sock_orphan() call to sk_set_socket() implies clearing the sock TX queue.
This might cause unexpected out-of-order transmit, as outstanding packets
can pick a different TX queue and bypass the ones already queued.
This is undesired in general. More specifically, it breaks the in-order
scheduling property guarantee for device-offloaded TLS sockets.
Introduce a function variation __sk_set_socket() that does not clear
the TX queue, and call it from sock_orphan().
All other callers of sk_set_socket() do not operate on an active socket,
so they do not need this change.
Fixes: e022f0b4a03f ("net: Introduce sk_tx_queue_mapping")
Signed-off-by: Tariq Toukan <tar...@mellanox.com>
Reviewed-by: Boris Pismenny <bor...@mellanox.com>
---
include/net/sock.h | 9 +++++++--
1 file changed, 7 insertions(+), 2 deletions(-)
Please queue for -stable.
diff --git a/include/net/sock.h b/include/net/sock.h
index c53cc42b5ab9..23e43f3d79f0 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -1846,10 +1846,15 @@ static inline int sk_rx_queue_get(const struct sock *sk)
}
#endif
+static inline void __sk_set_socket(struct sock *sk, struct socket *sock)
+{
+ sk->sk_socket = sock;
+}
+
static inline void sk_set_socket(struct sock *sk, struct socket *sock)
{
sk_tx_queue_clear(sk);
- sk->sk_socket = sock;
+ __sk_set_socket(sk, sock);
}
static inline wait_queue_head_t *sk_sleep(struct sock *sk)
@@ -1868,7 +1873,7 @@ static inline void sock_orphan(struct sock *sk)
{
write_lock_bh(&sk->sk_callback_lock);
sock_set_flag(sk, SOCK_DEAD);
- sk_set_socket(sk, NULL);
+ __sk_set_socket(sk, NULL);
sk->sk_wq = NULL;
write_unlock_bh(&sk->sk_callback_lock);
}
--
1.8.3.1
