From: Ilpo Järvinen ij@kernel.org

Prepare for AccECN that needs to have access here on IP ECN field value which is only available after INET_ECN_xmit().

No functional changes.

Signed-off-by: Ilpo Järvinen ij@kernel.org Signed-off-by: Chia-Yu Chang chia-yu.chang@nokia-bell-labs.com Reviewed-by: Eric Dumazet edumazet@google.com --- net/ipv4/tcp_output.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c index 3ac8d2d17e1f..2ede3c190293 100644 --- a/net/ipv4/tcp_output.c +++ b/net/ipv4/tcp_output.c @@ -350,10 +350,11 @@ static void tcp_ecn_send_syn(struct sock *sk, struct sk_buff *skb) tp->ecn_flags = 0;

if (use_ecn) { - TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ECE | TCPHDR_CWR; - tcp_ecn_mode_set(tp, TCP_ECN_MODE_RFC3168); if (tcp_ca_needs_ecn(sk) || bpf_needs_ecn) INET_ECN_xmit(sk); + + TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ECE | TCPHDR_CWR; + tcp_ecn_mode_set(tp, TCP_ECN_MODE_RFC3168); } }

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 29f59d50dc73..191e69e274a4 100644 --- a/include/linux/tcp.h +++ b/include/linux/tcp.h @@ -286,6 +286,8 @@ 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 */ @@ -304,8 +306,6 @@ 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 f64f8276a73c..86872ff89a27 100644 --- a/net/ipv4/tcp.c +++ b/net/ipv4/tcp.c @@ -5126,7 +5126,7 @@ 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); @@ -5143,7 +5143,7 @@ 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

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 9bf507c349a0..eef8700085de 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 4e33ac298191..c25110f36756 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; @@ -376,7 +380,9 @@ 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 */ @@ -392,6 +398,7 @@ 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 cc28255deef7..5fce4b49e132 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 ec91c63d2ea6..16e8f866f984 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 13d9c6f3d2cc..f14eb93cdb80 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -401,14 +401,114 @@ 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); } @@ -3828,8 +3928,18 @@ 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); @@ -3859,7 +3969,7 @@ 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); } }

@@ -4026,7 +4136,7 @@ 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; @@ -6021,8 +6131,7 @@ 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); @@ -6064,6 +6173,7 @@ 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. */ @@ -6156,7 +6266,7 @@ 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; } @@ -6167,16 +6277,23 @@ 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; } @@ -6389,6 +6506,12 @@ 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); @@ -6641,7 +6764,8 @@ 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); @@ -6713,7 +6837,7 @@ 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; }

@@ -6772,7 +6896,7 @@ 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); @@ -6954,7 +7078,7 @@ 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; } @@ -7001,9 +7125,16 @@ 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: { @@ -7173,6 +7304,15 @@ 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;

@@ -7180,7 +7320,8 @@ 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; @@ -7198,6 +7339,9 @@ 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 aba9e2eeeb59..5f3243612044 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. */ @@ -3414,7 +3457,10 @@ 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 ccaa768db145..3e7834bd7d68 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -1186,6 +1186,7 @@ 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; @@ -1547,7 +1548,7 @@ 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); @@ -1607,6 +1608,7 @@ 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;

/* Lost marker hint past SACKed? Tweak RFC3517 cnt */ if (tp->lost_skb_hint && @@ -1648,7 +1650,7 @@ 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);

@@ -1940,6 +1942,7 @@ 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) @@ -3556,6 +3559,8 @@ 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)) @@ -3659,6 +3664,10 @@ static int tcp_clean_rtx_queue(struct sock *sk, const struct sk_buff *ack_skb, delta = prior_sacked - tp->sacked_out; tp->lost_cnt_hint -= min(tp->lost_cnt_hint, delta); } + + 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))) { @@ -4121,6 +4130,7 @@ 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

The Accurate ECN allows echoing back the sum of bytes for each IP ECN field value in the received packets using AccECN option. This change implements AccECN option tx & rx side processing without option send control related features that are added by a later change.

Based on specification: https://tools.ietf.org/id/draft-ietf-tcpm-accurate-ecn-28.txt (Some features of the spec will be added in the later changes rather than in this one).

A full-length AccECN option is always attempted but if it does not fit, the minimum length is selected based on the counters that have changed since the last update. The AccECN option (with 24-bit fields) often ends in odd sizes so the option write code tries to take advantage of some nop used to pad the other TCP options.

The delivered_ecn_bytes pairs with received_ecn_bytes similar to how delivered_ce pairs with received_ce. In contrast to ACE field, however, the option is not always available to update delivered_ecn_bytes. For ACK w/o AccECN option, the delivered bytes calculated based on the cumulative ACK+SACK information are assigned to one of the counters using an estimation heuristic to select the most likely ECN byte counter. Any estimation error is corrected when the next AccECN option arrives. It may occur that the heuristic gets too confused when there are enough different byte counter deltas between ACKs with the AccECN option in which case the heuristic just gives up on updating the counters for a while.

tcp_ecn_option sysctl can be used to select option sending mode for AccECN: TCP_ECN_OPTION_DISABLED, TCP_ECN_OPTION_MINIMUM, and TCP_ECN_OPTION_FULL.

This patch increases the size of tcp_info struct, as there is no existing holes for new u32 variables. Below are the pahole outcomes before and after this patch:

[BEFORE THIS PATCH] struct tcp_info { [...] __u32 tcpi_total_rto_time; /* 244 4 */

/* size: 248, cachelines: 4, members: 61 */ }

[AFTER THIS PATCH] struct tcp_info { [...] __u32 tcpi_total_rto_time; /* 244 4 */ __u32 tcpi_received_ce; /* 248 4 */ __u32 tcpi_delivered_e1_bytes; /* 252 4 */ __u32 tcpi_delivered_e0_bytes; /* 256 4 */ __u32 tcpi_delivered_ce_bytes; /* 260 4 */ __u32 tcpi_received_e1_bytes; /* 264 4 */ __u32 tcpi_received_e0_bytes; /* 268 4 */ __u32 tcpi_received_ce_bytes; /* 272 4 */

/* size: 280, cachelines: 5, members: 68 */ }

This patch uses the existing 1-byte holes in the tcp_sock_write_txrx group for new u8 members, but adds a 4-byte hole in tcp_sock_write_rx group after the new u32 delivered_ecn_bytes[3] member. Therefore, the group size of tcp_sock_write_rx is increased from 96 to 112. Below are the pahole outcomes before and after this patch:

[BEFORE THIS PATCH] struct tcp_sock { [...] u8 received_ce_pending:4; /* 2586: 0 1 */ u8 unused2:4; /* 2586: 4 1 */ /* XXX 1 byte hole, try to pack */

[...] u32 rcv_rtt_last_tsecr; /* 2732 4 */

[...] __cacheline_group_end__tcp_sock_write_rx[0]; /* 2792 0 */

[...] /* size: 3264, cachelines: 51, members: 170 */ }

[AFTER THIS PATCH] struct tcp_sock { [...] 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 */

[...] u32 rcv_rtt_last_tsecr; /* 2732 4 */ u32 delivered_ecn_bytes[3];/* 2736 12 */ /* XXX 4 bytes hole, try to pack */

[...] __cacheline_group_end__tcp_sock_write_rx[0]; /* 2808 0 */

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

Signed-off-by: Ilpo Järvinen ij@kernel.org Signed-off-by: Neal Cardwell ncardwell@google.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

--- v8: - Reset leftover_size to 2 once leftover_bytes is used --- .../networking/net_cachelines/tcp_sock.rst | 3 + include/linux/tcp.h | 9 +- include/net/netns/ipv4.h | 1 + include/net/tcp.h | 20 +++ include/uapi/linux/tcp.h | 7 + net/ipv4/sysctl_net_ipv4.c | 9 + net/ipv4/tcp.c | 15 +- net/ipv4/tcp_input.c | 166 +++++++++++++++++- net/ipv4/tcp_ipv4.c | 1 + net/ipv4/tcp_output.c | 152 +++++++++++++++- 10 files changed, 370 insertions(+), 13 deletions(-)

diff --git a/Documentation/networking/net_cachelines/tcp_sock.rst b/Documentation/networking/net_cachelines/tcp_sock.rst index 64306ae096cd..c74f939f32cc 100644 --- a/Documentation/networking/net_cachelines/tcp_sock.rst +++ b/Documentation/networking/net_cachelines/tcp_sock.rst @@ -104,9 +104,12 @@ u32 delivered_ce read_mostly read_w u32 received_ce read_mostly read_write 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 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 +u8:2 accecn_minlen write_mostly read_write +u8:2 est_ecnfield 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 98af3ce97767..f2fefe6b3643 100644 --- a/include/linux/tcp.h +++ b/include/linux/tcp.h @@ -122,8 +122,9 @@ struct tcp_options_received { smc_ok : 1, /* SMC seen on SYN packet */ snd_wscale : 4, /* Window scaling received from sender */ rcv_wscale : 4; /* Window scaling to send to receiver */ - u8 saw_unknown:1, /* Received unknown option */ - unused:7; + u8 accecn:6, /* AccECN index in header, 0=no options */ + saw_unknown:1, /* Received unknown option */ + unused:1; u8 num_sacks; /* Number of SACK blocks */ u16 user_mss; /* mss requested by user in ioctl */ u16 mss_clamp; /* Maximal mss, negotiated at connection setup */ @@ -294,6 +295,9 @@ struct tcp_sock { rate_app_limited:1; /* rate_{delivered,interval_us} limited? */ u8 received_ce_pending:4, /* Not yet transmit cnt of received_ce */ unused2:4; + u8 accecn_minlen:2,/* Minimum length of AccECN option sent */ + est_ecnfield:2,/* ECN field for AccECN delivered estimates */ + unused3:4; __be32 pred_flags; u64 tcp_clock_cache; /* cache last tcp_clock_ns() (see tcp_mstamp_refresh()) */ u64 tcp_mstamp; /* most recent packet received/sent */ @@ -338,6 +342,7 @@ struct tcp_sock { u32 rate_delivered; /* saved rate sample: packets delivered */ u32 rate_interval_us; /* saved rate sample: time elapsed */ u32 rcv_rtt_last_tsecr; + u32 delivered_ecn_bytes[3]; u64 first_tx_mstamp; /* start of window send phase */ u64 delivered_mstamp; /* time we reached "delivered" */ u64 bytes_acked; /* RFC4898 tcpEStatsAppHCThruOctetsAcked diff --git a/include/net/netns/ipv4.h b/include/net/netns/ipv4.h index 6373e3f17da8..4569a9ef4fb8 100644 --- a/include/net/netns/ipv4.h +++ b/include/net/netns/ipv4.h @@ -148,6 +148,7 @@ struct netns_ipv4 { struct local_ports ip_local_ports;

u8 sysctl_tcp_ecn; + u8 sysctl_tcp_ecn_option; u8 sysctl_tcp_ecn_fallback;

u8 sysctl_ip_default_ttl; diff --git a/include/net/tcp.h b/include/net/tcp.h index 0110df8b5571..88643eb9bbcc 100644 --- a/include/net/tcp.h +++ b/include/net/tcp.h @@ -204,6 +204,8 @@ static_assert((1 << ATO_BITS) > TCP_DELACK_MAX); #define TCPOPT_AO 29 /* Authentication Option (RFC5925) */ #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */ #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ +#define TCPOPT_ACCECN0 172 /* 0xAC: Accurate ECN Order 0 */ +#define TCPOPT_ACCECN1 174 /* 0xAE: Accurate ECN Order 1 */ #define TCPOPT_EXP 254 /* Experimental */ /* Magic number to be after the option value for sharing TCP * experimental options. See draft-ietf-tcpm-experimental-options-00.txt @@ -221,6 +223,7 @@ static_assert((1 << ATO_BITS) > TCP_DELACK_MAX); #define TCPOLEN_TIMESTAMP 10 #define TCPOLEN_MD5SIG 18 #define TCPOLEN_FASTOPEN_BASE 2 +#define TCPOLEN_ACCECN_BASE 2 #define TCPOLEN_EXP_FASTOPEN_BASE 4 #define TCPOLEN_EXP_SMC_BASE 6

@@ -234,6 +237,13 @@ static_assert((1 << ATO_BITS) > TCP_DELACK_MAX); #define TCPOLEN_MD5SIG_ALIGNED 20 #define TCPOLEN_MSS_ALIGNED 4 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 +#define TCPOLEN_ACCECN_PERFIELD 3 + +/* Maximum number of byte counters in AccECN option + size */ +#define TCP_ACCECN_NUMFIELDS 3 +#define TCP_ACCECN_MAXSIZE (TCPOLEN_ACCECN_BASE + \ + TCPOLEN_ACCECN_PERFIELD * \ + TCP_ACCECN_NUMFIELDS)

/* tp->accecn_fail_mode */ #define TCP_ACCECN_ACE_FAIL_SEND BIT(0) @@ -1066,11 +1076,21 @@ enum tcp_ecn_mode { TCP_ECN_IN_ACCECN_OUT_NOECN = 5, };

+/* AccECN option sending when AccECN has been successfully negotiated */ +enum tcp_ecn_option { + TCP_ECN_OPTION_DISABLED = 0, + TCP_ECN_OPTION_MINIMUM = 1, + TCP_ECN_OPTION_FULL = 2, +}; + static inline void tcp_accecn_init_counters(struct tcp_sock *tp) { tp->received_ce = 0; tp->received_ce_pending = 0; __tcp_accecn_init_bytes_counters(tp->received_ecn_bytes); + __tcp_accecn_init_bytes_counters(tp->delivered_ecn_bytes); + tp->accecn_minlen = 0; + tp->est_ecnfield = 0; }

/* State flags for sacked in struct tcp_skb_cb */ diff --git a/include/uapi/linux/tcp.h b/include/uapi/linux/tcp.h index bdac8c42fa82..53e0e85b52be 100644 --- a/include/uapi/linux/tcp.h +++ b/include/uapi/linux/tcp.h @@ -316,6 +316,13 @@ struct tcp_info { * in milliseconds, including any * unfinished recovery. */ + __u32 tcpi_received_ce; /* # of CE marks received */ + __u32 tcpi_delivered_e1_bytes; /* Accurate ECN byte counters */ + __u32 tcpi_delivered_e0_bytes; + __u32 tcpi_delivered_ce_bytes; + __u32 tcpi_received_e1_bytes; + __u32 tcpi_received_e0_bytes; + __u32 tcpi_received_ce_bytes; };

/* netlink attributes types for SCM_TIMESTAMPING_OPT_STATS */ diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c index 75ec1a599b52..1d7fd86ca7b9 100644 --- a/net/ipv4/sysctl_net_ipv4.c +++ b/net/ipv4/sysctl_net_ipv4.c @@ -731,6 +731,15 @@ static struct ctl_table ipv4_net_table[] = { .extra1 = SYSCTL_ZERO, .extra2 = &tcp_ecn_mode_max, }, + { + .procname = "tcp_ecn_option", + .data = &init_net.ipv4.sysctl_tcp_ecn_option, + .maxlen = sizeof(u8), + .mode = 0644, + .proc_handler = proc_dou8vec_minmax, + .extra1 = SYSCTL_ZERO, + .extra2 = SYSCTL_TWO, + }, { .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 632a69126f6f..b0574b2a2e81 100644 --- a/net/ipv4/tcp.c +++ b/net/ipv4/tcp.c @@ -270,6 +270,7 @@

#include <net/icmp.h> #include <net/inet_common.h> +#include <net/inet_ecn.h> #include <net/tcp.h> #include <net/mptcp.h> #include <net/proto_memory.h> @@ -4138,6 +4139,9 @@ void tcp_get_info(struct sock *sk, struct tcp_info *info) { const struct tcp_sock *tp = tcp_sk(sk); /* iff sk_type == SOCK_STREAM */ const struct inet_connection_sock *icsk = inet_csk(sk); + const u8 ect1_idx = INET_ECN_ECT_1 - 1; + const u8 ect0_idx = INET_ECN_ECT_0 - 1; + const u8 ce_idx = INET_ECN_CE - 1; unsigned long rate; u32 now; u64 rate64; @@ -4264,6 +4268,14 @@ void tcp_get_info(struct sock *sk, struct tcp_info *info) if (tp->rto_stamp) info->tcpi_total_rto_time += tcp_clock_ms() - tp->rto_stamp;

+ info->tcpi_received_ce = tp->received_ce; + info->tcpi_delivered_e1_bytes = tp->delivered_ecn_bytes[ect1_idx]; + info->tcpi_delivered_e0_bytes = tp->delivered_ecn_bytes[ect0_idx]; + info->tcpi_delivered_ce_bytes = tp->delivered_ecn_bytes[ce_idx]; + info->tcpi_received_e1_bytes = tp->received_ecn_bytes[ect1_idx]; + info->tcpi_received_e0_bytes = tp->received_ecn_bytes[ect0_idx]; + info->tcpi_received_ce_bytes = tp->received_ecn_bytes[ce_idx]; + unlock_sock_fast(sk, slow); } EXPORT_SYMBOL_GPL(tcp_get_info); @@ -5143,12 +5155,13 @@ static void __init tcp_struct_check(void) CACHELINE_ASSERT_GROUP_MEMBER(struct tcp_sock, tcp_sock_write_rx, rate_delivered); 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, 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); 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, 96); + CACHELINE_ASSERT_GROUP_SIZE(struct tcp_sock, tcp_sock_write_rx, 112); }

void __init tcp_init(void) diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index 3e7834bd7d68..c0898b2a2394 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -70,6 +70,7 @@ #include <linux/sysctl.h> #include <linux/kernel.h> #include <linux/prefetch.h> +#include <linux/bitops.h> #include <net/dst.h> #include <net/tcp.h> #include <net/proto_memory.h> @@ -520,6 +521,139 @@ static bool tcp_ecn_rcv_ecn_echo(const struct tcp_sock *tp, const struct tcphdr return false; }

+/* Maps IP ECN field ECT/CE code point to AccECN option field number, given + * we are sending fields with Accurate ECN Order 1: ECT(1), CE, ECT(0). + */ +static u8 tcp_ecnfield_to_accecn_optfield(u8 ecnfield) +{ + switch (ecnfield & INET_ECN_MASK) { + case INET_ECN_NOT_ECT: + return 0; /* AccECN does not send counts of NOT_ECT */ + case INET_ECN_ECT_1: + return 1; + case INET_ECN_CE: + return 2; + case INET_ECN_ECT_0: + return 3; + } + return 0; +} + +/* Maps IP ECN field ECT/CE code point to AccECN option field value offset. + * Some fields do not start from zero, to detect zeroing by middleboxes. + */ +static u32 tcp_accecn_field_init_offset(u8 ecnfield) +{ + switch (ecnfield & INET_ECN_MASK) { + case INET_ECN_NOT_ECT: + return 0; /* AccECN does not send counts of NOT_ECT */ + case INET_ECN_ECT_1: + return TCP_ACCECN_E1B_INIT_OFFSET; + case INET_ECN_CE: + return TCP_ACCECN_CEB_INIT_OFFSET; + case INET_ECN_ECT_0: + return TCP_ACCECN_E0B_INIT_OFFSET; + } + return 0; +} + +/* Maps AccECN option field #nr to IP ECN field ECT/CE bits */ +static unsigned int tcp_accecn_optfield_to_ecnfield(unsigned int optfield, + bool order) +{ + u8 tmp; + + optfield = order ? 2 - optfield : optfield; + tmp = optfield + 2; + + return (tmp + (tmp >> 2)) & INET_ECN_MASK; +} + +/* Handles AccECN option ECT and CE 24-bit byte counters update into + * the u32 value in tcp_sock. As we're processing TCP options, it is + * safe to access from - 1. + */ +static s32 tcp_update_ecn_bytes(u32 *cnt, const char *from, u32 init_offset) +{ + u32 truncated = (get_unaligned_be32(from - 1) - init_offset) & + 0xFFFFFFU; + u32 delta = (truncated - *cnt) & 0xFFFFFFU; + + /* If delta has the highest bit set (24th bit) indicating + * negative, sign extend to correct an estimation using + * sign_extend32(delta, 24 - 1) + */ + delta = sign_extend32(delta, 23); + *cnt += delta; + return (s32)delta; +} + +/* Returns true if the byte counters can be used */ +static bool tcp_accecn_process_option(struct tcp_sock *tp, + const struct sk_buff *skb, + u32 delivered_bytes, int flag) +{ + u8 estimate_ecnfield = tp->est_ecnfield; + bool ambiguous_ecn_bytes_incr = false; + bool first_changed = false; + unsigned int optlen; + bool order1, res; + unsigned int i; + u8 *ptr; + + if (!(flag & FLAG_SLOWPATH) || !tp->rx_opt.accecn) { + if (estimate_ecnfield) { + u8 ecnfield = estimate_ecnfield - 1; + + tp->delivered_ecn_bytes[ecnfield] += delivered_bytes; + return true; + } + return false; + } + + ptr = skb_transport_header(skb) + tp->rx_opt.accecn; + optlen = ptr[1] - 2; + WARN_ON_ONCE(ptr[0] != TCPOPT_ACCECN0 && ptr[0] != TCPOPT_ACCECN1); + order1 = (ptr[0] == TCPOPT_ACCECN1); + ptr += 2; + + res = !!estimate_ecnfield; + for (i = 0; i < 3; i++) { + if (optlen < TCPOLEN_ACCECN_PERFIELD) + continue; + + u32 init_offset; + u8 ecnfield; + s32 delta; + u32 *cnt; + + ecnfield = tcp_accecn_optfield_to_ecnfield(i, order1); + init_offset = tcp_accecn_field_init_offset(ecnfield); + cnt = &tp->delivered_ecn_bytes[ecnfield - 1]; + delta = tcp_update_ecn_bytes(cnt, ptr, init_offset); + if (delta && delta < 0) { + res = false; + ambiguous_ecn_bytes_incr = true; + } + if (delta && ecnfield != estimate_ecnfield) { + if (!first_changed) { + tp->est_ecnfield = ecnfield; + first_changed = true; + } else { + res = false; + ambiguous_ecn_bytes_incr = true; + } + } + + optlen -= TCPOLEN_ACCECN_PERFIELD; + ptr += TCPOLEN_ACCECN_PERFIELD; + } + if (ambiguous_ecn_bytes_incr) + tp->est_ecnfield = 0; + + return res; +} + static void tcp_count_delivered_ce(struct tcp_sock *tp, u32 ecn_count) { tp->delivered_ce += ecn_count; @@ -536,7 +670,8 @@ static void tcp_count_delivered(struct tcp_sock *tp, u32 delivered,

/* Returns the ECN CE delta */ static u32 __tcp_accecn_process(struct sock *sk, const struct sk_buff *skb, - u32 delivered_pkts, int flag) + u32 delivered_pkts, u32 delivered_bytes, + int flag) { const struct tcphdr *th = tcp_hdr(skb); struct tcp_sock *tp = tcp_sk(sk); @@ -547,6 +682,8 @@ static u32 __tcp_accecn_process(struct sock *sk, const struct sk_buff *skb, if (!(flag & (FLAG_FORWARD_PROGRESS | FLAG_TS_PROGRESS))) return 0;

+ tcp_accecn_process_option(tp, skb, delivered_bytes, flag); + if (!(flag & FLAG_SLOWPATH)) { /* AccECN counter might overflow on large ACKs */ if (delivered_pkts <= TCP_ACCECN_CEP_ACE_MASK) @@ -572,12 +709,14 @@ static u32 __tcp_accecn_process(struct sock *sk, const struct sk_buff *skb, }

static u32 tcp_accecn_process(struct sock *sk, const struct sk_buff *skb, - u32 delivered_pkts, int *flag) + u32 delivered_pkts, u32 delivered_bytes, + int *flag) { struct tcp_sock *tp = tcp_sk(sk); u32 delta;

- delta = __tcp_accecn_process(sk, skb, delivered_pkts, *flag); + delta = __tcp_accecn_process(sk, skb, delivered_pkts, + delivered_bytes, *flag); if (delta > 0) { tcp_count_delivered_ce(tp, delta); *flag |= FLAG_ECE; @@ -4236,6 +4375,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag) if (tcp_ecn_mode_accecn(tp)) ecn_count = tcp_accecn_process(sk, skb, tp->delivered - delivered, + sack_state.delivered_bytes, &flag);

tcp_in_ack_event(sk, flag); @@ -4275,6 +4415,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag) if (tcp_ecn_mode_accecn(tp)) ecn_count = tcp_accecn_process(sk, skb, tp->delivered - delivered, + sack_state.delivered_bytes, &flag); tcp_in_ack_event(sk, flag); /* If data was DSACKed, see if we can undo a cwnd reduction. */ @@ -4402,6 +4543,7 @@ void tcp_parse_options(const struct net *net,

ptr = (const unsigned char *)(th + 1); opt_rx->saw_tstamp = 0; + opt_rx->accecn = 0; opt_rx->saw_unknown = 0;

while (length > 0) { @@ -4493,6 +4635,12 @@ void tcp_parse_options(const struct net *net, ptr, th->syn, foc, false); break;

+ case TCPOPT_ACCECN0: + case TCPOPT_ACCECN1: + /* Save offset of AccECN option in TCP header */ + opt_rx->accecn = (ptr - 2) - (__u8 *)th; + break; + case TCPOPT_EXP: /* Fast Open option shares code 254 using a * 16 bits magic number. @@ -4553,11 +4701,14 @@ static bool tcp_fast_parse_options(const struct net *net, */ if (th->doff == (sizeof(*th) / 4)) { tp->rx_opt.saw_tstamp = 0; + tp->rx_opt.accecn = 0; return false; } else if (tp->rx_opt.tstamp_ok && th->doff == ((sizeof(*th) + TCPOLEN_TSTAMP_ALIGNED) / 4)) { - if (tcp_parse_aligned_timestamp(tp, th)) + if (tcp_parse_aligned_timestamp(tp, th)) { + tp->rx_opt.accecn = 0; return true; + } }

tcp_parse_options(net, skb, &tp->rx_opt, 1, NULL); @@ -6166,8 +6317,12 @@ void tcp_ecn_received_counters(struct sock *sk, const struct sk_buff *skb, tp->received_ce_pending = min(tp->received_ce_pending + pcount, 0xfU);

- if (payload_len > 0) + if (payload_len > 0) { + u8 minlen = tcp_ecnfield_to_accecn_optfield(ecnfield); tp->received_ecn_bytes[ecnfield - 1] += payload_len; + tp->accecn_minlen = max_t(u8, tp->accecn_minlen, + minlen); + } } }

@@ -6387,6 +6542,7 @@ void tcp_rcv_established(struct sock *sk, struct sk_buff *skb) */

tp->rx_opt.saw_tstamp = 0; + tp->rx_opt.accecn = 0;

/* pred_flags is 0xS?10 << 16 + snd_wnd * if header_prediction is to be made diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c index c5b93802d7c0..387cf8994202 100644 --- a/net/ipv4/tcp_ipv4.c +++ b/net/ipv4/tcp_ipv4.c @@ -3450,6 +3450,7 @@ 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 = TCP_ECN_IN_ECN_OUT_NOECN; + net->ipv4.sysctl_tcp_ecn_option = TCP_ECN_OPTION_FULL; net->ipv4.sysctl_tcp_ecn_fallback = 1;

net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS; diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c index 78e78bca0db9..8869e49c07ec 100644 --- a/net/ipv4/tcp_output.c +++ b/net/ipv4/tcp_output.c @@ -492,6 +492,7 @@ static inline bool tcp_urg_mode(const struct tcp_sock *tp) #define OPTION_SMC BIT(9) #define OPTION_MPTCP BIT(10) #define OPTION_AO BIT(11) +#define OPTION_ACCECN BIT(12)

static void smc_options_write(__be32 *ptr, u16 *options) { @@ -513,12 +514,14 @@ struct tcp_out_options { u16 mss; /* 0 to disable */ u8 ws; /* window scale, 0 to disable */ u8 num_sack_blocks; /* number of SACK blocks to include */ + u8 num_accecn_fields; /* number of AccECN fields needed */ u8 hash_size; /* bytes in hash_location */ u8 bpf_opt_len; /* length of BPF hdr option */ __u8 *hash_location; /* temporary pointer, overloaded */ __u32 tsval, tsecr; /* need to include OPTION_TS */ struct tcp_fastopen_cookie *fastopen_cookie; /* Fast open cookie */ struct mptcp_out_options mptcp; + u32 *ecn_bytes; /* AccECN ECT/CE byte counters */ };

static void mptcp_options_write(struct tcphdr *th, __be32 *ptr, @@ -710,6 +713,8 @@ static __be32 *process_tcp_ao_options(struct tcp_sock *tp, return ptr; }

+#define NOP_LEFTOVER ((TCPOPT_NOP << 8) | TCPOPT_NOP) + /* Write previously computed TCP options to the packet. * * Beware: Something in the Internet is very sensitive to the ordering of @@ -728,8 +733,10 @@ static void tcp_options_write(struct tcphdr *th, struct tcp_sock *tp, struct tcp_out_options *opts, struct tcp_key *key) { + u16 leftover_bytes = NOP_LEFTOVER; /* replace next NOPs if avail */ __be32 *ptr = (__be32 *)(th + 1); u16 options = opts->options; /* mungable copy */ + int leftover_size = 2;

if (tcp_key_is_md5(key)) { *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | @@ -763,18 +770,65 @@ static void tcp_options_write(struct tcphdr *th, struct tcp_sock *tp, *ptr++ = htonl(opts->tsecr); }

+ if (OPTION_ACCECN & options) { + const u8 ect0_idx = INET_ECN_ECT_0 - 1; + const u8 ect1_idx = INET_ECN_ECT_1 - 1; + const u8 ce_idx = INET_ECN_CE - 1; + u32 e0b; + u32 e1b; + u32 ceb; + u8 len; + + e0b = opts->ecn_bytes[ect0_idx] + TCP_ACCECN_E0B_INIT_OFFSET; + e1b = opts->ecn_bytes[ect1_idx] + TCP_ACCECN_E1B_INIT_OFFSET; + ceb = opts->ecn_bytes[ce_idx] + TCP_ACCECN_CEB_INIT_OFFSET; + len = TCPOLEN_ACCECN_BASE + + opts->num_accecn_fields * TCPOLEN_ACCECN_PERFIELD; + + if (opts->num_accecn_fields == 2) { + *ptr++ = htonl((TCPOPT_ACCECN1 << 24) | (len << 16) | + ((e1b >> 8) & 0xffff)); + *ptr++ = htonl(((e1b & 0xff) << 24) | + (ceb & 0xffffff)); + } else if (opts->num_accecn_fields == 1) { + *ptr++ = htonl((TCPOPT_ACCECN1 << 24) | (len << 16) | + ((e1b >> 8) & 0xffff)); + leftover_bytes = ((e1b & 0xff) << 8) | + TCPOPT_NOP; + leftover_size = 1; + } else if (opts->num_accecn_fields == 0) { + leftover_bytes = (TCPOPT_ACCECN1 << 8) | len; + leftover_size = 2; + } else if (opts->num_accecn_fields == 3) { + *ptr++ = htonl((TCPOPT_ACCECN1 << 24) | (len << 16) | + ((e1b >> 8) & 0xffff)); + *ptr++ = htonl(((e1b & 0xff) << 24) | + (ceb & 0xffffff)); + *ptr++ = htonl(((e0b & 0xffffff) << 8) | + TCPOPT_NOP); + } + if (tp) + tp->accecn_minlen = 0; + } + if (unlikely(OPTION_SACK_ADVERTISE & options)) { - *ptr++ = htonl((TCPOPT_NOP << 24) | - (TCPOPT_NOP << 16) | + *ptr++ = htonl((leftover_bytes << 16) | (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM); + leftover_bytes = NOP_LEFTOVER; + leftover_size = 2; }

if (unlikely(OPTION_WSCALE & options)) { - *ptr++ = htonl((TCPOPT_NOP << 24) | + u8 highbyte = TCPOPT_NOP; + + if (unlikely(leftover_size == 1)) + highbyte = leftover_bytes >> 8; + *ptr++ = htonl((highbyte << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | opts->ws); + leftover_bytes = NOP_LEFTOVER; }

if (unlikely(opts->num_sack_blocks)) { @@ -782,8 +836,7 @@ static void tcp_options_write(struct tcphdr *th, struct tcp_sock *tp, tp->duplicate_sack : tp->selective_acks; int this_sack;

- *ptr++ = htonl((TCPOPT_NOP << 24) | - (TCPOPT_NOP << 16) | + *ptr++ = htonl((leftover_bytes << 16) | (TCPOPT_SACK << 8) | (TCPOLEN_SACK_BASE + (opts->num_sack_blocks * TCPOLEN_SACK_PERBLOCK))); @@ -795,6 +848,10 @@ static void tcp_options_write(struct tcphdr *th, struct tcp_sock *tp, }

tp->rx_opt.dsack = 0; + } else if (unlikely(leftover_bytes != NOP_LEFTOVER)) { + *ptr++ = htonl((leftover_bytes << 16) | + (TCPOPT_NOP << 8) | + TCPOPT_NOP); }

if (unlikely(OPTION_FAST_OPEN_COOKIE & options)) { @@ -875,6 +932,60 @@ static void mptcp_set_option_cond(const struct request_sock *req, } }

+/* Initial values for AccECN option, ordered is based on ECN field bits + * similar to received_ecn_bytes. Used for SYN/ACK AccECN option. + */ +static u32 synack_ecn_bytes[3] = { 0, 0, 0 }; + +static u32 tcp_synack_options_combine_saving(struct tcp_out_options *opts) +{ + /* How much there's room for combining with the alignment padding? */ + if ((opts->options & (OPTION_SACK_ADVERTISE | OPTION_TS)) == + OPTION_SACK_ADVERTISE) + return 2; + else if (opts->options & OPTION_WSCALE) + return 1; + return 0; +} + +/* Calculates how long AccECN option will fit to @remaining option space. + * + * AccECN option can sometimes replace NOPs used for alignment of other + * TCP options (up to @max_combine_saving available). + * + * Only solutions with at least @required AccECN fields are accepted. + * + * Returns: The size of the AccECN option excluding space repurposed from + * the alignment of the other options. + */ +static int tcp_options_fit_accecn(struct tcp_out_options *opts, int required, + int remaining, int max_combine_saving) +{ + int size = TCP_ACCECN_MAXSIZE; + + 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) { + size -= leftover_size; + break; + } + + opts->num_accecn_fields--; + size -= TCPOLEN_ACCECN_PERFIELD; + } + if (opts->num_accecn_fields < required) + return 0; + + opts->options |= OPTION_ACCECN; + return size; +} + /* Compute TCP options for SYN packets. This is not the final * network wire format yet. */ @@ -957,6 +1068,17 @@ static unsigned int tcp_syn_options(struct sock *sk, struct sk_buff *skb, } }

+ /* Simultaneous open SYN/ACK needs AccECN option but not SYN */ + if (unlikely((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK) && + tcp_ecn_mode_accecn(tp) && + sock_net(sk)->ipv4.sysctl_tcp_ecn_option && + remaining >= TCPOLEN_ACCECN_BASE)) { + u32 saving = tcp_synack_options_combine_saving(opts); + + opts->ecn_bytes = synack_ecn_bytes; + remaining -= tcp_options_fit_accecn(opts, 0, remaining, saving); + } + bpf_skops_hdr_opt_len(sk, skb, NULL, NULL, 0, opts, &remaining);

return MAX_TCP_OPTION_SPACE - remaining; @@ -974,6 +1096,7 @@ static unsigned int tcp_synack_options(const struct sock *sk, { struct inet_request_sock *ireq = inet_rsk(req); unsigned int remaining = MAX_TCP_OPTION_SPACE; + struct tcp_request_sock *treq = tcp_rsk(req);

if (tcp_key_is_md5(key)) { opts->options |= OPTION_MD5; @@ -1036,6 +1159,14 @@ static unsigned int tcp_synack_options(const struct sock *sk,

smc_set_option_cond(tcp_sk(sk), ireq, opts, &remaining);

+ if (treq->accecn_ok && sock_net(sk)->ipv4.sysctl_tcp_ecn_option && + remaining >= TCPOLEN_ACCECN_BASE) { + u32 saving = tcp_synack_options_combine_saving(opts); + + opts->ecn_bytes = synack_ecn_bytes; + remaining -= tcp_options_fit_accecn(opts, 0, remaining, saving); + } + bpf_skops_hdr_opt_len((struct sock *)sk, skb, req, syn_skb, synack_type, opts, &remaining);

@@ -1108,6 +1239,17 @@ static unsigned int tcp_established_options(struct sock *sk, struct sk_buff *skb opts->num_sack_blocks = 0; }

+ if (tcp_ecn_mode_accecn(tp) && + sock_net(sk)->ipv4.sysctl_tcp_ecn_option) { + int saving = opts->num_sack_blocks > 0 ? 2 : 0; + int remaining = MAX_TCP_OPTION_SPACE - size; + + opts->ecn_bytes = tp->received_ecn_bytes; + size += tcp_options_fit_accecn(opts, tp->accecn_minlen, + remaining, + saving); + } + if (unlikely(BPF_SOCK_OPS_TEST_FLAG(tp, BPF_SOCK_OPS_WRITE_HDR_OPT_CB_FLAG))) { unsigned int remaining = MAX_TCP_OPTION_SPACE - size;

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 c74f939f32cc..af76d4f9b33a 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 f2fefe6b3643..5d3743ef5d91 100644 --- a/include/linux/tcp.h +++ b/include/linux/tcp.h @@ -276,6 +276,7 @@ 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 @@ -297,7 +298,8 @@ 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 88643eb9bbcc..96cd8d99ff6f 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 b0574b2a2e81..2514eef12dcd 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)); @@ -5115,11 +5117,12 @@ 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 c0898b2a2394..37c4c7995934 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -487,6 +487,7 @@ 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)) { @@ -507,6 +508,7 @@ 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); } } @@ -6299,6 +6301,15 @@ 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) @@ -6306,6 +6317,7 @@ 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); @@ -6319,9 +6331,33 @@ 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; } } } @@ -6454,8 +6490,10 @@ 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 8869e49c07ec..3de6641c776e 100644 --- a/net/ipv4/tcp_output.c +++ b/net/ipv4/tcp_output.c @@ -807,8 +807,12 @@ 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)) { @@ -986,6 +990,18 @@ 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. */ @@ -1240,7 +1256,10 @@ 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;

@@ -2962,6 +2981,11 @@ 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

Armed with ceb delta from option, delivered bytes, and delivered packets it is possible to estimate how many times ACE field wrapped.

This calculation is necessary only if more than one wrap is possible. Without SACK, delivered bytes and packets are not always trustworthy in which case TCP falls back to the simpler no-or-all wraps ceb algorithm.

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 | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+)

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index b0492f85c4a7..6fb845e9b3aa 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -748,6 +748,24 @@ static u32 __tcp_accecn_process(struct sock *sk, const struct sk_buff *skb, d_ceb = tp->delivered_ecn_bytes[INET_ECN_CE - 1] - old_ceb; if (!d_ceb) return delta; + + if ((delivered_pkts >= (TCP_ACCECN_CEP_ACE_MASK + 1) * 2) && + (tcp_is_sack(tp) || + ((1 << inet_csk(sk)->icsk_ca_state) & + (TCPF_CA_Open | TCPF_CA_CWR)))) { + u32 est_d_cep; + + if (delivered_bytes <= d_ceb) + return safe_delta; + + est_d_cep = DIV_ROUND_UP_ULL((u64)d_ceb * + delivered_pkts, + delivered_bytes); + return min(safe_delta, + delta + + (est_d_cep & ~TCP_ACCECN_CEP_ACE_MASK)); + } + if (d_ceb > delta * tp->mss_cache) return safe_delta; if (d_ceb <

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 af76d4f9b33a..2e6ef8290aff 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 91d60fd4b229..9fa2a1d0e2f4 100644 --- a/include/linux/tcp.h +++ b/include/linux/tcp.h @@ -346,6 +346,7 @@ 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 15286f6c6504..61525b626c68 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)); @@ -5160,6 +5161,7 @@ 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 6fb845e9b3aa..f509e432e2a8 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -704,6 +704,10 @@ 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, @@ -723,6 +727,19 @@ 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) @@ -771,7 +788,8 @@ 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; }