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_fastopen; 242 extern int sysctl_tcp_retrans_collapse; 243 extern int sysctl_tcp_stdurg; 244 extern int sysctl_tcp_rfc1337; 245 extern int sysctl_tcp_abort_on_overflow; 246 extern int sysctl_tcp_max_orphans; 247 extern int sysctl_tcp_fack; 248 extern int sysctl_tcp_reordering; 249 extern int sysctl_tcp_max_reordering; 250 extern int sysctl_tcp_dsack; 251 extern long sysctl_tcp_mem[3]; 252 extern int sysctl_tcp_wmem[3]; 253 extern int sysctl_tcp_rmem[3]; 254 extern int sysctl_tcp_app_win; 255 extern int sysctl_tcp_adv_win_scale; 256 extern int sysctl_tcp_tw_reuse; 257 extern int sysctl_tcp_frto; 258 extern int sysctl_tcp_low_latency; 259 extern int sysctl_tcp_nometrics_save; 260 extern int sysctl_tcp_moderate_rcvbuf; 261 extern int sysctl_tcp_tso_win_divisor; 262 extern int sysctl_tcp_workaround_signed_windows; 263 extern int sysctl_tcp_slow_start_after_idle; 264 extern int sysctl_tcp_thin_linear_timeouts; 265 extern int sysctl_tcp_thin_dupack; 266 extern int sysctl_tcp_early_retrans; 267 extern int sysctl_tcp_limit_output_bytes; 268 extern int sysctl_tcp_challenge_ack_limit; 269 extern int sysctl_tcp_min_tso_segs; 270 extern int sysctl_tcp_min_rtt_wlen; 271 extern int sysctl_tcp_autocorking; 272 extern int sysctl_tcp_invalid_ratelimit; 273 extern int sysctl_tcp_pacing_ss_ratio; 274 extern int sysctl_tcp_pacing_ca_ratio; 275 276 extern atomic_long_t tcp_memory_allocated; 277 extern struct percpu_counter tcp_sockets_allocated; 278 extern int tcp_memory_pressure; 279 280 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 281 static inline bool tcp_under_memory_pressure(const struct sock *sk) 282 { 283 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 284 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 285 return true; 286 287 return tcp_memory_pressure; 288 } 289 /* 290 * The next routines deal with comparing 32 bit unsigned ints 291 * and worry about wraparound (automatic with unsigned arithmetic). 292 */ 293 294 static inline bool before(__u32 seq1, __u32 seq2) 295 { 296 return (__s32)(seq1-seq2) < 0; 297 } 298 #define after(seq2, seq1) before(seq1, seq2) 299 300 /* is s2<=s1<=s3 ? */ 301 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 302 { 303 return seq3 - seq2 >= seq1 - seq2; 304 } 305 306 static inline bool tcp_out_of_memory(struct sock *sk) 307 { 308 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 309 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 310 return true; 311 return false; 312 } 313 314 void sk_forced_mem_schedule(struct sock *sk, int size); 315 316 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 317 { 318 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 319 int orphans = percpu_counter_read_positive(ocp); 320 321 if (orphans << shift > sysctl_tcp_max_orphans) { 322 orphans = percpu_counter_sum_positive(ocp); 323 if (orphans << shift > sysctl_tcp_max_orphans) 324 return true; 325 } 326 return false; 327 } 328 329 bool tcp_check_oom(struct sock *sk, int shift); 330 331 332 extern struct proto tcp_prot; 333 334 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 335 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 336 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 337 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 338 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 339 340 void tcp_tasklet_init(void); 341 342 void tcp_v4_err(struct sk_buff *skb, u32); 343 344 void tcp_shutdown(struct sock *sk, int how); 345 346 void tcp_v4_early_demux(struct sk_buff *skb); 347 int tcp_v4_rcv(struct sk_buff *skb); 348 349 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 350 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 351 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 352 int flags); 353 void tcp_release_cb(struct sock *sk); 354 void tcp_wfree(struct sk_buff *skb); 355 void tcp_write_timer_handler(struct sock *sk); 356 void tcp_delack_timer_handler(struct sock *sk); 357 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 358 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 359 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 360 const struct tcphdr *th, unsigned int len); 361 void tcp_rcv_space_adjust(struct sock *sk); 362 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 363 void tcp_twsk_destructor(struct sock *sk); 364 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 365 struct pipe_inode_info *pipe, size_t len, 366 unsigned int flags); 367 368 static inline void tcp_dec_quickack_mode(struct sock *sk, 369 const unsigned int pkts) 370 { 371 struct inet_connection_sock *icsk = inet_csk(sk); 372 373 if (icsk->icsk_ack.quick) { 374 if (pkts >= icsk->icsk_ack.quick) { 375 icsk->icsk_ack.quick = 0; 376 /* Leaving quickack mode we deflate ATO. */ 377 icsk->icsk_ack.ato = TCP_ATO_MIN; 378 } else 379 icsk->icsk_ack.quick -= pkts; 380 } 381 } 382 383 #define TCP_ECN_OK 1 384 #define TCP_ECN_QUEUE_CWR 2 385 #define TCP_ECN_DEMAND_CWR 4 386 #define TCP_ECN_SEEN 8 387 388 enum tcp_tw_status { 389 TCP_TW_SUCCESS = 0, 390 TCP_TW_RST = 1, 391 TCP_TW_ACK = 2, 392 TCP_TW_SYN = 3 393 }; 394 395 396 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 397 struct sk_buff *skb, 398 const struct tcphdr *th); 399 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 400 struct request_sock *req, bool fastopen); 401 int tcp_child_process(struct sock *parent, struct sock *child, 402 struct sk_buff *skb); 403 void tcp_enter_loss(struct sock *sk); 404 void tcp_clear_retrans(struct tcp_sock *tp); 405 void tcp_update_metrics(struct sock *sk); 406 void tcp_init_metrics(struct sock *sk); 407 void tcp_metrics_init(void); 408 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, 409 bool paws_check, bool timestamps); 410 bool tcp_remember_stamp(struct sock *sk); 411 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); 412 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); 413 void tcp_disable_fack(struct tcp_sock *tp); 414 void tcp_close(struct sock *sk, long timeout); 415 void tcp_init_sock(struct sock *sk); 416 unsigned int tcp_poll(struct file *file, struct socket *sock, 417 struct poll_table_struct *wait); 418 int tcp_getsockopt(struct sock *sk, int level, int optname, 419 char __user *optval, int __user *optlen); 420 int tcp_setsockopt(struct sock *sk, int level, int optname, 421 char __user *optval, unsigned int optlen); 422 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 423 char __user *optval, int __user *optlen); 424 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 425 char __user *optval, unsigned int optlen); 426 void tcp_set_keepalive(struct sock *sk, int val); 427 void tcp_syn_ack_timeout(const struct request_sock *req); 428 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 429 int flags, int *addr_len); 430 void tcp_parse_options(const struct sk_buff *skb, 431 struct tcp_options_received *opt_rx, 432 int estab, struct tcp_fastopen_cookie *foc); 433 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 434 435 /* 436 * TCP v4 functions exported for the inet6 API 437 */ 438 439 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 440 void tcp_v4_mtu_reduced(struct sock *sk); 441 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 442 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 443 struct sock *tcp_create_openreq_child(const struct sock *sk, 444 struct request_sock *req, 445 struct sk_buff *skb); 446 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 447 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 448 struct request_sock *req, 449 struct dst_entry *dst, 450 struct request_sock *req_unhash, 451 bool *own_req); 452 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 453 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 454 int tcp_connect(struct sock *sk); 455 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 456 struct request_sock *req, 457 struct tcp_fastopen_cookie *foc, 458 bool attach_req); 459 int tcp_disconnect(struct sock *sk, int flags); 460 461 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 462 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 463 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 464 465 /* From syncookies.c */ 466 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 467 struct request_sock *req, 468 struct dst_entry *dst); 469 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 470 u32 cookie); 471 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 472 #ifdef CONFIG_SYN_COOKIES 473 474 /* Syncookies use a monotonic timer which increments every 60 seconds. 475 * This counter is used both as a hash input and partially encoded into 476 * the cookie value. A cookie is only validated further if the delta 477 * between the current counter value and the encoded one is less than this, 478 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 479 * the counter advances immediately after a cookie is generated). 480 */ 481 #define MAX_SYNCOOKIE_AGE 2 482 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 483 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 484 485 /* syncookies: remember time of last synqueue overflow 486 * But do not dirty this field too often (once per second is enough) 487 * It is racy as we do not hold a lock, but race is very minor. 488 */ 489 static inline void tcp_synq_overflow(const struct sock *sk) 490 { 491 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 492 unsigned long now = jiffies; 493 494 if (time_after(now, last_overflow + HZ)) 495 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 496 } 497 498 /* syncookies: no recent synqueue overflow on this listening socket? */ 499 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 500 { 501 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 502 503 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID); 504 } 505 506 static inline u32 tcp_cookie_time(void) 507 { 508 u64 val = get_jiffies_64(); 509 510 do_div(val, TCP_SYNCOOKIE_PERIOD); 511 return val; 512 } 513 514 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 515 u16 *mssp); 516 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 517 __u32 cookie_init_timestamp(struct request_sock *req); 518 bool cookie_timestamp_decode(struct tcp_options_received *opt); 519 bool cookie_ecn_ok(const struct tcp_options_received *opt, 520 const struct net *net, const struct dst_entry *dst); 521 522 /* From net/ipv6/syncookies.c */ 523 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 524 u32 cookie); 525 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 526 527 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 528 const struct tcphdr *th, u16 *mssp); 529 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 530 #endif 531 /* tcp_output.c */ 532 533 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 534 int nonagle); 535 bool tcp_may_send_now(struct sock *sk); 536 int __tcp_retransmit_skb(struct sock *, struct sk_buff *); 537 int tcp_retransmit_skb(struct sock *, struct sk_buff *); 538 void tcp_retransmit_timer(struct sock *sk); 539 void tcp_xmit_retransmit_queue(struct sock *); 540 void tcp_simple_retransmit(struct sock *); 541 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 542 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t); 543 544 void tcp_send_probe0(struct sock *); 545 void tcp_send_partial(struct sock *); 546 int tcp_write_wakeup(struct sock *, int mib); 547 void tcp_send_fin(struct sock *sk); 548 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 549 int tcp_send_synack(struct sock *); 550 void tcp_push_one(struct sock *, unsigned int mss_now); 551 void tcp_send_ack(struct sock *sk); 552 void tcp_send_delayed_ack(struct sock *sk); 553 void tcp_send_loss_probe(struct sock *sk); 554 bool tcp_schedule_loss_probe(struct sock *sk); 555 556 /* tcp_input.c */ 557 void tcp_resume_early_retransmit(struct sock *sk); 558 void tcp_rearm_rto(struct sock *sk); 559 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 560 void tcp_reset(struct sock *sk); 561 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 562 void tcp_fin(struct sock *sk); 563 564 /* tcp_timer.c */ 565 void tcp_init_xmit_timers(struct sock *); 566 static inline void tcp_clear_xmit_timers(struct sock *sk) 567 { 568 inet_csk_clear_xmit_timers(sk); 569 } 570 571 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 572 unsigned int tcp_current_mss(struct sock *sk); 573 574 /* Bound MSS / TSO packet size with the half of the window */ 575 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 576 { 577 int cutoff; 578 579 /* When peer uses tiny windows, there is no use in packetizing 580 * to sub-MSS pieces for the sake of SWS or making sure there 581 * are enough packets in the pipe for fast recovery. 582 * 583 * On the other hand, for extremely large MSS devices, handling 584 * smaller than MSS windows in this way does make sense. 585 */ 586 if (tp->max_window >= 512) 587 cutoff = (tp->max_window >> 1); 588 else 589 cutoff = tp->max_window; 590 591 if (cutoff && pktsize > cutoff) 592 return max_t(int, cutoff, 68U - tp->tcp_header_len); 593 else 594 return pktsize; 595 } 596 597 /* tcp.c */ 598 void tcp_get_info(struct sock *, struct tcp_info *); 599 600 /* Read 'sendfile()'-style from a TCP socket */ 601 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 602 unsigned int, size_t); 603 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 604 sk_read_actor_t recv_actor); 605 606 void tcp_initialize_rcv_mss(struct sock *sk); 607 608 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 609 int tcp_mss_to_mtu(struct sock *sk, int mss); 610 void tcp_mtup_init(struct sock *sk); 611 void tcp_init_buffer_space(struct sock *sk); 612 613 static inline void tcp_bound_rto(const struct sock *sk) 614 { 615 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 616 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 617 } 618 619 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 620 { 621 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 622 } 623 624 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 625 { 626 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 627 ntohl(TCP_FLAG_ACK) | 628 snd_wnd); 629 } 630 631 static inline void tcp_fast_path_on(struct tcp_sock *tp) 632 { 633 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 634 } 635 636 static inline void tcp_fast_path_check(struct sock *sk) 637 { 638 struct tcp_sock *tp = tcp_sk(sk); 639 640 if (skb_queue_empty(&tp->out_of_order_queue) && 641 tp->rcv_wnd && 642 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 643 !tp->urg_data) 644 tcp_fast_path_on(tp); 645 } 646 647 /* Compute the actual rto_min value */ 648 static inline u32 tcp_rto_min(struct sock *sk) 649 { 650 const struct dst_entry *dst = __sk_dst_get(sk); 651 u32 rto_min = TCP_RTO_MIN; 652 653 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 654 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 655 return rto_min; 656 } 657 658 static inline u32 tcp_rto_min_us(struct sock *sk) 659 { 660 return jiffies_to_usecs(tcp_rto_min(sk)); 661 } 662 663 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 664 { 665 return dst_metric_locked(dst, RTAX_CC_ALGO); 666 } 667 668 /* Minimum RTT in usec. ~0 means not available. */ 669 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 670 { 671 return tp->rtt_min[0].rtt; 672 } 673 674 /* Compute the actual receive window we are currently advertising. 675 * Rcv_nxt can be after the window if our peer push more data 676 * than the offered window. 677 */ 678 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 679 { 680 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 681 682 if (win < 0) 683 win = 0; 684 return (u32) win; 685 } 686 687 /* Choose a new window, without checks for shrinking, and without 688 * scaling applied to the result. The caller does these things 689 * if necessary. This is a "raw" window selection. 690 */ 691 u32 __tcp_select_window(struct sock *sk); 692 693 void tcp_send_window_probe(struct sock *sk); 694 695 /* TCP timestamps are only 32-bits, this causes a slight 696 * complication on 64-bit systems since we store a snapshot 697 * of jiffies in the buffer control blocks below. We decided 698 * to use only the low 32-bits of jiffies and hide the ugly 699 * casts with the following macro. 700 */ 701 #define tcp_time_stamp ((__u32)(jiffies)) 702 703 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 704 { 705 return skb->skb_mstamp.stamp_jiffies; 706 } 707 708 709 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 710 711 #define TCPHDR_FIN 0x01 712 #define TCPHDR_SYN 0x02 713 #define TCPHDR_RST 0x04 714 #define TCPHDR_PSH 0x08 715 #define TCPHDR_ACK 0x10 716 #define TCPHDR_URG 0x20 717 #define TCPHDR_ECE 0x40 718 #define TCPHDR_CWR 0x80 719 720 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 721 722 /* This is what the send packet queuing engine uses to pass 723 * TCP per-packet control information to the transmission code. 724 * We also store the host-order sequence numbers in here too. 725 * This is 44 bytes if IPV6 is enabled. 726 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 727 */ 728 struct tcp_skb_cb { 729 __u32 seq; /* Starting sequence number */ 730 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 731 union { 732 /* Note : tcp_tw_isn is used in input path only 733 * (isn chosen by tcp_timewait_state_process()) 734 * 735 * tcp_gso_segs/size are used in write queue only, 736 * cf tcp_skb_pcount()/tcp_skb_mss() 737 */ 738 __u32 tcp_tw_isn; 739 struct { 740 u16 tcp_gso_segs; 741 u16 tcp_gso_size; 742 }; 743 }; 744 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 745 746 __u8 sacked; /* State flags for SACK/FACK. */ 747 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 748 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 749 #define TCPCB_LOST 0x04 /* SKB is lost */ 750 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 751 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 752 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 753 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 754 TCPCB_REPAIRED) 755 756 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 757 /* 1 byte hole */ 758 __u32 ack_seq; /* Sequence number ACK'd */ 759 union { 760 struct inet_skb_parm h4; 761 #if IS_ENABLED(CONFIG_IPV6) 762 struct inet6_skb_parm h6; 763 #endif 764 } header; /* For incoming frames */ 765 }; 766 767 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 768 769 770 #if IS_ENABLED(CONFIG_IPV6) 771 /* This is the variant of inet6_iif() that must be used by TCP, 772 * as TCP moves IP6CB into a different location in skb->cb[] 773 */ 774 static inline int tcp_v6_iif(const struct sk_buff *skb) 775 { 776 return TCP_SKB_CB(skb)->header.h6.iif; 777 } 778 #endif 779 780 /* Due to TSO, an SKB can be composed of multiple actual 781 * packets. To keep these tracked properly, we use this. 782 */ 783 static inline int tcp_skb_pcount(const struct sk_buff *skb) 784 { 785 return TCP_SKB_CB(skb)->tcp_gso_segs; 786 } 787 788 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 789 { 790 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 791 } 792 793 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 794 { 795 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 796 } 797 798 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 799 static inline int tcp_skb_mss(const struct sk_buff *skb) 800 { 801 return TCP_SKB_CB(skb)->tcp_gso_size; 802 } 803 804 /* Events passed to congestion control interface */ 805 enum tcp_ca_event { 806 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 807 CA_EVENT_CWND_RESTART, /* congestion window restart */ 808 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 809 CA_EVENT_LOSS, /* loss timeout */ 810 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 811 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 812 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 813 CA_EVENT_NON_DELAYED_ACK, 814 }; 815 816 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 817 enum tcp_ca_ack_event_flags { 818 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 819 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 820 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 821 }; 822 823 /* 824 * Interface for adding new TCP congestion control handlers 825 */ 826 #define TCP_CA_NAME_MAX 16 827 #define TCP_CA_MAX 128 828 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 829 830 #define TCP_CA_UNSPEC 0 831 832 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 833 #define TCP_CONG_NON_RESTRICTED 0x1 834 /* Requires ECN/ECT set on all packets */ 835 #define TCP_CONG_NEEDS_ECN 0x2 836 837 union tcp_cc_info; 838 839 struct tcp_congestion_ops { 840 struct list_head list; 841 u32 key; 842 u32 flags; 843 844 /* initialize private data (optional) */ 845 void (*init)(struct sock *sk); 846 /* cleanup private data (optional) */ 847 void (*release)(struct sock *sk); 848 849 /* return slow start threshold (required) */ 850 u32 (*ssthresh)(struct sock *sk); 851 /* do new cwnd calculation (required) */ 852 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 853 /* call before changing ca_state (optional) */ 854 void (*set_state)(struct sock *sk, u8 new_state); 855 /* call when cwnd event occurs (optional) */ 856 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 857 /* call when ack arrives (optional) */ 858 void (*in_ack_event)(struct sock *sk, u32 flags); 859 /* new value of cwnd after loss (optional) */ 860 u32 (*undo_cwnd)(struct sock *sk); 861 /* hook for packet ack accounting (optional) */ 862 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 863 /* get info for inet_diag (optional) */ 864 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 865 union tcp_cc_info *info); 866 867 char name[TCP_CA_NAME_MAX]; 868 struct module *owner; 869 }; 870 871 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 872 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 873 874 void tcp_assign_congestion_control(struct sock *sk); 875 void tcp_init_congestion_control(struct sock *sk); 876 void tcp_cleanup_congestion_control(struct sock *sk); 877 int tcp_set_default_congestion_control(const char *name); 878 void tcp_get_default_congestion_control(char *name); 879 void tcp_get_available_congestion_control(char *buf, size_t len); 880 void tcp_get_allowed_congestion_control(char *buf, size_t len); 881 int tcp_set_allowed_congestion_control(char *allowed); 882 int tcp_set_congestion_control(struct sock *sk, const char *name); 883 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 884 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 885 886 u32 tcp_reno_ssthresh(struct sock *sk); 887 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 888 extern struct tcp_congestion_ops tcp_reno; 889 890 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 891 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca); 892 #ifdef CONFIG_INET 893 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 894 #else 895 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 896 { 897 return NULL; 898 } 899 #endif 900 901 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 902 { 903 const struct inet_connection_sock *icsk = inet_csk(sk); 904 905 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 906 } 907 908 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 909 { 910 struct inet_connection_sock *icsk = inet_csk(sk); 911 912 if (icsk->icsk_ca_ops->set_state) 913 icsk->icsk_ca_ops->set_state(sk, ca_state); 914 icsk->icsk_ca_state = ca_state; 915 } 916 917 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 918 { 919 const struct inet_connection_sock *icsk = inet_csk(sk); 920 921 if (icsk->icsk_ca_ops->cwnd_event) 922 icsk->icsk_ca_ops->cwnd_event(sk, event); 923 } 924 925 /* These functions determine how the current flow behaves in respect of SACK 926 * handling. SACK is negotiated with the peer, and therefore it can vary 927 * between different flows. 928 * 929 * tcp_is_sack - SACK enabled 930 * tcp_is_reno - No SACK 931 * tcp_is_fack - FACK enabled, implies SACK enabled 932 */ 933 static inline int tcp_is_sack(const struct tcp_sock *tp) 934 { 935 return tp->rx_opt.sack_ok; 936 } 937 938 static inline bool tcp_is_reno(const struct tcp_sock *tp) 939 { 940 return !tcp_is_sack(tp); 941 } 942 943 static inline bool tcp_is_fack(const struct tcp_sock *tp) 944 { 945 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 946 } 947 948 static inline void tcp_enable_fack(struct tcp_sock *tp) 949 { 950 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 951 } 952 953 /* TCP early-retransmit (ER) is similar to but more conservative than 954 * the thin-dupack feature. Enable ER only if thin-dupack is disabled. 955 */ 956 static inline void tcp_enable_early_retrans(struct tcp_sock *tp) 957 { 958 struct net *net = sock_net((struct sock *)tp); 959 960 tp->do_early_retrans = sysctl_tcp_early_retrans && 961 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack && 962 net->ipv4.sysctl_tcp_reordering == 3; 963 } 964 965 static inline void tcp_disable_early_retrans(struct tcp_sock *tp) 966 { 967 tp->do_early_retrans = 0; 968 } 969 970 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 971 { 972 return tp->sacked_out + tp->lost_out; 973 } 974 975 /* This determines how many packets are "in the network" to the best 976 * of our knowledge. In many cases it is conservative, but where 977 * detailed information is available from the receiver (via SACK 978 * blocks etc.) we can make more aggressive calculations. 979 * 980 * Use this for decisions involving congestion control, use just 981 * tp->packets_out to determine if the send queue is empty or not. 982 * 983 * Read this equation as: 984 * 985 * "Packets sent once on transmission queue" MINUS 986 * "Packets left network, but not honestly ACKed yet" PLUS 987 * "Packets fast retransmitted" 988 */ 989 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 990 { 991 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 992 } 993 994 #define TCP_INFINITE_SSTHRESH 0x7fffffff 995 996 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 997 { 998 return tp->snd_cwnd < tp->snd_ssthresh; 999 } 1000 1001 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1002 { 1003 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1004 } 1005 1006 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1007 { 1008 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1009 (1 << inet_csk(sk)->icsk_ca_state); 1010 } 1011 1012 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1013 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1014 * ssthresh. 1015 */ 1016 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1017 { 1018 const struct tcp_sock *tp = tcp_sk(sk); 1019 1020 if (tcp_in_cwnd_reduction(sk)) 1021 return tp->snd_ssthresh; 1022 else 1023 return max(tp->snd_ssthresh, 1024 ((tp->snd_cwnd >> 1) + 1025 (tp->snd_cwnd >> 2))); 1026 } 1027 1028 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1029 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1030 1031 void tcp_enter_cwr(struct sock *sk); 1032 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1033 1034 /* The maximum number of MSS of available cwnd for which TSO defers 1035 * sending if not using sysctl_tcp_tso_win_divisor. 1036 */ 1037 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1038 { 1039 return 3; 1040 } 1041 1042 /* Slow start with delack produces 3 packets of burst, so that 1043 * it is safe "de facto". This will be the default - same as 1044 * the default reordering threshold - but if reordering increases, 1045 * we must be able to allow cwnd to burst at least this much in order 1046 * to not pull it back when holes are filled. 1047 */ 1048 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 1049 { 1050 return tp->reordering; 1051 } 1052 1053 /* Returns end sequence number of the receiver's advertised window */ 1054 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1055 { 1056 return tp->snd_una + tp->snd_wnd; 1057 } 1058 1059 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1060 * flexible approach. The RFC suggests cwnd should not be raised unless 1061 * it was fully used previously. And that's exactly what we do in 1062 * congestion avoidance mode. But in slow start we allow cwnd to grow 1063 * as long as the application has used half the cwnd. 1064 * Example : 1065 * cwnd is 10 (IW10), but application sends 9 frames. 1066 * We allow cwnd to reach 18 when all frames are ACKed. 1067 * This check is safe because it's as aggressive as slow start which already 1068 * risks 100% overshoot. The advantage is that we discourage application to 1069 * either send more filler packets or data to artificially blow up the cwnd 1070 * usage, and allow application-limited process to probe bw more aggressively. 1071 */ 1072 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1073 { 1074 const struct tcp_sock *tp = tcp_sk(sk); 1075 1076 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1077 if (tcp_in_slow_start(tp)) 1078 return tp->snd_cwnd < 2 * tp->max_packets_out; 1079 1080 return tp->is_cwnd_limited; 1081 } 1082 1083 /* Something is really bad, we could not queue an additional packet, 1084 * because qdisc is full or receiver sent a 0 window. 1085 * We do not want to add fuel to the fire, or abort too early, 1086 * so make sure the timer we arm now is at least 200ms in the future, 1087 * regardless of current icsk_rto value (as it could be ~2ms) 1088 */ 1089 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1090 { 1091 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1092 } 1093 1094 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1095 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1096 unsigned long max_when) 1097 { 1098 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1099 1100 return (unsigned long)min_t(u64, when, max_when); 1101 } 1102 1103 static inline void tcp_check_probe_timer(struct sock *sk) 1104 { 1105 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1106 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1107 tcp_probe0_base(sk), TCP_RTO_MAX); 1108 } 1109 1110 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1111 { 1112 tp->snd_wl1 = seq; 1113 } 1114 1115 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1116 { 1117 tp->snd_wl1 = seq; 1118 } 1119 1120 /* 1121 * Calculate(/check) TCP checksum 1122 */ 1123 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1124 __be32 daddr, __wsum base) 1125 { 1126 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1127 } 1128 1129 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1130 { 1131 return __skb_checksum_complete(skb); 1132 } 1133 1134 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1135 { 1136 return !skb_csum_unnecessary(skb) && 1137 __tcp_checksum_complete(skb); 1138 } 1139 1140 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1141 1142 static inline void tcp_prequeue_init(struct tcp_sock *tp) 1143 { 1144 tp->ucopy.task = NULL; 1145 tp->ucopy.len = 0; 1146 tp->ucopy.memory = 0; 1147 skb_queue_head_init(&tp->ucopy.prequeue); 1148 } 1149 1150 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1151 1152 #undef STATE_TRACE 1153 1154 #ifdef STATE_TRACE 1155 static const char *statename[]={ 1156 "Unused","Established","Syn Sent","Syn Recv", 1157 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1158 "Close Wait","Last ACK","Listen","Closing" 1159 }; 1160 #endif 1161 void tcp_set_state(struct sock *sk, int state); 1162 1163 void tcp_done(struct sock *sk); 1164 1165 int tcp_abort(struct sock *sk, int err); 1166 1167 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1168 { 1169 rx_opt->dsack = 0; 1170 rx_opt->num_sacks = 0; 1171 } 1172 1173 u32 tcp_default_init_rwnd(u32 mss); 1174 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1175 1176 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1177 { 1178 struct tcp_sock *tp = tcp_sk(sk); 1179 s32 delta; 1180 1181 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out) 1182 return; 1183 delta = tcp_time_stamp - tp->lsndtime; 1184 if (delta > inet_csk(sk)->icsk_rto) 1185 tcp_cwnd_restart(sk, delta); 1186 } 1187 1188 /* Determine a window scaling and initial window to offer. */ 1189 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1190 __u32 *window_clamp, int wscale_ok, 1191 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1192 1193 static inline int tcp_win_from_space(int space) 1194 { 1195 return sysctl_tcp_adv_win_scale<=0 ? 1196 (space>>(-sysctl_tcp_adv_win_scale)) : 1197 space - (space>>sysctl_tcp_adv_win_scale); 1198 } 1199 1200 /* Note: caller must be prepared to deal with negative returns */ 1201 static inline int tcp_space(const struct sock *sk) 1202 { 1203 return tcp_win_from_space(sk->sk_rcvbuf - 1204 atomic_read(&sk->sk_rmem_alloc)); 1205 } 1206 1207 static inline int tcp_full_space(const struct sock *sk) 1208 { 1209 return tcp_win_from_space(sk->sk_rcvbuf); 1210 } 1211 1212 extern void tcp_openreq_init_rwin(struct request_sock *req, 1213 const struct sock *sk_listener, 1214 const struct dst_entry *dst); 1215 1216 void tcp_enter_memory_pressure(struct sock *sk); 1217 1218 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1219 { 1220 struct net *net = sock_net((struct sock *)tp); 1221 1222 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1223 } 1224 1225 static inline int keepalive_time_when(const struct tcp_sock *tp) 1226 { 1227 struct net *net = sock_net((struct sock *)tp); 1228 1229 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1230 } 1231 1232 static inline int keepalive_probes(const struct tcp_sock *tp) 1233 { 1234 struct net *net = sock_net((struct sock *)tp); 1235 1236 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1237 } 1238 1239 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1240 { 1241 const struct inet_connection_sock *icsk = &tp->inet_conn; 1242 1243 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1244 tcp_time_stamp - tp->rcv_tstamp); 1245 } 1246 1247 static inline int tcp_fin_time(const struct sock *sk) 1248 { 1249 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1250 const int rto = inet_csk(sk)->icsk_rto; 1251 1252 if (fin_timeout < (rto << 2) - (rto >> 1)) 1253 fin_timeout = (rto << 2) - (rto >> 1); 1254 1255 return fin_timeout; 1256 } 1257 1258 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1259 int paws_win) 1260 { 1261 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1262 return true; 1263 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1264 return true; 1265 /* 1266 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1267 * then following tcp messages have valid values. Ignore 0 value, 1268 * or else 'negative' tsval might forbid us to accept their packets. 1269 */ 1270 if (!rx_opt->ts_recent) 1271 return true; 1272 return false; 1273 } 1274 1275 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1276 int rst) 1277 { 1278 if (tcp_paws_check(rx_opt, 0)) 1279 return false; 1280 1281 /* RST segments are not recommended to carry timestamp, 1282 and, if they do, it is recommended to ignore PAWS because 1283 "their cleanup function should take precedence over timestamps." 1284 Certainly, it is mistake. It is necessary to understand the reasons 1285 of this constraint to relax it: if peer reboots, clock may go 1286 out-of-sync and half-open connections will not be reset. 1287 Actually, the problem would be not existing if all 1288 the implementations followed draft about maintaining clock 1289 via reboots. Linux-2.2 DOES NOT! 1290 1291 However, we can relax time bounds for RST segments to MSL. 1292 */ 1293 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1294 return false; 1295 return true; 1296 } 1297 1298 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1299 int mib_idx, u32 *last_oow_ack_time); 1300 1301 static inline void tcp_mib_init(struct net *net) 1302 { 1303 /* See RFC 2012 */ 1304 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1305 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1306 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1307 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1308 } 1309 1310 /* from STCP */ 1311 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1312 { 1313 tp->lost_skb_hint = NULL; 1314 } 1315 1316 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1317 { 1318 tcp_clear_retrans_hints_partial(tp); 1319 tp->retransmit_skb_hint = NULL; 1320 } 1321 1322 union tcp_md5_addr { 1323 struct in_addr a4; 1324 #if IS_ENABLED(CONFIG_IPV6) 1325 struct in6_addr a6; 1326 #endif 1327 }; 1328 1329 /* - key database */ 1330 struct tcp_md5sig_key { 1331 struct hlist_node node; 1332 u8 keylen; 1333 u8 family; /* AF_INET or AF_INET6 */ 1334 union tcp_md5_addr addr; 1335 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1336 struct rcu_head rcu; 1337 }; 1338 1339 /* - sock block */ 1340 struct tcp_md5sig_info { 1341 struct hlist_head head; 1342 struct rcu_head rcu; 1343 }; 1344 1345 /* - pseudo header */ 1346 struct tcp4_pseudohdr { 1347 __be32 saddr; 1348 __be32 daddr; 1349 __u8 pad; 1350 __u8 protocol; 1351 __be16 len; 1352 }; 1353 1354 struct tcp6_pseudohdr { 1355 struct in6_addr saddr; 1356 struct in6_addr daddr; 1357 __be32 len; 1358 __be32 protocol; /* including padding */ 1359 }; 1360 1361 union tcp_md5sum_block { 1362 struct tcp4_pseudohdr ip4; 1363 #if IS_ENABLED(CONFIG_IPV6) 1364 struct tcp6_pseudohdr ip6; 1365 #endif 1366 }; 1367 1368 /* - pool: digest algorithm, hash description and scratch buffer */ 1369 struct tcp_md5sig_pool { 1370 struct ahash_request *md5_req; 1371 union tcp_md5sum_block md5_blk; 1372 }; 1373 1374 /* - functions */ 1375 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1376 const struct sock *sk, const struct sk_buff *skb); 1377 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1378 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); 1379 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1380 int family); 1381 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1382 const struct sock *addr_sk); 1383 1384 #ifdef CONFIG_TCP_MD5SIG 1385 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1386 const union tcp_md5_addr *addr, 1387 int family); 1388 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1389 #else 1390 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1391 const union tcp_md5_addr *addr, 1392 int family) 1393 { 1394 return NULL; 1395 } 1396 #define tcp_twsk_md5_key(twsk) NULL 1397 #endif 1398 1399 bool tcp_alloc_md5sig_pool(void); 1400 1401 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1402 static inline void tcp_put_md5sig_pool(void) 1403 { 1404 local_bh_enable(); 1405 } 1406 1407 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1408 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1409 unsigned int header_len); 1410 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1411 const struct tcp_md5sig_key *key); 1412 1413 /* From tcp_fastopen.c */ 1414 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1415 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1416 unsigned long *last_syn_loss); 1417 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1418 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1419 u16 try_exp); 1420 struct tcp_fastopen_request { 1421 /* Fast Open cookie. Size 0 means a cookie request */ 1422 struct tcp_fastopen_cookie cookie; 1423 struct msghdr *data; /* data in MSG_FASTOPEN */ 1424 size_t size; 1425 int copied; /* queued in tcp_connect() */ 1426 }; 1427 void tcp_free_fastopen_req(struct tcp_sock *tp); 1428 1429 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1430 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1431 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1432 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1433 struct request_sock *req, 1434 struct tcp_fastopen_cookie *foc, 1435 struct dst_entry *dst); 1436 void tcp_fastopen_init_key_once(bool publish); 1437 #define TCP_FASTOPEN_KEY_LENGTH 16 1438 1439 /* Fastopen key context */ 1440 struct tcp_fastopen_context { 1441 struct crypto_cipher *tfm; 1442 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1443 struct rcu_head rcu; 1444 }; 1445 1446 /* write queue abstraction */ 1447 static inline void tcp_write_queue_purge(struct sock *sk) 1448 { 1449 struct sk_buff *skb; 1450 1451 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1452 sk_wmem_free_skb(sk, skb); 1453 sk_mem_reclaim(sk); 1454 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1455 } 1456 1457 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1458 { 1459 return skb_peek(&sk->sk_write_queue); 1460 } 1461 1462 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1463 { 1464 return skb_peek_tail(&sk->sk_write_queue); 1465 } 1466 1467 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1468 const struct sk_buff *skb) 1469 { 1470 return skb_queue_next(&sk->sk_write_queue, skb); 1471 } 1472 1473 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1474 const struct sk_buff *skb) 1475 { 1476 return skb_queue_prev(&sk->sk_write_queue, skb); 1477 } 1478 1479 #define tcp_for_write_queue(skb, sk) \ 1480 skb_queue_walk(&(sk)->sk_write_queue, skb) 1481 1482 #define tcp_for_write_queue_from(skb, sk) \ 1483 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1484 1485 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1486 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1487 1488 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1489 { 1490 return sk->sk_send_head; 1491 } 1492 1493 static inline bool tcp_skb_is_last(const struct sock *sk, 1494 const struct sk_buff *skb) 1495 { 1496 return skb_queue_is_last(&sk->sk_write_queue, skb); 1497 } 1498 1499 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1500 { 1501 if (tcp_skb_is_last(sk, skb)) 1502 sk->sk_send_head = NULL; 1503 else 1504 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1505 } 1506 1507 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1508 { 1509 if (sk->sk_send_head == skb_unlinked) 1510 sk->sk_send_head = NULL; 1511 } 1512 1513 static inline void tcp_init_send_head(struct sock *sk) 1514 { 1515 sk->sk_send_head = NULL; 1516 } 1517 1518 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1519 { 1520 __skb_queue_tail(&sk->sk_write_queue, skb); 1521 } 1522 1523 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1524 { 1525 __tcp_add_write_queue_tail(sk, skb); 1526 1527 /* Queue it, remembering where we must start sending. */ 1528 if (sk->sk_send_head == NULL) { 1529 sk->sk_send_head = skb; 1530 1531 if (tcp_sk(sk)->highest_sack == NULL) 1532 tcp_sk(sk)->highest_sack = skb; 1533 } 1534 } 1535 1536 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1537 { 1538 __skb_queue_head(&sk->sk_write_queue, skb); 1539 } 1540 1541 /* Insert buff after skb on the write queue of sk. */ 1542 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1543 struct sk_buff *buff, 1544 struct sock *sk) 1545 { 1546 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1547 } 1548 1549 /* Insert new before skb on the write queue of sk. */ 1550 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1551 struct sk_buff *skb, 1552 struct sock *sk) 1553 { 1554 __skb_queue_before(&sk->sk_write_queue, skb, new); 1555 1556 if (sk->sk_send_head == skb) 1557 sk->sk_send_head = new; 1558 } 1559 1560 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1561 { 1562 __skb_unlink(skb, &sk->sk_write_queue); 1563 } 1564 1565 static inline bool tcp_write_queue_empty(struct sock *sk) 1566 { 1567 return skb_queue_empty(&sk->sk_write_queue); 1568 } 1569 1570 static inline void tcp_push_pending_frames(struct sock *sk) 1571 { 1572 if (tcp_send_head(sk)) { 1573 struct tcp_sock *tp = tcp_sk(sk); 1574 1575 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1576 } 1577 } 1578 1579 /* Start sequence of the skb just after the highest skb with SACKed 1580 * bit, valid only if sacked_out > 0 or when the caller has ensured 1581 * validity by itself. 1582 */ 1583 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1584 { 1585 if (!tp->sacked_out) 1586 return tp->snd_una; 1587 1588 if (tp->highest_sack == NULL) 1589 return tp->snd_nxt; 1590 1591 return TCP_SKB_CB(tp->highest_sack)->seq; 1592 } 1593 1594 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1595 { 1596 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1597 tcp_write_queue_next(sk, skb); 1598 } 1599 1600 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1601 { 1602 return tcp_sk(sk)->highest_sack; 1603 } 1604 1605 static inline void tcp_highest_sack_reset(struct sock *sk) 1606 { 1607 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1608 } 1609 1610 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1611 static inline void tcp_highest_sack_combine(struct sock *sk, 1612 struct sk_buff *old, 1613 struct sk_buff *new) 1614 { 1615 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1616 tcp_sk(sk)->highest_sack = new; 1617 } 1618 1619 /* This helper checks if socket has IP_TRANSPARENT set */ 1620 static inline bool inet_sk_transparent(const struct sock *sk) 1621 { 1622 switch (sk->sk_state) { 1623 case TCP_TIME_WAIT: 1624 return inet_twsk(sk)->tw_transparent; 1625 case TCP_NEW_SYN_RECV: 1626 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1627 } 1628 return inet_sk(sk)->transparent; 1629 } 1630 1631 /* Determines whether this is a thin stream (which may suffer from 1632 * increased latency). Used to trigger latency-reducing mechanisms. 1633 */ 1634 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1635 { 1636 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1637 } 1638 1639 /* /proc */ 1640 enum tcp_seq_states { 1641 TCP_SEQ_STATE_LISTENING, 1642 TCP_SEQ_STATE_ESTABLISHED, 1643 }; 1644 1645 int tcp_seq_open(struct inode *inode, struct file *file); 1646 1647 struct tcp_seq_afinfo { 1648 char *name; 1649 sa_family_t family; 1650 const struct file_operations *seq_fops; 1651 struct seq_operations seq_ops; 1652 }; 1653 1654 struct tcp_iter_state { 1655 struct seq_net_private p; 1656 sa_family_t family; 1657 enum tcp_seq_states state; 1658 struct sock *syn_wait_sk; 1659 int bucket, offset, sbucket, num; 1660 loff_t last_pos; 1661 }; 1662 1663 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1664 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1665 1666 extern struct request_sock_ops tcp_request_sock_ops; 1667 extern struct request_sock_ops tcp6_request_sock_ops; 1668 1669 void tcp_v4_destroy_sock(struct sock *sk); 1670 1671 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1672 netdev_features_t features); 1673 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1674 int tcp_gro_complete(struct sk_buff *skb); 1675 1676 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1677 1678 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1679 { 1680 struct net *net = sock_net((struct sock *)tp); 1681 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1682 } 1683 1684 static inline bool tcp_stream_memory_free(const struct sock *sk) 1685 { 1686 const struct tcp_sock *tp = tcp_sk(sk); 1687 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1688 1689 return notsent_bytes < tcp_notsent_lowat(tp); 1690 } 1691 1692 #ifdef CONFIG_PROC_FS 1693 int tcp4_proc_init(void); 1694 void tcp4_proc_exit(void); 1695 #endif 1696 1697 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1698 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1699 const struct tcp_request_sock_ops *af_ops, 1700 struct sock *sk, struct sk_buff *skb); 1701 1702 /* TCP af-specific functions */ 1703 struct tcp_sock_af_ops { 1704 #ifdef CONFIG_TCP_MD5SIG 1705 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1706 const struct sock *addr_sk); 1707 int (*calc_md5_hash)(char *location, 1708 const struct tcp_md5sig_key *md5, 1709 const struct sock *sk, 1710 const struct sk_buff *skb); 1711 int (*md5_parse)(struct sock *sk, 1712 char __user *optval, 1713 int optlen); 1714 #endif 1715 }; 1716 1717 struct tcp_request_sock_ops { 1718 u16 mss_clamp; 1719 #ifdef CONFIG_TCP_MD5SIG 1720 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1721 const struct sock *addr_sk); 1722 int (*calc_md5_hash) (char *location, 1723 const struct tcp_md5sig_key *md5, 1724 const struct sock *sk, 1725 const struct sk_buff *skb); 1726 #endif 1727 void (*init_req)(struct request_sock *req, 1728 const struct sock *sk_listener, 1729 struct sk_buff *skb); 1730 #ifdef CONFIG_SYN_COOKIES 1731 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1732 __u16 *mss); 1733 #endif 1734 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1735 const struct request_sock *req, 1736 bool *strict); 1737 __u32 (*init_seq)(const struct sk_buff *skb); 1738 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1739 struct flowi *fl, struct request_sock *req, 1740 struct tcp_fastopen_cookie *foc, 1741 bool attach_req); 1742 }; 1743 1744 #ifdef CONFIG_SYN_COOKIES 1745 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1746 const struct sock *sk, struct sk_buff *skb, 1747 __u16 *mss) 1748 { 1749 tcp_synq_overflow(sk); 1750 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1751 return ops->cookie_init_seq(skb, mss); 1752 } 1753 #else 1754 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1755 const struct sock *sk, struct sk_buff *skb, 1756 __u16 *mss) 1757 { 1758 return 0; 1759 } 1760 #endif 1761 1762 int tcpv4_offload_init(void); 1763 1764 void tcp_v4_init(void); 1765 void tcp_init(void); 1766 1767 /* tcp_recovery.c */ 1768 1769 /* Flags to enable various loss recovery features. See below */ 1770 extern int sysctl_tcp_recovery; 1771 1772 /* Use TCP RACK to detect (some) tail and retransmit losses */ 1773 #define TCP_RACK_LOST_RETRANS 0x1 1774 1775 extern int tcp_rack_mark_lost(struct sock *sk); 1776 1777 extern void tcp_rack_advance(struct tcp_sock *tp, 1778 const struct skb_mstamp *xmit_time, u8 sacked); 1779 1780 /* 1781 * Save and compile IPv4 options, return a pointer to it 1782 */ 1783 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb) 1784 { 1785 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1786 struct ip_options_rcu *dopt = NULL; 1787 1788 if (opt->optlen) { 1789 int opt_size = sizeof(*dopt) + opt->optlen; 1790 1791 dopt = kmalloc(opt_size, GFP_ATOMIC); 1792 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) { 1793 kfree(dopt); 1794 dopt = NULL; 1795 } 1796 } 1797 return dopt; 1798 } 1799 1800 /* locally generated TCP pure ACKs have skb->truesize == 2 1801 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1802 * This is much faster than dissecting the packet to find out. 1803 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1804 */ 1805 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1806 { 1807 return skb->truesize == 2; 1808 } 1809 1810 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1811 { 1812 skb->truesize = 2; 1813 } 1814 1815 static inline int tcp_inq(struct sock *sk) 1816 { 1817 struct tcp_sock *tp = tcp_sk(sk); 1818 int answ; 1819 1820 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1821 answ = 0; 1822 } else if (sock_flag(sk, SOCK_URGINLINE) || 1823 !tp->urg_data || 1824 before(tp->urg_seq, tp->copied_seq) || 1825 !before(tp->urg_seq, tp->rcv_nxt)) { 1826 1827 answ = tp->rcv_nxt - tp->copied_seq; 1828 1829 /* Subtract 1, if FIN was received */ 1830 if (answ && sock_flag(sk, SOCK_DONE)) 1831 answ--; 1832 } else { 1833 answ = tp->urg_seq - tp->copied_seq; 1834 } 1835 1836 return answ; 1837 } 1838 1839 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 1840 { 1841 u16 segs_in; 1842 1843 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 1844 tp->segs_in += segs_in; 1845 if (skb->len > tcp_hdrlen(skb)) 1846 tp->data_segs_in += segs_in; 1847 } 1848 1849 #endif /* _TCP_H */ 1850