 21, 2013 at 11:20 AM, Oleg Nesterov <oleg@redhat.com> wrote:
> > Can't really comment the patch, just a nit:
> >
> > On 08/21, Andy Lutomirski wrote:
> >>
> >> +static bool may_flink(const struct path *path)
> >> +{
> >> +     bool ret;
> >> +     struct inode *inode = path->dentry->d_inode;
> >> +
> >> +     /*
> >> +      * This is racy: I_LINKABLE could be cleared between this check
> >> +      * and the actual link operation.
> >
> > OK,
> >
> >> +     spin_lock(&inode->i_lock);
> >> +     ret = !!(inode->i_state & I_LINKABLE);
> >> +     spin_unlock(&inode->i_lock);
> >
> > so why do we need to take a lock ?
> >
>
> We probably don't.  But other accesses to this field take that lock,

Not if you only need to check ->i_lock,

> (In principle, someone could take the lock, write I_LINKABLE, clear
> it, and unlock, and we'd get confused if we didn't take the lock
> ourselves.)

Or it can be cleared right after we drop i_lock, I do not think there
is any difference.

But OK, I won't argue.

Oleg.