in-kernel PM: re-establish subflows after "network" errors

[matttbe]

In case of network issue (blackhole, wrong path, etc.), it might be interesting to re-established subflows that got closed after a TCP-level error (timeout, RST, etc.) as it was maybe a temp issue or a change in the network (e.g. NAT dropping connections a bit too quickly)

[arinc9]

On top of that, it'd be great if newly added subflows were established on existing MPTCP connections. For example, a cellular network interface that was described as a subflow path may go down and up. In that case, even though mptcpd adds it back as a subflow, that subflow won't be re-established on the existing MPTCP connection. Unless I'm wrong?

[matttbe]

For example, a cellular network interface that was described as a subflow path may go down and up. In that case, even though mptcpd adds it back as a subflow, that subflow won't be re-established on the existing MPTCP connection. Unless I'm wrong?

If the corresponding endpoint is removed and re-added, the subflow should be re-created. If not, that's a bug.

[matttbe]

For the initial issue, such patch can be used as a starting point:

diff --git a/net/mptcp/pm.c b/net/mptcp/pm.c
index 98b0b31e3b8d..c64fa3fe4779 100644
--- a/net/mptcp/pm.c
+++ b/net/mptcp/pm.c
@@ -202,6 +202,8 @@ void mptcp_pm_subflow_check_next(struct mptcp_sock *msk,
 	if (!READ_ONCE(pm->work_pending) && !update_subflows)
 		return;
 
+	mptcp_pm_nl_close_subflow(msk, subflow); /* or before the return just above? */
+
 	spin_lock_bh(&pm->lock);
 	if (update_subflows)
 		__mptcp_pm_close_subflow(msk);
diff --git a/net/mptcp/pm_netlink.c b/net/mptcp/pm_netlink.c
index d353cb007090..5a7ccefdb36a 100644
--- a/net/mptcp/pm_netlink.c
+++ b/net/mptcp/pm_netlink.c
@@ -782,6 +787,15 @@ static bool mptcp_local_id_match(const struct mptcp_sock *msk, u8 local_id, u8 i
 	return local_id == id || (!local_id && msk->mpc_endpoint_id == id);
 }
 
+void mptcp_pm_nl_close_subflow(struct mptcp_sock *msk,
+			       const struct mptcp_subflow_context *rm_subflow)
+{
+	struct sock *rm_ssk = mptcp_subflow_tcp_sock(rm_subflow);
+
+	if (rm_ssk->sk_err == TODO || rm_subflow->reset_reason == TODO)
+		/* TODO: recreate subflow */
+}
+
 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk,
 					   const struct mptcp_rm_list *rm_list,
 					   enum linux_mptcp_mib_field rm_type)
diff --git a/net/mptcp/protocol.h b/net/mptcp/protocol.h
index cee0a8098b41..b7ed9a2db7ea 100644
--- a/net/mptcp/protocol.h
+++ b/net/mptcp/protocol.h
@@ -1145,6 +1145,9 @@ unsigned int mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock *msk);
 unsigned int mptcp_pm_get_subflows_max(const struct mptcp_sock *msk);
 unsigned int mptcp_pm_get_local_addr_max(const struct mptcp_sock *msk);
 
+void mptcp_pm_nl_close_subflow(struct mptcp_sock *msk,
+			       const struct mptcp_subflow_context *rm_subflow);
+
 /* called under PM lock */
 static inline void __mptcp_pm_close_subflow(struct mptcp_sock *msk)
 {