net_sched: sch_fq: handle non connected flows

FQ packet scheduler assumed that packets could be classified
based on their owning socket.

This means that if a UDP server uses one UDP socket to send
packets to different destinations, packets all land
in one FQ flow.

This is unfair, since each TCP flow has a unique bucket, meaning
that in case of pressure (fully utilised uplink), TCP flows
have more share of the bandwidth.

If we instead detect unconnected sockets, we can use a stochastic
hash based on the 4-tuple hash.

This also means a QUIC server using one UDP socket will properly
spread the outgoing packets to different buckets, and in-kernel
pacing based on EDT model will no longer risk having big rb-tree on
one flow.

Note that UDP application might provide the skb->hash in an
ancillary message at sendmsg() time to avoid the cost of a dissection
in fq packet scheduler.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/net/sched/sch_fq.c b/net/sched/sch_fq.c
index ee138365ec45ee01cb10f149ae5b1d7635fa1185..26a94e5cd5dfae34109649b04a1ebcaafa0f545b 100644 (file)
--- a/net/sched/sch_fq.c
+++ b/net/sched/sch_fq.c
@@ -270,6 +270,17 @@ static struct fq_flow *fq_classify(struct sk_buff *skb, struct fq_sched_data *q)
                 */
                sk = (struct sock *)((hash << 1) | 1UL);
                skb_orphan(skb);
+       } else if (sk->sk_state == TCP_CLOSE) {
+               unsigned long hash = skb_get_hash(skb) & q->orphan_mask;
+               /*
+                * Sockets in TCP_CLOSE are non connected.
+                * Typical use case is UDP sockets, they can send packets
+                * with sendto() to many different destinations.
+                * We probably could use a generic bit advertising
+                * non connected sockets, instead of sk_state == TCP_CLOSE,
+                * if we care enough.
+                */
+               sk = (struct sock *)((hash << 1) | 1UL);
        }
 
        root = &q->fq_root[hash_ptr(sk, q->fq_trees_log)];
@@ -290,7 +301,7 @@ static struct fq_flow *fq_classify(struct sk_buff *skb, struct fq_sched_data *q)
                         * It not, we need to refill credit with
                         * initial quantum
                         */
-                       if (unlikely(skb->sk &&
+                       if (unlikely(skb->sk == sk &&
                                     f->socket_hash != sk->sk_hash)) {
                                f->credit = q->initial_quantum;
                                f->socket_hash = sk->sk_hash;
@@ -315,7 +326,7 @@ static struct fq_flow *fq_classify(struct sk_buff *skb, struct fq_sched_data *q)
 
        fq_flow_set_detached(f);
        f->sk = sk;
-       if (skb->sk)
+       if (skb->sk == sk)
                f->socket_hash = sk->sk_hash;
        f->credit = q->initial_quantum;