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/cryptohash.h> 31 #include <linux/kref.h> 32 #include <linux/ktime.h> 33 34 #include <net/inet_connection_sock.h> 35 #include <net/inet_timewait_sock.h> 36 #include <net/inet_hashtables.h> 37 #include <net/checksum.h> 38 #include <net/request_sock.h> 39 #include <net/sock.h> 40 #include <net/snmp.h> 41 #include <net/ip.h> 42 #include <net/tcp_states.h> 43 #include <net/inet_ecn.h> 44 #include <net/dst.h> 45 46 #include <linux/seq_file.h> 47 #include <linux/memcontrol.h> 48 49 extern struct inet_hashinfo tcp_hashinfo; 50 51 extern struct percpu_counter tcp_orphan_count; 52 void tcp_time_wait(struct sock *sk, int state, int timeo); 53 54 #define MAX_TCP_HEADER (128 + MAX_HEADER) 55 #define MAX_TCP_OPTION_SPACE 40 56 57 /* 58 * Never offer a window over 32767 without using window scaling. Some 59 * poor stacks do signed 16bit maths! 60 */ 61 #define MAX_TCP_WINDOW 32767U 62 63 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 64 #define TCP_MIN_MSS 88U 65 66 /* The least MTU to use for probing */ 67 #define TCP_BASE_MSS 1024 68 69 /* probing interval, default to 10 minutes as per RFC4821 */ 70 #define TCP_PROBE_INTERVAL 600 71 72 /* Specify interval when tcp mtu probing will stop */ 73 #define TCP_PROBE_THRESHOLD 8 74 75 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 76 #define TCP_FASTRETRANS_THRESH 3 77 78 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 79 #define TCP_MAX_QUICKACKS 16U 80 81 /* urg_data states */ 82 #define TCP_URG_VALID 0x0100 83 #define TCP_URG_NOTYET 0x0200 84 #define TCP_URG_READ 0x0400 85 86 #define TCP_RETR1 3 /* 87 * This is how many retries it does before it 88 * tries to figure out if the gateway is 89 * down. Minimal RFC value is 3; it corresponds 90 * to ~3sec-8min depending on RTO. 91 */ 92 93 #define TCP_RETR2 15 /* 94 * This should take at least 95 * 90 minutes to time out. 96 * RFC1122 says that the limit is 100 sec. 97 * 15 is ~13-30min depending on RTO. 98 */ 99 100 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 101 * when active opening a connection. 102 * RFC1122 says the minimum retry MUST 103 * be at least 180secs. Nevertheless 104 * this value is corresponding to 105 * 63secs of retransmission with the 106 * current initial RTO. 107 */ 108 109 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 110 * when passive opening a connection. 111 * This is corresponding to 31secs of 112 * retransmission with the current 113 * initial RTO. 114 */ 115 116 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 117 * state, about 60 seconds */ 118 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 119 /* BSD style FIN_WAIT2 deadlock breaker. 120 * It used to be 3min, new value is 60sec, 121 * to combine FIN-WAIT-2 timeout with 122 * TIME-WAIT timer. 123 */ 124 125 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 126 #if HZ >= 100 127 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 128 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 129 #else 130 #define TCP_DELACK_MIN 4U 131 #define TCP_ATO_MIN 4U 132 #endif 133 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 134 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 135 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 136 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 137 * used as a fallback RTO for the 138 * initial data transmission if no 139 * valid RTT sample has been acquired, 140 * most likely due to retrans in 3WHS. 141 */ 142 143 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 144 * for local resources. 145 */ 146 147 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 148 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 149 #define TCP_KEEPALIVE_INTVL (75*HZ) 150 151 #define MAX_TCP_KEEPIDLE 32767 152 #define MAX_TCP_KEEPINTVL 32767 153 #define MAX_TCP_KEEPCNT 127 154 #define MAX_TCP_SYNCNT 127 155 156 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 157 158 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 159 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 160 * after this time. It should be equal 161 * (or greater than) TCP_TIMEWAIT_LEN 162 * to provide reliability equal to one 163 * provided by timewait state. 164 */ 165 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 166 * timestamps. It must be less than 167 * minimal timewait lifetime. 168 */ 169 /* 170 * TCP option 171 */ 172 173 #define TCPOPT_NOP 1 /* Padding */ 174 #define TCPOPT_EOL 0 /* End of options */ 175 #define TCPOPT_MSS 2 /* Segment size negotiating */ 176 #define TCPOPT_WINDOW 3 /* Window scaling */ 177 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 178 #define TCPOPT_SACK 5 /* SACK Block */ 179 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 180 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 181 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 182 #define TCPOPT_EXP 254 /* Experimental */ 183 /* Magic number to be after the option value for sharing TCP 184 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 185 */ 186 #define TCPOPT_FASTOPEN_MAGIC 0xF989 187 188 /* 189 * TCP option lengths 190 */ 191 192 #define TCPOLEN_MSS 4 193 #define TCPOLEN_WINDOW 3 194 #define TCPOLEN_SACK_PERM 2 195 #define TCPOLEN_TIMESTAMP 10 196 #define TCPOLEN_MD5SIG 18 197 #define TCPOLEN_FASTOPEN_BASE 2 198 #define TCPOLEN_EXP_FASTOPEN_BASE 4 199 200 /* But this is what stacks really send out. */ 201 #define TCPOLEN_TSTAMP_ALIGNED 12 202 #define TCPOLEN_WSCALE_ALIGNED 4 203 #define TCPOLEN_SACKPERM_ALIGNED 4 204 #define TCPOLEN_SACK_BASE 2 205 #define TCPOLEN_SACK_BASE_ALIGNED 4 206 #define TCPOLEN_SACK_PERBLOCK 8 207 #define TCPOLEN_MD5SIG_ALIGNED 20 208 #define TCPOLEN_MSS_ALIGNED 4 209 210 /* Flags in tp->nonagle */ 211 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 212 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 213 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 214 215 /* TCP thin-stream limits */ 216 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 217 218 /* TCP initial congestion window as per rfc6928 */ 219 #define TCP_INIT_CWND 10 220 221 /* Bit Flags for sysctl_tcp_fastopen */ 222 #define TFO_CLIENT_ENABLE 1 223 #define TFO_SERVER_ENABLE 2 224 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 225 226 /* Accept SYN data w/o any cookie option */ 227 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 228 229 /* Force enable TFO on all listeners, i.e., not requiring the 230 * TCP_FASTOPEN socket option. 231 */ 232 #define TFO_SERVER_WO_SOCKOPT1 0x400 233 234 extern struct inet_timewait_death_row tcp_death_row; 235 236 /* sysctl variables for tcp */ 237 extern int sysctl_tcp_timestamps; 238 extern int sysctl_tcp_window_scaling; 239 extern int sysctl_tcp_sack; 240 extern int sysctl_tcp_fastopen; 241 extern int sysctl_tcp_retrans_collapse; 242 extern int sysctl_tcp_stdurg; 243 extern int sysctl_tcp_rfc1337; 244 extern int sysctl_tcp_abort_on_overflow; 245 extern int sysctl_tcp_max_orphans; 246 extern int sysctl_tcp_fack; 247 extern int sysctl_tcp_reordering; 248 extern int sysctl_tcp_max_reordering; 249 extern int sysctl_tcp_dsack; 250 extern long sysctl_tcp_mem[3]; 251 extern int sysctl_tcp_wmem[3]; 252 extern int sysctl_tcp_rmem[3]; 253 extern int sysctl_tcp_app_win; 254 extern int sysctl_tcp_adv_win_scale; 255 extern int sysctl_tcp_tw_reuse; 256 extern int sysctl_tcp_frto; 257 extern int sysctl_tcp_low_latency; 258 extern int sysctl_tcp_nometrics_save; 259 extern int sysctl_tcp_moderate_rcvbuf; 260 extern int sysctl_tcp_tso_win_divisor; 261 extern int sysctl_tcp_workaround_signed_windows; 262 extern int sysctl_tcp_slow_start_after_idle; 263 extern int sysctl_tcp_thin_linear_timeouts; 264 extern int sysctl_tcp_thin_dupack; 265 extern int sysctl_tcp_early_retrans; 266 extern int sysctl_tcp_limit_output_bytes; 267 extern int sysctl_tcp_challenge_ack_limit; 268 extern int sysctl_tcp_min_tso_segs; 269 extern int sysctl_tcp_min_rtt_wlen; 270 extern int sysctl_tcp_autocorking; 271 extern int sysctl_tcp_invalid_ratelimit; 272 extern int sysctl_tcp_pacing_ss_ratio; 273 extern int sysctl_tcp_pacing_ca_ratio; 274 275 extern atomic_long_t tcp_memory_allocated; 276 extern struct percpu_counter tcp_sockets_allocated; 277 extern int tcp_memory_pressure; 278 279 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 280 static inline bool tcp_under_memory_pressure(const struct sock *sk) 281 { 282 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 283 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 284 return true; 285 286 return tcp_memory_pressure; 287 } 288 /* 289 * The next routines deal with comparing 32 bit unsigned ints 290 * and worry about wraparound (automatic with unsigned arithmetic). 291 */ 292 293 static inline bool before(__u32 seq1, __u32 seq2) 294 { 295 return (__s32)(seq1-seq2) < 0; 296 } 297 #define after(seq2, seq1) before(seq1, seq2) 298 299 /* is s2<=s1<=s3 ? */ 300 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 301 { 302 return seq3 - seq2 >= seq1 - seq2; 303 } 304 305 static inline bool tcp_out_of_memory(struct sock *sk) 306 { 307 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 308 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 309 return true; 310 return false; 311 } 312 313 void sk_forced_mem_schedule(struct sock *sk, int size); 314 315 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 316 { 317 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 318 int orphans = percpu_counter_read_positive(ocp); 319 320 if (orphans << shift > sysctl_tcp_max_orphans) { 321 orphans = percpu_counter_sum_positive(ocp); 322 if (orphans << shift > sysctl_tcp_max_orphans) 323 return true; 324 } 325 return false; 326 } 327 328 bool tcp_check_oom(struct sock *sk, int shift); 329 330 331 extern struct proto tcp_prot; 332 333 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 334 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 335 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 336 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 337 338 void tcp_tasklet_init(void); 339 340 void tcp_v4_err(struct sk_buff *skb, u32); 341 342 void tcp_shutdown(struct sock *sk, int how); 343 344 void tcp_v4_early_demux(struct sk_buff *skb); 345 int tcp_v4_rcv(struct sk_buff *skb); 346 347 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 348 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 349 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 350 int flags); 351 void tcp_release_cb(struct sock *sk); 352 void tcp_wfree(struct sk_buff *skb); 353 void tcp_write_timer_handler(struct sock *sk); 354 void tcp_delack_timer_handler(struct sock *sk); 355 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 356 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 357 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 358 const struct tcphdr *th, unsigned int len); 359 void tcp_rcv_space_adjust(struct sock *sk); 360 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 361 void tcp_twsk_destructor(struct sock *sk); 362 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 363 struct pipe_inode_info *pipe, size_t len, 364 unsigned int flags); 365 366 static inline void tcp_dec_quickack_mode(struct sock *sk, 367 const unsigned int pkts) 368 { 369 struct inet_connection_sock *icsk = inet_csk(sk); 370 371 if (icsk->icsk_ack.quick) { 372 if (pkts >= icsk->icsk_ack.quick) { 373 icsk->icsk_ack.quick = 0; 374 /* Leaving quickack mode we deflate ATO. */ 375 icsk->icsk_ack.ato = TCP_ATO_MIN; 376 } else 377 icsk->icsk_ack.quick -= pkts; 378 } 379 } 380 381 #define TCP_ECN_OK 1 382 #define TCP_ECN_QUEUE_CWR 2 383 #define TCP_ECN_DEMAND_CWR 4 384 #define TCP_ECN_SEEN 8 385 386 enum tcp_tw_status { 387 TCP_TW_SUCCESS = 0, 388 TCP_TW_RST = 1, 389 TCP_TW_ACK = 2, 390 TCP_TW_SYN = 3 391 }; 392 393 394 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 395 struct sk_buff *skb, 396 const struct tcphdr *th); 397 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 398 struct request_sock *req, bool fastopen); 399 int tcp_child_process(struct sock *parent, struct sock *child, 400 struct sk_buff *skb); 401 void tcp_enter_loss(struct sock *sk); 402 void tcp_clear_retrans(struct tcp_sock *tp); 403 void tcp_update_metrics(struct sock *sk); 404 void tcp_init_metrics(struct sock *sk); 405 void tcp_metrics_init(void); 406 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, 407 bool paws_check, bool timestamps); 408 bool tcp_remember_stamp(struct sock *sk); 409 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); 410 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); 411 void tcp_disable_fack(struct tcp_sock *tp); 412 void tcp_close(struct sock *sk, long timeout); 413 void tcp_init_sock(struct sock *sk); 414 unsigned int tcp_poll(struct file *file, struct socket *sock, 415 struct poll_table_struct *wait); 416 int tcp_getsockopt(struct sock *sk, int level, int optname, 417 char __user *optval, int __user *optlen); 418 int tcp_setsockopt(struct sock *sk, int level, int optname, 419 char __user *optval, unsigned int optlen); 420 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 421 char __user *optval, int __user *optlen); 422 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 423 char __user *optval, unsigned int optlen); 424 void tcp_set_keepalive(struct sock *sk, int val); 425 void tcp_syn_ack_timeout(const struct request_sock *req); 426 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 427 int flags, int *addr_len); 428 void tcp_parse_options(const struct sk_buff *skb, 429 struct tcp_options_received *opt_rx, 430 int estab, struct tcp_fastopen_cookie *foc); 431 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 432 433 /* 434 * TCP v4 functions exported for the inet6 API 435 */ 436 437 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 438 void tcp_v4_mtu_reduced(struct sock *sk); 439 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 440 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 441 struct sock *tcp_create_openreq_child(const struct sock *sk, 442 struct request_sock *req, 443 struct sk_buff *skb); 444 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 445 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 446 struct request_sock *req, 447 struct dst_entry *dst, 448 struct request_sock *req_unhash, 449 bool *own_req); 450 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 451 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 452 int tcp_connect(struct sock *sk); 453 enum tcp_synack_type { 454 TCP_SYNACK_NORMAL, 455 TCP_SYNACK_FASTOPEN, 456 TCP_SYNACK_COOKIE, 457 }; 458 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 459 struct request_sock *req, 460 struct tcp_fastopen_cookie *foc, 461 enum tcp_synack_type synack_type); 462 int tcp_disconnect(struct sock *sk, int flags); 463 464 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 465 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 466 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 467 468 /* From syncookies.c */ 469 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 470 struct request_sock *req, 471 struct dst_entry *dst); 472 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 473 u32 cookie); 474 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 475 #ifdef CONFIG_SYN_COOKIES 476 477 /* Syncookies use a monotonic timer which increments every 60 seconds. 478 * This counter is used both as a hash input and partially encoded into 479 * the cookie value. A cookie is only validated further if the delta 480 * between the current counter value and the encoded one is less than this, 481 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 482 * the counter advances immediately after a cookie is generated). 483 */ 484 #define MAX_SYNCOOKIE_AGE 2 485 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 486 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 487 488 /* syncookies: remember time of last synqueue overflow 489 * But do not dirty this field too often (once per second is enough) 490 * It is racy as we do not hold a lock, but race is very minor. 491 */ 492 static inline void tcp_synq_overflow(const struct sock *sk) 493 { 494 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 495 unsigned long now = jiffies; 496 497 if (time_after(now, last_overflow + HZ)) 498 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 499 } 500 501 /* syncookies: no recent synqueue overflow on this listening socket? */ 502 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 503 { 504 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 505 506 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID); 507 } 508 509 static inline u32 tcp_cookie_time(void) 510 { 511 u64 val = get_jiffies_64(); 512 513 do_div(val, TCP_SYNCOOKIE_PERIOD); 514 return val; 515 } 516 517 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 518 u16 *mssp); 519 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 520 __u32 cookie_init_timestamp(struct request_sock *req); 521 bool cookie_timestamp_decode(struct tcp_options_received *opt); 522 bool cookie_ecn_ok(const struct tcp_options_received *opt, 523 const struct net *net, const struct dst_entry *dst); 524 525 /* From net/ipv6/syncookies.c */ 526 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 527 u32 cookie); 528 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 529 530 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 531 const struct tcphdr *th, u16 *mssp); 532 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 533 #endif 534 /* tcp_output.c */ 535 536 u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now, 537 int min_tso_segs); 538 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 539 int nonagle); 540 bool tcp_may_send_now(struct sock *sk); 541 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 542 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 543 void tcp_retransmit_timer(struct sock *sk); 544 void tcp_xmit_retransmit_queue(struct sock *); 545 void tcp_simple_retransmit(struct sock *); 546 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 547 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t); 548 549 void tcp_send_probe0(struct sock *); 550 void tcp_send_partial(struct sock *); 551 int tcp_write_wakeup(struct sock *, int mib); 552 void tcp_send_fin(struct sock *sk); 553 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 554 int tcp_send_synack(struct sock *); 555 void tcp_push_one(struct sock *, unsigned int mss_now); 556 void tcp_send_ack(struct sock *sk); 557 void tcp_send_delayed_ack(struct sock *sk); 558 void tcp_send_loss_probe(struct sock *sk); 559 bool tcp_schedule_loss_probe(struct sock *sk); 560 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 561 const struct sk_buff *next_skb); 562 563 /* tcp_input.c */ 564 void tcp_resume_early_retransmit(struct sock *sk); 565 void tcp_rearm_rto(struct sock *sk); 566 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 567 void tcp_reset(struct sock *sk); 568 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 569 void tcp_fin(struct sock *sk); 570 571 /* tcp_timer.c */ 572 void tcp_init_xmit_timers(struct sock *); 573 static inline void tcp_clear_xmit_timers(struct sock *sk) 574 { 575 inet_csk_clear_xmit_timers(sk); 576 } 577 578 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 579 unsigned int tcp_current_mss(struct sock *sk); 580 581 /* Bound MSS / TSO packet size with the half of the window */ 582 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 583 { 584 int cutoff; 585 586 /* When peer uses tiny windows, there is no use in packetizing 587 * to sub-MSS pieces for the sake of SWS or making sure there 588 * are enough packets in the pipe for fast recovery. 589 * 590 * On the other hand, for extremely large MSS devices, handling 591 * smaller than MSS windows in this way does make sense. 592 */ 593 if (tp->max_window > TCP_MSS_DEFAULT) 594 cutoff = (tp->max_window >> 1); 595 else 596 cutoff = tp->max_window; 597 598 if (cutoff && pktsize > cutoff) 599 return max_t(int, cutoff, 68U - tp->tcp_header_len); 600 else 601 return pktsize; 602 } 603 604 /* tcp.c */ 605 void tcp_get_info(struct sock *, struct tcp_info *); 606 607 /* Read 'sendfile()'-style from a TCP socket */ 608 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 609 sk_read_actor_t recv_actor); 610 611 void tcp_initialize_rcv_mss(struct sock *sk); 612 613 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 614 int tcp_mss_to_mtu(struct sock *sk, int mss); 615 void tcp_mtup_init(struct sock *sk); 616 void tcp_init_buffer_space(struct sock *sk); 617 618 static inline void tcp_bound_rto(const struct sock *sk) 619 { 620 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 621 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 622 } 623 624 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 625 { 626 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 627 } 628 629 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 630 { 631 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 632 ntohl(TCP_FLAG_ACK) | 633 snd_wnd); 634 } 635 636 static inline void tcp_fast_path_on(struct tcp_sock *tp) 637 { 638 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 639 } 640 641 static inline void tcp_fast_path_check(struct sock *sk) 642 { 643 struct tcp_sock *tp = tcp_sk(sk); 644 645 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 646 tp->rcv_wnd && 647 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 648 !tp->urg_data) 649 tcp_fast_path_on(tp); 650 } 651 652 /* Compute the actual rto_min value */ 653 static inline u32 tcp_rto_min(struct sock *sk) 654 { 655 const struct dst_entry *dst = __sk_dst_get(sk); 656 u32 rto_min = TCP_RTO_MIN; 657 658 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 659 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 660 return rto_min; 661 } 662 663 static inline u32 tcp_rto_min_us(struct sock *sk) 664 { 665 return jiffies_to_usecs(tcp_rto_min(sk)); 666 } 667 668 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 669 { 670 return dst_metric_locked(dst, RTAX_CC_ALGO); 671 } 672 673 /* Minimum RTT in usec. ~0 means not available. */ 674 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 675 { 676 return minmax_get(&tp->rtt_min); 677 } 678 679 /* Compute the actual receive window we are currently advertising. 680 * Rcv_nxt can be after the window if our peer push more data 681 * than the offered window. 682 */ 683 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 684 { 685 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 686 687 if (win < 0) 688 win = 0; 689 return (u32) win; 690 } 691 692 /* Choose a new window, without checks for shrinking, and without 693 * scaling applied to the result. The caller does these things 694 * if necessary. This is a "raw" window selection. 695 */ 696 u32 __tcp_select_window(struct sock *sk); 697 698 void tcp_send_window_probe(struct sock *sk); 699 700 /* TCP timestamps are only 32-bits, this causes a slight 701 * complication on 64-bit systems since we store a snapshot 702 * of jiffies in the buffer control blocks below. We decided 703 * to use only the low 32-bits of jiffies and hide the ugly 704 * casts with the following macro. 705 */ 706 #define tcp_time_stamp ((__u32)(jiffies)) 707 708 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 709 { 710 return skb->skb_mstamp.stamp_jiffies; 711 } 712 713 714 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 715 716 #define TCPHDR_FIN 0x01 717 #define TCPHDR_SYN 0x02 718 #define TCPHDR_RST 0x04 719 #define TCPHDR_PSH 0x08 720 #define TCPHDR_ACK 0x10 721 #define TCPHDR_URG 0x20 722 #define TCPHDR_ECE 0x40 723 #define TCPHDR_CWR 0x80 724 725 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 726 727 /* This is what the send packet queuing engine uses to pass 728 * TCP per-packet control information to the transmission code. 729 * We also store the host-order sequence numbers in here too. 730 * This is 44 bytes if IPV6 is enabled. 731 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 732 */ 733 struct tcp_skb_cb { 734 __u32 seq; /* Starting sequence number */ 735 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 736 union { 737 /* Note : tcp_tw_isn is used in input path only 738 * (isn chosen by tcp_timewait_state_process()) 739 * 740 * tcp_gso_segs/size are used in write queue only, 741 * cf tcp_skb_pcount()/tcp_skb_mss() 742 */ 743 __u32 tcp_tw_isn; 744 struct { 745 u16 tcp_gso_segs; 746 u16 tcp_gso_size; 747 }; 748 }; 749 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 750 751 __u8 sacked; /* State flags for SACK/FACK. */ 752 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 753 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 754 #define TCPCB_LOST 0x04 /* SKB is lost */ 755 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 756 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 757 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 758 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 759 TCPCB_REPAIRED) 760 761 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 762 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 763 eor:1, /* Is skb MSG_EOR marked? */ 764 unused:6; 765 __u32 ack_seq; /* Sequence number ACK'd */ 766 union { 767 struct { 768 /* There is space for up to 24 bytes */ 769 __u32 in_flight:30,/* Bytes in flight at transmit */ 770 is_app_limited:1, /* cwnd not fully used? */ 771 unused:1; 772 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 773 __u32 delivered; 774 /* start of send pipeline phase */ 775 struct skb_mstamp first_tx_mstamp; 776 /* when we reached the "delivered" count */ 777 struct skb_mstamp delivered_mstamp; 778 } tx; /* only used for outgoing skbs */ 779 union { 780 struct inet_skb_parm h4; 781 #if IS_ENABLED(CONFIG_IPV6) 782 struct inet6_skb_parm h6; 783 #endif 784 } header; /* For incoming skbs */ 785 }; 786 }; 787 788 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 789 790 791 #if IS_ENABLED(CONFIG_IPV6) 792 /* This is the variant of inet6_iif() that must be used by TCP, 793 * as TCP moves IP6CB into a different location in skb->cb[] 794 */ 795 static inline int tcp_v6_iif(const struct sk_buff *skb) 796 { 797 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 798 799 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 800 } 801 #endif 802 803 /* TCP_SKB_CB reference means this can not be used from early demux */ 804 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 805 { 806 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 807 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 808 skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 809 return true; 810 #endif 811 return false; 812 } 813 814 /* Due to TSO, an SKB can be composed of multiple actual 815 * packets. To keep these tracked properly, we use this. 816 */ 817 static inline int tcp_skb_pcount(const struct sk_buff *skb) 818 { 819 return TCP_SKB_CB(skb)->tcp_gso_segs; 820 } 821 822 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 823 { 824 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 825 } 826 827 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 828 { 829 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 830 } 831 832 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 833 static inline int tcp_skb_mss(const struct sk_buff *skb) 834 { 835 return TCP_SKB_CB(skb)->tcp_gso_size; 836 } 837 838 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 839 { 840 return likely(!TCP_SKB_CB(skb)->eor); 841 } 842 843 /* Events passed to congestion control interface */ 844 enum tcp_ca_event { 845 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 846 CA_EVENT_CWND_RESTART, /* congestion window restart */ 847 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 848 CA_EVENT_LOSS, /* loss timeout */ 849 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 850 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 851 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 852 CA_EVENT_NON_DELAYED_ACK, 853 }; 854 855 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 856 enum tcp_ca_ack_event_flags { 857 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 858 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 859 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 860 }; 861 862 /* 863 * Interface for adding new TCP congestion control handlers 864 */ 865 #define TCP_CA_NAME_MAX 16 866 #define TCP_CA_MAX 128 867 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 868 869 #define TCP_CA_UNSPEC 0 870 871 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 872 #define TCP_CONG_NON_RESTRICTED 0x1 873 /* Requires ECN/ECT set on all packets */ 874 #define TCP_CONG_NEEDS_ECN 0x2 875 876 union tcp_cc_info; 877 878 struct ack_sample { 879 u32 pkts_acked; 880 s32 rtt_us; 881 u32 in_flight; 882 }; 883 884 /* A rate sample measures the number of (original/retransmitted) data 885 * packets delivered "delivered" over an interval of time "interval_us". 886 * The tcp_rate.c code fills in the rate sample, and congestion 887 * control modules that define a cong_control function to run at the end 888 * of ACK processing can optionally chose to consult this sample when 889 * setting cwnd and pacing rate. 890 * A sample is invalid if "delivered" or "interval_us" is negative. 891 */ 892 struct rate_sample { 893 struct skb_mstamp prior_mstamp; /* starting timestamp for interval */ 894 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 895 s32 delivered; /* number of packets delivered over interval */ 896 long interval_us; /* time for tp->delivered to incr "delivered" */ 897 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 898 int losses; /* number of packets marked lost upon ACK */ 899 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 900 u32 prior_in_flight; /* in flight before this ACK */ 901 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 902 bool is_retrans; /* is sample from retransmission? */ 903 }; 904 905 struct tcp_congestion_ops { 906 struct list_head list; 907 u32 key; 908 u32 flags; 909 910 /* initialize private data (optional) */ 911 void (*init)(struct sock *sk); 912 /* cleanup private data (optional) */ 913 void (*release)(struct sock *sk); 914 915 /* return slow start threshold (required) */ 916 u32 (*ssthresh)(struct sock *sk); 917 /* do new cwnd calculation (required) */ 918 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 919 /* call before changing ca_state (optional) */ 920 void (*set_state)(struct sock *sk, u8 new_state); 921 /* call when cwnd event occurs (optional) */ 922 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 923 /* call when ack arrives (optional) */ 924 void (*in_ack_event)(struct sock *sk, u32 flags); 925 /* new value of cwnd after loss (optional) */ 926 u32 (*undo_cwnd)(struct sock *sk); 927 /* hook for packet ack accounting (optional) */ 928 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 929 /* suggest number of segments for each skb to transmit (optional) */ 930 u32 (*tso_segs_goal)(struct sock *sk); 931 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 932 u32 (*sndbuf_expand)(struct sock *sk); 933 /* call when packets are delivered to update cwnd and pacing rate, 934 * after all the ca_state processing. (optional) 935 */ 936 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 937 /* get info for inet_diag (optional) */ 938 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 939 union tcp_cc_info *info); 940 941 char name[TCP_CA_NAME_MAX]; 942 struct module *owner; 943 }; 944 945 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 946 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 947 948 void tcp_assign_congestion_control(struct sock *sk); 949 void tcp_init_congestion_control(struct sock *sk); 950 void tcp_cleanup_congestion_control(struct sock *sk); 951 int tcp_set_default_congestion_control(const char *name); 952 void tcp_get_default_congestion_control(char *name); 953 void tcp_get_available_congestion_control(char *buf, size_t len); 954 void tcp_get_allowed_congestion_control(char *buf, size_t len); 955 int tcp_set_allowed_congestion_control(char *allowed); 956 int tcp_set_congestion_control(struct sock *sk, const char *name); 957 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 958 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 959 960 u32 tcp_reno_ssthresh(struct sock *sk); 961 u32 tcp_reno_undo_cwnd(struct sock *sk); 962 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 963 extern struct tcp_congestion_ops tcp_reno; 964 965 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 966 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca); 967 #ifdef CONFIG_INET 968 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 969 #else 970 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 971 { 972 return NULL; 973 } 974 #endif 975 976 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 977 { 978 const struct inet_connection_sock *icsk = inet_csk(sk); 979 980 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 981 } 982 983 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 984 { 985 struct inet_connection_sock *icsk = inet_csk(sk); 986 987 if (icsk->icsk_ca_ops->set_state) 988 icsk->icsk_ca_ops->set_state(sk, ca_state); 989 icsk->icsk_ca_state = ca_state; 990 } 991 992 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 993 { 994 const struct inet_connection_sock *icsk = inet_csk(sk); 995 996 if (icsk->icsk_ca_ops->cwnd_event) 997 icsk->icsk_ca_ops->cwnd_event(sk, event); 998 } 999 1000 /* From tcp_rate.c */ 1001 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1002 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1003 struct rate_sample *rs); 1004 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1005 struct skb_mstamp *now, struct rate_sample *rs); 1006 void tcp_rate_check_app_limited(struct sock *sk); 1007 1008 /* These functions determine how the current flow behaves in respect of SACK 1009 * handling. SACK is negotiated with the peer, and therefore it can vary 1010 * between different flows. 1011 * 1012 * tcp_is_sack - SACK enabled 1013 * tcp_is_reno - No SACK 1014 * tcp_is_fack - FACK enabled, implies SACK enabled 1015 */ 1016 static inline int tcp_is_sack(const struct tcp_sock *tp) 1017 { 1018 return tp->rx_opt.sack_ok; 1019 } 1020 1021 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1022 { 1023 return !tcp_is_sack(tp); 1024 } 1025 1026 static inline bool tcp_is_fack(const struct tcp_sock *tp) 1027 { 1028 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 1029 } 1030 1031 static inline void tcp_enable_fack(struct tcp_sock *tp) 1032 { 1033 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 1034 } 1035 1036 /* TCP early-retransmit (ER) is similar to but more conservative than 1037 * the thin-dupack feature. Enable ER only if thin-dupack is disabled. 1038 */ 1039 static inline void tcp_enable_early_retrans(struct tcp_sock *tp) 1040 { 1041 struct net *net = sock_net((struct sock *)tp); 1042 1043 tp->do_early_retrans = sysctl_tcp_early_retrans && 1044 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack && 1045 net->ipv4.sysctl_tcp_reordering == 3; 1046 } 1047 1048 static inline void tcp_disable_early_retrans(struct tcp_sock *tp) 1049 { 1050 tp->do_early_retrans = 0; 1051 } 1052 1053 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1054 { 1055 return tp->sacked_out + tp->lost_out; 1056 } 1057 1058 /* This determines how many packets are "in the network" to the best 1059 * of our knowledge. In many cases it is conservative, but where 1060 * detailed information is available from the receiver (via SACK 1061 * blocks etc.) we can make more aggressive calculations. 1062 * 1063 * Use this for decisions involving congestion control, use just 1064 * tp->packets_out to determine if the send queue is empty or not. 1065 * 1066 * Read this equation as: 1067 * 1068 * "Packets sent once on transmission queue" MINUS 1069 * "Packets left network, but not honestly ACKed yet" PLUS 1070 * "Packets fast retransmitted" 1071 */ 1072 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1073 { 1074 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1075 } 1076 1077 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1078 1079 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1080 { 1081 return tp->snd_cwnd < tp->snd_ssthresh; 1082 } 1083 1084 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1085 { 1086 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1087 } 1088 1089 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1090 { 1091 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1092 (1 << inet_csk(sk)->icsk_ca_state); 1093 } 1094 1095 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1096 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1097 * ssthresh. 1098 */ 1099 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1100 { 1101 const struct tcp_sock *tp = tcp_sk(sk); 1102 1103 if (tcp_in_cwnd_reduction(sk)) 1104 return tp->snd_ssthresh; 1105 else 1106 return max(tp->snd_ssthresh, 1107 ((tp->snd_cwnd >> 1) + 1108 (tp->snd_cwnd >> 2))); 1109 } 1110 1111 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1112 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1113 1114 void tcp_enter_cwr(struct sock *sk); 1115 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1116 1117 /* The maximum number of MSS of available cwnd for which TSO defers 1118 * sending if not using sysctl_tcp_tso_win_divisor. 1119 */ 1120 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1121 { 1122 return 3; 1123 } 1124 1125 /* Returns end sequence number of the receiver's advertised window */ 1126 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1127 { 1128 return tp->snd_una + tp->snd_wnd; 1129 } 1130 1131 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1132 * flexible approach. The RFC suggests cwnd should not be raised unless 1133 * it was fully used previously. And that's exactly what we do in 1134 * congestion avoidance mode. But in slow start we allow cwnd to grow 1135 * as long as the application has used half the cwnd. 1136 * Example : 1137 * cwnd is 10 (IW10), but application sends 9 frames. 1138 * We allow cwnd to reach 18 when all frames are ACKed. 1139 * This check is safe because it's as aggressive as slow start which already 1140 * risks 100% overshoot. The advantage is that we discourage application to 1141 * either send more filler packets or data to artificially blow up the cwnd 1142 * usage, and allow application-limited process to probe bw more aggressively. 1143 */ 1144 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1145 { 1146 const struct tcp_sock *tp = tcp_sk(sk); 1147 1148 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1149 if (tcp_in_slow_start(tp)) 1150 return tp->snd_cwnd < 2 * tp->max_packets_out; 1151 1152 return tp->is_cwnd_limited; 1153 } 1154 1155 /* Something is really bad, we could not queue an additional packet, 1156 * because qdisc is full or receiver sent a 0 window. 1157 * We do not want to add fuel to the fire, or abort too early, 1158 * so make sure the timer we arm now is at least 200ms in the future, 1159 * regardless of current icsk_rto value (as it could be ~2ms) 1160 */ 1161 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1162 { 1163 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1164 } 1165 1166 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1167 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1168 unsigned long max_when) 1169 { 1170 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1171 1172 return (unsigned long)min_t(u64, when, max_when); 1173 } 1174 1175 static inline void tcp_check_probe_timer(struct sock *sk) 1176 { 1177 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1178 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1179 tcp_probe0_base(sk), TCP_RTO_MAX); 1180 } 1181 1182 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1183 { 1184 tp->snd_wl1 = seq; 1185 } 1186 1187 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1188 { 1189 tp->snd_wl1 = seq; 1190 } 1191 1192 /* 1193 * Calculate(/check) TCP checksum 1194 */ 1195 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1196 __be32 daddr, __wsum base) 1197 { 1198 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1199 } 1200 1201 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1202 { 1203 return __skb_checksum_complete(skb); 1204 } 1205 1206 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1207 { 1208 return !skb_csum_unnecessary(skb) && 1209 __tcp_checksum_complete(skb); 1210 } 1211 1212 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1213 1214 static inline void tcp_prequeue_init(struct tcp_sock *tp) 1215 { 1216 tp->ucopy.task = NULL; 1217 tp->ucopy.len = 0; 1218 tp->ucopy.memory = 0; 1219 skb_queue_head_init(&tp->ucopy.prequeue); 1220 } 1221 1222 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1223 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1224 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1225 1226 #undef STATE_TRACE 1227 1228 #ifdef STATE_TRACE 1229 static const char *statename[]={ 1230 "Unused","Established","Syn Sent","Syn Recv", 1231 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1232 "Close Wait","Last ACK","Listen","Closing" 1233 }; 1234 #endif 1235 void tcp_set_state(struct sock *sk, int state); 1236 1237 void tcp_done(struct sock *sk); 1238 1239 int tcp_abort(struct sock *sk, int err); 1240 1241 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1242 { 1243 rx_opt->dsack = 0; 1244 rx_opt->num_sacks = 0; 1245 } 1246 1247 u32 tcp_default_init_rwnd(u32 mss); 1248 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1249 1250 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1251 { 1252 struct tcp_sock *tp = tcp_sk(sk); 1253 s32 delta; 1254 1255 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out) 1256 return; 1257 delta = tcp_time_stamp - tp->lsndtime; 1258 if (delta > inet_csk(sk)->icsk_rto) 1259 tcp_cwnd_restart(sk, delta); 1260 } 1261 1262 /* Determine a window scaling and initial window to offer. */ 1263 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1264 __u32 *window_clamp, int wscale_ok, 1265 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1266 1267 static inline int tcp_win_from_space(int space) 1268 { 1269 return sysctl_tcp_adv_win_scale<=0 ? 1270 (space>>(-sysctl_tcp_adv_win_scale)) : 1271 space - (space>>sysctl_tcp_adv_win_scale); 1272 } 1273 1274 /* Note: caller must be prepared to deal with negative returns */ 1275 static inline int tcp_space(const struct sock *sk) 1276 { 1277 return tcp_win_from_space(sk->sk_rcvbuf - 1278 atomic_read(&sk->sk_rmem_alloc)); 1279 } 1280 1281 static inline int tcp_full_space(const struct sock *sk) 1282 { 1283 return tcp_win_from_space(sk->sk_rcvbuf); 1284 } 1285 1286 extern void tcp_openreq_init_rwin(struct request_sock *req, 1287 const struct sock *sk_listener, 1288 const struct dst_entry *dst); 1289 1290 void tcp_enter_memory_pressure(struct sock *sk); 1291 1292 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1293 { 1294 struct net *net = sock_net((struct sock *)tp); 1295 1296 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1297 } 1298 1299 static inline int keepalive_time_when(const struct tcp_sock *tp) 1300 { 1301 struct net *net = sock_net((struct sock *)tp); 1302 1303 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1304 } 1305 1306 static inline int keepalive_probes(const struct tcp_sock *tp) 1307 { 1308 struct net *net = sock_net((struct sock *)tp); 1309 1310 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1311 } 1312 1313 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1314 { 1315 const struct inet_connection_sock *icsk = &tp->inet_conn; 1316 1317 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1318 tcp_time_stamp - tp->rcv_tstamp); 1319 } 1320 1321 static inline int tcp_fin_time(const struct sock *sk) 1322 { 1323 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1324 const int rto = inet_csk(sk)->icsk_rto; 1325 1326 if (fin_timeout < (rto << 2) - (rto >> 1)) 1327 fin_timeout = (rto << 2) - (rto >> 1); 1328 1329 return fin_timeout; 1330 } 1331 1332 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1333 int paws_win) 1334 { 1335 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1336 return true; 1337 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1338 return true; 1339 /* 1340 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1341 * then following tcp messages have valid values. Ignore 0 value, 1342 * or else 'negative' tsval might forbid us to accept their packets. 1343 */ 1344 if (!rx_opt->ts_recent) 1345 return true; 1346 return false; 1347 } 1348 1349 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1350 int rst) 1351 { 1352 if (tcp_paws_check(rx_opt, 0)) 1353 return false; 1354 1355 /* RST segments are not recommended to carry timestamp, 1356 and, if they do, it is recommended to ignore PAWS because 1357 "their cleanup function should take precedence over timestamps." 1358 Certainly, it is mistake. It is necessary to understand the reasons 1359 of this constraint to relax it: if peer reboots, clock may go 1360 out-of-sync and half-open connections will not be reset. 1361 Actually, the problem would be not existing if all 1362 the implementations followed draft about maintaining clock 1363 via reboots. Linux-2.2 DOES NOT! 1364 1365 However, we can relax time bounds for RST segments to MSL. 1366 */ 1367 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1368 return false; 1369 return true; 1370 } 1371 1372 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1373 int mib_idx, u32 *last_oow_ack_time); 1374 1375 static inline void tcp_mib_init(struct net *net) 1376 { 1377 /* See RFC 2012 */ 1378 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1379 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1380 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1381 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1382 } 1383 1384 /* from STCP */ 1385 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1386 { 1387 tp->lost_skb_hint = NULL; 1388 } 1389 1390 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1391 { 1392 tcp_clear_retrans_hints_partial(tp); 1393 tp->retransmit_skb_hint = NULL; 1394 } 1395 1396 union tcp_md5_addr { 1397 struct in_addr a4; 1398 #if IS_ENABLED(CONFIG_IPV6) 1399 struct in6_addr a6; 1400 #endif 1401 }; 1402 1403 /* - key database */ 1404 struct tcp_md5sig_key { 1405 struct hlist_node node; 1406 u8 keylen; 1407 u8 family; /* AF_INET or AF_INET6 */ 1408 union tcp_md5_addr addr; 1409 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1410 struct rcu_head rcu; 1411 }; 1412 1413 /* - sock block */ 1414 struct tcp_md5sig_info { 1415 struct hlist_head head; 1416 struct rcu_head rcu; 1417 }; 1418 1419 /* - pseudo header */ 1420 struct tcp4_pseudohdr { 1421 __be32 saddr; 1422 __be32 daddr; 1423 __u8 pad; 1424 __u8 protocol; 1425 __be16 len; 1426 }; 1427 1428 struct tcp6_pseudohdr { 1429 struct in6_addr saddr; 1430 struct in6_addr daddr; 1431 __be32 len; 1432 __be32 protocol; /* including padding */ 1433 }; 1434 1435 union tcp_md5sum_block { 1436 struct tcp4_pseudohdr ip4; 1437 #if IS_ENABLED(CONFIG_IPV6) 1438 struct tcp6_pseudohdr ip6; 1439 #endif 1440 }; 1441 1442 /* - pool: digest algorithm, hash description and scratch buffer */ 1443 struct tcp_md5sig_pool { 1444 struct ahash_request *md5_req; 1445 void *scratch; 1446 }; 1447 1448 /* - functions */ 1449 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1450 const struct sock *sk, const struct sk_buff *skb); 1451 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1452 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); 1453 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1454 int family); 1455 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1456 const struct sock *addr_sk); 1457 1458 #ifdef CONFIG_TCP_MD5SIG 1459 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1460 const union tcp_md5_addr *addr, 1461 int family); 1462 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1463 #else 1464 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1465 const union tcp_md5_addr *addr, 1466 int family) 1467 { 1468 return NULL; 1469 } 1470 #define tcp_twsk_md5_key(twsk) NULL 1471 #endif 1472 1473 bool tcp_alloc_md5sig_pool(void); 1474 1475 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1476 static inline void tcp_put_md5sig_pool(void) 1477 { 1478 local_bh_enable(); 1479 } 1480 1481 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1482 unsigned int header_len); 1483 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1484 const struct tcp_md5sig_key *key); 1485 1486 /* From tcp_fastopen.c */ 1487 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1488 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1489 unsigned long *last_syn_loss); 1490 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1491 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1492 u16 try_exp); 1493 struct tcp_fastopen_request { 1494 /* Fast Open cookie. Size 0 means a cookie request */ 1495 struct tcp_fastopen_cookie cookie; 1496 struct msghdr *data; /* data in MSG_FASTOPEN */ 1497 size_t size; 1498 int copied; /* queued in tcp_connect() */ 1499 }; 1500 void tcp_free_fastopen_req(struct tcp_sock *tp); 1501 1502 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1503 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1504 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1505 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1506 struct request_sock *req, 1507 struct tcp_fastopen_cookie *foc, 1508 struct dst_entry *dst); 1509 void tcp_fastopen_init_key_once(bool publish); 1510 #define TCP_FASTOPEN_KEY_LENGTH 16 1511 1512 /* Fastopen key context */ 1513 struct tcp_fastopen_context { 1514 struct crypto_cipher *tfm; 1515 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1516 struct rcu_head rcu; 1517 }; 1518 1519 /* Latencies incurred by various limits for a sender. They are 1520 * chronograph-like stats that are mutually exclusive. 1521 */ 1522 enum tcp_chrono { 1523 TCP_CHRONO_UNSPEC, 1524 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1525 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1526 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1527 __TCP_CHRONO_MAX, 1528 }; 1529 1530 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1531 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1532 1533 /* write queue abstraction */ 1534 static inline void tcp_write_queue_purge(struct sock *sk) 1535 { 1536 struct sk_buff *skb; 1537 1538 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1539 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1540 sk_wmem_free_skb(sk, skb); 1541 sk_mem_reclaim(sk); 1542 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1543 } 1544 1545 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1546 { 1547 return skb_peek(&sk->sk_write_queue); 1548 } 1549 1550 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1551 { 1552 return skb_peek_tail(&sk->sk_write_queue); 1553 } 1554 1555 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1556 const struct sk_buff *skb) 1557 { 1558 return skb_queue_next(&sk->sk_write_queue, skb); 1559 } 1560 1561 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1562 const struct sk_buff *skb) 1563 { 1564 return skb_queue_prev(&sk->sk_write_queue, skb); 1565 } 1566 1567 #define tcp_for_write_queue(skb, sk) \ 1568 skb_queue_walk(&(sk)->sk_write_queue, skb) 1569 1570 #define tcp_for_write_queue_from(skb, sk) \ 1571 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1572 1573 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1574 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1575 1576 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1577 { 1578 return sk->sk_send_head; 1579 } 1580 1581 static inline bool tcp_skb_is_last(const struct sock *sk, 1582 const struct sk_buff *skb) 1583 { 1584 return skb_queue_is_last(&sk->sk_write_queue, skb); 1585 } 1586 1587 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1588 { 1589 if (tcp_skb_is_last(sk, skb)) 1590 sk->sk_send_head = NULL; 1591 else 1592 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1593 } 1594 1595 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1596 { 1597 if (sk->sk_send_head == skb_unlinked) { 1598 sk->sk_send_head = NULL; 1599 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1600 } 1601 if (tcp_sk(sk)->highest_sack == skb_unlinked) 1602 tcp_sk(sk)->highest_sack = NULL; 1603 } 1604 1605 static inline void tcp_init_send_head(struct sock *sk) 1606 { 1607 sk->sk_send_head = NULL; 1608 } 1609 1610 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1611 { 1612 __skb_queue_tail(&sk->sk_write_queue, skb); 1613 } 1614 1615 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1616 { 1617 __tcp_add_write_queue_tail(sk, skb); 1618 1619 /* Queue it, remembering where we must start sending. */ 1620 if (sk->sk_send_head == NULL) { 1621 sk->sk_send_head = skb; 1622 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1623 1624 if (tcp_sk(sk)->highest_sack == NULL) 1625 tcp_sk(sk)->highest_sack = skb; 1626 } 1627 } 1628 1629 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1630 { 1631 __skb_queue_head(&sk->sk_write_queue, skb); 1632 } 1633 1634 /* Insert buff after skb on the write queue of sk. */ 1635 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1636 struct sk_buff *buff, 1637 struct sock *sk) 1638 { 1639 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1640 } 1641 1642 /* Insert new before skb on the write queue of sk. */ 1643 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1644 struct sk_buff *skb, 1645 struct sock *sk) 1646 { 1647 __skb_queue_before(&sk->sk_write_queue, skb, new); 1648 1649 if (sk->sk_send_head == skb) 1650 sk->sk_send_head = new; 1651 } 1652 1653 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1654 { 1655 __skb_unlink(skb, &sk->sk_write_queue); 1656 } 1657 1658 static inline bool tcp_write_queue_empty(struct sock *sk) 1659 { 1660 return skb_queue_empty(&sk->sk_write_queue); 1661 } 1662 1663 static inline void tcp_push_pending_frames(struct sock *sk) 1664 { 1665 if (tcp_send_head(sk)) { 1666 struct tcp_sock *tp = tcp_sk(sk); 1667 1668 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1669 } 1670 } 1671 1672 /* Start sequence of the skb just after the highest skb with SACKed 1673 * bit, valid only if sacked_out > 0 or when the caller has ensured 1674 * validity by itself. 1675 */ 1676 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1677 { 1678 if (!tp->sacked_out) 1679 return tp->snd_una; 1680 1681 if (tp->highest_sack == NULL) 1682 return tp->snd_nxt; 1683 1684 return TCP_SKB_CB(tp->highest_sack)->seq; 1685 } 1686 1687 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1688 { 1689 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1690 tcp_write_queue_next(sk, skb); 1691 } 1692 1693 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1694 { 1695 return tcp_sk(sk)->highest_sack; 1696 } 1697 1698 static inline void tcp_highest_sack_reset(struct sock *sk) 1699 { 1700 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1701 } 1702 1703 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1704 static inline void tcp_highest_sack_combine(struct sock *sk, 1705 struct sk_buff *old, 1706 struct sk_buff *new) 1707 { 1708 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1709 tcp_sk(sk)->highest_sack = new; 1710 } 1711 1712 /* This helper checks if socket has IP_TRANSPARENT set */ 1713 static inline bool inet_sk_transparent(const struct sock *sk) 1714 { 1715 switch (sk->sk_state) { 1716 case TCP_TIME_WAIT: 1717 return inet_twsk(sk)->tw_transparent; 1718 case TCP_NEW_SYN_RECV: 1719 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1720 } 1721 return inet_sk(sk)->transparent; 1722 } 1723 1724 /* Determines whether this is a thin stream (which may suffer from 1725 * increased latency). Used to trigger latency-reducing mechanisms. 1726 */ 1727 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1728 { 1729 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1730 } 1731 1732 /* /proc */ 1733 enum tcp_seq_states { 1734 TCP_SEQ_STATE_LISTENING, 1735 TCP_SEQ_STATE_ESTABLISHED, 1736 }; 1737 1738 int tcp_seq_open(struct inode *inode, struct file *file); 1739 1740 struct tcp_seq_afinfo { 1741 char *name; 1742 sa_family_t family; 1743 const struct file_operations *seq_fops; 1744 struct seq_operations seq_ops; 1745 }; 1746 1747 struct tcp_iter_state { 1748 struct seq_net_private p; 1749 sa_family_t family; 1750 enum tcp_seq_states state; 1751 struct sock *syn_wait_sk; 1752 int bucket, offset, sbucket, num; 1753 loff_t last_pos; 1754 }; 1755 1756 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1757 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1758 1759 extern struct request_sock_ops tcp_request_sock_ops; 1760 extern struct request_sock_ops tcp6_request_sock_ops; 1761 1762 void tcp_v4_destroy_sock(struct sock *sk); 1763 1764 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1765 netdev_features_t features); 1766 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1767 int tcp_gro_complete(struct sk_buff *skb); 1768 1769 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1770 1771 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1772 { 1773 struct net *net = sock_net((struct sock *)tp); 1774 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1775 } 1776 1777 static inline bool tcp_stream_memory_free(const struct sock *sk) 1778 { 1779 const struct tcp_sock *tp = tcp_sk(sk); 1780 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1781 1782 return notsent_bytes < tcp_notsent_lowat(tp); 1783 } 1784 1785 #ifdef CONFIG_PROC_FS 1786 int tcp4_proc_init(void); 1787 void tcp4_proc_exit(void); 1788 #endif 1789 1790 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1791 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1792 const struct tcp_request_sock_ops *af_ops, 1793 struct sock *sk, struct sk_buff *skb); 1794 1795 /* TCP af-specific functions */ 1796 struct tcp_sock_af_ops { 1797 #ifdef CONFIG_TCP_MD5SIG 1798 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1799 const struct sock *addr_sk); 1800 int (*calc_md5_hash)(char *location, 1801 const struct tcp_md5sig_key *md5, 1802 const struct sock *sk, 1803 const struct sk_buff *skb); 1804 int (*md5_parse)(struct sock *sk, 1805 char __user *optval, 1806 int optlen); 1807 #endif 1808 }; 1809 1810 struct tcp_request_sock_ops { 1811 u16 mss_clamp; 1812 #ifdef CONFIG_TCP_MD5SIG 1813 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1814 const struct sock *addr_sk); 1815 int (*calc_md5_hash) (char *location, 1816 const struct tcp_md5sig_key *md5, 1817 const struct sock *sk, 1818 const struct sk_buff *skb); 1819 #endif 1820 void (*init_req)(struct request_sock *req, 1821 const struct sock *sk_listener, 1822 struct sk_buff *skb); 1823 #ifdef CONFIG_SYN_COOKIES 1824 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1825 __u16 *mss); 1826 #endif 1827 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1828 const struct request_sock *req, 1829 bool *strict); 1830 __u32 (*init_seq)(const struct sk_buff *skb, u32 *tsoff); 1831 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1832 struct flowi *fl, struct request_sock *req, 1833 struct tcp_fastopen_cookie *foc, 1834 enum tcp_synack_type synack_type); 1835 }; 1836 1837 #ifdef CONFIG_SYN_COOKIES 1838 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1839 const struct sock *sk, struct sk_buff *skb, 1840 __u16 *mss) 1841 { 1842 tcp_synq_overflow(sk); 1843 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1844 return ops->cookie_init_seq(skb, mss); 1845 } 1846 #else 1847 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1848 const struct sock *sk, struct sk_buff *skb, 1849 __u16 *mss) 1850 { 1851 return 0; 1852 } 1853 #endif 1854 1855 int tcpv4_offload_init(void); 1856 1857 void tcp_v4_init(void); 1858 void tcp_init(void); 1859 1860 /* tcp_recovery.c */ 1861 1862 /* Flags to enable various loss recovery features. See below */ 1863 extern int sysctl_tcp_recovery; 1864 1865 /* Use TCP RACK to detect (some) tail and retransmit losses */ 1866 #define TCP_RACK_LOST_RETRANS 0x1 1867 1868 extern int tcp_rack_mark_lost(struct sock *sk); 1869 1870 extern void tcp_rack_advance(struct tcp_sock *tp, 1871 const struct skb_mstamp *xmit_time, u8 sacked); 1872 1873 /* 1874 * Save and compile IPv4 options, return a pointer to it 1875 */ 1876 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb) 1877 { 1878 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1879 struct ip_options_rcu *dopt = NULL; 1880 1881 if (opt->optlen) { 1882 int opt_size = sizeof(*dopt) + opt->optlen; 1883 1884 dopt = kmalloc(opt_size, GFP_ATOMIC); 1885 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) { 1886 kfree(dopt); 1887 dopt = NULL; 1888 } 1889 } 1890 return dopt; 1891 } 1892 1893 /* locally generated TCP pure ACKs have skb->truesize == 2 1894 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1895 * This is much faster than dissecting the packet to find out. 1896 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1897 */ 1898 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1899 { 1900 return skb->truesize == 2; 1901 } 1902 1903 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1904 { 1905 skb->truesize = 2; 1906 } 1907 1908 static inline int tcp_inq(struct sock *sk) 1909 { 1910 struct tcp_sock *tp = tcp_sk(sk); 1911 int answ; 1912 1913 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1914 answ = 0; 1915 } else if (sock_flag(sk, SOCK_URGINLINE) || 1916 !tp->urg_data || 1917 before(tp->urg_seq, tp->copied_seq) || 1918 !before(tp->urg_seq, tp->rcv_nxt)) { 1919 1920 answ = tp->rcv_nxt - tp->copied_seq; 1921 1922 /* Subtract 1, if FIN was received */ 1923 if (answ && sock_flag(sk, SOCK_DONE)) 1924 answ--; 1925 } else { 1926 answ = tp->urg_seq - tp->copied_seq; 1927 } 1928 1929 return answ; 1930 } 1931 1932 int tcp_peek_len(struct socket *sock); 1933 1934 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 1935 { 1936 u16 segs_in; 1937 1938 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 1939 tp->segs_in += segs_in; 1940 if (skb->len > tcp_hdrlen(skb)) 1941 tp->data_segs_in += segs_in; 1942 } 1943 1944 /* 1945 * TCP listen path runs lockless. 1946 * We forced "struct sock" to be const qualified to make sure 1947 * we don't modify one of its field by mistake. 1948 * Here, we increment sk_drops which is an atomic_t, so we can safely 1949 * make sock writable again. 1950 */ 1951 static inline void tcp_listendrop(const struct sock *sk) 1952 { 1953 atomic_inc(&((struct sock *)sk)->sk_drops); 1954 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 1955 } 1956 1957 #endif /* _TCP_H */ 1958