1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Definitions for the TCP protocol. 8 * 9 * Version: @(#)tcp.h 1.0.2 04/28/93 10 * 11 * Author: Fred N. van Kempen, <[email protected]> 12 */ 13 #ifndef _LINUX_TCP_H 14 #define _LINUX_TCP_H 15 16 17 #include <linux/skbuff.h> 18 #include <linux/win_minmax.h> 19 #include <net/sock.h> 20 #include <net/inet_connection_sock.h> 21 #include <net/inet_timewait_sock.h> 22 #include <uapi/linux/tcp.h> 23 24 static inline struct tcphdr *tcp_hdr(const struct sk_buff *skb) 25 { 26 return (struct tcphdr *)skb_transport_header(skb); 27 } 28 29 static inline unsigned int __tcp_hdrlen(const struct tcphdr *th) 30 { 31 return th->doff * 4; 32 } 33 34 static inline unsigned int tcp_hdrlen(const struct sk_buff *skb) 35 { 36 return __tcp_hdrlen(tcp_hdr(skb)); 37 } 38 39 static inline struct tcphdr *inner_tcp_hdr(const struct sk_buff *skb) 40 { 41 return (struct tcphdr *)skb_inner_transport_header(skb); 42 } 43 44 static inline unsigned int inner_tcp_hdrlen(const struct sk_buff *skb) 45 { 46 return inner_tcp_hdr(skb)->doff * 4; 47 } 48 49 static inline unsigned int tcp_optlen(const struct sk_buff *skb) 50 { 51 return (tcp_hdr(skb)->doff - 5) * 4; 52 } 53 54 /* TCP Fast Open */ 55 #define TCP_FASTOPEN_COOKIE_MIN 4 /* Min Fast Open Cookie size in bytes */ 56 #define TCP_FASTOPEN_COOKIE_MAX 16 /* Max Fast Open Cookie size in bytes */ 57 #define TCP_FASTOPEN_COOKIE_SIZE 8 /* the size employed by this impl. */ 58 59 /* TCP Fast Open Cookie as stored in memory */ 60 struct tcp_fastopen_cookie { 61 __le64 val[DIV_ROUND_UP(TCP_FASTOPEN_COOKIE_MAX, sizeof(u64))]; 62 s8 len; 63 bool exp; /* In RFC6994 experimental option format */ 64 }; 65 66 /* This defines a selective acknowledgement block. */ 67 struct tcp_sack_block_wire { 68 __be32 start_seq; 69 __be32 end_seq; 70 }; 71 72 struct tcp_sack_block { 73 u32 start_seq; 74 u32 end_seq; 75 }; 76 77 /*These are used to set the sack_ok field in struct tcp_options_received */ 78 #define TCP_SACK_SEEN (1 << 0) /*1 = peer is SACK capable, */ 79 #define TCP_DSACK_SEEN (1 << 2) /*1 = DSACK was received from peer*/ 80 81 #if IS_ENABLED(CONFIG_MPTCP) 82 struct mptcp_options_received { 83 u64 sndr_key; 84 u64 rcvr_key; 85 u64 data_ack; 86 u64 data_seq; 87 u32 subflow_seq; 88 u16 data_len; 89 u8 mp_capable : 1, 90 mp_join : 1, 91 dss : 1; 92 u8 use_map:1, 93 dsn64:1, 94 data_fin:1, 95 use_ack:1, 96 ack64:1, 97 mpc_map:1, 98 __unused:2; 99 }; 100 #endif 101 102 struct tcp_options_received { 103 /* PAWS/RTTM data */ 104 int ts_recent_stamp;/* Time we stored ts_recent (for aging) */ 105 u32 ts_recent; /* Time stamp to echo next */ 106 u32 rcv_tsval; /* Time stamp value */ 107 u32 rcv_tsecr; /* Time stamp echo reply */ 108 u16 saw_tstamp : 1, /* Saw TIMESTAMP on last packet */ 109 tstamp_ok : 1, /* TIMESTAMP seen on SYN packet */ 110 dsack : 1, /* D-SACK is scheduled */ 111 wscale_ok : 1, /* Wscale seen on SYN packet */ 112 sack_ok : 3, /* SACK seen on SYN packet */ 113 smc_ok : 1, /* SMC seen on SYN packet */ 114 snd_wscale : 4, /* Window scaling received from sender */ 115 rcv_wscale : 4; /* Window scaling to send to receiver */ 116 u8 num_sacks; /* Number of SACK blocks */ 117 u16 user_mss; /* mss requested by user in ioctl */ 118 u16 mss_clamp; /* Maximal mss, negotiated at connection setup */ 119 #if IS_ENABLED(CONFIG_MPTCP) 120 struct mptcp_options_received mptcp; 121 #endif 122 }; 123 124 static inline void tcp_clear_options(struct tcp_options_received *rx_opt) 125 { 126 rx_opt->tstamp_ok = rx_opt->sack_ok = 0; 127 rx_opt->wscale_ok = rx_opt->snd_wscale = 0; 128 #if IS_ENABLED(CONFIG_SMC) 129 rx_opt->smc_ok = 0; 130 #endif 131 #if IS_ENABLED(CONFIG_MPTCP) 132 rx_opt->mptcp.mp_capable = 0; 133 rx_opt->mptcp.mp_join = 0; 134 rx_opt->mptcp.dss = 0; 135 #endif 136 } 137 138 /* This is the max number of SACKS that we'll generate and process. It's safe 139 * to increase this, although since: 140 * size = TCPOLEN_SACK_BASE_ALIGNED (4) + n * TCPOLEN_SACK_PERBLOCK (8) 141 * only four options will fit in a standard TCP header */ 142 #define TCP_NUM_SACKS 4 143 144 struct tcp_request_sock_ops; 145 146 struct tcp_request_sock { 147 struct inet_request_sock req; 148 const struct tcp_request_sock_ops *af_specific; 149 u64 snt_synack; /* first SYNACK sent time */ 150 bool tfo_listener; 151 bool is_mptcp; 152 u32 txhash; 153 u32 rcv_isn; 154 u32 snt_isn; 155 u32 ts_off; 156 u32 last_oow_ack_time; /* last SYNACK */ 157 u32 rcv_nxt; /* the ack # by SYNACK. For 158 * FastOpen it's the seq# 159 * after data-in-SYN. 160 */ 161 }; 162 163 static inline struct tcp_request_sock *tcp_rsk(const struct request_sock *req) 164 { 165 return (struct tcp_request_sock *)req; 166 } 167 168 struct tcp_sock { 169 /* inet_connection_sock has to be the first member of tcp_sock */ 170 struct inet_connection_sock inet_conn; 171 u16 tcp_header_len; /* Bytes of tcp header to send */ 172 u16 gso_segs; /* Max number of segs per GSO packet */ 173 174 /* 175 * Header prediction flags 176 * 0x5?10 << 16 + snd_wnd in net byte order 177 */ 178 __be32 pred_flags; 179 180 /* 181 * RFC793 variables by their proper names. This means you can 182 * read the code and the spec side by side (and laugh ...) 183 * See RFC793 and RFC1122. The RFC writes these in capitals. 184 */ 185 u64 bytes_received; /* RFC4898 tcpEStatsAppHCThruOctetsReceived 186 * sum(delta(rcv_nxt)), or how many bytes 187 * were acked. 188 */ 189 u32 segs_in; /* RFC4898 tcpEStatsPerfSegsIn 190 * total number of segments in. 191 */ 192 u32 data_segs_in; /* RFC4898 tcpEStatsPerfDataSegsIn 193 * total number of data segments in. 194 */ 195 u32 rcv_nxt; /* What we want to receive next */ 196 u32 copied_seq; /* Head of yet unread data */ 197 u32 rcv_wup; /* rcv_nxt on last window update sent */ 198 u32 snd_nxt; /* Next sequence we send */ 199 u32 segs_out; /* RFC4898 tcpEStatsPerfSegsOut 200 * The total number of segments sent. 201 */ 202 u32 data_segs_out; /* RFC4898 tcpEStatsPerfDataSegsOut 203 * total number of data segments sent. 204 */ 205 u64 bytes_sent; /* RFC4898 tcpEStatsPerfHCDataOctetsOut 206 * total number of data bytes sent. 207 */ 208 u64 bytes_acked; /* RFC4898 tcpEStatsAppHCThruOctetsAcked 209 * sum(delta(snd_una)), or how many bytes 210 * were acked. 211 */ 212 u32 dsack_dups; /* RFC4898 tcpEStatsStackDSACKDups 213 * total number of DSACK blocks received 214 */ 215 u32 snd_una; /* First byte we want an ack for */ 216 u32 snd_sml; /* Last byte of the most recently transmitted small packet */ 217 u32 rcv_tstamp; /* timestamp of last received ACK (for keepalives) */ 218 u32 lsndtime; /* timestamp of last sent data packet (for restart window) */ 219 u32 last_oow_ack_time; /* timestamp of last out-of-window ACK */ 220 u32 compressed_ack_rcv_nxt; 221 222 u32 tsoffset; /* timestamp offset */ 223 224 struct list_head tsq_node; /* anchor in tsq_tasklet.head list */ 225 struct list_head tsorted_sent_queue; /* time-sorted sent but un-SACKed skbs */ 226 227 u32 snd_wl1; /* Sequence for window update */ 228 u32 snd_wnd; /* The window we expect to receive */ 229 u32 max_window; /* Maximal window ever seen from peer */ 230 u32 mss_cache; /* Cached effective mss, not including SACKS */ 231 232 u32 window_clamp; /* Maximal window to advertise */ 233 u32 rcv_ssthresh; /* Current window clamp */ 234 235 /* Information of the most recently (s)acked skb */ 236 struct tcp_rack { 237 u64 mstamp; /* (Re)sent time of the skb */ 238 u32 rtt_us; /* Associated RTT */ 239 u32 end_seq; /* Ending TCP sequence of the skb */ 240 u32 last_delivered; /* tp->delivered at last reo_wnd adj */ 241 u8 reo_wnd_steps; /* Allowed reordering window */ 242 #define TCP_RACK_RECOVERY_THRESH 16 243 u8 reo_wnd_persist:5, /* No. of recovery since last adj */ 244 dsack_seen:1, /* Whether DSACK seen after last adj */ 245 advanced:1; /* mstamp advanced since last lost marking */ 246 } rack; 247 u16 advmss; /* Advertised MSS */ 248 u8 compressed_ack; 249 u32 chrono_start; /* Start time in jiffies of a TCP chrono */ 250 u32 chrono_stat[3]; /* Time in jiffies for chrono_stat stats */ 251 u8 chrono_type:2, /* current chronograph type */ 252 rate_app_limited:1, /* rate_{delivered,interval_us} limited? */ 253 fastopen_connect:1, /* FASTOPEN_CONNECT sockopt */ 254 fastopen_no_cookie:1, /* Allow send/recv SYN+data without a cookie */ 255 is_sack_reneg:1, /* in recovery from loss with SACK reneg? */ 256 fastopen_client_fail:2; /* reason why fastopen failed */ 257 u8 nonagle : 4,/* Disable Nagle algorithm? */ 258 thin_lto : 1,/* Use linear timeouts for thin streams */ 259 recvmsg_inq : 1,/* Indicate # of bytes in queue upon recvmsg */ 260 repair : 1, 261 frto : 1;/* F-RTO (RFC5682) activated in CA_Loss */ 262 u8 repair_queue; 263 u8 syn_data:1, /* SYN includes data */ 264 syn_fastopen:1, /* SYN includes Fast Open option */ 265 syn_fastopen_exp:1,/* SYN includes Fast Open exp. option */ 266 syn_fastopen_ch:1, /* Active TFO re-enabling probe */ 267 syn_data_acked:1,/* data in SYN is acked by SYN-ACK */ 268 save_syn:1, /* Save headers of SYN packet */ 269 is_cwnd_limited:1,/* forward progress limited by snd_cwnd? */ 270 syn_smc:1; /* SYN includes SMC */ 271 u32 tlp_high_seq; /* snd_nxt at the time of TLP retransmit. */ 272 273 u32 tcp_tx_delay; /* delay (in usec) added to TX packets */ 274 u64 tcp_wstamp_ns; /* departure time for next sent data packet */ 275 u64 tcp_clock_cache; /* cache last tcp_clock_ns() (see tcp_mstamp_refresh()) */ 276 277 /* RTT measurement */ 278 u64 tcp_mstamp; /* most recent packet received/sent */ 279 u32 srtt_us; /* smoothed round trip time << 3 in usecs */ 280 u32 mdev_us; /* medium deviation */ 281 u32 mdev_max_us; /* maximal mdev for the last rtt period */ 282 u32 rttvar_us; /* smoothed mdev_max */ 283 u32 rtt_seq; /* sequence number to update rttvar */ 284 struct minmax rtt_min; 285 286 u32 packets_out; /* Packets which are "in flight" */ 287 u32 retrans_out; /* Retransmitted packets out */ 288 u32 max_packets_out; /* max packets_out in last window */ 289 u32 max_packets_seq; /* right edge of max_packets_out flight */ 290 291 u16 urg_data; /* Saved octet of OOB data and control flags */ 292 u8 ecn_flags; /* ECN status bits. */ 293 u8 keepalive_probes; /* num of allowed keep alive probes */ 294 u32 reordering; /* Packet reordering metric. */ 295 u32 reord_seen; /* number of data packet reordering events */ 296 u32 snd_up; /* Urgent pointer */ 297 298 /* 299 * Options received (usually on last packet, some only on SYN packets). 300 */ 301 struct tcp_options_received rx_opt; 302 303 /* 304 * Slow start and congestion control (see also Nagle, and Karn & Partridge) 305 */ 306 u32 snd_ssthresh; /* Slow start size threshold */ 307 u32 snd_cwnd; /* Sending congestion window */ 308 u32 snd_cwnd_cnt; /* Linear increase counter */ 309 u32 snd_cwnd_clamp; /* Do not allow snd_cwnd to grow above this */ 310 u32 snd_cwnd_used; 311 u32 snd_cwnd_stamp; 312 u32 prior_cwnd; /* cwnd right before starting loss recovery */ 313 u32 prr_delivered; /* Number of newly delivered packets to 314 * receiver in Recovery. */ 315 u32 prr_out; /* Total number of pkts sent during Recovery. */ 316 u32 delivered; /* Total data packets delivered incl. rexmits */ 317 u32 delivered_ce; /* Like the above but only ECE marked packets */ 318 u32 lost; /* Total data packets lost incl. rexmits */ 319 u32 app_limited; /* limited until "delivered" reaches this val */ 320 u64 first_tx_mstamp; /* start of window send phase */ 321 u64 delivered_mstamp; /* time we reached "delivered" */ 322 u32 rate_delivered; /* saved rate sample: packets delivered */ 323 u32 rate_interval_us; /* saved rate sample: time elapsed */ 324 325 u32 rcv_wnd; /* Current receiver window */ 326 u32 write_seq; /* Tail(+1) of data held in tcp send buffer */ 327 u32 notsent_lowat; /* TCP_NOTSENT_LOWAT */ 328 u32 pushed_seq; /* Last pushed seq, required to talk to windows */ 329 u32 lost_out; /* Lost packets */ 330 u32 sacked_out; /* SACK'd packets */ 331 332 struct hrtimer pacing_timer; 333 struct hrtimer compressed_ack_timer; 334 335 /* from STCP, retrans queue hinting */ 336 struct sk_buff* lost_skb_hint; 337 struct sk_buff *retransmit_skb_hint; 338 339 /* OOO segments go in this rbtree. Socket lock must be held. */ 340 struct rb_root out_of_order_queue; 341 struct sk_buff *ooo_last_skb; /* cache rb_last(out_of_order_queue) */ 342 343 /* SACKs data, these 2 need to be together (see tcp_options_write) */ 344 struct tcp_sack_block duplicate_sack[1]; /* D-SACK block */ 345 struct tcp_sack_block selective_acks[4]; /* The SACKS themselves*/ 346 347 struct tcp_sack_block recv_sack_cache[4]; 348 349 struct sk_buff *highest_sack; /* skb just after the highest 350 * skb with SACKed bit set 351 * (validity guaranteed only if 352 * sacked_out > 0) 353 */ 354 355 int lost_cnt_hint; 356 357 u32 prior_ssthresh; /* ssthresh saved at recovery start */ 358 u32 high_seq; /* snd_nxt at onset of congestion */ 359 360 u32 retrans_stamp; /* Timestamp of the last retransmit, 361 * also used in SYN-SENT to remember stamp of 362 * the first SYN. */ 363 u32 undo_marker; /* snd_una upon a new recovery episode. */ 364 int undo_retrans; /* number of undoable retransmissions. */ 365 u64 bytes_retrans; /* RFC4898 tcpEStatsPerfOctetsRetrans 366 * Total data bytes retransmitted 367 */ 368 u32 total_retrans; /* Total retransmits for entire connection */ 369 370 u32 urg_seq; /* Seq of received urgent pointer */ 371 unsigned int keepalive_time; /* time before keep alive takes place */ 372 unsigned int keepalive_intvl; /* time interval between keep alive probes */ 373 374 int linger2; 375 376 377 /* Sock_ops bpf program related variables */ 378 #ifdef CONFIG_BPF 379 u8 bpf_sock_ops_cb_flags; /* Control calling BPF programs 380 * values defined in uapi/linux/tcp.h 381 */ 382 #define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) (TP->bpf_sock_ops_cb_flags & ARG) 383 #else 384 #define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) 0 385 #endif 386 387 u16 timeout_rehash; /* Timeout-triggered rehash attempts */ 388 389 u32 rcv_ooopack; /* Received out-of-order packets, for tcpinfo */ 390 391 /* Receiver side RTT estimation */ 392 u32 rcv_rtt_last_tsecr; 393 struct { 394 u32 rtt_us; 395 u32 seq; 396 u64 time; 397 } rcv_rtt_est; 398 399 /* Receiver queue space */ 400 struct { 401 u32 space; 402 u32 seq; 403 u64 time; 404 } rcvq_space; 405 406 /* TCP-specific MTU probe information. */ 407 struct { 408 u32 probe_seq_start; 409 u32 probe_seq_end; 410 } mtu_probe; 411 u32 mtu_info; /* We received an ICMP_FRAG_NEEDED / ICMPV6_PKT_TOOBIG 412 * while socket was owned by user. 413 */ 414 #if IS_ENABLED(CONFIG_MPTCP) 415 bool is_mptcp; 416 #endif 417 418 #ifdef CONFIG_TCP_MD5SIG 419 /* TCP AF-Specific parts; only used by MD5 Signature support so far */ 420 const struct tcp_sock_af_ops *af_specific; 421 422 /* TCP MD5 Signature Option information */ 423 struct tcp_md5sig_info __rcu *md5sig_info; 424 #endif 425 426 /* TCP fastopen related information */ 427 struct tcp_fastopen_request *fastopen_req; 428 /* fastopen_rsk points to request_sock that resulted in this big 429 * socket. Used to retransmit SYNACKs etc. 430 */ 431 struct request_sock __rcu *fastopen_rsk; 432 u32 *saved_syn; 433 }; 434 435 enum tsq_enum { 436 TSQ_THROTTLED, 437 TSQ_QUEUED, 438 TCP_TSQ_DEFERRED, /* tcp_tasklet_func() found socket was owned */ 439 TCP_WRITE_TIMER_DEFERRED, /* tcp_write_timer() found socket was owned */ 440 TCP_DELACK_TIMER_DEFERRED, /* tcp_delack_timer() found socket was owned */ 441 TCP_MTU_REDUCED_DEFERRED, /* tcp_v{4|6}_err() could not call 442 * tcp_v{4|6}_mtu_reduced() 443 */ 444 }; 445 446 enum tsq_flags { 447 TSQF_THROTTLED = (1UL << TSQ_THROTTLED), 448 TSQF_QUEUED = (1UL << TSQ_QUEUED), 449 TCPF_TSQ_DEFERRED = (1UL << TCP_TSQ_DEFERRED), 450 TCPF_WRITE_TIMER_DEFERRED = (1UL << TCP_WRITE_TIMER_DEFERRED), 451 TCPF_DELACK_TIMER_DEFERRED = (1UL << TCP_DELACK_TIMER_DEFERRED), 452 TCPF_MTU_REDUCED_DEFERRED = (1UL << TCP_MTU_REDUCED_DEFERRED), 453 }; 454 455 static inline struct tcp_sock *tcp_sk(const struct sock *sk) 456 { 457 return (struct tcp_sock *)sk; 458 } 459 460 struct tcp_timewait_sock { 461 struct inet_timewait_sock tw_sk; 462 #define tw_rcv_nxt tw_sk.__tw_common.skc_tw_rcv_nxt 463 #define tw_snd_nxt tw_sk.__tw_common.skc_tw_snd_nxt 464 u32 tw_rcv_wnd; 465 u32 tw_ts_offset; 466 u32 tw_ts_recent; 467 468 /* The time we sent the last out-of-window ACK: */ 469 u32 tw_last_oow_ack_time; 470 471 int tw_ts_recent_stamp; 472 u32 tw_tx_delay; 473 #ifdef CONFIG_TCP_MD5SIG 474 struct tcp_md5sig_key *tw_md5_key; 475 #endif 476 }; 477 478 static inline struct tcp_timewait_sock *tcp_twsk(const struct sock *sk) 479 { 480 return (struct tcp_timewait_sock *)sk; 481 } 482 483 static inline bool tcp_passive_fastopen(const struct sock *sk) 484 { 485 return sk->sk_state == TCP_SYN_RECV && 486 rcu_access_pointer(tcp_sk(sk)->fastopen_rsk) != NULL; 487 } 488 489 static inline void fastopen_queue_tune(struct sock *sk, int backlog) 490 { 491 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue; 492 int somaxconn = READ_ONCE(sock_net(sk)->core.sysctl_somaxconn); 493 494 queue->fastopenq.max_qlen = min_t(unsigned int, backlog, somaxconn); 495 } 496 497 static inline void tcp_move_syn(struct tcp_sock *tp, 498 struct request_sock *req) 499 { 500 tp->saved_syn = req->saved_syn; 501 req->saved_syn = NULL; 502 } 503 504 static inline void tcp_saved_syn_free(struct tcp_sock *tp) 505 { 506 kfree(tp->saved_syn); 507 tp->saved_syn = NULL; 508 } 509 510 struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk); 511 512 static inline u16 tcp_mss_clamp(const struct tcp_sock *tp, u16 mss) 513 { 514 /* We use READ_ONCE() here because socket might not be locked. 515 * This happens for listeners. 516 */ 517 u16 user_mss = READ_ONCE(tp->rx_opt.user_mss); 518 519 return (user_mss && user_mss < mss) ? user_mss : mss; 520 } 521 522 int tcp_skb_shift(struct sk_buff *to, struct sk_buff *from, int pcount, 523 int shiftlen); 524 525 #endif /* _LINUX_TCP_H */ 526