From: Ilpo Järvinen ij@kernel.org

The following patch will use tcp_ecn_mode_accecn(), TCP_ACCECN_CEP_INIT_OFFSET, TCP_ACCECN_CEP_ACE_MASK in __tcp_fast_path_on() to make new flag for AccECN.

No functional changes.

Signed-off-by: Ilpo Järvinen ij@kernel.org Signed-off-by: Chia-Yu Chang chia-yu.chang@nokia-bell-labs.com --- include/net/tcp.h | 54 +++++++++++++++++++++++------------------------ 1 file changed, 27 insertions(+), 27 deletions(-)

diff --git a/include/net/tcp.h b/include/net/tcp.h index 761c4a0ad386..ed01ae4be7a9 100644 --- a/include/net/tcp.h +++ b/include/net/tcp.h @@ -812,33 +812,6 @@ static inline u32 __tcp_set_rto(const struct tcp_sock *tp) return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); }

-static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) -{ - /* mptcp hooks are only on the slow path */ - if (sk_is_mptcp((struct sock *)tp)) - return; - - tp->pred_flags = htonl((tp->tcp_header_len << 26) | - ntohl(TCP_FLAG_ACK) | - snd_wnd); -} - -static inline void tcp_fast_path_on(struct tcp_sock *tp) -{ - __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); -} - -static inline void tcp_fast_path_check(struct sock *sk) -{ - struct tcp_sock *tp = tcp_sk(sk); - - if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && - tp->rcv_wnd && - atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && - !tp->urg_data) - tcp_fast_path_on(tp); -} - u32 tcp_delack_max(const struct sock *sk);

/* Compute the actual rto_min value */ @@ -1798,6 +1771,33 @@ static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, return true; }

+static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) +{ + /* mptcp hooks are only on the slow path */ + if (sk_is_mptcp((struct sock *)tp)) + return; + + tp->pred_flags = htonl((tp->tcp_header_len << 26) | + ntohl(TCP_FLAG_ACK) | + snd_wnd); +} + +static inline void tcp_fast_path_on(struct tcp_sock *tp) +{ + __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); +} + +static inline void tcp_fast_path_check(struct sock *sk) +{ + struct tcp_sock *tp = tcp_sk(sk); + + if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && + tp->rcv_wnd && + atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && + !tp->urg_data) + tcp_fast_path_on(tp); +} + bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, int mib_idx, u32 *last_oow_ack_time);

From: Chia-Yu Chang chia-yu.chang@nokia-bell-labs.com

Use the first 3-byte hole at the beginning of the tcp_sock_write_txrx group for 'noneagle'/'rate_app_limited' to fill in the existing hole in later patches. Therefore, the group size of tcp_sock_write_txrx is reduced from 92 + 4 to 91 + 4. In addition, the group size of tcp_sock_write_rx is changed to 96 to fit in the pahole outcome. 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]; /* 2585 0 */ /* XXX 3 bytes hole, try to pack */

[...] struct tcp_options_received rx_opt; /* 2652 24 */ u8 nonagle:4; /* 2676: 0 1 */ u8 rate_app_limited:1; /* 2676: 4 1 */ /* XXX 3 bits hole, try to pack */

__cacheline_group_end__tcp_sock_write_txrx[0]; /* 2677 0 */ /* XXX 3 bytes hole, try to pack */

__cacheline_group_begin__tcp_sock_write_rx[0] __attribute__((__aligned__(8))); /* 2680 0 */

[...] __cacheline_group_end__tcp_sock_write_rx[0]; /* 2776 0 */

[...] /* size: 3264, cachelines: 51, members: 163 */ }

[AFTER THIS PATCH] struct tcp_sock { [...] __cacheline_group_begin__tcp_sock_write_txrx[0]; /* 2585 0 */ u8 nonagle:4; /* 2585: 0 1 */ u8 rate_app_limited:1; /* 2585: 4 1 */ /* XXX 3 bits hole, try to pack */ /* XXX 2 bytes hole, try to pack */

[...] struct tcp_options_received rx_opt; /* 2652 24 */

__cacheline_group_end__tcp_sock_write_txrx[0]; /* 2676 0 */ /* XXX 4 bytes hole, try to pack */

__cacheline_group_begin__tcp_sock_write_rx[0] __attribute__((__aligned__(8))); /* 2680 0 */

[...] __cacheline_group_end__tcp_sock_write_rx[0]; /* 2776 0 */

[...] /* size: 3264, cachelines: 51, members: 163 */ }

Signed-off-by: Chia-Yu Chang chia-yu.chang@nokia-bell-labs.com --- include/linux/tcp.h | 4 ++-- net/ipv4/tcp.c | 4 ++-- 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/include/linux/tcp.h b/include/linux/tcp.h index 1a5737b3753d..bd692764d039 100644 --- a/include/linux/tcp.h +++ b/include/linux/tcp.h @@ -285,6 +285,8 @@ struct tcp_sock { * Header prediction flags * 0x5?10 << 16 + snd_wnd in net byte order */ + u8 nonagle : 4,/* Disable Nagle algorithm? */ + rate_app_limited:1; /* rate_{delivered,interval_us} limited? */ __be32 pred_flags; u64 tcp_clock_cache; /* cache last tcp_clock_ns() (see tcp_mstamp_refresh()) */ u64 tcp_mstamp; /* most recent packet received/sent */ @@ -303,8 +305,6 @@ struct tcp_sock { * Options received (usually on last packet, some only on SYN packets). */ struct tcp_options_received rx_opt; - u8 nonagle : 4,/* Disable Nagle algorithm? */ - rate_app_limited:1; /* rate_{delivered,interval_us} limited? */ __cacheline_group_end(tcp_sock_write_txrx);

/* RX read-write hotpath cache lines */ diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c index 8a3c99246d2e..18b6cb7dc02e 100644 --- a/net/ipv4/tcp.c +++ b/net/ipv4/tcp.c @@ -5125,7 +5125,7 @@ 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, 92 + 4); + CACHELINE_ASSERT_GROUP_SIZE(struct tcp_sock, tcp_sock_write_txrx, 91 + 4);

/* RX read-write hotpath cache lines */ CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_rx, bytes_received); @@ -5142,7 +5142,7 @@ static void __init tcp_struct_check(void) CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_rx, bytes_acked); CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_rx, rcv_rtt_est); CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_rx, rcvq_space); - CACHELINE_ASSERT_GROUP_SIZE(struct tcp_sock, tcp_sock_write_rx, 99); + CACHELINE_ASSERT_GROUP_SIZE(struct tcp_sock, tcp_sock_write_rx, 96); }

void __init tcp_init(void)

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]; /* 2585 0 */ u8 nonagle:4; /* 2585: 0 1 */ u8 rate_app_limited:1; /* 2585: 4 1 */ /* XXX 3 bits hole, try to pack */ /* XXX 2 bytes hole, try to pack */

[...] u32 delivered_ce; /* 2640 4 */ u32 app_limited; /* 2644 4 */ u32 rcv_wnd; /* 2648 4 */ struct tcp_options_received rx_opt; /* 2652 24 */ __cacheline_group_end__tcp_sock_write_txrx[0]; /* 2676 0 */ /* XXX 4 bytes hole, try to pack */

[...] /* size: 3264, cachelines: 51, members: 163 */ }

[AFTER THIS PATCH] struct tcp_sock { [...] __cacheline_group_begin__tcp_sock_write_txrx[0]; /* 2585 0 */ u8 nonagle:4; /* 2585: 0 1 */ u8 rate_app_limited:1; /* 2585: 4 1 */ /* XXX 3 bits hole, try to pack */

/* Force alignment to the next boundary: */ u8 :0; u8 received_ce_pending:4;/* 2586: 0 1 */ u8 unused2:4; /* 2586: 4 1 */ /* XXX 1 byte hole, try to pack */

[...] u32 delivered_ce; /* 2640 4 */ u32 received_ce; /* 2644 4 */ u32 app_limited; /* 2648 4 */ u32 rcv_wnd; /* 2652 4 */ struct tcp_options_received rx_opt; /* 2656 24 */ __cacheline_group_end__tcp_sock_write_txrx[0]; /* 2680 0 */

[...] /* size: 3264, cachelines: 51, members: 166 */ }

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

--- 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 | 26 ++++ net/ipv4/tcp.c | 4 +- net/ipv4/tcp_input.c | 132 ++++++++++++++++-- net/ipv4/tcp_output.c | 20 ++- 6 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 bd692764d039..874e0e45dfad 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 ed01ae4be7a9..6cf5cea992e3 100644 --- a/include/net/tcp.h +++ b/include/net/tcp.h @@ -415,6 +415,11 @@ static inline void tcp_ecn_mode_set(struct tcp_sock *tp, u8 mode) tp->ecn_flags |= mode; }

+static inline u8 tcp_accecn_ace(const struct tcphdr *th) +{ + return (th->ae << 2) | (th->cwr << 1) | th->ece; +} + enum tcp_tw_status { TCP_TW_SUCCESS = 0, TCP_TW_RST = 1, @@ -964,6 +969,20 @@ 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 + +static inline void tcp_accecn_init_counters(struct tcp_sock *tp) +{ + tp->received_ce = 0; + tp->received_ce_pending = 0; +} + /* 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 */ @@ -1773,11 +1792,18 @@ 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/net/ipv4/tcp.c b/net/ipv4/tcp.c index 18b6cb7dc02e..8e0e8d784b1c 100644 --- a/net/ipv4/tcp.c +++ b/net/ipv4/tcp.c @@ -3392,6 +3392,7 @@ 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)); @@ -5118,6 +5119,7 @@ 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); @@ -5125,7 +5127,7 @@ 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 19a1542883df..0fa3803b353d 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -341,14 +341,17 @@ static bool tcp_in_quickack_mode(struct sock *sk)

static 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; }

static 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 @@ -384,16 +387,25 @@ 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; } @@ -428,10 +440,64 @@ 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. @@ -3787,7 +3853,8 @@ 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); @@ -3795,8 +3862,12 @@ 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; } @@ -3817,6 +3888,7 @@ 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; @@ -3924,6 +3996,11 @@ 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) @@ -3948,7 +4025,8 @@ 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); @@ -3957,12 +4035,16 @@ 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 @@ -3983,7 +4065,7 @@ 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); }

@@ -5821,6 +5903,31 @@ static void tcp_urg(struct sock *sk, struct sk_buff *skb, const struct tcphdr *t } }

+/* Updates Accurate ECN received counters from the received IP ECN field */ +static 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); + } +} + /* Accept RST for rcv_nxt - 1 after a FIN. * When tcp connections are abruptly terminated from Mac OSX (via ^C), a * FIN is sent followed by a RST packet. The RST is sent with the same @@ -6083,6 +6190,8 @@ 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. */ @@ -6127,6 +6236,7 @@ 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); @@ -6167,6 +6277,8 @@ 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 974e4062be88..4b32aa619594 100644 --- a/net/ipv4/tcp_output.c +++ b/net/ipv4/tcp_output.c @@ -374,6 +374,17 @@ tcp_ecn_make_synack(const struct request_sock *req, struct tcphdr *th) th->ece = 1; }

+static 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; +} + /* Set up ECN state for a packet on a ESTABLISHED socket that is about to * be sent. */ @@ -382,7 +393,14 @@ 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)) {

From: Ilpo Järvinen ij@kernel.org

Accurate ECN negotiation parts based on the specification: https://tools.ietf.org/id/draft-ietf-tcpm-accurate-ecn-28.txt

Accurate ECN is negotiated using ECE, CWR and AE flags in the TCP header. TCP falls back into using RFC3168 ECN if one of the ends supports only RFC3168-style ECN.

The AccECN negotiation includes reflecting IP ECN field value seen in SYN and SYNACK back using the same bits as negotiation to allow responding to SYN CE marks and to detect ECN field mangling. CE marks should not occur currently because SYN=1 segments are sent with Non-ECT in IP ECN field (but proposal exists to remove this restriction).

Reflecting SYN IP ECN field in SYNACK is relatively simple. Reflecting SYNACK IP ECN field in the final/third ACK of the handshake is more challenging. Linux TCP code is not well prepared for using the final/third ACK a signalling channel which makes things somewhat complicated here.

tcp_ecn sysctl can be used to select the highest ECN variant (Accurate ECN, ECN, No ECN) that is attemped to be negotiated and requested for incoming connection and outgoing connection: TCP_ECN_IN_NOECN_OUT_NOECN, TCP_ECN_IN_ECN_OUT_ECN, TCP_ECN_IN_ECN_OUT_NOECN, TCP_ECN_IN_ACCECN_OUT_ACCECN, TCP_ECN_IN_ACCECN_OUT_ECN, and TCP_ECN_IN_ACCECN_OUT_NOECN.

After this patch, the size of tcp_request_sock remains unchanged and no new holes are added. Below are the pahole outcomes before and after this patch:

[BEFORE THIS PATCH] struct tcp_request_sock { [...] u32 rcv_nxt; /* 352 4 */ u8 syn_tos; /* 356 1 */

/* size: 360, cachelines: 6, members: 16 */ }

[AFTER THIS PATCH] struct tcp_request_sock { [...] u32 rcv_nxt; /* 352 4 */ u8 syn_tos; /* 356 1 */ bool accecn_ok; /* 357 1 */ u8 syn_ect_snt:2; /* 358: 0 1 */ u8 syn_ect_rcv:2; /* 358: 2 1 */ u8 accecn_fail_mode:4; /* 358: 4 1 */

/* size: 360, cachelines: 6, members: 20 */ }

After this patch, the size of tcp_sock remains unchanged and no new holes are added. Also, 4 bits of the existing 2-byte hole are exploited. Below are the pahole outcomes before and after this patch:

[BEFORE THIS PATCH] struct tcp_sock { [...] u8 dup_ack_counter:2; /* 2825: 0 1 */ u8 tlp_retrans:1; /* 2825: 2 1 */ u8 unused:5; /* 2825: 3 1 */ u8 thin_lto:1; /* 2826: 0 1 */ u8 fastopen_connect:1; /* 2826: 1 1 */ u8 fastopen_no_cookie:1; /* 2826: 2 1 */ u8 fastopen_client_fail:2; /* 2826: 3 1 */ u8 frto:1; /* 2826: 5 1 */ /* XXX 2 bits hole, try to pack */

[...] u8 keepalive_probes; /* 2829 1 */ /* XXX 2 bytes hole, try to pack */

[...] /* size: 3264, cachelines: 51, members: 166 */ }

[AFTER THIS PATCH] struct tcp_sock { [...] u8 dup_ack_counter:2; /* 2825: 0 1 */ u8 tlp_retrans:1; /* 2825: 2 1 */ u8 syn_ect_snt:2; /* 2825: 3 1 */ u8 syn_ect_rcv:2; /* 2825: 5 1 */ u8 wait_third_ack:1; /* 2825: 7 1 */ u8 thin_lto:1; /* 2826: 0 1 */ u8 fastopen_connect:1; /* 2826: 1 1 */ u8 fastopen_no_cookie:1; /* 2826: 2 1 */ u8 fastopen_client_fail:2; /* 2826: 3 1 */ u8 frto:1; /* 2826: 5 1 */ /* XXX 2 bits hole, try to pack */

[...] u8 keepalive_probes; /* 2829 1 */ u8 accecn_fail_mode:4; /* 2830: 0 1 */ /* XXX 4 bits hole, try to pack */ /* XXX 1 byte hole, try to pack */

[...] /* size: 3264, cachelines: 51, members: 169 */ }

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 --- .../networking/net_cachelines/tcp_sock.rst | 4 + include/linux/tcp.h | 9 +- include/net/tcp.h | 93 ++++++++- net/ipv4/syncookies.c | 3 + net/ipv4/sysctl_net_ipv4.c | 3 +- net/ipv4/tcp.c | 2 + net/ipv4/tcp_input.c | 178 ++++++++++++++++-- net/ipv4/tcp_ipv4.c | 5 +- net/ipv4/tcp_minisocks.c | 51 ++++- net/ipv4/tcp_output.c | 78 ++++++-- net/ipv6/syncookies.c | 1 + net/ipv6/tcp_ipv6.c | 1 + 12 files changed, 384 insertions(+), 44 deletions(-)

diff --git a/Documentation/networking/net_cachelines/tcp_sock.rst b/Documentation/networking/net_cachelines/tcp_sock.rst index 31313a9adccc..22ac668fe6c7 100644 --- a/Documentation/networking/net_cachelines/tcp_sock.rst +++ b/Documentation/networking/net_cachelines/tcp_sock.rst @@ -103,6 +103,10 @@ u32 delivered read_mostly read_w 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 +u8:2 syn_ect_snt write_mostly read_write +u8:2 syn_ect_rcv read_mostly read_write +u8:1 wait_third_ack read_write +u8:4 accecn_fail_mode 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 874e0e45dfad..1d8301f2883c 100644 --- a/include/linux/tcp.h +++ b/include/linux/tcp.h @@ -168,6 +168,10 @@ struct tcp_request_sock { * after data-in-SYN. */ u8 syn_tos; + bool accecn_ok; + u8 syn_ect_snt: 2, + syn_ect_rcv: 2, + accecn_fail_mode:4; #ifdef CONFIG_TCP_AO u8 ao_keyid; u8 ao_rcv_next; @@ -375,7 +379,9 @@ struct tcp_sock { u8 compressed_ack; u8 dup_ack_counter:2, tlp_retrans:1, /* TLP is a retransmission */ - unused:5; + syn_ect_snt:2, /* AccECN ECT memory, only */ + syn_ect_rcv:2, /* ... needed durign 3WHS + first seqno */ + wait_third_ack:1; /* Wait 3rd ACK in simultaneous open */ u8 thin_lto : 1,/* Use linear timeouts for thin streams */ fastopen_connect:1, /* FASTOPEN_CONNECT sockopt */ fastopen_no_cookie:1, /* Allow send/recv SYN+data without a cookie */ @@ -391,6 +397,7 @@ struct tcp_sock { syn_fastopen_child:1; /* created TFO passive child socket */

u8 keepalive_probes; /* num of allowed keep alive probes */ + u8 accecn_fail_mode:4; /* AccECN failure handling */ u32 tcp_tx_delay; /* delay (in usec) added to TX packets */

/* RTT measurement */ diff --git a/include/net/tcp.h b/include/net/tcp.h index 6cf5cea992e3..4d6325fa3f67 100644 --- a/include/net/tcp.h +++ b/include/net/tcp.h @@ -27,6 +27,7 @@ #include <linux/ktime.h> #include <linux/indirect_call_wrapper.h> #include <linux/bits.h> +#include <linux/bitfield.h>

#include <net/inet_connection_sock.h> #include <net/inet_timewait_sock.h> @@ -234,6 +235,37 @@ static_assert((1 << ATO_BITS) > TCP_DELACK_MAX); #define TCPOLEN_MSS_ALIGNED 4 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8

+/* tp->accecn_fail_mode */ +#define TCP_ACCECN_ACE_FAIL_SEND BIT(0) +#define TCP_ACCECN_ACE_FAIL_RECV BIT(1) +#define TCP_ACCECN_OPT_FAIL_SEND BIT(2) +#define TCP_ACCECN_OPT_FAIL_RECV BIT(3) + +static inline bool tcp_accecn_ace_fail_send(const struct tcp_sock *tp) +{ + return tp->accecn_fail_mode & TCP_ACCECN_ACE_FAIL_SEND; +} + +static inline bool tcp_accecn_ace_fail_recv(const struct tcp_sock *tp) +{ + return tp->accecn_fail_mode & TCP_ACCECN_ACE_FAIL_RECV; +} + +static inline bool tcp_accecn_opt_fail_send(const struct tcp_sock *tp) +{ + return tp->accecn_fail_mode & TCP_ACCECN_OPT_FAIL_SEND; +} + +static inline bool tcp_accecn_opt_fail_recv(const struct tcp_sock *tp) +{ + return tp->accecn_fail_mode & TCP_ACCECN_OPT_FAIL_RECV; +} + +static inline void tcp_accecn_fail_mode_set(struct tcp_sock *tp, u8 mode) +{ + tp->accecn_fail_mode |= mode; +} + /* Flags in tp->nonagle */ #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ #define TCP_NAGLE_CORK 2 /* Socket is corked */ @@ -420,6 +452,23 @@ static inline u8 tcp_accecn_ace(const struct tcphdr *th) return (th->ae << 2) | (th->cwr << 1) | th->ece; }

+/* Infer the ECT value our SYN arrived with from the echoed ACE field */ +static inline int tcp_accecn_extract_syn_ect(u8 ace) +{ + if (ace & 0x1) + return INET_ECN_ECT_1; + if (!(ace & 0x2)) + return INET_ECN_ECT_0; + if (ace & 0x4) + return INET_ECN_CE; + return INET_ECN_NOT_ECT; +} + +bool tcp_accecn_validate_syn_feedback(struct sock *sk, u8 ace, u8 sent_ect); +void tcp_accecn_third_ack(struct sock *sk, const struct sk_buff *skb, + u8 syn_ect_snt); +void tcp_ecn_received_counters(struct sock *sk, const struct sk_buff *skb); + enum tcp_tw_status { TCP_TW_SUCCESS = 0, TCP_TW_RST = 1, @@ -657,6 +706,15 @@ static inline bool cookie_ecn_ok(const struct net *net, const struct dst_entry * dst_feature(dst, RTAX_FEATURE_ECN); }

+/* AccECN specification, 5.1: [...] a server can determine that it + * negotiated AccECN as [...] if the ACK contains an ACE field with + * the value 0b010 to 0b111 (decimal 2 to 7). + */ +static inline bool cookie_accecn_ok(const struct tcphdr *th) +{ + return tcp_accecn_ace(th) > 0x1; +} + #if IS_ENABLED(CONFIG_BPF) static inline bool cookie_bpf_ok(struct sk_buff *skb) { @@ -968,6 +1026,7 @@ 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 TCPHDR_SYNACK_ACCECN (TCPHDR_SYN | TCPHDR_ACK | TCPHDR_CWR)

#define TCP_ACCECN_CEP_ACE_MASK 0x7 #define TCP_ACCECN_ACE_MAX_DELTA 6 @@ -977,6 +1036,19 @@ static inline u32 tcp_rsk_tsval(const struct tcp_request_sock *treq) */ #define TCP_ACCECN_CEP_INIT_OFFSET 5

+/* The highest ECN variant (Accurate ECN, ECN, or no ECN) that is + * attemped to be negotiated and requested for incoming connection + * and outgoing connection, respectively. + */ +enum tcp_ecn_mode { + TCP_ECN_IN_NOECN_OUT_NOECN = 0, + TCP_ECN_IN_ECN_OUT_ECN = 1, + TCP_ECN_IN_ECN_OUT_NOECN = 2, + TCP_ECN_IN_ACCECN_OUT_ACCECN = 3, + TCP_ECN_IN_ACCECN_OUT_ECN = 4, + TCP_ECN_IN_ACCECN_OUT_NOECN = 5, +}; + static inline void tcp_accecn_init_counters(struct tcp_sock *tp) { tp->received_ce = 0; @@ -1051,6 +1123,15 @@ struct tcp_skb_cb {

#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))

+static inline u16 tcp_accecn_reflector_flags(u8 ect) +{ + u32 flags = ect + 2; + + if (ect == 3) + flags++; + return FIELD_PREP(TCPHDR_ACE, flags); +} + extern const struct inet_connection_sock_af_ops ipv4_specific;

#if IS_ENABLED(CONFIG_IPV6) @@ -1173,7 +1254,10 @@ enum tcp_ca_ack_event_flags { #define TCP_CONG_NON_RESTRICTED BIT(0) /* Requires ECN/ECT set on all packets */ #define TCP_CONG_NEEDS_ECN BIT(1) -#define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) +/* Require successfully negotiated AccECN capability */ +#define TCP_CONG_NEEDS_ACCECN BIT(2) +#define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN | \ + TCP_CONG_NEEDS_ACCECN)

union tcp_cc_info;

@@ -1305,6 +1389,13 @@ static inline bool tcp_ca_needs_ecn(const struct sock *sk) return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; }

+static inline bool tcp_ca_needs_accecn(const struct sock *sk) +{ + const struct inet_connection_sock *icsk = inet_csk(sk); + + return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ACCECN; +} + static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) { const struct inet_connection_sock *icsk = inet_csk(sk); diff --git a/net/ipv4/syncookies.c b/net/ipv4/syncookies.c index 5459a78b9809..3a44eb9c1d1a 100644 --- a/net/ipv4/syncookies.c +++ b/net/ipv4/syncookies.c @@ -403,6 +403,7 @@ struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb) struct tcp_sock *tp = tcp_sk(sk); struct inet_request_sock *ireq; struct net *net = sock_net(sk); + struct tcp_request_sock *treq; struct request_sock *req; struct sock *ret = sk; struct flowi4 fl4; @@ -428,6 +429,7 @@ struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb) }

ireq = inet_rsk(req); + treq = tcp_rsk(req);

sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr); sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr); @@ -482,6 +484,7 @@ struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb) if (!req->syncookie) ireq->rcv_wscale = rcv_wscale; ireq->ecn_ok &= cookie_ecn_ok(net, &rt->dst); + treq->accecn_ok = ireq->ecn_ok && cookie_accecn_ok(th);

ret = tcp_get_cookie_sock(sk, skb, req, &rt->dst); /* ip_queue_xmit() depends on our flow being setup diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c index 3a43010d726f..75ec1a599b52 100644 --- a/net/ipv4/sysctl_net_ipv4.c +++ b/net/ipv4/sysctl_net_ipv4.c @@ -47,6 +47,7 @@ static unsigned int udp_child_hash_entries_max = UDP_HTABLE_SIZE_MAX; static int tcp_plb_max_rounds = 31; static int tcp_plb_max_cong_thresh = 256; static unsigned int tcp_tw_reuse_delay_max = TCP_PAWS_MSL * MSEC_PER_SEC; +static int tcp_ecn_mode_max = 5;

/* obsolete */ static int sysctl_tcp_low_latency __read_mostly; @@ -728,7 +729,7 @@ static struct ctl_table ipv4_net_table[] = { .mode = 0644, .proc_handler = proc_dou8vec_minmax, .extra1 = SYSCTL_ZERO, - .extra2 = SYSCTL_TWO, + .extra2 = &tcp_ecn_mode_max, }, { .procname = "tcp_ecn_fallback", diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c index 8e0e8d784b1c..e6d7b5420c88 100644 --- a/net/ipv4/tcp.c +++ b/net/ipv4/tcp.c @@ -3392,6 +3392,8 @@ int tcp_disconnect(struct sock *sk, int flags) tp->window_clamp = 0; tp->delivered = 0; tp->delivered_ce = 0; + tp->wait_third_ack = 0; + tp->accecn_fail_mode = 0; tcp_accecn_init_counters(tp); if (icsk->icsk_ca_initialized && icsk->icsk_ca_ops->release) icsk->icsk_ca_ops->release(sk); diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index 0fa3803b353d..c986452302cb 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -411,14 +411,114 @@ static void tcp_data_ecn_check(struct sock *sk, const struct sk_buff *skb) } }

-static void tcp_ecn_rcv_synack(struct tcp_sock *tp, const struct tcphdr *th) +/* AccECN specificaiton, 3.1.2: If a TCP server that implements AccECN + * receives a SYN with the three TCP header flags (AE, CWR and ECE) set + * to any combination other than 000, 011 or 111, it MUST negotiate the + * use of AccECN as if they had been set to 111. + */ +static bool tcp_accecn_syn_requested(const struct tcphdr *th) +{ + u8 ace = tcp_accecn_ace(th); + + return ace && ace != 0x3; +} + +/* Check ECN field transition to detect invalid transitions */ +static bool tcp_ect_transition_valid(u8 snt, u8 rcv) { - if (tcp_ecn_mode_rfc3168(tp) && (!th->ece || th->cwr)) + if (rcv == snt) + return true; + + /* Non-ECT altered to something or something became non-ECT */ + if (snt == INET_ECN_NOT_ECT || rcv == INET_ECN_NOT_ECT) + return false; + /* CE -> ECT(0/1)? */ + if (snt == INET_ECN_CE) + return false; + return true; +} + +bool tcp_accecn_validate_syn_feedback(struct sock *sk, u8 ace, u8 sent_ect) +{ + u8 ect = tcp_accecn_extract_syn_ect(ace); + struct tcp_sock *tp = tcp_sk(sk); + + if (!sock_net(sk)->ipv4.sysctl_tcp_ecn_fallback) + return true; + + if (!tcp_ect_transition_valid(sent_ect, ect)) { + tcp_accecn_fail_mode_set(tp, TCP_ACCECN_ACE_FAIL_RECV); + return false; + } + + return true; +} + +/* See Table 2 of the AccECN draft */ + +static void tcp_ecn_rcv_synack(struct sock *sk, const struct tcphdr *th, + u8 ip_dsfield) +{ + struct tcp_sock *tp = tcp_sk(sk); + u8 ace = tcp_accecn_ace(th); + + switch (ace) { + case 0x0: + case 0x7: + /* +========+========+============+=============+ + * | A | B | SYN/ACK | Feedback | + * | | | B->A | Mode of A | + * | | | AE CWR ECE | | + * +========+========+============+=============+ + * | AccECN | No ECN | 0 0 0 | Not ECN | + * | AccECN | Broken | 1 1 1 | Not ECN | + * +--------+--------+------------+-------------+ + */ tcp_ecn_mode_set(tp, TCP_ECN_DISABLED); + break; + case 0x1: + case 0x5: + /* +========+========+============+=============+ + * | A | B | SYN/ACK | Feedback | + * | | | B->A | Mode of A | + * | | | AE CWR ECE | | + * +========+========+============+=============+ + * | AccECN | Nonce | 1 0 1 | (Reserved) | + * | AccECN | ECN | 0 0 1 | Classic ECN | + * | Nonce | AccECN | 0 0 1 | Classic ECN | + * | ECN | AccECN | 0 0 1 | Classic ECN | + * +--------+--------+------------+-------------+ + */ + if (tcp_ecn_mode_pending(tp)) + /* Downgrade from AccECN, or requested initially */ + tcp_ecn_mode_set(tp, TCP_ECN_MODE_RFC3168); + break; + default: + tcp_ecn_mode_set(tp, TCP_ECN_MODE_ACCECN); + tp->syn_ect_rcv = ip_dsfield & INET_ECN_MASK; + if (INET_ECN_is_ce(ip_dsfield) && + tcp_accecn_validate_syn_feedback(sk, ace, + tp->syn_ect_snt)) { + tp->received_ce++; + tp->received_ce_pending++; + } + break; + } }

-static void tcp_ecn_rcv_syn(struct tcp_sock *tp, const struct tcphdr *th) +static void tcp_ecn_rcv_syn(struct tcp_sock *tp, const struct tcphdr *th, + const struct sk_buff *skb) { + if (tcp_ecn_mode_pending(tp)) { + if (!tcp_accecn_syn_requested(th)) { + /* Downgrade to classic ECN feedback */ + tcp_ecn_mode_set(tp, TCP_ECN_MODE_RFC3168); + } else { + tp->syn_ect_rcv = TCP_SKB_CB(skb)->ip_dsfield & + INET_ECN_MASK; + tcp_ecn_mode_set(tp, TCP_ECN_MODE_ACCECN); + } + } if (tcp_ecn_mode_rfc3168(tp) && (!th->ece || !th->cwr)) tcp_ecn_mode_set(tp, TCP_ECN_DISABLED); } @@ -3711,8 +3811,18 @@ bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, return __tcp_oow_rate_limited(net, mib_idx, last_oow_ack_time); }

+static void tcp_send_ack_reflect_ect(struct sock *sk, bool accecn_reflector) +{ + struct tcp_sock *tp = tcp_sk(sk); + u16 flags = 0; + + if (accecn_reflector) + flags = tcp_accecn_reflector_flags(tp->syn_ect_rcv); + __tcp_send_ack(sk, tp->rcv_nxt, flags); +} + /* RFC 5961 7 [ACK Throttling] */ -static void tcp_send_challenge_ack(struct sock *sk) +static void tcp_send_challenge_ack(struct sock *sk, bool accecn_reflector) { struct tcp_sock *tp = tcp_sk(sk); struct net *net = sock_net(sk); @@ -3742,7 +3852,7 @@ static void tcp_send_challenge_ack(struct sock *sk) WRITE_ONCE(net->ipv4.tcp_challenge_count, count - 1); send_ack: NET_INC_STATS(net, LINUX_MIB_TCPCHALLENGEACK); - tcp_send_ack(sk); + tcp_send_ack_reflect_ect(sk, accecn_reflector); } }

@@ -3909,7 +4019,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag) /* RFC 5961 5.2 [Blind Data Injection Attack].[Mitigation] */ if (before(ack, prior_snd_una - max_window)) { if (!(flag & FLAG_NO_CHALLENGE_ACK)) - tcp_send_challenge_ack(sk); + tcp_send_challenge_ack(sk, false); return -SKB_DROP_REASON_TCP_TOO_OLD_ACK; } goto old_ack; @@ -5904,8 +6014,7 @@ static void tcp_urg(struct sock *sk, struct sk_buff *skb, const struct tcphdr *t }

/* Updates Accurate ECN received counters from the received IP ECN field */ -static void tcp_ecn_received_counters(struct sock *sk, - const struct sk_buff *skb) +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); @@ -5952,6 +6061,7 @@ static bool tcp_validate_incoming(struct sock *sk, struct sk_buff *skb, const struct tcphdr *th, int syn_inerr) { struct tcp_sock *tp = tcp_sk(sk); + bool accecn_reflector = false; SKB_DR(reason);

/* RFC1323: H1. Apply PAWS check first. */ @@ -6044,7 +6154,7 @@ static bool tcp_validate_incoming(struct sock *sk, struct sk_buff *skb, if (tp->syn_fastopen && !tp->data_segs_in && sk->sk_state == TCP_ESTABLISHED) tcp_fastopen_active_disable(sk); - tcp_send_challenge_ack(sk); + tcp_send_challenge_ack(sk, false); SKB_DR_SET(reason, TCP_RESET); goto discard; } @@ -6055,16 +6165,23 @@ static bool tcp_validate_incoming(struct sock *sk, struct sk_buff *skb, * RFC 5961 4.2 : Send a challenge ack */ if (th->syn) { + if (tcp_ecn_mode_accecn(tp)) + accecn_reflector = true; if (sk->sk_state == TCP_SYN_RECV && sk->sk_socket && th->ack && TCP_SKB_CB(skb)->seq + 1 == TCP_SKB_CB(skb)->end_seq && TCP_SKB_CB(skb)->seq + 1 == tp->rcv_nxt && - TCP_SKB_CB(skb)->ack_seq == tp->snd_nxt) + TCP_SKB_CB(skb)->ack_seq == tp->snd_nxt) { + if (!tcp_ecn_disabled(tp)) { + tp->wait_third_ack = true; + tcp_send_ack_reflect_ect(sk, accecn_reflector); + } goto pass; + } syn_challenge: if (syn_inerr) TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS); NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPSYNCHALLENGE); - tcp_send_challenge_ack(sk); + tcp_send_challenge_ack(sk, accecn_reflector); SKB_DR_SET(reason, TCP_INVALID_SYN); goto discard; } @@ -6277,6 +6394,12 @@ void tcp_rcv_established(struct sock *sk, struct sk_buff *skb) return;

step5: + if (unlikely(tp->wait_third_ack)) { + tp->wait_third_ack = 0; + if (tcp_ecn_mode_accecn(tp)) + tcp_accecn_third_ack(sk, skb, tp->syn_ect_snt); + tcp_fast_path_on(tp); + } tcp_ecn_received_counters(sk, skb);

reason = tcp_ack(sk, skb, FLAG_SLOWPATH | FLAG_UPDATE_TS_RECENT); @@ -6529,7 +6652,8 @@ static int tcp_rcv_synsent_state_process(struct sock *sk, struct sk_buff *skb, * state to ESTABLISHED..." */

- tcp_ecn_rcv_synack(tp, th); + if (tcp_ecn_mode_any(tp)) + tcp_ecn_rcv_synack(sk, th, TCP_SKB_CB(skb)->ip_dsfield);

tcp_init_wl(tp, TCP_SKB_CB(skb)->seq); tcp_try_undo_spurious_syn(sk); @@ -6601,7 +6725,7 @@ static int tcp_rcv_synsent_state_process(struct sock *sk, struct sk_buff *skb, TCP_DELACK_MAX, false); goto consume; } - tcp_send_ack(sk); + tcp_send_ack_reflect_ect(sk, tcp_ecn_mode_accecn(tp)); return -1; }

@@ -6660,7 +6784,7 @@ static int tcp_rcv_synsent_state_process(struct sock *sk, struct sk_buff *skb, tp->snd_wl1 = TCP_SKB_CB(skb)->seq; tp->max_window = tp->snd_wnd;

- tcp_ecn_rcv_syn(tp, th); + tcp_ecn_rcv_syn(tp, th, skb);

tcp_mtup_init(sk); tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); @@ -6842,7 +6966,7 @@ tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb) } /* accept old ack during closing */ if ((int)reason < 0) { - tcp_send_challenge_ack(sk); + tcp_send_challenge_ack(sk, false); reason = -reason; goto discard; } @@ -6889,9 +7013,16 @@ tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb) tp->lsndtime = tcp_jiffies32;

tcp_initialize_rcv_mss(sk); - tcp_fast_path_on(tp); + if (likely(!tp->wait_third_ack)) { + if (tcp_ecn_mode_accecn(tp)) + tcp_accecn_third_ack(sk, skb, tp->syn_ect_snt); + tcp_fast_path_on(tp); + } if (sk->sk_shutdown & SEND_SHUTDOWN) tcp_shutdown(sk, SEND_SHUTDOWN); + + if (sk->sk_socket && tp->wait_third_ack) + goto consume; break;

case TCP_FIN_WAIT1: { @@ -7061,6 +7192,15 @@ static void tcp_ecn_create_request(struct request_sock *req, bool ect, ecn_ok; u32 ecn_ok_dst;

+ if (tcp_accecn_syn_requested(th) && + (net->ipv4.sysctl_tcp_ecn >= 3 || tcp_ca_needs_accecn(listen_sk))) { + inet_rsk(req)->ecn_ok = 1; + tcp_rsk(req)->accecn_ok = 1; + tcp_rsk(req)->syn_ect_rcv = TCP_SKB_CB(skb)->ip_dsfield & + INET_ECN_MASK; + return; + } + if (!th_ecn) return;

@@ -7068,7 +7208,8 @@ static void tcp_ecn_create_request(struct request_sock *req, ecn_ok_dst = dst_feature(dst, DST_FEATURE_ECN_MASK); ecn_ok = READ_ONCE(net->ipv4.sysctl_tcp_ecn) || ecn_ok_dst;

- if (((!ect || th->res1) && ecn_ok) || tcp_ca_needs_ecn(listen_sk) || + if (((!ect || th->res1 || th->ae) && ecn_ok) || + tcp_ca_needs_ecn(listen_sk) || (ecn_ok_dst & DST_FEATURE_ECN_CA) || tcp_bpf_ca_needs_ecn((struct sock *)req)) inet_rsk(req)->ecn_ok = 1; @@ -7086,6 +7227,9 @@ static void tcp_openreq_init(struct request_sock *req, tcp_rsk(req)->snt_synack = 0; tcp_rsk(req)->snt_tsval_first = 0; tcp_rsk(req)->last_oow_ack_time = 0; + tcp_rsk(req)->accecn_ok = 0; + tcp_rsk(req)->syn_ect_rcv = 0; + tcp_rsk(req)->syn_ect_snt = 0; req->mss = rx_opt->mss_clamp; req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; ireq->tstamp_ok = rx_opt->tstamp_ok; diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c index 6a14f9e6fef6..c5b93802d7c0 100644 --- a/net/ipv4/tcp_ipv4.c +++ b/net/ipv4/tcp_ipv4.c @@ -1189,7 +1189,7 @@ static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst, enum tcp_synack_type synack_type, struct sk_buff *syn_skb) { - const struct inet_request_sock *ireq = inet_rsk(req); + struct inet_request_sock *ireq = inet_rsk(req); struct flowi4 fl4; int err = -1; struct sk_buff *skb; @@ -1202,6 +1202,7 @@ static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst, skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb);

if (skb) { + tcp_rsk(req)->syn_ect_snt = inet_sk(sk)->tos & INET_ECN_MASK; __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);

tos = READ_ONCE(inet_sk(sk)->tos); @@ -3448,7 +3449,7 @@ static void __net_init tcp_set_hashinfo(struct net *net)

static int __net_init tcp_sk_init(struct net *net) { - net->ipv4.sysctl_tcp_ecn = 2; + net->ipv4.sysctl_tcp_ecn = TCP_ECN_IN_ECN_OUT_NOECN; net->ipv4.sysctl_tcp_ecn_fallback = 1;

net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS; diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c index 43d7852ce07e..779a206a5ca6 100644 --- a/net/ipv4/tcp_minisocks.c +++ b/net/ipv4/tcp_minisocks.c @@ -461,12 +461,51 @@ void tcp_openreq_init_rwin(struct request_sock *req, ireq->rcv_wscale = rcv_wscale; }

-static void tcp_ecn_openreq_child(struct tcp_sock *tp, - const struct request_sock *req) +void tcp_accecn_third_ack(struct sock *sk, const struct sk_buff *skb, + u8 syn_ect_snt) { - tcp_ecn_mode_set(tp, inet_rsk(req)->ecn_ok ? - TCP_ECN_MODE_RFC3168 : - TCP_ECN_DISABLED); + u8 ace = tcp_accecn_ace(tcp_hdr(skb)); + struct tcp_sock *tp = tcp_sk(sk); + + switch (ace) { + case 0x0: + tcp_accecn_fail_mode_set(tp, TCP_ACCECN_ACE_FAIL_RECV); + break; + case 0x7: + case 0x5: + case 0x1: + /* Unused but legal values */ + break; + default: + /* Validation only applies to first non-data packet */ + if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq && + !TCP_SKB_CB(skb)->sacked && + tcp_accecn_validate_syn_feedback(sk, ace, syn_ect_snt)) { + if ((tcp_accecn_extract_syn_ect(ace) == INET_ECN_CE) && + !tp->delivered_ce) + tp->delivered_ce++; + } + break; + } +} + +static void tcp_ecn_openreq_child(struct sock *sk, + const struct request_sock *req, + const struct sk_buff *skb) +{ + const struct tcp_request_sock *treq = tcp_rsk(req); + struct tcp_sock *tp = tcp_sk(sk); + + if (treq->accecn_ok) { + tcp_ecn_mode_set(tp, TCP_ECN_MODE_ACCECN); + tp->syn_ect_snt = treq->syn_ect_snt; + tcp_accecn_third_ack(sk, skb, treq->syn_ect_snt); + tcp_ecn_received_counters(sk, skb); + } else { + tcp_ecn_mode_set(tp, inet_rsk(req)->ecn_ok ? + TCP_ECN_MODE_RFC3168 : + TCP_ECN_DISABLED); + } }

void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst) @@ -631,7 +670,7 @@ struct sock *tcp_create_openreq_child(const struct sock *sk, if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len) newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len; newtp->rx_opt.mss_clamp = req->mss; - tcp_ecn_openreq_child(newtp, req); + tcp_ecn_openreq_child(newsk, req, skb); newtp->fastopen_req = NULL; RCU_INIT_POINTER(newtp->fastopen_rsk, NULL);

diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c index 4b32aa619594..79ad69dfd9d6 100644 --- a/net/ipv4/tcp_output.c +++ b/net/ipv4/tcp_output.c @@ -322,7 +322,7 @@ static u16 tcp_select_window(struct sock *sk) /* Packet ECN state for a SYN-ACK */ static void tcp_ecn_send_synack(struct sock *sk, struct sk_buff *skb) { - const struct tcp_sock *tp = tcp_sk(sk); + struct tcp_sock *tp = tcp_sk(sk);

TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_CWR; if (tcp_ecn_disabled(tp)) @@ -330,6 +330,13 @@ static void tcp_ecn_send_synack(struct sock *sk, struct sk_buff *skb) else if (tcp_ca_needs_ecn(sk) || tcp_bpf_ca_needs_ecn(sk)) INET_ECN_xmit(sk); + + if (tp->ecn_flags & TCP_ECN_MODE_ACCECN) { + TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_ACE; + TCP_SKB_CB(skb)->tcp_flags |= + tcp_accecn_reflector_flags(tp->syn_ect_rcv); + tp->syn_ect_snt = inet_sk(sk)->tos & INET_ECN_MASK; + } }

/* Packet ECN state for a SYN. */ @@ -337,8 +344,20 @@ static void tcp_ecn_send_syn(struct sock *sk, struct sk_buff *skb) { struct tcp_sock *tp = tcp_sk(sk); bool bpf_needs_ecn = tcp_bpf_ca_needs_ecn(sk); - bool use_ecn = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_ecn) == 1 || - tcp_ca_needs_ecn(sk) || bpf_needs_ecn; + bool use_ecn, use_accecn; + u8 tcp_ecn = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_ecn); + + /* ============== ========================== + * tcp_ecn values Outgoing connections + * ============== ========================== + * 0,2,5 Do not request ECN + * 1,4 Request ECN connection + * 3 Request AccECN connection + * ============== ========================== + */ + use_accecn = tcp_ecn == 3 || tcp_ca_needs_accecn(sk); + use_ecn = tcp_ecn == 1 || tcp_ecn == 4 || + tcp_ca_needs_ecn(sk) || bpf_needs_ecn || use_accecn;

if (!use_ecn) { const struct dst_entry *dst = __sk_dst_get(sk); @@ -354,35 +373,58 @@ static void tcp_ecn_send_syn(struct sock *sk, struct sk_buff *skb) INET_ECN_xmit(sk);

TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ECE | TCPHDR_CWR; - tcp_ecn_mode_set(tp, TCP_ECN_MODE_RFC3168); + if (use_accecn) { + TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_AE; + tcp_ecn_mode_set(tp, TCP_ECN_MODE_PENDING); + tp->syn_ect_snt = inet_sk(sk)->tos & INET_ECN_MASK; + } else { + tcp_ecn_mode_set(tp, TCP_ECN_MODE_RFC3168); + } } }

static void tcp_ecn_clear_syn(struct sock *sk, struct sk_buff *skb) { - if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_ecn_fallback)) + if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_ecn_fallback)) { /* tp->ecn_flags are cleared at a later point in time when * SYN ACK is ultimatively being received. */ - TCP_SKB_CB(skb)->tcp_flags &= ~(TCPHDR_ECE | TCPHDR_CWR); + TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_ACE; + } +} + +static void tcp_accecn_echo_syn_ect(struct tcphdr *th, u8 ect) +{ + th->ae = !!(ect & INET_ECN_ECT_0); + th->cwr = ect != INET_ECN_ECT_0; + th->ece = ect == INET_ECN_ECT_1; }

static void tcp_ecn_make_synack(const struct request_sock *req, struct tcphdr *th) { - if (inet_rsk(req)->ecn_ok) + if (tcp_rsk(req)->accecn_ok) + tcp_accecn_echo_syn_ect(th, tcp_rsk(req)->syn_ect_rcv); + else if (inet_rsk(req)->ecn_ok) th->ece = 1; }

-static void tcp_accecn_set_ace(struct tcphdr *th, struct tcp_sock *tp) +static void tcp_accecn_set_ace(struct tcp_sock *tp, struct sk_buff *skb, + struct tcphdr *th) { 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; + /* The final packet of the 3WHS or anything like it must reflect + * the SYN/ACK ECT instead of putting CEP into ACE field, such + * case show up in tcp_flags. + */ + if (likely(!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_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; + } }

/* Set up ECN state for a packet on a ESTABLISHED socket that is about to @@ -397,8 +439,9 @@ static void tcp_ecn_send(struct sock *sk, struct sk_buff *skb, return;

if (tcp_ecn_mode_accecn(tp)) { - INET_ECN_xmit(sk); - tcp_accecn_set_ace(th, tp); + if (!tcp_accecn_ace_fail_recv(tp)) + INET_ECN_xmit(sk); + tcp_accecn_set_ace(tp, skb, th); skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ACCECN; } else { /* Not-retransmitted data segment: set ECT and inject CWR. */ @@ -3408,7 +3451,10 @@ int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs) tcp_retrans_try_collapse(sk, skb, avail_wnd); }

- /* RFC3168, section 6.1.1.1. ECN fallback */ + /* RFC3168, section 6.1.1.1. ECN fallback + * As AccECN uses the same SYN flags (+ AE), this check covers both + * cases. + */ if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN_ECN) == TCPHDR_SYN_ECN) tcp_ecn_clear_syn(sk, skb);

diff --git a/net/ipv6/syncookies.c b/net/ipv6/syncookies.c index 9d83eadd308b..50046460ee0b 100644 --- a/net/ipv6/syncookies.c +++ b/net/ipv6/syncookies.c @@ -264,6 +264,7 @@ struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb) if (!req->syncookie) ireq->rcv_wscale = rcv_wscale; ireq->ecn_ok &= cookie_ecn_ok(net, dst); + tcp_rsk(req)->accecn_ok = ireq->ecn_ok && cookie_accecn_ok(th);

ret = tcp_get_cookie_sock(sk, skb, req, dst); if (!ret) { diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c index e8e68a142649..eb7dc7fb08cc 100644 --- a/net/ipv6/tcp_ipv6.c +++ b/net/ipv6/tcp_ipv6.c @@ -544,6 +544,7 @@ static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst, skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb);

if (skb) { + tcp_rsk(req)->syn_ect_snt = np->tclass & INET_ECN_MASK; __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr, &ireq->ir_v6_rmt_addr);

From: Ilpo Järvinen ij@kernel.org

AccECN byte counter estimation requires delivered bytes which can be calculated while processing SACK blocks and cumulative ACK. The delivered bytes will be used to estimate the byte counters between AccECN option (on ACKs w/o the option).

Non-SACK calculation is quite annoying, inaccurate, and likely bogus.

Signed-off-by: Ilpo Järvinen ij@kernel.org Signed-off-by: Chia-Yu Chang chia-yu.chang@nokia-bell-labs.com --- net/ipv4/tcp_input.c | 14 ++++++++++++-- 1 file changed, 12 insertions(+), 2 deletions(-)

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index 5c0159cc0083..a327fc14b207 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -1196,6 +1196,7 @@ struct tcp_sacktag_state { u64 last_sackt; u32 reord; u32 sack_delivered; + u32 delivered_bytes; int flag; unsigned int mss_now; struct rate_sample *rate; @@ -1557,7 +1558,7 @@ static int tcp_match_skb_to_sack(struct sock *sk, struct sk_buff *skb, static u8 tcp_sacktag_one(struct sock *sk, struct tcp_sacktag_state *state, u8 sacked, u32 start_seq, u32 end_seq, - int dup_sack, int pcount, + int dup_sack, int pcount, u32 plen, u64 xmit_time) { struct tcp_sock *tp = tcp_sk(sk); @@ -1617,6 +1618,7 @@ static u8 tcp_sacktag_one(struct sock *sk, tp->sacked_out += pcount; /* Out-of-order packets delivered */ state->sack_delivered += pcount; + state->delivered_bytes += plen; }

/* D-SACK. We can detect redundant retransmission in S|R and plain R @@ -1653,7 +1655,7 @@ static bool tcp_shifted_skb(struct sock *sk, struct sk_buff *prev, * tcp_highest_sack_seq() when skb is highest_sack. */ tcp_sacktag_one(sk, state, TCP_SKB_CB(skb)->sacked, - start_seq, end_seq, dup_sack, pcount, + start_seq, end_seq, dup_sack, pcount, skb->len, tcp_skb_timestamp_us(skb)); tcp_rate_skb_delivered(sk, skb, state->rate);

@@ -1938,6 +1940,7 @@ static struct sk_buff *tcp_sacktag_walk(struct sk_buff *skb, struct sock *sk, TCP_SKB_CB(skb)->end_seq, dup_sack, tcp_skb_pcount(skb), + skb->len, tcp_skb_timestamp_us(skb)); tcp_rate_skb_delivered(sk, skb, state->rate); if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) @@ -3446,6 +3449,8 @@ static int tcp_clean_rtx_queue(struct sock *sk, const struct sk_buff *ack_skb,

if (sacked & TCPCB_SACKED_ACKED) { tp->sacked_out -= acked_pcount; + /* snd_una delta covers these skbs */ + sack->delivered_bytes -= skb->len; } else if (tcp_is_sack(tp)) { tcp_count_delivered(tp, acked_pcount, ece_ack); if (!tcp_skb_spurious_retrans(tp, skb)) @@ -3542,6 +3547,10 @@ static int tcp_clean_rtx_queue(struct sock *sk, const struct sk_buff *ack_skb, if (before(reord, prior_fack)) tcp_check_sack_reordering(sk, reord, 0); } + + sack->delivered_bytes = (skb ? + TCP_SKB_CB(skb)->seq : tp->snd_una) - + prior_snd_una; } else if (skb && rtt_update && sack_rtt_us >= 0 && sack_rtt_us > tcp_stamp_us_delta(tp->tcp_mstamp, tcp_skb_timestamp_us(skb))) { @@ -4004,6 +4013,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag) sack_state.first_sackt = 0; sack_state.rate = &rs; sack_state.sack_delivered = 0; + sack_state.delivered_bytes = 0;

/* We very likely will need to access rtx queue. */ prefetch(sk->tcp_rtx_queue.rb_node);

From: Ilpo Järvinen ij@kernel.org

Instead of sending the option in every ACK, limit sending to those ACKs where the option is necessary: - Handshake - "Change-triggered ACK" + the ACK following it. The 2nd ACK is necessary to unambiguously indicate which of the ECN byte counters in increasing. The first ACK has two counters increasing due to the ecnfield edge. - ACKs with CE to allow CEP delta validations to take advantage of the option. - Force option to be sent every at least once per 2^22 bytes. The check is done using the bit edges of the byte counters (avoids need for extra variables). - AccECN option beacon to send a few times per RTT even if nothing in the ECN state requires that. The default is 3 times per RTT, and its period can be set via sysctl_tcp_ecn_option_beacon.

Below are the pahole outcomes before and after this patch, in which the group size of tcp_sock_write_tx is increased from 89 to 97 due to the new u64 accecn_opt_tstamp member:

[BEFORE THIS PATCH] struct tcp_sock { [...] u64 tcp_wstamp_ns; /* 2552 8 */ struct list_head tsorted_sent_queue; /* 2560 16 */

[...] __cacheline_group_end__tcp_sock_write_tx[0]; /* 2585 0 */ __cacheline_group_begin__tcp_sock_write_txrx[0]; /* 2585 0 */ u8 nonagle:4; /* 2585: 0 1 */ u8 rate_app_limited:1; /* 2585: 4 1 */ /* XXX 3 bits hole, try to pack */

/* Force alignment to the next boundary: */ u8 :0; u8 received_ce_pending:4;/* 2586: 0 1 */ u8 unused2:4; /* 2586: 4 1 */ u8 accecn_minlen:2; /* 2587: 0 1 */ u8 est_ecnfield:2; /* 2587: 2 1 */ u8 unused3:4; /* 2587: 4 1 */

[...] __cacheline_group_end__tcp_sock_write_txrx[0]; /* 2692 0 */

[...] /* size: 3264, cachelines: 51, members: 174 */ }

[AFTER THIS PATCH] struct tcp_sock { [...] u64 tcp_wstamp_ns; /* 2552 8 */ u64 accecn_opt_tstamp; /* 2560 8 */ struct list_head tsorted_sent_queue; /* 2568 16 */

[...] __cacheline_group_end__tcp_sock_write_tx[0]; /* 2593 0 */ __cacheline_group_begin__tcp_sock_write_txrx[0]; /* 2593 0 */ u8 nonagle:4; /* 2593: 0 1 */ u8 rate_app_limited:1; /* 2593: 4 1 */ /* XXX 3 bits hole, try to pack */

/* Force alignment to the next boundary: */ u8 :0; u8 received_ce_pending:4;/* 2594: 0 1 */ u8 unused2:4; /* 2594: 4 1 */ u8 accecn_minlen:2; /* 2595: 0 1 */ u8 est_ecnfield:2; /* 2595: 2 1 */ u8 accecn_opt_demand:2; /* 2595: 4 1 */ u8 prev_ecnfield:2; /* 2595: 6 1 */

[...] __cacheline_group_end__tcp_sock_write_txrx[0]; /* 2700 0 */

[...] /* size: 3264, cachelines: 51, members: 176 */ }

Signed-off-by: Ilpo Järvinen ij@kernel.org 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

--- v8: - Add new helper function tcp_accecn_opt_demand_min() - Remove white space only change --- .../networking/net_cachelines/tcp_sock.rst | 3 ++ include/linux/tcp.h | 4 +- include/net/netns/ipv4.h | 1 + include/net/tcp.h | 5 +++ net/ipv4/sysctl_net_ipv4.c | 9 +++++ net/ipv4/tcp.c | 5 ++- net/ipv4/tcp_input.c | 40 ++++++++++++++++++- net/ipv4/tcp_ipv4.c | 1 + net/ipv4/tcp_minisocks.c | 2 + net/ipv4/tcp_output.c | 28 ++++++++++++- 10 files changed, 93 insertions(+), 5 deletions(-)

diff --git a/Documentation/networking/net_cachelines/tcp_sock.rst b/Documentation/networking/net_cachelines/tcp_sock.rst index ed6af7d0110c..88c668f9c68b 100644 --- a/Documentation/networking/net_cachelines/tcp_sock.rst +++ b/Documentation/networking/net_cachelines/tcp_sock.rst @@ -110,6 +110,9 @@ u8:2 syn_ect_rcv read_mostly read_w u8:1 wait_third_ack read_write u8:2 accecn_minlen write_mostly read_write u8:2 est_ecnfield read_write +u8:2 accecn_opt_demand read_mostly read_write +u8:2 prev_ecnfield read_write +u64 accecn_opt_tstamp read_write u8:4 accecn_fail_mode 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) diff --git a/include/linux/tcp.h b/include/linux/tcp.h index 0008b8190ea0..2c0a2db6e6dd 100644 --- a/include/linux/tcp.h +++ b/include/linux/tcp.h @@ -275,6 +275,7 @@ struct tcp_sock { u32 mdev_us; /* medium deviation */ u32 rtt_seq; /* sequence number to update rttvar */ u64 tcp_wstamp_ns; /* departure time for next sent data packet */ + u64 accecn_opt_tstamp; /* Last AccECN option sent timestamp */ struct list_head tsorted_sent_queue; /* time-sorted sent but un-SACKed skbs */ struct sk_buff *highest_sack; /* skb just after the highest * skb with SACKed bit set @@ -296,7 +297,8 @@ struct tcp_sock { unused2:4; u8 accecn_minlen:2,/* Minimum length of AccECN option sent */ est_ecnfield:2,/* ECN field for AccECN delivered estimates */ - unused3:4; + accecn_opt_demand:2,/* Demand AccECN option for n next ACKs */ + prev_ecnfield:2; /* ECN bits from the previous segment */ __be32 pred_flags; u64 tcp_clock_cache; /* cache last tcp_clock_ns() (see tcp_mstamp_refresh()) */ u64 tcp_mstamp; /* most recent packet received/sent */ diff --git a/include/net/netns/ipv4.h b/include/net/netns/ipv4.h index 4569a9ef4fb8..ff8b5b56ad00 100644 --- a/include/net/netns/ipv4.h +++ b/include/net/netns/ipv4.h @@ -149,6 +149,7 @@ struct netns_ipv4 {

u8 sysctl_tcp_ecn; u8 sysctl_tcp_ecn_option; + u8 sysctl_tcp_ecn_option_beacon; u8 sysctl_tcp_ecn_fallback;

u8 sysctl_ip_default_ttl; diff --git a/include/net/tcp.h b/include/net/tcp.h index f7ba314eb7f9..fb0d69ee3ce1 100644 --- a/include/net/tcp.h +++ b/include/net/tcp.h @@ -91,6 +91,9 @@ void tcp_time_wait(struct sock *sk, int state, int timeo); /* Maximal number of window scale according to RFC1323 */ #define TCP_MAX_WSCALE 14U

+/* Default sending frequency of accurate ECN option per RTT */ +#define TCP_ECN_OPTION_BEACON 3 + /* urg_data states */ #define TCP_URG_VALID 0x0100 #define TCP_URG_NOTYET 0x0200 @@ -477,6 +480,7 @@ static inline int tcp_accecn_extract_syn_ect(u8 ace) bool tcp_accecn_validate_syn_feedback(struct sock *sk, u8 ace, u8 sent_ect); void tcp_accecn_third_ack(struct sock *sk, const struct sk_buff *skb, u8 syn_ect_snt); +void tcp_accecn_opt_demand_min(struct sock *sk, u8 opt_demand_min); void tcp_ecn_received_counters_payload(struct sock *sk, const struct sk_buff *skb); void tcp_ecn_received_counters(struct sock *sk, const struct sk_buff *skb, @@ -1090,6 +1094,7 @@ static inline void tcp_accecn_init_counters(struct tcp_sock *tp) __tcp_accecn_init_bytes_counters(tp->received_ecn_bytes); __tcp_accecn_init_bytes_counters(tp->delivered_ecn_bytes); tp->accecn_minlen = 0; + tp->accecn_opt_demand = 0; tp->est_ecnfield = 0; }

diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c index 1d7fd86ca7b9..169a393374b3 100644 --- a/net/ipv4/sysctl_net_ipv4.c +++ b/net/ipv4/sysctl_net_ipv4.c @@ -740,6 +740,15 @@ static struct ctl_table ipv4_net_table[] = { .extra1 = SYSCTL_ZERO, .extra2 = SYSCTL_TWO, }, + { + .procname = "tcp_ecn_option_beacon", + .data = &init_net.ipv4.sysctl_tcp_ecn_option_beacon, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_THREE, + }, { .procname = "tcp_ecn_fallback", .data = &init_net.ipv4.sysctl_tcp_ecn_fallback, diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c index 1c1d4cbdda1c..f7a85192c86f 100644 --- a/net/ipv4/tcp.c +++ b/net/ipv4/tcp.c @@ -3396,6 +3396,8 @@ int tcp_disconnect(struct sock *sk, int flags) tp->wait_third_ack = 0; tp->accecn_fail_mode = 0; tcp_accecn_init_counters(tp); + tp->prev_ecnfield = 0; + tp->accecn_opt_tstamp = 0; 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)); @@ -5114,11 +5116,12 @@ static void __init tcp_struct_check(void) CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_tx, lsndtime); CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_tx, mdev_us); CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_tx, tcp_wstamp_ns); + CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_tx, accecn_opt_tstamp); CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_tx, rtt_seq); CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_tx, tsorted_sent_queue); CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_tx, highest_sack); CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_tx, ecn_flags); - CACHELINE_ASSERT_GROUP_SIZE(struct tcp_sock, tcp_sock_write_tx, 89); + CACHELINE_ASSERT_GROUP_SIZE(struct tcp_sock, tcp_sock_write_tx, 97);

/* TXRX read-write hotpath cache lines */ CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_txrx, pred_flags); diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index 66390966cc41..3cc9c57d9983 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -497,6 +497,7 @@ static void tcp_ecn_rcv_synack(struct sock *sk, const struct tcphdr *th, default: tcp_ecn_mode_set(tp, TCP_ECN_MODE_ACCECN); tp->syn_ect_rcv = ip_dsfield & INET_ECN_MASK; + tp->accecn_opt_demand = 2; if (INET_ECN_is_ce(ip_dsfield) && tcp_accecn_validate_syn_feedback(sk, ace, tp->syn_ect_snt)) { @@ -517,6 +518,7 @@ static void tcp_ecn_rcv_syn(struct tcp_sock *tp, const struct tcphdr *th, } else { tp->syn_ect_rcv = TCP_SKB_CB(skb)->ip_dsfield & INET_ECN_MASK; + tp->prev_ecnfield = tp->syn_ect_rcv; tcp_ecn_mode_set(tp, TCP_ECN_MODE_ACCECN); } } @@ -6182,6 +6184,15 @@ void tcp_ecn_received_counters_payload(struct sock *sk, tcp_ecn_received_counters(sk, skb, skb->len - th->doff * 4); }

+void tcp_accecn_opt_demand_min(struct sock *sk, u8 opt_demand_min) +{ + struct tcp_sock *tp = tcp_sk(sk); + u8 opt_demand; + + opt_demand = max_t(u8, opt_demand_min, tp->accecn_opt_demand); + tp->accecn_opt_demand = opt_demand; +} + /* Updates Accurate ECN received counters from the received IP ECN field */ void tcp_ecn_received_counters(struct sock *sk, const struct sk_buff *skb, u32 payload_len) @@ -6189,6 +6200,7 @@ 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); + bool ecn_edge;

if (!INET_ECN_is_not_ect(ecnfield)) { u32 pcount = is_ce * max_t(u16, 1, skb_shinfo(skb)->gso_segs); @@ -6207,9 +6219,33 @@ void tcp_ecn_received_counters(struct sock *sk, const struct sk_buff *skb,

if (payload_len > 0) { u8 minlen = tcp_ecnfield_to_accecn_optfield(ecnfield); + u32 oldbytes = tp->received_ecn_bytes[ecnfield - 1]; + tp->received_ecn_bytes[ecnfield - 1] += payload_len; tp->accecn_minlen = max_t(u8, tp->accecn_minlen, minlen); + + /* Demand AccECN option at least every 2^22 bytes to + * avoid overflowing the ECN byte counters. + */ + if ((tp->received_ecn_bytes[ecnfield - 1] ^ oldbytes) & + ~((1 << 22) - 1)) { + tcp_accecn_opt_demand_min(sk, 1); + } + } + } + + ecn_edge = tp->prev_ecnfield != ecnfield; + if (ecn_edge || is_ce) { + tp->prev_ecnfield = ecnfield; + /* Demand Accurate ECN change-triggered ACKs. Two ACK are + * demanded to indicate unambiguously the ecnfield value + * in the latter ACK. + */ + if (tcp_ecn_mode_accecn(tp)) { + if (ecn_edge) + inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_NOW; + tp->accecn_opt_demand = 2; } } } @@ -6342,8 +6378,10 @@ static bool tcp_validate_incoming(struct sock *sk, struct sk_buff *skb, * RFC 5961 4.2 : Send a challenge ack */ if (th->syn) { - if (tcp_ecn_mode_accecn(tp)) + if (tcp_ecn_mode_accecn(tp)) { accecn_reflector = true; + tcp_accecn_opt_demand_min(sk, 1); + } if (sk->sk_state == TCP_SYN_RECV && sk->sk_socket && th->ack && TCP_SKB_CB(skb)->seq + 1 == TCP_SKB_CB(skb)->end_seq && TCP_SKB_CB(skb)->seq + 1 == tp->rcv_nxt && diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c index 387cf8994202..7d4660378378 100644 --- a/net/ipv4/tcp_ipv4.c +++ b/net/ipv4/tcp_ipv4.c @@ -3451,6 +3451,7 @@ static int __net_init tcp_sk_init(struct net *net) { net->ipv4.sysctl_tcp_ecn = TCP_ECN_IN_ECN_OUT_NOECN; net->ipv4.sysctl_tcp_ecn_option = TCP_ECN_OPTION_FULL; + net->ipv4.sysctl_tcp_ecn_option_beacon = TCP_ECN_OPTION_BEACON; net->ipv4.sysctl_tcp_ecn_fallback = 1;

net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS; diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c index f1e698c323f3..5b0f703f23a5 100644 --- a/net/ipv4/tcp_minisocks.c +++ b/net/ipv4/tcp_minisocks.c @@ -500,6 +500,8 @@ static void tcp_ecn_openreq_child(struct sock *sk, tcp_ecn_mode_set(tp, TCP_ECN_MODE_ACCECN); tp->syn_ect_snt = treq->syn_ect_snt; tcp_accecn_third_ack(sk, skb, treq->syn_ect_snt); + tp->prev_ecnfield = treq->syn_ect_rcv; + tp->accecn_opt_demand = 1; tcp_ecn_received_counters_payload(sk, skb); } else { tcp_ecn_mode_set(tp, inet_rsk(req)->ecn_ok ? diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c index ca36f5d775f7..a4c80be115e5 100644 --- a/net/ipv4/tcp_output.c +++ b/net/ipv4/tcp_output.c @@ -816,8 +816,12 @@ static void tcp_options_write(struct tcphdr *th, struct tcp_sock *tp, *ptr++ = htonl(((e0b & 0xffffff) << 8) | TCPOPT_NOP); } - if (tp) + if (tp) { tp->accecn_minlen = 0; + tp->accecn_opt_tstamp = tp->tcp_mstamp; + if (tp->accecn_opt_demand) + tp->accecn_opt_demand--; + } }

if (unlikely(OPTION_SACK_ADVERTISE & options)) { @@ -1004,6 +1008,18 @@ static int tcp_options_fit_accecn(struct tcp_out_options *opts, int required, return size; }

+static bool tcp_accecn_option_beacon_check(const struct sock *sk) +{ + const struct tcp_sock *tp = tcp_sk(sk); + + if (!sock_net(sk)->ipv4.sysctl_tcp_ecn_option_beacon) + return false; + + return tcp_stamp_us_delta(tp->tcp_mstamp, tp->accecn_opt_tstamp) * + sock_net(sk)->ipv4.sysctl_tcp_ecn_option_beacon >= + (tp->srtt_us >> 3); +} + /* Compute TCP options for SYN packets. This is not the final * network wire format yet. */ @@ -1258,7 +1274,10 @@ static unsigned int tcp_established_options(struct sock *sk, struct sk_buff *skb }

if (tcp_ecn_mode_accecn(tp) && - sock_net(sk)->ipv4.sysctl_tcp_ecn_option) { + sock_net(sk)->ipv4.sysctl_tcp_ecn_option && + (sock_net(sk)->ipv4.sysctl_tcp_ecn_option >= TCP_ECN_OPTION_FULL || + tp->accecn_opt_demand || + tcp_accecn_option_beacon_check(sk))) { int saving = opts->num_sack_blocks > 0 ? 2 : 0; int remaining = MAX_TCP_OPTION_SPACE - size;

@@ -2975,6 +2994,11 @@ static bool tcp_write_xmit(struct sock *sk, unsigned int mss_now, int nonagle, sent_pkts = 0;

tcp_mstamp_refresh(tp); + + /* AccECN option beacon depends on mstamp, it may change mss */ + if (tcp_ecn_mode_accecn(tp) && tcp_accecn_option_beacon_check(sk)) + mss_now = tcp_current_mss(sk); + if (!push_one) { /* Do MTU probing. */ result = tcp_mtu_probe(sk);

From: Ilpo Järvinen ij@kernel.org

The heuristic algorithm from draft-11 Appendix A.2.2 to mitigate against false ACE field overflows.

Signed-off-by: Ilpo Järvinen ij@kernel.org Signed-off-by: Chia-Yu Chang chia-yu.chang@nokia-bell-labs.com --- include/net/tcp.h | 1 + net/ipv4/tcp_input.c | 18 ++++++++++++++++-- 2 files changed, 17 insertions(+), 2 deletions(-)

diff --git a/include/net/tcp.h b/include/net/tcp.h index 7ee85f6b733a..a122784d844b 100644 --- a/include/net/tcp.h +++ b/include/net/tcp.h @@ -247,6 +247,7 @@ static_assert((1 << ATO_BITS) > TCP_DELACK_MAX); #define TCP_ACCECN_MAXSIZE (TCPOLEN_ACCECN_BASE + \ TCPOLEN_ACCECN_PERFIELD * \ TCP_ACCECN_NUMFIELDS) +#define TCP_ACCECN_SAFETY_SHIFT 1 /* SAFETY_FACTOR in accecn draft */

/* tp->accecn_fail_mode */ #define TCP_ACCECN_ACE_FAIL_SEND BIT(0) diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index f415ed549339..3f8cac22efda 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -719,16 +719,19 @@ static u32 __tcp_accecn_process(struct sock *sk, const struct sk_buff *skb, u32 delivered_pkts, u32 delivered_bytes, int flag) { + u32 old_ceb = tcp_sk(sk)->delivered_ecn_bytes[INET_ECN_CE - 1]; const struct tcphdr *th = tcp_hdr(skb); struct tcp_sock *tp = tcp_sk(sk); - u32 delta, safe_delta; + u32 delta, safe_delta, d_ceb; + bool opt_deltas_valid; 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;

- tcp_accecn_process_option(tp, skb, delivered_bytes, flag); + opt_deltas_valid = tcp_accecn_process_option(tp, skb, + delivered_bytes, flag);

if (!(flag & FLAG_SLOWPATH)) { /* AccECN counter might overflow on large ACKs */ @@ -751,6 +754,17 @@ static u32 __tcp_accecn_process(struct sock *sk, const struct sk_buff *skb, safe_delta = delivered_pkts - ((delivered_pkts - delta) & TCP_ACCECN_CEP_ACE_MASK);

+ if (opt_deltas_valid) { + d_ceb = tp->delivered_ecn_bytes[INET_ECN_CE - 1] - old_ceb; + if (!d_ceb) + return delta; + if (d_ceb > delta * tp->mss_cache) + return safe_delta; + if (d_ceb < + safe_delta * tp->mss_cache >> TCP_ACCECN_SAFETY_SHIFT) + return delta; + } + return safe_delta; }

From: Chia-Yu Chang chia-yu.chang@nokia-bell-labs.com

As SACK blocks tend to eat all option space when there are many holes, it is useful to compromise on sending many SACK blocks in every ACK and attempt to fit the AccECN option there by reducing the number of SACK blocks. However, it will never go below two SACK blocks because of the AccECN option.

As the AccECN option is often not put to every ACK, the space hijack is usually only temporary. Depending on the length of AccECN option (can be either 11, 8, 5, or 2 bytes, cf. Table 5 in AccECN spec) and the NOPs used for alignment of other TCP options, up to two SACK blocks will be reduced.

Co-developed-by: Ilpo Järvinen ij@kernel.org Signed-off-by: Ilpo Järvinen ij@kernel.org Signed-off-by: Chia-Yu Chang chia-yu.chang@nokia-bell-labs.com

--- v8: - Update tcp_options_fit_accecn() to avoid using recursion --- net/ipv4/tcp_output.c | 22 ++++++++++++++++++++-- 1 file changed, 20 insertions(+), 2 deletions(-)

diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c index 4a87380037cb..001c883c2940 100644 --- a/net/ipv4/tcp_output.c +++ b/net/ipv4/tcp_output.c @@ -984,23 +984,41 @@ static int tcp_options_fit_accecn(struct tcp_out_options *opts, int required, int remaining, int max_combine_saving) { int size = TCP_ACCECN_MAXSIZE; + int sack_blocks_reduce = 0; + int rem = remaining;

opts->num_accecn_fields = TCP_ACCECN_NUMFIELDS; - while (opts->num_accecn_fields >= required) { int leftover_size = size & 0x3; /* Pad to dword if cannot combine */ if (leftover_size > max_combine_saving) leftover_size = -((4 - leftover_size) & 0x3);

- if (remaining >= size - leftover_size) { + if (rem >= size - leftover_size) { size -= leftover_size; break; + } else if (opts->num_accecn_fields == required && + opts->num_sack_blocks > 2 && + required > 0) { + /* Try to fit the option by removing one SACK block */ + opts->num_sack_blocks--; + sack_blocks_reduce++; + rem = rem + TCPOLEN_SACK_PERBLOCK; + + opts->num_accecn_fields = TCP_ACCECN_NUMFIELDS; + size = TCP_ACCECN_MAXSIZE; + continue; }

opts->num_accecn_fields--; size -= TCPOLEN_ACCECN_PERFIELD; } + if (sack_blocks_reduce > 0) { + if (opts->num_accecn_fields >= required) + size -= sack_blocks_reduce * TCPOLEN_SACK_PERBLOCK; + else + opts->num_sack_blocks += sack_blocks_reduce; + } if (opts->num_accecn_fields < required) return 0;

From: Ilpo Järvinen ij@kernel.org

Add newly acked pkts EWMA. When ACK thinning occurs, select between safer and unsafe cep delta in AccECN processing based on it. If the packets ACKed per ACK tends to be large, don't conservatively assume ACE field overflow.

This patch uses the existing 2-byte holes in the rx group for new u16 variables withtout creating more holes. Below are the pahole outcomes before and after this patch:

[BEFORE THIS PATCH] struct tcp_sock { [...] u32 delivered_ecn_bytes[3]; /* 2744 12 */ /* XXX 4 bytes hole, try to pack */

[...] __cacheline_group_end__tcp_sock_write_rx[0]; /* 2816 0 */

[...] /* size: 3264, cachelines: 51, members: 177 */ }

[AFTER THIS PATCH] struct tcp_sock { [...] u32 delivered_ecn_bytes[3]; /* 2744 12 */ u16 pkts_acked_ewma; /* 2756 2 */ /* XXX 2 bytes hole, try to pack */

[...] __cacheline_group_end__tcp_sock_write_rx[0]; /* 2816 0 */

[...] /* size: 3264, cachelines: 51, members: 178 */ }

Signed-off-by: Ilpo Järvinen ij@kernel.org 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 --- .../networking/net_cachelines/tcp_sock.rst | 1 + include/linux/tcp.h | 1 + net/ipv4/tcp.c | 2 ++ net/ipv4/tcp_input.c | 20 ++++++++++++++++++- 4 files changed, 23 insertions(+), 1 deletion(-)

diff --git a/Documentation/networking/net_cachelines/tcp_sock.rst b/Documentation/networking/net_cachelines/tcp_sock.rst index 88c668f9c68b..f431ff1eb932 100644 --- a/Documentation/networking/net_cachelines/tcp_sock.rst +++ b/Documentation/networking/net_cachelines/tcp_sock.rst @@ -105,6 +105,7 @@ u32 received_ce read_mostly read_w u32[3] received_ecn_bytes read_mostly read_write u8:4 received_ce_pending read_mostly read_write u32[3] delivered_ecn_bytes read_write +u16 pkts_acked_ewma read_write u8:2 syn_ect_snt write_mostly read_write u8:2 syn_ect_rcv read_mostly read_write u8:1 wait_third_ack read_write diff --git a/include/linux/tcp.h b/include/linux/tcp.h index 8609fb138108..c3b9ecc3f2e2 100644 --- a/include/linux/tcp.h +++ b/include/linux/tcp.h @@ -345,6 +345,7 @@ struct tcp_sock { u32 rate_interval_us; /* saved rate sample: time elapsed */ u32 rcv_rtt_last_tsecr; u32 delivered_ecn_bytes[3]; + u16 pkts_acked_ewma;/* Pkts acked EWMA for AccECN cep heuristic */ u64 first_tx_mstamp; /* start of window send phase */ u64 delivered_mstamp; /* time we reached "delivered" */ u64 bytes_acked; /* RFC4898 tcpEStatsAppHCThruOctetsAcked diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c index a9e30c72d71b..40fdfba1c04a 100644 --- a/net/ipv4/tcp.c +++ b/net/ipv4/tcp.c @@ -3399,6 +3399,7 @@ int tcp_disconnect(struct sock *sk, int flags) tcp_accecn_init_counters(tp); tp->prev_ecnfield = 0; tp->accecn_opt_tstamp = 0; + tp->pkts_acked_ewma = 0; 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)); @@ -5159,6 +5160,7 @@ static void __init tcp_struct_check(void) CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_rx, rate_interval_us); CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_rx, rcv_rtt_last_tsecr); CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_rx, delivered_ecn_bytes); + CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_rx, pkts_acked_ewma); CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_rx, first_tx_mstamp); CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_rx, delivered_mstamp); CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_rx, bytes_acked); diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index 97b5fe52d8fd..f15ebee4a641 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -714,6 +714,10 @@ static void tcp_count_delivered(struct tcp_sock *tp, u32 delivered, tcp_count_delivered_ce(tp, delivered); }

+#define PKTS_ACKED_WEIGHT 6 +#define PKTS_ACKED_PREC 6 +#define ACK_COMP_THRESH 4 + /* Returns the ECN CE delta */ static u32 __tcp_accecn_process(struct sock *sk, const struct sk_buff *skb, u32 delivered_pkts, u32 delivered_bytes, @@ -733,6 +737,19 @@ static u32 __tcp_accecn_process(struct sock *sk, const struct sk_buff *skb, opt_deltas_valid = tcp_accecn_process_option(tp, skb, delivered_bytes, flag);

+ if (delivered_pkts) { + if (!tp->pkts_acked_ewma) { + tp->pkts_acked_ewma = delivered_pkts << PKTS_ACKED_PREC; + } else { + u32 ewma = tp->pkts_acked_ewma; + + ewma = (((ewma << PKTS_ACKED_WEIGHT) - ewma) + + (delivered_pkts << PKTS_ACKED_PREC)) >> + PKTS_ACKED_WEIGHT; + tp->pkts_acked_ewma = min_t(u32, ewma, 0xFFFFU); + } + } + if (!(flag & FLAG_SLOWPATH)) { /* AccECN counter might overflow on large ACKs */ if (delivered_pkts <= TCP_ACCECN_CEP_ACE_MASK) @@ -781,7 +798,8 @@ static u32 __tcp_accecn_process(struct sock *sk, const struct sk_buff *skb, if (d_ceb < safe_delta * tp->mss_cache >> TCP_ACCECN_SAFETY_SHIFT) return delta; - } + } else if (tp->pkts_acked_ewma > (ACK_COMP_THRESH << PKTS_ACKED_PREC)) + return delta;

return safe_delta; }