Re: [PATCH] Re: Possible FS race condition between iterate_dir and d_alloc_parallel

[Al Viro <viro@zeniv.linux.org.uk>]

On Sun, Sep 15, 2019 at 10:58:50AM -0700, Linus Torvalds wrote:
> On Sun, Sep 15, 2019 at 9:02 AM Al Viro <viro@zeniv.linux.org.uk> wrote:
> >
> > Could be done, AFAICS.  I'm not even sure we need a flag per se - we
> > have two cases when the damn thing is not in the list and "before
> > everything" case doesn't really need to be distinguished from post-EOF
> > one.
> 
> Agreed, it looks like we could just look at f_pos and use that
> (together with whether we have a cursor or not) as the flag:
> 
>  - no cursor: f_pos < 2 means beginning, otherwise EOF
> 
>  - otherwise: cursor points to position

FWIW, #next.dcache has the straight conversion to hlist.  It definitely
wants at least nfsd, er... misconception dealt with, though: list_head
or hlist, this
static void nfsdfs_remove_files(struct dentry *root)
{
        struct dentry *dentry;
        struct hlist_node *n;

        hlist_for_each_entry_safe(dentry, n, &root->d_children, d_sibling) {
                if (!simple_positive(dentry)) {
                        WARN_ON_ONCE(1); /* I think this can't happen? */
                        continue;
                }
                nfsdfs_remove_file(d_inode(root), dentry);
        }
}
is wrong, for obvious reasons (have the victim directory opened before that
gets called and watch the fireworks)...

No "take cursors out of the list" parts yet.