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 /* Maximal number of window scale according to RFC1323 */ 82 #define TCP_MAX_WSCALE 14U 83 84 /* urg_data states */ 85 #define TCP_URG_VALID 0x0100 86 #define TCP_URG_NOTYET 0x0200 87 #define TCP_URG_READ 0x0400 88 89 #define TCP_RETR1 3 /* 90 * This is how many retries it does before it 91 * tries to figure out if the gateway is 92 * down. Minimal RFC value is 3; it corresponds 93 * to ~3sec-8min depending on RTO. 94 */ 95 96 #define TCP_RETR2 15 /* 97 * This should take at least 98 * 90 minutes to time out. 99 * RFC1122 says that the limit is 100 sec. 100 * 15 is ~13-30min depending on RTO. 101 */ 102 103 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 104 * when active opening a connection. 105 * RFC1122 says the minimum retry MUST 106 * be at least 180secs. Nevertheless 107 * this value is corresponding to 108 * 63secs of retransmission with the 109 * current initial RTO. 110 */ 111 112 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 113 * when passive opening a connection. 114 * This is corresponding to 31secs of 115 * retransmission with the current 116 * initial RTO. 117 */ 118 119 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 120 * state, about 60 seconds */ 121 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 122 /* BSD style FIN_WAIT2 deadlock breaker. 123 * It used to be 3min, new value is 60sec, 124 * to combine FIN-WAIT-2 timeout with 125 * TIME-WAIT timer. 126 */ 127 128 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 129 #if HZ >= 100 130 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 131 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 132 #else 133 #define TCP_DELACK_MIN 4U 134 #define TCP_ATO_MIN 4U 135 #endif 136 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 137 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 138 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 139 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 140 * used as a fallback RTO for the 141 * initial data transmission if no 142 * valid RTT sample has been acquired, 143 * most likely due to retrans in 3WHS. 144 */ 145 146 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 147 * for local resources. 148 */ 149 #define TCP_REO_TIMEOUT_MIN (2000) /* Min RACK reordering timeout in usec */ 150 151 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 152 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 153 #define TCP_KEEPALIVE_INTVL (75*HZ) 154 155 #define MAX_TCP_KEEPIDLE 32767 156 #define MAX_TCP_KEEPINTVL 32767 157 #define MAX_TCP_KEEPCNT 127 158 #define MAX_TCP_SYNCNT 127 159 160 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 161 162 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 163 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 164 * after this time. It should be equal 165 * (or greater than) TCP_TIMEWAIT_LEN 166 * to provide reliability equal to one 167 * provided by timewait state. 168 */ 169 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 170 * timestamps. It must be less than 171 * minimal timewait lifetime. 172 */ 173 /* 174 * TCP option 175 */ 176 177 #define TCPOPT_NOP 1 /* Padding */ 178 #define TCPOPT_EOL 0 /* End of options */ 179 #define TCPOPT_MSS 2 /* Segment size negotiating */ 180 #define TCPOPT_WINDOW 3 /* Window scaling */ 181 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 182 #define TCPOPT_SACK 5 /* SACK Block */ 183 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 184 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 185 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 186 #define TCPOPT_EXP 254 /* Experimental */ 187 /* Magic number to be after the option value for sharing TCP 188 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 189 */ 190 #define TCPOPT_FASTOPEN_MAGIC 0xF989 191 192 /* 193 * TCP option lengths 194 */ 195 196 #define TCPOLEN_MSS 4 197 #define TCPOLEN_WINDOW 3 198 #define TCPOLEN_SACK_PERM 2 199 #define TCPOLEN_TIMESTAMP 10 200 #define TCPOLEN_MD5SIG 18 201 #define TCPOLEN_FASTOPEN_BASE 2 202 #define TCPOLEN_EXP_FASTOPEN_BASE 4 203 204 /* But this is what stacks really send out. */ 205 #define TCPOLEN_TSTAMP_ALIGNED 12 206 #define TCPOLEN_WSCALE_ALIGNED 4 207 #define TCPOLEN_SACKPERM_ALIGNED 4 208 #define TCPOLEN_SACK_BASE 2 209 #define TCPOLEN_SACK_BASE_ALIGNED 4 210 #define TCPOLEN_SACK_PERBLOCK 8 211 #define TCPOLEN_MD5SIG_ALIGNED 20 212 #define TCPOLEN_MSS_ALIGNED 4 213 214 /* Flags in tp->nonagle */ 215 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 216 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 217 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 218 219 /* TCP thin-stream limits */ 220 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 221 222 /* TCP initial congestion window as per rfc6928 */ 223 #define TCP_INIT_CWND 10 224 225 /* Bit Flags for sysctl_tcp_fastopen */ 226 #define TFO_CLIENT_ENABLE 1 227 #define TFO_SERVER_ENABLE 2 228 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 229 230 /* Accept SYN data w/o any cookie option */ 231 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 232 233 /* Force enable TFO on all listeners, i.e., not requiring the 234 * TCP_FASTOPEN socket option. 235 */ 236 #define TFO_SERVER_WO_SOCKOPT1 0x400 237 238 239 /* sysctl variables for tcp */ 240 extern int sysctl_tcp_timestamps; 241 extern int sysctl_tcp_window_scaling; 242 extern int sysctl_tcp_sack; 243 extern int sysctl_tcp_fastopen; 244 extern int sysctl_tcp_retrans_collapse; 245 extern int sysctl_tcp_stdurg; 246 extern int sysctl_tcp_rfc1337; 247 extern int sysctl_tcp_abort_on_overflow; 248 extern int sysctl_tcp_max_orphans; 249 extern int sysctl_tcp_fack; 250 extern int sysctl_tcp_reordering; 251 extern int sysctl_tcp_max_reordering; 252 extern int sysctl_tcp_dsack; 253 extern long sysctl_tcp_mem[3]; 254 extern int sysctl_tcp_wmem[3]; 255 extern int sysctl_tcp_rmem[3]; 256 extern int sysctl_tcp_app_win; 257 extern int sysctl_tcp_adv_win_scale; 258 extern int sysctl_tcp_frto; 259 extern int sysctl_tcp_low_latency; 260 extern int sysctl_tcp_nometrics_save; 261 extern int sysctl_tcp_moderate_rcvbuf; 262 extern int sysctl_tcp_tso_win_divisor; 263 extern int sysctl_tcp_workaround_signed_windows; 264 extern int sysctl_tcp_slow_start_after_idle; 265 extern int sysctl_tcp_thin_linear_timeouts; 266 extern int sysctl_tcp_thin_dupack; 267 extern int sysctl_tcp_early_retrans; 268 extern int sysctl_tcp_recovery; 269 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 270 271 extern int sysctl_tcp_limit_output_bytes; 272 extern int sysctl_tcp_challenge_ack_limit; 273 extern int sysctl_tcp_min_tso_segs; 274 extern int sysctl_tcp_min_rtt_wlen; 275 extern int sysctl_tcp_autocorking; 276 extern int sysctl_tcp_invalid_ratelimit; 277 extern int sysctl_tcp_pacing_ss_ratio; 278 extern int sysctl_tcp_pacing_ca_ratio; 279 280 extern atomic_long_t tcp_memory_allocated; 281 extern struct percpu_counter tcp_sockets_allocated; 282 extern int tcp_memory_pressure; 283 284 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 285 static inline bool tcp_under_memory_pressure(const struct sock *sk) 286 { 287 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 288 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 289 return true; 290 291 return tcp_memory_pressure; 292 } 293 /* 294 * The next routines deal with comparing 32 bit unsigned ints 295 * and worry about wraparound (automatic with unsigned arithmetic). 296 */ 297 298 static inline bool before(__u32 seq1, __u32 seq2) 299 { 300 return (__s32)(seq1-seq2) < 0; 301 } 302 #define after(seq2, seq1) before(seq1, seq2) 303 304 /* is s2<=s1<=s3 ? */ 305 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 306 { 307 return seq3 - seq2 >= seq1 - seq2; 308 } 309 310 static inline bool tcp_out_of_memory(struct sock *sk) 311 { 312 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 313 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 314 return true; 315 return false; 316 } 317 318 void sk_forced_mem_schedule(struct sock *sk, int size); 319 320 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 321 { 322 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 323 int orphans = percpu_counter_read_positive(ocp); 324 325 if (orphans << shift > sysctl_tcp_max_orphans) { 326 orphans = percpu_counter_sum_positive(ocp); 327 if (orphans << shift > sysctl_tcp_max_orphans) 328 return true; 329 } 330 return false; 331 } 332 333 bool tcp_check_oom(struct sock *sk, int shift); 334 335 336 extern struct proto tcp_prot; 337 338 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 339 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 340 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 341 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 342 343 void tcp_tasklet_init(void); 344 345 void tcp_v4_err(struct sk_buff *skb, u32); 346 347 void tcp_shutdown(struct sock *sk, int how); 348 349 void tcp_v4_early_demux(struct sk_buff *skb); 350 int tcp_v4_rcv(struct sk_buff *skb); 351 352 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 353 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 354 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 355 int flags); 356 void tcp_release_cb(struct sock *sk); 357 void tcp_wfree(struct sk_buff *skb); 358 void tcp_write_timer_handler(struct sock *sk); 359 void tcp_delack_timer_handler(struct sock *sk); 360 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 361 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 362 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 363 const struct tcphdr *th, unsigned int len); 364 void tcp_rcv_space_adjust(struct sock *sk); 365 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 366 void tcp_twsk_destructor(struct sock *sk); 367 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 368 struct pipe_inode_info *pipe, size_t len, 369 unsigned int flags); 370 371 static inline void tcp_dec_quickack_mode(struct sock *sk, 372 const unsigned int pkts) 373 { 374 struct inet_connection_sock *icsk = inet_csk(sk); 375 376 if (icsk->icsk_ack.quick) { 377 if (pkts >= icsk->icsk_ack.quick) { 378 icsk->icsk_ack.quick = 0; 379 /* Leaving quickack mode we deflate ATO. */ 380 icsk->icsk_ack.ato = TCP_ATO_MIN; 381 } else 382 icsk->icsk_ack.quick -= pkts; 383 } 384 } 385 386 #define TCP_ECN_OK 1 387 #define TCP_ECN_QUEUE_CWR 2 388 #define TCP_ECN_DEMAND_CWR 4 389 #define TCP_ECN_SEEN 8 390 391 enum tcp_tw_status { 392 TCP_TW_SUCCESS = 0, 393 TCP_TW_RST = 1, 394 TCP_TW_ACK = 2, 395 TCP_TW_SYN = 3 396 }; 397 398 399 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 400 struct sk_buff *skb, 401 const struct tcphdr *th); 402 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 403 struct request_sock *req, bool fastopen); 404 int tcp_child_process(struct sock *parent, struct sock *child, 405 struct sk_buff *skb); 406 void tcp_enter_loss(struct sock *sk); 407 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag); 408 void tcp_clear_retrans(struct tcp_sock *tp); 409 void tcp_update_metrics(struct sock *sk); 410 void tcp_init_metrics(struct sock *sk); 411 void tcp_metrics_init(void); 412 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 413 void tcp_disable_fack(struct tcp_sock *tp); 414 void tcp_close(struct sock *sk, long timeout); 415 void tcp_init_sock(struct sock *sk); 416 unsigned int tcp_poll(struct file *file, struct socket *sock, 417 struct poll_table_struct *wait); 418 int tcp_getsockopt(struct sock *sk, int level, int optname, 419 char __user *optval, int __user *optlen); 420 int tcp_setsockopt(struct sock *sk, int level, int optname, 421 char __user *optval, unsigned int optlen); 422 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 423 char __user *optval, int __user *optlen); 424 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 425 char __user *optval, unsigned int optlen); 426 void tcp_set_keepalive(struct sock *sk, int val); 427 void tcp_syn_ack_timeout(const struct request_sock *req); 428 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 429 int flags, int *addr_len); 430 void tcp_parse_options(const struct sk_buff *skb, 431 struct tcp_options_received *opt_rx, 432 int estab, struct tcp_fastopen_cookie *foc); 433 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 434 435 /* 436 * TCP v4 functions exported for the inet6 API 437 */ 438 439 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 440 void tcp_v4_mtu_reduced(struct sock *sk); 441 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 442 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 443 struct sock *tcp_create_openreq_child(const struct sock *sk, 444 struct request_sock *req, 445 struct sk_buff *skb); 446 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 447 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 448 struct request_sock *req, 449 struct dst_entry *dst, 450 struct request_sock *req_unhash, 451 bool *own_req); 452 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 453 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 454 int tcp_connect(struct sock *sk); 455 enum tcp_synack_type { 456 TCP_SYNACK_NORMAL, 457 TCP_SYNACK_FASTOPEN, 458 TCP_SYNACK_COOKIE, 459 }; 460 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 461 struct request_sock *req, 462 struct tcp_fastopen_cookie *foc, 463 enum tcp_synack_type synack_type); 464 int tcp_disconnect(struct sock *sk, int flags); 465 466 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 467 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 468 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 469 470 /* From syncookies.c */ 471 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 472 struct request_sock *req, 473 struct dst_entry *dst); 474 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 475 u32 cookie); 476 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 477 #ifdef CONFIG_SYN_COOKIES 478 479 /* Syncookies use a monotonic timer which increments every 60 seconds. 480 * This counter is used both as a hash input and partially encoded into 481 * the cookie value. A cookie is only validated further if the delta 482 * between the current counter value and the encoded one is less than this, 483 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 484 * the counter advances immediately after a cookie is generated). 485 */ 486 #define MAX_SYNCOOKIE_AGE 2 487 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 488 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 489 490 /* syncookies: remember time of last synqueue overflow 491 * But do not dirty this field too often (once per second is enough) 492 * It is racy as we do not hold a lock, but race is very minor. 493 */ 494 static inline void tcp_synq_overflow(const struct sock *sk) 495 { 496 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 497 unsigned long now = jiffies; 498 499 if (time_after(now, last_overflow + HZ)) 500 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 501 } 502 503 /* syncookies: no recent synqueue overflow on this listening socket? */ 504 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 505 { 506 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 507 508 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID); 509 } 510 511 static inline u32 tcp_cookie_time(void) 512 { 513 u64 val = get_jiffies_64(); 514 515 do_div(val, TCP_SYNCOOKIE_PERIOD); 516 return val; 517 } 518 519 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 520 u16 *mssp); 521 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 522 __u32 cookie_init_timestamp(struct request_sock *req); 523 bool cookie_timestamp_decode(struct tcp_options_received *opt); 524 bool cookie_ecn_ok(const struct tcp_options_received *opt, 525 const struct net *net, const struct dst_entry *dst); 526 527 /* From net/ipv6/syncookies.c */ 528 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 529 u32 cookie); 530 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 531 532 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 533 const struct tcphdr *th, u16 *mssp); 534 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 535 #endif 536 /* tcp_output.c */ 537 538 u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now, 539 int min_tso_segs); 540 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 541 int nonagle); 542 bool tcp_may_send_now(struct sock *sk); 543 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 544 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 545 void tcp_retransmit_timer(struct sock *sk); 546 void tcp_xmit_retransmit_queue(struct sock *); 547 void tcp_simple_retransmit(struct sock *); 548 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 549 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 550 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t); 551 552 void tcp_send_probe0(struct sock *); 553 void tcp_send_partial(struct sock *); 554 int tcp_write_wakeup(struct sock *, int mib); 555 void tcp_send_fin(struct sock *sk); 556 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 557 int tcp_send_synack(struct sock *); 558 void tcp_push_one(struct sock *, unsigned int mss_now); 559 void tcp_send_ack(struct sock *sk); 560 void tcp_send_delayed_ack(struct sock *sk); 561 void tcp_send_loss_probe(struct sock *sk); 562 bool tcp_schedule_loss_probe(struct sock *sk); 563 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 564 const struct sk_buff *next_skb); 565 566 /* tcp_input.c */ 567 void tcp_rearm_rto(struct sock *sk); 568 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 569 void tcp_reset(struct sock *sk); 570 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 571 void tcp_fin(struct sock *sk); 572 573 /* tcp_timer.c */ 574 void tcp_init_xmit_timers(struct sock *); 575 static inline void tcp_clear_xmit_timers(struct sock *sk) 576 { 577 inet_csk_clear_xmit_timers(sk); 578 } 579 580 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 581 unsigned int tcp_current_mss(struct sock *sk); 582 583 /* Bound MSS / TSO packet size with the half of the window */ 584 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 585 { 586 int cutoff; 587 588 /* When peer uses tiny windows, there is no use in packetizing 589 * to sub-MSS pieces for the sake of SWS or making sure there 590 * are enough packets in the pipe for fast recovery. 591 * 592 * On the other hand, for extremely large MSS devices, handling 593 * smaller than MSS windows in this way does make sense. 594 */ 595 if (tp->max_window > TCP_MSS_DEFAULT) 596 cutoff = (tp->max_window >> 1); 597 else 598 cutoff = tp->max_window; 599 600 if (cutoff && pktsize > cutoff) 601 return max_t(int, cutoff, 68U - tp->tcp_header_len); 602 else 603 return pktsize; 604 } 605 606 /* tcp.c */ 607 void tcp_get_info(struct sock *, struct tcp_info *); 608 609 /* Read 'sendfile()'-style from a TCP socket */ 610 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 611 sk_read_actor_t recv_actor); 612 613 void tcp_initialize_rcv_mss(struct sock *sk); 614 615 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 616 int tcp_mss_to_mtu(struct sock *sk, int mss); 617 void tcp_mtup_init(struct sock *sk); 618 void tcp_init_buffer_space(struct sock *sk); 619 620 static inline void tcp_bound_rto(const struct sock *sk) 621 { 622 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 623 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 624 } 625 626 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 627 { 628 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 629 } 630 631 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 632 { 633 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 634 ntohl(TCP_FLAG_ACK) | 635 snd_wnd); 636 } 637 638 static inline void tcp_fast_path_on(struct tcp_sock *tp) 639 { 640 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 641 } 642 643 static inline void tcp_fast_path_check(struct sock *sk) 644 { 645 struct tcp_sock *tp = tcp_sk(sk); 646 647 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 648 tp->rcv_wnd && 649 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 650 !tp->urg_data) 651 tcp_fast_path_on(tp); 652 } 653 654 /* Compute the actual rto_min value */ 655 static inline u32 tcp_rto_min(struct sock *sk) 656 { 657 const struct dst_entry *dst = __sk_dst_get(sk); 658 u32 rto_min = TCP_RTO_MIN; 659 660 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 661 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 662 return rto_min; 663 } 664 665 static inline u32 tcp_rto_min_us(struct sock *sk) 666 { 667 return jiffies_to_usecs(tcp_rto_min(sk)); 668 } 669 670 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 671 { 672 return dst_metric_locked(dst, RTAX_CC_ALGO); 673 } 674 675 /* Minimum RTT in usec. ~0 means not available. */ 676 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 677 { 678 return minmax_get(&tp->rtt_min); 679 } 680 681 /* Compute the actual receive window we are currently advertising. 682 * Rcv_nxt can be after the window if our peer push more data 683 * than the offered window. 684 */ 685 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 686 { 687 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 688 689 if (win < 0) 690 win = 0; 691 return (u32) win; 692 } 693 694 /* Choose a new window, without checks for shrinking, and without 695 * scaling applied to the result. The caller does these things 696 * if necessary. This is a "raw" window selection. 697 */ 698 u32 __tcp_select_window(struct sock *sk); 699 700 void tcp_send_window_probe(struct sock *sk); 701 702 /* TCP timestamps are only 32-bits, this causes a slight 703 * complication on 64-bit systems since we store a snapshot 704 * of jiffies in the buffer control blocks below. We decided 705 * to use only the low 32-bits of jiffies and hide the ugly 706 * casts with the following macro. 707 */ 708 #define tcp_time_stamp ((__u32)(jiffies)) 709 710 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 711 { 712 return skb->skb_mstamp.stamp_jiffies; 713 } 714 715 716 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 717 718 #define TCPHDR_FIN 0x01 719 #define TCPHDR_SYN 0x02 720 #define TCPHDR_RST 0x04 721 #define TCPHDR_PSH 0x08 722 #define TCPHDR_ACK 0x10 723 #define TCPHDR_URG 0x20 724 #define TCPHDR_ECE 0x40 725 #define TCPHDR_CWR 0x80 726 727 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 728 729 /* This is what the send packet queuing engine uses to pass 730 * TCP per-packet control information to the transmission code. 731 * We also store the host-order sequence numbers in here too. 732 * This is 44 bytes if IPV6 is enabled. 733 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 734 */ 735 struct tcp_skb_cb { 736 __u32 seq; /* Starting sequence number */ 737 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 738 union { 739 /* Note : tcp_tw_isn is used in input path only 740 * (isn chosen by tcp_timewait_state_process()) 741 * 742 * tcp_gso_segs/size are used in write queue only, 743 * cf tcp_skb_pcount()/tcp_skb_mss() 744 */ 745 __u32 tcp_tw_isn; 746 struct { 747 u16 tcp_gso_segs; 748 u16 tcp_gso_size; 749 }; 750 }; 751 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 752 753 __u8 sacked; /* State flags for SACK/FACK. */ 754 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 755 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 756 #define TCPCB_LOST 0x04 /* SKB is lost */ 757 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 758 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 759 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 760 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 761 TCPCB_REPAIRED) 762 763 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 764 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 765 eor:1, /* Is skb MSG_EOR marked? */ 766 unused:6; 767 __u32 ack_seq; /* Sequence number ACK'd */ 768 union { 769 struct { 770 /* There is space for up to 24 bytes */ 771 __u32 in_flight:30,/* Bytes in flight at transmit */ 772 is_app_limited:1, /* cwnd not fully used? */ 773 unused:1; 774 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 775 __u32 delivered; 776 /* start of send pipeline phase */ 777 struct skb_mstamp first_tx_mstamp; 778 /* when we reached the "delivered" count */ 779 struct skb_mstamp delivered_mstamp; 780 } tx; /* only used for outgoing skbs */ 781 union { 782 struct inet_skb_parm h4; 783 #if IS_ENABLED(CONFIG_IPV6) 784 struct inet6_skb_parm h6; 785 #endif 786 } header; /* For incoming skbs */ 787 }; 788 }; 789 790 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 791 792 793 #if IS_ENABLED(CONFIG_IPV6) 794 /* This is the variant of inet6_iif() that must be used by TCP, 795 * as TCP moves IP6CB into a different location in skb->cb[] 796 */ 797 static inline int tcp_v6_iif(const struct sk_buff *skb) 798 { 799 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 800 801 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 802 } 803 #endif 804 805 /* TCP_SKB_CB reference means this can not be used from early demux */ 806 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 807 { 808 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 809 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 810 skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 811 return true; 812 #endif 813 return false; 814 } 815 816 /* Due to TSO, an SKB can be composed of multiple actual 817 * packets. To keep these tracked properly, we use this. 818 */ 819 static inline int tcp_skb_pcount(const struct sk_buff *skb) 820 { 821 return TCP_SKB_CB(skb)->tcp_gso_segs; 822 } 823 824 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 825 { 826 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 827 } 828 829 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 830 { 831 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 832 } 833 834 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 835 static inline int tcp_skb_mss(const struct sk_buff *skb) 836 { 837 return TCP_SKB_CB(skb)->tcp_gso_size; 838 } 839 840 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 841 { 842 return likely(!TCP_SKB_CB(skb)->eor); 843 } 844 845 /* Events passed to congestion control interface */ 846 enum tcp_ca_event { 847 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 848 CA_EVENT_CWND_RESTART, /* congestion window restart */ 849 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 850 CA_EVENT_LOSS, /* loss timeout */ 851 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 852 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 853 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 854 CA_EVENT_NON_DELAYED_ACK, 855 }; 856 857 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 858 enum tcp_ca_ack_event_flags { 859 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 860 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 861 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 862 }; 863 864 /* 865 * Interface for adding new TCP congestion control handlers 866 */ 867 #define TCP_CA_NAME_MAX 16 868 #define TCP_CA_MAX 128 869 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 870 871 #define TCP_CA_UNSPEC 0 872 873 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 874 #define TCP_CONG_NON_RESTRICTED 0x1 875 /* Requires ECN/ECT set on all packets */ 876 #define TCP_CONG_NEEDS_ECN 0x2 877 878 union tcp_cc_info; 879 880 struct ack_sample { 881 u32 pkts_acked; 882 s32 rtt_us; 883 u32 in_flight; 884 }; 885 886 /* A rate sample measures the number of (original/retransmitted) data 887 * packets delivered "delivered" over an interval of time "interval_us". 888 * The tcp_rate.c code fills in the rate sample, and congestion 889 * control modules that define a cong_control function to run at the end 890 * of ACK processing can optionally chose to consult this sample when 891 * setting cwnd and pacing rate. 892 * A sample is invalid if "delivered" or "interval_us" is negative. 893 */ 894 struct rate_sample { 895 struct skb_mstamp prior_mstamp; /* starting timestamp for interval */ 896 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 897 s32 delivered; /* number of packets delivered over interval */ 898 long interval_us; /* time for tp->delivered to incr "delivered" */ 899 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 900 int losses; /* number of packets marked lost upon ACK */ 901 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 902 u32 prior_in_flight; /* in flight before this ACK */ 903 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 904 bool is_retrans; /* is sample from retransmission? */ 905 }; 906 907 struct tcp_congestion_ops { 908 struct list_head list; 909 u32 key; 910 u32 flags; 911 912 /* initialize private data (optional) */ 913 void (*init)(struct sock *sk); 914 /* cleanup private data (optional) */ 915 void (*release)(struct sock *sk); 916 917 /* return slow start threshold (required) */ 918 u32 (*ssthresh)(struct sock *sk); 919 /* do new cwnd calculation (required) */ 920 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 921 /* call before changing ca_state (optional) */ 922 void (*set_state)(struct sock *sk, u8 new_state); 923 /* call when cwnd event occurs (optional) */ 924 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 925 /* call when ack arrives (optional) */ 926 void (*in_ack_event)(struct sock *sk, u32 flags); 927 /* new value of cwnd after loss (optional) */ 928 u32 (*undo_cwnd)(struct sock *sk); 929 /* hook for packet ack accounting (optional) */ 930 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 931 /* suggest number of segments for each skb to transmit (optional) */ 932 u32 (*tso_segs_goal)(struct sock *sk); 933 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 934 u32 (*sndbuf_expand)(struct sock *sk); 935 /* call when packets are delivered to update cwnd and pacing rate, 936 * after all the ca_state processing. (optional) 937 */ 938 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 939 /* get info for inet_diag (optional) */ 940 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 941 union tcp_cc_info *info); 942 943 char name[TCP_CA_NAME_MAX]; 944 struct module *owner; 945 }; 946 947 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 948 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 949 950 void tcp_assign_congestion_control(struct sock *sk); 951 void tcp_init_congestion_control(struct sock *sk); 952 void tcp_cleanup_congestion_control(struct sock *sk); 953 int tcp_set_default_congestion_control(const char *name); 954 void tcp_get_default_congestion_control(char *name); 955 void tcp_get_available_congestion_control(char *buf, size_t len); 956 void tcp_get_allowed_congestion_control(char *buf, size_t len); 957 int tcp_set_allowed_congestion_control(char *allowed); 958 int tcp_set_congestion_control(struct sock *sk, const char *name); 959 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 960 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 961 962 u32 tcp_reno_ssthresh(struct sock *sk); 963 u32 tcp_reno_undo_cwnd(struct sock *sk); 964 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 965 extern struct tcp_congestion_ops tcp_reno; 966 967 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 968 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca); 969 #ifdef CONFIG_INET 970 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 971 #else 972 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 973 { 974 return NULL; 975 } 976 #endif 977 978 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 979 { 980 const struct inet_connection_sock *icsk = inet_csk(sk); 981 982 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 983 } 984 985 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 986 { 987 struct inet_connection_sock *icsk = inet_csk(sk); 988 989 if (icsk->icsk_ca_ops->set_state) 990 icsk->icsk_ca_ops->set_state(sk, ca_state); 991 icsk->icsk_ca_state = ca_state; 992 } 993 994 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 995 { 996 const struct inet_connection_sock *icsk = inet_csk(sk); 997 998 if (icsk->icsk_ca_ops->cwnd_event) 999 icsk->icsk_ca_ops->cwnd_event(sk, event); 1000 } 1001 1002 /* From tcp_rate.c */ 1003 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1004 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1005 struct rate_sample *rs); 1006 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1007 struct rate_sample *rs); 1008 void tcp_rate_check_app_limited(struct sock *sk); 1009 1010 /* These functions determine how the current flow behaves in respect of SACK 1011 * handling. SACK is negotiated with the peer, and therefore it can vary 1012 * between different flows. 1013 * 1014 * tcp_is_sack - SACK enabled 1015 * tcp_is_reno - No SACK 1016 * tcp_is_fack - FACK enabled, implies SACK enabled 1017 */ 1018 static inline int tcp_is_sack(const struct tcp_sock *tp) 1019 { 1020 return tp->rx_opt.sack_ok; 1021 } 1022 1023 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1024 { 1025 return !tcp_is_sack(tp); 1026 } 1027 1028 static inline bool tcp_is_fack(const struct tcp_sock *tp) 1029 { 1030 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 1031 } 1032 1033 static inline void tcp_enable_fack(struct tcp_sock *tp) 1034 { 1035 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 1036 } 1037 1038 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1039 { 1040 return tp->sacked_out + tp->lost_out; 1041 } 1042 1043 /* This determines how many packets are "in the network" to the best 1044 * of our knowledge. In many cases it is conservative, but where 1045 * detailed information is available from the receiver (via SACK 1046 * blocks etc.) we can make more aggressive calculations. 1047 * 1048 * Use this for decisions involving congestion control, use just 1049 * tp->packets_out to determine if the send queue is empty or not. 1050 * 1051 * Read this equation as: 1052 * 1053 * "Packets sent once on transmission queue" MINUS 1054 * "Packets left network, but not honestly ACKed yet" PLUS 1055 * "Packets fast retransmitted" 1056 */ 1057 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1058 { 1059 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1060 } 1061 1062 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1063 1064 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1065 { 1066 return tp->snd_cwnd < tp->snd_ssthresh; 1067 } 1068 1069 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1070 { 1071 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1072 } 1073 1074 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1075 { 1076 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1077 (1 << inet_csk(sk)->icsk_ca_state); 1078 } 1079 1080 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1081 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1082 * ssthresh. 1083 */ 1084 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1085 { 1086 const struct tcp_sock *tp = tcp_sk(sk); 1087 1088 if (tcp_in_cwnd_reduction(sk)) 1089 return tp->snd_ssthresh; 1090 else 1091 return max(tp->snd_ssthresh, 1092 ((tp->snd_cwnd >> 1) + 1093 (tp->snd_cwnd >> 2))); 1094 } 1095 1096 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1097 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1098 1099 void tcp_enter_cwr(struct sock *sk); 1100 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1101 1102 /* The maximum number of MSS of available cwnd for which TSO defers 1103 * sending if not using sysctl_tcp_tso_win_divisor. 1104 */ 1105 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1106 { 1107 return 3; 1108 } 1109 1110 /* Returns end sequence number of the receiver's advertised window */ 1111 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1112 { 1113 return tp->snd_una + tp->snd_wnd; 1114 } 1115 1116 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1117 * flexible approach. The RFC suggests cwnd should not be raised unless 1118 * it was fully used previously. And that's exactly what we do in 1119 * congestion avoidance mode. But in slow start we allow cwnd to grow 1120 * as long as the application has used half the cwnd. 1121 * Example : 1122 * cwnd is 10 (IW10), but application sends 9 frames. 1123 * We allow cwnd to reach 18 when all frames are ACKed. 1124 * This check is safe because it's as aggressive as slow start which already 1125 * risks 100% overshoot. The advantage is that we discourage application to 1126 * either send more filler packets or data to artificially blow up the cwnd 1127 * usage, and allow application-limited process to probe bw more aggressively. 1128 */ 1129 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1130 { 1131 const struct tcp_sock *tp = tcp_sk(sk); 1132 1133 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1134 if (tcp_in_slow_start(tp)) 1135 return tp->snd_cwnd < 2 * tp->max_packets_out; 1136 1137 return tp->is_cwnd_limited; 1138 } 1139 1140 /* Something is really bad, we could not queue an additional packet, 1141 * because qdisc is full or receiver sent a 0 window. 1142 * We do not want to add fuel to the fire, or abort too early, 1143 * so make sure the timer we arm now is at least 200ms in the future, 1144 * regardless of current icsk_rto value (as it could be ~2ms) 1145 */ 1146 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1147 { 1148 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1149 } 1150 1151 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1152 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1153 unsigned long max_when) 1154 { 1155 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1156 1157 return (unsigned long)min_t(u64, when, max_when); 1158 } 1159 1160 static inline void tcp_check_probe_timer(struct sock *sk) 1161 { 1162 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1163 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1164 tcp_probe0_base(sk), TCP_RTO_MAX); 1165 } 1166 1167 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1168 { 1169 tp->snd_wl1 = seq; 1170 } 1171 1172 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1173 { 1174 tp->snd_wl1 = seq; 1175 } 1176 1177 /* 1178 * Calculate(/check) TCP checksum 1179 */ 1180 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1181 __be32 daddr, __wsum base) 1182 { 1183 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1184 } 1185 1186 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1187 { 1188 return __skb_checksum_complete(skb); 1189 } 1190 1191 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1192 { 1193 return !skb_csum_unnecessary(skb) && 1194 __tcp_checksum_complete(skb); 1195 } 1196 1197 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1198 1199 static inline void tcp_prequeue_init(struct tcp_sock *tp) 1200 { 1201 tp->ucopy.task = NULL; 1202 tp->ucopy.len = 0; 1203 tp->ucopy.memory = 0; 1204 skb_queue_head_init(&tp->ucopy.prequeue); 1205 } 1206 1207 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1208 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1209 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1210 1211 #undef STATE_TRACE 1212 1213 #ifdef STATE_TRACE 1214 static const char *statename[]={ 1215 "Unused","Established","Syn Sent","Syn Recv", 1216 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1217 "Close Wait","Last ACK","Listen","Closing" 1218 }; 1219 #endif 1220 void tcp_set_state(struct sock *sk, int state); 1221 1222 void tcp_done(struct sock *sk); 1223 1224 int tcp_abort(struct sock *sk, int err); 1225 1226 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1227 { 1228 rx_opt->dsack = 0; 1229 rx_opt->num_sacks = 0; 1230 } 1231 1232 u32 tcp_default_init_rwnd(u32 mss); 1233 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1234 1235 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1236 { 1237 struct tcp_sock *tp = tcp_sk(sk); 1238 s32 delta; 1239 1240 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out) 1241 return; 1242 delta = tcp_time_stamp - tp->lsndtime; 1243 if (delta > inet_csk(sk)->icsk_rto) 1244 tcp_cwnd_restart(sk, delta); 1245 } 1246 1247 /* Determine a window scaling and initial window to offer. */ 1248 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1249 __u32 *window_clamp, int wscale_ok, 1250 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1251 1252 static inline int tcp_win_from_space(int space) 1253 { 1254 int tcp_adv_win_scale = sysctl_tcp_adv_win_scale; 1255 1256 return tcp_adv_win_scale <= 0 ? 1257 (space>>(-tcp_adv_win_scale)) : 1258 space - (space>>tcp_adv_win_scale); 1259 } 1260 1261 /* Note: caller must be prepared to deal with negative returns */ 1262 static inline int tcp_space(const struct sock *sk) 1263 { 1264 return tcp_win_from_space(sk->sk_rcvbuf - 1265 atomic_read(&sk->sk_rmem_alloc)); 1266 } 1267 1268 static inline int tcp_full_space(const struct sock *sk) 1269 { 1270 return tcp_win_from_space(sk->sk_rcvbuf); 1271 } 1272 1273 extern void tcp_openreq_init_rwin(struct request_sock *req, 1274 const struct sock *sk_listener, 1275 const struct dst_entry *dst); 1276 1277 void tcp_enter_memory_pressure(struct sock *sk); 1278 1279 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1280 { 1281 struct net *net = sock_net((struct sock *)tp); 1282 1283 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1284 } 1285 1286 static inline int keepalive_time_when(const struct tcp_sock *tp) 1287 { 1288 struct net *net = sock_net((struct sock *)tp); 1289 1290 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1291 } 1292 1293 static inline int keepalive_probes(const struct tcp_sock *tp) 1294 { 1295 struct net *net = sock_net((struct sock *)tp); 1296 1297 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1298 } 1299 1300 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1301 { 1302 const struct inet_connection_sock *icsk = &tp->inet_conn; 1303 1304 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 1305 tcp_time_stamp - tp->rcv_tstamp); 1306 } 1307 1308 static inline int tcp_fin_time(const struct sock *sk) 1309 { 1310 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1311 const int rto = inet_csk(sk)->icsk_rto; 1312 1313 if (fin_timeout < (rto << 2) - (rto >> 1)) 1314 fin_timeout = (rto << 2) - (rto >> 1); 1315 1316 return fin_timeout; 1317 } 1318 1319 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1320 int paws_win) 1321 { 1322 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1323 return true; 1324 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1325 return true; 1326 /* 1327 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1328 * then following tcp messages have valid values. Ignore 0 value, 1329 * or else 'negative' tsval might forbid us to accept their packets. 1330 */ 1331 if (!rx_opt->ts_recent) 1332 return true; 1333 return false; 1334 } 1335 1336 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1337 int rst) 1338 { 1339 if (tcp_paws_check(rx_opt, 0)) 1340 return false; 1341 1342 /* RST segments are not recommended to carry timestamp, 1343 and, if they do, it is recommended to ignore PAWS because 1344 "their cleanup function should take precedence over timestamps." 1345 Certainly, it is mistake. It is necessary to understand the reasons 1346 of this constraint to relax it: if peer reboots, clock may go 1347 out-of-sync and half-open connections will not be reset. 1348 Actually, the problem would be not existing if all 1349 the implementations followed draft about maintaining clock 1350 via reboots. Linux-2.2 DOES NOT! 1351 1352 However, we can relax time bounds for RST segments to MSL. 1353 */ 1354 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1355 return false; 1356 return true; 1357 } 1358 1359 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1360 int mib_idx, u32 *last_oow_ack_time); 1361 1362 static inline void tcp_mib_init(struct net *net) 1363 { 1364 /* See RFC 2012 */ 1365 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1366 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1367 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1368 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1369 } 1370 1371 /* from STCP */ 1372 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1373 { 1374 tp->lost_skb_hint = NULL; 1375 } 1376 1377 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1378 { 1379 tcp_clear_retrans_hints_partial(tp); 1380 tp->retransmit_skb_hint = NULL; 1381 } 1382 1383 union tcp_md5_addr { 1384 struct in_addr a4; 1385 #if IS_ENABLED(CONFIG_IPV6) 1386 struct in6_addr a6; 1387 #endif 1388 }; 1389 1390 /* - key database */ 1391 struct tcp_md5sig_key { 1392 struct hlist_node node; 1393 u8 keylen; 1394 u8 family; /* AF_INET or AF_INET6 */ 1395 union tcp_md5_addr addr; 1396 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1397 struct rcu_head rcu; 1398 }; 1399 1400 /* - sock block */ 1401 struct tcp_md5sig_info { 1402 struct hlist_head head; 1403 struct rcu_head rcu; 1404 }; 1405 1406 /* - pseudo header */ 1407 struct tcp4_pseudohdr { 1408 __be32 saddr; 1409 __be32 daddr; 1410 __u8 pad; 1411 __u8 protocol; 1412 __be16 len; 1413 }; 1414 1415 struct tcp6_pseudohdr { 1416 struct in6_addr saddr; 1417 struct in6_addr daddr; 1418 __be32 len; 1419 __be32 protocol; /* including padding */ 1420 }; 1421 1422 union tcp_md5sum_block { 1423 struct tcp4_pseudohdr ip4; 1424 #if IS_ENABLED(CONFIG_IPV6) 1425 struct tcp6_pseudohdr ip6; 1426 #endif 1427 }; 1428 1429 /* - pool: digest algorithm, hash description and scratch buffer */ 1430 struct tcp_md5sig_pool { 1431 struct ahash_request *md5_req; 1432 void *scratch; 1433 }; 1434 1435 /* - functions */ 1436 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1437 const struct sock *sk, const struct sk_buff *skb); 1438 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1439 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); 1440 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1441 int family); 1442 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1443 const struct sock *addr_sk); 1444 1445 #ifdef CONFIG_TCP_MD5SIG 1446 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1447 const union tcp_md5_addr *addr, 1448 int family); 1449 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1450 #else 1451 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1452 const union tcp_md5_addr *addr, 1453 int family) 1454 { 1455 return NULL; 1456 } 1457 #define tcp_twsk_md5_key(twsk) NULL 1458 #endif 1459 1460 bool tcp_alloc_md5sig_pool(void); 1461 1462 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1463 static inline void tcp_put_md5sig_pool(void) 1464 { 1465 local_bh_enable(); 1466 } 1467 1468 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1469 unsigned int header_len); 1470 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1471 const struct tcp_md5sig_key *key); 1472 1473 /* From tcp_fastopen.c */ 1474 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1475 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1476 unsigned long *last_syn_loss); 1477 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1478 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1479 u16 try_exp); 1480 struct tcp_fastopen_request { 1481 /* Fast Open cookie. Size 0 means a cookie request */ 1482 struct tcp_fastopen_cookie cookie; 1483 struct msghdr *data; /* data in MSG_FASTOPEN */ 1484 size_t size; 1485 int copied; /* queued in tcp_connect() */ 1486 }; 1487 void tcp_free_fastopen_req(struct tcp_sock *tp); 1488 1489 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1490 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1491 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1492 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1493 struct request_sock *req, 1494 struct tcp_fastopen_cookie *foc, 1495 struct dst_entry *dst); 1496 void tcp_fastopen_init_key_once(bool publish); 1497 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1498 struct tcp_fastopen_cookie *cookie); 1499 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1500 #define TCP_FASTOPEN_KEY_LENGTH 16 1501 1502 /* Fastopen key context */ 1503 struct tcp_fastopen_context { 1504 struct crypto_cipher *tfm; 1505 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1506 struct rcu_head rcu; 1507 }; 1508 1509 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1510 void tcp_fastopen_active_disable(struct sock *sk); 1511 bool tcp_fastopen_active_should_disable(struct sock *sk); 1512 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1513 void tcp_fastopen_active_timeout_reset(void); 1514 1515 /* Latencies incurred by various limits for a sender. They are 1516 * chronograph-like stats that are mutually exclusive. 1517 */ 1518 enum tcp_chrono { 1519 TCP_CHRONO_UNSPEC, 1520 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1521 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1522 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1523 __TCP_CHRONO_MAX, 1524 }; 1525 1526 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1527 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1528 1529 /* write queue abstraction */ 1530 static inline void tcp_write_queue_purge(struct sock *sk) 1531 { 1532 struct sk_buff *skb; 1533 1534 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1535 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1536 sk_wmem_free_skb(sk, skb); 1537 sk_mem_reclaim(sk); 1538 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1539 } 1540 1541 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1542 { 1543 return skb_peek(&sk->sk_write_queue); 1544 } 1545 1546 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1547 { 1548 return skb_peek_tail(&sk->sk_write_queue); 1549 } 1550 1551 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1552 const struct sk_buff *skb) 1553 { 1554 return skb_queue_next(&sk->sk_write_queue, skb); 1555 } 1556 1557 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1558 const struct sk_buff *skb) 1559 { 1560 return skb_queue_prev(&sk->sk_write_queue, skb); 1561 } 1562 1563 #define tcp_for_write_queue(skb, sk) \ 1564 skb_queue_walk(&(sk)->sk_write_queue, skb) 1565 1566 #define tcp_for_write_queue_from(skb, sk) \ 1567 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1568 1569 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1570 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1571 1572 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1573 { 1574 return sk->sk_send_head; 1575 } 1576 1577 static inline bool tcp_skb_is_last(const struct sock *sk, 1578 const struct sk_buff *skb) 1579 { 1580 return skb_queue_is_last(&sk->sk_write_queue, skb); 1581 } 1582 1583 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1584 { 1585 if (tcp_skb_is_last(sk, skb)) 1586 sk->sk_send_head = NULL; 1587 else 1588 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1589 } 1590 1591 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1592 { 1593 if (sk->sk_send_head == skb_unlinked) { 1594 sk->sk_send_head = NULL; 1595 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1596 } 1597 if (tcp_sk(sk)->highest_sack == skb_unlinked) 1598 tcp_sk(sk)->highest_sack = NULL; 1599 } 1600 1601 static inline void tcp_init_send_head(struct sock *sk) 1602 { 1603 sk->sk_send_head = NULL; 1604 } 1605 1606 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1607 { 1608 __skb_queue_tail(&sk->sk_write_queue, skb); 1609 } 1610 1611 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1612 { 1613 __tcp_add_write_queue_tail(sk, skb); 1614 1615 /* Queue it, remembering where we must start sending. */ 1616 if (sk->sk_send_head == NULL) { 1617 sk->sk_send_head = skb; 1618 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1619 1620 if (tcp_sk(sk)->highest_sack == NULL) 1621 tcp_sk(sk)->highest_sack = skb; 1622 } 1623 } 1624 1625 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1626 { 1627 __skb_queue_head(&sk->sk_write_queue, skb); 1628 } 1629 1630 /* Insert buff after skb on the write queue of sk. */ 1631 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1632 struct sk_buff *buff, 1633 struct sock *sk) 1634 { 1635 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1636 } 1637 1638 /* Insert new before skb on the write queue of sk. */ 1639 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1640 struct sk_buff *skb, 1641 struct sock *sk) 1642 { 1643 __skb_queue_before(&sk->sk_write_queue, skb, new); 1644 1645 if (sk->sk_send_head == skb) 1646 sk->sk_send_head = new; 1647 } 1648 1649 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1650 { 1651 __skb_unlink(skb, &sk->sk_write_queue); 1652 } 1653 1654 static inline bool tcp_write_queue_empty(struct sock *sk) 1655 { 1656 return skb_queue_empty(&sk->sk_write_queue); 1657 } 1658 1659 static inline void tcp_push_pending_frames(struct sock *sk) 1660 { 1661 if (tcp_send_head(sk)) { 1662 struct tcp_sock *tp = tcp_sk(sk); 1663 1664 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1665 } 1666 } 1667 1668 /* Start sequence of the skb just after the highest skb with SACKed 1669 * bit, valid only if sacked_out > 0 or when the caller has ensured 1670 * validity by itself. 1671 */ 1672 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1673 { 1674 if (!tp->sacked_out) 1675 return tp->snd_una; 1676 1677 if (tp->highest_sack == NULL) 1678 return tp->snd_nxt; 1679 1680 return TCP_SKB_CB(tp->highest_sack)->seq; 1681 } 1682 1683 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1684 { 1685 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1686 tcp_write_queue_next(sk, skb); 1687 } 1688 1689 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1690 { 1691 return tcp_sk(sk)->highest_sack; 1692 } 1693 1694 static inline void tcp_highest_sack_reset(struct sock *sk) 1695 { 1696 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1697 } 1698 1699 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1700 static inline void tcp_highest_sack_combine(struct sock *sk, 1701 struct sk_buff *old, 1702 struct sk_buff *new) 1703 { 1704 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1705 tcp_sk(sk)->highest_sack = new; 1706 } 1707 1708 /* This helper checks if socket has IP_TRANSPARENT set */ 1709 static inline bool inet_sk_transparent(const struct sock *sk) 1710 { 1711 switch (sk->sk_state) { 1712 case TCP_TIME_WAIT: 1713 return inet_twsk(sk)->tw_transparent; 1714 case TCP_NEW_SYN_RECV: 1715 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1716 } 1717 return inet_sk(sk)->transparent; 1718 } 1719 1720 /* Determines whether this is a thin stream (which may suffer from 1721 * increased latency). Used to trigger latency-reducing mechanisms. 1722 */ 1723 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1724 { 1725 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1726 } 1727 1728 /* /proc */ 1729 enum tcp_seq_states { 1730 TCP_SEQ_STATE_LISTENING, 1731 TCP_SEQ_STATE_ESTABLISHED, 1732 }; 1733 1734 int tcp_seq_open(struct inode *inode, struct file *file); 1735 1736 struct tcp_seq_afinfo { 1737 char *name; 1738 sa_family_t family; 1739 const struct file_operations *seq_fops; 1740 struct seq_operations seq_ops; 1741 }; 1742 1743 struct tcp_iter_state { 1744 struct seq_net_private p; 1745 sa_family_t family; 1746 enum tcp_seq_states state; 1747 struct sock *syn_wait_sk; 1748 int bucket, offset, sbucket, num; 1749 loff_t last_pos; 1750 }; 1751 1752 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1753 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1754 1755 extern struct request_sock_ops tcp_request_sock_ops; 1756 extern struct request_sock_ops tcp6_request_sock_ops; 1757 1758 void tcp_v4_destroy_sock(struct sock *sk); 1759 1760 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1761 netdev_features_t features); 1762 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1763 int tcp_gro_complete(struct sk_buff *skb); 1764 1765 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1766 1767 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1768 { 1769 struct net *net = sock_net((struct sock *)tp); 1770 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1771 } 1772 1773 static inline bool tcp_stream_memory_free(const struct sock *sk) 1774 { 1775 const struct tcp_sock *tp = tcp_sk(sk); 1776 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1777 1778 return notsent_bytes < tcp_notsent_lowat(tp); 1779 } 1780 1781 #ifdef CONFIG_PROC_FS 1782 int tcp4_proc_init(void); 1783 void tcp4_proc_exit(void); 1784 #endif 1785 1786 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1787 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1788 const struct tcp_request_sock_ops *af_ops, 1789 struct sock *sk, struct sk_buff *skb); 1790 1791 /* TCP af-specific functions */ 1792 struct tcp_sock_af_ops { 1793 #ifdef CONFIG_TCP_MD5SIG 1794 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1795 const struct sock *addr_sk); 1796 int (*calc_md5_hash)(char *location, 1797 const struct tcp_md5sig_key *md5, 1798 const struct sock *sk, 1799 const struct sk_buff *skb); 1800 int (*md5_parse)(struct sock *sk, 1801 char __user *optval, 1802 int optlen); 1803 #endif 1804 }; 1805 1806 struct tcp_request_sock_ops { 1807 u16 mss_clamp; 1808 #ifdef CONFIG_TCP_MD5SIG 1809 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1810 const struct sock *addr_sk); 1811 int (*calc_md5_hash) (char *location, 1812 const struct tcp_md5sig_key *md5, 1813 const struct sock *sk, 1814 const struct sk_buff *skb); 1815 #endif 1816 void (*init_req)(struct request_sock *req, 1817 const struct sock *sk_listener, 1818 struct sk_buff *skb); 1819 #ifdef CONFIG_SYN_COOKIES 1820 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1821 __u16 *mss); 1822 #endif 1823 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1824 const struct request_sock *req); 1825 __u32 (*init_seq_tsoff)(const struct sk_buff *skb, u32 *tsoff); 1826 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1827 struct flowi *fl, struct request_sock *req, 1828 struct tcp_fastopen_cookie *foc, 1829 enum tcp_synack_type synack_type); 1830 }; 1831 1832 #ifdef CONFIG_SYN_COOKIES 1833 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1834 const struct sock *sk, struct sk_buff *skb, 1835 __u16 *mss) 1836 { 1837 tcp_synq_overflow(sk); 1838 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1839 return ops->cookie_init_seq(skb, mss); 1840 } 1841 #else 1842 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1843 const struct sock *sk, struct sk_buff *skb, 1844 __u16 *mss) 1845 { 1846 return 0; 1847 } 1848 #endif 1849 1850 int tcpv4_offload_init(void); 1851 1852 void tcp_v4_init(void); 1853 void tcp_init(void); 1854 1855 /* tcp_recovery.c */ 1856 extern void tcp_rack_mark_lost(struct sock *sk); 1857 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 1858 const struct skb_mstamp *xmit_time); 1859 extern void tcp_rack_reo_timeout(struct sock *sk); 1860 1861 /* 1862 * Save and compile IPv4 options, return a pointer to it 1863 */ 1864 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb) 1865 { 1866 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1867 struct ip_options_rcu *dopt = NULL; 1868 1869 if (opt->optlen) { 1870 int opt_size = sizeof(*dopt) + opt->optlen; 1871 1872 dopt = kmalloc(opt_size, GFP_ATOMIC); 1873 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) { 1874 kfree(dopt); 1875 dopt = NULL; 1876 } 1877 } 1878 return dopt; 1879 } 1880 1881 /* locally generated TCP pure ACKs have skb->truesize == 2 1882 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1883 * This is much faster than dissecting the packet to find out. 1884 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1885 */ 1886 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1887 { 1888 return skb->truesize == 2; 1889 } 1890 1891 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1892 { 1893 skb->truesize = 2; 1894 } 1895 1896 static inline int tcp_inq(struct sock *sk) 1897 { 1898 struct tcp_sock *tp = tcp_sk(sk); 1899 int answ; 1900 1901 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1902 answ = 0; 1903 } else if (sock_flag(sk, SOCK_URGINLINE) || 1904 !tp->urg_data || 1905 before(tp->urg_seq, tp->copied_seq) || 1906 !before(tp->urg_seq, tp->rcv_nxt)) { 1907 1908 answ = tp->rcv_nxt - tp->copied_seq; 1909 1910 /* Subtract 1, if FIN was received */ 1911 if (answ && sock_flag(sk, SOCK_DONE)) 1912 answ--; 1913 } else { 1914 answ = tp->urg_seq - tp->copied_seq; 1915 } 1916 1917 return answ; 1918 } 1919 1920 int tcp_peek_len(struct socket *sock); 1921 1922 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 1923 { 1924 u16 segs_in; 1925 1926 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 1927 tp->segs_in += segs_in; 1928 if (skb->len > tcp_hdrlen(skb)) 1929 tp->data_segs_in += segs_in; 1930 } 1931 1932 /* 1933 * TCP listen path runs lockless. 1934 * We forced "struct sock" to be const qualified to make sure 1935 * we don't modify one of its field by mistake. 1936 * Here, we increment sk_drops which is an atomic_t, so we can safely 1937 * make sock writable again. 1938 */ 1939 static inline void tcp_listendrop(const struct sock *sk) 1940 { 1941 atomic_inc(&((struct sock *)sk)->sk_drops); 1942 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 1943 } 1944 1945 #endif /* _TCP_H */ 1946