/* Marks a packet lost, if some packet sent later has been (s)acked.
* The underlying idea is similar to the traditional dupthresh and FACK
* but they look at different metrics:
*
* dupthresh: 3 OOO packets delivered (packet count)
* FACK: sequence delta to highest sacked sequence (sequence space)
* RACK: sent time delta to the latest delivered packet (time domain)
*
* The advantage of RACK is it applies to both original and retransmitted
* packet and therefore is robust against tail losses. Another advantage
* is being more resilient to reordering by simply allowing some
* "settling delay", instead of tweaking the dupthresh.
*
* The current version is only used after recovery starts but can be
* easily extended to detect the first loss.
*/
int tcp_rack_mark_lost(struct sock *sk)
{
struct tcp_sock *tp = tcp_sk(sk);
struct sk_buff *skb;
u32 reo_wnd, prior_retrans = tp->retrans_out;
if (inet_csk(sk)->icsk_ca_state < TCP_CA_Recovery || !tp->rack.advanced)
return 0;
/* Reset the advanced flag to avoid unnecessary queue scanning */
tp->rack.advanced = 0;
/* To be more reordering resilient, allow min_rtt/4 settling delay
* (lower-bounded to 1000uS). We use min_rtt instead of the smoothed
* RTT because reordering is often a path property and less related
* to queuing or delayed ACKs.
*
* TODO: measure and adapt to the observed reordering delay, and
* use a timer to retransmit like the delayed early retransmit.
*/
reo_wnd = 1000;
if (tp->rack.reord && tcp_min_rtt(tp) != ~0U)
reo_wnd = max(tcp_min_rtt(tp) >> 2, reo_wnd);
tcp_for_write_queue(skb, sk) {
struct tcp_skb_cb *scb = TCP_SKB_CB(skb);
if (skb == tcp_send_head(sk))
break;
/* Skip ones already (s)acked */
if (!after(scb->end_seq, tp->snd_una) ||
scb->sacked & TCPCB_SACKED_ACKED)
continue;
if (skb_mstamp_after(&tp->rack.mstamp, &skb->skb_mstamp)) {
if (skb_mstamp_us_delta(&tp->rack.mstamp,
&skb->skb_mstamp) <= reo_wnd)
continue;
/* skb is lost if packet sent later is sacked */
tcp_skb_mark_lost_uncond_verify(tp, skb);
if (scb->sacked & TCPCB_SACKED_RETRANS) {
scb->sacked &= ~TCPCB_SACKED_RETRANS;
tp->retrans_out -= tcp_skb_pcount(skb);
NET_INC_STATS_BH(sock_net(sk),
LINUX_MIB_TCPLOSTRETRANSMIT);
}
} else if (!(scb->sacked & TCPCB_RETRANS)) {
/* Original data are sent sequentially so stop early
* b/c the rest are all sent after rack_sent
*/
break;
}
}
return prior_retrans - tp->retrans_out;
}
/**
* Cleans the meta-socket retransmission queue and the reinject-queue.
* @sk must be the metasocket.
*/
static void mptcp_clean_rtx_queue(struct sock *meta_sk, u32 prior_snd_una)
{
struct sk_buff *skb, *tmp;
struct tcp_sock *meta_tp = tcp_sk(meta_sk);
struct mptcp_cb *mpcb = meta_tp->mpcb;
int acked = 0;
while ((skb = tcp_write_queue_head(meta_sk)) &&
skb != tcp_send_head(meta_sk)) {
if (before(meta_tp->snd_una, TCP_SKB_CB(skb)->end_seq))
break;
tcp_unlink_write_queue(skb, meta_sk);
if (mptcp_is_data_fin(skb)) {
struct sock *sk_it;
/* DATA_FIN has been acknowledged - now we can close
* the subflows
*/
mptcp_for_each_sk(mpcb, sk_it) {
unsigned long delay = 0;
/* If we are the passive closer, don't trigger
* subflow-fin until the subflow has been finned
* by the peer - thus we add a delay.
*/
if (mpcb->passive_close &&
sk_it->sk_state == TCP_ESTABLISHED)
delay = inet_csk(sk_it)->icsk_rto << 3;
mptcp_sub_close(sk_it, delay);
}
}
meta_tp->packets_out -= tcp_skb_pcount(skb);
sk_wmem_free_skb(meta_sk, skb);
acked = 1;
}
