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 module. 7 * 8 * Version: @(#)tcp.h 1.0.5 05/23/93 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * 13 * This program is free software; you can redistribute it and/or 14 * modify it under the terms of the GNU General Public License 15 * as published by the Free Software Foundation; either version 16 * 2 of the License, or (at your option) any later version. 17 */ 18 #ifndef _TCP_H 19 #define _TCP_H 20 21 #define FASTRETRANS_DEBUG 1 22 23 #include <linux/list.h> 24 #include <linux/tcp.h> 25 #include <linux/bug.h> 26 #include <linux/slab.h> 27 #include <linux/cache.h> 28 #include <linux/percpu.h> 29 #include <linux/skbuff.h> 30 #include <linux/cryptohash.h> 31 #include <linux/kref.h> 32 #include <linux/ktime.h> 33 34 #include <net/inet_connection_sock.h> 35 #include <net/inet_timewait_sock.h> 36 #include <net/inet_hashtables.h> 37 #include <net/checksum.h> 38 #include <net/request_sock.h> 39 #include <net/sock.h> 40 #include <net/snmp.h> 41 #include <net/ip.h> 42 #include <net/tcp_states.h> 43 #include <net/inet_ecn.h> 44 #include <net/dst.h> 45 46 #include <linux/seq_file.h> 47 #include <linux/memcontrol.h> 48 49 extern struct inet_hashinfo tcp_hashinfo; 50 51 extern struct percpu_counter tcp_orphan_count; 52 void tcp_time_wait(struct sock *sk, int state, int timeo); 53 54 #define MAX_TCP_HEADER (128 + MAX_HEADER) 55 #define MAX_TCP_OPTION_SPACE 40 56 57 /* 58 * Never offer a window over 32767 without using window scaling. Some 59 * poor stacks do signed 16bit maths! 60 */ 61 #define MAX_TCP_WINDOW 32767U 62 63 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 64 #define TCP_MIN_MSS 88U 65 66 /* The least MTU to use for probing */ 67 #define TCP_BASE_MSS 1024 68 69 /* probing interval, default to 10 minutes as per RFC4821 */ 70 #define TCP_PROBE_INTERVAL 600 71 72 /* Specify interval when tcp mtu probing will stop */ 73 #define TCP_PROBE_THRESHOLD 8 74 75 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 76 #define TCP_FASTRETRANS_THRESH 3 77 78 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 79 #define TCP_MAX_QUICKACKS 16U 80 81 /* urg_data states */ 82 #define TCP_URG_VALID 0x0100 83 #define TCP_URG_NOTYET 0x0200 84 #define TCP_URG_READ 0x0400 85 86 #define TCP_RETR1 3 /* 87 * This is how many retries it does before it 88 * tries to figure out if the gateway is 89 * down. Minimal RFC value is 3; it corresponds 90 * to ~3sec-8min depending on RTO. 91 */ 92 93 #define TCP_RETR2 15 /* 94 * This should take at least 95 * 90 minutes to time out. 96 * RFC1122 says that the limit is 100 sec. 97 * 15 is ~13-30min depending on RTO. 98 */ 99 100 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 101 * when active opening a connection. 102 * RFC1122 says the minimum retry MUST 103 * be at least 180secs. Nevertheless 104 * this value is corresponding to 105 * 63secs of retransmission with the 106 * current initial RTO. 107 */ 108 109 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 110 * when passive opening a connection. 111 * This is corresponding to 31secs of 112 * retransmission with the current 113 * initial RTO. 114 */ 115 116 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 117 * state, about 60 seconds */ 118 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 119 /* BSD style FIN_WAIT2 deadlock breaker. 120 * It used to be 3min, new value is 60sec, 121 * to combine FIN-WAIT-2 timeout with 122 * TIME-WAIT timer. 123 */ 124 125 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 126 #if HZ >= 100 127 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 128 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 129 #else 130 #define TCP_DELACK_MIN 4U 131 #define TCP_ATO_MIN 4U 132 #endif 133 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 134 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 135 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 136 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 137 * used as a fallback RTO for the 138 * initial data transmission if no 139 * valid RTT sample has been acquired, 140 * most likely due to retrans in 3WHS. 141 */ 142 143 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 144 * for local resources. 145 */ 146 147 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 148 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 149 #define TCP_KEEPALIVE_INTVL (75*HZ) 150 151 #define MAX_TCP_KEEPIDLE 32767 152 #define MAX_TCP_KEEPINTVL 32767 153 #define MAX_TCP_KEEPCNT 127 154 #define MAX_TCP_SYNCNT 127 155 156 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 157 158 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 159 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 160 * after this time. It should be equal 161 * (or greater than) TCP_TIMEWAIT_LEN 162 * to provide reliability equal to one 163 * provided by timewait state. 164 */ 165 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 166 * timestamps. It must be less than 167 * minimal timewait lifetime. 168 */ 169 /* 170 * TCP option 171 */ 172 173 #define TCPOPT_NOP 1 /* Padding */ 174 #define TCPOPT_EOL 0 /* End of options */ 175 #define TCPOPT_MSS 2 /* Segment size negotiating */ 176 #define TCPOPT_WINDOW 3 /* Window scaling */ 177 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 178 #define TCPOPT_SACK 5 /* SACK Block */ 179 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 180 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 181 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 182 #define TCPOPT_EXP 254 /* Experimental */ 183 /* Magic number to be after the option value for sharing TCP 184 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 185 */ 186 #define TCPOPT_FASTOPEN_MAGIC 0xF989 187 188 /* 189 * TCP option lengths 190 */ 191 192 #define TCPOLEN_MSS 4 193 #define TCPOLEN_WINDOW 3 194 #define TCPOLEN_SACK_PERM 2 195 #define TCPOLEN_TIMESTAMP 10 196 #define TCPOLEN_MD5SIG 18 197 #define TCPOLEN_FASTOPEN_BASE 2 198 #define TCPOLEN_EXP_FASTOPEN_BASE 4 199 200 /* But this is what stacks really send out. */ 201 #define TCPOLEN_TSTAMP_ALIGNED 12 202 #define TCPOLEN_WSCALE_ALIGNED 4 203 #define TCPOLEN_SACKPERM_ALIGNED 4 204 #define TCPOLEN_SACK_BASE 2 205 #define TCPOLEN_SACK_BASE_ALIGNED 4 206 #define TCPOLEN_SACK_PERBLOCK 8 207 #define TCPOLEN_MD5SIG_ALIGNED 20 208 #define TCPOLEN_MSS_ALIGNED 4 209 210 /* Flags in tp->nonagle */ 211 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 212 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 213 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 214 215 /* TCP thin-stream limits */ 216 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 217 218 /* TCP initial congestion window as per rfc6928 */ 219 #define TCP_INIT_CWND 10 220 221 /* Bit Flags for sysctl_tcp_fastopen */ 222 #define TFO_CLIENT_ENABLE 1 223 #define TFO_SERVER_ENABLE 2 224 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 225 226 /* Accept SYN data w/o any cookie option */ 227 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 228 229 /* Force enable TFO on all listeners, i.e., not requiring the 230 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen. 231 */ 232 #define TFO_SERVER_WO_SOCKOPT1 0x400 233 #define TFO_SERVER_WO_SOCKOPT2 0x800 234 235 extern struct inet_timewait_death_row tcp_death_row; 236 237 /* sysctl variables for tcp */ 238 extern int sysctl_tcp_timestamps; 239 extern int sysctl_tcp_window_scaling; 240 extern int sysctl_tcp_sack; 241 extern int sysctl_tcp_fin_timeout; 242 extern int sysctl_tcp_syn_retries; 243 extern int sysctl_tcp_synack_retries; 244 extern int sysctl_tcp_retries1; 245 extern int sysctl_tcp_retries2; 246 extern int sysctl_tcp_orphan_retries; 247 extern int sysctl_tcp_syncookies; 248 extern int sysctl_tcp_fastopen; 249 extern int sysctl_tcp_retrans_collapse; 250 extern int sysctl_tcp_stdurg; 251 extern int sysctl_tcp_rfc1337; 252 extern int sysctl_tcp_abort_on_overflow; 253 extern int sysctl_tcp_max_orphans; 254 extern int sysctl_tcp_fack; 255 extern int sysctl_tcp_reordering; 256 extern int sysctl_tcp_max_reordering; 257 extern int sysctl_tcp_dsack; 258 extern long sysctl_tcp_mem[3]; 259 extern int sysctl_tcp_wmem[3]; 260 extern int sysctl_tcp_rmem[3]; 261 extern int sysctl_tcp_app_win; 262 extern int sysctl_tcp_adv_win_scale; 263 extern int sysctl_tcp_tw_reuse; 264 extern int sysctl_tcp_frto; 265 extern int sysctl_tcp_low_latency; 266 extern int sysctl_tcp_nometrics_save; 267 extern int sysctl_tcp_moderate_rcvbuf; 268 extern int sysctl_tcp_tso_win_divisor; 269 extern int sysctl_tcp_workaround_signed_windows; 270 extern int sysctl_tcp_slow_start_after_idle; 271 extern int sysctl_tcp_thin_linear_timeouts; 272 extern int sysctl_tcp_thin_dupack; 273 extern int sysctl_tcp_early_retrans; 274 extern int sysctl_tcp_limit_output_bytes; 275 extern int sysctl_tcp_challenge_ack_limit; 276 extern unsigned int sysctl_tcp_notsent_lowat; 277 extern int sysctl_tcp_min_tso_segs; 278 extern int sysctl_tcp_min_rtt_wlen; 279 extern int sysctl_tcp_autocorking; 280 extern int sysctl_tcp_invalid_ratelimit; 281 extern int sysctl_tcp_pacing_ss_ratio; 282 extern int sysctl_tcp_pacing_ca_ratio; 283 284 extern atomic_long_t tcp_memory_allocated; 285 extern struct percpu_counter tcp_sockets_allocated; 286 extern int tcp_memory_pressure; 287 288 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 289 static inline bool tcp_under_memory_pressure(const struct sock *sk) 290 { 291 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 292 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 293 return true; 294 295 return tcp_memory_pressure; 296 } 297 /* 298 * The next routines deal with comparing 32 bit unsigned ints 299 * and worry about wraparound (automatic with unsigned arithmetic). 300 */ 301 302 static inline bool before(__u32 seq1, __u32 seq2) 303 { 304 return (__s32)(seq1-seq2) < 0; 305 } 306 #define after(seq2, seq1) before(seq1, seq2) 307 308 /* is s2<=s1<=s3 ? */ 309 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 310 { 311 return seq3 - seq2 >= seq1 - seq2; 312 } 313 314 static inline bool tcp_out_of_memory(struct sock *sk) 315 { 316 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 317 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 318 return true; 319 return false; 320 } 321 322 void sk_forced_mem_schedule(struct sock *sk, int size); 323 324 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 325 { 326 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 327 int orphans = percpu_counter_read_positive(ocp); 328 329 if (orphans << shift > sysctl_tcp_max_orphans) { 330 orphans = percpu_counter_sum_positive(ocp); 331 if (orphans << shift > sysctl_tcp_max_orphans) 332 return true; 333 } 334 return false; 335 } 336 337 bool tcp_check_oom(struct sock *sk, int shift); 338 339 340 extern struct proto tcp_prot; 341 342 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 343 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 344 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 345 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 346 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 347 348 void tcp_tasklet_init(void); 349 350 void tcp_v4_err(struct sk_buff *skb, u32); 351 352 void tcp_shutdown(struct sock *sk, int how); 353 354 void tcp_v4_early_demux(struct sk_buff *skb); 355 int tcp_v4_rcv(struct sk_buff *skb); 356 357 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 358 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 359 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 360 int flags); 361 void tcp_release_cb(struct sock *sk); 362 void tcp_wfree(struct sk_buff *skb); 363 void tcp_write_timer_handler(struct sock *sk); 364 void tcp_delack_timer_handler(struct sock *sk); 365 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 366 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 367 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 368 const struct tcphdr *th, unsigned int len); 369 void tcp_rcv_space_adjust(struct sock *sk); 370 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 371 void tcp_twsk_destructor(struct sock *sk); 372 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 373 struct pipe_inode_info *pipe, size_t len, 374 unsigned int flags); 375 376 static inline void tcp_dec_quickack_mode(struct sock *sk, 377 const unsigned int pkts) 378 { 379 struct inet_connection_sock *icsk = inet_csk(sk); 380 381 if (icsk->icsk_ack.quick) { 382 if (pkts >= icsk->icsk_ack.quick) { 383 icsk->icsk_ack.quick = 0; 384 /* Leaving quickack mode we deflate ATO. */ 385 icsk->icsk_ack.ato = TCP_ATO_MIN; 386 } else 387 icsk->icsk_ack.quick -= pkts; 388 } 389 } 390 391 #define TCP_ECN_OK 1 392 #define TCP_ECN_QUEUE_CWR 2 393 #define TCP_ECN_DEMAND_CWR 4 394 #define TCP_ECN_SEEN 8 395 396 enum tcp_tw_status { 397 TCP_TW_SUCCESS = 0, 398 TCP_TW_RST = 1, 399 TCP_TW_ACK = 2, 400 TCP_TW_SYN = 3 401 }; 402 403 404 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 405 struct sk_buff *skb, 406 const struct tcphdr *th); 407 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 408 struct request_sock *req, bool fastopen); 409 int tcp_child_process(struct sock *parent, struct sock *child, 410 struct sk_buff *skb); 411 void tcp_enter_loss(struct sock *sk); 412 void tcp_clear_retrans(struct tcp_sock *tp); 413 void tcp_update_metrics(struct sock *sk); 414 void tcp_init_metrics(struct sock *sk); 415 void tcp_metrics_init(void); 416 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, 417 bool paws_check, bool timestamps); 418 bool tcp_remember_stamp(struct sock *sk); 419 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); 420 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); 421 void tcp_disable_fack(struct tcp_sock *tp); 422 void tcp_close(struct sock *sk, long timeout); 423 void tcp_init_sock(struct sock *sk); 424 unsigned int tcp_poll(struct file *file, struct socket *sock, 425 struct poll_table_struct *wait); 426 int tcp_getsockopt(struct sock *sk, int level, int optname, 427 char __user *optval, int __user *optlen); 428 int tcp_setsockopt(struct sock *sk, int level, int optname, 429 char __user *optval, unsigned int optlen); 430 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 431 char __user *optval, int __user *optlen); 432 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 433 char __user *optval, unsigned int optlen); 434 void tcp_set_keepalive(struct sock *sk, int val); 435 void tcp_syn_ack_timeout(const struct request_sock *req); 436 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 437 int flags, int *addr_len); 438 void tcp_parse_options(const struct sk_buff *skb, 439 struct tcp_options_received *opt_rx, 440 int estab, struct tcp_fastopen_cookie *foc); 441 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 442 443 /* 444 * TCP v4 functions exported for the inet6 API 445 */ 446 447 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 448 void tcp_v4_mtu_reduced(struct sock *sk); 449 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 450 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 451 struct sock *tcp_create_openreq_child(const struct sock *sk, 452 struct request_sock *req, 453 struct sk_buff *skb); 454 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 455 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 456 struct request_sock *req, 457 struct dst_entry *dst, 458 struct request_sock *req_unhash, 459 bool *own_req); 460 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 461 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 462 int tcp_connect(struct sock *sk); 463 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 464 struct request_sock *req, 465 struct tcp_fastopen_cookie *foc, 466 bool attach_req); 467 int tcp_disconnect(struct sock *sk, int flags); 468 469 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 470 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 471 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 472 473 /* From syncookies.c */ 474 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 475 struct request_sock *req, 476 struct dst_entry *dst); 477 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 478 u32 cookie); 479 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 480 #ifdef CONFIG_SYN_COOKIES 481 482 /* Syncookies use a monotonic timer which increments every 60 seconds. 483 * This counter is used both as a hash input and partially encoded into 484 * the cookie value. A cookie is only validated further if the delta 485 * between the current counter value and the encoded one is less than this, 486 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 487 * the counter advances immediately after a cookie is generated). 488 */ 489 #define MAX_SYNCOOKIE_AGE 2 490 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 491 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 492 493 /* syncookies: remember time of last synqueue overflow 494 * But do not dirty this field too often (once per second is enough) 495 * It is racy as we do not hold a lock, but race is very minor. 496 */ 497 static inline void tcp_synq_overflow(const struct sock *sk) 498 { 499 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 500 unsigned long now = jiffies; 501 502 if (time_after(now, last_overflow + HZ)) 503 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 504 } 505 506 /* syncookies: no recent synqueue overflow on this listening socket? */ 507 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 508 { 509 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 510 511 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID); 512 } 513 514 static inline u32 tcp_cookie_time(void) 515 { 516 u64 val = get_jiffies_64(); 517 518 do_div(val, TCP_SYNCOOKIE_PERIOD); 519 return val; 520 } 521 522 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 523 u16 *mssp); 524 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 525 __u32 cookie_init_timestamp(struct request_sock *req); 526 bool cookie_timestamp_decode(struct tcp_options_received *opt); 527 bool cookie_ecn_ok(const struct tcp_options_received *opt, 528 const struct net *net, const struct dst_entry *dst); 529 530 /* From net/ipv6/syncookies.c */ 531 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 532 u32 cookie); 533 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 534 535 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 536 const struct tcphdr *th, u16 *mssp); 537 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 538 #endif 539 /* tcp_output.c */ 540 541 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 542 int nonagle); 543 bool tcp_may_send_now(struct sock *sk); 544 int __tcp_retransmit_skb(struct sock *, struct sk_buff *); 545 int tcp_retransmit_skb(struct sock *, struct sk_buff *); 546 void tcp_retransmit_timer(struct sock *sk); 547 void tcp_xmit_retransmit_queue(struct sock *); 548 void tcp_simple_retransmit(struct sock *); 549 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 550 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t); 551 552 void tcp_send_probe0(struct sock *); 553 void tcp_send_partial(struct sock *); 554 int tcp_write_wakeup(struct sock *, int mib); 555 void tcp_send_fin(struct sock *sk); 556 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 557 int tcp_send_synack(struct sock *); 558 void tcp_push_one(struct sock *, unsigned int mss_now); 559 void tcp_send_ack(struct sock *sk); 560 void tcp_send_delayed_ack(struct sock *sk); 561 void tcp_send_loss_probe(struct sock *sk); 562 bool tcp_schedule_loss_probe(struct sock *sk); 563 564 /* tcp_input.c */ 565 void tcp_resume_early_retransmit(struct sock *sk); 566 void tcp_rearm_rto(struct sock *sk); 567 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 568 void tcp_reset(struct sock *sk); 569 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 570 571 /* tcp_timer.c */ 572 void tcp_init_xmit_timers(struct sock *); 573 static inline void tcp_clear_xmit_timers(struct sock *sk) 574 { 575 inet_csk_clear_xmit_timers(sk); 576 } 577 578 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 579 unsigned int tcp_current_mss(struct sock *sk); 580 581 /* Bound MSS / TSO packet size with the half of the window */ 582 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 583 { 584 int cutoff; 585 586 /* When peer uses tiny windows, there is no use in packetizing 587 * to sub-MSS pieces for the sake of SWS or making sure there 588 * are enough packets in the pipe for fast recovery. 589 * 590 * On the other hand, for extremely large MSS devices, handling 591 * smaller than MSS windows in this way does make sense. 592 */ 593 if (tp->max_window >= 512) 594 cutoff = (tp->max_window >> 1); 595 else 596 cutoff = tp->max_window; 597 598 if (cutoff && pktsize > cutoff) 599 return max_t(int, cutoff, 68U - tp->tcp_header_len); 600 else 601 return pktsize; 602 } 603 604 /* tcp.c */ 605 void tcp_get_info(struct sock *, struct tcp_info *); 606 607 /* Read 'sendfile()'-style from a TCP socket */ 608 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 609 unsigned int, size_t); 610 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 611 sk_read_actor_t recv_actor); 612 613 void tcp_initialize_rcv_mss(struct sock *sk); 614 615 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 616 int tcp_mss_to_mtu(struct sock *sk, int mss); 617 void tcp_mtup_init(struct sock *sk); 618 void tcp_init_buffer_space(struct sock *sk); 619 620 static inline void tcp_bound_rto(const struct sock *sk) 621 { 622 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 623 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 624 } 625 626 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 627 { 628 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 629 } 630 631 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 632 { 633 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 634 ntohl(TCP_FLAG_ACK) | 635 snd_wnd); 636 } 637 638 static inline void tcp_fast_path_on(struct tcp_sock *tp) 639 { 640 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 641 } 642 643 static inline void tcp_fast_path_check(struct sock *sk) 644 { 645 struct tcp_sock *tp = tcp_sk(sk); 646 647 if (skb_queue_empty(&tp->out_of_order_queue) && 648 tp->rcv_wnd && 649 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 650 !tp->urg_data) 651 tcp_fast_path_on(tp); 652 } 653 654 /* Compute the actual rto_min value */ 655 static inline u32 tcp_rto_min(struct sock *sk) 656 { 657 const struct dst_entry *dst = __sk_dst_get(sk); 658 u32 rto_min = TCP_RTO_MIN; 659 660 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 661 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 662 return rto_min; 663 } 664 665 static inline u32 tcp_rto_min_us(struct sock *sk) 666 { 667 return jiffies_to_usecs(tcp_rto_min(sk)); 668 } 669 670 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 671 { 672 return dst_metric_locked(dst, RTAX_CC_ALGO); 673 } 674 675 /* Minimum RTT in usec. ~0 means not available. */ 676 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 677 { 678 return tp->rtt_min[0].rtt; 679 } 680 681 /* Compute the actual receive window we are currently advertising. 682 * Rcv_nxt can be after the window if our peer push more data 683 * than the offered window. 684 */ 685 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 686 { 687 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 688 689 if (win < 0) 690 win = 0; 691 return (u32) win; 692 } 693 694 /* Choose a new window, without checks for shrinking, and without 695 * scaling applied to the result. The caller does these things 696 * if necessary. This is a "raw" window selection. 697 */ 698 u32 __tcp_select_window(struct sock *sk); 699 700 void tcp_send_window_probe(struct sock *sk); 701 702 /* TCP timestamps are only 32-bits, this causes a slight 703 * complication on 64-bit systems since we store a snapshot 704 * of jiffies in the buffer control blocks below. We decided 705 * to use only the low 32-bits of jiffies and hide the ugly 706 * casts with the following macro. 707 */ 708 #define tcp_time_stamp ((__u32)(jiffies)) 709 710 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 711 { 712 return skb->skb_mstamp.stamp_jiffies; 713 } 714 715 716 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 717 718 #define TCPHDR_FIN 0x01 719 #define TCPHDR_SYN 0x02 720 #define TCPHDR_RST 0x04 721 #define TCPHDR_PSH 0x08 722 #define TCPHDR_ACK 0x10 723 #define TCPHDR_URG 0x20 724 #define TCPHDR_ECE 0x40 725 #define TCPHDR_CWR 0x80 726 727 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 728 729 /* This is what the send packet queuing engine uses to pass 730 * TCP per-packet control information to the transmission code. 731 * We also store the host-order sequence numbers in here too. 732 * This is 44 bytes if IPV6 is enabled. 733 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 734 */ 735 struct tcp_skb_cb { 736 __u32 seq; /* Starting sequence number */ 737 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 738 union { 739 /* Note : tcp_tw_isn is used in input path only 740 * (isn chosen by tcp_timewait_state_process()) 741 * 742 * tcp_gso_segs/size are used in write queue only, 743 * cf tcp_skb_pcount()/tcp_skb_mss() 744 */ 745 __u32 tcp_tw_isn; 746 struct { 747 u16 tcp_gso_segs; 748 u16 tcp_gso_size; 749 }; 750 }; 751 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 752 753 __u8 sacked; /* State flags for SACK/FACK. */ 754 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 755 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 756 #define TCPCB_LOST 0x04 /* SKB is lost */ 757 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 758 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 759 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 760 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 761 TCPCB_REPAIRED) 762 763 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 764 /* 1 byte hole */ 765 __u32 ack_seq; /* Sequence number ACK'd */ 766 union { 767 struct inet_skb_parm h4; 768 #if IS_ENABLED(CONFIG_IPV6) 769 struct inet6_skb_parm h6; 770 #endif 771 } header; /* For incoming frames */ 772 }; 773 774 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 775 776 777 #if IS_ENABLED(CONFIG_IPV6) 778 /* This is the variant of inet6_iif() that must be used by TCP, 779 * as TCP moves IP6CB into a different location in skb->cb[] 780 */ 781 static inline int tcp_v6_iif(const struct sk_buff *skb) 782 { 783 return TCP_SKB_CB(skb)->header.h6.iif; 784 } 785 #endif 786 787 /* Due to TSO, an SKB can be composed of multiple actual 788 * packets. To keep these tracked properly, we use this. 789 */ 790 static inline int tcp_skb_pcount(const struct sk_buff *skb) 791 { 792 return TCP_SKB_CB(skb)->tcp_gso_segs; 793 } 794 795 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 796 { 797 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 798 } 799 800 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 801 { 802 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 803 } 804 805 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 806 static inline int tcp_skb_mss(const struct sk_buff *skb) 807 { 808 return TCP_SKB_CB(skb)->tcp_gso_size; 809 } 810 811 /* Events passed to congestion control interface */ 812 enum tcp_ca_event { 813 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 814 CA_EVENT_CWND_RESTART, /* congestion window restart */ 815 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 816 CA_EVENT_LOSS, /* loss timeout */ 817 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 818 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 819 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 820 CA_EVENT_NON_DELAYED_ACK, 821 }; 822 823 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 824 enum tcp_ca_ack_event_flags { 825 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 826 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 827 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 828 }; 829 830 /* 831 * Interface for adding new TCP congestion control handlers 832 */ 833 #define TCP_CA_NAME_MAX 16 834 #define TCP_CA_MAX 128 835 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 836 837 #define TCP_CA_UNSPEC 0 838 839 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 840 #define TCP_CONG_NON_RESTRICTED 0x1 841 /* Requires ECN/ECT set on all packets */ 842 #define TCP_CONG_NEEDS_ECN 0x2 843 844 union tcp_cc_info; 845 846 struct tcp_congestion_ops { 847 struct list_head list; 848 u32 key; 849 u32 flags; 850 851 /* initialize private data (optional) */ 852 void (*init)(struct sock *sk); 853 /* cleanup private data (optional) */ 854 void (*release)(struct sock *sk); 855 856 /* return slow start threshold (required) */ 857 u32 (*ssthresh)(struct sock *sk); 858 /* do new cwnd calculation (required) */ 859 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 860 /* call before changing ca_state (optional) */ 861 void (*set_state)(struct sock *sk, u8 new_state); 862 /* call when cwnd event occurs (optional) */ 863 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 864 /* call when ack arrives (optional) */ 865 void (*in_ack_event)(struct sock *sk, u32 flags); 866 /* new value of cwnd after loss (optional) */ 867 u32 (*undo_cwnd)(struct sock *sk); 868 /* hook for packet ack accounting (optional) */ 869 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 870 /* get info for inet_diag (optional) */ 871 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 872 union tcp_cc_info *info); 873 874 char name[TCP_CA_NAME_MAX]; 875 struct module *owner; 876 }; 877 878 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 879 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 880 881 void tcp_assign_congestion_control(struct sock *sk); 882 void tcp_init_congestion_control(struct sock *sk); 883 void tcp_cleanup_congestion_control(struct sock *sk); 884 int tcp_set_default_congestion_control(const char *name); 885 void tcp_get_default_congestion_control(char *name); 886 void tcp_get_available_congestion_control(char *buf, size_t len); 887 void tcp_get_allowed_congestion_control(char *buf, size_t len); 888 int tcp_set_allowed_congestion_control(char *allowed); 889 int tcp_set_congestion_control(struct sock *sk, const char *name); 890 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 891 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 892 893 u32 tcp_reno_ssthresh(struct sock *sk); 894 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 895 extern struct tcp_congestion_ops tcp_reno; 896 897 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 898 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca); 899 #ifdef CONFIG_INET 900 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 901 #else 902 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 903 { 904 return NULL; 905 } 906 #endif 907 908 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 909 { 910 const struct inet_connection_sock *icsk = inet_csk(sk); 911 912 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 913 } 914 915 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 916 { 917 struct inet_connection_sock *icsk = inet_csk(sk); 918 919 if (icsk->icsk_ca_ops->set_state) 920 icsk->icsk_ca_ops->set_state(sk, ca_state); 921 icsk->icsk_ca_state = ca_state; 922 } 923 924 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 925 { 926 const struct inet_connection_sock *icsk = inet_csk(sk); 927 928 if (icsk->icsk_ca_ops->cwnd_event) 929 icsk->icsk_ca_ops->cwnd_event(sk, event); 930 } 931 932 /* These functions determine how the current flow behaves in respect of SACK 933 * handling. SACK is negotiated with the peer, and therefore it can vary 934 * between different flows. 935 * 936 * tcp_is_sack - SACK enabled 937 * tcp_is_reno - No SACK 938 * tcp_is_fack - FACK enabled, implies SACK enabled 939 */ 940 static inline int tcp_is_sack(const struct tcp_sock *tp) 941 { 942 return tp->rx_opt.sack_ok; 943 } 944 945 static inline bool tcp_is_reno(const struct tcp_sock *tp) 946 { 947 return !tcp_is_sack(tp); 948 } 949 950 static inline bool tcp_is_fack(const struct tcp_sock *tp) 951 { 952 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 953 } 954 955 static inline void tcp_enable_fack(struct tcp_sock *tp) 956 { 957 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 958 } 959 960 /* TCP early-retransmit (ER) is similar to but more conservative than 961 * the thin-dupack feature. Enable ER only if thin-dupack is disabled. 962 */ 963 static inline void tcp_enable_early_retrans(struct tcp_sock *tp) 964 { 965 tp->do_early_retrans = sysctl_tcp_early_retrans && 966 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack && 967 sysctl_tcp_reordering == 3; 968 } 969 970 static inline void tcp_disable_early_retrans(struct tcp_sock *tp) 971 { 972 tp->do_early_retrans = 0; 973 } 974 975 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 976 { 977 return tp->sacked_out + tp->lost_out; 978 } 979 980 /* This determines how many packets are "in the network" to the best 981 * of our knowledge. In many cases it is conservative, but where 982 * detailed information is available from the receiver (via SACK 983 * blocks etc.) we can make more aggressive calculations. 984 * 985 * Use this for decisions involving congestion control, use just 986 * tp->packets_out to determine if the send queue is empty or not. 987 * 988 * Read this equation as: 989 * 990 * "Packets sent once on transmission queue" MINUS 991 * "Packets left network, but not honestly ACKed yet" PLUS 992 * "Packets fast retransmitted" 993 */ 994 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 995 { 996 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 997 } 998 999 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1000 1001 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1002 { 1003 return tp->snd_cwnd < tp->snd_ssthresh; 1004 } 1005 1006 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1007 { 1008 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1009 } 1010 1011 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1012 { 1013 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1014 (1 << inet_csk(sk)->icsk_ca_state); 1015 } 1016 1017 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1018 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1019 * ssthresh. 1020 */ 1021 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1022 { 1023 const struct tcp_sock *tp = tcp_sk(sk); 1024 1025 if (tcp_in_cwnd_reduction(sk)) 1026 return tp->snd_ssthresh; 1027 else 1028 return max(tp->snd_ssthresh, 1029 ((tp->snd_cwnd >> 1) + 1030 (tp->snd_cwnd >> 2))); 1031 } 1032 1033 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1034 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1035 1036 void tcp_enter_cwr(struct sock *sk); 1037 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1038 1039 /* The maximum number of MSS of available cwnd for which TSO defers 1040 * sending if not using sysctl_tcp_tso_win_divisor. 1041 */ 1042 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1043 { 1044 return 3; 1045 } 1046 1047 /* Slow start with delack produces 3 packets of burst, so that 1048 * it is safe "de facto". This will be the default - same as 1049 * the default reordering threshold - but if reordering increases, 1050 * we must be able to allow cwnd to burst at least this much in order 1051 * to not pull it back when holes are filled. 1052 */ 1053 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 1054 { 1055 return tp->reordering; 1056 } 1057 1058 /* Returns end sequence number of the receiver's advertised window */ 1059 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1060 { 1061 return tp->snd_una + tp->snd_wnd; 1062 } 1063 1064 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1065 * flexible approach. The RFC suggests cwnd should not be raised unless 1066 * it was fully used previously. And that's exactly what we do in 1067 * congestion avoidance mode. But in slow start we allow cwnd to grow 1068 * as long as the application has used half the cwnd. 1069 * Example : 1070 * cwnd is 10 (IW10), but application sends 9 frames. 1071 * We allow cwnd to reach 18 when all frames are ACKed. 1072 * This check is safe because it's as aggressive as slow start which already 1073 * risks 100% overshoot. The advantage is that we discourage application to 1074 * either send more filler packets or data to artificially blow up the cwnd 1075 * usage, and allow application-limited process to probe bw more aggressively. 1076 */ 1077 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1078 { 1079 const struct tcp_sock *tp = tcp_sk(sk); 1080 1081 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1082 if (tcp_in_slow_start(tp)) 1083 return tp->snd_cwnd < 2 * tp->max_packets_out; 1084 1085 return tp->is_cwnd_limited; 1086 } 1087 1088 /* Something is really bad, we could not queue an additional packet, 1089 * because qdisc is full or receiver sent a 0 window. 1090 * We do not want to add fuel to the fire, or abort too early, 1091 * so make sure the timer we arm now is at least 200ms in the future, 1092 * regardless of current icsk_rto value (as it could be ~2ms) 1093 */ 1094 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1095 { 1096 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1097 } 1098 1099 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1100 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1101 unsigned long max_when) 1102 { 1103 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1104 1105 return (unsigned long)min_t(u64, when, max_when); 1106 } 1107 1108 static inline void tcp_check_probe_timer(struct sock *sk) 1109 { 1110 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1111 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1112 tcp_probe0_base(sk), TCP_RTO_MAX); 1113 } 1114 1115 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1116 { 1117 tp->snd_wl1 = seq; 1118 } 1119 1120 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1121 { 1122 tp->snd_wl1 = seq; 1123 } 1124 1125 /* 1126 * Calculate(/check) TCP checksum 1127 */ 1128 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1129 __be32 daddr, __wsum base) 1130 { 1131 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1132 } 1133 1134 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1135 { 1136 return __skb_checksum_complete(skb); 1137 } 1138 1139 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1140 { 1141 return !skb_csum_unnecessary(skb) && 1142 __tcp_checksum_complete(skb); 1143 } 1144 1145 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1146 1147 static inline void tcp_prequeue_init(struct tcp_sock *tp) 1148 { 1149 tp->ucopy.task = NULL; 1150 tp->ucopy.len = 0; 1151 tp->ucopy.memory = 0; 1152 skb_queue_head_init(&tp->ucopy.prequeue); 1153 } 1154 1155 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1156 1157 #undef STATE_TRACE 1158 1159 #ifdef STATE_TRACE 1160 static const char *statename[]={ 1161 "Unused","Established","Syn Sent","Syn Recv", 1162 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1163 "Close Wait","Last ACK","Listen","Closing" 1164 }; 1165 #endif 1166 void tcp_set_state(struct sock *sk, int state); 1167 1168 void tcp_done(struct sock *sk); 1169 1170 int tcp_abort(struct sock *sk, int err); 1171 1172 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1173 { 1174 rx_opt->dsack = 0; 1175 rx_opt->num_sacks = 0; 1176 } 1177 1178 u32 tcp_default_init_rwnd(u32 mss); 1179 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1180 1181 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1182 { 1183 struct tcp_sock *tp = tcp_sk(sk); 1184 s32 delta; 1185 1186 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out) 1187 return; 1188 delta = tcp_time_stamp - tp->lsndtime; 1189 if (delta > inet_csk(sk)->icsk_rto) 1190 tcp_cwnd_restart(sk, delta); 1191 } 1192 1193 /* Determine a window scaling and initial window to offer. */ 1194 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1195 __u32 *window_clamp, int wscale_ok, 1196 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1197 1198 static inline int tcp_win_from_space(int space) 1199 { 1200 return sysctl_tcp_adv_win_scale<=0 ? 1201 (space>>(-sysctl_tcp_adv_win_scale)) : 1202 space - (space>>sysctl_tcp_adv_win_scale); 1203 } 1204 1205 /* Note: caller must be prepared to deal with negative returns */ 1206 static inline int tcp_space(const struct sock *sk) 1207 { 1208 return tcp_win_from_space(sk->sk_rcvbuf - 1209 atomic_read(&sk->sk_rmem_alloc)); 1210 } 1211 1212 static inline int tcp_full_space(const struct sock *sk) 1213 { 1214 return tcp_win_from_space(sk->sk_rcvbuf); 1215 } 1216 1217 extern void tcp_openreq_init_rwin(struct request_sock *req, 1218 const struct sock *sk_listener, 1219 const struct dst_entry *dst); 1220 1221 void tcp_enter_memory_pressure(struct sock *sk); 1222 1223 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1224 { 1225 struct net *net = sock_net((struct sock *)tp); 1226 1227 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1228 } 1229 1230 static inline int keepalive_time_when(const struct tcp_sock *tp) 1231 { 1232 struct net *net = sock_net((struct sock *)tp); 1233 1234 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1235 } 1236 1237 static inline int keepalive_probes(const struct tcp_sock *tp) 1238 { 1239 struct net *net = sock_net((struct sock *)tp); 1240 1241 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1242 } 1243 1244 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1245 { 1246 const struct inet_connection_sock *icsk = &tp->inet_conn; 1247 1248 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1249 tcp_time_stamp - tp->rcv_tstamp); 1250 } 1251 1252 static inline int tcp_fin_time(const struct sock *sk) 1253 { 1254 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1255 const int rto = inet_csk(sk)->icsk_rto; 1256 1257 if (fin_timeout < (rto << 2) - (rto >> 1)) 1258 fin_timeout = (rto << 2) - (rto >> 1); 1259 1260 return fin_timeout; 1261 } 1262 1263 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1264 int paws_win) 1265 { 1266 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1267 return true; 1268 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1269 return true; 1270 /* 1271 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1272 * then following tcp messages have valid values. Ignore 0 value, 1273 * or else 'negative' tsval might forbid us to accept their packets. 1274 */ 1275 if (!rx_opt->ts_recent) 1276 return true; 1277 return false; 1278 } 1279 1280 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1281 int rst) 1282 { 1283 if (tcp_paws_check(rx_opt, 0)) 1284 return false; 1285 1286 /* RST segments are not recommended to carry timestamp, 1287 and, if they do, it is recommended to ignore PAWS because 1288 "their cleanup function should take precedence over timestamps." 1289 Certainly, it is mistake. It is necessary to understand the reasons 1290 of this constraint to relax it: if peer reboots, clock may go 1291 out-of-sync and half-open connections will not be reset. 1292 Actually, the problem would be not existing if all 1293 the implementations followed draft about maintaining clock 1294 via reboots. Linux-2.2 DOES NOT! 1295 1296 However, we can relax time bounds for RST segments to MSL. 1297 */ 1298 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1299 return false; 1300 return true; 1301 } 1302 1303 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1304 int mib_idx, u32 *last_oow_ack_time); 1305 1306 static inline void tcp_mib_init(struct net *net) 1307 { 1308 /* See RFC 2012 */ 1309 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1310 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1311 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1312 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1313 } 1314 1315 /* from STCP */ 1316 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1317 { 1318 tp->lost_skb_hint = NULL; 1319 } 1320 1321 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1322 { 1323 tcp_clear_retrans_hints_partial(tp); 1324 tp->retransmit_skb_hint = NULL; 1325 } 1326 1327 union tcp_md5_addr { 1328 struct in_addr a4; 1329 #if IS_ENABLED(CONFIG_IPV6) 1330 struct in6_addr a6; 1331 #endif 1332 }; 1333 1334 /* - key database */ 1335 struct tcp_md5sig_key { 1336 struct hlist_node node; 1337 u8 keylen; 1338 u8 family; /* AF_INET or AF_INET6 */ 1339 union tcp_md5_addr addr; 1340 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1341 struct rcu_head rcu; 1342 }; 1343 1344 /* - sock block */ 1345 struct tcp_md5sig_info { 1346 struct hlist_head head; 1347 struct rcu_head rcu; 1348 }; 1349 1350 /* - pseudo header */ 1351 struct tcp4_pseudohdr { 1352 __be32 saddr; 1353 __be32 daddr; 1354 __u8 pad; 1355 __u8 protocol; 1356 __be16 len; 1357 }; 1358 1359 struct tcp6_pseudohdr { 1360 struct in6_addr saddr; 1361 struct in6_addr daddr; 1362 __be32 len; 1363 __be32 protocol; /* including padding */ 1364 }; 1365 1366 union tcp_md5sum_block { 1367 struct tcp4_pseudohdr ip4; 1368 #if IS_ENABLED(CONFIG_IPV6) 1369 struct tcp6_pseudohdr ip6; 1370 #endif 1371 }; 1372 1373 /* - pool: digest algorithm, hash description and scratch buffer */ 1374 struct tcp_md5sig_pool { 1375 struct ahash_request *md5_req; 1376 union tcp_md5sum_block md5_blk; 1377 }; 1378 1379 /* - functions */ 1380 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1381 const struct sock *sk, const struct sk_buff *skb); 1382 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1383 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); 1384 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1385 int family); 1386 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1387 const struct sock *addr_sk); 1388 1389 #ifdef CONFIG_TCP_MD5SIG 1390 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1391 const union tcp_md5_addr *addr, 1392 int family); 1393 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1394 #else 1395 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1396 const union tcp_md5_addr *addr, 1397 int family) 1398 { 1399 return NULL; 1400 } 1401 #define tcp_twsk_md5_key(twsk) NULL 1402 #endif 1403 1404 bool tcp_alloc_md5sig_pool(void); 1405 1406 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1407 static inline void tcp_put_md5sig_pool(void) 1408 { 1409 local_bh_enable(); 1410 } 1411 1412 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1413 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1414 unsigned int header_len); 1415 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1416 const struct tcp_md5sig_key *key); 1417 1418 /* From tcp_fastopen.c */ 1419 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1420 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1421 unsigned long *last_syn_loss); 1422 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1423 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1424 u16 try_exp); 1425 struct tcp_fastopen_request { 1426 /* Fast Open cookie. Size 0 means a cookie request */ 1427 struct tcp_fastopen_cookie cookie; 1428 struct msghdr *data; /* data in MSG_FASTOPEN */ 1429 size_t size; 1430 int copied; /* queued in tcp_connect() */ 1431 }; 1432 void tcp_free_fastopen_req(struct tcp_sock *tp); 1433 1434 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1435 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1436 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1437 struct request_sock *req, 1438 struct tcp_fastopen_cookie *foc, 1439 struct dst_entry *dst); 1440 void tcp_fastopen_init_key_once(bool publish); 1441 #define TCP_FASTOPEN_KEY_LENGTH 16 1442 1443 /* Fastopen key context */ 1444 struct tcp_fastopen_context { 1445 struct crypto_cipher *tfm; 1446 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1447 struct rcu_head rcu; 1448 }; 1449 1450 /* write queue abstraction */ 1451 static inline void tcp_write_queue_purge(struct sock *sk) 1452 { 1453 struct sk_buff *skb; 1454 1455 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1456 sk_wmem_free_skb(sk, skb); 1457 sk_mem_reclaim(sk); 1458 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1459 } 1460 1461 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1462 { 1463 return skb_peek(&sk->sk_write_queue); 1464 } 1465 1466 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1467 { 1468 return skb_peek_tail(&sk->sk_write_queue); 1469 } 1470 1471 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1472 const struct sk_buff *skb) 1473 { 1474 return skb_queue_next(&sk->sk_write_queue, skb); 1475 } 1476 1477 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1478 const struct sk_buff *skb) 1479 { 1480 return skb_queue_prev(&sk->sk_write_queue, skb); 1481 } 1482 1483 #define tcp_for_write_queue(skb, sk) \ 1484 skb_queue_walk(&(sk)->sk_write_queue, skb) 1485 1486 #define tcp_for_write_queue_from(skb, sk) \ 1487 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1488 1489 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1490 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1491 1492 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1493 { 1494 return sk->sk_send_head; 1495 } 1496 1497 static inline bool tcp_skb_is_last(const struct sock *sk, 1498 const struct sk_buff *skb) 1499 { 1500 return skb_queue_is_last(&sk->sk_write_queue, skb); 1501 } 1502 1503 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1504 { 1505 if (tcp_skb_is_last(sk, skb)) 1506 sk->sk_send_head = NULL; 1507 else 1508 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1509 } 1510 1511 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1512 { 1513 if (sk->sk_send_head == skb_unlinked) 1514 sk->sk_send_head = NULL; 1515 } 1516 1517 static inline void tcp_init_send_head(struct sock *sk) 1518 { 1519 sk->sk_send_head = NULL; 1520 } 1521 1522 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1523 { 1524 __skb_queue_tail(&sk->sk_write_queue, skb); 1525 } 1526 1527 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1528 { 1529 __tcp_add_write_queue_tail(sk, skb); 1530 1531 /* Queue it, remembering where we must start sending. */ 1532 if (sk->sk_send_head == NULL) { 1533 sk->sk_send_head = skb; 1534 1535 if (tcp_sk(sk)->highest_sack == NULL) 1536 tcp_sk(sk)->highest_sack = skb; 1537 } 1538 } 1539 1540 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1541 { 1542 __skb_queue_head(&sk->sk_write_queue, skb); 1543 } 1544 1545 /* Insert buff after skb on the write queue of sk. */ 1546 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1547 struct sk_buff *buff, 1548 struct sock *sk) 1549 { 1550 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1551 } 1552 1553 /* Insert new before skb on the write queue of sk. */ 1554 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1555 struct sk_buff *skb, 1556 struct sock *sk) 1557 { 1558 __skb_queue_before(&sk->sk_write_queue, skb, new); 1559 1560 if (sk->sk_send_head == skb) 1561 sk->sk_send_head = new; 1562 } 1563 1564 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1565 { 1566 __skb_unlink(skb, &sk->sk_write_queue); 1567 } 1568 1569 static inline bool tcp_write_queue_empty(struct sock *sk) 1570 { 1571 return skb_queue_empty(&sk->sk_write_queue); 1572 } 1573 1574 static inline void tcp_push_pending_frames(struct sock *sk) 1575 { 1576 if (tcp_send_head(sk)) { 1577 struct tcp_sock *tp = tcp_sk(sk); 1578 1579 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1580 } 1581 } 1582 1583 /* Start sequence of the skb just after the highest skb with SACKed 1584 * bit, valid only if sacked_out > 0 or when the caller has ensured 1585 * validity by itself. 1586 */ 1587 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1588 { 1589 if (!tp->sacked_out) 1590 return tp->snd_una; 1591 1592 if (tp->highest_sack == NULL) 1593 return tp->snd_nxt; 1594 1595 return TCP_SKB_CB(tp->highest_sack)->seq; 1596 } 1597 1598 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1599 { 1600 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1601 tcp_write_queue_next(sk, skb); 1602 } 1603 1604 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1605 { 1606 return tcp_sk(sk)->highest_sack; 1607 } 1608 1609 static inline void tcp_highest_sack_reset(struct sock *sk) 1610 { 1611 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1612 } 1613 1614 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1615 static inline void tcp_highest_sack_combine(struct sock *sk, 1616 struct sk_buff *old, 1617 struct sk_buff *new) 1618 { 1619 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1620 tcp_sk(sk)->highest_sack = new; 1621 } 1622 1623 /* This helper checks if socket has IP_TRANSPARENT set */ 1624 static inline bool inet_sk_transparent(const struct sock *sk) 1625 { 1626 switch (sk->sk_state) { 1627 case TCP_TIME_WAIT: 1628 return inet_twsk(sk)->tw_transparent; 1629 case TCP_NEW_SYN_RECV: 1630 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1631 } 1632 return inet_sk(sk)->transparent; 1633 } 1634 1635 /* Determines whether this is a thin stream (which may suffer from 1636 * increased latency). Used to trigger latency-reducing mechanisms. 1637 */ 1638 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1639 { 1640 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1641 } 1642 1643 /* /proc */ 1644 enum tcp_seq_states { 1645 TCP_SEQ_STATE_LISTENING, 1646 TCP_SEQ_STATE_ESTABLISHED, 1647 }; 1648 1649 int tcp_seq_open(struct inode *inode, struct file *file); 1650 1651 struct tcp_seq_afinfo { 1652 char *name; 1653 sa_family_t family; 1654 const struct file_operations *seq_fops; 1655 struct seq_operations seq_ops; 1656 }; 1657 1658 struct tcp_iter_state { 1659 struct seq_net_private p; 1660 sa_family_t family; 1661 enum tcp_seq_states state; 1662 struct sock *syn_wait_sk; 1663 int bucket, offset, sbucket, num; 1664 loff_t last_pos; 1665 }; 1666 1667 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1668 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1669 1670 extern struct request_sock_ops tcp_request_sock_ops; 1671 extern struct request_sock_ops tcp6_request_sock_ops; 1672 1673 void tcp_v4_destroy_sock(struct sock *sk); 1674 1675 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1676 netdev_features_t features); 1677 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1678 int tcp_gro_complete(struct sk_buff *skb); 1679 1680 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1681 1682 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1683 { 1684 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat; 1685 } 1686 1687 static inline bool tcp_stream_memory_free(const struct sock *sk) 1688 { 1689 const struct tcp_sock *tp = tcp_sk(sk); 1690 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1691 1692 return notsent_bytes < tcp_notsent_lowat(tp); 1693 } 1694 1695 #ifdef CONFIG_PROC_FS 1696 int tcp4_proc_init(void); 1697 void tcp4_proc_exit(void); 1698 #endif 1699 1700 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1701 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1702 const struct tcp_request_sock_ops *af_ops, 1703 struct sock *sk, struct sk_buff *skb); 1704 1705 /* TCP af-specific functions */ 1706 struct tcp_sock_af_ops { 1707 #ifdef CONFIG_TCP_MD5SIG 1708 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1709 const struct sock *addr_sk); 1710 int (*calc_md5_hash)(char *location, 1711 const struct tcp_md5sig_key *md5, 1712 const struct sock *sk, 1713 const struct sk_buff *skb); 1714 int (*md5_parse)(struct sock *sk, 1715 char __user *optval, 1716 int optlen); 1717 #endif 1718 }; 1719 1720 struct tcp_request_sock_ops { 1721 u16 mss_clamp; 1722 #ifdef CONFIG_TCP_MD5SIG 1723 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1724 const struct sock *addr_sk); 1725 int (*calc_md5_hash) (char *location, 1726 const struct tcp_md5sig_key *md5, 1727 const struct sock *sk, 1728 const struct sk_buff *skb); 1729 #endif 1730 void (*init_req)(struct request_sock *req, 1731 const struct sock *sk_listener, 1732 struct sk_buff *skb); 1733 #ifdef CONFIG_SYN_COOKIES 1734 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1735 __u16 *mss); 1736 #endif 1737 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1738 const struct request_sock *req, 1739 bool *strict); 1740 __u32 (*init_seq)(const struct sk_buff *skb); 1741 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1742 struct flowi *fl, struct request_sock *req, 1743 struct tcp_fastopen_cookie *foc, 1744 bool attach_req); 1745 }; 1746 1747 #ifdef CONFIG_SYN_COOKIES 1748 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1749 const struct sock *sk, struct sk_buff *skb, 1750 __u16 *mss) 1751 { 1752 tcp_synq_overflow(sk); 1753 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1754 return ops->cookie_init_seq(skb, mss); 1755 } 1756 #else 1757 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1758 const struct sock *sk, struct sk_buff *skb, 1759 __u16 *mss) 1760 { 1761 return 0; 1762 } 1763 #endif 1764 1765 int tcpv4_offload_init(void); 1766 1767 void tcp_v4_init(void); 1768 void tcp_init(void); 1769 1770 /* tcp_recovery.c */ 1771 1772 /* Flags to enable various loss recovery features. See below */ 1773 extern int sysctl_tcp_recovery; 1774 1775 /* Use TCP RACK to detect (some) tail and retransmit losses */ 1776 #define TCP_RACK_LOST_RETRANS 0x1 1777 1778 extern int tcp_rack_mark_lost(struct sock *sk); 1779 1780 extern void tcp_rack_advance(struct tcp_sock *tp, 1781 const struct skb_mstamp *xmit_time, u8 sacked); 1782 1783 /* 1784 * Save and compile IPv4 options, return a pointer to it 1785 */ 1786 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb) 1787 { 1788 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1789 struct ip_options_rcu *dopt = NULL; 1790 1791 if (opt->optlen) { 1792 int opt_size = sizeof(*dopt) + opt->optlen; 1793 1794 dopt = kmalloc(opt_size, GFP_ATOMIC); 1795 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) { 1796 kfree(dopt); 1797 dopt = NULL; 1798 } 1799 } 1800 return dopt; 1801 } 1802 1803 /* locally generated TCP pure ACKs have skb->truesize == 2 1804 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1805 * This is much faster than dissecting the packet to find out. 1806 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1807 */ 1808 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1809 { 1810 return skb->truesize == 2; 1811 } 1812 1813 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1814 { 1815 skb->truesize = 2; 1816 } 1817 1818 #endif /* _TCP_H */ 1819