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 #include <linux/ktime.h> 35 36 #include <net/inet_connection_sock.h> 37 #include <net/inet_timewait_sock.h> 38 #include <net/inet_hashtables.h> 39 #include <net/checksum.h> 40 #include <net/request_sock.h> 41 #include <net/sock.h> 42 #include <net/snmp.h> 43 #include <net/ip.h> 44 #include <net/tcp_states.h> 45 #include <net/inet_ecn.h> 46 #include <net/dst.h> 47 48 #include <linux/seq_file.h> 49 #include <linux/memcontrol.h> 50 51 extern struct inet_hashinfo tcp_hashinfo; 52 53 extern struct percpu_counter tcp_orphan_count; 54 void tcp_time_wait(struct sock *sk, int state, int timeo); 55 56 #define MAX_TCP_HEADER (128 + MAX_HEADER) 57 #define MAX_TCP_OPTION_SPACE 40 58 59 /* 60 * Never offer a window over 32767 without using window scaling. Some 61 * poor stacks do signed 16bit maths! 62 */ 63 #define MAX_TCP_WINDOW 32767U 64 65 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 66 #define TCP_MIN_MSS 88U 67 68 /* The least MTU to use for probing */ 69 #define TCP_BASE_MSS 512 70 71 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 72 #define TCP_FASTRETRANS_THRESH 3 73 74 /* Maximal reordering. */ 75 #define TCP_MAX_REORDERING 127 76 77 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 78 #define TCP_MAX_QUICKACKS 16U 79 80 /* urg_data states */ 81 #define TCP_URG_VALID 0x0100 82 #define TCP_URG_NOTYET 0x0200 83 #define TCP_URG_READ 0x0400 84 85 #define TCP_RETR1 3 /* 86 * This is how many retries it does before it 87 * tries to figure out if the gateway is 88 * down. Minimal RFC value is 3; it corresponds 89 * to ~3sec-8min depending on RTO. 90 */ 91 92 #define TCP_RETR2 15 /* 93 * This should take at least 94 * 90 minutes to time out. 95 * RFC1122 says that the limit is 100 sec. 96 * 15 is ~13-30min depending on RTO. 97 */ 98 99 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 100 * when active opening a connection. 101 * RFC1122 says the minimum retry MUST 102 * be at least 180secs. Nevertheless 103 * this value is corresponding to 104 * 63secs of retransmission with the 105 * current initial RTO. 106 */ 107 108 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 109 * when passive opening a connection. 110 * This is corresponding to 31secs of 111 * retransmission with the current 112 * initial RTO. 113 */ 114 115 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 116 * state, about 60 seconds */ 117 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 118 /* BSD style FIN_WAIT2 deadlock breaker. 119 * It used to be 3min, new value is 60sec, 120 * to combine FIN-WAIT-2 timeout with 121 * TIME-WAIT timer. 122 */ 123 124 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 125 #if HZ >= 100 126 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 127 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 128 #else 129 #define TCP_DELACK_MIN 4U 130 #define TCP_ATO_MIN 4U 131 #endif 132 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 133 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 134 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 135 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 136 * used as a fallback RTO for the 137 * initial data transmission if no 138 * valid RTT sample has been acquired, 139 * most likely due to retrans in 3WHS. 140 */ 141 142 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 143 * for local resources. 144 */ 145 146 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 147 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 148 #define TCP_KEEPALIVE_INTVL (75*HZ) 149 150 #define MAX_TCP_KEEPIDLE 32767 151 #define MAX_TCP_KEEPINTVL 32767 152 #define MAX_TCP_KEEPCNT 127 153 #define MAX_TCP_SYNCNT 127 154 155 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 156 157 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 158 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 159 * after this time. It should be equal 160 * (or greater than) TCP_TIMEWAIT_LEN 161 * to provide reliability equal to one 162 * provided by timewait state. 163 */ 164 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 165 * timestamps. It must be less than 166 * minimal timewait lifetime. 167 */ 168 /* 169 * TCP option 170 */ 171 172 #define TCPOPT_NOP 1 /* Padding */ 173 #define TCPOPT_EOL 0 /* End of options */ 174 #define TCPOPT_MSS 2 /* Segment size negotiating */ 175 #define TCPOPT_WINDOW 3 /* Window scaling */ 176 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 177 #define TCPOPT_SACK 5 /* SACK Block */ 178 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 179 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 180 #define TCPOPT_EXP 254 /* Experimental */ 181 /* Magic number to be after the option value for sharing TCP 182 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 183 */ 184 #define TCPOPT_FASTOPEN_MAGIC 0xF989 185 186 /* 187 * TCP option lengths 188 */ 189 190 #define TCPOLEN_MSS 4 191 #define TCPOLEN_WINDOW 3 192 #define TCPOLEN_SACK_PERM 2 193 #define TCPOLEN_TIMESTAMP 10 194 #define TCPOLEN_MD5SIG 18 195 #define TCPOLEN_EXP_FASTOPEN_BASE 4 196 197 /* But this is what stacks really send out. */ 198 #define TCPOLEN_TSTAMP_ALIGNED 12 199 #define TCPOLEN_WSCALE_ALIGNED 4 200 #define TCPOLEN_SACKPERM_ALIGNED 4 201 #define TCPOLEN_SACK_BASE 2 202 #define TCPOLEN_SACK_BASE_ALIGNED 4 203 #define TCPOLEN_SACK_PERBLOCK 8 204 #define TCPOLEN_MD5SIG_ALIGNED 20 205 #define TCPOLEN_MSS_ALIGNED 4 206 207 /* Flags in tp->nonagle */ 208 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 209 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 210 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 211 212 /* TCP thin-stream limits */ 213 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 214 215 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */ 216 #define TCP_INIT_CWND 10 217 218 /* Bit Flags for sysctl_tcp_fastopen */ 219 #define TFO_CLIENT_ENABLE 1 220 #define TFO_SERVER_ENABLE 2 221 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 222 223 /* Process SYN data but skip cookie validation */ 224 #define TFO_SERVER_COOKIE_NOT_CHKED 0x100 225 /* Accept SYN data w/o any cookie option */ 226 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 227 228 /* Force enable TFO on all listeners, i.e., not requiring the 229 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen. 230 */ 231 #define TFO_SERVER_WO_SOCKOPT1 0x400 232 #define TFO_SERVER_WO_SOCKOPT2 0x800 233 /* Always create TFO child sockets on a TFO listener even when 234 * cookie/data not present. (For testing purpose!) 235 */ 236 #define TFO_SERVER_ALWAYS 0x1000 237 238 extern struct inet_timewait_death_row tcp_death_row; 239 240 /* sysctl variables for tcp */ 241 extern int sysctl_tcp_timestamps; 242 extern int sysctl_tcp_window_scaling; 243 extern int sysctl_tcp_sack; 244 extern int sysctl_tcp_fin_timeout; 245 extern int sysctl_tcp_keepalive_time; 246 extern int sysctl_tcp_keepalive_probes; 247 extern int sysctl_tcp_keepalive_intvl; 248 extern int sysctl_tcp_syn_retries; 249 extern int sysctl_tcp_synack_retries; 250 extern int sysctl_tcp_retries1; 251 extern int sysctl_tcp_retries2; 252 extern int sysctl_tcp_orphan_retries; 253 extern int sysctl_tcp_syncookies; 254 extern int sysctl_tcp_fastopen; 255 extern int sysctl_tcp_retrans_collapse; 256 extern int sysctl_tcp_stdurg; 257 extern int sysctl_tcp_rfc1337; 258 extern int sysctl_tcp_abort_on_overflow; 259 extern int sysctl_tcp_max_orphans; 260 extern int sysctl_tcp_fack; 261 extern int sysctl_tcp_reordering; 262 extern int sysctl_tcp_dsack; 263 extern long sysctl_tcp_mem[3]; 264 extern int sysctl_tcp_wmem[3]; 265 extern int sysctl_tcp_rmem[3]; 266 extern int sysctl_tcp_app_win; 267 extern int sysctl_tcp_adv_win_scale; 268 extern int sysctl_tcp_tw_reuse; 269 extern int sysctl_tcp_frto; 270 extern int sysctl_tcp_low_latency; 271 extern int sysctl_tcp_dma_copybreak; 272 extern int sysctl_tcp_nometrics_save; 273 extern int sysctl_tcp_moderate_rcvbuf; 274 extern int sysctl_tcp_tso_win_divisor; 275 extern int sysctl_tcp_mtu_probing; 276 extern int sysctl_tcp_base_mss; 277 extern int sysctl_tcp_workaround_signed_windows; 278 extern int sysctl_tcp_slow_start_after_idle; 279 extern int sysctl_tcp_thin_linear_timeouts; 280 extern int sysctl_tcp_thin_dupack; 281 extern int sysctl_tcp_early_retrans; 282 extern int sysctl_tcp_limit_output_bytes; 283 extern int sysctl_tcp_challenge_ack_limit; 284 extern unsigned int sysctl_tcp_notsent_lowat; 285 extern int sysctl_tcp_min_tso_segs; 286 extern int sysctl_tcp_autocorking; 287 288 extern atomic_long_t tcp_memory_allocated; 289 extern struct percpu_counter tcp_sockets_allocated; 290 extern int tcp_memory_pressure; 291 292 /* 293 * The next routines deal with comparing 32 bit unsigned ints 294 * and worry about wraparound (automatic with unsigned arithmetic). 295 */ 296 297 static inline bool before(__u32 seq1, __u32 seq2) 298 { 299 return (__s32)(seq1-seq2) < 0; 300 } 301 #define after(seq2, seq1) before(seq1, seq2) 302 303 /* is s2<=s1<=s3 ? */ 304 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 305 { 306 return seq3 - seq2 >= seq1 - seq2; 307 } 308 309 static inline bool tcp_out_of_memory(struct sock *sk) 310 { 311 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 312 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 313 return true; 314 return false; 315 } 316 317 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 318 { 319 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 320 int orphans = percpu_counter_read_positive(ocp); 321 322 if (orphans << shift > sysctl_tcp_max_orphans) { 323 orphans = percpu_counter_sum_positive(ocp); 324 if (orphans << shift > sysctl_tcp_max_orphans) 325 return true; 326 } 327 return false; 328 } 329 330 bool tcp_check_oom(struct sock *sk, int shift); 331 332 /* syncookies: remember time of last synqueue overflow */ 333 static inline void tcp_synq_overflow(struct sock *sk) 334 { 335 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies; 336 } 337 338 /* syncookies: no recent synqueue overflow on this listening socket? */ 339 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 340 { 341 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 342 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK); 343 } 344 345 extern struct proto tcp_prot; 346 347 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 348 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 349 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 350 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 351 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 352 353 void tcp_tasklet_init(void); 354 355 void tcp_v4_err(struct sk_buff *skb, u32); 356 357 void tcp_shutdown(struct sock *sk, int how); 358 359 void tcp_v4_early_demux(struct sk_buff *skb); 360 int tcp_v4_rcv(struct sk_buff *skb); 361 362 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 363 int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 364 size_t size); 365 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 366 int flags); 367 void tcp_release_cb(struct sock *sk); 368 void tcp_wfree(struct sk_buff *skb); 369 void tcp_write_timer_handler(struct sock *sk); 370 void tcp_delack_timer_handler(struct sock *sk); 371 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 372 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 373 const struct tcphdr *th, unsigned int len); 374 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 375 const struct tcphdr *th, unsigned int len); 376 void tcp_rcv_space_adjust(struct sock *sk); 377 void tcp_cleanup_rbuf(struct sock *sk, int copied); 378 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 379 void tcp_twsk_destructor(struct sock *sk); 380 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 381 struct pipe_inode_info *pipe, size_t len, 382 unsigned int flags); 383 384 static inline void tcp_dec_quickack_mode(struct sock *sk, 385 const unsigned int pkts) 386 { 387 struct inet_connection_sock *icsk = inet_csk(sk); 388 389 if (icsk->icsk_ack.quick) { 390 if (pkts >= icsk->icsk_ack.quick) { 391 icsk->icsk_ack.quick = 0; 392 /* Leaving quickack mode we deflate ATO. */ 393 icsk->icsk_ack.ato = TCP_ATO_MIN; 394 } else 395 icsk->icsk_ack.quick -= pkts; 396 } 397 } 398 399 #define TCP_ECN_OK 1 400 #define TCP_ECN_QUEUE_CWR 2 401 #define TCP_ECN_DEMAND_CWR 4 402 #define TCP_ECN_SEEN 8 403 404 enum tcp_tw_status { 405 TCP_TW_SUCCESS = 0, 406 TCP_TW_RST = 1, 407 TCP_TW_ACK = 2, 408 TCP_TW_SYN = 3 409 }; 410 411 412 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 413 struct sk_buff *skb, 414 const struct tcphdr *th); 415 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 416 struct request_sock *req, struct request_sock **prev, 417 bool fastopen); 418 int tcp_child_process(struct sock *parent, struct sock *child, 419 struct sk_buff *skb); 420 void tcp_enter_loss(struct sock *sk, int how); 421 void tcp_clear_retrans(struct tcp_sock *tp); 422 void tcp_update_metrics(struct sock *sk); 423 void tcp_init_metrics(struct sock *sk); 424 void tcp_metrics_init(void); 425 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, 426 bool paws_check); 427 bool tcp_remember_stamp(struct sock *sk); 428 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); 429 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); 430 void tcp_disable_fack(struct tcp_sock *tp); 431 void tcp_close(struct sock *sk, long timeout); 432 void tcp_init_sock(struct sock *sk); 433 unsigned int tcp_poll(struct file *file, struct socket *sock, 434 struct poll_table_struct *wait); 435 int tcp_getsockopt(struct sock *sk, int level, int optname, 436 char __user *optval, int __user *optlen); 437 int tcp_setsockopt(struct sock *sk, int level, int optname, 438 char __user *optval, unsigned int optlen); 439 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 440 char __user *optval, int __user *optlen); 441 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 442 char __user *optval, unsigned int optlen); 443 void tcp_set_keepalive(struct sock *sk, int val); 444 void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req); 445 int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 446 size_t len, int nonblock, int flags, int *addr_len); 447 void tcp_parse_options(const struct sk_buff *skb, 448 struct tcp_options_received *opt_rx, 449 int estab, struct tcp_fastopen_cookie *foc); 450 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 451 452 /* 453 * TCP v4 functions exported for the inet6 API 454 */ 455 456 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 457 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 458 struct sock *tcp_create_openreq_child(struct sock *sk, 459 struct request_sock *req, 460 struct sk_buff *skb); 461 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 462 struct request_sock *req, 463 struct dst_entry *dst); 464 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 465 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 466 int tcp_connect(struct sock *sk); 467 struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst, 468 struct request_sock *req, 469 struct tcp_fastopen_cookie *foc); 470 int tcp_disconnect(struct sock *sk, int flags); 471 472 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 473 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 474 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 475 476 /* From syncookies.c */ 477 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 478 u32 cookie); 479 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 480 struct ip_options *opt); 481 #ifdef CONFIG_SYN_COOKIES 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_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 usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 623 } 624 625 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 626 { 627 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 628 ntohl(TCP_FLAG_ACK) | 629 snd_wnd); 630 } 631 632 static inline void tcp_fast_path_on(struct tcp_sock *tp) 633 { 634 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 635 } 636 637 static inline void tcp_fast_path_check(struct sock *sk) 638 { 639 struct tcp_sock *tp = tcp_sk(sk); 640 641 if (skb_queue_empty(&tp->out_of_order_queue) && 642 tp->rcv_wnd && 643 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 644 !tp->urg_data) 645 tcp_fast_path_on(tp); 646 } 647 648 /* Compute the actual rto_min value */ 649 static inline u32 tcp_rto_min(struct sock *sk) 650 { 651 const struct dst_entry *dst = __sk_dst_get(sk); 652 u32 rto_min = TCP_RTO_MIN; 653 654 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 655 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 656 return rto_min; 657 } 658 659 static inline u32 tcp_rto_min_us(struct sock *sk) 660 { 661 return jiffies_to_usecs(tcp_rto_min(sk)); 662 } 663 664 /* Compute the actual receive window we are currently advertising. 665 * Rcv_nxt can be after the window if our peer push more data 666 * than the offered window. 667 */ 668 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 669 { 670 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 671 672 if (win < 0) 673 win = 0; 674 return (u32) win; 675 } 676 677 /* Choose a new window, without checks for shrinking, and without 678 * scaling applied to the result. The caller does these things 679 * if necessary. This is a "raw" window selection. 680 */ 681 u32 __tcp_select_window(struct sock *sk); 682 683 void tcp_send_window_probe(struct sock *sk); 684 685 /* TCP timestamps are only 32-bits, this causes a slight 686 * complication on 64-bit systems since we store a snapshot 687 * of jiffies in the buffer control blocks below. We decided 688 * to use only the low 32-bits of jiffies and hide the ugly 689 * casts with the following macro. 690 */ 691 #define tcp_time_stamp ((__u32)(jiffies)) 692 693 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 694 695 #define TCPHDR_FIN 0x01 696 #define TCPHDR_SYN 0x02 697 #define TCPHDR_RST 0x04 698 #define TCPHDR_PSH 0x08 699 #define TCPHDR_ACK 0x10 700 #define TCPHDR_URG 0x20 701 #define TCPHDR_ECE 0x40 702 #define TCPHDR_CWR 0x80 703 704 /* This is what the send packet queuing engine uses to pass 705 * TCP per-packet control information to the transmission code. 706 * We also store the host-order sequence numbers in here too. 707 * This is 44 bytes if IPV6 is enabled. 708 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 709 */ 710 struct tcp_skb_cb { 711 union { 712 struct inet_skb_parm h4; 713 #if IS_ENABLED(CONFIG_IPV6) 714 struct inet6_skb_parm h6; 715 #endif 716 } header; /* For incoming frames */ 717 __u32 seq; /* Starting sequence number */ 718 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 719 __u32 when; /* used to compute rtt's */ 720 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 721 722 __u8 sacked; /* State flags for SACK/FACK. */ 723 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 724 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 725 #define TCPCB_LOST 0x04 /* SKB is lost */ 726 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 727 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 728 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 729 730 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 731 /* 1 byte hole */ 732 __u32 ack_seq; /* Sequence number ACK'd */ 733 }; 734 735 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 736 737 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set 738 * 739 * If we receive a SYN packet with these bits set, it means a network is 740 * playing bad games with TOS bits. In order to avoid possible false congestion 741 * notifications, we disable TCP ECN negociation. 742 */ 743 static inline void 744 TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb, 745 struct net *net) 746 { 747 const struct tcphdr *th = tcp_hdr(skb); 748 749 if (net->ipv4.sysctl_tcp_ecn && th->ece && th->cwr && 750 INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield)) 751 inet_rsk(req)->ecn_ok = 1; 752 } 753 754 /* Due to TSO, an SKB can be composed of multiple actual 755 * packets. To keep these tracked properly, we use this. 756 */ 757 static inline int tcp_skb_pcount(const struct sk_buff *skb) 758 { 759 return skb_shinfo(skb)->gso_segs; 760 } 761 762 /* This is valid iff tcp_skb_pcount() > 1. */ 763 static inline int tcp_skb_mss(const struct sk_buff *skb) 764 { 765 return skb_shinfo(skb)->gso_size; 766 } 767 768 /* Events passed to congestion control interface */ 769 enum tcp_ca_event { 770 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 771 CA_EVENT_CWND_RESTART, /* congestion window restart */ 772 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 773 CA_EVENT_LOSS, /* loss timeout */ 774 CA_EVENT_FAST_ACK, /* in sequence ack */ 775 CA_EVENT_SLOW_ACK, /* other ack */ 776 }; 777 778 /* 779 * Interface for adding new TCP congestion control handlers 780 */ 781 #define TCP_CA_NAME_MAX 16 782 #define TCP_CA_MAX 128 783 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 784 785 #define TCP_CONG_NON_RESTRICTED 0x1 786 787 struct tcp_congestion_ops { 788 struct list_head list; 789 unsigned long flags; 790 791 /* initialize private data (optional) */ 792 void (*init)(struct sock *sk); 793 /* cleanup private data (optional) */ 794 void (*release)(struct sock *sk); 795 796 /* return slow start threshold (required) */ 797 u32 (*ssthresh)(struct sock *sk); 798 /* do new cwnd calculation (required) */ 799 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 800 /* call before changing ca_state (optional) */ 801 void (*set_state)(struct sock *sk, u8 new_state); 802 /* call when cwnd event occurs (optional) */ 803 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 804 /* new value of cwnd after loss (optional) */ 805 u32 (*undo_cwnd)(struct sock *sk); 806 /* hook for packet ack accounting (optional) */ 807 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 808 /* get info for inet_diag (optional) */ 809 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 810 811 char name[TCP_CA_NAME_MAX]; 812 struct module *owner; 813 }; 814 815 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 816 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 817 818 void tcp_init_congestion_control(struct sock *sk); 819 void tcp_cleanup_congestion_control(struct sock *sk); 820 int tcp_set_default_congestion_control(const char *name); 821 void tcp_get_default_congestion_control(char *name); 822 void tcp_get_available_congestion_control(char *buf, size_t len); 823 void tcp_get_allowed_congestion_control(char *buf, size_t len); 824 int tcp_set_allowed_congestion_control(char *allowed); 825 int tcp_set_congestion_control(struct sock *sk, const char *name); 826 int tcp_slow_start(struct tcp_sock *tp, u32 acked); 827 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 828 829 extern struct tcp_congestion_ops tcp_init_congestion_ops; 830 u32 tcp_reno_ssthresh(struct sock *sk); 831 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 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 } 977 978 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 979 * flexible approach. The RFC suggests cwnd should not be raised unless 980 * it was fully used previously. But we allow cwnd to grow as long as the 981 * application has used half the cwnd. 982 * Example : 983 * cwnd is 10 (IW10), but application sends 9 frames. 984 * We allow cwnd to reach 18 when all frames are ACKed. 985 * This check is safe because it's as aggressive as slow start which already 986 * risks 100% overshoot. The advantage is that we discourage application to 987 * either send more filler packets or data to artificially blow up the cwnd 988 * usage, and allow application-limited process to probe bw more aggressively. 989 */ 990 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 991 { 992 const struct tcp_sock *tp = tcp_sk(sk); 993 994 return tp->snd_cwnd < 2 * tp->lsnd_pending; 995 } 996 997 static inline void tcp_check_probe_timer(struct sock *sk) 998 { 999 const struct tcp_sock *tp = tcp_sk(sk); 1000 const struct inet_connection_sock *icsk = inet_csk(sk); 1001 1002 if (!tp->packets_out && !icsk->icsk_pending) 1003 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1004 icsk->icsk_rto, TCP_RTO_MAX); 1005 } 1006 1007 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1008 { 1009 tp->snd_wl1 = seq; 1010 } 1011 1012 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1013 { 1014 tp->snd_wl1 = seq; 1015 } 1016 1017 /* 1018 * Calculate(/check) TCP checksum 1019 */ 1020 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1021 __be32 daddr, __wsum base) 1022 { 1023 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1024 } 1025 1026 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1027 { 1028 return __skb_checksum_complete(skb); 1029 } 1030 1031 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1032 { 1033 return !skb_csum_unnecessary(skb) && 1034 __tcp_checksum_complete(skb); 1035 } 1036 1037 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1038 1039 static inline void tcp_prequeue_init(struct tcp_sock *tp) 1040 { 1041 tp->ucopy.task = NULL; 1042 tp->ucopy.len = 0; 1043 tp->ucopy.memory = 0; 1044 skb_queue_head_init(&tp->ucopy.prequeue); 1045 #ifdef CONFIG_NET_DMA 1046 tp->ucopy.dma_chan = NULL; 1047 tp->ucopy.wakeup = 0; 1048 tp->ucopy.pinned_list = NULL; 1049 tp->ucopy.dma_cookie = 0; 1050 #endif 1051 } 1052 1053 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1054 1055 #undef STATE_TRACE 1056 1057 #ifdef STATE_TRACE 1058 static const char *statename[]={ 1059 "Unused","Established","Syn Sent","Syn Recv", 1060 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1061 "Close Wait","Last ACK","Listen","Closing" 1062 }; 1063 #endif 1064 void tcp_set_state(struct sock *sk, int state); 1065 1066 void tcp_done(struct sock *sk); 1067 1068 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1069 { 1070 rx_opt->dsack = 0; 1071 rx_opt->num_sacks = 0; 1072 } 1073 1074 u32 tcp_default_init_rwnd(u32 mss); 1075 1076 /* Determine a window scaling and initial window to offer. */ 1077 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1078 __u32 *window_clamp, int wscale_ok, 1079 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1080 1081 static inline int tcp_win_from_space(int space) 1082 { 1083 return sysctl_tcp_adv_win_scale<=0 ? 1084 (space>>(-sysctl_tcp_adv_win_scale)) : 1085 space - (space>>sysctl_tcp_adv_win_scale); 1086 } 1087 1088 /* Note: caller must be prepared to deal with negative returns */ 1089 static inline int tcp_space(const struct sock *sk) 1090 { 1091 return tcp_win_from_space(sk->sk_rcvbuf - 1092 atomic_read(&sk->sk_rmem_alloc)); 1093 } 1094 1095 static inline int tcp_full_space(const struct sock *sk) 1096 { 1097 return tcp_win_from_space(sk->sk_rcvbuf); 1098 } 1099 1100 static inline void tcp_openreq_init(struct request_sock *req, 1101 struct tcp_options_received *rx_opt, 1102 struct sk_buff *skb) 1103 { 1104 struct inet_request_sock *ireq = inet_rsk(req); 1105 1106 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 1107 req->cookie_ts = 0; 1108 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 1109 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1; 1110 tcp_rsk(req)->snt_synack = 0; 1111 req->mss = rx_opt->mss_clamp; 1112 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 1113 ireq->tstamp_ok = rx_opt->tstamp_ok; 1114 ireq->sack_ok = rx_opt->sack_ok; 1115 ireq->snd_wscale = rx_opt->snd_wscale; 1116 ireq->wscale_ok = rx_opt->wscale_ok; 1117 ireq->acked = 0; 1118 ireq->ecn_ok = 0; 1119 ireq->ir_rmt_port = tcp_hdr(skb)->source; 1120 ireq->ir_num = ntohs(tcp_hdr(skb)->dest); 1121 } 1122 1123 void tcp_enter_memory_pressure(struct sock *sk); 1124 1125 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1126 { 1127 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1128 } 1129 1130 static inline int keepalive_time_when(const struct tcp_sock *tp) 1131 { 1132 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1133 } 1134 1135 static inline int keepalive_probes(const struct tcp_sock *tp) 1136 { 1137 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1138 } 1139 1140 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1141 { 1142 const struct inet_connection_sock *icsk = &tp->inet_conn; 1143 1144 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1145 tcp_time_stamp - tp->rcv_tstamp); 1146 } 1147 1148 static inline int tcp_fin_time(const struct sock *sk) 1149 { 1150 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1151 const int rto = inet_csk(sk)->icsk_rto; 1152 1153 if (fin_timeout < (rto << 2) - (rto >> 1)) 1154 fin_timeout = (rto << 2) - (rto >> 1); 1155 1156 return fin_timeout; 1157 } 1158 1159 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1160 int paws_win) 1161 { 1162 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1163 return true; 1164 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1165 return true; 1166 /* 1167 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1168 * then following tcp messages have valid values. Ignore 0 value, 1169 * or else 'negative' tsval might forbid us to accept their packets. 1170 */ 1171 if (!rx_opt->ts_recent) 1172 return true; 1173 return false; 1174 } 1175 1176 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1177 int rst) 1178 { 1179 if (tcp_paws_check(rx_opt, 0)) 1180 return false; 1181 1182 /* RST segments are not recommended to carry timestamp, 1183 and, if they do, it is recommended to ignore PAWS because 1184 "their cleanup function should take precedence over timestamps." 1185 Certainly, it is mistake. It is necessary to understand the reasons 1186 of this constraint to relax it: if peer reboots, clock may go 1187 out-of-sync and half-open connections will not be reset. 1188 Actually, the problem would be not existing if all 1189 the implementations followed draft about maintaining clock 1190 via reboots. Linux-2.2 DOES NOT! 1191 1192 However, we can relax time bounds for RST segments to MSL. 1193 */ 1194 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1195 return false; 1196 return true; 1197 } 1198 1199 static inline void tcp_mib_init(struct net *net) 1200 { 1201 /* See RFC 2012 */ 1202 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1203 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1204 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1205 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1206 } 1207 1208 /* from STCP */ 1209 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1210 { 1211 tp->lost_skb_hint = NULL; 1212 } 1213 1214 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1215 { 1216 tcp_clear_retrans_hints_partial(tp); 1217 tp->retransmit_skb_hint = NULL; 1218 } 1219 1220 /* MD5 Signature */ 1221 struct crypto_hash; 1222 1223 union tcp_md5_addr { 1224 struct in_addr a4; 1225 #if IS_ENABLED(CONFIG_IPV6) 1226 struct in6_addr a6; 1227 #endif 1228 }; 1229 1230 /* - key database */ 1231 struct tcp_md5sig_key { 1232 struct hlist_node node; 1233 u8 keylen; 1234 u8 family; /* AF_INET or AF_INET6 */ 1235 union tcp_md5_addr addr; 1236 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1237 struct rcu_head rcu; 1238 }; 1239 1240 /* - sock block */ 1241 struct tcp_md5sig_info { 1242 struct hlist_head head; 1243 struct rcu_head rcu; 1244 }; 1245 1246 /* - pseudo header */ 1247 struct tcp4_pseudohdr { 1248 __be32 saddr; 1249 __be32 daddr; 1250 __u8 pad; 1251 __u8 protocol; 1252 __be16 len; 1253 }; 1254 1255 struct tcp6_pseudohdr { 1256 struct in6_addr saddr; 1257 struct in6_addr daddr; 1258 __be32 len; 1259 __be32 protocol; /* including padding */ 1260 }; 1261 1262 union tcp_md5sum_block { 1263 struct tcp4_pseudohdr ip4; 1264 #if IS_ENABLED(CONFIG_IPV6) 1265 struct tcp6_pseudohdr ip6; 1266 #endif 1267 }; 1268 1269 /* - pool: digest algorithm, hash description and scratch buffer */ 1270 struct tcp_md5sig_pool { 1271 struct hash_desc md5_desc; 1272 union tcp_md5sum_block md5_blk; 1273 }; 1274 1275 /* - functions */ 1276 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1277 const struct sock *sk, const struct request_sock *req, 1278 const struct sk_buff *skb); 1279 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1280 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); 1281 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1282 int family); 1283 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1284 struct sock *addr_sk); 1285 1286 #ifdef CONFIG_TCP_MD5SIG 1287 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1288 const union tcp_md5_addr *addr, 1289 int family); 1290 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1291 #else 1292 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1293 const union tcp_md5_addr *addr, 1294 int family) 1295 { 1296 return NULL; 1297 } 1298 #define tcp_twsk_md5_key(twsk) NULL 1299 #endif 1300 1301 bool tcp_alloc_md5sig_pool(void); 1302 1303 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1304 static inline void tcp_put_md5sig_pool(void) 1305 { 1306 local_bh_enable(); 1307 } 1308 1309 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1310 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1311 unsigned int header_len); 1312 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1313 const struct tcp_md5sig_key *key); 1314 1315 /* From tcp_fastopen.c */ 1316 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1317 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1318 unsigned long *last_syn_loss); 1319 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1320 struct tcp_fastopen_cookie *cookie, bool syn_lost); 1321 struct tcp_fastopen_request { 1322 /* Fast Open cookie. Size 0 means a cookie request */ 1323 struct tcp_fastopen_cookie cookie; 1324 struct msghdr *data; /* data in MSG_FASTOPEN */ 1325 size_t size; 1326 int copied; /* queued in tcp_connect() */ 1327 }; 1328 void tcp_free_fastopen_req(struct tcp_sock *tp); 1329 1330 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1331 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1332 void tcp_fastopen_cookie_gen(__be32 src, __be32 dst, 1333 struct tcp_fastopen_cookie *foc); 1334 void tcp_fastopen_init_key_once(bool publish); 1335 #define TCP_FASTOPEN_KEY_LENGTH 16 1336 1337 /* Fastopen key context */ 1338 struct tcp_fastopen_context { 1339 struct crypto_cipher *tfm; 1340 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1341 struct rcu_head rcu; 1342 }; 1343 1344 /* write queue abstraction */ 1345 static inline void tcp_write_queue_purge(struct sock *sk) 1346 { 1347 struct sk_buff *skb; 1348 1349 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1350 sk_wmem_free_skb(sk, skb); 1351 sk_mem_reclaim(sk); 1352 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1353 } 1354 1355 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1356 { 1357 return skb_peek(&sk->sk_write_queue); 1358 } 1359 1360 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1361 { 1362 return skb_peek_tail(&sk->sk_write_queue); 1363 } 1364 1365 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1366 const struct sk_buff *skb) 1367 { 1368 return skb_queue_next(&sk->sk_write_queue, skb); 1369 } 1370 1371 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1372 const struct sk_buff *skb) 1373 { 1374 return skb_queue_prev(&sk->sk_write_queue, skb); 1375 } 1376 1377 #define tcp_for_write_queue(skb, sk) \ 1378 skb_queue_walk(&(sk)->sk_write_queue, skb) 1379 1380 #define tcp_for_write_queue_from(skb, sk) \ 1381 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1382 1383 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1384 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1385 1386 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1387 { 1388 return sk->sk_send_head; 1389 } 1390 1391 static inline bool tcp_skb_is_last(const struct sock *sk, 1392 const struct sk_buff *skb) 1393 { 1394 return skb_queue_is_last(&sk->sk_write_queue, skb); 1395 } 1396 1397 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1398 { 1399 if (tcp_skb_is_last(sk, skb)) 1400 sk->sk_send_head = NULL; 1401 else 1402 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1403 } 1404 1405 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1406 { 1407 if (sk->sk_send_head == skb_unlinked) 1408 sk->sk_send_head = NULL; 1409 } 1410 1411 static inline void tcp_init_send_head(struct sock *sk) 1412 { 1413 sk->sk_send_head = NULL; 1414 } 1415 1416 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1417 { 1418 __skb_queue_tail(&sk->sk_write_queue, skb); 1419 } 1420 1421 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1422 { 1423 __tcp_add_write_queue_tail(sk, skb); 1424 1425 /* Queue it, remembering where we must start sending. */ 1426 if (sk->sk_send_head == NULL) { 1427 sk->sk_send_head = skb; 1428 1429 if (tcp_sk(sk)->highest_sack == NULL) 1430 tcp_sk(sk)->highest_sack = skb; 1431 } 1432 } 1433 1434 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1435 { 1436 __skb_queue_head(&sk->sk_write_queue, skb); 1437 } 1438 1439 /* Insert buff after skb on the write queue of sk. */ 1440 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1441 struct sk_buff *buff, 1442 struct sock *sk) 1443 { 1444 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1445 } 1446 1447 /* Insert new before skb on the write queue of sk. */ 1448 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1449 struct sk_buff *skb, 1450 struct sock *sk) 1451 { 1452 __skb_queue_before(&sk->sk_write_queue, skb, new); 1453 1454 if (sk->sk_send_head == skb) 1455 sk->sk_send_head = new; 1456 } 1457 1458 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1459 { 1460 __skb_unlink(skb, &sk->sk_write_queue); 1461 } 1462 1463 static inline bool tcp_write_queue_empty(struct sock *sk) 1464 { 1465 return skb_queue_empty(&sk->sk_write_queue); 1466 } 1467 1468 static inline void tcp_push_pending_frames(struct sock *sk) 1469 { 1470 if (tcp_send_head(sk)) { 1471 struct tcp_sock *tp = tcp_sk(sk); 1472 1473 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1474 } 1475 } 1476 1477 /* Start sequence of the skb just after the highest skb with SACKed 1478 * bit, valid only if sacked_out > 0 or when the caller has ensured 1479 * validity by itself. 1480 */ 1481 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1482 { 1483 if (!tp->sacked_out) 1484 return tp->snd_una; 1485 1486 if (tp->highest_sack == NULL) 1487 return tp->snd_nxt; 1488 1489 return TCP_SKB_CB(tp->highest_sack)->seq; 1490 } 1491 1492 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1493 { 1494 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1495 tcp_write_queue_next(sk, skb); 1496 } 1497 1498 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1499 { 1500 return tcp_sk(sk)->highest_sack; 1501 } 1502 1503 static inline void tcp_highest_sack_reset(struct sock *sk) 1504 { 1505 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1506 } 1507 1508 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1509 static inline void tcp_highest_sack_combine(struct sock *sk, 1510 struct sk_buff *old, 1511 struct sk_buff *new) 1512 { 1513 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1514 tcp_sk(sk)->highest_sack = new; 1515 } 1516 1517 /* Determines whether this is a thin stream (which may suffer from 1518 * increased latency). Used to trigger latency-reducing mechanisms. 1519 */ 1520 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1521 { 1522 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1523 } 1524 1525 /* /proc */ 1526 enum tcp_seq_states { 1527 TCP_SEQ_STATE_LISTENING, 1528 TCP_SEQ_STATE_OPENREQ, 1529 TCP_SEQ_STATE_ESTABLISHED, 1530 }; 1531 1532 int tcp_seq_open(struct inode *inode, struct file *file); 1533 1534 struct tcp_seq_afinfo { 1535 char *name; 1536 sa_family_t family; 1537 const struct file_operations *seq_fops; 1538 struct seq_operations seq_ops; 1539 }; 1540 1541 struct tcp_iter_state { 1542 struct seq_net_private p; 1543 sa_family_t family; 1544 enum tcp_seq_states state; 1545 struct sock *syn_wait_sk; 1546 int bucket, offset, sbucket, num; 1547 kuid_t uid; 1548 loff_t last_pos; 1549 }; 1550 1551 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1552 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1553 1554 extern struct request_sock_ops tcp_request_sock_ops; 1555 extern struct request_sock_ops tcp6_request_sock_ops; 1556 1557 void tcp_v4_destroy_sock(struct sock *sk); 1558 1559 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1560 netdev_features_t features); 1561 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1562 int tcp_gro_complete(struct sk_buff *skb); 1563 1564 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1565 1566 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1567 { 1568 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat; 1569 } 1570 1571 static inline bool tcp_stream_memory_free(const struct sock *sk) 1572 { 1573 const struct tcp_sock *tp = tcp_sk(sk); 1574 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1575 1576 return notsent_bytes < tcp_notsent_lowat(tp); 1577 } 1578 1579 #ifdef CONFIG_PROC_FS 1580 int tcp4_proc_init(void); 1581 void tcp4_proc_exit(void); 1582 #endif 1583 1584 /* TCP af-specific functions */ 1585 struct tcp_sock_af_ops { 1586 #ifdef CONFIG_TCP_MD5SIG 1587 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1588 struct sock *addr_sk); 1589 int (*calc_md5_hash) (char *location, 1590 struct tcp_md5sig_key *md5, 1591 const struct sock *sk, 1592 const struct request_sock *req, 1593 const struct sk_buff *skb); 1594 int (*md5_parse) (struct sock *sk, 1595 char __user *optval, 1596 int optlen); 1597 #endif 1598 }; 1599 1600 struct tcp_request_sock_ops { 1601 #ifdef CONFIG_TCP_MD5SIG 1602 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1603 struct request_sock *req); 1604 int (*calc_md5_hash) (char *location, 1605 struct tcp_md5sig_key *md5, 1606 const struct sock *sk, 1607 const struct request_sock *req, 1608 const struct sk_buff *skb); 1609 #endif 1610 }; 1611 1612 int tcpv4_offload_init(void); 1613 1614 void tcp_v4_init(void); 1615 void tcp_init(void); 1616 1617 #endif /* _TCP_H */ 1618