Re: [PATCH] ceph: is_root_ceph_dentry() cleanup

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

On Mon, Jan 27, 2025 at 05:10:23PM -0800, Viacheslav Dubeyko wrote:
> From: Viacheslav Dubeyko <Slava.Dubeyko@ibm.com>
> 
> This patch introduces CEPH_HIDDEN_DIR_NAME. It
> declares name of the hidden directory .ceph in
> the include/linux/ceph/ceph_fs.h instead of hiding
> it in dir.c file. Also hardcoded length of the name
> is changed on strlen(CEPH_HIDDEN_DIR_NAME).

Hmm...

Speaking of that area
	* how the hell could ceph_lookup() ever be called with dentry
that is *NOT* negative?  VFS certainly won't do that; I'm not sure about
ceph_handle_notrace_create(), but it doesn't look like that's possible
without server being malicious (if it's possible at all).

	* speaking of malicious servers, what happens if
it gets CEPH_MDS_OP_LOOKUP and it returns a normal reply to positive
lookup, but with cpu_to_le32(-ENOENT) shoved into head->result?
	AFAICS, ceph_handle_snapdir() will be called with dentry
that is already made positive; results will not be pretty...