On Sun, 18 Feb 2024 at 09:54, Christian Brauner <brauner@xxxxxxxxxx> wrote:
>
> So, I'm running out of time today so I'm appending the draft I jotted
> down now (untested).
This looks quite nice to me. Written like that with those well-named
helpers it is much more readable than the old nsfs code, honestly.
The only ugliness I see is the one that comes from the original code -
I'm not thrilled about the "return -EAGAIN" part, and I think that if
we found a previously stashed entry after all, we should loop.
But I think that whole horror comes from a fear of an endless loop
when the dentry is marked dead, and another CPU still sees the old
value (so you can't re-use it, and yet it's not NULL).
Which probably never happens, but I didn't think about it too much.
Anyway, that issue is pre-existing code, and your patch looks like an
improvement regardless, so that is *not* in any way a NAK, it's just
me musing on the code.
Al's email address has been bouncing for me for the last week or so,
so presumably he is not getting this and commenting on it. Oh well.
Anyway, ACK from me on this approach (assuming it passes your testing
when you get back, of course).
Linus
