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, <waltje@uWalt.NL.Mugnet.ORG> 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 <net/sock.h> 23#include <net/inet_connection_sock.h> 24#include <net/inet_timewait_sock.h> 25#include <uapi/linux/tcp.h> 26 27static inline struct tcphdr *tcp_hdr(const struct sk_buff *skb) 28{ 29 return (struct tcphdr *)skb_transport_header(skb); 30} 31 32static inline unsigned int tcp_hdrlen(const struct sk_buff *skb) 33{ 34 return tcp_hdr(skb)->doff * 4; 35} 36 37static inline struct tcphdr *inner_tcp_hdr(const struct sk_buff *skb) 38{ 39 return (struct tcphdr *)skb_inner_transport_header(skb); 40} 41 42static inline unsigned int inner_tcp_hdrlen(const struct sk_buff *skb) 43{ 44 return inner_tcp_hdr(skb)->doff * 4; 45} 46 47static inline unsigned int tcp_optlen(const struct sk_buff *skb) 48{ 49 return (tcp_hdr(skb)->doff - 5) * 4; 50} 51 52/* TCP Fast Open */ 53#define TCP_FASTOPEN_COOKIE_MIN 4 /* Min Fast Open Cookie size in bytes */ 54#define TCP_FASTOPEN_COOKIE_MAX 16 /* Max Fast Open Cookie size in bytes */ 55#define TCP_FASTOPEN_COOKIE_SIZE 8 /* the size employed by this impl. */ 56 57/* TCP Fast Open Cookie as stored in memory */ 58struct tcp_fastopen_cookie { 59 s8 len; 60 u8 val[TCP_FASTOPEN_COOKIE_MAX]; 61 bool exp; /* In RFC6994 experimental option format */ 62}; 63 64/* This defines a selective acknowledgement block. */ 65struct tcp_sack_block_wire { 66 __be32 start_seq; 67 __be32 end_seq; 68}; 69 70struct tcp_sack_block { 71 u32 start_seq; 72 u32 end_seq; 73}; 74 75/*These are used to set the sack_ok field in struct tcp_options_received */ 76#define TCP_SACK_SEEN (1 << 0) /*1 = peer is SACK capable, */ 77#define TCP_FACK_ENABLED (1 << 1) /*1 = FACK is enabled locally*/ 78#define TCP_DSACK_SEEN (1 << 2) /*1 = DSACK was received from peer*/ 79 80struct tcp_options_received { 81/* PAWS/RTTM data */ 82 long ts_recent_stamp;/* Time we stored ts_recent (for aging) */ 83 u32 ts_recent; /* Time stamp to echo next */ 84 u32 rcv_tsval; /* Time stamp value */ 85 u32 rcv_tsecr; /* Time stamp echo reply */ 86 u16 saw_tstamp : 1, /* Saw TIMESTAMP on last packet */ 87 tstamp_ok : 1, /* TIMESTAMP seen on SYN packet */ 88 dsack : 1, /* D-SACK is scheduled */ 89 wscale_ok : 1, /* Wscale seen on SYN packet */ 90 sack_ok : 4, /* SACK seen on SYN packet */ 91 snd_wscale : 4, /* Window scaling received from sender */ 92 rcv_wscale : 4; /* Window scaling to send to receiver */ 93 u8 num_sacks; /* Number of SACK blocks */ 94 u16 user_mss; /* mss requested by user in ioctl */ 95 u16 mss_clamp; /* Maximal mss, negotiated at connection setup */ 96}; 97 98static inline void tcp_clear_options(struct tcp_options_received *rx_opt) 99{ 100 rx_opt->tstamp_ok = rx_opt->sack_ok = 0; 101 rx_opt->wscale_ok = rx_opt->snd_wscale = 0; 102} 103 104/* This is the max number of SACKS that we'll generate and process. It's safe 105 * to increase this, although since: 106 * size = TCPOLEN_SACK_BASE_ALIGNED (4) + n * TCPOLEN_SACK_PERBLOCK (8) 107 * only four options will fit in a standard TCP header */ 108#define TCP_NUM_SACKS 4 109 110struct tcp_request_sock_ops; 111 112struct tcp_request_sock { 113 struct inet_request_sock req; 114 const struct tcp_request_sock_ops *af_specific; 115 bool tfo_listener; 116 u32 rcv_isn; 117 u32 snt_isn; 118 u32 snt_synack; /* synack sent time */ 119 u32 last_oow_ack_time; /* last SYNACK */ 120 u32 rcv_nxt; /* the ack # by SYNACK. For 121 * FastOpen it's the seq# 122 * after data-in-SYN. 123 */ 124}; 125 126static inline struct tcp_request_sock *tcp_rsk(const struct request_sock *req) 127{ 128 return (struct tcp_request_sock *)req; 129} 130 131struct tcp_sock { 132 /* inet_connection_sock has to be the first member of tcp_sock */ 133 struct inet_connection_sock inet_conn; 134 u16 tcp_header_len; /* Bytes of tcp header to send */ 135 u16 gso_segs; /* Max number of segs per GSO packet */ 136 137/* 138 * Header prediction flags 139 * 0x5?10 << 16 + snd_wnd in net byte order 140 */ 141 __be32 pred_flags; 142 143/* 144 * RFC793 variables by their proper names. This means you can 145 * read the code and the spec side by side (and laugh ...) 146 * See RFC793 and RFC1122. The RFC writes these in capitals. 147 */ 148 u64 bytes_received; /* RFC4898 tcpEStatsAppHCThruOctetsReceived 149 * sum(delta(rcv_nxt)), or how many bytes 150 * were acked. 151 */ 152 u32 rcv_nxt; /* What we want to receive next */ 153 u32 copied_seq; /* Head of yet unread data */ 154 u32 rcv_wup; /* rcv_nxt on last window update sent */ 155 u32 snd_nxt; /* Next sequence we send */ 156 157 u64 bytes_acked; /* RFC4898 tcpEStatsAppHCThruOctetsAcked 158 * sum(delta(snd_una)), or how many bytes 159 * were acked. 160 */ 161 struct u64_stats_sync syncp; /* protects 64bit vars (cf tcp_get_info()) */ 162 163 u32 snd_una; /* First byte we want an ack for */ 164 u32 snd_sml; /* Last byte of the most recently transmitted small packet */ 165 u32 rcv_tstamp; /* timestamp of last received ACK (for keepalives) */ 166 u32 lsndtime; /* timestamp of last sent data packet (for restart window) */ 167 u32 last_oow_ack_time; /* timestamp of last out-of-window ACK */ 168 169 u32 tsoffset; /* timestamp offset */ 170 171 struct list_head tsq_node; /* anchor in tsq_tasklet.head list */ 172 unsigned long tsq_flags; 173 174 /* Data for direct copy to user */ 175 struct { 176 struct sk_buff_head prequeue; 177 struct task_struct *task; 178 struct msghdr *msg; 179 int memory; 180 int len; 181 } ucopy; 182 183 u32 snd_wl1; /* Sequence for window update */ 184 u32 snd_wnd; /* The window we expect to receive */ 185 u32 max_window; /* Maximal window ever seen from peer */ 186 u32 mss_cache; /* Cached effective mss, not including SACKS */ 187 188 u32 window_clamp; /* Maximal window to advertise */ 189 u32 rcv_ssthresh; /* Current window clamp */ 190 191 u16 advmss; /* Advertised MSS */ 192 u8 unused; 193 u8 nonagle : 4,/* Disable Nagle algorithm? */ 194 thin_lto : 1,/* Use linear timeouts for thin streams */ 195 thin_dupack : 1,/* Fast retransmit on first dupack */ 196 repair : 1, 197 frto : 1;/* F-RTO (RFC5682) activated in CA_Loss */ 198 u8 repair_queue; 199 u8 do_early_retrans:1,/* Enable RFC5827 early-retransmit */ 200 syn_data:1, /* SYN includes data */ 201 syn_fastopen:1, /* SYN includes Fast Open option */ 202 syn_fastopen_exp:1,/* SYN includes Fast Open exp. option */ 203 syn_data_acked:1,/* data in SYN is acked by SYN-ACK */ 204 is_cwnd_limited:1;/* forward progress limited by snd_cwnd? */ 205 u32 tlp_high_seq; /* snd_nxt at the time of TLP retransmit. */ 206 207/* RTT measurement */ 208 u32 srtt_us; /* smoothed round trip time << 3 in usecs */ 209 u32 mdev_us; /* medium deviation */ 210 u32 mdev_max_us; /* maximal mdev for the last rtt period */ 211 u32 rttvar_us; /* smoothed mdev_max */ 212 u32 rtt_seq; /* sequence number to update rttvar */ 213 214 u32 packets_out; /* Packets which are "in flight" */ 215 u32 retrans_out; /* Retransmitted packets out */ 216 u32 max_packets_out; /* max packets_out in last window */ 217 u32 max_packets_seq; /* right edge of max_packets_out flight */ 218 219 u16 urg_data; /* Saved octet of OOB data and control flags */ 220 u8 ecn_flags; /* ECN status bits. */ 221 u8 keepalive_probes; /* num of allowed keep alive probes */ 222 u32 reordering; /* Packet reordering metric. */ 223 u32 snd_up; /* Urgent pointer */ 224 225/* 226 * Options received (usually on last packet, some only on SYN packets). 227 */ 228 struct tcp_options_received rx_opt; 229 230/* 231 * Slow start and congestion control (see also Nagle, and Karn & Partridge) 232 */ 233 u32 snd_ssthresh; /* Slow start size threshold */ 234 u32 snd_cwnd; /* Sending congestion window */ 235 u32 snd_cwnd_cnt; /* Linear increase counter */ 236 u32 snd_cwnd_clamp; /* Do not allow snd_cwnd to grow above this */ 237 u32 snd_cwnd_used; 238 u32 snd_cwnd_stamp; 239 u32 prior_cwnd; /* Congestion window at start of Recovery. */ 240 u32 prr_delivered; /* Number of newly delivered packets to 241 * receiver in Recovery. */ 242 u32 prr_out; /* Total number of pkts sent during Recovery. */ 243 244 u32 rcv_wnd; /* Current receiver window */ 245 u32 write_seq; /* Tail(+1) of data held in tcp send buffer */ 246 u32 notsent_lowat; /* TCP_NOTSENT_LOWAT */ 247 u32 pushed_seq; /* Last pushed seq, required to talk to windows */ 248 u32 lost_out; /* Lost packets */ 249 u32 sacked_out; /* SACK'd packets */ 250 u32 fackets_out; /* FACK'd packets */ 251 252 /* from STCP, retrans queue hinting */ 253 struct sk_buff* lost_skb_hint; 254 struct sk_buff *retransmit_skb_hint; 255 256 /* OOO segments go in this list. Note that socket lock must be held, 257 * as we do not use sk_buff_head lock. 258 */ 259 struct sk_buff_head out_of_order_queue; 260 261 /* SACKs data, these 2 need to be together (see tcp_options_write) */ 262 struct tcp_sack_block duplicate_sack[1]; /* D-SACK block */ 263 struct tcp_sack_block selective_acks[4]; /* The SACKS themselves*/ 264 265 struct tcp_sack_block recv_sack_cache[4]; 266 267 struct sk_buff *highest_sack; /* skb just after the highest 268 * skb with SACKed bit set 269 * (validity guaranteed only if 270 * sacked_out > 0) 271 */ 272 273 int lost_cnt_hint; 274 u32 retransmit_high; /* L-bits may be on up to this seqno */ 275 276 u32 lost_retrans_low; /* Sent seq after any rxmit (lowest) */ 277 278 u32 prior_ssthresh; /* ssthresh saved at recovery start */ 279 u32 high_seq; /* snd_nxt at onset of congestion */ 280 281 u32 retrans_stamp; /* Timestamp of the last retransmit, 282 * also used in SYN-SENT to remember stamp of 283 * the first SYN. */ 284 u32 undo_marker; /* snd_una upon a new recovery episode. */ 285 int undo_retrans; /* number of undoable retransmissions. */ 286 u32 total_retrans; /* Total retransmits for entire connection */ 287 288 u32 urg_seq; /* Seq of received urgent pointer */ 289 unsigned int keepalive_time; /* time before keep alive takes place */ 290 unsigned int keepalive_intvl; /* time interval between keep alive probes */ 291 292 int linger2; 293 294/* Receiver side RTT estimation */ 295 struct { 296 u32 rtt; 297 u32 seq; 298 u32 time; 299 } rcv_rtt_est; 300 301/* Receiver queue space */ 302 struct { 303 int space; 304 u32 seq; 305 u32 time; 306 } rcvq_space; 307 308/* TCP-specific MTU probe information. */ 309 struct { 310 u32 probe_seq_start; 311 u32 probe_seq_end; 312 } mtu_probe; 313 u32 mtu_info; /* We received an ICMP_FRAG_NEEDED / ICMPV6_PKT_TOOBIG 314 * while socket was owned by user. 315 */ 316 317#ifdef CONFIG_TCP_MD5SIG 318/* TCP AF-Specific parts; only used by MD5 Signature support so far */ 319 const struct tcp_sock_af_ops *af_specific; 320 321/* TCP MD5 Signature Option information */ 322 struct tcp_md5sig_info __rcu *md5sig_info; 323#endif 324 325/* TCP fastopen related information */ 326 struct tcp_fastopen_request *fastopen_req; 327 /* fastopen_rsk points to request_sock that resulted in this big 328 * socket. Used to retransmit SYNACKs etc. 329 */ 330 struct request_sock *fastopen_rsk; 331}; 332 333enum tsq_flags { 334 TSQ_THROTTLED, 335 TSQ_QUEUED, 336 TCP_TSQ_DEFERRED, /* tcp_tasklet_func() found socket was owned */ 337 TCP_WRITE_TIMER_DEFERRED, /* tcp_write_timer() found socket was owned */ 338 TCP_DELACK_TIMER_DEFERRED, /* tcp_delack_timer() found socket was owned */ 339 TCP_MTU_REDUCED_DEFERRED, /* tcp_v{4|6}_err() could not call 340 * tcp_v{4|6}_mtu_reduced() 341 */ 342}; 343 344static inline struct tcp_sock *tcp_sk(const struct sock *sk) 345{ 346 return (struct tcp_sock *)sk; 347} 348 349struct tcp_timewait_sock { 350 struct inet_timewait_sock tw_sk; 351 u32 tw_rcv_nxt; 352 u32 tw_snd_nxt; 353 u32 tw_rcv_wnd; 354 u32 tw_ts_offset; 355 u32 tw_ts_recent; 356 357 /* The time we sent the last out-of-window ACK: */ 358 u32 tw_last_oow_ack_time; 359 360 long tw_ts_recent_stamp; 361#ifdef CONFIG_TCP_MD5SIG 362 struct tcp_md5sig_key *tw_md5_key; 363#endif 364}; 365 366static inline struct tcp_timewait_sock *tcp_twsk(const struct sock *sk) 367{ 368 return (struct tcp_timewait_sock *)sk; 369} 370 371static inline bool tcp_passive_fastopen(const struct sock *sk) 372{ 373 return (sk->sk_state == TCP_SYN_RECV && 374 tcp_sk(sk)->fastopen_rsk != NULL); 375} 376 377extern void tcp_sock_destruct(struct sock *sk); 378 379static inline int fastopen_init_queue(struct sock *sk, int backlog) 380{ 381 struct request_sock_queue *queue = 382 &inet_csk(sk)->icsk_accept_queue; 383 384 if (queue->fastopenq == NULL) { 385 queue->fastopenq = kzalloc( 386 sizeof(struct fastopen_queue), 387 sk->sk_allocation); 388 if (queue->fastopenq == NULL) 389 return -ENOMEM; 390 391 sk->sk_destruct = tcp_sock_destruct; 392 spin_lock_init(&queue->fastopenq->lock); 393 } 394 queue->fastopenq->max_qlen = backlog; 395 return 0; 396} 397 398#endif /* _LINUX_TCP_H */ 399