Re: [PATCH v3 0/5] Store overlay real upper file in ovl_file

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

On Thu, Oct 17, 2024 at 03:52:10PM +0200, Miklos Szeredi wrote:
> On Thu, 17 Oct 2024 at 10:18, Amir Goldstein <amir73il@gmail.com> wrote:
> 
> > It has been like that since the first upstream version.
> > My guess is that it is an attempt to avoid turning wdentry
> > into a negative dentry, which is not expected to be useful in
> > ovl_clear_empty() situations, but this is just a guess.
> 
> Yes.  This was discussed in a private thread before merging overlayfs upstream.

10 years ago ;-/  Sorry, memories swapped out...

|> BTW, why on the Earth are you pinning that ->__upperdentry twice?  The
|> comment about d_delete() makes no sense whatsoever - anything other than
|> overlayfs itself would have to grab a reference to call that d_delete(),

[snip]

Out of curiosity, which callers pass anyone's upperdentry to ovl_cleanup()?
Note that if we get dentry reference not from ovl_dentry_upper() et.al.,
we'll get that protection for free.  From quick look it seems that the only
such callchain is from ovl_clear_empty()...

Oh, well...  Shifting it over there would be asking for trouble later on.
Might be worth a comment in ovl_cleanup() - it is subtle enough to be
confusing for readers.