kcm: Fix error handling for SOCK_DGRAM in kcm_sendmsg().

[ Upstream commit a22730b ]

syzkaller found a memory leak in kcm_sendmsg(), and commit c821a88
("kcm: Fix memory leak in error path of kcm_sendmsg()") suppressed it by
updating kcm_tx_msg(head)->last_skb if partial data is copied so that the
following sendmsg() will resume from the skb.

However, we cannot know how many bytes were copied when we get the error.
Thus, we could mess up the MSG_MORE queue.

When kcm_sendmsg() fails for SOCK_DGRAM, we should purge the queue as we
do so for UDP by udp_flush_pending_frames().

Even without this change, when the error occurred, the following sendmsg()
resumed from a wrong skb and the queue was messed up.  However, we have
yet to get such a report, and only syzkaller stumbled on it.  So, this
can be changed safely.

Note this does not change SOCK_SEQPACKET behaviour.

Fixes: c821a88 ("kcm: Fix memory leak in error path of kcm_sendmsg()")
Fixes: ab7ac4e ("kcm: Kernel Connection Multiplexor module")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://lore.kernel.org/r/20230912022753.33327-1-kuniyu@amazon.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

net/kcm/kcmsock.c

@@ -1065,17 +1065,18 @@ static int kcm_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
 out_error:
 	kcm_push(kcm);
 
-	if (copied && sock->type == SOCK_SEQPACKET) {
+	if (sock->type == SOCK_SEQPACKET) {
 		/* Wrote some bytes before encountering an
 		 * error, return partial success.
 		 */
-		goto partial_message;
-	}
-
-	if (head != kcm->seq_skb)
+		if (copied)
+			goto partial_message;
+		if (head != kcm->seq_skb)
+			kfree_skb(head);
+	} else {
 		kfree_skb(head);
-	else if (copied)
-		kcm_tx_msg(head)->last_skb = skb;
+		kcm->seq_skb = NULL;
+	}
 
 	err = sk_stream_error(sk, msg->msg_flags, err);