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/crypto.h> 31 #include <linux/cryptohash.h> 32 #include <linux/kref.h> 33 #include <linux/ktime.h> 34 35 #include <net/inet_connection_sock.h> 36 #include <net/inet_timewait_sock.h> 37 #include <net/inet_hashtables.h> 38 #include <net/checksum.h> 39 #include <net/request_sock.h> 40 #include <net/sock.h> 41 #include <net/snmp.h> 42 #include <net/ip.h> 43 #include <net/tcp_states.h> 44 #include <net/inet_ecn.h> 45 #include <net/dst.h> 46 47 #include <linux/seq_file.h> 48 #include <linux/memcontrol.h> 49 50 extern struct inet_hashinfo tcp_hashinfo; 51 52 extern struct percpu_counter tcp_orphan_count; 53 void tcp_time_wait(struct sock *sk, int state, int timeo); 54 55 #define MAX_TCP_HEADER (128 + MAX_HEADER) 56 #define MAX_TCP_OPTION_SPACE 40 57 58 /* 59 * Never offer a window over 32767 without using window scaling. Some 60 * poor stacks do signed 16bit maths! 61 */ 62 #define MAX_TCP_WINDOW 32767U 63 64 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 65 #define TCP_MIN_MSS 88U 66 67 /* The least MTU to use for probing */ 68 #define TCP_BASE_MSS 512 69 70 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 71 #define TCP_FASTRETRANS_THRESH 3 72 73 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 74 #define TCP_MAX_QUICKACKS 16U 75 76 /* urg_data states */ 77 #define TCP_URG_VALID 0x0100 78 #define TCP_URG_NOTYET 0x0200 79 #define TCP_URG_READ 0x0400 80 81 #define TCP_RETR1 3 /* 82 * This is how many retries it does before it 83 * tries to figure out if the gateway is 84 * down. Minimal RFC value is 3; it corresponds 85 * to ~3sec-8min depending on RTO. 86 */ 87 88 #define TCP_RETR2 15 /* 89 * This should take at least 90 * 90 minutes to time out. 91 * RFC1122 says that the limit is 100 sec. 92 * 15 is ~13-30min depending on RTO. 93 */ 94 95 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 96 * when active opening a connection. 97 * RFC1122 says the minimum retry MUST 98 * be at least 180secs. Nevertheless 99 * this value is corresponding to 100 * 63secs of retransmission with the 101 * current initial RTO. 102 */ 103 104 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 105 * when passive opening a connection. 106 * This is corresponding to 31secs of 107 * retransmission with the current 108 * initial RTO. 109 */ 110 111 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 112 * state, about 60 seconds */ 113 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 114 /* BSD style FIN_WAIT2 deadlock breaker. 115 * It used to be 3min, new value is 60sec, 116 * to combine FIN-WAIT-2 timeout with 117 * TIME-WAIT timer. 118 */ 119 120 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 121 #if HZ >= 100 122 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 123 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 124 #else 125 #define TCP_DELACK_MIN 4U 126 #define TCP_ATO_MIN 4U 127 #endif 128 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 129 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 130 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 131 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 132 * used as a fallback RTO for the 133 * initial data transmission if no 134 * valid RTT sample has been acquired, 135 * most likely due to retrans in 3WHS. 136 */ 137 138 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 139 * for local resources. 140 */ 141 142 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 143 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 144 #define TCP_KEEPALIVE_INTVL (75*HZ) 145 146 #define MAX_TCP_KEEPIDLE 32767 147 #define MAX_TCP_KEEPINTVL 32767 148 #define MAX_TCP_KEEPCNT 127 149 #define MAX_TCP_SYNCNT 127 150 151 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 152 153 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 154 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 155 * after this time. It should be equal 156 * (or greater than) TCP_TIMEWAIT_LEN 157 * to provide reliability equal to one 158 * provided by timewait state. 159 */ 160 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 161 * timestamps. It must be less than 162 * minimal timewait lifetime. 163 */ 164 /* 165 * TCP option 166 */ 167 168 #define TCPOPT_NOP 1 /* Padding */ 169 #define TCPOPT_EOL 0 /* End of options */ 170 #define TCPOPT_MSS 2 /* Segment size negotiating */ 171 #define TCPOPT_WINDOW 3 /* Window scaling */ 172 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 173 #define TCPOPT_SACK 5 /* SACK Block */ 174 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 175 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 176 #define TCPOPT_EXP 254 /* Experimental */ 177 /* Magic number to be after the option value for sharing TCP 178 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 179 */ 180 #define TCPOPT_FASTOPEN_MAGIC 0xF989 181 182 /* 183 * TCP option lengths 184 */ 185 186 #define TCPOLEN_MSS 4 187 #define TCPOLEN_WINDOW 3 188 #define TCPOLEN_SACK_PERM 2 189 #define TCPOLEN_TIMESTAMP 10 190 #define TCPOLEN_MD5SIG 18 191 #define TCPOLEN_EXP_FASTOPEN_BASE 4 192 193 /* But this is what stacks really send out. */ 194 #define TCPOLEN_TSTAMP_ALIGNED 12 195 #define TCPOLEN_WSCALE_ALIGNED 4 196 #define TCPOLEN_SACKPERM_ALIGNED 4 197 #define TCPOLEN_SACK_BASE 2 198 #define TCPOLEN_SACK_BASE_ALIGNED 4 199 #define TCPOLEN_SACK_PERBLOCK 8 200 #define TCPOLEN_MD5SIG_ALIGNED 20 201 #define TCPOLEN_MSS_ALIGNED 4 202 203 /* Flags in tp->nonagle */ 204 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 205 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 206 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 207 208 /* TCP thin-stream limits */ 209 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 210 211 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */ 212 #define TCP_INIT_CWND 10 213 214 /* Bit Flags for sysctl_tcp_fastopen */ 215 #define TFO_CLIENT_ENABLE 1 216 #define TFO_SERVER_ENABLE 2 217 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 218 219 /* Accept SYN data w/o any cookie option */ 220 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 221 222 /* Force enable TFO on all listeners, i.e., not requiring the 223 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen. 224 */ 225 #define TFO_SERVER_WO_SOCKOPT1 0x400 226 #define TFO_SERVER_WO_SOCKOPT2 0x800 227 228 extern struct inet_timewait_death_row tcp_death_row; 229 230 /* sysctl variables for tcp */ 231 extern int sysctl_tcp_timestamps; 232 extern int sysctl_tcp_window_scaling; 233 extern int sysctl_tcp_sack; 234 extern int sysctl_tcp_fin_timeout; 235 extern int sysctl_tcp_keepalive_time; 236 extern int sysctl_tcp_keepalive_probes; 237 extern int sysctl_tcp_keepalive_intvl; 238 extern int sysctl_tcp_syn_retries; 239 extern int sysctl_tcp_synack_retries; 240 extern int sysctl_tcp_retries1; 241 extern int sysctl_tcp_retries2; 242 extern int sysctl_tcp_orphan_retries; 243 extern int sysctl_tcp_syncookies; 244 extern int sysctl_tcp_fastopen; 245 extern int sysctl_tcp_retrans_collapse; 246 extern int sysctl_tcp_stdurg; 247 extern int sysctl_tcp_rfc1337; 248 extern int sysctl_tcp_abort_on_overflow; 249 extern int sysctl_tcp_max_orphans; 250 extern int sysctl_tcp_fack; 251 extern int sysctl_tcp_reordering; 252 extern int sysctl_tcp_max_reordering; 253 extern int sysctl_tcp_dsack; 254 extern long sysctl_tcp_mem[3]; 255 extern int sysctl_tcp_wmem[3]; 256 extern int sysctl_tcp_rmem[3]; 257 extern int sysctl_tcp_app_win; 258 extern int sysctl_tcp_adv_win_scale; 259 extern int sysctl_tcp_tw_reuse; 260 extern int sysctl_tcp_frto; 261 extern int sysctl_tcp_low_latency; 262 extern int sysctl_tcp_nometrics_save; 263 extern int sysctl_tcp_moderate_rcvbuf; 264 extern int sysctl_tcp_tso_win_divisor; 265 extern int sysctl_tcp_mtu_probing; 266 extern int sysctl_tcp_base_mss; 267 extern int sysctl_tcp_workaround_signed_windows; 268 extern int sysctl_tcp_slow_start_after_idle; 269 extern int sysctl_tcp_thin_linear_timeouts; 270 extern int sysctl_tcp_thin_dupack; 271 extern int sysctl_tcp_early_retrans; 272 extern int sysctl_tcp_limit_output_bytes; 273 extern int sysctl_tcp_challenge_ack_limit; 274 extern unsigned int sysctl_tcp_notsent_lowat; 275 extern int sysctl_tcp_min_tso_segs; 276 extern int sysctl_tcp_autocorking; 277 278 extern atomic_long_t tcp_memory_allocated; 279 extern struct percpu_counter tcp_sockets_allocated; 280 extern int tcp_memory_pressure; 281 282 /* 283 * The next routines deal with comparing 32 bit unsigned ints 284 * and worry about wraparound (automatic with unsigned arithmetic). 285 */ 286 287 static inline bool before(__u32 seq1, __u32 seq2) 288 { 289 return (__s32)(seq1-seq2) < 0; 290 } 291 #define after(seq2, seq1) before(seq1, seq2) 292 293 /* is s2<=s1<=s3 ? */ 294 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 295 { 296 return seq3 - seq2 >= seq1 - seq2; 297 } 298 299 static inline bool tcp_out_of_memory(struct sock *sk) 300 { 301 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 302 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 303 return true; 304 return false; 305 } 306 307 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 308 { 309 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 310 int orphans = percpu_counter_read_positive(ocp); 311 312 if (orphans << shift > sysctl_tcp_max_orphans) { 313 orphans = percpu_counter_sum_positive(ocp); 314 if (orphans << shift > sysctl_tcp_max_orphans) 315 return true; 316 } 317 return false; 318 } 319 320 bool tcp_check_oom(struct sock *sk, int shift); 321 322 /* syncookies: remember time of last synqueue overflow */ 323 static inline void tcp_synq_overflow(struct sock *sk) 324 { 325 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies; 326 } 327 328 /* syncookies: no recent synqueue overflow on this listening socket? */ 329 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 330 { 331 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 332 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK); 333 } 334 335 extern struct proto tcp_prot; 336 337 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 338 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 339 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 340 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 341 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 342 343 void tcp_tasklet_init(void); 344 345 void tcp_v4_err(struct sk_buff *skb, u32); 346 347 void tcp_shutdown(struct sock *sk, int how); 348 349 void tcp_v4_early_demux(struct sk_buff *skb); 350 int tcp_v4_rcv(struct sk_buff *skb); 351 352 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 353 int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 354 size_t size); 355 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 356 int flags); 357 void tcp_release_cb(struct sock *sk); 358 void tcp_wfree(struct sk_buff *skb); 359 void tcp_write_timer_handler(struct sock *sk); 360 void tcp_delack_timer_handler(struct sock *sk); 361 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 362 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 363 const struct tcphdr *th, unsigned int len); 364 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 365 const struct tcphdr *th, unsigned int len); 366 void tcp_rcv_space_adjust(struct sock *sk); 367 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 368 void tcp_twsk_destructor(struct sock *sk); 369 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 370 struct pipe_inode_info *pipe, size_t len, 371 unsigned int flags); 372 373 static inline void tcp_dec_quickack_mode(struct sock *sk, 374 const unsigned int pkts) 375 { 376 struct inet_connection_sock *icsk = inet_csk(sk); 377 378 if (icsk->icsk_ack.quick) { 379 if (pkts >= icsk->icsk_ack.quick) { 380 icsk->icsk_ack.quick = 0; 381 /* Leaving quickack mode we deflate ATO. */ 382 icsk->icsk_ack.ato = TCP_ATO_MIN; 383 } else 384 icsk->icsk_ack.quick -= pkts; 385 } 386 } 387 388 #define TCP_ECN_OK 1 389 #define TCP_ECN_QUEUE_CWR 2 390 #define TCP_ECN_DEMAND_CWR 4 391 #define TCP_ECN_SEEN 8 392 393 enum tcp_tw_status { 394 TCP_TW_SUCCESS = 0, 395 TCP_TW_RST = 1, 396 TCP_TW_ACK = 2, 397 TCP_TW_SYN = 3 398 }; 399 400 401 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 402 struct sk_buff *skb, 403 const struct tcphdr *th); 404 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 405 struct request_sock *req, struct request_sock **prev, 406 bool fastopen); 407 int tcp_child_process(struct sock *parent, struct sock *child, 408 struct sk_buff *skb); 409 void tcp_enter_loss(struct sock *sk); 410 void tcp_clear_retrans(struct tcp_sock *tp); 411 void tcp_update_metrics(struct sock *sk); 412 void tcp_init_metrics(struct sock *sk); 413 void tcp_metrics_init(void); 414 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, 415 bool paws_check, bool timestamps); 416 bool tcp_remember_stamp(struct sock *sk); 417 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); 418 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); 419 void tcp_disable_fack(struct tcp_sock *tp); 420 void tcp_close(struct sock *sk, long timeout); 421 void tcp_init_sock(struct sock *sk); 422 unsigned int tcp_poll(struct file *file, struct socket *sock, 423 struct poll_table_struct *wait); 424 int tcp_getsockopt(struct sock *sk, int level, int optname, 425 char __user *optval, int __user *optlen); 426 int tcp_setsockopt(struct sock *sk, int level, int optname, 427 char __user *optval, unsigned int optlen); 428 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 429 char __user *optval, int __user *optlen); 430 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 431 char __user *optval, unsigned int optlen); 432 void tcp_set_keepalive(struct sock *sk, int val); 433 void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req); 434 int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 435 size_t len, int nonblock, int flags, int *addr_len); 436 void tcp_parse_options(const struct sk_buff *skb, 437 struct tcp_options_received *opt_rx, 438 int estab, struct tcp_fastopen_cookie *foc); 439 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 440 441 /* 442 * TCP v4 functions exported for the inet6 API 443 */ 444 445 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 446 void tcp_v4_mtu_reduced(struct sock *sk); 447 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 448 struct sock *tcp_create_openreq_child(struct sock *sk, 449 struct request_sock *req, 450 struct sk_buff *skb); 451 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 452 struct request_sock *req, 453 struct dst_entry *dst); 454 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 455 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 456 int tcp_connect(struct sock *sk); 457 struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst, 458 struct request_sock *req, 459 struct tcp_fastopen_cookie *foc); 460 int tcp_disconnect(struct sock *sk, int flags); 461 462 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 463 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 464 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 465 466 /* From syncookies.c */ 467 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 468 u32 cookie); 469 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 470 #ifdef CONFIG_SYN_COOKIES 471 472 /* Syncookies use a monotonic timer which increments every 60 seconds. 473 * This counter is used both as a hash input and partially encoded into 474 * the cookie value. A cookie is only validated further if the delta 475 * between the current counter value and the encoded one is less than this, 476 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 477 * the counter advances immediately after a cookie is generated). 478 */ 479 #define MAX_SYNCOOKIE_AGE 2 480 481 static inline u32 tcp_cookie_time(void) 482 { 483 u64 val = get_jiffies_64(); 484 485 do_div(val, 60 * HZ); 486 return val; 487 } 488 489 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 490 u16 *mssp); 491 __u32 cookie_v4_init_sequence(struct sock *sk, const struct sk_buff *skb, 492 __u16 *mss); 493 __u32 cookie_init_timestamp(struct request_sock *req); 494 bool cookie_timestamp_decode(struct tcp_options_received *opt); 495 bool cookie_ecn_ok(const struct tcp_options_received *opt, 496 const struct net *net, const struct dst_entry *dst); 497 498 /* From net/ipv6/syncookies.c */ 499 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 500 u32 cookie); 501 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 502 503 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 504 const struct tcphdr *th, u16 *mssp); 505 __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb, 506 __u16 *mss); 507 #endif 508 /* tcp_output.c */ 509 510 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 511 int nonagle); 512 bool tcp_may_send_now(struct sock *sk); 513 int __tcp_retransmit_skb(struct sock *, struct sk_buff *); 514 int tcp_retransmit_skb(struct sock *, struct sk_buff *); 515 void tcp_retransmit_timer(struct sock *sk); 516 void tcp_xmit_retransmit_queue(struct sock *); 517 void tcp_simple_retransmit(struct sock *); 518 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 519 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t); 520 521 void tcp_send_probe0(struct sock *); 522 void tcp_send_partial(struct sock *); 523 int tcp_write_wakeup(struct sock *); 524 void tcp_send_fin(struct sock *sk); 525 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 526 int tcp_send_synack(struct sock *); 527 bool tcp_syn_flood_action(struct sock *sk, const struct sk_buff *skb, 528 const char *proto); 529 void tcp_push_one(struct sock *, unsigned int mss_now); 530 void tcp_send_ack(struct sock *sk); 531 void tcp_send_delayed_ack(struct sock *sk); 532 void tcp_send_loss_probe(struct sock *sk); 533 bool tcp_schedule_loss_probe(struct sock *sk); 534 535 /* tcp_input.c */ 536 void tcp_resume_early_retransmit(struct sock *sk); 537 void tcp_rearm_rto(struct sock *sk); 538 void tcp_reset(struct sock *sk); 539 540 /* tcp_timer.c */ 541 void tcp_init_xmit_timers(struct sock *); 542 static inline void tcp_clear_xmit_timers(struct sock *sk) 543 { 544 inet_csk_clear_xmit_timers(sk); 545 } 546 547 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 548 unsigned int tcp_current_mss(struct sock *sk); 549 550 /* Bound MSS / TSO packet size with the half of the window */ 551 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 552 { 553 int cutoff; 554 555 /* When peer uses tiny windows, there is no use in packetizing 556 * to sub-MSS pieces for the sake of SWS or making sure there 557 * are enough packets in the pipe for fast recovery. 558 * 559 * On the other hand, for extremely large MSS devices, handling 560 * smaller than MSS windows in this way does make sense. 561 */ 562 if (tp->max_window >= 512) 563 cutoff = (tp->max_window >> 1); 564 else 565 cutoff = tp->max_window; 566 567 if (cutoff && pktsize > cutoff) 568 return max_t(int, cutoff, 68U - tp->tcp_header_len); 569 else 570 return pktsize; 571 } 572 573 /* tcp.c */ 574 void tcp_get_info(const struct sock *, struct tcp_info *); 575 576 /* Read 'sendfile()'-style from a TCP socket */ 577 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 578 unsigned int, size_t); 579 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 580 sk_read_actor_t recv_actor); 581 582 void tcp_initialize_rcv_mss(struct sock *sk); 583 584 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 585 int tcp_mss_to_mtu(struct sock *sk, int mss); 586 void tcp_mtup_init(struct sock *sk); 587 void tcp_init_buffer_space(struct sock *sk); 588 589 static inline void tcp_bound_rto(const struct sock *sk) 590 { 591 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 592 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 593 } 594 595 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 596 { 597 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 598 } 599 600 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 601 { 602 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 603 ntohl(TCP_FLAG_ACK) | 604 snd_wnd); 605 } 606 607 static inline void tcp_fast_path_on(struct tcp_sock *tp) 608 { 609 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 610 } 611 612 static inline void tcp_fast_path_check(struct sock *sk) 613 { 614 struct tcp_sock *tp = tcp_sk(sk); 615 616 if (skb_queue_empty(&tp->out_of_order_queue) && 617 tp->rcv_wnd && 618 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 619 !tp->urg_data) 620 tcp_fast_path_on(tp); 621 } 622 623 /* Compute the actual rto_min value */ 624 static inline u32 tcp_rto_min(struct sock *sk) 625 { 626 const struct dst_entry *dst = __sk_dst_get(sk); 627 u32 rto_min = TCP_RTO_MIN; 628 629 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 630 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 631 return rto_min; 632 } 633 634 static inline u32 tcp_rto_min_us(struct sock *sk) 635 { 636 return jiffies_to_usecs(tcp_rto_min(sk)); 637 } 638 639 /* Compute the actual receive window we are currently advertising. 640 * Rcv_nxt can be after the window if our peer push more data 641 * than the offered window. 642 */ 643 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 644 { 645 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 646 647 if (win < 0) 648 win = 0; 649 return (u32) win; 650 } 651 652 /* Choose a new window, without checks for shrinking, and without 653 * scaling applied to the result. The caller does these things 654 * if necessary. This is a "raw" window selection. 655 */ 656 u32 __tcp_select_window(struct sock *sk); 657 658 void tcp_send_window_probe(struct sock *sk); 659 660 /* TCP timestamps are only 32-bits, this causes a slight 661 * complication on 64-bit systems since we store a snapshot 662 * of jiffies in the buffer control blocks below. We decided 663 * to use only the low 32-bits of jiffies and hide the ugly 664 * casts with the following macro. 665 */ 666 #define tcp_time_stamp ((__u32)(jiffies)) 667 668 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 669 { 670 return skb->skb_mstamp.stamp_jiffies; 671 } 672 673 674 #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 __u32 seq; /* Starting sequence number */ 693 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 694 union { 695 /* Note : tcp_tw_isn is used in input path only 696 * (isn chosen by tcp_timewait_state_process()) 697 * 698 * tcp_gso_segs is used in write queue only, 699 * cf tcp_skb_pcount() 700 */ 701 __u32 tcp_tw_isn; 702 __u32 tcp_gso_segs; 703 }; 704 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 705 706 __u8 sacked; /* State flags for SACK/FACK. */ 707 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 708 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 709 #define TCPCB_LOST 0x04 /* SKB is lost */ 710 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 711 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 712 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 713 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 714 TCPCB_REPAIRED) 715 716 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 717 /* 1 byte hole */ 718 __u32 ack_seq; /* Sequence number ACK'd */ 719 union { 720 struct inet_skb_parm h4; 721 #if IS_ENABLED(CONFIG_IPV6) 722 struct inet6_skb_parm h6; 723 #endif 724 } header; /* For incoming frames */ 725 }; 726 727 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 728 729 730 #if IS_ENABLED(CONFIG_IPV6) 731 /* This is the variant of inet6_iif() that must be used by TCP, 732 * as TCP moves IP6CB into a different location in skb->cb[] 733 */ 734 static inline int tcp_v6_iif(const struct sk_buff *skb) 735 { 736 return TCP_SKB_CB(skb)->header.h6.iif; 737 } 738 #endif 739 740 /* Due to TSO, an SKB can be composed of multiple actual 741 * packets. To keep these tracked properly, we use this. 742 */ 743 static inline int tcp_skb_pcount(const struct sk_buff *skb) 744 { 745 return TCP_SKB_CB(skb)->tcp_gso_segs; 746 } 747 748 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 749 { 750 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 751 } 752 753 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 754 { 755 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 756 } 757 758 /* This is valid iff tcp_skb_pcount() > 1. */ 759 static inline int tcp_skb_mss(const struct sk_buff *skb) 760 { 761 return skb_shinfo(skb)->gso_size; 762 } 763 764 /* Events passed to congestion control interface */ 765 enum tcp_ca_event { 766 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 767 CA_EVENT_CWND_RESTART, /* congestion window restart */ 768 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 769 CA_EVENT_LOSS, /* loss timeout */ 770 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 771 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 772 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 773 CA_EVENT_NON_DELAYED_ACK, 774 }; 775 776 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 777 enum tcp_ca_ack_event_flags { 778 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 779 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 780 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 781 }; 782 783 /* 784 * Interface for adding new TCP congestion control handlers 785 */ 786 #define TCP_CA_NAME_MAX 16 787 #define TCP_CA_MAX 128 788 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 789 790 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 791 #define TCP_CONG_NON_RESTRICTED 0x1 792 /* Requires ECN/ECT set on all packets */ 793 #define TCP_CONG_NEEDS_ECN 0x2 794 795 struct tcp_congestion_ops { 796 struct list_head list; 797 unsigned long flags; 798 799 /* initialize private data (optional) */ 800 void (*init)(struct sock *sk); 801 /* cleanup private data (optional) */ 802 void (*release)(struct sock *sk); 803 804 /* return slow start threshold (required) */ 805 u32 (*ssthresh)(struct sock *sk); 806 /* do new cwnd calculation (required) */ 807 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 808 /* call before changing ca_state (optional) */ 809 void (*set_state)(struct sock *sk, u8 new_state); 810 /* call when cwnd event occurs (optional) */ 811 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 812 /* call when ack arrives (optional) */ 813 void (*in_ack_event)(struct sock *sk, u32 flags); 814 /* new value of cwnd after loss (optional) */ 815 u32 (*undo_cwnd)(struct sock *sk); 816 /* hook for packet ack accounting (optional) */ 817 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 818 /* get info for inet_diag (optional) */ 819 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 820 821 char name[TCP_CA_NAME_MAX]; 822 struct module *owner; 823 }; 824 825 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 826 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 827 828 void tcp_assign_congestion_control(struct sock *sk); 829 void tcp_init_congestion_control(struct sock *sk); 830 void tcp_cleanup_congestion_control(struct sock *sk); 831 int tcp_set_default_congestion_control(const char *name); 832 void tcp_get_default_congestion_control(char *name); 833 void tcp_get_available_congestion_control(char *buf, size_t len); 834 void tcp_get_allowed_congestion_control(char *buf, size_t len); 835 int tcp_set_allowed_congestion_control(char *allowed); 836 int tcp_set_congestion_control(struct sock *sk, const char *name); 837 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 838 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 839 840 u32 tcp_reno_ssthresh(struct sock *sk); 841 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 842 extern struct tcp_congestion_ops tcp_reno; 843 844 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 845 { 846 const struct inet_connection_sock *icsk = inet_csk(sk); 847 848 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 849 } 850 851 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 852 { 853 struct inet_connection_sock *icsk = inet_csk(sk); 854 855 if (icsk->icsk_ca_ops->set_state) 856 icsk->icsk_ca_ops->set_state(sk, ca_state); 857 icsk->icsk_ca_state = ca_state; 858 } 859 860 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 861 { 862 const struct inet_connection_sock *icsk = inet_csk(sk); 863 864 if (icsk->icsk_ca_ops->cwnd_event) 865 icsk->icsk_ca_ops->cwnd_event(sk, event); 866 } 867 868 /* These functions determine how the current flow behaves in respect of SACK 869 * handling. SACK is negotiated with the peer, and therefore it can vary 870 * between different flows. 871 * 872 * tcp_is_sack - SACK enabled 873 * tcp_is_reno - No SACK 874 * tcp_is_fack - FACK enabled, implies SACK enabled 875 */ 876 static inline int tcp_is_sack(const struct tcp_sock *tp) 877 { 878 return tp->rx_opt.sack_ok; 879 } 880 881 static inline bool tcp_is_reno(const struct tcp_sock *tp) 882 { 883 return !tcp_is_sack(tp); 884 } 885 886 static inline bool tcp_is_fack(const struct tcp_sock *tp) 887 { 888 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 889 } 890 891 static inline void tcp_enable_fack(struct tcp_sock *tp) 892 { 893 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 894 } 895 896 /* TCP early-retransmit (ER) is similar to but more conservative than 897 * the thin-dupack feature. Enable ER only if thin-dupack is disabled. 898 */ 899 static inline void tcp_enable_early_retrans(struct tcp_sock *tp) 900 { 901 tp->do_early_retrans = sysctl_tcp_early_retrans && 902 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack && 903 sysctl_tcp_reordering == 3; 904 } 905 906 static inline void tcp_disable_early_retrans(struct tcp_sock *tp) 907 { 908 tp->do_early_retrans = 0; 909 } 910 911 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 912 { 913 return tp->sacked_out + tp->lost_out; 914 } 915 916 /* This determines how many packets are "in the network" to the best 917 * of our knowledge. In many cases it is conservative, but where 918 * detailed information is available from the receiver (via SACK 919 * blocks etc.) we can make more aggressive calculations. 920 * 921 * Use this for decisions involving congestion control, use just 922 * tp->packets_out to determine if the send queue is empty or not. 923 * 924 * Read this equation as: 925 * 926 * "Packets sent once on transmission queue" MINUS 927 * "Packets left network, but not honestly ACKed yet" PLUS 928 * "Packets fast retransmitted" 929 */ 930 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 931 { 932 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 933 } 934 935 #define TCP_INFINITE_SSTHRESH 0x7fffffff 936 937 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 938 { 939 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 940 } 941 942 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 943 { 944 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 945 (1 << inet_csk(sk)->icsk_ca_state); 946 } 947 948 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 949 * The exception is cwnd reduction phase, when cwnd is decreasing towards 950 * ssthresh. 951 */ 952 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 953 { 954 const struct tcp_sock *tp = tcp_sk(sk); 955 956 if (tcp_in_cwnd_reduction(sk)) 957 return tp->snd_ssthresh; 958 else 959 return max(tp->snd_ssthresh, 960 ((tp->snd_cwnd >> 1) + 961 (tp->snd_cwnd >> 2))); 962 } 963 964 /* Use define here intentionally to get WARN_ON location shown at the caller */ 965 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 966 967 void tcp_enter_cwr(struct sock *sk); 968 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 969 970 /* The maximum number of MSS of available cwnd for which TSO defers 971 * sending if not using sysctl_tcp_tso_win_divisor. 972 */ 973 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 974 { 975 return 3; 976 } 977 978 /* Slow start with delack produces 3 packets of burst, so that 979 * it is safe "de facto". This will be the default - same as 980 * the default reordering threshold - but if reordering increases, 981 * we must be able to allow cwnd to burst at least this much in order 982 * to not pull it back when holes are filled. 983 */ 984 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 985 { 986 return tp->reordering; 987 } 988 989 /* Returns end sequence number of the receiver's advertised window */ 990 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 991 { 992 return tp->snd_una + tp->snd_wnd; 993 } 994 995 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 996 * flexible approach. The RFC suggests cwnd should not be raised unless 997 * it was fully used previously. And that's exactly what we do in 998 * congestion avoidance mode. But in slow start we allow cwnd to grow 999 * as long as the application has used half the cwnd. 1000 * Example : 1001 * cwnd is 10 (IW10), but application sends 9 frames. 1002 * We allow cwnd to reach 18 when all frames are ACKed. 1003 * This check is safe because it's as aggressive as slow start which already 1004 * risks 100% overshoot. The advantage is that we discourage application to 1005 * either send more filler packets or data to artificially blow up the cwnd 1006 * usage, and allow application-limited process to probe bw more aggressively. 1007 */ 1008 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1009 { 1010 const struct tcp_sock *tp = tcp_sk(sk); 1011 1012 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1013 if (tp->snd_cwnd <= tp->snd_ssthresh) 1014 return tp->snd_cwnd < 2 * tp->max_packets_out; 1015 1016 return tp->is_cwnd_limited; 1017 } 1018 1019 static inline void tcp_check_probe_timer(struct sock *sk) 1020 { 1021 const struct tcp_sock *tp = tcp_sk(sk); 1022 const struct inet_connection_sock *icsk = inet_csk(sk); 1023 1024 if (!tp->packets_out && !icsk->icsk_pending) 1025 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1026 icsk->icsk_rto, TCP_RTO_MAX); 1027 } 1028 1029 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1030 { 1031 tp->snd_wl1 = seq; 1032 } 1033 1034 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1035 { 1036 tp->snd_wl1 = seq; 1037 } 1038 1039 /* 1040 * Calculate(/check) TCP checksum 1041 */ 1042 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1043 __be32 daddr, __wsum base) 1044 { 1045 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1046 } 1047 1048 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1049 { 1050 return __skb_checksum_complete(skb); 1051 } 1052 1053 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1054 { 1055 return !skb_csum_unnecessary(skb) && 1056 __tcp_checksum_complete(skb); 1057 } 1058 1059 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1060 1061 static inline void tcp_prequeue_init(struct tcp_sock *tp) 1062 { 1063 tp->ucopy.task = NULL; 1064 tp->ucopy.len = 0; 1065 tp->ucopy.memory = 0; 1066 skb_queue_head_init(&tp->ucopy.prequeue); 1067 } 1068 1069 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1070 1071 #undef STATE_TRACE 1072 1073 #ifdef STATE_TRACE 1074 static const char *statename[]={ 1075 "Unused","Established","Syn Sent","Syn Recv", 1076 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1077 "Close Wait","Last ACK","Listen","Closing" 1078 }; 1079 #endif 1080 void tcp_set_state(struct sock *sk, int state); 1081 1082 void tcp_done(struct sock *sk); 1083 1084 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1085 { 1086 rx_opt->dsack = 0; 1087 rx_opt->num_sacks = 0; 1088 } 1089 1090 u32 tcp_default_init_rwnd(u32 mss); 1091 1092 /* Determine a window scaling and initial window to offer. */ 1093 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1094 __u32 *window_clamp, int wscale_ok, 1095 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1096 1097 static inline int tcp_win_from_space(int space) 1098 { 1099 return sysctl_tcp_adv_win_scale<=0 ? 1100 (space>>(-sysctl_tcp_adv_win_scale)) : 1101 space - (space>>sysctl_tcp_adv_win_scale); 1102 } 1103 1104 /* Note: caller must be prepared to deal with negative returns */ 1105 static inline int tcp_space(const struct sock *sk) 1106 { 1107 return tcp_win_from_space(sk->sk_rcvbuf - 1108 atomic_read(&sk->sk_rmem_alloc)); 1109 } 1110 1111 static inline int tcp_full_space(const struct sock *sk) 1112 { 1113 return tcp_win_from_space(sk->sk_rcvbuf); 1114 } 1115 1116 static inline void tcp_openreq_init(struct request_sock *req, 1117 struct tcp_options_received *rx_opt, 1118 struct sk_buff *skb, struct sock *sk) 1119 { 1120 struct inet_request_sock *ireq = inet_rsk(req); 1121 1122 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 1123 req->cookie_ts = 0; 1124 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 1125 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1; 1126 tcp_rsk(req)->snt_synack = tcp_time_stamp; 1127 req->mss = rx_opt->mss_clamp; 1128 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 1129 ireq->tstamp_ok = rx_opt->tstamp_ok; 1130 ireq->sack_ok = rx_opt->sack_ok; 1131 ireq->snd_wscale = rx_opt->snd_wscale; 1132 ireq->wscale_ok = rx_opt->wscale_ok; 1133 ireq->acked = 0; 1134 ireq->ecn_ok = 0; 1135 ireq->ir_rmt_port = tcp_hdr(skb)->source; 1136 ireq->ir_num = ntohs(tcp_hdr(skb)->dest); 1137 ireq->ir_mark = inet_request_mark(sk, skb); 1138 } 1139 1140 extern void tcp_openreq_init_rwin(struct request_sock *req, 1141 struct sock *sk, struct dst_entry *dst); 1142 1143 void tcp_enter_memory_pressure(struct sock *sk); 1144 1145 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1146 { 1147 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1148 } 1149 1150 static inline int keepalive_time_when(const struct tcp_sock *tp) 1151 { 1152 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1153 } 1154 1155 static inline int keepalive_probes(const struct tcp_sock *tp) 1156 { 1157 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1158 } 1159 1160 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1161 { 1162 const struct inet_connection_sock *icsk = &tp->inet_conn; 1163 1164 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1165 tcp_time_stamp - tp->rcv_tstamp); 1166 } 1167 1168 static inline int tcp_fin_time(const struct sock *sk) 1169 { 1170 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1171 const int rto = inet_csk(sk)->icsk_rto; 1172 1173 if (fin_timeout < (rto << 2) - (rto >> 1)) 1174 fin_timeout = (rto << 2) - (rto >> 1); 1175 1176 return fin_timeout; 1177 } 1178 1179 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1180 int paws_win) 1181 { 1182 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1183 return true; 1184 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1185 return true; 1186 /* 1187 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1188 * then following tcp messages have valid values. Ignore 0 value, 1189 * or else 'negative' tsval might forbid us to accept their packets. 1190 */ 1191 if (!rx_opt->ts_recent) 1192 return true; 1193 return false; 1194 } 1195 1196 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1197 int rst) 1198 { 1199 if (tcp_paws_check(rx_opt, 0)) 1200 return false; 1201 1202 /* RST segments are not recommended to carry timestamp, 1203 and, if they do, it is recommended to ignore PAWS because 1204 "their cleanup function should take precedence over timestamps." 1205 Certainly, it is mistake. It is necessary to understand the reasons 1206 of this constraint to relax it: if peer reboots, clock may go 1207 out-of-sync and half-open connections will not be reset. 1208 Actually, the problem would be not existing if all 1209 the implementations followed draft about maintaining clock 1210 via reboots. Linux-2.2 DOES NOT! 1211 1212 However, we can relax time bounds for RST segments to MSL. 1213 */ 1214 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1215 return false; 1216 return true; 1217 } 1218 1219 static inline void tcp_mib_init(struct net *net) 1220 { 1221 /* See RFC 2012 */ 1222 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1223 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1224 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1225 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1226 } 1227 1228 /* from STCP */ 1229 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1230 { 1231 tp->lost_skb_hint = NULL; 1232 } 1233 1234 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1235 { 1236 tcp_clear_retrans_hints_partial(tp); 1237 tp->retransmit_skb_hint = NULL; 1238 } 1239 1240 /* MD5 Signature */ 1241 struct crypto_hash; 1242 1243 union tcp_md5_addr { 1244 struct in_addr a4; 1245 #if IS_ENABLED(CONFIG_IPV6) 1246 struct in6_addr a6; 1247 #endif 1248 }; 1249 1250 /* - key database */ 1251 struct tcp_md5sig_key { 1252 struct hlist_node node; 1253 u8 keylen; 1254 u8 family; /* AF_INET or AF_INET6 */ 1255 union tcp_md5_addr addr; 1256 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1257 struct rcu_head rcu; 1258 }; 1259 1260 /* - sock block */ 1261 struct tcp_md5sig_info { 1262 struct hlist_head head; 1263 struct rcu_head rcu; 1264 }; 1265 1266 /* - pseudo header */ 1267 struct tcp4_pseudohdr { 1268 __be32 saddr; 1269 __be32 daddr; 1270 __u8 pad; 1271 __u8 protocol; 1272 __be16 len; 1273 }; 1274 1275 struct tcp6_pseudohdr { 1276 struct in6_addr saddr; 1277 struct in6_addr daddr; 1278 __be32 len; 1279 __be32 protocol; /* including padding */ 1280 }; 1281 1282 union tcp_md5sum_block { 1283 struct tcp4_pseudohdr ip4; 1284 #if IS_ENABLED(CONFIG_IPV6) 1285 struct tcp6_pseudohdr ip6; 1286 #endif 1287 }; 1288 1289 /* - pool: digest algorithm, hash description and scratch buffer */ 1290 struct tcp_md5sig_pool { 1291 struct hash_desc md5_desc; 1292 union tcp_md5sum_block md5_blk; 1293 }; 1294 1295 /* - functions */ 1296 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1297 const struct sock *sk, const struct request_sock *req, 1298 const struct sk_buff *skb); 1299 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1300 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); 1301 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1302 int family); 1303 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1304 struct sock *addr_sk); 1305 1306 #ifdef CONFIG_TCP_MD5SIG 1307 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1308 const union tcp_md5_addr *addr, 1309 int family); 1310 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1311 #else 1312 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1313 const union tcp_md5_addr *addr, 1314 int family) 1315 { 1316 return NULL; 1317 } 1318 #define tcp_twsk_md5_key(twsk) NULL 1319 #endif 1320 1321 bool tcp_alloc_md5sig_pool(void); 1322 1323 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1324 static inline void tcp_put_md5sig_pool(void) 1325 { 1326 local_bh_enable(); 1327 } 1328 1329 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1330 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1331 unsigned int header_len); 1332 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1333 const struct tcp_md5sig_key *key); 1334 1335 /* From tcp_fastopen.c */ 1336 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1337 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1338 unsigned long *last_syn_loss); 1339 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1340 struct tcp_fastopen_cookie *cookie, bool syn_lost); 1341 struct tcp_fastopen_request { 1342 /* Fast Open cookie. Size 0 means a cookie request */ 1343 struct tcp_fastopen_cookie cookie; 1344 struct msghdr *data; /* data in MSG_FASTOPEN */ 1345 size_t size; 1346 int copied; /* queued in tcp_connect() */ 1347 }; 1348 void tcp_free_fastopen_req(struct tcp_sock *tp); 1349 1350 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1351 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1352 bool tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1353 struct request_sock *req, 1354 struct tcp_fastopen_cookie *foc, 1355 struct dst_entry *dst); 1356 void tcp_fastopen_init_key_once(bool publish); 1357 #define TCP_FASTOPEN_KEY_LENGTH 16 1358 1359 /* Fastopen key context */ 1360 struct tcp_fastopen_context { 1361 struct crypto_cipher *tfm; 1362 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1363 struct rcu_head rcu; 1364 }; 1365 1366 /* write queue abstraction */ 1367 static inline void tcp_write_queue_purge(struct sock *sk) 1368 { 1369 struct sk_buff *skb; 1370 1371 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1372 sk_wmem_free_skb(sk, skb); 1373 sk_mem_reclaim(sk); 1374 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1375 } 1376 1377 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1378 { 1379 return skb_peek(&sk->sk_write_queue); 1380 } 1381 1382 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1383 { 1384 return skb_peek_tail(&sk->sk_write_queue); 1385 } 1386 1387 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1388 const struct sk_buff *skb) 1389 { 1390 return skb_queue_next(&sk->sk_write_queue, skb); 1391 } 1392 1393 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1394 const struct sk_buff *skb) 1395 { 1396 return skb_queue_prev(&sk->sk_write_queue, skb); 1397 } 1398 1399 #define tcp_for_write_queue(skb, sk) \ 1400 skb_queue_walk(&(sk)->sk_write_queue, skb) 1401 1402 #define tcp_for_write_queue_from(skb, sk) \ 1403 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1404 1405 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1406 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1407 1408 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1409 { 1410 return sk->sk_send_head; 1411 } 1412 1413 static inline bool tcp_skb_is_last(const struct sock *sk, 1414 const struct sk_buff *skb) 1415 { 1416 return skb_queue_is_last(&sk->sk_write_queue, skb); 1417 } 1418 1419 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1420 { 1421 if (tcp_skb_is_last(sk, skb)) 1422 sk->sk_send_head = NULL; 1423 else 1424 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1425 } 1426 1427 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1428 { 1429 if (sk->sk_send_head == skb_unlinked) 1430 sk->sk_send_head = NULL; 1431 } 1432 1433 static inline void tcp_init_send_head(struct sock *sk) 1434 { 1435 sk->sk_send_head = NULL; 1436 } 1437 1438 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1439 { 1440 __skb_queue_tail(&sk->sk_write_queue, skb); 1441 } 1442 1443 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1444 { 1445 __tcp_add_write_queue_tail(sk, skb); 1446 1447 /* Queue it, remembering where we must start sending. */ 1448 if (sk->sk_send_head == NULL) { 1449 sk->sk_send_head = skb; 1450 1451 if (tcp_sk(sk)->highest_sack == NULL) 1452 tcp_sk(sk)->highest_sack = skb; 1453 } 1454 } 1455 1456 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1457 { 1458 __skb_queue_head(&sk->sk_write_queue, skb); 1459 } 1460 1461 /* Insert buff after skb on the write queue of sk. */ 1462 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1463 struct sk_buff *buff, 1464 struct sock *sk) 1465 { 1466 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1467 } 1468 1469 /* Insert new before skb on the write queue of sk. */ 1470 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1471 struct sk_buff *skb, 1472 struct sock *sk) 1473 { 1474 __skb_queue_before(&sk->sk_write_queue, skb, new); 1475 1476 if (sk->sk_send_head == skb) 1477 sk->sk_send_head = new; 1478 } 1479 1480 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1481 { 1482 __skb_unlink(skb, &sk->sk_write_queue); 1483 } 1484 1485 static inline bool tcp_write_queue_empty(struct sock *sk) 1486 { 1487 return skb_queue_empty(&sk->sk_write_queue); 1488 } 1489 1490 static inline void tcp_push_pending_frames(struct sock *sk) 1491 { 1492 if (tcp_send_head(sk)) { 1493 struct tcp_sock *tp = tcp_sk(sk); 1494 1495 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1496 } 1497 } 1498 1499 /* Start sequence of the skb just after the highest skb with SACKed 1500 * bit, valid only if sacked_out > 0 or when the caller has ensured 1501 * validity by itself. 1502 */ 1503 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1504 { 1505 if (!tp->sacked_out) 1506 return tp->snd_una; 1507 1508 if (tp->highest_sack == NULL) 1509 return tp->snd_nxt; 1510 1511 return TCP_SKB_CB(tp->highest_sack)->seq; 1512 } 1513 1514 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1515 { 1516 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1517 tcp_write_queue_next(sk, skb); 1518 } 1519 1520 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1521 { 1522 return tcp_sk(sk)->highest_sack; 1523 } 1524 1525 static inline void tcp_highest_sack_reset(struct sock *sk) 1526 { 1527 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1528 } 1529 1530 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1531 static inline void tcp_highest_sack_combine(struct sock *sk, 1532 struct sk_buff *old, 1533 struct sk_buff *new) 1534 { 1535 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1536 tcp_sk(sk)->highest_sack = new; 1537 } 1538 1539 /* Determines whether this is a thin stream (which may suffer from 1540 * increased latency). Used to trigger latency-reducing mechanisms. 1541 */ 1542 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1543 { 1544 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1545 } 1546 1547 /* /proc */ 1548 enum tcp_seq_states { 1549 TCP_SEQ_STATE_LISTENING, 1550 TCP_SEQ_STATE_OPENREQ, 1551 TCP_SEQ_STATE_ESTABLISHED, 1552 }; 1553 1554 int tcp_seq_open(struct inode *inode, struct file *file); 1555 1556 struct tcp_seq_afinfo { 1557 char *name; 1558 sa_family_t family; 1559 const struct file_operations *seq_fops; 1560 struct seq_operations seq_ops; 1561 }; 1562 1563 struct tcp_iter_state { 1564 struct seq_net_private p; 1565 sa_family_t family; 1566 enum tcp_seq_states state; 1567 struct sock *syn_wait_sk; 1568 int bucket, offset, sbucket, num; 1569 kuid_t uid; 1570 loff_t last_pos; 1571 }; 1572 1573 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1574 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1575 1576 extern struct request_sock_ops tcp_request_sock_ops; 1577 extern struct request_sock_ops tcp6_request_sock_ops; 1578 1579 void tcp_v4_destroy_sock(struct sock *sk); 1580 1581 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1582 netdev_features_t features); 1583 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1584 int tcp_gro_complete(struct sk_buff *skb); 1585 1586 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1587 1588 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1589 { 1590 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat; 1591 } 1592 1593 static inline bool tcp_stream_memory_free(const struct sock *sk) 1594 { 1595 const struct tcp_sock *tp = tcp_sk(sk); 1596 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1597 1598 return notsent_bytes < tcp_notsent_lowat(tp); 1599 } 1600 1601 #ifdef CONFIG_PROC_FS 1602 int tcp4_proc_init(void); 1603 void tcp4_proc_exit(void); 1604 #endif 1605 1606 int tcp_rtx_synack(struct sock *sk, struct request_sock *req); 1607 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1608 const struct tcp_request_sock_ops *af_ops, 1609 struct sock *sk, struct sk_buff *skb); 1610 1611 /* TCP af-specific functions */ 1612 struct tcp_sock_af_ops { 1613 #ifdef CONFIG_TCP_MD5SIG 1614 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1615 struct sock *addr_sk); 1616 int (*calc_md5_hash) (char *location, 1617 struct tcp_md5sig_key *md5, 1618 const struct sock *sk, 1619 const struct request_sock *req, 1620 const struct sk_buff *skb); 1621 int (*md5_parse) (struct sock *sk, 1622 char __user *optval, 1623 int optlen); 1624 #endif 1625 }; 1626 1627 struct tcp_request_sock_ops { 1628 u16 mss_clamp; 1629 #ifdef CONFIG_TCP_MD5SIG 1630 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1631 struct request_sock *req); 1632 int (*calc_md5_hash) (char *location, 1633 struct tcp_md5sig_key *md5, 1634 const struct sock *sk, 1635 const struct request_sock *req, 1636 const struct sk_buff *skb); 1637 #endif 1638 void (*init_req)(struct request_sock *req, struct sock *sk, 1639 struct sk_buff *skb); 1640 #ifdef CONFIG_SYN_COOKIES 1641 __u32 (*cookie_init_seq)(struct sock *sk, const struct sk_buff *skb, 1642 __u16 *mss); 1643 #endif 1644 struct dst_entry *(*route_req)(struct sock *sk, struct flowi *fl, 1645 const struct request_sock *req, 1646 bool *strict); 1647 __u32 (*init_seq)(const struct sk_buff *skb); 1648 int (*send_synack)(struct sock *sk, struct dst_entry *dst, 1649 struct flowi *fl, struct request_sock *req, 1650 u16 queue_mapping, struct tcp_fastopen_cookie *foc); 1651 void (*queue_hash_add)(struct sock *sk, struct request_sock *req, 1652 const unsigned long timeout); 1653 }; 1654 1655 #ifdef CONFIG_SYN_COOKIES 1656 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1657 struct sock *sk, struct sk_buff *skb, 1658 __u16 *mss) 1659 { 1660 return ops->cookie_init_seq(sk, skb, mss); 1661 } 1662 #else 1663 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1664 struct sock *sk, struct sk_buff *skb, 1665 __u16 *mss) 1666 { 1667 return 0; 1668 } 1669 #endif 1670 1671 int tcpv4_offload_init(void); 1672 1673 void tcp_v4_init(void); 1674 void tcp_init(void); 1675 1676 /* 1677 * Save and compile IPv4 options, return a pointer to it 1678 */ 1679 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb) 1680 { 1681 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1682 struct ip_options_rcu *dopt = NULL; 1683 1684 if (opt->optlen) { 1685 int opt_size = sizeof(*dopt) + opt->optlen; 1686 1687 dopt = kmalloc(opt_size, GFP_ATOMIC); 1688 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) { 1689 kfree(dopt); 1690 dopt = NULL; 1691 } 1692 } 1693 return dopt; 1694 } 1695 1696 #endif /* _TCP_H */ 1697