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