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_fastopen; 241 extern int sysctl_tcp_retrans_collapse; 242 extern int sysctl_tcp_stdurg; 243 extern int sysctl_tcp_rfc1337; 244 extern int sysctl_tcp_abort_on_overflow; 245 extern int sysctl_tcp_max_orphans; 246 extern int sysctl_tcp_fack; 247 extern int sysctl_tcp_reordering; 248 extern int sysctl_tcp_max_reordering; 249 extern int sysctl_tcp_dsack; 250 extern long sysctl_tcp_mem[3]; 251 extern int sysctl_tcp_wmem[3]; 252 extern int sysctl_tcp_rmem[3]; 253 extern int sysctl_tcp_app_win; 254 extern int sysctl_tcp_adv_win_scale; 255 extern int sysctl_tcp_frto; 256 extern int sysctl_tcp_low_latency; 257 extern int sysctl_tcp_nometrics_save; 258 extern int sysctl_tcp_moderate_rcvbuf; 259 extern int sysctl_tcp_tso_win_divisor; 260 extern int sysctl_tcp_workaround_signed_windows; 261 extern int sysctl_tcp_slow_start_after_idle; 262 extern int sysctl_tcp_thin_linear_timeouts; 263 extern int sysctl_tcp_thin_dupack; 264 extern int sysctl_tcp_early_retrans; 265 extern int sysctl_tcp_recovery; 266 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 267 268 extern int sysctl_tcp_limit_output_bytes; 269 extern int sysctl_tcp_challenge_ack_limit; 270 extern int sysctl_tcp_min_tso_segs; 271 extern int sysctl_tcp_min_rtt_wlen; 272 extern int sysctl_tcp_autocorking; 273 extern int sysctl_tcp_invalid_ratelimit; 274 extern int sysctl_tcp_pacing_ss_ratio; 275 extern int sysctl_tcp_pacing_ca_ratio; 276 277 extern atomic_long_t tcp_memory_allocated; 278 extern struct percpu_counter tcp_sockets_allocated; 279 extern unsigned long tcp_memory_pressure; 280 281 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 282 static inline bool tcp_under_memory_pressure(const struct sock *sk) 283 { 284 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 285 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 286 return true; 287 288 return tcp_memory_pressure; 289 } 290 /* 291 * The next routines deal with comparing 32 bit unsigned ints 292 * and worry about wraparound (automatic with unsigned arithmetic). 293 */ 294 295 static inline bool before(__u32 seq1, __u32 seq2) 296 { 297 return (__s32)(seq1-seq2) < 0; 298 } 299 #define after(seq2, seq1) before(seq1, seq2) 300 301 /* is s2<=s1<=s3 ? */ 302 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 303 { 304 return seq3 - seq2 >= seq1 - seq2; 305 } 306 307 static inline bool tcp_out_of_memory(struct sock *sk) 308 { 309 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 310 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 311 return true; 312 return false; 313 } 314 315 void sk_forced_mem_schedule(struct sock *sk, int size); 316 317 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 318 { 319 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 320 int orphans = percpu_counter_read_positive(ocp); 321 322 if (orphans << shift > sysctl_tcp_max_orphans) { 323 orphans = percpu_counter_sum_positive(ocp); 324 if (orphans << shift > sysctl_tcp_max_orphans) 325 return true; 326 } 327 return false; 328 } 329 330 bool tcp_check_oom(struct sock *sk, int shift); 331 332 333 extern struct proto tcp_prot; 334 335 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 336 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 337 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 338 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 339 340 void tcp_tasklet_init(void); 341 342 void tcp_v4_err(struct sk_buff *skb, u32); 343 344 void tcp_shutdown(struct sock *sk, int how); 345 346 void tcp_v4_early_demux(struct sk_buff *skb); 347 int tcp_v4_rcv(struct sk_buff *skb); 348 349 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 350 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 351 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 352 int flags); 353 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, 354 size_t size, int flags); 355 void tcp_release_cb(struct sock *sk); 356 void tcp_wfree(struct sk_buff *skb); 357 void tcp_write_timer_handler(struct sock *sk); 358 void tcp_delack_timer_handler(struct sock *sk); 359 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 360 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 361 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 362 const struct tcphdr *th, unsigned int len); 363 void tcp_rcv_space_adjust(struct sock *sk); 364 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 365 void tcp_twsk_destructor(struct sock *sk); 366 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 367 struct pipe_inode_info *pipe, size_t len, 368 unsigned int flags); 369 370 static inline void tcp_dec_quickack_mode(struct sock *sk, 371 const unsigned int pkts) 372 { 373 struct inet_connection_sock *icsk = inet_csk(sk); 374 375 if (icsk->icsk_ack.quick) { 376 if (pkts >= icsk->icsk_ack.quick) { 377 icsk->icsk_ack.quick = 0; 378 /* Leaving quickack mode we deflate ATO. */ 379 icsk->icsk_ack.ato = TCP_ATO_MIN; 380 } else 381 icsk->icsk_ack.quick -= pkts; 382 } 383 } 384 385 #define TCP_ECN_OK 1 386 #define TCP_ECN_QUEUE_CWR 2 387 #define TCP_ECN_DEMAND_CWR 4 388 #define TCP_ECN_SEEN 8 389 390 enum tcp_tw_status { 391 TCP_TW_SUCCESS = 0, 392 TCP_TW_RST = 1, 393 TCP_TW_ACK = 2, 394 TCP_TW_SYN = 3 395 }; 396 397 398 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 399 struct sk_buff *skb, 400 const struct tcphdr *th); 401 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 402 struct request_sock *req, bool fastopen); 403 int tcp_child_process(struct sock *parent, struct sock *child, 404 struct sk_buff *skb); 405 void tcp_enter_loss(struct sock *sk); 406 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag); 407 void tcp_clear_retrans(struct tcp_sock *tp); 408 void tcp_update_metrics(struct sock *sk); 409 void tcp_init_metrics(struct sock *sk); 410 void tcp_metrics_init(void); 411 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 412 void tcp_disable_fack(struct tcp_sock *tp); 413 void tcp_close(struct sock *sk, long timeout); 414 void tcp_init_sock(struct sock *sk); 415 unsigned int tcp_poll(struct file *file, struct socket *sock, 416 struct poll_table_struct *wait); 417 int tcp_getsockopt(struct sock *sk, int level, int optname, 418 char __user *optval, int __user *optlen); 419 int tcp_setsockopt(struct sock *sk, int level, int optname, 420 char __user *optval, unsigned int optlen); 421 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 422 char __user *optval, int __user *optlen); 423 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 424 char __user *optval, unsigned int optlen); 425 void tcp_set_keepalive(struct sock *sk, int val); 426 void tcp_syn_ack_timeout(const struct request_sock *req); 427 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 428 int flags, int *addr_len); 429 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 430 struct tcp_options_received *opt_rx, 431 int estab, struct tcp_fastopen_cookie *foc); 432 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 433 434 /* 435 * TCP v4 functions exported for the inet6 API 436 */ 437 438 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 439 void tcp_v4_mtu_reduced(struct sock *sk); 440 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 441 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 442 struct sock *tcp_create_openreq_child(const struct sock *sk, 443 struct request_sock *req, 444 struct sk_buff *skb); 445 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 446 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 447 struct request_sock *req, 448 struct dst_entry *dst, 449 struct request_sock *req_unhash, 450 bool *own_req); 451 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 452 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 453 int tcp_connect(struct sock *sk); 454 enum tcp_synack_type { 455 TCP_SYNACK_NORMAL, 456 TCP_SYNACK_FASTOPEN, 457 TCP_SYNACK_COOKIE, 458 }; 459 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 460 struct request_sock *req, 461 struct tcp_fastopen_cookie *foc, 462 enum tcp_synack_type synack_type); 463 int tcp_disconnect(struct sock *sk, int flags); 464 465 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 466 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 467 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 468 469 /* From syncookies.c */ 470 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 471 struct request_sock *req, 472 struct dst_entry *dst, u32 tsoff); 473 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 474 u32 cookie); 475 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 476 #ifdef CONFIG_SYN_COOKIES 477 478 /* Syncookies use a monotonic timer which increments every 60 seconds. 479 * This counter is used both as a hash input and partially encoded into 480 * the cookie value. A cookie is only validated further if the delta 481 * between the current counter value and the encoded one is less than this, 482 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 483 * the counter advances immediately after a cookie is generated). 484 */ 485 #define MAX_SYNCOOKIE_AGE 2 486 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 487 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 488 489 /* syncookies: remember time of last synqueue overflow 490 * But do not dirty this field too often (once per second is enough) 491 * It is racy as we do not hold a lock, but race is very minor. 492 */ 493 static inline void tcp_synq_overflow(const struct sock *sk) 494 { 495 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 496 unsigned long now = jiffies; 497 498 if (time_after(now, last_overflow + HZ)) 499 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 500 } 501 502 /* syncookies: no recent synqueue overflow on this listening socket? */ 503 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 504 { 505 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 506 507 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID); 508 } 509 510 static inline u32 tcp_cookie_time(void) 511 { 512 u64 val = get_jiffies_64(); 513 514 do_div(val, TCP_SYNCOOKIE_PERIOD); 515 return val; 516 } 517 518 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 519 u16 *mssp); 520 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 521 u64 cookie_init_timestamp(struct request_sock *req); 522 bool cookie_timestamp_decode(const struct net *net, 523 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 hrtimer_cancel(&tcp_sk(sk)->pacing_timer); 578 inet_csk_clear_xmit_timers(sk); 579 } 580 581 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 582 unsigned int tcp_current_mss(struct sock *sk); 583 584 /* Bound MSS / TSO packet size with the half of the window */ 585 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 586 { 587 int cutoff; 588 589 /* When peer uses tiny windows, there is no use in packetizing 590 * to sub-MSS pieces for the sake of SWS or making sure there 591 * are enough packets in the pipe for fast recovery. 592 * 593 * On the other hand, for extremely large MSS devices, handling 594 * smaller than MSS windows in this way does make sense. 595 */ 596 if (tp->max_window > TCP_MSS_DEFAULT) 597 cutoff = (tp->max_window >> 1); 598 else 599 cutoff = tp->max_window; 600 601 if (cutoff && pktsize > cutoff) 602 return max_t(int, cutoff, 68U - tp->tcp_header_len); 603 else 604 return pktsize; 605 } 606 607 /* tcp.c */ 608 void tcp_get_info(struct sock *, struct tcp_info *); 609 610 /* Read 'sendfile()'-style from a TCP socket */ 611 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 612 sk_read_actor_t recv_actor); 613 614 void tcp_initialize_rcv_mss(struct sock *sk); 615 616 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 617 int tcp_mss_to_mtu(struct sock *sk, int mss); 618 void tcp_mtup_init(struct sock *sk); 619 void tcp_init_buffer_space(struct sock *sk); 620 621 static inline void tcp_bound_rto(const struct sock *sk) 622 { 623 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 624 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 625 } 626 627 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 628 { 629 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 630 } 631 632 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 633 { 634 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 635 ntohl(TCP_FLAG_ACK) | 636 snd_wnd); 637 } 638 639 static inline void tcp_fast_path_on(struct tcp_sock *tp) 640 { 641 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 642 } 643 644 static inline void tcp_fast_path_check(struct sock *sk) 645 { 646 struct tcp_sock *tp = tcp_sk(sk); 647 648 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 649 tp->rcv_wnd && 650 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 651 !tp->urg_data) 652 tcp_fast_path_on(tp); 653 } 654 655 /* Compute the actual rto_min value */ 656 static inline u32 tcp_rto_min(struct sock *sk) 657 { 658 const struct dst_entry *dst = __sk_dst_get(sk); 659 u32 rto_min = TCP_RTO_MIN; 660 661 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 662 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 663 return rto_min; 664 } 665 666 static inline u32 tcp_rto_min_us(struct sock *sk) 667 { 668 return jiffies_to_usecs(tcp_rto_min(sk)); 669 } 670 671 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 672 { 673 return dst_metric_locked(dst, RTAX_CC_ALGO); 674 } 675 676 /* Minimum RTT in usec. ~0 means not available. */ 677 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 678 { 679 return minmax_get(&tp->rtt_min); 680 } 681 682 /* Compute the actual receive window we are currently advertising. 683 * Rcv_nxt can be after the window if our peer push more data 684 * than the offered window. 685 */ 686 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 687 { 688 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 689 690 if (win < 0) 691 win = 0; 692 return (u32) win; 693 } 694 695 /* Choose a new window, without checks for shrinking, and without 696 * scaling applied to the result. The caller does these things 697 * if necessary. This is a "raw" window selection. 698 */ 699 u32 __tcp_select_window(struct sock *sk); 700 701 void tcp_send_window_probe(struct sock *sk); 702 703 /* TCP uses 32bit jiffies to save some space. 704 * Note that this is different from tcp_time_stamp, which 705 * historically has been the same until linux-4.13. 706 */ 707 #define tcp_jiffies32 ((u32)jiffies) 708 709 /* 710 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 711 * It is no longer tied to jiffies, but to 1 ms clock. 712 * Note: double check if you want to use tcp_jiffies32 instead of this. 713 */ 714 #define TCP_TS_HZ 1000 715 716 static inline u64 tcp_clock_ns(void) 717 { 718 return local_clock(); 719 } 720 721 static inline u64 tcp_clock_us(void) 722 { 723 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 724 } 725 726 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 727 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 728 { 729 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 730 } 731 732 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 733 static inline u32 tcp_time_stamp_raw(void) 734 { 735 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ); 736 } 737 738 739 /* Refresh 1us clock of a TCP socket, 740 * ensuring monotically increasing values. 741 */ 742 static inline void tcp_mstamp_refresh(struct tcp_sock *tp) 743 { 744 u64 val = tcp_clock_us(); 745 746 if (val > tp->tcp_mstamp) 747 tp->tcp_mstamp = val; 748 } 749 750 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 751 { 752 return max_t(s64, t1 - t0, 0); 753 } 754 755 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 756 { 757 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ); 758 } 759 760 761 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 762 763 #define TCPHDR_FIN 0x01 764 #define TCPHDR_SYN 0x02 765 #define TCPHDR_RST 0x04 766 #define TCPHDR_PSH 0x08 767 #define TCPHDR_ACK 0x10 768 #define TCPHDR_URG 0x20 769 #define TCPHDR_ECE 0x40 770 #define TCPHDR_CWR 0x80 771 772 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 773 774 /* This is what the send packet queuing engine uses to pass 775 * TCP per-packet control information to the transmission code. 776 * We also store the host-order sequence numbers in here too. 777 * This is 44 bytes if IPV6 is enabled. 778 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 779 */ 780 struct tcp_skb_cb { 781 __u32 seq; /* Starting sequence number */ 782 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 783 union { 784 /* Note : tcp_tw_isn is used in input path only 785 * (isn chosen by tcp_timewait_state_process()) 786 * 787 * tcp_gso_segs/size are used in write queue only, 788 * cf tcp_skb_pcount()/tcp_skb_mss() 789 */ 790 __u32 tcp_tw_isn; 791 struct { 792 u16 tcp_gso_segs; 793 u16 tcp_gso_size; 794 }; 795 }; 796 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 797 798 __u8 sacked; /* State flags for SACK/FACK. */ 799 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 800 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 801 #define TCPCB_LOST 0x04 /* SKB is lost */ 802 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 803 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 804 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 805 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 806 TCPCB_REPAIRED) 807 808 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 809 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 810 eor:1, /* Is skb MSG_EOR marked? */ 811 unused:6; 812 __u32 ack_seq; /* Sequence number ACK'd */ 813 union { 814 struct { 815 /* There is space for up to 24 bytes */ 816 __u32 in_flight:30,/* Bytes in flight at transmit */ 817 is_app_limited:1, /* cwnd not fully used? */ 818 unused:1; 819 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 820 __u32 delivered; 821 /* start of send pipeline phase */ 822 u64 first_tx_mstamp; 823 /* when we reached the "delivered" count */ 824 u64 delivered_mstamp; 825 } tx; /* only used for outgoing skbs */ 826 union { 827 struct inet_skb_parm h4; 828 #if IS_ENABLED(CONFIG_IPV6) 829 struct inet6_skb_parm h6; 830 #endif 831 } header; /* For incoming skbs */ 832 }; 833 }; 834 835 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 836 837 838 #if IS_ENABLED(CONFIG_IPV6) 839 /* This is the variant of inet6_iif() that must be used by TCP, 840 * as TCP moves IP6CB into a different location in skb->cb[] 841 */ 842 static inline int tcp_v6_iif(const struct sk_buff *skb) 843 { 844 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 845 846 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 847 } 848 #endif 849 850 /* TCP_SKB_CB reference means this can not be used from early demux */ 851 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 852 { 853 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 854 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 855 skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 856 return true; 857 #endif 858 return false; 859 } 860 861 /* Due to TSO, an SKB can be composed of multiple actual 862 * packets. To keep these tracked properly, we use this. 863 */ 864 static inline int tcp_skb_pcount(const struct sk_buff *skb) 865 { 866 return TCP_SKB_CB(skb)->tcp_gso_segs; 867 } 868 869 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 870 { 871 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 872 } 873 874 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 875 { 876 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 877 } 878 879 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 880 static inline int tcp_skb_mss(const struct sk_buff *skb) 881 { 882 return TCP_SKB_CB(skb)->tcp_gso_size; 883 } 884 885 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 886 { 887 return likely(!TCP_SKB_CB(skb)->eor); 888 } 889 890 /* Events passed to congestion control interface */ 891 enum tcp_ca_event { 892 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 893 CA_EVENT_CWND_RESTART, /* congestion window restart */ 894 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 895 CA_EVENT_LOSS, /* loss timeout */ 896 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 897 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 898 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 899 CA_EVENT_NON_DELAYED_ACK, 900 }; 901 902 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 903 enum tcp_ca_ack_event_flags { 904 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 905 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 906 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 907 }; 908 909 /* 910 * Interface for adding new TCP congestion control handlers 911 */ 912 #define TCP_CA_NAME_MAX 16 913 #define TCP_CA_MAX 128 914 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 915 916 #define TCP_CA_UNSPEC 0 917 918 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 919 #define TCP_CONG_NON_RESTRICTED 0x1 920 /* Requires ECN/ECT set on all packets */ 921 #define TCP_CONG_NEEDS_ECN 0x2 922 923 union tcp_cc_info; 924 925 struct ack_sample { 926 u32 pkts_acked; 927 s32 rtt_us; 928 u32 in_flight; 929 }; 930 931 /* A rate sample measures the number of (original/retransmitted) data 932 * packets delivered "delivered" over an interval of time "interval_us". 933 * The tcp_rate.c code fills in the rate sample, and congestion 934 * control modules that define a cong_control function to run at the end 935 * of ACK processing can optionally chose to consult this sample when 936 * setting cwnd and pacing rate. 937 * A sample is invalid if "delivered" or "interval_us" is negative. 938 */ 939 struct rate_sample { 940 u64 prior_mstamp; /* starting timestamp for interval */ 941 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 942 s32 delivered; /* number of packets delivered over interval */ 943 long interval_us; /* time for tp->delivered to incr "delivered" */ 944 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 945 int losses; /* number of packets marked lost upon ACK */ 946 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 947 u32 prior_in_flight; /* in flight before this ACK */ 948 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 949 bool is_retrans; /* is sample from retransmission? */ 950 }; 951 952 struct tcp_congestion_ops { 953 struct list_head list; 954 u32 key; 955 u32 flags; 956 957 /* initialize private data (optional) */ 958 void (*init)(struct sock *sk); 959 /* cleanup private data (optional) */ 960 void (*release)(struct sock *sk); 961 962 /* return slow start threshold (required) */ 963 u32 (*ssthresh)(struct sock *sk); 964 /* do new cwnd calculation (required) */ 965 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 966 /* call before changing ca_state (optional) */ 967 void (*set_state)(struct sock *sk, u8 new_state); 968 /* call when cwnd event occurs (optional) */ 969 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 970 /* call when ack arrives (optional) */ 971 void (*in_ack_event)(struct sock *sk, u32 flags); 972 /* new value of cwnd after loss (required) */ 973 u32 (*undo_cwnd)(struct sock *sk); 974 /* hook for packet ack accounting (optional) */ 975 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 976 /* suggest number of segments for each skb to transmit (optional) */ 977 u32 (*tso_segs_goal)(struct sock *sk); 978 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 979 u32 (*sndbuf_expand)(struct sock *sk); 980 /* call when packets are delivered to update cwnd and pacing rate, 981 * after all the ca_state processing. (optional) 982 */ 983 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 984 /* get info for inet_diag (optional) */ 985 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 986 union tcp_cc_info *info); 987 988 char name[TCP_CA_NAME_MAX]; 989 struct module *owner; 990 }; 991 992 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 993 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 994 995 void tcp_assign_congestion_control(struct sock *sk); 996 void tcp_init_congestion_control(struct sock *sk); 997 void tcp_cleanup_congestion_control(struct sock *sk); 998 int tcp_set_default_congestion_control(const char *name); 999 void tcp_get_default_congestion_control(char *name); 1000 void tcp_get_available_congestion_control(char *buf, size_t len); 1001 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1002 int tcp_set_allowed_congestion_control(char *allowed); 1003 int tcp_set_congestion_control(struct sock *sk, const char *name); 1004 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1005 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1006 1007 u32 tcp_reno_ssthresh(struct sock *sk); 1008 u32 tcp_reno_undo_cwnd(struct sock *sk); 1009 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1010 extern struct tcp_congestion_ops tcp_reno; 1011 1012 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1013 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca); 1014 #ifdef CONFIG_INET 1015 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1016 #else 1017 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1018 { 1019 return NULL; 1020 } 1021 #endif 1022 1023 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1024 { 1025 const struct inet_connection_sock *icsk = inet_csk(sk); 1026 1027 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1028 } 1029 1030 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1031 { 1032 struct inet_connection_sock *icsk = inet_csk(sk); 1033 1034 if (icsk->icsk_ca_ops->set_state) 1035 icsk->icsk_ca_ops->set_state(sk, ca_state); 1036 icsk->icsk_ca_state = ca_state; 1037 } 1038 1039 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1040 { 1041 const struct inet_connection_sock *icsk = inet_csk(sk); 1042 1043 if (icsk->icsk_ca_ops->cwnd_event) 1044 icsk->icsk_ca_ops->cwnd_event(sk, event); 1045 } 1046 1047 /* From tcp_rate.c */ 1048 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1049 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1050 struct rate_sample *rs); 1051 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1052 struct rate_sample *rs); 1053 void tcp_rate_check_app_limited(struct sock *sk); 1054 1055 /* These functions determine how the current flow behaves in respect of SACK 1056 * handling. SACK is negotiated with the peer, and therefore it can vary 1057 * between different flows. 1058 * 1059 * tcp_is_sack - SACK enabled 1060 * tcp_is_reno - No SACK 1061 * tcp_is_fack - FACK enabled, implies SACK enabled 1062 */ 1063 static inline int tcp_is_sack(const struct tcp_sock *tp) 1064 { 1065 return tp->rx_opt.sack_ok; 1066 } 1067 1068 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1069 { 1070 return !tcp_is_sack(tp); 1071 } 1072 1073 static inline bool tcp_is_fack(const struct tcp_sock *tp) 1074 { 1075 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 1076 } 1077 1078 static inline void tcp_enable_fack(struct tcp_sock *tp) 1079 { 1080 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 1081 } 1082 1083 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1084 { 1085 return tp->sacked_out + tp->lost_out; 1086 } 1087 1088 /* This determines how many packets are "in the network" to the best 1089 * of our knowledge. In many cases it is conservative, but where 1090 * detailed information is available from the receiver (via SACK 1091 * blocks etc.) we can make more aggressive calculations. 1092 * 1093 * Use this for decisions involving congestion control, use just 1094 * tp->packets_out to determine if the send queue is empty or not. 1095 * 1096 * Read this equation as: 1097 * 1098 * "Packets sent once on transmission queue" MINUS 1099 * "Packets left network, but not honestly ACKed yet" PLUS 1100 * "Packets fast retransmitted" 1101 */ 1102 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1103 { 1104 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1105 } 1106 1107 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1108 1109 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1110 { 1111 return tp->snd_cwnd < tp->snd_ssthresh; 1112 } 1113 1114 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1115 { 1116 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1117 } 1118 1119 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1120 { 1121 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1122 (1 << inet_csk(sk)->icsk_ca_state); 1123 } 1124 1125 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1126 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1127 * ssthresh. 1128 */ 1129 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1130 { 1131 const struct tcp_sock *tp = tcp_sk(sk); 1132 1133 if (tcp_in_cwnd_reduction(sk)) 1134 return tp->snd_ssthresh; 1135 else 1136 return max(tp->snd_ssthresh, 1137 ((tp->snd_cwnd >> 1) + 1138 (tp->snd_cwnd >> 2))); 1139 } 1140 1141 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1142 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1143 1144 void tcp_enter_cwr(struct sock *sk); 1145 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1146 1147 /* The maximum number of MSS of available cwnd for which TSO defers 1148 * sending if not using sysctl_tcp_tso_win_divisor. 1149 */ 1150 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1151 { 1152 return 3; 1153 } 1154 1155 /* Returns end sequence number of the receiver's advertised window */ 1156 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1157 { 1158 return tp->snd_una + tp->snd_wnd; 1159 } 1160 1161 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1162 * flexible approach. The RFC suggests cwnd should not be raised unless 1163 * it was fully used previously. And that's exactly what we do in 1164 * congestion avoidance mode. But in slow start we allow cwnd to grow 1165 * as long as the application has used half the cwnd. 1166 * Example : 1167 * cwnd is 10 (IW10), but application sends 9 frames. 1168 * We allow cwnd to reach 18 when all frames are ACKed. 1169 * This check is safe because it's as aggressive as slow start which already 1170 * risks 100% overshoot. The advantage is that we discourage application to 1171 * either send more filler packets or data to artificially blow up the cwnd 1172 * usage, and allow application-limited process to probe bw more aggressively. 1173 */ 1174 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1175 { 1176 const struct tcp_sock *tp = tcp_sk(sk); 1177 1178 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1179 if (tcp_in_slow_start(tp)) 1180 return tp->snd_cwnd < 2 * tp->max_packets_out; 1181 1182 return tp->is_cwnd_limited; 1183 } 1184 1185 /* Something is really bad, we could not queue an additional packet, 1186 * because qdisc is full or receiver sent a 0 window. 1187 * We do not want to add fuel to the fire, or abort too early, 1188 * so make sure the timer we arm now is at least 200ms in the future, 1189 * regardless of current icsk_rto value (as it could be ~2ms) 1190 */ 1191 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1192 { 1193 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1194 } 1195 1196 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1197 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1198 unsigned long max_when) 1199 { 1200 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1201 1202 return (unsigned long)min_t(u64, when, max_when); 1203 } 1204 1205 static inline void tcp_check_probe_timer(struct sock *sk) 1206 { 1207 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1208 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1209 tcp_probe0_base(sk), TCP_RTO_MAX); 1210 } 1211 1212 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1213 { 1214 tp->snd_wl1 = seq; 1215 } 1216 1217 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1218 { 1219 tp->snd_wl1 = seq; 1220 } 1221 1222 /* 1223 * Calculate(/check) TCP checksum 1224 */ 1225 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1226 __be32 daddr, __wsum base) 1227 { 1228 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1229 } 1230 1231 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1232 { 1233 return __skb_checksum_complete(skb); 1234 } 1235 1236 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1237 { 1238 return !skb_csum_unnecessary(skb) && 1239 __tcp_checksum_complete(skb); 1240 } 1241 1242 /* Prequeue for VJ style copy to user, combined with checksumming. */ 1243 1244 static inline void tcp_prequeue_init(struct tcp_sock *tp) 1245 { 1246 tp->ucopy.task = NULL; 1247 tp->ucopy.len = 0; 1248 tp->ucopy.memory = 0; 1249 skb_queue_head_init(&tp->ucopy.prequeue); 1250 } 1251 1252 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 1253 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1254 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1255 1256 #undef STATE_TRACE 1257 1258 #ifdef STATE_TRACE 1259 static const char *statename[]={ 1260 "Unused","Established","Syn Sent","Syn Recv", 1261 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1262 "Close Wait","Last ACK","Listen","Closing" 1263 }; 1264 #endif 1265 void tcp_set_state(struct sock *sk, int state); 1266 1267 void tcp_done(struct sock *sk); 1268 1269 int tcp_abort(struct sock *sk, int err); 1270 1271 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1272 { 1273 rx_opt->dsack = 0; 1274 rx_opt->num_sacks = 0; 1275 } 1276 1277 u32 tcp_default_init_rwnd(u32 mss); 1278 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1279 1280 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1281 { 1282 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1283 struct tcp_sock *tp = tcp_sk(sk); 1284 s32 delta; 1285 1286 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out || 1287 ca_ops->cong_control) 1288 return; 1289 delta = tcp_jiffies32 - tp->lsndtime; 1290 if (delta > inet_csk(sk)->icsk_rto) 1291 tcp_cwnd_restart(sk, delta); 1292 } 1293 1294 /* Determine a window scaling and initial window to offer. */ 1295 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1296 __u32 *window_clamp, int wscale_ok, 1297 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1298 1299 static inline int tcp_win_from_space(int space) 1300 { 1301 int tcp_adv_win_scale = sysctl_tcp_adv_win_scale; 1302 1303 return tcp_adv_win_scale <= 0 ? 1304 (space>>(-tcp_adv_win_scale)) : 1305 space - (space>>tcp_adv_win_scale); 1306 } 1307 1308 /* Note: caller must be prepared to deal with negative returns */ 1309 static inline int tcp_space(const struct sock *sk) 1310 { 1311 return tcp_win_from_space(sk->sk_rcvbuf - 1312 atomic_read(&sk->sk_rmem_alloc)); 1313 } 1314 1315 static inline int tcp_full_space(const struct sock *sk) 1316 { 1317 return tcp_win_from_space(sk->sk_rcvbuf); 1318 } 1319 1320 extern void tcp_openreq_init_rwin(struct request_sock *req, 1321 const struct sock *sk_listener, 1322 const struct dst_entry *dst); 1323 1324 void tcp_enter_memory_pressure(struct sock *sk); 1325 void tcp_leave_memory_pressure(struct sock *sk); 1326 1327 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1328 { 1329 struct net *net = sock_net((struct sock *)tp); 1330 1331 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1332 } 1333 1334 static inline int keepalive_time_when(const struct tcp_sock *tp) 1335 { 1336 struct net *net = sock_net((struct sock *)tp); 1337 1338 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1339 } 1340 1341 static inline int keepalive_probes(const struct tcp_sock *tp) 1342 { 1343 struct net *net = sock_net((struct sock *)tp); 1344 1345 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1346 } 1347 1348 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1349 { 1350 const struct inet_connection_sock *icsk = &tp->inet_conn; 1351 1352 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1353 tcp_jiffies32 - tp->rcv_tstamp); 1354 } 1355 1356 static inline int tcp_fin_time(const struct sock *sk) 1357 { 1358 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1359 const int rto = inet_csk(sk)->icsk_rto; 1360 1361 if (fin_timeout < (rto << 2) - (rto >> 1)) 1362 fin_timeout = (rto << 2) - (rto >> 1); 1363 1364 return fin_timeout; 1365 } 1366 1367 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1368 int paws_win) 1369 { 1370 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1371 return true; 1372 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1373 return true; 1374 /* 1375 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1376 * then following tcp messages have valid values. Ignore 0 value, 1377 * or else 'negative' tsval might forbid us to accept their packets. 1378 */ 1379 if (!rx_opt->ts_recent) 1380 return true; 1381 return false; 1382 } 1383 1384 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1385 int rst) 1386 { 1387 if (tcp_paws_check(rx_opt, 0)) 1388 return false; 1389 1390 /* RST segments are not recommended to carry timestamp, 1391 and, if they do, it is recommended to ignore PAWS because 1392 "their cleanup function should take precedence over timestamps." 1393 Certainly, it is mistake. It is necessary to understand the reasons 1394 of this constraint to relax it: if peer reboots, clock may go 1395 out-of-sync and half-open connections will not be reset. 1396 Actually, the problem would be not existing if all 1397 the implementations followed draft about maintaining clock 1398 via reboots. Linux-2.2 DOES NOT! 1399 1400 However, we can relax time bounds for RST segments to MSL. 1401 */ 1402 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1403 return false; 1404 return true; 1405 } 1406 1407 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1408 int mib_idx, u32 *last_oow_ack_time); 1409 1410 static inline void tcp_mib_init(struct net *net) 1411 { 1412 /* See RFC 2012 */ 1413 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1414 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1415 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1416 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1417 } 1418 1419 /* from STCP */ 1420 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1421 { 1422 tp->lost_skb_hint = NULL; 1423 } 1424 1425 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1426 { 1427 tcp_clear_retrans_hints_partial(tp); 1428 tp->retransmit_skb_hint = NULL; 1429 } 1430 1431 union tcp_md5_addr { 1432 struct in_addr a4; 1433 #if IS_ENABLED(CONFIG_IPV6) 1434 struct in6_addr a6; 1435 #endif 1436 }; 1437 1438 /* - key database */ 1439 struct tcp_md5sig_key { 1440 struct hlist_node node; 1441 u8 keylen; 1442 u8 family; /* AF_INET or AF_INET6 */ 1443 union tcp_md5_addr addr; 1444 u8 prefixlen; 1445 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1446 struct rcu_head rcu; 1447 }; 1448 1449 /* - sock block */ 1450 struct tcp_md5sig_info { 1451 struct hlist_head head; 1452 struct rcu_head rcu; 1453 }; 1454 1455 /* - pseudo header */ 1456 struct tcp4_pseudohdr { 1457 __be32 saddr; 1458 __be32 daddr; 1459 __u8 pad; 1460 __u8 protocol; 1461 __be16 len; 1462 }; 1463 1464 struct tcp6_pseudohdr { 1465 struct in6_addr saddr; 1466 struct in6_addr daddr; 1467 __be32 len; 1468 __be32 protocol; /* including padding */ 1469 }; 1470 1471 union tcp_md5sum_block { 1472 struct tcp4_pseudohdr ip4; 1473 #if IS_ENABLED(CONFIG_IPV6) 1474 struct tcp6_pseudohdr ip6; 1475 #endif 1476 }; 1477 1478 /* - pool: digest algorithm, hash description and scratch buffer */ 1479 struct tcp_md5sig_pool { 1480 struct ahash_request *md5_req; 1481 void *scratch; 1482 }; 1483 1484 /* - functions */ 1485 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1486 const struct sock *sk, const struct sk_buff *skb); 1487 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1488 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen, 1489 gfp_t gfp); 1490 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1491 int family, u8 prefixlen); 1492 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1493 const struct sock *addr_sk); 1494 1495 #ifdef CONFIG_TCP_MD5SIG 1496 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1497 const union tcp_md5_addr *addr, 1498 int family); 1499 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1500 #else 1501 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1502 const union tcp_md5_addr *addr, 1503 int family) 1504 { 1505 return NULL; 1506 } 1507 #define tcp_twsk_md5_key(twsk) NULL 1508 #endif 1509 1510 bool tcp_alloc_md5sig_pool(void); 1511 1512 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1513 static inline void tcp_put_md5sig_pool(void) 1514 { 1515 local_bh_enable(); 1516 } 1517 1518 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1519 unsigned int header_len); 1520 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1521 const struct tcp_md5sig_key *key); 1522 1523 /* From tcp_fastopen.c */ 1524 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1525 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1526 unsigned long *last_syn_loss); 1527 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1528 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1529 u16 try_exp); 1530 struct tcp_fastopen_request { 1531 /* Fast Open cookie. Size 0 means a cookie request */ 1532 struct tcp_fastopen_cookie cookie; 1533 struct msghdr *data; /* data in MSG_FASTOPEN */ 1534 size_t size; 1535 int copied; /* queued in tcp_connect() */ 1536 }; 1537 void tcp_free_fastopen_req(struct tcp_sock *tp); 1538 1539 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1540 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1541 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1542 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1543 struct request_sock *req, 1544 struct tcp_fastopen_cookie *foc, 1545 struct dst_entry *dst); 1546 void tcp_fastopen_init_key_once(bool publish); 1547 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1548 struct tcp_fastopen_cookie *cookie); 1549 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1550 #define TCP_FASTOPEN_KEY_LENGTH 16 1551 1552 /* Fastopen key context */ 1553 struct tcp_fastopen_context { 1554 struct crypto_cipher *tfm; 1555 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1556 struct rcu_head rcu; 1557 }; 1558 1559 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1560 void tcp_fastopen_active_disable(struct sock *sk); 1561 bool tcp_fastopen_active_should_disable(struct sock *sk); 1562 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1563 void tcp_fastopen_active_timeout_reset(void); 1564 1565 /* Latencies incurred by various limits for a sender. They are 1566 * chronograph-like stats that are mutually exclusive. 1567 */ 1568 enum tcp_chrono { 1569 TCP_CHRONO_UNSPEC, 1570 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1571 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1572 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1573 __TCP_CHRONO_MAX, 1574 }; 1575 1576 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1577 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1578 1579 /* write queue abstraction */ 1580 static inline void tcp_write_queue_purge(struct sock *sk) 1581 { 1582 struct sk_buff *skb; 1583 1584 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1585 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1586 sk_wmem_free_skb(sk, skb); 1587 sk_mem_reclaim(sk); 1588 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1589 } 1590 1591 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1592 { 1593 return skb_peek(&sk->sk_write_queue); 1594 } 1595 1596 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1597 { 1598 return skb_peek_tail(&sk->sk_write_queue); 1599 } 1600 1601 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1602 const struct sk_buff *skb) 1603 { 1604 return skb_queue_next(&sk->sk_write_queue, skb); 1605 } 1606 1607 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1608 const struct sk_buff *skb) 1609 { 1610 return skb_queue_prev(&sk->sk_write_queue, skb); 1611 } 1612 1613 #define tcp_for_write_queue(skb, sk) \ 1614 skb_queue_walk(&(sk)->sk_write_queue, skb) 1615 1616 #define tcp_for_write_queue_from(skb, sk) \ 1617 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1618 1619 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1620 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1621 1622 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1623 { 1624 return sk->sk_send_head; 1625 } 1626 1627 static inline bool tcp_skb_is_last(const struct sock *sk, 1628 const struct sk_buff *skb) 1629 { 1630 return skb_queue_is_last(&sk->sk_write_queue, skb); 1631 } 1632 1633 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1634 { 1635 if (tcp_skb_is_last(sk, skb)) 1636 sk->sk_send_head = NULL; 1637 else 1638 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1639 } 1640 1641 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1642 { 1643 if (sk->sk_send_head == skb_unlinked) { 1644 sk->sk_send_head = NULL; 1645 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1646 } 1647 if (tcp_sk(sk)->highest_sack == skb_unlinked) 1648 tcp_sk(sk)->highest_sack = NULL; 1649 } 1650 1651 static inline void tcp_init_send_head(struct sock *sk) 1652 { 1653 sk->sk_send_head = NULL; 1654 } 1655 1656 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1657 { 1658 __skb_queue_tail(&sk->sk_write_queue, skb); 1659 } 1660 1661 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1662 { 1663 __tcp_add_write_queue_tail(sk, skb); 1664 1665 /* Queue it, remembering where we must start sending. */ 1666 if (sk->sk_send_head == NULL) { 1667 sk->sk_send_head = skb; 1668 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1669 1670 if (tcp_sk(sk)->highest_sack == NULL) 1671 tcp_sk(sk)->highest_sack = skb; 1672 } 1673 } 1674 1675 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1676 { 1677 __skb_queue_head(&sk->sk_write_queue, skb); 1678 } 1679 1680 /* Insert buff after skb on the write queue of sk. */ 1681 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1682 struct sk_buff *buff, 1683 struct sock *sk) 1684 { 1685 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1686 } 1687 1688 /* Insert new before skb on the write queue of sk. */ 1689 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1690 struct sk_buff *skb, 1691 struct sock *sk) 1692 { 1693 __skb_queue_before(&sk->sk_write_queue, skb, new); 1694 1695 if (sk->sk_send_head == skb) 1696 sk->sk_send_head = new; 1697 } 1698 1699 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1700 { 1701 __skb_unlink(skb, &sk->sk_write_queue); 1702 } 1703 1704 static inline bool tcp_write_queue_empty(struct sock *sk) 1705 { 1706 return skb_queue_empty(&sk->sk_write_queue); 1707 } 1708 1709 static inline void tcp_push_pending_frames(struct sock *sk) 1710 { 1711 if (tcp_send_head(sk)) { 1712 struct tcp_sock *tp = tcp_sk(sk); 1713 1714 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1715 } 1716 } 1717 1718 /* Start sequence of the skb just after the highest skb with SACKed 1719 * bit, valid only if sacked_out > 0 or when the caller has ensured 1720 * validity by itself. 1721 */ 1722 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1723 { 1724 if (!tp->sacked_out) 1725 return tp->snd_una; 1726 1727 if (tp->highest_sack == NULL) 1728 return tp->snd_nxt; 1729 1730 return TCP_SKB_CB(tp->highest_sack)->seq; 1731 } 1732 1733 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1734 { 1735 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1736 tcp_write_queue_next(sk, skb); 1737 } 1738 1739 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1740 { 1741 return tcp_sk(sk)->highest_sack; 1742 } 1743 1744 static inline void tcp_highest_sack_reset(struct sock *sk) 1745 { 1746 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1747 } 1748 1749 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1750 static inline void tcp_highest_sack_combine(struct sock *sk, 1751 struct sk_buff *old, 1752 struct sk_buff *new) 1753 { 1754 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1755 tcp_sk(sk)->highest_sack = new; 1756 } 1757 1758 /* This helper checks if socket has IP_TRANSPARENT set */ 1759 static inline bool inet_sk_transparent(const struct sock *sk) 1760 { 1761 switch (sk->sk_state) { 1762 case TCP_TIME_WAIT: 1763 return inet_twsk(sk)->tw_transparent; 1764 case TCP_NEW_SYN_RECV: 1765 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1766 } 1767 return inet_sk(sk)->transparent; 1768 } 1769 1770 /* Determines whether this is a thin stream (which may suffer from 1771 * increased latency). Used to trigger latency-reducing mechanisms. 1772 */ 1773 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1774 { 1775 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1776 } 1777 1778 /* /proc */ 1779 enum tcp_seq_states { 1780 TCP_SEQ_STATE_LISTENING, 1781 TCP_SEQ_STATE_ESTABLISHED, 1782 }; 1783 1784 int tcp_seq_open(struct inode *inode, struct file *file); 1785 1786 struct tcp_seq_afinfo { 1787 char *name; 1788 sa_family_t family; 1789 const struct file_operations *seq_fops; 1790 struct seq_operations seq_ops; 1791 }; 1792 1793 struct tcp_iter_state { 1794 struct seq_net_private p; 1795 sa_family_t family; 1796 enum tcp_seq_states state; 1797 struct sock *syn_wait_sk; 1798 int bucket, offset, sbucket, num; 1799 loff_t last_pos; 1800 }; 1801 1802 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1803 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1804 1805 extern struct request_sock_ops tcp_request_sock_ops; 1806 extern struct request_sock_ops tcp6_request_sock_ops; 1807 1808 void tcp_v4_destroy_sock(struct sock *sk); 1809 1810 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1811 netdev_features_t features); 1812 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1813 int tcp_gro_complete(struct sk_buff *skb); 1814 1815 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1816 1817 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1818 { 1819 struct net *net = sock_net((struct sock *)tp); 1820 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1821 } 1822 1823 static inline bool tcp_stream_memory_free(const struct sock *sk) 1824 { 1825 const struct tcp_sock *tp = tcp_sk(sk); 1826 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1827 1828 return notsent_bytes < tcp_notsent_lowat(tp); 1829 } 1830 1831 #ifdef CONFIG_PROC_FS 1832 int tcp4_proc_init(void); 1833 void tcp4_proc_exit(void); 1834 #endif 1835 1836 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1837 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1838 const struct tcp_request_sock_ops *af_ops, 1839 struct sock *sk, struct sk_buff *skb); 1840 1841 /* TCP af-specific functions */ 1842 struct tcp_sock_af_ops { 1843 #ifdef CONFIG_TCP_MD5SIG 1844 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1845 const struct sock *addr_sk); 1846 int (*calc_md5_hash)(char *location, 1847 const struct tcp_md5sig_key *md5, 1848 const struct sock *sk, 1849 const struct sk_buff *skb); 1850 int (*md5_parse)(struct sock *sk, 1851 int optname, 1852 char __user *optval, 1853 int optlen); 1854 #endif 1855 }; 1856 1857 struct tcp_request_sock_ops { 1858 u16 mss_clamp; 1859 #ifdef CONFIG_TCP_MD5SIG 1860 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1861 const struct sock *addr_sk); 1862 int (*calc_md5_hash) (char *location, 1863 const struct tcp_md5sig_key *md5, 1864 const struct sock *sk, 1865 const struct sk_buff *skb); 1866 #endif 1867 void (*init_req)(struct request_sock *req, 1868 const struct sock *sk_listener, 1869 struct sk_buff *skb); 1870 #ifdef CONFIG_SYN_COOKIES 1871 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1872 __u16 *mss); 1873 #endif 1874 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1875 const struct request_sock *req); 1876 u32 (*init_seq)(const struct sk_buff *skb); 1877 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 1878 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1879 struct flowi *fl, struct request_sock *req, 1880 struct tcp_fastopen_cookie *foc, 1881 enum tcp_synack_type synack_type); 1882 }; 1883 1884 #ifdef CONFIG_SYN_COOKIES 1885 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1886 const struct sock *sk, struct sk_buff *skb, 1887 __u16 *mss) 1888 { 1889 tcp_synq_overflow(sk); 1890 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1891 return ops->cookie_init_seq(skb, mss); 1892 } 1893 #else 1894 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1895 const struct sock *sk, struct sk_buff *skb, 1896 __u16 *mss) 1897 { 1898 return 0; 1899 } 1900 #endif 1901 1902 int tcpv4_offload_init(void); 1903 1904 void tcp_v4_init(void); 1905 void tcp_init(void); 1906 1907 /* tcp_recovery.c */ 1908 extern void tcp_rack_mark_lost(struct sock *sk); 1909 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 1910 u64 xmit_time); 1911 extern void tcp_rack_reo_timeout(struct sock *sk); 1912 1913 /* 1914 * Save and compile IPv4 options, return a pointer to it 1915 */ 1916 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb) 1917 { 1918 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1919 struct ip_options_rcu *dopt = NULL; 1920 1921 if (opt->optlen) { 1922 int opt_size = sizeof(*dopt) + opt->optlen; 1923 1924 dopt = kmalloc(opt_size, GFP_ATOMIC); 1925 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) { 1926 kfree(dopt); 1927 dopt = NULL; 1928 } 1929 } 1930 return dopt; 1931 } 1932 1933 /* locally generated TCP pure ACKs have skb->truesize == 2 1934 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1935 * This is much faster than dissecting the packet to find out. 1936 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1937 */ 1938 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1939 { 1940 return skb->truesize == 2; 1941 } 1942 1943 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1944 { 1945 skb->truesize = 2; 1946 } 1947 1948 static inline int tcp_inq(struct sock *sk) 1949 { 1950 struct tcp_sock *tp = tcp_sk(sk); 1951 int answ; 1952 1953 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1954 answ = 0; 1955 } else if (sock_flag(sk, SOCK_URGINLINE) || 1956 !tp->urg_data || 1957 before(tp->urg_seq, tp->copied_seq) || 1958 !before(tp->urg_seq, tp->rcv_nxt)) { 1959 1960 answ = tp->rcv_nxt - tp->copied_seq; 1961 1962 /* Subtract 1, if FIN was received */ 1963 if (answ && sock_flag(sk, SOCK_DONE)) 1964 answ--; 1965 } else { 1966 answ = tp->urg_seq - tp->copied_seq; 1967 } 1968 1969 return answ; 1970 } 1971 1972 int tcp_peek_len(struct socket *sock); 1973 1974 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 1975 { 1976 u16 segs_in; 1977 1978 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 1979 tp->segs_in += segs_in; 1980 if (skb->len > tcp_hdrlen(skb)) 1981 tp->data_segs_in += segs_in; 1982 } 1983 1984 /* 1985 * TCP listen path runs lockless. 1986 * We forced "struct sock" to be const qualified to make sure 1987 * we don't modify one of its field by mistake. 1988 * Here, we increment sk_drops which is an atomic_t, so we can safely 1989 * make sock writable again. 1990 */ 1991 static inline void tcp_listendrop(const struct sock *sk) 1992 { 1993 atomic_inc(&((struct sock *)sk)->sk_drops); 1994 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 1995 } 1996 1997 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 1998 1999 /* 2000 * Interface for adding Upper Level Protocols over TCP 2001 */ 2002 2003 #define TCP_ULP_NAME_MAX 16 2004 #define TCP_ULP_MAX 128 2005 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2006 2007 struct tcp_ulp_ops { 2008 struct list_head list; 2009 2010 /* initialize ulp */ 2011 int (*init)(struct sock *sk); 2012 /* cleanup ulp */ 2013 void (*release)(struct sock *sk); 2014 2015 char name[TCP_ULP_NAME_MAX]; 2016 struct module *owner; 2017 }; 2018 int tcp_register_ulp(struct tcp_ulp_ops *type); 2019 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2020 int tcp_set_ulp(struct sock *sk, const char *name); 2021 void tcp_get_available_ulp(char *buf, size_t len); 2022 void tcp_cleanup_ulp(struct sock *sk); 2023 2024 #endif /* _TCP_H */ 2025