From: Gang Yan <yangang@kylinos.cn>

Extend MPTCP's sendmsg handling to recognize and honor the MSG_EOR flag,
which marks the end of a record for application-level message boundaries.

Data fragments tagged with MSG_EOR are explicitly marked in the
mptcp_data_frag structure and skb context to prevent unintended
coalescing with subsequent data chunks. This ensures the intent of
applications using MSG_EOR is preserved across MPTCP subflows,
maintaining consistent message segmentation behavior.

Signed-off-by: Gang Yan <yangang@kylinos.cn>
---
 net/mptcp/protocol.c | 24 +++++++++++++++++++++---
 net/mptcp/protocol.h |  3 ++-
 2 files changed, 23 insertions(+), 4 deletions(-)

diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
index 01690a84ea6d..dafa178f43c5 100644
--- a/net/mptcp/protocol.c
+++ b/net/mptcp/protocol.c
@@ -1032,7 +1032,8 @@ static bool mptcp_frag_can_collapse_to(const struct mptcp_sock *msk,
 				       const struct page_frag *pfrag,
 				       const struct mptcp_data_frag *df)
 {
-	return df && pfrag->page == df->page &&
+	return df && !df->eor &&
+		pfrag->page == df->page &&
 		pfrag->size - pfrag->offset > 0 &&
 		pfrag->offset == (df->offset + df->data_len) &&
 		df->data_seq + df->data_len == msk->write_seq;
@@ -1174,6 +1175,7 @@ mptcp_carve_data_frag(const struct mptcp_sock *msk, struct page_frag *pfrag,
 	dfrag->offset = offset + sizeof(struct mptcp_data_frag);
 	dfrag->already_sent = 0;
 	dfrag->page = pfrag->page;
+	dfrag->eor = 0;
 
 	return dfrag;
 }
@@ -1435,6 +1437,13 @@ static int mptcp_sendmsg_frag(struct sock *sk, struct sock *ssk,
 		mptcp_update_infinite_map(msk, ssk, mpext);
 	trace_mptcp_sendmsg_frag(mpext);
 	mptcp_subflow_ctx(ssk)->rel_write_seq += copy;
+
+	/* if this is the last chunk of a dfrag with MSG_EOR set,
+	 * mark the skb to prevent coalescing with subsequent data.
+	 */
+	if (dfrag->eor && info->sent + copy >= dfrag->data_len)
+		TCP_SKB_CB(skb)->eor = 1;
+
 	return copy;
 }
 
@@ -1895,7 +1904,8 @@ static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
 	long timeo;
 
 	/* silently ignore everything else */
-	msg->msg_flags &= MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | MSG_FASTOPEN;
+	msg->msg_flags &= MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
+			  MSG_FASTOPEN | MSG_EOR;
 
 	lock_sock(sk);
 
@@ -2002,8 +2012,16 @@ static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
 			goto do_error;
 	}
 
-	if (copied)
+	if (copied) {
+		/* mark the last dfrag with EOR if MSG_EOR was set */
+		if (msg->msg_flags & MSG_EOR) {
+			struct mptcp_data_frag *dfrag = mptcp_pending_tail(sk);
+
+			if (dfrag)
+				dfrag->eor = 1;
+		}
 		__mptcp_push_pending(sk, msg->msg_flags);
+	}
 
 out:
 	release_sock(sk);
diff --git a/net/mptcp/protocol.h b/net/mptcp/protocol.h
index afdead91a4d7..db96f2945cbd 100644
--- a/net/mptcp/protocol.h
+++ b/net/mptcp/protocol.h
@@ -265,7 +265,8 @@ struct mptcp_data_frag {
 	u16 data_len;
 	u16 offset;
 	u8 overhead;
-	u8 __unused;
+	u8 eor:1,
+	   __unused:7;
 	u16 already_sent;
 	struct page *page;
 };
-- 
2.43.0

On Tue, 31 Mar 2026 17:08:09 +0800, Gang Yan <gang.yan@linux.dev> wrote:
> diff --git a/net/mptcp/protocol.h b/net/mptcp/protocol.h
> index afdead91a4d7..db96f2945cbd 100644
> --- a/net/mptcp/protocol.h
> +++ b/net/mptcp/protocol.h
> @@ -265,7 +265,8 @@ struct mptcp_data_frag {
>  	u16 data_len;
>  	u16 offset;
>  	u8 overhead;
> -	u8 __unused;
> +	u8 eor:1,
> +	   __unused:7;

Here, we could also use the whole u8 than using only one bit which is a
bit more costly to read/write, and we would avoid any KMSAN warnings (if
any).

I can also do this modification when applying the patch if that's OK:

  u8 eor;	/* currently using 1 bit */

WDYT?

Reviewed-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>

-- 
Matthieu Baerts (NGI0) <matttbe@kernel.org>