That's the behaviour with the default packet scheduler.

In some early design, the default scheduler was supposed to take into
account only the received backup flags, but it ended up not being the
case, and setting the flag would also affect outgoing data.

Suggested-by: Mat Martineau <martineau@kernel.org>
Acked-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
---
 man/man8/ip-mptcp.8 | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

diff --git a/man/man8/ip-mptcp.8 b/man/man8/ip-mptcp.8
index 11df43ce..f3d09bab 100644
--- a/man/man8/ip-mptcp.8
+++ b/man/man8/ip-mptcp.8
@@ -170,10 +170,12 @@ typically do this.
 If this is a
 .BR subflow
 endpoint, the subflows created using this endpoint will have the backup
-flag set during the connection process. This flag instructs the peer to
-only send data on a given subflow when all non-backup subflows are
-unavailable. This does not affect outgoing data, where subflow priority
-is determined by the backup/non-backup flag received from the peer
+flag set during the connection process. This flag instructs the remote
+peer to only send data on a given subflow when all non-backup subflows
+are unavailable. When using the default packet scheduler with a 'backup'
+endpoint, outgoing data from the local peer is also affected: packets
+will only be sent from this endpoint when all non-backup subflows are
+unavailable.
 
 .TP
 .BR fullmesh

-- 
2.45.2