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 draft-hkchu-tcpm-initcwnd-01 */ 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_fin_timeout; 243 extern int sysctl_tcp_keepalive_time; 244 extern int sysctl_tcp_keepalive_probes; 245 extern int sysctl_tcp_keepalive_intvl; 246 extern int sysctl_tcp_syn_retries; 247 extern int sysctl_tcp_synack_retries; 248 extern int sysctl_tcp_retries1; 249 extern int sysctl_tcp_retries2; 250 extern int sysctl_tcp_orphan_retries; 251 extern int sysctl_tcp_syncookies; 252 extern int sysctl_tcp_fastopen; 253 extern int sysctl_tcp_retrans_collapse; 254 extern int sysctl_tcp_stdurg; 255 extern int sysctl_tcp_rfc1337; 256 extern int sysctl_tcp_abort_on_overflow; 257 extern int sysctl_tcp_max_orphans; 258 extern int sysctl_tcp_fack; 259 extern int sysctl_tcp_reordering; 260 extern int sysctl_tcp_max_reordering; 261 extern int sysctl_tcp_dsack; 262 extern long sysctl_tcp_mem[3]; 263 extern int sysctl_tcp_wmem[3]; 264 extern int sysctl_tcp_rmem[3]; 265 extern int sysctl_tcp_app_win; 266 extern int sysctl_tcp_adv_win_scale; 267 extern int sysctl_tcp_tw_reuse; 268 extern int sysctl_tcp_frto; 269 extern int sysctl_tcp_low_latency; 270 extern int sysctl_tcp_nometrics_save; 271 extern int sysctl_tcp_moderate_rcvbuf; 272 extern int sysctl_tcp_tso_win_divisor; 273 extern int sysctl_tcp_workaround_signed_windows; 274 extern int sysctl_tcp_slow_start_after_idle; 275 extern int sysctl_tcp_thin_linear_timeouts; 276 extern int sysctl_tcp_thin_dupack; 277 extern int sysctl_tcp_early_retrans; 278 extern int sysctl_tcp_limit_output_bytes; 279 extern int sysctl_tcp_challenge_ack_limit; 280 extern unsigned int sysctl_tcp_notsent_lowat; 281 extern int sysctl_tcp_min_tso_segs; 282 extern int sysctl_tcp_min_rtt_wlen; 283 extern int sysctl_tcp_autocorking; 284 extern int sysctl_tcp_invalid_ratelimit; 285 extern int sysctl_tcp_pacing_ss_ratio; 286 extern int sysctl_tcp_pacing_ca_ratio; 287 288 extern atomic_long_t tcp_memory_allocated; 289 extern struct percpu_counter tcp_sockets_allocated; 290 extern int tcp_memory_pressure; 291 292 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 293 static inline bool tcp_under_memory_pressure(const struct sock *sk) 294 { 295 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 296 return !!sk->sk_cgrp->memory_pressure; 297 298 return tcp_memory_pressure; 299 } 300 /* 301 * The next routines deal with comparing 32 bit unsigned ints 302 * and worry about wraparound (automatic with unsigned arithmetic). 303 */ 304 305 static inline bool before(__u32 seq1, __u32 seq2) 306 { 307 return (__s32)(seq1-seq2) < 0; 308 } 309 #define after(seq2, seq1) before(seq1, seq2) 310 311 /* is s2<=s1<=s3 ? */ 312 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 313 { 314 return seq3 - seq2 >= seq1 - seq2; 315 } 316 317 static inline bool tcp_out_of_memory(struct sock *sk) 318 { 319 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 320 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 321 return true; 322 return false; 323 } 324 325 void sk_forced_mem_schedule(struct sock *sk, int size); 326 327 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 328 { 329 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 330 int orphans = percpu_counter_read_positive(ocp); 331 332 if (orphans << shift > sysctl_tcp_max_orphans) { 333 orphans = percpu_counter_sum_positive(ocp); 334 if (orphans << shift > sysctl_tcp_max_orphans) 335 return true; 336 } 337 return false; 338 } 339 340 bool tcp_check_oom(struct sock *sk, int shift); 341 342 343 extern struct proto tcp_prot; 344 345 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 346 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 347 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 348 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 349 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 350 351 void tcp_tasklet_init(void); 352 353 void tcp_v4_err(struct sk_buff *skb, u32); 354 355 void tcp_shutdown(struct sock *sk, int how); 356 357 void tcp_v4_early_demux(struct sk_buff *skb); 358 int tcp_v4_rcv(struct sk_buff *skb); 359 360 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 361 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 362 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 363 int flags); 364 void tcp_release_cb(struct sock *sk); 365 void tcp_wfree(struct sk_buff *skb); 366 void tcp_write_timer_handler(struct sock *sk); 367 void tcp_delack_timer_handler(struct sock *sk); 368 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 369 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 370 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 371 const struct tcphdr *th, unsigned int len); 372 void tcp_rcv_space_adjust(struct sock *sk); 373 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 374 void tcp_twsk_destructor(struct sock *sk); 375 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 376 struct pipe_inode_info *pipe, size_t len, 377 unsigned int flags); 378 379 static inline void tcp_dec_quickack_mode(struct sock *sk, 380 const unsigned int pkts) 381 { 382 struct inet_connection_sock *icsk = inet_csk(sk); 383 384 if (icsk->icsk_ack.quick) { 385 if (pkts >= icsk->icsk_ack.quick) { 386 icsk->icsk_ack.quick = 0; 387 /* Leaving quickack mode we deflate ATO. */ 388 icsk->icsk_ack.ato = TCP_ATO_MIN; 389 } else 390 icsk->icsk_ack.quick -= pkts; 391 } 392 } 393 394 #define TCP_ECN_OK 1 395 #define TCP_ECN_QUEUE_CWR 2 396 #define TCP_ECN_DEMAND_CWR 4 397 #define TCP_ECN_SEEN 8 398 399 enum tcp_tw_status { 400 TCP_TW_SUCCESS = 0, 401 TCP_TW_RST = 1, 402 TCP_TW_ACK = 2, 403 TCP_TW_SYN = 3 404 }; 405 406 407 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 408 struct sk_buff *skb, 409 const struct tcphdr *th); 410 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 411 struct request_sock *req, bool fastopen); 412 int tcp_child_process(struct sock *parent, struct sock *child, 413 struct sk_buff *skb); 414 void tcp_enter_loss(struct sock *sk); 415 void tcp_clear_retrans(struct tcp_sock *tp); 416 void tcp_update_metrics(struct sock *sk); 417 void tcp_init_metrics(struct sock *sk); 418 void tcp_metrics_init(void); 419 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, 420 bool paws_check, bool timestamps); 421 bool tcp_remember_stamp(struct sock *sk); 422 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); 423 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); 424 void tcp_disable_fack(struct tcp_sock *tp); 425 void tcp_close(struct sock *sk, long timeout); 426 void tcp_init_sock(struct sock *sk); 427 unsigned int tcp_poll(struct file *file, struct socket *sock, 428 struct poll_table_struct *wait); 429 int tcp_getsockopt(struct sock *sk, int level, int optname, 430 char __user *optval, int __user *optlen); 431 int tcp_setsockopt(struct sock *sk, int level, int optname, 432 char __user *optval, unsigned int optlen); 433 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 434 char __user *optval, int __user *optlen); 435 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 436 char __user *optval, unsigned int optlen); 437 void tcp_set_keepalive(struct sock *sk, int val); 438 void tcp_syn_ack_timeout(const struct request_sock *req); 439 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 440 int flags, int *addr_len); 441 void tcp_parse_options(const struct sk_buff *skb, 442 struct tcp_options_received *opt_rx, 443 int estab, struct tcp_fastopen_cookie *foc); 444 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 445 446 /* 447 * TCP v4 functions exported for the inet6 API 448 */ 449 450 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 451 void tcp_v4_mtu_reduced(struct sock *sk); 452 void tcp_req_err(struct sock *sk, u32 seq); 453 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 454 struct sock *tcp_create_openreq_child(const struct sock *sk, 455 struct request_sock *req, 456 struct sk_buff *skb); 457 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 458 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 459 struct request_sock *req, 460 struct dst_entry *dst, 461 struct request_sock *req_unhash, 462 bool *own_req); 463 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 464 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 465 int tcp_connect(struct sock *sk); 466 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 467 struct request_sock *req, 468 struct tcp_fastopen_cookie *foc, 469 bool attach_req); 470 int tcp_disconnect(struct sock *sk, int flags); 471 472 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 473 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 474 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 475 476 /* From syncookies.c */ 477 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 478 struct request_sock *req, 479 struct dst_entry *dst); 480 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 481 u32 cookie); 482 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 483 #ifdef CONFIG_SYN_COOKIES 484 485 /* Syncookies use a monotonic timer which increments every 60 seconds. 486 * This counter is used both as a hash input and partially encoded into 487 * the cookie value. A cookie is only validated further if the delta 488 * between the current counter value and the encoded one is less than this, 489 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 490 * the counter advances immediately after a cookie is generated). 491 */ 492 #define MAX_SYNCOOKIE_AGE 2 493 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 494 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 495 496 /* syncookies: remember time of last synqueue overflow 497 * But do not dirty this field too often (once per second is enough) 498 * It is racy as we do not hold a lock, but race is very minor. 499 */ 500 static inline void tcp_synq_overflow(const struct sock *sk) 501 { 502 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 503 unsigned long now = jiffies; 504 505 if (time_after(now, last_overflow + HZ)) 506 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 507 } 508 509 /* syncookies: no recent synqueue overflow on this listening socket? */ 510 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 511 { 512 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 513 514 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID); 515 } 516 517 static inline u32 tcp_cookie_time(void) 518 { 519 u64 val = get_jiffies_64(); 520 521 do_div(val, TCP_SYNCOOKIE_PERIOD); 522 return val; 523 } 524 525 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 526 u16 *mssp); 527 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 528 __u32 cookie_init_timestamp(struct request_sock *req); 529 bool cookie_timestamp_decode(struct tcp_options_received *opt); 530 bool cookie_ecn_ok(const struct tcp_options_received *opt, 531 const struct net *net, const struct dst_entry *dst); 532 533 /* From net/ipv6/syncookies.c */ 534 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 535 u32 cookie); 536 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 537 538 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 539 const struct tcphdr *th, u16 *mssp); 540 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 541 #endif 542 /* tcp_output.c */ 543 544 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 545 int nonagle); 546 bool tcp_may_send_now(struct sock *sk); 547 int __tcp_retransmit_skb(struct sock *, struct sk_buff *); 548 int tcp_retransmit_skb(struct sock *, struct sk_buff *); 549 void tcp_retransmit_timer(struct sock *sk); 550 void tcp_xmit_retransmit_queue(struct sock *); 551 void tcp_simple_retransmit(struct sock *); 552 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 553 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t); 554 555 void tcp_send_probe0(struct sock *); 556 void tcp_send_partial(struct sock *); 557 int tcp_write_wakeup(struct sock *, int mib); 558 void tcp_send_fin(struct sock *sk); 559 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 560 int tcp_send_synack(struct sock *); 561 void tcp_push_one(struct sock *, unsigned int mss_now); 562 void tcp_send_ack(struct sock *sk); 563 void tcp_send_delayed_ack(struct sock *sk); 564 void tcp_send_loss_probe(struct sock *sk); 565 bool tcp_schedule_loss_probe(struct sock *sk); 566 567 /* tcp_input.c */ 568 void tcp_resume_early_retransmit(struct sock *sk); 569 void tcp_rearm_rto(struct sock *sk); 570 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 571 void tcp_reset(struct sock *sk); 572 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 573 574 /* tcp_timer.c */ 575 void tcp_init_xmit_timers(struct sock *); 576 static inline void tcp_clear_xmit_timers(struct sock *sk) 577 { 578 inet_csk_clear_xmit_timers(sk); 579 } 580 581 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 582 unsigned int tcp_current_mss(struct sock *sk); 583 584 /* Bound MSS / TSO packet size with the half of the window */ 585 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 586 { 587 int cutoff; 588 589 /* When peer uses tiny windows, there is no use in packetizing 590 * to sub-MSS pieces for the sake of SWS or making sure there 591 * are enough packets in the pipe for fast recovery. 592 * 593 * On the other hand, for extremely large MSS devices, handling 594 * smaller than MSS windows in this way does make sense. 595 */ 596 if (tp->max_window >= 512) 597 cutoff = (tp->max_window >> 1); 598 else 599 cutoff = tp->max_window; 600 601 if (cutoff && pktsize > cutoff) 602 return max_t(int, cutoff, 68U - tp->tcp_header_len); 603 else 604 return pktsize; 605 } 606 607 /* tcp.c */ 608 void tcp_get_info(struct sock *, struct tcp_info *); 609 610 /* Read 'sendfile()'-style from a TCP socket */ 611 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 612 unsigned int, size_t); 613 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 614 sk_read_actor_t recv_actor); 615 616 void tcp_initialize_rcv_mss(struct sock *sk); 617 618 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 619 int tcp_mss_to_mtu(struct sock *sk, int mss); 620 void tcp_mtup_init(struct sock *sk); 621 void tcp_init_buffer_space(struct sock *sk); 622 623 static inline void tcp_bound_rto(const struct sock *sk) 624 { 625 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 626 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 627 } 628 629 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 630 { 631 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 632 } 633 634 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 635 { 636 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 637 ntohl(TCP_FLAG_ACK) | 638 snd_wnd); 639 } 640 641 static inline void tcp_fast_path_on(struct tcp_sock *tp) 642 { 643 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 644 } 645 646 static inline void tcp_fast_path_check(struct sock *sk) 647 { 648 struct tcp_sock *tp = tcp_sk(sk); 649 650 if (skb_queue_empty(&tp->out_of_order_queue) && 651 tp->rcv_wnd && 652 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 653 !tp->urg_data) 654 tcp_fast_path_on(tp); 655 } 656 657 /* Compute the actual rto_min value */ 658 static inline u32 tcp_rto_min(struct sock *sk) 659 { 660 const struct dst_entry *dst = __sk_dst_get(sk); 661 u32 rto_min = TCP_RTO_MIN; 662 663 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 664 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 665 return rto_min; 666 } 667 668 static inline u32 tcp_rto_min_us(struct sock *sk) 669 { 670 return jiffies_to_usecs(tcp_rto_min(sk)); 671 } 672 673 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 674 { 675 return dst_metric_locked(dst, RTAX_CC_ALGO); 676 } 677 678 /* Minimum RTT in usec. ~0 means not available. */ 679 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 680 { 681 return tp->rtt_min[0].rtt; 682 } 683 684 /* Compute the actual receive window we are currently advertising. 685 * Rcv_nxt can be after the window if our peer push more data 686 * than the offered window. 687 */ 688 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 689 { 690 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 691 692 if (win < 0) 693 win = 0; 694 return (u32) win; 695 } 696 697 /* Choose a new window, without checks for shrinking, and without 698 * scaling applied to the result. The caller does these things 699 * if necessary. This is a "raw" window selection. 700 */ 701 u32 __tcp_select_window(struct sock *sk); 702 703 void tcp_send_window_probe(struct sock *sk); 704 705 /* TCP timestamps are only 32-bits, this causes a slight 706 * complication on 64-bit systems since we store a snapshot 707 * of jiffies in the buffer control blocks below. We decided 708 * to use only the low 32-bits of jiffies and hide the ugly 709 * casts with the following macro. 710 */ 711 #define tcp_time_stamp ((__u32)(jiffies)) 712 713 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 714 { 715 return skb->skb_mstamp.stamp_jiffies; 716 } 717 718 719 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 720 721 #define TCPHDR_FIN 0x01 722 #define TCPHDR_SYN 0x02 723 #define TCPHDR_RST 0x04 724 #define TCPHDR_PSH 0x08 725 #define TCPHDR_ACK 0x10 726 #define TCPHDR_URG 0x20 727 #define TCPHDR_ECE 0x40 728 #define TCPHDR_CWR 0x80 729 730 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 731 732 /* This is what the send packet queuing engine uses to pass 733 * TCP per-packet control information to the transmission code. 734 * We also store the host-order sequence numbers in here too. 735 * This is 44 bytes if IPV6 is enabled. 736 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 737 */ 738 struct tcp_skb_cb { 739 __u32 seq; /* Starting sequence number */ 740 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 741 union { 742 /* Note : tcp_tw_isn is used in input path only 743 * (isn chosen by tcp_timewait_state_process()) 744 * 745 * tcp_gso_segs/size are used in write queue only, 746 * cf tcp_skb_pcount()/tcp_skb_mss() 747 */ 748 __u32 tcp_tw_isn; 749 struct { 750 u16 tcp_gso_segs; 751 u16 tcp_gso_size; 752 }; 753 }; 754 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 755 756 __u8 sacked; /* State flags for SACK/FACK. */ 757 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 758 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 759 #define TCPCB_LOST 0x04 /* SKB is lost */ 760 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 761 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 762 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 763 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 764 TCPCB_REPAIRED) 765 766 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 767 /* 1 byte hole */ 768 __u32 ack_seq; /* Sequence number ACK'd */ 769 union { 770 struct inet_skb_parm h4; 771 #if IS_ENABLED(CONFIG_IPV6) 772 struct inet6_skb_parm h6; 773 #endif 774 } header; /* For incoming frames */ 775 }; 776 777 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 778 779 780 #if IS_ENABLED(CONFIG_IPV6) 781 /* This is the variant of inet6_iif() that must be used by TCP, 782 * as TCP moves IP6CB into a different location in skb->cb[] 783 */ 784 static inline int tcp_v6_iif(const struct sk_buff *skb) 785 { 786 return TCP_SKB_CB(skb)->header.h6.iif; 787 } 788 #endif 789 790 /* Due to TSO, an SKB can be composed of multiple actual 791 * packets. To keep these tracked properly, we use this. 792 */ 793 static inline int tcp_skb_pcount(const struct sk_buff *skb) 794 { 795 return TCP_SKB_CB(skb)->tcp_gso_segs; 796 } 797 798 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 799 { 800 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 801 } 802 803 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 804 { 805 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 806 } 807 808 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 809 static inline int tcp_skb_mss(const struct sk_buff *skb) 810 { 811 return TCP_SKB_CB(skb)->tcp_gso_size; 812 } 813 814 /* Events passed to congestion control interface */ 815 enum tcp_ca_event { 816 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 817 CA_EVENT_CWND_RESTART, /* congestion window restart */ 818 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 819 CA_EVENT_LOSS, /* loss timeout */ 820 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 821 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 822 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 823 CA_EVENT_NON_DELAYED_ACK, 824 }; 825 826 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 827 enum tcp_ca_ack_event_flags { 828 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 829 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 830 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 831 }; 832 833 /* 834 * Interface for adding new TCP congestion control handlers 835 */ 836 #define TCP_CA_NAME_MAX 16 837 #define TCP_CA_MAX 128 838 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 839 840 #define TCP_CA_UNSPEC 0 841 842 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 843 #define TCP_CONG_NON_RESTRICTED 0x1 844 /* Requires ECN/ECT set on all packets */ 845 #define TCP_CONG_NEEDS_ECN 0x2 846 847 union tcp_cc_info; 848 849 struct tcp_congestion_ops { 850 struct list_head list; 851 u32 key; 852 u32 flags; 853 854 /* initialize private data (optional) */ 855 void (*init)(struct sock *sk); 856 /* cleanup private data (optional) */ 857 void (*release)(struct sock *sk); 858 859 /* return slow start threshold (required) */ 860 u32 (*ssthresh)(struct sock *sk); 861 /* do new cwnd calculation (required) */ 862 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 863 /* call before changing ca_state (optional) */ 864 void (*set_state)(struct sock *sk, u8 new_state); 865 /* call when cwnd event occurs (optional) */ 866 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 867 /* call when ack arrives (optional) */ 868 void (*in_ack_event)(struct sock *sk, u32 flags); 869 /* new value of cwnd after loss (optional) */ 870 u32 (*undo_cwnd)(struct sock *sk); 871 /* hook for packet ack accounting (optional) */ 872 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 873 /* get info for inet_diag (optional) */ 874 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 875 union tcp_cc_info *info); 876 877 char name[TCP_CA_NAME_MAX]; 878 struct module *owner; 879 }; 880 881 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 882 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 883 884 void tcp_assign_congestion_control(struct sock *sk); 885 void tcp_init_congestion_control(struct sock *sk); 886 void tcp_cleanup_congestion_control(struct sock *sk); 887 int tcp_set_default_congestion_control(const char *name); 888 void tcp_get_default_congestion_control(char *name); 889 void tcp_get_available_congestion_control(char *buf, size_t len); 890 void tcp_get_allowed_congestion_control(char *buf, size_t len); 891 int tcp_set_allowed_congestion_control(char *allowed); 892 int tcp_set_congestion_control(struct sock *sk, const char *name); 893 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 894 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 895 896 u32 tcp_reno_ssthresh(struct sock *sk); 897 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 898 extern struct tcp_congestion_ops tcp_reno; 899 900 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 901 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca); 902 #ifdef CONFIG_INET 903 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 904 #else 905 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 906 { 907 return NULL; 908 } 909 #endif 910 911 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 912 { 913 const struct inet_connection_sock *icsk = inet_csk(sk); 914 915 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 916 } 917 918 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 919 { 920 struct inet_connection_sock *icsk = inet_csk(sk); 921 922 if (icsk->icsk_ca_ops->set_state) 923 icsk->icsk_ca_ops->set_state(sk, ca_state); 924 icsk->icsk_ca_state = ca_state; 925 } 926 927 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 928 { 929 const struct inet_connection_sock *icsk = inet_csk(sk); 930 931 if (icsk->icsk_ca_ops->cwnd_event) 932 icsk->icsk_ca_ops->cwnd_event(sk, event); 933 } 934 935 /* These functions determine how the current flow behaves in respect of SACK 936 * handling. SACK is negotiated with the peer, and therefore it can vary 937 * between different flows. 938 * 939 * tcp_is_sack - SACK enabled 940 * tcp_is_reno - No SACK 941 * tcp_is_fack - FACK enabled, implies SACK enabled 942 */ 943 static inline int tcp_is_sack(const struct tcp_sock *tp) 944 { 945 return tp->rx_opt.sack_ok; 946 } 947 948 static inline bool tcp_is_reno(const struct tcp_sock *tp) 949 { 950 return !tcp_is_sack(tp); 951 } 952 953 static inline bool tcp_is_fack(const struct tcp_sock *tp) 954 { 955 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 956 } 957 958 static inline void tcp_enable_fack(struct tcp_sock *tp) 959 { 960 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 961 } 962 963 /* TCP early-retransmit (ER) is similar to but more conservative than 964 * the thin-dupack feature. Enable ER only if thin-dupack is disabled. 965 */ 966 static inline void tcp_enable_early_retrans(struct tcp_sock *tp) 967 { 968 tp->do_early_retrans = sysctl_tcp_early_retrans && 969 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack && 970 sysctl_tcp_reordering == 3; 971 } 972 973 static inline void tcp_disable_early_retrans(struct tcp_sock *tp) 974 { 975 tp->do_early_retrans = 0; 976 } 977 978 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 979 { 980 return tp->sacked_out + tp->lost_out; 981 } 982 983 /* This determines how many packets are "in the network" to the best 984 * of our knowledge. In many cases it is conservative, but where 985 * detailed information is available from the receiver (via SACK 986 * blocks etc.) we can make more aggressive calculations. 987 * 988 * Use this for decisions involving congestion control, use just 989 * tp->packets_out to determine if the send queue is empty or not. 990 * 991 * Read this equation as: 992 * 993 * "Packets sent once on transmission queue" MINUS 994 * "Packets left network, but not honestly ACKed yet" PLUS 995 * "Packets fast retransmitted" 996 */ 997 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 998 { 999 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1000 } 1001 1002 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1003 1004 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1005 { 1006 return tp->snd_cwnd < tp->snd_ssthresh; 1007 } 1008 1009 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1010 { 1011 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1012 } 1013 1014 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1015 { 1016 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1017 (1 << inet_csk(sk)->icsk_ca_state); 1018 } 1019 1020 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1021 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1022 * ssthresh. 1023 */ 1024 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1025 { 1026 const struct tcp_sock *tp = tcp_sk(sk); 1027 1028 if (tcp_in_cwnd_reduction(sk)) 1029 return tp->snd_ssthresh; 1030 else 1031 return max(tp->snd_ssthresh, 1032 ((tp->snd_cwnd >> 1) + 1033 (tp->snd_cwnd >> 2))); 1034 } 1035 1036 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1037 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1038 1039 void tcp_enter_cwr(struct sock *sk); 1040 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1041 1042 /* The maximum number of MSS of available cwnd for which TSO defers 1043 * sending if not using sysctl_tcp_tso_win_divisor. 1044 */ 1045 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1046 { 1047 return 3; 1048 } 1049 1050 /* Slow start with delack produces 3 packets of burst, so that 1051 * it is safe "de facto". This will be the default - same as 1052 * the default reordering threshold - but if reordering increases, 1053 * we must be able to allow cwnd to burst at least this much in order 1054 * to not pull it back when holes are filled. 1055 */ 1056 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 1057 { 1058 return tp->reordering; 1059 } 1060 1061 /* Returns end sequence number of the receiver's advertised window */ 1062 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1063 { 1064 return tp->snd_una + tp->snd_wnd; 1065 } 1066 1067 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1068 * flexible approach. The RFC suggests cwnd should not be raised unless 1069 * it was fully used previously. And that's exactly what we do in 1070 * congestion avoidance mode. But in slow start we allow cwnd to grow 1071 * as long as the application has used half the cwnd. 1072 * Example : 1073 * cwnd is 10 (IW10), but application sends 9 frames. 1074 * We allow cwnd to reach 18 when all frames are ACKed. 1075 * This check is safe because it's as aggressive as slow start which already 1076 * risks 100% overshoot. The advantage is that we discourage application to 1077 * either send more filler packets or data to artificially blow up the cwnd 1078 * usage, and allow application-limited process to probe bw more aggressively. 1079 */ 1080 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1081 { 1082 const struct tcp_sock *tp = tcp_sk(sk); 1083 1084 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1085 if (tcp_in_slow_start(tp)) 1086 return tp->snd_cwnd < 2 * tp->max_packets_out; 1087 1088 return tp->is_cwnd_limited; 1089 } 1090 1091 /* Something is really bad, we could not queue an additional packet, 1092 * because qdisc is full or receiver sent a 0 window. 1093 * We do not want to add fuel to the fire, or abort too early, 1094 * so make sure the timer we arm now is at least 200ms in the future, 1095 * regardless of current icsk_rto value (as it could be ~2ms) 1096 */ 1097 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1098 { 1099 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1100 } 1101 1102 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1103 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1104 unsigned long max_when) 1105 { 1106 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1107 1108 return (unsigned long)min_t(u64, when, max_when); 1109 } 1110 1111 static inline void tcp_check_probe_timer(struct sock *sk) 1112 { 1113 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1114 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1115 tcp_probe0_base(sk), TCP_RTO_MAX); 1116 } 1117 1118 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1119 { 1120 tp->snd_wl1 = seq; 1121 } 1122 1123 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1124 { 1125 tp->snd_wl1 = seq; 1126 } 1127 1128 /* 1129 * Calculate(/check) TCP checksum 1130 */ 1131 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1132 __be32 daddr, __wsum base) 1133 { 1134 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1135 } 1136 1137 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1138 { 1139 return __skb_checksum_complete(skb); 1140 } 1141 1142 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1143 { 1144 return !skb_csum_unnecessary(skb) && 1145 __tcp_checksum_complete(skb); 1146 } 1147 1148 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1149 1150 static inline void tcp_prequeue_init(struct tcp_sock *tp) 1151 { 1152 tp->ucopy.task = NULL; 1153 tp->ucopy.len = 0; 1154 tp->ucopy.memory = 0; 1155 skb_queue_head_init(&tp->ucopy.prequeue); 1156 } 1157 1158 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1159 1160 #undef STATE_TRACE 1161 1162 #ifdef STATE_TRACE 1163 static const char *statename[]={ 1164 "Unused","Established","Syn Sent","Syn Recv", 1165 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1166 "Close Wait","Last ACK","Listen","Closing" 1167 }; 1168 #endif 1169 void tcp_set_state(struct sock *sk, int state); 1170 1171 void tcp_done(struct sock *sk); 1172 1173 int tcp_abort(struct sock *sk, int err); 1174 1175 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1176 { 1177 rx_opt->dsack = 0; 1178 rx_opt->num_sacks = 0; 1179 } 1180 1181 u32 tcp_default_init_rwnd(u32 mss); 1182 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1183 1184 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1185 { 1186 struct tcp_sock *tp = tcp_sk(sk); 1187 s32 delta; 1188 1189 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out) 1190 return; 1191 delta = tcp_time_stamp - tp->lsndtime; 1192 if (delta > inet_csk(sk)->icsk_rto) 1193 tcp_cwnd_restart(sk, delta); 1194 } 1195 1196 /* Determine a window scaling and initial window to offer. */ 1197 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1198 __u32 *window_clamp, int wscale_ok, 1199 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1200 1201 static inline int tcp_win_from_space(int space) 1202 { 1203 return sysctl_tcp_adv_win_scale<=0 ? 1204 (space>>(-sysctl_tcp_adv_win_scale)) : 1205 space - (space>>sysctl_tcp_adv_win_scale); 1206 } 1207 1208 /* Note: caller must be prepared to deal with negative returns */ 1209 static inline int tcp_space(const struct sock *sk) 1210 { 1211 return tcp_win_from_space(sk->sk_rcvbuf - 1212 atomic_read(&sk->sk_rmem_alloc)); 1213 } 1214 1215 static inline int tcp_full_space(const struct sock *sk) 1216 { 1217 return tcp_win_from_space(sk->sk_rcvbuf); 1218 } 1219 1220 extern void tcp_openreq_init_rwin(struct request_sock *req, 1221 const struct sock *sk_listener, 1222 const struct dst_entry *dst); 1223 1224 void tcp_enter_memory_pressure(struct sock *sk); 1225 1226 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1227 { 1228 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1229 } 1230 1231 static inline int keepalive_time_when(const struct tcp_sock *tp) 1232 { 1233 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1234 } 1235 1236 static inline int keepalive_probes(const struct tcp_sock *tp) 1237 { 1238 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1239 } 1240 1241 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1242 { 1243 const struct inet_connection_sock *icsk = &tp->inet_conn; 1244 1245 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1246 tcp_time_stamp - tp->rcv_tstamp); 1247 } 1248 1249 static inline int tcp_fin_time(const struct sock *sk) 1250 { 1251 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1252 const int rto = inet_csk(sk)->icsk_rto; 1253 1254 if (fin_timeout < (rto << 2) - (rto >> 1)) 1255 fin_timeout = (rto << 2) - (rto >> 1); 1256 1257 return fin_timeout; 1258 } 1259 1260 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1261 int paws_win) 1262 { 1263 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1264 return true; 1265 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1266 return true; 1267 /* 1268 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1269 * then following tcp messages have valid values. Ignore 0 value, 1270 * or else 'negative' tsval might forbid us to accept their packets. 1271 */ 1272 if (!rx_opt->ts_recent) 1273 return true; 1274 return false; 1275 } 1276 1277 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1278 int rst) 1279 { 1280 if (tcp_paws_check(rx_opt, 0)) 1281 return false; 1282 1283 /* RST segments are not recommended to carry timestamp, 1284 and, if they do, it is recommended to ignore PAWS because 1285 "their cleanup function should take precedence over timestamps." 1286 Certainly, it is mistake. It is necessary to understand the reasons 1287 of this constraint to relax it: if peer reboots, clock may go 1288 out-of-sync and half-open connections will not be reset. 1289 Actually, the problem would be not existing if all 1290 the implementations followed draft about maintaining clock 1291 via reboots. Linux-2.2 DOES NOT! 1292 1293 However, we can relax time bounds for RST segments to MSL. 1294 */ 1295 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1296 return false; 1297 return true; 1298 } 1299 1300 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1301 int mib_idx, u32 *last_oow_ack_time); 1302 1303 static inline void tcp_mib_init(struct net *net) 1304 { 1305 /* See RFC 2012 */ 1306 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1307 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1308 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1309 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1310 } 1311 1312 /* from STCP */ 1313 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1314 { 1315 tp->lost_skb_hint = NULL; 1316 } 1317 1318 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1319 { 1320 tcp_clear_retrans_hints_partial(tp); 1321 tp->retransmit_skb_hint = NULL; 1322 } 1323 1324 /* MD5 Signature */ 1325 struct crypto_hash; 1326 1327 union tcp_md5_addr { 1328 struct in_addr a4; 1329 #if IS_ENABLED(CONFIG_IPV6) 1330 struct in6_addr a6; 1331 #endif 1332 }; 1333 1334 /* - key database */ 1335 struct tcp_md5sig_key { 1336 struct hlist_node node; 1337 u8 keylen; 1338 u8 family; /* AF_INET or AF_INET6 */ 1339 union tcp_md5_addr addr; 1340 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1341 struct rcu_head rcu; 1342 }; 1343 1344 /* - sock block */ 1345 struct tcp_md5sig_info { 1346 struct hlist_head head; 1347 struct rcu_head rcu; 1348 }; 1349 1350 /* - pseudo header */ 1351 struct tcp4_pseudohdr { 1352 __be32 saddr; 1353 __be32 daddr; 1354 __u8 pad; 1355 __u8 protocol; 1356 __be16 len; 1357 }; 1358 1359 struct tcp6_pseudohdr { 1360 struct in6_addr saddr; 1361 struct in6_addr daddr; 1362 __be32 len; 1363 __be32 protocol; /* including padding */ 1364 }; 1365 1366 union tcp_md5sum_block { 1367 struct tcp4_pseudohdr ip4; 1368 #if IS_ENABLED(CONFIG_IPV6) 1369 struct tcp6_pseudohdr ip6; 1370 #endif 1371 }; 1372 1373 /* - pool: digest algorithm, hash description and scratch buffer */ 1374 struct tcp_md5sig_pool { 1375 struct hash_desc md5_desc; 1376 union tcp_md5sum_block md5_blk; 1377 }; 1378 1379 /* - functions */ 1380 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1381 const struct sock *sk, const struct sk_buff *skb); 1382 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1383 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); 1384 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1385 int family); 1386 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1387 const struct sock *addr_sk); 1388 1389 #ifdef CONFIG_TCP_MD5SIG 1390 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1391 const union tcp_md5_addr *addr, 1392 int family); 1393 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1394 #else 1395 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1396 const union tcp_md5_addr *addr, 1397 int family) 1398 { 1399 return NULL; 1400 } 1401 #define tcp_twsk_md5_key(twsk) NULL 1402 #endif 1403 1404 bool tcp_alloc_md5sig_pool(void); 1405 1406 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1407 static inline void tcp_put_md5sig_pool(void) 1408 { 1409 local_bh_enable(); 1410 } 1411 1412 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1413 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1414 unsigned int header_len); 1415 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1416 const struct tcp_md5sig_key *key); 1417 1418 /* From tcp_fastopen.c */ 1419 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1420 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1421 unsigned long *last_syn_loss); 1422 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1423 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1424 u16 try_exp); 1425 struct tcp_fastopen_request { 1426 /* Fast Open cookie. Size 0 means a cookie request */ 1427 struct tcp_fastopen_cookie cookie; 1428 struct msghdr *data; /* data in MSG_FASTOPEN */ 1429 size_t size; 1430 int copied; /* queued in tcp_connect() */ 1431 }; 1432 void tcp_free_fastopen_req(struct tcp_sock *tp); 1433 1434 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1435 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1436 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1437 struct request_sock *req, 1438 struct tcp_fastopen_cookie *foc, 1439 struct dst_entry *dst); 1440 void tcp_fastopen_init_key_once(bool publish); 1441 #define TCP_FASTOPEN_KEY_LENGTH 16 1442 1443 /* Fastopen key context */ 1444 struct tcp_fastopen_context { 1445 struct crypto_cipher *tfm; 1446 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1447 struct rcu_head rcu; 1448 }; 1449 1450 /* write queue abstraction */ 1451 static inline void tcp_write_queue_purge(struct sock *sk) 1452 { 1453 struct sk_buff *skb; 1454 1455 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1456 sk_wmem_free_skb(sk, skb); 1457 sk_mem_reclaim(sk); 1458 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1459 } 1460 1461 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1462 { 1463 return skb_peek(&sk->sk_write_queue); 1464 } 1465 1466 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1467 { 1468 return skb_peek_tail(&sk->sk_write_queue); 1469 } 1470 1471 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1472 const struct sk_buff *skb) 1473 { 1474 return skb_queue_next(&sk->sk_write_queue, skb); 1475 } 1476 1477 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1478 const struct sk_buff *skb) 1479 { 1480 return skb_queue_prev(&sk->sk_write_queue, skb); 1481 } 1482 1483 #define tcp_for_write_queue(skb, sk) \ 1484 skb_queue_walk(&(sk)->sk_write_queue, skb) 1485 1486 #define tcp_for_write_queue_from(skb, sk) \ 1487 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1488 1489 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1490 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1491 1492 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1493 { 1494 return sk->sk_send_head; 1495 } 1496 1497 static inline bool tcp_skb_is_last(const struct sock *sk, 1498 const struct sk_buff *skb) 1499 { 1500 return skb_queue_is_last(&sk->sk_write_queue, skb); 1501 } 1502 1503 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1504 { 1505 if (tcp_skb_is_last(sk, skb)) 1506 sk->sk_send_head = NULL; 1507 else 1508 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1509 } 1510 1511 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1512 { 1513 if (sk->sk_send_head == skb_unlinked) 1514 sk->sk_send_head = NULL; 1515 } 1516 1517 static inline void tcp_init_send_head(struct sock *sk) 1518 { 1519 sk->sk_send_head = NULL; 1520 } 1521 1522 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1523 { 1524 __skb_queue_tail(&sk->sk_write_queue, skb); 1525 } 1526 1527 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1528 { 1529 __tcp_add_write_queue_tail(sk, skb); 1530 1531 /* Queue it, remembering where we must start sending. */ 1532 if (sk->sk_send_head == NULL) { 1533 sk->sk_send_head = skb; 1534 1535 if (tcp_sk(sk)->highest_sack == NULL) 1536 tcp_sk(sk)->highest_sack = skb; 1537 } 1538 } 1539 1540 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1541 { 1542 __skb_queue_head(&sk->sk_write_queue, skb); 1543 } 1544 1545 /* Insert buff after skb on the write queue of sk. */ 1546 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1547 struct sk_buff *buff, 1548 struct sock *sk) 1549 { 1550 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1551 } 1552 1553 /* Insert new before skb on the write queue of sk. */ 1554 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1555 struct sk_buff *skb, 1556 struct sock *sk) 1557 { 1558 __skb_queue_before(&sk->sk_write_queue, skb, new); 1559 1560 if (sk->sk_send_head == skb) 1561 sk->sk_send_head = new; 1562 } 1563 1564 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1565 { 1566 __skb_unlink(skb, &sk->sk_write_queue); 1567 } 1568 1569 static inline bool tcp_write_queue_empty(struct sock *sk) 1570 { 1571 return skb_queue_empty(&sk->sk_write_queue); 1572 } 1573 1574 static inline void tcp_push_pending_frames(struct sock *sk) 1575 { 1576 if (tcp_send_head(sk)) { 1577 struct tcp_sock *tp = tcp_sk(sk); 1578 1579 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1580 } 1581 } 1582 1583 /* Start sequence of the skb just after the highest skb with SACKed 1584 * bit, valid only if sacked_out > 0 or when the caller has ensured 1585 * validity by itself. 1586 */ 1587 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1588 { 1589 if (!tp->sacked_out) 1590 return tp->snd_una; 1591 1592 if (tp->highest_sack == NULL) 1593 return tp->snd_nxt; 1594 1595 return TCP_SKB_CB(tp->highest_sack)->seq; 1596 } 1597 1598 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1599 { 1600 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1601 tcp_write_queue_next(sk, skb); 1602 } 1603 1604 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1605 { 1606 return tcp_sk(sk)->highest_sack; 1607 } 1608 1609 static inline void tcp_highest_sack_reset(struct sock *sk) 1610 { 1611 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1612 } 1613 1614 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1615 static inline void tcp_highest_sack_combine(struct sock *sk, 1616 struct sk_buff *old, 1617 struct sk_buff *new) 1618 { 1619 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1620 tcp_sk(sk)->highest_sack = new; 1621 } 1622 1623 /* This helper checks if socket has IP_TRANSPARENT set */ 1624 static inline bool inet_sk_transparent(const struct sock *sk) 1625 { 1626 switch (sk->sk_state) { 1627 case TCP_TIME_WAIT: 1628 return inet_twsk(sk)->tw_transparent; 1629 case TCP_NEW_SYN_RECV: 1630 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1631 } 1632 return inet_sk(sk)->transparent; 1633 } 1634 1635 /* Determines whether this is a thin stream (which may suffer from 1636 * increased latency). Used to trigger latency-reducing mechanisms. 1637 */ 1638 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1639 { 1640 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1641 } 1642 1643 /* /proc */ 1644 enum tcp_seq_states { 1645 TCP_SEQ_STATE_LISTENING, 1646 TCP_SEQ_STATE_ESTABLISHED, 1647 }; 1648 1649 int tcp_seq_open(struct inode *inode, struct file *file); 1650 1651 struct tcp_seq_afinfo { 1652 char *name; 1653 sa_family_t family; 1654 const struct file_operations *seq_fops; 1655 struct seq_operations seq_ops; 1656 }; 1657 1658 struct tcp_iter_state { 1659 struct seq_net_private p; 1660 sa_family_t family; 1661 enum tcp_seq_states state; 1662 struct sock *syn_wait_sk; 1663 int bucket, offset, sbucket, num; 1664 loff_t last_pos; 1665 }; 1666 1667 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1668 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1669 1670 extern struct request_sock_ops tcp_request_sock_ops; 1671 extern struct request_sock_ops tcp6_request_sock_ops; 1672 1673 void tcp_v4_destroy_sock(struct sock *sk); 1674 1675 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1676 netdev_features_t features); 1677 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1678 int tcp_gro_complete(struct sk_buff *skb); 1679 1680 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1681 1682 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1683 { 1684 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat; 1685 } 1686 1687 static inline bool tcp_stream_memory_free(const struct sock *sk) 1688 { 1689 const struct tcp_sock *tp = tcp_sk(sk); 1690 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1691 1692 return notsent_bytes < tcp_notsent_lowat(tp); 1693 } 1694 1695 #ifdef CONFIG_PROC_FS 1696 int tcp4_proc_init(void); 1697 void tcp4_proc_exit(void); 1698 #endif 1699 1700 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1701 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1702 const struct tcp_request_sock_ops *af_ops, 1703 struct sock *sk, struct sk_buff *skb); 1704 1705 /* TCP af-specific functions */ 1706 struct tcp_sock_af_ops { 1707 #ifdef CONFIG_TCP_MD5SIG 1708 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1709 const struct sock *addr_sk); 1710 int (*calc_md5_hash)(char *location, 1711 const struct tcp_md5sig_key *md5, 1712 const struct sock *sk, 1713 const struct sk_buff *skb); 1714 int (*md5_parse)(struct sock *sk, 1715 char __user *optval, 1716 int optlen); 1717 #endif 1718 }; 1719 1720 struct tcp_request_sock_ops { 1721 u16 mss_clamp; 1722 #ifdef CONFIG_TCP_MD5SIG 1723 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1724 const struct sock *addr_sk); 1725 int (*calc_md5_hash) (char *location, 1726 const struct tcp_md5sig_key *md5, 1727 const struct sock *sk, 1728 const struct sk_buff *skb); 1729 #endif 1730 void (*init_req)(struct request_sock *req, 1731 const struct sock *sk_listener, 1732 struct sk_buff *skb); 1733 #ifdef CONFIG_SYN_COOKIES 1734 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1735 __u16 *mss); 1736 #endif 1737 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1738 const struct request_sock *req, 1739 bool *strict); 1740 __u32 (*init_seq)(const struct sk_buff *skb); 1741 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1742 struct flowi *fl, struct request_sock *req, 1743 struct tcp_fastopen_cookie *foc, 1744 bool attach_req); 1745 }; 1746 1747 #ifdef CONFIG_SYN_COOKIES 1748 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1749 const struct sock *sk, struct sk_buff *skb, 1750 __u16 *mss) 1751 { 1752 tcp_synq_overflow(sk); 1753 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1754 return ops->cookie_init_seq(skb, mss); 1755 } 1756 #else 1757 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1758 const struct sock *sk, struct sk_buff *skb, 1759 __u16 *mss) 1760 { 1761 return 0; 1762 } 1763 #endif 1764 1765 int tcpv4_offload_init(void); 1766 1767 void tcp_v4_init(void); 1768 void tcp_init(void); 1769 1770 /* tcp_recovery.c */ 1771 1772 /* Flags to enable various loss recovery features. See below */ 1773 extern int sysctl_tcp_recovery; 1774 1775 /* Use TCP RACK to detect (some) tail and retransmit losses */ 1776 #define TCP_RACK_LOST_RETRANS 0x1 1777 1778 extern int tcp_rack_mark_lost(struct sock *sk); 1779 1780 extern void tcp_rack_advance(struct tcp_sock *tp, 1781 const struct skb_mstamp *xmit_time, u8 sacked); 1782 1783 /* 1784 * Save and compile IPv4 options, return a pointer to it 1785 */ 1786 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb) 1787 { 1788 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1789 struct ip_options_rcu *dopt = NULL; 1790 1791 if (opt->optlen) { 1792 int opt_size = sizeof(*dopt) + opt->optlen; 1793 1794 dopt = kmalloc(opt_size, GFP_ATOMIC); 1795 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) { 1796 kfree(dopt); 1797 dopt = NULL; 1798 } 1799 } 1800 return dopt; 1801 } 1802 1803 /* locally generated TCP pure ACKs have skb->truesize == 2 1804 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1805 * This is much faster than dissecting the packet to find out. 1806 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1807 */ 1808 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1809 { 1810 return skb->truesize == 2; 1811 } 1812 1813 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1814 { 1815 skb->truesize = 2; 1816 } 1817 1818 #endif /* _TCP_H */ 1819