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