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