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, int how); 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); 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 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 452 struct sock *tcp_create_openreq_child(struct sock *sk, 453 struct request_sock *req, 454 struct sk_buff *skb); 455 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 456 struct request_sock *req, 457 struct dst_entry *dst); 458 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 459 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 460 int tcp_connect(struct sock *sk); 461 struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst, 462 struct request_sock *req, 463 struct tcp_fastopen_cookie *foc); 464 int tcp_disconnect(struct sock *sk, int flags); 465 466 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 467 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 468 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 469 470 /* From syncookies.c */ 471 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 472 u32 cookie); 473 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 474 struct ip_options *opt); 475 #ifdef CONFIG_SYN_COOKIES 476 477 /* Syncookies use a monotonic timer which increments every 60 seconds. 478 * This counter is used both as a hash input and partially encoded into 479 * the cookie value. A cookie is only validated further if the delta 480 * between the current counter value and the encoded one is less than this, 481 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 482 * the counter advances immediately after a cookie is generated). 483 */ 484 #define MAX_SYNCOOKIE_AGE 2 485 486 static inline u32 tcp_cookie_time(void) 487 { 488 u64 val = get_jiffies_64(); 489 490 do_div(val, 60 * HZ); 491 return val; 492 } 493 494 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 495 u16 *mssp); 496 __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, __u16 *mss); 497 #else 498 static inline __u32 cookie_v4_init_sequence(struct sock *sk, 499 struct sk_buff *skb, 500 __u16 *mss) 501 { 502 return 0; 503 } 504 #endif 505 506 __u32 cookie_init_timestamp(struct request_sock *req); 507 bool cookie_check_timestamp(struct tcp_options_received *opt, struct net *net, 508 bool *ecn_ok); 509 510 /* From net/ipv6/syncookies.c */ 511 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 512 u32 cookie); 513 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 514 #ifdef CONFIG_SYN_COOKIES 515 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 516 const struct tcphdr *th, u16 *mssp); 517 __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb, 518 __u16 *mss); 519 #else 520 static inline __u32 cookie_v6_init_sequence(struct sock *sk, 521 struct sk_buff *skb, 522 __u16 *mss) 523 { 524 return 0; 525 } 526 #endif 527 /* tcp_output.c */ 528 529 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 530 int nonagle); 531 bool tcp_may_send_now(struct sock *sk); 532 int __tcp_retransmit_skb(struct sock *, struct sk_buff *); 533 int tcp_retransmit_skb(struct sock *, struct sk_buff *); 534 void tcp_retransmit_timer(struct sock *sk); 535 void tcp_xmit_retransmit_queue(struct sock *); 536 void tcp_simple_retransmit(struct sock *); 537 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 538 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t); 539 540 void tcp_send_probe0(struct sock *); 541 void tcp_send_partial(struct sock *); 542 int tcp_write_wakeup(struct sock *); 543 void tcp_send_fin(struct sock *sk); 544 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 545 int tcp_send_synack(struct sock *); 546 bool tcp_syn_flood_action(struct sock *sk, const struct sk_buff *skb, 547 const char *proto); 548 void tcp_push_one(struct sock *, unsigned int mss_now); 549 void tcp_send_ack(struct sock *sk); 550 void tcp_send_delayed_ack(struct sock *sk); 551 void tcp_send_loss_probe(struct sock *sk); 552 bool tcp_schedule_loss_probe(struct sock *sk); 553 554 /* tcp_input.c */ 555 void tcp_resume_early_retransmit(struct sock *sk); 556 void tcp_rearm_rto(struct sock *sk); 557 void tcp_reset(struct sock *sk); 558 559 /* tcp_timer.c */ 560 void tcp_init_xmit_timers(struct sock *); 561 static inline void tcp_clear_xmit_timers(struct sock *sk) 562 { 563 inet_csk_clear_xmit_timers(sk); 564 } 565 566 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 567 unsigned int tcp_current_mss(struct sock *sk); 568 569 /* Bound MSS / TSO packet size with the half of the window */ 570 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 571 { 572 int cutoff; 573 574 /* When peer uses tiny windows, there is no use in packetizing 575 * to sub-MSS pieces for the sake of SWS or making sure there 576 * are enough packets in the pipe for fast recovery. 577 * 578 * On the other hand, for extremely large MSS devices, handling 579 * smaller than MSS windows in this way does make sense. 580 */ 581 if (tp->max_window >= 512) 582 cutoff = (tp->max_window >> 1); 583 else 584 cutoff = tp->max_window; 585 586 if (cutoff && pktsize > cutoff) 587 return max_t(int, cutoff, 68U - tp->tcp_header_len); 588 else 589 return pktsize; 590 } 591 592 /* tcp.c */ 593 void tcp_get_info(const struct sock *, struct tcp_info *); 594 595 /* Read 'sendfile()'-style from a TCP socket */ 596 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 597 unsigned int, size_t); 598 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 599 sk_read_actor_t recv_actor); 600 601 void tcp_initialize_rcv_mss(struct sock *sk); 602 603 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 604 int tcp_mss_to_mtu(struct sock *sk, int mss); 605 void tcp_mtup_init(struct sock *sk); 606 void tcp_init_buffer_space(struct sock *sk); 607 608 static inline void tcp_bound_rto(const struct sock *sk) 609 { 610 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 611 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 612 } 613 614 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 615 { 616 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 617 } 618 619 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 620 { 621 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 622 ntohl(TCP_FLAG_ACK) | 623 snd_wnd); 624 } 625 626 static inline void tcp_fast_path_on(struct tcp_sock *tp) 627 { 628 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 629 } 630 631 static inline void tcp_fast_path_check(struct sock *sk) 632 { 633 struct tcp_sock *tp = tcp_sk(sk); 634 635 if (skb_queue_empty(&tp->out_of_order_queue) && 636 tp->rcv_wnd && 637 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 638 !tp->urg_data) 639 tcp_fast_path_on(tp); 640 } 641 642 /* Compute the actual rto_min value */ 643 static inline u32 tcp_rto_min(struct sock *sk) 644 { 645 const struct dst_entry *dst = __sk_dst_get(sk); 646 u32 rto_min = TCP_RTO_MIN; 647 648 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 649 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 650 return rto_min; 651 } 652 653 static inline u32 tcp_rto_min_us(struct sock *sk) 654 { 655 return jiffies_to_usecs(tcp_rto_min(sk)); 656 } 657 658 /* Compute the actual receive window we are currently advertising. 659 * Rcv_nxt can be after the window if our peer push more data 660 * than the offered window. 661 */ 662 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 663 { 664 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 665 666 if (win < 0) 667 win = 0; 668 return (u32) win; 669 } 670 671 /* Choose a new window, without checks for shrinking, and without 672 * scaling applied to the result. The caller does these things 673 * if necessary. This is a "raw" window selection. 674 */ 675 u32 __tcp_select_window(struct sock *sk); 676 677 void tcp_send_window_probe(struct sock *sk); 678 679 /* TCP timestamps are only 32-bits, this causes a slight 680 * complication on 64-bit systems since we store a snapshot 681 * of jiffies in the buffer control blocks below. We decided 682 * to use only the low 32-bits of jiffies and hide the ugly 683 * casts with the following macro. 684 */ 685 #define tcp_time_stamp ((__u32)(jiffies)) 686 687 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 688 689 #define TCPHDR_FIN 0x01 690 #define TCPHDR_SYN 0x02 691 #define TCPHDR_RST 0x04 692 #define TCPHDR_PSH 0x08 693 #define TCPHDR_ACK 0x10 694 #define TCPHDR_URG 0x20 695 #define TCPHDR_ECE 0x40 696 #define TCPHDR_CWR 0x80 697 698 /* This is what the send packet queuing engine uses to pass 699 * TCP per-packet control information to the transmission code. 700 * We also store the host-order sequence numbers in here too. 701 * This is 44 bytes if IPV6 is enabled. 702 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 703 */ 704 struct tcp_skb_cb { 705 union { 706 struct inet_skb_parm h4; 707 #if IS_ENABLED(CONFIG_IPV6) 708 struct inet6_skb_parm h6; 709 #endif 710 } header; /* For incoming frames */ 711 __u32 seq; /* Starting sequence number */ 712 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 713 __u32 when; /* used to compute rtt's */ 714 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 715 716 __u8 sacked; /* State flags for SACK/FACK. */ 717 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 718 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 719 #define TCPCB_LOST 0x04 /* SKB is lost */ 720 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 721 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 722 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 723 724 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 725 /* 1 byte hole */ 726 __u32 ack_seq; /* Sequence number ACK'd */ 727 }; 728 729 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 730 731 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set 732 * 733 * If we receive a SYN packet with these bits set, it means a network is 734 * playing bad games with TOS bits. In order to avoid possible false congestion 735 * notifications, we disable TCP ECN negociation. 736 */ 737 static inline void 738 TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb, 739 struct net *net) 740 { 741 const struct tcphdr *th = tcp_hdr(skb); 742 743 if (net->ipv4.sysctl_tcp_ecn && th->ece && th->cwr && 744 INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield)) 745 inet_rsk(req)->ecn_ok = 1; 746 } 747 748 /* Due to TSO, an SKB can be composed of multiple actual 749 * packets. To keep these tracked properly, we use this. 750 */ 751 static inline int tcp_skb_pcount(const struct sk_buff *skb) 752 { 753 return skb_shinfo(skb)->gso_segs; 754 } 755 756 /* This is valid iff tcp_skb_pcount() > 1. */ 757 static inline int tcp_skb_mss(const struct sk_buff *skb) 758 { 759 return skb_shinfo(skb)->gso_size; 760 } 761 762 /* Events passed to congestion control interface */ 763 enum tcp_ca_event { 764 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 765 CA_EVENT_CWND_RESTART, /* congestion window restart */ 766 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 767 CA_EVENT_LOSS, /* loss timeout */ 768 CA_EVENT_FAST_ACK, /* in sequence ack */ 769 CA_EVENT_SLOW_ACK, /* other ack */ 770 }; 771 772 /* 773 * Interface for adding new TCP congestion control handlers 774 */ 775 #define TCP_CA_NAME_MAX 16 776 #define TCP_CA_MAX 128 777 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 778 779 #define TCP_CONG_NON_RESTRICTED 0x1 780 781 struct tcp_congestion_ops { 782 struct list_head list; 783 unsigned long flags; 784 785 /* initialize private data (optional) */ 786 void (*init)(struct sock *sk); 787 /* cleanup private data (optional) */ 788 void (*release)(struct sock *sk); 789 790 /* return slow start threshold (required) */ 791 u32 (*ssthresh)(struct sock *sk); 792 /* do new cwnd calculation (required) */ 793 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 794 /* call before changing ca_state (optional) */ 795 void (*set_state)(struct sock *sk, u8 new_state); 796 /* call when cwnd event occurs (optional) */ 797 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 798 /* new value of cwnd after loss (optional) */ 799 u32 (*undo_cwnd)(struct sock *sk); 800 /* hook for packet ack accounting (optional) */ 801 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 802 /* get info for inet_diag (optional) */ 803 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 804 805 char name[TCP_CA_NAME_MAX]; 806 struct module *owner; 807 }; 808 809 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 810 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 811 812 void tcp_init_congestion_control(struct sock *sk); 813 void tcp_cleanup_congestion_control(struct sock *sk); 814 int tcp_set_default_congestion_control(const char *name); 815 void tcp_get_default_congestion_control(char *name); 816 void tcp_get_available_congestion_control(char *buf, size_t len); 817 void tcp_get_allowed_congestion_control(char *buf, size_t len); 818 int tcp_set_allowed_congestion_control(char *allowed); 819 int tcp_set_congestion_control(struct sock *sk, const char *name); 820 int tcp_slow_start(struct tcp_sock *tp, u32 acked); 821 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 822 823 extern struct tcp_congestion_ops tcp_init_congestion_ops; 824 u32 tcp_reno_ssthresh(struct sock *sk); 825 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 826 extern struct tcp_congestion_ops tcp_reno; 827 828 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 829 { 830 struct inet_connection_sock *icsk = inet_csk(sk); 831 832 if (icsk->icsk_ca_ops->set_state) 833 icsk->icsk_ca_ops->set_state(sk, ca_state); 834 icsk->icsk_ca_state = ca_state; 835 } 836 837 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 838 { 839 const struct inet_connection_sock *icsk = inet_csk(sk); 840 841 if (icsk->icsk_ca_ops->cwnd_event) 842 icsk->icsk_ca_ops->cwnd_event(sk, event); 843 } 844 845 /* These functions determine how the current flow behaves in respect of SACK 846 * handling. SACK is negotiated with the peer, and therefore it can vary 847 * between different flows. 848 * 849 * tcp_is_sack - SACK enabled 850 * tcp_is_reno - No SACK 851 * tcp_is_fack - FACK enabled, implies SACK enabled 852 */ 853 static inline int tcp_is_sack(const struct tcp_sock *tp) 854 { 855 return tp->rx_opt.sack_ok; 856 } 857 858 static inline bool tcp_is_reno(const struct tcp_sock *tp) 859 { 860 return !tcp_is_sack(tp); 861 } 862 863 static inline bool tcp_is_fack(const struct tcp_sock *tp) 864 { 865 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 866 } 867 868 static inline void tcp_enable_fack(struct tcp_sock *tp) 869 { 870 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 871 } 872 873 /* TCP early-retransmit (ER) is similar to but more conservative than 874 * the thin-dupack feature. Enable ER only if thin-dupack is disabled. 875 */ 876 static inline void tcp_enable_early_retrans(struct tcp_sock *tp) 877 { 878 tp->do_early_retrans = sysctl_tcp_early_retrans && 879 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack && 880 sysctl_tcp_reordering == 3; 881 } 882 883 static inline void tcp_disable_early_retrans(struct tcp_sock *tp) 884 { 885 tp->do_early_retrans = 0; 886 } 887 888 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 889 { 890 return tp->sacked_out + tp->lost_out; 891 } 892 893 /* This determines how many packets are "in the network" to the best 894 * of our knowledge. In many cases it is conservative, but where 895 * detailed information is available from the receiver (via SACK 896 * blocks etc.) we can make more aggressive calculations. 897 * 898 * Use this for decisions involving congestion control, use just 899 * tp->packets_out to determine if the send queue is empty or not. 900 * 901 * Read this equation as: 902 * 903 * "Packets sent once on transmission queue" MINUS 904 * "Packets left network, but not honestly ACKed yet" PLUS 905 * "Packets fast retransmitted" 906 */ 907 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 908 { 909 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 910 } 911 912 #define TCP_INFINITE_SSTHRESH 0x7fffffff 913 914 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 915 { 916 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 917 } 918 919 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 920 { 921 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 922 (1 << inet_csk(sk)->icsk_ca_state); 923 } 924 925 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 926 * The exception is cwnd reduction phase, when cwnd is decreasing towards 927 * ssthresh. 928 */ 929 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 930 { 931 const struct tcp_sock *tp = tcp_sk(sk); 932 933 if (tcp_in_cwnd_reduction(sk)) 934 return tp->snd_ssthresh; 935 else 936 return max(tp->snd_ssthresh, 937 ((tp->snd_cwnd >> 1) + 938 (tp->snd_cwnd >> 2))); 939 } 940 941 /* Use define here intentionally to get WARN_ON location shown at the caller */ 942 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 943 944 void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 945 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 946 947 /* The maximum number of MSS of available cwnd for which TSO defers 948 * sending if not using sysctl_tcp_tso_win_divisor. 949 */ 950 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 951 { 952 return 3; 953 } 954 955 /* Slow start with delack produces 3 packets of burst, so that 956 * it is safe "de facto". This will be the default - same as 957 * the default reordering threshold - but if reordering increases, 958 * we must be able to allow cwnd to burst at least this much in order 959 * to not pull it back when holes are filled. 960 */ 961 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 962 { 963 return tp->reordering; 964 } 965 966 /* Returns end sequence number of the receiver's advertised window */ 967 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 968 { 969 return tp->snd_una + tp->snd_wnd; 970 } 971 972 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 973 * flexible approach. The RFC suggests cwnd should not be raised unless 974 * it was fully used previously. And that's exactly what we do in 975 * congestion avoidance mode. But in slow start we allow cwnd to grow 976 * as long as the application has used half the cwnd. 977 * Example : 978 * cwnd is 10 (IW10), but application sends 9 frames. 979 * We allow cwnd to reach 18 when all frames are ACKed. 980 * This check is safe because it's as aggressive as slow start which already 981 * risks 100% overshoot. The advantage is that we discourage application to 982 * either send more filler packets or data to artificially blow up the cwnd 983 * usage, and allow application-limited process to probe bw more aggressively. 984 */ 985 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 986 { 987 const struct tcp_sock *tp = tcp_sk(sk); 988 989 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 990 if (tp->snd_cwnd <= tp->snd_ssthresh) 991 return tp->snd_cwnd < 2 * tp->max_packets_out; 992 993 return tp->is_cwnd_limited; 994 } 995 996 static inline void tcp_check_probe_timer(struct sock *sk) 997 { 998 const struct tcp_sock *tp = tcp_sk(sk); 999 const struct inet_connection_sock *icsk = inet_csk(sk); 1000 1001 if (!tp->packets_out && !icsk->icsk_pending) 1002 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1003 icsk->icsk_rto, TCP_RTO_MAX); 1004 } 1005 1006 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1007 { 1008 tp->snd_wl1 = seq; 1009 } 1010 1011 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1012 { 1013 tp->snd_wl1 = seq; 1014 } 1015 1016 /* 1017 * Calculate(/check) TCP checksum 1018 */ 1019 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1020 __be32 daddr, __wsum base) 1021 { 1022 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1023 } 1024 1025 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1026 { 1027 return __skb_checksum_complete(skb); 1028 } 1029 1030 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1031 { 1032 return !skb_csum_unnecessary(skb) && 1033 __tcp_checksum_complete(skb); 1034 } 1035 1036 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1037 1038 static inline void tcp_prequeue_init(struct tcp_sock *tp) 1039 { 1040 tp->ucopy.task = NULL; 1041 tp->ucopy.len = 0; 1042 tp->ucopy.memory = 0; 1043 skb_queue_head_init(&tp->ucopy.prequeue); 1044 #ifdef CONFIG_NET_DMA 1045 tp->ucopy.dma_chan = NULL; 1046 tp->ucopy.wakeup = 0; 1047 tp->ucopy.pinned_list = NULL; 1048 tp->ucopy.dma_cookie = 0; 1049 #endif 1050 } 1051 1052 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1053 1054 #undef STATE_TRACE 1055 1056 #ifdef STATE_TRACE 1057 static const char *statename[]={ 1058 "Unused","Established","Syn Sent","Syn Recv", 1059 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1060 "Close Wait","Last ACK","Listen","Closing" 1061 }; 1062 #endif 1063 void tcp_set_state(struct sock *sk, int state); 1064 1065 void tcp_done(struct sock *sk); 1066 1067 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1068 { 1069 rx_opt->dsack = 0; 1070 rx_opt->num_sacks = 0; 1071 } 1072 1073 u32 tcp_default_init_rwnd(u32 mss); 1074 1075 /* Determine a window scaling and initial window to offer. */ 1076 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1077 __u32 *window_clamp, int wscale_ok, 1078 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1079 1080 static inline int tcp_win_from_space(int space) 1081 { 1082 return sysctl_tcp_adv_win_scale<=0 ? 1083 (space>>(-sysctl_tcp_adv_win_scale)) : 1084 space - (space>>sysctl_tcp_adv_win_scale); 1085 } 1086 1087 /* Note: caller must be prepared to deal with negative returns */ 1088 static inline int tcp_space(const struct sock *sk) 1089 { 1090 return tcp_win_from_space(sk->sk_rcvbuf - 1091 atomic_read(&sk->sk_rmem_alloc)); 1092 } 1093 1094 static inline int tcp_full_space(const struct sock *sk) 1095 { 1096 return tcp_win_from_space(sk->sk_rcvbuf); 1097 } 1098 1099 static inline void tcp_openreq_init(struct request_sock *req, 1100 struct tcp_options_received *rx_opt, 1101 struct sk_buff *skb) 1102 { 1103 struct inet_request_sock *ireq = inet_rsk(req); 1104 1105 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 1106 req->cookie_ts = 0; 1107 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 1108 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1; 1109 tcp_rsk(req)->snt_synack = 0; 1110 req->mss = rx_opt->mss_clamp; 1111 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 1112 ireq->tstamp_ok = rx_opt->tstamp_ok; 1113 ireq->sack_ok = rx_opt->sack_ok; 1114 ireq->snd_wscale = rx_opt->snd_wscale; 1115 ireq->wscale_ok = rx_opt->wscale_ok; 1116 ireq->acked = 0; 1117 ireq->ecn_ok = 0; 1118 ireq->ir_rmt_port = tcp_hdr(skb)->source; 1119 ireq->ir_num = ntohs(tcp_hdr(skb)->dest); 1120 } 1121 1122 extern void tcp_openreq_init_rwin(struct request_sock *req, 1123 struct sock *sk, struct dst_entry *dst); 1124 1125 void tcp_enter_memory_pressure(struct sock *sk); 1126 1127 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1128 { 1129 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1130 } 1131 1132 static inline int keepalive_time_when(const struct tcp_sock *tp) 1133 { 1134 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1135 } 1136 1137 static inline int keepalive_probes(const struct tcp_sock *tp) 1138 { 1139 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1140 } 1141 1142 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1143 { 1144 const struct inet_connection_sock *icsk = &tp->inet_conn; 1145 1146 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1147 tcp_time_stamp - tp->rcv_tstamp); 1148 } 1149 1150 static inline int tcp_fin_time(const struct sock *sk) 1151 { 1152 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1153 const int rto = inet_csk(sk)->icsk_rto; 1154 1155 if (fin_timeout < (rto << 2) - (rto >> 1)) 1156 fin_timeout = (rto << 2) - (rto >> 1); 1157 1158 return fin_timeout; 1159 } 1160 1161 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1162 int paws_win) 1163 { 1164 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1165 return true; 1166 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1167 return true; 1168 /* 1169 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1170 * then following tcp messages have valid values. Ignore 0 value, 1171 * or else 'negative' tsval might forbid us to accept their packets. 1172 */ 1173 if (!rx_opt->ts_recent) 1174 return true; 1175 return false; 1176 } 1177 1178 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1179 int rst) 1180 { 1181 if (tcp_paws_check(rx_opt, 0)) 1182 return false; 1183 1184 /* RST segments are not recommended to carry timestamp, 1185 and, if they do, it is recommended to ignore PAWS because 1186 "their cleanup function should take precedence over timestamps." 1187 Certainly, it is mistake. It is necessary to understand the reasons 1188 of this constraint to relax it: if peer reboots, clock may go 1189 out-of-sync and half-open connections will not be reset. 1190 Actually, the problem would be not existing if all 1191 the implementations followed draft about maintaining clock 1192 via reboots. Linux-2.2 DOES NOT! 1193 1194 However, we can relax time bounds for RST segments to MSL. 1195 */ 1196 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1197 return false; 1198 return true; 1199 } 1200 1201 static inline void tcp_mib_init(struct net *net) 1202 { 1203 /* See RFC 2012 */ 1204 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1205 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1206 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1207 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1208 } 1209 1210 /* from STCP */ 1211 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1212 { 1213 tp->lost_skb_hint = NULL; 1214 } 1215 1216 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1217 { 1218 tcp_clear_retrans_hints_partial(tp); 1219 tp->retransmit_skb_hint = NULL; 1220 } 1221 1222 /* MD5 Signature */ 1223 struct crypto_hash; 1224 1225 union tcp_md5_addr { 1226 struct in_addr a4; 1227 #if IS_ENABLED(CONFIG_IPV6) 1228 struct in6_addr a6; 1229 #endif 1230 }; 1231 1232 /* - key database */ 1233 struct tcp_md5sig_key { 1234 struct hlist_node node; 1235 u8 keylen; 1236 u8 family; /* AF_INET or AF_INET6 */ 1237 union tcp_md5_addr addr; 1238 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1239 struct rcu_head rcu; 1240 }; 1241 1242 /* - sock block */ 1243 struct tcp_md5sig_info { 1244 struct hlist_head head; 1245 struct rcu_head rcu; 1246 }; 1247 1248 /* - pseudo header */ 1249 struct tcp4_pseudohdr { 1250 __be32 saddr; 1251 __be32 daddr; 1252 __u8 pad; 1253 __u8 protocol; 1254 __be16 len; 1255 }; 1256 1257 struct tcp6_pseudohdr { 1258 struct in6_addr saddr; 1259 struct in6_addr daddr; 1260 __be32 len; 1261 __be32 protocol; /* including padding */ 1262 }; 1263 1264 union tcp_md5sum_block { 1265 struct tcp4_pseudohdr ip4; 1266 #if IS_ENABLED(CONFIG_IPV6) 1267 struct tcp6_pseudohdr ip6; 1268 #endif 1269 }; 1270 1271 /* - pool: digest algorithm, hash description and scratch buffer */ 1272 struct tcp_md5sig_pool { 1273 struct hash_desc md5_desc; 1274 union tcp_md5sum_block md5_blk; 1275 }; 1276 1277 /* - functions */ 1278 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1279 const struct sock *sk, const struct request_sock *req, 1280 const struct sk_buff *skb); 1281 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1282 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); 1283 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1284 int family); 1285 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1286 struct sock *addr_sk); 1287 1288 #ifdef CONFIG_TCP_MD5SIG 1289 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1290 const union tcp_md5_addr *addr, 1291 int family); 1292 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1293 #else 1294 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1295 const union tcp_md5_addr *addr, 1296 int family) 1297 { 1298 return NULL; 1299 } 1300 #define tcp_twsk_md5_key(twsk) NULL 1301 #endif 1302 1303 bool tcp_alloc_md5sig_pool(void); 1304 1305 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1306 static inline void tcp_put_md5sig_pool(void) 1307 { 1308 local_bh_enable(); 1309 } 1310 1311 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1312 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1313 unsigned int header_len); 1314 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1315 const struct tcp_md5sig_key *key); 1316 1317 /* From tcp_fastopen.c */ 1318 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1319 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1320 unsigned long *last_syn_loss); 1321 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1322 struct tcp_fastopen_cookie *cookie, bool syn_lost); 1323 struct tcp_fastopen_request { 1324 /* Fast Open cookie. Size 0 means a cookie request */ 1325 struct tcp_fastopen_cookie cookie; 1326 struct msghdr *data; /* data in MSG_FASTOPEN */ 1327 size_t size; 1328 int copied; /* queued in tcp_connect() */ 1329 }; 1330 void tcp_free_fastopen_req(struct tcp_sock *tp); 1331 1332 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1333 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1334 bool tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1335 struct request_sock *req, 1336 struct tcp_fastopen_cookie *foc, 1337 struct dst_entry *dst); 1338 void tcp_fastopen_init_key_once(bool publish); 1339 #define TCP_FASTOPEN_KEY_LENGTH 16 1340 1341 /* Fastopen key context */ 1342 struct tcp_fastopen_context { 1343 struct crypto_cipher *tfm; 1344 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1345 struct rcu_head rcu; 1346 }; 1347 1348 /* write queue abstraction */ 1349 static inline void tcp_write_queue_purge(struct sock *sk) 1350 { 1351 struct sk_buff *skb; 1352 1353 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1354 sk_wmem_free_skb(sk, skb); 1355 sk_mem_reclaim(sk); 1356 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1357 } 1358 1359 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1360 { 1361 return skb_peek(&sk->sk_write_queue); 1362 } 1363 1364 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1365 { 1366 return skb_peek_tail(&sk->sk_write_queue); 1367 } 1368 1369 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1370 const struct sk_buff *skb) 1371 { 1372 return skb_queue_next(&sk->sk_write_queue, skb); 1373 } 1374 1375 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1376 const struct sk_buff *skb) 1377 { 1378 return skb_queue_prev(&sk->sk_write_queue, skb); 1379 } 1380 1381 #define tcp_for_write_queue(skb, sk) \ 1382 skb_queue_walk(&(sk)->sk_write_queue, skb) 1383 1384 #define tcp_for_write_queue_from(skb, sk) \ 1385 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1386 1387 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1388 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1389 1390 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1391 { 1392 return sk->sk_send_head; 1393 } 1394 1395 static inline bool tcp_skb_is_last(const struct sock *sk, 1396 const struct sk_buff *skb) 1397 { 1398 return skb_queue_is_last(&sk->sk_write_queue, skb); 1399 } 1400 1401 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1402 { 1403 if (tcp_skb_is_last(sk, skb)) 1404 sk->sk_send_head = NULL; 1405 else 1406 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1407 } 1408 1409 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1410 { 1411 if (sk->sk_send_head == skb_unlinked) 1412 sk->sk_send_head = NULL; 1413 } 1414 1415 static inline void tcp_init_send_head(struct sock *sk) 1416 { 1417 sk->sk_send_head = NULL; 1418 } 1419 1420 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1421 { 1422 __skb_queue_tail(&sk->sk_write_queue, skb); 1423 } 1424 1425 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1426 { 1427 __tcp_add_write_queue_tail(sk, skb); 1428 1429 /* Queue it, remembering where we must start sending. */ 1430 if (sk->sk_send_head == NULL) { 1431 sk->sk_send_head = skb; 1432 1433 if (tcp_sk(sk)->highest_sack == NULL) 1434 tcp_sk(sk)->highest_sack = skb; 1435 } 1436 } 1437 1438 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1439 { 1440 __skb_queue_head(&sk->sk_write_queue, skb); 1441 } 1442 1443 /* Insert buff after skb on the write queue of sk. */ 1444 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1445 struct sk_buff *buff, 1446 struct sock *sk) 1447 { 1448 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1449 } 1450 1451 /* Insert new before skb on the write queue of sk. */ 1452 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1453 struct sk_buff *skb, 1454 struct sock *sk) 1455 { 1456 __skb_queue_before(&sk->sk_write_queue, skb, new); 1457 1458 if (sk->sk_send_head == skb) 1459 sk->sk_send_head = new; 1460 } 1461 1462 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1463 { 1464 __skb_unlink(skb, &sk->sk_write_queue); 1465 } 1466 1467 static inline bool tcp_write_queue_empty(struct sock *sk) 1468 { 1469 return skb_queue_empty(&sk->sk_write_queue); 1470 } 1471 1472 static inline void tcp_push_pending_frames(struct sock *sk) 1473 { 1474 if (tcp_send_head(sk)) { 1475 struct tcp_sock *tp = tcp_sk(sk); 1476 1477 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1478 } 1479 } 1480 1481 /* Start sequence of the skb just after the highest skb with SACKed 1482 * bit, valid only if sacked_out > 0 or when the caller has ensured 1483 * validity by itself. 1484 */ 1485 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1486 { 1487 if (!tp->sacked_out) 1488 return tp->snd_una; 1489 1490 if (tp->highest_sack == NULL) 1491 return tp->snd_nxt; 1492 1493 return TCP_SKB_CB(tp->highest_sack)->seq; 1494 } 1495 1496 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1497 { 1498 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1499 tcp_write_queue_next(sk, skb); 1500 } 1501 1502 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1503 { 1504 return tcp_sk(sk)->highest_sack; 1505 } 1506 1507 static inline void tcp_highest_sack_reset(struct sock *sk) 1508 { 1509 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1510 } 1511 1512 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1513 static inline void tcp_highest_sack_combine(struct sock *sk, 1514 struct sk_buff *old, 1515 struct sk_buff *new) 1516 { 1517 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1518 tcp_sk(sk)->highest_sack = new; 1519 } 1520 1521 /* Determines whether this is a thin stream (which may suffer from 1522 * increased latency). Used to trigger latency-reducing mechanisms. 1523 */ 1524 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1525 { 1526 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1527 } 1528 1529 /* /proc */ 1530 enum tcp_seq_states { 1531 TCP_SEQ_STATE_LISTENING, 1532 TCP_SEQ_STATE_OPENREQ, 1533 TCP_SEQ_STATE_ESTABLISHED, 1534 }; 1535 1536 int tcp_seq_open(struct inode *inode, struct file *file); 1537 1538 struct tcp_seq_afinfo { 1539 char *name; 1540 sa_family_t family; 1541 const struct file_operations *seq_fops; 1542 struct seq_operations seq_ops; 1543 }; 1544 1545 struct tcp_iter_state { 1546 struct seq_net_private p; 1547 sa_family_t family; 1548 enum tcp_seq_states state; 1549 struct sock *syn_wait_sk; 1550 int bucket, offset, sbucket, num; 1551 kuid_t uid; 1552 loff_t last_pos; 1553 }; 1554 1555 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1556 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1557 1558 extern struct request_sock_ops tcp_request_sock_ops; 1559 extern struct request_sock_ops tcp6_request_sock_ops; 1560 1561 void tcp_v4_destroy_sock(struct sock *sk); 1562 1563 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1564 netdev_features_t features); 1565 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1566 int tcp_gro_complete(struct sk_buff *skb); 1567 1568 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1569 1570 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1571 { 1572 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat; 1573 } 1574 1575 static inline bool tcp_stream_memory_free(const struct sock *sk) 1576 { 1577 const struct tcp_sock *tp = tcp_sk(sk); 1578 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1579 1580 return notsent_bytes < tcp_notsent_lowat(tp); 1581 } 1582 1583 #ifdef CONFIG_PROC_FS 1584 int tcp4_proc_init(void); 1585 void tcp4_proc_exit(void); 1586 #endif 1587 1588 /* TCP af-specific functions */ 1589 struct tcp_sock_af_ops { 1590 #ifdef CONFIG_TCP_MD5SIG 1591 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1592 struct sock *addr_sk); 1593 int (*calc_md5_hash) (char *location, 1594 struct tcp_md5sig_key *md5, 1595 const struct sock *sk, 1596 const struct request_sock *req, 1597 const struct sk_buff *skb); 1598 int (*md5_parse) (struct sock *sk, 1599 char __user *optval, 1600 int optlen); 1601 #endif 1602 }; 1603 1604 struct tcp_request_sock_ops { 1605 #ifdef CONFIG_TCP_MD5SIG 1606 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1607 struct request_sock *req); 1608 int (*calc_md5_hash) (char *location, 1609 struct tcp_md5sig_key *md5, 1610 const struct sock *sk, 1611 const struct request_sock *req, 1612 const struct sk_buff *skb); 1613 #endif 1614 }; 1615 1616 int tcpv4_offload_init(void); 1617 1618 void tcp_v4_init(void); 1619 void tcp_init(void); 1620 1621 #endif /* _TCP_H */ 1622