On Mon, Sep 16, 2013 at 09:03:25PM +0200, Miklos Szeredi wrote:
> On Mon, Sep 16, 2013 at 8:19 PM, Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:
> > On Mon, Sep 16, 2013 at 02:51:56PM +0200, Miklos Szeredi wrote:
> >> From: Miklos Szeredi <mszeredi@xxxxxxx>
> >>
> >> commit b6f4bee02f "fs/9p: Fix atomic_open" fixed the O_EXCL behavior, but
> >> results in a dentry leak if v9fs_vfs_lookup() returns non-NULL.
> >
> > Frankly, I would prefer to deal with that in fs/namei.c:atomic_open()
> > instead. I.e. let it call finish_no_open() as it used to do and
> > turn
> > if (create_error && dentry->d_inode == NULL) {
> > error = create_error;
> > goto out;
> > }
> > in fs/namei.c:atomic_open() into
> > if (!dentry->d_inode) {
> > if (create_error) {
> > error = create_error;
> > goto out;
> > }
> > } else if ((open_flag & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) {
> > error = -EEXIST;
> > goto out;
> > }
> >
> > rather than try to deal with that crap in each instance of ->atomic_open()...
> > Objections?
>
> ->atomic_open() could be any one of
>
> lookup
> lookup+create
> lookup+create+open
>
> If it's the second one then the above is wrong. Sure, we could check
> FILE_CREATED as well, and if file wasn't created yet dentry is
> positive then we return EEXIST. But for that to be correct we need
> the last patch in the series, preventing FILE_CREATED from being set
> unconditionally.
You mean, lookup + create + return finish_no_open()? Does anything actually
do that? I agree that we want your "deal with setting FILE_CREATED in
filesystems", BTW, and I'm fine with putting it in front of the rest of
the queue.
I would definitely prefer EEXIST logics dealt with in fs/namei.c - if nothing
else, it had been done wrong in too many instances...
--
To unsubscribe from this list: send the line "unsubscribe stable" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html