[PATCH mptcp-next v7 3/4] mptcp: support MSG_ERRQUEUE on the parent socket

[David Carlier <devnexen@gmail.com>]

Splice pending err skbs from each subflow's error queue onto the parent
msk's error queue at error-report time, so poll() and recvmsg(MSG_ERRQUEUE)
on the parent socket observe TX timestamps and MSG_ZEROCOPY completion
notifications through the standard inet ABI.

The splice filters by SO_EE_ORIGIN: TIMESTAMPING / ZEROCOPY / LOCAL
events forward to the parent because they are tied to user-handed data,
not to a specific path; subflow-level ICMP errors are dropped because
the legacy RECVERR ABI cannot meaningfully convey their per-subflow peer
identity to single-path-aware userspace. Such events will be carried by
a future MPTCP_RECERR channel.

mptcp_recv_error() retries the splice on the pull side: if
sock_queue_err_skb() previously failed under rmem pressure, the skb
stays on the subflow queue, and the next recvmsg(MSG_ERRQUEUE) splices
it once the parent's queue has been drained.

Suggested-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David Carlier <devnexen@gmail.com>
---
 net/mptcp/protocol.c | 92 +++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 86 insertions(+), 6 deletions(-)

diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
index 93e7a42fc65c..53abb8dc2c0f 100644
--- a/net/mptcp/protocol.c
+++ b/net/mptcp/protocol.c
@@ -11,6 +11,7 @@
 #include <linux/netdevice.h>
 #include <linux/sched/signal.h>
 #include <linux/atomic.h>
+#include <linux/errqueue.h>
 #include <net/aligned_data.h>
 #include <net/rps.h>
 #include <net/sock.h>
@@ -815,21 +816,52 @@ static bool __mptcp_ofo_queue(struct mptcp_sock *msk)
 	return moved;
 }
 
+static bool mptcp_errqueue_skb_forwardable(const struct sk_buff *skb)
+{
+	u8 origin = SKB_EXT_ERR(skb)->ee.ee_origin;
+
+	return origin == SO_EE_ORIGIN_TIMESTAMPING ||
+		origin == SO_EE_ORIGIN_ZEROCOPY ||
+		origin == SO_EE_ORIGIN_LOCAL;
+}
+
+static bool __mptcp_subflow_splice_errqueue(struct sock *sk, struct sock *ssk)
+{
+	struct sk_buff *skb;
+	bool moved = false;
+
+	while ((skb = skb_dequeue(&ssk->sk_error_queue))) {
+		if (!mptcp_errqueue_skb_forwardable(skb)) {
+			kfree_skb(skb);  /* path-specific (ICMP) — belongs in MPTCP_RECERR */
+			continue;
+		}
+		if (sock_queue_err_skb(sk, skb)) {
+			skb_queue_head(&ssk->sk_error_queue, skb);
+			break;
+		}
+		moved = true;
+	}
+
+	return moved;
+}
+
 static bool __mptcp_subflow_error_report(struct sock *sk, struct sock *ssk)
 {
 	int ssk_state;
+	bool report;
 	int err;
 
+	report = __mptcp_subflow_splice_errqueue(sk, ssk);
+
 	/* only propagate errors on fallen-back sockets or
 	 * on MPC connect
 	 */
 	if (sk->sk_state != TCP_SYN_SENT && !__mptcp_check_fallback(mptcp_sk(sk)))
-		return false;
+		goto out;
 
 	err = sock_error(ssk);
 	if (!err)
-		return false;
-
+		goto out;
 	/* We need to propagate only transition to CLOSE state.
 	 * Orphaned socket will see such state change via
 	 * subflow_sched_work_if_closed() and that path will properly
@@ -839,6 +871,11 @@ static bool __mptcp_subflow_error_report(struct sock *sk, struct sock *ssk)
 	if (ssk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DEAD))
 		mptcp_set_state(sk, ssk_state);
 	WRITE_ONCE(sk->sk_err, -err);
+	report = true;
+
+out:
+	if (!report)
+		return false;
 
 	/* This barrier is coupled with smp_rmb() in mptcp_poll() */
 	smp_wmb();
@@ -2286,6 +2323,35 @@ static unsigned int mptcp_inq_hint(const struct sock *sk)
 	return 0;
 }
 
+static int mptcp_recv_error(struct sock *sk, struct msghdr *msg, int len)
+{
+	struct mptcp_sock *msk = mptcp_sk(sk);
+	struct mptcp_subflow_context *subflow;
+	bool moved = false;
+	int ret;
+
+	/* Drain the parent first: a previous splice may have failed under
+	 * rmem pressure and the skb stayed on a subflow. Freeing space here
+	 * lets the splice below succeed; sock_queue_err_skb() then re-asserts
+	 * EPOLLERR so userspace knows to drain again on the next poll.
+	 */
+	ret = inet_recv_error(sk, msg, len);
+
+	lock_sock(sk);
+	mptcp_for_each_subflow(msk, subflow) {
+		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
+
+		if (!skb_queue_empty_lockless(&ssk->sk_error_queue))
+			moved |= __mptcp_subflow_splice_errqueue(sk, ssk);
+	}
+	release_sock(sk);
+
+	if (ret == -EAGAIN && moved)
+		ret = inet_recv_error(sk, msg, len);
+
+	return ret;
+}
+
 static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
 			 int flags)
 {
@@ -2295,9 +2361,8 @@ static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
 	int target;
 	long timeo;
 
-	/* MSG_ERRQUEUE is really a no-op till we support IP_RECVERR */
 	if (unlikely(flags & MSG_ERRQUEUE))
-		return inet_recv_error(sk, msg, len);
+		return mptcp_recv_error(sk, msg, len);
 
 	lock_sock(sk);
 	if (unlikely(sk->sk_state == TCP_LISTEN)) {
@@ -4298,6 +4363,19 @@ static __poll_t mptcp_check_writeable(struct mptcp_sock *msk)
 	return 0;
 }
 
+static bool mptcp_subflow_errqueue_pending(const struct mptcp_sock *msk)
+{
+	struct mptcp_subflow_context *subflow;
+
+	mptcp_for_each_subflow(msk, subflow) {
+		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
+
+		if (!skb_queue_empty_lockless(&ssk->sk_error_queue))
+			return true;
+	}
+	return false;
+}
+
 static __poll_t mptcp_poll(struct file *file, struct socket *sock,
 			   struct poll_table_struct *wait)
 {
@@ -4341,7 +4419,9 @@ static __poll_t mptcp_poll(struct file *file, struct socket *sock,
 
 	/* This barrier is coupled with smp_wmb() in __mptcp_error_report() */
 	smp_rmb();
-	if (READ_ONCE(sk->sk_err))
+	if (READ_ONCE(sk->sk_err) ||
+	    !skb_queue_empty_lockless(&sk->sk_error_queue) ||
+	    mptcp_subflow_errqueue_pending(msk))
 		mask |= EPOLLERR;
 
 	return mask;
-- 
2.53.0