Re: [PATCH 5/7] fanotify: record new parent and name in MOVED_FROM event

From: Jan Kara <jack@suse.cz>
To: Amir Goldstein <amir73il@gmail.com>
Cc: Jan Kara <jack@suse.cz>,
	Matthew Bobrowski <mbobrowski@mbobrowski.org>,
	linux-fsdevel@vger.kernel.org
Subject: Re: [PATCH 5/7] fanotify: record new parent and name in MOVED_FROM event
Date: Fri, 12 Nov 2021 17:48:24 +0100	[thread overview]
Message-ID: <20211112164824.GB30295@quack2.suse.cz> (raw)
In-Reply-To: <20211029114028.569755-6-amir73il@gmail.com>

On Fri 29-10-21 14:40:26, Amir Goldstein wrote:
> In the special case of MOVED_FROM event, if we are recording the child
> fid due to FAN_REPORT_TARGET_FID init flag, we also record the new
> parent and name.
> 
> Signed-off-by: Amir Goldstein <amir73il@gmail.com>
...
> diff --git a/fs/notify/fanotify/fanotify.c b/fs/notify/fanotify/fanotify.c
> index 795bedcb6f9b..d1adcb3437c5 100644
> --- a/fs/notify/fanotify/fanotify.c
> +++ b/fs/notify/fanotify/fanotify.c
> @@ -592,21 +592,30 @@ static struct fanotify_event *fanotify_alloc_name_event(struct inode *id,
>  							__kernel_fsid_t *fsid,
>  							const struct qstr *name,
>  							struct inode *child,
> +							struct dentry *moved,
>  							unsigned int *hash,
>  							gfp_t gfp)
>  {
>  	struct fanotify_name_event *fne;
>  	struct fanotify_info *info;
>  	struct fanotify_fh *dfh, *ffh;
> +	struct inode *dir2 = moved ? d_inode(moved->d_parent) : NULL;

I think we need to be more careful here (like dget_parent()?). Fsnotify is
called after everything is unlocked after rename so dentry can be changing
under us, cannot it? Also does that mean that we could actually get a wrong
parent here if the dentry is renamed immediately after we unlock things and
before we report fsnotify event?

								Honza
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR