Hello,

On Thu, 23 Jun 2022 11:30:04 +0200 Paolo Abeni wrote:

quoted

When the link layer connection is broken, the rose->neighbour is
set to null. But rose->neighbour could be used by rose_connection()
and rose_release() later, because there is no synchronization among
them. As a result, the null-ptr-deref bugs will happen.

One of the null-ptr-deref bugs is shown below:

    (thread 1)                  |        (thread 2)
                                |  rose_connect
rose_kill_by_neigh              |    lock_sock(sk)
  spin_lock_bh(&rose_list_lock) |    if (!rose->neighbour)
  rose->neighbour = NULL;//(1)  |
                                |    rose->neighbour->use++;//(2)

The rose->neighbour is set to null in position (1) and dereferenced
in position (2).

The KASAN report triggered by POC is shown below:

KASAN: null-ptr-deref in range [0x0000000000000028-0x000000000000002f]
...
RIP: 0010:rose_connect+0x6c2/0xf30
RSP: 0018:ffff88800ab47d60 EFLAGS: 00000206
RAX: 0000000000000005 RBX: 000000000000002a RCX: 0000000000000000
RDX: ffff88800ab38000 RSI: ffff88800ab47e48 RDI: ffff88800ab38309
RBP: dffffc0000000000 R08: 0000000000000000 R09: ffffed1001567062
R10: dfffe91001567063 R11: 1ffff11001567061 R12: 1ffff11000d17cd0
R13: ffff8880068be680 R14: 0000000000000002 R15: 1ffff11000d17cd0
...
Call Trace:
  <TASK>
  ? __local_bh_enable_ip+0x54/0x80
  ? selinux_netlbl_socket_connect+0x26/0x30
  ? rose_bind+0x5b0/0x5b0
  __sys_connect+0x216/0x280
  __x64_sys_connect+0x71/0x80
  do_syscall_64+0x43/0x90
  entry_SYSCALL_64_after_hwframe+0x46/0xb0

This patch adds lock_sock() in rose_kill_by_neigh() in order to
synchronize with rose_connect() and rose_release().

Meanwhile, this patch adds sock_hold() protected by rose_list_lock
that could synchronize with rose_remove_socket() in order to mitigate
UAF bug caused by lock_sock() we add.

What's more, there is no need using rose_neigh_list_lock to protect
rose_kill_by_neigh(). Because we have already used rose_neigh_list_lock
to protect the state change of rose_neigh in rose_link_failed(), which
is well synchronized.

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Signed-off-by: Duoming Zhou <redacted>
---
 net/rose/af_rose.c    | 5 +++++
 net/rose/rose_route.c | 2 ++
 2 files changed, 7 insertions(+)

diff --git a/net/rose/af_rose.c b/net/rose/af_rose.c
index bf2d986a6bc..dece637e274 100644
--- a/net/rose/af_rose.c
+++ b/net/rose/af_rose.c

@@ -169,9 +169,14 @@ void rose_kill_by_neigh(struct rose_neigh *neigh)
 		struct rose_sock *rose = rose_sk(s);
 
 		if (rose->neighbour == neigh) {
+			sock_hold(s);
 			rose_disconnect(s, ENETUNREACH, ROSE_OUT_OF_ORDER, 0);
 			rose->neighbour->use--;
+			spin_unlock_bh(&rose_list_lock);

You can't release the lock protecting the list traversal, then re-
acquire it and keep traversing using the same iterator. The list could
be modified in-between.

I think release the lock and then reacquire it is ok. Because we have held the
refcount of sock and called rose_disconnect() to change the state of sock with
the protection of rose_list_lock which could synchronize with rose_destroy_socket().

If the sock is removed from the list by rose_destroy_socket(), there is
no rose->neighbour equals to neigh and the rose_kill_by_neigh() will return.

If there is a rose->neighbour equals to neigh, we held the refcount of sock
and called the rose_disconnect() to change the state of it with the protection
of rose_list_lock. Even if the sock could be removed from the rose_list by
rose_destroy_socket() during the time of unlocking, but the sock will not be 
deallocated because we have held the refcount of sock. When we reacquire the 
rose_list_lock, we only do sock_put() in order to deallocate the sock.

@@ -169,9 +169,15 @@ void rose_kill_by_neigh(struct rose_neigh *neigh)
                struct rose_sock *rose = rose_sk(s);
 
                if (rose->neighbour == neigh) {
+                       sock_hold(s);
                        rose_disconnect(s, ENETUNREACH, ROSE_OUT_OF_ORDER, 0);
                        rose->neighbour->use--;
+                       spin_unlock_bh(&rose_list_lock);
+                       lock_sock(s);
                        rose->neighbour = NULL;
+                       release_sock(s);
+                       spin_lock_bh(&rose_list_lock);
+                       sock_put(s);
                }
        }
        spin_unlock_bh(&rose_list_lock);

Instead you could build a local list containing the relevant sockets
(under the rose_list_lock protection), additionally acquiring a
reference to each of them

Then traverse such list outside the rose_list_lock, acquire the socket
lock on each of them, do the neigh clearing and release the reference.

If we build a local list contain the relevant sockets and only acquire a reference
to each of them with the protection of rose_list_lock, the socket could be removed
by rose_destroy_socket() after we release the rose_list_lock. Then if we traverse 
such list outside the rose_list_lock we could not find the socket, as a result,
the neigh clearing and the refcount releasing operations will not be executed.

Best regards,
Duoming Zhou