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/dmaengine.h> 31 #include <linux/crypto.h> 32 #include <linux/cryptohash.h> 33 #include <linux/kref.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 /* Process SYN data but skip cookie validation */ 223 #define TFO_SERVER_COOKIE_NOT_CHKED 0x100 224 /* Accept SYN data w/o any cookie option */ 225 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 226 227 /* Force enable TFO on all listeners, i.e., not requiring the 228 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen. 229 */ 230 #define TFO_SERVER_WO_SOCKOPT1 0x400 231 #define TFO_SERVER_WO_SOCKOPT2 0x800 232 /* Always create TFO child sockets on a TFO listener even when 233 * cookie/data not present. (For testing purpose!) 234 */ 235 #define TFO_SERVER_ALWAYS 0x1000 236 237 extern struct inet_timewait_death_row tcp_death_row; 238 239 /* sysctl variables for tcp */ 240 extern int sysctl_tcp_timestamps; 241 extern int sysctl_tcp_window_scaling; 242 extern int sysctl_tcp_sack; 243 extern int sysctl_tcp_fin_timeout; 244 extern int sysctl_tcp_keepalive_time; 245 extern int sysctl_tcp_keepalive_probes; 246 extern int sysctl_tcp_keepalive_intvl; 247 extern int sysctl_tcp_syn_retries; 248 extern int sysctl_tcp_synack_retries; 249 extern int sysctl_tcp_retries1; 250 extern int sysctl_tcp_retries2; 251 extern int sysctl_tcp_orphan_retries; 252 extern int sysctl_tcp_syncookies; 253 extern int sysctl_tcp_fastopen; 254 extern int sysctl_tcp_retrans_collapse; 255 extern int sysctl_tcp_stdurg; 256 extern int sysctl_tcp_rfc1337; 257 extern int sysctl_tcp_abort_on_overflow; 258 extern int sysctl_tcp_max_orphans; 259 extern int sysctl_tcp_fack; 260 extern int sysctl_tcp_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_dma_copybreak; 271 extern int sysctl_tcp_nometrics_save; 272 extern int sysctl_tcp_moderate_rcvbuf; 273 extern int sysctl_tcp_tso_win_divisor; 274 extern int sysctl_tcp_mtu_probing; 275 extern int sysctl_tcp_base_mss; 276 extern int sysctl_tcp_workaround_signed_windows; 277 extern int sysctl_tcp_slow_start_after_idle; 278 extern int sysctl_tcp_thin_linear_timeouts; 279 extern int sysctl_tcp_thin_dupack; 280 extern int sysctl_tcp_early_retrans; 281 extern int sysctl_tcp_limit_output_bytes; 282 extern int sysctl_tcp_challenge_ack_limit; 283 extern unsigned int sysctl_tcp_notsent_lowat; 284 extern int sysctl_tcp_min_tso_segs; 285 extern int sysctl_tcp_autocorking; 286 287 extern atomic_long_t tcp_memory_allocated; 288 extern struct percpu_counter tcp_sockets_allocated; 289 extern int tcp_memory_pressure; 290 291 /* 292 * The next routines deal with comparing 32 bit unsigned ints 293 * and worry about wraparound (automatic with unsigned arithmetic). 294 */ 295 296 static inline bool before(__u32 seq1, __u32 seq2) 297 { 298 return (__s32)(seq1-seq2) < 0; 299 } 300 #define after(seq2, seq1) before(seq1, seq2) 301 302 /* is s2<=s1<=s3 ? */ 303 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 304 { 305 return seq3 - seq2 >= seq1 - seq2; 306 } 307 308 static inline bool tcp_out_of_memory(struct sock *sk) 309 { 310 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 311 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 312 return true; 313 return false; 314 } 315 316 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 317 { 318 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 319 int orphans = percpu_counter_read_positive(ocp); 320 321 if (orphans << shift > sysctl_tcp_max_orphans) { 322 orphans = percpu_counter_sum_positive(ocp); 323 if (orphans << shift > sysctl_tcp_max_orphans) 324 return true; 325 } 326 return false; 327 } 328 329 bool tcp_check_oom(struct sock *sk, int shift); 330 331 /* syncookies: remember time of last synqueue overflow */ 332 static inline void tcp_synq_overflow(struct sock *sk) 333 { 334 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies; 335 } 336 337 /* syncookies: no recent synqueue overflow on this listening socket? */ 338 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 339 { 340 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 341 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK); 342 } 343 344 extern struct proto tcp_prot; 345 346 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 347 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 348 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 349 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 350 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 351 352 void tcp_tasklet_init(void); 353 354 void tcp_v4_err(struct sk_buff *skb, u32); 355 356 void tcp_shutdown(struct sock *sk, int how); 357 358 void tcp_v4_early_demux(struct sk_buff *skb); 359 int tcp_v4_rcv(struct sk_buff *skb); 360 361 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 362 int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 363 size_t size); 364 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 365 int flags); 366 void tcp_release_cb(struct sock *sk); 367 void tcp_wfree(struct sk_buff *skb); 368 void tcp_write_timer_handler(struct sock *sk); 369 void tcp_delack_timer_handler(struct sock *sk); 370 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 371 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 372 const struct tcphdr *th, unsigned int len); 373 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 374 const struct tcphdr *th, unsigned int len); 375 void tcp_rcv_space_adjust(struct sock *sk); 376 void tcp_cleanup_rbuf(struct sock *sk, int copied); 377 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 378 void tcp_twsk_destructor(struct sock *sk); 379 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 380 struct pipe_inode_info *pipe, size_t len, 381 unsigned int flags); 382 383 static inline void tcp_dec_quickack_mode(struct sock *sk, 384 const unsigned int pkts) 385 { 386 struct inet_connection_sock *icsk = inet_csk(sk); 387 388 if (icsk->icsk_ack.quick) { 389 if (pkts >= icsk->icsk_ack.quick) { 390 icsk->icsk_ack.quick = 0; 391 /* Leaving quickack mode we deflate ATO. */ 392 icsk->icsk_ack.ato = TCP_ATO_MIN; 393 } else 394 icsk->icsk_ack.quick -= pkts; 395 } 396 } 397 398 #define TCP_ECN_OK 1 399 #define TCP_ECN_QUEUE_CWR 2 400 #define TCP_ECN_DEMAND_CWR 4 401 #define TCP_ECN_SEEN 8 402 403 enum tcp_tw_status { 404 TCP_TW_SUCCESS = 0, 405 TCP_TW_RST = 1, 406 TCP_TW_ACK = 2, 407 TCP_TW_SYN = 3 408 }; 409 410 411 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 412 struct sk_buff *skb, 413 const struct tcphdr *th); 414 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 415 struct request_sock *req, struct request_sock **prev, 416 bool fastopen); 417 int tcp_child_process(struct sock *parent, struct sock *child, 418 struct sk_buff *skb); 419 void tcp_enter_loss(struct sock *sk, int how); 420 void tcp_clear_retrans(struct tcp_sock *tp); 421 void tcp_update_metrics(struct sock *sk); 422 void tcp_init_metrics(struct sock *sk); 423 void tcp_metrics_init(void); 424 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, 425 bool paws_check); 426 bool tcp_remember_stamp(struct sock *sk); 427 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); 428 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); 429 void tcp_disable_fack(struct tcp_sock *tp); 430 void tcp_close(struct sock *sk, long timeout); 431 void tcp_init_sock(struct sock *sk); 432 unsigned int tcp_poll(struct file *file, struct socket *sock, 433 struct poll_table_struct *wait); 434 int tcp_getsockopt(struct sock *sk, int level, int optname, 435 char __user *optval, int __user *optlen); 436 int tcp_setsockopt(struct sock *sk, int level, int optname, 437 char __user *optval, unsigned int optlen); 438 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 439 char __user *optval, int __user *optlen); 440 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 441 char __user *optval, unsigned int optlen); 442 void tcp_set_keepalive(struct sock *sk, int val); 443 void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req); 444 int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 445 size_t len, int nonblock, int flags, int *addr_len); 446 void tcp_parse_options(const struct sk_buff *skb, 447 struct tcp_options_received *opt_rx, 448 int estab, struct tcp_fastopen_cookie *foc); 449 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 450 451 /* 452 * TCP v4 functions exported for the inet6 API 453 */ 454 455 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 456 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 457 struct sock *tcp_create_openreq_child(struct sock *sk, 458 struct request_sock *req, 459 struct sk_buff *skb); 460 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 461 struct request_sock *req, 462 struct dst_entry *dst); 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(struct sock *sk, struct dst_entry *dst, 467 struct request_sock *req, 468 struct tcp_fastopen_cookie *foc); 469 int tcp_disconnect(struct sock *sk, int flags); 470 471 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 472 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 473 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 474 475 /* From syncookies.c */ 476 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 477 u32 cookie); 478 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 479 struct ip_options *opt); 480 #ifdef CONFIG_SYN_COOKIES 481 #include <linux/ktime.h> 482 483 /* Syncookies use a monotonic timer which increments every 60 seconds. 484 * This counter is used both as a hash input and partially encoded into 485 * the cookie value. A cookie is only validated further if the delta 486 * between the current counter value and the encoded one is less than this, 487 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 488 * the counter advances immediately after a cookie is generated). 489 */ 490 #define MAX_SYNCOOKIE_AGE 2 491 492 static inline u32 tcp_cookie_time(void) 493 { 494 u64 val = get_jiffies_64(); 495 496 do_div(val, 60 * HZ); 497 return val; 498 } 499 500 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 501 u16 *mssp); 502 __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, __u16 *mss); 503 #else 504 static inline __u32 cookie_v4_init_sequence(struct sock *sk, 505 struct sk_buff *skb, 506 __u16 *mss) 507 { 508 return 0; 509 } 510 #endif 511 512 __u32 cookie_init_timestamp(struct request_sock *req); 513 bool cookie_check_timestamp(struct tcp_options_received *opt, struct net *net, 514 bool *ecn_ok); 515 516 /* From net/ipv6/syncookies.c */ 517 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 518 u32 cookie); 519 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 520 #ifdef CONFIG_SYN_COOKIES 521 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 522 const struct tcphdr *th, u16 *mssp); 523 __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb, 524 __u16 *mss); 525 #else 526 static inline __u32 cookie_v6_init_sequence(struct sock *sk, 527 struct sk_buff *skb, 528 __u16 *mss) 529 { 530 return 0; 531 } 532 #endif 533 /* tcp_output.c */ 534 535 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 536 int nonagle); 537 bool tcp_may_send_now(struct sock *sk); 538 int __tcp_retransmit_skb(struct sock *, struct sk_buff *); 539 int tcp_retransmit_skb(struct sock *, struct sk_buff *); 540 void tcp_retransmit_timer(struct sock *sk); 541 void tcp_xmit_retransmit_queue(struct sock *); 542 void tcp_simple_retransmit(struct sock *); 543 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 544 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 545 546 void tcp_send_probe0(struct sock *); 547 void tcp_send_partial(struct sock *); 548 int tcp_write_wakeup(struct sock *); 549 void tcp_send_fin(struct sock *sk); 550 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 551 int tcp_send_synack(struct sock *); 552 bool tcp_syn_flood_action(struct sock *sk, const struct sk_buff *skb, 553 const char *proto); 554 void tcp_push_one(struct sock *, unsigned int mss_now); 555 void tcp_send_ack(struct sock *sk); 556 void tcp_send_delayed_ack(struct sock *sk); 557 void tcp_send_loss_probe(struct sock *sk); 558 bool tcp_schedule_loss_probe(struct sock *sk); 559 560 /* tcp_input.c */ 561 void tcp_cwnd_application_limited(struct sock *sk); 562 void tcp_resume_early_retransmit(struct sock *sk); 563 void tcp_rearm_rto(struct sock *sk); 564 void tcp_reset(struct sock *sk); 565 566 /* tcp_timer.c */ 567 void tcp_init_xmit_timers(struct sock *); 568 static inline void tcp_clear_xmit_timers(struct sock *sk) 569 { 570 inet_csk_clear_xmit_timers(sk); 571 } 572 573 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 574 unsigned int tcp_current_mss(struct sock *sk); 575 576 /* Bound MSS / TSO packet size with the half of the window */ 577 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 578 { 579 int cutoff; 580 581 /* When peer uses tiny windows, there is no use in packetizing 582 * to sub-MSS pieces for the sake of SWS or making sure there 583 * are enough packets in the pipe for fast recovery. 584 * 585 * On the other hand, for extremely large MSS devices, handling 586 * smaller than MSS windows in this way does make sense. 587 */ 588 if (tp->max_window >= 512) 589 cutoff = (tp->max_window >> 1); 590 else 591 cutoff = tp->max_window; 592 593 if (cutoff && pktsize > cutoff) 594 return max_t(int, cutoff, 68U - tp->tcp_header_len); 595 else 596 return pktsize; 597 } 598 599 /* tcp.c */ 600 void tcp_get_info(const struct sock *, struct tcp_info *); 601 602 /* Read 'sendfile()'-style from a TCP socket */ 603 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 604 unsigned int, size_t); 605 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 606 sk_read_actor_t recv_actor); 607 608 void tcp_initialize_rcv_mss(struct sock *sk); 609 610 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 611 int tcp_mss_to_mtu(struct sock *sk, int mss); 612 void tcp_mtup_init(struct sock *sk); 613 void tcp_init_buffer_space(struct sock *sk); 614 615 static inline void tcp_bound_rto(const struct sock *sk) 616 { 617 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 618 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 619 } 620 621 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 622 { 623 return (tp->srtt >> 3) + tp->rttvar; 624 } 625 626 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 627 { 628 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 629 ntohl(TCP_FLAG_ACK) | 630 snd_wnd); 631 } 632 633 static inline void tcp_fast_path_on(struct tcp_sock *tp) 634 { 635 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 636 } 637 638 static inline void tcp_fast_path_check(struct sock *sk) 639 { 640 struct tcp_sock *tp = tcp_sk(sk); 641 642 if (skb_queue_empty(&tp->out_of_order_queue) && 643 tp->rcv_wnd && 644 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 645 !tp->urg_data) 646 tcp_fast_path_on(tp); 647 } 648 649 /* Compute the actual rto_min value */ 650 static inline u32 tcp_rto_min(struct sock *sk) 651 { 652 const struct dst_entry *dst = __sk_dst_get(sk); 653 u32 rto_min = TCP_RTO_MIN; 654 655 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 656 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 657 return rto_min; 658 } 659 660 /* Compute the actual receive window we are currently advertising. 661 * Rcv_nxt can be after the window if our peer push more data 662 * than the offered window. 663 */ 664 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 665 { 666 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 667 668 if (win < 0) 669 win = 0; 670 return (u32) win; 671 } 672 673 /* Choose a new window, without checks for shrinking, and without 674 * scaling applied to the result. The caller does these things 675 * if necessary. This is a "raw" window selection. 676 */ 677 u32 __tcp_select_window(struct sock *sk); 678 679 void tcp_send_window_probe(struct sock *sk); 680 681 /* TCP timestamps are only 32-bits, this causes a slight 682 * complication on 64-bit systems since we store a snapshot 683 * of jiffies in the buffer control blocks below. We decided 684 * to use only the low 32-bits of jiffies and hide the ugly 685 * casts with the following macro. 686 */ 687 #define tcp_time_stamp ((__u32)(jiffies)) 688 689 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 690 691 #define TCPHDR_FIN 0x01 692 #define TCPHDR_SYN 0x02 693 #define TCPHDR_RST 0x04 694 #define TCPHDR_PSH 0x08 695 #define TCPHDR_ACK 0x10 696 #define TCPHDR_URG 0x20 697 #define TCPHDR_ECE 0x40 698 #define TCPHDR_CWR 0x80 699 700 /* This is what the send packet queuing engine uses to pass 701 * TCP per-packet control information to the transmission code. 702 * We also store the host-order sequence numbers in here too. 703 * This is 44 bytes if IPV6 is enabled. 704 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 705 */ 706 struct tcp_skb_cb { 707 union { 708 struct inet_skb_parm h4; 709 #if IS_ENABLED(CONFIG_IPV6) 710 struct inet6_skb_parm h6; 711 #endif 712 } header; /* For incoming frames */ 713 __u32 seq; /* Starting sequence number */ 714 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 715 __u32 when; /* used to compute rtt's */ 716 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 717 718 __u8 sacked; /* State flags for SACK/FACK. */ 719 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 720 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 721 #define TCPCB_LOST 0x04 /* SKB is lost */ 722 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 723 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 724 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 725 726 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 727 /* 1 byte hole */ 728 __u32 ack_seq; /* Sequence number ACK'd */ 729 }; 730 731 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 732 733 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set 734 * 735 * If we receive a SYN packet with these bits set, it means a network is 736 * playing bad games with TOS bits. In order to avoid possible false congestion 737 * notifications, we disable TCP ECN negociation. 738 */ 739 static inline void 740 TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb, 741 struct net *net) 742 { 743 const struct tcphdr *th = tcp_hdr(skb); 744 745 if (net->ipv4.sysctl_tcp_ecn && th->ece && th->cwr && 746 INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield)) 747 inet_rsk(req)->ecn_ok = 1; 748 } 749 750 /* Due to TSO, an SKB can be composed of multiple actual 751 * packets. To keep these tracked properly, we use this. 752 */ 753 static inline int tcp_skb_pcount(const struct sk_buff *skb) 754 { 755 return skb_shinfo(skb)->gso_segs; 756 } 757 758 /* This is valid iff tcp_skb_pcount() > 1. */ 759 static inline int tcp_skb_mss(const struct sk_buff *skb) 760 { 761 return skb_shinfo(skb)->gso_size; 762 } 763 764 /* Events passed to congestion control interface */ 765 enum tcp_ca_event { 766 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 767 CA_EVENT_CWND_RESTART, /* congestion window restart */ 768 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 769 CA_EVENT_LOSS, /* loss timeout */ 770 CA_EVENT_FAST_ACK, /* in sequence ack */ 771 CA_EVENT_SLOW_ACK, /* other ack */ 772 }; 773 774 /* 775 * Interface for adding new TCP congestion control handlers 776 */ 777 #define TCP_CA_NAME_MAX 16 778 #define TCP_CA_MAX 128 779 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 780 781 #define TCP_CONG_NON_RESTRICTED 0x1 782 #define TCP_CONG_RTT_STAMP 0x2 783 784 struct tcp_congestion_ops { 785 struct list_head list; 786 unsigned long flags; 787 788 /* initialize private data (optional) */ 789 void (*init)(struct sock *sk); 790 /* cleanup private data (optional) */ 791 void (*release)(struct sock *sk); 792 793 /* return slow start threshold (required) */ 794 u32 (*ssthresh)(struct sock *sk); 795 /* lower bound for congestion window (optional) */ 796 u32 (*min_cwnd)(const struct sock *sk); 797 /* do new cwnd calculation (required) */ 798 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked, u32 in_flight); 799 /* call before changing ca_state (optional) */ 800 void (*set_state)(struct sock *sk, u8 new_state); 801 /* call when cwnd event occurs (optional) */ 802 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 803 /* new value of cwnd after loss (optional) */ 804 u32 (*undo_cwnd)(struct sock *sk); 805 /* hook for packet ack accounting (optional) */ 806 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 807 /* get info for inet_diag (optional) */ 808 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 809 810 char name[TCP_CA_NAME_MAX]; 811 struct module *owner; 812 }; 813 814 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 815 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 816 817 void tcp_init_congestion_control(struct sock *sk); 818 void tcp_cleanup_congestion_control(struct sock *sk); 819 int tcp_set_default_congestion_control(const char *name); 820 void tcp_get_default_congestion_control(char *name); 821 void tcp_get_available_congestion_control(char *buf, size_t len); 822 void tcp_get_allowed_congestion_control(char *buf, size_t len); 823 int tcp_set_allowed_congestion_control(char *allowed); 824 int tcp_set_congestion_control(struct sock *sk, const char *name); 825 int tcp_slow_start(struct tcp_sock *tp, u32 acked); 826 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 827 828 extern struct tcp_congestion_ops tcp_init_congestion_ops; 829 u32 tcp_reno_ssthresh(struct sock *sk); 830 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked, u32 in_flight); 831 u32 tcp_reno_min_cwnd(const struct sock *sk); 832 extern struct tcp_congestion_ops tcp_reno; 833 834 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 835 { 836 struct inet_connection_sock *icsk = inet_csk(sk); 837 838 if (icsk->icsk_ca_ops->set_state) 839 icsk->icsk_ca_ops->set_state(sk, ca_state); 840 icsk->icsk_ca_state = ca_state; 841 } 842 843 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 844 { 845 const struct inet_connection_sock *icsk = inet_csk(sk); 846 847 if (icsk->icsk_ca_ops->cwnd_event) 848 icsk->icsk_ca_ops->cwnd_event(sk, event); 849 } 850 851 /* These functions determine how the current flow behaves in respect of SACK 852 * handling. SACK is negotiated with the peer, and therefore it can vary 853 * between different flows. 854 * 855 * tcp_is_sack - SACK enabled 856 * tcp_is_reno - No SACK 857 * tcp_is_fack - FACK enabled, implies SACK enabled 858 */ 859 static inline int tcp_is_sack(const struct tcp_sock *tp) 860 { 861 return tp->rx_opt.sack_ok; 862 } 863 864 static inline bool tcp_is_reno(const struct tcp_sock *tp) 865 { 866 return !tcp_is_sack(tp); 867 } 868 869 static inline bool tcp_is_fack(const struct tcp_sock *tp) 870 { 871 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 872 } 873 874 static inline void tcp_enable_fack(struct tcp_sock *tp) 875 { 876 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 877 } 878 879 /* TCP early-retransmit (ER) is similar to but more conservative than 880 * the thin-dupack feature. Enable ER only if thin-dupack is disabled. 881 */ 882 static inline void tcp_enable_early_retrans(struct tcp_sock *tp) 883 { 884 tp->do_early_retrans = sysctl_tcp_early_retrans && 885 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack && 886 sysctl_tcp_reordering == 3; 887 } 888 889 static inline void tcp_disable_early_retrans(struct tcp_sock *tp) 890 { 891 tp->do_early_retrans = 0; 892 } 893 894 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 895 { 896 return tp->sacked_out + tp->lost_out; 897 } 898 899 /* This determines how many packets are "in the network" to the best 900 * of our knowledge. In many cases it is conservative, but where 901 * detailed information is available from the receiver (via SACK 902 * blocks etc.) we can make more aggressive calculations. 903 * 904 * Use this for decisions involving congestion control, use just 905 * tp->packets_out to determine if the send queue is empty or not. 906 * 907 * Read this equation as: 908 * 909 * "Packets sent once on transmission queue" MINUS 910 * "Packets left network, but not honestly ACKed yet" PLUS 911 * "Packets fast retransmitted" 912 */ 913 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 914 { 915 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 916 } 917 918 #define TCP_INFINITE_SSTHRESH 0x7fffffff 919 920 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 921 { 922 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 923 } 924 925 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 926 { 927 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 928 (1 << inet_csk(sk)->icsk_ca_state); 929 } 930 931 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 932 * The exception is cwnd reduction phase, when cwnd is decreasing towards 933 * ssthresh. 934 */ 935 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 936 { 937 const struct tcp_sock *tp = tcp_sk(sk); 938 939 if (tcp_in_cwnd_reduction(sk)) 940 return tp->snd_ssthresh; 941 else 942 return max(tp->snd_ssthresh, 943 ((tp->snd_cwnd >> 1) + 944 (tp->snd_cwnd >> 2))); 945 } 946 947 /* Use define here intentionally to get WARN_ON location shown at the caller */ 948 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 949 950 void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 951 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 952 953 /* The maximum number of MSS of available cwnd for which TSO defers 954 * sending if not using sysctl_tcp_tso_win_divisor. 955 */ 956 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 957 { 958 return 3; 959 } 960 961 /* Slow start with delack produces 3 packets of burst, so that 962 * it is safe "de facto". This will be the default - same as 963 * the default reordering threshold - but if reordering increases, 964 * we must be able to allow cwnd to burst at least this much in order 965 * to not pull it back when holes are filled. 966 */ 967 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 968 { 969 return tp->reordering; 970 } 971 972 /* Returns end sequence number of the receiver's advertised window */ 973 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 974 { 975 return tp->snd_una + tp->snd_wnd; 976 } 977 bool tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight); 978 979 static inline void tcp_check_probe_timer(struct sock *sk) 980 { 981 const struct tcp_sock *tp = tcp_sk(sk); 982 const struct inet_connection_sock *icsk = inet_csk(sk); 983 984 if (!tp->packets_out && !icsk->icsk_pending) 985 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 986 icsk->icsk_rto, TCP_RTO_MAX); 987 } 988 989 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 990 { 991 tp->snd_wl1 = seq; 992 } 993 994 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 995 { 996 tp->snd_wl1 = seq; 997 } 998 999 /* 1000 * Calculate(/check) TCP checksum 1001 */ 1002 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1003 __be32 daddr, __wsum base) 1004 { 1005 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1006 } 1007 1008 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1009 { 1010 return __skb_checksum_complete(skb); 1011 } 1012 1013 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1014 { 1015 return !skb_csum_unnecessary(skb) && 1016 __tcp_checksum_complete(skb); 1017 } 1018 1019 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1020 1021 static inline void tcp_prequeue_init(struct tcp_sock *tp) 1022 { 1023 tp->ucopy.task = NULL; 1024 tp->ucopy.len = 0; 1025 tp->ucopy.memory = 0; 1026 skb_queue_head_init(&tp->ucopy.prequeue); 1027 #ifdef CONFIG_NET_DMA 1028 tp->ucopy.dma_chan = NULL; 1029 tp->ucopy.wakeup = 0; 1030 tp->ucopy.pinned_list = NULL; 1031 tp->ucopy.dma_cookie = 0; 1032 #endif 1033 } 1034 1035 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1036 1037 #undef STATE_TRACE 1038 1039 #ifdef STATE_TRACE 1040 static const char *statename[]={ 1041 "Unused","Established","Syn Sent","Syn Recv", 1042 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1043 "Close Wait","Last ACK","Listen","Closing" 1044 }; 1045 #endif 1046 void tcp_set_state(struct sock *sk, int state); 1047 1048 void tcp_done(struct sock *sk); 1049 1050 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1051 { 1052 rx_opt->dsack = 0; 1053 rx_opt->num_sacks = 0; 1054 } 1055 1056 u32 tcp_default_init_rwnd(u32 mss); 1057 1058 /* Determine a window scaling and initial window to offer. */ 1059 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1060 __u32 *window_clamp, int wscale_ok, 1061 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1062 1063 static inline int tcp_win_from_space(int space) 1064 { 1065 return sysctl_tcp_adv_win_scale<=0 ? 1066 (space>>(-sysctl_tcp_adv_win_scale)) : 1067 space - (space>>sysctl_tcp_adv_win_scale); 1068 } 1069 1070 /* Note: caller must be prepared to deal with negative returns */ 1071 static inline int tcp_space(const struct sock *sk) 1072 { 1073 return tcp_win_from_space(sk->sk_rcvbuf - 1074 atomic_read(&sk->sk_rmem_alloc)); 1075 } 1076 1077 static inline int tcp_full_space(const struct sock *sk) 1078 { 1079 return tcp_win_from_space(sk->sk_rcvbuf); 1080 } 1081 1082 static inline void tcp_openreq_init(struct request_sock *req, 1083 struct tcp_options_received *rx_opt, 1084 struct sk_buff *skb) 1085 { 1086 struct inet_request_sock *ireq = inet_rsk(req); 1087 1088 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 1089 req->cookie_ts = 0; 1090 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 1091 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1; 1092 tcp_rsk(req)->snt_synack = 0; 1093 req->mss = rx_opt->mss_clamp; 1094 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 1095 ireq->tstamp_ok = rx_opt->tstamp_ok; 1096 ireq->sack_ok = rx_opt->sack_ok; 1097 ireq->snd_wscale = rx_opt->snd_wscale; 1098 ireq->wscale_ok = rx_opt->wscale_ok; 1099 ireq->acked = 0; 1100 ireq->ecn_ok = 0; 1101 ireq->ir_rmt_port = tcp_hdr(skb)->source; 1102 ireq->ir_num = ntohs(tcp_hdr(skb)->dest); 1103 } 1104 1105 void tcp_enter_memory_pressure(struct sock *sk); 1106 1107 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1108 { 1109 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1110 } 1111 1112 static inline int keepalive_time_when(const struct tcp_sock *tp) 1113 { 1114 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1115 } 1116 1117 static inline int keepalive_probes(const struct tcp_sock *tp) 1118 { 1119 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1120 } 1121 1122 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1123 { 1124 const struct inet_connection_sock *icsk = &tp->inet_conn; 1125 1126 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1127 tcp_time_stamp - tp->rcv_tstamp); 1128 } 1129 1130 static inline int tcp_fin_time(const struct sock *sk) 1131 { 1132 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1133 const int rto = inet_csk(sk)->icsk_rto; 1134 1135 if (fin_timeout < (rto << 2) - (rto >> 1)) 1136 fin_timeout = (rto << 2) - (rto >> 1); 1137 1138 return fin_timeout; 1139 } 1140 1141 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1142 int paws_win) 1143 { 1144 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1145 return true; 1146 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1147 return true; 1148 /* 1149 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1150 * then following tcp messages have valid values. Ignore 0 value, 1151 * or else 'negative' tsval might forbid us to accept their packets. 1152 */ 1153 if (!rx_opt->ts_recent) 1154 return true; 1155 return false; 1156 } 1157 1158 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1159 int rst) 1160 { 1161 if (tcp_paws_check(rx_opt, 0)) 1162 return false; 1163 1164 /* RST segments are not recommended to carry timestamp, 1165 and, if they do, it is recommended to ignore PAWS because 1166 "their cleanup function should take precedence over timestamps." 1167 Certainly, it is mistake. It is necessary to understand the reasons 1168 of this constraint to relax it: if peer reboots, clock may go 1169 out-of-sync and half-open connections will not be reset. 1170 Actually, the problem would be not existing if all 1171 the implementations followed draft about maintaining clock 1172 via reboots. Linux-2.2 DOES NOT! 1173 1174 However, we can relax time bounds for RST segments to MSL. 1175 */ 1176 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1177 return false; 1178 return true; 1179 } 1180 1181 static inline void tcp_mib_init(struct net *net) 1182 { 1183 /* See RFC 2012 */ 1184 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1185 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1186 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1187 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1188 } 1189 1190 /* from STCP */ 1191 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1192 { 1193 tp->lost_skb_hint = NULL; 1194 } 1195 1196 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1197 { 1198 tcp_clear_retrans_hints_partial(tp); 1199 tp->retransmit_skb_hint = NULL; 1200 } 1201 1202 /* MD5 Signature */ 1203 struct crypto_hash; 1204 1205 union tcp_md5_addr { 1206 struct in_addr a4; 1207 #if IS_ENABLED(CONFIG_IPV6) 1208 struct in6_addr a6; 1209 #endif 1210 }; 1211 1212 /* - key database */ 1213 struct tcp_md5sig_key { 1214 struct hlist_node node; 1215 u8 keylen; 1216 u8 family; /* AF_INET or AF_INET6 */ 1217 union tcp_md5_addr addr; 1218 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1219 struct rcu_head rcu; 1220 }; 1221 1222 /* - sock block */ 1223 struct tcp_md5sig_info { 1224 struct hlist_head head; 1225 struct rcu_head rcu; 1226 }; 1227 1228 /* - pseudo header */ 1229 struct tcp4_pseudohdr { 1230 __be32 saddr; 1231 __be32 daddr; 1232 __u8 pad; 1233 __u8 protocol; 1234 __be16 len; 1235 }; 1236 1237 struct tcp6_pseudohdr { 1238 struct in6_addr saddr; 1239 struct in6_addr daddr; 1240 __be32 len; 1241 __be32 protocol; /* including padding */ 1242 }; 1243 1244 union tcp_md5sum_block { 1245 struct tcp4_pseudohdr ip4; 1246 #if IS_ENABLED(CONFIG_IPV6) 1247 struct tcp6_pseudohdr ip6; 1248 #endif 1249 }; 1250 1251 /* - pool: digest algorithm, hash description and scratch buffer */ 1252 struct tcp_md5sig_pool { 1253 struct hash_desc md5_desc; 1254 union tcp_md5sum_block md5_blk; 1255 }; 1256 1257 /* - functions */ 1258 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1259 const struct sock *sk, const struct request_sock *req, 1260 const struct sk_buff *skb); 1261 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1262 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); 1263 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1264 int family); 1265 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1266 struct sock *addr_sk); 1267 1268 #ifdef CONFIG_TCP_MD5SIG 1269 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1270 const union tcp_md5_addr *addr, 1271 int family); 1272 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1273 #else 1274 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1275 const union tcp_md5_addr *addr, 1276 int family) 1277 { 1278 return NULL; 1279 } 1280 #define tcp_twsk_md5_key(twsk) NULL 1281 #endif 1282 1283 bool tcp_alloc_md5sig_pool(void); 1284 1285 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1286 static inline void tcp_put_md5sig_pool(void) 1287 { 1288 local_bh_enable(); 1289 } 1290 1291 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1292 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1293 unsigned int header_len); 1294 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1295 const struct tcp_md5sig_key *key); 1296 1297 /* From tcp_fastopen.c */ 1298 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1299 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1300 unsigned long *last_syn_loss); 1301 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1302 struct tcp_fastopen_cookie *cookie, bool syn_lost); 1303 struct tcp_fastopen_request { 1304 /* Fast Open cookie. Size 0 means a cookie request */ 1305 struct tcp_fastopen_cookie cookie; 1306 struct msghdr *data; /* data in MSG_FASTOPEN */ 1307 size_t size; 1308 int copied; /* queued in tcp_connect() */ 1309 }; 1310 void tcp_free_fastopen_req(struct tcp_sock *tp); 1311 1312 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1313 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1314 void tcp_fastopen_cookie_gen(__be32 src, __be32 dst, 1315 struct tcp_fastopen_cookie *foc); 1316 void tcp_fastopen_init_key_once(bool publish); 1317 #define TCP_FASTOPEN_KEY_LENGTH 16 1318 1319 /* Fastopen key context */ 1320 struct tcp_fastopen_context { 1321 struct crypto_cipher *tfm; 1322 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1323 struct rcu_head rcu; 1324 }; 1325 1326 /* write queue abstraction */ 1327 static inline void tcp_write_queue_purge(struct sock *sk) 1328 { 1329 struct sk_buff *skb; 1330 1331 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1332 sk_wmem_free_skb(sk, skb); 1333 sk_mem_reclaim(sk); 1334 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1335 } 1336 1337 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1338 { 1339 return skb_peek(&sk->sk_write_queue); 1340 } 1341 1342 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1343 { 1344 return skb_peek_tail(&sk->sk_write_queue); 1345 } 1346 1347 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1348 const struct sk_buff *skb) 1349 { 1350 return skb_queue_next(&sk->sk_write_queue, skb); 1351 } 1352 1353 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1354 const struct sk_buff *skb) 1355 { 1356 return skb_queue_prev(&sk->sk_write_queue, skb); 1357 } 1358 1359 #define tcp_for_write_queue(skb, sk) \ 1360 skb_queue_walk(&(sk)->sk_write_queue, skb) 1361 1362 #define tcp_for_write_queue_from(skb, sk) \ 1363 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1364 1365 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1366 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1367 1368 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1369 { 1370 return sk->sk_send_head; 1371 } 1372 1373 static inline bool tcp_skb_is_last(const struct sock *sk, 1374 const struct sk_buff *skb) 1375 { 1376 return skb_queue_is_last(&sk->sk_write_queue, skb); 1377 } 1378 1379 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1380 { 1381 if (tcp_skb_is_last(sk, skb)) 1382 sk->sk_send_head = NULL; 1383 else 1384 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1385 } 1386 1387 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1388 { 1389 if (sk->sk_send_head == skb_unlinked) 1390 sk->sk_send_head = NULL; 1391 } 1392 1393 static inline void tcp_init_send_head(struct sock *sk) 1394 { 1395 sk->sk_send_head = NULL; 1396 } 1397 1398 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1399 { 1400 __skb_queue_tail(&sk->sk_write_queue, skb); 1401 } 1402 1403 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1404 { 1405 __tcp_add_write_queue_tail(sk, skb); 1406 1407 /* Queue it, remembering where we must start sending. */ 1408 if (sk->sk_send_head == NULL) { 1409 sk->sk_send_head = skb; 1410 1411 if (tcp_sk(sk)->highest_sack == NULL) 1412 tcp_sk(sk)->highest_sack = skb; 1413 } 1414 } 1415 1416 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1417 { 1418 __skb_queue_head(&sk->sk_write_queue, skb); 1419 } 1420 1421 /* Insert buff after skb on the write queue of sk. */ 1422 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1423 struct sk_buff *buff, 1424 struct sock *sk) 1425 { 1426 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1427 } 1428 1429 /* Insert new before skb on the write queue of sk. */ 1430 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1431 struct sk_buff *skb, 1432 struct sock *sk) 1433 { 1434 __skb_queue_before(&sk->sk_write_queue, skb, new); 1435 1436 if (sk->sk_send_head == skb) 1437 sk->sk_send_head = new; 1438 } 1439 1440 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1441 { 1442 __skb_unlink(skb, &sk->sk_write_queue); 1443 } 1444 1445 static inline bool tcp_write_queue_empty(struct sock *sk) 1446 { 1447 return skb_queue_empty(&sk->sk_write_queue); 1448 } 1449 1450 static inline void tcp_push_pending_frames(struct sock *sk) 1451 { 1452 if (tcp_send_head(sk)) { 1453 struct tcp_sock *tp = tcp_sk(sk); 1454 1455 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1456 } 1457 } 1458 1459 /* Start sequence of the skb just after the highest skb with SACKed 1460 * bit, valid only if sacked_out > 0 or when the caller has ensured 1461 * validity by itself. 1462 */ 1463 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1464 { 1465 if (!tp->sacked_out) 1466 return tp->snd_una; 1467 1468 if (tp->highest_sack == NULL) 1469 return tp->snd_nxt; 1470 1471 return TCP_SKB_CB(tp->highest_sack)->seq; 1472 } 1473 1474 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1475 { 1476 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1477 tcp_write_queue_next(sk, skb); 1478 } 1479 1480 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1481 { 1482 return tcp_sk(sk)->highest_sack; 1483 } 1484 1485 static inline void tcp_highest_sack_reset(struct sock *sk) 1486 { 1487 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1488 } 1489 1490 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1491 static inline void tcp_highest_sack_combine(struct sock *sk, 1492 struct sk_buff *old, 1493 struct sk_buff *new) 1494 { 1495 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1496 tcp_sk(sk)->highest_sack = new; 1497 } 1498 1499 /* Determines whether this is a thin stream (which may suffer from 1500 * increased latency). Used to trigger latency-reducing mechanisms. 1501 */ 1502 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1503 { 1504 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1505 } 1506 1507 /* /proc */ 1508 enum tcp_seq_states { 1509 TCP_SEQ_STATE_LISTENING, 1510 TCP_SEQ_STATE_OPENREQ, 1511 TCP_SEQ_STATE_ESTABLISHED, 1512 }; 1513 1514 int tcp_seq_open(struct inode *inode, struct file *file); 1515 1516 struct tcp_seq_afinfo { 1517 char *name; 1518 sa_family_t family; 1519 const struct file_operations *seq_fops; 1520 struct seq_operations seq_ops; 1521 }; 1522 1523 struct tcp_iter_state { 1524 struct seq_net_private p; 1525 sa_family_t family; 1526 enum tcp_seq_states state; 1527 struct sock *syn_wait_sk; 1528 int bucket, offset, sbucket, num; 1529 kuid_t uid; 1530 loff_t last_pos; 1531 }; 1532 1533 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1534 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1535 1536 extern struct request_sock_ops tcp_request_sock_ops; 1537 extern struct request_sock_ops tcp6_request_sock_ops; 1538 1539 void tcp_v4_destroy_sock(struct sock *sk); 1540 1541 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1542 netdev_features_t features); 1543 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1544 int tcp_gro_complete(struct sk_buff *skb); 1545 1546 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1547 1548 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1549 { 1550 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat; 1551 } 1552 1553 static inline bool tcp_stream_memory_free(const struct sock *sk) 1554 { 1555 const struct tcp_sock *tp = tcp_sk(sk); 1556 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1557 1558 return notsent_bytes < tcp_notsent_lowat(tp); 1559 } 1560 1561 #ifdef CONFIG_PROC_FS 1562 int tcp4_proc_init(void); 1563 void tcp4_proc_exit(void); 1564 #endif 1565 1566 /* TCP af-specific functions */ 1567 struct tcp_sock_af_ops { 1568 #ifdef CONFIG_TCP_MD5SIG 1569 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1570 struct sock *addr_sk); 1571 int (*calc_md5_hash) (char *location, 1572 struct tcp_md5sig_key *md5, 1573 const struct sock *sk, 1574 const struct request_sock *req, 1575 const struct sk_buff *skb); 1576 int (*md5_parse) (struct sock *sk, 1577 char __user *optval, 1578 int optlen); 1579 #endif 1580 }; 1581 1582 struct tcp_request_sock_ops { 1583 #ifdef CONFIG_TCP_MD5SIG 1584 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1585 struct request_sock *req); 1586 int (*calc_md5_hash) (char *location, 1587 struct tcp_md5sig_key *md5, 1588 const struct sock *sk, 1589 const struct request_sock *req, 1590 const struct sk_buff *skb); 1591 #endif 1592 }; 1593 1594 int tcpv4_offload_init(void); 1595 1596 void tcp_v4_init(void); 1597 void tcp_init(void); 1598 1599 #endif /* _TCP_H */ 1600