1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * Definitions for the TCP module. 7 * 8 * Version: @(#)tcp.h 1.0.5 05/23/93 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * 13 * This program is free software; you can redistribute it and/or 14 * modify it under the terms of the GNU General Public License 15 * as published by the Free Software Foundation; either version 16 * 2 of the License, or (at your option) any later version. 17 */ 18 #ifndef _TCP_H 19 #define _TCP_H 20 21 #define FASTRETRANS_DEBUG 1 22 23 #include <linux/list.h> 24 #include <linux/tcp.h> 25 #include <linux/bug.h> 26 #include <linux/slab.h> 27 #include <linux/cache.h> 28 #include <linux/percpu.h> 29 #include <linux/skbuff.h> 30 #include <linux/dmaengine.h> 31 #include <linux/crypto.h> 32 #include <linux/cryptohash.h> 33 #include <linux/kref.h> 34 35 #include <net/inet_connection_sock.h> 36 #include <net/inet_timewait_sock.h> 37 #include <net/inet_hashtables.h> 38 #include <net/checksum.h> 39 #include <net/request_sock.h> 40 #include <net/sock.h> 41 #include <net/snmp.h> 42 #include <net/ip.h> 43 #include <net/tcp_states.h> 44 #include <net/inet_ecn.h> 45 #include <net/dst.h> 46 47 #include <linux/seq_file.h> 48 #include <linux/memcontrol.h> 49 50 extern struct inet_hashinfo tcp_hashinfo; 51 52 extern struct percpu_counter tcp_orphan_count; 53 void tcp_time_wait(struct sock *sk, int state, int timeo); 54 55 #define MAX_TCP_HEADER (128 + MAX_HEADER) 56 #define MAX_TCP_OPTION_SPACE 40 57 58 /* 59 * Never offer a window over 32767 without using window scaling. Some 60 * poor stacks do signed 16bit maths! 61 */ 62 #define MAX_TCP_WINDOW 32767U 63 64 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 65 #define TCP_MIN_MSS 88U 66 67 /* The least MTU to use for probing */ 68 #define TCP_BASE_MSS 512 69 70 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 71 #define TCP_FASTRETRANS_THRESH 3 72 73 /* Maximal reordering. */ 74 #define TCP_MAX_REORDERING 127 75 76 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 77 #define TCP_MAX_QUICKACKS 16U 78 79 /* urg_data states */ 80 #define TCP_URG_VALID 0x0100 81 #define TCP_URG_NOTYET 0x0200 82 #define TCP_URG_READ 0x0400 83 84 #define TCP_RETR1 3 /* 85 * This is how many retries it does before it 86 * tries to figure out if the gateway is 87 * down. Minimal RFC value is 3; it corresponds 88 * to ~3sec-8min depending on RTO. 89 */ 90 91 #define TCP_RETR2 15 /* 92 * This should take at least 93 * 90 minutes to time out. 94 * RFC1122 says that the limit is 100 sec. 95 * 15 is ~13-30min depending on RTO. 96 */ 97 98 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 99 * when active opening a connection. 100 * RFC1122 says the minimum retry MUST 101 * be at least 180secs. Nevertheless 102 * this value is corresponding to 103 * 63secs of retransmission with the 104 * current initial RTO. 105 */ 106 107 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 108 * when passive opening a connection. 109 * This is corresponding to 31secs of 110 * retransmission with the current 111 * initial RTO. 112 */ 113 114 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 115 * state, about 60 seconds */ 116 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 117 /* BSD style FIN_WAIT2 deadlock breaker. 118 * It used to be 3min, new value is 60sec, 119 * to combine FIN-WAIT-2 timeout with 120 * TIME-WAIT timer. 121 */ 122 123 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 124 #if HZ >= 100 125 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 126 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 127 #else 128 #define TCP_DELACK_MIN 4U 129 #define TCP_ATO_MIN 4U 130 #endif 131 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 132 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 133 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 134 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 135 * used as a fallback RTO for the 136 * initial data transmission if no 137 * valid RTT sample has been acquired, 138 * most likely due to retrans in 3WHS. 139 */ 140 141 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 142 * for local resources. 143 */ 144 145 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 146 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 147 #define TCP_KEEPALIVE_INTVL (75*HZ) 148 149 #define MAX_TCP_KEEPIDLE 32767 150 #define MAX_TCP_KEEPINTVL 32767 151 #define MAX_TCP_KEEPCNT 127 152 #define MAX_TCP_SYNCNT 127 153 154 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 155 156 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 157 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 158 * after this time. It should be equal 159 * (or greater than) TCP_TIMEWAIT_LEN 160 * to provide reliability equal to one 161 * provided by timewait state. 162 */ 163 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 164 * timestamps. It must be less than 165 * minimal timewait lifetime. 166 */ 167 /* 168 * TCP option 169 */ 170 171 #define TCPOPT_NOP 1 /* Padding */ 172 #define TCPOPT_EOL 0 /* End of options */ 173 #define TCPOPT_MSS 2 /* Segment size negotiating */ 174 #define TCPOPT_WINDOW 3 /* Window scaling */ 175 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 176 #define TCPOPT_SACK 5 /* SACK Block */ 177 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 178 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 179 #define TCPOPT_EXP 254 /* Experimental */ 180 /* Magic number to be after the option value for sharing TCP 181 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 182 */ 183 #define TCPOPT_FASTOPEN_MAGIC 0xF989 184 185 /* 186 * TCP option lengths 187 */ 188 189 #define TCPOLEN_MSS 4 190 #define TCPOLEN_WINDOW 3 191 #define TCPOLEN_SACK_PERM 2 192 #define TCPOLEN_TIMESTAMP 10 193 #define TCPOLEN_MD5SIG 18 194 #define TCPOLEN_EXP_FASTOPEN_BASE 4 195 196 /* But this is what stacks really send out. */ 197 #define TCPOLEN_TSTAMP_ALIGNED 12 198 #define TCPOLEN_WSCALE_ALIGNED 4 199 #define TCPOLEN_SACKPERM_ALIGNED 4 200 #define TCPOLEN_SACK_BASE 2 201 #define TCPOLEN_SACK_BASE_ALIGNED 4 202 #define TCPOLEN_SACK_PERBLOCK 8 203 #define TCPOLEN_MD5SIG_ALIGNED 20 204 #define TCPOLEN_MSS_ALIGNED 4 205 206 /* Flags in tp->nonagle */ 207 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 208 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 209 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 210 211 /* TCP thin-stream limits */ 212 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 213 214 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */ 215 #define TCP_INIT_CWND 10 216 217 /* Bit Flags for sysctl_tcp_fastopen */ 218 #define TFO_CLIENT_ENABLE 1 219 #define TFO_SERVER_ENABLE 2 220 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 221 222 /* Process SYN data but skip cookie validation */ 223 #define TFO_SERVER_COOKIE_NOT_CHKED 0x100 224 /* Accept SYN data w/o any cookie option */ 225 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 226 227 /* Force enable TFO on all listeners, i.e., not requiring the 228 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen. 229 */ 230 #define TFO_SERVER_WO_SOCKOPT1 0x400 231 #define TFO_SERVER_WO_SOCKOPT2 0x800 232 /* Always create TFO child sockets on a TFO listener even when 233 * cookie/data not present. (For testing purpose!) 234 */ 235 #define TFO_SERVER_ALWAYS 0x1000 236 237 extern struct inet_timewait_death_row tcp_death_row; 238 239 /* sysctl variables for tcp */ 240 extern int sysctl_tcp_timestamps; 241 extern int sysctl_tcp_window_scaling; 242 extern int sysctl_tcp_sack; 243 extern int sysctl_tcp_fin_timeout; 244 extern int sysctl_tcp_keepalive_time; 245 extern int sysctl_tcp_keepalive_probes; 246 extern int sysctl_tcp_keepalive_intvl; 247 extern int sysctl_tcp_syn_retries; 248 extern int sysctl_tcp_synack_retries; 249 extern int sysctl_tcp_retries1; 250 extern int sysctl_tcp_retries2; 251 extern int sysctl_tcp_orphan_retries; 252 extern int sysctl_tcp_syncookies; 253 extern int sysctl_tcp_fastopen; 254 extern int sysctl_tcp_retrans_collapse; 255 extern int sysctl_tcp_stdurg; 256 extern int sysctl_tcp_rfc1337; 257 extern int sysctl_tcp_abort_on_overflow; 258 extern int sysctl_tcp_max_orphans; 259 extern int sysctl_tcp_fack; 260 extern int sysctl_tcp_reordering; 261 extern int sysctl_tcp_dsack; 262 extern long sysctl_tcp_mem[3]; 263 extern int sysctl_tcp_wmem[3]; 264 extern int sysctl_tcp_rmem[3]; 265 extern int sysctl_tcp_app_win; 266 extern int sysctl_tcp_adv_win_scale; 267 extern int sysctl_tcp_tw_reuse; 268 extern int sysctl_tcp_frto; 269 extern int sysctl_tcp_low_latency; 270 extern int sysctl_tcp_dma_copybreak; 271 extern int sysctl_tcp_nometrics_save; 272 extern int sysctl_tcp_moderate_rcvbuf; 273 extern int sysctl_tcp_tso_win_divisor; 274 extern int sysctl_tcp_mtu_probing; 275 extern int sysctl_tcp_base_mss; 276 extern int sysctl_tcp_workaround_signed_windows; 277 extern int sysctl_tcp_slow_start_after_idle; 278 extern int sysctl_tcp_thin_linear_timeouts; 279 extern int sysctl_tcp_thin_dupack; 280 extern int sysctl_tcp_early_retrans; 281 extern int sysctl_tcp_limit_output_bytes; 282 extern int sysctl_tcp_challenge_ack_limit; 283 extern unsigned int sysctl_tcp_notsent_lowat; 284 extern int sysctl_tcp_min_tso_segs; 285 286 extern atomic_long_t tcp_memory_allocated; 287 extern struct percpu_counter tcp_sockets_allocated; 288 extern int tcp_memory_pressure; 289 290 /* 291 * The next routines deal with comparing 32 bit unsigned ints 292 * and worry about wraparound (automatic with unsigned arithmetic). 293 */ 294 295 static inline bool before(__u32 seq1, __u32 seq2) 296 { 297 return (__s32)(seq1-seq2) < 0; 298 } 299 #define after(seq2, seq1) before(seq1, seq2) 300 301 /* is s2<=s1<=s3 ? */ 302 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 303 { 304 return seq3 - seq2 >= seq1 - seq2; 305 } 306 307 static inline bool tcp_out_of_memory(struct sock *sk) 308 { 309 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 310 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 311 return true; 312 return false; 313 } 314 315 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 316 { 317 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 318 int orphans = percpu_counter_read_positive(ocp); 319 320 if (orphans << shift > sysctl_tcp_max_orphans) { 321 orphans = percpu_counter_sum_positive(ocp); 322 if (orphans << shift > sysctl_tcp_max_orphans) 323 return true; 324 } 325 return false; 326 } 327 328 bool tcp_check_oom(struct sock *sk, int shift); 329 330 /* syncookies: remember time of last synqueue overflow */ 331 static inline void tcp_synq_overflow(struct sock *sk) 332 { 333 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies; 334 } 335 336 /* syncookies: no recent synqueue overflow on this listening socket? */ 337 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 338 { 339 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 340 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK); 341 } 342 343 extern struct proto tcp_prot; 344 345 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 346 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 347 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 348 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 349 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 350 351 void tcp_tasklet_init(void); 352 353 void tcp_v4_err(struct sk_buff *skb, u32); 354 355 void tcp_shutdown(struct sock *sk, int how); 356 357 void tcp_v4_early_demux(struct sk_buff *skb); 358 int tcp_v4_rcv(struct sk_buff *skb); 359 360 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 361 int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 362 size_t size); 363 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 364 int flags); 365 void tcp_release_cb(struct sock *sk); 366 void tcp_wfree(struct sk_buff *skb); 367 void tcp_write_timer_handler(struct sock *sk); 368 void tcp_delack_timer_handler(struct sock *sk); 369 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 370 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 371 const struct tcphdr *th, unsigned int len); 372 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 373 const struct tcphdr *th, unsigned int len); 374 void tcp_rcv_space_adjust(struct sock *sk); 375 void tcp_cleanup_rbuf(struct sock *sk, int copied); 376 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 377 void tcp_twsk_destructor(struct sock *sk); 378 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 379 struct pipe_inode_info *pipe, size_t len, 380 unsigned int flags); 381 382 static inline void tcp_dec_quickack_mode(struct sock *sk, 383 const unsigned int pkts) 384 { 385 struct inet_connection_sock *icsk = inet_csk(sk); 386 387 if (icsk->icsk_ack.quick) { 388 if (pkts >= icsk->icsk_ack.quick) { 389 icsk->icsk_ack.quick = 0; 390 /* Leaving quickack mode we deflate ATO. */ 391 icsk->icsk_ack.ato = TCP_ATO_MIN; 392 } else 393 icsk->icsk_ack.quick -= pkts; 394 } 395 } 396 397 #define TCP_ECN_OK 1 398 #define TCP_ECN_QUEUE_CWR 2 399 #define TCP_ECN_DEMAND_CWR 4 400 #define TCP_ECN_SEEN 8 401 402 enum tcp_tw_status { 403 TCP_TW_SUCCESS = 0, 404 TCP_TW_RST = 1, 405 TCP_TW_ACK = 2, 406 TCP_TW_SYN = 3 407 }; 408 409 410 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 411 struct sk_buff *skb, 412 const struct tcphdr *th); 413 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 414 struct request_sock *req, struct request_sock **prev, 415 bool fastopen); 416 int tcp_child_process(struct sock *parent, struct sock *child, 417 struct sk_buff *skb); 418 void tcp_enter_loss(struct sock *sk, int how); 419 void tcp_clear_retrans(struct tcp_sock *tp); 420 void tcp_update_metrics(struct sock *sk); 421 void tcp_init_metrics(struct sock *sk); 422 void tcp_metrics_init(void); 423 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, 424 bool paws_check); 425 bool tcp_remember_stamp(struct sock *sk); 426 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); 427 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); 428 void tcp_disable_fack(struct tcp_sock *tp); 429 void tcp_close(struct sock *sk, long timeout); 430 void tcp_init_sock(struct sock *sk); 431 unsigned int tcp_poll(struct file *file, struct socket *sock, 432 struct poll_table_struct *wait); 433 int tcp_getsockopt(struct sock *sk, int level, int optname, 434 char __user *optval, int __user *optlen); 435 int tcp_setsockopt(struct sock *sk, int level, int optname, 436 char __user *optval, unsigned int optlen); 437 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 438 char __user *optval, int __user *optlen); 439 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 440 char __user *optval, unsigned int optlen); 441 void tcp_set_keepalive(struct sock *sk, int val); 442 void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req); 443 int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 444 size_t len, int nonblock, int flags, int *addr_len); 445 void tcp_parse_options(const struct sk_buff *skb, 446 struct tcp_options_received *opt_rx, 447 int estab, struct tcp_fastopen_cookie *foc); 448 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 449 450 /* 451 * TCP v4 functions exported for the inet6 API 452 */ 453 454 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 455 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 456 struct sock *tcp_create_openreq_child(struct sock *sk, 457 struct request_sock *req, 458 struct sk_buff *skb); 459 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 460 struct request_sock *req, 461 struct dst_entry *dst); 462 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 463 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 464 int tcp_connect(struct sock *sk); 465 struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst, 466 struct request_sock *req, 467 struct tcp_fastopen_cookie *foc); 468 int tcp_disconnect(struct sock *sk, int flags); 469 470 void tcp_connect_init(struct sock *sk); 471 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 472 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 473 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 474 475 /* From syncookies.c */ 476 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 477 u32 cookie); 478 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 479 struct ip_options *opt); 480 #ifdef CONFIG_SYN_COOKIES 481 #include <linux/ktime.h> 482 483 /* Syncookies use a monotonic timer which increments every 64 seconds. 484 * This counter is used both as a hash input and partially encoded into 485 * the cookie value. A cookie is only validated further if the delta 486 * between the current counter value and the encoded one is less than this, 487 * i.e. a sent cookie is valid only at most for 128 seconds (or less if 488 * the counter advances immediately after a cookie is generated). 489 */ 490 #define MAX_SYNCOOKIE_AGE 2 491 492 static inline u32 tcp_cookie_time(void) 493 { 494 struct timespec now; 495 getnstimeofday(&now); 496 return now.tv_sec >> 6; /* 64 seconds granularity */ 497 } 498 499 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 500 u16 *mssp); 501 __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, __u16 *mss); 502 #else 503 static inline __u32 cookie_v4_init_sequence(struct sock *sk, 504 struct sk_buff *skb, 505 __u16 *mss) 506 { 507 return 0; 508 } 509 #endif 510 511 __u32 cookie_init_timestamp(struct request_sock *req); 512 bool cookie_check_timestamp(struct tcp_options_received *opt, struct net *net, 513 bool *ecn_ok); 514 515 /* From net/ipv6/syncookies.c */ 516 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 517 u32 cookie); 518 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 519 #ifdef CONFIG_SYN_COOKIES 520 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 521 const struct tcphdr *th, u16 *mssp); 522 __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb, 523 __u16 *mss); 524 #else 525 static inline __u32 cookie_v6_init_sequence(struct sock *sk, 526 struct sk_buff *skb, 527 __u16 *mss) 528 { 529 return 0; 530 } 531 #endif 532 /* tcp_output.c */ 533 534 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 535 int nonagle); 536 bool tcp_may_send_now(struct sock *sk); 537 int __tcp_retransmit_skb(struct sock *, struct sk_buff *); 538 int tcp_retransmit_skb(struct sock *, struct sk_buff *); 539 void tcp_retransmit_timer(struct sock *sk); 540 void tcp_xmit_retransmit_queue(struct sock *); 541 void tcp_simple_retransmit(struct sock *); 542 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 543 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 544 545 void tcp_send_probe0(struct sock *); 546 void tcp_send_partial(struct sock *); 547 int tcp_write_wakeup(struct sock *); 548 void tcp_send_fin(struct sock *sk); 549 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 550 int tcp_send_synack(struct sock *); 551 bool tcp_syn_flood_action(struct sock *sk, const struct sk_buff *skb, 552 const char *proto); 553 void tcp_push_one(struct sock *, unsigned int mss_now); 554 void tcp_send_ack(struct sock *sk); 555 void tcp_send_delayed_ack(struct sock *sk); 556 void tcp_send_loss_probe(struct sock *sk); 557 bool tcp_schedule_loss_probe(struct sock *sk); 558 559 /* tcp_input.c */ 560 void tcp_cwnd_application_limited(struct sock *sk); 561 void tcp_resume_early_retransmit(struct sock *sk); 562 void tcp_rearm_rto(struct sock *sk); 563 void tcp_reset(struct sock *sk); 564 565 /* tcp_timer.c */ 566 void tcp_init_xmit_timers(struct sock *); 567 static inline void tcp_clear_xmit_timers(struct sock *sk) 568 { 569 inet_csk_clear_xmit_timers(sk); 570 } 571 572 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 573 unsigned int tcp_current_mss(struct sock *sk); 574 575 /* Bound MSS / TSO packet size with the half of the window */ 576 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 577 { 578 int cutoff; 579 580 /* When peer uses tiny windows, there is no use in packetizing 581 * to sub-MSS pieces for the sake of SWS or making sure there 582 * are enough packets in the pipe for fast recovery. 583 * 584 * On the other hand, for extremely large MSS devices, handling 585 * smaller than MSS windows in this way does make sense. 586 */ 587 if (tp->max_window >= 512) 588 cutoff = (tp->max_window >> 1); 589 else 590 cutoff = tp->max_window; 591 592 if (cutoff && pktsize > cutoff) 593 return max_t(int, cutoff, 68U - tp->tcp_header_len); 594 else 595 return pktsize; 596 } 597 598 /* tcp.c */ 599 void tcp_get_info(const struct sock *, struct tcp_info *); 600 601 /* Read 'sendfile()'-style from a TCP socket */ 602 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 603 unsigned int, size_t); 604 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 605 sk_read_actor_t recv_actor); 606 607 void tcp_initialize_rcv_mss(struct sock *sk); 608 609 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 610 int tcp_mss_to_mtu(struct sock *sk, int mss); 611 void tcp_mtup_init(struct sock *sk); 612 void tcp_init_buffer_space(struct sock *sk); 613 614 static inline void tcp_bound_rto(const struct sock *sk) 615 { 616 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 617 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 618 } 619 620 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 621 { 622 return (tp->srtt >> 3) + tp->rttvar; 623 } 624 625 void tcp_set_rto(struct sock *sk); 626 627 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 628 { 629 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 630 ntohl(TCP_FLAG_ACK) | 631 snd_wnd); 632 } 633 634 static inline void tcp_fast_path_on(struct tcp_sock *tp) 635 { 636 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 637 } 638 639 static inline void tcp_fast_path_check(struct sock *sk) 640 { 641 struct tcp_sock *tp = tcp_sk(sk); 642 643 if (skb_queue_empty(&tp->out_of_order_queue) && 644 tp->rcv_wnd && 645 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 646 !tp->urg_data) 647 tcp_fast_path_on(tp); 648 } 649 650 /* Compute the actual rto_min value */ 651 static inline u32 tcp_rto_min(struct sock *sk) 652 { 653 const struct dst_entry *dst = __sk_dst_get(sk); 654 u32 rto_min = TCP_RTO_MIN; 655 656 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 657 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 658 return rto_min; 659 } 660 661 /* Compute the actual receive window we are currently advertising. 662 * Rcv_nxt can be after the window if our peer push more data 663 * than the offered window. 664 */ 665 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 666 { 667 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 668 669 if (win < 0) 670 win = 0; 671 return (u32) win; 672 } 673 674 /* Choose a new window, without checks for shrinking, and without 675 * scaling applied to the result. The caller does these things 676 * if necessary. This is a "raw" window selection. 677 */ 678 u32 __tcp_select_window(struct sock *sk); 679 680 void tcp_send_window_probe(struct sock *sk); 681 682 /* TCP timestamps are only 32-bits, this causes a slight 683 * complication on 64-bit systems since we store a snapshot 684 * of jiffies in the buffer control blocks below. We decided 685 * to use only the low 32-bits of jiffies and hide the ugly 686 * casts with the following macro. 687 */ 688 #define tcp_time_stamp ((__u32)(jiffies)) 689 690 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 691 692 #define TCPHDR_FIN 0x01 693 #define TCPHDR_SYN 0x02 694 #define TCPHDR_RST 0x04 695 #define TCPHDR_PSH 0x08 696 #define TCPHDR_ACK 0x10 697 #define TCPHDR_URG 0x20 698 #define TCPHDR_ECE 0x40 699 #define TCPHDR_CWR 0x80 700 701 /* This is what the send packet queuing engine uses to pass 702 * TCP per-packet control information to the transmission code. 703 * We also store the host-order sequence numbers in here too. 704 * This is 44 bytes if IPV6 is enabled. 705 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 706 */ 707 struct tcp_skb_cb { 708 union { 709 struct inet_skb_parm h4; 710 #if IS_ENABLED(CONFIG_IPV6) 711 struct inet6_skb_parm h6; 712 #endif 713 } header; /* For incoming frames */ 714 __u32 seq; /* Starting sequence number */ 715 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 716 __u32 when; /* used to compute rtt's */ 717 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 718 719 __u8 sacked; /* State flags for SACK/FACK. */ 720 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 721 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 722 #define TCPCB_LOST 0x04 /* SKB is lost */ 723 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 724 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 725 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 726 727 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 728 /* 1 byte hole */ 729 __u32 ack_seq; /* Sequence number ACK'd */ 730 }; 731 732 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 733 734 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set 735 * 736 * If we receive a SYN packet with these bits set, it means a network is 737 * playing bad games with TOS bits. In order to avoid possible false congestion 738 * notifications, we disable TCP ECN negociation. 739 */ 740 static inline void 741 TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb, 742 struct net *net) 743 { 744 const struct tcphdr *th = tcp_hdr(skb); 745 746 if (net->ipv4.sysctl_tcp_ecn && th->ece && th->cwr && 747 INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield)) 748 inet_rsk(req)->ecn_ok = 1; 749 } 750 751 /* Due to TSO, an SKB can be composed of multiple actual 752 * packets. To keep these tracked properly, we use this. 753 */ 754 static inline int tcp_skb_pcount(const struct sk_buff *skb) 755 { 756 return skb_shinfo(skb)->gso_segs; 757 } 758 759 /* This is valid iff tcp_skb_pcount() > 1. */ 760 static inline int tcp_skb_mss(const struct sk_buff *skb) 761 { 762 return skb_shinfo(skb)->gso_size; 763 } 764 765 /* Events passed to congestion control interface */ 766 enum tcp_ca_event { 767 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 768 CA_EVENT_CWND_RESTART, /* congestion window restart */ 769 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 770 CA_EVENT_LOSS, /* loss timeout */ 771 CA_EVENT_FAST_ACK, /* in sequence ack */ 772 CA_EVENT_SLOW_ACK, /* other ack */ 773 }; 774 775 /* 776 * Interface for adding new TCP congestion control handlers 777 */ 778 #define TCP_CA_NAME_MAX 16 779 #define TCP_CA_MAX 128 780 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 781 782 #define TCP_CONG_NON_RESTRICTED 0x1 783 #define TCP_CONG_RTT_STAMP 0x2 784 785 struct tcp_congestion_ops { 786 struct list_head list; 787 unsigned long flags; 788 789 /* initialize private data (optional) */ 790 void (*init)(struct sock *sk); 791 /* cleanup private data (optional) */ 792 void (*release)(struct sock *sk); 793 794 /* return slow start threshold (required) */ 795 u32 (*ssthresh)(struct sock *sk); 796 /* lower bound for congestion window (optional) */ 797 u32 (*min_cwnd)(const struct sock *sk); 798 /* do new cwnd calculation (required) */ 799 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked, u32 in_flight); 800 /* call before changing ca_state (optional) */ 801 void (*set_state)(struct sock *sk, u8 new_state); 802 /* call when cwnd event occurs (optional) */ 803 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 804 /* new value of cwnd after loss (optional) */ 805 u32 (*undo_cwnd)(struct sock *sk); 806 /* hook for packet ack accounting (optional) */ 807 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 808 /* get info for inet_diag (optional) */ 809 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 810 811 char name[TCP_CA_NAME_MAX]; 812 struct module *owner; 813 }; 814 815 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 816 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 817 818 void tcp_init_congestion_control(struct sock *sk); 819 void tcp_cleanup_congestion_control(struct sock *sk); 820 int tcp_set_default_congestion_control(const char *name); 821 void tcp_get_default_congestion_control(char *name); 822 void tcp_get_available_congestion_control(char *buf, size_t len); 823 void tcp_get_allowed_congestion_control(char *buf, size_t len); 824 int tcp_set_allowed_congestion_control(char *allowed); 825 int tcp_set_congestion_control(struct sock *sk, const char *name); 826 int tcp_slow_start(struct tcp_sock *tp, u32 acked); 827 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 828 829 extern struct tcp_congestion_ops tcp_init_congestion_ops; 830 u32 tcp_reno_ssthresh(struct sock *sk); 831 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked, u32 in_flight); 832 u32 tcp_reno_min_cwnd(const struct sock *sk); 833 extern struct tcp_congestion_ops tcp_reno; 834 835 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 836 { 837 struct inet_connection_sock *icsk = inet_csk(sk); 838 839 if (icsk->icsk_ca_ops->set_state) 840 icsk->icsk_ca_ops->set_state(sk, ca_state); 841 icsk->icsk_ca_state = ca_state; 842 } 843 844 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 845 { 846 const struct inet_connection_sock *icsk = inet_csk(sk); 847 848 if (icsk->icsk_ca_ops->cwnd_event) 849 icsk->icsk_ca_ops->cwnd_event(sk, event); 850 } 851 852 /* These functions determine how the current flow behaves in respect of SACK 853 * handling. SACK is negotiated with the peer, and therefore it can vary 854 * between different flows. 855 * 856 * tcp_is_sack - SACK enabled 857 * tcp_is_reno - No SACK 858 * tcp_is_fack - FACK enabled, implies SACK enabled 859 */ 860 static inline int tcp_is_sack(const struct tcp_sock *tp) 861 { 862 return tp->rx_opt.sack_ok; 863 } 864 865 static inline bool tcp_is_reno(const struct tcp_sock *tp) 866 { 867 return !tcp_is_sack(tp); 868 } 869 870 static inline bool tcp_is_fack(const struct tcp_sock *tp) 871 { 872 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 873 } 874 875 static inline void tcp_enable_fack(struct tcp_sock *tp) 876 { 877 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 878 } 879 880 /* TCP early-retransmit (ER) is similar to but more conservative than 881 * the thin-dupack feature. Enable ER only if thin-dupack is disabled. 882 */ 883 static inline void tcp_enable_early_retrans(struct tcp_sock *tp) 884 { 885 tp->do_early_retrans = sysctl_tcp_early_retrans && 886 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack && 887 sysctl_tcp_reordering == 3; 888 } 889 890 static inline void tcp_disable_early_retrans(struct tcp_sock *tp) 891 { 892 tp->do_early_retrans = 0; 893 } 894 895 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 896 { 897 return tp->sacked_out + tp->lost_out; 898 } 899 900 /* This determines how many packets are "in the network" to the best 901 * of our knowledge. In many cases it is conservative, but where 902 * detailed information is available from the receiver (via SACK 903 * blocks etc.) we can make more aggressive calculations. 904 * 905 * Use this for decisions involving congestion control, use just 906 * tp->packets_out to determine if the send queue is empty or not. 907 * 908 * Read this equation as: 909 * 910 * "Packets sent once on transmission queue" MINUS 911 * "Packets left network, but not honestly ACKed yet" PLUS 912 * "Packets fast retransmitted" 913 */ 914 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 915 { 916 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 917 } 918 919 #define TCP_INFINITE_SSTHRESH 0x7fffffff 920 921 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 922 { 923 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 924 } 925 926 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 927 { 928 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 929 (1 << inet_csk(sk)->icsk_ca_state); 930 } 931 932 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 933 * The exception is cwnd reduction phase, when cwnd is decreasing towards 934 * ssthresh. 935 */ 936 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 937 { 938 const struct tcp_sock *tp = tcp_sk(sk); 939 940 if (tcp_in_cwnd_reduction(sk)) 941 return tp->snd_ssthresh; 942 else 943 return max(tp->snd_ssthresh, 944 ((tp->snd_cwnd >> 1) + 945 (tp->snd_cwnd >> 2))); 946 } 947 948 /* Use define here intentionally to get WARN_ON location shown at the caller */ 949 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 950 951 void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 952 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 953 954 /* The maximum number of MSS of available cwnd for which TSO defers 955 * sending if not using sysctl_tcp_tso_win_divisor. 956 */ 957 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 958 { 959 return 3; 960 } 961 962 /* Slow start with delack produces 3 packets of burst, so that 963 * it is safe "de facto". This will be the default - same as 964 * the default reordering threshold - but if reordering increases, 965 * we must be able to allow cwnd to burst at least this much in order 966 * to not pull it back when holes are filled. 967 */ 968 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 969 { 970 return tp->reordering; 971 } 972 973 /* Returns end sequence number of the receiver's advertised window */ 974 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 975 { 976 return tp->snd_una + tp->snd_wnd; 977 } 978 bool tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight); 979 980 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss, 981 const struct sk_buff *skb) 982 { 983 if (skb->len < mss) 984 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 985 } 986 987 static inline void tcp_check_probe_timer(struct sock *sk) 988 { 989 const struct tcp_sock *tp = tcp_sk(sk); 990 const struct inet_connection_sock *icsk = inet_csk(sk); 991 992 if (!tp->packets_out && !icsk->icsk_pending) 993 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 994 icsk->icsk_rto, TCP_RTO_MAX); 995 } 996 997 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 998 { 999 tp->snd_wl1 = seq; 1000 } 1001 1002 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1003 { 1004 tp->snd_wl1 = seq; 1005 } 1006 1007 /* 1008 * Calculate(/check) TCP checksum 1009 */ 1010 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1011 __be32 daddr, __wsum base) 1012 { 1013 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1014 } 1015 1016 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1017 { 1018 return __skb_checksum_complete(skb); 1019 } 1020 1021 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1022 { 1023 return !skb_csum_unnecessary(skb) && 1024 __tcp_checksum_complete(skb); 1025 } 1026 1027 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1028 1029 static inline void tcp_prequeue_init(struct tcp_sock *tp) 1030 { 1031 tp->ucopy.task = NULL; 1032 tp->ucopy.len = 0; 1033 tp->ucopy.memory = 0; 1034 skb_queue_head_init(&tp->ucopy.prequeue); 1035 #ifdef CONFIG_NET_DMA 1036 tp->ucopy.dma_chan = NULL; 1037 tp->ucopy.wakeup = 0; 1038 tp->ucopy.pinned_list = NULL; 1039 tp->ucopy.dma_cookie = 0; 1040 #endif 1041 } 1042 1043 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1044 1045 #undef STATE_TRACE 1046 1047 #ifdef STATE_TRACE 1048 static const char *statename[]={ 1049 "Unused","Established","Syn Sent","Syn Recv", 1050 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1051 "Close Wait","Last ACK","Listen","Closing" 1052 }; 1053 #endif 1054 void tcp_set_state(struct sock *sk, int state); 1055 1056 void tcp_done(struct sock *sk); 1057 1058 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1059 { 1060 rx_opt->dsack = 0; 1061 rx_opt->num_sacks = 0; 1062 } 1063 1064 u32 tcp_default_init_rwnd(u32 mss); 1065 1066 /* Determine a window scaling and initial window to offer. */ 1067 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1068 __u32 *window_clamp, int wscale_ok, 1069 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1070 1071 static inline int tcp_win_from_space(int space) 1072 { 1073 return sysctl_tcp_adv_win_scale<=0 ? 1074 (space>>(-sysctl_tcp_adv_win_scale)) : 1075 space - (space>>sysctl_tcp_adv_win_scale); 1076 } 1077 1078 /* Note: caller must be prepared to deal with negative returns */ 1079 static inline int tcp_space(const struct sock *sk) 1080 { 1081 return tcp_win_from_space(sk->sk_rcvbuf - 1082 atomic_read(&sk->sk_rmem_alloc)); 1083 } 1084 1085 static inline int tcp_full_space(const struct sock *sk) 1086 { 1087 return tcp_win_from_space(sk->sk_rcvbuf); 1088 } 1089 1090 static inline void tcp_openreq_init(struct request_sock *req, 1091 struct tcp_options_received *rx_opt, 1092 struct sk_buff *skb) 1093 { 1094 struct inet_request_sock *ireq = inet_rsk(req); 1095 1096 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 1097 req->cookie_ts = 0; 1098 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 1099 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1; 1100 tcp_rsk(req)->snt_synack = 0; 1101 req->mss = rx_opt->mss_clamp; 1102 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 1103 ireq->tstamp_ok = rx_opt->tstamp_ok; 1104 ireq->sack_ok = rx_opt->sack_ok; 1105 ireq->snd_wscale = rx_opt->snd_wscale; 1106 ireq->wscale_ok = rx_opt->wscale_ok; 1107 ireq->acked = 0; 1108 ireq->ecn_ok = 0; 1109 ireq->ir_rmt_port = tcp_hdr(skb)->source; 1110 ireq->ir_num = ntohs(tcp_hdr(skb)->dest); 1111 } 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 u16 copied; /* queued in tcp_connect() */ 1316 }; 1317 void tcp_free_fastopen_req(struct tcp_sock *tp); 1318 1319 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1320 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1321 void tcp_fastopen_cookie_gen(__be32 src, __be32 dst, 1322 struct tcp_fastopen_cookie *foc); 1323 void tcp_fastopen_init_key_once(bool publish); 1324 #define TCP_FASTOPEN_KEY_LENGTH 16 1325 1326 /* Fastopen key context */ 1327 struct tcp_fastopen_context { 1328 struct crypto_cipher *tfm; 1329 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1330 struct rcu_head rcu; 1331 }; 1332 1333 /* write queue abstraction */ 1334 static inline void tcp_write_queue_purge(struct sock *sk) 1335 { 1336 struct sk_buff *skb; 1337 1338 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1339 sk_wmem_free_skb(sk, skb); 1340 sk_mem_reclaim(sk); 1341 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1342 } 1343 1344 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1345 { 1346 return skb_peek(&sk->sk_write_queue); 1347 } 1348 1349 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1350 { 1351 return skb_peek_tail(&sk->sk_write_queue); 1352 } 1353 1354 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1355 const struct sk_buff *skb) 1356 { 1357 return skb_queue_next(&sk->sk_write_queue, skb); 1358 } 1359 1360 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1361 const struct sk_buff *skb) 1362 { 1363 return skb_queue_prev(&sk->sk_write_queue, skb); 1364 } 1365 1366 #define tcp_for_write_queue(skb, sk) \ 1367 skb_queue_walk(&(sk)->sk_write_queue, skb) 1368 1369 #define tcp_for_write_queue_from(skb, sk) \ 1370 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1371 1372 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1373 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1374 1375 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1376 { 1377 return sk->sk_send_head; 1378 } 1379 1380 static inline bool tcp_skb_is_last(const struct sock *sk, 1381 const struct sk_buff *skb) 1382 { 1383 return skb_queue_is_last(&sk->sk_write_queue, skb); 1384 } 1385 1386 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1387 { 1388 if (tcp_skb_is_last(sk, skb)) 1389 sk->sk_send_head = NULL; 1390 else 1391 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1392 } 1393 1394 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1395 { 1396 if (sk->sk_send_head == skb_unlinked) 1397 sk->sk_send_head = NULL; 1398 } 1399 1400 static inline void tcp_init_send_head(struct sock *sk) 1401 { 1402 sk->sk_send_head = NULL; 1403 } 1404 1405 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1406 { 1407 __skb_queue_tail(&sk->sk_write_queue, skb); 1408 } 1409 1410 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1411 { 1412 __tcp_add_write_queue_tail(sk, skb); 1413 1414 /* Queue it, remembering where we must start sending. */ 1415 if (sk->sk_send_head == NULL) { 1416 sk->sk_send_head = skb; 1417 1418 if (tcp_sk(sk)->highest_sack == NULL) 1419 tcp_sk(sk)->highest_sack = skb; 1420 } 1421 } 1422 1423 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1424 { 1425 __skb_queue_head(&sk->sk_write_queue, skb); 1426 } 1427 1428 /* Insert buff after skb on the write queue of sk. */ 1429 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1430 struct sk_buff *buff, 1431 struct sock *sk) 1432 { 1433 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1434 } 1435 1436 /* Insert new before skb on the write queue of sk. */ 1437 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1438 struct sk_buff *skb, 1439 struct sock *sk) 1440 { 1441 __skb_queue_before(&sk->sk_write_queue, skb, new); 1442 1443 if (sk->sk_send_head == skb) 1444 sk->sk_send_head = new; 1445 } 1446 1447 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1448 { 1449 __skb_unlink(skb, &sk->sk_write_queue); 1450 } 1451 1452 static inline bool tcp_write_queue_empty(struct sock *sk) 1453 { 1454 return skb_queue_empty(&sk->sk_write_queue); 1455 } 1456 1457 static inline void tcp_push_pending_frames(struct sock *sk) 1458 { 1459 if (tcp_send_head(sk)) { 1460 struct tcp_sock *tp = tcp_sk(sk); 1461 1462 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1463 } 1464 } 1465 1466 /* Start sequence of the skb just after the highest skb with SACKed 1467 * bit, valid only if sacked_out > 0 or when the caller has ensured 1468 * validity by itself. 1469 */ 1470 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1471 { 1472 if (!tp->sacked_out) 1473 return tp->snd_una; 1474 1475 if (tp->highest_sack == NULL) 1476 return tp->snd_nxt; 1477 1478 return TCP_SKB_CB(tp->highest_sack)->seq; 1479 } 1480 1481 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1482 { 1483 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1484 tcp_write_queue_next(sk, skb); 1485 } 1486 1487 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1488 { 1489 return tcp_sk(sk)->highest_sack; 1490 } 1491 1492 static inline void tcp_highest_sack_reset(struct sock *sk) 1493 { 1494 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1495 } 1496 1497 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1498 static inline void tcp_highest_sack_combine(struct sock *sk, 1499 struct sk_buff *old, 1500 struct sk_buff *new) 1501 { 1502 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1503 tcp_sk(sk)->highest_sack = new; 1504 } 1505 1506 /* Determines whether this is a thin stream (which may suffer from 1507 * increased latency). Used to trigger latency-reducing mechanisms. 1508 */ 1509 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1510 { 1511 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1512 } 1513 1514 /* /proc */ 1515 enum tcp_seq_states { 1516 TCP_SEQ_STATE_LISTENING, 1517 TCP_SEQ_STATE_OPENREQ, 1518 TCP_SEQ_STATE_ESTABLISHED, 1519 }; 1520 1521 int tcp_seq_open(struct inode *inode, struct file *file); 1522 1523 struct tcp_seq_afinfo { 1524 char *name; 1525 sa_family_t family; 1526 const struct file_operations *seq_fops; 1527 struct seq_operations seq_ops; 1528 }; 1529 1530 struct tcp_iter_state { 1531 struct seq_net_private p; 1532 sa_family_t family; 1533 enum tcp_seq_states state; 1534 struct sock *syn_wait_sk; 1535 int bucket, offset, sbucket, num; 1536 kuid_t uid; 1537 loff_t last_pos; 1538 }; 1539 1540 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1541 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1542 1543 extern struct request_sock_ops tcp_request_sock_ops; 1544 extern struct request_sock_ops tcp6_request_sock_ops; 1545 1546 void tcp_v4_destroy_sock(struct sock *sk); 1547 1548 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1549 netdev_features_t features); 1550 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1551 int tcp_gro_complete(struct sk_buff *skb); 1552 1553 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1554 1555 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1556 { 1557 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat; 1558 } 1559 1560 static inline bool tcp_stream_memory_free(const struct sock *sk) 1561 { 1562 const struct tcp_sock *tp = tcp_sk(sk); 1563 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1564 1565 return notsent_bytes < tcp_notsent_lowat(tp); 1566 } 1567 1568 #ifdef CONFIG_PROC_FS 1569 int tcp4_proc_init(void); 1570 void tcp4_proc_exit(void); 1571 #endif 1572 1573 /* TCP af-specific functions */ 1574 struct tcp_sock_af_ops { 1575 #ifdef CONFIG_TCP_MD5SIG 1576 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1577 struct sock *addr_sk); 1578 int (*calc_md5_hash) (char *location, 1579 struct tcp_md5sig_key *md5, 1580 const struct sock *sk, 1581 const struct request_sock *req, 1582 const struct sk_buff *skb); 1583 int (*md5_parse) (struct sock *sk, 1584 char __user *optval, 1585 int optlen); 1586 #endif 1587 }; 1588 1589 struct tcp_request_sock_ops { 1590 #ifdef CONFIG_TCP_MD5SIG 1591 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1592 struct request_sock *req); 1593 int (*calc_md5_hash) (char *location, 1594 struct tcp_md5sig_key *md5, 1595 const struct sock *sk, 1596 const struct request_sock *req, 1597 const struct sk_buff *skb); 1598 #endif 1599 }; 1600 1601 int tcpv4_offload_init(void); 1602 1603 void tcp_v4_init(void); 1604 void tcp_init(void); 1605 1606 #endif /* _TCP_H */ 1607