1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * Definitions for the TCP module. 7 * 8 * Version: @(#)tcp.h 1.0.5 05/23/93 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * 13 * This program is free software; you can redistribute it and/or 14 * modify it under the terms of the GNU General Public License 15 * as published by the Free Software Foundation; either version 16 * 2 of the License, or (at your option) any later version. 17 */ 18 #ifndef _TCP_H 19 #define _TCP_H 20 21 #define FASTRETRANS_DEBUG 1 22 23 #include <linux/list.h> 24 #include <linux/tcp.h> 25 #include <linux/bug.h> 26 #include <linux/slab.h> 27 #include <linux/cache.h> 28 #include <linux/percpu.h> 29 #include <linux/skbuff.h> 30 #include <linux/crypto.h> 31 #include <linux/cryptohash.h> 32 #include <linux/kref.h> 33 #include <linux/ktime.h> 34 35 #include <net/inet_connection_sock.h> 36 #include <net/inet_timewait_sock.h> 37 #include <net/inet_hashtables.h> 38 #include <net/checksum.h> 39 #include <net/request_sock.h> 40 #include <net/sock.h> 41 #include <net/snmp.h> 42 #include <net/ip.h> 43 #include <net/tcp_states.h> 44 #include <net/inet_ecn.h> 45 #include <net/dst.h> 46 47 #include <linux/seq_file.h> 48 #include <linux/memcontrol.h> 49 50 extern struct inet_hashinfo tcp_hashinfo; 51 52 extern struct percpu_counter tcp_orphan_count; 53 void tcp_time_wait(struct sock *sk, int state, int timeo); 54 55 #define MAX_TCP_HEADER (128 + MAX_HEADER) 56 #define MAX_TCP_OPTION_SPACE 40 57 58 /* 59 * Never offer a window over 32767 without using window scaling. Some 60 * poor stacks do signed 16bit maths! 61 */ 62 #define MAX_TCP_WINDOW 32767U 63 64 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 65 #define TCP_MIN_MSS 88U 66 67 /* The least MTU to use for probing */ 68 #define TCP_BASE_MSS 1024 69 70 /* probing interval, default to 10 minutes as per RFC4821 */ 71 #define TCP_PROBE_INTERVAL 600 72 73 /* Specify interval when tcp mtu probing will stop */ 74 #define TCP_PROBE_THRESHOLD 8 75 76 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 77 #define TCP_FASTRETRANS_THRESH 3 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 6 /* This is how many retries are done 102 * when active opening a connection. 103 * RFC1122 says the minimum retry MUST 104 * be at least 180secs. Nevertheless 105 * this value is corresponding to 106 * 63secs of retransmission with the 107 * current initial RTO. 108 */ 109 110 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 111 * when passive opening a connection. 112 * This is corresponding to 31secs of 113 * retransmission with the current 114 * initial RTO. 115 */ 116 117 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 118 * state, about 60 seconds */ 119 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 120 /* BSD style FIN_WAIT2 deadlock breaker. 121 * It used to be 3min, new value is 60sec, 122 * to combine FIN-WAIT-2 timeout with 123 * TIME-WAIT timer. 124 */ 125 126 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 127 #if HZ >= 100 128 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 129 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 130 #else 131 #define TCP_DELACK_MIN 4U 132 #define TCP_ATO_MIN 4U 133 #endif 134 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 135 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 136 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 137 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 138 * used as a fallback RTO for the 139 * initial data transmission if no 140 * valid RTT sample has been acquired, 141 * most likely due to retrans in 3WHS. 142 */ 143 144 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 145 * for local resources. 146 */ 147 148 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 149 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 150 #define TCP_KEEPALIVE_INTVL (75*HZ) 151 152 #define MAX_TCP_KEEPIDLE 32767 153 #define MAX_TCP_KEEPINTVL 32767 154 #define MAX_TCP_KEEPCNT 127 155 #define MAX_TCP_SYNCNT 127 156 157 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 158 159 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 160 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 161 * after this time. It should be equal 162 * (or greater than) TCP_TIMEWAIT_LEN 163 * to provide reliability equal to one 164 * provided by timewait state. 165 */ 166 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 167 * timestamps. It must be less than 168 * minimal timewait lifetime. 169 */ 170 /* 171 * TCP option 172 */ 173 174 #define TCPOPT_NOP 1 /* Padding */ 175 #define TCPOPT_EOL 0 /* End of options */ 176 #define TCPOPT_MSS 2 /* Segment size negotiating */ 177 #define TCPOPT_WINDOW 3 /* Window scaling */ 178 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 179 #define TCPOPT_SACK 5 /* SACK Block */ 180 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 181 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 182 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 183 #define TCPOPT_EXP 254 /* Experimental */ 184 /* Magic number to be after the option value for sharing TCP 185 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 186 */ 187 #define TCPOPT_FASTOPEN_MAGIC 0xF989 188 189 /* 190 * TCP option lengths 191 */ 192 193 #define TCPOLEN_MSS 4 194 #define TCPOLEN_WINDOW 3 195 #define TCPOLEN_SACK_PERM 2 196 #define TCPOLEN_TIMESTAMP 10 197 #define TCPOLEN_MD5SIG 18 198 #define TCPOLEN_FASTOPEN_BASE 2 199 #define TCPOLEN_EXP_FASTOPEN_BASE 4 200 201 /* But this is what stacks really send out. */ 202 #define TCPOLEN_TSTAMP_ALIGNED 12 203 #define TCPOLEN_WSCALE_ALIGNED 4 204 #define TCPOLEN_SACKPERM_ALIGNED 4 205 #define TCPOLEN_SACK_BASE 2 206 #define TCPOLEN_SACK_BASE_ALIGNED 4 207 #define TCPOLEN_SACK_PERBLOCK 8 208 #define TCPOLEN_MD5SIG_ALIGNED 20 209 #define TCPOLEN_MSS_ALIGNED 4 210 211 /* Flags in tp->nonagle */ 212 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 213 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 214 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 215 216 /* TCP thin-stream limits */ 217 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 218 219 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */ 220 #define TCP_INIT_CWND 10 221 222 /* Bit Flags for sysctl_tcp_fastopen */ 223 #define TFO_CLIENT_ENABLE 1 224 #define TFO_SERVER_ENABLE 2 225 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 226 227 /* Accept SYN data w/o any cookie option */ 228 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 229 230 /* Force enable TFO on all listeners, i.e., not requiring the 231 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen. 232 */ 233 #define TFO_SERVER_WO_SOCKOPT1 0x400 234 #define TFO_SERVER_WO_SOCKOPT2 0x800 235 236 extern struct inet_timewait_death_row tcp_death_row; 237 238 /* sysctl variables for tcp */ 239 extern int sysctl_tcp_timestamps; 240 extern int sysctl_tcp_window_scaling; 241 extern int sysctl_tcp_sack; 242 extern int sysctl_tcp_fin_timeout; 243 extern int sysctl_tcp_keepalive_time; 244 extern int sysctl_tcp_keepalive_probes; 245 extern int sysctl_tcp_keepalive_intvl; 246 extern int sysctl_tcp_syn_retries; 247 extern int sysctl_tcp_synack_retries; 248 extern int sysctl_tcp_retries1; 249 extern int sysctl_tcp_retries2; 250 extern int sysctl_tcp_orphan_retries; 251 extern int sysctl_tcp_syncookies; 252 extern int sysctl_tcp_fastopen; 253 extern int sysctl_tcp_retrans_collapse; 254 extern int sysctl_tcp_stdurg; 255 extern int sysctl_tcp_rfc1337; 256 extern int sysctl_tcp_abort_on_overflow; 257 extern int sysctl_tcp_max_orphans; 258 extern int sysctl_tcp_fack; 259 extern int sysctl_tcp_reordering; 260 extern int sysctl_tcp_max_reordering; 261 extern int sysctl_tcp_dsack; 262 extern long sysctl_tcp_mem[3]; 263 extern int sysctl_tcp_wmem[3]; 264 extern int sysctl_tcp_rmem[3]; 265 extern int sysctl_tcp_app_win; 266 extern int sysctl_tcp_adv_win_scale; 267 extern int sysctl_tcp_tw_reuse; 268 extern int sysctl_tcp_frto; 269 extern int sysctl_tcp_low_latency; 270 extern int sysctl_tcp_nometrics_save; 271 extern int sysctl_tcp_moderate_rcvbuf; 272 extern int sysctl_tcp_tso_win_divisor; 273 extern int sysctl_tcp_workaround_signed_windows; 274 extern int sysctl_tcp_slow_start_after_idle; 275 extern int sysctl_tcp_thin_linear_timeouts; 276 extern int sysctl_tcp_thin_dupack; 277 extern int sysctl_tcp_early_retrans; 278 extern int sysctl_tcp_limit_output_bytes; 279 extern int sysctl_tcp_challenge_ack_limit; 280 extern unsigned int sysctl_tcp_notsent_lowat; 281 extern int sysctl_tcp_min_tso_segs; 282 extern int sysctl_tcp_autocorking; 283 extern int sysctl_tcp_invalid_ratelimit; 284 285 extern atomic_long_t tcp_memory_allocated; 286 extern struct percpu_counter tcp_sockets_allocated; 287 extern int tcp_memory_pressure; 288 289 /* 290 * The next routines deal with comparing 32 bit unsigned ints 291 * and worry about wraparound (automatic with unsigned arithmetic). 292 */ 293 294 static inline bool before(__u32 seq1, __u32 seq2) 295 { 296 return (__s32)(seq1-seq2) < 0; 297 } 298 #define after(seq2, seq1) before(seq1, seq2) 299 300 /* is s2<=s1<=s3 ? */ 301 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 302 { 303 return seq3 - seq2 >= seq1 - seq2; 304 } 305 306 static inline bool tcp_out_of_memory(struct sock *sk) 307 { 308 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 309 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 310 return true; 311 return false; 312 } 313 314 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 315 { 316 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 317 int orphans = percpu_counter_read_positive(ocp); 318 319 if (orphans << shift > sysctl_tcp_max_orphans) { 320 orphans = percpu_counter_sum_positive(ocp); 321 if (orphans << shift > sysctl_tcp_max_orphans) 322 return true; 323 } 324 return false; 325 } 326 327 bool tcp_check_oom(struct sock *sk, int shift); 328 329 /* syncookies: remember time of last synqueue overflow */ 330 static inline void tcp_synq_overflow(struct sock *sk) 331 { 332 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies; 333 } 334 335 /* syncookies: no recent synqueue overflow on this listening socket? */ 336 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 337 { 338 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 339 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK); 340 } 341 342 extern struct proto tcp_prot; 343 344 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 345 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 346 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 347 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 348 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 349 350 void tcp_tasklet_init(void); 351 352 void tcp_v4_err(struct sk_buff *skb, u32); 353 354 void tcp_shutdown(struct sock *sk, int how); 355 356 void tcp_v4_early_demux(struct sk_buff *skb); 357 int tcp_v4_rcv(struct sk_buff *skb); 358 359 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 360 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 361 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 362 int flags); 363 void tcp_release_cb(struct sock *sk); 364 void tcp_wfree(struct sk_buff *skb); 365 void tcp_write_timer_handler(struct sock *sk); 366 void tcp_delack_timer_handler(struct sock *sk); 367 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 368 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 369 const struct tcphdr *th, unsigned int len); 370 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 371 const struct tcphdr *th, unsigned int len); 372 void tcp_rcv_space_adjust(struct sock *sk); 373 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 374 void tcp_twsk_destructor(struct sock *sk); 375 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 376 struct pipe_inode_info *pipe, size_t len, 377 unsigned int flags); 378 379 static inline void tcp_dec_quickack_mode(struct sock *sk, 380 const unsigned int pkts) 381 { 382 struct inet_connection_sock *icsk = inet_csk(sk); 383 384 if (icsk->icsk_ack.quick) { 385 if (pkts >= icsk->icsk_ack.quick) { 386 icsk->icsk_ack.quick = 0; 387 /* Leaving quickack mode we deflate ATO. */ 388 icsk->icsk_ack.ato = TCP_ATO_MIN; 389 } else 390 icsk->icsk_ack.quick -= pkts; 391 } 392 } 393 394 #define TCP_ECN_OK 1 395 #define TCP_ECN_QUEUE_CWR 2 396 #define TCP_ECN_DEMAND_CWR 4 397 #define TCP_ECN_SEEN 8 398 399 enum tcp_tw_status { 400 TCP_TW_SUCCESS = 0, 401 TCP_TW_RST = 1, 402 TCP_TW_ACK = 2, 403 TCP_TW_SYN = 3 404 }; 405 406 407 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 408 struct sk_buff *skb, 409 const struct tcphdr *th); 410 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 411 struct request_sock *req, bool fastopen); 412 int tcp_child_process(struct sock *parent, struct sock *child, 413 struct sk_buff *skb); 414 void tcp_enter_loss(struct sock *sk); 415 void tcp_clear_retrans(struct tcp_sock *tp); 416 void tcp_update_metrics(struct sock *sk); 417 void tcp_init_metrics(struct sock *sk); 418 void tcp_metrics_init(void); 419 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, 420 bool paws_check, bool timestamps); 421 bool tcp_remember_stamp(struct sock *sk); 422 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); 423 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); 424 void tcp_disable_fack(struct tcp_sock *tp); 425 void tcp_close(struct sock *sk, long timeout); 426 void tcp_init_sock(struct sock *sk); 427 unsigned int tcp_poll(struct file *file, struct socket *sock, 428 struct poll_table_struct *wait); 429 int tcp_getsockopt(struct sock *sk, int level, int optname, 430 char __user *optval, int __user *optlen); 431 int tcp_setsockopt(struct sock *sk, int level, int optname, 432 char __user *optval, unsigned int optlen); 433 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 434 char __user *optval, int __user *optlen); 435 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 436 char __user *optval, unsigned int optlen); 437 void tcp_set_keepalive(struct sock *sk, int val); 438 void tcp_syn_ack_timeout(const struct request_sock *req); 439 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 440 int flags, int *addr_len); 441 void tcp_parse_options(const struct sk_buff *skb, 442 struct tcp_options_received *opt_rx, 443 int estab, struct tcp_fastopen_cookie *foc); 444 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 445 446 /* 447 * TCP v4 functions exported for the inet6 API 448 */ 449 450 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 451 void tcp_v4_mtu_reduced(struct sock *sk); 452 void tcp_req_err(struct sock *sk, u32 seq); 453 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 454 struct sock *tcp_create_openreq_child(struct sock *sk, 455 struct request_sock *req, 456 struct sk_buff *skb); 457 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 458 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 459 struct request_sock *req, 460 struct dst_entry *dst); 461 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 462 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 463 int tcp_connect(struct sock *sk); 464 struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst, 465 struct request_sock *req, 466 struct tcp_fastopen_cookie *foc); 467 int tcp_disconnect(struct sock *sk, int flags); 468 469 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 470 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 471 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 472 473 /* From syncookies.c */ 474 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 475 u32 cookie); 476 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 477 #ifdef CONFIG_SYN_COOKIES 478 479 /* Syncookies use a monotonic timer which increments every 60 seconds. 480 * This counter is used both as a hash input and partially encoded into 481 * the cookie value. A cookie is only validated further if the delta 482 * between the current counter value and the encoded one is less than this, 483 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 484 * the counter advances immediately after a cookie is generated). 485 */ 486 #define MAX_SYNCOOKIE_AGE 2 487 488 static inline u32 tcp_cookie_time(void) 489 { 490 u64 val = get_jiffies_64(); 491 492 do_div(val, 60 * HZ); 493 return val; 494 } 495 496 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 497 u16 *mssp); 498 __u32 cookie_v4_init_sequence(struct sock *sk, const struct sk_buff *skb, 499 __u16 *mss); 500 __u32 cookie_init_timestamp(struct request_sock *req); 501 bool cookie_timestamp_decode(struct tcp_options_received *opt); 502 bool cookie_ecn_ok(const struct tcp_options_received *opt, 503 const struct net *net, const struct dst_entry *dst); 504 505 /* From net/ipv6/syncookies.c */ 506 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 507 u32 cookie); 508 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 509 510 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 511 const struct tcphdr *th, u16 *mssp); 512 __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb, 513 __u16 *mss); 514 #endif 515 /* tcp_output.c */ 516 517 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 518 int nonagle); 519 bool tcp_may_send_now(struct sock *sk); 520 int __tcp_retransmit_skb(struct sock *, struct sk_buff *); 521 int tcp_retransmit_skb(struct sock *, struct sk_buff *); 522 void tcp_retransmit_timer(struct sock *sk); 523 void tcp_xmit_retransmit_queue(struct sock *); 524 void tcp_simple_retransmit(struct sock *); 525 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 526 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t); 527 528 void tcp_send_probe0(struct sock *); 529 void tcp_send_partial(struct sock *); 530 int tcp_write_wakeup(struct sock *); 531 void tcp_send_fin(struct sock *sk); 532 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 533 int tcp_send_synack(struct sock *); 534 void tcp_push_one(struct sock *, unsigned int mss_now); 535 void tcp_send_ack(struct sock *sk); 536 void tcp_send_delayed_ack(struct sock *sk); 537 void tcp_send_loss_probe(struct sock *sk); 538 bool tcp_schedule_loss_probe(struct sock *sk); 539 540 /* tcp_input.c */ 541 void tcp_resume_early_retransmit(struct sock *sk); 542 void tcp_rearm_rto(struct sock *sk); 543 void tcp_reset(struct sock *sk); 544 545 /* tcp_timer.c */ 546 void tcp_init_xmit_timers(struct sock *); 547 static inline void tcp_clear_xmit_timers(struct sock *sk) 548 { 549 inet_csk_clear_xmit_timers(sk); 550 } 551 552 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 553 unsigned int tcp_current_mss(struct sock *sk); 554 555 /* Bound MSS / TSO packet size with the half of the window */ 556 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 557 { 558 int cutoff; 559 560 /* When peer uses tiny windows, there is no use in packetizing 561 * to sub-MSS pieces for the sake of SWS or making sure there 562 * are enough packets in the pipe for fast recovery. 563 * 564 * On the other hand, for extremely large MSS devices, handling 565 * smaller than MSS windows in this way does make sense. 566 */ 567 if (tp->max_window >= 512) 568 cutoff = (tp->max_window >> 1); 569 else 570 cutoff = tp->max_window; 571 572 if (cutoff && pktsize > cutoff) 573 return max_t(int, cutoff, 68U - tp->tcp_header_len); 574 else 575 return pktsize; 576 } 577 578 /* tcp.c */ 579 void tcp_get_info(const struct sock *, struct tcp_info *); 580 581 /* Read 'sendfile()'-style from a TCP socket */ 582 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 583 unsigned int, size_t); 584 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 585 sk_read_actor_t recv_actor); 586 587 void tcp_initialize_rcv_mss(struct sock *sk); 588 589 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 590 int tcp_mss_to_mtu(struct sock *sk, int mss); 591 void tcp_mtup_init(struct sock *sk); 592 void tcp_init_buffer_space(struct sock *sk); 593 594 static inline void tcp_bound_rto(const struct sock *sk) 595 { 596 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 597 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 598 } 599 600 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 601 { 602 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 603 } 604 605 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 606 { 607 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 608 ntohl(TCP_FLAG_ACK) | 609 snd_wnd); 610 } 611 612 static inline void tcp_fast_path_on(struct tcp_sock *tp) 613 { 614 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 615 } 616 617 static inline void tcp_fast_path_check(struct sock *sk) 618 { 619 struct tcp_sock *tp = tcp_sk(sk); 620 621 if (skb_queue_empty(&tp->out_of_order_queue) && 622 tp->rcv_wnd && 623 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 624 !tp->urg_data) 625 tcp_fast_path_on(tp); 626 } 627 628 /* Compute the actual rto_min value */ 629 static inline u32 tcp_rto_min(struct sock *sk) 630 { 631 const struct dst_entry *dst = __sk_dst_get(sk); 632 u32 rto_min = TCP_RTO_MIN; 633 634 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 635 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 636 return rto_min; 637 } 638 639 static inline u32 tcp_rto_min_us(struct sock *sk) 640 { 641 return jiffies_to_usecs(tcp_rto_min(sk)); 642 } 643 644 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 645 { 646 return dst_metric_locked(dst, RTAX_CC_ALGO); 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 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 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 679 { 680 return skb->skb_mstamp.stamp_jiffies; 681 } 682 683 684 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 685 686 #define TCPHDR_FIN 0x01 687 #define TCPHDR_SYN 0x02 688 #define TCPHDR_RST 0x04 689 #define TCPHDR_PSH 0x08 690 #define TCPHDR_ACK 0x10 691 #define TCPHDR_URG 0x20 692 #define TCPHDR_ECE 0x40 693 #define TCPHDR_CWR 0x80 694 695 /* This is what the send packet queuing engine uses to pass 696 * TCP per-packet control information to the transmission code. 697 * We also store the host-order sequence numbers in here too. 698 * This is 44 bytes if IPV6 is enabled. 699 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 700 */ 701 struct tcp_skb_cb { 702 __u32 seq; /* Starting sequence number */ 703 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 704 union { 705 /* Note : tcp_tw_isn is used in input path only 706 * (isn chosen by tcp_timewait_state_process()) 707 * 708 * tcp_gso_segs is used in write queue only, 709 * cf tcp_skb_pcount() 710 */ 711 __u32 tcp_tw_isn; 712 __u32 tcp_gso_segs; 713 }; 714 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 715 716 __u8 sacked; /* State flags for SACK/FACK. */ 717 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 718 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 719 #define TCPCB_LOST 0x04 /* SKB is lost */ 720 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 721 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 722 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 723 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 724 TCPCB_REPAIRED) 725 726 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 727 /* 1 byte hole */ 728 __u32 ack_seq; /* Sequence number ACK'd */ 729 union { 730 struct inet_skb_parm h4; 731 #if IS_ENABLED(CONFIG_IPV6) 732 struct inet6_skb_parm h6; 733 #endif 734 } header; /* For incoming frames */ 735 }; 736 737 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 738 739 740 #if IS_ENABLED(CONFIG_IPV6) 741 /* This is the variant of inet6_iif() that must be used by TCP, 742 * as TCP moves IP6CB into a different location in skb->cb[] 743 */ 744 static inline int tcp_v6_iif(const struct sk_buff *skb) 745 { 746 return TCP_SKB_CB(skb)->header.h6.iif; 747 } 748 #endif 749 750 /* Due to TSO, an SKB can be composed of multiple actual 751 * packets. To keep these tracked properly, we use this. 752 */ 753 static inline int tcp_skb_pcount(const struct sk_buff *skb) 754 { 755 return TCP_SKB_CB(skb)->tcp_gso_segs; 756 } 757 758 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 759 { 760 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 761 } 762 763 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 764 { 765 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 766 } 767 768 /* This is valid iff tcp_skb_pcount() > 1. */ 769 static inline int tcp_skb_mss(const struct sk_buff *skb) 770 { 771 return skb_shinfo(skb)->gso_size; 772 } 773 774 /* Events passed to congestion control interface */ 775 enum tcp_ca_event { 776 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 777 CA_EVENT_CWND_RESTART, /* congestion window restart */ 778 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 779 CA_EVENT_LOSS, /* loss timeout */ 780 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 781 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 782 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 783 CA_EVENT_NON_DELAYED_ACK, 784 }; 785 786 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 787 enum tcp_ca_ack_event_flags { 788 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 789 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 790 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 791 }; 792 793 /* 794 * Interface for adding new TCP congestion control handlers 795 */ 796 #define TCP_CA_NAME_MAX 16 797 #define TCP_CA_MAX 128 798 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 799 800 #define TCP_CA_UNSPEC 0 801 802 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 803 #define TCP_CONG_NON_RESTRICTED 0x1 804 /* Requires ECN/ECT set on all packets */ 805 #define TCP_CONG_NEEDS_ECN 0x2 806 807 struct tcp_congestion_ops { 808 struct list_head list; 809 u32 key; 810 u32 flags; 811 812 /* initialize private data (optional) */ 813 void (*init)(struct sock *sk); 814 /* cleanup private data (optional) */ 815 void (*release)(struct sock *sk); 816 817 /* return slow start threshold (required) */ 818 u32 (*ssthresh)(struct sock *sk); 819 /* do new cwnd calculation (required) */ 820 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 821 /* call before changing ca_state (optional) */ 822 void (*set_state)(struct sock *sk, u8 new_state); 823 /* call when cwnd event occurs (optional) */ 824 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 825 /* call when ack arrives (optional) */ 826 void (*in_ack_event)(struct sock *sk, u32 flags); 827 /* new value of cwnd after loss (optional) */ 828 u32 (*undo_cwnd)(struct sock *sk); 829 /* hook for packet ack accounting (optional) */ 830 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 831 /* get info for inet_diag (optional) */ 832 int (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 833 834 char name[TCP_CA_NAME_MAX]; 835 struct module *owner; 836 }; 837 838 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 839 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 840 841 void tcp_assign_congestion_control(struct sock *sk); 842 void tcp_init_congestion_control(struct sock *sk); 843 void tcp_cleanup_congestion_control(struct sock *sk); 844 int tcp_set_default_congestion_control(const char *name); 845 void tcp_get_default_congestion_control(char *name); 846 void tcp_get_available_congestion_control(char *buf, size_t len); 847 void tcp_get_allowed_congestion_control(char *buf, size_t len); 848 int tcp_set_allowed_congestion_control(char *allowed); 849 int tcp_set_congestion_control(struct sock *sk, const char *name); 850 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 851 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 852 853 u32 tcp_reno_ssthresh(struct sock *sk); 854 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 855 extern struct tcp_congestion_ops tcp_reno; 856 857 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 858 u32 tcp_ca_get_key_by_name(const char *name); 859 #ifdef CONFIG_INET 860 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 861 #else 862 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 863 { 864 return NULL; 865 } 866 #endif 867 868 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 869 { 870 const struct inet_connection_sock *icsk = inet_csk(sk); 871 872 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 873 } 874 875 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 876 { 877 struct inet_connection_sock *icsk = inet_csk(sk); 878 879 if (icsk->icsk_ca_ops->set_state) 880 icsk->icsk_ca_ops->set_state(sk, ca_state); 881 icsk->icsk_ca_state = ca_state; 882 } 883 884 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 885 { 886 const struct inet_connection_sock *icsk = inet_csk(sk); 887 888 if (icsk->icsk_ca_ops->cwnd_event) 889 icsk->icsk_ca_ops->cwnd_event(sk, event); 890 } 891 892 /* These functions determine how the current flow behaves in respect of SACK 893 * handling. SACK is negotiated with the peer, and therefore it can vary 894 * between different flows. 895 * 896 * tcp_is_sack - SACK enabled 897 * tcp_is_reno - No SACK 898 * tcp_is_fack - FACK enabled, implies SACK enabled 899 */ 900 static inline int tcp_is_sack(const struct tcp_sock *tp) 901 { 902 return tp->rx_opt.sack_ok; 903 } 904 905 static inline bool tcp_is_reno(const struct tcp_sock *tp) 906 { 907 return !tcp_is_sack(tp); 908 } 909 910 static inline bool tcp_is_fack(const struct tcp_sock *tp) 911 { 912 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 913 } 914 915 static inline void tcp_enable_fack(struct tcp_sock *tp) 916 { 917 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 918 } 919 920 /* TCP early-retransmit (ER) is similar to but more conservative than 921 * the thin-dupack feature. Enable ER only if thin-dupack is disabled. 922 */ 923 static inline void tcp_enable_early_retrans(struct tcp_sock *tp) 924 { 925 tp->do_early_retrans = sysctl_tcp_early_retrans && 926 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack && 927 sysctl_tcp_reordering == 3; 928 } 929 930 static inline void tcp_disable_early_retrans(struct tcp_sock *tp) 931 { 932 tp->do_early_retrans = 0; 933 } 934 935 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 936 { 937 return tp->sacked_out + tp->lost_out; 938 } 939 940 /* This determines how many packets are "in the network" to the best 941 * of our knowledge. In many cases it is conservative, but where 942 * detailed information is available from the receiver (via SACK 943 * blocks etc.) we can make more aggressive calculations. 944 * 945 * Use this for decisions involving congestion control, use just 946 * tp->packets_out to determine if the send queue is empty or not. 947 * 948 * Read this equation as: 949 * 950 * "Packets sent once on transmission queue" MINUS 951 * "Packets left network, but not honestly ACKed yet" PLUS 952 * "Packets fast retransmitted" 953 */ 954 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 955 { 956 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 957 } 958 959 #define TCP_INFINITE_SSTHRESH 0x7fffffff 960 961 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 962 { 963 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 964 } 965 966 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 967 { 968 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 969 (1 << inet_csk(sk)->icsk_ca_state); 970 } 971 972 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 973 * The exception is cwnd reduction phase, when cwnd is decreasing towards 974 * ssthresh. 975 */ 976 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 977 { 978 const struct tcp_sock *tp = tcp_sk(sk); 979 980 if (tcp_in_cwnd_reduction(sk)) 981 return tp->snd_ssthresh; 982 else 983 return max(tp->snd_ssthresh, 984 ((tp->snd_cwnd >> 1) + 985 (tp->snd_cwnd >> 2))); 986 } 987 988 /* Use define here intentionally to get WARN_ON location shown at the caller */ 989 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 990 991 void tcp_enter_cwr(struct sock *sk); 992 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 993 994 /* The maximum number of MSS of available cwnd for which TSO defers 995 * sending if not using sysctl_tcp_tso_win_divisor. 996 */ 997 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 998 { 999 return 3; 1000 } 1001 1002 /* Slow start with delack produces 3 packets of burst, so that 1003 * it is safe "de facto". This will be the default - same as 1004 * the default reordering threshold - but if reordering increases, 1005 * we must be able to allow cwnd to burst at least this much in order 1006 * to not pull it back when holes are filled. 1007 */ 1008 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 1009 { 1010 return tp->reordering; 1011 } 1012 1013 /* Returns end sequence number of the receiver's advertised window */ 1014 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1015 { 1016 return tp->snd_una + tp->snd_wnd; 1017 } 1018 1019 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1020 * flexible approach. The RFC suggests cwnd should not be raised unless 1021 * it was fully used previously. And that's exactly what we do in 1022 * congestion avoidance mode. But in slow start we allow cwnd to grow 1023 * as long as the application has used half the cwnd. 1024 * Example : 1025 * cwnd is 10 (IW10), but application sends 9 frames. 1026 * We allow cwnd to reach 18 when all frames are ACKed. 1027 * This check is safe because it's as aggressive as slow start which already 1028 * risks 100% overshoot. The advantage is that we discourage application to 1029 * either send more filler packets or data to artificially blow up the cwnd 1030 * usage, and allow application-limited process to probe bw more aggressively. 1031 */ 1032 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1033 { 1034 const struct tcp_sock *tp = tcp_sk(sk); 1035 1036 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1037 if (tp->snd_cwnd <= tp->snd_ssthresh) 1038 return tp->snd_cwnd < 2 * tp->max_packets_out; 1039 1040 return tp->is_cwnd_limited; 1041 } 1042 1043 static inline void tcp_check_probe_timer(struct sock *sk) 1044 { 1045 const struct tcp_sock *tp = tcp_sk(sk); 1046 const struct inet_connection_sock *icsk = inet_csk(sk); 1047 1048 if (!tp->packets_out && !icsk->icsk_pending) 1049 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1050 icsk->icsk_rto, TCP_RTO_MAX); 1051 } 1052 1053 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1054 { 1055 tp->snd_wl1 = seq; 1056 } 1057 1058 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1059 { 1060 tp->snd_wl1 = seq; 1061 } 1062 1063 /* 1064 * Calculate(/check) TCP checksum 1065 */ 1066 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1067 __be32 daddr, __wsum base) 1068 { 1069 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1070 } 1071 1072 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1073 { 1074 return __skb_checksum_complete(skb); 1075 } 1076 1077 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1078 { 1079 return !skb_csum_unnecessary(skb) && 1080 __tcp_checksum_complete(skb); 1081 } 1082 1083 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1084 1085 static inline void tcp_prequeue_init(struct tcp_sock *tp) 1086 { 1087 tp->ucopy.task = NULL; 1088 tp->ucopy.len = 0; 1089 tp->ucopy.memory = 0; 1090 skb_queue_head_init(&tp->ucopy.prequeue); 1091 } 1092 1093 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1094 1095 #undef STATE_TRACE 1096 1097 #ifdef STATE_TRACE 1098 static const char *statename[]={ 1099 "Unused","Established","Syn Sent","Syn Recv", 1100 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1101 "Close Wait","Last ACK","Listen","Closing" 1102 }; 1103 #endif 1104 void tcp_set_state(struct sock *sk, int state); 1105 1106 void tcp_done(struct sock *sk); 1107 1108 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1109 { 1110 rx_opt->dsack = 0; 1111 rx_opt->num_sacks = 0; 1112 } 1113 1114 u32 tcp_default_init_rwnd(u32 mss); 1115 1116 /* Determine a window scaling and initial window to offer. */ 1117 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1118 __u32 *window_clamp, int wscale_ok, 1119 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1120 1121 static inline int tcp_win_from_space(int space) 1122 { 1123 return sysctl_tcp_adv_win_scale<=0 ? 1124 (space>>(-sysctl_tcp_adv_win_scale)) : 1125 space - (space>>sysctl_tcp_adv_win_scale); 1126 } 1127 1128 /* Note: caller must be prepared to deal with negative returns */ 1129 static inline int tcp_space(const struct sock *sk) 1130 { 1131 return tcp_win_from_space(sk->sk_rcvbuf - 1132 atomic_read(&sk->sk_rmem_alloc)); 1133 } 1134 1135 static inline int tcp_full_space(const struct sock *sk) 1136 { 1137 return tcp_win_from_space(sk->sk_rcvbuf); 1138 } 1139 1140 extern void tcp_openreq_init_rwin(struct request_sock *req, 1141 struct sock *sk, struct dst_entry *dst); 1142 1143 void tcp_enter_memory_pressure(struct sock *sk); 1144 1145 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1146 { 1147 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 1148 } 1149 1150 static inline int keepalive_time_when(const struct tcp_sock *tp) 1151 { 1152 return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 1153 } 1154 1155 static inline int keepalive_probes(const struct tcp_sock *tp) 1156 { 1157 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1158 } 1159 1160 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1161 { 1162 const struct inet_connection_sock *icsk = &tp->inet_conn; 1163 1164 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1165 tcp_time_stamp - tp->rcv_tstamp); 1166 } 1167 1168 static inline int tcp_fin_time(const struct sock *sk) 1169 { 1170 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1171 const int rto = inet_csk(sk)->icsk_rto; 1172 1173 if (fin_timeout < (rto << 2) - (rto >> 1)) 1174 fin_timeout = (rto << 2) - (rto >> 1); 1175 1176 return fin_timeout; 1177 } 1178 1179 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1180 int paws_win) 1181 { 1182 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1183 return true; 1184 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1185 return true; 1186 /* 1187 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1188 * then following tcp messages have valid values. Ignore 0 value, 1189 * or else 'negative' tsval might forbid us to accept their packets. 1190 */ 1191 if (!rx_opt->ts_recent) 1192 return true; 1193 return false; 1194 } 1195 1196 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1197 int rst) 1198 { 1199 if (tcp_paws_check(rx_opt, 0)) 1200 return false; 1201 1202 /* RST segments are not recommended to carry timestamp, 1203 and, if they do, it is recommended to ignore PAWS because 1204 "their cleanup function should take precedence over timestamps." 1205 Certainly, it is mistake. It is necessary to understand the reasons 1206 of this constraint to relax it: if peer reboots, clock may go 1207 out-of-sync and half-open connections will not be reset. 1208 Actually, the problem would be not existing if all 1209 the implementations followed draft about maintaining clock 1210 via reboots. Linux-2.2 DOES NOT! 1211 1212 However, we can relax time bounds for RST segments to MSL. 1213 */ 1214 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1215 return false; 1216 return true; 1217 } 1218 1219 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1220 int mib_idx, u32 *last_oow_ack_time); 1221 1222 static inline void tcp_mib_init(struct net *net) 1223 { 1224 /* See RFC 2012 */ 1225 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1226 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1227 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1228 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 1229 } 1230 1231 /* from STCP */ 1232 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1233 { 1234 tp->lost_skb_hint = NULL; 1235 } 1236 1237 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1238 { 1239 tcp_clear_retrans_hints_partial(tp); 1240 tp->retransmit_skb_hint = NULL; 1241 } 1242 1243 /* MD5 Signature */ 1244 struct crypto_hash; 1245 1246 union tcp_md5_addr { 1247 struct in_addr a4; 1248 #if IS_ENABLED(CONFIG_IPV6) 1249 struct in6_addr a6; 1250 #endif 1251 }; 1252 1253 /* - key database */ 1254 struct tcp_md5sig_key { 1255 struct hlist_node node; 1256 u8 keylen; 1257 u8 family; /* AF_INET or AF_INET6 */ 1258 union tcp_md5_addr addr; 1259 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1260 struct rcu_head rcu; 1261 }; 1262 1263 /* - sock block */ 1264 struct tcp_md5sig_info { 1265 struct hlist_head head; 1266 struct rcu_head rcu; 1267 }; 1268 1269 /* - pseudo header */ 1270 struct tcp4_pseudohdr { 1271 __be32 saddr; 1272 __be32 daddr; 1273 __u8 pad; 1274 __u8 protocol; 1275 __be16 len; 1276 }; 1277 1278 struct tcp6_pseudohdr { 1279 struct in6_addr saddr; 1280 struct in6_addr daddr; 1281 __be32 len; 1282 __be32 protocol; /* including padding */ 1283 }; 1284 1285 union tcp_md5sum_block { 1286 struct tcp4_pseudohdr ip4; 1287 #if IS_ENABLED(CONFIG_IPV6) 1288 struct tcp6_pseudohdr ip6; 1289 #endif 1290 }; 1291 1292 /* - pool: digest algorithm, hash description and scratch buffer */ 1293 struct tcp_md5sig_pool { 1294 struct hash_desc md5_desc; 1295 union tcp_md5sum_block md5_blk; 1296 }; 1297 1298 /* - functions */ 1299 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1300 const struct sock *sk, const struct sk_buff *skb); 1301 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1302 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); 1303 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1304 int family); 1305 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1306 const struct sock *addr_sk); 1307 1308 #ifdef CONFIG_TCP_MD5SIG 1309 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1310 const union tcp_md5_addr *addr, 1311 int family); 1312 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1313 #else 1314 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1315 const union tcp_md5_addr *addr, 1316 int family) 1317 { 1318 return NULL; 1319 } 1320 #define tcp_twsk_md5_key(twsk) NULL 1321 #endif 1322 1323 bool tcp_alloc_md5sig_pool(void); 1324 1325 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1326 static inline void tcp_put_md5sig_pool(void) 1327 { 1328 local_bh_enable(); 1329 } 1330 1331 int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 1332 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1333 unsigned int header_len); 1334 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1335 const struct tcp_md5sig_key *key); 1336 1337 /* From tcp_fastopen.c */ 1338 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1339 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1340 unsigned long *last_syn_loss); 1341 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1342 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1343 u16 try_exp); 1344 struct tcp_fastopen_request { 1345 /* Fast Open cookie. Size 0 means a cookie request */ 1346 struct tcp_fastopen_cookie cookie; 1347 struct msghdr *data; /* data in MSG_FASTOPEN */ 1348 size_t size; 1349 int copied; /* queued in tcp_connect() */ 1350 }; 1351 void tcp_free_fastopen_req(struct tcp_sock *tp); 1352 1353 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1354 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1355 bool tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1356 struct request_sock *req, 1357 struct tcp_fastopen_cookie *foc, 1358 struct dst_entry *dst); 1359 void tcp_fastopen_init_key_once(bool publish); 1360 #define TCP_FASTOPEN_KEY_LENGTH 16 1361 1362 /* Fastopen key context */ 1363 struct tcp_fastopen_context { 1364 struct crypto_cipher *tfm; 1365 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1366 struct rcu_head rcu; 1367 }; 1368 1369 /* write queue abstraction */ 1370 static inline void tcp_write_queue_purge(struct sock *sk) 1371 { 1372 struct sk_buff *skb; 1373 1374 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1375 sk_wmem_free_skb(sk, skb); 1376 sk_mem_reclaim(sk); 1377 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1378 } 1379 1380 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1381 { 1382 return skb_peek(&sk->sk_write_queue); 1383 } 1384 1385 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1386 { 1387 return skb_peek_tail(&sk->sk_write_queue); 1388 } 1389 1390 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1391 const struct sk_buff *skb) 1392 { 1393 return skb_queue_next(&sk->sk_write_queue, skb); 1394 } 1395 1396 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1397 const struct sk_buff *skb) 1398 { 1399 return skb_queue_prev(&sk->sk_write_queue, skb); 1400 } 1401 1402 #define tcp_for_write_queue(skb, sk) \ 1403 skb_queue_walk(&(sk)->sk_write_queue, skb) 1404 1405 #define tcp_for_write_queue_from(skb, sk) \ 1406 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1407 1408 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1409 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1410 1411 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1412 { 1413 return sk->sk_send_head; 1414 } 1415 1416 static inline bool tcp_skb_is_last(const struct sock *sk, 1417 const struct sk_buff *skb) 1418 { 1419 return skb_queue_is_last(&sk->sk_write_queue, skb); 1420 } 1421 1422 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1423 { 1424 if (tcp_skb_is_last(sk, skb)) 1425 sk->sk_send_head = NULL; 1426 else 1427 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1428 } 1429 1430 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1431 { 1432 if (sk->sk_send_head == skb_unlinked) 1433 sk->sk_send_head = NULL; 1434 } 1435 1436 static inline void tcp_init_send_head(struct sock *sk) 1437 { 1438 sk->sk_send_head = NULL; 1439 } 1440 1441 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1442 { 1443 __skb_queue_tail(&sk->sk_write_queue, skb); 1444 } 1445 1446 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1447 { 1448 __tcp_add_write_queue_tail(sk, skb); 1449 1450 /* Queue it, remembering where we must start sending. */ 1451 if (sk->sk_send_head == NULL) { 1452 sk->sk_send_head = skb; 1453 1454 if (tcp_sk(sk)->highest_sack == NULL) 1455 tcp_sk(sk)->highest_sack = skb; 1456 } 1457 } 1458 1459 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1460 { 1461 __skb_queue_head(&sk->sk_write_queue, skb); 1462 } 1463 1464 /* Insert buff after skb on the write queue of sk. */ 1465 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1466 struct sk_buff *buff, 1467 struct sock *sk) 1468 { 1469 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1470 } 1471 1472 /* Insert new before skb on the write queue of sk. */ 1473 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1474 struct sk_buff *skb, 1475 struct sock *sk) 1476 { 1477 __skb_queue_before(&sk->sk_write_queue, skb, new); 1478 1479 if (sk->sk_send_head == skb) 1480 sk->sk_send_head = new; 1481 } 1482 1483 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1484 { 1485 __skb_unlink(skb, &sk->sk_write_queue); 1486 } 1487 1488 static inline bool tcp_write_queue_empty(struct sock *sk) 1489 { 1490 return skb_queue_empty(&sk->sk_write_queue); 1491 } 1492 1493 static inline void tcp_push_pending_frames(struct sock *sk) 1494 { 1495 if (tcp_send_head(sk)) { 1496 struct tcp_sock *tp = tcp_sk(sk); 1497 1498 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1499 } 1500 } 1501 1502 /* Start sequence of the skb just after the highest skb with SACKed 1503 * bit, valid only if sacked_out > 0 or when the caller has ensured 1504 * validity by itself. 1505 */ 1506 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1507 { 1508 if (!tp->sacked_out) 1509 return tp->snd_una; 1510 1511 if (tp->highest_sack == NULL) 1512 return tp->snd_nxt; 1513 1514 return TCP_SKB_CB(tp->highest_sack)->seq; 1515 } 1516 1517 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1518 { 1519 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1520 tcp_write_queue_next(sk, skb); 1521 } 1522 1523 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1524 { 1525 return tcp_sk(sk)->highest_sack; 1526 } 1527 1528 static inline void tcp_highest_sack_reset(struct sock *sk) 1529 { 1530 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1531 } 1532 1533 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1534 static inline void tcp_highest_sack_combine(struct sock *sk, 1535 struct sk_buff *old, 1536 struct sk_buff *new) 1537 { 1538 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1539 tcp_sk(sk)->highest_sack = new; 1540 } 1541 1542 /* Determines whether this is a thin stream (which may suffer from 1543 * increased latency). Used to trigger latency-reducing mechanisms. 1544 */ 1545 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1546 { 1547 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1548 } 1549 1550 /* /proc */ 1551 enum tcp_seq_states { 1552 TCP_SEQ_STATE_LISTENING, 1553 TCP_SEQ_STATE_OPENREQ, 1554 TCP_SEQ_STATE_ESTABLISHED, 1555 }; 1556 1557 int tcp_seq_open(struct inode *inode, struct file *file); 1558 1559 struct tcp_seq_afinfo { 1560 char *name; 1561 sa_family_t family; 1562 const struct file_operations *seq_fops; 1563 struct seq_operations seq_ops; 1564 }; 1565 1566 struct tcp_iter_state { 1567 struct seq_net_private p; 1568 sa_family_t family; 1569 enum tcp_seq_states state; 1570 struct sock *syn_wait_sk; 1571 int bucket, offset, sbucket, num; 1572 kuid_t uid; 1573 loff_t last_pos; 1574 }; 1575 1576 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1577 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1578 1579 extern struct request_sock_ops tcp_request_sock_ops; 1580 extern struct request_sock_ops tcp6_request_sock_ops; 1581 1582 void tcp_v4_destroy_sock(struct sock *sk); 1583 1584 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1585 netdev_features_t features); 1586 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1587 int tcp_gro_complete(struct sk_buff *skb); 1588 1589 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1590 1591 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1592 { 1593 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat; 1594 } 1595 1596 static inline bool tcp_stream_memory_free(const struct sock *sk) 1597 { 1598 const struct tcp_sock *tp = tcp_sk(sk); 1599 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1600 1601 return notsent_bytes < tcp_notsent_lowat(tp); 1602 } 1603 1604 #ifdef CONFIG_PROC_FS 1605 int tcp4_proc_init(void); 1606 void tcp4_proc_exit(void); 1607 #endif 1608 1609 int tcp_rtx_synack(struct sock *sk, struct request_sock *req); 1610 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1611 const struct tcp_request_sock_ops *af_ops, 1612 struct sock *sk, struct sk_buff *skb); 1613 1614 /* TCP af-specific functions */ 1615 struct tcp_sock_af_ops { 1616 #ifdef CONFIG_TCP_MD5SIG 1617 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1618 const struct sock *addr_sk); 1619 int (*calc_md5_hash)(char *location, 1620 const struct tcp_md5sig_key *md5, 1621 const struct sock *sk, 1622 const struct sk_buff *skb); 1623 int (*md5_parse)(struct sock *sk, 1624 char __user *optval, 1625 int optlen); 1626 #endif 1627 }; 1628 1629 struct tcp_request_sock_ops { 1630 u16 mss_clamp; 1631 #ifdef CONFIG_TCP_MD5SIG 1632 struct tcp_md5sig_key *(*req_md5_lookup)(struct sock *sk, 1633 const struct sock *addr_sk); 1634 int (*calc_md5_hash) (char *location, 1635 const struct tcp_md5sig_key *md5, 1636 const struct sock *sk, 1637 const struct sk_buff *skb); 1638 #endif 1639 void (*init_req)(struct request_sock *req, struct sock *sk, 1640 struct sk_buff *skb); 1641 #ifdef CONFIG_SYN_COOKIES 1642 __u32 (*cookie_init_seq)(struct sock *sk, const struct sk_buff *skb, 1643 __u16 *mss); 1644 #endif 1645 struct dst_entry *(*route_req)(struct sock *sk, struct flowi *fl, 1646 const struct request_sock *req, 1647 bool *strict); 1648 __u32 (*init_seq)(const struct sk_buff *skb); 1649 int (*send_synack)(struct sock *sk, struct dst_entry *dst, 1650 struct flowi *fl, struct request_sock *req, 1651 u16 queue_mapping, struct tcp_fastopen_cookie *foc); 1652 void (*queue_hash_add)(struct sock *sk, struct request_sock *req, 1653 const unsigned long timeout); 1654 }; 1655 1656 #ifdef CONFIG_SYN_COOKIES 1657 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1658 struct sock *sk, struct sk_buff *skb, 1659 __u16 *mss) 1660 { 1661 return ops->cookie_init_seq(sk, skb, mss); 1662 } 1663 #else 1664 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1665 struct sock *sk, struct sk_buff *skb, 1666 __u16 *mss) 1667 { 1668 return 0; 1669 } 1670 #endif 1671 1672 int tcpv4_offload_init(void); 1673 1674 void tcp_v4_init(void); 1675 void tcp_init(void); 1676 1677 /* 1678 * Save and compile IPv4 options, return a pointer to it 1679 */ 1680 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb) 1681 { 1682 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1683 struct ip_options_rcu *dopt = NULL; 1684 1685 if (opt->optlen) { 1686 int opt_size = sizeof(*dopt) + opt->optlen; 1687 1688 dopt = kmalloc(opt_size, GFP_ATOMIC); 1689 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) { 1690 kfree(dopt); 1691 dopt = NULL; 1692 } 1693 } 1694 return dopt; 1695 } 1696 1697 /* locally generated TCP pure ACKs have skb->truesize == 2 1698 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1699 * This is much faster than dissecting the packet to find out. 1700 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1701 */ 1702 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1703 { 1704 return skb->truesize == 2; 1705 } 1706 1707 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1708 { 1709 skb->truesize = 2; 1710 } 1711 1712 #endif /* _TCP_H */ 1713