tcp: accecn: detect loss ACK w/ AccECN option and add TCP_ACCECN_OPTION_PERSIST

Detect spurious retransmission of a previously sent ACK carrying the AccECN option after the second retransmission. Since this might be caused by the middlebox dropping ACK with options it does not recognize, disable the sending of the AccECN option in all subsequent ACKs. This patch follows Section 3.2.3.2.2 of AccECN spec (RFC9768), and a new field (accecn_opt_sent_w_dsack) is added to indicate that an AccECN option was sent with duplicate SACK info. Also, a new AccECN option sending mode is added to tcp_ecn_option sysctl: (TCP_ECN_OPTION_PERSIST), which ignores the AccECN fallback policy and persistently sends AccECN option once it fits into TCP option space.

Chia-Yu Chang committed 9 months ago

Commit 1247fb11 parent 2ed66126 files · +26 −5Browse files

Unified · Split

6 files

Documentation/networking/ip-sysctl.rst +3 −1
include/linux/tcp.h +2 −1
include/net/tcp_ecn.h +2 −0
net/ipv4/sysctl_net_ipv4.c +1 −1
net/ipv4/tcp_input.c +12 −1
net/ipv4/tcp_output.c +6 −1

Documentation/networking/ip-sysctl.rst+3−1

@@ -482,7 +482,9 @@ tcp_ecn_option - INTEGER · 481 unmodified lines
482	482	1 Send AccECN option sparingly according to the minimum option
483	483	rules outlined in draft-ietf-tcpm-accurate-ecn.
484	484	2 Send AccECN option on every packet whenever it fits into TCP
485		− option space.
	485	+ option space except when AccECN fallback is triggered.
	486	+ 3 Send AccECN option on every packet whenever it fits into TCP
	487	+ option space even when AccECN fallback is triggered.
486	488	= ============================================================
487	489
488	490	Default: 2

include/linux/tcp.h+2−1

@@ -291,7 +291,8 @@ struct tcp_sock { · 290 unmodified lines
291	291	u8 nonagle : 4,/* Disable Nagle algorithm? */
292	292	rate_app_limited:1; /* rate_{delivered,interval_us} limited? */
293	293	u8 received_ce_pending:4, /* Not yet transmit cnt of received_ce */
294		− unused2:4;
	294	+ accecn_opt_sent_w_dsack:1,/* Sent ACCECN opt in previous ACK w/ D-SACK */
	295	+ unused2:3;
295	296	u8 accecn_minlen:2,/* Minimum length of AccECN option sent */
296	297	est_ecnfield:2,/* ECN field for AccECN delivered estimates */
297	298	accecn_opt_demand:2,/* Demand AccECN option for n next ACKs */

include/net/tcp_ecn.h+2−0

@@ -29,6 +29,7 @@ enum tcp_accecn_option { · 28 unmodified lines
29	29	TCP_ACCECN_OPTION_DISABLED = 0,
30	30	TCP_ACCECN_OPTION_MINIMUM = 1,
31	31	TCP_ACCECN_OPTION_FULL = 2,
	32	+ TCP_ACCECN_OPTION_PERSIST = 3,
32	33	};
33	34
34	35	/* Apply either ECT(0) or ECT(1) based on TCP_CONG_ECT_1_NEGOTIATION flag */
@@ -406,6 +407,7 @@ static inline void tcp_accecn_init_counters(struct tcp_sock *tp) · 371 unmodified lines
406	407	tp->received_ce_pending = 0;
407	408	__tcp_accecn_init_bytes_counters(tp->received_ecn_bytes);
408	409	__tcp_accecn_init_bytes_counters(tp->delivered_ecn_bytes);
	410	+ tp->accecn_opt_sent_w_dsack = 0;
409	411	tp->accecn_minlen = 0;
410	412	tp->accecn_opt_demand = 0;
411	413	tp->est_ecnfield = 0;

net/ipv4/sysctl_net_ipv4.c+1−1

@@ -749,7 +749,7 @@ static struct ctl_table ipv4_net_table[] = { · 748 unmodified lines
749	749	.mode = 0644,
750	750	.proc_handler = proc_dou8vec_minmax,
751	751	.extra1 = SYSCTL_ZERO,
752		− .extra2 = SYSCTL_TWO,
	752	+ .extra2 = SYSCTL_THREE,
753	753	},
754	754	{
755	755	.procname = "tcp_ecn_option_beacon",

net/ipv4/tcp_input.c+12−1

@@ -5046,8 +5046,11 @@ static void tcp_dsack_extend(struct sock *sk, u32 seq, u32 end_seq) · 5045 unmodified lines
5046	5046	tcp_sack_extend(tp->duplicate_sack, seq, end_seq);
5047	5047	}
5048	5048
5049		−static void tcp_rcv_spurious_retrans(struct sock sk, const struct sk_buff skb)
	5049	+static void tcp_rcv_spurious_retrans(struct sock *sk,
	5050	+ const struct sk_buff *skb)
5050	5051	{
	5052	+ struct tcp_sock *tp = tcp_sk(sk);
	5053	+
5051	5054	/* When the ACK path fails or drops most ACKs, the sender would
5052	5055	* timeout and spuriously retransmit the same segment repeatedly.
5053	5056	* If it seems our ACKs are not reaching the other side,
@@ -5067,6 +5070,14 @@ static void tcp_rcv_spurious_retrans(struct sock sk, const struct sk_buff skb) · 13 unmodified lines
5067	5070	/* Save last flowlabel after a spurious retrans. */
5068	5071	tcp_save_lrcv_flowlabel(sk, skb);
5069	5072	#endif
	5073	+ /* Check DSACK info to detect that the previous ACK carrying the
	5074	+ * AccECN option was lost after the second retransmision, and then
	5075	+ * stop sending AccECN option in all subsequent ACKs.
	5076	+ */
	5077	+ if (tcp_ecn_mode_accecn(tp) &&
	5078	+ tp->accecn_opt_sent_w_dsack &&
	5079	+ TCP_SKB_CB(skb)->seq == tp->duplicate_sack[0].start_seq)
	5080	+ tcp_accecn_fail_mode_set(tp, TCP_ACCECN_OPT_FAIL_SEND);
5070	5081	}
5071	5082
5072	5083	static void tcp_send_dupack(struct sock sk, const struct sk_buff skb)

net/ipv4/tcp_output.c+6−1

@@ -715,9 +715,12 @@ static void tcp_options_write(struct tcphdr th, struct tcp_sock tp, · 714 unmodified lines
715	715	if (tp) {
716	716	tp->accecn_minlen = 0;
717	717	tp->accecn_opt_tstamp = tp->tcp_mstamp;
	718	+ tp->accecn_opt_sent_w_dsack = tp->rx_opt.dsack;
718	719	if (tp->accecn_opt_demand)
719	720	tp->accecn_opt_demand--;
720	721	}
	722	+ } else if (tp) {
	723	+ tp->accecn_opt_sent_w_dsack = 0;
721	724	}
722	725
723	726	if (unlikely(OPTION_SACK_ADVERTISE & options)) {
@@ -1189,7 +1192,9 @@ static unsigned int tcp_established_options(struct sock sk, struct sk_buff skb · 465 unmodified lines
1189	1192	if (tcp_ecn_mode_accecn(tp)) {
1190	1193	int ecn_opt = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_ecn_option);
1191	1194
1192		− if (ecn_opt && tp->saw_accecn_opt && !tcp_accecn_opt_fail_send(tp) &&
	1195	+ if (ecn_opt && tp->saw_accecn_opt &&
	1196	+ (ecn_opt >= TCP_ACCECN_OPTION_PERSIST \|\|
	1197	+ !tcp_accecn_opt_fail_send(tp)) &&
1193	1198	(ecn_opt >= TCP_ACCECN_OPTION_FULL \|\| tp->accecn_opt_demand \|\|
1194	1199	tcp_accecn_option_beacon_check(sk))) {
1195	1200	opts->use_synack_ecn_bytes = 0;