1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef __BPF_TCP_HELPERS_H 3 #define __BPF_TCP_HELPERS_H 4 5 #include <stdbool.h> 6 #include <linux/types.h> 7 #include <bpf/bpf_helpers.h> 8 #include <bpf/bpf_core_read.h> 9 #include <bpf/bpf_tracing.h> 10 11 #define BPF_STRUCT_OPS(name, args...) \ 12 SEC("struct_ops/"#name) \ 13 BPF_PROG(name, args) 14 15 #define tcp_jiffies32 ((__u32)bpf_jiffies64()) 16 17 struct sock_common { 18 unsigned char skc_state; 19 __u16 skc_num; 20 } __attribute__((preserve_access_index)); 21 22 enum sk_pacing { 23 SK_PACING_NONE = 0, 24 SK_PACING_NEEDED = 1, 25 SK_PACING_FQ = 2, 26 }; 27 28 struct sock { 29 struct sock_common __sk_common; 30 unsigned long sk_pacing_rate; 31 __u32 sk_pacing_status; /* see enum sk_pacing */ 32 } __attribute__((preserve_access_index)); 33 34 struct inet_sock { 35 struct sock sk; 36 } __attribute__((preserve_access_index)); 37 38 struct inet_connection_sock { 39 struct inet_sock icsk_inet; 40 __u8 icsk_ca_state:6, 41 icsk_ca_setsockopt:1, 42 icsk_ca_dst_locked:1; 43 struct { 44 __u8 pending; 45 } icsk_ack; 46 __u64 icsk_ca_priv[104 / sizeof(__u64)]; 47 } __attribute__((preserve_access_index)); 48 49 struct request_sock { 50 struct sock_common __req_common; 51 } __attribute__((preserve_access_index)); 52 53 struct tcp_sock { 54 struct inet_connection_sock inet_conn; 55 56 __u32 rcv_nxt; 57 __u32 snd_nxt; 58 __u32 snd_una; 59 __u32 window_clamp; 60 __u8 ecn_flags; 61 __u32 delivered; 62 __u32 delivered_ce; 63 __u32 snd_cwnd; 64 __u32 snd_cwnd_cnt; 65 __u32 snd_cwnd_clamp; 66 __u32 snd_ssthresh; 67 __u8 syn_data:1, /* SYN includes data */ 68 syn_fastopen:1, /* SYN includes Fast Open option */ 69 syn_fastopen_exp:1,/* SYN includes Fast Open exp. option */ 70 syn_fastopen_ch:1, /* Active TFO re-enabling probe */ 71 syn_data_acked:1,/* data in SYN is acked by SYN-ACK */ 72 save_syn:1, /* Save headers of SYN packet */ 73 is_cwnd_limited:1,/* forward progress limited by snd_cwnd? */ 74 syn_smc:1; /* SYN includes SMC */ 75 __u32 max_packets_out; 76 __u32 lsndtime; 77 __u32 prior_cwnd; 78 __u64 tcp_mstamp; /* most recent packet received/sent */ 79 } __attribute__((preserve_access_index)); 80 81 static __always_inline struct inet_connection_sock *inet_csk(const struct sock *sk) 82 { 83 return (struct inet_connection_sock *)sk; 84 } 85 86 static __always_inline void *inet_csk_ca(const struct sock *sk) 87 { 88 return (void *)inet_csk(sk)->icsk_ca_priv; 89 } 90 91 static __always_inline struct tcp_sock *tcp_sk(const struct sock *sk) 92 { 93 return (struct tcp_sock *)sk; 94 } 95 96 static __always_inline bool before(__u32 seq1, __u32 seq2) 97 { 98 return (__s32)(seq1-seq2) < 0; 99 } 100 #define after(seq2, seq1) before(seq1, seq2) 101 102 #define TCP_ECN_OK 1 103 #define TCP_ECN_QUEUE_CWR 2 104 #define TCP_ECN_DEMAND_CWR 4 105 #define TCP_ECN_SEEN 8 106 107 enum inet_csk_ack_state_t { 108 ICSK_ACK_SCHED = 1, 109 ICSK_ACK_TIMER = 2, 110 ICSK_ACK_PUSHED = 4, 111 ICSK_ACK_PUSHED2 = 8, 112 ICSK_ACK_NOW = 16 /* Send the next ACK immediately (once) */ 113 }; 114 115 enum tcp_ca_event { 116 CA_EVENT_TX_START = 0, 117 CA_EVENT_CWND_RESTART = 1, 118 CA_EVENT_COMPLETE_CWR = 2, 119 CA_EVENT_LOSS = 3, 120 CA_EVENT_ECN_NO_CE = 4, 121 CA_EVENT_ECN_IS_CE = 5, 122 }; 123 124 struct ack_sample { 125 __u32 pkts_acked; 126 __s32 rtt_us; 127 __u32 in_flight; 128 } __attribute__((preserve_access_index)); 129 130 struct rate_sample { 131 __u64 prior_mstamp; /* starting timestamp for interval */ 132 __u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 133 __s32 delivered; /* number of packets delivered over interval */ 134 long interval_us; /* time for tp->delivered to incr "delivered" */ 135 __u32 snd_interval_us; /* snd interval for delivered packets */ 136 __u32 rcv_interval_us; /* rcv interval for delivered packets */ 137 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 138 int losses; /* number of packets marked lost upon ACK */ 139 __u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 140 __u32 prior_in_flight; /* in flight before this ACK */ 141 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 142 bool is_retrans; /* is sample from retransmission? */ 143 bool is_ack_delayed; /* is this (likely) a delayed ACK? */ 144 } __attribute__((preserve_access_index)); 145 146 #define TCP_CA_NAME_MAX 16 147 #define TCP_CONG_NEEDS_ECN 0x2 148 149 struct tcp_congestion_ops { 150 char name[TCP_CA_NAME_MAX]; 151 __u32 flags; 152 153 /* initialize private data (optional) */ 154 void (*init)(struct sock *sk); 155 /* cleanup private data (optional) */ 156 void (*release)(struct sock *sk); 157 158 /* return slow start threshold (required) */ 159 __u32 (*ssthresh)(struct sock *sk); 160 /* do new cwnd calculation (required) */ 161 void (*cong_avoid)(struct sock *sk, __u32 ack, __u32 acked); 162 /* call before changing ca_state (optional) */ 163 void (*set_state)(struct sock *sk, __u8 new_state); 164 /* call when cwnd event occurs (optional) */ 165 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 166 /* call when ack arrives (optional) */ 167 void (*in_ack_event)(struct sock *sk, __u32 flags); 168 /* new value of cwnd after loss (required) */ 169 __u32 (*undo_cwnd)(struct sock *sk); 170 /* hook for packet ack accounting (optional) */ 171 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 172 /* override sysctl_tcp_min_tso_segs */ 173 __u32 (*min_tso_segs)(struct sock *sk); 174 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 175 __u32 (*sndbuf_expand)(struct sock *sk); 176 /* call when packets are delivered to update cwnd and pacing rate, 177 * after all the ca_state processing. (optional) 178 */ 179 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 180 void *owner; 181 }; 182 183 #define min(a, b) ((a) < (b) ? (a) : (b)) 184 #define max(a, b) ((a) > (b) ? (a) : (b)) 185 #define min_not_zero(x, y) ({ \ 186 typeof(x) __x = (x); \ 187 typeof(y) __y = (y); \ 188 __x == 0 ? __y : ((__y == 0) ? __x : min(__x, __y)); }) 189 190 static __always_inline __u32 tcp_slow_start(struct tcp_sock *tp, __u32 acked) 191 { 192 __u32 cwnd = min(tp->snd_cwnd + acked, tp->snd_ssthresh); 193 194 acked -= cwnd - tp->snd_cwnd; 195 tp->snd_cwnd = min(cwnd, tp->snd_cwnd_clamp); 196 197 return acked; 198 } 199 200 static __always_inline bool tcp_in_slow_start(const struct tcp_sock *tp) 201 { 202 return tp->snd_cwnd < tp->snd_ssthresh; 203 } 204 205 static __always_inline bool tcp_is_cwnd_limited(const struct sock *sk) 206 { 207 const struct tcp_sock *tp = tcp_sk(sk); 208 209 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 210 if (tcp_in_slow_start(tp)) 211 return tp->snd_cwnd < 2 * tp->max_packets_out; 212 213 return !!BPF_CORE_READ_BITFIELD(tp, is_cwnd_limited); 214 } 215 216 static __always_inline void tcp_cong_avoid_ai(struct tcp_sock *tp, __u32 w, __u32 acked) 217 { 218 /* If credits accumulated at a higher w, apply them gently now. */ 219 if (tp->snd_cwnd_cnt >= w) { 220 tp->snd_cwnd_cnt = 0; 221 tp->snd_cwnd++; 222 } 223 224 tp->snd_cwnd_cnt += acked; 225 if (tp->snd_cwnd_cnt >= w) { 226 __u32 delta = tp->snd_cwnd_cnt / w; 227 228 tp->snd_cwnd_cnt -= delta * w; 229 tp->snd_cwnd += delta; 230 } 231 tp->snd_cwnd = min(tp->snd_cwnd, tp->snd_cwnd_clamp); 232 } 233 234 #endif 235