Re: [PATCH] vfs: avoid spurious dentry ref/unref cycle on open

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

On Wed, Aug 07, 2024 at 11:50:50AM +0200, Mateusz Guzik wrote:

> tripping ip:
> vfs_tmpfile+0x162/0x230:
> fsnotify_parent at include/linux/fsnotify.h:81
> (inlined by) fsnotify_file at include/linux/fsnotify.h:131
> (inlined by) fsnotify_open at include/linux/fsnotify.h:401
> (inlined by) vfs_tmpfile at fs/namei.c:3781

Try this for incremental; missed the fact that finish_open() is
used by ->tmpfile() instances, not just ->atomic_open().

Al, crawling back to sleep...

diff --git a/fs/namei.c b/fs/namei.c
index 95345a5beb3a..0536907e8e79 100644
--- a/fs/namei.c
+++ b/fs/namei.c
@@ -3776,7 +3776,10 @@ int vfs_tmpfile(struct mnt_idmap *idmap,
 	file->f_path.dentry = child;
 	mode = vfs_prepare_mode(idmap, dir, mode, mode, mode);
 	error = dir->i_op->tmpfile(idmap, dir, file, mode);
-	dput(child);
+	if (file->f_mode & FMODE_OPENED)
+		mntget(parentpath->mnt);
+	else
+		dput(child);
 	if (file->f_mode & FMODE_OPENED)
 		fsnotify_open(file);
 	if (error)