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 #include <linux/bpf.h> 50 #include <linux/filter.h> 51 #include <linux/bpf-cgroup.h> 52 53 extern struct inet_hashinfo tcp_hashinfo; 54 55 extern struct percpu_counter tcp_orphan_count; 56 void tcp_time_wait(struct sock *sk, int state, int timeo); 57 58 #define MAX_TCP_HEADER (128 + MAX_HEADER) 59 #define MAX_TCP_OPTION_SPACE 40 60 61 /* 62 * Never offer a window over 32767 without using window scaling. Some 63 * poor stacks do signed 16bit maths! 64 */ 65 #define MAX_TCP_WINDOW 32767U 66 67 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 68 #define TCP_MIN_MSS 88U 69 70 /* The least MTU to use for probing */ 71 #define TCP_BASE_MSS 1024 72 73 /* probing interval, default to 10 minutes as per RFC4821 */ 74 #define TCP_PROBE_INTERVAL 600 75 76 /* Specify interval when tcp mtu probing will stop */ 77 #define TCP_PROBE_THRESHOLD 8 78 79 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 80 #define TCP_FASTRETRANS_THRESH 3 81 82 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 83 #define TCP_MAX_QUICKACKS 16U 84 85 /* Maximal number of window scale according to RFC1323 */ 86 #define TCP_MAX_WSCALE 14U 87 88 /* urg_data states */ 89 #define TCP_URG_VALID 0x0100 90 #define TCP_URG_NOTYET 0x0200 91 #define TCP_URG_READ 0x0400 92 93 #define TCP_RETR1 3 /* 94 * This is how many retries it does before it 95 * tries to figure out if the gateway is 96 * down. Minimal RFC value is 3; it corresponds 97 * to ~3sec-8min depending on RTO. 98 */ 99 100 #define TCP_RETR2 15 /* 101 * This should take at least 102 * 90 minutes to time out. 103 * RFC1122 says that the limit is 100 sec. 104 * 15 is ~13-30min depending on RTO. 105 */ 106 107 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 108 * when active opening a connection. 109 * RFC1122 says the minimum retry MUST 110 * be at least 180secs. Nevertheless 111 * this value is corresponding to 112 * 63secs of retransmission with the 113 * current initial RTO. 114 */ 115 116 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 117 * when passive opening a connection. 118 * This is corresponding to 31secs of 119 * retransmission with the current 120 * initial RTO. 121 */ 122 123 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 124 * state, about 60 seconds */ 125 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 126 /* BSD style FIN_WAIT2 deadlock breaker. 127 * It used to be 3min, new value is 60sec, 128 * to combine FIN-WAIT-2 timeout with 129 * TIME-WAIT timer. 130 */ 131 132 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 133 #if HZ >= 100 134 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 135 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 136 #else 137 #define TCP_DELACK_MIN 4U 138 #define TCP_ATO_MIN 4U 139 #endif 140 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 141 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 142 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 143 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 144 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 145 * used as a fallback RTO for the 146 * initial data transmission if no 147 * valid RTT sample has been acquired, 148 * most likely due to retrans in 3WHS. 149 */ 150 151 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 152 * for local resources. 153 */ 154 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 155 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 156 #define TCP_KEEPALIVE_INTVL (75*HZ) 157 158 #define MAX_TCP_KEEPIDLE 32767 159 #define MAX_TCP_KEEPINTVL 32767 160 #define MAX_TCP_KEEPCNT 127 161 #define MAX_TCP_SYNCNT 127 162 163 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 164 165 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 166 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 167 * after this time. It should be equal 168 * (or greater than) TCP_TIMEWAIT_LEN 169 * to provide reliability equal to one 170 * provided by timewait state. 171 */ 172 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 173 * timestamps. It must be less than 174 * minimal timewait lifetime. 175 */ 176 /* 177 * TCP option 178 */ 179 180 #define TCPOPT_NOP 1 /* Padding */ 181 #define TCPOPT_EOL 0 /* End of options */ 182 #define TCPOPT_MSS 2 /* Segment size negotiating */ 183 #define TCPOPT_WINDOW 3 /* Window scaling */ 184 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 185 #define TCPOPT_SACK 5 /* SACK Block */ 186 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 187 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 188 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 189 #define TCPOPT_EXP 254 /* Experimental */ 190 /* Magic number to be after the option value for sharing TCP 191 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 192 */ 193 #define TCPOPT_FASTOPEN_MAGIC 0xF989 194 195 /* 196 * TCP option lengths 197 */ 198 199 #define TCPOLEN_MSS 4 200 #define TCPOLEN_WINDOW 3 201 #define TCPOLEN_SACK_PERM 2 202 #define TCPOLEN_TIMESTAMP 10 203 #define TCPOLEN_MD5SIG 18 204 #define TCPOLEN_FASTOPEN_BASE 2 205 #define TCPOLEN_EXP_FASTOPEN_BASE 4 206 207 /* But this is what stacks really send out. */ 208 #define TCPOLEN_TSTAMP_ALIGNED 12 209 #define TCPOLEN_WSCALE_ALIGNED 4 210 #define TCPOLEN_SACKPERM_ALIGNED 4 211 #define TCPOLEN_SACK_BASE 2 212 #define TCPOLEN_SACK_BASE_ALIGNED 4 213 #define TCPOLEN_SACK_PERBLOCK 8 214 #define TCPOLEN_MD5SIG_ALIGNED 20 215 #define TCPOLEN_MSS_ALIGNED 4 216 217 /* Flags in tp->nonagle */ 218 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 219 #define TCP_NAGLE_CORK 2 /* Socket is corked */ 220 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 221 222 /* TCP thin-stream limits */ 223 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 224 225 /* TCP initial congestion window as per rfc6928 */ 226 #define TCP_INIT_CWND 10 227 228 /* Bit Flags for sysctl_tcp_fastopen */ 229 #define TFO_CLIENT_ENABLE 1 230 #define TFO_SERVER_ENABLE 2 231 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 232 233 /* Accept SYN data w/o any cookie option */ 234 #define TFO_SERVER_COOKIE_NOT_REQD 0x200 235 236 /* Force enable TFO on all listeners, i.e., not requiring the 237 * TCP_FASTOPEN socket option. 238 */ 239 #define TFO_SERVER_WO_SOCKOPT1 0x400 240 241 242 /* sysctl variables for tcp */ 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_nometrics_save; 260 extern int sysctl_tcp_moderate_rcvbuf; 261 extern int sysctl_tcp_tso_win_divisor; 262 extern int sysctl_tcp_workaround_signed_windows; 263 extern int sysctl_tcp_slow_start_after_idle; 264 extern int sysctl_tcp_thin_linear_timeouts; 265 extern int sysctl_tcp_thin_dupack; 266 extern int sysctl_tcp_early_retrans; 267 extern int sysctl_tcp_recovery; 268 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 269 270 extern int sysctl_tcp_limit_output_bytes; 271 extern int sysctl_tcp_challenge_ack_limit; 272 extern int sysctl_tcp_min_tso_segs; 273 extern int sysctl_tcp_min_rtt_wlen; 274 extern int sysctl_tcp_autocorking; 275 extern int sysctl_tcp_invalid_ratelimit; 276 extern int sysctl_tcp_pacing_ss_ratio; 277 extern int sysctl_tcp_pacing_ca_ratio; 278 279 extern atomic_long_t tcp_memory_allocated; 280 extern struct percpu_counter tcp_sockets_allocated; 281 extern unsigned long tcp_memory_pressure; 282 283 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 284 static inline bool tcp_under_memory_pressure(const struct sock *sk) 285 { 286 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 287 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 288 return true; 289 290 return tcp_memory_pressure; 291 } 292 /* 293 * The next routines deal with comparing 32 bit unsigned ints 294 * and worry about wraparound (automatic with unsigned arithmetic). 295 */ 296 297 static inline bool before(__u32 seq1, __u32 seq2) 298 { 299 return (__s32)(seq1-seq2) < 0; 300 } 301 #define after(seq2, seq1) before(seq1, seq2) 302 303 /* is s2<=s1<=s3 ? */ 304 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 305 { 306 return seq3 - seq2 >= seq1 - seq2; 307 } 308 309 static inline bool tcp_out_of_memory(struct sock *sk) 310 { 311 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 312 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 313 return true; 314 return false; 315 } 316 317 void sk_forced_mem_schedule(struct sock *sk, int size); 318 319 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 320 { 321 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 322 int orphans = percpu_counter_read_positive(ocp); 323 324 if (orphans << shift > sysctl_tcp_max_orphans) { 325 orphans = percpu_counter_sum_positive(ocp); 326 if (orphans << shift > sysctl_tcp_max_orphans) 327 return true; 328 } 329 return false; 330 } 331 332 bool tcp_check_oom(struct sock *sk, int shift); 333 334 335 extern struct proto tcp_prot; 336 337 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 338 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 339 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 340 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 341 342 void tcp_tasklet_init(void); 343 344 void tcp_v4_err(struct sk_buff *skb, u32); 345 346 void tcp_shutdown(struct sock *sk, int how); 347 348 void tcp_v4_early_demux(struct sk_buff *skb); 349 int tcp_v4_rcv(struct sk_buff *skb); 350 351 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 352 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 353 int tcp_sendmsg_locked(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 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, 357 size_t size, int flags); 358 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, 359 size_t size, int flags); 360 void tcp_release_cb(struct sock *sk); 361 void tcp_wfree(struct sk_buff *skb); 362 void tcp_write_timer_handler(struct sock *sk); 363 void tcp_delack_timer_handler(struct sock *sk); 364 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 365 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 366 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 367 const struct tcphdr *th); 368 void tcp_rcv_space_adjust(struct sock *sk); 369 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 370 void tcp_twsk_destructor(struct sock *sk); 371 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 372 struct pipe_inode_info *pipe, size_t len, 373 unsigned int flags); 374 375 static inline void tcp_dec_quickack_mode(struct sock *sk, 376 const unsigned int pkts) 377 { 378 struct inet_connection_sock *icsk = inet_csk(sk); 379 380 if (icsk->icsk_ack.quick) { 381 if (pkts >= icsk->icsk_ack.quick) { 382 icsk->icsk_ack.quick = 0; 383 /* Leaving quickack mode we deflate ATO. */ 384 icsk->icsk_ack.ato = TCP_ATO_MIN; 385 } else 386 icsk->icsk_ack.quick -= pkts; 387 } 388 } 389 390 #define TCP_ECN_OK 1 391 #define TCP_ECN_QUEUE_CWR 2 392 #define TCP_ECN_DEMAND_CWR 4 393 #define TCP_ECN_SEEN 8 394 395 enum tcp_tw_status { 396 TCP_TW_SUCCESS = 0, 397 TCP_TW_RST = 1, 398 TCP_TW_ACK = 2, 399 TCP_TW_SYN = 3 400 }; 401 402 403 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 404 struct sk_buff *skb, 405 const struct tcphdr *th); 406 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 407 struct request_sock *req, bool fastopen); 408 int tcp_child_process(struct sock *parent, struct sock *child, 409 struct sk_buff *skb); 410 void tcp_enter_loss(struct sock *sk); 411 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag); 412 void tcp_clear_retrans(struct tcp_sock *tp); 413 void tcp_update_metrics(struct sock *sk); 414 void tcp_init_metrics(struct sock *sk); 415 void tcp_metrics_init(void); 416 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 417 void tcp_disable_fack(struct tcp_sock *tp); 418 void tcp_close(struct sock *sk, long timeout); 419 void tcp_init_sock(struct sock *sk); 420 unsigned int tcp_poll(struct file *file, struct socket *sock, 421 struct poll_table_struct *wait); 422 int tcp_getsockopt(struct sock *sk, int level, int optname, 423 char __user *optval, int __user *optlen); 424 int tcp_setsockopt(struct sock *sk, int level, int optname, 425 char __user *optval, unsigned int optlen); 426 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 427 char __user *optval, int __user *optlen); 428 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 429 char __user *optval, unsigned int optlen); 430 void tcp_set_keepalive(struct sock *sk, int val); 431 void tcp_syn_ack_timeout(const struct request_sock *req); 432 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 433 int flags, int *addr_len); 434 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 435 struct tcp_options_received *opt_rx, 436 int estab, struct tcp_fastopen_cookie *foc); 437 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 438 439 /* 440 * TCP v4 functions exported for the inet6 API 441 */ 442 443 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 444 void tcp_v4_mtu_reduced(struct sock *sk); 445 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 446 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 447 struct sock *tcp_create_openreq_child(const struct sock *sk, 448 struct request_sock *req, 449 struct sk_buff *skb); 450 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 451 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 452 struct request_sock *req, 453 struct dst_entry *dst, 454 struct request_sock *req_unhash, 455 bool *own_req); 456 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 457 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 458 int tcp_connect(struct sock *sk); 459 enum tcp_synack_type { 460 TCP_SYNACK_NORMAL, 461 TCP_SYNACK_FASTOPEN, 462 TCP_SYNACK_COOKIE, 463 }; 464 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 465 struct request_sock *req, 466 struct tcp_fastopen_cookie *foc, 467 enum tcp_synack_type synack_type); 468 int tcp_disconnect(struct sock *sk, int flags); 469 470 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 471 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 472 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 473 474 /* From syncookies.c */ 475 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 476 struct request_sock *req, 477 struct dst_entry *dst, u32 tsoff); 478 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 479 u32 cookie); 480 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 481 #ifdef CONFIG_SYN_COOKIES 482 483 /* Syncookies use a monotonic timer which increments every 60 seconds. 484 * This counter is used both as a hash input and partially encoded into 485 * the cookie value. A cookie is only validated further if the delta 486 * between the current counter value and the encoded one is less than this, 487 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 488 * the counter advances immediately after a cookie is generated). 489 */ 490 #define MAX_SYNCOOKIE_AGE 2 491 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 492 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 493 494 /* syncookies: remember time of last synqueue overflow 495 * But do not dirty this field too often (once per second is enough) 496 * It is racy as we do not hold a lock, but race is very minor. 497 */ 498 static inline void tcp_synq_overflow(const struct sock *sk) 499 { 500 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 501 unsigned long now = jiffies; 502 503 if (time_after(now, last_overflow + HZ)) 504 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 505 } 506 507 /* syncookies: no recent synqueue overflow on this listening socket? */ 508 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 509 { 510 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 511 512 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID); 513 } 514 515 static inline u32 tcp_cookie_time(void) 516 { 517 u64 val = get_jiffies_64(); 518 519 do_div(val, TCP_SYNCOOKIE_PERIOD); 520 return val; 521 } 522 523 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 524 u16 *mssp); 525 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 526 u64 cookie_init_timestamp(struct request_sock *req); 527 bool cookie_timestamp_decode(const struct net *net, 528 struct tcp_options_received *opt); 529 bool cookie_ecn_ok(const struct tcp_options_received *opt, 530 const struct net *net, const struct dst_entry *dst); 531 532 /* From net/ipv6/syncookies.c */ 533 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 534 u32 cookie); 535 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 536 537 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 538 const struct tcphdr *th, u16 *mssp); 539 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 540 #endif 541 /* tcp_output.c */ 542 543 u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now, 544 int min_tso_segs); 545 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 546 int nonagle); 547 bool tcp_may_send_now(struct sock *sk); 548 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 549 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 550 void tcp_retransmit_timer(struct sock *sk); 551 void tcp_xmit_retransmit_queue(struct sock *); 552 void tcp_simple_retransmit(struct sock *); 553 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 554 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 555 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t); 556 557 void tcp_send_probe0(struct sock *); 558 void tcp_send_partial(struct sock *); 559 int tcp_write_wakeup(struct sock *, int mib); 560 void tcp_send_fin(struct sock *sk); 561 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 562 int tcp_send_synack(struct sock *); 563 void tcp_push_one(struct sock *, unsigned int mss_now); 564 void tcp_send_ack(struct sock *sk); 565 void tcp_send_delayed_ack(struct sock *sk); 566 void tcp_send_loss_probe(struct sock *sk); 567 bool tcp_schedule_loss_probe(struct sock *sk); 568 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 569 const struct sk_buff *next_skb); 570 571 /* tcp_input.c */ 572 void tcp_rearm_rto(struct sock *sk); 573 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 574 void tcp_reset(struct sock *sk); 575 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 576 void tcp_fin(struct sock *sk); 577 578 /* tcp_timer.c */ 579 void tcp_init_xmit_timers(struct sock *); 580 static inline void tcp_clear_xmit_timers(struct sock *sk) 581 { 582 hrtimer_cancel(&tcp_sk(sk)->pacing_timer); 583 inet_csk_clear_xmit_timers(sk); 584 } 585 586 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 587 unsigned int tcp_current_mss(struct sock *sk); 588 589 /* Bound MSS / TSO packet size with the half of the window */ 590 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 591 { 592 int cutoff; 593 594 /* When peer uses tiny windows, there is no use in packetizing 595 * to sub-MSS pieces for the sake of SWS or making sure there 596 * are enough packets in the pipe for fast recovery. 597 * 598 * On the other hand, for extremely large MSS devices, handling 599 * smaller than MSS windows in this way does make sense. 600 */ 601 if (tp->max_window > TCP_MSS_DEFAULT) 602 cutoff = (tp->max_window >> 1); 603 else 604 cutoff = tp->max_window; 605 606 if (cutoff && pktsize > cutoff) 607 return max_t(int, cutoff, 68U - tp->tcp_header_len); 608 else 609 return pktsize; 610 } 611 612 /* tcp.c */ 613 void tcp_get_info(struct sock *, struct tcp_info *); 614 615 /* Read 'sendfile()'-style from a TCP socket */ 616 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 617 sk_read_actor_t recv_actor); 618 619 void tcp_initialize_rcv_mss(struct sock *sk); 620 621 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 622 int tcp_mss_to_mtu(struct sock *sk, int mss); 623 void tcp_mtup_init(struct sock *sk); 624 void tcp_init_buffer_space(struct sock *sk); 625 626 static inline void tcp_bound_rto(const struct sock *sk) 627 { 628 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 629 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 630 } 631 632 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 633 { 634 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 635 } 636 637 /* Compute the actual rto_min value */ 638 static inline u32 tcp_rto_min(struct sock *sk) 639 { 640 const struct dst_entry *dst = __sk_dst_get(sk); 641 u32 rto_min = TCP_RTO_MIN; 642 643 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 644 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 645 return rto_min; 646 } 647 648 static inline u32 tcp_rto_min_us(struct sock *sk) 649 { 650 return jiffies_to_usecs(tcp_rto_min(sk)); 651 } 652 653 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 654 { 655 return dst_metric_locked(dst, RTAX_CC_ALGO); 656 } 657 658 /* Minimum RTT in usec. ~0 means not available. */ 659 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 660 { 661 return minmax_get(&tp->rtt_min); 662 } 663 664 /* Compute the actual receive window we are currently advertising. 665 * Rcv_nxt can be after the window if our peer push more data 666 * than the offered window. 667 */ 668 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 669 { 670 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 671 672 if (win < 0) 673 win = 0; 674 return (u32) win; 675 } 676 677 /* Choose a new window, without checks for shrinking, and without 678 * scaling applied to the result. The caller does these things 679 * if necessary. This is a "raw" window selection. 680 */ 681 u32 __tcp_select_window(struct sock *sk); 682 683 void tcp_send_window_probe(struct sock *sk); 684 685 /* TCP uses 32bit jiffies to save some space. 686 * Note that this is different from tcp_time_stamp, which 687 * historically has been the same until linux-4.13. 688 */ 689 #define tcp_jiffies32 ((u32)jiffies) 690 691 /* 692 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 693 * It is no longer tied to jiffies, but to 1 ms clock. 694 * Note: double check if you want to use tcp_jiffies32 instead of this. 695 */ 696 #define TCP_TS_HZ 1000 697 698 static inline u64 tcp_clock_ns(void) 699 { 700 return local_clock(); 701 } 702 703 static inline u64 tcp_clock_us(void) 704 { 705 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 706 } 707 708 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 709 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 710 { 711 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 712 } 713 714 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 715 static inline u32 tcp_time_stamp_raw(void) 716 { 717 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ); 718 } 719 720 721 /* Refresh 1us clock of a TCP socket, 722 * ensuring monotically increasing values. 723 */ 724 static inline void tcp_mstamp_refresh(struct tcp_sock *tp) 725 { 726 u64 val = tcp_clock_us(); 727 728 if (val > tp->tcp_mstamp) 729 tp->tcp_mstamp = val; 730 } 731 732 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 733 { 734 return max_t(s64, t1 - t0, 0); 735 } 736 737 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 738 { 739 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ); 740 } 741 742 743 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 744 745 #define TCPHDR_FIN 0x01 746 #define TCPHDR_SYN 0x02 747 #define TCPHDR_RST 0x04 748 #define TCPHDR_PSH 0x08 749 #define TCPHDR_ACK 0x10 750 #define TCPHDR_URG 0x20 751 #define TCPHDR_ECE 0x40 752 #define TCPHDR_CWR 0x80 753 754 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 755 756 /* This is what the send packet queuing engine uses to pass 757 * TCP per-packet control information to the transmission code. 758 * We also store the host-order sequence numbers in here too. 759 * This is 44 bytes if IPV6 is enabled. 760 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 761 */ 762 struct tcp_skb_cb { 763 __u32 seq; /* Starting sequence number */ 764 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 765 union { 766 /* Note : tcp_tw_isn is used in input path only 767 * (isn chosen by tcp_timewait_state_process()) 768 * 769 * tcp_gso_segs/size are used in write queue only, 770 * cf tcp_skb_pcount()/tcp_skb_mss() 771 */ 772 __u32 tcp_tw_isn; 773 struct { 774 u16 tcp_gso_segs; 775 u16 tcp_gso_size; 776 }; 777 778 /* Used to stash the receive timestamp while this skb is in the 779 * out of order queue, as skb->tstamp is overwritten by the 780 * rbnode. 781 */ 782 ktime_t swtstamp; 783 }; 784 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 785 786 __u8 sacked; /* State flags for SACK/FACK. */ 787 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 788 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 789 #define TCPCB_LOST 0x04 /* SKB is lost */ 790 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 791 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 792 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 793 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 794 TCPCB_REPAIRED) 795 796 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 797 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 798 eor:1, /* Is skb MSG_EOR marked? */ 799 has_rxtstamp:1, /* SKB has a RX timestamp */ 800 unused:5; 801 __u32 ack_seq; /* Sequence number ACK'd */ 802 union { 803 struct { 804 /* There is space for up to 24 bytes */ 805 __u32 in_flight:30,/* Bytes in flight at transmit */ 806 is_app_limited:1, /* cwnd not fully used? */ 807 unused:1; 808 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 809 __u32 delivered; 810 /* start of send pipeline phase */ 811 u64 first_tx_mstamp; 812 /* when we reached the "delivered" count */ 813 u64 delivered_mstamp; 814 } tx; /* only used for outgoing skbs */ 815 union { 816 struct inet_skb_parm h4; 817 #if IS_ENABLED(CONFIG_IPV6) 818 struct inet6_skb_parm h6; 819 #endif 820 } header; /* For incoming skbs */ 821 }; 822 }; 823 824 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 825 826 827 #if IS_ENABLED(CONFIG_IPV6) 828 /* This is the variant of inet6_iif() that must be used by TCP, 829 * as TCP moves IP6CB into a different location in skb->cb[] 830 */ 831 static inline int tcp_v6_iif(const struct sk_buff *skb) 832 { 833 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 834 835 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 836 } 837 838 /* TCP_SKB_CB reference means this can not be used from early demux */ 839 static inline int tcp_v6_sdif(const struct sk_buff *skb) 840 { 841 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 842 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 843 return TCP_SKB_CB(skb)->header.h6.iif; 844 #endif 845 return 0; 846 } 847 #endif 848 849 /* TCP_SKB_CB reference means this can not be used from early demux */ 850 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) 851 { 852 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 853 if (!net->ipv4.sysctl_tcp_l3mdev_accept && 854 skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 855 return true; 856 #endif 857 return false; 858 } 859 860 /* TCP_SKB_CB reference means this can not be used from early demux */ 861 static inline int tcp_v4_sdif(struct sk_buff *skb) 862 { 863 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 864 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 865 return TCP_SKB_CB(skb)->header.h4.iif; 866 #endif 867 return 0; 868 } 869 870 /* Due to TSO, an SKB can be composed of multiple actual 871 * packets. To keep these tracked properly, we use this. 872 */ 873 static inline int tcp_skb_pcount(const struct sk_buff *skb) 874 { 875 return TCP_SKB_CB(skb)->tcp_gso_segs; 876 } 877 878 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 879 { 880 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 881 } 882 883 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 884 { 885 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 886 } 887 888 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 889 static inline int tcp_skb_mss(const struct sk_buff *skb) 890 { 891 return TCP_SKB_CB(skb)->tcp_gso_size; 892 } 893 894 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 895 { 896 return likely(!TCP_SKB_CB(skb)->eor); 897 } 898 899 /* Events passed to congestion control interface */ 900 enum tcp_ca_event { 901 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 902 CA_EVENT_CWND_RESTART, /* congestion window restart */ 903 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 904 CA_EVENT_LOSS, /* loss timeout */ 905 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 906 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 907 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 908 CA_EVENT_NON_DELAYED_ACK, 909 }; 910 911 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 912 enum tcp_ca_ack_event_flags { 913 CA_ACK_WIN_UPDATE = (1 << 0), /* ACK updated window */ 914 CA_ACK_ECE = (1 << 1), /* ECE bit is set on ack */ 915 }; 916 917 /* 918 * Interface for adding new TCP congestion control handlers 919 */ 920 #define TCP_CA_NAME_MAX 16 921 #define TCP_CA_MAX 128 922 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 923 924 #define TCP_CA_UNSPEC 0 925 926 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 927 #define TCP_CONG_NON_RESTRICTED 0x1 928 /* Requires ECN/ECT set on all packets */ 929 #define TCP_CONG_NEEDS_ECN 0x2 930 931 union tcp_cc_info; 932 933 struct ack_sample { 934 u32 pkts_acked; 935 s32 rtt_us; 936 u32 in_flight; 937 }; 938 939 /* A rate sample measures the number of (original/retransmitted) data 940 * packets delivered "delivered" over an interval of time "interval_us". 941 * The tcp_rate.c code fills in the rate sample, and congestion 942 * control modules that define a cong_control function to run at the end 943 * of ACK processing can optionally chose to consult this sample when 944 * setting cwnd and pacing rate. 945 * A sample is invalid if "delivered" or "interval_us" is negative. 946 */ 947 struct rate_sample { 948 u64 prior_mstamp; /* starting timestamp for interval */ 949 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 950 s32 delivered; /* number of packets delivered over interval */ 951 long interval_us; /* time for tp->delivered to incr "delivered" */ 952 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 953 int losses; /* number of packets marked lost upon ACK */ 954 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 955 u32 prior_in_flight; /* in flight before this ACK */ 956 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 957 bool is_retrans; /* is sample from retransmission? */ 958 }; 959 960 struct tcp_congestion_ops { 961 struct list_head list; 962 u32 key; 963 u32 flags; 964 965 /* initialize private data (optional) */ 966 void (*init)(struct sock *sk); 967 /* cleanup private data (optional) */ 968 void (*release)(struct sock *sk); 969 970 /* return slow start threshold (required) */ 971 u32 (*ssthresh)(struct sock *sk); 972 /* do new cwnd calculation (required) */ 973 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 974 /* call before changing ca_state (optional) */ 975 void (*set_state)(struct sock *sk, u8 new_state); 976 /* call when cwnd event occurs (optional) */ 977 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 978 /* call when ack arrives (optional) */ 979 void (*in_ack_event)(struct sock *sk, u32 flags); 980 /* new value of cwnd after loss (required) */ 981 u32 (*undo_cwnd)(struct sock *sk); 982 /* hook for packet ack accounting (optional) */ 983 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 984 /* suggest number of segments for each skb to transmit (optional) */ 985 u32 (*tso_segs_goal)(struct sock *sk); 986 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 987 u32 (*sndbuf_expand)(struct sock *sk); 988 /* call when packets are delivered to update cwnd and pacing rate, 989 * after all the ca_state processing. (optional) 990 */ 991 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 992 /* get info for inet_diag (optional) */ 993 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 994 union tcp_cc_info *info); 995 996 char name[TCP_CA_NAME_MAX]; 997 struct module *owner; 998 }; 999 1000 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1001 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1002 1003 void tcp_assign_congestion_control(struct sock *sk); 1004 void tcp_init_congestion_control(struct sock *sk); 1005 void tcp_cleanup_congestion_control(struct sock *sk); 1006 int tcp_set_default_congestion_control(const char *name); 1007 void tcp_get_default_congestion_control(char *name); 1008 void tcp_get_available_congestion_control(char *buf, size_t len); 1009 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1010 int tcp_set_allowed_congestion_control(char *allowed); 1011 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load); 1012 void tcp_reinit_congestion_control(struct sock *sk, 1013 const struct tcp_congestion_ops *ca); 1014 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 1015 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 1016 1017 u32 tcp_reno_ssthresh(struct sock *sk); 1018 u32 tcp_reno_undo_cwnd(struct sock *sk); 1019 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 1020 extern struct tcp_congestion_ops tcp_reno; 1021 1022 struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 1023 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca); 1024 #ifdef CONFIG_INET 1025 char *tcp_ca_get_name_by_key(u32 key, char *buffer); 1026 #else 1027 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 1028 { 1029 return NULL; 1030 } 1031 #endif 1032 1033 static inline bool tcp_ca_needs_ecn(const struct sock *sk) 1034 { 1035 const struct inet_connection_sock *icsk = inet_csk(sk); 1036 1037 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 1038 } 1039 1040 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 1041 { 1042 struct inet_connection_sock *icsk = inet_csk(sk); 1043 1044 if (icsk->icsk_ca_ops->set_state) 1045 icsk->icsk_ca_ops->set_state(sk, ca_state); 1046 icsk->icsk_ca_state = ca_state; 1047 } 1048 1049 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 1050 { 1051 const struct inet_connection_sock *icsk = inet_csk(sk); 1052 1053 if (icsk->icsk_ca_ops->cwnd_event) 1054 icsk->icsk_ca_ops->cwnd_event(sk, event); 1055 } 1056 1057 /* From tcp_rate.c */ 1058 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); 1059 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, 1060 struct rate_sample *rs); 1061 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, 1062 struct rate_sample *rs); 1063 void tcp_rate_check_app_limited(struct sock *sk); 1064 1065 /* These functions determine how the current flow behaves in respect of SACK 1066 * handling. SACK is negotiated with the peer, and therefore it can vary 1067 * between different flows. 1068 * 1069 * tcp_is_sack - SACK enabled 1070 * tcp_is_reno - No SACK 1071 * tcp_is_fack - FACK enabled, implies SACK enabled 1072 */ 1073 static inline int tcp_is_sack(const struct tcp_sock *tp) 1074 { 1075 return tp->rx_opt.sack_ok; 1076 } 1077 1078 static inline bool tcp_is_reno(const struct tcp_sock *tp) 1079 { 1080 return !tcp_is_sack(tp); 1081 } 1082 1083 static inline bool tcp_is_fack(const struct tcp_sock *tp) 1084 { 1085 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 1086 } 1087 1088 static inline void tcp_enable_fack(struct tcp_sock *tp) 1089 { 1090 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 1091 } 1092 1093 static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 1094 { 1095 return tp->sacked_out + tp->lost_out; 1096 } 1097 1098 /* This determines how many packets are "in the network" to the best 1099 * of our knowledge. In many cases it is conservative, but where 1100 * detailed information is available from the receiver (via SACK 1101 * blocks etc.) we can make more aggressive calculations. 1102 * 1103 * Use this for decisions involving congestion control, use just 1104 * tp->packets_out to determine if the send queue is empty or not. 1105 * 1106 * Read this equation as: 1107 * 1108 * "Packets sent once on transmission queue" MINUS 1109 * "Packets left network, but not honestly ACKed yet" PLUS 1110 * "Packets fast retransmitted" 1111 */ 1112 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 1113 { 1114 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 1115 } 1116 1117 #define TCP_INFINITE_SSTHRESH 0x7fffffff 1118 1119 static inline bool tcp_in_slow_start(const struct tcp_sock *tp) 1120 { 1121 return tp->snd_cwnd < tp->snd_ssthresh; 1122 } 1123 1124 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 1125 { 1126 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 1127 } 1128 1129 static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 1130 { 1131 return (TCPF_CA_CWR | TCPF_CA_Recovery) & 1132 (1 << inet_csk(sk)->icsk_ca_state); 1133 } 1134 1135 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 1136 * The exception is cwnd reduction phase, when cwnd is decreasing towards 1137 * ssthresh. 1138 */ 1139 static inline __u32 tcp_current_ssthresh(const struct sock *sk) 1140 { 1141 const struct tcp_sock *tp = tcp_sk(sk); 1142 1143 if (tcp_in_cwnd_reduction(sk)) 1144 return tp->snd_ssthresh; 1145 else 1146 return max(tp->snd_ssthresh, 1147 ((tp->snd_cwnd >> 1) + 1148 (tp->snd_cwnd >> 2))); 1149 } 1150 1151 /* Use define here intentionally to get WARN_ON location shown at the caller */ 1152 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 1153 1154 void tcp_enter_cwr(struct sock *sk); 1155 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 1156 1157 /* The maximum number of MSS of available cwnd for which TSO defers 1158 * sending if not using sysctl_tcp_tso_win_divisor. 1159 */ 1160 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 1161 { 1162 return 3; 1163 } 1164 1165 /* Returns end sequence number of the receiver's advertised window */ 1166 static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 1167 { 1168 return tp->snd_una + tp->snd_wnd; 1169 } 1170 1171 /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1172 * flexible approach. The RFC suggests cwnd should not be raised unless 1173 * it was fully used previously. And that's exactly what we do in 1174 * congestion avoidance mode. But in slow start we allow cwnd to grow 1175 * as long as the application has used half the cwnd. 1176 * Example : 1177 * cwnd is 10 (IW10), but application sends 9 frames. 1178 * We allow cwnd to reach 18 when all frames are ACKed. 1179 * This check is safe because it's as aggressive as slow start which already 1180 * risks 100% overshoot. The advantage is that we discourage application to 1181 * either send more filler packets or data to artificially blow up the cwnd 1182 * usage, and allow application-limited process to probe bw more aggressively. 1183 */ 1184 static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1185 { 1186 const struct tcp_sock *tp = tcp_sk(sk); 1187 1188 /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1189 if (tcp_in_slow_start(tp)) 1190 return tp->snd_cwnd < 2 * tp->max_packets_out; 1191 1192 return tp->is_cwnd_limited; 1193 } 1194 1195 /* Something is really bad, we could not queue an additional packet, 1196 * because qdisc is full or receiver sent a 0 window. 1197 * We do not want to add fuel to the fire, or abort too early, 1198 * so make sure the timer we arm now is at least 200ms in the future, 1199 * regardless of current icsk_rto value (as it could be ~2ms) 1200 */ 1201 static inline unsigned long tcp_probe0_base(const struct sock *sk) 1202 { 1203 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); 1204 } 1205 1206 /* Variant of inet_csk_rto_backoff() used for zero window probes */ 1207 static inline unsigned long tcp_probe0_when(const struct sock *sk, 1208 unsigned long max_when) 1209 { 1210 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; 1211 1212 return (unsigned long)min_t(u64, when, max_when); 1213 } 1214 1215 static inline void tcp_check_probe_timer(struct sock *sk) 1216 { 1217 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) 1218 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 1219 tcp_probe0_base(sk), TCP_RTO_MAX); 1220 } 1221 1222 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 1223 { 1224 tp->snd_wl1 = seq; 1225 } 1226 1227 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 1228 { 1229 tp->snd_wl1 = seq; 1230 } 1231 1232 /* 1233 * Calculate(/check) TCP checksum 1234 */ 1235 static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1236 __be32 daddr, __wsum base) 1237 { 1238 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 1239 } 1240 1241 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 1242 { 1243 return __skb_checksum_complete(skb); 1244 } 1245 1246 static inline bool tcp_checksum_complete(struct sk_buff *skb) 1247 { 1248 return !skb_csum_unnecessary(skb) && 1249 __tcp_checksum_complete(skb); 1250 } 1251 1252 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); 1253 int tcp_filter(struct sock *sk, struct sk_buff *skb); 1254 1255 #undef STATE_TRACE 1256 1257 #ifdef STATE_TRACE 1258 static const char *statename[]={ 1259 "Unused","Established","Syn Sent","Syn Recv", 1260 "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 1261 "Close Wait","Last ACK","Listen","Closing" 1262 }; 1263 #endif 1264 void tcp_set_state(struct sock *sk, int state); 1265 1266 void tcp_done(struct sock *sk); 1267 1268 int tcp_abort(struct sock *sk, int err); 1269 1270 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 1271 { 1272 rx_opt->dsack = 0; 1273 rx_opt->num_sacks = 0; 1274 } 1275 1276 u32 tcp_default_init_rwnd(u32 mss); 1277 void tcp_cwnd_restart(struct sock *sk, s32 delta); 1278 1279 static inline void tcp_slow_start_after_idle_check(struct sock *sk) 1280 { 1281 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; 1282 struct tcp_sock *tp = tcp_sk(sk); 1283 s32 delta; 1284 1285 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out || 1286 ca_ops->cong_control) 1287 return; 1288 delta = tcp_jiffies32 - tp->lsndtime; 1289 if (delta > inet_csk(sk)->icsk_rto) 1290 tcp_cwnd_restart(sk, delta); 1291 } 1292 1293 /* Determine a window scaling and initial window to offer. */ 1294 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 1295 __u32 *window_clamp, int wscale_ok, 1296 __u8 *rcv_wscale, __u32 init_rcv_wnd); 1297 1298 static inline int tcp_win_from_space(int space) 1299 { 1300 int tcp_adv_win_scale = sysctl_tcp_adv_win_scale; 1301 1302 return tcp_adv_win_scale <= 0 ? 1303 (space>>(-tcp_adv_win_scale)) : 1304 space - (space>>tcp_adv_win_scale); 1305 } 1306 1307 /* Note: caller must be prepared to deal with negative returns */ 1308 static inline int tcp_space(const struct sock *sk) 1309 { 1310 return tcp_win_from_space(sk->sk_rcvbuf - 1311 atomic_read(&sk->sk_rmem_alloc)); 1312 } 1313 1314 static inline int tcp_full_space(const struct sock *sk) 1315 { 1316 return tcp_win_from_space(sk->sk_rcvbuf); 1317 } 1318 1319 extern void tcp_openreq_init_rwin(struct request_sock *req, 1320 const struct sock *sk_listener, 1321 const struct dst_entry *dst); 1322 1323 void tcp_enter_memory_pressure(struct sock *sk); 1324 void tcp_leave_memory_pressure(struct sock *sk); 1325 1326 static inline int keepalive_intvl_when(const struct tcp_sock *tp) 1327 { 1328 struct net *net = sock_net((struct sock *)tp); 1329 1330 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; 1331 } 1332 1333 static inline int keepalive_time_when(const struct tcp_sock *tp) 1334 { 1335 struct net *net = sock_net((struct sock *)tp); 1336 1337 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; 1338 } 1339 1340 static inline int keepalive_probes(const struct tcp_sock *tp) 1341 { 1342 struct net *net = sock_net((struct sock *)tp); 1343 1344 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; 1345 } 1346 1347 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 1348 { 1349 const struct inet_connection_sock *icsk = &tp->inet_conn; 1350 1351 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, 1352 tcp_jiffies32 - tp->rcv_tstamp); 1353 } 1354 1355 static inline int tcp_fin_time(const struct sock *sk) 1356 { 1357 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; 1358 const int rto = inet_csk(sk)->icsk_rto; 1359 1360 if (fin_timeout < (rto << 2) - (rto >> 1)) 1361 fin_timeout = (rto << 2) - (rto >> 1); 1362 1363 return fin_timeout; 1364 } 1365 1366 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1367 int paws_win) 1368 { 1369 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1370 return true; 1371 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1372 return true; 1373 /* 1374 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1375 * then following tcp messages have valid values. Ignore 0 value, 1376 * or else 'negative' tsval might forbid us to accept their packets. 1377 */ 1378 if (!rx_opt->ts_recent) 1379 return true; 1380 return false; 1381 } 1382 1383 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1384 int rst) 1385 { 1386 if (tcp_paws_check(rx_opt, 0)) 1387 return false; 1388 1389 /* RST segments are not recommended to carry timestamp, 1390 and, if they do, it is recommended to ignore PAWS because 1391 "their cleanup function should take precedence over timestamps." 1392 Certainly, it is mistake. It is necessary to understand the reasons 1393 of this constraint to relax it: if peer reboots, clock may go 1394 out-of-sync and half-open connections will not be reset. 1395 Actually, the problem would be not existing if all 1396 the implementations followed draft about maintaining clock 1397 via reboots. Linux-2.2 DOES NOT! 1398 1399 However, we can relax time bounds for RST segments to MSL. 1400 */ 1401 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1402 return false; 1403 return true; 1404 } 1405 1406 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 1407 int mib_idx, u32 *last_oow_ack_time); 1408 1409 static inline void tcp_mib_init(struct net *net) 1410 { 1411 /* See RFC 2012 */ 1412 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); 1413 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1414 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1415 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); 1416 } 1417 1418 /* from STCP */ 1419 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 1420 { 1421 tp->lost_skb_hint = NULL; 1422 } 1423 1424 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1425 { 1426 tcp_clear_retrans_hints_partial(tp); 1427 tp->retransmit_skb_hint = NULL; 1428 } 1429 1430 union tcp_md5_addr { 1431 struct in_addr a4; 1432 #if IS_ENABLED(CONFIG_IPV6) 1433 struct in6_addr a6; 1434 #endif 1435 }; 1436 1437 /* - key database */ 1438 struct tcp_md5sig_key { 1439 struct hlist_node node; 1440 u8 keylen; 1441 u8 family; /* AF_INET or AF_INET6 */ 1442 union tcp_md5_addr addr; 1443 u8 prefixlen; 1444 u8 key[TCP_MD5SIG_MAXKEYLEN]; 1445 struct rcu_head rcu; 1446 }; 1447 1448 /* - sock block */ 1449 struct tcp_md5sig_info { 1450 struct hlist_head head; 1451 struct rcu_head rcu; 1452 }; 1453 1454 /* - pseudo header */ 1455 struct tcp4_pseudohdr { 1456 __be32 saddr; 1457 __be32 daddr; 1458 __u8 pad; 1459 __u8 protocol; 1460 __be16 len; 1461 }; 1462 1463 struct tcp6_pseudohdr { 1464 struct in6_addr saddr; 1465 struct in6_addr daddr; 1466 __be32 len; 1467 __be32 protocol; /* including padding */ 1468 }; 1469 1470 union tcp_md5sum_block { 1471 struct tcp4_pseudohdr ip4; 1472 #if IS_ENABLED(CONFIG_IPV6) 1473 struct tcp6_pseudohdr ip6; 1474 #endif 1475 }; 1476 1477 /* - pool: digest algorithm, hash description and scratch buffer */ 1478 struct tcp_md5sig_pool { 1479 struct ahash_request *md5_req; 1480 void *scratch; 1481 }; 1482 1483 /* - functions */ 1484 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 1485 const struct sock *sk, const struct sk_buff *skb); 1486 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 1487 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen, 1488 gfp_t gfp); 1489 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1490 int family, u8 prefixlen); 1491 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, 1492 const struct sock *addr_sk); 1493 1494 #ifdef CONFIG_TCP_MD5SIG 1495 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1496 const union tcp_md5_addr *addr, 1497 int family); 1498 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 1499 #else 1500 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, 1501 const union tcp_md5_addr *addr, 1502 int family) 1503 { 1504 return NULL; 1505 } 1506 #define tcp_twsk_md5_key(twsk) NULL 1507 #endif 1508 1509 bool tcp_alloc_md5sig_pool(void); 1510 1511 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 1512 static inline void tcp_put_md5sig_pool(void) 1513 { 1514 local_bh_enable(); 1515 } 1516 1517 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 1518 unsigned int header_len); 1519 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1520 const struct tcp_md5sig_key *key); 1521 1522 /* From tcp_fastopen.c */ 1523 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 1524 struct tcp_fastopen_cookie *cookie, int *syn_loss, 1525 unsigned long *last_syn_loss); 1526 void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1527 struct tcp_fastopen_cookie *cookie, bool syn_lost, 1528 u16 try_exp); 1529 struct tcp_fastopen_request { 1530 /* Fast Open cookie. Size 0 means a cookie request */ 1531 struct tcp_fastopen_cookie cookie; 1532 struct msghdr *data; /* data in MSG_FASTOPEN */ 1533 size_t size; 1534 int copied; /* queued in tcp_connect() */ 1535 }; 1536 void tcp_free_fastopen_req(struct tcp_sock *tp); 1537 1538 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 1539 int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1540 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1541 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1542 struct request_sock *req, 1543 struct tcp_fastopen_cookie *foc); 1544 void tcp_fastopen_init_key_once(bool publish); 1545 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1546 struct tcp_fastopen_cookie *cookie); 1547 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1548 #define TCP_FASTOPEN_KEY_LENGTH 16 1549 1550 /* Fastopen key context */ 1551 struct tcp_fastopen_context { 1552 struct crypto_cipher *tfm; 1553 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1554 struct rcu_head rcu; 1555 }; 1556 1557 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1558 void tcp_fastopen_active_disable(struct sock *sk); 1559 bool tcp_fastopen_active_should_disable(struct sock *sk); 1560 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1561 void tcp_fastopen_active_timeout_reset(void); 1562 1563 /* Latencies incurred by various limits for a sender. They are 1564 * chronograph-like stats that are mutually exclusive. 1565 */ 1566 enum tcp_chrono { 1567 TCP_CHRONO_UNSPEC, 1568 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1569 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1570 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1571 __TCP_CHRONO_MAX, 1572 }; 1573 1574 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1575 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1576 1577 /* write queue abstraction */ 1578 static inline void tcp_write_queue_purge(struct sock *sk) 1579 { 1580 struct sk_buff *skb; 1581 1582 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1583 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 1584 sk_wmem_free_skb(sk, skb); 1585 sk_mem_reclaim(sk); 1586 tcp_clear_all_retrans_hints(tcp_sk(sk)); 1587 } 1588 1589 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1590 { 1591 return skb_peek(&sk->sk_write_queue); 1592 } 1593 1594 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1595 { 1596 return skb_peek_tail(&sk->sk_write_queue); 1597 } 1598 1599 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1600 const struct sk_buff *skb) 1601 { 1602 return skb_queue_next(&sk->sk_write_queue, skb); 1603 } 1604 1605 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1606 const struct sk_buff *skb) 1607 { 1608 return skb_queue_prev(&sk->sk_write_queue, skb); 1609 } 1610 1611 #define tcp_for_write_queue(skb, sk) \ 1612 skb_queue_walk(&(sk)->sk_write_queue, skb) 1613 1614 #define tcp_for_write_queue_from(skb, sk) \ 1615 skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1616 1617 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1618 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1619 1620 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1621 { 1622 return sk->sk_send_head; 1623 } 1624 1625 static inline bool tcp_skb_is_last(const struct sock *sk, 1626 const struct sk_buff *skb) 1627 { 1628 return skb_queue_is_last(&sk->sk_write_queue, skb); 1629 } 1630 1631 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1632 { 1633 if (tcp_skb_is_last(sk, skb)) 1634 sk->sk_send_head = NULL; 1635 else 1636 sk->sk_send_head = tcp_write_queue_next(sk, skb); 1637 } 1638 1639 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1640 { 1641 if (sk->sk_send_head == skb_unlinked) { 1642 sk->sk_send_head = NULL; 1643 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1644 } 1645 if (tcp_sk(sk)->highest_sack == skb_unlinked) 1646 tcp_sk(sk)->highest_sack = NULL; 1647 } 1648 1649 static inline void tcp_init_send_head(struct sock *sk) 1650 { 1651 sk->sk_send_head = NULL; 1652 } 1653 1654 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1655 { 1656 __skb_queue_tail(&sk->sk_write_queue, skb); 1657 } 1658 1659 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1660 { 1661 __tcp_add_write_queue_tail(sk, skb); 1662 1663 /* Queue it, remembering where we must start sending. */ 1664 if (sk->sk_send_head == NULL) { 1665 sk->sk_send_head = skb; 1666 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1667 1668 if (tcp_sk(sk)->highest_sack == NULL) 1669 tcp_sk(sk)->highest_sack = skb; 1670 } 1671 } 1672 1673 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1674 { 1675 __skb_queue_head(&sk->sk_write_queue, skb); 1676 } 1677 1678 /* Insert buff after skb on the write queue of sk. */ 1679 static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1680 struct sk_buff *buff, 1681 struct sock *sk) 1682 { 1683 __skb_queue_after(&sk->sk_write_queue, skb, buff); 1684 } 1685 1686 /* Insert new before skb on the write queue of sk. */ 1687 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1688 struct sk_buff *skb, 1689 struct sock *sk) 1690 { 1691 __skb_queue_before(&sk->sk_write_queue, skb, new); 1692 1693 if (sk->sk_send_head == skb) 1694 sk->sk_send_head = new; 1695 } 1696 1697 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1698 { 1699 __skb_unlink(skb, &sk->sk_write_queue); 1700 } 1701 1702 static inline bool tcp_write_queue_empty(struct sock *sk) 1703 { 1704 return skb_queue_empty(&sk->sk_write_queue); 1705 } 1706 1707 static inline void tcp_push_pending_frames(struct sock *sk) 1708 { 1709 if (tcp_send_head(sk)) { 1710 struct tcp_sock *tp = tcp_sk(sk); 1711 1712 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1713 } 1714 } 1715 1716 /* Start sequence of the skb just after the highest skb with SACKed 1717 * bit, valid only if sacked_out > 0 or when the caller has ensured 1718 * validity by itself. 1719 */ 1720 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1721 { 1722 if (!tp->sacked_out) 1723 return tp->snd_una; 1724 1725 if (tp->highest_sack == NULL) 1726 return tp->snd_nxt; 1727 1728 return TCP_SKB_CB(tp->highest_sack)->seq; 1729 } 1730 1731 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1732 { 1733 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 1734 tcp_write_queue_next(sk, skb); 1735 } 1736 1737 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1738 { 1739 return tcp_sk(sk)->highest_sack; 1740 } 1741 1742 static inline void tcp_highest_sack_reset(struct sock *sk) 1743 { 1744 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 1745 } 1746 1747 /* Called when old skb is about to be deleted (to be combined with new skb) */ 1748 static inline void tcp_highest_sack_combine(struct sock *sk, 1749 struct sk_buff *old, 1750 struct sk_buff *new) 1751 { 1752 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 1753 tcp_sk(sk)->highest_sack = new; 1754 } 1755 1756 /* This helper checks if socket has IP_TRANSPARENT set */ 1757 static inline bool inet_sk_transparent(const struct sock *sk) 1758 { 1759 switch (sk->sk_state) { 1760 case TCP_TIME_WAIT: 1761 return inet_twsk(sk)->tw_transparent; 1762 case TCP_NEW_SYN_RECV: 1763 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1764 } 1765 return inet_sk(sk)->transparent; 1766 } 1767 1768 /* Determines whether this is a thin stream (which may suffer from 1769 * increased latency). Used to trigger latency-reducing mechanisms. 1770 */ 1771 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1772 { 1773 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1774 } 1775 1776 /* /proc */ 1777 enum tcp_seq_states { 1778 TCP_SEQ_STATE_LISTENING, 1779 TCP_SEQ_STATE_ESTABLISHED, 1780 }; 1781 1782 int tcp_seq_open(struct inode *inode, struct file *file); 1783 1784 struct tcp_seq_afinfo { 1785 char *name; 1786 sa_family_t family; 1787 const struct file_operations *seq_fops; 1788 struct seq_operations seq_ops; 1789 }; 1790 1791 struct tcp_iter_state { 1792 struct seq_net_private p; 1793 sa_family_t family; 1794 enum tcp_seq_states state; 1795 struct sock *syn_wait_sk; 1796 int bucket, offset, sbucket, num; 1797 loff_t last_pos; 1798 }; 1799 1800 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1801 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1802 1803 extern struct request_sock_ops tcp_request_sock_ops; 1804 extern struct request_sock_ops tcp6_request_sock_ops; 1805 1806 void tcp_v4_destroy_sock(struct sock *sk); 1807 1808 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1809 netdev_features_t features); 1810 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1811 int tcp_gro_complete(struct sk_buff *skb); 1812 1813 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1814 1815 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1816 { 1817 struct net *net = sock_net((struct sock *)tp); 1818 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1819 } 1820 1821 static inline bool tcp_stream_memory_free(const struct sock *sk) 1822 { 1823 const struct tcp_sock *tp = tcp_sk(sk); 1824 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1825 1826 return notsent_bytes < tcp_notsent_lowat(tp); 1827 } 1828 1829 #ifdef CONFIG_PROC_FS 1830 int tcp4_proc_init(void); 1831 void tcp4_proc_exit(void); 1832 #endif 1833 1834 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1835 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1836 const struct tcp_request_sock_ops *af_ops, 1837 struct sock *sk, struct sk_buff *skb); 1838 1839 /* TCP af-specific functions */ 1840 struct tcp_sock_af_ops { 1841 #ifdef CONFIG_TCP_MD5SIG 1842 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1843 const struct sock *addr_sk); 1844 int (*calc_md5_hash)(char *location, 1845 const struct tcp_md5sig_key *md5, 1846 const struct sock *sk, 1847 const struct sk_buff *skb); 1848 int (*md5_parse)(struct sock *sk, 1849 int optname, 1850 char __user *optval, 1851 int optlen); 1852 #endif 1853 }; 1854 1855 struct tcp_request_sock_ops { 1856 u16 mss_clamp; 1857 #ifdef CONFIG_TCP_MD5SIG 1858 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1859 const struct sock *addr_sk); 1860 int (*calc_md5_hash) (char *location, 1861 const struct tcp_md5sig_key *md5, 1862 const struct sock *sk, 1863 const struct sk_buff *skb); 1864 #endif 1865 void (*init_req)(struct request_sock *req, 1866 const struct sock *sk_listener, 1867 struct sk_buff *skb); 1868 #ifdef CONFIG_SYN_COOKIES 1869 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1870 __u16 *mss); 1871 #endif 1872 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1873 const struct request_sock *req); 1874 u32 (*init_seq)(const struct sk_buff *skb); 1875 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 1876 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1877 struct flowi *fl, struct request_sock *req, 1878 struct tcp_fastopen_cookie *foc, 1879 enum tcp_synack_type synack_type); 1880 }; 1881 1882 #ifdef CONFIG_SYN_COOKIES 1883 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1884 const struct sock *sk, struct sk_buff *skb, 1885 __u16 *mss) 1886 { 1887 tcp_synq_overflow(sk); 1888 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1889 return ops->cookie_init_seq(skb, mss); 1890 } 1891 #else 1892 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1893 const struct sock *sk, struct sk_buff *skb, 1894 __u16 *mss) 1895 { 1896 return 0; 1897 } 1898 #endif 1899 1900 int tcpv4_offload_init(void); 1901 1902 void tcp_v4_init(void); 1903 void tcp_init(void); 1904 1905 /* tcp_recovery.c */ 1906 extern void tcp_rack_mark_lost(struct sock *sk); 1907 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 1908 u64 xmit_time); 1909 extern void tcp_rack_reo_timeout(struct sock *sk); 1910 1911 /* At how many usecs into the future should the RTO fire? */ 1912 static inline s64 tcp_rto_delta_us(const struct sock *sk) 1913 { 1914 const struct sk_buff *skb = tcp_write_queue_head(sk); 1915 u32 rto = inet_csk(sk)->icsk_rto; 1916 u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto); 1917 1918 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 1919 } 1920 1921 /* 1922 * Save and compile IPv4 options, return a pointer to it 1923 */ 1924 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 1925 struct sk_buff *skb) 1926 { 1927 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1928 struct ip_options_rcu *dopt = NULL; 1929 1930 if (opt->optlen) { 1931 int opt_size = sizeof(*dopt) + opt->optlen; 1932 1933 dopt = kmalloc(opt_size, GFP_ATOMIC); 1934 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 1935 kfree(dopt); 1936 dopt = NULL; 1937 } 1938 } 1939 return dopt; 1940 } 1941 1942 /* locally generated TCP pure ACKs have skb->truesize == 2 1943 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1944 * This is much faster than dissecting the packet to find out. 1945 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1946 */ 1947 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1948 { 1949 return skb->truesize == 2; 1950 } 1951 1952 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1953 { 1954 skb->truesize = 2; 1955 } 1956 1957 static inline int tcp_inq(struct sock *sk) 1958 { 1959 struct tcp_sock *tp = tcp_sk(sk); 1960 int answ; 1961 1962 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1963 answ = 0; 1964 } else if (sock_flag(sk, SOCK_URGINLINE) || 1965 !tp->urg_data || 1966 before(tp->urg_seq, tp->copied_seq) || 1967 !before(tp->urg_seq, tp->rcv_nxt)) { 1968 1969 answ = tp->rcv_nxt - tp->copied_seq; 1970 1971 /* Subtract 1, if FIN was received */ 1972 if (answ && sock_flag(sk, SOCK_DONE)) 1973 answ--; 1974 } else { 1975 answ = tp->urg_seq - tp->copied_seq; 1976 } 1977 1978 return answ; 1979 } 1980 1981 int tcp_peek_len(struct socket *sock); 1982 1983 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 1984 { 1985 u16 segs_in; 1986 1987 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 1988 tp->segs_in += segs_in; 1989 if (skb->len > tcp_hdrlen(skb)) 1990 tp->data_segs_in += segs_in; 1991 } 1992 1993 /* 1994 * TCP listen path runs lockless. 1995 * We forced "struct sock" to be const qualified to make sure 1996 * we don't modify one of its field by mistake. 1997 * Here, we increment sk_drops which is an atomic_t, so we can safely 1998 * make sock writable again. 1999 */ 2000 static inline void tcp_listendrop(const struct sock *sk) 2001 { 2002 atomic_inc(&((struct sock *)sk)->sk_drops); 2003 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2004 } 2005 2006 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2007 2008 /* 2009 * Interface for adding Upper Level Protocols over TCP 2010 */ 2011 2012 #define TCP_ULP_NAME_MAX 16 2013 #define TCP_ULP_MAX 128 2014 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2015 2016 struct tcp_ulp_ops { 2017 struct list_head list; 2018 2019 /* initialize ulp */ 2020 int (*init)(struct sock *sk); 2021 /* cleanup ulp */ 2022 void (*release)(struct sock *sk); 2023 2024 char name[TCP_ULP_NAME_MAX]; 2025 struct module *owner; 2026 }; 2027 int tcp_register_ulp(struct tcp_ulp_ops *type); 2028 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2029 int tcp_set_ulp(struct sock *sk, const char *name); 2030 void tcp_get_available_ulp(char *buf, size_t len); 2031 void tcp_cleanup_ulp(struct sock *sk); 2032 2033 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2034 * is < 0, then the BPF op failed (for example if the loaded BPF 2035 * program does not support the chosen operation or there is no BPF 2036 * program loaded). 2037 */ 2038 #ifdef CONFIG_BPF 2039 static inline int tcp_call_bpf(struct sock *sk, int op) 2040 { 2041 struct bpf_sock_ops_kern sock_ops; 2042 int ret; 2043 2044 if (sk_fullsock(sk)) 2045 sock_owned_by_me(sk); 2046 2047 memset(&sock_ops, 0, sizeof(sock_ops)); 2048 sock_ops.sk = sk; 2049 sock_ops.op = op; 2050 2051 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2052 if (ret == 0) 2053 ret = sock_ops.reply; 2054 else 2055 ret = -1; 2056 return ret; 2057 } 2058 #else 2059 static inline int tcp_call_bpf(struct sock *sk, int op) 2060 { 2061 return -EPERM; 2062 } 2063 #endif 2064 2065 static inline u32 tcp_timeout_init(struct sock *sk) 2066 { 2067 int timeout; 2068 2069 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT); 2070 2071 if (timeout <= 0) 2072 timeout = TCP_TIMEOUT_INIT; 2073 return timeout; 2074 } 2075 2076 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2077 { 2078 int rwnd; 2079 2080 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT); 2081 2082 if (rwnd < 0) 2083 rwnd = 0; 2084 return rwnd; 2085 } 2086 2087 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2088 { 2089 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN) == 1); 2090 } 2091 #endif /* _TCP_H */ 2092