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