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 #include <linux/bpf-cgroup.h> 49 50 extern struct inet_hashinfo tcp_hashinfo; 51 52 extern struct percpu_counter tcp_orphan_count; 53 void tcp_time_wait(struct sock *sk, int state, int timeo); 54 55 #define MAX_TCP_HEADER (128 + MAX_HEADER) 56 #define MAX_TCP_OPTION_SPACE 40 57 58 /* 59 * Never offer a window over 32767 without using window scaling. Some 60 * poor stacks do signed 16bit maths! 61 */ 62 #define MAX_TCP_WINDOW 32767U 63 64 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 65 #define TCP_MIN_MSS 88U 66 67 /* The least MTU to use for probing */ 68 #define TCP_BASE_MSS 1024 69 70 /* probing interval, default to 10 minutes as per RFC4821 */ 71 #define TCP_PROBE_INTERVAL 600 72 73 /* Specify interval when tcp mtu probing will stop */ 74 #define TCP_PROBE_THRESHOLD 8 75 76 /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 77 #define TCP_FASTRETRANS_THRESH 3 78 79 /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 80 #define TCP_MAX_QUICKACKS 16U 81 82 /* Maximal number of window scale according to RFC1323 */ 83 #define TCP_MAX_WSCALE 14U 84 85 /* urg_data states */ 86 #define TCP_URG_VALID 0x0100 87 #define TCP_URG_NOTYET 0x0200 88 #define TCP_URG_READ 0x0400 89 90 #define TCP_RETR1 3 /* 91 * This is how many retries it does before it 92 * tries to figure out if the gateway is 93 * down. Minimal RFC value is 3; it corresponds 94 * to ~3sec-8min depending on RTO. 95 */ 96 97 #define TCP_RETR2 15 /* 98 * This should take at least 99 * 90 minutes to time out. 100 * RFC1122 says that the limit is 100 sec. 101 * 15 is ~13-30min depending on RTO. 102 */ 103 104 #define TCP_SYN_RETRIES 6 /* This is how many retries are done 105 * when active opening a connection. 106 * RFC1122 says the minimum retry MUST 107 * be at least 180secs. Nevertheless 108 * this value is corresponding to 109 * 63secs of retransmission with the 110 * current initial RTO. 111 */ 112 113 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 114 * when passive opening a connection. 115 * This is corresponding to 31secs of 116 * retransmission with the current 117 * initial RTO. 118 */ 119 120 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 121 * state, about 60 seconds */ 122 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 123 /* BSD style FIN_WAIT2 deadlock breaker. 124 * It used to be 3min, new value is 60sec, 125 * to combine FIN-WAIT-2 timeout with 126 * TIME-WAIT timer. 127 */ 128 129 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 130 #if HZ >= 100 131 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 132 #define TCP_ATO_MIN ((unsigned)(HZ/25)) 133 #else 134 #define TCP_DELACK_MIN 4U 135 #define TCP_ATO_MIN 4U 136 #endif 137 #define TCP_RTO_MAX ((unsigned)(120*HZ)) 138 #define TCP_RTO_MIN ((unsigned)(HZ/5)) 139 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ 140 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 141 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 142 * used as a fallback RTO for the 143 * initial data transmission if no 144 * valid RTT sample has been acquired, 145 * most likely due to retrans in 3WHS. 146 */ 147 148 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 149 * for local resources. 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 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 192 193 /* 194 * TCP option lengths 195 */ 196 197 #define TCPOLEN_MSS 4 198 #define TCPOLEN_WINDOW 3 199 #define TCPOLEN_SACK_PERM 2 200 #define TCPOLEN_TIMESTAMP 10 201 #define TCPOLEN_MD5SIG 18 202 #define TCPOLEN_FASTOPEN_BASE 2 203 #define TCPOLEN_EXP_FASTOPEN_BASE 4 204 #define TCPOLEN_EXP_SMC_BASE 6 205 206 /* But this is what stacks really send out. */ 207 #define TCPOLEN_TSTAMP_ALIGNED 12 208 #define TCPOLEN_WSCALE_ALIGNED 4 209 #define TCPOLEN_SACKPERM_ALIGNED 4 210 #define TCPOLEN_SACK_BASE 2 211 #define TCPOLEN_SACK_BASE_ALIGNED 4 212 #define TCPOLEN_SACK_PERBLOCK 8 213 #define TCPOLEN_MD5SIG_ALIGNED 20 214 #define TCPOLEN_MSS_ALIGNED 4 215 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 216 217 /* 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_max_orphans; 244 extern long sysctl_tcp_mem[3]; 245 extern int sysctl_tcp_wmem[3]; 246 extern int sysctl_tcp_rmem[3]; 247 248 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ 249 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ 250 251 extern atomic_long_t tcp_memory_allocated; 252 extern struct percpu_counter tcp_sockets_allocated; 253 extern unsigned long tcp_memory_pressure; 254 255 /* optimized version of sk_under_memory_pressure() for TCP sockets */ 256 static inline bool tcp_under_memory_pressure(const struct sock *sk) 257 { 258 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 259 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 260 return true; 261 262 return tcp_memory_pressure; 263 } 264 /* 265 * The next routines deal with comparing 32 bit unsigned ints 266 * and worry about wraparound (automatic with unsigned arithmetic). 267 */ 268 269 static inline bool before(__u32 seq1, __u32 seq2) 270 { 271 return (__s32)(seq1-seq2) < 0; 272 } 273 #define after(seq2, seq1) before(seq1, seq2) 274 275 /* is s2<=s1<=s3 ? */ 276 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 277 { 278 return seq3 - seq2 >= seq1 - seq2; 279 } 280 281 static inline bool tcp_out_of_memory(struct sock *sk) 282 { 283 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 284 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 285 return true; 286 return false; 287 } 288 289 void sk_forced_mem_schedule(struct sock *sk, int size); 290 291 static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 292 { 293 struct percpu_counter *ocp = sk->sk_prot->orphan_count; 294 int orphans = percpu_counter_read_positive(ocp); 295 296 if (orphans << shift > sysctl_tcp_max_orphans) { 297 orphans = percpu_counter_sum_positive(ocp); 298 if (orphans << shift > sysctl_tcp_max_orphans) 299 return true; 300 } 301 return false; 302 } 303 304 bool tcp_check_oom(struct sock *sk, int shift); 305 306 307 extern struct proto tcp_prot; 308 309 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 310 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) 311 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 312 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 313 314 void tcp_tasklet_init(void); 315 316 void tcp_v4_err(struct sk_buff *skb, u32); 317 318 void tcp_shutdown(struct sock *sk, int how); 319 320 int tcp_v4_early_demux(struct sk_buff *skb); 321 int tcp_v4_rcv(struct sk_buff *skb); 322 323 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 324 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 325 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); 326 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 327 int flags); 328 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, 329 size_t size, int flags); 330 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, 331 size_t size, int flags); 332 void tcp_release_cb(struct sock *sk); 333 void tcp_wfree(struct sk_buff *skb); 334 void tcp_write_timer_handler(struct sock *sk); 335 void tcp_delack_timer_handler(struct sock *sk); 336 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 337 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); 338 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 339 const struct tcphdr *th); 340 void tcp_rcv_space_adjust(struct sock *sk); 341 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 342 void tcp_twsk_destructor(struct sock *sk); 343 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 344 struct pipe_inode_info *pipe, size_t len, 345 unsigned int flags); 346 347 static inline void tcp_dec_quickack_mode(struct sock *sk, 348 const unsigned int pkts) 349 { 350 struct inet_connection_sock *icsk = inet_csk(sk); 351 352 if (icsk->icsk_ack.quick) { 353 if (pkts >= icsk->icsk_ack.quick) { 354 icsk->icsk_ack.quick = 0; 355 /* Leaving quickack mode we deflate ATO. */ 356 icsk->icsk_ack.ato = TCP_ATO_MIN; 357 } else 358 icsk->icsk_ack.quick -= pkts; 359 } 360 } 361 362 #define TCP_ECN_OK 1 363 #define TCP_ECN_QUEUE_CWR 2 364 #define TCP_ECN_DEMAND_CWR 4 365 #define TCP_ECN_SEEN 8 366 367 enum tcp_tw_status { 368 TCP_TW_SUCCESS = 0, 369 TCP_TW_RST = 1, 370 TCP_TW_ACK = 2, 371 TCP_TW_SYN = 3 372 }; 373 374 375 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 376 struct sk_buff *skb, 377 const struct tcphdr *th); 378 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 379 struct request_sock *req, bool fastopen); 380 int tcp_child_process(struct sock *parent, struct sock *child, 381 struct sk_buff *skb); 382 void tcp_enter_loss(struct sock *sk); 383 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag); 384 void tcp_clear_retrans(struct tcp_sock *tp); 385 void tcp_update_metrics(struct sock *sk); 386 void tcp_init_metrics(struct sock *sk); 387 void tcp_metrics_init(void); 388 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); 389 void tcp_disable_fack(struct tcp_sock *tp); 390 void tcp_close(struct sock *sk, long timeout); 391 void tcp_init_sock(struct sock *sk); 392 void tcp_init_transfer(struct sock *sk, int bpf_op); 393 unsigned int tcp_poll(struct file *file, struct socket *sock, 394 struct poll_table_struct *wait); 395 int tcp_getsockopt(struct sock *sk, int level, int optname, 396 char __user *optval, int __user *optlen); 397 int tcp_setsockopt(struct sock *sk, int level, int optname, 398 char __user *optval, unsigned int optlen); 399 int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 400 char __user *optval, int __user *optlen); 401 int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 402 char __user *optval, unsigned int optlen); 403 void tcp_set_keepalive(struct sock *sk, int val); 404 void tcp_syn_ack_timeout(const struct request_sock *req); 405 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 406 int flags, int *addr_len); 407 void tcp_parse_options(const struct net *net, const struct sk_buff *skb, 408 struct tcp_options_received *opt_rx, 409 int estab, struct tcp_fastopen_cookie *foc); 410 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 411 412 /* 413 * TCP v4 functions exported for the inet6 API 414 */ 415 416 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 417 void tcp_v4_mtu_reduced(struct sock *sk); 418 void tcp_req_err(struct sock *sk, u32 seq, bool abort); 419 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 420 struct sock *tcp_create_openreq_child(const struct sock *sk, 421 struct request_sock *req, 422 struct sk_buff *skb); 423 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 424 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, 425 struct request_sock *req, 426 struct dst_entry *dst, 427 struct request_sock *req_unhash, 428 bool *own_req); 429 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 430 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 431 int tcp_connect(struct sock *sk); 432 enum tcp_synack_type { 433 TCP_SYNACK_NORMAL, 434 TCP_SYNACK_FASTOPEN, 435 TCP_SYNACK_COOKIE, 436 }; 437 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, 438 struct request_sock *req, 439 struct tcp_fastopen_cookie *foc, 440 enum tcp_synack_type synack_type); 441 int tcp_disconnect(struct sock *sk, int flags); 442 443 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 444 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 445 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 446 447 /* From syncookies.c */ 448 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, 449 struct request_sock *req, 450 struct dst_entry *dst, u32 tsoff); 451 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 452 u32 cookie); 453 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 454 #ifdef CONFIG_SYN_COOKIES 455 456 /* Syncookies use a monotonic timer which increments every 60 seconds. 457 * This counter is used both as a hash input and partially encoded into 458 * the cookie value. A cookie is only validated further if the delta 459 * between the current counter value and the encoded one is less than this, 460 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 461 * the counter advances immediately after a cookie is generated). 462 */ 463 #define MAX_SYNCOOKIE_AGE 2 464 #define TCP_SYNCOOKIE_PERIOD (60 * HZ) 465 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) 466 467 /* syncookies: remember time of last synqueue overflow 468 * But do not dirty this field too often (once per second is enough) 469 * It is racy as we do not hold a lock, but race is very minor. 470 */ 471 static inline void tcp_synq_overflow(const struct sock *sk) 472 { 473 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 474 unsigned long now = jiffies; 475 476 if (time_after(now, last_overflow + HZ)) 477 tcp_sk(sk)->rx_opt.ts_recent_stamp = now; 478 } 479 480 /* syncookies: no recent synqueue overflow on this listening socket? */ 481 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 482 { 483 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 484 485 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID); 486 } 487 488 static inline u32 tcp_cookie_time(void) 489 { 490 u64 val = get_jiffies_64(); 491 492 do_div(val, TCP_SYNCOOKIE_PERIOD); 493 return val; 494 } 495 496 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 497 u16 *mssp); 498 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); 499 u64 cookie_init_timestamp(struct request_sock *req); 500 bool cookie_timestamp_decode(const struct net *net, 501 struct tcp_options_received *opt); 502 bool cookie_ecn_ok(const struct tcp_options_received *opt, 503 const struct net *net, const struct dst_entry *dst); 504 505 /* From net/ipv6/syncookies.c */ 506 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 507 u32 cookie); 508 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 509 510 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 511 const struct tcphdr *th, u16 *mssp); 512 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); 513 #endif 514 /* tcp_output.c */ 515 516 u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now, 517 int min_tso_segs); 518 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 519 int nonagle); 520 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 521 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); 522 void tcp_retransmit_timer(struct sock *sk); 523 void tcp_xmit_retransmit_queue(struct sock *); 524 void tcp_simple_retransmit(struct sock *); 525 void tcp_enter_recovery(struct sock *sk, bool ece_ack); 526 int tcp_trim_head(struct sock *, struct sk_buff *, u32); 527 enum tcp_queue { 528 TCP_FRAG_IN_WRITE_QUEUE, 529 TCP_FRAG_IN_RTX_QUEUE, 530 }; 531 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, 532 struct sk_buff *skb, u32 len, 533 unsigned int mss_now, gfp_t gfp); 534 535 void tcp_send_probe0(struct sock *); 536 void tcp_send_partial(struct sock *); 537 int tcp_write_wakeup(struct sock *, int mib); 538 void tcp_send_fin(struct sock *sk); 539 void tcp_send_active_reset(struct sock *sk, gfp_t priority); 540 int tcp_send_synack(struct sock *); 541 void tcp_push_one(struct sock *, unsigned int mss_now); 542 void tcp_send_ack(struct sock *sk); 543 void tcp_send_delayed_ack(struct sock *sk); 544 void tcp_send_loss_probe(struct sock *sk); 545 bool tcp_schedule_loss_probe(struct sock *sk); 546 void tcp_skb_collapse_tstamp(struct sk_buff *skb, 547 const struct sk_buff *next_skb); 548 549 /* tcp_input.c */ 550 void tcp_rearm_rto(struct sock *sk); 551 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); 552 void tcp_reset(struct sock *sk); 553 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); 554 void tcp_fin(struct sock *sk); 555 556 /* tcp_timer.c */ 557 void tcp_init_xmit_timers(struct sock *); 558 static inline void tcp_clear_xmit_timers(struct sock *sk) 559 { 560 hrtimer_cancel(&tcp_sk(sk)->pacing_timer); 561 inet_csk_clear_xmit_timers(sk); 562 } 563 564 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 565 unsigned int tcp_current_mss(struct sock *sk); 566 567 /* Bound MSS / TSO packet size with the half of the window */ 568 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 569 { 570 int cutoff; 571 572 /* When peer uses tiny windows, there is no use in packetizing 573 * to sub-MSS pieces for the sake of SWS or making sure there 574 * are enough packets in the pipe for fast recovery. 575 * 576 * On the other hand, for extremely large MSS devices, handling 577 * smaller than MSS windows in this way does make sense. 578 */ 579 if (tp->max_window > TCP_MSS_DEFAULT) 580 cutoff = (tp->max_window >> 1); 581 else 582 cutoff = tp->max_window; 583 584 if (cutoff && pktsize > cutoff) 585 return max_t(int, cutoff, 68U - tp->tcp_header_len); 586 else 587 return pktsize; 588 } 589 590 /* tcp.c */ 591 void tcp_get_info(struct sock *, struct tcp_info *); 592 593 /* Read 'sendfile()'-style from a TCP socket */ 594 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 595 sk_read_actor_t recv_actor); 596 597 void tcp_initialize_rcv_mss(struct sock *sk); 598 599 int tcp_mtu_to_mss(struct sock *sk, int pmtu); 600 int tcp_mss_to_mtu(struct sock *sk, int mss); 601 void tcp_mtup_init(struct sock *sk); 602 void tcp_init_buffer_space(struct sock *sk); 603 604 static inline void tcp_bound_rto(const struct sock *sk) 605 { 606 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 607 inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 608 } 609 610 static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 611 { 612 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 613 } 614 615 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 616 { 617 tp->pred_flags = htonl((tp->tcp_header_len << 26) | 618 ntohl(TCP_FLAG_ACK) | 619 snd_wnd); 620 } 621 622 static inline void tcp_fast_path_on(struct tcp_sock *tp) 623 { 624 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 625 } 626 627 static inline void tcp_fast_path_check(struct sock *sk) 628 { 629 struct tcp_sock *tp = tcp_sk(sk); 630 631 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && 632 tp->rcv_wnd && 633 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 634 !tp->urg_data) 635 tcp_fast_path_on(tp); 636 } 637 638 /* Compute the actual rto_min value */ 639 static inline u32 tcp_rto_min(struct sock *sk) 640 { 641 const struct dst_entry *dst = __sk_dst_get(sk); 642 u32 rto_min = TCP_RTO_MIN; 643 644 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 645 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 646 return rto_min; 647 } 648 649 static inline u32 tcp_rto_min_us(struct sock *sk) 650 { 651 return jiffies_to_usecs(tcp_rto_min(sk)); 652 } 653 654 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 655 { 656 return dst_metric_locked(dst, RTAX_CC_ALGO); 657 } 658 659 /* Minimum RTT in usec. ~0 means not available. */ 660 static inline u32 tcp_min_rtt(const struct tcp_sock *tp) 661 { 662 return minmax_get(&tp->rtt_min); 663 } 664 665 /* Compute the actual receive window we are currently advertising. 666 * Rcv_nxt can be after the window if our peer push more data 667 * than the offered window. 668 */ 669 static inline u32 tcp_receive_window(const struct tcp_sock *tp) 670 { 671 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 672 673 if (win < 0) 674 win = 0; 675 return (u32) win; 676 } 677 678 /* Choose a new window, without checks for shrinking, and without 679 * scaling applied to the result. The caller does these things 680 * if necessary. This is a "raw" window selection. 681 */ 682 u32 __tcp_select_window(struct sock *sk); 683 684 void tcp_send_window_probe(struct sock *sk); 685 686 /* TCP uses 32bit jiffies to save some space. 687 * Note that this is different from tcp_time_stamp, which 688 * historically has been the same until linux-4.13. 689 */ 690 #define tcp_jiffies32 ((u32)jiffies) 691 692 /* 693 * Deliver a 32bit value for TCP timestamp option (RFC 7323) 694 * It is no longer tied to jiffies, but to 1 ms clock. 695 * Note: double check if you want to use tcp_jiffies32 instead of this. 696 */ 697 #define TCP_TS_HZ 1000 698 699 static inline u64 tcp_clock_ns(void) 700 { 701 return local_clock(); 702 } 703 704 static inline u64 tcp_clock_us(void) 705 { 706 return div_u64(tcp_clock_ns(), NSEC_PER_USEC); 707 } 708 709 /* This should only be used in contexts where tp->tcp_mstamp is up to date */ 710 static inline u32 tcp_time_stamp(const struct tcp_sock *tp) 711 { 712 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); 713 } 714 715 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ 716 static inline u32 tcp_time_stamp_raw(void) 717 { 718 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ); 719 } 720 721 722 /* Refresh 1us clock of a TCP socket, 723 * ensuring monotically increasing values. 724 */ 725 static inline void tcp_mstamp_refresh(struct tcp_sock *tp) 726 { 727 u64 val = tcp_clock_us(); 728 729 if (val > tp->tcp_mstamp) 730 tp->tcp_mstamp = val; 731 } 732 733 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) 734 { 735 return max_t(s64, t1 - t0, 0); 736 } 737 738 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 739 { 740 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ); 741 } 742 743 744 #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 745 746 #define TCPHDR_FIN 0x01 747 #define TCPHDR_SYN 0x02 748 #define TCPHDR_RST 0x04 749 #define TCPHDR_PSH 0x08 750 #define TCPHDR_ACK 0x10 751 #define TCPHDR_URG 0x20 752 #define TCPHDR_ECE 0x40 753 #define TCPHDR_CWR 0x80 754 755 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) 756 757 /* This is what the send packet queuing engine uses to pass 758 * TCP per-packet control information to the transmission code. 759 * We also store the host-order sequence numbers in here too. 760 * This is 44 bytes if IPV6 is enabled. 761 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 762 */ 763 struct tcp_skb_cb { 764 __u32 seq; /* Starting sequence number */ 765 __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 766 union { 767 /* Note : tcp_tw_isn is used in input path only 768 * (isn chosen by tcp_timewait_state_process()) 769 * 770 * tcp_gso_segs/size are used in write queue only, 771 * cf tcp_skb_pcount()/tcp_skb_mss() 772 */ 773 __u32 tcp_tw_isn; 774 struct { 775 u16 tcp_gso_segs; 776 u16 tcp_gso_size; 777 }; 778 }; 779 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 780 781 __u8 sacked; /* State flags for SACK/FACK. */ 782 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 783 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 784 #define TCPCB_LOST 0x04 /* SKB is lost */ 785 #define TCPCB_TAGBITS 0x07 /* All tag bits */ 786 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 787 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 788 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 789 TCPCB_REPAIRED) 790 791 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 792 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ 793 eor:1, /* Is skb MSG_EOR marked? */ 794 has_rxtstamp:1, /* SKB has a RX timestamp */ 795 unused:5; 796 __u32 ack_seq; /* Sequence number ACK'd */ 797 union { 798 struct { 799 /* There is space for up to 24 bytes */ 800 __u32 in_flight:30,/* Bytes in flight at transmit */ 801 is_app_limited:1, /* cwnd not fully used? */ 802 unused:1; 803 /* pkts S/ACKed so far upon tx of skb, incl retrans: */ 804 __u32 delivered; 805 /* start of send pipeline phase */ 806 u64 first_tx_mstamp; 807 /* when we reached the "delivered" count */ 808 u64 delivered_mstamp; 809 } tx; /* only used for outgoing skbs */ 810 union { 811 struct inet_skb_parm h4; 812 #if IS_ENABLED(CONFIG_IPV6) 813 struct inet6_skb_parm h6; 814 #endif 815 } header; /* For incoming skbs */ 816 struct { 817 __u32 key; 818 __u32 flags; 819 struct bpf_map *map; 820 void *data_end; 821 } bpf; 822 }; 823 }; 824 825 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 826 827 828 #if IS_ENABLED(CONFIG_IPV6) 829 /* This is the variant of inet6_iif() that must be used by TCP, 830 * as TCP moves IP6CB into a different location in skb->cb[] 831 */ 832 static inline int tcp_v6_iif(const struct sk_buff *skb) 833 { 834 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); 835 836 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; 837 } 838 839 /* TCP_SKB_CB reference means this can not be used from early demux */ 840 static inline int tcp_v6_sdif(const struct sk_buff *skb) 841 { 842 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 843 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) 844 return TCP_SKB_CB(skb)->header.h6.iif; 845 #endif 846 return 0; 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 /* TCP_SKB_CB reference means this can not be used from early demux */ 862 static inline int tcp_v4_sdif(struct sk_buff *skb) 863 { 864 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) 865 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) 866 return TCP_SKB_CB(skb)->header.h4.iif; 867 #endif 868 return 0; 869 } 870 871 /* Due to TSO, an SKB can be composed of multiple actual 872 * packets. To keep these tracked properly, we use this. 873 */ 874 static inline int tcp_skb_pcount(const struct sk_buff *skb) 875 { 876 return TCP_SKB_CB(skb)->tcp_gso_segs; 877 } 878 879 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 880 { 881 TCP_SKB_CB(skb)->tcp_gso_segs = segs; 882 } 883 884 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 885 { 886 TCP_SKB_CB(skb)->tcp_gso_segs += segs; 887 } 888 889 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ 890 static inline int tcp_skb_mss(const struct sk_buff *skb) 891 { 892 return TCP_SKB_CB(skb)->tcp_gso_size; 893 } 894 895 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) 896 { 897 return likely(!TCP_SKB_CB(skb)->eor); 898 } 899 900 /* Events passed to congestion control interface */ 901 enum tcp_ca_event { 902 CA_EVENT_TX_START, /* first transmit when no packets in flight */ 903 CA_EVENT_CWND_RESTART, /* congestion window restart */ 904 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 905 CA_EVENT_LOSS, /* loss timeout */ 906 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 907 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 908 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 909 CA_EVENT_NON_DELAYED_ACK, 910 }; 911 912 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 913 enum tcp_ca_ack_event_flags { 914 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 915 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 916 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 917 }; 918 919 /* 920 * Interface for adding new TCP congestion control handlers 921 */ 922 #define TCP_CA_NAME_MAX 16 923 #define TCP_CA_MAX 128 924 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 925 926 #define TCP_CA_UNSPEC 0 927 928 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 929 #define TCP_CONG_NON_RESTRICTED 0x1 930 /* Requires ECN/ECT set on all packets */ 931 #define TCP_CONG_NEEDS_ECN 0x2 932 933 union tcp_cc_info; 934 935 struct ack_sample { 936 u32 pkts_acked; 937 s32 rtt_us; 938 u32 in_flight; 939 }; 940 941 /* A rate sample measures the number of (original/retransmitted) data 942 * packets delivered "delivered" over an interval of time "interval_us". 943 * The tcp_rate.c code fills in the rate sample, and congestion 944 * control modules that define a cong_control function to run at the end 945 * of ACK processing can optionally chose to consult this sample when 946 * setting cwnd and pacing rate. 947 * A sample is invalid if "delivered" or "interval_us" is negative. 948 */ 949 struct rate_sample { 950 u64 prior_mstamp; /* starting timestamp for interval */ 951 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ 952 s32 delivered; /* number of packets delivered over interval */ 953 long interval_us; /* time for tp->delivered to incr "delivered" */ 954 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ 955 int losses; /* number of packets marked lost upon ACK */ 956 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ 957 u32 prior_in_flight; /* in flight before this ACK */ 958 bool is_app_limited; /* is sample from packet with bubble in pipe? */ 959 bool is_retrans; /* is sample from retransmission? */ 960 }; 961 962 struct tcp_congestion_ops { 963 struct list_head list; 964 u32 key; 965 u32 flags; 966 967 /* initialize private data (optional) */ 968 void (*init)(struct sock *sk); 969 /* cleanup private data (optional) */ 970 void (*release)(struct sock *sk); 971 972 /* return slow start threshold (required) */ 973 u32 (*ssthresh)(struct sock *sk); 974 /* do new cwnd calculation (required) */ 975 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 976 /* call before changing ca_state (optional) */ 977 void (*set_state)(struct sock *sk, u8 new_state); 978 /* call when cwnd event occurs (optional) */ 979 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 980 /* call when ack arrives (optional) */ 981 void (*in_ack_event)(struct sock *sk, u32 flags); 982 /* new value of cwnd after loss (required) */ 983 u32 (*undo_cwnd)(struct sock *sk); 984 /* hook for packet ack accounting (optional) */ 985 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); 986 /* suggest number of segments for each skb to transmit (optional) */ 987 u32 (*tso_segs_goal)(struct sock *sk); 988 /* returns the multiplier used in tcp_sndbuf_expand (optional) */ 989 u32 (*sndbuf_expand)(struct sock *sk); 990 /* call when packets are delivered to update cwnd and pacing rate, 991 * after all the ca_state processing. (optional) 992 */ 993 void (*cong_control)(struct sock *sk, const struct rate_sample *rs); 994 /* get info for inet_diag (optional) */ 995 size_t (*get_info)(struct sock *sk, u32 ext, int *attr, 996 union tcp_cc_info *info); 997 998 char name[TCP_CA_NAME_MAX]; 999 struct module *owner; 1000 }; 1001 1002 int tcp_register_congestion_control(struct tcp_congestion_ops *type); 1003 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 1004 1005 void tcp_assign_congestion_control(struct sock *sk); 1006 void tcp_init_congestion_control(struct sock *sk); 1007 void tcp_cleanup_congestion_control(struct sock *sk); 1008 int tcp_set_default_congestion_control(const char *name); 1009 void tcp_get_default_congestion_control(char *name); 1010 void tcp_get_available_congestion_control(char *buf, size_t len); 1011 void tcp_get_allowed_congestion_control(char *buf, size_t len); 1012 int tcp_set_allowed_congestion_control(char *allowed); 1013 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit); 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 (!sock_net(sk)->ipv4.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(const struct sock *sk, int __space, 1295 __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(const struct sock *sk, int space) 1300 { 1301 int tcp_adv_win_scale = sock_net(sk)->ipv4.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->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->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 void tcp_fastopen_destroy_cipher(struct sock *sk); 1539 void tcp_fastopen_ctx_destroy(struct net *net); 1540 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, 1541 void *key, unsigned int len); 1542 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); 1543 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 1544 struct request_sock *req, 1545 struct tcp_fastopen_cookie *foc, 1546 const struct dst_entry *dst); 1547 void tcp_fastopen_init_key_once(struct net *net); 1548 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, 1549 struct tcp_fastopen_cookie *cookie); 1550 bool tcp_fastopen_defer_connect(struct sock *sk, int *err); 1551 #define TCP_FASTOPEN_KEY_LENGTH 16 1552 1553 /* Fastopen key context */ 1554 struct tcp_fastopen_context { 1555 struct crypto_cipher *tfm; 1556 __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 1557 struct rcu_head rcu; 1558 }; 1559 1560 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; 1561 void tcp_fastopen_active_disable(struct sock *sk); 1562 bool tcp_fastopen_active_should_disable(struct sock *sk); 1563 void tcp_fastopen_active_disable_ofo_check(struct sock *sk); 1564 void tcp_fastopen_active_timeout_reset(void); 1565 1566 /* Latencies incurred by various limits for a sender. They are 1567 * chronograph-like stats that are mutually exclusive. 1568 */ 1569 enum tcp_chrono { 1570 TCP_CHRONO_UNSPEC, 1571 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ 1572 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ 1573 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ 1574 __TCP_CHRONO_MAX, 1575 }; 1576 1577 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); 1578 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); 1579 1580 /* This helper is needed, because skb->tcp_tsorted_anchor uses 1581 * the same memory storage than skb->destructor/_skb_refdst 1582 */ 1583 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) 1584 { 1585 skb->destructor = NULL; 1586 skb->_skb_refdst = 0UL; 1587 } 1588 1589 #define tcp_skb_tsorted_save(skb) { \ 1590 unsigned long _save = skb->_skb_refdst; \ 1591 skb->_skb_refdst = 0UL; 1592 1593 #define tcp_skb_tsorted_restore(skb) \ 1594 skb->_skb_refdst = _save; \ 1595 } 1596 1597 void tcp_write_queue_purge(struct sock *sk); 1598 1599 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) 1600 { 1601 return skb_rb_first(&sk->tcp_rtx_queue); 1602 } 1603 1604 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1605 { 1606 return skb_peek(&sk->sk_write_queue); 1607 } 1608 1609 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1610 { 1611 return skb_peek_tail(&sk->sk_write_queue); 1612 } 1613 1614 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1615 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1616 1617 static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1618 { 1619 return skb_peek(&sk->sk_write_queue); 1620 } 1621 1622 static inline bool tcp_skb_is_last(const struct sock *sk, 1623 const struct sk_buff *skb) 1624 { 1625 return skb_queue_is_last(&sk->sk_write_queue, skb); 1626 } 1627 1628 static inline bool tcp_write_queue_empty(const struct sock *sk) 1629 { 1630 return skb_queue_empty(&sk->sk_write_queue); 1631 } 1632 1633 static inline bool tcp_rtx_queue_empty(const struct sock *sk) 1634 { 1635 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); 1636 } 1637 1638 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) 1639 { 1640 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); 1641 } 1642 1643 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1644 { 1645 if (tcp_write_queue_empty(sk)) 1646 tcp_chrono_stop(sk, TCP_CHRONO_BUSY); 1647 1648 if (tcp_sk(sk)->highest_sack == skb_unlinked) 1649 tcp_sk(sk)->highest_sack = NULL; 1650 } 1651 1652 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1653 { 1654 __skb_queue_tail(&sk->sk_write_queue, skb); 1655 } 1656 1657 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1658 { 1659 __tcp_add_write_queue_tail(sk, skb); 1660 1661 /* Queue it, remembering where we must start sending. */ 1662 if (sk->sk_write_queue.next == skb) { 1663 tcp_chrono_start(sk, TCP_CHRONO_BUSY); 1664 1665 if (tcp_sk(sk)->highest_sack == NULL) 1666 tcp_sk(sk)->highest_sack = skb; 1667 } 1668 } 1669 1670 /* Insert new before skb on the write queue of sk. */ 1671 static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1672 struct sk_buff *skb, 1673 struct sock *sk) 1674 { 1675 __skb_queue_before(&sk->sk_write_queue, skb, new); 1676 } 1677 1678 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1679 { 1680 tcp_skb_tsorted_anchor_cleanup(skb); 1681 __skb_unlink(skb, &sk->sk_write_queue); 1682 } 1683 1684 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); 1685 1686 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) 1687 { 1688 tcp_skb_tsorted_anchor_cleanup(skb); 1689 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); 1690 } 1691 1692 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) 1693 { 1694 list_del(&skb->tcp_tsorted_anchor); 1695 tcp_rtx_queue_unlink(skb, sk); 1696 sk_wmem_free_skb(sk, skb); 1697 } 1698 1699 static inline void tcp_push_pending_frames(struct sock *sk) 1700 { 1701 if (tcp_send_head(sk)) { 1702 struct tcp_sock *tp = tcp_sk(sk); 1703 1704 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 1705 } 1706 } 1707 1708 /* Start sequence of the skb just after the highest skb with SACKed 1709 * bit, valid only if sacked_out > 0 or when the caller has ensured 1710 * validity by itself. 1711 */ 1712 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1713 { 1714 if (!tp->sacked_out) 1715 return tp->snd_una; 1716 1717 if (tp->highest_sack == NULL) 1718 return tp->snd_nxt; 1719 1720 return TCP_SKB_CB(tp->highest_sack)->seq; 1721 } 1722 1723 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 1724 { 1725 struct sk_buff *next = skb_rb_next(skb); 1726 1727 tcp_sk(sk)->highest_sack = next ?: tcp_send_head(sk); 1728 } 1729 1730 static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 1731 { 1732 return tcp_sk(sk)->highest_sack; 1733 } 1734 1735 static inline void tcp_highest_sack_reset(struct sock *sk) 1736 { 1737 struct sk_buff *skb = tcp_rtx_queue_head(sk); 1738 1739 tcp_sk(sk)->highest_sack = skb ?: tcp_send_head(sk); 1740 } 1741 1742 /* Called when old skb is about to be deleted and replaced by new skb */ 1743 static inline void tcp_highest_sack_replace(struct sock *sk, 1744 struct sk_buff *old, 1745 struct sk_buff *new) 1746 { 1747 if (old == tcp_highest_sack(sk)) 1748 tcp_sk(sk)->highest_sack = new; 1749 } 1750 1751 /* This helper checks if socket has IP_TRANSPARENT set */ 1752 static inline bool inet_sk_transparent(const struct sock *sk) 1753 { 1754 switch (sk->sk_state) { 1755 case TCP_TIME_WAIT: 1756 return inet_twsk(sk)->tw_transparent; 1757 case TCP_NEW_SYN_RECV: 1758 return inet_rsk(inet_reqsk(sk))->no_srccheck; 1759 } 1760 return inet_sk(sk)->transparent; 1761 } 1762 1763 /* Determines whether this is a thin stream (which may suffer from 1764 * increased latency). Used to trigger latency-reducing mechanisms. 1765 */ 1766 static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 1767 { 1768 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 1769 } 1770 1771 /* /proc */ 1772 enum tcp_seq_states { 1773 TCP_SEQ_STATE_LISTENING, 1774 TCP_SEQ_STATE_ESTABLISHED, 1775 }; 1776 1777 int tcp_seq_open(struct inode *inode, struct file *file); 1778 1779 struct tcp_seq_afinfo { 1780 char *name; 1781 sa_family_t family; 1782 const struct file_operations *seq_fops; 1783 struct seq_operations seq_ops; 1784 }; 1785 1786 struct tcp_iter_state { 1787 struct seq_net_private p; 1788 sa_family_t family; 1789 enum tcp_seq_states state; 1790 struct sock *syn_wait_sk; 1791 int bucket, offset, sbucket, num; 1792 loff_t last_pos; 1793 }; 1794 1795 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 1796 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 1797 1798 extern struct request_sock_ops tcp_request_sock_ops; 1799 extern struct request_sock_ops tcp6_request_sock_ops; 1800 1801 void tcp_v4_destroy_sock(struct sock *sk); 1802 1803 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1804 netdev_features_t features); 1805 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 1806 int tcp_gro_complete(struct sk_buff *skb); 1807 1808 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1809 1810 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1811 { 1812 struct net *net = sock_net((struct sock *)tp); 1813 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; 1814 } 1815 1816 static inline bool tcp_stream_memory_free(const struct sock *sk) 1817 { 1818 const struct tcp_sock *tp = tcp_sk(sk); 1819 u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1820 1821 return notsent_bytes < tcp_notsent_lowat(tp); 1822 } 1823 1824 #ifdef CONFIG_PROC_FS 1825 int tcp4_proc_init(void); 1826 void tcp4_proc_exit(void); 1827 #endif 1828 1829 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); 1830 int tcp_conn_request(struct request_sock_ops *rsk_ops, 1831 const struct tcp_request_sock_ops *af_ops, 1832 struct sock *sk, struct sk_buff *skb); 1833 1834 /* TCP af-specific functions */ 1835 struct tcp_sock_af_ops { 1836 #ifdef CONFIG_TCP_MD5SIG 1837 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, 1838 const struct sock *addr_sk); 1839 int (*calc_md5_hash)(char *location, 1840 const struct tcp_md5sig_key *md5, 1841 const struct sock *sk, 1842 const struct sk_buff *skb); 1843 int (*md5_parse)(struct sock *sk, 1844 int optname, 1845 char __user *optval, 1846 int optlen); 1847 #endif 1848 }; 1849 1850 struct tcp_request_sock_ops { 1851 u16 mss_clamp; 1852 #ifdef CONFIG_TCP_MD5SIG 1853 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, 1854 const struct sock *addr_sk); 1855 int (*calc_md5_hash) (char *location, 1856 const struct tcp_md5sig_key *md5, 1857 const struct sock *sk, 1858 const struct sk_buff *skb); 1859 #endif 1860 void (*init_req)(struct request_sock *req, 1861 const struct sock *sk_listener, 1862 struct sk_buff *skb); 1863 #ifdef CONFIG_SYN_COOKIES 1864 __u32 (*cookie_init_seq)(const struct sk_buff *skb, 1865 __u16 *mss); 1866 #endif 1867 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, 1868 const struct request_sock *req); 1869 u32 (*init_seq)(const struct sk_buff *skb); 1870 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); 1871 int (*send_synack)(const struct sock *sk, struct dst_entry *dst, 1872 struct flowi *fl, struct request_sock *req, 1873 struct tcp_fastopen_cookie *foc, 1874 enum tcp_synack_type synack_type); 1875 }; 1876 1877 #ifdef CONFIG_SYN_COOKIES 1878 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1879 const struct sock *sk, struct sk_buff *skb, 1880 __u16 *mss) 1881 { 1882 tcp_synq_overflow(sk); 1883 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 1884 return ops->cookie_init_seq(skb, mss); 1885 } 1886 #else 1887 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1888 const struct sock *sk, struct sk_buff *skb, 1889 __u16 *mss) 1890 { 1891 return 0; 1892 } 1893 #endif 1894 1895 int tcpv4_offload_init(void); 1896 1897 void tcp_v4_init(void); 1898 void tcp_init(void); 1899 1900 /* tcp_recovery.c */ 1901 extern void tcp_rack_mark_lost(struct sock *sk); 1902 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, 1903 u64 xmit_time); 1904 extern void tcp_rack_reo_timeout(struct sock *sk); 1905 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); 1906 1907 /* At how many usecs into the future should the RTO fire? */ 1908 static inline s64 tcp_rto_delta_us(const struct sock *sk) 1909 { 1910 const struct sk_buff *skb = tcp_rtx_queue_head(sk); 1911 u32 rto = inet_csk(sk)->icsk_rto; 1912 u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto); 1913 1914 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; 1915 } 1916 1917 /* 1918 * Save and compile IPv4 options, return a pointer to it 1919 */ 1920 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, 1921 struct sk_buff *skb) 1922 { 1923 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1924 struct ip_options_rcu *dopt = NULL; 1925 1926 if (opt->optlen) { 1927 int opt_size = sizeof(*dopt) + opt->optlen; 1928 1929 dopt = kmalloc(opt_size, GFP_ATOMIC); 1930 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { 1931 kfree(dopt); 1932 dopt = NULL; 1933 } 1934 } 1935 return dopt; 1936 } 1937 1938 /* locally generated TCP pure ACKs have skb->truesize == 2 1939 * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 1940 * This is much faster than dissecting the packet to find out. 1941 * (Think of GRE encapsulations, IPv4, IPv6, ...) 1942 */ 1943 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 1944 { 1945 return skb->truesize == 2; 1946 } 1947 1948 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 1949 { 1950 skb->truesize = 2; 1951 } 1952 1953 static inline int tcp_inq(struct sock *sk) 1954 { 1955 struct tcp_sock *tp = tcp_sk(sk); 1956 int answ; 1957 1958 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1959 answ = 0; 1960 } else if (sock_flag(sk, SOCK_URGINLINE) || 1961 !tp->urg_data || 1962 before(tp->urg_seq, tp->copied_seq) || 1963 !before(tp->urg_seq, tp->rcv_nxt)) { 1964 1965 answ = tp->rcv_nxt - tp->copied_seq; 1966 1967 /* Subtract 1, if FIN was received */ 1968 if (answ && sock_flag(sk, SOCK_DONE)) 1969 answ--; 1970 } else { 1971 answ = tp->urg_seq - tp->copied_seq; 1972 } 1973 1974 return answ; 1975 } 1976 1977 int tcp_peek_len(struct socket *sock); 1978 1979 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) 1980 { 1981 u16 segs_in; 1982 1983 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 1984 tp->segs_in += segs_in; 1985 if (skb->len > tcp_hdrlen(skb)) 1986 tp->data_segs_in += segs_in; 1987 } 1988 1989 /* 1990 * TCP listen path runs lockless. 1991 * We forced "struct sock" to be const qualified to make sure 1992 * we don't modify one of its field by mistake. 1993 * Here, we increment sk_drops which is an atomic_t, so we can safely 1994 * make sock writable again. 1995 */ 1996 static inline void tcp_listendrop(const struct sock *sk) 1997 { 1998 atomic_inc(&((struct sock *)sk)->sk_drops); 1999 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); 2000 } 2001 2002 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); 2003 2004 /* 2005 * Interface for adding Upper Level Protocols over TCP 2006 */ 2007 2008 #define TCP_ULP_NAME_MAX 16 2009 #define TCP_ULP_MAX 128 2010 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) 2011 2012 struct tcp_ulp_ops { 2013 struct list_head list; 2014 2015 /* initialize ulp */ 2016 int (*init)(struct sock *sk); 2017 /* cleanup ulp */ 2018 void (*release)(struct sock *sk); 2019 2020 char name[TCP_ULP_NAME_MAX]; 2021 struct module *owner; 2022 }; 2023 int tcp_register_ulp(struct tcp_ulp_ops *type); 2024 void tcp_unregister_ulp(struct tcp_ulp_ops *type); 2025 int tcp_set_ulp(struct sock *sk, const char *name); 2026 void tcp_get_available_ulp(char *buf, size_t len); 2027 void tcp_cleanup_ulp(struct sock *sk); 2028 2029 /* Call BPF_SOCK_OPS program that returns an int. If the return value 2030 * is < 0, then the BPF op failed (for example if the loaded BPF 2031 * program does not support the chosen operation or there is no BPF 2032 * program loaded). 2033 */ 2034 #ifdef CONFIG_BPF 2035 static inline int tcp_call_bpf(struct sock *sk, int op) 2036 { 2037 struct bpf_sock_ops_kern sock_ops; 2038 int ret; 2039 2040 if (sk_fullsock(sk)) 2041 sock_owned_by_me(sk); 2042 2043 memset(&sock_ops, 0, sizeof(sock_ops)); 2044 sock_ops.sk = sk; 2045 sock_ops.op = op; 2046 2047 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); 2048 if (ret == 0) 2049 ret = sock_ops.reply; 2050 else 2051 ret = -1; 2052 return ret; 2053 } 2054 #else 2055 static inline int tcp_call_bpf(struct sock *sk, int op) 2056 { 2057 return -EPERM; 2058 } 2059 #endif 2060 2061 static inline u32 tcp_timeout_init(struct sock *sk) 2062 { 2063 int timeout; 2064 2065 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT); 2066 2067 if (timeout <= 0) 2068 timeout = TCP_TIMEOUT_INIT; 2069 return timeout; 2070 } 2071 2072 static inline u32 tcp_rwnd_init_bpf(struct sock *sk) 2073 { 2074 int rwnd; 2075 2076 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT); 2077 2078 if (rwnd < 0) 2079 rwnd = 0; 2080 return rwnd; 2081 } 2082 2083 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) 2084 { 2085 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN) == 1); 2086 } 2087 2088 #if IS_ENABLED(CONFIG_SMC) 2089 extern struct static_key_false tcp_have_smc; 2090 #endif 2091 #endif /* _TCP_H */ 2092