Re: [PATCH v2 1/3] get rid of impossible checks in detach_attrs()/configfs_detach_item()

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

On Wed, Jun 03, 2026 at 10:09:37AM +0200, Christian Brauner wrote:
> On Wed, Jun 03, 2026 at 08:53:40AM +0100, Al Viro wrote:
> > 	Argh...  Please disregard the 1--3/3 part of the thread -
> > I've missed the stale (and really ancient - it is from back in
> > 2019) files sitting around.  Sorry, should've checked what the glob
> > had expanded to... ;-/
> 
> I'm more worried about what time it is where you currently are... ;)

4am ;-/

FWIW, the followups in #more.configfs switch the sucker to
..._recursive_removal() and massage it to building subtrees
detached from the main dentry tree; as the result, the locking
is _much_ milder there - all lockdep magic goes away and so
does the configfs_dirent_is_ready() crap.

Diffstat of that part:
 Documentation/filesystems/porting.rst |   8 +
 fs/configfs/configfs_internal.h       |  12 +-
 fs/configfs/dir.c                     | 705 +++++++-------------------
 fs/configfs/file.c                    |  43 --
 fs/configfs/inode.c                   |  41 +-
 fs/configfs/symlink.c                 |   9 -
 fs/dcache.c                           |   2 +-
 fs/libfs.c                            |  61 ++-
 include/linux/fs.h                    |   2 +
 include/linux/fsnotify.h              |  13 -
 net/sunrpc/rpc_pipe.c                 |   2 +-
 11 files changed, 243 insertions(+), 655 deletions(-)

and it even seems to survive light testing.  There still are bits
in old branches that need to be ported on top of that, but most
of the tricky massage is already in.  Unlike #work.configfs that's
the next cycle fodder, though.

I suspect that #fixes-configfs (first two commits on #work.configfs)
might be worth merging before -final; the rest of #work.configfs is
-next fodder if nobody has problems with it.  #more.configfs is
going to be rebased on top of 7.2-rc1 at the end of merge window...

Anyway, I'm really too sleepy right now - will post other stuff
intended for -next (starting with updated #work.dcache) after I get
some sleep.