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 64 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 128 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 struct timespec now; 495 getnstimeofday(&now); 496 return now.tv_sec >> 6; /* 64 seconds granularity */ 497 } 498 499 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 500 u16 *mssp); 501 __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, __u16 *mss); 502 #else 503 static inline __u32 cookie_v4_init_sequence(struct sock *sk, 504 struct sk_buff *skb, 505 __u16 *mss) 506 { 507 return 0; 508 } 509 #endif 510 511 __u32 cookie_init_timestamp(struct request_sock *req); 512 bool cookie_check_timestamp(struct tcp_options_received *opt, struct net *net, 513 bool *ecn_ok); 514 515 /* From net/ipv6/syncookies.c */ 516 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 517 u32 cookie); 518 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 519 #ifdef CONFIG_SYN_COOKIES 520 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 521 const struct tcphdr *th, u16 *mssp); 522 __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb, 523 __u16 *mss); 524 #else 525 static inline __u32 cookie_v6_init_sequence(struct sock *sk, 526 struct sk_buff *skb, 527 __u16 *mss) 528 { 529 return 0; 530 } 531 #endif 532 /* tcp_output.c */ 533 534 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 535 int nonagle); 536 bool tcp_may_send_now(struct sock *sk); 537 int __tcp_retransmit_skb(struct sock *, struct sk_buff *); 538 int tcp_retransmit_skb(struct sock *, struct sk_buff *); 539 void tcp_retransmit_timer(struct sock *sk); 540 void tcp_xmit_retransmit_queue(struct sock *); 541 void tcp_simple_retransmit(struct sock *); 542 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 543 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 544 545 void tcp_send_probe0(struct sock *); 546 void tcp_send_partial(struct sock *); 547 int tcp_write_wakeup(struct sock *); 548 void tcp_send_fin(struct sock *sk); 549 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 550 int tcp_send_synack(struct sock *); 551 bool tcp_syn_flood_action(struct sock *sk, const struct sk_buff *skb, 552 const char *proto); 553 void tcp_push_one(struct sock *, unsigned int mss_now); 554 void tcp_send_ack(struct sock *sk); 555 void tcp_send_delayed_ack(struct sock *sk); 556 void tcp_send_loss_probe(struct sock *sk); 557 bool tcp_schedule_loss_probe(struct sock *sk); 558 559 /* tcp_input.c */ 560 void tcp_cwnd_application_limited(struct sock *sk); 561 void tcp_resume_early_retransmit(struct sock *sk); 562 void tcp_rearm_rto(struct sock *sk); 563 void tcp_reset(struct sock *sk); 564 565 /* tcp_timer.c */ 566 void tcp_init_xmit_timers(struct sock *); 567 static inline void tcp_clear_xmit_timers(struct sock *sk) 568 { 569 inet_csk_clear_xmit_timers(sk); 570 } 571 572 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 573 unsigned int tcp_current_mss(struct sock *sk); 574 575 /* Bound MSS / TSO packet size with the half of the window */ 576 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 577 { 578 int cutoff; 579 580 /* When peer uses tiny windows, there is no use in packetizing 581 * to sub-MSS pieces for the sake of SWS or making sure there 582 * are enough packets in the pipe for fast recovery. 583 * 584 * On the other hand, for extremely large MSS devices, handling 585 * smaller than MSS windows in this way does make sense. 586 */ 587 if (tp->max_window >= 512) 588 cutoff = (tp->max_window >> 1); 589 else 590 cutoff = tp->max_window; 591 592 if (cutoff && pktsize > cutoff) 593 return max_t(int, cutoff, 68U - tp->tcp_header_len); 594 else 595 return pktsize; 596 } 597 598 /* tcp.c */ 599 void tcp_get_info(const struct sock *, struct tcp_info *); 600 601 /* Read 'sendfile()'-style from a TCP socket */ 602 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 603 unsigned int, size_t); 604 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 605 sk_read_actor_t recv_actor); 606 607 void tcp_initialize_rcv_mss(struct sock *sk); 608 609 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 610 int tcp_mss_to_mtu(struct sock *sk, int mss); 611 void tcp_mtup_init(struct sock *sk); 612 void tcp_init_buffer_space(struct sock *sk); 613 614 static inline void tcp_bound_rto(const struct sock *sk) 615 { 616 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 617 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 618 } 619 620 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 621 { 622 return (tp->srtt >> 3) + tp->rttvar; 623 } 624 625 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 626 { 627 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 628 ntohl(TCP_FLAG_ACK) | 629 snd_wnd); 630 } 631 632 static inline void tcp_fast_path_on(struct tcp_sock *tp) 633 { 634 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 635 } 636 637 static inline void tcp_fast_path_check(struct sock *sk) 638 { 639 struct tcp_sock *tp = tcp_sk(sk); 640 641 if (skb_queue_empty(&tp->out_of_order_queue) && 642 tp->rcv_wnd && 643 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 644 !tp->urg_data) 645 tcp_fast_path_on(tp); 646 } 647 648 /* Compute the actual rto_min value */ 649 static inline u32 tcp_rto_min(struct sock *sk) 650 { 651 const struct dst_entry *dst = __sk_dst_get(sk); 652 u32 rto_min = TCP_RTO_MIN; 653 654 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 655 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 656 return rto_min; 657 } 658 659 /* Compute the actual receive window we are currently advertising. 660 * Rcv_nxt can be after the window if our peer push more data 661 * than the offered window. 662 */ 663 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 664 { 665 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 666 667 if (win < 0) 668 win = 0; 669 return (u32) win; 670 } 671 672 /* Choose a new window, without checks for shrinking, and without 673 * scaling applied to the result. The caller does these things 674 * if necessary. This is a "raw" window selection. 675 */ 676 u32 __tcp_select_window(struct sock *sk); 677 678 void tcp_send_window_probe(struct sock *sk); 679 680 /* TCP timestamps are only 32-bits, this causes a slight 681 * complication on 64-bit systems since we store a snapshot 682 * of jiffies in the buffer control blocks below. We decided 683 * to use only the low 32-bits of jiffies and hide the ugly 684 * casts with the following macro. 685 */ 686 #define tcp_time_stamp ((__u32)(jiffies)) 687 688 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 689 690 #define TCPHDR_FIN 0x01 691 #define TCPHDR_SYN 0x02 692 #define TCPHDR_RST 0x04 693 #define TCPHDR_PSH 0x08 694 #define TCPHDR_ACK 0x10 695 #define TCPHDR_URG 0x20 696 #define TCPHDR_ECE 0x40 697 #define TCPHDR_CWR 0x80 698 699 /* This is what the send packet queuing engine uses to pass 700 * TCP per-packet control information to the transmission code. 701 * We also store the host-order sequence numbers in here too. 702 * This is 44 bytes if IPV6 is enabled. 703 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 704 */ 705 struct tcp_skb_cb { 706 union { 707 struct inet_skb_parm h4; 708 #if IS_ENABLED(CONFIG_IPV6) 709 struct inet6_skb_parm h6; 710 #endif 711 } header; /* For incoming frames */ 712 __u32 seq; /* Starting sequence number */ 713 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 714 __u32 when; /* used to compute rtt's */ 715 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 716 717 __u8 sacked; /* State flags for SACK/FACK. */ 718 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 719 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 720 #define TCPCB_LOST 0x04 /* SKB is lost */ 721 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 722 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 723 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 724 725 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 726 /* 1 byte hole */ 727 __u32 ack_seq; /* Sequence number ACK'd */ 728 }; 729 730 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 731 732 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set 733 * 734 * If we receive a SYN packet with these bits set, it means a network is 735 * playing bad games with TOS bits. In order to avoid possible false congestion 736 * notifications, we disable TCP ECN negociation. 737 */ 738 static inline void 739 TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb, 740 struct net *net) 741 { 742 const struct tcphdr *th = tcp_hdr(skb); 743 744 if (net->ipv4.sysctl_tcp_ecn && th->ece && th->cwr && 745 INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield)) 746 inet_rsk(req)->ecn_ok = 1; 747 } 748 749 /* Due to TSO, an SKB can be composed of multiple actual 750 * packets. To keep these tracked properly, we use this. 751 */ 752 static inline int tcp_skb_pcount(const struct sk_buff *skb) 753 { 754 return skb_shinfo(skb)->gso_segs; 755 } 756 757 /* This is valid iff tcp_skb_pcount() > 1. */ 758 static inline int tcp_skb_mss(const struct sk_buff *skb) 759 { 760 return skb_shinfo(skb)->gso_size; 761 } 762 763 /* Events passed to congestion control interface */ 764 enum tcp_ca_event { 765 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 766 CA_EVENT_CWND_RESTART, /* congestion window restart */ 767 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 768 CA_EVENT_LOSS, /* loss timeout */ 769 CA_EVENT_FAST_ACK, /* in sequence ack */ 770 CA_EVENT_SLOW_ACK, /* other ack */ 771 }; 772 773 /* 774 * Interface for adding new TCP congestion control handlers 775 */ 776 #define TCP_CA_NAME_MAX 16 777 #define TCP_CA_MAX 128 778 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 779 780 #define TCP_CONG_NON_RESTRICTED 0x1 781 #define TCP_CONG_RTT_STAMP 0x2 782 783 struct tcp_congestion_ops { 784 struct list_head list; 785 unsigned long flags; 786 787 /* initialize private data (optional) */ 788 void (*init)(struct sock *sk); 789 /* cleanup private data (optional) */ 790 void (*release)(struct sock *sk); 791 792 /* return slow start threshold (required) */ 793 u32 (*ssthresh)(struct sock *sk); 794 /* lower bound for congestion window (optional) */ 795 u32 (*min_cwnd)(const struct sock *sk); 796 /* do new cwnd calculation (required) */ 797 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked, u32 in_flight); 798 /* call before changing ca_state (optional) */ 799 void (*set_state)(struct sock *sk, u8 new_state); 800 /* call when cwnd event occurs (optional) */ 801 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 802 /* new value of cwnd after loss (optional) */ 803 u32 (*undo_cwnd)(struct sock *sk); 804 /* hook for packet ack accounting (optional) */ 805 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 806 /* get info for inet_diag (optional) */ 807 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 808 809 char name[TCP_CA_NAME_MAX]; 810 struct module *owner; 811 }; 812 813 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 814 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 815 816 void tcp_init_congestion_control(struct sock *sk); 817 void tcp_cleanup_congestion_control(struct sock *sk); 818 int tcp_set_default_congestion_control(const char *name); 819 void tcp_get_default_congestion_control(char *name); 820 void tcp_get_available_congestion_control(char *buf, size_t len); 821 void tcp_get_allowed_congestion_control(char *buf, size_t len); 822 int tcp_set_allowed_congestion_control(char *allowed); 823 int tcp_set_congestion_control(struct sock *sk, const char *name); 824 int tcp_slow_start(struct tcp_sock *tp, u32 acked); 825 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 826 827 extern struct tcp_congestion_ops tcp_init_congestion_ops; 828 u32 tcp_reno_ssthresh(struct sock *sk); 829 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked, u32 in_flight); 830 u32 tcp_reno_min_cwnd(const struct sock *sk); 831 extern struct tcp_congestion_ops tcp_reno; 832 833 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 834 { 835 struct inet_connection_sock *icsk = inet_csk(sk); 836 837 if (icsk->icsk_ca_ops->set_state) 838 icsk->icsk_ca_ops->set_state(sk, ca_state); 839 icsk->icsk_ca_state = ca_state; 840 } 841 842 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 843 { 844 const struct inet_connection_sock *icsk = inet_csk(sk); 845 846 if (icsk->icsk_ca_ops->cwnd_event) 847 icsk->icsk_ca_ops->cwnd_event(sk, event); 848 } 849 850 /* These functions determine how the current flow behaves in respect of SACK 851 * handling. SACK is negotiated with the peer, and therefore it can vary 852 * between different flows. 853 * 854 * tcp_is_sack - SACK enabled 855 * tcp_is_reno - No SACK 856 * tcp_is_fack - FACK enabled, implies SACK enabled 857 */ 858 static inline int tcp_is_sack(const struct tcp_sock *tp) 859 { 860 return tp->rx_opt.sack_ok; 861 } 862 863 static inline bool tcp_is_reno(const struct tcp_sock *tp) 864 { 865 return !tcp_is_sack(tp); 866 } 867 868 static inline bool tcp_is_fack(const struct tcp_sock *tp) 869 { 870 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 871 } 872 873 static inline void tcp_enable_fack(struct tcp_sock *tp) 874 { 875 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 876 } 877 878 /* TCP early-retransmit (ER) is similar to but more conservative than 879 * the thin-dupack feature. Enable ER only if thin-dupack is disabled. 880 */ 881 static inline void tcp_enable_early_retrans(struct tcp_sock *tp) 882 { 883 tp->do_early_retrans = sysctl_tcp_early_retrans && 884 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack && 885 sysctl_tcp_reordering == 3; 886 } 887 888 static inline void tcp_disable_early_retrans(struct tcp_sock *tp) 889 { 890 tp->do_early_retrans = 0; 891 } 892 893 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 894 { 895 return tp->sacked_out + tp->lost_out; 896 } 897 898 /* This determines how many packets are "in the network" to the best 899 * of our knowledge. In many cases it is conservative, but where 900 * detailed information is available from the receiver (via SACK 901 * blocks etc.) we can make more aggressive calculations. 902 * 903 * Use this for decisions involving congestion control, use just 904 * tp->packets_out to determine if the send queue is empty or not. 905 * 906 * Read this equation as: 907 * 908 * "Packets sent once on transmission queue" MINUS 909 * "Packets left network, but not honestly ACKed yet" PLUS 910 * "Packets fast retransmitted" 911 */ 912 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 913 { 914 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 915 } 916 917 #define TCP_INFINITE_SSTHRESH 0x7fffffff 918 919 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 920 { 921 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 922 } 923 924 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 925 { 926 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 927 (1 << inet_csk(sk)->icsk_ca_state); 928 } 929 930 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 931 * The exception is cwnd reduction phase, when cwnd is decreasing towards 932 * ssthresh. 933 */ 934 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 935 { 936 const struct tcp_sock *tp = tcp_sk(sk); 937 938 if (tcp_in_cwnd_reduction(sk)) 939 return tp->snd_ssthresh; 940 else 941 return max(tp->snd_ssthresh, 942 ((tp->snd_cwnd >> 1) + 943 (tp->snd_cwnd >> 2))); 944 } 945 946 /* Use define here intentionally to get WARN_ON location shown at the caller */ 947 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 948 949 void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 950 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 951 952 /* The maximum number of MSS of available cwnd for which TSO defers 953 * sending if not using sysctl_tcp_tso_win_divisor. 954 */ 955 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 956 { 957 return 3; 958 } 959 960 /* Slow start with delack produces 3 packets of burst, so that 961 * it is safe "de facto". This will be the default - same as 962 * the default reordering threshold - but if reordering increases, 963 * we must be able to allow cwnd to burst at least this much in order 964 * to not pull it back when holes are filled. 965 */ 966 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 967 { 968 return tp->reordering; 969 } 970 971 /* Returns end sequence number of the receiver's advertised window */ 972 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 973 { 974 return tp->snd_una + tp->snd_wnd; 975 } 976 bool tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight); 977 978 static inline void tcp_check_probe_timer(struct sock *sk) 979 { 980 const struct tcp_sock *tp = tcp_sk(sk); 981 const struct inet_connection_sock *icsk = inet_csk(sk); 982 983 if (!tp->packets_out && !icsk->icsk_pending) 984 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 985 icsk->icsk_rto, TCP_RTO_MAX); 986 } 987 988 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 989 { 990 tp->snd_wl1 = seq; 991 } 992 993 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 994 { 995 tp->snd_wl1 = seq; 996 } 997 998 /* 999 * Calculate(/check) TCP checksum 1000 */ 1001 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1002 __be32 daddr, __wsum base) 1003 { 1004 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1005 } 1006 1007 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1008 { 1009 return __skb_checksum_complete(skb); 1010 } 1011 1012 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1013 { 1014 return !skb_csum_unnecessary(skb) && 1015 __tcp_checksum_complete(skb); 1016 } 1017 1018 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1019 1020 static inline void tcp_prequeue_init(struct tcp_sock *tp) 1021 { 1022 tp->ucopy.task = NULL; 1023 tp->ucopy.len = 0; 1024 tp->ucopy.memory = 0; 1025 skb_queue_head_init(&tp->ucopy.prequeue); 1026 #ifdef CONFIG_NET_DMA 1027 tp->ucopy.dma_chan = NULL; 1028 tp->ucopy.wakeup = 0; 1029 tp->ucopy.pinned_list = NULL; 1030 tp->ucopy.dma_cookie = 0; 1031 #endif 1032 } 1033 1034 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1035 1036 #undef STATE_TRACE 1037 1038 #ifdef STATE_TRACE 1039 static const char *statename[]={ 1040 "Unused","Established","Syn Sent","Syn Recv", 1041 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1042 "Close Wait","Last ACK","Listen","Closing" 1043 }; 1044 #endif 1045 void tcp_set_state(struct sock *sk, int state); 1046 1047 void tcp_done(struct sock *sk); 1048 1049 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1050 { 1051 rx_opt->dsack = 0; 1052 rx_opt->num_sacks = 0; 1053 } 1054 1055 u32 tcp_default_init_rwnd(u32 mss); 1056 1057 /* Determine a window scaling and initial window to offer. */ 1058 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1059 __u32 *window_clamp, int wscale_ok, 1060 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1061 1062 static inline int tcp_win_from_space(int space) 1063 { 1064 return sysctl_tcp_adv_win_scale<=0 ? 1065 (space>>(-sysctl_tcp_adv_win_scale)) : 1066 space - (space>>sysctl_tcp_adv_win_scale); 1067 } 1068 1069 /* Note: caller must be prepared to deal with negative returns */ 1070 static inline int tcp_space(const struct sock *sk) 1071 { 1072 return tcp_win_from_space(sk->sk_rcvbuf - 1073 atomic_read(&sk->sk_rmem_alloc)); 1074 } 1075 1076 static inline int tcp_full_space(const struct sock *sk) 1077 { 1078 return tcp_win_from_space(sk->sk_rcvbuf); 1079 } 1080 1081 static inline void tcp_openreq_init(struct request_sock *req, 1082 struct tcp_options_received *rx_opt, 1083 struct sk_buff *skb) 1084 { 1085 struct inet_request_sock *ireq = inet_rsk(req); 1086 1087 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 1088 req->cookie_ts = 0; 1089 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 1090 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1; 1091 tcp_rsk(req)->snt_synack = 0; 1092 req->mss = rx_opt->mss_clamp; 1093 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 1094 ireq->tstamp_ok = rx_opt->tstamp_ok; 1095 ireq->sack_ok = rx_opt->sack_ok; 1096 ireq->snd_wscale = rx_opt->snd_wscale; 1097 ireq->wscale_ok = rx_opt->wscale_ok; 1098 ireq->acked = 0; 1099 ireq->ecn_ok = 0; 1100 ireq->ir_rmt_port = tcp_hdr(skb)->source; 1101 ireq->ir_num = ntohs(tcp_hdr(skb)->dest); 1102 } 1103 1104 void tcp_enter_memory_pressure(struct sock *sk); 1105 1106 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1107 { 1108 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1109 } 1110 1111 static inline int keepalive_time_when(const struct tcp_sock *tp) 1112 { 1113 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1114 } 1115 1116 static inline int keepalive_probes(const struct tcp_sock *tp) 1117 { 1118 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1119 } 1120 1121 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1122 { 1123 const struct inet_connection_sock *icsk = &tp->inet_conn; 1124 1125 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1126 tcp_time_stamp - tp->rcv_tstamp); 1127 } 1128 1129 static inline int tcp_fin_time(const struct sock *sk) 1130 { 1131 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1132 const int rto = inet_csk(sk)->icsk_rto; 1133 1134 if (fin_timeout < (rto << 2) - (rto >> 1)) 1135 fin_timeout = (rto << 2) - (rto >> 1); 1136 1137 return fin_timeout; 1138 } 1139 1140 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1141 int paws_win) 1142 { 1143 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1144 return true; 1145 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1146 return true; 1147 /* 1148 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1149 * then following tcp messages have valid values. Ignore 0 value, 1150 * or else 'negative' tsval might forbid us to accept their packets. 1151 */ 1152 if (!rx_opt->ts_recent) 1153 return true; 1154 return false; 1155 } 1156 1157 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1158 int rst) 1159 { 1160 if (tcp_paws_check(rx_opt, 0)) 1161 return false; 1162 1163 /* RST segments are not recommended to carry timestamp, 1164 and, if they do, it is recommended to ignore PAWS because 1165 "their cleanup function should take precedence over timestamps." 1166 Certainly, it is mistake. It is necessary to understand the reasons 1167 of this constraint to relax it: if peer reboots, clock may go 1168 out-of-sync and half-open connections will not be reset. 1169 Actually, the problem would be not existing if all 1170 the implementations followed draft about maintaining clock 1171 via reboots. Linux-2.2 DOES NOT! 1172 1173 However, we can relax time bounds for RST segments to MSL. 1174 */ 1175 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1176 return false; 1177 return true; 1178 } 1179 1180 static inline void tcp_mib_init(struct net *net) 1181 { 1182 /* See RFC 2012 */ 1183 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1184 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1185 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1186 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1187 } 1188 1189 /* from STCP */ 1190 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1191 { 1192 tp->lost_skb_hint = NULL; 1193 } 1194 1195 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1196 { 1197 tcp_clear_retrans_hints_partial(tp); 1198 tp->retransmit_skb_hint = NULL; 1199 } 1200 1201 /* MD5 Signature */ 1202 struct crypto_hash; 1203 1204 union tcp_md5_addr { 1205 struct in_addr a4; 1206 #if IS_ENABLED(CONFIG_IPV6) 1207 struct in6_addr a6; 1208 #endif 1209 }; 1210 1211 /* - key database */ 1212 struct tcp_md5sig_key { 1213 struct hlist_node node; 1214 u8 keylen; 1215 u8 family; /* AF_INET or AF_INET6 */ 1216 union tcp_md5_addr addr; 1217 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1218 struct rcu_head rcu; 1219 }; 1220 1221 /* - sock block */ 1222 struct tcp_md5sig_info { 1223 struct hlist_head head; 1224 struct rcu_head rcu; 1225 }; 1226 1227 /* - pseudo header */ 1228 struct tcp4_pseudohdr { 1229 __be32 saddr; 1230 __be32 daddr; 1231 __u8 pad; 1232 __u8 protocol; 1233 __be16 len; 1234 }; 1235 1236 struct tcp6_pseudohdr { 1237 struct in6_addr saddr; 1238 struct in6_addr daddr; 1239 __be32 len; 1240 __be32 protocol; /* including padding */ 1241 }; 1242 1243 union tcp_md5sum_block { 1244 struct tcp4_pseudohdr ip4; 1245 #if IS_ENABLED(CONFIG_IPV6) 1246 struct tcp6_pseudohdr ip6; 1247 #endif 1248 }; 1249 1250 /* - pool: digest algorithm, hash description and scratch buffer */ 1251 struct tcp_md5sig_pool { 1252 struct hash_desc md5_desc; 1253 union tcp_md5sum_block md5_blk; 1254 }; 1255 1256 /* - functions */ 1257 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1258 const struct sock *sk, const struct request_sock *req, 1259 const struct sk_buff *skb); 1260 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1261 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); 1262 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1263 int family); 1264 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1265 struct sock *addr_sk); 1266 1267 #ifdef CONFIG_TCP_MD5SIG 1268 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1269 const union tcp_md5_addr *addr, 1270 int family); 1271 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1272 #else 1273 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1274 const union tcp_md5_addr *addr, 1275 int family) 1276 { 1277 return NULL; 1278 } 1279 #define tcp_twsk_md5_key(twsk) NULL 1280 #endif 1281 1282 bool tcp_alloc_md5sig_pool(void); 1283 1284 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1285 static inline void tcp_put_md5sig_pool(void) 1286 { 1287 local_bh_enable(); 1288 } 1289 1290 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1291 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1292 unsigned int header_len); 1293 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1294 const struct tcp_md5sig_key *key); 1295 1296 /* From tcp_fastopen.c */ 1297 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1298 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1299 unsigned long *last_syn_loss); 1300 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1301 struct tcp_fastopen_cookie *cookie, bool syn_lost); 1302 struct tcp_fastopen_request { 1303 /* Fast Open cookie. Size 0 means a cookie request */ 1304 struct tcp_fastopen_cookie cookie; 1305 struct msghdr *data; /* data in MSG_FASTOPEN */ 1306 u16 copied; /* queued in tcp_connect() */ 1307 }; 1308 void tcp_free_fastopen_req(struct tcp_sock *tp); 1309 1310 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1311 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1312 void tcp_fastopen_cookie_gen(__be32 src, __be32 dst, 1313 struct tcp_fastopen_cookie *foc); 1314 void tcp_fastopen_init_key_once(bool publish); 1315 #define TCP_FASTOPEN_KEY_LENGTH 16 1316 1317 /* Fastopen key context */ 1318 struct tcp_fastopen_context { 1319 struct crypto_cipher *tfm; 1320 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1321 struct rcu_head rcu; 1322 }; 1323 1324 /* write queue abstraction */ 1325 static inline void tcp_write_queue_purge(struct sock *sk) 1326 { 1327 struct sk_buff *skb; 1328 1329 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1330 sk_wmem_free_skb(sk, skb); 1331 sk_mem_reclaim(sk); 1332 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1333 } 1334 1335 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1336 { 1337 return skb_peek(&sk->sk_write_queue); 1338 } 1339 1340 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1341 { 1342 return skb_peek_tail(&sk->sk_write_queue); 1343 } 1344 1345 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1346 const struct sk_buff *skb) 1347 { 1348 return skb_queue_next(&sk->sk_write_queue, skb); 1349 } 1350 1351 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1352 const struct sk_buff *skb) 1353 { 1354 return skb_queue_prev(&sk->sk_write_queue, skb); 1355 } 1356 1357 #define tcp_for_write_queue(skb, sk) \ 1358 skb_queue_walk(&(sk)->sk_write_queue, skb) 1359 1360 #define tcp_for_write_queue_from(skb, sk) \ 1361 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1362 1363 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1364 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1365 1366 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1367 { 1368 return sk->sk_send_head; 1369 } 1370 1371 static inline bool tcp_skb_is_last(const struct sock *sk, 1372 const struct sk_buff *skb) 1373 { 1374 return skb_queue_is_last(&sk->sk_write_queue, skb); 1375 } 1376 1377 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1378 { 1379 if (tcp_skb_is_last(sk, skb)) 1380 sk->sk_send_head = NULL; 1381 else 1382 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1383 } 1384 1385 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1386 { 1387 if (sk->sk_send_head == skb_unlinked) 1388 sk->sk_send_head = NULL; 1389 } 1390 1391 static inline void tcp_init_send_head(struct sock *sk) 1392 { 1393 sk->sk_send_head = NULL; 1394 } 1395 1396 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1397 { 1398 __skb_queue_tail(&sk->sk_write_queue, skb); 1399 } 1400 1401 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1402 { 1403 __tcp_add_write_queue_tail(sk, skb); 1404 1405 /* Queue it, remembering where we must start sending. */ 1406 if (sk->sk_send_head == NULL) { 1407 sk->sk_send_head = skb; 1408 1409 if (tcp_sk(sk)->highest_sack == NULL) 1410 tcp_sk(sk)->highest_sack = skb; 1411 } 1412 } 1413 1414 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1415 { 1416 __skb_queue_head(&sk->sk_write_queue, skb); 1417 } 1418 1419 /* Insert buff after skb on the write queue of sk. */ 1420 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1421 struct sk_buff *buff, 1422 struct sock *sk) 1423 { 1424 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1425 } 1426 1427 /* Insert new before skb on the write queue of sk. */ 1428 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1429 struct sk_buff *skb, 1430 struct sock *sk) 1431 { 1432 __skb_queue_before(&sk->sk_write_queue, skb, new); 1433 1434 if (sk->sk_send_head == skb) 1435 sk->sk_send_head = new; 1436 } 1437 1438 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1439 { 1440 __skb_unlink(skb, &sk->sk_write_queue); 1441 } 1442 1443 static inline bool tcp_write_queue_empty(struct sock *sk) 1444 { 1445 return skb_queue_empty(&sk->sk_write_queue); 1446 } 1447 1448 static inline void tcp_push_pending_frames(struct sock *sk) 1449 { 1450 if (tcp_send_head(sk)) { 1451 struct tcp_sock *tp = tcp_sk(sk); 1452 1453 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1454 } 1455 } 1456 1457 /* Start sequence of the skb just after the highest skb with SACKed 1458 * bit, valid only if sacked_out > 0 or when the caller has ensured 1459 * validity by itself. 1460 */ 1461 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1462 { 1463 if (!tp->sacked_out) 1464 return tp->snd_una; 1465 1466 if (tp->highest_sack == NULL) 1467 return tp->snd_nxt; 1468 1469 return TCP_SKB_CB(tp->highest_sack)->seq; 1470 } 1471 1472 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1473 { 1474 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1475 tcp_write_queue_next(sk, skb); 1476 } 1477 1478 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1479 { 1480 return tcp_sk(sk)->highest_sack; 1481 } 1482 1483 static inline void tcp_highest_sack_reset(struct sock *sk) 1484 { 1485 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1486 } 1487 1488 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1489 static inline void tcp_highest_sack_combine(struct sock *sk, 1490 struct sk_buff *old, 1491 struct sk_buff *new) 1492 { 1493 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1494 tcp_sk(sk)->highest_sack = new; 1495 } 1496 1497 /* Determines whether this is a thin stream (which may suffer from 1498 * increased latency). Used to trigger latency-reducing mechanisms. 1499 */ 1500 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1501 { 1502 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1503 } 1504 1505 /* /proc */ 1506 enum tcp_seq_states { 1507 TCP_SEQ_STATE_LISTENING, 1508 TCP_SEQ_STATE_OPENREQ, 1509 TCP_SEQ_STATE_ESTABLISHED, 1510 }; 1511 1512 int tcp_seq_open(struct inode *inode, struct file *file); 1513 1514 struct tcp_seq_afinfo { 1515 char *name; 1516 sa_family_t family; 1517 const struct file_operations *seq_fops; 1518 struct seq_operations seq_ops; 1519 }; 1520 1521 struct tcp_iter_state { 1522 struct seq_net_private p; 1523 sa_family_t family; 1524 enum tcp_seq_states state; 1525 struct sock *syn_wait_sk; 1526 int bucket, offset, sbucket, num; 1527 kuid_t uid; 1528 loff_t last_pos; 1529 }; 1530 1531 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1532 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1533 1534 extern struct request_sock_ops tcp_request_sock_ops; 1535 extern struct request_sock_ops tcp6_request_sock_ops; 1536 1537 void tcp_v4_destroy_sock(struct sock *sk); 1538 1539 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1540 netdev_features_t features); 1541 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1542 int tcp_gro_complete(struct sk_buff *skb); 1543 1544 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1545 1546 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1547 { 1548 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat; 1549 } 1550 1551 static inline bool tcp_stream_memory_free(const struct sock *sk) 1552 { 1553 const struct tcp_sock *tp = tcp_sk(sk); 1554 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1555 1556 return notsent_bytes < tcp_notsent_lowat(tp); 1557 } 1558 1559 #ifdef CONFIG_PROC_FS 1560 int tcp4_proc_init(void); 1561 void tcp4_proc_exit(void); 1562 #endif 1563 1564 /* TCP af-specific functions */ 1565 struct tcp_sock_af_ops { 1566 #ifdef CONFIG_TCP_MD5SIG 1567 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1568 struct sock *addr_sk); 1569 int (*calc_md5_hash) (char *location, 1570 struct tcp_md5sig_key *md5, 1571 const struct sock *sk, 1572 const struct request_sock *req, 1573 const struct sk_buff *skb); 1574 int (*md5_parse) (struct sock *sk, 1575 char __user *optval, 1576 int optlen); 1577 #endif 1578 }; 1579 1580 struct tcp_request_sock_ops { 1581 #ifdef CONFIG_TCP_MD5SIG 1582 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1583 struct request_sock *req); 1584 int (*calc_md5_hash) (char *location, 1585 struct tcp_md5sig_key *md5, 1586 const struct sock *sk, 1587 const struct request_sock *req, 1588 const struct sk_buff *skb); 1589 #endif 1590 }; 1591 1592 int tcpv4_offload_init(void); 1593 1594 void tcp_v4_init(void); 1595 void tcp_init(void); 1596 1597 #endif /* _TCP_H */ 1598