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 /* Accept SYN data w/o any cookie option */ 224 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 225 226 /* Force enable TFO on all listeners, i.e., not requiring the 227 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen. 228 */ 229 #define TFO_SERVER_WO_SOCKOPT1 0x400 230 #define TFO_SERVER_WO_SOCKOPT2 0x800 231 232 extern struct inet_timewait_death_row tcp_death_row; 233 234 /* sysctl variables for tcp */ 235 extern int sysctl_tcp_timestamps; 236 extern int sysctl_tcp_window_scaling; 237 extern int sysctl_tcp_sack; 238 extern int sysctl_tcp_fin_timeout; 239 extern int sysctl_tcp_keepalive_time; 240 extern int sysctl_tcp_keepalive_probes; 241 extern int sysctl_tcp_keepalive_intvl; 242 extern int sysctl_tcp_syn_retries; 243 extern int sysctl_tcp_synack_retries; 244 extern int sysctl_tcp_retries1; 245 extern int sysctl_tcp_retries2; 246 extern int sysctl_tcp_orphan_retries; 247 extern int sysctl_tcp_syncookies; 248 extern int sysctl_tcp_fastopen; 249 extern int sysctl_tcp_retrans_collapse; 250 extern int sysctl_tcp_stdurg; 251 extern int sysctl_tcp_rfc1337; 252 extern int sysctl_tcp_abort_on_overflow; 253 extern int sysctl_tcp_max_orphans; 254 extern int sysctl_tcp_fack; 255 extern int sysctl_tcp_reordering; 256 extern int sysctl_tcp_dsack; 257 extern long sysctl_tcp_mem[3]; 258 extern int sysctl_tcp_wmem[3]; 259 extern int sysctl_tcp_rmem[3]; 260 extern int sysctl_tcp_app_win; 261 extern int sysctl_tcp_adv_win_scale; 262 extern int sysctl_tcp_tw_reuse; 263 extern int sysctl_tcp_frto; 264 extern int sysctl_tcp_low_latency; 265 extern int sysctl_tcp_dma_copybreak; 266 extern int sysctl_tcp_nometrics_save; 267 extern int sysctl_tcp_moderate_rcvbuf; 268 extern int sysctl_tcp_tso_win_divisor; 269 extern int sysctl_tcp_mtu_probing; 270 extern int sysctl_tcp_base_mss; 271 extern int sysctl_tcp_workaround_signed_windows; 272 extern int sysctl_tcp_slow_start_after_idle; 273 extern int sysctl_tcp_thin_linear_timeouts; 274 extern int sysctl_tcp_thin_dupack; 275 extern int sysctl_tcp_early_retrans; 276 extern int sysctl_tcp_limit_output_bytes; 277 extern int sysctl_tcp_challenge_ack_limit; 278 extern unsigned int sysctl_tcp_notsent_lowat; 279 extern int sysctl_tcp_min_tso_segs; 280 extern int sysctl_tcp_autocorking; 281 282 extern atomic_long_t tcp_memory_allocated; 283 extern struct percpu_counter tcp_sockets_allocated; 284 extern int tcp_memory_pressure; 285 286 /* 287 * The next routines deal with comparing 32 bit unsigned ints 288 * and worry about wraparound (automatic with unsigned arithmetic). 289 */ 290 291 static inline bool before(__u32 seq1, __u32 seq2) 292 { 293 return (__s32)(seq1-seq2) < 0; 294 } 295 #define after(seq2, seq1) before(seq1, seq2) 296 297 /* is s2<=s1<=s3 ? */ 298 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 299 { 300 return seq3 - seq2 >= seq1 - seq2; 301 } 302 303 static inline bool tcp_out_of_memory(struct sock *sk) 304 { 305 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 306 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 307 return true; 308 return false; 309 } 310 311 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 312 { 313 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 314 int orphans = percpu_counter_read_positive(ocp); 315 316 if (orphans << shift > sysctl_tcp_max_orphans) { 317 orphans = percpu_counter_sum_positive(ocp); 318 if (orphans << shift > sysctl_tcp_max_orphans) 319 return true; 320 } 321 return false; 322 } 323 324 bool tcp_check_oom(struct sock *sk, int shift); 325 326 /* syncookies: remember time of last synqueue overflow */ 327 static inline void tcp_synq_overflow(struct sock *sk) 328 { 329 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies; 330 } 331 332 /* syncookies: no recent synqueue overflow on this listening socket? */ 333 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 334 { 335 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 336 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK); 337 } 338 339 extern struct proto tcp_prot; 340 341 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 342 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 343 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 344 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 345 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 346 347 void tcp_tasklet_init(void); 348 349 void tcp_v4_err(struct sk_buff *skb, u32); 350 351 void tcp_shutdown(struct sock *sk, int how); 352 353 void tcp_v4_early_demux(struct sk_buff *skb); 354 int tcp_v4_rcv(struct sk_buff *skb); 355 356 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 357 int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 358 size_t size); 359 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 360 int flags); 361 void tcp_release_cb(struct sock *sk); 362 void tcp_wfree(struct sk_buff *skb); 363 void tcp_write_timer_handler(struct sock *sk); 364 void tcp_delack_timer_handler(struct sock *sk); 365 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 366 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 367 const struct tcphdr *th, unsigned int len); 368 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 369 const struct tcphdr *th, unsigned int len); 370 void tcp_rcv_space_adjust(struct sock *sk); 371 void tcp_cleanup_rbuf(struct sock *sk, int copied); 372 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 373 void tcp_twsk_destructor(struct sock *sk); 374 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 375 struct pipe_inode_info *pipe, size_t len, 376 unsigned int flags); 377 378 static inline void tcp_dec_quickack_mode(struct sock *sk, 379 const unsigned int pkts) 380 { 381 struct inet_connection_sock *icsk = inet_csk(sk); 382 383 if (icsk->icsk_ack.quick) { 384 if (pkts >= icsk->icsk_ack.quick) { 385 icsk->icsk_ack.quick = 0; 386 /* Leaving quickack mode we deflate ATO. */ 387 icsk->icsk_ack.ato = TCP_ATO_MIN; 388 } else 389 icsk->icsk_ack.quick -= pkts; 390 } 391 } 392 393 #define TCP_ECN_OK 1 394 #define TCP_ECN_QUEUE_CWR 2 395 #define TCP_ECN_DEMAND_CWR 4 396 #define TCP_ECN_SEEN 8 397 398 enum tcp_tw_status { 399 TCP_TW_SUCCESS = 0, 400 TCP_TW_RST = 1, 401 TCP_TW_ACK = 2, 402 TCP_TW_SYN = 3 403 }; 404 405 406 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 407 struct sk_buff *skb, 408 const struct tcphdr *th); 409 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 410 struct request_sock *req, struct request_sock **prev, 411 bool fastopen); 412 int tcp_child_process(struct sock *parent, struct sock *child, 413 struct sk_buff *skb); 414 void tcp_enter_loss(struct sock *sk); 415 void tcp_clear_retrans(struct tcp_sock *tp); 416 void tcp_update_metrics(struct sock *sk); 417 void tcp_init_metrics(struct sock *sk); 418 void tcp_metrics_init(void); 419 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, 420 bool paws_check, bool timestamps); 421 bool tcp_remember_stamp(struct sock *sk); 422 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); 423 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); 424 void tcp_disable_fack(struct tcp_sock *tp); 425 void tcp_close(struct sock *sk, long timeout); 426 void tcp_init_sock(struct sock *sk); 427 unsigned int tcp_poll(struct file *file, struct socket *sock, 428 struct poll_table_struct *wait); 429 int tcp_getsockopt(struct sock *sk, int level, int optname, 430 char __user *optval, int __user *optlen); 431 int tcp_setsockopt(struct sock *sk, int level, int optname, 432 char __user *optval, unsigned int optlen); 433 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 434 char __user *optval, int __user *optlen); 435 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 436 char __user *optval, unsigned int optlen); 437 void tcp_set_keepalive(struct sock *sk, int val); 438 void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req); 439 int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 440 size_t len, int nonblock, int flags, int *addr_len); 441 void tcp_parse_options(const struct sk_buff *skb, 442 struct tcp_options_received *opt_rx, 443 int estab, struct tcp_fastopen_cookie *foc); 444 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 445 446 /* 447 * TCP v4 functions exported for the inet6 API 448 */ 449 450 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 451 void tcp_v4_mtu_reduced(struct sock *sk); 452 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 453 struct sock *tcp_create_openreq_child(struct sock *sk, 454 struct request_sock *req, 455 struct sk_buff *skb); 456 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 457 struct request_sock *req, 458 struct dst_entry *dst); 459 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 460 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 461 int tcp_connect(struct sock *sk); 462 struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst, 463 struct request_sock *req, 464 struct tcp_fastopen_cookie *foc); 465 int tcp_disconnect(struct sock *sk, int flags); 466 467 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 468 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 469 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 470 471 /* From syncookies.c */ 472 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 473 u32 cookie); 474 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 475 struct ip_options *opt); 476 #ifdef CONFIG_SYN_COOKIES 477 478 /* Syncookies use a monotonic timer which increments every 60 seconds. 479 * This counter is used both as a hash input and partially encoded into 480 * the cookie value. A cookie is only validated further if the delta 481 * between the current counter value and the encoded one is less than this, 482 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 483 * the counter advances immediately after a cookie is generated). 484 */ 485 #define MAX_SYNCOOKIE_AGE 2 486 487 static inline u32 tcp_cookie_time(void) 488 { 489 u64 val = get_jiffies_64(); 490 491 do_div(val, 60 * HZ); 492 return val; 493 } 494 495 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 496 u16 *mssp); 497 __u32 cookie_v4_init_sequence(struct sock *sk, const struct sk_buff *skb, 498 __u16 *mss); 499 #endif 500 501 __u32 cookie_init_timestamp(struct request_sock *req); 502 bool cookie_check_timestamp(struct tcp_options_received *opt, struct net *net, 503 bool *ecn_ok); 504 505 /* From net/ipv6/syncookies.c */ 506 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 507 u32 cookie); 508 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 509 #ifdef CONFIG_SYN_COOKIES 510 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 511 const struct tcphdr *th, u16 *mssp); 512 __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb, 513 __u16 *mss); 514 #endif 515 /* tcp_output.c */ 516 517 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 518 int nonagle); 519 bool tcp_may_send_now(struct sock *sk); 520 int __tcp_retransmit_skb(struct sock *, struct sk_buff *); 521 int tcp_retransmit_skb(struct sock *, struct sk_buff *); 522 void tcp_retransmit_timer(struct sock *sk); 523 void tcp_xmit_retransmit_queue(struct sock *); 524 void tcp_simple_retransmit(struct sock *); 525 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 526 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t); 527 528 void tcp_send_probe0(struct sock *); 529 void tcp_send_partial(struct sock *); 530 int tcp_write_wakeup(struct sock *); 531 void tcp_send_fin(struct sock *sk); 532 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 533 int tcp_send_synack(struct sock *); 534 bool tcp_syn_flood_action(struct sock *sk, const struct sk_buff *skb, 535 const char *proto); 536 void tcp_push_one(struct sock *, unsigned int mss_now); 537 void tcp_send_ack(struct sock *sk); 538 void tcp_send_delayed_ack(struct sock *sk); 539 void tcp_send_loss_probe(struct sock *sk); 540 bool tcp_schedule_loss_probe(struct sock *sk); 541 542 /* tcp_input.c */ 543 void tcp_resume_early_retransmit(struct sock *sk); 544 void tcp_rearm_rto(struct sock *sk); 545 void tcp_reset(struct sock *sk); 546 547 /* tcp_timer.c */ 548 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 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 555 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 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 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 587 sk_read_actor_t recv_actor); 588 589 void tcp_initialize_rcv_mss(struct sock *sk); 590 591 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 592 int tcp_mss_to_mtu(struct sock *sk, int mss); 593 void tcp_mtup_init(struct sock *sk); 594 void tcp_init_buffer_space(struct sock *sk); 595 596 static inline void tcp_bound_rto(const struct sock *sk) 597 { 598 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 599 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 600 } 601 602 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 603 { 604 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 605 } 606 607 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 608 { 609 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 610 ntohl(TCP_FLAG_ACK) | 611 snd_wnd); 612 } 613 614 static inline void tcp_fast_path_on(struct tcp_sock *tp) 615 { 616 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 617 } 618 619 static inline void tcp_fast_path_check(struct sock *sk) 620 { 621 struct tcp_sock *tp = tcp_sk(sk); 622 623 if (skb_queue_empty(&tp->out_of_order_queue) && 624 tp->rcv_wnd && 625 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 626 !tp->urg_data) 627 tcp_fast_path_on(tp); 628 } 629 630 /* Compute the actual rto_min value */ 631 static inline u32 tcp_rto_min(struct sock *sk) 632 { 633 const struct dst_entry *dst = __sk_dst_get(sk); 634 u32 rto_min = TCP_RTO_MIN; 635 636 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 637 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 638 return rto_min; 639 } 640 641 static inline u32 tcp_rto_min_us(struct sock *sk) 642 { 643 return jiffies_to_usecs(tcp_rto_min(sk)); 644 } 645 646 /* Compute the actual receive window we are currently advertising. 647 * Rcv_nxt can be after the window if our peer push more data 648 * than the offered window. 649 */ 650 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 651 { 652 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 653 654 if (win < 0) 655 win = 0; 656 return (u32) win; 657 } 658 659 /* Choose a new window, without checks for shrinking, and without 660 * scaling applied to the result. The caller does these things 661 * if necessary. This is a "raw" window selection. 662 */ 663 u32 __tcp_select_window(struct sock *sk); 664 665 void tcp_send_window_probe(struct sock *sk); 666 667 /* TCP timestamps are only 32-bits, this causes a slight 668 * complication on 64-bit systems since we store a snapshot 669 * of jiffies in the buffer control blocks below. We decided 670 * to use only the low 32-bits of jiffies and hide the ugly 671 * casts with the following macro. 672 */ 673 #define tcp_time_stamp ((__u32)(jiffies)) 674 675 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 676 677 #define TCPHDR_FIN 0x01 678 #define TCPHDR_SYN 0x02 679 #define TCPHDR_RST 0x04 680 #define TCPHDR_PSH 0x08 681 #define TCPHDR_ACK 0x10 682 #define TCPHDR_URG 0x20 683 #define TCPHDR_ECE 0x40 684 #define TCPHDR_CWR 0x80 685 686 /* This is what the send packet queuing engine uses to pass 687 * TCP per-packet control information to the transmission code. 688 * We also store the host-order sequence numbers in here too. 689 * This is 44 bytes if IPV6 is enabled. 690 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 691 */ 692 struct tcp_skb_cb { 693 union { 694 struct inet_skb_parm h4; 695 #if IS_ENABLED(CONFIG_IPV6) 696 struct inet6_skb_parm h6; 697 #endif 698 } header; /* For incoming frames */ 699 __u32 seq; /* Starting sequence number */ 700 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 701 __u32 when; /* used to compute rtt's */ 702 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 703 704 __u8 sacked; /* State flags for SACK/FACK. */ 705 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 706 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 707 #define TCPCB_LOST 0x04 /* SKB is lost */ 708 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 709 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 710 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 711 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 712 TCPCB_REPAIRED) 713 714 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 715 /* 1 byte hole */ 716 __u32 ack_seq; /* Sequence number ACK'd */ 717 }; 718 719 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 720 721 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set 722 * 723 * If we receive a SYN packet with these bits set, it means a network is 724 * playing bad games with TOS bits. In order to avoid possible false congestion 725 * notifications, we disable TCP ECN negociation. 726 */ 727 static inline void 728 TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb, 729 struct net *net) 730 { 731 const struct tcphdr *th = tcp_hdr(skb); 732 733 if (net->ipv4.sysctl_tcp_ecn && th->ece && th->cwr && 734 INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield)) 735 inet_rsk(req)->ecn_ok = 1; 736 } 737 738 /* Due to TSO, an SKB can be composed of multiple actual 739 * packets. To keep these tracked properly, we use this. 740 */ 741 static inline int tcp_skb_pcount(const struct sk_buff *skb) 742 { 743 return skb_shinfo(skb)->gso_segs; 744 } 745 746 /* This is valid iff tcp_skb_pcount() > 1. */ 747 static inline int tcp_skb_mss(const struct sk_buff *skb) 748 { 749 return skb_shinfo(skb)->gso_size; 750 } 751 752 /* Events passed to congestion control interface */ 753 enum tcp_ca_event { 754 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 755 CA_EVENT_CWND_RESTART, /* congestion window restart */ 756 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 757 CA_EVENT_LOSS, /* loss timeout */ 758 CA_EVENT_FAST_ACK, /* in sequence ack */ 759 CA_EVENT_SLOW_ACK, /* other ack */ 760 }; 761 762 /* 763 * Interface for adding new TCP congestion control handlers 764 */ 765 #define TCP_CA_NAME_MAX 16 766 #define TCP_CA_MAX 128 767 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 768 769 #define TCP_CONG_NON_RESTRICTED 0x1 770 771 struct tcp_congestion_ops { 772 struct list_head list; 773 unsigned long flags; 774 775 /* initialize private data (optional) */ 776 void (*init)(struct sock *sk); 777 /* cleanup private data (optional) */ 778 void (*release)(struct sock *sk); 779 780 /* return slow start threshold (required) */ 781 u32 (*ssthresh)(struct sock *sk); 782 /* do new cwnd calculation (required) */ 783 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 784 /* call before changing ca_state (optional) */ 785 void (*set_state)(struct sock *sk, u8 new_state); 786 /* call when cwnd event occurs (optional) */ 787 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 788 /* new value of cwnd after loss (optional) */ 789 u32 (*undo_cwnd)(struct sock *sk); 790 /* hook for packet ack accounting (optional) */ 791 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 792 /* get info for inet_diag (optional) */ 793 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 794 795 char name[TCP_CA_NAME_MAX]; 796 struct module *owner; 797 }; 798 799 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 800 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 801 802 void tcp_init_congestion_control(struct sock *sk); 803 void tcp_cleanup_congestion_control(struct sock *sk); 804 int tcp_set_default_congestion_control(const char *name); 805 void tcp_get_default_congestion_control(char *name); 806 void tcp_get_available_congestion_control(char *buf, size_t len); 807 void tcp_get_allowed_congestion_control(char *buf, size_t len); 808 int tcp_set_allowed_congestion_control(char *allowed); 809 int tcp_set_congestion_control(struct sock *sk, const char *name); 810 int tcp_slow_start(struct tcp_sock *tp, u32 acked); 811 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 812 813 extern struct tcp_congestion_ops tcp_init_congestion_ops; 814 u32 tcp_reno_ssthresh(struct sock *sk); 815 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 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 void tcp_enter_cwr(struct sock *sk); 935 __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 } 961 962 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 963 * flexible approach. The RFC suggests cwnd should not be raised unless 964 * it was fully used previously. And that's exactly what we do in 965 * congestion avoidance mode. But in slow start we allow cwnd to grow 966 * as long as the application has used half the cwnd. 967 * Example : 968 * cwnd is 10 (IW10), but application sends 9 frames. 969 * We allow cwnd to reach 18 when all frames are ACKed. 970 * This check is safe because it's as aggressive as slow start which already 971 * risks 100% overshoot. The advantage is that we discourage application to 972 * either send more filler packets or data to artificially blow up the cwnd 973 * usage, and allow application-limited process to probe bw more aggressively. 974 */ 975 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 976 { 977 const struct tcp_sock *tp = tcp_sk(sk); 978 979 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 980 if (tp->snd_cwnd <= tp->snd_ssthresh) 981 return tp->snd_cwnd < 2 * tp->max_packets_out; 982 983 return tp->is_cwnd_limited; 984 } 985 986 static inline void tcp_check_probe_timer(struct sock *sk) 987 { 988 const struct tcp_sock *tp = tcp_sk(sk); 989 const struct inet_connection_sock *icsk = inet_csk(sk); 990 991 if (!tp->packets_out && !icsk->icsk_pending) 992 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 993 icsk->icsk_rto, TCP_RTO_MAX); 994 } 995 996 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 997 { 998 tp->snd_wl1 = seq; 999 } 1000 1001 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1002 { 1003 tp->snd_wl1 = seq; 1004 } 1005 1006 /* 1007 * Calculate(/check) TCP checksum 1008 */ 1009 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1010 __be32 daddr, __wsum base) 1011 { 1012 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1013 } 1014 1015 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1016 { 1017 return __skb_checksum_complete(skb); 1018 } 1019 1020 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1021 { 1022 return !skb_csum_unnecessary(skb) && 1023 __tcp_checksum_complete(skb); 1024 } 1025 1026 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1027 1028 static inline void tcp_prequeue_init(struct tcp_sock *tp) 1029 { 1030 tp->ucopy.task = NULL; 1031 tp->ucopy.len = 0; 1032 tp->ucopy.memory = 0; 1033 skb_queue_head_init(&tp->ucopy.prequeue); 1034 #ifdef CONFIG_NET_DMA 1035 tp->ucopy.dma_chan = NULL; 1036 tp->ucopy.wakeup = 0; 1037 tp->ucopy.pinned_list = NULL; 1038 tp->ucopy.dma_cookie = 0; 1039 #endif 1040 } 1041 1042 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1043 1044 #undef STATE_TRACE 1045 1046 #ifdef STATE_TRACE 1047 static const char *statename[]={ 1048 "Unused","Established","Syn Sent","Syn Recv", 1049 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1050 "Close Wait","Last ACK","Listen","Closing" 1051 }; 1052 #endif 1053 void tcp_set_state(struct sock *sk, int state); 1054 1055 void tcp_done(struct sock *sk); 1056 1057 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1058 { 1059 rx_opt->dsack = 0; 1060 rx_opt->num_sacks = 0; 1061 } 1062 1063 u32 tcp_default_init_rwnd(u32 mss); 1064 1065 /* Determine a window scaling and initial window to offer. */ 1066 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1067 __u32 *window_clamp, int wscale_ok, 1068 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1069 1070 static inline int tcp_win_from_space(int space) 1071 { 1072 return sysctl_tcp_adv_win_scale<=0 ? 1073 (space>>(-sysctl_tcp_adv_win_scale)) : 1074 space - (space>>sysctl_tcp_adv_win_scale); 1075 } 1076 1077 /* Note: caller must be prepared to deal with negative returns */ 1078 static inline int tcp_space(const struct sock *sk) 1079 { 1080 return tcp_win_from_space(sk->sk_rcvbuf - 1081 atomic_read(&sk->sk_rmem_alloc)); 1082 } 1083 1084 static inline int tcp_full_space(const struct sock *sk) 1085 { 1086 return tcp_win_from_space(sk->sk_rcvbuf); 1087 } 1088 1089 static inline void tcp_openreq_init(struct request_sock *req, 1090 struct tcp_options_received *rx_opt, 1091 struct sk_buff *skb, struct sock *sk) 1092 { 1093 struct inet_request_sock *ireq = inet_rsk(req); 1094 1095 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 1096 req->cookie_ts = 0; 1097 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 1098 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1; 1099 tcp_rsk(req)->snt_synack = tcp_time_stamp; 1100 req->mss = rx_opt->mss_clamp; 1101 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 1102 ireq->tstamp_ok = rx_opt->tstamp_ok; 1103 ireq->sack_ok = rx_opt->sack_ok; 1104 ireq->snd_wscale = rx_opt->snd_wscale; 1105 ireq->wscale_ok = rx_opt->wscale_ok; 1106 ireq->acked = 0; 1107 ireq->ecn_ok = 0; 1108 ireq->ir_rmt_port = tcp_hdr(skb)->source; 1109 ireq->ir_num = ntohs(tcp_hdr(skb)->dest); 1110 ireq->ir_mark = inet_request_mark(sk, skb); 1111 } 1112 1113 extern void tcp_openreq_init_rwin(struct request_sock *req, 1114 struct sock *sk, struct dst_entry *dst); 1115 1116 void tcp_enter_memory_pressure(struct sock *sk); 1117 1118 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1119 { 1120 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1121 } 1122 1123 static inline int keepalive_time_when(const struct tcp_sock *tp) 1124 { 1125 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1126 } 1127 1128 static inline int keepalive_probes(const struct tcp_sock *tp) 1129 { 1130 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1131 } 1132 1133 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1134 { 1135 const struct inet_connection_sock *icsk = &tp->inet_conn; 1136 1137 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1138 tcp_time_stamp - tp->rcv_tstamp); 1139 } 1140 1141 static inline int tcp_fin_time(const struct sock *sk) 1142 { 1143 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1144 const int rto = inet_csk(sk)->icsk_rto; 1145 1146 if (fin_timeout < (rto << 2) - (rto >> 1)) 1147 fin_timeout = (rto << 2) - (rto >> 1); 1148 1149 return fin_timeout; 1150 } 1151 1152 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1153 int paws_win) 1154 { 1155 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1156 return true; 1157 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1158 return true; 1159 /* 1160 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1161 * then following tcp messages have valid values. Ignore 0 value, 1162 * or else 'negative' tsval might forbid us to accept their packets. 1163 */ 1164 if (!rx_opt->ts_recent) 1165 return true; 1166 return false; 1167 } 1168 1169 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1170 int rst) 1171 { 1172 if (tcp_paws_check(rx_opt, 0)) 1173 return false; 1174 1175 /* RST segments are not recommended to carry timestamp, 1176 and, if they do, it is recommended to ignore PAWS because 1177 "their cleanup function should take precedence over timestamps." 1178 Certainly, it is mistake. It is necessary to understand the reasons 1179 of this constraint to relax it: if peer reboots, clock may go 1180 out-of-sync and half-open connections will not be reset. 1181 Actually, the problem would be not existing if all 1182 the implementations followed draft about maintaining clock 1183 via reboots. Linux-2.2 DOES NOT! 1184 1185 However, we can relax time bounds for RST segments to MSL. 1186 */ 1187 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1188 return false; 1189 return true; 1190 } 1191 1192 static inline void tcp_mib_init(struct net *net) 1193 { 1194 /* See RFC 2012 */ 1195 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1196 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1197 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1198 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1199 } 1200 1201 /* from STCP */ 1202 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1203 { 1204 tp->lost_skb_hint = NULL; 1205 } 1206 1207 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1208 { 1209 tcp_clear_retrans_hints_partial(tp); 1210 tp->retransmit_skb_hint = NULL; 1211 } 1212 1213 /* MD5 Signature */ 1214 struct crypto_hash; 1215 1216 union tcp_md5_addr { 1217 struct in_addr a4; 1218 #if IS_ENABLED(CONFIG_IPV6) 1219 struct in6_addr a6; 1220 #endif 1221 }; 1222 1223 /* - key database */ 1224 struct tcp_md5sig_key { 1225 struct hlist_node node; 1226 u8 keylen; 1227 u8 family; /* AF_INET or AF_INET6 */ 1228 union tcp_md5_addr addr; 1229 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1230 struct rcu_head rcu; 1231 }; 1232 1233 /* - sock block */ 1234 struct tcp_md5sig_info { 1235 struct hlist_head head; 1236 struct rcu_head rcu; 1237 }; 1238 1239 /* - pseudo header */ 1240 struct tcp4_pseudohdr { 1241 __be32 saddr; 1242 __be32 daddr; 1243 __u8 pad; 1244 __u8 protocol; 1245 __be16 len; 1246 }; 1247 1248 struct tcp6_pseudohdr { 1249 struct in6_addr saddr; 1250 struct in6_addr daddr; 1251 __be32 len; 1252 __be32 protocol; /* including padding */ 1253 }; 1254 1255 union tcp_md5sum_block { 1256 struct tcp4_pseudohdr ip4; 1257 #if IS_ENABLED(CONFIG_IPV6) 1258 struct tcp6_pseudohdr ip6; 1259 #endif 1260 }; 1261 1262 /* - pool: digest algorithm, hash description and scratch buffer */ 1263 struct tcp_md5sig_pool { 1264 struct hash_desc md5_desc; 1265 union tcp_md5sum_block md5_blk; 1266 }; 1267 1268 /* - functions */ 1269 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1270 const struct sock *sk, const struct request_sock *req, 1271 const struct sk_buff *skb); 1272 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1273 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); 1274 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1275 int family); 1276 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1277 struct sock *addr_sk); 1278 1279 #ifdef CONFIG_TCP_MD5SIG 1280 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1281 const union tcp_md5_addr *addr, 1282 int family); 1283 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1284 #else 1285 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1286 const union tcp_md5_addr *addr, 1287 int family) 1288 { 1289 return NULL; 1290 } 1291 #define tcp_twsk_md5_key(twsk) NULL 1292 #endif 1293 1294 bool tcp_alloc_md5sig_pool(void); 1295 1296 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1297 static inline void tcp_put_md5sig_pool(void) 1298 { 1299 local_bh_enable(); 1300 } 1301 1302 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1303 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1304 unsigned int header_len); 1305 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1306 const struct tcp_md5sig_key *key); 1307 1308 /* From tcp_fastopen.c */ 1309 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1310 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1311 unsigned long *last_syn_loss); 1312 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1313 struct tcp_fastopen_cookie *cookie, bool syn_lost); 1314 struct tcp_fastopen_request { 1315 /* Fast Open cookie. Size 0 means a cookie request */ 1316 struct tcp_fastopen_cookie cookie; 1317 struct msghdr *data; /* data in MSG_FASTOPEN */ 1318 size_t size; 1319 int copied; /* queued in tcp_connect() */ 1320 }; 1321 void tcp_free_fastopen_req(struct tcp_sock *tp); 1322 1323 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1324 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1325 bool tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1326 struct request_sock *req, 1327 struct tcp_fastopen_cookie *foc, 1328 struct dst_entry *dst); 1329 void tcp_fastopen_init_key_once(bool publish); 1330 #define TCP_FASTOPEN_KEY_LENGTH 16 1331 1332 /* Fastopen key context */ 1333 struct tcp_fastopen_context { 1334 struct crypto_cipher *tfm; 1335 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1336 struct rcu_head rcu; 1337 }; 1338 1339 /* write queue abstraction */ 1340 static inline void tcp_write_queue_purge(struct sock *sk) 1341 { 1342 struct sk_buff *skb; 1343 1344 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1345 sk_wmem_free_skb(sk, skb); 1346 sk_mem_reclaim(sk); 1347 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1348 } 1349 1350 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1351 { 1352 return skb_peek(&sk->sk_write_queue); 1353 } 1354 1355 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1356 { 1357 return skb_peek_tail(&sk->sk_write_queue); 1358 } 1359 1360 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1361 const struct sk_buff *skb) 1362 { 1363 return skb_queue_next(&sk->sk_write_queue, skb); 1364 } 1365 1366 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1367 const struct sk_buff *skb) 1368 { 1369 return skb_queue_prev(&sk->sk_write_queue, skb); 1370 } 1371 1372 #define tcp_for_write_queue(skb, sk) \ 1373 skb_queue_walk(&(sk)->sk_write_queue, skb) 1374 1375 #define tcp_for_write_queue_from(skb, sk) \ 1376 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1377 1378 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1379 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1380 1381 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1382 { 1383 return sk->sk_send_head; 1384 } 1385 1386 static inline bool tcp_skb_is_last(const struct sock *sk, 1387 const struct sk_buff *skb) 1388 { 1389 return skb_queue_is_last(&sk->sk_write_queue, skb); 1390 } 1391 1392 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1393 { 1394 if (tcp_skb_is_last(sk, skb)) 1395 sk->sk_send_head = NULL; 1396 else 1397 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1398 } 1399 1400 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1401 { 1402 if (sk->sk_send_head == skb_unlinked) 1403 sk->sk_send_head = NULL; 1404 } 1405 1406 static inline void tcp_init_send_head(struct sock *sk) 1407 { 1408 sk->sk_send_head = NULL; 1409 } 1410 1411 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1412 { 1413 __skb_queue_tail(&sk->sk_write_queue, skb); 1414 } 1415 1416 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1417 { 1418 __tcp_add_write_queue_tail(sk, skb); 1419 1420 /* Queue it, remembering where we must start sending. */ 1421 if (sk->sk_send_head == NULL) { 1422 sk->sk_send_head = skb; 1423 1424 if (tcp_sk(sk)->highest_sack == NULL) 1425 tcp_sk(sk)->highest_sack = skb; 1426 } 1427 } 1428 1429 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1430 { 1431 __skb_queue_head(&sk->sk_write_queue, skb); 1432 } 1433 1434 /* Insert buff after skb on the write queue of sk. */ 1435 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1436 struct sk_buff *buff, 1437 struct sock *sk) 1438 { 1439 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1440 } 1441 1442 /* Insert new before skb on the write queue of sk. */ 1443 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1444 struct sk_buff *skb, 1445 struct sock *sk) 1446 { 1447 __skb_queue_before(&sk->sk_write_queue, skb, new); 1448 1449 if (sk->sk_send_head == skb) 1450 sk->sk_send_head = new; 1451 } 1452 1453 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1454 { 1455 __skb_unlink(skb, &sk->sk_write_queue); 1456 } 1457 1458 static inline bool tcp_write_queue_empty(struct sock *sk) 1459 { 1460 return skb_queue_empty(&sk->sk_write_queue); 1461 } 1462 1463 static inline void tcp_push_pending_frames(struct sock *sk) 1464 { 1465 if (tcp_send_head(sk)) { 1466 struct tcp_sock *tp = tcp_sk(sk); 1467 1468 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1469 } 1470 } 1471 1472 /* Start sequence of the skb just after the highest skb with SACKed 1473 * bit, valid only if sacked_out > 0 or when the caller has ensured 1474 * validity by itself. 1475 */ 1476 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1477 { 1478 if (!tp->sacked_out) 1479 return tp->snd_una; 1480 1481 if (tp->highest_sack == NULL) 1482 return tp->snd_nxt; 1483 1484 return TCP_SKB_CB(tp->highest_sack)->seq; 1485 } 1486 1487 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1488 { 1489 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1490 tcp_write_queue_next(sk, skb); 1491 } 1492 1493 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1494 { 1495 return tcp_sk(sk)->highest_sack; 1496 } 1497 1498 static inline void tcp_highest_sack_reset(struct sock *sk) 1499 { 1500 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1501 } 1502 1503 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1504 static inline void tcp_highest_sack_combine(struct sock *sk, 1505 struct sk_buff *old, 1506 struct sk_buff *new) 1507 { 1508 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1509 tcp_sk(sk)->highest_sack = new; 1510 } 1511 1512 /* Determines whether this is a thin stream (which may suffer from 1513 * increased latency). Used to trigger latency-reducing mechanisms. 1514 */ 1515 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1516 { 1517 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1518 } 1519 1520 /* /proc */ 1521 enum tcp_seq_states { 1522 TCP_SEQ_STATE_LISTENING, 1523 TCP_SEQ_STATE_OPENREQ, 1524 TCP_SEQ_STATE_ESTABLISHED, 1525 }; 1526 1527 int tcp_seq_open(struct inode *inode, struct file *file); 1528 1529 struct tcp_seq_afinfo { 1530 char *name; 1531 sa_family_t family; 1532 const struct file_operations *seq_fops; 1533 struct seq_operations seq_ops; 1534 }; 1535 1536 struct tcp_iter_state { 1537 struct seq_net_private p; 1538 sa_family_t family; 1539 enum tcp_seq_states state; 1540 struct sock *syn_wait_sk; 1541 int bucket, offset, sbucket, num; 1542 kuid_t uid; 1543 loff_t last_pos; 1544 }; 1545 1546 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1547 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1548 1549 extern struct request_sock_ops tcp_request_sock_ops; 1550 extern struct request_sock_ops tcp6_request_sock_ops; 1551 1552 void tcp_v4_destroy_sock(struct sock *sk); 1553 1554 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1555 netdev_features_t features); 1556 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1557 int tcp_gro_complete(struct sk_buff *skb); 1558 1559 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1560 1561 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1562 { 1563 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat; 1564 } 1565 1566 static inline bool tcp_stream_memory_free(const struct sock *sk) 1567 { 1568 const struct tcp_sock *tp = tcp_sk(sk); 1569 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1570 1571 return notsent_bytes < tcp_notsent_lowat(tp); 1572 } 1573 1574 #ifdef CONFIG_PROC_FS 1575 int tcp4_proc_init(void); 1576 void tcp4_proc_exit(void); 1577 #endif 1578 1579 int tcp_rtx_synack(struct sock *sk, struct request_sock *req); 1580 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1581 const struct tcp_request_sock_ops *af_ops, 1582 struct sock *sk, struct sk_buff *skb); 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 u16 mss_clamp; 1602 #ifdef CONFIG_TCP_MD5SIG 1603 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1604 struct request_sock *req); 1605 int (*calc_md5_hash) (char *location, 1606 struct tcp_md5sig_key *md5, 1607 const struct sock *sk, 1608 const struct request_sock *req, 1609 const struct sk_buff *skb); 1610 #endif 1611 void (*init_req)(struct request_sock *req, struct sock *sk, 1612 struct sk_buff *skb); 1613 #ifdef CONFIG_SYN_COOKIES 1614 __u32 (*cookie_init_seq)(struct sock *sk, const struct sk_buff *skb, 1615 __u16 *mss); 1616 #endif 1617 struct dst_entry *(*route_req)(struct sock *sk, struct flowi *fl, 1618 const struct request_sock *req, 1619 bool *strict); 1620 __u32 (*init_seq)(const struct sk_buff *skb); 1621 int (*send_synack)(struct sock *sk, struct dst_entry *dst, 1622 struct flowi *fl, struct request_sock *req, 1623 u16 queue_mapping, struct tcp_fastopen_cookie *foc); 1624 void (*queue_hash_add)(struct sock *sk, struct request_sock *req, 1625 const unsigned long timeout); 1626 }; 1627 1628 #ifdef CONFIG_SYN_COOKIES 1629 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1630 struct sock *sk, struct sk_buff *skb, 1631 __u16 *mss) 1632 { 1633 return ops->cookie_init_seq(sk, skb, mss); 1634 } 1635 #else 1636 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1637 struct sock *sk, struct sk_buff *skb, 1638 __u16 *mss) 1639 { 1640 return 0; 1641 } 1642 #endif 1643 1644 int tcpv4_offload_init(void); 1645 1646 void tcp_v4_init(void); 1647 void tcp_init(void); 1648 1649 #endif /* _TCP_H */ 1650