There are two paths to access mptcp_pm_del_add_timer, result in a race
condition:

     CPU1				CPU2
     ====                               ====
     net_rx_action
     napi_poll                          netlink_sendmsg
     __napi_poll                        netlink_unicast
     process_backlog                    netlink_unicast_kernel
     __netif_receive_skb                genl_rcv
     __netif_receive_skb_one_core       netlink_rcv_skb
     NF_HOOK                            genl_rcv_msg
     ip_local_deliver_finish            genl_family_rcv_msg
     ip_protocol_deliver_rcu            genl_family_rcv_msg_doit
     tcp_v4_rcv                         mptcp_pm_nl_flush_addrs_doit
     tcp_v4_do_rcv                      mptcp_nl_remove_addrs_list
     tcp_rcv_established                mptcp_pm_remove_addrs_and_subflows
     tcp_data_queue                     remove_anno_list_by_saddr
     mptcp_incoming_options             mptcp_pm_del_add_timer
     mptcp_pm_del_add_timer             kfree(entry)

In remove_anno_list_by_saddr(running on CPU2), after leaving the critical
zone protected by "pm.lock", the entry will be released, which leads to the
occurrence of uaf in the mptcp_pm_del_add_timer(running on CPU1).

Keeping a reference to add_timer inside the lock, and calling
sk_stop_timer_sync() with this reference, instead of "entry->add_timer".

Move list_del(&entry->list) to mptcp_pm_del_add_timer and inside the pm lock,
do not directly access any members of the entry outside the pm lock, which
can avoid similar "entry->x" uaf.

Fixes: 00cfd77b9063 ("mptcp: retransmit ADD_ADDR when timeout")
Cc: stable@vger.kernel.org
Reported-and-tested-by: syzbot+f3a31fb909db9b2a5c4d@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=f3a31fb909db9b2a5c4d
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Signed-off-by: Edward Adam Davis <eadavis@qq.com>
---
 net/mptcp/pm_netlink.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)

diff --git a/net/mptcp/pm_netlink.c b/net/mptcp/pm_netlink.c
index 3e4ad801786f..f195b577c367 100644
--- a/net/mptcp/pm_netlink.c
+++ b/net/mptcp/pm_netlink.c
@@ -331,15 +331,21 @@ mptcp_pm_del_add_timer(struct mptcp_sock *msk,
 {
 	struct mptcp_pm_add_entry *entry;
 	struct sock *sk = (struct sock *)msk;
+	struct timer_list *add_timer = NULL;
 
 	spin_lock_bh(&msk->pm.lock);
 	entry = mptcp_lookup_anno_list_by_saddr(msk, addr);
-	if (entry && (!check_id || entry->addr.id == addr->id))
+	if (entry && (!check_id || entry->addr.id == addr->id)) {
 		entry->retrans_times = ADD_ADDR_RETRANS_MAX;
+		add_timer = &entry->add_timer;
+	}
+	if (!check_id && entry)
+		list_del(&entry->list);
 	spin_unlock_bh(&msk->pm.lock);
 
-	if (entry && (!check_id || entry->addr.id == addr->id))
-		sk_stop_timer_sync(sk, &entry->add_timer);
+	/* no lock, because sk_stop_timer_sync() is calling del_timer_sync() */
+	if (add_timer)
+		sk_stop_timer_sync(sk, add_timer);
 
 	return entry;
 }
@@ -1430,7 +1436,6 @@ static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
 
 	entry = mptcp_pm_del_add_timer(msk, addr, false);
 	if (entry) {
-		list_del(&entry->list);
 		kfree(entry);
 		return true;
 	}
-- 
2.43.0

Hi Edward,

On 10/09/2024 11:58, Edward Adam Davis wrote:
> There are two paths to access mptcp_pm_del_add_timer, result in a race
> condition:
> 
>      CPU1				CPU2
>      ====                               ====
>      net_rx_action
>      napi_poll                          netlink_sendmsg
>      __napi_poll                        netlink_unicast
>      process_backlog                    netlink_unicast_kernel
>      __netif_receive_skb                genl_rcv
>      __netif_receive_skb_one_core       netlink_rcv_skb
>      NF_HOOK                            genl_rcv_msg
>      ip_local_deliver_finish            genl_family_rcv_msg
>      ip_protocol_deliver_rcu            genl_family_rcv_msg_doit
>      tcp_v4_rcv                         mptcp_pm_nl_flush_addrs_doit
>      tcp_v4_do_rcv                      mptcp_nl_remove_addrs_list
>      tcp_rcv_established                mptcp_pm_remove_addrs_and_subflows
>      tcp_data_queue                     remove_anno_list_by_saddr
>      mptcp_incoming_options             mptcp_pm_del_add_timer
>      mptcp_pm_del_add_timer             kfree(entry)
> 
> In remove_anno_list_by_saddr(running on CPU2), after leaving the critical
> zone protected by "pm.lock", the entry will be released, which leads to the
> occurrence of uaf in the mptcp_pm_del_add_timer(running on CPU1).
> 
> Keeping a reference to add_timer inside the lock, and calling
> sk_stop_timer_sync() with this reference, instead of "entry->add_timer".
> 
> Move list_del(&entry->list) to mptcp_pm_del_add_timer and inside the pm lock,
> do not directly access any members of the entry outside the pm lock, which
> can avoid similar "entry->x" uaf.
> 
> Fixes: 00cfd77b9063 ("mptcp: retransmit ADD_ADDR when timeout")
> Cc: stable@vger.kernel.org
> Reported-and-tested-by: syzbot+f3a31fb909db9b2a5c4d@syzkaller.appspotmail.com
> Closes: https://syzkaller.appspot.com/bug?extid=f3a31fb909db9b2a5c4d
> Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
> Signed-off-by: Edward Adam Davis <eadavis@qq.com>
It looks also good to me, but no need for me to add a reviewed-by tag I
suppose.

This v4 can be applied in netdev directly.

Cheers,
Matt
-- 
Sponsored by the NGI0 Core fund.

On 9/10/24 11:58, Edward Adam Davis wrote:
> There are two paths to access mptcp_pm_del_add_timer, result in a race
> condition:
> 
>       CPU1				CPU2
>       ====                               ====
>       net_rx_action
>       napi_poll                          netlink_sendmsg
>       __napi_poll                        netlink_unicast
>       process_backlog                    netlink_unicast_kernel
>       __netif_receive_skb                genl_rcv
>       __netif_receive_skb_one_core       netlink_rcv_skb
>       NF_HOOK                            genl_rcv_msg
>       ip_local_deliver_finish            genl_family_rcv_msg
>       ip_protocol_deliver_rcu            genl_family_rcv_msg_doit
>       tcp_v4_rcv                         mptcp_pm_nl_flush_addrs_doit
>       tcp_v4_do_rcv                      mptcp_nl_remove_addrs_list
>       tcp_rcv_established                mptcp_pm_remove_addrs_and_subflows
>       tcp_data_queue                     remove_anno_list_by_saddr
>       mptcp_incoming_options             mptcp_pm_del_add_timer
>       mptcp_pm_del_add_timer             kfree(entry)
> 
> In remove_anno_list_by_saddr(running on CPU2), after leaving the critical
> zone protected by "pm.lock", the entry will be released, which leads to the
> occurrence of uaf in the mptcp_pm_del_add_timer(running on CPU1).
> 
> Keeping a reference to add_timer inside the lock, and calling
> sk_stop_timer_sync() with this reference, instead of "entry->add_timer".
> 
> Move list_del(&entry->list) to mptcp_pm_del_add_timer and inside the pm lock,
> do not directly access any members of the entry outside the pm lock, which
> can avoid similar "entry->x" uaf.
> 
> Fixes: 00cfd77b9063 ("mptcp: retransmit ADD_ADDR when timeout")
> Cc: stable@vger.kernel.org
> Reported-and-tested-by: syzbot+f3a31fb909db9b2a5c4d@syzkaller.appspotmail.com
> Closes: https://syzkaller.appspot.com/bug?extid=f3a31fb909db9b2a5c4d
> Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
> Signed-off-by: Edward Adam Davis <eadavis@qq.com>

Acked-by: Paolo Abeni <pabeni@redhat.com>