From: Al Viro <viro@xxxxxxxxxxxxxxxx> On Behalf Of Al Viro
Sent: Friday, August 28, 2020 6:56 PM
> On Fri, Aug 28, 2020 at 07:39:32AM -0700, Konstantin Komarov wrote:
>
> > +static int ntfs_atomic_open(struct inode *dir, struct dentry *dentry,
> > + struct file *file, u32 flags, umode_t mode)
> > +{
[]
> > + if (d_really_is_positive(dentry)) {
> > + if (file->f_mode & FMODE_OPENED) {
>
> How do we get FMODE_OPENED here?
>
> > + dput(d);
> > + err = 0;
> > + } else
> > + err = finish_no_open(file, d);
> > + goto out1;
> > + }
> > + WARN_ON(d);
> > + }
> > +
> > + if (!(flags & O_CREAT)) {
> > + err = -ENOENT;
> > + goto out1;
> > + }
>
> Just return finish_no_open() in that case. And let the caller handle
> that.
>
> > + err = ntfs_create_inode(dir, dentry, file, mode, 0, NULL, 0, excl, fnd,
> > + &inode);
> > +
> > +out1:
> > + fnd_put(fnd);
> > +out:
> > + ni_unlock(ni);
> > +
> > + return err;
> > +}
>
> BTW, what's the point of that ni_lock() here? d_in_lookup() is stable
> regardless of that and any attempts to create something in the parent
> are serialized by ->i_rwsem. If you want it around the actual file
> creation, why not take it just there, and replace the open-coded
> ntfs_lookup() with the call of the real thing? As in
> if (d_in_lookup(dentry)) {
> d = ntfs_lookup(....);
> if (IS_ERR(d))
> return d;
> if (d)
> dentry = d;
> }
> if (!(flags & O_CREAT) || d_really_is_positive(dentry))
> return finish_no_open(file, d);
> /* deal with creation of file */
> ni_lock(...);
> ....
Thanks for the feedback! We refactored the atomic_open() based on
your concerns. Will be published in v4 patch today.
