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/crypto.h> 31 #include <linux/cryptohash.h> 32 #include <linux/kref.h> 33 #include <linux/ktime.h> 34 35 #include <net/inet_connection_sock.h> 36 #include <net/inet_timewait_sock.h> 37 #include <net/inet_hashtables.h> 38 #include <net/checksum.h> 39 #include <net/request_sock.h> 40 #include <net/sock.h> 41 #include <net/snmp.h> 42 #include <net/ip.h> 43 #include <net/tcp_states.h> 44 #include <net/inet_ecn.h> 45 #include <net/dst.h> 46 47 #include <linux/seq_file.h> 48 #include <linux/memcontrol.h> 49 50 extern struct inet_hashinfo tcp_hashinfo; 51 52 extern struct percpu_counter tcp_orphan_count; 53 void tcp_time_wait(struct sock *sk, int state, int timeo); 54 55 #define MAX_TCP_HEADER (128 + MAX_HEADER) 56 #define MAX_TCP_OPTION_SPACE 40 57 58 /* 59 * Never offer a window over 32767 without using window scaling. Some 60 * poor stacks do signed 16bit maths! 61 */ 62 #define MAX_TCP_WINDOW 32767U 63 64 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 65 #define TCP_MIN_MSS 88U 66 67 /* The least MTU to use for probing */ 68 #define TCP_BASE_MSS 1024 69 70 /* probing interval, default to 10 minutes as per RFC4821 */ 71 #define TCP_PROBE_INTERVAL 600 72 73 /* Specify interval when tcp mtu probing will stop */ 74 #define TCP_PROBE_THRESHOLD 8 75 76 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 77 #define TCP_FASTRETRANS_THRESH 3 78 79 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 80 #define TCP_MAX_QUICKACKS 16U 81 82 /* urg_data states */ 83 #define TCP_URG_VALID 0x0100 84 #define TCP_URG_NOTYET 0x0200 85 #define TCP_URG_READ 0x0400 86 87 #define TCP_RETR1 3 /* 88 * This is how many retries it does before it 89 * tries to figure out if the gateway is 90 * down. Minimal RFC value is 3; it corresponds 91 * to ~3sec-8min depending on RTO. 92 */ 93 94 #define TCP_RETR2 15 /* 95 * This should take at least 96 * 90 minutes to time out. 97 * RFC1122 says that the limit is 100 sec. 98 * 15 is ~13-30min depending on RTO. 99 */ 100 101 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 102 * when active opening a connection. 103 * RFC1122 says the minimum retry MUST 104 * be at least 180secs. Nevertheless 105 * this value is corresponding to 106 * 63secs of retransmission with the 107 * current initial RTO. 108 */ 109 110 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 111 * when passive opening a connection. 112 * This is corresponding to 31secs of 113 * retransmission with the current 114 * initial RTO. 115 */ 116 117 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 118 * state, about 60 seconds */ 119 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 120 /* BSD style FIN_WAIT2 deadlock breaker. 121 * It used to be 3min, new value is 60sec, 122 * to combine FIN-WAIT-2 timeout with 123 * TIME-WAIT timer. 124 */ 125 126 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 127 #if HZ >= 100 128 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 129 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 130 #else 131 #define TCP_DELACK_MIN 4U 132 #define TCP_ATO_MIN 4U 133 #endif 134 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 135 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 136 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 137 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 138 * used as a fallback RTO for the 139 * initial data transmission if no 140 * valid RTT sample has been acquired, 141 * most likely due to retrans in 3WHS. 142 */ 143 144 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 145 * for local resources. 146 */ 147 148 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 149 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 150 #define TCP_KEEPALIVE_INTVL (75*HZ) 151 152 #define MAX_TCP_KEEPIDLE 32767 153 #define MAX_TCP_KEEPINTVL 32767 154 #define MAX_TCP_KEEPCNT 127 155 #define MAX_TCP_SYNCNT 127 156 157 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 158 159 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 160 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 161 * after this time. It should be equal 162 * (or greater than) TCP_TIMEWAIT_LEN 163 * to provide reliability equal to one 164 * provided by timewait state. 165 */ 166 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 167 * timestamps. It must be less than 168 * minimal timewait lifetime. 169 */ 170 /* 171 * TCP option 172 */ 173 174 #define TCPOPT_NOP 1 /* Padding */ 175 #define TCPOPT_EOL 0 /* End of options */ 176 #define TCPOPT_MSS 2 /* Segment size negotiating */ 177 #define TCPOPT_WINDOW 3 /* Window scaling */ 178 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 179 #define TCPOPT_SACK 5 /* SACK Block */ 180 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 181 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 182 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 183 #define TCPOPT_EXP 254 /* Experimental */ 184 /* Magic number to be after the option value for sharing TCP 185 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 186 */ 187 #define TCPOPT_FASTOPEN_MAGIC 0xF989 188 189 /* 190 * TCP option lengths 191 */ 192 193 #define TCPOLEN_MSS 4 194 #define TCPOLEN_WINDOW 3 195 #define TCPOLEN_SACK_PERM 2 196 #define TCPOLEN_TIMESTAMP 10 197 #define TCPOLEN_MD5SIG 18 198 #define TCPOLEN_FASTOPEN_BASE 2 199 #define TCPOLEN_EXP_FASTOPEN_BASE 4 200 201 /* But this is what stacks really send out. */ 202 #define TCPOLEN_TSTAMP_ALIGNED 12 203 #define TCPOLEN_WSCALE_ALIGNED 4 204 #define TCPOLEN_SACKPERM_ALIGNED 4 205 #define TCPOLEN_SACK_BASE 2 206 #define TCPOLEN_SACK_BASE_ALIGNED 4 207 #define TCPOLEN_SACK_PERBLOCK 8 208 #define TCPOLEN_MD5SIG_ALIGNED 20 209 #define TCPOLEN_MSS_ALIGNED 4 210 211 /* Flags in tp->nonagle */ 212 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 213 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 214 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 215 216 /* TCP thin-stream limits */ 217 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 218 219 /* TCP initial congestion window as per rfc6928 */ 220 #define TCP_INIT_CWND 10 221 222 /* Bit Flags for sysctl_tcp_fastopen */ 223 #define TFO_CLIENT_ENABLE 1 224 #define TFO_SERVER_ENABLE 2 225 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 226 227 /* Accept SYN data w/o any cookie option */ 228 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 229 230 /* Force enable TFO on all listeners, i.e., not requiring the 231 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen. 232 */ 233 #define TFO_SERVER_WO_SOCKOPT1 0x400 234 #define TFO_SERVER_WO_SOCKOPT2 0x800 235 236 extern struct inet_timewait_death_row tcp_death_row; 237 238 /* sysctl variables for tcp */ 239 extern int sysctl_tcp_timestamps; 240 extern int sysctl_tcp_window_scaling; 241 extern int sysctl_tcp_sack; 242 extern int sysctl_tcp_fastopen; 243 extern int sysctl_tcp_retrans_collapse; 244 extern int sysctl_tcp_stdurg; 245 extern int sysctl_tcp_rfc1337; 246 extern int sysctl_tcp_abort_on_overflow; 247 extern int sysctl_tcp_max_orphans; 248 extern int sysctl_tcp_fack; 249 extern int sysctl_tcp_reordering; 250 extern int sysctl_tcp_max_reordering; 251 extern int sysctl_tcp_dsack; 252 extern long sysctl_tcp_mem[3]; 253 extern int sysctl_tcp_wmem[3]; 254 extern int sysctl_tcp_rmem[3]; 255 extern int sysctl_tcp_app_win; 256 extern int sysctl_tcp_adv_win_scale; 257 extern int sysctl_tcp_tw_reuse; 258 extern int sysctl_tcp_frto; 259 extern int sysctl_tcp_low_latency; 260 extern int sysctl_tcp_nometrics_save; 261 extern int sysctl_tcp_moderate_rcvbuf; 262 extern int sysctl_tcp_tso_win_divisor; 263 extern int sysctl_tcp_workaround_signed_windows; 264 extern int sysctl_tcp_slow_start_after_idle; 265 extern int sysctl_tcp_thin_linear_timeouts; 266 extern int sysctl_tcp_thin_dupack; 267 extern int sysctl_tcp_early_retrans; 268 extern int sysctl_tcp_limit_output_bytes; 269 extern int sysctl_tcp_challenge_ack_limit; 270 extern int sysctl_tcp_min_tso_segs; 271 extern int sysctl_tcp_min_rtt_wlen; 272 extern int sysctl_tcp_autocorking; 273 extern int sysctl_tcp_invalid_ratelimit; 274 extern int sysctl_tcp_pacing_ss_ratio; 275 extern int sysctl_tcp_pacing_ca_ratio; 276 277 extern atomic_long_t tcp_memory_allocated; 278 extern struct percpu_counter tcp_sockets_allocated; 279 extern int tcp_memory_pressure; 280 281 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 282 static inline bool tcp_under_memory_pressure(const struct sock *sk) 283 { 284 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 285 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 286 return true; 287 288 return tcp_memory_pressure; 289 } 290 /* 291 * The next routines deal with comparing 32 bit unsigned ints 292 * and worry about wraparound (automatic with unsigned arithmetic). 293 */ 294 295 static inline bool before(__u32 seq1, __u32 seq2) 296 { 297 return (__s32)(seq1-seq2) < 0; 298 } 299 #define after(seq2, seq1) before(seq1, seq2) 300 301 /* is s2<=s1<=s3 ? */ 302 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 303 { 304 return seq3 - seq2 >= seq1 - seq2; 305 } 306 307 static inline bool tcp_out_of_memory(struct sock *sk) 308 { 309 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 310 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 311 return true; 312 return false; 313 } 314 315 void sk_forced_mem_schedule(struct sock *sk, int size); 316 317 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 318 { 319 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 320 int orphans = percpu_counter_read_positive(ocp); 321 322 if (orphans << shift > sysctl_tcp_max_orphans) { 323 orphans = percpu_counter_sum_positive(ocp); 324 if (orphans << shift > sysctl_tcp_max_orphans) 325 return true; 326 } 327 return false; 328 } 329 330 bool tcp_check_oom(struct sock *sk, int shift); 331 332 333 extern struct proto tcp_prot; 334 335 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 336 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 337 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 338 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 339 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 340 341 void tcp_tasklet_init(void); 342 343 void tcp_v4_err(struct sk_buff *skb, u32); 344 345 void tcp_shutdown(struct sock *sk, int how); 346 347 void tcp_v4_early_demux(struct sk_buff *skb); 348 int tcp_v4_rcv(struct sk_buff *skb); 349 350 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 351 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 352 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 353 int flags); 354 void tcp_release_cb(struct sock *sk); 355 void tcp_wfree(struct sk_buff *skb); 356 void tcp_write_timer_handler(struct sock *sk); 357 void tcp_delack_timer_handler(struct sock *sk); 358 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 359 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 360 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 361 const struct tcphdr *th, unsigned int len); 362 void tcp_rcv_space_adjust(struct sock *sk); 363 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 364 void tcp_twsk_destructor(struct sock *sk); 365 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 366 struct pipe_inode_info *pipe, size_t len, 367 unsigned int flags); 368 369 static inline void tcp_dec_quickack_mode(struct sock *sk, 370 const unsigned int pkts) 371 { 372 struct inet_connection_sock *icsk = inet_csk(sk); 373 374 if (icsk->icsk_ack.quick) { 375 if (pkts >= icsk->icsk_ack.quick) { 376 icsk->icsk_ack.quick = 0; 377 /* Leaving quickack mode we deflate ATO. */ 378 icsk->icsk_ack.ato = TCP_ATO_MIN; 379 } else 380 icsk->icsk_ack.quick -= pkts; 381 } 382 } 383 384 #define TCP_ECN_OK 1 385 #define TCP_ECN_QUEUE_CWR 2 386 #define TCP_ECN_DEMAND_CWR 4 387 #define TCP_ECN_SEEN 8 388 389 enum tcp_tw_status { 390 TCP_TW_SUCCESS = 0, 391 TCP_TW_RST = 1, 392 TCP_TW_ACK = 2, 393 TCP_TW_SYN = 3 394 }; 395 396 397 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 398 struct sk_buff *skb, 399 const struct tcphdr *th); 400 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 401 struct request_sock *req, bool fastopen); 402 int tcp_child_process(struct sock *parent, struct sock *child, 403 struct sk_buff *skb); 404 void tcp_enter_loss(struct sock *sk); 405 void tcp_clear_retrans(struct tcp_sock *tp); 406 void tcp_update_metrics(struct sock *sk); 407 void tcp_init_metrics(struct sock *sk); 408 void tcp_metrics_init(void); 409 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, 410 bool paws_check, bool timestamps); 411 bool tcp_remember_stamp(struct sock *sk); 412 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); 413 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); 414 void tcp_disable_fack(struct tcp_sock *tp); 415 void tcp_close(struct sock *sk, long timeout); 416 void tcp_init_sock(struct sock *sk); 417 unsigned int tcp_poll(struct file *file, struct socket *sock, 418 struct poll_table_struct *wait); 419 int tcp_getsockopt(struct sock *sk, int level, int optname, 420 char __user *optval, int __user *optlen); 421 int tcp_setsockopt(struct sock *sk, int level, int optname, 422 char __user *optval, unsigned int optlen); 423 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 424 char __user *optval, int __user *optlen); 425 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 426 char __user *optval, unsigned int optlen); 427 void tcp_set_keepalive(struct sock *sk, int val); 428 void tcp_syn_ack_timeout(const struct request_sock *req); 429 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 430 int flags, int *addr_len); 431 void tcp_parse_options(const struct sk_buff *skb, 432 struct tcp_options_received *opt_rx, 433 int estab, struct tcp_fastopen_cookie *foc); 434 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 435 436 /* 437 * TCP v4 functions exported for the inet6 API 438 */ 439 440 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 441 void tcp_v4_mtu_reduced(struct sock *sk); 442 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 443 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 444 struct sock *tcp_create_openreq_child(const struct sock *sk, 445 struct request_sock *req, 446 struct sk_buff *skb); 447 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 448 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 449 struct request_sock *req, 450 struct dst_entry *dst, 451 struct request_sock *req_unhash, 452 bool *own_req); 453 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 454 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 455 int tcp_connect(struct sock *sk); 456 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 457 struct request_sock *req, 458 struct tcp_fastopen_cookie *foc, 459 bool attach_req); 460 int tcp_disconnect(struct sock *sk, int flags); 461 462 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 463 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 464 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 465 466 /* From syncookies.c */ 467 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 468 struct request_sock *req, 469 struct dst_entry *dst); 470 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 471 u32 cookie); 472 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 473 #ifdef CONFIG_SYN_COOKIES 474 475 /* Syncookies use a monotonic timer which increments every 60 seconds. 476 * This counter is used both as a hash input and partially encoded into 477 * the cookie value. A cookie is only validated further if the delta 478 * between the current counter value and the encoded one is less than this, 479 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 480 * the counter advances immediately after a cookie is generated). 481 */ 482 #define MAX_SYNCOOKIE_AGE 2 483 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 484 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 485 486 /* syncookies: remember time of last synqueue overflow 487 * But do not dirty this field too often (once per second is enough) 488 * It is racy as we do not hold a lock, but race is very minor. 489 */ 490 static inline void tcp_synq_overflow(const struct sock *sk) 491 { 492 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 493 unsigned long now = jiffies; 494 495 if (time_after(now, last_overflow + HZ)) 496 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 497 } 498 499 /* syncookies: no recent synqueue overflow on this listening socket? */ 500 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 501 { 502 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 503 504 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID); 505 } 506 507 static inline u32 tcp_cookie_time(void) 508 { 509 u64 val = get_jiffies_64(); 510 511 do_div(val, TCP_SYNCOOKIE_PERIOD); 512 return val; 513 } 514 515 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 516 u16 *mssp); 517 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 518 __u32 cookie_init_timestamp(struct request_sock *req); 519 bool cookie_timestamp_decode(struct tcp_options_received *opt); 520 bool cookie_ecn_ok(const struct tcp_options_received *opt, 521 const struct net *net, const struct dst_entry *dst); 522 523 /* From net/ipv6/syncookies.c */ 524 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 525 u32 cookie); 526 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 527 528 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 529 const struct tcphdr *th, u16 *mssp); 530 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 531 #endif 532 /* tcp_output.c */ 533 534 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 535 int nonagle); 536 bool tcp_may_send_now(struct sock *sk); 537 int __tcp_retransmit_skb(struct sock *, struct sk_buff *); 538 int tcp_retransmit_skb(struct sock *, struct sk_buff *); 539 void tcp_retransmit_timer(struct sock *sk); 540 void tcp_xmit_retransmit_queue(struct sock *); 541 void tcp_simple_retransmit(struct sock *); 542 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 543 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t); 544 545 void tcp_send_probe0(struct sock *); 546 void tcp_send_partial(struct sock *); 547 int tcp_write_wakeup(struct sock *, int mib); 548 void tcp_send_fin(struct sock *sk); 549 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 550 int tcp_send_synack(struct sock *); 551 void tcp_push_one(struct sock *, unsigned int mss_now); 552 void tcp_send_ack(struct sock *sk); 553 void tcp_send_delayed_ack(struct sock *sk); 554 void tcp_send_loss_probe(struct sock *sk); 555 bool tcp_schedule_loss_probe(struct sock *sk); 556 557 /* tcp_input.c */ 558 void tcp_resume_early_retransmit(struct sock *sk); 559 void tcp_rearm_rto(struct sock *sk); 560 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 561 void tcp_reset(struct sock *sk); 562 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 563 void tcp_fin(struct sock *sk); 564 565 /* tcp_timer.c */ 566 void tcp_init_xmit_timers(struct sock *); 567 static inline void tcp_clear_xmit_timers(struct sock *sk) 568 { 569 inet_csk_clear_xmit_timers(sk); 570 } 571 572 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 573 unsigned int tcp_current_mss(struct sock *sk); 574 575 /* Bound MSS / TSO packet size with the half of the window */ 576 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 577 { 578 int cutoff; 579 580 /* When peer uses tiny windows, there is no use in packetizing 581 * to sub-MSS pieces for the sake of SWS or making sure there 582 * are enough packets in the pipe for fast recovery. 583 * 584 * On the other hand, for extremely large MSS devices, handling 585 * smaller than MSS windows in this way does make sense. 586 */ 587 if (tp->max_window >= 512) 588 cutoff = (tp->max_window >> 1); 589 else 590 cutoff = tp->max_window; 591 592 if (cutoff && pktsize > cutoff) 593 return max_t(int, cutoff, 68U - tp->tcp_header_len); 594 else 595 return pktsize; 596 } 597 598 /* tcp.c */ 599 void tcp_get_info(struct sock *, struct tcp_info *); 600 601 /* Read 'sendfile()'-style from a TCP socket */ 602 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 603 unsigned int, size_t); 604 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 605 sk_read_actor_t recv_actor); 606 607 void tcp_initialize_rcv_mss(struct sock *sk); 608 609 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 610 int tcp_mss_to_mtu(struct sock *sk, int mss); 611 void tcp_mtup_init(struct sock *sk); 612 void tcp_init_buffer_space(struct sock *sk); 613 614 static inline void tcp_bound_rto(const struct sock *sk) 615 { 616 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 617 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 618 } 619 620 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 621 { 622 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 623 } 624 625 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 626 { 627 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 628 ntohl(TCP_FLAG_ACK) | 629 snd_wnd); 630 } 631 632 static inline void tcp_fast_path_on(struct tcp_sock *tp) 633 { 634 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 635 } 636 637 static inline void tcp_fast_path_check(struct sock *sk) 638 { 639 struct tcp_sock *tp = tcp_sk(sk); 640 641 if (skb_queue_empty(&tp->out_of_order_queue) && 642 tp->rcv_wnd && 643 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 644 !tp->urg_data) 645 tcp_fast_path_on(tp); 646 } 647 648 /* Compute the actual rto_min value */ 649 static inline u32 tcp_rto_min(struct sock *sk) 650 { 651 const struct dst_entry *dst = __sk_dst_get(sk); 652 u32 rto_min = TCP_RTO_MIN; 653 654 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 655 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 656 return rto_min; 657 } 658 659 static inline u32 tcp_rto_min_us(struct sock *sk) 660 { 661 return jiffies_to_usecs(tcp_rto_min(sk)); 662 } 663 664 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 665 { 666 return dst_metric_locked(dst, RTAX_CC_ALGO); 667 } 668 669 /* Minimum RTT in usec. ~0 means not available. */ 670 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 671 { 672 return tp->rtt_min[0].rtt; 673 } 674 675 /* Compute the actual receive window we are currently advertising. 676 * Rcv_nxt can be after the window if our peer push more data 677 * than the offered window. 678 */ 679 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 680 { 681 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 682 683 if (win < 0) 684 win = 0; 685 return (u32) win; 686 } 687 688 /* Choose a new window, without checks for shrinking, and without 689 * scaling applied to the result. The caller does these things 690 * if necessary. This is a "raw" window selection. 691 */ 692 u32 __tcp_select_window(struct sock *sk); 693 694 void tcp_send_window_probe(struct sock *sk); 695 696 /* TCP timestamps are only 32-bits, this causes a slight 697 * complication on 64-bit systems since we store a snapshot 698 * of jiffies in the buffer control blocks below. We decided 699 * to use only the low 32-bits of jiffies and hide the ugly 700 * casts with the following macro. 701 */ 702 #define tcp_time_stamp ((__u32)(jiffies)) 703 704 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 705 { 706 return skb->skb_mstamp.stamp_jiffies; 707 } 708 709 710 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 711 712 #define TCPHDR_FIN 0x01 713 #define TCPHDR_SYN 0x02 714 #define TCPHDR_RST 0x04 715 #define TCPHDR_PSH 0x08 716 #define TCPHDR_ACK 0x10 717 #define TCPHDR_URG 0x20 718 #define TCPHDR_ECE 0x40 719 #define TCPHDR_CWR 0x80 720 721 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 722 723 /* This is what the send packet queuing engine uses to pass 724 * TCP per-packet control information to the transmission code. 725 * We also store the host-order sequence numbers in here too. 726 * This is 44 bytes if IPV6 is enabled. 727 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 728 */ 729 struct tcp_skb_cb { 730 __u32 seq; /* Starting sequence number */ 731 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 732 union { 733 /* Note : tcp_tw_isn is used in input path only 734 * (isn chosen by tcp_timewait_state_process()) 735 * 736 * tcp_gso_segs/size are used in write queue only, 737 * cf tcp_skb_pcount()/tcp_skb_mss() 738 */ 739 __u32 tcp_tw_isn; 740 struct { 741 u16 tcp_gso_segs; 742 u16 tcp_gso_size; 743 }; 744 }; 745 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 746 747 __u8 sacked; /* State flags for SACK/FACK. */ 748 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 749 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 750 #define TCPCB_LOST 0x04 /* SKB is lost */ 751 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 752 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 753 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 754 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 755 TCPCB_REPAIRED) 756 757 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 758 /* 1 byte hole */ 759 __u32 ack_seq; /* Sequence number ACK'd */ 760 union { 761 struct inet_skb_parm h4; 762 #if IS_ENABLED(CONFIG_IPV6) 763 struct inet6_skb_parm h6; 764 #endif 765 } header; /* For incoming frames */ 766 }; 767 768 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 769 770 771 #if IS_ENABLED(CONFIG_IPV6) 772 /* This is the variant of inet6_iif() that must be used by TCP, 773 * as TCP moves IP6CB into a different location in skb->cb[] 774 */ 775 static inline int tcp_v6_iif(const struct sk_buff *skb) 776 { 777 return TCP_SKB_CB(skb)->header.h6.iif; 778 } 779 #endif 780 781 /* Due to TSO, an SKB can be composed of multiple actual 782 * packets. To keep these tracked properly, we use this. 783 */ 784 static inline int tcp_skb_pcount(const struct sk_buff *skb) 785 { 786 return TCP_SKB_CB(skb)->tcp_gso_segs; 787 } 788 789 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 790 { 791 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 792 } 793 794 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 795 { 796 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 797 } 798 799 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 800 static inline int tcp_skb_mss(const struct sk_buff *skb) 801 { 802 return TCP_SKB_CB(skb)->tcp_gso_size; 803 } 804 805 /* Events passed to congestion control interface */ 806 enum tcp_ca_event { 807 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 808 CA_EVENT_CWND_RESTART, /* congestion window restart */ 809 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 810 CA_EVENT_LOSS, /* loss timeout */ 811 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 812 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 813 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 814 CA_EVENT_NON_DELAYED_ACK, 815 }; 816 817 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 818 enum tcp_ca_ack_event_flags { 819 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 820 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 821 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 822 }; 823 824 /* 825 * Interface for adding new TCP congestion control handlers 826 */ 827 #define TCP_CA_NAME_MAX 16 828 #define TCP_CA_MAX 128 829 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 830 831 #define TCP_CA_UNSPEC 0 832 833 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 834 #define TCP_CONG_NON_RESTRICTED 0x1 835 /* Requires ECN/ECT set on all packets */ 836 #define TCP_CONG_NEEDS_ECN 0x2 837 838 union tcp_cc_info; 839 840 struct tcp_congestion_ops { 841 struct list_head list; 842 u32 key; 843 u32 flags; 844 845 /* initialize private data (optional) */ 846 void (*init)(struct sock *sk); 847 /* cleanup private data (optional) */ 848 void (*release)(struct sock *sk); 849 850 /* return slow start threshold (required) */ 851 u32 (*ssthresh)(struct sock *sk); 852 /* do new cwnd calculation (required) */ 853 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 854 /* call before changing ca_state (optional) */ 855 void (*set_state)(struct sock *sk, u8 new_state); 856 /* call when cwnd event occurs (optional) */ 857 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 858 /* call when ack arrives (optional) */ 859 void (*in_ack_event)(struct sock *sk, u32 flags); 860 /* new value of cwnd after loss (optional) */ 861 u32 (*undo_cwnd)(struct sock *sk); 862 /* hook for packet ack accounting (optional) */ 863 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 864 /* get info for inet_diag (optional) */ 865 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 866 union tcp_cc_info *info); 867 868 char name[TCP_CA_NAME_MAX]; 869 struct module *owner; 870 }; 871 872 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 873 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 874 875 void tcp_assign_congestion_control(struct sock *sk); 876 void tcp_init_congestion_control(struct sock *sk); 877 void tcp_cleanup_congestion_control(struct sock *sk); 878 int tcp_set_default_congestion_control(const char *name); 879 void tcp_get_default_congestion_control(char *name); 880 void tcp_get_available_congestion_control(char *buf, size_t len); 881 void tcp_get_allowed_congestion_control(char *buf, size_t len); 882 int tcp_set_allowed_congestion_control(char *allowed); 883 int tcp_set_congestion_control(struct sock *sk, const char *name); 884 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 885 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 886 887 u32 tcp_reno_ssthresh(struct sock *sk); 888 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 889 extern struct tcp_congestion_ops tcp_reno; 890 891 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 892 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca); 893 #ifdef CONFIG_INET 894 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 895 #else 896 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 897 { 898 return NULL; 899 } 900 #endif 901 902 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 903 { 904 const struct inet_connection_sock *icsk = inet_csk(sk); 905 906 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 907 } 908 909 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 910 { 911 struct inet_connection_sock *icsk = inet_csk(sk); 912 913 if (icsk->icsk_ca_ops->set_state) 914 icsk->icsk_ca_ops->set_state(sk, ca_state); 915 icsk->icsk_ca_state = ca_state; 916 } 917 918 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 919 { 920 const struct inet_connection_sock *icsk = inet_csk(sk); 921 922 if (icsk->icsk_ca_ops->cwnd_event) 923 icsk->icsk_ca_ops->cwnd_event(sk, event); 924 } 925 926 /* These functions determine how the current flow behaves in respect of SACK 927 * handling. SACK is negotiated with the peer, and therefore it can vary 928 * between different flows. 929 * 930 * tcp_is_sack - SACK enabled 931 * tcp_is_reno - No SACK 932 * tcp_is_fack - FACK enabled, implies SACK enabled 933 */ 934 static inline int tcp_is_sack(const struct tcp_sock *tp) 935 { 936 return tp->rx_opt.sack_ok; 937 } 938 939 static inline bool tcp_is_reno(const struct tcp_sock *tp) 940 { 941 return !tcp_is_sack(tp); 942 } 943 944 static inline bool tcp_is_fack(const struct tcp_sock *tp) 945 { 946 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 947 } 948 949 static inline void tcp_enable_fack(struct tcp_sock *tp) 950 { 951 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 952 } 953 954 /* TCP early-retransmit (ER) is similar to but more conservative than 955 * the thin-dupack feature. Enable ER only if thin-dupack is disabled. 956 */ 957 static inline void tcp_enable_early_retrans(struct tcp_sock *tp) 958 { 959 struct net *net = sock_net((struct sock *)tp); 960 961 tp->do_early_retrans = sysctl_tcp_early_retrans && 962 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack && 963 net->ipv4.sysctl_tcp_reordering == 3; 964 } 965 966 static inline void tcp_disable_early_retrans(struct tcp_sock *tp) 967 { 968 tp->do_early_retrans = 0; 969 } 970 971 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 972 { 973 return tp->sacked_out + tp->lost_out; 974 } 975 976 /* This determines how many packets are "in the network" to the best 977 * of our knowledge. In many cases it is conservative, but where 978 * detailed information is available from the receiver (via SACK 979 * blocks etc.) we can make more aggressive calculations. 980 * 981 * Use this for decisions involving congestion control, use just 982 * tp->packets_out to determine if the send queue is empty or not. 983 * 984 * Read this equation as: 985 * 986 * "Packets sent once on transmission queue" MINUS 987 * "Packets left network, but not honestly ACKed yet" PLUS 988 * "Packets fast retransmitted" 989 */ 990 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 991 { 992 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 993 } 994 995 #define TCP_INFINITE_SSTHRESH 0x7fffffff 996 997 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 998 { 999 return tp->snd_cwnd < tp->snd_ssthresh; 1000 } 1001 1002 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1003 { 1004 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1005 } 1006 1007 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1008 { 1009 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1010 (1 << inet_csk(sk)->icsk_ca_state); 1011 } 1012 1013 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1014 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1015 * ssthresh. 1016 */ 1017 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1018 { 1019 const struct tcp_sock *tp = tcp_sk(sk); 1020 1021 if (tcp_in_cwnd_reduction(sk)) 1022 return tp->snd_ssthresh; 1023 else 1024 return max(tp->snd_ssthresh, 1025 ((tp->snd_cwnd >> 1) + 1026 (tp->snd_cwnd >> 2))); 1027 } 1028 1029 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1030 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1031 1032 void tcp_enter_cwr(struct sock *sk); 1033 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1034 1035 /* The maximum number of MSS of available cwnd for which TSO defers 1036 * sending if not using sysctl_tcp_tso_win_divisor. 1037 */ 1038 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1039 { 1040 return 3; 1041 } 1042 1043 /* Slow start with delack produces 3 packets of burst, so that 1044 * it is safe "de facto". This will be the default - same as 1045 * the default reordering threshold - but if reordering increases, 1046 * we must be able to allow cwnd to burst at least this much in order 1047 * to not pull it back when holes are filled. 1048 */ 1049 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 1050 { 1051 return tp->reordering; 1052 } 1053 1054 /* Returns end sequence number of the receiver's advertised window */ 1055 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1056 { 1057 return tp->snd_una + tp->snd_wnd; 1058 } 1059 1060 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1061 * flexible approach. The RFC suggests cwnd should not be raised unless 1062 * it was fully used previously. And that's exactly what we do in 1063 * congestion avoidance mode. But in slow start we allow cwnd to grow 1064 * as long as the application has used half the cwnd. 1065 * Example : 1066 * cwnd is 10 (IW10), but application sends 9 frames. 1067 * We allow cwnd to reach 18 when all frames are ACKed. 1068 * This check is safe because it's as aggressive as slow start which already 1069 * risks 100% overshoot. The advantage is that we discourage application to 1070 * either send more filler packets or data to artificially blow up the cwnd 1071 * usage, and allow application-limited process to probe bw more aggressively. 1072 */ 1073 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1074 { 1075 const struct tcp_sock *tp = tcp_sk(sk); 1076 1077 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1078 if (tcp_in_slow_start(tp)) 1079 return tp->snd_cwnd < 2 * tp->max_packets_out; 1080 1081 return tp->is_cwnd_limited; 1082 } 1083 1084 /* Something is really bad, we could not queue an additional packet, 1085 * because qdisc is full or receiver sent a 0 window. 1086 * We do not want to add fuel to the fire, or abort too early, 1087 * so make sure the timer we arm now is at least 200ms in the future, 1088 * regardless of current icsk_rto value (as it could be ~2ms) 1089 */ 1090 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1091 { 1092 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1093 } 1094 1095 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1096 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1097 unsigned long max_when) 1098 { 1099 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1100 1101 return (unsigned long)min_t(u64, when, max_when); 1102 } 1103 1104 static inline void tcp_check_probe_timer(struct sock *sk) 1105 { 1106 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1107 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1108 tcp_probe0_base(sk), TCP_RTO_MAX); 1109 } 1110 1111 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1112 { 1113 tp->snd_wl1 = seq; 1114 } 1115 1116 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1117 { 1118 tp->snd_wl1 = seq; 1119 } 1120 1121 /* 1122 * Calculate(/check) TCP checksum 1123 */ 1124 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1125 __be32 daddr, __wsum base) 1126 { 1127 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1128 } 1129 1130 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1131 { 1132 return __skb_checksum_complete(skb); 1133 } 1134 1135 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1136 { 1137 return !skb_csum_unnecessary(skb) && 1138 __tcp_checksum_complete(skb); 1139 } 1140 1141 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1142 1143 static inline void tcp_prequeue_init(struct tcp_sock *tp) 1144 { 1145 tp->ucopy.task = NULL; 1146 tp->ucopy.len = 0; 1147 tp->ucopy.memory = 0; 1148 skb_queue_head_init(&tp->ucopy.prequeue); 1149 } 1150 1151 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1152 1153 #undef STATE_TRACE 1154 1155 #ifdef STATE_TRACE 1156 static const char *statename[]={ 1157 "Unused","Established","Syn Sent","Syn Recv", 1158 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1159 "Close Wait","Last ACK","Listen","Closing" 1160 }; 1161 #endif 1162 void tcp_set_state(struct sock *sk, int state); 1163 1164 void tcp_done(struct sock *sk); 1165 1166 int tcp_abort(struct sock *sk, int err); 1167 1168 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1169 { 1170 rx_opt->dsack = 0; 1171 rx_opt->num_sacks = 0; 1172 } 1173 1174 u32 tcp_default_init_rwnd(u32 mss); 1175 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1176 1177 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1178 { 1179 struct tcp_sock *tp = tcp_sk(sk); 1180 s32 delta; 1181 1182 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out) 1183 return; 1184 delta = tcp_time_stamp - tp->lsndtime; 1185 if (delta > inet_csk(sk)->icsk_rto) 1186 tcp_cwnd_restart(sk, delta); 1187 } 1188 1189 /* Determine a window scaling and initial window to offer. */ 1190 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1191 __u32 *window_clamp, int wscale_ok, 1192 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1193 1194 static inline int tcp_win_from_space(int space) 1195 { 1196 return sysctl_tcp_adv_win_scale<=0 ? 1197 (space>>(-sysctl_tcp_adv_win_scale)) : 1198 space - (space>>sysctl_tcp_adv_win_scale); 1199 } 1200 1201 /* Note: caller must be prepared to deal with negative returns */ 1202 static inline int tcp_space(const struct sock *sk) 1203 { 1204 return tcp_win_from_space(sk->sk_rcvbuf - 1205 atomic_read(&sk->sk_rmem_alloc)); 1206 } 1207 1208 static inline int tcp_full_space(const struct sock *sk) 1209 { 1210 return tcp_win_from_space(sk->sk_rcvbuf); 1211 } 1212 1213 extern void tcp_openreq_init_rwin(struct request_sock *req, 1214 const struct sock *sk_listener, 1215 const struct dst_entry *dst); 1216 1217 void tcp_enter_memory_pressure(struct sock *sk); 1218 1219 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1220 { 1221 struct net *net = sock_net((struct sock *)tp); 1222 1223 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1224 } 1225 1226 static inline int keepalive_time_when(const struct tcp_sock *tp) 1227 { 1228 struct net *net = sock_net((struct sock *)tp); 1229 1230 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1231 } 1232 1233 static inline int keepalive_probes(const struct tcp_sock *tp) 1234 { 1235 struct net *net = sock_net((struct sock *)tp); 1236 1237 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1238 } 1239 1240 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1241 { 1242 const struct inet_connection_sock *icsk = &tp->inet_conn; 1243 1244 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1245 tcp_time_stamp - tp->rcv_tstamp); 1246 } 1247 1248 static inline int tcp_fin_time(const struct sock *sk) 1249 { 1250 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1251 const int rto = inet_csk(sk)->icsk_rto; 1252 1253 if (fin_timeout < (rto << 2) - (rto >> 1)) 1254 fin_timeout = (rto << 2) - (rto >> 1); 1255 1256 return fin_timeout; 1257 } 1258 1259 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1260 int paws_win) 1261 { 1262 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1263 return true; 1264 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1265 return true; 1266 /* 1267 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1268 * then following tcp messages have valid values. Ignore 0 value, 1269 * or else 'negative' tsval might forbid us to accept their packets. 1270 */ 1271 if (!rx_opt->ts_recent) 1272 return true; 1273 return false; 1274 } 1275 1276 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1277 int rst) 1278 { 1279 if (tcp_paws_check(rx_opt, 0)) 1280 return false; 1281 1282 /* RST segments are not recommended to carry timestamp, 1283 and, if they do, it is recommended to ignore PAWS because 1284 "their cleanup function should take precedence over timestamps." 1285 Certainly, it is mistake. It is necessary to understand the reasons 1286 of this constraint to relax it: if peer reboots, clock may go 1287 out-of-sync and half-open connections will not be reset. 1288 Actually, the problem would be not existing if all 1289 the implementations followed draft about maintaining clock 1290 via reboots. Linux-2.2 DOES NOT! 1291 1292 However, we can relax time bounds for RST segments to MSL. 1293 */ 1294 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1295 return false; 1296 return true; 1297 } 1298 1299 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1300 int mib_idx, u32 *last_oow_ack_time); 1301 1302 static inline void tcp_mib_init(struct net *net) 1303 { 1304 /* See RFC 2012 */ 1305 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1306 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1307 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1308 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1309 } 1310 1311 /* from STCP */ 1312 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1313 { 1314 tp->lost_skb_hint = NULL; 1315 } 1316 1317 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1318 { 1319 tcp_clear_retrans_hints_partial(tp); 1320 tp->retransmit_skb_hint = NULL; 1321 } 1322 1323 /* MD5 Signature */ 1324 struct crypto_hash; 1325 1326 union tcp_md5_addr { 1327 struct in_addr a4; 1328 #if IS_ENABLED(CONFIG_IPV6) 1329 struct in6_addr a6; 1330 #endif 1331 }; 1332 1333 /* - key database */ 1334 struct tcp_md5sig_key { 1335 struct hlist_node node; 1336 u8 keylen; 1337 u8 family; /* AF_INET or AF_INET6 */ 1338 union tcp_md5_addr addr; 1339 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1340 struct rcu_head rcu; 1341 }; 1342 1343 /* - sock block */ 1344 struct tcp_md5sig_info { 1345 struct hlist_head head; 1346 struct rcu_head rcu; 1347 }; 1348 1349 /* - pseudo header */ 1350 struct tcp4_pseudohdr { 1351 __be32 saddr; 1352 __be32 daddr; 1353 __u8 pad; 1354 __u8 protocol; 1355 __be16 len; 1356 }; 1357 1358 struct tcp6_pseudohdr { 1359 struct in6_addr saddr; 1360 struct in6_addr daddr; 1361 __be32 len; 1362 __be32 protocol; /* including padding */ 1363 }; 1364 1365 union tcp_md5sum_block { 1366 struct tcp4_pseudohdr ip4; 1367 #if IS_ENABLED(CONFIG_IPV6) 1368 struct tcp6_pseudohdr ip6; 1369 #endif 1370 }; 1371 1372 /* - pool: digest algorithm, hash description and scratch buffer */ 1373 struct tcp_md5sig_pool { 1374 struct hash_desc md5_desc; 1375 union tcp_md5sum_block md5_blk; 1376 }; 1377 1378 /* - functions */ 1379 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1380 const struct sock *sk, const struct sk_buff *skb); 1381 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1382 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); 1383 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1384 int family); 1385 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1386 const struct sock *addr_sk); 1387 1388 #ifdef CONFIG_TCP_MD5SIG 1389 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1390 const union tcp_md5_addr *addr, 1391 int family); 1392 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1393 #else 1394 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1395 const union tcp_md5_addr *addr, 1396 int family) 1397 { 1398 return NULL; 1399 } 1400 #define tcp_twsk_md5_key(twsk) NULL 1401 #endif 1402 1403 bool tcp_alloc_md5sig_pool(void); 1404 1405 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1406 static inline void tcp_put_md5sig_pool(void) 1407 { 1408 local_bh_enable(); 1409 } 1410 1411 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1412 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1413 unsigned int header_len); 1414 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1415 const struct tcp_md5sig_key *key); 1416 1417 /* From tcp_fastopen.c */ 1418 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1419 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1420 unsigned long *last_syn_loss); 1421 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1422 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1423 u16 try_exp); 1424 struct tcp_fastopen_request { 1425 /* Fast Open cookie. Size 0 means a cookie request */ 1426 struct tcp_fastopen_cookie cookie; 1427 struct msghdr *data; /* data in MSG_FASTOPEN */ 1428 size_t size; 1429 int copied; /* queued in tcp_connect() */ 1430 }; 1431 void tcp_free_fastopen_req(struct tcp_sock *tp); 1432 1433 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1434 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1435 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 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 struct net *net = sock_net((struct sock *)tp); 1685 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1686 } 1687 1688 static inline bool tcp_stream_memory_free(const struct sock *sk) 1689 { 1690 const struct tcp_sock *tp = tcp_sk(sk); 1691 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1692 1693 return notsent_bytes < tcp_notsent_lowat(tp); 1694 } 1695 1696 #ifdef CONFIG_PROC_FS 1697 int tcp4_proc_init(void); 1698 void tcp4_proc_exit(void); 1699 #endif 1700 1701 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1702 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1703 const struct tcp_request_sock_ops *af_ops, 1704 struct sock *sk, struct sk_buff *skb); 1705 1706 /* TCP af-specific functions */ 1707 struct tcp_sock_af_ops { 1708 #ifdef CONFIG_TCP_MD5SIG 1709 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1710 const struct sock *addr_sk); 1711 int (*calc_md5_hash)(char *location, 1712 const struct tcp_md5sig_key *md5, 1713 const struct sock *sk, 1714 const struct sk_buff *skb); 1715 int (*md5_parse)(struct sock *sk, 1716 char __user *optval, 1717 int optlen); 1718 #endif 1719 }; 1720 1721 struct tcp_request_sock_ops { 1722 u16 mss_clamp; 1723 #ifdef CONFIG_TCP_MD5SIG 1724 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1725 const struct sock *addr_sk); 1726 int (*calc_md5_hash) (char *location, 1727 const struct tcp_md5sig_key *md5, 1728 const struct sock *sk, 1729 const struct sk_buff *skb); 1730 #endif 1731 void (*init_req)(struct request_sock *req, 1732 const struct sock *sk_listener, 1733 struct sk_buff *skb); 1734 #ifdef CONFIG_SYN_COOKIES 1735 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1736 __u16 *mss); 1737 #endif 1738 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1739 const struct request_sock *req, 1740 bool *strict); 1741 __u32 (*init_seq)(const struct sk_buff *skb); 1742 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1743 struct flowi *fl, struct request_sock *req, 1744 struct tcp_fastopen_cookie *foc, 1745 bool attach_req); 1746 }; 1747 1748 #ifdef CONFIG_SYN_COOKIES 1749 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1750 const struct sock *sk, struct sk_buff *skb, 1751 __u16 *mss) 1752 { 1753 tcp_synq_overflow(sk); 1754 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1755 return ops->cookie_init_seq(skb, mss); 1756 } 1757 #else 1758 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1759 const struct sock *sk, struct sk_buff *skb, 1760 __u16 *mss) 1761 { 1762 return 0; 1763 } 1764 #endif 1765 1766 int tcpv4_offload_init(void); 1767 1768 void tcp_v4_init(void); 1769 void tcp_init(void); 1770 1771 /* tcp_recovery.c */ 1772 1773 /* Flags to enable various loss recovery features. See below */ 1774 extern int sysctl_tcp_recovery; 1775 1776 /* Use TCP RACK to detect (some) tail and retransmit losses */ 1777 #define TCP_RACK_LOST_RETRANS 0x1 1778 1779 extern int tcp_rack_mark_lost(struct sock *sk); 1780 1781 extern void tcp_rack_advance(struct tcp_sock *tp, 1782 const struct skb_mstamp *xmit_time, u8 sacked); 1783 1784 /* 1785 * Save and compile IPv4 options, return a pointer to it 1786 */ 1787 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb) 1788 { 1789 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1790 struct ip_options_rcu *dopt = NULL; 1791 1792 if (opt->optlen) { 1793 int opt_size = sizeof(*dopt) + opt->optlen; 1794 1795 dopt = kmalloc(opt_size, GFP_ATOMIC); 1796 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) { 1797 kfree(dopt); 1798 dopt = NULL; 1799 } 1800 } 1801 return dopt; 1802 } 1803 1804 /* locally generated TCP pure ACKs have skb->truesize == 2 1805 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1806 * This is much faster than dissecting the packet to find out. 1807 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1808 */ 1809 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1810 { 1811 return skb->truesize == 2; 1812 } 1813 1814 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1815 { 1816 skb->truesize = 2; 1817 } 1818 1819 #endif /* _TCP_H */ 1820