From: Ilpo Järvinen ij@kernel.org
This change implements Accurate ECN without negotiation and AccECN Option (that will be added by later changes). Based on AccECN specifications: https://tools.ietf.org/id/draft-ietf-tcpm-accurate-ecn-28.txt
Accurate ECN allows feeding back the number of CE (congestion experienced) marks accurately to the sender in contrast to RFC3168 ECN that can only signal one marks-seen-yes/no per RTT. Congestion control algorithms can take advantage of the accurate ECN information to fine-tune their congestion response to avoid drastic rate reduction when only mild congestion is encountered.
With Accurate ECN, tp->received_ce (r.cep in AccECN spec) keeps track of how many segments have arrived with a CE mark. Accurate ECN uses ACE field (ECE, CWR, AE) to communicate the value back to the sender which updates tp->delivered_ce (s.cep) based on the feedback. This signalling channel is lossy when ACE field overflow occurs.
Conservative strategy is selected here to deal with the ACE overflow, however, some strategies using the AccECN option later in the overall patchset mitigate against false overflows detected.
The ACE field values on the wire are offset by TCP_ACCECN_CEP_INIT_OFFSET. Delivered_ce/received_ce count the real CE marks rather than forcing all downstream users to adapt to the wire offset.
This patch uses the first 1-byte hole and the last 4-byte hole of the tcp_sock_write_txrx for 'received_ce_pending' and 'received_ce'. Also, the group size of tcp_sock_write_txrx is increased from 91 + 4 to 95 + 4 due to the new u32 received_ce member. Below are the trimmed pahole outcomes before and after this patch.
[BEFORE THIS PATCH] struct tcp_sock { [...] __cacheline_group_begin__tcp_sock_write_txrx[0]; /* 2521 0 */ u8 nonagle:4; /* 2521: 0 1 */ u8 rate_app_limited:1; /* 2521: 4 1 */ /* XXX 3 bits hole, try to pack */ /* XXX 2 bytes hole, try to pack */
[...] u32 delivered_ce; /* 2576 4 */ u32 app_limited; /* 2580 4 */ u32 rcv_wnd; /* 2684 4 */ struct tcp_options_received rx_opt; /* 2688 24 */ __cacheline_group_end__tcp_sock_write_txrx[0]; /* 2612 0 */ /* XXX 4 bytes hole, try to pack */
[...] /* size: 3200, cachelines: 50, members: 161 */ }
[AFTER THIS PATCH] struct tcp_sock { [...] __cacheline_group_begin__tcp_sock_write_txrx[0]; /* 2521 0 */ u8 nonagle:4; /* 2521: 0 1 */ u8 rate_app_limited:1; /* 2521: 4 1 */ /* XXX 3 bits hole, try to pack */
/* Force alignment to the next boundary: */ u8 :0; u8 received_ce_pending:4;/* 2522: 0 1 */ u8 unused2:4; /* 2522: 4 1 */ /* XXX 1 byte hole, try to pack */
[...] u32 delivered_ce; /* 2576 4 */ u32 received_ce; /* 2580 4 */ u32 app_limited; /* 2584 4 */ u32 rcv_wnd; /* 2588 4 */ struct tcp_options_received rx_opt; /* 2592 24 */ __cacheline_group_end__tcp_sock_write_txrx[0]; /* 2616 0 */
[...] /* size: 3200, cachelines: 50, members: 164 */ }
Signed-off-by: Ilpo Järvinen ij@kernel.org Co-developed-by: Olivier Tilmans olivier.tilmans@nokia.com Signed-off-by: Olivier Tilmans olivier.tilmans@nokia.com Co-developed-by: Chia-Yu Chang chia-yu.chang@nokia-bell-labs.com Signed-off-by: Chia-Yu Chang chia-yu.chang@nokia-bell-labs.com Reviewed-by: Eric Dumazet edumazet@google.com
--- v9: - Use tcp_data_ecn_check() to set TCP_ECN_SEE flag only for RFC3168 ECN - Add comments about setting TCP_ECN_SEEN flag for RFC3168 and Accruate ECN --- .../networking/net_cachelines/tcp_sock.rst | 2 + include/linux/tcp.h | 3 + include/net/tcp.h | 15 +++ include/net/tcp_ecn.h | 53 +++++++++- net/ipv4/tcp.c | 5 +- net/ipv4/tcp_input.c | 100 ++++++++++++++++-- net/ipv4/tcp_output.c | 9 +- 7 files changed, 175 insertions(+), 12 deletions(-)
diff --git a/Documentation/networking/net_cachelines/tcp_sock.rst b/Documentation/networking/net_cachelines/tcp_sock.rst index 7bbda5944ee2..31313a9adccc 100644 --- a/Documentation/networking/net_cachelines/tcp_sock.rst +++ b/Documentation/networking/net_cachelines/tcp_sock.rst @@ -101,6 +101,8 @@ u32 prr_delivered u32 prr_out read_mostly read_mostly tcp_rate_skb_sent,tcp_newly_delivered(tx);tcp_ack,tcp_rate_gen,tcp_clean_rtx_queue(rx) u32 delivered read_mostly read_write tcp_rate_skb_sent, tcp_newly_delivered(tx);tcp_ack, tcp_rate_gen, tcp_clean_rtx_queue (rx) u32 delivered_ce read_mostly read_write tcp_rate_skb_sent(tx);tcp_rate_gen(rx) +u32 received_ce read_mostly read_write +u8:4 received_ce_pending read_mostly read_write u32 lost read_mostly tcp_ack u32 app_limited read_write read_mostly tcp_rate_check_app_limited,tcp_rate_skb_sent(tx);tcp_rate_gen(rx) u64 first_tx_mstamp read_write tcp_rate_skb_sent diff --git a/include/linux/tcp.h b/include/linux/tcp.h index d103cc0e7a35..90cee6e53527 100644 --- a/include/linux/tcp.h +++ b/include/linux/tcp.h @@ -287,6 +287,8 @@ struct tcp_sock { */ u8 nonagle : 4,/* Disable Nagle algorithm? */ rate_app_limited:1; /* rate_{delivered,interval_us} limited? */ + u8 received_ce_pending:4, /* Not yet transmit cnt of received_ce */ + unused2:4; __be32 pred_flags; u64 tcp_clock_cache; /* cache last tcp_clock_ns() (see tcp_mstamp_refresh()) */ u64 tcp_mstamp; /* most recent packet received/sent */ @@ -299,6 +301,7 @@ struct tcp_sock { u32 snd_up; /* Urgent pointer */ u32 delivered; /* Total data packets delivered incl. rexmits */ u32 delivered_ce; /* Like the above but only ECE marked packets */ + u32 received_ce; /* Like the above but for rcvd CE marked pkts */ u32 app_limited; /* limited until "delivered" reaches this val */ u32 rcv_wnd; /* Current receiver window */ /* diff --git a/include/net/tcp.h b/include/net/tcp.h index 905fd63be74d..3a25c8f0b99e 100644 --- a/include/net/tcp.h +++ b/include/net/tcp.h @@ -973,6 +973,14 @@ static inline u32 tcp_rsk_tsval(const struct tcp_request_sock *treq) #define TCPHDR_ACE (TCPHDR_ECE | TCPHDR_CWR | TCPHDR_AE) #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
+#define TCP_ACCECN_CEP_ACE_MASK 0x7 +#define TCP_ACCECN_ACE_MAX_DELTA 6 + +/* To avoid/detect middlebox interference, not all counters start at 0. + * See draft-ietf-tcpm-accurate-ecn for the latest values. + */ +#define TCP_ACCECN_CEP_INIT_OFFSET 5 + /* State flags for sacked in struct tcp_skb_cb */ enum tcp_skb_cb_sacked_flags { TCPCB_SACKED_ACKED = (1 << 0), /* SKB ACK'd by a SACK block */ @@ -1782,11 +1790,18 @@ static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) { + u32 ace; + /* mptcp hooks are only on the slow path */ if (sk_is_mptcp((struct sock *)tp)) return;
+ ace = tcp_ecn_mode_accecn(tp) ? + ((tp->delivered_ce + TCP_ACCECN_CEP_INIT_OFFSET) & + TCP_ACCECN_CEP_ACE_MASK) : 0; + tp->pred_flags = htonl((tp->tcp_header_len << 26) | + (ace << 22) | ntohl(TCP_FLAG_ACK) | snd_wnd); } diff --git a/include/net/tcp_ecn.h b/include/net/tcp_ecn.h index b3430557676b..b0ed89dbad41 100644 --- a/include/net/tcp_ecn.h +++ b/include/net/tcp_ecn.h @@ -12,6 +12,7 @@
static inline void tcp_ecn_queue_cwr(struct tcp_sock *tp) { + /* Do not set CWR if in AccECN mode! */ if (tcp_ecn_mode_rfc3168(tp)) tp->ecn_flags |= TCP_ECN_QUEUE_CWR; } @@ -19,8 +20,10 @@ static inline void tcp_ecn_queue_cwr(struct tcp_sock *tp) static inline void tcp_ecn_accept_cwr(struct sock *sk, const struct sk_buff *skb) { - if (tcp_hdr(skb)->cwr) { - tcp_sk(sk)->ecn_flags &= ~TCP_ECN_DEMAND_CWR; + struct tcp_sock *tp = tcp_sk(sk); + + if (tcp_ecn_mode_rfc3168(tp) && tcp_hdr(skb)->cwr) { + tp->ecn_flags &= ~TCP_ECN_DEMAND_CWR;
/* If the sender is telling us it has entered CWR, then its * cwnd may be very low (even just 1 packet), so we should ACK @@ -36,6 +39,52 @@ static inline void tcp_ecn_withdraw_cwr(struct tcp_sock *tp) tp->ecn_flags &= ~TCP_ECN_QUEUE_CWR; }
+static inline u8 tcp_accecn_ace(const struct tcphdr *th) +{ + return (th->ae << 2) | (th->cwr << 1) | th->ece; +} + +static inline void tcp_accecn_init_counters(struct tcp_sock *tp) +{ + tp->received_ce = 0; + tp->received_ce_pending = 0; +} + +/* Updates Accurate ECN received counters from the received IP ECN field */ +static inline void tcp_ecn_received_counters(struct sock *sk, const struct sk_buff *skb) +{ + u8 ecnfield = TCP_SKB_CB(skb)->ip_dsfield & INET_ECN_MASK; + u8 is_ce = INET_ECN_is_ce(ecnfield); + struct tcp_sock *tp = tcp_sk(sk); + + if (!INET_ECN_is_not_ect(ecnfield)) { + u32 pcount = is_ce * max_t(u16, 1, skb_shinfo(skb)->gso_segs); + + /* As for accurate ECN, the TCP_ECN_SEEN flag is set by + * tcp_ecn_received_counters() when the ECN codepoint of + * received TCP data or ACK contains ECT(0), ECT(1), or CE. + */ + if (!tcp_ecn_mode_rfc3168(tp)) + tp->ecn_flags |= TCP_ECN_SEEN; + + /* ACE counter tracks *all* segments including pure ACKs */ + tp->received_ce += pcount; + tp->received_ce_pending = min(tp->received_ce_pending + pcount, + 0xfU); + } +} + +static inline void tcp_accecn_set_ace(struct tcphdr *th, struct tcp_sock *tp) +{ + u32 wire_ace; + + wire_ace = tp->received_ce + TCP_ACCECN_CEP_INIT_OFFSET; + th->ece = !!(wire_ace & 0x1); + th->cwr = !!(wire_ace & 0x2); + th->ae = !!(wire_ace & 0x4); + tp->received_ce_pending = 0; +} + static inline void tcp_ecn_rcv_synack(struct tcp_sock *tp, const struct tcphdr *th) { diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c index 897cd28d2d8c..56b957efdee1 100644 --- a/net/ipv4/tcp.c +++ b/net/ipv4/tcp.c @@ -271,6 +271,7 @@ #include <net/icmp.h> #include <net/inet_common.h> #include <net/tcp.h> +#include <net/tcp_ecn.h> #include <net/mptcp.h> #include <net/proto_memory.h> #include <net/xfrm.h> @@ -3390,6 +3391,7 @@ int tcp_disconnect(struct sock *sk, int flags) tp->window_clamp = 0; tp->delivered = 0; tp->delivered_ce = 0; + tcp_accecn_init_counters(tp); if (icsk->icsk_ca_initialized && icsk->icsk_ca_ops->release) icsk->icsk_ca_ops->release(sk); memset(icsk->icsk_ca_priv, 0, sizeof(icsk->icsk_ca_priv)); @@ -5122,6 +5124,7 @@ static void __init tcp_struct_check(void) CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_txrx, snd_up); CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_txrx, delivered); CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_txrx, delivered_ce); + CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_txrx, received_ce); CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_txrx, app_limited); CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_txrx, rcv_wnd); CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_txrx, rx_opt); @@ -5129,7 +5132,7 @@ static void __init tcp_struct_check(void) /* 32bit arches with 8byte alignment on u64 fields might need padding * before tcp_clock_cache. */ - CACHELINE_ASSERT_GROUP_SIZE(struct tcp_sock, tcp_sock_write_txrx, 91 + 4); + CACHELINE_ASSERT_GROUP_SIZE(struct tcp_sock, tcp_sock_write_txrx, 95 + 4);
/* RX read-write hotpath cache lines */ CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_rx, bytes_received); diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index b2793e749cfd..98782134c2f4 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -360,16 +360,25 @@ static void tcp_data_ecn_check(struct sock *sk, const struct sk_buff *skb) if (tcp_ca_needs_ecn(sk)) tcp_ca_event(sk, CA_EVENT_ECN_IS_CE);
- if (!(tp->ecn_flags & TCP_ECN_DEMAND_CWR)) { + if (!(tp->ecn_flags & TCP_ECN_DEMAND_CWR) && + tcp_ecn_mode_rfc3168(tp)) { /* Better not delay acks, sender can have a very low cwnd */ tcp_enter_quickack_mode(sk, 2); tp->ecn_flags |= TCP_ECN_DEMAND_CWR; } + /* As for RFC3168 ECN, the TCP_ECN_SEEN flag is set by + * tcp_data_ecn_check() when the ECN codepoint of + * received TCP data contains ECT(0), ECT(1), or CE. + */ + if (!tcp_ecn_mode_rfc3168(tp)) + break; tp->ecn_flags |= TCP_ECN_SEEN; break; default: if (tcp_ca_needs_ecn(sk)) tcp_ca_event(sk, CA_EVENT_ECN_NO_CE); + if (!tcp_ecn_mode_rfc3168(tp)) + break; tp->ecn_flags |= TCP_ECN_SEEN; break; } @@ -385,10 +394,64 @@ static void tcp_count_delivered(struct tcp_sock *tp, u32 delivered, bool ece_ack) { tp->delivered += delivered; - if (ece_ack) + if (tcp_ecn_mode_rfc3168(tp) && ece_ack) tcp_count_delivered_ce(tp, delivered); }
+/* Returns the ECN CE delta */ +static u32 __tcp_accecn_process(struct sock *sk, const struct sk_buff *skb, + u32 delivered_pkts, int flag) +{ + const struct tcphdr *th = tcp_hdr(skb); + struct tcp_sock *tp = tcp_sk(sk); + u32 delta, safe_delta; + u32 corrected_ace; + + /* Reordered ACK or uncertain due to lack of data to send and ts */ + if (!(flag & (FLAG_FORWARD_PROGRESS | FLAG_TS_PROGRESS))) + return 0; + + if (!(flag & FLAG_SLOWPATH)) { + /* AccECN counter might overflow on large ACKs */ + if (delivered_pkts <= TCP_ACCECN_CEP_ACE_MASK) + return 0; + } + + /* ACE field is not available during handshake */ + if (flag & FLAG_SYN_ACKED) + return 0; + + if (tp->received_ce_pending >= TCP_ACCECN_ACE_MAX_DELTA) + inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_NOW; + + corrected_ace = tcp_accecn_ace(th) - TCP_ACCECN_CEP_INIT_OFFSET; + delta = (corrected_ace - tp->delivered_ce) & TCP_ACCECN_CEP_ACE_MASK; + if (delivered_pkts <= TCP_ACCECN_CEP_ACE_MASK) + return delta; + + safe_delta = delivered_pkts - + ((delivered_pkts - delta) & TCP_ACCECN_CEP_ACE_MASK); + + return safe_delta; +} + +static u32 tcp_accecn_process(struct sock *sk, const struct sk_buff *skb, + u32 delivered_pkts, int *flag) +{ + struct tcp_sock *tp = tcp_sk(sk); + u32 delta; + + delta = __tcp_accecn_process(sk, skb, delivered_pkts, *flag); + if (delta > 0) { + tcp_count_delivered_ce(tp, delta); + *flag |= FLAG_ECE; + /* Recalculate header predictor */ + if (tp->pred_flags) + tcp_fast_path_on(tp); + } + return delta; +} + /* Buffer size and advertised window tuning. * * 1. Tuning sk->sk_sndbuf, when connection enters established state. @@ -3744,7 +3807,8 @@ static void tcp_xmit_recovery(struct sock *sk, int rexmit) }
/* Returns the number of packets newly acked or sacked by the current ACK */ -static u32 tcp_newly_delivered(struct sock *sk, u32 prior_delivered, int flag) +static u32 tcp_newly_delivered(struct sock *sk, u32 prior_delivered, + u32 ecn_count, int flag) { const struct net *net = sock_net(sk); struct tcp_sock *tp = tcp_sk(sk); @@ -3752,8 +3816,12 @@ static u32 tcp_newly_delivered(struct sock *sk, u32 prior_delivered, int flag)
delivered = tp->delivered - prior_delivered; NET_ADD_STATS(net, LINUX_MIB_TCPDELIVERED, delivered); - if (flag & FLAG_ECE) - NET_ADD_STATS(net, LINUX_MIB_TCPDELIVEREDCE, delivered); + + if (flag & FLAG_ECE) { + if (tcp_ecn_mode_rfc3168(tp)) + ecn_count = delivered; + NET_ADD_STATS(net, LINUX_MIB_TCPDELIVEREDCE, ecn_count); + }
return delivered; } @@ -3774,6 +3842,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag) u32 delivered = tp->delivered; u32 lost = tp->lost; int rexmit = REXMIT_NONE; /* Flag to (re)transmit to recover losses */ + u32 ecn_count = 0; /* Did we receive ECE/an AccECN ACE update? */ u32 prior_fack;
sack_state.first_sackt = 0; @@ -3881,6 +3950,11 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag)
tcp_rack_update_reo_wnd(sk, &rs);
+ if (tcp_ecn_mode_accecn(tp)) + ecn_count = tcp_accecn_process(sk, skb, + tp->delivered - delivered, + &flag); + tcp_in_ack_event(sk, flag);
if (tp->tlp_high_seq) @@ -3905,7 +3979,8 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag) if ((flag & FLAG_FORWARD_PROGRESS) || !(flag & FLAG_NOT_DUP)) sk_dst_confirm(sk);
- delivered = tcp_newly_delivered(sk, delivered, flag); + delivered = tcp_newly_delivered(sk, delivered, ecn_count, flag); + lost = tp->lost - lost; /* freshly marked lost */ rs.is_ack_delayed = !!(flag & FLAG_ACK_MAYBE_DELAYED); tcp_rate_gen(sk, delivered, lost, is_sack_reneg, sack_state.rate); @@ -3914,12 +3989,16 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag) return 1;
no_queue: + if (tcp_ecn_mode_accecn(tp)) + ecn_count = tcp_accecn_process(sk, skb, + tp->delivered - delivered, + &flag); tcp_in_ack_event(sk, flag); /* If data was DSACKed, see if we can undo a cwnd reduction. */ if (flag & FLAG_DSACKING_ACK) { tcp_fastretrans_alert(sk, prior_snd_una, num_dupack, &flag, &rexmit); - tcp_newly_delivered(sk, delivered, flag); + tcp_newly_delivered(sk, delivered, ecn_count, flag); } /* If this ack opens up a zero window, clear backoff. It was * being used to time the probes, and is probably far higher than @@ -3940,7 +4019,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag) &sack_state); tcp_fastretrans_alert(sk, prior_snd_una, num_dupack, &flag, &rexmit); - tcp_newly_delivered(sk, delivered, flag); + tcp_newly_delivered(sk, delivered, ecn_count, flag); tcp_xmit_recovery(sk, rexmit); }
@@ -6071,6 +6150,8 @@ void tcp_rcv_established(struct sock *sk, struct sk_buff *skb) flag |= __tcp_replace_ts_recent(tp, delta);
+ tcp_ecn_received_counters(sk, skb); + /* We know that such packets are checksummed * on entry. */ @@ -6119,6 +6200,7 @@ void tcp_rcv_established(struct sock *sk, struct sk_buff *skb) /* Bulk data transfer: receiver */ tcp_cleanup_skb(skb); __skb_pull(skb, tcp_header_len); + tcp_ecn_received_counters(sk, skb); eaten = tcp_queue_rcv(sk, skb, &fragstolen);
tcp_event_data_recv(sk, skb); @@ -6159,6 +6241,8 @@ void tcp_rcv_established(struct sock *sk, struct sk_buff *skb) return;
step5: + tcp_ecn_received_counters(sk, skb); + reason = tcp_ack(sk, skb, FLAG_SLOWPATH | FLAG_UPDATE_TS_RECENT); if ((int)reason < 0) { reason = -reason; diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c index be8ceefa5332..a3a6d3e91d84 100644 --- a/net/ipv4/tcp_output.c +++ b/net/ipv4/tcp_output.c @@ -328,7 +328,14 @@ static void tcp_ecn_send(struct sock *sk, struct sk_buff *skb, { struct tcp_sock *tp = tcp_sk(sk);
- if (tcp_ecn_mode_rfc3168(tp)) { + if (!tcp_ecn_mode_any(tp)) + return; + + if (tcp_ecn_mode_accecn(tp)) { + INET_ECN_xmit(sk); + tcp_accecn_set_ace(th, tp); + skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ACCECN; + } else { /* Not-retransmitted data segment: set ECT and inject CWR. */ if (skb->len != tcp_header_len && !before(TCP_SKB_CB(skb)->seq, tp->snd_nxt)) {