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 512 69 70 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 71 #define TCP_FASTRETRANS_THRESH 3 72 73 /* Maximal reordering. */ 74 #define TCP_MAX_REORDERING 127 75 76 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 77 #define TCP_MAX_QUICKACKS 16U 78 79 /* urg_data states */ 80 #define TCP_URG_VALID 0x0100 81 #define TCP_URG_NOTYET 0x0200 82 #define TCP_URG_READ 0x0400 83 84 #define TCP_RETR1 3 /* 85 * This is how many retries it does before it 86 * tries to figure out if the gateway is 87 * down. Minimal RFC value is 3; it corresponds 88 * to ~3sec-8min depending on RTO. 89 */ 90 91 #define TCP_RETR2 15 /* 92 * This should take at least 93 * 90 minutes to time out. 94 * RFC1122 says that the limit is 100 sec. 95 * 15 is ~13-30min depending on RTO. 96 */ 97 98 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 99 * when active opening a connection. 100 * RFC1122 says the minimum retry MUST 101 * be at least 180secs. Nevertheless 102 * this value is corresponding to 103 * 63secs of retransmission with the 104 * current initial RTO. 105 */ 106 107 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 108 * when passive opening a connection. 109 * This is corresponding to 31secs of 110 * retransmission with the current 111 * initial RTO. 112 */ 113 114 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 115 * state, about 60 seconds */ 116 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 117 /* BSD style FIN_WAIT2 deadlock breaker. 118 * It used to be 3min, new value is 60sec, 119 * to combine FIN-WAIT-2 timeout with 120 * TIME-WAIT timer. 121 */ 122 123 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 124 #if HZ >= 100 125 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 126 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 127 #else 128 #define TCP_DELACK_MIN 4U 129 #define TCP_ATO_MIN 4U 130 #endif 131 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 132 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 133 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 134 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 135 * used as a fallback RTO for the 136 * initial data transmission if no 137 * valid RTT sample has been acquired, 138 * most likely due to retrans in 3WHS. 139 */ 140 141 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 142 * for local resources. 143 */ 144 145 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 146 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 147 #define TCP_KEEPALIVE_INTVL (75*HZ) 148 149 #define MAX_TCP_KEEPIDLE 32767 150 #define MAX_TCP_KEEPINTVL 32767 151 #define MAX_TCP_KEEPCNT 127 152 #define MAX_TCP_SYNCNT 127 153 154 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 155 156 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 157 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 158 * after this time. It should be equal 159 * (or greater than) TCP_TIMEWAIT_LEN 160 * to provide reliability equal to one 161 * provided by timewait state. 162 */ 163 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 164 * timestamps. It must be less than 165 * minimal timewait lifetime. 166 */ 167 /* 168 * TCP option 169 */ 170 171 #define TCPOPT_NOP 1 /* Padding */ 172 #define TCPOPT_EOL 0 /* End of options */ 173 #define TCPOPT_MSS 2 /* Segment size negotiating */ 174 #define TCPOPT_WINDOW 3 /* Window scaling */ 175 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 176 #define TCPOPT_SACK 5 /* SACK Block */ 177 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 178 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 179 #define TCPOPT_EXP 254 /* Experimental */ 180 /* Magic number to be after the option value for sharing TCP 181 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 182 */ 183 #define TCPOPT_FASTOPEN_MAGIC 0xF989 184 185 /* 186 * TCP option lengths 187 */ 188 189 #define TCPOLEN_MSS 4 190 #define TCPOLEN_WINDOW 3 191 #define TCPOLEN_SACK_PERM 2 192 #define TCPOLEN_TIMESTAMP 10 193 #define TCPOLEN_MD5SIG 18 194 #define TCPOLEN_EXP_FASTOPEN_BASE 4 195 196 /* But this is what stacks really send out. */ 197 #define TCPOLEN_TSTAMP_ALIGNED 12 198 #define TCPOLEN_WSCALE_ALIGNED 4 199 #define TCPOLEN_SACKPERM_ALIGNED 4 200 #define TCPOLEN_SACK_BASE 2 201 #define TCPOLEN_SACK_BASE_ALIGNED 4 202 #define TCPOLEN_SACK_PERBLOCK 8 203 #define TCPOLEN_MD5SIG_ALIGNED 20 204 #define TCPOLEN_MSS_ALIGNED 4 205 206 /* Flags in tp->nonagle */ 207 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 208 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 209 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 210 211 /* TCP thin-stream limits */ 212 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 213 214 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */ 215 #define TCP_INIT_CWND 10 216 217 /* Bit Flags for sysctl_tcp_fastopen */ 218 #define TFO_CLIENT_ENABLE 1 219 #define TFO_SERVER_ENABLE 2 220 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 221 222 /* Accept SYN data w/o any cookie option */ 223 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 224 225 /* Force enable TFO on all listeners, i.e., not requiring the 226 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen. 227 */ 228 #define TFO_SERVER_WO_SOCKOPT1 0x400 229 #define TFO_SERVER_WO_SOCKOPT2 0x800 230 231 extern struct inet_timewait_death_row tcp_death_row; 232 233 /* sysctl variables for tcp */ 234 extern int sysctl_tcp_timestamps; 235 extern int sysctl_tcp_window_scaling; 236 extern int sysctl_tcp_sack; 237 extern int sysctl_tcp_fin_timeout; 238 extern int sysctl_tcp_keepalive_time; 239 extern int sysctl_tcp_keepalive_probes; 240 extern int sysctl_tcp_keepalive_intvl; 241 extern int sysctl_tcp_syn_retries; 242 extern int sysctl_tcp_synack_retries; 243 extern int sysctl_tcp_retries1; 244 extern int sysctl_tcp_retries2; 245 extern int sysctl_tcp_orphan_retries; 246 extern int sysctl_tcp_syncookies; 247 extern int sysctl_tcp_fastopen; 248 extern int sysctl_tcp_retrans_collapse; 249 extern int sysctl_tcp_stdurg; 250 extern int sysctl_tcp_rfc1337; 251 extern int sysctl_tcp_abort_on_overflow; 252 extern int sysctl_tcp_max_orphans; 253 extern int sysctl_tcp_fack; 254 extern int sysctl_tcp_reordering; 255 extern int sysctl_tcp_dsack; 256 extern long sysctl_tcp_mem[3]; 257 extern int sysctl_tcp_wmem[3]; 258 extern int sysctl_tcp_rmem[3]; 259 extern int sysctl_tcp_app_win; 260 extern int sysctl_tcp_adv_win_scale; 261 extern int sysctl_tcp_tw_reuse; 262 extern int sysctl_tcp_frto; 263 extern int sysctl_tcp_low_latency; 264 extern int sysctl_tcp_nometrics_save; 265 extern int sysctl_tcp_moderate_rcvbuf; 266 extern int sysctl_tcp_tso_win_divisor; 267 extern int sysctl_tcp_mtu_probing; 268 extern int sysctl_tcp_base_mss; 269 extern int sysctl_tcp_workaround_signed_windows; 270 extern int sysctl_tcp_slow_start_after_idle; 271 extern int sysctl_tcp_thin_linear_timeouts; 272 extern int sysctl_tcp_thin_dupack; 273 extern int sysctl_tcp_early_retrans; 274 extern int sysctl_tcp_limit_output_bytes; 275 extern int sysctl_tcp_challenge_ack_limit; 276 extern unsigned int sysctl_tcp_notsent_lowat; 277 extern int sysctl_tcp_min_tso_segs; 278 extern int sysctl_tcp_autocorking; 279 280 extern atomic_long_t tcp_memory_allocated; 281 extern struct percpu_counter tcp_sockets_allocated; 282 extern int tcp_memory_pressure; 283 284 /* 285 * The next routines deal with comparing 32 bit unsigned ints 286 * and worry about wraparound (automatic with unsigned arithmetic). 287 */ 288 289 static inline bool before(__u32 seq1, __u32 seq2) 290 { 291 return (__s32)(seq1-seq2) < 0; 292 } 293 #define after(seq2, seq1) before(seq1, seq2) 294 295 /* is s2<=s1<=s3 ? */ 296 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 297 { 298 return seq3 - seq2 >= seq1 - seq2; 299 } 300 301 static inline bool tcp_out_of_memory(struct sock *sk) 302 { 303 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 304 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 305 return true; 306 return false; 307 } 308 309 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 310 { 311 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 312 int orphans = percpu_counter_read_positive(ocp); 313 314 if (orphans << shift > sysctl_tcp_max_orphans) { 315 orphans = percpu_counter_sum_positive(ocp); 316 if (orphans << shift > sysctl_tcp_max_orphans) 317 return true; 318 } 319 return false; 320 } 321 322 bool tcp_check_oom(struct sock *sk, int shift); 323 324 /* syncookies: remember time of last synqueue overflow */ 325 static inline void tcp_synq_overflow(struct sock *sk) 326 { 327 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies; 328 } 329 330 /* syncookies: no recent synqueue overflow on this listening socket? */ 331 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 332 { 333 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 334 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK); 335 } 336 337 extern struct proto tcp_prot; 338 339 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 340 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 341 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 342 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 343 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 344 345 void tcp_tasklet_init(void); 346 347 void tcp_v4_err(struct sk_buff *skb, u32); 348 349 void tcp_shutdown(struct sock *sk, int how); 350 351 void tcp_v4_early_demux(struct sk_buff *skb); 352 int tcp_v4_rcv(struct sk_buff *skb); 353 354 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 355 int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 356 size_t size); 357 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 358 int flags); 359 void tcp_release_cb(struct sock *sk); 360 void tcp_wfree(struct sk_buff *skb); 361 void tcp_write_timer_handler(struct sock *sk); 362 void tcp_delack_timer_handler(struct sock *sk); 363 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 364 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 365 const struct tcphdr *th, unsigned int len); 366 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 367 const struct tcphdr *th, unsigned int len); 368 void tcp_rcv_space_adjust(struct sock *sk); 369 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 370 void tcp_twsk_destructor(struct sock *sk); 371 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 372 struct pipe_inode_info *pipe, size_t len, 373 unsigned int flags); 374 375 static inline void tcp_dec_quickack_mode(struct sock *sk, 376 const unsigned int pkts) 377 { 378 struct inet_connection_sock *icsk = inet_csk(sk); 379 380 if (icsk->icsk_ack.quick) { 381 if (pkts >= icsk->icsk_ack.quick) { 382 icsk->icsk_ack.quick = 0; 383 /* Leaving quickack mode we deflate ATO. */ 384 icsk->icsk_ack.ato = TCP_ATO_MIN; 385 } else 386 icsk->icsk_ack.quick -= pkts; 387 } 388 } 389 390 #define TCP_ECN_OK 1 391 #define TCP_ECN_QUEUE_CWR 2 392 #define TCP_ECN_DEMAND_CWR 4 393 #define TCP_ECN_SEEN 8 394 395 enum tcp_tw_status { 396 TCP_TW_SUCCESS = 0, 397 TCP_TW_RST = 1, 398 TCP_TW_ACK = 2, 399 TCP_TW_SYN = 3 400 }; 401 402 403 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 404 struct sk_buff *skb, 405 const struct tcphdr *th); 406 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 407 struct request_sock *req, struct request_sock **prev, 408 bool fastopen); 409 int tcp_child_process(struct sock *parent, struct sock *child, 410 struct sk_buff *skb); 411 void tcp_enter_loss(struct sock *sk); 412 void tcp_clear_retrans(struct tcp_sock *tp); 413 void tcp_update_metrics(struct sock *sk); 414 void tcp_init_metrics(struct sock *sk); 415 void tcp_metrics_init(void); 416 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, 417 bool paws_check, bool timestamps); 418 bool tcp_remember_stamp(struct sock *sk); 419 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); 420 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); 421 void tcp_disable_fack(struct tcp_sock *tp); 422 void tcp_close(struct sock *sk, long timeout); 423 void tcp_init_sock(struct sock *sk); 424 unsigned int tcp_poll(struct file *file, struct socket *sock, 425 struct poll_table_struct *wait); 426 int tcp_getsockopt(struct sock *sk, int level, int optname, 427 char __user *optval, int __user *optlen); 428 int tcp_setsockopt(struct sock *sk, int level, int optname, 429 char __user *optval, unsigned int optlen); 430 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 431 char __user *optval, int __user *optlen); 432 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 433 char __user *optval, unsigned int optlen); 434 void tcp_set_keepalive(struct sock *sk, int val); 435 void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req); 436 int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 437 size_t len, int nonblock, int flags, int *addr_len); 438 void tcp_parse_options(const struct sk_buff *skb, 439 struct tcp_options_received *opt_rx, 440 int estab, struct tcp_fastopen_cookie *foc); 441 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 442 443 /* 444 * TCP v4 functions exported for the inet6 API 445 */ 446 447 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 448 void tcp_v4_mtu_reduced(struct sock *sk); 449 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 450 struct sock *tcp_create_openreq_child(struct sock *sk, 451 struct request_sock *req, 452 struct sk_buff *skb); 453 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 454 struct request_sock *req, 455 struct dst_entry *dst); 456 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 457 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 458 int tcp_connect(struct sock *sk); 459 struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst, 460 struct request_sock *req, 461 struct tcp_fastopen_cookie *foc); 462 int tcp_disconnect(struct sock *sk, int flags); 463 464 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 465 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 466 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 467 468 /* From syncookies.c */ 469 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 470 u32 cookie); 471 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 472 struct ip_options *opt); 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 2 483 484 static inline u32 tcp_cookie_time(void) 485 { 486 u64 val = get_jiffies_64(); 487 488 do_div(val, 60 * HZ); 489 return val; 490 } 491 492 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 493 u16 *mssp); 494 __u32 cookie_v4_init_sequence(struct sock *sk, const struct sk_buff *skb, 495 __u16 *mss); 496 #endif 497 498 __u32 cookie_init_timestamp(struct request_sock *req); 499 bool cookie_check_timestamp(struct tcp_options_received *opt, struct net *net, 500 bool *ecn_ok); 501 502 /* From net/ipv6/syncookies.c */ 503 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 504 u32 cookie); 505 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 506 #ifdef CONFIG_SYN_COOKIES 507 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 508 const struct tcphdr *th, u16 *mssp); 509 __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb, 510 __u16 *mss); 511 #endif 512 /* tcp_output.c */ 513 514 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 515 int nonagle); 516 bool tcp_may_send_now(struct sock *sk); 517 int __tcp_retransmit_skb(struct sock *, struct sk_buff *); 518 int tcp_retransmit_skb(struct sock *, struct sk_buff *); 519 void tcp_retransmit_timer(struct sock *sk); 520 void tcp_xmit_retransmit_queue(struct sock *); 521 void tcp_simple_retransmit(struct sock *); 522 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 523 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t); 524 525 void tcp_send_probe0(struct sock *); 526 void tcp_send_partial(struct sock *); 527 int tcp_write_wakeup(struct sock *); 528 void tcp_send_fin(struct sock *sk); 529 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 530 int tcp_send_synack(struct sock *); 531 bool tcp_syn_flood_action(struct sock *sk, const struct sk_buff *skb, 532 const char *proto); 533 void tcp_push_one(struct sock *, unsigned int mss_now); 534 void tcp_send_ack(struct sock *sk); 535 void tcp_send_delayed_ack(struct sock *sk); 536 void tcp_send_loss_probe(struct sock *sk); 537 bool tcp_schedule_loss_probe(struct sock *sk); 538 539 /* tcp_input.c */ 540 void tcp_resume_early_retransmit(struct sock *sk); 541 void tcp_rearm_rto(struct sock *sk); 542 void tcp_reset(struct sock *sk); 543 544 /* tcp_timer.c */ 545 void tcp_init_xmit_timers(struct sock *); 546 static inline void tcp_clear_xmit_timers(struct sock *sk) 547 { 548 inet_csk_clear_xmit_timers(sk); 549 } 550 551 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 552 unsigned int tcp_current_mss(struct sock *sk); 553 554 /* Bound MSS / TSO packet size with the half of the window */ 555 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 556 { 557 int cutoff; 558 559 /* When peer uses tiny windows, there is no use in packetizing 560 * to sub-MSS pieces for the sake of SWS or making sure there 561 * are enough packets in the pipe for fast recovery. 562 * 563 * On the other hand, for extremely large MSS devices, handling 564 * smaller than MSS windows in this way does make sense. 565 */ 566 if (tp->max_window >= 512) 567 cutoff = (tp->max_window >> 1); 568 else 569 cutoff = tp->max_window; 570 571 if (cutoff && pktsize > cutoff) 572 return max_t(int, cutoff, 68U - tp->tcp_header_len); 573 else 574 return pktsize; 575 } 576 577 /* tcp.c */ 578 void tcp_get_info(const struct sock *, struct tcp_info *); 579 580 /* Read 'sendfile()'-style from a TCP socket */ 581 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 582 unsigned int, size_t); 583 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 584 sk_read_actor_t recv_actor); 585 586 void tcp_initialize_rcv_mss(struct sock *sk); 587 588 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 589 int tcp_mss_to_mtu(struct sock *sk, int mss); 590 void tcp_mtup_init(struct sock *sk); 591 void tcp_init_buffer_space(struct sock *sk); 592 593 static inline void tcp_bound_rto(const struct sock *sk) 594 { 595 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 596 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 597 } 598 599 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 600 { 601 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 602 } 603 604 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 605 { 606 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 607 ntohl(TCP_FLAG_ACK) | 608 snd_wnd); 609 } 610 611 static inline void tcp_fast_path_on(struct tcp_sock *tp) 612 { 613 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 614 } 615 616 static inline void tcp_fast_path_check(struct sock *sk) 617 { 618 struct tcp_sock *tp = tcp_sk(sk); 619 620 if (skb_queue_empty(&tp->out_of_order_queue) && 621 tp->rcv_wnd && 622 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 623 !tp->urg_data) 624 tcp_fast_path_on(tp); 625 } 626 627 /* Compute the actual rto_min value */ 628 static inline u32 tcp_rto_min(struct sock *sk) 629 { 630 const struct dst_entry *dst = __sk_dst_get(sk); 631 u32 rto_min = TCP_RTO_MIN; 632 633 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 634 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 635 return rto_min; 636 } 637 638 static inline u32 tcp_rto_min_us(struct sock *sk) 639 { 640 return jiffies_to_usecs(tcp_rto_min(sk)); 641 } 642 643 /* Compute the actual receive window we are currently advertising. 644 * Rcv_nxt can be after the window if our peer push more data 645 * than the offered window. 646 */ 647 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 648 { 649 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 650 651 if (win < 0) 652 win = 0; 653 return (u32) win; 654 } 655 656 /* Choose a new window, without checks for shrinking, and without 657 * scaling applied to the result. The caller does these things 658 * if necessary. This is a "raw" window selection. 659 */ 660 u32 __tcp_select_window(struct sock *sk); 661 662 void tcp_send_window_probe(struct sock *sk); 663 664 /* TCP timestamps are only 32-bits, this causes a slight 665 * complication on 64-bit systems since we store a snapshot 666 * of jiffies in the buffer control blocks below. We decided 667 * to use only the low 32-bits of jiffies and hide the ugly 668 * casts with the following macro. 669 */ 670 #define tcp_time_stamp ((__u32)(jiffies)) 671 672 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 673 { 674 return skb->skb_mstamp.stamp_jiffies; 675 } 676 677 678 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 679 680 #define TCPHDR_FIN 0x01 681 #define TCPHDR_SYN 0x02 682 #define TCPHDR_RST 0x04 683 #define TCPHDR_PSH 0x08 684 #define TCPHDR_ACK 0x10 685 #define TCPHDR_URG 0x20 686 #define TCPHDR_ECE 0x40 687 #define TCPHDR_CWR 0x80 688 689 /* This is what the send packet queuing engine uses to pass 690 * TCP per-packet control information to the transmission code. 691 * We also store the host-order sequence numbers in here too. 692 * This is 44 bytes if IPV6 is enabled. 693 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 694 */ 695 struct tcp_skb_cb { 696 __u32 seq; /* Starting sequence number */ 697 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 698 union { 699 /* Note : tcp_tw_isn is used in input path only 700 * (isn chosen by tcp_timewait_state_process()) 701 * 702 * tcp_gso_segs is used in write queue only, 703 * cf tcp_skb_pcount() 704 */ 705 __u32 tcp_tw_isn; 706 __u32 tcp_gso_segs; 707 }; 708 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 709 710 __u8 sacked; /* State flags for SACK/FACK. */ 711 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 712 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 713 #define TCPCB_LOST 0x04 /* SKB is lost */ 714 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 715 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 716 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 717 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 718 TCPCB_REPAIRED) 719 720 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 721 /* 1 byte hole */ 722 __u32 ack_seq; /* Sequence number ACK'd */ 723 union { 724 struct inet_skb_parm h4; 725 #if IS_ENABLED(CONFIG_IPV6) 726 struct inet6_skb_parm h6; 727 #endif 728 } header; /* For incoming frames */ 729 }; 730 731 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 732 733 /* Due to TSO, an SKB can be composed of multiple actual 734 * packets. To keep these tracked properly, we use this. 735 */ 736 static inline int tcp_skb_pcount(const struct sk_buff *skb) 737 { 738 return TCP_SKB_CB(skb)->tcp_gso_segs; 739 } 740 741 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 742 { 743 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 744 } 745 746 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 747 { 748 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 749 } 750 751 /* This is valid iff tcp_skb_pcount() > 1. */ 752 static inline int tcp_skb_mss(const struct sk_buff *skb) 753 { 754 return skb_shinfo(skb)->gso_size; 755 } 756 757 /* Events passed to congestion control interface */ 758 enum tcp_ca_event { 759 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 760 CA_EVENT_CWND_RESTART, /* congestion window restart */ 761 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 762 CA_EVENT_LOSS, /* loss timeout */ 763 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 764 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 765 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 766 CA_EVENT_NON_DELAYED_ACK, 767 }; 768 769 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 770 enum tcp_ca_ack_event_flags { 771 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 772 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 773 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 774 }; 775 776 /* 777 * Interface for adding new TCP congestion control handlers 778 */ 779 #define TCP_CA_NAME_MAX 16 780 #define TCP_CA_MAX 128 781 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 782 783 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 784 #define TCP_CONG_NON_RESTRICTED 0x1 785 /* Requires ECN/ECT set on all packets */ 786 #define TCP_CONG_NEEDS_ECN 0x2 787 788 struct tcp_congestion_ops { 789 struct list_head list; 790 unsigned long flags; 791 792 /* initialize private data (optional) */ 793 void (*init)(struct sock *sk); 794 /* cleanup private data (optional) */ 795 void (*release)(struct sock *sk); 796 797 /* return slow start threshold (required) */ 798 u32 (*ssthresh)(struct sock *sk); 799 /* do new cwnd calculation (required) */ 800 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 801 /* call before changing ca_state (optional) */ 802 void (*set_state)(struct sock *sk, u8 new_state); 803 /* call when cwnd event occurs (optional) */ 804 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 805 /* call when ack arrives (optional) */ 806 void (*in_ack_event)(struct sock *sk, u32 flags); 807 /* new value of cwnd after loss (optional) */ 808 u32 (*undo_cwnd)(struct sock *sk); 809 /* hook for packet ack accounting (optional) */ 810 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 811 /* get info for inet_diag (optional) */ 812 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 813 814 char name[TCP_CA_NAME_MAX]; 815 struct module *owner; 816 }; 817 818 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 819 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 820 821 void tcp_assign_congestion_control(struct sock *sk); 822 void tcp_init_congestion_control(struct sock *sk); 823 void tcp_cleanup_congestion_control(struct sock *sk); 824 int tcp_set_default_congestion_control(const char *name); 825 void tcp_get_default_congestion_control(char *name); 826 void tcp_get_available_congestion_control(char *buf, size_t len); 827 void tcp_get_allowed_congestion_control(char *buf, size_t len); 828 int tcp_set_allowed_congestion_control(char *allowed); 829 int tcp_set_congestion_control(struct sock *sk, const char *name); 830 void tcp_slow_start(struct tcp_sock *tp, u32 acked); 831 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 832 833 u32 tcp_reno_ssthresh(struct sock *sk); 834 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 835 extern struct tcp_congestion_ops tcp_reno; 836 837 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 838 { 839 const struct inet_connection_sock *icsk = inet_csk(sk); 840 841 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 842 } 843 844 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 845 { 846 struct inet_connection_sock *icsk = inet_csk(sk); 847 848 if (icsk->icsk_ca_ops->set_state) 849 icsk->icsk_ca_ops->set_state(sk, ca_state); 850 icsk->icsk_ca_state = ca_state; 851 } 852 853 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 854 { 855 const struct inet_connection_sock *icsk = inet_csk(sk); 856 857 if (icsk->icsk_ca_ops->cwnd_event) 858 icsk->icsk_ca_ops->cwnd_event(sk, event); 859 } 860 861 /* These functions determine how the current flow behaves in respect of SACK 862 * handling. SACK is negotiated with the peer, and therefore it can vary 863 * between different flows. 864 * 865 * tcp_is_sack - SACK enabled 866 * tcp_is_reno - No SACK 867 * tcp_is_fack - FACK enabled, implies SACK enabled 868 */ 869 static inline int tcp_is_sack(const struct tcp_sock *tp) 870 { 871 return tp->rx_opt.sack_ok; 872 } 873 874 static inline bool tcp_is_reno(const struct tcp_sock *tp) 875 { 876 return !tcp_is_sack(tp); 877 } 878 879 static inline bool tcp_is_fack(const struct tcp_sock *tp) 880 { 881 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 882 } 883 884 static inline void tcp_enable_fack(struct tcp_sock *tp) 885 { 886 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 887 } 888 889 /* TCP early-retransmit (ER) is similar to but more conservative than 890 * the thin-dupack feature. Enable ER only if thin-dupack is disabled. 891 */ 892 static inline void tcp_enable_early_retrans(struct tcp_sock *tp) 893 { 894 tp->do_early_retrans = sysctl_tcp_early_retrans && 895 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack && 896 sysctl_tcp_reordering == 3; 897 } 898 899 static inline void tcp_disable_early_retrans(struct tcp_sock *tp) 900 { 901 tp->do_early_retrans = 0; 902 } 903 904 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 905 { 906 return tp->sacked_out + tp->lost_out; 907 } 908 909 /* This determines how many packets are "in the network" to the best 910 * of our knowledge. In many cases it is conservative, but where 911 * detailed information is available from the receiver (via SACK 912 * blocks etc.) we can make more aggressive calculations. 913 * 914 * Use this for decisions involving congestion control, use just 915 * tp->packets_out to determine if the send queue is empty or not. 916 * 917 * Read this equation as: 918 * 919 * "Packets sent once on transmission queue" MINUS 920 * "Packets left network, but not honestly ACKed yet" PLUS 921 * "Packets fast retransmitted" 922 */ 923 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 924 { 925 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 926 } 927 928 #define TCP_INFINITE_SSTHRESH 0x7fffffff 929 930 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 931 { 932 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 933 } 934 935 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 936 { 937 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 938 (1 << inet_csk(sk)->icsk_ca_state); 939 } 940 941 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 942 * The exception is cwnd reduction phase, when cwnd is decreasing towards 943 * ssthresh. 944 */ 945 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 946 { 947 const struct tcp_sock *tp = tcp_sk(sk); 948 949 if (tcp_in_cwnd_reduction(sk)) 950 return tp->snd_ssthresh; 951 else 952 return max(tp->snd_ssthresh, 953 ((tp->snd_cwnd >> 1) + 954 (tp->snd_cwnd >> 2))); 955 } 956 957 /* Use define here intentionally to get WARN_ON location shown at the caller */ 958 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 959 960 void tcp_enter_cwr(struct sock *sk); 961 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 962 963 /* The maximum number of MSS of available cwnd for which TSO defers 964 * sending if not using sysctl_tcp_tso_win_divisor. 965 */ 966 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 967 { 968 return 3; 969 } 970 971 /* Slow start with delack produces 3 packets of burst, so that 972 * it is safe "de facto". This will be the default - same as 973 * the default reordering threshold - but if reordering increases, 974 * we must be able to allow cwnd to burst at least this much in order 975 * to not pull it back when holes are filled. 976 */ 977 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 978 { 979 return tp->reordering; 980 } 981 982 /* Returns end sequence number of the receiver's advertised window */ 983 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 984 { 985 return tp->snd_una + tp->snd_wnd; 986 } 987 988 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 989 * flexible approach. The RFC suggests cwnd should not be raised unless 990 * it was fully used previously. And that's exactly what we do in 991 * congestion avoidance mode. But in slow start we allow cwnd to grow 992 * as long as the application has used half the cwnd. 993 * Example : 994 * cwnd is 10 (IW10), but application sends 9 frames. 995 * We allow cwnd to reach 18 when all frames are ACKed. 996 * This check is safe because it's as aggressive as slow start which already 997 * risks 100% overshoot. The advantage is that we discourage application to 998 * either send more filler packets or data to artificially blow up the cwnd 999 * usage, and allow application-limited process to probe bw more aggressively. 1000 */ 1001 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1002 { 1003 const struct tcp_sock *tp = tcp_sk(sk); 1004 1005 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1006 if (tp->snd_cwnd <= tp->snd_ssthresh) 1007 return tp->snd_cwnd < 2 * tp->max_packets_out; 1008 1009 return tp->is_cwnd_limited; 1010 } 1011 1012 static inline void tcp_check_probe_timer(struct sock *sk) 1013 { 1014 const struct tcp_sock *tp = tcp_sk(sk); 1015 const struct inet_connection_sock *icsk = inet_csk(sk); 1016 1017 if (!tp->packets_out && !icsk->icsk_pending) 1018 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1019 icsk->icsk_rto, TCP_RTO_MAX); 1020 } 1021 1022 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1023 { 1024 tp->snd_wl1 = seq; 1025 } 1026 1027 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1028 { 1029 tp->snd_wl1 = seq; 1030 } 1031 1032 /* 1033 * Calculate(/check) TCP checksum 1034 */ 1035 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1036 __be32 daddr, __wsum base) 1037 { 1038 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1039 } 1040 1041 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1042 { 1043 return __skb_checksum_complete(skb); 1044 } 1045 1046 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1047 { 1048 return !skb_csum_unnecessary(skb) && 1049 __tcp_checksum_complete(skb); 1050 } 1051 1052 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1053 1054 static inline void tcp_prequeue_init(struct tcp_sock *tp) 1055 { 1056 tp->ucopy.task = NULL; 1057 tp->ucopy.len = 0; 1058 tp->ucopy.memory = 0; 1059 skb_queue_head_init(&tp->ucopy.prequeue); 1060 } 1061 1062 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1063 1064 #undef STATE_TRACE 1065 1066 #ifdef STATE_TRACE 1067 static const char *statename[]={ 1068 "Unused","Established","Syn Sent","Syn Recv", 1069 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1070 "Close Wait","Last ACK","Listen","Closing" 1071 }; 1072 #endif 1073 void tcp_set_state(struct sock *sk, int state); 1074 1075 void tcp_done(struct sock *sk); 1076 1077 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1078 { 1079 rx_opt->dsack = 0; 1080 rx_opt->num_sacks = 0; 1081 } 1082 1083 u32 tcp_default_init_rwnd(u32 mss); 1084 1085 /* Determine a window scaling and initial window to offer. */ 1086 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1087 __u32 *window_clamp, int wscale_ok, 1088 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1089 1090 static inline int tcp_win_from_space(int space) 1091 { 1092 return sysctl_tcp_adv_win_scale<=0 ? 1093 (space>>(-sysctl_tcp_adv_win_scale)) : 1094 space - (space>>sysctl_tcp_adv_win_scale); 1095 } 1096 1097 /* Note: caller must be prepared to deal with negative returns */ 1098 static inline int tcp_space(const struct sock *sk) 1099 { 1100 return tcp_win_from_space(sk->sk_rcvbuf - 1101 atomic_read(&sk->sk_rmem_alloc)); 1102 } 1103 1104 static inline int tcp_full_space(const struct sock *sk) 1105 { 1106 return tcp_win_from_space(sk->sk_rcvbuf); 1107 } 1108 1109 static inline void tcp_openreq_init(struct request_sock *req, 1110 struct tcp_options_received *rx_opt, 1111 struct sk_buff *skb, struct sock *sk) 1112 { 1113 struct inet_request_sock *ireq = inet_rsk(req); 1114 1115 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 1116 req->cookie_ts = 0; 1117 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 1118 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1; 1119 tcp_rsk(req)->snt_synack = tcp_time_stamp; 1120 req->mss = rx_opt->mss_clamp; 1121 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 1122 ireq->tstamp_ok = rx_opt->tstamp_ok; 1123 ireq->sack_ok = rx_opt->sack_ok; 1124 ireq->snd_wscale = rx_opt->snd_wscale; 1125 ireq->wscale_ok = rx_opt->wscale_ok; 1126 ireq->acked = 0; 1127 ireq->ecn_ok = 0; 1128 ireq->ir_rmt_port = tcp_hdr(skb)->source; 1129 ireq->ir_num = ntohs(tcp_hdr(skb)->dest); 1130 ireq->ir_mark = inet_request_mark(sk, skb); 1131 } 1132 1133 extern void tcp_openreq_init_rwin(struct request_sock *req, 1134 struct sock *sk, struct dst_entry *dst); 1135 1136 void tcp_enter_memory_pressure(struct sock *sk); 1137 1138 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1139 { 1140 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1141 } 1142 1143 static inline int keepalive_time_when(const struct tcp_sock *tp) 1144 { 1145 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1146 } 1147 1148 static inline int keepalive_probes(const struct tcp_sock *tp) 1149 { 1150 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1151 } 1152 1153 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1154 { 1155 const struct inet_connection_sock *icsk = &tp->inet_conn; 1156 1157 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1158 tcp_time_stamp - tp->rcv_tstamp); 1159 } 1160 1161 static inline int tcp_fin_time(const struct sock *sk) 1162 { 1163 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1164 const int rto = inet_csk(sk)->icsk_rto; 1165 1166 if (fin_timeout < (rto << 2) - (rto >> 1)) 1167 fin_timeout = (rto << 2) - (rto >> 1); 1168 1169 return fin_timeout; 1170 } 1171 1172 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1173 int paws_win) 1174 { 1175 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1176 return true; 1177 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1178 return true; 1179 /* 1180 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1181 * then following tcp messages have valid values. Ignore 0 value, 1182 * or else 'negative' tsval might forbid us to accept their packets. 1183 */ 1184 if (!rx_opt->ts_recent) 1185 return true; 1186 return false; 1187 } 1188 1189 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1190 int rst) 1191 { 1192 if (tcp_paws_check(rx_opt, 0)) 1193 return false; 1194 1195 /* RST segments are not recommended to carry timestamp, 1196 and, if they do, it is recommended to ignore PAWS because 1197 "their cleanup function should take precedence over timestamps." 1198 Certainly, it is mistake. It is necessary to understand the reasons 1199 of this constraint to relax it: if peer reboots, clock may go 1200 out-of-sync and half-open connections will not be reset. 1201 Actually, the problem would be not existing if all 1202 the implementations followed draft about maintaining clock 1203 via reboots. Linux-2.2 DOES NOT! 1204 1205 However, we can relax time bounds for RST segments to MSL. 1206 */ 1207 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1208 return false; 1209 return true; 1210 } 1211 1212 static inline void tcp_mib_init(struct net *net) 1213 { 1214 /* See RFC 2012 */ 1215 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1216 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1217 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1218 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1219 } 1220 1221 /* from STCP */ 1222 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1223 { 1224 tp->lost_skb_hint = NULL; 1225 } 1226 1227 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1228 { 1229 tcp_clear_retrans_hints_partial(tp); 1230 tp->retransmit_skb_hint = NULL; 1231 } 1232 1233 /* MD5 Signature */ 1234 struct crypto_hash; 1235 1236 union tcp_md5_addr { 1237 struct in_addr a4; 1238 #if IS_ENABLED(CONFIG_IPV6) 1239 struct in6_addr a6; 1240 #endif 1241 }; 1242 1243 /* - key database */ 1244 struct tcp_md5sig_key { 1245 struct hlist_node node; 1246 u8 keylen; 1247 u8 family; /* AF_INET or AF_INET6 */ 1248 union tcp_md5_addr addr; 1249 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1250 struct rcu_head rcu; 1251 }; 1252 1253 /* - sock block */ 1254 struct tcp_md5sig_info { 1255 struct hlist_head head; 1256 struct rcu_head rcu; 1257 }; 1258 1259 /* - pseudo header */ 1260 struct tcp4_pseudohdr { 1261 __be32 saddr; 1262 __be32 daddr; 1263 __u8 pad; 1264 __u8 protocol; 1265 __be16 len; 1266 }; 1267 1268 struct tcp6_pseudohdr { 1269 struct in6_addr saddr; 1270 struct in6_addr daddr; 1271 __be32 len; 1272 __be32 protocol; /* including padding */ 1273 }; 1274 1275 union tcp_md5sum_block { 1276 struct tcp4_pseudohdr ip4; 1277 #if IS_ENABLED(CONFIG_IPV6) 1278 struct tcp6_pseudohdr ip6; 1279 #endif 1280 }; 1281 1282 /* - pool: digest algorithm, hash description and scratch buffer */ 1283 struct tcp_md5sig_pool { 1284 struct hash_desc md5_desc; 1285 union tcp_md5sum_block md5_blk; 1286 }; 1287 1288 /* - functions */ 1289 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1290 const struct sock *sk, const struct request_sock *req, 1291 const struct sk_buff *skb); 1292 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1293 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); 1294 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1295 int family); 1296 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1297 struct sock *addr_sk); 1298 1299 #ifdef CONFIG_TCP_MD5SIG 1300 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1301 const union tcp_md5_addr *addr, 1302 int family); 1303 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1304 #else 1305 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1306 const union tcp_md5_addr *addr, 1307 int family) 1308 { 1309 return NULL; 1310 } 1311 #define tcp_twsk_md5_key(twsk) NULL 1312 #endif 1313 1314 bool tcp_alloc_md5sig_pool(void); 1315 1316 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1317 static inline void tcp_put_md5sig_pool(void) 1318 { 1319 local_bh_enable(); 1320 } 1321 1322 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1323 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1324 unsigned int header_len); 1325 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1326 const struct tcp_md5sig_key *key); 1327 1328 /* From tcp_fastopen.c */ 1329 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1330 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1331 unsigned long *last_syn_loss); 1332 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1333 struct tcp_fastopen_cookie *cookie, bool syn_lost); 1334 struct tcp_fastopen_request { 1335 /* Fast Open cookie. Size 0 means a cookie request */ 1336 struct tcp_fastopen_cookie cookie; 1337 struct msghdr *data; /* data in MSG_FASTOPEN */ 1338 size_t size; 1339 int copied; /* queued in tcp_connect() */ 1340 }; 1341 void tcp_free_fastopen_req(struct tcp_sock *tp); 1342 1343 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1344 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1345 bool tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1346 struct request_sock *req, 1347 struct tcp_fastopen_cookie *foc, 1348 struct dst_entry *dst); 1349 void tcp_fastopen_init_key_once(bool publish); 1350 #define TCP_FASTOPEN_KEY_LENGTH 16 1351 1352 /* Fastopen key context */ 1353 struct tcp_fastopen_context { 1354 struct crypto_cipher *tfm; 1355 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1356 struct rcu_head rcu; 1357 }; 1358 1359 /* write queue abstraction */ 1360 static inline void tcp_write_queue_purge(struct sock *sk) 1361 { 1362 struct sk_buff *skb; 1363 1364 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1365 sk_wmem_free_skb(sk, skb); 1366 sk_mem_reclaim(sk); 1367 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1368 } 1369 1370 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1371 { 1372 return skb_peek(&sk->sk_write_queue); 1373 } 1374 1375 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1376 { 1377 return skb_peek_tail(&sk->sk_write_queue); 1378 } 1379 1380 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1381 const struct sk_buff *skb) 1382 { 1383 return skb_queue_next(&sk->sk_write_queue, skb); 1384 } 1385 1386 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1387 const struct sk_buff *skb) 1388 { 1389 return skb_queue_prev(&sk->sk_write_queue, skb); 1390 } 1391 1392 #define tcp_for_write_queue(skb, sk) \ 1393 skb_queue_walk(&(sk)->sk_write_queue, skb) 1394 1395 #define tcp_for_write_queue_from(skb, sk) \ 1396 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1397 1398 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1399 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1400 1401 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1402 { 1403 return sk->sk_send_head; 1404 } 1405 1406 static inline bool tcp_skb_is_last(const struct sock *sk, 1407 const struct sk_buff *skb) 1408 { 1409 return skb_queue_is_last(&sk->sk_write_queue, skb); 1410 } 1411 1412 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1413 { 1414 if (tcp_skb_is_last(sk, skb)) 1415 sk->sk_send_head = NULL; 1416 else 1417 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1418 } 1419 1420 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1421 { 1422 if (sk->sk_send_head == skb_unlinked) 1423 sk->sk_send_head = NULL; 1424 } 1425 1426 static inline void tcp_init_send_head(struct sock *sk) 1427 { 1428 sk->sk_send_head = NULL; 1429 } 1430 1431 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1432 { 1433 __skb_queue_tail(&sk->sk_write_queue, skb); 1434 } 1435 1436 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1437 { 1438 __tcp_add_write_queue_tail(sk, skb); 1439 1440 /* Queue it, remembering where we must start sending. */ 1441 if (sk->sk_send_head == NULL) { 1442 sk->sk_send_head = skb; 1443 1444 if (tcp_sk(sk)->highest_sack == NULL) 1445 tcp_sk(sk)->highest_sack = skb; 1446 } 1447 } 1448 1449 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1450 { 1451 __skb_queue_head(&sk->sk_write_queue, skb); 1452 } 1453 1454 /* Insert buff after skb on the write queue of sk. */ 1455 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1456 struct sk_buff *buff, 1457 struct sock *sk) 1458 { 1459 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1460 } 1461 1462 /* Insert new before skb on the write queue of sk. */ 1463 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1464 struct sk_buff *skb, 1465 struct sock *sk) 1466 { 1467 __skb_queue_before(&sk->sk_write_queue, skb, new); 1468 1469 if (sk->sk_send_head == skb) 1470 sk->sk_send_head = new; 1471 } 1472 1473 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1474 { 1475 __skb_unlink(skb, &sk->sk_write_queue); 1476 } 1477 1478 static inline bool tcp_write_queue_empty(struct sock *sk) 1479 { 1480 return skb_queue_empty(&sk->sk_write_queue); 1481 } 1482 1483 static inline void tcp_push_pending_frames(struct sock *sk) 1484 { 1485 if (tcp_send_head(sk)) { 1486 struct tcp_sock *tp = tcp_sk(sk); 1487 1488 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1489 } 1490 } 1491 1492 /* Start sequence of the skb just after the highest skb with SACKed 1493 * bit, valid only if sacked_out > 0 or when the caller has ensured 1494 * validity by itself. 1495 */ 1496 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1497 { 1498 if (!tp->sacked_out) 1499 return tp->snd_una; 1500 1501 if (tp->highest_sack == NULL) 1502 return tp->snd_nxt; 1503 1504 return TCP_SKB_CB(tp->highest_sack)->seq; 1505 } 1506 1507 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1508 { 1509 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1510 tcp_write_queue_next(sk, skb); 1511 } 1512 1513 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1514 { 1515 return tcp_sk(sk)->highest_sack; 1516 } 1517 1518 static inline void tcp_highest_sack_reset(struct sock *sk) 1519 { 1520 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1521 } 1522 1523 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1524 static inline void tcp_highest_sack_combine(struct sock *sk, 1525 struct sk_buff *old, 1526 struct sk_buff *new) 1527 { 1528 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1529 tcp_sk(sk)->highest_sack = new; 1530 } 1531 1532 /* Determines whether this is a thin stream (which may suffer from 1533 * increased latency). Used to trigger latency-reducing mechanisms. 1534 */ 1535 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1536 { 1537 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1538 } 1539 1540 /* /proc */ 1541 enum tcp_seq_states { 1542 TCP_SEQ_STATE_LISTENING, 1543 TCP_SEQ_STATE_OPENREQ, 1544 TCP_SEQ_STATE_ESTABLISHED, 1545 }; 1546 1547 int tcp_seq_open(struct inode *inode, struct file *file); 1548 1549 struct tcp_seq_afinfo { 1550 char *name; 1551 sa_family_t family; 1552 const struct file_operations *seq_fops; 1553 struct seq_operations seq_ops; 1554 }; 1555 1556 struct tcp_iter_state { 1557 struct seq_net_private p; 1558 sa_family_t family; 1559 enum tcp_seq_states state; 1560 struct sock *syn_wait_sk; 1561 int bucket, offset, sbucket, num; 1562 kuid_t uid; 1563 loff_t last_pos; 1564 }; 1565 1566 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1567 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1568 1569 extern struct request_sock_ops tcp_request_sock_ops; 1570 extern struct request_sock_ops tcp6_request_sock_ops; 1571 1572 void tcp_v4_destroy_sock(struct sock *sk); 1573 1574 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1575 netdev_features_t features); 1576 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1577 int tcp_gro_complete(struct sk_buff *skb); 1578 1579 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1580 1581 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1582 { 1583 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat; 1584 } 1585 1586 static inline bool tcp_stream_memory_free(const struct sock *sk) 1587 { 1588 const struct tcp_sock *tp = tcp_sk(sk); 1589 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1590 1591 return notsent_bytes < tcp_notsent_lowat(tp); 1592 } 1593 1594 #ifdef CONFIG_PROC_FS 1595 int tcp4_proc_init(void); 1596 void tcp4_proc_exit(void); 1597 #endif 1598 1599 int tcp_rtx_synack(struct sock *sk, struct request_sock *req); 1600 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1601 const struct tcp_request_sock_ops *af_ops, 1602 struct sock *sk, struct sk_buff *skb); 1603 1604 /* TCP af-specific functions */ 1605 struct tcp_sock_af_ops { 1606 #ifdef CONFIG_TCP_MD5SIG 1607 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1608 struct sock *addr_sk); 1609 int (*calc_md5_hash) (char *location, 1610 struct tcp_md5sig_key *md5, 1611 const struct sock *sk, 1612 const struct request_sock *req, 1613 const struct sk_buff *skb); 1614 int (*md5_parse) (struct sock *sk, 1615 char __user *optval, 1616 int optlen); 1617 #endif 1618 }; 1619 1620 struct tcp_request_sock_ops { 1621 u16 mss_clamp; 1622 #ifdef CONFIG_TCP_MD5SIG 1623 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1624 struct request_sock *req); 1625 int (*calc_md5_hash) (char *location, 1626 struct tcp_md5sig_key *md5, 1627 const struct sock *sk, 1628 const struct request_sock *req, 1629 const struct sk_buff *skb); 1630 #endif 1631 void (*init_req)(struct request_sock *req, struct sock *sk, 1632 struct sk_buff *skb); 1633 #ifdef CONFIG_SYN_COOKIES 1634 __u32 (*cookie_init_seq)(struct sock *sk, const struct sk_buff *skb, 1635 __u16 *mss); 1636 #endif 1637 struct dst_entry *(*route_req)(struct sock *sk, struct flowi *fl, 1638 const struct request_sock *req, 1639 bool *strict); 1640 __u32 (*init_seq)(const struct sk_buff *skb); 1641 int (*send_synack)(struct sock *sk, struct dst_entry *dst, 1642 struct flowi *fl, struct request_sock *req, 1643 u16 queue_mapping, struct tcp_fastopen_cookie *foc); 1644 void (*queue_hash_add)(struct sock *sk, struct request_sock *req, 1645 const unsigned long timeout); 1646 }; 1647 1648 #ifdef CONFIG_SYN_COOKIES 1649 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1650 struct sock *sk, struct sk_buff *skb, 1651 __u16 *mss) 1652 { 1653 return ops->cookie_init_seq(sk, skb, mss); 1654 } 1655 #else 1656 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1657 struct sock *sk, struct sk_buff *skb, 1658 __u16 *mss) 1659 { 1660 return 0; 1661 } 1662 #endif 1663 1664 int tcpv4_offload_init(void); 1665 1666 void tcp_v4_init(void); 1667 void tcp_init(void); 1668 1669 #endif /* _TCP_H */ 1670