Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Jann Horn <jannh@google.com>
To:     Al Viro <viro@zeniv.linux.org.uk>
Cc:     Linus Torvalds <torvalds@linux-foundation.org>,
        Miklos Szeredi <mszeredi@redhat.com>,
        linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org,
        Will Deacon <will@kernel.org>
Subject: [PATCH] fs: add memory barrier in __fget_light()
Date:   Mon, 31 Oct 2022 18:13:07 +0100
Message-Id: <20221031171307.2784981-1-jannh@google.com>
MIME-Version: 1.0
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

We must prevent the CPU from reordering the files->count read with the
FD table access like this, on architectures where read-read reordering is
possible:

    files_lookup_fd_raw()
                                  close_fd()
                                  put_files_struct()
    atomic_read(&files->count)

I would like to mark this for stable, but the stable rules explicitly say
"no theoretical races", and given that the FD table pointer and
files->count are explicitly stored in the same cacheline, this sort of
reordering seems quite unlikely in practice...

If this is too expensive on platforms like arm64, I guess the more
performant alternative would be to add another flags field that tracks
whether the fs_struct was ever shared and check that instead of the
reference count in __fget_light().

Signed-off-by: Jann Horn <jannh@google.com>
---
 fs/file.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/fs/file.c b/fs/file.c
index 5f9c802a5d8d3..6144287ddc0fe 100644
--- a/fs/file.c
+++ b/fs/file.c
@@ -1004,6 +1004,18 @@ static unsigned long __fget_light(unsigned int fd, f=
mode_t mask)
 	struct file *file;
=20
 	if (atomic_read(&files->count) =3D=3D 1) {
+		/*
+		 * If another thread is concurrently calling close_fd() followed
+		 * by put_files_struct(), we must not observe the old table
+		 * entry combined with the new refcount - otherwise we could
+		 * return a file that is concurrently being freed.
+		 *
+		 * Pairs with atomic_dec_and_test() in put_files_struct().
+		 * An alternative to using a barrier here would be to use a
+		 * separate field in files_struct to track whether it was ever
+		 * shared.
+		 */
+		smp_rmb();
 		file =3D files_lookup_fd_raw(files, fd);
 		if (!file || unlikely(file->f_mode & mask))
 			return 0;

base-commit: 30a0b95b1335e12efef89dd78518ed3e4a71a763
--=20
2.38.1.273.g43a17bfeac-goog