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 extern 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 /* Offer an initial receive window of 10 mss. */ 65 #define TCP_DEFAULT_INIT_RCVWND 10 66 67 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 68 #define TCP_MIN_MSS 88U 69 70 /* The least MTU to use for probing */ 71 #define TCP_BASE_MSS 512 72 73 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 74 #define TCP_FASTRETRANS_THRESH 3 75 76 /* Maximal reordering. */ 77 #define TCP_MAX_REORDERING 127 78 79 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 80 #define TCP_MAX_QUICKACKS 16U 81 82 /* urg_data states */ 83 #define TCP_URG_VALID 0x0100 84 #define TCP_URG_NOTYET 0x0200 85 #define TCP_URG_READ 0x0400 86 87 #define TCP_RETR1 3 /* 88 * This is how many retries it does before it 89 * tries to figure out if the gateway is 90 * down. Minimal RFC value is 3; it corresponds 91 * to ~3sec-8min depending on RTO. 92 */ 93 94 #define TCP_RETR2 15 /* 95 * This should take at least 96 * 90 minutes to time out. 97 * RFC1122 says that the limit is 100 sec. 98 * 15 is ~13-30min depending on RTO. 99 */ 100 101 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a 102 * connection: ~180sec is RFC minimum */ 103 104 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a 105 * connection: ~180sec is RFC minimum */ 106 107 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 108 * state, about 60 seconds */ 109 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 110 /* BSD style FIN_WAIT2 deadlock breaker. 111 * It used to be 3min, new value is 60sec, 112 * to combine FIN-WAIT-2 timeout with 113 * TIME-WAIT timer. 114 */ 115 116 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 117 #if HZ >= 100 118 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 119 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 120 #else 121 #define TCP_DELACK_MIN 4U 122 #define TCP_ATO_MIN 4U 123 #endif 124 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 125 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 126 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 127 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 128 * used as a fallback RTO for the 129 * initial data transmission if no 130 * valid RTT sample has been acquired, 131 * most likely due to retrans in 3WHS. 132 */ 133 134 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 135 * for local resources. 136 */ 137 138 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 139 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 140 #define TCP_KEEPALIVE_INTVL (75*HZ) 141 142 #define MAX_TCP_KEEPIDLE 32767 143 #define MAX_TCP_KEEPINTVL 32767 144 #define MAX_TCP_KEEPCNT 127 145 #define MAX_TCP_SYNCNT 127 146 147 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 148 149 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 150 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 151 * after this time. It should be equal 152 * (or greater than) TCP_TIMEWAIT_LEN 153 * to provide reliability equal to one 154 * provided by timewait state. 155 */ 156 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 157 * timestamps. It must be less than 158 * minimal timewait lifetime. 159 */ 160 /* 161 * TCP option 162 */ 163 164 #define TCPOPT_NOP 1 /* Padding */ 165 #define TCPOPT_EOL 0 /* End of options */ 166 #define TCPOPT_MSS 2 /* Segment size negotiating */ 167 #define TCPOPT_WINDOW 3 /* Window scaling */ 168 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 169 #define TCPOPT_SACK 5 /* SACK Block */ 170 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 171 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 172 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */ 173 174 /* 175 * TCP option lengths 176 */ 177 178 #define TCPOLEN_MSS 4 179 #define TCPOLEN_WINDOW 3 180 #define TCPOLEN_SACK_PERM 2 181 #define TCPOLEN_TIMESTAMP 10 182 #define TCPOLEN_MD5SIG 18 183 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */ 184 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */ 185 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN) 186 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX) 187 188 /* But this is what stacks really send out. */ 189 #define TCPOLEN_TSTAMP_ALIGNED 12 190 #define TCPOLEN_WSCALE_ALIGNED 4 191 #define TCPOLEN_SACKPERM_ALIGNED 4 192 #define TCPOLEN_SACK_BASE 2 193 #define TCPOLEN_SACK_BASE_ALIGNED 4 194 #define TCPOLEN_SACK_PERBLOCK 8 195 #define TCPOLEN_MD5SIG_ALIGNED 20 196 #define TCPOLEN_MSS_ALIGNED 4 197 198 /* Flags in tp->nonagle */ 199 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 200 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 201 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 202 203 /* TCP thin-stream limits */ 204 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 205 206 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */ 207 #define TCP_INIT_CWND 10 208 209 extern struct inet_timewait_death_row tcp_death_row; 210 211 /* sysctl variables for tcp */ 212 extern int sysctl_tcp_timestamps; 213 extern int sysctl_tcp_window_scaling; 214 extern int sysctl_tcp_sack; 215 extern int sysctl_tcp_fin_timeout; 216 extern int sysctl_tcp_keepalive_time; 217 extern int sysctl_tcp_keepalive_probes; 218 extern int sysctl_tcp_keepalive_intvl; 219 extern int sysctl_tcp_syn_retries; 220 extern int sysctl_tcp_synack_retries; 221 extern int sysctl_tcp_retries1; 222 extern int sysctl_tcp_retries2; 223 extern int sysctl_tcp_orphan_retries; 224 extern int sysctl_tcp_syncookies; 225 extern int sysctl_tcp_retrans_collapse; 226 extern int sysctl_tcp_stdurg; 227 extern int sysctl_tcp_rfc1337; 228 extern int sysctl_tcp_abort_on_overflow; 229 extern int sysctl_tcp_max_orphans; 230 extern int sysctl_tcp_fack; 231 extern int sysctl_tcp_reordering; 232 extern int sysctl_tcp_ecn; 233 extern int sysctl_tcp_dsack; 234 extern int sysctl_tcp_wmem[3]; 235 extern int sysctl_tcp_rmem[3]; 236 extern int sysctl_tcp_app_win; 237 extern int sysctl_tcp_adv_win_scale; 238 extern int sysctl_tcp_tw_reuse; 239 extern int sysctl_tcp_frto; 240 extern int sysctl_tcp_frto_response; 241 extern int sysctl_tcp_low_latency; 242 extern int sysctl_tcp_dma_copybreak; 243 extern int sysctl_tcp_nometrics_save; 244 extern int sysctl_tcp_moderate_rcvbuf; 245 extern int sysctl_tcp_tso_win_divisor; 246 extern int sysctl_tcp_abc; 247 extern int sysctl_tcp_mtu_probing; 248 extern int sysctl_tcp_base_mss; 249 extern int sysctl_tcp_workaround_signed_windows; 250 extern int sysctl_tcp_slow_start_after_idle; 251 extern int sysctl_tcp_max_ssthresh; 252 extern int sysctl_tcp_cookie_size; 253 extern int sysctl_tcp_thin_linear_timeouts; 254 extern int sysctl_tcp_thin_dupack; 255 extern int sysctl_tcp_early_retrans; 256 extern int sysctl_tcp_limit_output_bytes; 257 258 extern atomic_long_t tcp_memory_allocated; 259 extern struct percpu_counter tcp_sockets_allocated; 260 extern int tcp_memory_pressure; 261 262 /* 263 * The next routines deal with comparing 32 bit unsigned ints 264 * and worry about wraparound (automatic with unsigned arithmetic). 265 */ 266 267 static inline bool before(__u32 seq1, __u32 seq2) 268 { 269 return (__s32)(seq1-seq2) < 0; 270 } 271 #define after(seq2, seq1) before(seq1, seq2) 272 273 /* is s2<=s1<=s3 ? */ 274 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 275 { 276 return seq3 - seq2 >= seq1 - seq2; 277 } 278 279 static inline bool tcp_out_of_memory(struct sock *sk) 280 { 281 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 282 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 283 return true; 284 return false; 285 } 286 287 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 288 { 289 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 290 int orphans = percpu_counter_read_positive(ocp); 291 292 if (orphans << shift > sysctl_tcp_max_orphans) { 293 orphans = percpu_counter_sum_positive(ocp); 294 if (orphans << shift > sysctl_tcp_max_orphans) 295 return true; 296 } 297 return false; 298 } 299 300 extern bool tcp_check_oom(struct sock *sk, int shift); 301 302 /* syncookies: remember time of last synqueue overflow */ 303 static inline void tcp_synq_overflow(struct sock *sk) 304 { 305 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies; 306 } 307 308 /* syncookies: no recent synqueue overflow on this listening socket? */ 309 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 310 { 311 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 312 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK); 313 } 314 315 extern struct proto tcp_prot; 316 317 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 318 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 319 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 320 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 321 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 322 323 extern void tcp_init_mem(struct net *net); 324 325 extern void tcp_tasklet_init(void); 326 327 extern void tcp_v4_err(struct sk_buff *skb, u32); 328 329 extern void tcp_shutdown (struct sock *sk, int how); 330 331 extern void tcp_v4_early_demux(struct sk_buff *skb); 332 extern int tcp_v4_rcv(struct sk_buff *skb); 333 334 extern struct inet_peer *tcp_v4_get_peer(struct sock *sk); 335 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 336 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 337 size_t size); 338 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset, 339 size_t size, int flags); 340 extern void tcp_release_cb(struct sock *sk); 341 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 342 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 343 const struct tcphdr *th, unsigned int len); 344 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 345 const struct tcphdr *th, unsigned int len); 346 extern void tcp_rcv_space_adjust(struct sock *sk); 347 extern void tcp_cleanup_rbuf(struct sock *sk, int copied); 348 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 349 extern void tcp_twsk_destructor(struct sock *sk); 350 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 351 struct pipe_inode_info *pipe, size_t len, 352 unsigned int flags); 353 354 static inline void tcp_dec_quickack_mode(struct sock *sk, 355 const unsigned int pkts) 356 { 357 struct inet_connection_sock *icsk = inet_csk(sk); 358 359 if (icsk->icsk_ack.quick) { 360 if (pkts >= icsk->icsk_ack.quick) { 361 icsk->icsk_ack.quick = 0; 362 /* Leaving quickack mode we deflate ATO. */ 363 icsk->icsk_ack.ato = TCP_ATO_MIN; 364 } else 365 icsk->icsk_ack.quick -= pkts; 366 } 367 } 368 369 #define TCP_ECN_OK 1 370 #define TCP_ECN_QUEUE_CWR 2 371 #define TCP_ECN_DEMAND_CWR 4 372 #define TCP_ECN_SEEN 8 373 374 enum tcp_tw_status { 375 TCP_TW_SUCCESS = 0, 376 TCP_TW_RST = 1, 377 TCP_TW_ACK = 2, 378 TCP_TW_SYN = 3 379 }; 380 381 382 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 383 struct sk_buff *skb, 384 const struct tcphdr *th); 385 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, 386 struct request_sock *req, 387 struct request_sock **prev); 388 extern int tcp_child_process(struct sock *parent, struct sock *child, 389 struct sk_buff *skb); 390 extern bool tcp_use_frto(struct sock *sk); 391 extern void tcp_enter_frto(struct sock *sk); 392 extern void tcp_enter_loss(struct sock *sk, int how); 393 extern void tcp_clear_retrans(struct tcp_sock *tp); 394 extern void tcp_update_metrics(struct sock *sk); 395 extern void tcp_init_metrics(struct sock *sk); 396 extern void tcp_metrics_init(void); 397 extern bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, bool paws_check); 398 extern bool tcp_remember_stamp(struct sock *sk); 399 extern bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); 400 extern void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); 401 extern void tcp_disable_fack(struct tcp_sock *tp); 402 extern void tcp_close(struct sock *sk, long timeout); 403 extern void tcp_init_sock(struct sock *sk); 404 extern unsigned int tcp_poll(struct file * file, struct socket *sock, 405 struct poll_table_struct *wait); 406 extern int tcp_getsockopt(struct sock *sk, int level, int optname, 407 char __user *optval, int __user *optlen); 408 extern int tcp_setsockopt(struct sock *sk, int level, int optname, 409 char __user *optval, unsigned int optlen); 410 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 411 char __user *optval, int __user *optlen); 412 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 413 char __user *optval, unsigned int optlen); 414 extern void tcp_set_keepalive(struct sock *sk, int val); 415 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req); 416 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 417 size_t len, int nonblock, int flags, int *addr_len); 418 extern void tcp_parse_options(const struct sk_buff *skb, 419 struct tcp_options_received *opt_rx, const u8 **hvpp, 420 int estab); 421 extern const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 422 423 /* 424 * TCP v4 functions exported for the inet6 API 425 */ 426 427 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 428 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 429 extern struct sock * tcp_create_openreq_child(struct sock *sk, 430 struct request_sock *req, 431 struct sk_buff *skb); 432 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 433 struct request_sock *req, 434 struct dst_entry *dst); 435 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 436 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, 437 int addr_len); 438 extern int tcp_connect(struct sock *sk); 439 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst, 440 struct request_sock *req, 441 struct request_values *rvp); 442 extern int tcp_disconnect(struct sock *sk, int flags); 443 444 void tcp_connect_init(struct sock *sk); 445 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 446 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 447 448 /* From syncookies.c */ 449 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS]; 450 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 451 struct ip_options *opt); 452 #ifdef CONFIG_SYN_COOKIES 453 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, 454 __u16 *mss); 455 #else 456 static inline __u32 cookie_v4_init_sequence(struct sock *sk, 457 struct sk_buff *skb, 458 __u16 *mss) 459 { 460 return 0; 461 } 462 #endif 463 464 extern __u32 cookie_init_timestamp(struct request_sock *req); 465 extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *); 466 467 /* From net/ipv6/syncookies.c */ 468 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 469 #ifdef CONFIG_SYN_COOKIES 470 extern __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb, 471 __u16 *mss); 472 #else 473 static inline __u32 cookie_v6_init_sequence(struct sock *sk, 474 struct sk_buff *skb, 475 __u16 *mss) 476 { 477 return 0; 478 } 479 #endif 480 /* tcp_output.c */ 481 482 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 483 int nonagle); 484 extern bool tcp_may_send_now(struct sock *sk); 485 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); 486 extern void tcp_retransmit_timer(struct sock *sk); 487 extern void tcp_xmit_retransmit_queue(struct sock *); 488 extern void tcp_simple_retransmit(struct sock *); 489 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); 490 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int); 491 492 extern void tcp_send_probe0(struct sock *); 493 extern void tcp_send_partial(struct sock *); 494 extern int tcp_write_wakeup(struct sock *); 495 extern void tcp_send_fin(struct sock *sk); 496 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority); 497 extern int tcp_send_synack(struct sock *); 498 extern bool tcp_syn_flood_action(struct sock *sk, 499 const struct sk_buff *skb, 500 const char *proto); 501 extern void tcp_push_one(struct sock *, unsigned int mss_now); 502 extern void tcp_send_ack(struct sock *sk); 503 extern void tcp_send_delayed_ack(struct sock *sk); 504 505 /* tcp_input.c */ 506 extern void tcp_cwnd_application_limited(struct sock *sk); 507 extern void tcp_resume_early_retransmit(struct sock *sk); 508 extern void tcp_rearm_rto(struct sock *sk); 509 510 /* tcp_timer.c */ 511 extern void tcp_init_xmit_timers(struct sock *); 512 static inline void tcp_clear_xmit_timers(struct sock *sk) 513 { 514 inet_csk_clear_xmit_timers(sk); 515 } 516 517 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 518 extern unsigned int tcp_current_mss(struct sock *sk); 519 520 /* Bound MSS / TSO packet size with the half of the window */ 521 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 522 { 523 int cutoff; 524 525 /* When peer uses tiny windows, there is no use in packetizing 526 * to sub-MSS pieces for the sake of SWS or making sure there 527 * are enough packets in the pipe for fast recovery. 528 * 529 * On the other hand, for extremely large MSS devices, handling 530 * smaller than MSS windows in this way does make sense. 531 */ 532 if (tp->max_window >= 512) 533 cutoff = (tp->max_window >> 1); 534 else 535 cutoff = tp->max_window; 536 537 if (cutoff && pktsize > cutoff) 538 return max_t(int, cutoff, 68U - tp->tcp_header_len); 539 else 540 return pktsize; 541 } 542 543 /* tcp.c */ 544 extern void tcp_get_info(const struct sock *, struct tcp_info *); 545 546 /* Read 'sendfile()'-style from a TCP socket */ 547 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 548 unsigned int, size_t); 549 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 550 sk_read_actor_t recv_actor); 551 552 extern void tcp_initialize_rcv_mss(struct sock *sk); 553 554 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu); 555 extern int tcp_mss_to_mtu(struct sock *sk, int mss); 556 extern void tcp_mtup_init(struct sock *sk); 557 extern void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt); 558 559 static inline void tcp_bound_rto(const struct sock *sk) 560 { 561 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 562 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 563 } 564 565 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 566 { 567 return (tp->srtt >> 3) + tp->rttvar; 568 } 569 570 extern void tcp_set_rto(struct sock *sk); 571 572 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 573 { 574 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 575 ntohl(TCP_FLAG_ACK) | 576 snd_wnd); 577 } 578 579 static inline void tcp_fast_path_on(struct tcp_sock *tp) 580 { 581 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 582 } 583 584 static inline void tcp_fast_path_check(struct sock *sk) 585 { 586 struct tcp_sock *tp = tcp_sk(sk); 587 588 if (skb_queue_empty(&tp->out_of_order_queue) && 589 tp->rcv_wnd && 590 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 591 !tp->urg_data) 592 tcp_fast_path_on(tp); 593 } 594 595 /* Compute the actual rto_min value */ 596 static inline u32 tcp_rto_min(struct sock *sk) 597 { 598 const struct dst_entry *dst = __sk_dst_get(sk); 599 u32 rto_min = TCP_RTO_MIN; 600 601 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 602 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 603 return rto_min; 604 } 605 606 /* Compute the actual receive window we are currently advertising. 607 * Rcv_nxt can be after the window if our peer push more data 608 * than the offered window. 609 */ 610 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 611 { 612 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 613 614 if (win < 0) 615 win = 0; 616 return (u32) win; 617 } 618 619 /* Choose a new window, without checks for shrinking, and without 620 * scaling applied to the result. The caller does these things 621 * if necessary. This is a "raw" window selection. 622 */ 623 extern u32 __tcp_select_window(struct sock *sk); 624 625 void tcp_send_window_probe(struct sock *sk); 626 627 /* TCP timestamps are only 32-bits, this causes a slight 628 * complication on 64-bit systems since we store a snapshot 629 * of jiffies in the buffer control blocks below. We decided 630 * to use only the low 32-bits of jiffies and hide the ugly 631 * casts with the following macro. 632 */ 633 #define tcp_time_stamp ((__u32)(jiffies)) 634 635 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 636 637 #define TCPHDR_FIN 0x01 638 #define TCPHDR_SYN 0x02 639 #define TCPHDR_RST 0x04 640 #define TCPHDR_PSH 0x08 641 #define TCPHDR_ACK 0x10 642 #define TCPHDR_URG 0x20 643 #define TCPHDR_ECE 0x40 644 #define TCPHDR_CWR 0x80 645 646 /* This is what the send packet queuing engine uses to pass 647 * TCP per-packet control information to the transmission code. 648 * We also store the host-order sequence numbers in here too. 649 * This is 44 bytes if IPV6 is enabled. 650 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 651 */ 652 struct tcp_skb_cb { 653 union { 654 struct inet_skb_parm h4; 655 #if IS_ENABLED(CONFIG_IPV6) 656 struct inet6_skb_parm h6; 657 #endif 658 } header; /* For incoming frames */ 659 __u32 seq; /* Starting sequence number */ 660 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 661 __u32 when; /* used to compute rtt's */ 662 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 663 664 __u8 sacked; /* State flags for SACK/FACK. */ 665 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 666 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 667 #define TCPCB_LOST 0x04 /* SKB is lost */ 668 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 669 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 670 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) 671 672 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 673 /* 1 byte hole */ 674 __u32 ack_seq; /* Sequence number ACK'd */ 675 }; 676 677 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 678 679 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set 680 * 681 * If we receive a SYN packet with these bits set, it means a network is 682 * playing bad games with TOS bits. In order to avoid possible false congestion 683 * notifications, we disable TCP ECN negociation. 684 */ 685 static inline void 686 TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb) 687 { 688 const struct tcphdr *th = tcp_hdr(skb); 689 690 if (sysctl_tcp_ecn && th->ece && th->cwr && 691 INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield)) 692 inet_rsk(req)->ecn_ok = 1; 693 } 694 695 /* Due to TSO, an SKB can be composed of multiple actual 696 * packets. To keep these tracked properly, we use this. 697 */ 698 static inline int tcp_skb_pcount(const struct sk_buff *skb) 699 { 700 return skb_shinfo(skb)->gso_segs; 701 } 702 703 /* This is valid iff tcp_skb_pcount() > 1. */ 704 static inline int tcp_skb_mss(const struct sk_buff *skb) 705 { 706 return skb_shinfo(skb)->gso_size; 707 } 708 709 /* Events passed to congestion control interface */ 710 enum tcp_ca_event { 711 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 712 CA_EVENT_CWND_RESTART, /* congestion window restart */ 713 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 714 CA_EVENT_FRTO, /* fast recovery timeout */ 715 CA_EVENT_LOSS, /* loss timeout */ 716 CA_EVENT_FAST_ACK, /* in sequence ack */ 717 CA_EVENT_SLOW_ACK, /* other ack */ 718 }; 719 720 /* 721 * Interface for adding new TCP congestion control handlers 722 */ 723 #define TCP_CA_NAME_MAX 16 724 #define TCP_CA_MAX 128 725 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 726 727 #define TCP_CONG_NON_RESTRICTED 0x1 728 #define TCP_CONG_RTT_STAMP 0x2 729 730 struct tcp_congestion_ops { 731 struct list_head list; 732 unsigned long flags; 733 734 /* initialize private data (optional) */ 735 void (*init)(struct sock *sk); 736 /* cleanup private data (optional) */ 737 void (*release)(struct sock *sk); 738 739 /* return slow start threshold (required) */ 740 u32 (*ssthresh)(struct sock *sk); 741 /* lower bound for congestion window (optional) */ 742 u32 (*min_cwnd)(const struct sock *sk); 743 /* do new cwnd calculation (required) */ 744 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight); 745 /* call before changing ca_state (optional) */ 746 void (*set_state)(struct sock *sk, u8 new_state); 747 /* call when cwnd event occurs (optional) */ 748 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 749 /* new value of cwnd after loss (optional) */ 750 u32 (*undo_cwnd)(struct sock *sk); 751 /* hook for packet ack accounting (optional) */ 752 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 753 /* get info for inet_diag (optional) */ 754 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 755 756 char name[TCP_CA_NAME_MAX]; 757 struct module *owner; 758 }; 759 760 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); 761 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 762 763 extern void tcp_init_congestion_control(struct sock *sk); 764 extern void tcp_cleanup_congestion_control(struct sock *sk); 765 extern int tcp_set_default_congestion_control(const char *name); 766 extern void tcp_get_default_congestion_control(char *name); 767 extern void tcp_get_available_congestion_control(char *buf, size_t len); 768 extern void tcp_get_allowed_congestion_control(char *buf, size_t len); 769 extern int tcp_set_allowed_congestion_control(char *allowed); 770 extern int tcp_set_congestion_control(struct sock *sk, const char *name); 771 extern void tcp_slow_start(struct tcp_sock *tp); 772 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w); 773 774 extern struct tcp_congestion_ops tcp_init_congestion_ops; 775 extern u32 tcp_reno_ssthresh(struct sock *sk); 776 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight); 777 extern u32 tcp_reno_min_cwnd(const struct sock *sk); 778 extern struct tcp_congestion_ops tcp_reno; 779 780 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 781 { 782 struct inet_connection_sock *icsk = inet_csk(sk); 783 784 if (icsk->icsk_ca_ops->set_state) 785 icsk->icsk_ca_ops->set_state(sk, ca_state); 786 icsk->icsk_ca_state = ca_state; 787 } 788 789 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 790 { 791 const struct inet_connection_sock *icsk = inet_csk(sk); 792 793 if (icsk->icsk_ca_ops->cwnd_event) 794 icsk->icsk_ca_ops->cwnd_event(sk, event); 795 } 796 797 /* These functions determine how the current flow behaves in respect of SACK 798 * handling. SACK is negotiated with the peer, and therefore it can vary 799 * between different flows. 800 * 801 * tcp_is_sack - SACK enabled 802 * tcp_is_reno - No SACK 803 * tcp_is_fack - FACK enabled, implies SACK enabled 804 */ 805 static inline int tcp_is_sack(const struct tcp_sock *tp) 806 { 807 return tp->rx_opt.sack_ok; 808 } 809 810 static inline bool tcp_is_reno(const struct tcp_sock *tp) 811 { 812 return !tcp_is_sack(tp); 813 } 814 815 static inline bool tcp_is_fack(const struct tcp_sock *tp) 816 { 817 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 818 } 819 820 static inline void tcp_enable_fack(struct tcp_sock *tp) 821 { 822 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 823 } 824 825 /* TCP early-retransmit (ER) is similar to but more conservative than 826 * the thin-dupack feature. Enable ER only if thin-dupack is disabled. 827 */ 828 static inline void tcp_enable_early_retrans(struct tcp_sock *tp) 829 { 830 tp->do_early_retrans = sysctl_tcp_early_retrans && 831 !sysctl_tcp_thin_dupack && sysctl_tcp_reordering == 3; 832 tp->early_retrans_delayed = 0; 833 } 834 835 static inline void tcp_disable_early_retrans(struct tcp_sock *tp) 836 { 837 tp->do_early_retrans = 0; 838 } 839 840 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 841 { 842 return tp->sacked_out + tp->lost_out; 843 } 844 845 /* This determines how many packets are "in the network" to the best 846 * of our knowledge. In many cases it is conservative, but where 847 * detailed information is available from the receiver (via SACK 848 * blocks etc.) we can make more aggressive calculations. 849 * 850 * Use this for decisions involving congestion control, use just 851 * tp->packets_out to determine if the send queue is empty or not. 852 * 853 * Read this equation as: 854 * 855 * "Packets sent once on transmission queue" MINUS 856 * "Packets left network, but not honestly ACKed yet" PLUS 857 * "Packets fast retransmitted" 858 */ 859 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 860 { 861 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 862 } 863 864 #define TCP_INFINITE_SSTHRESH 0x7fffffff 865 866 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 867 { 868 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 869 } 870 871 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 872 * The exception is rate halving phase, when cwnd is decreasing towards 873 * ssthresh. 874 */ 875 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 876 { 877 const struct tcp_sock *tp = tcp_sk(sk); 878 879 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery)) 880 return tp->snd_ssthresh; 881 else 882 return max(tp->snd_ssthresh, 883 ((tp->snd_cwnd >> 1) + 884 (tp->snd_cwnd >> 2))); 885 } 886 887 /* Use define here intentionally to get WARN_ON location shown at the caller */ 888 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 889 890 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh); 891 extern __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 892 893 /* The maximum number of MSS of available cwnd for which TSO defers 894 * sending if not using sysctl_tcp_tso_win_divisor. 895 */ 896 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 897 { 898 return 3; 899 } 900 901 /* Slow start with delack produces 3 packets of burst, so that 902 * it is safe "de facto". This will be the default - same as 903 * the default reordering threshold - but if reordering increases, 904 * we must be able to allow cwnd to burst at least this much in order 905 * to not pull it back when holes are filled. 906 */ 907 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 908 { 909 return tp->reordering; 910 } 911 912 /* Returns end sequence number of the receiver's advertised window */ 913 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 914 { 915 return tp->snd_una + tp->snd_wnd; 916 } 917 extern bool tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight); 918 919 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss, 920 const struct sk_buff *skb) 921 { 922 if (skb->len < mss) 923 tp->snd_sml = TCP_SKB_CB(skb)->end_seq; 924 } 925 926 static inline void tcp_check_probe_timer(struct sock *sk) 927 { 928 const struct tcp_sock *tp = tcp_sk(sk); 929 const struct inet_connection_sock *icsk = inet_csk(sk); 930 931 if (!tp->packets_out && !icsk->icsk_pending) 932 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 933 icsk->icsk_rto, TCP_RTO_MAX); 934 } 935 936 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 937 { 938 tp->snd_wl1 = seq; 939 } 940 941 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 942 { 943 tp->snd_wl1 = seq; 944 } 945 946 /* 947 * Calculate(/check) TCP checksum 948 */ 949 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 950 __be32 daddr, __wsum base) 951 { 952 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 953 } 954 955 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 956 { 957 return __skb_checksum_complete(skb); 958 } 959 960 static inline bool tcp_checksum_complete(struct sk_buff *skb) 961 { 962 return !skb_csum_unnecessary(skb) && 963 __tcp_checksum_complete(skb); 964 } 965 966 /* Prequeue for VJ style copy to user, combined with checksumming. */ 967 968 static inline void tcp_prequeue_init(struct tcp_sock *tp) 969 { 970 tp->ucopy.task = NULL; 971 tp->ucopy.len = 0; 972 tp->ucopy.memory = 0; 973 skb_queue_head_init(&tp->ucopy.prequeue); 974 #ifdef CONFIG_NET_DMA 975 tp->ucopy.dma_chan = NULL; 976 tp->ucopy.wakeup = 0; 977 tp->ucopy.pinned_list = NULL; 978 tp->ucopy.dma_cookie = 0; 979 #endif 980 } 981 982 /* Packet is added to VJ-style prequeue for processing in process 983 * context, if a reader task is waiting. Apparently, this exciting 984 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) 985 * failed somewhere. Latency? Burstiness? Well, at least now we will 986 * see, why it failed. 8)8) --ANK 987 * 988 * NOTE: is this not too big to inline? 989 */ 990 static inline bool tcp_prequeue(struct sock *sk, struct sk_buff *skb) 991 { 992 struct tcp_sock *tp = tcp_sk(sk); 993 994 if (sysctl_tcp_low_latency || !tp->ucopy.task) 995 return false; 996 997 __skb_queue_tail(&tp->ucopy.prequeue, skb); 998 tp->ucopy.memory += skb->truesize; 999 if (tp->ucopy.memory > sk->sk_rcvbuf) { 1000 struct sk_buff *skb1; 1001 1002 BUG_ON(sock_owned_by_user(sk)); 1003 1004 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { 1005 sk_backlog_rcv(sk, skb1); 1006 NET_INC_STATS_BH(sock_net(sk), 1007 LINUX_MIB_TCPPREQUEUEDROPPED); 1008 } 1009 1010 tp->ucopy.memory = 0; 1011 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { 1012 wake_up_interruptible_sync_poll(sk_sleep(sk), 1013 POLLIN | POLLRDNORM | POLLRDBAND); 1014 if (!inet_csk_ack_scheduled(sk)) 1015 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, 1016 (3 * tcp_rto_min(sk)) / 4, 1017 TCP_RTO_MAX); 1018 } 1019 return true; 1020 } 1021 1022 1023 #undef STATE_TRACE 1024 1025 #ifdef STATE_TRACE 1026 static const char *statename[]={ 1027 "Unused","Established","Syn Sent","Syn Recv", 1028 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1029 "Close Wait","Last ACK","Listen","Closing" 1030 }; 1031 #endif 1032 extern void tcp_set_state(struct sock *sk, int state); 1033 1034 extern void tcp_done(struct sock *sk); 1035 1036 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1037 { 1038 rx_opt->dsack = 0; 1039 rx_opt->num_sacks = 0; 1040 } 1041 1042 /* Determine a window scaling and initial window to offer. */ 1043 extern void tcp_select_initial_window(int __space, __u32 mss, 1044 __u32 *rcv_wnd, __u32 *window_clamp, 1045 int wscale_ok, __u8 *rcv_wscale, 1046 __u32 init_rcv_wnd); 1047 1048 static inline int tcp_win_from_space(int space) 1049 { 1050 return sysctl_tcp_adv_win_scale<=0 ? 1051 (space>>(-sysctl_tcp_adv_win_scale)) : 1052 space - (space>>sysctl_tcp_adv_win_scale); 1053 } 1054 1055 /* Note: caller must be prepared to deal with negative returns */ 1056 static inline int tcp_space(const struct sock *sk) 1057 { 1058 return tcp_win_from_space(sk->sk_rcvbuf - 1059 atomic_read(&sk->sk_rmem_alloc)); 1060 } 1061 1062 static inline int tcp_full_space(const struct sock *sk) 1063 { 1064 return tcp_win_from_space(sk->sk_rcvbuf); 1065 } 1066 1067 static inline void tcp_openreq_init(struct request_sock *req, 1068 struct tcp_options_received *rx_opt, 1069 struct sk_buff *skb) 1070 { 1071 struct inet_request_sock *ireq = inet_rsk(req); 1072 1073 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ 1074 req->cookie_ts = 0; 1075 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; 1076 req->mss = rx_opt->mss_clamp; 1077 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; 1078 ireq->tstamp_ok = rx_opt->tstamp_ok; 1079 ireq->sack_ok = rx_opt->sack_ok; 1080 ireq->snd_wscale = rx_opt->snd_wscale; 1081 ireq->wscale_ok = rx_opt->wscale_ok; 1082 ireq->acked = 0; 1083 ireq->ecn_ok = 0; 1084 ireq->rmt_port = tcp_hdr(skb)->source; 1085 ireq->loc_port = tcp_hdr(skb)->dest; 1086 } 1087 1088 extern void tcp_enter_memory_pressure(struct sock *sk); 1089 1090 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1091 { 1092 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1093 } 1094 1095 static inline int keepalive_time_when(const struct tcp_sock *tp) 1096 { 1097 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1098 } 1099 1100 static inline int keepalive_probes(const struct tcp_sock *tp) 1101 { 1102 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1103 } 1104 1105 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1106 { 1107 const struct inet_connection_sock *icsk = &tp->inet_conn; 1108 1109 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1110 tcp_time_stamp - tp->rcv_tstamp); 1111 } 1112 1113 static inline int tcp_fin_time(const struct sock *sk) 1114 { 1115 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1116 const int rto = inet_csk(sk)->icsk_rto; 1117 1118 if (fin_timeout < (rto << 2) - (rto >> 1)) 1119 fin_timeout = (rto << 2) - (rto >> 1); 1120 1121 return fin_timeout; 1122 } 1123 1124 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1125 int paws_win) 1126 { 1127 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1128 return true; 1129 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1130 return true; 1131 /* 1132 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1133 * then following tcp messages have valid values. Ignore 0 value, 1134 * or else 'negative' tsval might forbid us to accept their packets. 1135 */ 1136 if (!rx_opt->ts_recent) 1137 return true; 1138 return false; 1139 } 1140 1141 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1142 int rst) 1143 { 1144 if (tcp_paws_check(rx_opt, 0)) 1145 return false; 1146 1147 /* RST segments are not recommended to carry timestamp, 1148 and, if they do, it is recommended to ignore PAWS because 1149 "their cleanup function should take precedence over timestamps." 1150 Certainly, it is mistake. It is necessary to understand the reasons 1151 of this constraint to relax it: if peer reboots, clock may go 1152 out-of-sync and half-open connections will not be reset. 1153 Actually, the problem would be not existing if all 1154 the implementations followed draft about maintaining clock 1155 via reboots. Linux-2.2 DOES NOT! 1156 1157 However, we can relax time bounds for RST segments to MSL. 1158 */ 1159 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1160 return false; 1161 return true; 1162 } 1163 1164 static inline void tcp_mib_init(struct net *net) 1165 { 1166 /* See RFC 2012 */ 1167 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1168 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1169 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1170 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1171 } 1172 1173 /* from STCP */ 1174 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1175 { 1176 tp->lost_skb_hint = NULL; 1177 tp->scoreboard_skb_hint = NULL; 1178 } 1179 1180 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1181 { 1182 tcp_clear_retrans_hints_partial(tp); 1183 tp->retransmit_skb_hint = NULL; 1184 } 1185 1186 /* MD5 Signature */ 1187 struct crypto_hash; 1188 1189 union tcp_md5_addr { 1190 struct in_addr a4; 1191 #if IS_ENABLED(CONFIG_IPV6) 1192 struct in6_addr a6; 1193 #endif 1194 }; 1195 1196 /* - key database */ 1197 struct tcp_md5sig_key { 1198 struct hlist_node node; 1199 u8 keylen; 1200 u8 family; /* AF_INET or AF_INET6 */ 1201 union tcp_md5_addr addr; 1202 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1203 struct rcu_head rcu; 1204 }; 1205 1206 /* - sock block */ 1207 struct tcp_md5sig_info { 1208 struct hlist_head head; 1209 struct rcu_head rcu; 1210 }; 1211 1212 /* - pseudo header */ 1213 struct tcp4_pseudohdr { 1214 __be32 saddr; 1215 __be32 daddr; 1216 __u8 pad; 1217 __u8 protocol; 1218 __be16 len; 1219 }; 1220 1221 struct tcp6_pseudohdr { 1222 struct in6_addr saddr; 1223 struct in6_addr daddr; 1224 __be32 len; 1225 __be32 protocol; /* including padding */ 1226 }; 1227 1228 union tcp_md5sum_block { 1229 struct tcp4_pseudohdr ip4; 1230 #if IS_ENABLED(CONFIG_IPV6) 1231 struct tcp6_pseudohdr ip6; 1232 #endif 1233 }; 1234 1235 /* - pool: digest algorithm, hash description and scratch buffer */ 1236 struct tcp_md5sig_pool { 1237 struct hash_desc md5_desc; 1238 union tcp_md5sum_block md5_blk; 1239 }; 1240 1241 /* - functions */ 1242 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, 1243 const struct sock *sk, 1244 const struct request_sock *req, 1245 const struct sk_buff *skb); 1246 extern int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1247 int family, const u8 *newkey, 1248 u8 newkeylen, gfp_t gfp); 1249 extern int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1250 int family); 1251 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1252 struct sock *addr_sk); 1253 1254 #ifdef CONFIG_TCP_MD5SIG 1255 extern struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1256 const union tcp_md5_addr *addr, int family); 1257 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1258 #else 1259 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1260 const union tcp_md5_addr *addr, 1261 int family) 1262 { 1263 return NULL; 1264 } 1265 #define tcp_twsk_md5_key(twsk) NULL 1266 #endif 1267 1268 extern struct tcp_md5sig_pool __percpu *tcp_alloc_md5sig_pool(struct sock *); 1269 extern void tcp_free_md5sig_pool(void); 1270 1271 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1272 extern void tcp_put_md5sig_pool(void); 1273 1274 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1275 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1276 unsigned int header_len); 1277 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1278 const struct tcp_md5sig_key *key); 1279 1280 /* write queue abstraction */ 1281 static inline void tcp_write_queue_purge(struct sock *sk) 1282 { 1283 struct sk_buff *skb; 1284 1285 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1286 sk_wmem_free_skb(sk, skb); 1287 sk_mem_reclaim(sk); 1288 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1289 } 1290 1291 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1292 { 1293 return skb_peek(&sk->sk_write_queue); 1294 } 1295 1296 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1297 { 1298 return skb_peek_tail(&sk->sk_write_queue); 1299 } 1300 1301 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1302 const struct sk_buff *skb) 1303 { 1304 return skb_queue_next(&sk->sk_write_queue, skb); 1305 } 1306 1307 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1308 const struct sk_buff *skb) 1309 { 1310 return skb_queue_prev(&sk->sk_write_queue, skb); 1311 } 1312 1313 #define tcp_for_write_queue(skb, sk) \ 1314 skb_queue_walk(&(sk)->sk_write_queue, skb) 1315 1316 #define tcp_for_write_queue_from(skb, sk) \ 1317 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1318 1319 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1320 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1321 1322 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1323 { 1324 return sk->sk_send_head; 1325 } 1326 1327 static inline bool tcp_skb_is_last(const struct sock *sk, 1328 const struct sk_buff *skb) 1329 { 1330 return skb_queue_is_last(&sk->sk_write_queue, skb); 1331 } 1332 1333 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1334 { 1335 if (tcp_skb_is_last(sk, skb)) 1336 sk->sk_send_head = NULL; 1337 else 1338 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1339 } 1340 1341 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1342 { 1343 if (sk->sk_send_head == skb_unlinked) 1344 sk->sk_send_head = NULL; 1345 } 1346 1347 static inline void tcp_init_send_head(struct sock *sk) 1348 { 1349 sk->sk_send_head = NULL; 1350 } 1351 1352 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1353 { 1354 __skb_queue_tail(&sk->sk_write_queue, skb); 1355 } 1356 1357 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1358 { 1359 __tcp_add_write_queue_tail(sk, skb); 1360 1361 /* Queue it, remembering where we must start sending. */ 1362 if (sk->sk_send_head == NULL) { 1363 sk->sk_send_head = skb; 1364 1365 if (tcp_sk(sk)->highest_sack == NULL) 1366 tcp_sk(sk)->highest_sack = skb; 1367 } 1368 } 1369 1370 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1371 { 1372 __skb_queue_head(&sk->sk_write_queue, skb); 1373 } 1374 1375 /* Insert buff after skb on the write queue of sk. */ 1376 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1377 struct sk_buff *buff, 1378 struct sock *sk) 1379 { 1380 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1381 } 1382 1383 /* Insert new before skb on the write queue of sk. */ 1384 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1385 struct sk_buff *skb, 1386 struct sock *sk) 1387 { 1388 __skb_queue_before(&sk->sk_write_queue, skb, new); 1389 1390 if (sk->sk_send_head == skb) 1391 sk->sk_send_head = new; 1392 } 1393 1394 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1395 { 1396 __skb_unlink(skb, &sk->sk_write_queue); 1397 } 1398 1399 static inline bool tcp_write_queue_empty(struct sock *sk) 1400 { 1401 return skb_queue_empty(&sk->sk_write_queue); 1402 } 1403 1404 static inline void tcp_push_pending_frames(struct sock *sk) 1405 { 1406 if (tcp_send_head(sk)) { 1407 struct tcp_sock *tp = tcp_sk(sk); 1408 1409 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1410 } 1411 } 1412 1413 /* Start sequence of the skb just after the highest skb with SACKed 1414 * bit, valid only if sacked_out > 0 or when the caller has ensured 1415 * validity by itself. 1416 */ 1417 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1418 { 1419 if (!tp->sacked_out) 1420 return tp->snd_una; 1421 1422 if (tp->highest_sack == NULL) 1423 return tp->snd_nxt; 1424 1425 return TCP_SKB_CB(tp->highest_sack)->seq; 1426 } 1427 1428 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1429 { 1430 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1431 tcp_write_queue_next(sk, skb); 1432 } 1433 1434 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1435 { 1436 return tcp_sk(sk)->highest_sack; 1437 } 1438 1439 static inline void tcp_highest_sack_reset(struct sock *sk) 1440 { 1441 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1442 } 1443 1444 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1445 static inline void tcp_highest_sack_combine(struct sock *sk, 1446 struct sk_buff *old, 1447 struct sk_buff *new) 1448 { 1449 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1450 tcp_sk(sk)->highest_sack = new; 1451 } 1452 1453 /* Determines whether this is a thin stream (which may suffer from 1454 * increased latency). Used to trigger latency-reducing mechanisms. 1455 */ 1456 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1457 { 1458 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1459 } 1460 1461 /* /proc */ 1462 enum tcp_seq_states { 1463 TCP_SEQ_STATE_LISTENING, 1464 TCP_SEQ_STATE_OPENREQ, 1465 TCP_SEQ_STATE_ESTABLISHED, 1466 TCP_SEQ_STATE_TIME_WAIT, 1467 }; 1468 1469 int tcp_seq_open(struct inode *inode, struct file *file); 1470 1471 struct tcp_seq_afinfo { 1472 char *name; 1473 sa_family_t family; 1474 const struct file_operations *seq_fops; 1475 struct seq_operations seq_ops; 1476 }; 1477 1478 struct tcp_iter_state { 1479 struct seq_net_private p; 1480 sa_family_t family; 1481 enum tcp_seq_states state; 1482 struct sock *syn_wait_sk; 1483 int bucket, offset, sbucket, num, uid; 1484 loff_t last_pos; 1485 }; 1486 1487 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1488 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1489 1490 extern struct request_sock_ops tcp_request_sock_ops; 1491 extern struct request_sock_ops tcp6_request_sock_ops; 1492 1493 extern void tcp_v4_destroy_sock(struct sock *sk); 1494 1495 extern int tcp_v4_gso_send_check(struct sk_buff *skb); 1496 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, 1497 netdev_features_t features); 1498 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head, 1499 struct sk_buff *skb); 1500 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head, 1501 struct sk_buff *skb); 1502 extern int tcp_gro_complete(struct sk_buff *skb); 1503 extern int tcp4_gro_complete(struct sk_buff *skb); 1504 1505 #ifdef CONFIG_PROC_FS 1506 extern int tcp4_proc_init(void); 1507 extern void tcp4_proc_exit(void); 1508 #endif 1509 1510 /* TCP af-specific functions */ 1511 struct tcp_sock_af_ops { 1512 #ifdef CONFIG_TCP_MD5SIG 1513 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1514 struct sock *addr_sk); 1515 int (*calc_md5_hash) (char *location, 1516 struct tcp_md5sig_key *md5, 1517 const struct sock *sk, 1518 const struct request_sock *req, 1519 const struct sk_buff *skb); 1520 int (*md5_parse) (struct sock *sk, 1521 char __user *optval, 1522 int optlen); 1523 #endif 1524 }; 1525 1526 struct tcp_request_sock_ops { 1527 #ifdef CONFIG_TCP_MD5SIG 1528 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1529 struct request_sock *req); 1530 int (*calc_md5_hash) (char *location, 1531 struct tcp_md5sig_key *md5, 1532 const struct sock *sk, 1533 const struct request_sock *req, 1534 const struct sk_buff *skb); 1535 #endif 1536 }; 1537 1538 /* Using SHA1 for now, define some constants. 1539 */ 1540 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS) 1541 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4) 1542 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS) 1543 1544 extern int tcp_cookie_generator(u32 *bakery); 1545 1546 /** 1547 * struct tcp_cookie_values - each socket needs extra space for the 1548 * cookies, together with (optional) space for any SYN data. 1549 * 1550 * A tcp_sock contains a pointer to the current value, and this is 1551 * cloned to the tcp_timewait_sock. 1552 * 1553 * @cookie_pair: variable data from the option exchange. 1554 * 1555 * @cookie_desired: user specified tcpct_cookie_desired. Zero 1556 * indicates default (sysctl_tcp_cookie_size). 1557 * After cookie sent, remembers size of cookie. 1558 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX. 1559 * 1560 * @s_data_desired: user specified tcpct_s_data_desired. When the 1561 * constant payload is specified (@s_data_constant), 1562 * holds its length instead. 1563 * Range 0 to TCP_MSS_DESIRED. 1564 * 1565 * @s_data_payload: constant data that is to be included in the 1566 * payload of SYN or SYNACK segments when the 1567 * cookie option is present. 1568 */ 1569 struct tcp_cookie_values { 1570 struct kref kref; 1571 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE]; 1572 u8 cookie_pair_size; 1573 u8 cookie_desired; 1574 u16 s_data_desired:11, 1575 s_data_constant:1, 1576 s_data_in:1, 1577 s_data_out:1, 1578 s_data_unused:2; 1579 u8 s_data_payload[0]; 1580 }; 1581 1582 static inline void tcp_cookie_values_release(struct kref *kref) 1583 { 1584 kfree(container_of(kref, struct tcp_cookie_values, kref)); 1585 } 1586 1587 /* The length of constant payload data. Note that s_data_desired is 1588 * overloaded, depending on s_data_constant: either the length of constant 1589 * data (returned here) or the limit on variable data. 1590 */ 1591 static inline int tcp_s_data_size(const struct tcp_sock *tp) 1592 { 1593 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant) 1594 ? tp->cookie_values->s_data_desired 1595 : 0; 1596 } 1597 1598 /** 1599 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace. 1600 * 1601 * As tcp_request_sock has already been extended in other places, the 1602 * only remaining method is to pass stack values along as function 1603 * parameters. These parameters are not needed after sending SYNACK. 1604 * 1605 * @cookie_bakery: cryptographic secret and message workspace. 1606 * 1607 * @cookie_plus: bytes in authenticator/cookie option, copied from 1608 * struct tcp_options_received (above). 1609 */ 1610 struct tcp_extend_values { 1611 struct request_values rv; 1612 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS]; 1613 u8 cookie_plus:6, 1614 cookie_out_never:1, 1615 cookie_in_always:1; 1616 }; 1617 1618 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp) 1619 { 1620 return (struct tcp_extend_values *)rvp; 1621 } 1622 1623 extern void tcp_v4_init(void); 1624 extern void tcp_init(void); 1625 1626 #endif /* _TCP_H */ 1627