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 int sysctl_tcp_wmem[3]; 263 extern int sysctl_tcp_rmem[3]; 264 extern int sysctl_tcp_app_win; 265 extern int sysctl_tcp_adv_win_scale; 266 extern int sysctl_tcp_tw_reuse; 267 extern int sysctl_tcp_frto; 268 extern int sysctl_tcp_low_latency; 269 extern int sysctl_tcp_dma_copybreak; 270 extern int sysctl_tcp_nometrics_save; 271 extern int sysctl_tcp_moderate_rcvbuf; 272 extern int sysctl_tcp_tso_win_divisor; 273 extern int sysctl_tcp_mtu_probing; 274 extern int sysctl_tcp_base_mss; 275 extern int sysctl_tcp_workaround_signed_windows; 276 extern int sysctl_tcp_slow_start_after_idle; 277 extern int sysctl_tcp_max_ssthresh; 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_init_mem(struct net *net); 352 353 void tcp_tasklet_init(void); 354 355 void tcp_v4_err(struct sk_buff *skb, u32); 356 357 void tcp_shutdown(struct sock *sk, int how); 358 359 void tcp_v4_early_demux(struct sk_buff *skb); 360 int tcp_v4_rcv(struct sk_buff *skb); 361 362 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 363 int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 364 size_t size); 365 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 366 int flags); 367 void tcp_release_cb(struct sock *sk); 368 void tcp_wfree(struct sk_buff *skb); 369 void tcp_write_timer_handler(struct sock *sk); 370 void tcp_delack_timer_handler(struct sock *sk); 371 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 372 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 373 const struct tcphdr *th, unsigned int len); 374 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 375 const struct tcphdr *th, unsigned int len); 376 void tcp_rcv_space_adjust(struct sock *sk); 377 void tcp_cleanup_rbuf(struct sock *sk, int copied); 378 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 379 void tcp_twsk_destructor(struct sock *sk); 380 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 381 struct pipe_inode_info *pipe, size_t len, 382 unsigned int flags); 383 384 static inline void tcp_dec_quickack_mode(struct sock *sk, 385 const unsigned int pkts) 386 { 387 struct inet_connection_sock *icsk = inet_csk(sk); 388 389 if (icsk->icsk_ack.quick) { 390 if (pkts >= icsk->icsk_ack.quick) { 391 icsk->icsk_ack.quick = 0; 392 /* Leaving quickack mode we deflate ATO. */ 393 icsk->icsk_ack.ato = TCP_ATO_MIN; 394 } else 395 icsk->icsk_ack.quick -= pkts; 396 } 397 } 398 399 #define TCP_ECN_OK 1 400 #define TCP_ECN_QUEUE_CWR 2 401 #define TCP_ECN_DEMAND_CWR 4 402 #define TCP_ECN_SEEN 8 403 404 enum tcp_tw_status { 405 TCP_TW_SUCCESS = 0, 406 TCP_TW_RST = 1, 407 TCP_TW_ACK = 2, 408 TCP_TW_SYN = 3 409 }; 410 411 412 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 413 struct sk_buff *skb, 414 const struct tcphdr *th); 415 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 416 struct request_sock *req, struct request_sock **prev, 417 bool fastopen); 418 int tcp_child_process(struct sock *parent, struct sock *child, 419 struct sk_buff *skb); 420 void tcp_enter_loss(struct sock *sk, int how); 421 void tcp_clear_retrans(struct tcp_sock *tp); 422 void tcp_update_metrics(struct sock *sk); 423 void tcp_init_metrics(struct sock *sk); 424 void tcp_metrics_init(void); 425 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, 426 bool paws_check); 427 bool tcp_remember_stamp(struct sock *sk); 428 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); 429 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); 430 void tcp_disable_fack(struct tcp_sock *tp); 431 void tcp_close(struct sock *sk, long timeout); 432 void tcp_init_sock(struct sock *sk); 433 unsigned int tcp_poll(struct file *file, struct socket *sock, 434 struct poll_table_struct *wait); 435 int tcp_getsockopt(struct sock *sk, int level, int optname, 436 char __user *optval, int __user *optlen); 437 int tcp_setsockopt(struct sock *sk, int level, int optname, 438 char __user *optval, unsigned int optlen); 439 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 440 char __user *optval, int __user *optlen); 441 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 442 char __user *optval, unsigned int optlen); 443 void tcp_set_keepalive(struct sock *sk, int val); 444 void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req); 445 int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 446 size_t len, int nonblock, int flags, int *addr_len); 447 void tcp_parse_options(const struct sk_buff *skb, 448 struct tcp_options_received *opt_rx, 449 int estab, struct tcp_fastopen_cookie *foc); 450 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 451 452 /* 453 * TCP v4 functions exported for the inet6 API 454 */ 455 456 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 457 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 458 struct sock *tcp_create_openreq_child(struct sock *sk, 459 struct request_sock *req, 460 struct sk_buff *skb); 461 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 462 struct request_sock *req, 463 struct dst_entry *dst); 464 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 465 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 466 int tcp_connect(struct sock *sk); 467 struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst, 468 struct request_sock *req, 469 struct tcp_fastopen_cookie *foc); 470 int tcp_disconnect(struct sock *sk, int flags); 471 472 void tcp_connect_init(struct sock *sk); 473 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 474 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 475 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 476 477 /* From syncookies.c */ 478 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS]; 479 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 480 u32 cookie); 481 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 482 struct ip_options *opt); 483 #ifdef CONFIG_SYN_COOKIES 484 #include <linux/ktime.h> 485 486 /* Syncookies use a monotonic timer which increments every 64 seconds. 487 * This counter is used both as a hash input and partially encoded into 488 * the cookie value. A cookie is only validated further if the delta 489 * between the current counter value and the encoded one is less than this, 490 * i.e. a sent cookie is valid only at most for 128 seconds (or less if 491 * the counter advances immediately after a cookie is generated). 492 */ 493 #define MAX_SYNCOOKIE_AGE 2 494 495 static inline u32 tcp_cookie_time(void) 496 { 497 struct timespec now; 498 getnstimeofday(&now); 499 return now.tv_sec >> 6; /* 64 seconds granularity */ 500 } 501 502 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 503 u16 *mssp); 504 __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, __u16 *mss); 505 #else 506 static inline __u32 cookie_v4_init_sequence(struct sock *sk, 507 struct sk_buff *skb, 508 __u16 *mss) 509 { 510 return 0; 511 } 512 #endif 513 514 __u32 cookie_init_timestamp(struct request_sock *req); 515 bool cookie_check_timestamp(struct tcp_options_received *opt, struct net *net, 516 bool *ecn_ok); 517 518 /* From net/ipv6/syncookies.c */ 519 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 520 u32 cookie); 521 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 522 #ifdef CONFIG_SYN_COOKIES 523 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 524 const struct tcphdr *th, u16 *mssp); 525 __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb, 526 __u16 *mss); 527 #else 528 static inline __u32 cookie_v6_init_sequence(struct sock *sk, 529 struct sk_buff *skb, 530 __u16 *mss) 531 { 532 return 0; 533 } 534 #endif 535 /* tcp_output.c */ 536 537 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 538 int nonagle); 539 bool tcp_may_send_now(struct sock *sk); 540 int __tcp_retransmit_skb(struct sock *, struct sk_buff *); 541 int tcp_retransmit_skb(struct sock *, struct sk_buff *); 542 void tcp_retransmit_timer(struct sock *sk); 543 void tcp_xmit_retransmit_queue(struct sock *); 544 void tcp_simple_retransmit(struct sock *); 545 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 546 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 547 548 void tcp_send_probe0(struct sock *); 549 void tcp_send_partial(struct sock *); 550 int tcp_write_wakeup(struct sock *); 551 void tcp_send_fin(struct sock *sk); 552 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 553 int tcp_send_synack(struct sock *); 554 bool tcp_syn_flood_action(struct sock *sk, const struct sk_buff *skb, 555 const char *proto); 556 void tcp_push_one(struct sock *, unsigned int mss_now); 557 void tcp_send_ack(struct sock *sk); 558 void tcp_send_delayed_ack(struct sock *sk); 559 void tcp_send_loss_probe(struct sock *sk); 560 bool tcp_schedule_loss_probe(struct sock *sk); 561 562 /* tcp_input.c */ 563 void tcp_cwnd_application_limited(struct sock *sk); 564 void tcp_resume_early_retransmit(struct sock *sk); 565 void tcp_rearm_rto(struct sock *sk); 566 void tcp_reset(struct sock *sk); 567 568 /* tcp_timer.c */ 569 void tcp_init_xmit_timers(struct sock *); 570 static inline void tcp_clear_xmit_timers(struct sock *sk) 571 { 572 inet_csk_clear_xmit_timers(sk); 573 } 574 575 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 576 unsigned int tcp_current_mss(struct sock *sk); 577 578 /* Bound MSS / TSO packet size with the half of the window */ 579 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 580 { 581 int cutoff; 582 583 /* When peer uses tiny windows, there is no use in packetizing 584 * to sub-MSS pieces for the sake of SWS or making sure there 585 * are enough packets in the pipe for fast recovery. 586 * 587 * On the other hand, for extremely large MSS devices, handling 588 * smaller than MSS windows in this way does make sense. 589 */ 590 if (tp->max_window >= 512) 591 cutoff = (tp->max_window >> 1); 592 else 593 cutoff = tp->max_window; 594 595 if (cutoff && pktsize > cutoff) 596 return max_t(int, cutoff, 68U - tp->tcp_header_len); 597 else 598 return pktsize; 599 } 600 601 /* tcp.c */ 602 void tcp_get_info(const struct sock *, struct tcp_info *); 603 604 /* Read 'sendfile()'-style from a TCP socket */ 605 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 606 unsigned int, size_t); 607 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 608 sk_read_actor_t recv_actor); 609 610 void tcp_initialize_rcv_mss(struct sock *sk); 611 612 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 613 int tcp_mss_to_mtu(struct sock *sk, int mss); 614 void tcp_mtup_init(struct sock *sk); 615 void tcp_init_buffer_space(struct sock *sk); 616 617 static inline void tcp_bound_rto(const struct sock *sk) 618 { 619 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 620 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 621 } 622 623 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 624 { 625 return (tp->srtt >> 3) + tp->rttvar; 626 } 627 628 void tcp_set_rto(struct sock *sk); 629 630 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 631 { 632 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 633 ntohl(TCP_FLAG_ACK) | 634 snd_wnd); 635 } 636 637 static inline void tcp_fast_path_on(struct tcp_sock *tp) 638 { 639 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 640 } 641 642 static inline void tcp_fast_path_check(struct sock *sk) 643 { 644 struct tcp_sock *tp = tcp_sk(sk); 645 646 if (skb_queue_empty(&tp->out_of_order_queue) && 647 tp->rcv_wnd && 648 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 649 !tp->urg_data) 650 tcp_fast_path_on(tp); 651 } 652 653 /* Compute the actual rto_min value */ 654 static inline u32 tcp_rto_min(struct sock *sk) 655 { 656 const struct dst_entry *dst = __sk_dst_get(sk); 657 u32 rto_min = TCP_RTO_MIN; 658 659 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 660 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 661 return rto_min; 662 } 663 664 /* Compute the actual receive window we are currently advertising. 665 * Rcv_nxt can be after the window if our peer push more data 666 * than the offered window. 667 */ 668 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 669 { 670 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 671 672 if (win < 0) 673 win = 0; 674 return (u32) win; 675 } 676 677 /* Choose a new window, without checks for shrinking, and without 678 * scaling applied to the result. The caller does these things 679 * if necessary. This is a "raw" window selection. 680 */ 681 u32 __tcp_select_window(struct sock *sk); 682 683 void tcp_send_window_probe(struct sock *sk); 684 685 /* TCP timestamps are only 32-bits, this causes a slight 686 * complication on 64-bit systems since we store a snapshot 687 * of jiffies in the buffer control blocks below. We decided 688 * to use only the low 32-bits of jiffies and hide the ugly 689 * casts with the following macro. 690 */ 691 #define tcp_time_stamp ((__u32)(jiffies)) 692 693 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 694 695 #define TCPHDR_FIN 0x01 696 #define TCPHDR_SYN 0x02 697 #define TCPHDR_RST 0x04 698 #define TCPHDR_PSH 0x08 699 #define TCPHDR_ACK 0x10 700 #define TCPHDR_URG 0x20 701 #define TCPHDR_ECE 0x40 702 #define TCPHDR_CWR 0x80 703 704 /* This is what the send packet queuing engine uses to pass 705 * TCP per-packet control information to the transmission code. 706 * We also store the host-order sequence numbers in here too. 707 * This is 44 bytes if IPV6 is enabled. 708 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 709 */ 710 struct tcp_skb_cb { 711 union { 712 struct inet_skb_parm h4; 713 #if IS_ENABLED(CONFIG_IPV6) 714 struct inet6_skb_parm h6; 715 #endif 716 } header; /* For incoming frames */ 717 __u32 seq; /* Starting sequence number */ 718 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 719 __u32 when; /* used to compute rtt's */ 720 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 721 722 __u8 sacked; /* State flags for SACK/FACK. */ 723 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 724 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 725 #define TCPCB_LOST 0x04 /* SKB is lost */ 726 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 727 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 728 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 729 730 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 731 /* 1 byte hole */ 732 __u32 ack_seq; /* Sequence number ACK'd */ 733 }; 734 735 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 736 737 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set 738 * 739 * If we receive a SYN packet with these bits set, it means a network is 740 * playing bad games with TOS bits. In order to avoid possible false congestion 741 * notifications, we disable TCP ECN negociation. 742 */ 743 static inline void 744 TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb, 745 struct net *net) 746 { 747 const struct tcphdr *th = tcp_hdr(skb); 748 749 if (net->ipv4.sysctl_tcp_ecn && th->ece && th->cwr && 750 INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield)) 751 inet_rsk(req)->ecn_ok = 1; 752 } 753 754 /* Due to TSO, an SKB can be composed of multiple actual 755 * packets. To keep these tracked properly, we use this. 756 */ 757 static inline int tcp_skb_pcount(const struct sk_buff *skb) 758 { 759 return skb_shinfo(skb)->gso_segs; 760 } 761 762 /* This is valid iff tcp_skb_pcount() > 1. */ 763 static inline int tcp_skb_mss(const struct sk_buff *skb) 764 { 765 return skb_shinfo(skb)->gso_size; 766 } 767 768 /* Events passed to congestion control interface */ 769 enum tcp_ca_event { 770 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 771 CA_EVENT_CWND_RESTART, /* congestion window restart */ 772 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 773 CA_EVENT_LOSS, /* loss timeout */ 774 CA_EVENT_FAST_ACK, /* in sequence ack */ 775 CA_EVENT_SLOW_ACK, /* other ack */ 776 }; 777 778 /* 779 * Interface for adding new TCP congestion control handlers 780 */ 781 #define TCP_CA_NAME_MAX 16 782 #define TCP_CA_MAX 128 783 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 784 785 #define TCP_CONG_NON_RESTRICTED 0x1 786 #define TCP_CONG_RTT_STAMP 0x2 787 788 struct tcp_congestion_ops { 789 struct list_head list; 790 unsigned long flags; 791 792 /* initialize private data (optional) */ 793 void (*init)(struct sock *sk); 794 /* cleanup private data (optional) */ 795 void (*release)(struct sock *sk); 796 797 /* return slow start threshold (required) */ 798 u32 (*ssthresh)(struct sock *sk); 799 /* lower bound for congestion window (optional) */ 800 u32 (*min_cwnd)(const struct sock *sk); 801 /* do new cwnd calculation (required) */ 802 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight); 803 /* call before changing ca_state (optional) */ 804 void (*set_state)(struct sock *sk, u8 new_state); 805 /* call when cwnd event occurs (optional) */ 806 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 807 /* new value of cwnd after loss (optional) */ 808 u32 (*undo_cwnd)(struct sock *sk); 809 /* hook for packet ack accounting (optional) */ 810 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 811 /* get info for inet_diag (optional) */ 812 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 813 814 char name[TCP_CA_NAME_MAX]; 815 struct module *owner; 816 }; 817 818 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 819 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 820 821 void tcp_init_congestion_control(struct sock *sk); 822 void tcp_cleanup_congestion_control(struct sock *sk); 823 int tcp_set_default_congestion_control(const char *name); 824 void tcp_get_default_congestion_control(char *name); 825 void tcp_get_available_congestion_control(char *buf, size_t len); 826 void tcp_get_allowed_congestion_control(char *buf, size_t len); 827 int tcp_set_allowed_congestion_control(char *allowed); 828 int tcp_set_congestion_control(struct sock *sk, const char *name); 829 void tcp_slow_start(struct tcp_sock *tp); 830 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 831 832 extern struct tcp_congestion_ops tcp_init_congestion_ops; 833 u32 tcp_reno_ssthresh(struct sock *sk); 834 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight); 835 u32 tcp_reno_min_cwnd(const struct sock *sk); 836 extern struct tcp_congestion_ops tcp_reno; 837 838 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 839 { 840 struct inet_connection_sock *icsk = inet_csk(sk); 841 842 if (icsk->icsk_ca_ops->set_state) 843 icsk->icsk_ca_ops->set_state(sk, ca_state); 844 icsk->icsk_ca_state = ca_state; 845 } 846 847 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 848 { 849 const struct inet_connection_sock *icsk = inet_csk(sk); 850 851 if (icsk->icsk_ca_ops->cwnd_event) 852 icsk->icsk_ca_ops->cwnd_event(sk, event); 853 } 854 855 /* These functions determine how the current flow behaves in respect of SACK 856 * handling. SACK is negotiated with the peer, and therefore it can vary 857 * between different flows. 858 * 859 * tcp_is_sack - SACK enabled 860 * tcp_is_reno - No SACK 861 * tcp_is_fack - FACK enabled, implies SACK enabled 862 */ 863 static inline int tcp_is_sack(const struct tcp_sock *tp) 864 { 865 return tp->rx_opt.sack_ok; 866 } 867 868 static inline bool tcp_is_reno(const struct tcp_sock *tp) 869 { 870 return !tcp_is_sack(tp); 871 } 872 873 static inline bool tcp_is_fack(const struct tcp_sock *tp) 874 { 875 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 876 } 877 878 static inline void tcp_enable_fack(struct tcp_sock *tp) 879 { 880 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 881 } 882 883 /* TCP early-retransmit (ER) is similar to but more conservative than 884 * the thin-dupack feature. Enable ER only if thin-dupack is disabled. 885 */ 886 static inline void tcp_enable_early_retrans(struct tcp_sock *tp) 887 { 888 tp->do_early_retrans = sysctl_tcp_early_retrans && 889 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack && 890 sysctl_tcp_reordering == 3; 891 } 892 893 static inline void tcp_disable_early_retrans(struct tcp_sock *tp) 894 { 895 tp->do_early_retrans = 0; 896 } 897 898 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 899 { 900 return tp->sacked_out + tp->lost_out; 901 } 902 903 /* This determines how many packets are "in the network" to the best 904 * of our knowledge. In many cases it is conservative, but where 905 * detailed information is available from the receiver (via SACK 906 * blocks etc.) we can make more aggressive calculations. 907 * 908 * Use this for decisions involving congestion control, use just 909 * tp->packets_out to determine if the send queue is empty or not. 910 * 911 * Read this equation as: 912 * 913 * "Packets sent once on transmission queue" MINUS 914 * "Packets left network, but not honestly ACKed yet" PLUS 915 * "Packets fast retransmitted" 916 */ 917 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 918 { 919 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 920 } 921 922 #define TCP_INFINITE_SSTHRESH 0x7fffffff 923 924 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 925 { 926 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 927 } 928 929 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 930 { 931 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 932 (1 << inet_csk(sk)->icsk_ca_state); 933 } 934 935 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 936 * The exception is cwnd reduction phase, when cwnd is decreasing towards 937 * ssthresh. 938 */ 939 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 940 { 941 const struct tcp_sock *tp = tcp_sk(sk); 942 943 if (tcp_in_cwnd_reduction(sk)) 944 return tp->snd_ssthresh; 945 else 946 return max(tp->snd_ssthresh, 947 ((tp->snd_cwnd >> 1) + 948 (tp->snd_cwnd >> 2))); 949 } 950 951 /* Use define here intentionally to get WARN_ON location shown at the caller */ 952 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 953 954 void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 955 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 956 957 /* The maximum number of MSS of available cwnd for which TSO defers 958 * sending if not using sysctl_tcp_tso_win_divisor. 959 */ 960 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 961 { 962 return 3; 963 } 964 965 /* Slow start with delack produces 3 packets of burst, so that 966 * it is safe "de facto". This will be the default - same as 967 * the default reordering threshold - but if reordering increases, 968 * we must be able to allow cwnd to burst at least this much in order 969 * to not pull it back when holes are filled. 970 */ 971 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 972 { 973 return tp->reordering; 974 } 975 976 /* Returns end sequence number of the receiver's advertised window */ 977 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 978 { 979 return tp->snd_una + tp->snd_wnd; 980 } 981 bool tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight); 982 983 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss, 984 const struct sk_buff *skb) 985 { 986 if (skb->len < mss) 987 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 988 } 989 990 static inline void tcp_check_probe_timer(struct sock *sk) 991 { 992 const struct tcp_sock *tp = tcp_sk(sk); 993 const struct inet_connection_sock *icsk = inet_csk(sk); 994 995 if (!tp->packets_out && !icsk->icsk_pending) 996 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 997 icsk->icsk_rto, TCP_RTO_MAX); 998 } 999 1000 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1001 { 1002 tp->snd_wl1 = seq; 1003 } 1004 1005 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1006 { 1007 tp->snd_wl1 = seq; 1008 } 1009 1010 /* 1011 * Calculate(/check) TCP checksum 1012 */ 1013 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1014 __be32 daddr, __wsum base) 1015 { 1016 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1017 } 1018 1019 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1020 { 1021 return __skb_checksum_complete(skb); 1022 } 1023 1024 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1025 { 1026 return !skb_csum_unnecessary(skb) && 1027 __tcp_checksum_complete(skb); 1028 } 1029 1030 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1031 1032 static inline void tcp_prequeue_init(struct tcp_sock *tp) 1033 { 1034 tp->ucopy.task = NULL; 1035 tp->ucopy.len = 0; 1036 tp->ucopy.memory = 0; 1037 skb_queue_head_init(&tp->ucopy.prequeue); 1038 #ifdef CONFIG_NET_DMA 1039 tp->ucopy.dma_chan = NULL; 1040 tp->ucopy.wakeup = 0; 1041 tp->ucopy.pinned_list = NULL; 1042 tp->ucopy.dma_cookie = 0; 1043 #endif 1044 } 1045 1046 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1047 1048 #undef STATE_TRACE 1049 1050 #ifdef STATE_TRACE 1051 static const char *statename[]={ 1052 "Unused","Established","Syn Sent","Syn Recv", 1053 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1054 "Close Wait","Last ACK","Listen","Closing" 1055 }; 1056 #endif 1057 void tcp_set_state(struct sock *sk, int state); 1058 1059 void tcp_done(struct sock *sk); 1060 1061 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1062 { 1063 rx_opt->dsack = 0; 1064 rx_opt->num_sacks = 0; 1065 } 1066 1067 u32 tcp_default_init_rwnd(u32 mss); 1068 1069 /* Determine a window scaling and initial window to offer. */ 1070 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1071 __u32 *window_clamp, int wscale_ok, 1072 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1073 1074 static inline int tcp_win_from_space(int space) 1075 { 1076 return sysctl_tcp_adv_win_scale<=0 ? 1077 (space>>(-sysctl_tcp_adv_win_scale)) : 1078 space - (space>>sysctl_tcp_adv_win_scale); 1079 } 1080 1081 /* Note: caller must be prepared to deal with negative returns */ 1082 static inline int tcp_space(const struct sock *sk) 1083 { 1084 return tcp_win_from_space(sk->sk_rcvbuf - 1085 atomic_read(&sk->sk_rmem_alloc)); 1086 } 1087 1088 static inline int tcp_full_space(const struct sock *sk) 1089 { 1090 return tcp_win_from_space(sk->sk_rcvbuf); 1091 } 1092 1093 static inline void tcp_openreq_init(struct request_sock *req, 1094 struct tcp_options_received *rx_opt, 1095 struct sk_buff *skb) 1096 { 1097 struct inet_request_sock *ireq = inet_rsk(req); 1098 1099 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 1100 req->cookie_ts = 0; 1101 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 1102 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1; 1103 tcp_rsk(req)->snt_synack = 0; 1104 req->mss = rx_opt->mss_clamp; 1105 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 1106 ireq->tstamp_ok = rx_opt->tstamp_ok; 1107 ireq->sack_ok = rx_opt->sack_ok; 1108 ireq->snd_wscale = rx_opt->snd_wscale; 1109 ireq->wscale_ok = rx_opt->wscale_ok; 1110 ireq->acked = 0; 1111 ireq->ecn_ok = 0; 1112 ireq->rmt_port = tcp_hdr(skb)->source; 1113 ireq->loc_port = tcp_hdr(skb)->dest; 1114 } 1115 1116 void tcp_enter_memory_pressure(struct sock *sk); 1117 1118 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1119 { 1120 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1121 } 1122 1123 static inline int keepalive_time_when(const struct tcp_sock *tp) 1124 { 1125 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1126 } 1127 1128 static inline int keepalive_probes(const struct tcp_sock *tp) 1129 { 1130 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1131 } 1132 1133 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1134 { 1135 const struct inet_connection_sock *icsk = &tp->inet_conn; 1136 1137 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1138 tcp_time_stamp - tp->rcv_tstamp); 1139 } 1140 1141 static inline int tcp_fin_time(const struct sock *sk) 1142 { 1143 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1144 const int rto = inet_csk(sk)->icsk_rto; 1145 1146 if (fin_timeout < (rto << 2) - (rto >> 1)) 1147 fin_timeout = (rto << 2) - (rto >> 1); 1148 1149 return fin_timeout; 1150 } 1151 1152 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1153 int paws_win) 1154 { 1155 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1156 return true; 1157 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1158 return true; 1159 /* 1160 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1161 * then following tcp messages have valid values. Ignore 0 value, 1162 * or else 'negative' tsval might forbid us to accept their packets. 1163 */ 1164 if (!rx_opt->ts_recent) 1165 return true; 1166 return false; 1167 } 1168 1169 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1170 int rst) 1171 { 1172 if (tcp_paws_check(rx_opt, 0)) 1173 return false; 1174 1175 /* RST segments are not recommended to carry timestamp, 1176 and, if they do, it is recommended to ignore PAWS because 1177 "their cleanup function should take precedence over timestamps." 1178 Certainly, it is mistake. It is necessary to understand the reasons 1179 of this constraint to relax it: if peer reboots, clock may go 1180 out-of-sync and half-open connections will not be reset. 1181 Actually, the problem would be not existing if all 1182 the implementations followed draft about maintaining clock 1183 via reboots. Linux-2.2 DOES NOT! 1184 1185 However, we can relax time bounds for RST segments to MSL. 1186 */ 1187 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1188 return false; 1189 return true; 1190 } 1191 1192 static inline void tcp_mib_init(struct net *net) 1193 { 1194 /* See RFC 2012 */ 1195 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1196 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1197 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1198 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1199 } 1200 1201 /* from STCP */ 1202 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1203 { 1204 tp->lost_skb_hint = NULL; 1205 } 1206 1207 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1208 { 1209 tcp_clear_retrans_hints_partial(tp); 1210 tp->retransmit_skb_hint = NULL; 1211 } 1212 1213 /* MD5 Signature */ 1214 struct crypto_hash; 1215 1216 union tcp_md5_addr { 1217 struct in_addr a4; 1218 #if IS_ENABLED(CONFIG_IPV6) 1219 struct in6_addr a6; 1220 #endif 1221 }; 1222 1223 /* - key database */ 1224 struct tcp_md5sig_key { 1225 struct hlist_node node; 1226 u8 keylen; 1227 u8 family; /* AF_INET or AF_INET6 */ 1228 union tcp_md5_addr addr; 1229 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1230 struct rcu_head rcu; 1231 }; 1232 1233 /* - sock block */ 1234 struct tcp_md5sig_info { 1235 struct hlist_head head; 1236 struct rcu_head rcu; 1237 }; 1238 1239 /* - pseudo header */ 1240 struct tcp4_pseudohdr { 1241 __be32 saddr; 1242 __be32 daddr; 1243 __u8 pad; 1244 __u8 protocol; 1245 __be16 len; 1246 }; 1247 1248 struct tcp6_pseudohdr { 1249 struct in6_addr saddr; 1250 struct in6_addr daddr; 1251 __be32 len; 1252 __be32 protocol; /* including padding */ 1253 }; 1254 1255 union tcp_md5sum_block { 1256 struct tcp4_pseudohdr ip4; 1257 #if IS_ENABLED(CONFIG_IPV6) 1258 struct tcp6_pseudohdr ip6; 1259 #endif 1260 }; 1261 1262 /* - pool: digest algorithm, hash description and scratch buffer */ 1263 struct tcp_md5sig_pool { 1264 struct hash_desc md5_desc; 1265 union tcp_md5sum_block md5_blk; 1266 }; 1267 1268 /* - functions */ 1269 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1270 const struct sock *sk, const struct request_sock *req, 1271 const struct sk_buff *skb); 1272 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1273 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); 1274 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1275 int family); 1276 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1277 struct sock *addr_sk); 1278 1279 #ifdef CONFIG_TCP_MD5SIG 1280 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1281 const union tcp_md5_addr *addr, 1282 int family); 1283 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1284 #else 1285 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1286 const union tcp_md5_addr *addr, 1287 int family) 1288 { 1289 return NULL; 1290 } 1291 #define tcp_twsk_md5_key(twsk) NULL 1292 #endif 1293 1294 bool tcp_alloc_md5sig_pool(void); 1295 1296 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1297 static inline void tcp_put_md5sig_pool(void) 1298 { 1299 local_bh_enable(); 1300 } 1301 1302 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1303 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1304 unsigned int header_len); 1305 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1306 const struct tcp_md5sig_key *key); 1307 1308 /* From tcp_fastopen.c */ 1309 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1310 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1311 unsigned long *last_syn_loss); 1312 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1313 struct tcp_fastopen_cookie *cookie, bool syn_lost); 1314 struct tcp_fastopen_request { 1315 /* Fast Open cookie. Size 0 means a cookie request */ 1316 struct tcp_fastopen_cookie cookie; 1317 struct msghdr *data; /* data in MSG_FASTOPEN */ 1318 u16 copied; /* queued in tcp_connect() */ 1319 }; 1320 void tcp_free_fastopen_req(struct tcp_sock *tp); 1321 1322 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1323 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1324 void tcp_fastopen_cookie_gen(__be32 src, __be32 dst, 1325 struct tcp_fastopen_cookie *foc); 1326 1327 #define TCP_FASTOPEN_KEY_LENGTH 16 1328 1329 /* Fastopen key context */ 1330 struct tcp_fastopen_context { 1331 struct crypto_cipher *tfm; 1332 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1333 struct rcu_head rcu; 1334 }; 1335 1336 /* write queue abstraction */ 1337 static inline void tcp_write_queue_purge(struct sock *sk) 1338 { 1339 struct sk_buff *skb; 1340 1341 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1342 sk_wmem_free_skb(sk, skb); 1343 sk_mem_reclaim(sk); 1344 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1345 } 1346 1347 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1348 { 1349 return skb_peek(&sk->sk_write_queue); 1350 } 1351 1352 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1353 { 1354 return skb_peek_tail(&sk->sk_write_queue); 1355 } 1356 1357 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1358 const struct sk_buff *skb) 1359 { 1360 return skb_queue_next(&sk->sk_write_queue, skb); 1361 } 1362 1363 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1364 const struct sk_buff *skb) 1365 { 1366 return skb_queue_prev(&sk->sk_write_queue, skb); 1367 } 1368 1369 #define tcp_for_write_queue(skb, sk) \ 1370 skb_queue_walk(&(sk)->sk_write_queue, skb) 1371 1372 #define tcp_for_write_queue_from(skb, sk) \ 1373 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1374 1375 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1376 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1377 1378 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1379 { 1380 return sk->sk_send_head; 1381 } 1382 1383 static inline bool tcp_skb_is_last(const struct sock *sk, 1384 const struct sk_buff *skb) 1385 { 1386 return skb_queue_is_last(&sk->sk_write_queue, skb); 1387 } 1388 1389 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1390 { 1391 if (tcp_skb_is_last(sk, skb)) 1392 sk->sk_send_head = NULL; 1393 else 1394 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1395 } 1396 1397 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1398 { 1399 if (sk->sk_send_head == skb_unlinked) 1400 sk->sk_send_head = NULL; 1401 } 1402 1403 static inline void tcp_init_send_head(struct sock *sk) 1404 { 1405 sk->sk_send_head = NULL; 1406 } 1407 1408 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1409 { 1410 __skb_queue_tail(&sk->sk_write_queue, skb); 1411 } 1412 1413 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1414 { 1415 __tcp_add_write_queue_tail(sk, skb); 1416 1417 /* Queue it, remembering where we must start sending. */ 1418 if (sk->sk_send_head == NULL) { 1419 sk->sk_send_head = skb; 1420 1421 if (tcp_sk(sk)->highest_sack == NULL) 1422 tcp_sk(sk)->highest_sack = skb; 1423 } 1424 } 1425 1426 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1427 { 1428 __skb_queue_head(&sk->sk_write_queue, skb); 1429 } 1430 1431 /* Insert buff after skb on the write queue of sk. */ 1432 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1433 struct sk_buff *buff, 1434 struct sock *sk) 1435 { 1436 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1437 } 1438 1439 /* Insert new before skb on the write queue of sk. */ 1440 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1441 struct sk_buff *skb, 1442 struct sock *sk) 1443 { 1444 __skb_queue_before(&sk->sk_write_queue, skb, new); 1445 1446 if (sk->sk_send_head == skb) 1447 sk->sk_send_head = new; 1448 } 1449 1450 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1451 { 1452 __skb_unlink(skb, &sk->sk_write_queue); 1453 } 1454 1455 static inline bool tcp_write_queue_empty(struct sock *sk) 1456 { 1457 return skb_queue_empty(&sk->sk_write_queue); 1458 } 1459 1460 static inline void tcp_push_pending_frames(struct sock *sk) 1461 { 1462 if (tcp_send_head(sk)) { 1463 struct tcp_sock *tp = tcp_sk(sk); 1464 1465 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1466 } 1467 } 1468 1469 /* Start sequence of the skb just after the highest skb with SACKed 1470 * bit, valid only if sacked_out > 0 or when the caller has ensured 1471 * validity by itself. 1472 */ 1473 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1474 { 1475 if (!tp->sacked_out) 1476 return tp->snd_una; 1477 1478 if (tp->highest_sack == NULL) 1479 return tp->snd_nxt; 1480 1481 return TCP_SKB_CB(tp->highest_sack)->seq; 1482 } 1483 1484 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1485 { 1486 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1487 tcp_write_queue_next(sk, skb); 1488 } 1489 1490 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1491 { 1492 return tcp_sk(sk)->highest_sack; 1493 } 1494 1495 static inline void tcp_highest_sack_reset(struct sock *sk) 1496 { 1497 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1498 } 1499 1500 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1501 static inline void tcp_highest_sack_combine(struct sock *sk, 1502 struct sk_buff *old, 1503 struct sk_buff *new) 1504 { 1505 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1506 tcp_sk(sk)->highest_sack = new; 1507 } 1508 1509 /* Determines whether this is a thin stream (which may suffer from 1510 * increased latency). Used to trigger latency-reducing mechanisms. 1511 */ 1512 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1513 { 1514 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1515 } 1516 1517 /* /proc */ 1518 enum tcp_seq_states { 1519 TCP_SEQ_STATE_LISTENING, 1520 TCP_SEQ_STATE_OPENREQ, 1521 TCP_SEQ_STATE_ESTABLISHED, 1522 TCP_SEQ_STATE_TIME_WAIT, 1523 }; 1524 1525 int tcp_seq_open(struct inode *inode, struct file *file); 1526 1527 struct tcp_seq_afinfo { 1528 char *name; 1529 sa_family_t family; 1530 const struct file_operations *seq_fops; 1531 struct seq_operations seq_ops; 1532 }; 1533 1534 struct tcp_iter_state { 1535 struct seq_net_private p; 1536 sa_family_t family; 1537 enum tcp_seq_states state; 1538 struct sock *syn_wait_sk; 1539 int bucket, offset, sbucket, num; 1540 kuid_t uid; 1541 loff_t last_pos; 1542 }; 1543 1544 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1545 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1546 1547 extern struct request_sock_ops tcp_request_sock_ops; 1548 extern struct request_sock_ops tcp6_request_sock_ops; 1549 1550 void tcp_v4_destroy_sock(struct sock *sk); 1551 1552 struct sk_buff *tcp_tso_segment(struct sk_buff *skb, 1553 netdev_features_t features); 1554 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1555 int tcp_gro_complete(struct sk_buff *skb); 1556 1557 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1558 1559 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1560 { 1561 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat; 1562 } 1563 1564 static inline bool tcp_stream_memory_free(const struct sock *sk) 1565 { 1566 const struct tcp_sock *tp = tcp_sk(sk); 1567 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1568 1569 return notsent_bytes < tcp_notsent_lowat(tp); 1570 } 1571 1572 #ifdef CONFIG_PROC_FS 1573 int tcp4_proc_init(void); 1574 void tcp4_proc_exit(void); 1575 #endif 1576 1577 /* TCP af-specific functions */ 1578 struct tcp_sock_af_ops { 1579 #ifdef CONFIG_TCP_MD5SIG 1580 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1581 struct sock *addr_sk); 1582 int (*calc_md5_hash) (char *location, 1583 struct tcp_md5sig_key *md5, 1584 const struct sock *sk, 1585 const struct request_sock *req, 1586 const struct sk_buff *skb); 1587 int (*md5_parse) (struct sock *sk, 1588 char __user *optval, 1589 int optlen); 1590 #endif 1591 }; 1592 1593 struct tcp_request_sock_ops { 1594 #ifdef CONFIG_TCP_MD5SIG 1595 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1596 struct request_sock *req); 1597 int (*calc_md5_hash) (char *location, 1598 struct tcp_md5sig_key *md5, 1599 const struct sock *sk, 1600 const struct request_sock *req, 1601 const struct sk_buff *skb); 1602 #endif 1603 }; 1604 1605 int tcpv4_offload_init(void); 1606 1607 void tcp_v4_init(void); 1608 void tcp_init(void); 1609 1610 #endif /* _TCP_H */ 1611