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 extern 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 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */ 196 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */ 197 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN) 198 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX) 199 200 /* But this is what stacks really send out. */ 201 #define TCPOLEN_TSTAMP_ALIGNED 12 202 #define TCPOLEN_WSCALE_ALIGNED 4 203 #define TCPOLEN_SACKPERM_ALIGNED 4 204 #define TCPOLEN_SACK_BASE 2 205 #define TCPOLEN_SACK_BASE_ALIGNED 4 206 #define TCPOLEN_SACK_PERBLOCK 8 207 #define TCPOLEN_MD5SIG_ALIGNED 20 208 #define TCPOLEN_MSS_ALIGNED 4 209 210 /* Flags in tp->nonagle */ 211 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 212 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 213 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 214 215 /* TCP thin-stream limits */ 216 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 217 218 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */ 219 #define TCP_INIT_CWND 10 220 221 /* Bit Flags for sysctl_tcp_fastopen */ 222 #define TFO_CLIENT_ENABLE 1 223 #define TFO_SERVER_ENABLE 2 224 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 225 226 /* Process SYN data but skip cookie validation */ 227 #define TFO_SERVER_COOKIE_NOT_CHKED 0x100 228 /* Accept SYN data w/o any cookie option */ 229 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 230 231 /* Force enable TFO on all listeners, i.e., not requiring the 232 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen. 233 */ 234 #define TFO_SERVER_WO_SOCKOPT1 0x400 235 #define TFO_SERVER_WO_SOCKOPT2 0x800 236 /* Always create TFO child sockets on a TFO listener even when 237 * cookie/data not present. (For testing purpose!) 238 */ 239 #define TFO_SERVER_ALWAYS 0x1000 240 241 extern struct inet_timewait_death_row tcp_death_row; 242 243 /* sysctl variables for tcp */ 244 extern int sysctl_tcp_timestamps; 245 extern int sysctl_tcp_window_scaling; 246 extern int sysctl_tcp_sack; 247 extern int sysctl_tcp_fin_timeout; 248 extern int sysctl_tcp_keepalive_time; 249 extern int sysctl_tcp_keepalive_probes; 250 extern int sysctl_tcp_keepalive_intvl; 251 extern int sysctl_tcp_syn_retries; 252 extern int sysctl_tcp_synack_retries; 253 extern int sysctl_tcp_retries1; 254 extern int sysctl_tcp_retries2; 255 extern int sysctl_tcp_orphan_retries; 256 extern int sysctl_tcp_syncookies; 257 extern int sysctl_tcp_fastopen; 258 extern int sysctl_tcp_retrans_collapse; 259 extern int sysctl_tcp_stdurg; 260 extern int sysctl_tcp_rfc1337; 261 extern int sysctl_tcp_abort_on_overflow; 262 extern int sysctl_tcp_max_orphans; 263 extern int sysctl_tcp_fack; 264 extern int sysctl_tcp_reordering; 265 extern int sysctl_tcp_dsack; 266 extern int sysctl_tcp_wmem[3]; 267 extern int sysctl_tcp_rmem[3]; 268 extern int sysctl_tcp_app_win; 269 extern int sysctl_tcp_adv_win_scale; 270 extern int sysctl_tcp_tw_reuse; 271 extern int sysctl_tcp_frto; 272 extern int sysctl_tcp_low_latency; 273 extern int sysctl_tcp_dma_copybreak; 274 extern int sysctl_tcp_nometrics_save; 275 extern int sysctl_tcp_moderate_rcvbuf; 276 extern int sysctl_tcp_tso_win_divisor; 277 extern int sysctl_tcp_mtu_probing; 278 extern int sysctl_tcp_base_mss; 279 extern int sysctl_tcp_workaround_signed_windows; 280 extern int sysctl_tcp_slow_start_after_idle; 281 extern int sysctl_tcp_max_ssthresh; 282 extern int sysctl_tcp_thin_linear_timeouts; 283 extern int sysctl_tcp_thin_dupack; 284 extern int sysctl_tcp_early_retrans; 285 extern int sysctl_tcp_limit_output_bytes; 286 extern int sysctl_tcp_challenge_ack_limit; 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 extern 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 extern void tcp_init_mem(struct net *net); 354 355 extern void tcp_tasklet_init(void); 356 357 extern void tcp_v4_err(struct sk_buff *skb, u32); 358 359 extern void tcp_shutdown (struct sock *sk, int how); 360 361 extern void tcp_v4_early_demux(struct sk_buff *skb); 362 extern int tcp_v4_rcv(struct sk_buff *skb); 363 364 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 365 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 366 size_t size); 367 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset, 368 size_t size, int flags); 369 extern void tcp_release_cb(struct sock *sk); 370 extern void tcp_wfree(struct sk_buff *skb); 371 extern void tcp_write_timer_handler(struct sock *sk); 372 extern void tcp_delack_timer_handler(struct sock *sk); 373 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 374 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 375 const struct tcphdr *th, unsigned int len); 376 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 377 const struct tcphdr *th, unsigned int len); 378 extern void tcp_rcv_space_adjust(struct sock *sk); 379 extern void tcp_cleanup_rbuf(struct sock *sk, int copied); 380 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 381 extern void tcp_twsk_destructor(struct sock *sk); 382 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 383 struct pipe_inode_info *pipe, size_t len, 384 unsigned int flags); 385 386 static inline void tcp_dec_quickack_mode(struct sock *sk, 387 const unsigned int pkts) 388 { 389 struct inet_connection_sock *icsk = inet_csk(sk); 390 391 if (icsk->icsk_ack.quick) { 392 if (pkts >= icsk->icsk_ack.quick) { 393 icsk->icsk_ack.quick = 0; 394 /* Leaving quickack mode we deflate ATO. */ 395 icsk->icsk_ack.ato = TCP_ATO_MIN; 396 } else 397 icsk->icsk_ack.quick -= pkts; 398 } 399 } 400 401 #define TCP_ECN_OK 1 402 #define TCP_ECN_QUEUE_CWR 2 403 #define TCP_ECN_DEMAND_CWR 4 404 #define TCP_ECN_SEEN 8 405 406 enum tcp_tw_status { 407 TCP_TW_SUCCESS = 0, 408 TCP_TW_RST = 1, 409 TCP_TW_ACK = 2, 410 TCP_TW_SYN = 3 411 }; 412 413 414 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 415 struct sk_buff *skb, 416 const struct tcphdr *th); 417 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, 418 struct request_sock *req, 419 struct request_sock **prev, 420 bool fastopen); 421 extern int tcp_child_process(struct sock *parent, struct sock *child, 422 struct sk_buff *skb); 423 extern void tcp_enter_loss(struct sock *sk, int how); 424 extern void tcp_clear_retrans(struct tcp_sock *tp); 425 extern void tcp_update_metrics(struct sock *sk); 426 extern void tcp_init_metrics(struct sock *sk); 427 extern void tcp_metrics_init(void); 428 extern bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, bool paws_check); 429 extern bool tcp_remember_stamp(struct sock *sk); 430 extern bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); 431 extern void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); 432 extern void tcp_disable_fack(struct tcp_sock *tp); 433 extern void tcp_close(struct sock *sk, long timeout); 434 extern void tcp_init_sock(struct sock *sk); 435 extern unsigned int tcp_poll(struct file * file, struct socket *sock, 436 struct poll_table_struct *wait); 437 extern int tcp_getsockopt(struct sock *sk, int level, int optname, 438 char __user *optval, int __user *optlen); 439 extern int tcp_setsockopt(struct sock *sk, int level, int optname, 440 char __user *optval, unsigned int optlen); 441 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 442 char __user *optval, int __user *optlen); 443 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 444 char __user *optval, unsigned int optlen); 445 extern void tcp_set_keepalive(struct sock *sk, int val); 446 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req); 447 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 448 size_t len, int nonblock, int flags, int *addr_len); 449 extern void tcp_parse_options(const struct sk_buff *skb, 450 struct tcp_options_received *opt_rx, 451 int estab, struct tcp_fastopen_cookie *foc); 452 extern const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 453 454 /* 455 * TCP v4 functions exported for the inet6 API 456 */ 457 458 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 459 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 460 extern struct sock * tcp_create_openreq_child(struct sock *sk, 461 struct request_sock *req, 462 struct sk_buff *skb); 463 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 464 struct request_sock *req, 465 struct dst_entry *dst); 466 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 467 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, 468 int addr_len); 469 extern int tcp_connect(struct sock *sk); 470 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst, 471 struct request_sock *req, 472 struct tcp_fastopen_cookie *foc); 473 extern int tcp_disconnect(struct sock *sk, int flags); 474 475 void tcp_connect_init(struct sock *sk); 476 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 477 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 478 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 479 480 /* From syncookies.c */ 481 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS]; 482 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 483 struct ip_options *opt); 484 #ifdef CONFIG_SYN_COOKIES 485 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 486 __u16 *mss); 487 #else 488 static inline __u32 cookie_v4_init_sequence(struct sock *sk, 489 struct sk_buff *skb, 490 __u16 *mss) 491 { 492 return 0; 493 } 494 #endif 495 496 extern __u32 cookie_init_timestamp(struct request_sock *req); 497 extern bool cookie_check_timestamp(struct tcp_options_received *opt, 498 struct net *net, bool *ecn_ok); 499 500 /* From net/ipv6/syncookies.c */ 501 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 502 #ifdef CONFIG_SYN_COOKIES 503 extern __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb, 504 __u16 *mss); 505 #else 506 static inline __u32 cookie_v6_init_sequence(struct sock *sk, 507 struct sk_buff *skb, 508 __u16 *mss) 509 { 510 return 0; 511 } 512 #endif 513 /* tcp_output.c */ 514 515 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 516 int nonagle); 517 extern bool tcp_may_send_now(struct sock *sk); 518 extern int __tcp_retransmit_skb(struct sock *, struct sk_buff *); 519 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 520 extern void tcp_retransmit_timer(struct sock *sk); 521 extern void tcp_xmit_retransmit_queue(struct sock *); 522 extern void tcp_simple_retransmit(struct sock *); 523 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); 524 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 525 526 extern void tcp_send_probe0(struct sock *); 527 extern void tcp_send_partial(struct sock *); 528 extern int tcp_write_wakeup(struct sock *); 529 extern void tcp_send_fin(struct sock *sk); 530 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority); 531 extern int tcp_send_synack(struct sock *); 532 extern bool tcp_syn_flood_action(struct sock *sk, 533 const struct sk_buff *skb, 534 const char *proto); 535 extern void tcp_push_one(struct sock *, unsigned int mss_now); 536 extern void tcp_send_ack(struct sock *sk); 537 extern void tcp_send_delayed_ack(struct sock *sk); 538 extern void tcp_send_loss_probe(struct sock *sk); 539 extern bool tcp_schedule_loss_probe(struct sock *sk); 540 541 /* tcp_input.c */ 542 extern void tcp_cwnd_application_limited(struct sock *sk); 543 extern void tcp_resume_early_retransmit(struct sock *sk); 544 extern void tcp_rearm_rto(struct sock *sk); 545 extern void tcp_reset(struct sock *sk); 546 547 /* tcp_timer.c */ 548 extern void tcp_init_xmit_timers(struct sock *); 549 static inline void tcp_clear_xmit_timers(struct sock *sk) 550 { 551 inet_csk_clear_xmit_timers(sk); 552 } 553 554 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 555 extern unsigned int tcp_current_mss(struct sock *sk); 556 557 /* Bound MSS / TSO packet size with the half of the window */ 558 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 559 { 560 int cutoff; 561 562 /* When peer uses tiny windows, there is no use in packetizing 563 * to sub-MSS pieces for the sake of SWS or making sure there 564 * are enough packets in the pipe for fast recovery. 565 * 566 * On the other hand, for extremely large MSS devices, handling 567 * smaller than MSS windows in this way does make sense. 568 */ 569 if (tp->max_window >= 512) 570 cutoff = (tp->max_window >> 1); 571 else 572 cutoff = tp->max_window; 573 574 if (cutoff && pktsize > cutoff) 575 return max_t(int, cutoff, 68U - tp->tcp_header_len); 576 else 577 return pktsize; 578 } 579 580 /* tcp.c */ 581 extern void tcp_get_info(const struct sock *, struct tcp_info *); 582 583 /* Read 'sendfile()'-style from a TCP socket */ 584 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 585 unsigned int, size_t); 586 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 587 sk_read_actor_t recv_actor); 588 589 extern void tcp_initialize_rcv_mss(struct sock *sk); 590 591 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu); 592 extern int tcp_mss_to_mtu(struct sock *sk, int mss); 593 extern void tcp_mtup_init(struct sock *sk); 594 extern void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt); 595 extern void tcp_init_buffer_space(struct sock *sk); 596 597 static inline void tcp_bound_rto(const struct sock *sk) 598 { 599 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 600 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 601 } 602 603 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 604 { 605 return (tp->srtt >> 3) + tp->rttvar; 606 } 607 608 extern void tcp_set_rto(struct sock *sk); 609 610 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 611 { 612 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 613 ntohl(TCP_FLAG_ACK) | 614 snd_wnd); 615 } 616 617 static inline void tcp_fast_path_on(struct tcp_sock *tp) 618 { 619 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 620 } 621 622 static inline void tcp_fast_path_check(struct sock *sk) 623 { 624 struct tcp_sock *tp = tcp_sk(sk); 625 626 if (skb_queue_empty(&tp->out_of_order_queue) && 627 tp->rcv_wnd && 628 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 629 !tp->urg_data) 630 tcp_fast_path_on(tp); 631 } 632 633 /* Compute the actual rto_min value */ 634 static inline u32 tcp_rto_min(struct sock *sk) 635 { 636 const struct dst_entry *dst = __sk_dst_get(sk); 637 u32 rto_min = TCP_RTO_MIN; 638 639 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 640 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 641 return rto_min; 642 } 643 644 /* Compute the actual receive window we are currently advertising. 645 * Rcv_nxt can be after the window if our peer push more data 646 * than the offered window. 647 */ 648 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 649 { 650 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 651 652 if (win < 0) 653 win = 0; 654 return (u32) win; 655 } 656 657 /* Choose a new window, without checks for shrinking, and without 658 * scaling applied to the result. The caller does these things 659 * if necessary. This is a "raw" window selection. 660 */ 661 extern u32 __tcp_select_window(struct sock *sk); 662 663 void tcp_send_window_probe(struct sock *sk); 664 665 /* TCP timestamps are only 32-bits, this causes a slight 666 * complication on 64-bit systems since we store a snapshot 667 * of jiffies in the buffer control blocks below. We decided 668 * to use only the low 32-bits of jiffies and hide the ugly 669 * casts with the following macro. 670 */ 671 #define tcp_time_stamp ((__u32)(jiffies)) 672 673 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 674 675 #define TCPHDR_FIN 0x01 676 #define TCPHDR_SYN 0x02 677 #define TCPHDR_RST 0x04 678 #define TCPHDR_PSH 0x08 679 #define TCPHDR_ACK 0x10 680 #define TCPHDR_URG 0x20 681 #define TCPHDR_ECE 0x40 682 #define TCPHDR_CWR 0x80 683 684 /* This is what the send packet queuing engine uses to pass 685 * TCP per-packet control information to the transmission code. 686 * We also store the host-order sequence numbers in here too. 687 * This is 44 bytes if IPV6 is enabled. 688 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 689 */ 690 struct tcp_skb_cb { 691 union { 692 struct inet_skb_parm h4; 693 #if IS_ENABLED(CONFIG_IPV6) 694 struct inet6_skb_parm h6; 695 #endif 696 } header; /* For incoming frames */ 697 __u32 seq; /* Starting sequence number */ 698 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 699 __u32 when; /* used to compute rtt's */ 700 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 701 702 __u8 sacked; /* State flags for SACK/FACK. */ 703 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 704 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 705 #define TCPCB_LOST 0x04 /* SKB is lost */ 706 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 707 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 708 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 709 710 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 711 /* 1 byte hole */ 712 __u32 ack_seq; /* Sequence number ACK'd */ 713 }; 714 715 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 716 717 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set 718 * 719 * If we receive a SYN packet with these bits set, it means a network is 720 * playing bad games with TOS bits. In order to avoid possible false congestion 721 * notifications, we disable TCP ECN negociation. 722 */ 723 static inline void 724 TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb, 725 struct net *net) 726 { 727 const struct tcphdr *th = tcp_hdr(skb); 728 729 if (net->ipv4.sysctl_tcp_ecn && th->ece && th->cwr && 730 INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield)) 731 inet_rsk(req)->ecn_ok = 1; 732 } 733 734 /* Due to TSO, an SKB can be composed of multiple actual 735 * packets. To keep these tracked properly, we use this. 736 */ 737 static inline int tcp_skb_pcount(const struct sk_buff *skb) 738 { 739 return skb_shinfo(skb)->gso_segs; 740 } 741 742 /* This is valid iff tcp_skb_pcount() > 1. */ 743 static inline int tcp_skb_mss(const struct sk_buff *skb) 744 { 745 return skb_shinfo(skb)->gso_size; 746 } 747 748 /* Events passed to congestion control interface */ 749 enum tcp_ca_event { 750 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 751 CA_EVENT_CWND_RESTART, /* congestion window restart */ 752 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 753 CA_EVENT_LOSS, /* loss timeout */ 754 CA_EVENT_FAST_ACK, /* in sequence ack */ 755 CA_EVENT_SLOW_ACK, /* other ack */ 756 }; 757 758 /* 759 * Interface for adding new TCP congestion control handlers 760 */ 761 #define TCP_CA_NAME_MAX 16 762 #define TCP_CA_MAX 128 763 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 764 765 #define TCP_CONG_NON_RESTRICTED 0x1 766 #define TCP_CONG_RTT_STAMP 0x2 767 768 struct tcp_congestion_ops { 769 struct list_head list; 770 unsigned long flags; 771 772 /* initialize private data (optional) */ 773 void (*init)(struct sock *sk); 774 /* cleanup private data (optional) */ 775 void (*release)(struct sock *sk); 776 777 /* return slow start threshold (required) */ 778 u32 (*ssthresh)(struct sock *sk); 779 /* lower bound for congestion window (optional) */ 780 u32 (*min_cwnd)(const struct sock *sk); 781 /* do new cwnd calculation (required) */ 782 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight); 783 /* call before changing ca_state (optional) */ 784 void (*set_state)(struct sock *sk, u8 new_state); 785 /* call when cwnd event occurs (optional) */ 786 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 787 /* new value of cwnd after loss (optional) */ 788 u32 (*undo_cwnd)(struct sock *sk); 789 /* hook for packet ack accounting (optional) */ 790 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 791 /* get info for inet_diag (optional) */ 792 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 793 794 char name[TCP_CA_NAME_MAX]; 795 struct module *owner; 796 }; 797 798 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); 799 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 800 801 extern void tcp_init_congestion_control(struct sock *sk); 802 extern void tcp_cleanup_congestion_control(struct sock *sk); 803 extern int tcp_set_default_congestion_control(const char *name); 804 extern void tcp_get_default_congestion_control(char *name); 805 extern void tcp_get_available_congestion_control(char *buf, size_t len); 806 extern void tcp_get_allowed_congestion_control(char *buf, size_t len); 807 extern int tcp_set_allowed_congestion_control(char *allowed); 808 extern int tcp_set_congestion_control(struct sock *sk, const char *name); 809 extern void tcp_slow_start(struct tcp_sock *tp); 810 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 811 812 extern struct tcp_congestion_ops tcp_init_congestion_ops; 813 extern u32 tcp_reno_ssthresh(struct sock *sk); 814 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight); 815 extern u32 tcp_reno_min_cwnd(const struct sock *sk); 816 extern struct tcp_congestion_ops tcp_reno; 817 818 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 819 { 820 struct inet_connection_sock *icsk = inet_csk(sk); 821 822 if (icsk->icsk_ca_ops->set_state) 823 icsk->icsk_ca_ops->set_state(sk, ca_state); 824 icsk->icsk_ca_state = ca_state; 825 } 826 827 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 828 { 829 const struct inet_connection_sock *icsk = inet_csk(sk); 830 831 if (icsk->icsk_ca_ops->cwnd_event) 832 icsk->icsk_ca_ops->cwnd_event(sk, event); 833 } 834 835 /* These functions determine how the current flow behaves in respect of SACK 836 * handling. SACK is negotiated with the peer, and therefore it can vary 837 * between different flows. 838 * 839 * tcp_is_sack - SACK enabled 840 * tcp_is_reno - No SACK 841 * tcp_is_fack - FACK enabled, implies SACK enabled 842 */ 843 static inline int tcp_is_sack(const struct tcp_sock *tp) 844 { 845 return tp->rx_opt.sack_ok; 846 } 847 848 static inline bool tcp_is_reno(const struct tcp_sock *tp) 849 { 850 return !tcp_is_sack(tp); 851 } 852 853 static inline bool tcp_is_fack(const struct tcp_sock *tp) 854 { 855 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 856 } 857 858 static inline void tcp_enable_fack(struct tcp_sock *tp) 859 { 860 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 861 } 862 863 /* TCP early-retransmit (ER) is similar to but more conservative than 864 * the thin-dupack feature. Enable ER only if thin-dupack is disabled. 865 */ 866 static inline void tcp_enable_early_retrans(struct tcp_sock *tp) 867 { 868 tp->do_early_retrans = sysctl_tcp_early_retrans && 869 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack && 870 sysctl_tcp_reordering == 3; 871 } 872 873 static inline void tcp_disable_early_retrans(struct tcp_sock *tp) 874 { 875 tp->do_early_retrans = 0; 876 } 877 878 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 879 { 880 return tp->sacked_out + tp->lost_out; 881 } 882 883 /* This determines how many packets are "in the network" to the best 884 * of our knowledge. In many cases it is conservative, but where 885 * detailed information is available from the receiver (via SACK 886 * blocks etc.) we can make more aggressive calculations. 887 * 888 * Use this for decisions involving congestion control, use just 889 * tp->packets_out to determine if the send queue is empty or not. 890 * 891 * Read this equation as: 892 * 893 * "Packets sent once on transmission queue" MINUS 894 * "Packets left network, but not honestly ACKed yet" PLUS 895 * "Packets fast retransmitted" 896 */ 897 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 898 { 899 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 900 } 901 902 #define TCP_INFINITE_SSTHRESH 0x7fffffff 903 904 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 905 { 906 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 907 } 908 909 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 910 { 911 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 912 (1 << inet_csk(sk)->icsk_ca_state); 913 } 914 915 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 916 * The exception is cwnd reduction phase, when cwnd is decreasing towards 917 * ssthresh. 918 */ 919 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 920 { 921 const struct tcp_sock *tp = tcp_sk(sk); 922 923 if (tcp_in_cwnd_reduction(sk)) 924 return tp->snd_ssthresh; 925 else 926 return max(tp->snd_ssthresh, 927 ((tp->snd_cwnd >> 1) + 928 (tp->snd_cwnd >> 2))); 929 } 930 931 /* Use define here intentionally to get WARN_ON location shown at the caller */ 932 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 933 934 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 935 extern __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 936 937 /* The maximum number of MSS of available cwnd for which TSO defers 938 * sending if not using sysctl_tcp_tso_win_divisor. 939 */ 940 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 941 { 942 return 3; 943 } 944 945 /* Slow start with delack produces 3 packets of burst, so that 946 * it is safe "de facto". This will be the default - same as 947 * the default reordering threshold - but if reordering increases, 948 * we must be able to allow cwnd to burst at least this much in order 949 * to not pull it back when holes are filled. 950 */ 951 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 952 { 953 return tp->reordering; 954 } 955 956 /* Returns end sequence number of the receiver's advertised window */ 957 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 958 { 959 return tp->snd_una + tp->snd_wnd; 960 } 961 extern bool tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight); 962 963 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss, 964 const struct sk_buff *skb) 965 { 966 if (skb->len < mss) 967 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 968 } 969 970 static inline void tcp_check_probe_timer(struct sock *sk) 971 { 972 const struct tcp_sock *tp = tcp_sk(sk); 973 const struct inet_connection_sock *icsk = inet_csk(sk); 974 975 if (!tp->packets_out && !icsk->icsk_pending) 976 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 977 icsk->icsk_rto, TCP_RTO_MAX); 978 } 979 980 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 981 { 982 tp->snd_wl1 = seq; 983 } 984 985 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 986 { 987 tp->snd_wl1 = seq; 988 } 989 990 /* 991 * Calculate(/check) TCP checksum 992 */ 993 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 994 __be32 daddr, __wsum base) 995 { 996 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 997 } 998 999 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1000 { 1001 return __skb_checksum_complete(skb); 1002 } 1003 1004 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1005 { 1006 return !skb_csum_unnecessary(skb) && 1007 __tcp_checksum_complete(skb); 1008 } 1009 1010 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1011 1012 static inline void tcp_prequeue_init(struct tcp_sock *tp) 1013 { 1014 tp->ucopy.task = NULL; 1015 tp->ucopy.len = 0; 1016 tp->ucopy.memory = 0; 1017 skb_queue_head_init(&tp->ucopy.prequeue); 1018 #ifdef CONFIG_NET_DMA 1019 tp->ucopy.dma_chan = NULL; 1020 tp->ucopy.wakeup = 0; 1021 tp->ucopy.pinned_list = NULL; 1022 tp->ucopy.dma_cookie = 0; 1023 #endif 1024 } 1025 1026 extern bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1027 1028 #undef STATE_TRACE 1029 1030 #ifdef STATE_TRACE 1031 static const char *statename[]={ 1032 "Unused","Established","Syn Sent","Syn Recv", 1033 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1034 "Close Wait","Last ACK","Listen","Closing" 1035 }; 1036 #endif 1037 extern void tcp_set_state(struct sock *sk, int state); 1038 1039 extern void tcp_done(struct sock *sk); 1040 1041 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1042 { 1043 rx_opt->dsack = 0; 1044 rx_opt->num_sacks = 0; 1045 } 1046 1047 extern u32 tcp_default_init_rwnd(u32 mss); 1048 1049 /* Determine a window scaling and initial window to offer. */ 1050 extern void tcp_select_initial_window(int __space, __u32 mss, 1051 __u32 *rcv_wnd, __u32 *window_clamp, 1052 int wscale_ok, __u8 *rcv_wscale, 1053 __u32 init_rcv_wnd); 1054 1055 static inline int tcp_win_from_space(int space) 1056 { 1057 return sysctl_tcp_adv_win_scale<=0 ? 1058 (space>>(-sysctl_tcp_adv_win_scale)) : 1059 space - (space>>sysctl_tcp_adv_win_scale); 1060 } 1061 1062 /* Note: caller must be prepared to deal with negative returns */ 1063 static inline int tcp_space(const struct sock *sk) 1064 { 1065 return tcp_win_from_space(sk->sk_rcvbuf - 1066 atomic_read(&sk->sk_rmem_alloc)); 1067 } 1068 1069 static inline int tcp_full_space(const struct sock *sk) 1070 { 1071 return tcp_win_from_space(sk->sk_rcvbuf); 1072 } 1073 1074 static inline void tcp_openreq_init(struct request_sock *req, 1075 struct tcp_options_received *rx_opt, 1076 struct sk_buff *skb) 1077 { 1078 struct inet_request_sock *ireq = inet_rsk(req); 1079 1080 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 1081 req->cookie_ts = 0; 1082 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 1083 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1; 1084 tcp_rsk(req)->snt_synack = 0; 1085 req->mss = rx_opt->mss_clamp; 1086 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 1087 ireq->tstamp_ok = rx_opt->tstamp_ok; 1088 ireq->sack_ok = rx_opt->sack_ok; 1089 ireq->snd_wscale = rx_opt->snd_wscale; 1090 ireq->wscale_ok = rx_opt->wscale_ok; 1091 ireq->acked = 0; 1092 ireq->ecn_ok = 0; 1093 ireq->rmt_port = tcp_hdr(skb)->source; 1094 ireq->loc_port = tcp_hdr(skb)->dest; 1095 } 1096 1097 /* Compute time elapsed between SYNACK and the ACK completing 3WHS */ 1098 static inline void tcp_synack_rtt_meas(struct sock *sk, 1099 struct request_sock *req) 1100 { 1101 if (tcp_rsk(req)->snt_synack) 1102 tcp_valid_rtt_meas(sk, 1103 tcp_time_stamp - tcp_rsk(req)->snt_synack); 1104 } 1105 1106 extern void tcp_enter_memory_pressure(struct sock *sk); 1107 1108 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1109 { 1110 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1111 } 1112 1113 static inline int keepalive_time_when(const struct tcp_sock *tp) 1114 { 1115 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1116 } 1117 1118 static inline int keepalive_probes(const struct tcp_sock *tp) 1119 { 1120 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1121 } 1122 1123 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1124 { 1125 const struct inet_connection_sock *icsk = &tp->inet_conn; 1126 1127 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1128 tcp_time_stamp - tp->rcv_tstamp); 1129 } 1130 1131 static inline int tcp_fin_time(const struct sock *sk) 1132 { 1133 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1134 const int rto = inet_csk(sk)->icsk_rto; 1135 1136 if (fin_timeout < (rto << 2) - (rto >> 1)) 1137 fin_timeout = (rto << 2) - (rto >> 1); 1138 1139 return fin_timeout; 1140 } 1141 1142 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1143 int paws_win) 1144 { 1145 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1146 return true; 1147 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1148 return true; 1149 /* 1150 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1151 * then following tcp messages have valid values. Ignore 0 value, 1152 * or else 'negative' tsval might forbid us to accept their packets. 1153 */ 1154 if (!rx_opt->ts_recent) 1155 return true; 1156 return false; 1157 } 1158 1159 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1160 int rst) 1161 { 1162 if (tcp_paws_check(rx_opt, 0)) 1163 return false; 1164 1165 /* RST segments are not recommended to carry timestamp, 1166 and, if they do, it is recommended to ignore PAWS because 1167 "their cleanup function should take precedence over timestamps." 1168 Certainly, it is mistake. It is necessary to understand the reasons 1169 of this constraint to relax it: if peer reboots, clock may go 1170 out-of-sync and half-open connections will not be reset. 1171 Actually, the problem would be not existing if all 1172 the implementations followed draft about maintaining clock 1173 via reboots. Linux-2.2 DOES NOT! 1174 1175 However, we can relax time bounds for RST segments to MSL. 1176 */ 1177 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1178 return false; 1179 return true; 1180 } 1181 1182 static inline void tcp_mib_init(struct net *net) 1183 { 1184 /* See RFC 2012 */ 1185 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1186 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1187 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1188 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1189 } 1190 1191 /* from STCP */ 1192 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1193 { 1194 tp->lost_skb_hint = NULL; 1195 } 1196 1197 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1198 { 1199 tcp_clear_retrans_hints_partial(tp); 1200 tp->retransmit_skb_hint = NULL; 1201 } 1202 1203 /* MD5 Signature */ 1204 struct crypto_hash; 1205 1206 union tcp_md5_addr { 1207 struct in_addr a4; 1208 #if IS_ENABLED(CONFIG_IPV6) 1209 struct in6_addr a6; 1210 #endif 1211 }; 1212 1213 /* - key database */ 1214 struct tcp_md5sig_key { 1215 struct hlist_node node; 1216 u8 keylen; 1217 u8 family; /* AF_INET or AF_INET6 */ 1218 union tcp_md5_addr addr; 1219 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1220 struct rcu_head rcu; 1221 }; 1222 1223 /* - sock block */ 1224 struct tcp_md5sig_info { 1225 struct hlist_head head; 1226 struct rcu_head rcu; 1227 }; 1228 1229 /* - pseudo header */ 1230 struct tcp4_pseudohdr { 1231 __be32 saddr; 1232 __be32 daddr; 1233 __u8 pad; 1234 __u8 protocol; 1235 __be16 len; 1236 }; 1237 1238 struct tcp6_pseudohdr { 1239 struct in6_addr saddr; 1240 struct in6_addr daddr; 1241 __be32 len; 1242 __be32 protocol; /* including padding */ 1243 }; 1244 1245 union tcp_md5sum_block { 1246 struct tcp4_pseudohdr ip4; 1247 #if IS_ENABLED(CONFIG_IPV6) 1248 struct tcp6_pseudohdr ip6; 1249 #endif 1250 }; 1251 1252 /* - pool: digest algorithm, hash description and scratch buffer */ 1253 struct tcp_md5sig_pool { 1254 struct hash_desc md5_desc; 1255 union tcp_md5sum_block md5_blk; 1256 }; 1257 1258 /* - functions */ 1259 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1260 const struct sock *sk, 1261 const struct request_sock *req, 1262 const struct sk_buff *skb); 1263 extern int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1264 int family, const u8 *newkey, 1265 u8 newkeylen, gfp_t gfp); 1266 extern int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1267 int family); 1268 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1269 struct sock *addr_sk); 1270 1271 #ifdef CONFIG_TCP_MD5SIG 1272 extern struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1273 const union tcp_md5_addr *addr, int family); 1274 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1275 #else 1276 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1277 const union tcp_md5_addr *addr, 1278 int family) 1279 { 1280 return NULL; 1281 } 1282 #define tcp_twsk_md5_key(twsk) NULL 1283 #endif 1284 1285 extern bool tcp_alloc_md5sig_pool(void); 1286 1287 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1288 static inline void tcp_put_md5sig_pool(void) 1289 { 1290 local_bh_enable(); 1291 } 1292 1293 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1294 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1295 unsigned int header_len); 1296 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1297 const struct tcp_md5sig_key *key); 1298 1299 /* From tcp_fastopen.c */ 1300 extern void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1301 struct tcp_fastopen_cookie *cookie, 1302 int *syn_loss, unsigned long *last_syn_loss); 1303 extern void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1304 struct tcp_fastopen_cookie *cookie, 1305 bool syn_lost); 1306 struct tcp_fastopen_request { 1307 /* Fast Open cookie. Size 0 means a cookie request */ 1308 struct tcp_fastopen_cookie cookie; 1309 struct msghdr *data; /* data in MSG_FASTOPEN */ 1310 u16 copied; /* queued in tcp_connect() */ 1311 }; 1312 void tcp_free_fastopen_req(struct tcp_sock *tp); 1313 1314 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1315 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1316 void tcp_fastopen_cookie_gen(__be32 addr, struct tcp_fastopen_cookie *foc); 1317 1318 #define TCP_FASTOPEN_KEY_LENGTH 16 1319 1320 /* Fastopen key context */ 1321 struct tcp_fastopen_context { 1322 struct crypto_cipher *tfm; 1323 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1324 struct rcu_head rcu; 1325 }; 1326 1327 /* write queue abstraction */ 1328 static inline void tcp_write_queue_purge(struct sock *sk) 1329 { 1330 struct sk_buff *skb; 1331 1332 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1333 sk_wmem_free_skb(sk, skb); 1334 sk_mem_reclaim(sk); 1335 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1336 } 1337 1338 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1339 { 1340 return skb_peek(&sk->sk_write_queue); 1341 } 1342 1343 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1344 { 1345 return skb_peek_tail(&sk->sk_write_queue); 1346 } 1347 1348 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1349 const struct sk_buff *skb) 1350 { 1351 return skb_queue_next(&sk->sk_write_queue, skb); 1352 } 1353 1354 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1355 const struct sk_buff *skb) 1356 { 1357 return skb_queue_prev(&sk->sk_write_queue, skb); 1358 } 1359 1360 #define tcp_for_write_queue(skb, sk) \ 1361 skb_queue_walk(&(sk)->sk_write_queue, skb) 1362 1363 #define tcp_for_write_queue_from(skb, sk) \ 1364 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1365 1366 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1367 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1368 1369 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1370 { 1371 return sk->sk_send_head; 1372 } 1373 1374 static inline bool tcp_skb_is_last(const struct sock *sk, 1375 const struct sk_buff *skb) 1376 { 1377 return skb_queue_is_last(&sk->sk_write_queue, skb); 1378 } 1379 1380 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1381 { 1382 if (tcp_skb_is_last(sk, skb)) 1383 sk->sk_send_head = NULL; 1384 else 1385 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1386 } 1387 1388 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1389 { 1390 if (sk->sk_send_head == skb_unlinked) 1391 sk->sk_send_head = NULL; 1392 } 1393 1394 static inline void tcp_init_send_head(struct sock *sk) 1395 { 1396 sk->sk_send_head = NULL; 1397 } 1398 1399 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1400 { 1401 __skb_queue_tail(&sk->sk_write_queue, skb); 1402 } 1403 1404 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1405 { 1406 __tcp_add_write_queue_tail(sk, skb); 1407 1408 /* Queue it, remembering where we must start sending. */ 1409 if (sk->sk_send_head == NULL) { 1410 sk->sk_send_head = skb; 1411 1412 if (tcp_sk(sk)->highest_sack == NULL) 1413 tcp_sk(sk)->highest_sack = skb; 1414 } 1415 } 1416 1417 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1418 { 1419 __skb_queue_head(&sk->sk_write_queue, skb); 1420 } 1421 1422 /* Insert buff after skb on the write queue of sk. */ 1423 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1424 struct sk_buff *buff, 1425 struct sock *sk) 1426 { 1427 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1428 } 1429 1430 /* Insert new before skb on the write queue of sk. */ 1431 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1432 struct sk_buff *skb, 1433 struct sock *sk) 1434 { 1435 __skb_queue_before(&sk->sk_write_queue, skb, new); 1436 1437 if (sk->sk_send_head == skb) 1438 sk->sk_send_head = new; 1439 } 1440 1441 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1442 { 1443 __skb_unlink(skb, &sk->sk_write_queue); 1444 } 1445 1446 static inline bool tcp_write_queue_empty(struct sock *sk) 1447 { 1448 return skb_queue_empty(&sk->sk_write_queue); 1449 } 1450 1451 static inline void tcp_push_pending_frames(struct sock *sk) 1452 { 1453 if (tcp_send_head(sk)) { 1454 struct tcp_sock *tp = tcp_sk(sk); 1455 1456 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1457 } 1458 } 1459 1460 /* Start sequence of the skb just after the highest skb with SACKed 1461 * bit, valid only if sacked_out > 0 or when the caller has ensured 1462 * validity by itself. 1463 */ 1464 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1465 { 1466 if (!tp->sacked_out) 1467 return tp->snd_una; 1468 1469 if (tp->highest_sack == NULL) 1470 return tp->snd_nxt; 1471 1472 return TCP_SKB_CB(tp->highest_sack)->seq; 1473 } 1474 1475 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1476 { 1477 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1478 tcp_write_queue_next(sk, skb); 1479 } 1480 1481 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1482 { 1483 return tcp_sk(sk)->highest_sack; 1484 } 1485 1486 static inline void tcp_highest_sack_reset(struct sock *sk) 1487 { 1488 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1489 } 1490 1491 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1492 static inline void tcp_highest_sack_combine(struct sock *sk, 1493 struct sk_buff *old, 1494 struct sk_buff *new) 1495 { 1496 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1497 tcp_sk(sk)->highest_sack = new; 1498 } 1499 1500 /* Determines whether this is a thin stream (which may suffer from 1501 * increased latency). Used to trigger latency-reducing mechanisms. 1502 */ 1503 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1504 { 1505 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1506 } 1507 1508 /* /proc */ 1509 enum tcp_seq_states { 1510 TCP_SEQ_STATE_LISTENING, 1511 TCP_SEQ_STATE_OPENREQ, 1512 TCP_SEQ_STATE_ESTABLISHED, 1513 TCP_SEQ_STATE_TIME_WAIT, 1514 }; 1515 1516 int tcp_seq_open(struct inode *inode, struct file *file); 1517 1518 struct tcp_seq_afinfo { 1519 char *name; 1520 sa_family_t family; 1521 const struct file_operations *seq_fops; 1522 struct seq_operations seq_ops; 1523 }; 1524 1525 struct tcp_iter_state { 1526 struct seq_net_private p; 1527 sa_family_t family; 1528 enum tcp_seq_states state; 1529 struct sock *syn_wait_sk; 1530 int bucket, offset, sbucket, num; 1531 kuid_t uid; 1532 loff_t last_pos; 1533 }; 1534 1535 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1536 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1537 1538 extern struct request_sock_ops tcp_request_sock_ops; 1539 extern struct request_sock_ops tcp6_request_sock_ops; 1540 1541 extern void tcp_v4_destroy_sock(struct sock *sk); 1542 1543 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, 1544 netdev_features_t features); 1545 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head, 1546 struct sk_buff *skb); 1547 extern int tcp_gro_complete(struct sk_buff *skb); 1548 1549 extern void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, 1550 __be32 daddr); 1551 1552 #ifdef CONFIG_PROC_FS 1553 extern int tcp4_proc_init(void); 1554 extern void tcp4_proc_exit(void); 1555 #endif 1556 1557 /* TCP af-specific functions */ 1558 struct tcp_sock_af_ops { 1559 #ifdef CONFIG_TCP_MD5SIG 1560 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1561 struct sock *addr_sk); 1562 int (*calc_md5_hash) (char *location, 1563 struct tcp_md5sig_key *md5, 1564 const struct sock *sk, 1565 const struct request_sock *req, 1566 const struct sk_buff *skb); 1567 int (*md5_parse) (struct sock *sk, 1568 char __user *optval, 1569 int optlen); 1570 #endif 1571 }; 1572 1573 struct tcp_request_sock_ops { 1574 #ifdef CONFIG_TCP_MD5SIG 1575 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1576 struct request_sock *req); 1577 int (*calc_md5_hash) (char *location, 1578 struct tcp_md5sig_key *md5, 1579 const struct sock *sk, 1580 const struct request_sock *req, 1581 const struct sk_buff *skb); 1582 #endif 1583 }; 1584 1585 extern int tcpv4_offload_init(void); 1586 1587 extern void tcp_v4_init(void); 1588 extern void tcp_init(void); 1589 1590 #endif /* _TCP_H */ 1591