1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * Definitions for the TCP protocol. 7 * 8 * Version: @(#)tcp.h 1.0.2 04/28/93 9 * 10 * Author: Fred N. van Kempen, <[email protected]> 11 * 12 * This program is free software; you can redistribute it and/or 13 * modify it under the terms of the GNU General Public License 14 * as published by the Free Software Foundation; either version 15 * 2 of the License, or (at your option) any later version. 16 */ 17 #ifndef _LINUX_TCP_H 18 #define _LINUX_TCP_H 19 20 21 #include <linux/skbuff.h> 22 #include <linux/win_minmax.h> 23 #include <net/sock.h> 24 #include <net/inet_connection_sock.h> 25 #include <net/inet_timewait_sock.h> 26 #include <uapi/linux/tcp.h> 27 28 static inline struct tcphdr *tcp_hdr(const struct sk_buff *skb) 29 { 30 return (struct tcphdr *)skb_transport_header(skb); 31 } 32 33 static inline unsigned int __tcp_hdrlen(const struct tcphdr *th) 34 { 35 return th->doff * 4; 36 } 37 38 static inline unsigned int tcp_hdrlen(const struct sk_buff *skb) 39 { 40 return __tcp_hdrlen(tcp_hdr(skb)); 41 } 42 43 static inline struct tcphdr *inner_tcp_hdr(const struct sk_buff *skb) 44 { 45 return (struct tcphdr *)skb_inner_transport_header(skb); 46 } 47 48 static inline unsigned int inner_tcp_hdrlen(const struct sk_buff *skb) 49 { 50 return inner_tcp_hdr(skb)->doff * 4; 51 } 52 53 static inline unsigned int tcp_optlen(const struct sk_buff *skb) 54 { 55 return (tcp_hdr(skb)->doff - 5) * 4; 56 } 57 58 /* TCP Fast Open */ 59 #define TCP_FASTOPEN_COOKIE_MIN 4 /* Min Fast Open Cookie size in bytes */ 60 #define TCP_FASTOPEN_COOKIE_MAX 16 /* Max Fast Open Cookie size in bytes */ 61 #define TCP_FASTOPEN_COOKIE_SIZE 8 /* the size employed by this impl. */ 62 63 /* TCP Fast Open Cookie as stored in memory */ 64 struct tcp_fastopen_cookie { 65 s8 len; 66 u8 val[TCP_FASTOPEN_COOKIE_MAX]; 67 bool exp; /* In RFC6994 experimental option format */ 68 }; 69 70 /* This defines a selective acknowledgement block. */ 71 struct tcp_sack_block_wire { 72 __be32 start_seq; 73 __be32 end_seq; 74 }; 75 76 struct tcp_sack_block { 77 u32 start_seq; 78 u32 end_seq; 79 }; 80 81 /*These are used to set the sack_ok field in struct tcp_options_received */ 82 #define TCP_SACK_SEEN (1 << 0) /*1 = peer is SACK capable, */ 83 #define TCP_FACK_ENABLED (1 << 1) /*1 = FACK is enabled locally*/ 84 #define TCP_DSACK_SEEN (1 << 2) /*1 = DSACK was received from peer*/ 85 86 struct tcp_options_received { 87 /* PAWS/RTTM data */ 88 long ts_recent_stamp;/* Time we stored ts_recent (for aging) */ 89 u32 ts_recent; /* Time stamp to echo next */ 90 u32 rcv_tsval; /* Time stamp value */ 91 u32 rcv_tsecr; /* Time stamp echo reply */ 92 u16 saw_tstamp : 1, /* Saw TIMESTAMP on last packet */ 93 tstamp_ok : 1, /* TIMESTAMP seen on SYN packet */ 94 dsack : 1, /* D-SACK is scheduled */ 95 wscale_ok : 1, /* Wscale seen on SYN packet */ 96 sack_ok : 4, /* SACK seen on SYN packet */ 97 snd_wscale : 4, /* Window scaling received from sender */ 98 rcv_wscale : 4; /* Window scaling to send to receiver */ 99 u8 num_sacks; /* Number of SACK blocks */ 100 u16 user_mss; /* mss requested by user in ioctl */ 101 u16 mss_clamp; /* Maximal mss, negotiated at connection setup */ 102 }; 103 104 static inline void tcp_clear_options(struct tcp_options_received *rx_opt) 105 { 106 rx_opt->tstamp_ok = rx_opt->sack_ok = 0; 107 rx_opt->wscale_ok = rx_opt->snd_wscale = 0; 108 } 109 110 /* This is the max number of SACKS that we'll generate and process. It's safe 111 * to increase this, although since: 112 * size = TCPOLEN_SACK_BASE_ALIGNED (4) + n * TCPOLEN_SACK_PERBLOCK (8) 113 * only four options will fit in a standard TCP header */ 114 #define TCP_NUM_SACKS 4 115 116 struct tcp_request_sock_ops; 117 118 struct tcp_request_sock { 119 struct inet_request_sock req; 120 const struct tcp_request_sock_ops *af_specific; 121 struct skb_mstamp snt_synack; /* first SYNACK sent time */ 122 bool tfo_listener; 123 u32 txhash; 124 u32 rcv_isn; 125 u32 snt_isn; 126 u32 ts_off; 127 u32 last_oow_ack_time; /* last SYNACK */ 128 u32 rcv_nxt; /* the ack # by SYNACK. For 129 * FastOpen it's the seq# 130 * after data-in-SYN. 131 */ 132 }; 133 134 static inline struct tcp_request_sock *tcp_rsk(const struct request_sock *req) 135 { 136 return (struct tcp_request_sock *)req; 137 } 138 139 struct tcp_sock { 140 /* inet_connection_sock has to be the first member of tcp_sock */ 141 struct inet_connection_sock inet_conn; 142 u16 tcp_header_len; /* Bytes of tcp header to send */ 143 u16 gso_segs; /* Max number of segs per GSO packet */ 144 145 /* 146 * Header prediction flags 147 * 0x5?10 << 16 + snd_wnd in net byte order 148 */ 149 __be32 pred_flags; 150 151 /* 152 * RFC793 variables by their proper names. This means you can 153 * read the code and the spec side by side (and laugh ...) 154 * See RFC793 and RFC1122. The RFC writes these in capitals. 155 */ 156 u64 bytes_received; /* RFC4898 tcpEStatsAppHCThruOctetsReceived 157 * sum(delta(rcv_nxt)), or how many bytes 158 * were acked. 159 */ 160 u32 segs_in; /* RFC4898 tcpEStatsPerfSegsIn 161 * total number of segments in. 162 */ 163 u32 data_segs_in; /* RFC4898 tcpEStatsPerfDataSegsIn 164 * total number of data segments in. 165 */ 166 u32 rcv_nxt; /* What we want to receive next */ 167 u32 copied_seq; /* Head of yet unread data */ 168 u32 rcv_wup; /* rcv_nxt on last window update sent */ 169 u32 snd_nxt; /* Next sequence we send */ 170 u32 segs_out; /* RFC4898 tcpEStatsPerfSegsOut 171 * The total number of segments sent. 172 */ 173 u32 data_segs_out; /* RFC4898 tcpEStatsPerfDataSegsOut 174 * total number of data segments sent. 175 */ 176 u64 bytes_acked; /* RFC4898 tcpEStatsAppHCThruOctetsAcked 177 * sum(delta(snd_una)), or how many bytes 178 * were acked. 179 */ 180 u32 snd_una; /* First byte we want an ack for */ 181 u32 snd_sml; /* Last byte of the most recently transmitted small packet */ 182 u32 rcv_tstamp; /* timestamp of last received ACK (for keepalives) */ 183 u32 lsndtime; /* timestamp of last sent data packet (for restart window) */ 184 u32 last_oow_ack_time; /* timestamp of last out-of-window ACK */ 185 186 u32 tsoffset; /* timestamp offset */ 187 188 struct list_head tsq_node; /* anchor in tsq_tasklet.head list */ 189 190 /* Data for direct copy to user */ 191 struct { 192 struct sk_buff_head prequeue; 193 struct task_struct *task; 194 struct msghdr *msg; 195 int memory; 196 int len; 197 } ucopy; 198 199 u32 snd_wl1; /* Sequence for window update */ 200 u32 snd_wnd; /* The window we expect to receive */ 201 u32 max_window; /* Maximal window ever seen from peer */ 202 u32 mss_cache; /* Cached effective mss, not including SACKS */ 203 204 u32 window_clamp; /* Maximal window to advertise */ 205 u32 rcv_ssthresh; /* Current window clamp */ 206 207 /* Information of the most recently (s)acked skb */ 208 struct tcp_rack { 209 struct skb_mstamp mstamp; /* (Re)sent time of the skb */ 210 u8 advanced; /* mstamp advanced since last lost marking */ 211 u8 reord; /* reordering detected */ 212 } rack; 213 u16 advmss; /* Advertised MSS */ 214 u32 chrono_start; /* Start time in jiffies of a TCP chrono */ 215 u32 chrono_stat[3]; /* Time in jiffies for chrono_stat stats */ 216 u8 chrono_type:2, /* current chronograph type */ 217 rate_app_limited:1, /* rate_{delivered,interval_us} limited? */ 218 unused:5; 219 u8 nonagle : 4,/* Disable Nagle algorithm? */ 220 thin_lto : 1,/* Use linear timeouts for thin streams */ 221 thin_dupack : 1,/* Fast retransmit on first dupack */ 222 repair : 1, 223 frto : 1;/* F-RTO (RFC5682) activated in CA_Loss */ 224 u8 repair_queue; 225 u8 do_early_retrans:1,/* Enable RFC5827 early-retransmit */ 226 syn_data:1, /* SYN includes data */ 227 syn_fastopen:1, /* SYN includes Fast Open option */ 228 syn_fastopen_exp:1,/* SYN includes Fast Open exp. option */ 229 syn_data_acked:1,/* data in SYN is acked by SYN-ACK */ 230 save_syn:1, /* Save headers of SYN packet */ 231 is_cwnd_limited:1;/* forward progress limited by snd_cwnd? */ 232 u32 tlp_high_seq; /* snd_nxt at the time of TLP retransmit. */ 233 234 /* RTT measurement */ 235 u32 srtt_us; /* smoothed round trip time << 3 in usecs */ 236 u32 mdev_us; /* medium deviation */ 237 u32 mdev_max_us; /* maximal mdev for the last rtt period */ 238 u32 rttvar_us; /* smoothed mdev_max */ 239 u32 rtt_seq; /* sequence number to update rttvar */ 240 struct minmax rtt_min; 241 242 u32 packets_out; /* Packets which are "in flight" */ 243 u32 retrans_out; /* Retransmitted packets out */ 244 u32 max_packets_out; /* max packets_out in last window */ 245 u32 max_packets_seq; /* right edge of max_packets_out flight */ 246 247 u16 urg_data; /* Saved octet of OOB data and control flags */ 248 u8 ecn_flags; /* ECN status bits. */ 249 u8 keepalive_probes; /* num of allowed keep alive probes */ 250 u32 reordering; /* Packet reordering metric. */ 251 u32 snd_up; /* Urgent pointer */ 252 253 /* 254 * Options received (usually on last packet, some only on SYN packets). 255 */ 256 struct tcp_options_received rx_opt; 257 258 /* 259 * Slow start and congestion control (see also Nagle, and Karn & Partridge) 260 */ 261 u32 snd_ssthresh; /* Slow start size threshold */ 262 u32 snd_cwnd; /* Sending congestion window */ 263 u32 snd_cwnd_cnt; /* Linear increase counter */ 264 u32 snd_cwnd_clamp; /* Do not allow snd_cwnd to grow above this */ 265 u32 snd_cwnd_used; 266 u32 snd_cwnd_stamp; 267 u32 prior_cwnd; /* Congestion window at start of Recovery. */ 268 u32 prr_delivered; /* Number of newly delivered packets to 269 * receiver in Recovery. */ 270 u32 prr_out; /* Total number of pkts sent during Recovery. */ 271 u32 delivered; /* Total data packets delivered incl. rexmits */ 272 u32 lost; /* Total data packets lost incl. rexmits */ 273 u32 app_limited; /* limited until "delivered" reaches this val */ 274 struct skb_mstamp first_tx_mstamp; /* start of window send phase */ 275 struct skb_mstamp delivered_mstamp; /* time we reached "delivered" */ 276 u32 rate_delivered; /* saved rate sample: packets delivered */ 277 u32 rate_interval_us; /* saved rate sample: time elapsed */ 278 279 u32 rcv_wnd; /* Current receiver window */ 280 u32 write_seq; /* Tail(+1) of data held in tcp send buffer */ 281 u32 notsent_lowat; /* TCP_NOTSENT_LOWAT */ 282 u32 pushed_seq; /* Last pushed seq, required to talk to windows */ 283 u32 lost_out; /* Lost packets */ 284 u32 sacked_out; /* SACK'd packets */ 285 u32 fackets_out; /* FACK'd packets */ 286 287 /* from STCP, retrans queue hinting */ 288 struct sk_buff* lost_skb_hint; 289 struct sk_buff *retransmit_skb_hint; 290 291 /* OOO segments go in this rbtree. Socket lock must be held. */ 292 struct rb_root out_of_order_queue; 293 struct sk_buff *ooo_last_skb; /* cache rb_last(out_of_order_queue) */ 294 295 /* SACKs data, these 2 need to be together (see tcp_options_write) */ 296 struct tcp_sack_block duplicate_sack[1]; /* D-SACK block */ 297 struct tcp_sack_block selective_acks[4]; /* The SACKS themselves*/ 298 299 struct tcp_sack_block recv_sack_cache[4]; 300 301 struct sk_buff *highest_sack; /* skb just after the highest 302 * skb with SACKed bit set 303 * (validity guaranteed only if 304 * sacked_out > 0) 305 */ 306 307 int lost_cnt_hint; 308 u32 retransmit_high; /* L-bits may be on up to this seqno */ 309 310 u32 prior_ssthresh; /* ssthresh saved at recovery start */ 311 u32 high_seq; /* snd_nxt at onset of congestion */ 312 313 u32 retrans_stamp; /* Timestamp of the last retransmit, 314 * also used in SYN-SENT to remember stamp of 315 * the first SYN. */ 316 u32 undo_marker; /* snd_una upon a new recovery episode. */ 317 int undo_retrans; /* number of undoable retransmissions. */ 318 u32 total_retrans; /* Total retransmits for entire connection */ 319 320 u32 urg_seq; /* Seq of received urgent pointer */ 321 unsigned int keepalive_time; /* time before keep alive takes place */ 322 unsigned int keepalive_intvl; /* time interval between keep alive probes */ 323 324 int linger2; 325 326 /* Receiver side RTT estimation */ 327 struct { 328 u32 rtt; 329 u32 seq; 330 u32 time; 331 } rcv_rtt_est; 332 333 /* Receiver queue space */ 334 struct { 335 int space; 336 u32 seq; 337 u32 time; 338 } rcvq_space; 339 340 /* TCP-specific MTU probe information. */ 341 struct { 342 u32 probe_seq_start; 343 u32 probe_seq_end; 344 } mtu_probe; 345 u32 mtu_info; /* We received an ICMP_FRAG_NEEDED / ICMPV6_PKT_TOOBIG 346 * while socket was owned by user. 347 */ 348 349 #ifdef CONFIG_TCP_MD5SIG 350 /* TCP AF-Specific parts; only used by MD5 Signature support so far */ 351 const struct tcp_sock_af_ops *af_specific; 352 353 /* TCP MD5 Signature Option information */ 354 struct tcp_md5sig_info __rcu *md5sig_info; 355 #endif 356 357 /* TCP fastopen related information */ 358 struct tcp_fastopen_request *fastopen_req; 359 /* fastopen_rsk points to request_sock that resulted in this big 360 * socket. Used to retransmit SYNACKs etc. 361 */ 362 struct request_sock *fastopen_rsk; 363 u32 *saved_syn; 364 }; 365 366 enum tsq_enum { 367 TSQ_THROTTLED, 368 TSQ_QUEUED, 369 TCP_TSQ_DEFERRED, /* tcp_tasklet_func() found socket was owned */ 370 TCP_WRITE_TIMER_DEFERRED, /* tcp_write_timer() found socket was owned */ 371 TCP_DELACK_TIMER_DEFERRED, /* tcp_delack_timer() found socket was owned */ 372 TCP_MTU_REDUCED_DEFERRED, /* tcp_v{4|6}_err() could not call 373 * tcp_v{4|6}_mtu_reduced() 374 */ 375 }; 376 377 enum tsq_flags { 378 TSQF_THROTTLED = (1UL << TSQ_THROTTLED), 379 TSQF_QUEUED = (1UL << TSQ_QUEUED), 380 TCPF_TSQ_DEFERRED = (1UL << TCP_TSQ_DEFERRED), 381 TCPF_WRITE_TIMER_DEFERRED = (1UL << TCP_WRITE_TIMER_DEFERRED), 382 TCPF_DELACK_TIMER_DEFERRED = (1UL << TCP_DELACK_TIMER_DEFERRED), 383 TCPF_MTU_REDUCED_DEFERRED = (1UL << TCP_MTU_REDUCED_DEFERRED), 384 }; 385 386 static inline struct tcp_sock *tcp_sk(const struct sock *sk) 387 { 388 return (struct tcp_sock *)sk; 389 } 390 391 struct tcp_timewait_sock { 392 struct inet_timewait_sock tw_sk; 393 #define tw_rcv_nxt tw_sk.__tw_common.skc_tw_rcv_nxt 394 #define tw_snd_nxt tw_sk.__tw_common.skc_tw_snd_nxt 395 u32 tw_rcv_wnd; 396 u32 tw_ts_offset; 397 u32 tw_ts_recent; 398 399 /* The time we sent the last out-of-window ACK: */ 400 u32 tw_last_oow_ack_time; 401 402 long tw_ts_recent_stamp; 403 #ifdef CONFIG_TCP_MD5SIG 404 struct tcp_md5sig_key *tw_md5_key; 405 #endif 406 }; 407 408 static inline struct tcp_timewait_sock *tcp_twsk(const struct sock *sk) 409 { 410 return (struct tcp_timewait_sock *)sk; 411 } 412 413 static inline bool tcp_passive_fastopen(const struct sock *sk) 414 { 415 return (sk->sk_state == TCP_SYN_RECV && 416 tcp_sk(sk)->fastopen_rsk != NULL); 417 } 418 419 static inline void fastopen_queue_tune(struct sock *sk, int backlog) 420 { 421 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue; 422 int somaxconn = READ_ONCE(sock_net(sk)->core.sysctl_somaxconn); 423 424 queue->fastopenq.max_qlen = min_t(unsigned int, backlog, somaxconn); 425 } 426 427 static inline void tcp_move_syn(struct tcp_sock *tp, 428 struct request_sock *req) 429 { 430 tp->saved_syn = req->saved_syn; 431 req->saved_syn = NULL; 432 } 433 434 static inline void tcp_saved_syn_free(struct tcp_sock *tp) 435 { 436 kfree(tp->saved_syn); 437 tp->saved_syn = NULL; 438 } 439 440 struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk); 441 442 #endif /* _LINUX_TCP_H */ 443