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); 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, const struct sk_buff *skb, 497 __u16 *mss); 498 #endif 499 500 __u32 cookie_init_timestamp(struct request_sock *req); 501 bool cookie_check_timestamp(struct tcp_options_received *opt, struct net *net, 502 bool *ecn_ok); 503 504 /* From net/ipv6/syncookies.c */ 505 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 506 u32 cookie); 507 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 508 #ifdef CONFIG_SYN_COOKIES 509 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 510 const struct tcphdr *th, u16 *mssp); 511 __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb, 512 __u16 *mss); 513 #endif 514 /* tcp_output.c */ 515 516 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 517 int nonagle); 518 bool tcp_may_send_now(struct sock *sk); 519 int __tcp_retransmit_skb(struct sock *, struct sk_buff *); 520 int tcp_retransmit_skb(struct sock *, struct sk_buff *); 521 void tcp_retransmit_timer(struct sock *sk); 522 void tcp_xmit_retransmit_queue(struct sock *); 523 void tcp_simple_retransmit(struct sock *); 524 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 525 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t); 526 527 void tcp_send_probe0(struct sock *); 528 void tcp_send_partial(struct sock *); 529 int tcp_write_wakeup(struct sock *); 530 void tcp_send_fin(struct sock *sk); 531 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 532 int tcp_send_synack(struct sock *); 533 bool tcp_syn_flood_action(struct sock *sk, const struct sk_buff *skb, 534 const char *proto); 535 void tcp_push_one(struct sock *, unsigned int mss_now); 536 void tcp_send_ack(struct sock *sk); 537 void tcp_send_delayed_ack(struct sock *sk); 538 void tcp_send_loss_probe(struct sock *sk); 539 bool tcp_schedule_loss_probe(struct sock *sk); 540 541 /* tcp_input.c */ 542 void tcp_resume_early_retransmit(struct sock *sk); 543 void tcp_rearm_rto(struct sock *sk); 544 void tcp_reset(struct sock *sk); 545 546 /* tcp_timer.c */ 547 void tcp_init_xmit_timers(struct sock *); 548 static inline void tcp_clear_xmit_timers(struct sock *sk) 549 { 550 inet_csk_clear_xmit_timers(sk); 551 } 552 553 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 554 unsigned int tcp_current_mss(struct sock *sk); 555 556 /* Bound MSS / TSO packet size with the half of the window */ 557 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 558 { 559 int cutoff; 560 561 /* When peer uses tiny windows, there is no use in packetizing 562 * to sub-MSS pieces for the sake of SWS or making sure there 563 * are enough packets in the pipe for fast recovery. 564 * 565 * On the other hand, for extremely large MSS devices, handling 566 * smaller than MSS windows in this way does make sense. 567 */ 568 if (tp->max_window >= 512) 569 cutoff = (tp->max_window >> 1); 570 else 571 cutoff = tp->max_window; 572 573 if (cutoff && pktsize > cutoff) 574 return max_t(int, cutoff, 68U - tp->tcp_header_len); 575 else 576 return pktsize; 577 } 578 579 /* tcp.c */ 580 void tcp_get_info(const struct sock *, struct tcp_info *); 581 582 /* Read 'sendfile()'-style from a TCP socket */ 583 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 584 unsigned int, size_t); 585 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 586 sk_read_actor_t recv_actor); 587 588 void tcp_initialize_rcv_mss(struct sock *sk); 589 590 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 591 int tcp_mss_to_mtu(struct sock *sk, int mss); 592 void tcp_mtup_init(struct sock *sk); 593 void tcp_init_buffer_space(struct sock *sk); 594 595 static inline void tcp_bound_rto(const struct sock *sk) 596 { 597 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 598 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 599 } 600 601 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 602 { 603 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 604 } 605 606 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 607 { 608 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 609 ntohl(TCP_FLAG_ACK) | 610 snd_wnd); 611 } 612 613 static inline void tcp_fast_path_on(struct tcp_sock *tp) 614 { 615 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 616 } 617 618 static inline void tcp_fast_path_check(struct sock *sk) 619 { 620 struct tcp_sock *tp = tcp_sk(sk); 621 622 if (skb_queue_empty(&tp->out_of_order_queue) && 623 tp->rcv_wnd && 624 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 625 !tp->urg_data) 626 tcp_fast_path_on(tp); 627 } 628 629 /* Compute the actual rto_min value */ 630 static inline u32 tcp_rto_min(struct sock *sk) 631 { 632 const struct dst_entry *dst = __sk_dst_get(sk); 633 u32 rto_min = TCP_RTO_MIN; 634 635 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 636 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 637 return rto_min; 638 } 639 640 static inline u32 tcp_rto_min_us(struct sock *sk) 641 { 642 return jiffies_to_usecs(tcp_rto_min(sk)); 643 } 644 645 /* Compute the actual receive window we are currently advertising. 646 * Rcv_nxt can be after the window if our peer push more data 647 * than the offered window. 648 */ 649 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 650 { 651 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 652 653 if (win < 0) 654 win = 0; 655 return (u32) win; 656 } 657 658 /* Choose a new window, without checks for shrinking, and without 659 * scaling applied to the result. The caller does these things 660 * if necessary. This is a "raw" window selection. 661 */ 662 u32 __tcp_select_window(struct sock *sk); 663 664 void tcp_send_window_probe(struct sock *sk); 665 666 /* TCP timestamps are only 32-bits, this causes a slight 667 * complication on 64-bit systems since we store a snapshot 668 * of jiffies in the buffer control blocks below. We decided 669 * to use only the low 32-bits of jiffies and hide the ugly 670 * casts with the following macro. 671 */ 672 #define tcp_time_stamp ((__u32)(jiffies)) 673 674 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 675 676 #define TCPHDR_FIN 0x01 677 #define TCPHDR_SYN 0x02 678 #define TCPHDR_RST 0x04 679 #define TCPHDR_PSH 0x08 680 #define TCPHDR_ACK 0x10 681 #define TCPHDR_URG 0x20 682 #define TCPHDR_ECE 0x40 683 #define TCPHDR_CWR 0x80 684 685 /* This is what the send packet queuing engine uses to pass 686 * TCP per-packet control information to the transmission code. 687 * We also store the host-order sequence numbers in here too. 688 * This is 44 bytes if IPV6 is enabled. 689 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 690 */ 691 struct tcp_skb_cb { 692 union { 693 struct inet_skb_parm h4; 694 #if IS_ENABLED(CONFIG_IPV6) 695 struct inet6_skb_parm h6; 696 #endif 697 } header; /* For incoming frames */ 698 __u32 seq; /* Starting sequence number */ 699 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 700 __u32 when; /* used to compute rtt's */ 701 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 702 703 __u8 sacked; /* State flags for SACK/FACK. */ 704 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 705 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 706 #define TCPCB_LOST 0x04 /* SKB is lost */ 707 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 708 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 709 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 710 711 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 712 /* 1 byte hole */ 713 __u32 ack_seq; /* Sequence number ACK'd */ 714 }; 715 716 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 717 718 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set 719 * 720 * If we receive a SYN packet with these bits set, it means a network is 721 * playing bad games with TOS bits. In order to avoid possible false congestion 722 * notifications, we disable TCP ECN negociation. 723 */ 724 static inline void 725 TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb, 726 struct net *net) 727 { 728 const struct tcphdr *th = tcp_hdr(skb); 729 730 if (net->ipv4.sysctl_tcp_ecn && th->ece && th->cwr && 731 INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield)) 732 inet_rsk(req)->ecn_ok = 1; 733 } 734 735 /* Due to TSO, an SKB can be composed of multiple actual 736 * packets. To keep these tracked properly, we use this. 737 */ 738 static inline int tcp_skb_pcount(const struct sk_buff *skb) 739 { 740 return skb_shinfo(skb)->gso_segs; 741 } 742 743 /* This is valid iff tcp_skb_pcount() > 1. */ 744 static inline int tcp_skb_mss(const struct sk_buff *skb) 745 { 746 return skb_shinfo(skb)->gso_size; 747 } 748 749 /* Events passed to congestion control interface */ 750 enum tcp_ca_event { 751 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 752 CA_EVENT_CWND_RESTART, /* congestion window restart */ 753 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 754 CA_EVENT_LOSS, /* loss timeout */ 755 CA_EVENT_FAST_ACK, /* in sequence ack */ 756 CA_EVENT_SLOW_ACK, /* other ack */ 757 }; 758 759 /* 760 * Interface for adding new TCP congestion control handlers 761 */ 762 #define TCP_CA_NAME_MAX 16 763 #define TCP_CA_MAX 128 764 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 765 766 #define TCP_CONG_NON_RESTRICTED 0x1 767 768 struct tcp_congestion_ops { 769 struct list_head list; 770 unsigned long flags; 771 772 /* initialize private data (optional) */ 773 void (*init)(struct sock *sk); 774 /* cleanup private data (optional) */ 775 void (*release)(struct sock *sk); 776 777 /* return slow start threshold (required) */ 778 u32 (*ssthresh)(struct sock *sk); 779 /* do new cwnd calculation (required) */ 780 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 781 /* call before changing ca_state (optional) */ 782 void (*set_state)(struct sock *sk, u8 new_state); 783 /* call when cwnd event occurs (optional) */ 784 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 785 /* new value of cwnd after loss (optional) */ 786 u32 (*undo_cwnd)(struct sock *sk); 787 /* hook for packet ack accounting (optional) */ 788 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 789 /* get info for inet_diag (optional) */ 790 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 791 792 char name[TCP_CA_NAME_MAX]; 793 struct module *owner; 794 }; 795 796 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 797 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 798 799 void tcp_init_congestion_control(struct sock *sk); 800 void tcp_cleanup_congestion_control(struct sock *sk); 801 int tcp_set_default_congestion_control(const char *name); 802 void tcp_get_default_congestion_control(char *name); 803 void tcp_get_available_congestion_control(char *buf, size_t len); 804 void tcp_get_allowed_congestion_control(char *buf, size_t len); 805 int tcp_set_allowed_congestion_control(char *allowed); 806 int tcp_set_congestion_control(struct sock *sk, const char *name); 807 int tcp_slow_start(struct tcp_sock *tp, u32 acked); 808 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 809 810 extern struct tcp_congestion_ops tcp_init_congestion_ops; 811 u32 tcp_reno_ssthresh(struct sock *sk); 812 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 813 extern struct tcp_congestion_ops tcp_reno; 814 815 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 816 { 817 struct inet_connection_sock *icsk = inet_csk(sk); 818 819 if (icsk->icsk_ca_ops->set_state) 820 icsk->icsk_ca_ops->set_state(sk, ca_state); 821 icsk->icsk_ca_state = ca_state; 822 } 823 824 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 825 { 826 const struct inet_connection_sock *icsk = inet_csk(sk); 827 828 if (icsk->icsk_ca_ops->cwnd_event) 829 icsk->icsk_ca_ops->cwnd_event(sk, event); 830 } 831 832 /* These functions determine how the current flow behaves in respect of SACK 833 * handling. SACK is negotiated with the peer, and therefore it can vary 834 * between different flows. 835 * 836 * tcp_is_sack - SACK enabled 837 * tcp_is_reno - No SACK 838 * tcp_is_fack - FACK enabled, implies SACK enabled 839 */ 840 static inline int tcp_is_sack(const struct tcp_sock *tp) 841 { 842 return tp->rx_opt.sack_ok; 843 } 844 845 static inline bool tcp_is_reno(const struct tcp_sock *tp) 846 { 847 return !tcp_is_sack(tp); 848 } 849 850 static inline bool tcp_is_fack(const struct tcp_sock *tp) 851 { 852 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 853 } 854 855 static inline void tcp_enable_fack(struct tcp_sock *tp) 856 { 857 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 858 } 859 860 /* TCP early-retransmit (ER) is similar to but more conservative than 861 * the thin-dupack feature. Enable ER only if thin-dupack is disabled. 862 */ 863 static inline void tcp_enable_early_retrans(struct tcp_sock *tp) 864 { 865 tp->do_early_retrans = sysctl_tcp_early_retrans && 866 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack && 867 sysctl_tcp_reordering == 3; 868 } 869 870 static inline void tcp_disable_early_retrans(struct tcp_sock *tp) 871 { 872 tp->do_early_retrans = 0; 873 } 874 875 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 876 { 877 return tp->sacked_out + tp->lost_out; 878 } 879 880 /* This determines how many packets are "in the network" to the best 881 * of our knowledge. In many cases it is conservative, but where 882 * detailed information is available from the receiver (via SACK 883 * blocks etc.) we can make more aggressive calculations. 884 * 885 * Use this for decisions involving congestion control, use just 886 * tp->packets_out to determine if the send queue is empty or not. 887 * 888 * Read this equation as: 889 * 890 * "Packets sent once on transmission queue" MINUS 891 * "Packets left network, but not honestly ACKed yet" PLUS 892 * "Packets fast retransmitted" 893 */ 894 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 895 { 896 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 897 } 898 899 #define TCP_INFINITE_SSTHRESH 0x7fffffff 900 901 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 902 { 903 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 904 } 905 906 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 907 { 908 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 909 (1 << inet_csk(sk)->icsk_ca_state); 910 } 911 912 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 913 * The exception is cwnd reduction phase, when cwnd is decreasing towards 914 * ssthresh. 915 */ 916 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 917 { 918 const struct tcp_sock *tp = tcp_sk(sk); 919 920 if (tcp_in_cwnd_reduction(sk)) 921 return tp->snd_ssthresh; 922 else 923 return max(tp->snd_ssthresh, 924 ((tp->snd_cwnd >> 1) + 925 (tp->snd_cwnd >> 2))); 926 } 927 928 /* Use define here intentionally to get WARN_ON location shown at the caller */ 929 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 930 931 void tcp_enter_cwr(struct sock *sk); 932 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 933 934 /* The maximum number of MSS of available cwnd for which TSO defers 935 * sending if not using sysctl_tcp_tso_win_divisor. 936 */ 937 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 938 { 939 return 3; 940 } 941 942 /* Slow start with delack produces 3 packets of burst, so that 943 * it is safe "de facto". This will be the default - same as 944 * the default reordering threshold - but if reordering increases, 945 * we must be able to allow cwnd to burst at least this much in order 946 * to not pull it back when holes are filled. 947 */ 948 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 949 { 950 return tp->reordering; 951 } 952 953 /* Returns end sequence number of the receiver's advertised window */ 954 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 955 { 956 return tp->snd_una + tp->snd_wnd; 957 } 958 959 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 960 * flexible approach. The RFC suggests cwnd should not be raised unless 961 * it was fully used previously. And that's exactly what we do in 962 * congestion avoidance mode. But in slow start we allow cwnd to grow 963 * as long as the application has used half the cwnd. 964 * Example : 965 * cwnd is 10 (IW10), but application sends 9 frames. 966 * We allow cwnd to reach 18 when all frames are ACKed. 967 * This check is safe because it's as aggressive as slow start which already 968 * risks 100% overshoot. The advantage is that we discourage application to 969 * either send more filler packets or data to artificially blow up the cwnd 970 * usage, and allow application-limited process to probe bw more aggressively. 971 */ 972 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 973 { 974 const struct tcp_sock *tp = tcp_sk(sk); 975 976 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 977 if (tp->snd_cwnd <= tp->snd_ssthresh) 978 return tp->snd_cwnd < 2 * tp->max_packets_out; 979 980 return tp->is_cwnd_limited; 981 } 982 983 static inline void tcp_check_probe_timer(struct sock *sk) 984 { 985 const struct tcp_sock *tp = tcp_sk(sk); 986 const struct inet_connection_sock *icsk = inet_csk(sk); 987 988 if (!tp->packets_out && !icsk->icsk_pending) 989 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 990 icsk->icsk_rto, TCP_RTO_MAX); 991 } 992 993 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 994 { 995 tp->snd_wl1 = seq; 996 } 997 998 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 999 { 1000 tp->snd_wl1 = seq; 1001 } 1002 1003 /* 1004 * Calculate(/check) TCP checksum 1005 */ 1006 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1007 __be32 daddr, __wsum base) 1008 { 1009 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1010 } 1011 1012 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1013 { 1014 return __skb_checksum_complete(skb); 1015 } 1016 1017 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1018 { 1019 return !skb_csum_unnecessary(skb) && 1020 __tcp_checksum_complete(skb); 1021 } 1022 1023 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1024 1025 static inline void tcp_prequeue_init(struct tcp_sock *tp) 1026 { 1027 tp->ucopy.task = NULL; 1028 tp->ucopy.len = 0; 1029 tp->ucopy.memory = 0; 1030 skb_queue_head_init(&tp->ucopy.prequeue); 1031 #ifdef CONFIG_NET_DMA 1032 tp->ucopy.dma_chan = NULL; 1033 tp->ucopy.wakeup = 0; 1034 tp->ucopy.pinned_list = NULL; 1035 tp->ucopy.dma_cookie = 0; 1036 #endif 1037 } 1038 1039 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1040 1041 #undef STATE_TRACE 1042 1043 #ifdef STATE_TRACE 1044 static const char *statename[]={ 1045 "Unused","Established","Syn Sent","Syn Recv", 1046 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1047 "Close Wait","Last ACK","Listen","Closing" 1048 }; 1049 #endif 1050 void tcp_set_state(struct sock *sk, int state); 1051 1052 void tcp_done(struct sock *sk); 1053 1054 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1055 { 1056 rx_opt->dsack = 0; 1057 rx_opt->num_sacks = 0; 1058 } 1059 1060 u32 tcp_default_init_rwnd(u32 mss); 1061 1062 /* Determine a window scaling and initial window to offer. */ 1063 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1064 __u32 *window_clamp, int wscale_ok, 1065 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1066 1067 static inline int tcp_win_from_space(int space) 1068 { 1069 return sysctl_tcp_adv_win_scale<=0 ? 1070 (space>>(-sysctl_tcp_adv_win_scale)) : 1071 space - (space>>sysctl_tcp_adv_win_scale); 1072 } 1073 1074 /* Note: caller must be prepared to deal with negative returns */ 1075 static inline int tcp_space(const struct sock *sk) 1076 { 1077 return tcp_win_from_space(sk->sk_rcvbuf - 1078 atomic_read(&sk->sk_rmem_alloc)); 1079 } 1080 1081 static inline int tcp_full_space(const struct sock *sk) 1082 { 1083 return tcp_win_from_space(sk->sk_rcvbuf); 1084 } 1085 1086 static inline void tcp_openreq_init(struct request_sock *req, 1087 struct tcp_options_received *rx_opt, 1088 struct sk_buff *skb, struct sock *sk) 1089 { 1090 struct inet_request_sock *ireq = inet_rsk(req); 1091 1092 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 1093 req->cookie_ts = 0; 1094 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 1095 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1; 1096 tcp_rsk(req)->snt_synack = tcp_time_stamp; 1097 req->mss = rx_opt->mss_clamp; 1098 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 1099 ireq->tstamp_ok = rx_opt->tstamp_ok; 1100 ireq->sack_ok = rx_opt->sack_ok; 1101 ireq->snd_wscale = rx_opt->snd_wscale; 1102 ireq->wscale_ok = rx_opt->wscale_ok; 1103 ireq->acked = 0; 1104 ireq->ecn_ok = 0; 1105 ireq->ir_rmt_port = tcp_hdr(skb)->source; 1106 ireq->ir_num = ntohs(tcp_hdr(skb)->dest); 1107 ireq->ir_mark = inet_request_mark(sk, skb); 1108 } 1109 1110 extern void tcp_openreq_init_rwin(struct request_sock *req, 1111 struct sock *sk, struct dst_entry *dst); 1112 1113 void tcp_enter_memory_pressure(struct sock *sk); 1114 1115 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1116 { 1117 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1118 } 1119 1120 static inline int keepalive_time_when(const struct tcp_sock *tp) 1121 { 1122 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1123 } 1124 1125 static inline int keepalive_probes(const struct tcp_sock *tp) 1126 { 1127 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1128 } 1129 1130 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1131 { 1132 const struct inet_connection_sock *icsk = &tp->inet_conn; 1133 1134 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1135 tcp_time_stamp - tp->rcv_tstamp); 1136 } 1137 1138 static inline int tcp_fin_time(const struct sock *sk) 1139 { 1140 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1141 const int rto = inet_csk(sk)->icsk_rto; 1142 1143 if (fin_timeout < (rto << 2) - (rto >> 1)) 1144 fin_timeout = (rto << 2) - (rto >> 1); 1145 1146 return fin_timeout; 1147 } 1148 1149 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1150 int paws_win) 1151 { 1152 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1153 return true; 1154 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1155 return true; 1156 /* 1157 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1158 * then following tcp messages have valid values. Ignore 0 value, 1159 * or else 'negative' tsval might forbid us to accept their packets. 1160 */ 1161 if (!rx_opt->ts_recent) 1162 return true; 1163 return false; 1164 } 1165 1166 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1167 int rst) 1168 { 1169 if (tcp_paws_check(rx_opt, 0)) 1170 return false; 1171 1172 /* RST segments are not recommended to carry timestamp, 1173 and, if they do, it is recommended to ignore PAWS because 1174 "their cleanup function should take precedence over timestamps." 1175 Certainly, it is mistake. It is necessary to understand the reasons 1176 of this constraint to relax it: if peer reboots, clock may go 1177 out-of-sync and half-open connections will not be reset. 1178 Actually, the problem would be not existing if all 1179 the implementations followed draft about maintaining clock 1180 via reboots. Linux-2.2 DOES NOT! 1181 1182 However, we can relax time bounds for RST segments to MSL. 1183 */ 1184 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1185 return false; 1186 return true; 1187 } 1188 1189 static inline void tcp_mib_init(struct net *net) 1190 { 1191 /* See RFC 2012 */ 1192 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1193 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1194 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1195 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1196 } 1197 1198 /* from STCP */ 1199 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1200 { 1201 tp->lost_skb_hint = NULL; 1202 } 1203 1204 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1205 { 1206 tcp_clear_retrans_hints_partial(tp); 1207 tp->retransmit_skb_hint = NULL; 1208 } 1209 1210 /* MD5 Signature */ 1211 struct crypto_hash; 1212 1213 union tcp_md5_addr { 1214 struct in_addr a4; 1215 #if IS_ENABLED(CONFIG_IPV6) 1216 struct in6_addr a6; 1217 #endif 1218 }; 1219 1220 /* - key database */ 1221 struct tcp_md5sig_key { 1222 struct hlist_node node; 1223 u8 keylen; 1224 u8 family; /* AF_INET or AF_INET6 */ 1225 union tcp_md5_addr addr; 1226 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1227 struct rcu_head rcu; 1228 }; 1229 1230 /* - sock block */ 1231 struct tcp_md5sig_info { 1232 struct hlist_head head; 1233 struct rcu_head rcu; 1234 }; 1235 1236 /* - pseudo header */ 1237 struct tcp4_pseudohdr { 1238 __be32 saddr; 1239 __be32 daddr; 1240 __u8 pad; 1241 __u8 protocol; 1242 __be16 len; 1243 }; 1244 1245 struct tcp6_pseudohdr { 1246 struct in6_addr saddr; 1247 struct in6_addr daddr; 1248 __be32 len; 1249 __be32 protocol; /* including padding */ 1250 }; 1251 1252 union tcp_md5sum_block { 1253 struct tcp4_pseudohdr ip4; 1254 #if IS_ENABLED(CONFIG_IPV6) 1255 struct tcp6_pseudohdr ip6; 1256 #endif 1257 }; 1258 1259 /* - pool: digest algorithm, hash description and scratch buffer */ 1260 struct tcp_md5sig_pool { 1261 struct hash_desc md5_desc; 1262 union tcp_md5sum_block md5_blk; 1263 }; 1264 1265 /* - functions */ 1266 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1267 const struct sock *sk, const struct request_sock *req, 1268 const struct sk_buff *skb); 1269 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1270 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); 1271 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1272 int family); 1273 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1274 struct sock *addr_sk); 1275 1276 #ifdef CONFIG_TCP_MD5SIG 1277 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1278 const union tcp_md5_addr *addr, 1279 int family); 1280 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1281 #else 1282 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1283 const union tcp_md5_addr *addr, 1284 int family) 1285 { 1286 return NULL; 1287 } 1288 #define tcp_twsk_md5_key(twsk) NULL 1289 #endif 1290 1291 bool tcp_alloc_md5sig_pool(void); 1292 1293 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1294 static inline void tcp_put_md5sig_pool(void) 1295 { 1296 local_bh_enable(); 1297 } 1298 1299 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1300 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1301 unsigned int header_len); 1302 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1303 const struct tcp_md5sig_key *key); 1304 1305 /* From tcp_fastopen.c */ 1306 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1307 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1308 unsigned long *last_syn_loss); 1309 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1310 struct tcp_fastopen_cookie *cookie, bool syn_lost); 1311 struct tcp_fastopen_request { 1312 /* Fast Open cookie. Size 0 means a cookie request */ 1313 struct tcp_fastopen_cookie cookie; 1314 struct msghdr *data; /* data in MSG_FASTOPEN */ 1315 size_t size; 1316 int copied; /* queued in tcp_connect() */ 1317 }; 1318 void tcp_free_fastopen_req(struct tcp_sock *tp); 1319 1320 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1321 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1322 bool tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1323 struct request_sock *req, 1324 struct tcp_fastopen_cookie *foc, 1325 struct dst_entry *dst); 1326 void tcp_fastopen_init_key_once(bool publish); 1327 #define TCP_FASTOPEN_KEY_LENGTH 16 1328 1329 /* Fastopen key context */ 1330 struct tcp_fastopen_context { 1331 struct crypto_cipher *tfm; 1332 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1333 struct rcu_head rcu; 1334 }; 1335 1336 /* write queue abstraction */ 1337 static inline void tcp_write_queue_purge(struct sock *sk) 1338 { 1339 struct sk_buff *skb; 1340 1341 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1342 sk_wmem_free_skb(sk, skb); 1343 sk_mem_reclaim(sk); 1344 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1345 } 1346 1347 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1348 { 1349 return skb_peek(&sk->sk_write_queue); 1350 } 1351 1352 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1353 { 1354 return skb_peek_tail(&sk->sk_write_queue); 1355 } 1356 1357 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1358 const struct sk_buff *skb) 1359 { 1360 return skb_queue_next(&sk->sk_write_queue, skb); 1361 } 1362 1363 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1364 const struct sk_buff *skb) 1365 { 1366 return skb_queue_prev(&sk->sk_write_queue, skb); 1367 } 1368 1369 #define tcp_for_write_queue(skb, sk) \ 1370 skb_queue_walk(&(sk)->sk_write_queue, skb) 1371 1372 #define tcp_for_write_queue_from(skb, sk) \ 1373 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1374 1375 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1376 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1377 1378 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1379 { 1380 return sk->sk_send_head; 1381 } 1382 1383 static inline bool tcp_skb_is_last(const struct sock *sk, 1384 const struct sk_buff *skb) 1385 { 1386 return skb_queue_is_last(&sk->sk_write_queue, skb); 1387 } 1388 1389 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1390 { 1391 if (tcp_skb_is_last(sk, skb)) 1392 sk->sk_send_head = NULL; 1393 else 1394 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1395 } 1396 1397 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1398 { 1399 if (sk->sk_send_head == skb_unlinked) 1400 sk->sk_send_head = NULL; 1401 } 1402 1403 static inline void tcp_init_send_head(struct sock *sk) 1404 { 1405 sk->sk_send_head = NULL; 1406 } 1407 1408 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1409 { 1410 __skb_queue_tail(&sk->sk_write_queue, skb); 1411 } 1412 1413 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1414 { 1415 __tcp_add_write_queue_tail(sk, skb); 1416 1417 /* Queue it, remembering where we must start sending. */ 1418 if (sk->sk_send_head == NULL) { 1419 sk->sk_send_head = skb; 1420 1421 if (tcp_sk(sk)->highest_sack == NULL) 1422 tcp_sk(sk)->highest_sack = skb; 1423 } 1424 } 1425 1426 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1427 { 1428 __skb_queue_head(&sk->sk_write_queue, skb); 1429 } 1430 1431 /* Insert buff after skb on the write queue of sk. */ 1432 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1433 struct sk_buff *buff, 1434 struct sock *sk) 1435 { 1436 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1437 } 1438 1439 /* Insert new before skb on the write queue of sk. */ 1440 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1441 struct sk_buff *skb, 1442 struct sock *sk) 1443 { 1444 __skb_queue_before(&sk->sk_write_queue, skb, new); 1445 1446 if (sk->sk_send_head == skb) 1447 sk->sk_send_head = new; 1448 } 1449 1450 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1451 { 1452 __skb_unlink(skb, &sk->sk_write_queue); 1453 } 1454 1455 static inline bool tcp_write_queue_empty(struct sock *sk) 1456 { 1457 return skb_queue_empty(&sk->sk_write_queue); 1458 } 1459 1460 static inline void tcp_push_pending_frames(struct sock *sk) 1461 { 1462 if (tcp_send_head(sk)) { 1463 struct tcp_sock *tp = tcp_sk(sk); 1464 1465 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1466 } 1467 } 1468 1469 /* Start sequence of the skb just after the highest skb with SACKed 1470 * bit, valid only if sacked_out > 0 or when the caller has ensured 1471 * validity by itself. 1472 */ 1473 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1474 { 1475 if (!tp->sacked_out) 1476 return tp->snd_una; 1477 1478 if (tp->highest_sack == NULL) 1479 return tp->snd_nxt; 1480 1481 return TCP_SKB_CB(tp->highest_sack)->seq; 1482 } 1483 1484 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1485 { 1486 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1487 tcp_write_queue_next(sk, skb); 1488 } 1489 1490 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1491 { 1492 return tcp_sk(sk)->highest_sack; 1493 } 1494 1495 static inline void tcp_highest_sack_reset(struct sock *sk) 1496 { 1497 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1498 } 1499 1500 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1501 static inline void tcp_highest_sack_combine(struct sock *sk, 1502 struct sk_buff *old, 1503 struct sk_buff *new) 1504 { 1505 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1506 tcp_sk(sk)->highest_sack = new; 1507 } 1508 1509 /* Determines whether this is a thin stream (which may suffer from 1510 * increased latency). Used to trigger latency-reducing mechanisms. 1511 */ 1512 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1513 { 1514 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1515 } 1516 1517 /* /proc */ 1518 enum tcp_seq_states { 1519 TCP_SEQ_STATE_LISTENING, 1520 TCP_SEQ_STATE_OPENREQ, 1521 TCP_SEQ_STATE_ESTABLISHED, 1522 }; 1523 1524 int tcp_seq_open(struct inode *inode, struct file *file); 1525 1526 struct tcp_seq_afinfo { 1527 char *name; 1528 sa_family_t family; 1529 const struct file_operations *seq_fops; 1530 struct seq_operations seq_ops; 1531 }; 1532 1533 struct tcp_iter_state { 1534 struct seq_net_private p; 1535 sa_family_t family; 1536 enum tcp_seq_states state; 1537 struct sock *syn_wait_sk; 1538 int bucket, offset, sbucket, num; 1539 kuid_t uid; 1540 loff_t last_pos; 1541 }; 1542 1543 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1544 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1545 1546 extern struct request_sock_ops tcp_request_sock_ops; 1547 extern struct request_sock_ops tcp6_request_sock_ops; 1548 1549 void tcp_v4_destroy_sock(struct sock *sk); 1550 1551 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1552 netdev_features_t features); 1553 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1554 int tcp_gro_complete(struct sk_buff *skb); 1555 1556 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1557 1558 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1559 { 1560 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat; 1561 } 1562 1563 static inline bool tcp_stream_memory_free(const struct sock *sk) 1564 { 1565 const struct tcp_sock *tp = tcp_sk(sk); 1566 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1567 1568 return notsent_bytes < tcp_notsent_lowat(tp); 1569 } 1570 1571 #ifdef CONFIG_PROC_FS 1572 int tcp4_proc_init(void); 1573 void tcp4_proc_exit(void); 1574 #endif 1575 1576 int tcp_rtx_synack(struct sock *sk, struct request_sock *req); 1577 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1578 const struct tcp_request_sock_ops *af_ops, 1579 struct sock *sk, struct sk_buff *skb); 1580 1581 /* TCP af-specific functions */ 1582 struct tcp_sock_af_ops { 1583 #ifdef CONFIG_TCP_MD5SIG 1584 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1585 struct sock *addr_sk); 1586 int (*calc_md5_hash) (char *location, 1587 struct tcp_md5sig_key *md5, 1588 const struct sock *sk, 1589 const struct request_sock *req, 1590 const struct sk_buff *skb); 1591 int (*md5_parse) (struct sock *sk, 1592 char __user *optval, 1593 int optlen); 1594 #endif 1595 }; 1596 1597 struct tcp_request_sock_ops { 1598 u16 mss_clamp; 1599 #ifdef CONFIG_TCP_MD5SIG 1600 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1601 struct request_sock *req); 1602 int (*calc_md5_hash) (char *location, 1603 struct tcp_md5sig_key *md5, 1604 const struct sock *sk, 1605 const struct request_sock *req, 1606 const struct sk_buff *skb); 1607 #endif 1608 void (*init_req)(struct request_sock *req, struct sock *sk, 1609 struct sk_buff *skb); 1610 #ifdef CONFIG_SYN_COOKIES 1611 __u32 (*cookie_init_seq)(struct sock *sk, const struct sk_buff *skb, 1612 __u16 *mss); 1613 #endif 1614 struct dst_entry *(*route_req)(struct sock *sk, struct flowi *fl, 1615 const struct request_sock *req, 1616 bool *strict); 1617 __u32 (*init_seq)(const struct sk_buff *skb); 1618 int (*send_synack)(struct sock *sk, struct dst_entry *dst, 1619 struct flowi *fl, struct request_sock *req, 1620 u16 queue_mapping, struct tcp_fastopen_cookie *foc); 1621 void (*queue_hash_add)(struct sock *sk, struct request_sock *req, 1622 const unsigned long timeout); 1623 }; 1624 1625 #ifdef CONFIG_SYN_COOKIES 1626 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1627 struct sock *sk, struct sk_buff *skb, 1628 __u16 *mss) 1629 { 1630 return ops->cookie_init_seq(sk, skb, mss); 1631 } 1632 #else 1633 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1634 struct sock *sk, struct sk_buff *skb, 1635 __u16 *mss) 1636 { 1637 return 0; 1638 } 1639 #endif 1640 1641 int tcpv4_offload_init(void); 1642 1643 void tcp_v4_init(void); 1644 void tcp_init(void); 1645 1646 #endif /* _TCP_H */ 1647